Index: head/sys/amd64/amd64/pmap.c =================================================================== --- head/sys/amd64/amd64/pmap.c (revision 298481) +++ head/sys/amd64/amd64/pmap.c (revision 298482) @@ -1,7147 +1,7147 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1994 John S. Dyson * All rights reserved. * Copyright (c) 1994 David Greenman * All rights reserved. * Copyright (c) 2003 Peter Wemm * All rights reserved. * Copyright (c) 2005-2010 Alan L. Cox * All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department and William Jolitz of UUNET Technologies Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)pmap.c 7.7 (Berkeley) 5/12/91 */ /*- * Copyright (c) 2003 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Jake Burkholder, * Safeport Network Services, and Network Associates Laboratories, the * Security Research Division of Network Associates, Inc. under * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA * CHATS research program. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #define AMD64_NPT_AWARE #include __FBSDID("$FreeBSD$"); /* * Manages physical address maps. * * Since the information managed by this module is * also stored by the logical address mapping module, * this module may throw away valid virtual-to-physical * mappings at almost any time. However, invalidations * of virtual-to-physical mappings must be done as * requested. * * In order to cope with hardware architectures which * make virtual-to-physical map invalidates expensive, * this module may delay invalidate or reduced protection * operations until such time as they are actually * necessary. This module is given full information as * to which processors are currently using which maps, * and to when physical maps must be made correct. */ #include "opt_pmap.h" #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SMP #include #endif static __inline boolean_t pmap_type_guest(pmap_t pmap) { return ((pmap->pm_type == PT_EPT) || (pmap->pm_type == PT_RVI)); } static __inline boolean_t pmap_emulate_ad_bits(pmap_t pmap) { return ((pmap->pm_flags & PMAP_EMULATE_AD_BITS) != 0); } static __inline pt_entry_t pmap_valid_bit(pmap_t pmap) { pt_entry_t mask; switch (pmap->pm_type) { case PT_X86: case PT_RVI: mask = X86_PG_V; break; case PT_EPT: if (pmap_emulate_ad_bits(pmap)) mask = EPT_PG_EMUL_V; else mask = EPT_PG_READ; break; default: panic("pmap_valid_bit: invalid pm_type %d", pmap->pm_type); } return (mask); } static __inline pt_entry_t pmap_rw_bit(pmap_t pmap) { pt_entry_t mask; switch (pmap->pm_type) { case PT_X86: case PT_RVI: mask = X86_PG_RW; break; case PT_EPT: if (pmap_emulate_ad_bits(pmap)) mask = EPT_PG_EMUL_RW; else mask = EPT_PG_WRITE; break; default: panic("pmap_rw_bit: invalid pm_type %d", pmap->pm_type); } return (mask); } static __inline pt_entry_t pmap_global_bit(pmap_t pmap) { pt_entry_t mask; switch (pmap->pm_type) { case PT_X86: mask = X86_PG_G; break; case PT_RVI: case PT_EPT: mask = 0; break; default: panic("pmap_global_bit: invalid pm_type %d", pmap->pm_type); } return (mask); } static __inline pt_entry_t pmap_accessed_bit(pmap_t pmap) { pt_entry_t mask; switch (pmap->pm_type) { case PT_X86: case PT_RVI: mask = X86_PG_A; break; case PT_EPT: if (pmap_emulate_ad_bits(pmap)) mask = EPT_PG_READ; else mask = EPT_PG_A; break; default: panic("pmap_accessed_bit: invalid pm_type %d", pmap->pm_type); } return (mask); } static __inline pt_entry_t pmap_modified_bit(pmap_t pmap) { pt_entry_t mask; switch (pmap->pm_type) { case PT_X86: case PT_RVI: mask = X86_PG_M; break; case PT_EPT: if (pmap_emulate_ad_bits(pmap)) mask = EPT_PG_WRITE; else mask = EPT_PG_M; break; default: panic("pmap_modified_bit: invalid pm_type %d", pmap->pm_type); } return (mask); } extern struct pcpu __pcpu[]; #if !defined(DIAGNOSTIC) #ifdef __GNUC_GNU_INLINE__ #define PMAP_INLINE __attribute__((__gnu_inline__)) inline #else #define PMAP_INLINE extern inline #endif #else #define PMAP_INLINE #endif #ifdef PV_STATS #define PV_STAT(x) do { x ; } while (0) #else #define PV_STAT(x) do { } while (0) #endif #define pa_index(pa) ((pa) >> PDRSHIFT) #define pa_to_pvh(pa) (&pv_table[pa_index(pa)]) #define NPV_LIST_LOCKS MAXCPU #define PHYS_TO_PV_LIST_LOCK(pa) \ (&pv_list_locks[pa_index(pa) % NPV_LIST_LOCKS]) #define CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa) do { \ struct rwlock **_lockp = (lockp); \ struct rwlock *_new_lock; \ \ _new_lock = PHYS_TO_PV_LIST_LOCK(pa); \ if (_new_lock != *_lockp) { \ if (*_lockp != NULL) \ rw_wunlock(*_lockp); \ *_lockp = _new_lock; \ rw_wlock(*_lockp); \ } \ } while (0) #define CHANGE_PV_LIST_LOCK_TO_VM_PAGE(lockp, m) \ CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, VM_PAGE_TO_PHYS(m)) #define RELEASE_PV_LIST_LOCK(lockp) do { \ struct rwlock **_lockp = (lockp); \ \ if (*_lockp != NULL) { \ rw_wunlock(*_lockp); \ *_lockp = NULL; \ } \ } while (0) #define VM_PAGE_TO_PV_LIST_LOCK(m) \ PHYS_TO_PV_LIST_LOCK(VM_PAGE_TO_PHYS(m)) struct pmap kernel_pmap_store; vm_offset_t virtual_avail; /* VA of first avail page (after kernel bss) */ vm_offset_t virtual_end; /* VA of last avail page (end of kernel AS) */ int nkpt; SYSCTL_INT(_machdep, OID_AUTO, nkpt, CTLFLAG_RD, &nkpt, 0, "Number of kernel page table pages allocated on bootup"); static int ndmpdp; vm_paddr_t dmaplimit; vm_offset_t kernel_vm_end = VM_MIN_KERNEL_ADDRESS; pt_entry_t pg_nx; static SYSCTL_NODE(_vm, OID_AUTO, pmap, CTLFLAG_RD, 0, "VM/pmap parameters"); static int pat_works = 1; SYSCTL_INT(_vm_pmap, OID_AUTO, pat_works, CTLFLAG_RD, &pat_works, 1, "Is page attribute table fully functional?"); static int pg_ps_enabled = 1; SYSCTL_INT(_vm_pmap, OID_AUTO, pg_ps_enabled, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &pg_ps_enabled, 0, "Are large page mappings enabled?"); #define PAT_INDEX_SIZE 8 static int pat_index[PAT_INDEX_SIZE]; /* cache mode to PAT index conversion */ static u_int64_t KPTphys; /* phys addr of kernel level 1 */ static u_int64_t KPDphys; /* phys addr of kernel level 2 */ u_int64_t KPDPphys; /* phys addr of kernel level 3 */ u_int64_t KPML4phys; /* phys addr of kernel level 4 */ static u_int64_t DMPDphys; /* phys addr of direct mapped level 2 */ static u_int64_t DMPDPphys; /* phys addr of direct mapped level 3 */ static int ndmpdpphys; /* number of DMPDPphys pages */ /* * pmap_mapdev support pre initialization (i.e. console) */ #define PMAP_PREINIT_MAPPING_COUNT 8 static struct pmap_preinit_mapping { vm_paddr_t pa; vm_offset_t va; vm_size_t sz; int mode; } pmap_preinit_mapping[PMAP_PREINIT_MAPPING_COUNT]; static int pmap_initialized; static struct rwlock_padalign pvh_global_lock; /* * Data for the pv entry allocation mechanism */ static TAILQ_HEAD(pch, pv_chunk) pv_chunks = TAILQ_HEAD_INITIALIZER(pv_chunks); static struct mtx pv_chunks_mutex; static struct rwlock pv_list_locks[NPV_LIST_LOCKS]; static struct md_page *pv_table; /* * All those kernel PT submaps that BSD is so fond of */ pt_entry_t *CMAP1 = 0; caddr_t CADDR1 = 0; static vm_offset_t qframe = 0; static struct mtx qframe_mtx; static int pmap_flags = PMAP_PDE_SUPERPAGE; /* flags for x86 pmaps */ int pmap_pcid_enabled = 1; SYSCTL_INT(_vm_pmap, OID_AUTO, pcid_enabled, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &pmap_pcid_enabled, 0, "Is TLB Context ID enabled ?"); int invpcid_works = 0; SYSCTL_INT(_vm_pmap, OID_AUTO, invpcid_works, CTLFLAG_RD, &invpcid_works, 0, "Is the invpcid instruction available ?"); static int pmap_pcid_save_cnt_proc(SYSCTL_HANDLER_ARGS) { int i; uint64_t res; res = 0; CPU_FOREACH(i) { res += cpuid_to_pcpu[i]->pc_pm_save_cnt; } return (sysctl_handle_64(oidp, &res, 0, req)); } SYSCTL_PROC(_vm_pmap, OID_AUTO, pcid_save_cnt, CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, pmap_pcid_save_cnt_proc, "QU", "Count of saved TLB context on switch"); /* * Crashdump maps. */ static caddr_t crashdumpmap; static void free_pv_chunk(struct pv_chunk *pc); static void free_pv_entry(pmap_t pmap, pv_entry_t pv); static pv_entry_t get_pv_entry(pmap_t pmap, struct rwlock **lockp); static int popcnt_pc_map_elem_pq(uint64_t elem); static vm_page_t reclaim_pv_chunk(pmap_t locked_pmap, struct rwlock **lockp); static void reserve_pv_entries(pmap_t pmap, int needed, struct rwlock **lockp); static void pmap_pv_demote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp); static boolean_t pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp); static void pmap_pv_promote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp); static void pmap_pvh_free(struct md_page *pvh, pmap_t pmap, vm_offset_t va); static pv_entry_t pmap_pvh_remove(struct md_page *pvh, pmap_t pmap, vm_offset_t va); static int pmap_change_attr_locked(vm_offset_t va, vm_size_t size, int mode); static boolean_t pmap_demote_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va); static boolean_t pmap_demote_pde_locked(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, struct rwlock **lockp); static boolean_t pmap_demote_pdpe(pmap_t pmap, pdp_entry_t *pdpe, vm_offset_t va); static boolean_t pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, struct rwlock **lockp); static vm_page_t pmap_enter_quick_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, vm_page_t mpte, struct rwlock **lockp); static void pmap_fill_ptp(pt_entry_t *firstpte, pt_entry_t newpte); static int pmap_insert_pt_page(pmap_t pmap, vm_page_t mpte); static void pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode); static vm_page_t pmap_lookup_pt_page(pmap_t pmap, vm_offset_t va); static void pmap_pde_attr(pd_entry_t *pde, int cache_bits, int mask); static void pmap_promote_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, struct rwlock **lockp); static boolean_t pmap_protect_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t sva, vm_prot_t prot); static void pmap_pte_attr(pt_entry_t *pte, int cache_bits, int mask); static int pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, struct spglist *free, struct rwlock **lockp); static int pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t sva, pd_entry_t ptepde, struct spglist *free, struct rwlock **lockp); static void pmap_remove_pt_page(pmap_t pmap, vm_page_t mpte); static void pmap_remove_page(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, struct spglist *free); static boolean_t pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va, vm_page_t m, struct rwlock **lockp); static void pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde); static void pmap_update_pde_invalidate(pmap_t, vm_offset_t va, pd_entry_t pde); static vm_page_t _pmap_allocpte(pmap_t pmap, vm_pindex_t ptepindex, struct rwlock **lockp); static vm_page_t pmap_allocpde(pmap_t pmap, vm_offset_t va, struct rwlock **lockp); static vm_page_t pmap_allocpte(pmap_t pmap, vm_offset_t va, struct rwlock **lockp); static void _pmap_unwire_ptp(pmap_t pmap, vm_offset_t va, vm_page_t m, struct spglist *free); static int pmap_unuse_pt(pmap_t, vm_offset_t, pd_entry_t, struct spglist *); static vm_offset_t pmap_kmem_choose(vm_offset_t addr); /* * Move the kernel virtual free pointer to the next * 2MB. This is used to help improve performance * by using a large (2MB) page for much of the kernel * (.text, .data, .bss) */ static vm_offset_t pmap_kmem_choose(vm_offset_t addr) { vm_offset_t newaddr = addr; newaddr = roundup2(addr, NBPDR); return (newaddr); } /********************/ /* Inline functions */ /********************/ /* Return a non-clipped PD index for a given VA */ static __inline vm_pindex_t pmap_pde_pindex(vm_offset_t va) { return (va >> PDRSHIFT); } /* Return various clipped indexes for a given VA */ static __inline vm_pindex_t pmap_pte_index(vm_offset_t va) { return ((va >> PAGE_SHIFT) & ((1ul << NPTEPGSHIFT) - 1)); } static __inline vm_pindex_t pmap_pde_index(vm_offset_t va) { return ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); } static __inline vm_pindex_t pmap_pdpe_index(vm_offset_t va) { return ((va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1)); } static __inline vm_pindex_t pmap_pml4e_index(vm_offset_t va) { return ((va >> PML4SHIFT) & ((1ul << NPML4EPGSHIFT) - 1)); } /* Return a pointer to the PML4 slot that corresponds to a VA */ static __inline pml4_entry_t * pmap_pml4e(pmap_t pmap, vm_offset_t va) { return (&pmap->pm_pml4[pmap_pml4e_index(va)]); } /* Return a pointer to the PDP slot that corresponds to a VA */ static __inline pdp_entry_t * pmap_pml4e_to_pdpe(pml4_entry_t *pml4e, vm_offset_t va) { pdp_entry_t *pdpe; pdpe = (pdp_entry_t *)PHYS_TO_DMAP(*pml4e & PG_FRAME); return (&pdpe[pmap_pdpe_index(va)]); } /* Return a pointer to the PDP slot that corresponds to a VA */ static __inline pdp_entry_t * pmap_pdpe(pmap_t pmap, vm_offset_t va) { pml4_entry_t *pml4e; pt_entry_t PG_V; PG_V = pmap_valid_bit(pmap); pml4e = pmap_pml4e(pmap, va); if ((*pml4e & PG_V) == 0) return (NULL); return (pmap_pml4e_to_pdpe(pml4e, va)); } /* Return a pointer to the PD slot that corresponds to a VA */ static __inline pd_entry_t * pmap_pdpe_to_pde(pdp_entry_t *pdpe, vm_offset_t va) { pd_entry_t *pde; pde = (pd_entry_t *)PHYS_TO_DMAP(*pdpe & PG_FRAME); return (&pde[pmap_pde_index(va)]); } /* Return a pointer to the PD slot that corresponds to a VA */ static __inline pd_entry_t * pmap_pde(pmap_t pmap, vm_offset_t va) { pdp_entry_t *pdpe; pt_entry_t PG_V; PG_V = pmap_valid_bit(pmap); pdpe = pmap_pdpe(pmap, va); if (pdpe == NULL || (*pdpe & PG_V) == 0) return (NULL); return (pmap_pdpe_to_pde(pdpe, va)); } /* Return a pointer to the PT slot that corresponds to a VA */ static __inline pt_entry_t * pmap_pde_to_pte(pd_entry_t *pde, vm_offset_t va) { pt_entry_t *pte; pte = (pt_entry_t *)PHYS_TO_DMAP(*pde & PG_FRAME); return (&pte[pmap_pte_index(va)]); } /* Return a pointer to the PT slot that corresponds to a VA */ static __inline pt_entry_t * pmap_pte(pmap_t pmap, vm_offset_t va) { pd_entry_t *pde; pt_entry_t PG_V; PG_V = pmap_valid_bit(pmap); pde = pmap_pde(pmap, va); if (pde == NULL || (*pde & PG_V) == 0) return (NULL); if ((*pde & PG_PS) != 0) /* compat with i386 pmap_pte() */ return ((pt_entry_t *)pde); return (pmap_pde_to_pte(pde, va)); } static __inline void pmap_resident_count_inc(pmap_t pmap, int count) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); pmap->pm_stats.resident_count += count; } static __inline void pmap_resident_count_dec(pmap_t pmap, int count) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); KASSERT(pmap->pm_stats.resident_count >= count, ("pmap %p resident count underflow %ld %d", pmap, pmap->pm_stats.resident_count, count)); pmap->pm_stats.resident_count -= count; } PMAP_INLINE pt_entry_t * vtopte(vm_offset_t va) { u_int64_t mask = ((1ul << (NPTEPGSHIFT + NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT)) - 1); KASSERT(va >= VM_MAXUSER_ADDRESS, ("vtopte on a uva/gpa 0x%0lx", va)); return (PTmap + ((va >> PAGE_SHIFT) & mask)); } static __inline pd_entry_t * vtopde(vm_offset_t va) { u_int64_t mask = ((1ul << (NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT)) - 1); KASSERT(va >= VM_MAXUSER_ADDRESS, ("vtopde on a uva/gpa 0x%0lx", va)); return (PDmap + ((va >> PDRSHIFT) & mask)); } static u_int64_t allocpages(vm_paddr_t *firstaddr, int n) { u_int64_t ret; ret = *firstaddr; bzero((void *)ret, n * PAGE_SIZE); *firstaddr += n * PAGE_SIZE; return (ret); } CTASSERT(powerof2(NDMPML4E)); /* number of kernel PDP slots */ -#define NKPDPE(ptpgs) howmany((ptpgs), NPDEPG) +#define NKPDPE(ptpgs) howmany(ptpgs, NPDEPG) static void nkpt_init(vm_paddr_t addr) { int pt_pages; #ifdef NKPT pt_pages = NKPT; #else pt_pages = howmany(addr, 1 << PDRSHIFT); pt_pages += NKPDPE(pt_pages); /* * Add some slop beyond the bare minimum required for bootstrapping * the kernel. * * This is quite important when allocating KVA for kernel modules. * The modules are required to be linked in the negative 2GB of * the address space. If we run out of KVA in this region then * pmap_growkernel() will need to allocate page table pages to map * the entire 512GB of KVA space which is an unnecessary tax on * physical memory. * * Secondly, device memory mapped as part of setting up the low- * level console(s) is taken from KVA, starting at virtual_avail. * This is because cninit() is called after pmap_bootstrap() but * before vm_init() and pmap_init(). 20MB for a frame buffer is * not uncommon. */ pt_pages += 32; /* 64MB additional slop. */ #endif nkpt = pt_pages; } static void create_pagetables(vm_paddr_t *firstaddr) { int i, j, ndm1g, nkpdpe; pt_entry_t *pt_p; pd_entry_t *pd_p; pdp_entry_t *pdp_p; pml4_entry_t *p4_p; /* Allocate page table pages for the direct map */ ndmpdp = (ptoa(Maxmem) + NBPDP - 1) >> PDPSHIFT; if (ndmpdp < 4) /* Minimum 4GB of dirmap */ ndmpdp = 4; ndmpdpphys = howmany(ndmpdp, NPDPEPG); if (ndmpdpphys > NDMPML4E) { /* * Each NDMPML4E allows 512 GB, so limit to that, * and then readjust ndmpdp and ndmpdpphys. */ printf("NDMPML4E limits system to %d GB\n", NDMPML4E * 512); Maxmem = atop(NDMPML4E * NBPML4); ndmpdpphys = NDMPML4E; ndmpdp = NDMPML4E * NPDEPG; } DMPDPphys = allocpages(firstaddr, ndmpdpphys); ndm1g = 0; if ((amd_feature & AMDID_PAGE1GB) != 0) ndm1g = ptoa(Maxmem) >> PDPSHIFT; if (ndm1g < ndmpdp) DMPDphys = allocpages(firstaddr, ndmpdp - ndm1g); dmaplimit = (vm_paddr_t)ndmpdp << PDPSHIFT; /* Allocate pages */ KPML4phys = allocpages(firstaddr, 1); KPDPphys = allocpages(firstaddr, NKPML4E); /* * Allocate the initial number of kernel page table pages required to * bootstrap. We defer this until after all memory-size dependent * allocations are done (e.g. direct map), so that we don't have to * build in too much slop in our estimate. * * Note that when NKPML4E > 1, we have an empty page underneath * all but the KPML4I'th one, so we need NKPML4E-1 extra (zeroed) * pages. (pmap_enter requires a PD page to exist for each KPML4E.) */ nkpt_init(*firstaddr); nkpdpe = NKPDPE(nkpt); KPTphys = allocpages(firstaddr, nkpt); KPDphys = allocpages(firstaddr, nkpdpe); /* Fill in the underlying page table pages */ /* Nominally read-only (but really R/W) from zero to physfree */ /* XXX not fully used, underneath 2M pages */ pt_p = (pt_entry_t *)KPTphys; for (i = 0; ptoa(i) < *firstaddr; i++) pt_p[i] = ptoa(i) | X86_PG_RW | X86_PG_V | X86_PG_G; /* Now map the page tables at their location within PTmap */ pd_p = (pd_entry_t *)KPDphys; for (i = 0; i < nkpt; i++) pd_p[i] = (KPTphys + ptoa(i)) | X86_PG_RW | X86_PG_V; /* Map from zero to end of allocations under 2M pages */ /* This replaces some of the KPTphys entries above */ for (i = 0; (i << PDRSHIFT) < *firstaddr; i++) pd_p[i] = (i << PDRSHIFT) | X86_PG_RW | X86_PG_V | PG_PS | X86_PG_G; /* And connect up the PD to the PDP (leaving room for L4 pages) */ pdp_p = (pdp_entry_t *)(KPDPphys + ptoa(KPML4I - KPML4BASE)); for (i = 0; i < nkpdpe; i++) pdp_p[i + KPDPI] = (KPDphys + ptoa(i)) | X86_PG_RW | X86_PG_V | PG_U; /* * Now, set up the direct map region using 2MB and/or 1GB pages. If * the end of physical memory is not aligned to a 1GB page boundary, * then the residual physical memory is mapped with 2MB pages. Later, * if pmap_mapdev{_attr}() uses the direct map for non-write-back * memory, pmap_change_attr() will demote any 2MB or 1GB page mappings * that are partially used. */ pd_p = (pd_entry_t *)DMPDphys; for (i = NPDEPG * ndm1g, j = 0; i < NPDEPG * ndmpdp; i++, j++) { pd_p[j] = (vm_paddr_t)i << PDRSHIFT; /* Preset PG_M and PG_A because demotion expects it. */ pd_p[j] |= X86_PG_RW | X86_PG_V | PG_PS | X86_PG_G | X86_PG_M | X86_PG_A; } pdp_p = (pdp_entry_t *)DMPDPphys; for (i = 0; i < ndm1g; i++) { pdp_p[i] = (vm_paddr_t)i << PDPSHIFT; /* Preset PG_M and PG_A because demotion expects it. */ pdp_p[i] |= X86_PG_RW | X86_PG_V | PG_PS | X86_PG_G | X86_PG_M | X86_PG_A; } for (j = 0; i < ndmpdp; i++, j++) { pdp_p[i] = DMPDphys + ptoa(j); pdp_p[i] |= X86_PG_RW | X86_PG_V | PG_U; } /* And recursively map PML4 to itself in order to get PTmap */ p4_p = (pml4_entry_t *)KPML4phys; p4_p[PML4PML4I] = KPML4phys; p4_p[PML4PML4I] |= X86_PG_RW | X86_PG_V | PG_U; /* Connect the Direct Map slot(s) up to the PML4. */ for (i = 0; i < ndmpdpphys; i++) { p4_p[DMPML4I + i] = DMPDPphys + ptoa(i); p4_p[DMPML4I + i] |= X86_PG_RW | X86_PG_V | PG_U; } /* Connect the KVA slots up to the PML4 */ for (i = 0; i < NKPML4E; i++) { p4_p[KPML4BASE + i] = KPDPphys + ptoa(i); p4_p[KPML4BASE + i] |= X86_PG_RW | X86_PG_V | PG_U; } } /* * Bootstrap the system enough to run with virtual memory. * * On amd64 this is called after mapping has already been enabled * and just syncs the pmap module with what has already been done. * [We can't call it easily with mapping off since the kernel is not * mapped with PA == VA, hence we would have to relocate every address * from the linked base (virtual) address "KERNBASE" to the actual * (physical) address starting relative to 0] */ void pmap_bootstrap(vm_paddr_t *firstaddr) { vm_offset_t va; pt_entry_t *pte; int i; /* * Create an initial set of page tables to run the kernel in. */ create_pagetables(firstaddr); /* * Add a physical memory segment (vm_phys_seg) corresponding to the * preallocated kernel page table pages so that vm_page structures * representing these pages will be created. The vm_page structures * are required for promotion of the corresponding kernel virtual * addresses to superpage mappings. */ vm_phys_add_seg(KPTphys, KPTphys + ptoa(nkpt)); virtual_avail = (vm_offset_t) KERNBASE + *firstaddr; virtual_avail = pmap_kmem_choose(virtual_avail); virtual_end = VM_MAX_KERNEL_ADDRESS; /* XXX do %cr0 as well */ load_cr4(rcr4() | CR4_PGE); load_cr3(KPML4phys); if (cpu_stdext_feature & CPUID_STDEXT_SMEP) load_cr4(rcr4() | CR4_SMEP); /* * Initialize the kernel pmap (which is statically allocated). */ PMAP_LOCK_INIT(kernel_pmap); kernel_pmap->pm_pml4 = (pdp_entry_t *)PHYS_TO_DMAP(KPML4phys); kernel_pmap->pm_cr3 = KPML4phys; CPU_FILL(&kernel_pmap->pm_active); /* don't allow deactivation */ TAILQ_INIT(&kernel_pmap->pm_pvchunk); kernel_pmap->pm_flags = pmap_flags; /* * Initialize the global pv list lock. */ rw_init(&pvh_global_lock, "pmap pv global"); /* * Reserve some special page table entries/VA space for temporary * mapping of pages. */ #define SYSMAP(c, p, v, n) \ v = (c)va; va += ((n)*PAGE_SIZE); p = pte; pte += (n); va = virtual_avail; pte = vtopte(va); /* * Crashdump maps. The first page is reused as CMAP1 for the * memory test. */ SYSMAP(caddr_t, CMAP1, crashdumpmap, MAXDUMPPGS) CADDR1 = crashdumpmap; virtual_avail = va; /* Initialize the PAT MSR. */ pmap_init_pat(); /* Initialize TLB Context Id. */ TUNABLE_INT_FETCH("vm.pmap.pcid_enabled", &pmap_pcid_enabled); if ((cpu_feature2 & CPUID2_PCID) != 0 && pmap_pcid_enabled) { /* Check for INVPCID support */ invpcid_works = (cpu_stdext_feature & CPUID_STDEXT_INVPCID) != 0; for (i = 0; i < MAXCPU; i++) { kernel_pmap->pm_pcids[i].pm_pcid = PMAP_PCID_KERN; kernel_pmap->pm_pcids[i].pm_gen = 1; } __pcpu[0].pc_pcid_next = PMAP_PCID_KERN + 1; __pcpu[0].pc_pcid_gen = 1; /* * pcpu area for APs is zeroed during AP startup. * pc_pcid_next and pc_pcid_gen are initialized by AP * during pcpu setup. */ load_cr4(rcr4() | CR4_PCIDE); } else { pmap_pcid_enabled = 0; } } /* * Setup the PAT MSR. */ void pmap_init_pat(void) { int pat_table[PAT_INDEX_SIZE]; uint64_t pat_msr; u_long cr0, cr4; int i; /* Bail if this CPU doesn't implement PAT. */ if ((cpu_feature & CPUID_PAT) == 0) panic("no PAT??"); /* Set default PAT index table. */ for (i = 0; i < PAT_INDEX_SIZE; i++) pat_table[i] = -1; pat_table[PAT_WRITE_BACK] = 0; pat_table[PAT_WRITE_THROUGH] = 1; pat_table[PAT_UNCACHEABLE] = 3; pat_table[PAT_WRITE_COMBINING] = 3; pat_table[PAT_WRITE_PROTECTED] = 3; pat_table[PAT_UNCACHED] = 3; /* Initialize default PAT entries. */ pat_msr = PAT_VALUE(0, PAT_WRITE_BACK) | PAT_VALUE(1, PAT_WRITE_THROUGH) | PAT_VALUE(2, PAT_UNCACHED) | PAT_VALUE(3, PAT_UNCACHEABLE) | PAT_VALUE(4, PAT_WRITE_BACK) | PAT_VALUE(5, PAT_WRITE_THROUGH) | PAT_VALUE(6, PAT_UNCACHED) | PAT_VALUE(7, PAT_UNCACHEABLE); if (pat_works) { /* * Leave the indices 0-3 at the default of WB, WT, UC-, and UC. * Program 5 and 6 as WP and WC. * Leave 4 and 7 as WB and UC. */ pat_msr &= ~(PAT_MASK(5) | PAT_MASK(6)); pat_msr |= PAT_VALUE(5, PAT_WRITE_PROTECTED) | PAT_VALUE(6, PAT_WRITE_COMBINING); pat_table[PAT_UNCACHED] = 2; pat_table[PAT_WRITE_PROTECTED] = 5; pat_table[PAT_WRITE_COMBINING] = 6; } else { /* * Just replace PAT Index 2 with WC instead of UC-. */ pat_msr &= ~PAT_MASK(2); pat_msr |= PAT_VALUE(2, PAT_WRITE_COMBINING); pat_table[PAT_WRITE_COMBINING] = 2; } /* Disable PGE. */ cr4 = rcr4(); load_cr4(cr4 & ~CR4_PGE); /* Disable caches (CD = 1, NW = 0). */ cr0 = rcr0(); load_cr0((cr0 & ~CR0_NW) | CR0_CD); /* Flushes caches and TLBs. */ wbinvd(); invltlb(); /* Update PAT and index table. */ wrmsr(MSR_PAT, pat_msr); for (i = 0; i < PAT_INDEX_SIZE; i++) pat_index[i] = pat_table[i]; /* Flush caches and TLBs again. */ wbinvd(); invltlb(); /* Restore caches and PGE. */ load_cr0(cr0); load_cr4(cr4); } /* * Initialize a vm_page's machine-dependent fields. */ void pmap_page_init(vm_page_t m) { TAILQ_INIT(&m->md.pv_list); m->md.pat_mode = PAT_WRITE_BACK; } /* * Initialize the pmap module. * Called by vm_init, to initialize any structures that the pmap * system needs to map virtual memory. */ void pmap_init(void) { struct pmap_preinit_mapping *ppim; vm_page_t mpte; vm_size_t s; int error, i, pv_npg; /* * Initialize the vm page array entries for the kernel pmap's * page table pages. */ for (i = 0; i < nkpt; i++) { mpte = PHYS_TO_VM_PAGE(KPTphys + (i << PAGE_SHIFT)); KASSERT(mpte >= vm_page_array && mpte < &vm_page_array[vm_page_array_size], ("pmap_init: page table page is out of range")); mpte->pindex = pmap_pde_pindex(KERNBASE) + i; mpte->phys_addr = KPTphys + (i << PAGE_SHIFT); } /* * If the kernel is running on a virtual machine, then it must assume * that MCA is enabled by the hypervisor. Moreover, the kernel must * be prepared for the hypervisor changing the vendor and family that * are reported by CPUID. Consequently, the workaround for AMD Family * 10h Erratum 383 is enabled if the processor's feature set does not * include at least one feature that is only supported by older Intel * or newer AMD processors. */ if (vm_guest == VM_GUEST_VM && (cpu_feature & CPUID_SS) == 0 && (cpu_feature2 & (CPUID2_SSSE3 | CPUID2_SSE41 | CPUID2_AESNI | CPUID2_AVX | CPUID2_XSAVE)) == 0 && (amd_feature2 & (AMDID2_XOP | AMDID2_FMA4)) == 0) workaround_erratum383 = 1; /* * Are large page mappings enabled? */ TUNABLE_INT_FETCH("vm.pmap.pg_ps_enabled", &pg_ps_enabled); if (pg_ps_enabled) { KASSERT(MAXPAGESIZES > 1 && pagesizes[1] == 0, ("pmap_init: can't assign to pagesizes[1]")); pagesizes[1] = NBPDR; } /* * Initialize the pv chunk list mutex. */ mtx_init(&pv_chunks_mutex, "pmap pv chunk list", NULL, MTX_DEF); /* * Initialize the pool of pv list locks. */ for (i = 0; i < NPV_LIST_LOCKS; i++) rw_init(&pv_list_locks[i], "pmap pv list"); /* * Calculate the size of the pv head table for superpages. */ pv_npg = howmany(vm_phys_segs[vm_phys_nsegs - 1].end, NBPDR); /* * Allocate memory for the pv head table for superpages. */ s = (vm_size_t)(pv_npg * sizeof(struct md_page)); s = round_page(s); pv_table = (struct md_page *)kmem_malloc(kernel_arena, s, M_WAITOK | M_ZERO); for (i = 0; i < pv_npg; i++) TAILQ_INIT(&pv_table[i].pv_list); pmap_initialized = 1; for (i = 0; i < PMAP_PREINIT_MAPPING_COUNT; i++) { ppim = pmap_preinit_mapping + i; if (ppim->va == 0) continue; /* Make the direct map consistent */ if (ppim->pa < dmaplimit && ppim->pa + ppim->sz < dmaplimit) { (void)pmap_change_attr(PHYS_TO_DMAP(ppim->pa), ppim->sz, ppim->mode); } if (!bootverbose) continue; printf("PPIM %u: PA=%#lx, VA=%#lx, size=%#lx, mode=%#x\n", i, ppim->pa, ppim->va, ppim->sz, ppim->mode); } mtx_init(&qframe_mtx, "qfrmlk", NULL, MTX_SPIN); error = vmem_alloc(kernel_arena, PAGE_SIZE, M_BESTFIT | M_WAITOK, (vmem_addr_t *)&qframe); if (error != 0) panic("qframe allocation failed"); } static SYSCTL_NODE(_vm_pmap, OID_AUTO, pde, CTLFLAG_RD, 0, "2MB page mapping counters"); static u_long pmap_pde_demotions; SYSCTL_ULONG(_vm_pmap_pde, OID_AUTO, demotions, CTLFLAG_RD, &pmap_pde_demotions, 0, "2MB page demotions"); static u_long pmap_pde_mappings; SYSCTL_ULONG(_vm_pmap_pde, OID_AUTO, mappings, CTLFLAG_RD, &pmap_pde_mappings, 0, "2MB page mappings"); static u_long pmap_pde_p_failures; SYSCTL_ULONG(_vm_pmap_pde, OID_AUTO, p_failures, CTLFLAG_RD, &pmap_pde_p_failures, 0, "2MB page promotion failures"); static u_long pmap_pde_promotions; SYSCTL_ULONG(_vm_pmap_pde, OID_AUTO, promotions, CTLFLAG_RD, &pmap_pde_promotions, 0, "2MB page promotions"); static SYSCTL_NODE(_vm_pmap, OID_AUTO, pdpe, CTLFLAG_RD, 0, "1GB page mapping counters"); static u_long pmap_pdpe_demotions; SYSCTL_ULONG(_vm_pmap_pdpe, OID_AUTO, demotions, CTLFLAG_RD, &pmap_pdpe_demotions, 0, "1GB page demotions"); /*************************************************** * Low level helper routines..... ***************************************************/ static pt_entry_t pmap_swap_pat(pmap_t pmap, pt_entry_t entry) { int x86_pat_bits = X86_PG_PTE_PAT | X86_PG_PDE_PAT; switch (pmap->pm_type) { case PT_X86: case PT_RVI: /* Verify that both PAT bits are not set at the same time */ KASSERT((entry & x86_pat_bits) != x86_pat_bits, ("Invalid PAT bits in entry %#lx", entry)); /* Swap the PAT bits if one of them is set */ if ((entry & x86_pat_bits) != 0) entry ^= x86_pat_bits; break; case PT_EPT: /* * Nothing to do - the memory attributes are represented * the same way for regular pages and superpages. */ break; default: panic("pmap_switch_pat_bits: bad pm_type %d", pmap->pm_type); } return (entry); } /* * Determine the appropriate bits to set in a PTE or PDE for a specified * caching mode. */ static int pmap_cache_bits(pmap_t pmap, int mode, boolean_t is_pde) { int cache_bits, pat_flag, pat_idx; if (mode < 0 || mode >= PAT_INDEX_SIZE || pat_index[mode] < 0) panic("Unknown caching mode %d\n", mode); switch (pmap->pm_type) { case PT_X86: case PT_RVI: /* The PAT bit is different for PTE's and PDE's. */ pat_flag = is_pde ? X86_PG_PDE_PAT : X86_PG_PTE_PAT; /* Map the caching mode to a PAT index. */ pat_idx = pat_index[mode]; /* Map the 3-bit index value into the PAT, PCD, and PWT bits. */ cache_bits = 0; if (pat_idx & 0x4) cache_bits |= pat_flag; if (pat_idx & 0x2) cache_bits |= PG_NC_PCD; if (pat_idx & 0x1) cache_bits |= PG_NC_PWT; break; case PT_EPT: cache_bits = EPT_PG_IGNORE_PAT | EPT_PG_MEMORY_TYPE(mode); break; default: panic("unsupported pmap type %d", pmap->pm_type); } return (cache_bits); } static int pmap_cache_mask(pmap_t pmap, boolean_t is_pde) { int mask; switch (pmap->pm_type) { case PT_X86: case PT_RVI: mask = is_pde ? X86_PG_PDE_CACHE : X86_PG_PTE_CACHE; break; case PT_EPT: mask = EPT_PG_IGNORE_PAT | EPT_PG_MEMORY_TYPE(0x7); break; default: panic("pmap_cache_mask: invalid pm_type %d", pmap->pm_type); } return (mask); } static __inline boolean_t pmap_ps_enabled(pmap_t pmap) { return (pg_ps_enabled && (pmap->pm_flags & PMAP_PDE_SUPERPAGE) != 0); } static void pmap_update_pde_store(pmap_t pmap, pd_entry_t *pde, pd_entry_t newpde) { switch (pmap->pm_type) { case PT_X86: break; case PT_RVI: case PT_EPT: /* * XXX * This is a little bogus since the generation number is * supposed to be bumped up when a region of the address * space is invalidated in the page tables. * * In this case the old PDE entry is valid but yet we want * to make sure that any mappings using the old entry are * invalidated in the TLB. * * The reason this works as expected is because we rendezvous * "all" host cpus and force any vcpu context to exit as a * side-effect. */ atomic_add_acq_long(&pmap->pm_eptgen, 1); break; default: panic("pmap_update_pde_store: bad pm_type %d", pmap->pm_type); } pde_store(pde, newpde); } /* * After changing the page size for the specified virtual address in the page * table, flush the corresponding entries from the processor's TLB. Only the * calling processor's TLB is affected. * * The calling thread must be pinned to a processor. */ static void pmap_update_pde_invalidate(pmap_t pmap, vm_offset_t va, pd_entry_t newpde) { pt_entry_t PG_G; if (pmap_type_guest(pmap)) return; KASSERT(pmap->pm_type == PT_X86, ("pmap_update_pde_invalidate: invalid type %d", pmap->pm_type)); PG_G = pmap_global_bit(pmap); if ((newpde & PG_PS) == 0) /* Demotion: flush a specific 2MB page mapping. */ invlpg(va); else if ((newpde & PG_G) == 0) /* * Promotion: flush every 4KB page mapping from the TLB * because there are too many to flush individually. */ invltlb(); else { /* * Promotion: flush every 4KB page mapping from the TLB, * including any global (PG_G) mappings. */ invltlb_glob(); } } #ifdef SMP /* * For SMP, these functions have to use the IPI mechanism for coherence. * * N.B.: Before calling any of the following TLB invalidation functions, * the calling processor must ensure that all stores updating a non- * kernel page table are globally performed. Otherwise, another * processor could cache an old, pre-update entry without being * invalidated. This can happen one of two ways: (1) The pmap becomes * active on another processor after its pm_active field is checked by * one of the following functions but before a store updating the page * table is globally performed. (2) The pmap becomes active on another * processor before its pm_active field is checked but due to * speculative loads one of the following functions stills reads the * pmap as inactive on the other processor. * * The kernel page table is exempt because its pm_active field is * immutable. The kernel page table is always active on every * processor. */ /* * Interrupt the cpus that are executing in the guest context. * This will force the vcpu to exit and the cached EPT mappings * will be invalidated by the host before the next vmresume. */ static __inline void pmap_invalidate_ept(pmap_t pmap) { int ipinum; sched_pin(); KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), ("pmap_invalidate_ept: absurd pm_active")); /* * The TLB mappings associated with a vcpu context are not * flushed each time a different vcpu is chosen to execute. * * This is in contrast with a process's vtop mappings that * are flushed from the TLB on each context switch. * * Therefore we need to do more than just a TLB shootdown on * the active cpus in 'pmap->pm_active'. To do this we keep * track of the number of invalidations performed on this pmap. * * Each vcpu keeps a cache of this counter and compares it * just before a vmresume. If the counter is out-of-date an * invept will be done to flush stale mappings from the TLB. */ atomic_add_acq_long(&pmap->pm_eptgen, 1); /* * Force the vcpu to exit and trap back into the hypervisor. */ ipinum = pmap->pm_flags & PMAP_NESTED_IPIMASK; ipi_selected(pmap->pm_active, ipinum); sched_unpin(); } void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { cpuset_t *mask; u_int cpuid, i; if (pmap_type_guest(pmap)) { pmap_invalidate_ept(pmap); return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_page: invalid type %d", pmap->pm_type)); sched_pin(); if (pmap == kernel_pmap) { invlpg(va); mask = &all_cpus; } else { cpuid = PCPU_GET(cpuid); if (pmap == PCPU_GET(curpmap)) invlpg(va); else if (pmap_pcid_enabled) pmap->pm_pcids[cpuid].pm_gen = 0; if (pmap_pcid_enabled) { CPU_FOREACH(i) { if (cpuid != i) pmap->pm_pcids[i].pm_gen = 0; } } mask = &pmap->pm_active; } smp_masked_invlpg(*mask, va); sched_unpin(); } /* 4k PTEs -- Chosen to exceed the total size of Broadwell L2 TLB */ #define PMAP_INVLPG_THRESHOLD (4 * 1024 * PAGE_SIZE) void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { cpuset_t *mask; vm_offset_t addr; u_int cpuid, i; if (eva - sva >= PMAP_INVLPG_THRESHOLD) { pmap_invalidate_all(pmap); return; } if (pmap_type_guest(pmap)) { pmap_invalidate_ept(pmap); return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_range: invalid type %d", pmap->pm_type)); sched_pin(); cpuid = PCPU_GET(cpuid); if (pmap == kernel_pmap) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); mask = &all_cpus; } else { if (pmap == PCPU_GET(curpmap)) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); } else if (pmap_pcid_enabled) { pmap->pm_pcids[cpuid].pm_gen = 0; } if (pmap_pcid_enabled) { CPU_FOREACH(i) { if (cpuid != i) pmap->pm_pcids[i].pm_gen = 0; } } mask = &pmap->pm_active; } smp_masked_invlpg_range(*mask, sva, eva); sched_unpin(); } void pmap_invalidate_all(pmap_t pmap) { cpuset_t *mask; struct invpcid_descr d; u_int cpuid, i; if (pmap_type_guest(pmap)) { pmap_invalidate_ept(pmap); return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_all: invalid type %d", pmap->pm_type)); sched_pin(); if (pmap == kernel_pmap) { if (pmap_pcid_enabled && invpcid_works) { bzero(&d, sizeof(d)); invpcid(&d, INVPCID_CTXGLOB); } else { invltlb_glob(); } mask = &all_cpus; } else { cpuid = PCPU_GET(cpuid); if (pmap == PCPU_GET(curpmap)) { if (pmap_pcid_enabled) { if (invpcid_works) { d.pcid = pmap->pm_pcids[cpuid].pm_pcid; d.pad = 0; d.addr = 0; invpcid(&d, INVPCID_CTX); } else { load_cr3(pmap->pm_cr3 | pmap->pm_pcids [PCPU_GET(cpuid)].pm_pcid); } } else { invltlb(); } } else if (pmap_pcid_enabled) { pmap->pm_pcids[cpuid].pm_gen = 0; } if (pmap_pcid_enabled) { CPU_FOREACH(i) { if (cpuid != i) pmap->pm_pcids[i].pm_gen = 0; } } mask = &pmap->pm_active; } smp_masked_invltlb(*mask, pmap); sched_unpin(); } void pmap_invalidate_cache(void) { sched_pin(); wbinvd(); smp_cache_flush(); sched_unpin(); } struct pde_action { cpuset_t invalidate; /* processors that invalidate their TLB */ pmap_t pmap; vm_offset_t va; pd_entry_t *pde; pd_entry_t newpde; u_int store; /* processor that updates the PDE */ }; static void pmap_update_pde_action(void *arg) { struct pde_action *act = arg; if (act->store == PCPU_GET(cpuid)) pmap_update_pde_store(act->pmap, act->pde, act->newpde); } static void pmap_update_pde_teardown(void *arg) { struct pde_action *act = arg; if (CPU_ISSET(PCPU_GET(cpuid), &act->invalidate)) pmap_update_pde_invalidate(act->pmap, act->va, act->newpde); } /* * Change the page size for the specified virtual address in a way that * prevents any possibility of the TLB ever having two entries that map the * same virtual address using different page sizes. This is the recommended * workaround for Erratum 383 on AMD Family 10h processors. It prevents a * machine check exception for a TLB state that is improperly diagnosed as a * hardware error. */ static void pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) { struct pde_action act; cpuset_t active, other_cpus; u_int cpuid; sched_pin(); cpuid = PCPU_GET(cpuid); other_cpus = all_cpus; CPU_CLR(cpuid, &other_cpus); if (pmap == kernel_pmap || pmap_type_guest(pmap)) active = all_cpus; else { active = pmap->pm_active; } if (CPU_OVERLAP(&active, &other_cpus)) { act.store = cpuid; act.invalidate = active; act.va = va; act.pmap = pmap; act.pde = pde; act.newpde = newpde; CPU_SET(cpuid, &active); smp_rendezvous_cpus(active, smp_no_rendevous_barrier, pmap_update_pde_action, pmap_update_pde_teardown, &act); } else { pmap_update_pde_store(pmap, pde, newpde); if (CPU_ISSET(cpuid, &active)) pmap_update_pde_invalidate(pmap, va, newpde); } sched_unpin(); } #else /* !SMP */ /* * Normal, non-SMP, invalidation functions. */ void pmap_invalidate_page(pmap_t pmap, vm_offset_t va) { if (pmap->pm_type == PT_RVI || pmap->pm_type == PT_EPT) { pmap->pm_eptgen++; return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_range: unknown type %d", pmap->pm_type)); if (pmap == kernel_pmap || pmap == PCPU_GET(curpmap)) invlpg(va); else if (pmap_pcid_enabled) pmap->pm_pcids[0].pm_gen = 0; } void pmap_invalidate_range(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { vm_offset_t addr; if (pmap->pm_type == PT_RVI || pmap->pm_type == PT_EPT) { pmap->pm_eptgen++; return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_range: unknown type %d", pmap->pm_type)); if (pmap == kernel_pmap || pmap == PCPU_GET(curpmap)) { for (addr = sva; addr < eva; addr += PAGE_SIZE) invlpg(addr); } else if (pmap_pcid_enabled) { pmap->pm_pcids[0].pm_gen = 0; } } void pmap_invalidate_all(pmap_t pmap) { struct invpcid_descr d; if (pmap->pm_type == PT_RVI || pmap->pm_type == PT_EPT) { pmap->pm_eptgen++; return; } KASSERT(pmap->pm_type == PT_X86, ("pmap_invalidate_all: unknown type %d", pmap->pm_type)); if (pmap == kernel_pmap) { if (pmap_pcid_enabled && invpcid_works) { bzero(&d, sizeof(d)); invpcid(&d, INVPCID_CTXGLOB); } else { invltlb_glob(); } } else if (pmap == PCPU_GET(curpmap)) { if (pmap_pcid_enabled) { if (invpcid_works) { d.pcid = pmap->pm_pcids[0].pm_pcid; d.pad = 0; d.addr = 0; invpcid(&d, INVPCID_CTX); } else { load_cr3(pmap->pm_cr3 | pmap->pm_pcids[0]. pm_pcid); } } else { invltlb(); } } else if (pmap_pcid_enabled) { pmap->pm_pcids[0].pm_gen = 0; } } PMAP_INLINE void pmap_invalidate_cache(void) { wbinvd(); } static void pmap_update_pde(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, pd_entry_t newpde) { pmap_update_pde_store(pmap, pde, newpde); if (pmap == kernel_pmap || pmap == PCPU_GET(curpmap)) pmap_update_pde_invalidate(pmap, va, newpde); else pmap->pm_pcids[0].pm_gen = 0; } #endif /* !SMP */ #define PMAP_CLFLUSH_THRESHOLD (2 * 1024 * 1024) void pmap_invalidate_cache_range(vm_offset_t sva, vm_offset_t eva, boolean_t force) { if (force) { sva &= ~(vm_offset_t)cpu_clflush_line_size; } else { KASSERT((sva & PAGE_MASK) == 0, ("pmap_invalidate_cache_range: sva not page-aligned")); KASSERT((eva & PAGE_MASK) == 0, ("pmap_invalidate_cache_range: eva not page-aligned")); } if ((cpu_feature & CPUID_SS) != 0 && !force) ; /* If "Self Snoop" is supported and allowed, do nothing. */ else if ((cpu_stdext_feature & CPUID_STDEXT_CLFLUSHOPT) != 0 && eva - sva < PMAP_CLFLUSH_THRESHOLD) { /* * XXX: Some CPUs fault, hang, or trash the local APIC * registers if we use CLFLUSH on the local APIC * range. The local APIC is always uncached, so we * don't need to flush for that range anyway. */ if (pmap_kextract(sva) == lapic_paddr) return; /* * Otherwise, do per-cache line flush. Use the mfence * instruction to insure that previous stores are * included in the write-back. The processor * propagates flush to other processors in the cache * coherence domain. */ mfence(); for (; sva < eva; sva += cpu_clflush_line_size) clflushopt(sva); mfence(); } else if ((cpu_feature & CPUID_CLFSH) != 0 && eva - sva < PMAP_CLFLUSH_THRESHOLD) { if (pmap_kextract(sva) == lapic_paddr) return; /* * Writes are ordered by CLFLUSH on Intel CPUs. */ if (cpu_vendor_id != CPU_VENDOR_INTEL) mfence(); for (; sva < eva; sva += cpu_clflush_line_size) clflush(sva); if (cpu_vendor_id != CPU_VENDOR_INTEL) mfence(); } else { /* * No targeted cache flush methods are supported by CPU, * or the supplied range is bigger than 2MB. * Globally invalidate cache. */ pmap_invalidate_cache(); } } /* * Remove the specified set of pages from the data and instruction caches. * * In contrast to pmap_invalidate_cache_range(), this function does not * rely on the CPU's self-snoop feature, because it is intended for use * when moving pages into a different cache domain. */ void pmap_invalidate_cache_pages(vm_page_t *pages, int count) { vm_offset_t daddr, eva; int i; bool useclflushopt; useclflushopt = (cpu_stdext_feature & CPUID_STDEXT_CLFLUSHOPT) != 0; if (count >= PMAP_CLFLUSH_THRESHOLD / PAGE_SIZE || ((cpu_feature & CPUID_CLFSH) == 0 && !useclflushopt)) pmap_invalidate_cache(); else { if (useclflushopt || cpu_vendor_id != CPU_VENDOR_INTEL) mfence(); for (i = 0; i < count; i++) { daddr = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pages[i])); eva = daddr + PAGE_SIZE; for (; daddr < eva; daddr += cpu_clflush_line_size) { if (useclflushopt) clflushopt(daddr); else clflush(daddr); } } if (useclflushopt || cpu_vendor_id != CPU_VENDOR_INTEL) mfence(); } } /* * Routine: pmap_extract * Function: * Extract the physical page address associated * with the given map/virtual_address pair. */ vm_paddr_t pmap_extract(pmap_t pmap, vm_offset_t va) { pdp_entry_t *pdpe; pd_entry_t *pde; pt_entry_t *pte, PG_V; vm_paddr_t pa; pa = 0; PG_V = pmap_valid_bit(pmap); PMAP_LOCK(pmap); pdpe = pmap_pdpe(pmap, va); if (pdpe != NULL && (*pdpe & PG_V) != 0) { if ((*pdpe & PG_PS) != 0) pa = (*pdpe & PG_PS_FRAME) | (va & PDPMASK); else { pde = pmap_pdpe_to_pde(pdpe, va); if ((*pde & PG_V) != 0) { if ((*pde & PG_PS) != 0) { pa = (*pde & PG_PS_FRAME) | (va & PDRMASK); } else { pte = pmap_pde_to_pte(pde, va); pa = (*pte & PG_FRAME) | (va & PAGE_MASK); } } } } PMAP_UNLOCK(pmap); return (pa); } /* * Routine: pmap_extract_and_hold * Function: * Atomically extract and hold the physical page * with the given pmap and virtual address pair * if that mapping permits the given protection. */ vm_page_t pmap_extract_and_hold(pmap_t pmap, vm_offset_t va, vm_prot_t prot) { pd_entry_t pde, *pdep; pt_entry_t pte, PG_RW, PG_V; vm_paddr_t pa; vm_page_t m; pa = 0; m = NULL; PG_RW = pmap_rw_bit(pmap); PG_V = pmap_valid_bit(pmap); PMAP_LOCK(pmap); retry: pdep = pmap_pde(pmap, va); if (pdep != NULL && (pde = *pdep)) { if (pde & PG_PS) { if ((pde & PG_RW) || (prot & VM_PROT_WRITE) == 0) { if (vm_page_pa_tryrelock(pmap, (pde & PG_PS_FRAME) | (va & PDRMASK), &pa)) goto retry; m = PHYS_TO_VM_PAGE((pde & PG_PS_FRAME) | (va & PDRMASK)); vm_page_hold(m); } } else { pte = *pmap_pde_to_pte(pdep, va); if ((pte & PG_V) && ((pte & PG_RW) || (prot & VM_PROT_WRITE) == 0)) { if (vm_page_pa_tryrelock(pmap, pte & PG_FRAME, &pa)) goto retry; m = PHYS_TO_VM_PAGE(pte & PG_FRAME); vm_page_hold(m); } } } PA_UNLOCK_COND(pa); PMAP_UNLOCK(pmap); return (m); } vm_paddr_t pmap_kextract(vm_offset_t va) { pd_entry_t pde; vm_paddr_t pa; if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) { pa = DMAP_TO_PHYS(va); } else { pde = *vtopde(va); if (pde & PG_PS) { pa = (pde & PG_PS_FRAME) | (va & PDRMASK); } else { /* * Beware of a concurrent promotion that changes the * PDE at this point! For example, vtopte() must not * be used to access the PTE because it would use the * new PDE. It is, however, safe to use the old PDE * because the page table page is preserved by the * promotion. */ pa = *pmap_pde_to_pte(&pde, va); pa = (pa & PG_FRAME) | (va & PAGE_MASK); } } return (pa); } /*************************************************** * Low level mapping routines..... ***************************************************/ /* * Add a wired page to the kva. * Note: not SMP coherent. */ PMAP_INLINE void pmap_kenter(vm_offset_t va, vm_paddr_t pa) { pt_entry_t *pte; pte = vtopte(va); pte_store(pte, pa | X86_PG_RW | X86_PG_V | X86_PG_G); } static __inline void pmap_kenter_attr(vm_offset_t va, vm_paddr_t pa, int mode) { pt_entry_t *pte; int cache_bits; pte = vtopte(va); cache_bits = pmap_cache_bits(kernel_pmap, mode, 0); pte_store(pte, pa | X86_PG_RW | X86_PG_V | X86_PG_G | cache_bits); } /* * Remove a page from the kernel pagetables. * Note: not SMP coherent. */ PMAP_INLINE void pmap_kremove(vm_offset_t va) { pt_entry_t *pte; pte = vtopte(va); pte_clear(pte); } /* * Used to map a range of physical addresses into kernel * virtual address space. * * The value passed in '*virt' is a suggested virtual address for * the mapping. Architectures which can support a direct-mapped * physical to virtual region can return the appropriate address * within that region, leaving '*virt' unchanged. Other * architectures should map the pages starting at '*virt' and * update '*virt' with the first usable address after the mapped * region. */ vm_offset_t pmap_map(vm_offset_t *virt, vm_paddr_t start, vm_paddr_t end, int prot) { return PHYS_TO_DMAP(start); } /* * Add a list of wired pages to the kva * this routine is only used for temporary * kernel mappings that do not need to have * page modification or references recorded. * Note that old mappings are simply written * over. The page *must* be wired. * Note: SMP coherent. Uses a ranged shootdown IPI. */ void pmap_qenter(vm_offset_t sva, vm_page_t *ma, int count) { pt_entry_t *endpte, oldpte, pa, *pte; vm_page_t m; int cache_bits; oldpte = 0; pte = vtopte(sva); endpte = pte + count; while (pte < endpte) { m = *ma++; cache_bits = pmap_cache_bits(kernel_pmap, m->md.pat_mode, 0); pa = VM_PAGE_TO_PHYS(m) | cache_bits; if ((*pte & (PG_FRAME | X86_PG_PTE_CACHE)) != pa) { oldpte |= *pte; pte_store(pte, pa | X86_PG_G | X86_PG_RW | X86_PG_V); } pte++; } if (__predict_false((oldpte & X86_PG_V) != 0)) pmap_invalidate_range(kernel_pmap, sva, sva + count * PAGE_SIZE); } /* * This routine tears out page mappings from the * kernel -- it is meant only for temporary mappings. * Note: SMP coherent. Uses a ranged shootdown IPI. */ void pmap_qremove(vm_offset_t sva, int count) { vm_offset_t va; va = sva; while (count-- > 0) { KASSERT(va >= VM_MIN_KERNEL_ADDRESS, ("usermode va %lx", va)); pmap_kremove(va); va += PAGE_SIZE; } pmap_invalidate_range(kernel_pmap, sva, va); } /*************************************************** * Page table page management routines..... ***************************************************/ static __inline void pmap_free_zero_pages(struct spglist *free) { vm_page_t m; while ((m = SLIST_FIRST(free)) != NULL) { SLIST_REMOVE_HEAD(free, plinks.s.ss); /* Preserve the page's PG_ZERO setting. */ vm_page_free_toq(m); } } /* * Schedule the specified unused page table page to be freed. Specifically, * add the page to the specified list of pages that will be released to the * physical memory manager after the TLB has been updated. */ static __inline void pmap_add_delayed_free_list(vm_page_t m, struct spglist *free, boolean_t set_PG_ZERO) { if (set_PG_ZERO) m->flags |= PG_ZERO; else m->flags &= ~PG_ZERO; SLIST_INSERT_HEAD(free, m, plinks.s.ss); } /* * Inserts the specified page table page into the specified pmap's collection * of idle page table pages. Each of a pmap's page table pages is responsible * for mapping a distinct range of virtual addresses. The pmap's collection is * ordered by this virtual address range. */ static __inline int pmap_insert_pt_page(pmap_t pmap, vm_page_t mpte) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); return (vm_radix_insert(&pmap->pm_root, mpte)); } /* * Looks for a page table page mapping the specified virtual address in the * specified pmap's collection of idle page table pages. Returns NULL if there * is no page table page corresponding to the specified virtual address. */ static __inline vm_page_t pmap_lookup_pt_page(pmap_t pmap, vm_offset_t va) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); return (vm_radix_lookup(&pmap->pm_root, pmap_pde_pindex(va))); } /* * Removes the specified page table page from the specified pmap's collection * of idle page table pages. The specified page table page must be a member of * the pmap's collection. */ static __inline void pmap_remove_pt_page(pmap_t pmap, vm_page_t mpte) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); vm_radix_remove(&pmap->pm_root, mpte->pindex); } /* * Decrements a page table page's wire count, which is used to record the * number of valid page table entries within the page. If the wire count * drops to zero, then the page table page is unmapped. Returns TRUE if the * page table page was unmapped and FALSE otherwise. */ static inline boolean_t pmap_unwire_ptp(pmap_t pmap, vm_offset_t va, vm_page_t m, struct spglist *free) { --m->wire_count; if (m->wire_count == 0) { _pmap_unwire_ptp(pmap, va, m, free); return (TRUE); } else return (FALSE); } static void _pmap_unwire_ptp(pmap_t pmap, vm_offset_t va, vm_page_t m, struct spglist *free) { PMAP_LOCK_ASSERT(pmap, MA_OWNED); /* * unmap the page table page */ if (m->pindex >= (NUPDE + NUPDPE)) { /* PDP page */ pml4_entry_t *pml4; pml4 = pmap_pml4e(pmap, va); *pml4 = 0; } else if (m->pindex >= NUPDE) { /* PD page */ pdp_entry_t *pdp; pdp = pmap_pdpe(pmap, va); *pdp = 0; } else { /* PTE page */ pd_entry_t *pd; pd = pmap_pde(pmap, va); *pd = 0; } pmap_resident_count_dec(pmap, 1); if (m->pindex < NUPDE) { /* We just released a PT, unhold the matching PD */ vm_page_t pdpg; pdpg = PHYS_TO_VM_PAGE(*pmap_pdpe(pmap, va) & PG_FRAME); pmap_unwire_ptp(pmap, va, pdpg, free); } if (m->pindex >= NUPDE && m->pindex < (NUPDE + NUPDPE)) { /* We just released a PD, unhold the matching PDP */ vm_page_t pdppg; pdppg = PHYS_TO_VM_PAGE(*pmap_pml4e(pmap, va) & PG_FRAME); pmap_unwire_ptp(pmap, va, pdppg, free); } /* * This is a release store so that the ordinary store unmapping * the page table page is globally performed before TLB shoot- * down is begun. */ atomic_subtract_rel_int(&vm_cnt.v_wire_count, 1); /* * Put page on a list so that it is released after * *ALL* TLB shootdown is done */ pmap_add_delayed_free_list(m, free, TRUE); } /* * After removing a page table entry, this routine is used to * conditionally free the page, and manage the hold/wire counts. */ static int pmap_unuse_pt(pmap_t pmap, vm_offset_t va, pd_entry_t ptepde, struct spglist *free) { vm_page_t mpte; if (va >= VM_MAXUSER_ADDRESS) return (0); KASSERT(ptepde != 0, ("pmap_unuse_pt: ptepde != 0")); mpte = PHYS_TO_VM_PAGE(ptepde & PG_FRAME); return (pmap_unwire_ptp(pmap, va, mpte, free)); } void pmap_pinit0(pmap_t pmap) { int i; PMAP_LOCK_INIT(pmap); pmap->pm_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); pmap->pm_cr3 = KPML4phys; pmap->pm_root.rt_root = 0; CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); pmap->pm_flags = pmap_flags; CPU_FOREACH(i) { pmap->pm_pcids[i].pm_pcid = PMAP_PCID_NONE; pmap->pm_pcids[i].pm_gen = 0; } PCPU_SET(curpmap, kernel_pmap); pmap_activate(curthread); CPU_FILL(&kernel_pmap->pm_active); } /* * Initialize a preallocated and zeroed pmap structure, * such as one in a vmspace structure. */ int pmap_pinit_type(pmap_t pmap, enum pmap_type pm_type, int flags) { vm_page_t pml4pg; vm_paddr_t pml4phys; int i; /* * allocate the page directory page */ while ((pml4pg = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO)) == NULL) VM_WAIT; pml4phys = VM_PAGE_TO_PHYS(pml4pg); pmap->pm_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(pml4phys); CPU_FOREACH(i) { pmap->pm_pcids[i].pm_pcid = PMAP_PCID_NONE; pmap->pm_pcids[i].pm_gen = 0; } pmap->pm_cr3 = ~0; /* initialize to an invalid value */ if ((pml4pg->flags & PG_ZERO) == 0) pagezero(pmap->pm_pml4); /* * Do not install the host kernel mappings in the nested page * tables. These mappings are meaningless in the guest physical * address space. */ if ((pmap->pm_type = pm_type) == PT_X86) { pmap->pm_cr3 = pml4phys; /* Wire in kernel global address entries. */ for (i = 0; i < NKPML4E; i++) { pmap->pm_pml4[KPML4BASE + i] = (KPDPphys + ptoa(i)) | X86_PG_RW | X86_PG_V | PG_U; } for (i = 0; i < ndmpdpphys; i++) { pmap->pm_pml4[DMPML4I + i] = (DMPDPphys + ptoa(i)) | X86_PG_RW | X86_PG_V | PG_U; } /* install self-referential address mapping entry(s) */ pmap->pm_pml4[PML4PML4I] = VM_PAGE_TO_PHYS(pml4pg) | X86_PG_V | X86_PG_RW | X86_PG_A | X86_PG_M; } pmap->pm_root.rt_root = 0; CPU_ZERO(&pmap->pm_active); TAILQ_INIT(&pmap->pm_pvchunk); bzero(&pmap->pm_stats, sizeof pmap->pm_stats); pmap->pm_flags = flags; pmap->pm_eptgen = 0; return (1); } int pmap_pinit(pmap_t pmap) { return (pmap_pinit_type(pmap, PT_X86, pmap_flags)); } /* * This routine is called if the desired page table page does not exist. * * If page table page allocation fails, this routine may sleep before * returning NULL. It sleeps only if a lock pointer was given. * * Note: If a page allocation fails at page table level two or three, * one or two pages may be held during the wait, only to be released * afterwards. This conservative approach is easily argued to avoid * race conditions. */ static vm_page_t _pmap_allocpte(pmap_t pmap, vm_pindex_t ptepindex, struct rwlock **lockp) { vm_page_t m, pdppg, pdpg; pt_entry_t PG_A, PG_M, PG_RW, PG_V; PMAP_LOCK_ASSERT(pmap, MA_OWNED); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); /* * Allocate a page table page. */ if ((m = vm_page_alloc(NULL, ptepindex, VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO)) == NULL) { if (lockp != NULL) { RELEASE_PV_LIST_LOCK(lockp); PMAP_UNLOCK(pmap); rw_runlock(&pvh_global_lock); VM_WAIT; rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); } /* * Indicate the need to retry. While waiting, the page table * page may have been allocated. */ return (NULL); } if ((m->flags & PG_ZERO) == 0) pmap_zero_page(m); /* * Map the pagetable page into the process address space, if * it isn't already there. */ if (ptepindex >= (NUPDE + NUPDPE)) { pml4_entry_t *pml4; vm_pindex_t pml4index; /* Wire up a new PDPE page */ pml4index = ptepindex - (NUPDE + NUPDPE); pml4 = &pmap->pm_pml4[pml4index]; *pml4 = VM_PAGE_TO_PHYS(m) | PG_U | PG_RW | PG_V | PG_A | PG_M; } else if (ptepindex >= NUPDE) { vm_pindex_t pml4index; vm_pindex_t pdpindex; pml4_entry_t *pml4; pdp_entry_t *pdp; /* Wire up a new PDE page */ pdpindex = ptepindex - NUPDE; pml4index = pdpindex >> NPML4EPGSHIFT; pml4 = &pmap->pm_pml4[pml4index]; if ((*pml4 & PG_V) == 0) { /* Have to allocate a new pdp, recurse */ if (_pmap_allocpte(pmap, NUPDE + NUPDPE + pml4index, lockp) == NULL) { --m->wire_count; atomic_subtract_int(&vm_cnt.v_wire_count, 1); vm_page_free_zero(m); return (NULL); } } else { /* Add reference to pdp page */ pdppg = PHYS_TO_VM_PAGE(*pml4 & PG_FRAME); pdppg->wire_count++; } pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); /* Now find the pdp page */ pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)]; *pdp = VM_PAGE_TO_PHYS(m) | PG_U | PG_RW | PG_V | PG_A | PG_M; } else { vm_pindex_t pml4index; vm_pindex_t pdpindex; pml4_entry_t *pml4; pdp_entry_t *pdp; pd_entry_t *pd; /* Wire up a new PTE page */ pdpindex = ptepindex >> NPDPEPGSHIFT; pml4index = pdpindex >> NPML4EPGSHIFT; /* First, find the pdp and check that its valid. */ pml4 = &pmap->pm_pml4[pml4index]; if ((*pml4 & PG_V) == 0) { /* Have to allocate a new pd, recurse */ if (_pmap_allocpte(pmap, NUPDE + pdpindex, lockp) == NULL) { --m->wire_count; atomic_subtract_int(&vm_cnt.v_wire_count, 1); vm_page_free_zero(m); return (NULL); } pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)]; } else { pdp = (pdp_entry_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); pdp = &pdp[pdpindex & ((1ul << NPDPEPGSHIFT) - 1)]; if ((*pdp & PG_V) == 0) { /* Have to allocate a new pd, recurse */ if (_pmap_allocpte(pmap, NUPDE + pdpindex, lockp) == NULL) { --m->wire_count; atomic_subtract_int(&vm_cnt.v_wire_count, 1); vm_page_free_zero(m); return (NULL); } } else { /* Add reference to the pd page */ pdpg = PHYS_TO_VM_PAGE(*pdp & PG_FRAME); pdpg->wire_count++; } } pd = (pd_entry_t *)PHYS_TO_DMAP(*pdp & PG_FRAME); /* Now we know where the page directory page is */ pd = &pd[ptepindex & ((1ul << NPDEPGSHIFT) - 1)]; *pd = VM_PAGE_TO_PHYS(m) | PG_U | PG_RW | PG_V | PG_A | PG_M; } pmap_resident_count_inc(pmap, 1); return (m); } static vm_page_t pmap_allocpde(pmap_t pmap, vm_offset_t va, struct rwlock **lockp) { vm_pindex_t pdpindex, ptepindex; pdp_entry_t *pdpe, PG_V; vm_page_t pdpg; PG_V = pmap_valid_bit(pmap); retry: pdpe = pmap_pdpe(pmap, va); if (pdpe != NULL && (*pdpe & PG_V) != 0) { /* Add a reference to the pd page. */ pdpg = PHYS_TO_VM_PAGE(*pdpe & PG_FRAME); pdpg->wire_count++; } else { /* Allocate a pd page. */ ptepindex = pmap_pde_pindex(va); pdpindex = ptepindex >> NPDPEPGSHIFT; pdpg = _pmap_allocpte(pmap, NUPDE + pdpindex, lockp); if (pdpg == NULL && lockp != NULL) goto retry; } return (pdpg); } static vm_page_t pmap_allocpte(pmap_t pmap, vm_offset_t va, struct rwlock **lockp) { vm_pindex_t ptepindex; pd_entry_t *pd, PG_V; vm_page_t m; PG_V = pmap_valid_bit(pmap); /* * Calculate pagetable page index */ ptepindex = pmap_pde_pindex(va); retry: /* * Get the page directory entry */ pd = pmap_pde(pmap, va); /* * This supports switching from a 2MB page to a * normal 4K page. */ if (pd != NULL && (*pd & (PG_PS | PG_V)) == (PG_PS | PG_V)) { if (!pmap_demote_pde_locked(pmap, pd, va, lockp)) { /* * Invalidation of the 2MB page mapping may have caused * the deallocation of the underlying PD page. */ pd = NULL; } } /* * If the page table page is mapped, we just increment the * hold count, and activate it. */ if (pd != NULL && (*pd & PG_V) != 0) { m = PHYS_TO_VM_PAGE(*pd & PG_FRAME); m->wire_count++; } else { /* * Here if the pte page isn't mapped, or if it has been * deallocated. */ m = _pmap_allocpte(pmap, ptepindex, lockp); if (m == NULL && lockp != NULL) goto retry; } return (m); } /*************************************************** * Pmap allocation/deallocation routines. ***************************************************/ /* * Release any resources held by the given physical map. * Called when a pmap initialized by pmap_pinit is being released. * Should only be called if the map contains no valid mappings. */ void pmap_release(pmap_t pmap) { vm_page_t m; int i; KASSERT(pmap->pm_stats.resident_count == 0, ("pmap_release: pmap resident count %ld != 0", pmap->pm_stats.resident_count)); KASSERT(vm_radix_is_empty(&pmap->pm_root), ("pmap_release: pmap has reserved page table page(s)")); KASSERT(CPU_EMPTY(&pmap->pm_active), ("releasing active pmap %p", pmap)); m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pmap->pm_pml4)); for (i = 0; i < NKPML4E; i++) /* KVA */ pmap->pm_pml4[KPML4BASE + i] = 0; for (i = 0; i < ndmpdpphys; i++)/* Direct Map */ pmap->pm_pml4[DMPML4I + i] = 0; pmap->pm_pml4[PML4PML4I] = 0; /* Recursive Mapping */ m->wire_count--; atomic_subtract_int(&vm_cnt.v_wire_count, 1); vm_page_free_zero(m); } static int kvm_size(SYSCTL_HANDLER_ARGS) { unsigned long ksize = VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS; return sysctl_handle_long(oidp, &ksize, 0, req); } SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG|CTLFLAG_RD, 0, 0, kvm_size, "LU", "Size of KVM"); static int kvm_free(SYSCTL_HANDLER_ARGS) { unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end; return sysctl_handle_long(oidp, &kfree, 0, req); } SYSCTL_PROC(_vm, OID_AUTO, kvm_free, CTLTYPE_LONG|CTLFLAG_RD, 0, 0, kvm_free, "LU", "Amount of KVM free"); /* * grow the number of kernel page table entries, if needed */ void pmap_growkernel(vm_offset_t addr) { vm_paddr_t paddr; vm_page_t nkpg; pd_entry_t *pde, newpdir; pdp_entry_t *pdpe; mtx_assert(&kernel_map->system_mtx, MA_OWNED); /* * Return if "addr" is within the range of kernel page table pages * that were preallocated during pmap bootstrap. Moreover, leave * "kernel_vm_end" and the kernel page table as they were. * * The correctness of this action is based on the following * argument: vm_map_insert() allocates contiguous ranges of the * kernel virtual address space. It calls this function if a range * ends after "kernel_vm_end". If the kernel is mapped between * "kernel_vm_end" and "addr", then the range cannot begin at * "kernel_vm_end". In fact, its beginning address cannot be less * than the kernel. Thus, there is no immediate need to allocate * any new kernel page table pages between "kernel_vm_end" and * "KERNBASE". */ if (KERNBASE < addr && addr <= KERNBASE + nkpt * NBPDR) return; addr = roundup2(addr, NBPDR); if (addr - 1 >= kernel_map->max_offset) addr = kernel_map->max_offset; while (kernel_vm_end < addr) { pdpe = pmap_pdpe(kernel_pmap, kernel_vm_end); if ((*pdpe & X86_PG_V) == 0) { /* We need a new PDP entry */ nkpg = vm_page_alloc(NULL, kernel_vm_end >> PDPSHIFT, VM_ALLOC_INTERRUPT | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO); if (nkpg == NULL) panic("pmap_growkernel: no memory to grow kernel"); if ((nkpg->flags & PG_ZERO) == 0) pmap_zero_page(nkpg); paddr = VM_PAGE_TO_PHYS(nkpg); *pdpe = (pdp_entry_t)(paddr | X86_PG_V | X86_PG_RW | X86_PG_A | X86_PG_M); continue; /* try again */ } pde = pmap_pdpe_to_pde(pdpe, kernel_vm_end); if ((*pde & X86_PG_V) != 0) { kernel_vm_end = (kernel_vm_end + NBPDR) & ~PDRMASK; if (kernel_vm_end - 1 >= kernel_map->max_offset) { kernel_vm_end = kernel_map->max_offset; break; } continue; } nkpg = vm_page_alloc(NULL, pmap_pde_pindex(kernel_vm_end), VM_ALLOC_INTERRUPT | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_ZERO); if (nkpg == NULL) panic("pmap_growkernel: no memory to grow kernel"); if ((nkpg->flags & PG_ZERO) == 0) pmap_zero_page(nkpg); paddr = VM_PAGE_TO_PHYS(nkpg); newpdir = paddr | X86_PG_V | X86_PG_RW | X86_PG_A | X86_PG_M; pde_store(pde, newpdir); kernel_vm_end = (kernel_vm_end + NBPDR) & ~PDRMASK; if (kernel_vm_end - 1 >= kernel_map->max_offset) { kernel_vm_end = kernel_map->max_offset; break; } } } /*************************************************** * page management routines. ***************************************************/ CTASSERT(sizeof(struct pv_chunk) == PAGE_SIZE); CTASSERT(_NPCM == 3); CTASSERT(_NPCPV == 168); static __inline struct pv_chunk * pv_to_chunk(pv_entry_t pv) { return ((struct pv_chunk *)((uintptr_t)pv & ~(uintptr_t)PAGE_MASK)); } #define PV_PMAP(pv) (pv_to_chunk(pv)->pc_pmap) #define PC_FREE0 0xfffffffffffffffful #define PC_FREE1 0xfffffffffffffffful #define PC_FREE2 0x000000fffffffffful static const uint64_t pc_freemask[_NPCM] = { PC_FREE0, PC_FREE1, PC_FREE2 }; #ifdef PV_STATS static int pc_chunk_count, pc_chunk_allocs, pc_chunk_frees, pc_chunk_tryfail; SYSCTL_INT(_vm_pmap, OID_AUTO, pc_chunk_count, CTLFLAG_RD, &pc_chunk_count, 0, "Current number of pv entry chunks"); SYSCTL_INT(_vm_pmap, OID_AUTO, pc_chunk_allocs, CTLFLAG_RD, &pc_chunk_allocs, 0, "Current number of pv entry chunks allocated"); SYSCTL_INT(_vm_pmap, OID_AUTO, pc_chunk_frees, CTLFLAG_RD, &pc_chunk_frees, 0, "Current number of pv entry chunks frees"); SYSCTL_INT(_vm_pmap, OID_AUTO, pc_chunk_tryfail, CTLFLAG_RD, &pc_chunk_tryfail, 0, "Number of times tried to get a chunk page but failed."); static long pv_entry_frees, pv_entry_allocs, pv_entry_count; static int pv_entry_spare; SYSCTL_LONG(_vm_pmap, OID_AUTO, pv_entry_frees, CTLFLAG_RD, &pv_entry_frees, 0, "Current number of pv entry frees"); SYSCTL_LONG(_vm_pmap, OID_AUTO, pv_entry_allocs, CTLFLAG_RD, &pv_entry_allocs, 0, "Current number of pv entry allocs"); SYSCTL_LONG(_vm_pmap, OID_AUTO, pv_entry_count, CTLFLAG_RD, &pv_entry_count, 0, "Current number of pv entries"); SYSCTL_INT(_vm_pmap, OID_AUTO, pv_entry_spare, CTLFLAG_RD, &pv_entry_spare, 0, "Current number of spare pv entries"); #endif /* * We are in a serious low memory condition. Resort to * drastic measures to free some pages so we can allocate * another pv entry chunk. * * Returns NULL if PV entries were reclaimed from the specified pmap. * * We do not, however, unmap 2mpages because subsequent accesses will * allocate per-page pv entries until repromotion occurs, thereby * exacerbating the shortage of free pv entries. */ static vm_page_t reclaim_pv_chunk(pmap_t locked_pmap, struct rwlock **lockp) { struct pch new_tail; struct pv_chunk *pc; struct md_page *pvh; pd_entry_t *pde; pmap_t pmap; pt_entry_t *pte, tpte; pt_entry_t PG_G, PG_A, PG_M, PG_RW; pv_entry_t pv; vm_offset_t va; vm_page_t m, m_pc; struct spglist free; uint64_t inuse; int bit, field, freed; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(locked_pmap, MA_OWNED); KASSERT(lockp != NULL, ("reclaim_pv_chunk: lockp is NULL")); pmap = NULL; m_pc = NULL; PG_G = PG_A = PG_M = PG_RW = 0; SLIST_INIT(&free); TAILQ_INIT(&new_tail); mtx_lock(&pv_chunks_mutex); while ((pc = TAILQ_FIRST(&pv_chunks)) != NULL && SLIST_EMPTY(&free)) { TAILQ_REMOVE(&pv_chunks, pc, pc_lru); mtx_unlock(&pv_chunks_mutex); if (pmap != pc->pc_pmap) { if (pmap != NULL) { pmap_invalidate_all(pmap); if (pmap != locked_pmap) PMAP_UNLOCK(pmap); } pmap = pc->pc_pmap; /* Avoid deadlock and lock recursion. */ if (pmap > locked_pmap) { RELEASE_PV_LIST_LOCK(lockp); PMAP_LOCK(pmap); } else if (pmap != locked_pmap && !PMAP_TRYLOCK(pmap)) { pmap = NULL; TAILQ_INSERT_TAIL(&new_tail, pc, pc_lru); mtx_lock(&pv_chunks_mutex); continue; } PG_G = pmap_global_bit(pmap); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); } /* * Destroy every non-wired, 4 KB page mapping in the chunk. */ freed = 0; for (field = 0; field < _NPCM; field++) { for (inuse = ~pc->pc_map[field] & pc_freemask[field]; inuse != 0; inuse &= ~(1UL << bit)) { bit = bsfq(inuse); pv = &pc->pc_pventry[field * 64 + bit]; va = pv->pv_va; pde = pmap_pde(pmap, va); if ((*pde & PG_PS) != 0) continue; pte = pmap_pde_to_pte(pde, va); if ((*pte & PG_W) != 0) continue; tpte = pte_load_clear(pte); if ((tpte & PG_G) != 0) pmap_invalidate_page(pmap, va); m = PHYS_TO_VM_PAGE(tpte & PG_FRAME); if ((tpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(m); if ((tpte & PG_A) != 0) vm_page_aflag_set(m, PGA_REFERENCED); CHANGE_PV_LIST_LOCK_TO_VM_PAGE(lockp, m); TAILQ_REMOVE(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; if (TAILQ_EMPTY(&m->md.pv_list) && (m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); if (TAILQ_EMPTY(&pvh->pv_list)) { vm_page_aflag_clear(m, PGA_WRITEABLE); } } pc->pc_map[field] |= 1UL << bit; pmap_unuse_pt(pmap, va, *pde, &free); freed++; } } if (freed == 0) { TAILQ_INSERT_TAIL(&new_tail, pc, pc_lru); mtx_lock(&pv_chunks_mutex); continue; } /* Every freed mapping is for a 4 KB page. */ pmap_resident_count_dec(pmap, freed); PV_STAT(atomic_add_long(&pv_entry_frees, freed)); PV_STAT(atomic_add_int(&pv_entry_spare, freed)); PV_STAT(atomic_subtract_long(&pv_entry_count, freed)); TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); if (pc->pc_map[0] == PC_FREE0 && pc->pc_map[1] == PC_FREE1 && pc->pc_map[2] == PC_FREE2) { PV_STAT(atomic_subtract_int(&pv_entry_spare, _NPCPV)); PV_STAT(atomic_subtract_int(&pc_chunk_count, 1)); PV_STAT(atomic_add_int(&pc_chunk_frees, 1)); /* Entire chunk is free; return it. */ m_pc = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pc)); dump_drop_page(m_pc->phys_addr); mtx_lock(&pv_chunks_mutex); break; } TAILQ_INSERT_HEAD(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_TAIL(&new_tail, pc, pc_lru); mtx_lock(&pv_chunks_mutex); /* One freed pv entry in locked_pmap is sufficient. */ if (pmap == locked_pmap) break; } TAILQ_CONCAT(&pv_chunks, &new_tail, pc_lru); mtx_unlock(&pv_chunks_mutex); if (pmap != NULL) { pmap_invalidate_all(pmap); if (pmap != locked_pmap) PMAP_UNLOCK(pmap); } if (m_pc == NULL && !SLIST_EMPTY(&free)) { m_pc = SLIST_FIRST(&free); SLIST_REMOVE_HEAD(&free, plinks.s.ss); /* Recycle a freed page table page. */ m_pc->wire_count = 1; atomic_add_int(&vm_cnt.v_wire_count, 1); } pmap_free_zero_pages(&free); return (m_pc); } /* * free the pv_entry back to the free list */ static void free_pv_entry(pmap_t pmap, pv_entry_t pv) { struct pv_chunk *pc; int idx, field, bit; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); PV_STAT(atomic_add_long(&pv_entry_frees, 1)); PV_STAT(atomic_add_int(&pv_entry_spare, 1)); PV_STAT(atomic_subtract_long(&pv_entry_count, 1)); pc = pv_to_chunk(pv); idx = pv - &pc->pc_pventry[0]; field = idx / 64; bit = idx % 64; pc->pc_map[field] |= 1ul << bit; if (pc->pc_map[0] != PC_FREE0 || pc->pc_map[1] != PC_FREE1 || pc->pc_map[2] != PC_FREE2) { /* 98% of the time, pc is already at the head of the list. */ if (__predict_false(pc != TAILQ_FIRST(&pmap->pm_pvchunk))) { TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_HEAD(&pmap->pm_pvchunk, pc, pc_list); } return; } TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); free_pv_chunk(pc); } static void free_pv_chunk(struct pv_chunk *pc) { vm_page_t m; mtx_lock(&pv_chunks_mutex); TAILQ_REMOVE(&pv_chunks, pc, pc_lru); mtx_unlock(&pv_chunks_mutex); PV_STAT(atomic_subtract_int(&pv_entry_spare, _NPCPV)); PV_STAT(atomic_subtract_int(&pc_chunk_count, 1)); PV_STAT(atomic_add_int(&pc_chunk_frees, 1)); /* entire chunk is free, return it */ m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pc)); dump_drop_page(m->phys_addr); vm_page_unwire(m, PQ_NONE); vm_page_free(m); } /* * Returns a new PV entry, allocating a new PV chunk from the system when * needed. If this PV chunk allocation fails and a PV list lock pointer was * given, a PV chunk is reclaimed from an arbitrary pmap. Otherwise, NULL is * returned. * * The given PV list lock may be released. */ static pv_entry_t get_pv_entry(pmap_t pmap, struct rwlock **lockp) { int bit, field; pv_entry_t pv; struct pv_chunk *pc; vm_page_t m; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); PV_STAT(atomic_add_long(&pv_entry_allocs, 1)); retry: pc = TAILQ_FIRST(&pmap->pm_pvchunk); if (pc != NULL) { for (field = 0; field < _NPCM; field++) { if (pc->pc_map[field]) { bit = bsfq(pc->pc_map[field]); break; } } if (field < _NPCM) { pv = &pc->pc_pventry[field * 64 + bit]; pc->pc_map[field] &= ~(1ul << bit); /* If this was the last item, move it to tail */ if (pc->pc_map[0] == 0 && pc->pc_map[1] == 0 && pc->pc_map[2] == 0) { TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_TAIL(&pmap->pm_pvchunk, pc, pc_list); } PV_STAT(atomic_add_long(&pv_entry_count, 1)); PV_STAT(atomic_subtract_int(&pv_entry_spare, 1)); return (pv); } } /* No free items, allocate another chunk */ m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED); if (m == NULL) { if (lockp == NULL) { PV_STAT(pc_chunk_tryfail++); return (NULL); } m = reclaim_pv_chunk(pmap, lockp); if (m == NULL) goto retry; } PV_STAT(atomic_add_int(&pc_chunk_count, 1)); PV_STAT(atomic_add_int(&pc_chunk_allocs, 1)); dump_add_page(m->phys_addr); pc = (void *)PHYS_TO_DMAP(m->phys_addr); pc->pc_pmap = pmap; pc->pc_map[0] = PC_FREE0 & ~1ul; /* preallocated bit 0 */ pc->pc_map[1] = PC_FREE1; pc->pc_map[2] = PC_FREE2; mtx_lock(&pv_chunks_mutex); TAILQ_INSERT_TAIL(&pv_chunks, pc, pc_lru); mtx_unlock(&pv_chunks_mutex); pv = &pc->pc_pventry[0]; TAILQ_INSERT_HEAD(&pmap->pm_pvchunk, pc, pc_list); PV_STAT(atomic_add_long(&pv_entry_count, 1)); PV_STAT(atomic_add_int(&pv_entry_spare, _NPCPV - 1)); return (pv); } /* * Returns the number of one bits within the given PV chunk map element. * * The erratas for Intel processors state that "POPCNT Instruction May * Take Longer to Execute Than Expected". It is believed that the * issue is the spurious dependency on the destination register. * Provide a hint to the register rename logic that the destination * value is overwritten, by clearing it, as suggested in the * optimization manual. It should be cheap for unaffected processors * as well. * * Reference numbers for erratas are * 4th Gen Core: HSD146 * 5th Gen Core: BDM85 */ static int popcnt_pc_map_elem_pq(uint64_t elem) { u_long result; __asm __volatile("xorl %k0,%k0;popcntq %1,%0" : "=&r" (result) : "rm" (elem)); return (result); } /* * Ensure that the number of spare PV entries in the specified pmap meets or * exceeds the given count, "needed". * * The given PV list lock may be released. */ static void reserve_pv_entries(pmap_t pmap, int needed, struct rwlock **lockp) { struct pch new_tail; struct pv_chunk *pc; int avail, free; vm_page_t m; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); KASSERT(lockp != NULL, ("reserve_pv_entries: lockp is NULL")); /* * Newly allocated PV chunks must be stored in a private list until * the required number of PV chunks have been allocated. Otherwise, * reclaim_pv_chunk() could recycle one of these chunks. In * contrast, these chunks must be added to the pmap upon allocation. */ TAILQ_INIT(&new_tail); retry: avail = 0; TAILQ_FOREACH(pc, &pmap->pm_pvchunk, pc_list) { #ifndef __POPCNT__ if ((cpu_feature2 & CPUID2_POPCNT) == 0) { free = bitcount64(pc->pc_map[0]); free += bitcount64(pc->pc_map[1]); free += bitcount64(pc->pc_map[2]); } else #endif { free = popcnt_pc_map_elem_pq(pc->pc_map[0]); free += popcnt_pc_map_elem_pq(pc->pc_map[1]); free += popcnt_pc_map_elem_pq(pc->pc_map[2]); } if (free == 0) break; avail += free; if (avail >= needed) break; } for (; avail < needed; avail += _NPCPV) { m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED); if (m == NULL) { m = reclaim_pv_chunk(pmap, lockp); if (m == NULL) goto retry; } PV_STAT(atomic_add_int(&pc_chunk_count, 1)); PV_STAT(atomic_add_int(&pc_chunk_allocs, 1)); dump_add_page(m->phys_addr); pc = (void *)PHYS_TO_DMAP(m->phys_addr); pc->pc_pmap = pmap; pc->pc_map[0] = PC_FREE0; pc->pc_map[1] = PC_FREE1; pc->pc_map[2] = PC_FREE2; TAILQ_INSERT_HEAD(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_TAIL(&new_tail, pc, pc_lru); PV_STAT(atomic_add_int(&pv_entry_spare, _NPCPV)); } if (!TAILQ_EMPTY(&new_tail)) { mtx_lock(&pv_chunks_mutex); TAILQ_CONCAT(&pv_chunks, &new_tail, pc_lru); mtx_unlock(&pv_chunks_mutex); } } /* * First find and then remove the pv entry for the specified pmap and virtual * address from the specified pv list. Returns the pv entry if found and NULL * otherwise. This operation can be performed on pv lists for either 4KB or * 2MB page mappings. */ static __inline pv_entry_t pmap_pvh_remove(struct md_page *pvh, pmap_t pmap, vm_offset_t va) { pv_entry_t pv; rw_assert(&pvh_global_lock, RA_LOCKED); TAILQ_FOREACH(pv, &pvh->pv_list, pv_next) { if (pmap == PV_PMAP(pv) && va == pv->pv_va) { TAILQ_REMOVE(&pvh->pv_list, pv, pv_next); pvh->pv_gen++; break; } } return (pv); } /* * After demotion from a 2MB page mapping to 512 4KB page mappings, * destroy the pv entry for the 2MB page mapping and reinstantiate the pv * entries for each of the 4KB page mappings. */ static void pmap_pv_demote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp) { struct md_page *pvh; struct pv_chunk *pc; pv_entry_t pv; vm_offset_t va_last; vm_page_t m; int bit, field; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); KASSERT((pa & PDRMASK) == 0, ("pmap_pv_demote_pde: pa is not 2mpage aligned")); CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa); /* * Transfer the 2mpage's pv entry for this mapping to the first * page's pv list. Once this transfer begins, the pv list lock * must not be released until the last pv entry is reinstantiated. */ pvh = pa_to_pvh(pa); va = trunc_2mpage(va); pv = pmap_pvh_remove(pvh, pmap, va); KASSERT(pv != NULL, ("pmap_pv_demote_pde: pv not found")); m = PHYS_TO_VM_PAGE(pa); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; /* Instantiate the remaining NPTEPG - 1 pv entries. */ PV_STAT(atomic_add_long(&pv_entry_allocs, NPTEPG - 1)); va_last = va + NBPDR - PAGE_SIZE; for (;;) { pc = TAILQ_FIRST(&pmap->pm_pvchunk); KASSERT(pc->pc_map[0] != 0 || pc->pc_map[1] != 0 || pc->pc_map[2] != 0, ("pmap_pv_demote_pde: missing spare")); for (field = 0; field < _NPCM; field++) { while (pc->pc_map[field]) { bit = bsfq(pc->pc_map[field]); pc->pc_map[field] &= ~(1ul << bit); pv = &pc->pc_pventry[field * 64 + bit]; va += PAGE_SIZE; pv->pv_va = va; m++; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_pv_demote_pde: page %p is not managed", m)); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; if (va == va_last) goto out; } } TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_TAIL(&pmap->pm_pvchunk, pc, pc_list); } out: if (pc->pc_map[0] == 0 && pc->pc_map[1] == 0 && pc->pc_map[2] == 0) { TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); TAILQ_INSERT_TAIL(&pmap->pm_pvchunk, pc, pc_list); } PV_STAT(atomic_add_long(&pv_entry_count, NPTEPG - 1)); PV_STAT(atomic_subtract_int(&pv_entry_spare, NPTEPG - 1)); } /* * After promotion from 512 4KB page mappings to a single 2MB page mapping, * replace the many pv entries for the 4KB page mappings by a single pv entry * for the 2MB page mapping. */ static void pmap_pv_promote_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp) { struct md_page *pvh; pv_entry_t pv; vm_offset_t va_last; vm_page_t m; rw_assert(&pvh_global_lock, RA_LOCKED); KASSERT((pa & PDRMASK) == 0, ("pmap_pv_promote_pde: pa is not 2mpage aligned")); CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa); /* * Transfer the first page's pv entry for this mapping to the 2mpage's * pv list. Aside from avoiding the cost of a call to get_pv_entry(), * a transfer avoids the possibility that get_pv_entry() calls * reclaim_pv_chunk() and that reclaim_pv_chunk() removes one of the * mappings that is being promoted. */ m = PHYS_TO_VM_PAGE(pa); va = trunc_2mpage(va); pv = pmap_pvh_remove(&m->md, pmap, va); KASSERT(pv != NULL, ("pmap_pv_promote_pde: pv not found")); pvh = pa_to_pvh(pa); TAILQ_INSERT_TAIL(&pvh->pv_list, pv, pv_next); pvh->pv_gen++; /* Free the remaining NPTEPG - 1 pv entries. */ va_last = va + NBPDR - PAGE_SIZE; do { m++; va += PAGE_SIZE; pmap_pvh_free(&m->md, pmap, va); } while (va < va_last); } /* * First find and then destroy the pv entry for the specified pmap and virtual * address. This operation can be performed on pv lists for either 4KB or 2MB * page mappings. */ static void pmap_pvh_free(struct md_page *pvh, pmap_t pmap, vm_offset_t va) { pv_entry_t pv; pv = pmap_pvh_remove(pvh, pmap, va); KASSERT(pv != NULL, ("pmap_pvh_free: pv not found")); free_pv_entry(pmap, pv); } /* * Conditionally create the PV entry for a 4KB page mapping if the required * memory can be allocated without resorting to reclamation. */ static boolean_t pmap_try_insert_pv_entry(pmap_t pmap, vm_offset_t va, vm_page_t m, struct rwlock **lockp) { pv_entry_t pv; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); /* Pass NULL instead of the lock pointer to disable reclamation. */ if ((pv = get_pv_entry(pmap, NULL)) != NULL) { pv->pv_va = va; CHANGE_PV_LIST_LOCK_TO_VM_PAGE(lockp, m); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; return (TRUE); } else return (FALSE); } /* * Conditionally create the PV entry for a 2MB page mapping if the required * memory can be allocated without resorting to reclamation. */ static boolean_t pmap_pv_insert_pde(pmap_t pmap, vm_offset_t va, vm_paddr_t pa, struct rwlock **lockp) { struct md_page *pvh; pv_entry_t pv; rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); /* Pass NULL instead of the lock pointer to disable reclamation. */ if ((pv = get_pv_entry(pmap, NULL)) != NULL) { pv->pv_va = va; CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, pa); pvh = pa_to_pvh(pa); TAILQ_INSERT_TAIL(&pvh->pv_list, pv, pv_next); pvh->pv_gen++; return (TRUE); } else return (FALSE); } /* * Fills a page table page with mappings to consecutive physical pages. */ static void pmap_fill_ptp(pt_entry_t *firstpte, pt_entry_t newpte) { pt_entry_t *pte; for (pte = firstpte; pte < firstpte + NPTEPG; pte++) { *pte = newpte; newpte += PAGE_SIZE; } } /* * Tries to demote a 2MB page mapping. If demotion fails, the 2MB page * mapping is invalidated. */ static boolean_t pmap_demote_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va) { struct rwlock *lock; boolean_t rv; lock = NULL; rv = pmap_demote_pde_locked(pmap, pde, va, &lock); if (lock != NULL) rw_wunlock(lock); return (rv); } static boolean_t pmap_demote_pde_locked(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, struct rwlock **lockp) { pd_entry_t newpde, oldpde; pt_entry_t *firstpte, newpte; pt_entry_t PG_A, PG_G, PG_M, PG_RW, PG_V; vm_paddr_t mptepa; vm_page_t mpte; struct spglist free; int PG_PTE_CACHE; PG_G = pmap_global_bit(pmap); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_PTE_CACHE = pmap_cache_mask(pmap, 0); PMAP_LOCK_ASSERT(pmap, MA_OWNED); oldpde = *pde; KASSERT((oldpde & (PG_PS | PG_V)) == (PG_PS | PG_V), ("pmap_demote_pde: oldpde is missing PG_PS and/or PG_V")); if ((oldpde & PG_A) != 0 && (mpte = pmap_lookup_pt_page(pmap, va)) != NULL) pmap_remove_pt_page(pmap, mpte); else { KASSERT((oldpde & PG_W) == 0, ("pmap_demote_pde: page table page for a wired mapping" " is missing")); /* * Invalidate the 2MB page mapping and return "failure" if the * mapping was never accessed or the allocation of the new * page table page fails. If the 2MB page mapping belongs to * the direct map region of the kernel's address space, then * the page allocation request specifies the highest possible * priority (VM_ALLOC_INTERRUPT). Otherwise, the priority is * normal. Page table pages are preallocated for every other * part of the kernel address space, so the direct map region * is the only part of the kernel address space that must be * handled here. */ if ((oldpde & PG_A) == 0 || (mpte = vm_page_alloc(NULL, pmap_pde_pindex(va), (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS ? VM_ALLOC_INTERRUPT : VM_ALLOC_NORMAL) | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) { SLIST_INIT(&free); pmap_remove_pde(pmap, pde, trunc_2mpage(va), &free, lockp); pmap_invalidate_page(pmap, trunc_2mpage(va)); pmap_free_zero_pages(&free); CTR2(KTR_PMAP, "pmap_demote_pde: failure for va %#lx" " in pmap %p", va, pmap); return (FALSE); } if (va < VM_MAXUSER_ADDRESS) pmap_resident_count_inc(pmap, 1); } mptepa = VM_PAGE_TO_PHYS(mpte); firstpte = (pt_entry_t *)PHYS_TO_DMAP(mptepa); newpde = mptepa | PG_M | PG_A | (oldpde & PG_U) | PG_RW | PG_V; KASSERT((oldpde & PG_A) != 0, ("pmap_demote_pde: oldpde is missing PG_A")); KASSERT((oldpde & (PG_M | PG_RW)) != PG_RW, ("pmap_demote_pde: oldpde is missing PG_M")); newpte = oldpde & ~PG_PS; newpte = pmap_swap_pat(pmap, newpte); /* * If the page table page is new, initialize it. */ if (mpte->wire_count == 1) { mpte->wire_count = NPTEPG; pmap_fill_ptp(firstpte, newpte); } KASSERT((*firstpte & PG_FRAME) == (newpte & PG_FRAME), ("pmap_demote_pde: firstpte and newpte map different physical" " addresses")); /* * If the mapping has changed attributes, update the page table * entries. */ if ((*firstpte & PG_PTE_PROMOTE) != (newpte & PG_PTE_PROMOTE)) pmap_fill_ptp(firstpte, newpte); /* * The spare PV entries must be reserved prior to demoting the * mapping, that is, prior to changing the PDE. Otherwise, the state * of the PDE and the PV lists will be inconsistent, which can result * in reclaim_pv_chunk() attempting to remove a PV entry from the * wrong PV list and pmap_pv_demote_pde() failing to find the expected * PV entry for the 2MB page mapping that is being demoted. */ if ((oldpde & PG_MANAGED) != 0) reserve_pv_entries(pmap, NPTEPG - 1, lockp); /* * Demote the mapping. This pmap is locked. The old PDE has * PG_A set. If the old PDE has PG_RW set, it also has PG_M * set. Thus, there is no danger of a race with another * processor changing the setting of PG_A and/or PG_M between * the read above and the store below. */ if (workaround_erratum383) pmap_update_pde(pmap, va, pde, newpde); else pde_store(pde, newpde); /* * Invalidate a stale recursive mapping of the page table page. */ if (va >= VM_MAXUSER_ADDRESS) pmap_invalidate_page(pmap, (vm_offset_t)vtopte(va)); /* * Demote the PV entry. */ if ((oldpde & PG_MANAGED) != 0) pmap_pv_demote_pde(pmap, va, oldpde & PG_PS_FRAME, lockp); atomic_add_long(&pmap_pde_demotions, 1); CTR2(KTR_PMAP, "pmap_demote_pde: success for va %#lx" " in pmap %p", va, pmap); return (TRUE); } /* * pmap_remove_kernel_pde: Remove a kernel superpage mapping. */ static void pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va) { pd_entry_t newpde; vm_paddr_t mptepa; vm_page_t mpte; KASSERT(pmap == kernel_pmap, ("pmap %p is not kernel_pmap", pmap)); PMAP_LOCK_ASSERT(pmap, MA_OWNED); mpte = pmap_lookup_pt_page(pmap, va); if (mpte == NULL) panic("pmap_remove_kernel_pde: Missing pt page."); pmap_remove_pt_page(pmap, mpte); mptepa = VM_PAGE_TO_PHYS(mpte); newpde = mptepa | X86_PG_M | X86_PG_A | X86_PG_RW | X86_PG_V; /* * Initialize the page table page. */ pagezero((void *)PHYS_TO_DMAP(mptepa)); /* * Demote the mapping. */ if (workaround_erratum383) pmap_update_pde(pmap, va, pde, newpde); else pde_store(pde, newpde); /* * Invalidate a stale recursive mapping of the page table page. */ pmap_invalidate_page(pmap, (vm_offset_t)vtopte(va)); } /* * pmap_remove_pde: do the things to unmap a superpage in a process */ static int pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, struct spglist *free, struct rwlock **lockp) { struct md_page *pvh; pd_entry_t oldpde; vm_offset_t eva, va; vm_page_t m, mpte; pt_entry_t PG_G, PG_A, PG_M, PG_RW; PG_G = pmap_global_bit(pmap); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); PMAP_LOCK_ASSERT(pmap, MA_OWNED); KASSERT((sva & PDRMASK) == 0, ("pmap_remove_pde: sva is not 2mpage aligned")); oldpde = pte_load_clear(pdq); if (oldpde & PG_W) pmap->pm_stats.wired_count -= NBPDR / PAGE_SIZE; /* * Machines that don't support invlpg, also don't support * PG_G. */ if (oldpde & PG_G) pmap_invalidate_page(kernel_pmap, sva); pmap_resident_count_dec(pmap, NBPDR / PAGE_SIZE); if (oldpde & PG_MANAGED) { CHANGE_PV_LIST_LOCK_TO_PHYS(lockp, oldpde & PG_PS_FRAME); pvh = pa_to_pvh(oldpde & PG_PS_FRAME); pmap_pvh_free(pvh, pmap, sva); eva = sva + NBPDR; for (va = sva, m = PHYS_TO_VM_PAGE(oldpde & PG_PS_FRAME); va < eva; va += PAGE_SIZE, m++) { if ((oldpde & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(m); if (oldpde & PG_A) vm_page_aflag_set(m, PGA_REFERENCED); if (TAILQ_EMPTY(&m->md.pv_list) && TAILQ_EMPTY(&pvh->pv_list)) vm_page_aflag_clear(m, PGA_WRITEABLE); } } if (pmap == kernel_pmap) { pmap_remove_kernel_pde(pmap, pdq, sva); } else { mpte = pmap_lookup_pt_page(pmap, sva); if (mpte != NULL) { pmap_remove_pt_page(pmap, mpte); pmap_resident_count_dec(pmap, 1); KASSERT(mpte->wire_count == NPTEPG, ("pmap_remove_pde: pte page wire count error")); mpte->wire_count = 0; pmap_add_delayed_free_list(mpte, free, FALSE); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } } return (pmap_unuse_pt(pmap, sva, *pmap_pdpe(pmap, sva), free)); } /* * pmap_remove_pte: do the things to unmap a page in a process */ static int pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t va, pd_entry_t ptepde, struct spglist *free, struct rwlock **lockp) { struct md_page *pvh; pt_entry_t oldpte, PG_A, PG_M, PG_RW; vm_page_t m; PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); PMAP_LOCK_ASSERT(pmap, MA_OWNED); oldpte = pte_load_clear(ptq); if (oldpte & PG_W) pmap->pm_stats.wired_count -= 1; pmap_resident_count_dec(pmap, 1); if (oldpte & PG_MANAGED) { m = PHYS_TO_VM_PAGE(oldpte & PG_FRAME); if ((oldpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(m); if (oldpte & PG_A) vm_page_aflag_set(m, PGA_REFERENCED); CHANGE_PV_LIST_LOCK_TO_VM_PAGE(lockp, m); pmap_pvh_free(&m->md, pmap, va); if (TAILQ_EMPTY(&m->md.pv_list) && (m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); if (TAILQ_EMPTY(&pvh->pv_list)) vm_page_aflag_clear(m, PGA_WRITEABLE); } } return (pmap_unuse_pt(pmap, va, ptepde, free)); } /* * Remove a single page from a process address space */ static void pmap_remove_page(pmap_t pmap, vm_offset_t va, pd_entry_t *pde, struct spglist *free) { struct rwlock *lock; pt_entry_t *pte, PG_V; PG_V = pmap_valid_bit(pmap); PMAP_LOCK_ASSERT(pmap, MA_OWNED); if ((*pde & PG_V) == 0) return; pte = pmap_pde_to_pte(pde, va); if ((*pte & PG_V) == 0) return; lock = NULL; pmap_remove_pte(pmap, pte, va, *pde, free, &lock); if (lock != NULL) rw_wunlock(lock); pmap_invalidate_page(pmap, va); } /* * Remove the given range of addresses from the specified map. * * It is assumed that the start and end are properly * rounded to the page size. */ void pmap_remove(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { struct rwlock *lock; vm_offset_t va, va_next; pml4_entry_t *pml4e; pdp_entry_t *pdpe; pd_entry_t ptpaddr, *pde; pt_entry_t *pte, PG_G, PG_V; struct spglist free; int anyvalid; PG_G = pmap_global_bit(pmap); PG_V = pmap_valid_bit(pmap); /* * Perform an unsynchronized read. This is, however, safe. */ if (pmap->pm_stats.resident_count == 0) return; anyvalid = 0; SLIST_INIT(&free); rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); /* * special handling of removing one page. a very * common operation and easy to short circuit some * code. */ if (sva + PAGE_SIZE == eva) { pde = pmap_pde(pmap, sva); if (pde && (*pde & PG_PS) == 0) { pmap_remove_page(pmap, sva, pde, &free); goto out; } } lock = NULL; for (; sva < eva; sva = va_next) { if (pmap->pm_stats.resident_count == 0) break; pml4e = pmap_pml4e(pmap, sva); if ((*pml4e & PG_V) == 0) { va_next = (sva + NBPML4) & ~PML4MASK; if (va_next < sva) va_next = eva; continue; } pdpe = pmap_pml4e_to_pdpe(pml4e, sva); if ((*pdpe & PG_V) == 0) { va_next = (sva + NBPDP) & ~PDPMASK; if (va_next < sva) va_next = eva; continue; } /* * Calculate index for next page table. */ va_next = (sva + NBPDR) & ~PDRMASK; if (va_next < sva) va_next = eva; pde = pmap_pdpe_to_pde(pdpe, sva); ptpaddr = *pde; /* * Weed out invalid mappings. */ if (ptpaddr == 0) continue; /* * Check for large page. */ if ((ptpaddr & PG_PS) != 0) { /* * Are we removing the entire large page? If not, * demote the mapping and fall through. */ if (sva + NBPDR == va_next && eva >= va_next) { /* * The TLB entry for a PG_G mapping is * invalidated by pmap_remove_pde(). */ if ((ptpaddr & PG_G) == 0) anyvalid = 1; pmap_remove_pde(pmap, pde, sva, &free, &lock); continue; } else if (!pmap_demote_pde_locked(pmap, pde, sva, &lock)) { /* The large page mapping was destroyed. */ continue; } else ptpaddr = *pde; } /* * Limit our scan to either the end of the va represented * by the current page table page, or to the end of the * range being removed. */ if (va_next > eva) va_next = eva; va = va_next; for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++, sva += PAGE_SIZE) { if (*pte == 0) { if (va != va_next) { pmap_invalidate_range(pmap, va, sva); va = va_next; } continue; } if ((*pte & PG_G) == 0) anyvalid = 1; else if (va == va_next) va = sva; if (pmap_remove_pte(pmap, pte, sva, ptpaddr, &free, &lock)) { sva += PAGE_SIZE; break; } } if (va != va_next) pmap_invalidate_range(pmap, va, sva); } if (lock != NULL) rw_wunlock(lock); out: if (anyvalid) pmap_invalidate_all(pmap); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); pmap_free_zero_pages(&free); } /* * Routine: pmap_remove_all * Function: * Removes this physical page from * all physical maps in which it resides. * Reflects back modify bits to the pager. * * Notes: * Original versions of this routine were very * inefficient because they iteratively called * pmap_remove (slow...) */ void pmap_remove_all(vm_page_t m) { struct md_page *pvh; pv_entry_t pv; pmap_t pmap; pt_entry_t *pte, tpte, PG_A, PG_M, PG_RW; pd_entry_t *pde; vm_offset_t va; struct spglist free; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_remove_all: page %p is not managed", m)); SLIST_INIT(&free); rw_wlock(&pvh_global_lock); if ((m->flags & PG_FICTITIOUS) != 0) goto small_mappings; pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); while ((pv = TAILQ_FIRST(&pvh->pv_list)) != NULL) { pmap = PV_PMAP(pv); PMAP_LOCK(pmap); va = pv->pv_va; pde = pmap_pde(pmap, va); (void)pmap_demote_pde(pmap, pde, va); PMAP_UNLOCK(pmap); } small_mappings: while ((pv = TAILQ_FIRST(&m->md.pv_list)) != NULL) { pmap = PV_PMAP(pv); PMAP_LOCK(pmap); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); pmap_resident_count_dec(pmap, 1); pde = pmap_pde(pmap, pv->pv_va); KASSERT((*pde & PG_PS) == 0, ("pmap_remove_all: found" " a 2mpage in page %p's pv list", m)); pte = pmap_pde_to_pte(pde, pv->pv_va); tpte = pte_load_clear(pte); if (tpte & PG_W) pmap->pm_stats.wired_count--; if (tpte & PG_A) vm_page_aflag_set(m, PGA_REFERENCED); /* * Update the vm_page_t clean and reference bits. */ if ((tpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(m); pmap_unuse_pt(pmap, pv->pv_va, *pde, &free); pmap_invalidate_page(pmap, pv->pv_va); TAILQ_REMOVE(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; free_pv_entry(pmap, pv); PMAP_UNLOCK(pmap); } vm_page_aflag_clear(m, PGA_WRITEABLE); rw_wunlock(&pvh_global_lock); pmap_free_zero_pages(&free); } /* * pmap_protect_pde: do the things to protect a 2mpage in a process */ static boolean_t pmap_protect_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t sva, vm_prot_t prot) { pd_entry_t newpde, oldpde; vm_offset_t eva, va; vm_page_t m; boolean_t anychanged; pt_entry_t PG_G, PG_M, PG_RW; PG_G = pmap_global_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); PMAP_LOCK_ASSERT(pmap, MA_OWNED); KASSERT((sva & PDRMASK) == 0, ("pmap_protect_pde: sva is not 2mpage aligned")); anychanged = FALSE; retry: oldpde = newpde = *pde; if (oldpde & PG_MANAGED) { eva = sva + NBPDR; for (va = sva, m = PHYS_TO_VM_PAGE(oldpde & PG_PS_FRAME); va < eva; va += PAGE_SIZE, m++) if ((oldpde & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(m); } if ((prot & VM_PROT_WRITE) == 0) newpde &= ~(PG_RW | PG_M); if ((prot & VM_PROT_EXECUTE) == 0) newpde |= pg_nx; if (newpde != oldpde) { if (!atomic_cmpset_long(pde, oldpde, newpde)) goto retry; if (oldpde & PG_G) pmap_invalidate_page(pmap, sva); else anychanged = TRUE; } return (anychanged); } /* * Set the physical protection on the * specified range of this map as requested. */ void pmap_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, vm_prot_t prot) { vm_offset_t va_next; pml4_entry_t *pml4e; pdp_entry_t *pdpe; pd_entry_t ptpaddr, *pde; pt_entry_t *pte, PG_G, PG_M, PG_RW, PG_V; boolean_t anychanged, pv_lists_locked; KASSERT((prot & ~VM_PROT_ALL) == 0, ("invalid prot %x", prot)); if (prot == VM_PROT_NONE) { pmap_remove(pmap, sva, eva); return; } if ((prot & (VM_PROT_WRITE|VM_PROT_EXECUTE)) == (VM_PROT_WRITE|VM_PROT_EXECUTE)) return; PG_G = pmap_global_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); pv_lists_locked = FALSE; resume: anychanged = FALSE; PMAP_LOCK(pmap); for (; sva < eva; sva = va_next) { pml4e = pmap_pml4e(pmap, sva); if ((*pml4e & PG_V) == 0) { va_next = (sva + NBPML4) & ~PML4MASK; if (va_next < sva) va_next = eva; continue; } pdpe = pmap_pml4e_to_pdpe(pml4e, sva); if ((*pdpe & PG_V) == 0) { va_next = (sva + NBPDP) & ~PDPMASK; if (va_next < sva) va_next = eva; continue; } va_next = (sva + NBPDR) & ~PDRMASK; if (va_next < sva) va_next = eva; pde = pmap_pdpe_to_pde(pdpe, sva); ptpaddr = *pde; /* * Weed out invalid mappings. */ if (ptpaddr == 0) continue; /* * Check for large page. */ if ((ptpaddr & PG_PS) != 0) { /* * Are we protecting the entire large page? If not, * demote the mapping and fall through. */ if (sva + NBPDR == va_next && eva >= va_next) { /* * The TLB entry for a PG_G mapping is * invalidated by pmap_protect_pde(). */ if (pmap_protect_pde(pmap, pde, sva, prot)) anychanged = TRUE; continue; } else { if (!pv_lists_locked) { pv_lists_locked = TRUE; if (!rw_try_rlock(&pvh_global_lock)) { if (anychanged) pmap_invalidate_all( pmap); PMAP_UNLOCK(pmap); rw_rlock(&pvh_global_lock); goto resume; } } if (!pmap_demote_pde(pmap, pde, sva)) { /* * The large page mapping was * destroyed. */ continue; } } } if (va_next > eva) va_next = eva; for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++, sva += PAGE_SIZE) { pt_entry_t obits, pbits; vm_page_t m; retry: obits = pbits = *pte; if ((pbits & PG_V) == 0) continue; if ((prot & VM_PROT_WRITE) == 0) { if ((pbits & (PG_MANAGED | PG_M | PG_RW)) == (PG_MANAGED | PG_M | PG_RW)) { m = PHYS_TO_VM_PAGE(pbits & PG_FRAME); vm_page_dirty(m); } pbits &= ~(PG_RW | PG_M); } if ((prot & VM_PROT_EXECUTE) == 0) pbits |= pg_nx; if (pbits != obits) { if (!atomic_cmpset_long(pte, obits, pbits)) goto retry; if (obits & PG_G) pmap_invalidate_page(pmap, sva); else anychanged = TRUE; } } } if (anychanged) pmap_invalidate_all(pmap); if (pv_lists_locked) rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); } /* * Tries to promote the 512, contiguous 4KB page mappings that are within a * single page table page (PTP) to a single 2MB page mapping. For promotion * to occur, two conditions must be met: (1) the 4KB page mappings must map * aligned, contiguous physical memory and (2) the 4KB page mappings must have * identical characteristics. */ static void pmap_promote_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, struct rwlock **lockp) { pd_entry_t newpde; pt_entry_t *firstpte, oldpte, pa, *pte; pt_entry_t PG_G, PG_A, PG_M, PG_RW, PG_V; vm_page_t mpte; int PG_PTE_CACHE; PG_A = pmap_accessed_bit(pmap); PG_G = pmap_global_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); PG_PTE_CACHE = pmap_cache_mask(pmap, 0); PMAP_LOCK_ASSERT(pmap, MA_OWNED); /* * Examine the first PTE in the specified PTP. Abort if this PTE is * either invalid, unused, or does not map the first 4KB physical page * within a 2MB page. */ firstpte = (pt_entry_t *)PHYS_TO_DMAP(*pde & PG_FRAME); setpde: newpde = *firstpte; if ((newpde & ((PG_FRAME & PDRMASK) | PG_A | PG_V)) != (PG_A | PG_V)) { atomic_add_long(&pmap_pde_p_failures, 1); CTR2(KTR_PMAP, "pmap_promote_pde: failure for va %#lx" " in pmap %p", va, pmap); return; } if ((newpde & (PG_M | PG_RW)) == PG_RW) { /* * When PG_M is already clear, PG_RW can be cleared without * a TLB invalidation. */ if (!atomic_cmpset_long(firstpte, newpde, newpde & ~PG_RW)) goto setpde; newpde &= ~PG_RW; } /* * Examine each of the other PTEs in the specified PTP. Abort if this * PTE maps an unexpected 4KB physical page or does not have identical * characteristics to the first PTE. */ pa = (newpde & (PG_PS_FRAME | PG_A | PG_V)) + NBPDR - PAGE_SIZE; for (pte = firstpte + NPTEPG - 1; pte > firstpte; pte--) { setpte: oldpte = *pte; if ((oldpte & (PG_FRAME | PG_A | PG_V)) != pa) { atomic_add_long(&pmap_pde_p_failures, 1); CTR2(KTR_PMAP, "pmap_promote_pde: failure for va %#lx" " in pmap %p", va, pmap); return; } if ((oldpte & (PG_M | PG_RW)) == PG_RW) { /* * When PG_M is already clear, PG_RW can be cleared * without a TLB invalidation. */ if (!atomic_cmpset_long(pte, oldpte, oldpte & ~PG_RW)) goto setpte; oldpte &= ~PG_RW; CTR2(KTR_PMAP, "pmap_promote_pde: protect for va %#lx" " in pmap %p", (oldpte & PG_FRAME & PDRMASK) | (va & ~PDRMASK), pmap); } if ((oldpte & PG_PTE_PROMOTE) != (newpde & PG_PTE_PROMOTE)) { atomic_add_long(&pmap_pde_p_failures, 1); CTR2(KTR_PMAP, "pmap_promote_pde: failure for va %#lx" " in pmap %p", va, pmap); return; } pa -= PAGE_SIZE; } /* * Save the page table page in its current state until the PDE * mapping the superpage is demoted by pmap_demote_pde() or * destroyed by pmap_remove_pde(). */ mpte = PHYS_TO_VM_PAGE(*pde & PG_FRAME); KASSERT(mpte >= vm_page_array && mpte < &vm_page_array[vm_page_array_size], ("pmap_promote_pde: page table page is out of range")); KASSERT(mpte->pindex == pmap_pde_pindex(va), ("pmap_promote_pde: page table page's pindex is wrong")); if (pmap_insert_pt_page(pmap, mpte)) { atomic_add_long(&pmap_pde_p_failures, 1); CTR2(KTR_PMAP, "pmap_promote_pde: failure for va %#lx in pmap %p", va, pmap); return; } /* * Promote the pv entries. */ if ((newpde & PG_MANAGED) != 0) pmap_pv_promote_pde(pmap, va, newpde & PG_PS_FRAME, lockp); /* * Propagate the PAT index to its proper position. */ newpde = pmap_swap_pat(pmap, newpde); /* * Map the superpage. */ if (workaround_erratum383) pmap_update_pde(pmap, va, pde, PG_PS | newpde); else pde_store(pde, PG_PS | newpde); atomic_add_long(&pmap_pde_promotions, 1); CTR2(KTR_PMAP, "pmap_promote_pde: success for va %#lx" " in pmap %p", va, pmap); } /* * Insert the given physical page (p) at * the specified virtual address (v) in the * target physical map with the protection requested. * * If specified, the page will be wired down, meaning * that the related pte can not be reclaimed. * * NB: This is the only routine which MAY NOT lazy-evaluate * or lose information. That is, this routine must actually * insert this page into the given map NOW. */ int pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, u_int flags, int8_t psind __unused) { struct rwlock *lock; pd_entry_t *pde; pt_entry_t *pte, PG_G, PG_A, PG_M, PG_RW, PG_V; pt_entry_t newpte, origpte; pv_entry_t pv; vm_paddr_t opa, pa; vm_page_t mpte, om; boolean_t nosleep; PG_A = pmap_accessed_bit(pmap); PG_G = pmap_global_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); va = trunc_page(va); KASSERT(va <= VM_MAX_KERNEL_ADDRESS, ("pmap_enter: toobig")); KASSERT(va < UPT_MIN_ADDRESS || va >= UPT_MAX_ADDRESS, ("pmap_enter: invalid to pmap_enter page table pages (va: 0x%lx)", va)); KASSERT((m->oflags & VPO_UNMANAGED) != 0 || va < kmi.clean_sva || va >= kmi.clean_eva, ("pmap_enter: managed mapping within the clean submap")); if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m)) VM_OBJECT_ASSERT_LOCKED(m->object); pa = VM_PAGE_TO_PHYS(m); newpte = (pt_entry_t)(pa | PG_A | PG_V); if ((flags & VM_PROT_WRITE) != 0) newpte |= PG_M; if ((prot & VM_PROT_WRITE) != 0) newpte |= PG_RW; KASSERT((newpte & (PG_M | PG_RW)) != PG_M, ("pmap_enter: flags includes VM_PROT_WRITE but prot doesn't")); if ((prot & VM_PROT_EXECUTE) == 0) newpte |= pg_nx; if ((flags & PMAP_ENTER_WIRED) != 0) newpte |= PG_W; if (va < VM_MAXUSER_ADDRESS) newpte |= PG_U; if (pmap == kernel_pmap) newpte |= PG_G; newpte |= pmap_cache_bits(pmap, m->md.pat_mode, 0); /* * Set modified bit gratuitously for writeable mappings if * the page is unmanaged. We do not want to take a fault * to do the dirty bit accounting for these mappings. */ if ((m->oflags & VPO_UNMANAGED) != 0) { if ((newpte & PG_RW) != 0) newpte |= PG_M; } mpte = NULL; lock = NULL; rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); /* * In the case that a page table page is not * resident, we are creating it here. */ retry: pde = pmap_pde(pmap, va); if (pde != NULL && (*pde & PG_V) != 0 && ((*pde & PG_PS) == 0 || pmap_demote_pde_locked(pmap, pde, va, &lock))) { pte = pmap_pde_to_pte(pde, va); if (va < VM_MAXUSER_ADDRESS && mpte == NULL) { mpte = PHYS_TO_VM_PAGE(*pde & PG_FRAME); mpte->wire_count++; } } else if (va < VM_MAXUSER_ADDRESS) { /* * Here if the pte page isn't mapped, or if it has been * deallocated. */ nosleep = (flags & PMAP_ENTER_NOSLEEP) != 0; mpte = _pmap_allocpte(pmap, pmap_pde_pindex(va), nosleep ? NULL : &lock); if (mpte == NULL && nosleep) { if (lock != NULL) rw_wunlock(lock); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); return (KERN_RESOURCE_SHORTAGE); } goto retry; } else panic("pmap_enter: invalid page directory va=%#lx", va); origpte = *pte; /* * Is the specified virtual address already mapped? */ if ((origpte & PG_V) != 0) { /* * Wiring change, just update stats. We don't worry about * wiring PT pages as they remain resident as long as there * are valid mappings in them. Hence, if a user page is wired, * the PT page will be also. */ if ((newpte & PG_W) != 0 && (origpte & PG_W) == 0) pmap->pm_stats.wired_count++; else if ((newpte & PG_W) == 0 && (origpte & PG_W) != 0) pmap->pm_stats.wired_count--; /* * Remove the extra PT page reference. */ if (mpte != NULL) { mpte->wire_count--; KASSERT(mpte->wire_count > 0, ("pmap_enter: missing reference to page table page," " va: 0x%lx", va)); } /* * Has the physical page changed? */ opa = origpte & PG_FRAME; if (opa == pa) { /* * No, might be a protection or wiring change. */ if ((origpte & PG_MANAGED) != 0) { newpte |= PG_MANAGED; if ((newpte & PG_RW) != 0) vm_page_aflag_set(m, PGA_WRITEABLE); } if (((origpte ^ newpte) & ~(PG_M | PG_A)) == 0) goto unchanged; goto validate; } } else { /* * Increment the counters. */ if ((newpte & PG_W) != 0) pmap->pm_stats.wired_count++; pmap_resident_count_inc(pmap, 1); } /* * Enter on the PV list if part of our managed memory. */ if ((m->oflags & VPO_UNMANAGED) == 0) { newpte |= PG_MANAGED; pv = get_pv_entry(pmap, &lock); pv->pv_va = va; CHANGE_PV_LIST_LOCK_TO_PHYS(&lock, pa); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; if ((newpte & PG_RW) != 0) vm_page_aflag_set(m, PGA_WRITEABLE); } /* * Update the PTE. */ if ((origpte & PG_V) != 0) { validate: origpte = pte_load_store(pte, newpte); opa = origpte & PG_FRAME; if (opa != pa) { if ((origpte & PG_MANAGED) != 0) { om = PHYS_TO_VM_PAGE(opa); if ((origpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) vm_page_dirty(om); if ((origpte & PG_A) != 0) vm_page_aflag_set(om, PGA_REFERENCED); CHANGE_PV_LIST_LOCK_TO_PHYS(&lock, opa); pmap_pvh_free(&om->md, pmap, va); if ((om->aflags & PGA_WRITEABLE) != 0 && TAILQ_EMPTY(&om->md.pv_list) && ((om->flags & PG_FICTITIOUS) != 0 || TAILQ_EMPTY(&pa_to_pvh(opa)->pv_list))) vm_page_aflag_clear(om, PGA_WRITEABLE); } } else if ((newpte & PG_M) == 0 && (origpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) { if ((origpte & PG_MANAGED) != 0) vm_page_dirty(m); /* * Although the PTE may still have PG_RW set, TLB * invalidation may nonetheless be required because * the PTE no longer has PG_M set. */ } else if ((origpte & PG_NX) != 0 || (newpte & PG_NX) == 0) { /* * This PTE change does not require TLB invalidation. */ goto unchanged; } if ((origpte & PG_A) != 0) pmap_invalidate_page(pmap, va); } else pte_store(pte, newpte); unchanged: /* * If both the page table page and the reservation are fully * populated, then attempt promotion. */ if ((mpte == NULL || mpte->wire_count == NPTEPG) && pmap_ps_enabled(pmap) && (m->flags & PG_FICTITIOUS) == 0 && vm_reserv_level_iffullpop(m) == 0) pmap_promote_pde(pmap, pde, va, &lock); if (lock != NULL) rw_wunlock(lock); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); return (KERN_SUCCESS); } /* * Tries to create a 2MB page mapping. Returns TRUE if successful and FALSE * otherwise. Fails if (1) a page table page cannot be allocated without * blocking, (2) a mapping already exists at the specified virtual address, or * (3) a pv entry cannot be allocated without reclaiming another pv entry. */ static boolean_t pmap_enter_pde(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, struct rwlock **lockp) { pd_entry_t *pde, newpde; pt_entry_t PG_V; vm_page_t mpde; struct spglist free; PG_V = pmap_valid_bit(pmap); rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); if ((mpde = pmap_allocpde(pmap, va, NULL)) == NULL) { CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx" " in pmap %p", va, pmap); return (FALSE); } pde = (pd_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(mpde)); pde = &pde[pmap_pde_index(va)]; if ((*pde & PG_V) != 0) { KASSERT(mpde->wire_count > 1, ("pmap_enter_pde: mpde's wire count is too low")); mpde->wire_count--; CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx" " in pmap %p", va, pmap); return (FALSE); } newpde = VM_PAGE_TO_PHYS(m) | pmap_cache_bits(pmap, m->md.pat_mode, 1) | PG_PS | PG_V; if ((m->oflags & VPO_UNMANAGED) == 0) { newpde |= PG_MANAGED; /* * Abort this mapping if its PV entry could not be created. */ if (!pmap_pv_insert_pde(pmap, va, VM_PAGE_TO_PHYS(m), lockp)) { SLIST_INIT(&free); if (pmap_unwire_ptp(pmap, va, mpde, &free)) { pmap_invalidate_page(pmap, va); pmap_free_zero_pages(&free); } CTR2(KTR_PMAP, "pmap_enter_pde: failure for va %#lx" " in pmap %p", va, pmap); return (FALSE); } } if ((prot & VM_PROT_EXECUTE) == 0) newpde |= pg_nx; if (va < VM_MAXUSER_ADDRESS) newpde |= PG_U; /* * Increment counters. */ pmap_resident_count_inc(pmap, NBPDR / PAGE_SIZE); /* * Map the superpage. */ pde_store(pde, newpde); atomic_add_long(&pmap_pde_mappings, 1); CTR2(KTR_PMAP, "pmap_enter_pde: success for va %#lx" " in pmap %p", va, pmap); return (TRUE); } /* * Maps a sequence of resident pages belonging to the same object. * The sequence begins with the given page m_start. This page is * mapped at the given virtual address start. Each subsequent page is * mapped at a virtual address that is offset from start by the same * amount as the page is offset from m_start within the object. The * last page in the sequence is the page with the largest offset from * m_start that can be mapped at a virtual address less than the given * virtual address end. Not every virtual page between start and end * is mapped; only those for which a resident page exists with the * corresponding offset from m_start are mapped. */ void pmap_enter_object(pmap_t pmap, vm_offset_t start, vm_offset_t end, vm_page_t m_start, vm_prot_t prot) { struct rwlock *lock; vm_offset_t va; vm_page_t m, mpte; vm_pindex_t diff, psize; VM_OBJECT_ASSERT_LOCKED(m_start->object); psize = atop(end - start); mpte = NULL; m = m_start; lock = NULL; rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) { va = start + ptoa(diff); if ((va & PDRMASK) == 0 && va + NBPDR <= end && m->psind == 1 && pmap_ps_enabled(pmap) && pmap_enter_pde(pmap, va, m, prot, &lock)) m = &m[NBPDR / PAGE_SIZE - 1]; else mpte = pmap_enter_quick_locked(pmap, va, m, prot, mpte, &lock); m = TAILQ_NEXT(m, listq); } if (lock != NULL) rw_wunlock(lock); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); } /* * this code makes some *MAJOR* assumptions: * 1. Current pmap & pmap exists. * 2. Not wired. * 3. Read access. * 4. No page table pages. * but is *MUCH* faster than pmap_enter... */ void pmap_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot) { struct rwlock *lock; lock = NULL; rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); (void)pmap_enter_quick_locked(pmap, va, m, prot, NULL, &lock); if (lock != NULL) rw_wunlock(lock); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); } static vm_page_t pmap_enter_quick_locked(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, vm_page_t mpte, struct rwlock **lockp) { struct spglist free; pt_entry_t *pte, PG_V; vm_paddr_t pa; KASSERT(va < kmi.clean_sva || va >= kmi.clean_eva || (m->oflags & VPO_UNMANAGED) != 0, ("pmap_enter_quick_locked: managed mapping within the clean submap")); PG_V = pmap_valid_bit(pmap); rw_assert(&pvh_global_lock, RA_LOCKED); PMAP_LOCK_ASSERT(pmap, MA_OWNED); /* * In the case that a page table page is not * resident, we are creating it here. */ if (va < VM_MAXUSER_ADDRESS) { vm_pindex_t ptepindex; pd_entry_t *ptepa; /* * Calculate pagetable page index */ ptepindex = pmap_pde_pindex(va); if (mpte && (mpte->pindex == ptepindex)) { mpte->wire_count++; } else { /* * Get the page directory entry */ ptepa = pmap_pde(pmap, va); /* * If the page table page is mapped, we just increment * the hold count, and activate it. Otherwise, we * attempt to allocate a page table page. If this * attempt fails, we don't retry. Instead, we give up. */ if (ptepa && (*ptepa & PG_V) != 0) { if (*ptepa & PG_PS) return (NULL); mpte = PHYS_TO_VM_PAGE(*ptepa & PG_FRAME); mpte->wire_count++; } else { /* * Pass NULL instead of the PV list lock * pointer, because we don't intend to sleep. */ mpte = _pmap_allocpte(pmap, ptepindex, NULL); if (mpte == NULL) return (mpte); } } pte = (pt_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(mpte)); pte = &pte[pmap_pte_index(va)]; } else { mpte = NULL; pte = vtopte(va); } if (*pte) { if (mpte != NULL) { mpte->wire_count--; mpte = NULL; } return (mpte); } /* * Enter on the PV list if part of our managed memory. */ if ((m->oflags & VPO_UNMANAGED) == 0 && !pmap_try_insert_pv_entry(pmap, va, m, lockp)) { if (mpte != NULL) { SLIST_INIT(&free); if (pmap_unwire_ptp(pmap, va, mpte, &free)) { pmap_invalidate_page(pmap, va); pmap_free_zero_pages(&free); } mpte = NULL; } return (mpte); } /* * Increment counters */ pmap_resident_count_inc(pmap, 1); pa = VM_PAGE_TO_PHYS(m) | pmap_cache_bits(pmap, m->md.pat_mode, 0); if ((prot & VM_PROT_EXECUTE) == 0) pa |= pg_nx; /* * Now validate mapping with RO protection */ if ((m->oflags & VPO_UNMANAGED) != 0) pte_store(pte, pa | PG_V | PG_U); else pte_store(pte, pa | PG_V | PG_U | PG_MANAGED); return (mpte); } /* * Make a temporary mapping for a physical address. This is only intended * to be used for panic dumps. */ void * pmap_kenter_temporary(vm_paddr_t pa, int i) { vm_offset_t va; va = (vm_offset_t)crashdumpmap + (i * PAGE_SIZE); pmap_kenter(va, pa); invlpg(va); return ((void *)crashdumpmap); } /* * This code maps large physical mmap regions into the * processor address space. Note that some shortcuts * are taken, but the code works. */ void pmap_object_init_pt(pmap_t pmap, vm_offset_t addr, vm_object_t object, vm_pindex_t pindex, vm_size_t size) { pd_entry_t *pde; pt_entry_t PG_A, PG_M, PG_RW, PG_V; vm_paddr_t pa, ptepa; vm_page_t p, pdpg; int pat_mode; PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG, ("pmap_object_init_pt: non-device object")); if ((addr & (NBPDR - 1)) == 0 && (size & (NBPDR - 1)) == 0) { if (!pmap_ps_enabled(pmap)) return; if (!vm_object_populate(object, pindex, pindex + atop(size))) return; p = vm_page_lookup(object, pindex); KASSERT(p->valid == VM_PAGE_BITS_ALL, ("pmap_object_init_pt: invalid page %p", p)); pat_mode = p->md.pat_mode; /* * Abort the mapping if the first page is not physically * aligned to a 2MB page boundary. */ ptepa = VM_PAGE_TO_PHYS(p); if (ptepa & (NBPDR - 1)) return; /* * Skip the first page. Abort the mapping if the rest of * the pages are not physically contiguous or have differing * memory attributes. */ p = TAILQ_NEXT(p, listq); for (pa = ptepa + PAGE_SIZE; pa < ptepa + size; pa += PAGE_SIZE) { KASSERT(p->valid == VM_PAGE_BITS_ALL, ("pmap_object_init_pt: invalid page %p", p)); if (pa != VM_PAGE_TO_PHYS(p) || pat_mode != p->md.pat_mode) return; p = TAILQ_NEXT(p, listq); } /* * Map using 2MB pages. Since "ptepa" is 2M aligned and * "size" is a multiple of 2M, adding the PAT setting to "pa" * will not affect the termination of this loop. */ PMAP_LOCK(pmap); for (pa = ptepa | pmap_cache_bits(pmap, pat_mode, 1); pa < ptepa + size; pa += NBPDR) { pdpg = pmap_allocpde(pmap, addr, NULL); if (pdpg == NULL) { /* * The creation of mappings below is only an * optimization. If a page directory page * cannot be allocated without blocking, * continue on to the next mapping rather than * blocking. */ addr += NBPDR; continue; } pde = (pd_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pdpg)); pde = &pde[pmap_pde_index(addr)]; if ((*pde & PG_V) == 0) { pde_store(pde, pa | PG_PS | PG_M | PG_A | PG_U | PG_RW | PG_V); pmap_resident_count_inc(pmap, NBPDR / PAGE_SIZE); atomic_add_long(&pmap_pde_mappings, 1); } else { /* Continue on if the PDE is already valid. */ pdpg->wire_count--; KASSERT(pdpg->wire_count > 0, ("pmap_object_init_pt: missing reference " "to page directory page, va: 0x%lx", addr)); } addr += NBPDR; } PMAP_UNLOCK(pmap); } } /* * Clear the wired attribute from the mappings for the specified range of * addresses in the given pmap. Every valid mapping within that range * must have the wired attribute set. In contrast, invalid mappings * cannot have the wired attribute set, so they are ignored. * * The wired attribute of the page table entry is not a hardware feature, * so there is no need to invalidate any TLB entries. */ void pmap_unwire(pmap_t pmap, vm_offset_t sva, vm_offset_t eva) { vm_offset_t va_next; pml4_entry_t *pml4e; pdp_entry_t *pdpe; pd_entry_t *pde; pt_entry_t *pte, PG_V; boolean_t pv_lists_locked; PG_V = pmap_valid_bit(pmap); pv_lists_locked = FALSE; resume: PMAP_LOCK(pmap); for (; sva < eva; sva = va_next) { pml4e = pmap_pml4e(pmap, sva); if ((*pml4e & PG_V) == 0) { va_next = (sva + NBPML4) & ~PML4MASK; if (va_next < sva) va_next = eva; continue; } pdpe = pmap_pml4e_to_pdpe(pml4e, sva); if ((*pdpe & PG_V) == 0) { va_next = (sva + NBPDP) & ~PDPMASK; if (va_next < sva) va_next = eva; continue; } va_next = (sva + NBPDR) & ~PDRMASK; if (va_next < sva) va_next = eva; pde = pmap_pdpe_to_pde(pdpe, sva); if ((*pde & PG_V) == 0) continue; if ((*pde & PG_PS) != 0) { if ((*pde & PG_W) == 0) panic("pmap_unwire: pde %#jx is missing PG_W", (uintmax_t)*pde); /* * Are we unwiring the entire large page? If not, * demote the mapping and fall through. */ if (sva + NBPDR == va_next && eva >= va_next) { atomic_clear_long(pde, PG_W); pmap->pm_stats.wired_count -= NBPDR / PAGE_SIZE; continue; } else { if (!pv_lists_locked) { pv_lists_locked = TRUE; if (!rw_try_rlock(&pvh_global_lock)) { PMAP_UNLOCK(pmap); rw_rlock(&pvh_global_lock); /* Repeat sva. */ goto resume; } } if (!pmap_demote_pde(pmap, pde, sva)) panic("pmap_unwire: demotion failed"); } } if (va_next > eva) va_next = eva; for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++, sva += PAGE_SIZE) { if ((*pte & PG_V) == 0) continue; if ((*pte & PG_W) == 0) panic("pmap_unwire: pte %#jx is missing PG_W", (uintmax_t)*pte); /* * PG_W must be cleared atomically. Although the pmap * lock synchronizes access to PG_W, another processor * could be setting PG_M and/or PG_A concurrently. */ atomic_clear_long(pte, PG_W); pmap->pm_stats.wired_count--; } } if (pv_lists_locked) rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); } /* * Copy the range specified by src_addr/len * from the source map to the range dst_addr/len * in the destination map. * * This routine is only advisory and need not do anything. */ void pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr) { struct rwlock *lock; struct spglist free; vm_offset_t addr; vm_offset_t end_addr = src_addr + len; vm_offset_t va_next; pt_entry_t PG_A, PG_M, PG_V; if (dst_addr != src_addr) return; if (dst_pmap->pm_type != src_pmap->pm_type) return; /* * EPT page table entries that require emulation of A/D bits are * sensitive to clearing the PG_A bit (aka EPT_PG_READ). Although * we clear PG_M (aka EPT_PG_WRITE) concomitantly, the PG_U bit * (aka EPT_PG_EXECUTE) could still be set. Since some EPT * implementations flag an EPT misconfiguration for exec-only * mappings we skip this function entirely for emulated pmaps. */ if (pmap_emulate_ad_bits(dst_pmap)) return; lock = NULL; rw_rlock(&pvh_global_lock); if (dst_pmap < src_pmap) { PMAP_LOCK(dst_pmap); PMAP_LOCK(src_pmap); } else { PMAP_LOCK(src_pmap); PMAP_LOCK(dst_pmap); } PG_A = pmap_accessed_bit(dst_pmap); PG_M = pmap_modified_bit(dst_pmap); PG_V = pmap_valid_bit(dst_pmap); for (addr = src_addr; addr < end_addr; addr = va_next) { pt_entry_t *src_pte, *dst_pte; vm_page_t dstmpde, dstmpte, srcmpte; pml4_entry_t *pml4e; pdp_entry_t *pdpe; pd_entry_t srcptepaddr, *pde; KASSERT(addr < UPT_MIN_ADDRESS, ("pmap_copy: invalid to pmap_copy page tables")); pml4e = pmap_pml4e(src_pmap, addr); if ((*pml4e & PG_V) == 0) { va_next = (addr + NBPML4) & ~PML4MASK; if (va_next < addr) va_next = end_addr; continue; } pdpe = pmap_pml4e_to_pdpe(pml4e, addr); if ((*pdpe & PG_V) == 0) { va_next = (addr + NBPDP) & ~PDPMASK; if (va_next < addr) va_next = end_addr; continue; } va_next = (addr + NBPDR) & ~PDRMASK; if (va_next < addr) va_next = end_addr; pde = pmap_pdpe_to_pde(pdpe, addr); srcptepaddr = *pde; if (srcptepaddr == 0) continue; if (srcptepaddr & PG_PS) { if ((addr & PDRMASK) != 0 || addr + NBPDR > end_addr) continue; dstmpde = pmap_allocpde(dst_pmap, addr, NULL); if (dstmpde == NULL) break; pde = (pd_entry_t *) PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dstmpde)); pde = &pde[pmap_pde_index(addr)]; if (*pde == 0 && ((srcptepaddr & PG_MANAGED) == 0 || pmap_pv_insert_pde(dst_pmap, addr, srcptepaddr & PG_PS_FRAME, &lock))) { *pde = srcptepaddr & ~PG_W; pmap_resident_count_inc(dst_pmap, NBPDR / PAGE_SIZE); atomic_add_long(&pmap_pde_mappings, 1); } else dstmpde->wire_count--; continue; } srcptepaddr &= PG_FRAME; srcmpte = PHYS_TO_VM_PAGE(srcptepaddr); KASSERT(srcmpte->wire_count > 0, ("pmap_copy: source page table page is unused")); if (va_next > end_addr) va_next = end_addr; src_pte = (pt_entry_t *)PHYS_TO_DMAP(srcptepaddr); src_pte = &src_pte[pmap_pte_index(addr)]; dstmpte = NULL; while (addr < va_next) { pt_entry_t ptetemp; ptetemp = *src_pte; /* * we only virtual copy managed pages */ if ((ptetemp & PG_MANAGED) != 0) { if (dstmpte != NULL && dstmpte->pindex == pmap_pde_pindex(addr)) dstmpte->wire_count++; else if ((dstmpte = pmap_allocpte(dst_pmap, addr, NULL)) == NULL) goto out; dst_pte = (pt_entry_t *) PHYS_TO_DMAP(VM_PAGE_TO_PHYS(dstmpte)); dst_pte = &dst_pte[pmap_pte_index(addr)]; if (*dst_pte == 0 && pmap_try_insert_pv_entry(dst_pmap, addr, PHYS_TO_VM_PAGE(ptetemp & PG_FRAME), &lock)) { /* * Clear the wired, modified, and * accessed (referenced) bits * during the copy. */ *dst_pte = ptetemp & ~(PG_W | PG_M | PG_A); pmap_resident_count_inc(dst_pmap, 1); } else { SLIST_INIT(&free); if (pmap_unwire_ptp(dst_pmap, addr, dstmpte, &free)) { pmap_invalidate_page(dst_pmap, addr); pmap_free_zero_pages(&free); } goto out; } if (dstmpte->wire_count >= srcmpte->wire_count) break; } addr += PAGE_SIZE; src_pte++; } } out: if (lock != NULL) rw_wunlock(lock); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(src_pmap); PMAP_UNLOCK(dst_pmap); } /* * pmap_zero_page zeros the specified hardware page by mapping * the page into KVM and using bzero to clear its contents. */ void pmap_zero_page(vm_page_t m) { vm_offset_t va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); pagezero((void *)va); } /* * pmap_zero_page_area zeros the specified hardware page by mapping * the page into KVM and using bzero to clear its contents. * * off and size may not cover an area beyond a single hardware page. */ void pmap_zero_page_area(vm_page_t m, int off, int size) { vm_offset_t va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); if (off == 0 && size == PAGE_SIZE) pagezero((void *)va); else bzero((char *)va + off, size); } /* * pmap_zero_page_idle zeros the specified hardware page by mapping * the page into KVM and using bzero to clear its contents. This * is intended to be called from the vm_pagezero process only and * outside of Giant. */ void pmap_zero_page_idle(vm_page_t m) { vm_offset_t va = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); pagezero((void *)va); } /* * pmap_copy_page copies the specified (machine independent) * page by mapping the page into virtual memory and using * bcopy to copy the page, one machine dependent page at a * time. */ void pmap_copy_page(vm_page_t msrc, vm_page_t mdst) { vm_offset_t src = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(msrc)); vm_offset_t dst = PHYS_TO_DMAP(VM_PAGE_TO_PHYS(mdst)); pagecopy((void *)src, (void *)dst); } int unmapped_buf_allowed = 1; void pmap_copy_pages(vm_page_t ma[], vm_offset_t a_offset, vm_page_t mb[], vm_offset_t b_offset, int xfersize) { void *a_cp, *b_cp; vm_page_t pages[2]; vm_offset_t vaddr[2], a_pg_offset, b_pg_offset; int cnt; boolean_t mapped; while (xfersize > 0) { a_pg_offset = a_offset & PAGE_MASK; pages[0] = ma[a_offset >> PAGE_SHIFT]; b_pg_offset = b_offset & PAGE_MASK; pages[1] = mb[b_offset >> PAGE_SHIFT]; cnt = min(xfersize, PAGE_SIZE - a_pg_offset); cnt = min(cnt, PAGE_SIZE - b_pg_offset); mapped = pmap_map_io_transient(pages, vaddr, 2, FALSE); a_cp = (char *)vaddr[0] + a_pg_offset; b_cp = (char *)vaddr[1] + b_pg_offset; bcopy(a_cp, b_cp, cnt); if (__predict_false(mapped)) pmap_unmap_io_transient(pages, vaddr, 2, FALSE); a_offset += cnt; b_offset += cnt; xfersize -= cnt; } } /* * Returns true if the pmap's pv is one of the first * 16 pvs linked to from this page. This count may * be changed upwards or downwards in the future; it * is only necessary that true be returned for a small * subset of pmaps for proper page aging. */ boolean_t pmap_page_exists_quick(pmap_t pmap, vm_page_t m) { struct md_page *pvh; struct rwlock *lock; pv_entry_t pv; int loops = 0; boolean_t rv; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_page_exists_quick: page %p is not managed", m)); rv = FALSE; rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); rw_rlock(lock); TAILQ_FOREACH(pv, &m->md.pv_list, pv_next) { if (PV_PMAP(pv) == pmap) { rv = TRUE; break; } loops++; if (loops >= 16) break; } if (!rv && loops < 16 && (m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); TAILQ_FOREACH(pv, &pvh->pv_list, pv_next) { if (PV_PMAP(pv) == pmap) { rv = TRUE; break; } loops++; if (loops >= 16) break; } } rw_runlock(lock); rw_runlock(&pvh_global_lock); return (rv); } /* * pmap_page_wired_mappings: * * Return the number of managed mappings to the given physical page * that are wired. */ int pmap_page_wired_mappings(vm_page_t m) { struct rwlock *lock; struct md_page *pvh; pmap_t pmap; pt_entry_t *pte; pv_entry_t pv; int count, md_gen, pvh_gen; if ((m->oflags & VPO_UNMANAGED) != 0) return (0); rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); rw_rlock(lock); restart: count = 0; TAILQ_FOREACH(pv, &m->md.pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { md_gen = m->md.pv_gen; rw_runlock(lock); PMAP_LOCK(pmap); rw_rlock(lock); if (md_gen != m->md.pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } pte = pmap_pte(pmap, pv->pv_va); if ((*pte & PG_W) != 0) count++; PMAP_UNLOCK(pmap); } if ((m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); TAILQ_FOREACH(pv, &pvh->pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { md_gen = m->md.pv_gen; pvh_gen = pvh->pv_gen; rw_runlock(lock); PMAP_LOCK(pmap); rw_rlock(lock); if (md_gen != m->md.pv_gen || pvh_gen != pvh->pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } pte = pmap_pde(pmap, pv->pv_va); if ((*pte & PG_W) != 0) count++; PMAP_UNLOCK(pmap); } } rw_runlock(lock); rw_runlock(&pvh_global_lock); return (count); } /* * Returns TRUE if the given page is mapped individually or as part of * a 2mpage. Otherwise, returns FALSE. */ boolean_t pmap_page_is_mapped(vm_page_t m) { struct rwlock *lock; boolean_t rv; if ((m->oflags & VPO_UNMANAGED) != 0) return (FALSE); rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); rw_rlock(lock); rv = !TAILQ_EMPTY(&m->md.pv_list) || ((m->flags & PG_FICTITIOUS) == 0 && !TAILQ_EMPTY(&pa_to_pvh(VM_PAGE_TO_PHYS(m))->pv_list)); rw_runlock(lock); rw_runlock(&pvh_global_lock); return (rv); } /* * Destroy all managed, non-wired mappings in the given user-space * pmap. This pmap cannot be active on any processor besides the * caller. * * This function cannot be applied to the kernel pmap. Moreover, it * is not intended for general use. It is only to be used during * process termination. Consequently, it can be implemented in ways * that make it faster than pmap_remove(). First, it can more quickly * destroy mappings by iterating over the pmap's collection of PV * entries, rather than searching the page table. Second, it doesn't * have to test and clear the page table entries atomically, because * no processor is currently accessing the user address space. In * particular, a page table entry's dirty bit won't change state once * this function starts. */ void pmap_remove_pages(pmap_t pmap) { pd_entry_t ptepde; pt_entry_t *pte, tpte; pt_entry_t PG_M, PG_RW, PG_V; struct spglist free; vm_page_t m, mpte, mt; pv_entry_t pv; struct md_page *pvh; struct pv_chunk *pc, *npc; struct rwlock *lock; int64_t bit; uint64_t inuse, bitmask; int allfree, field, freed, idx; boolean_t superpage; vm_paddr_t pa; /* * Assert that the given pmap is only active on the current * CPU. Unfortunately, we cannot block another CPU from * activating the pmap while this function is executing. */ KASSERT(pmap == PCPU_GET(curpmap), ("non-current pmap %p", pmap)); #ifdef INVARIANTS { cpuset_t other_cpus; other_cpus = all_cpus; critical_enter(); CPU_CLR(PCPU_GET(cpuid), &other_cpus); CPU_AND(&other_cpus, &pmap->pm_active); critical_exit(); KASSERT(CPU_EMPTY(&other_cpus), ("pmap active %p", pmap)); } #endif lock = NULL; PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); SLIST_INIT(&free); rw_rlock(&pvh_global_lock); PMAP_LOCK(pmap); TAILQ_FOREACH_SAFE(pc, &pmap->pm_pvchunk, pc_list, npc) { allfree = 1; freed = 0; for (field = 0; field < _NPCM; field++) { inuse = ~pc->pc_map[field] & pc_freemask[field]; while (inuse != 0) { bit = bsfq(inuse); bitmask = 1UL << bit; idx = field * 64 + bit; pv = &pc->pc_pventry[idx]; inuse &= ~bitmask; pte = pmap_pdpe(pmap, pv->pv_va); ptepde = *pte; pte = pmap_pdpe_to_pde(pte, pv->pv_va); tpte = *pte; if ((tpte & (PG_PS | PG_V)) == PG_V) { superpage = FALSE; ptepde = tpte; pte = (pt_entry_t *)PHYS_TO_DMAP(tpte & PG_FRAME); pte = &pte[pmap_pte_index(pv->pv_va)]; tpte = *pte; } else { /* * Keep track whether 'tpte' is a * superpage explicitly instead of * relying on PG_PS being set. * * This is because PG_PS is numerically * identical to PG_PTE_PAT and thus a * regular page could be mistaken for * a superpage. */ superpage = TRUE; } if ((tpte & PG_V) == 0) { panic("bad pte va %lx pte %lx", pv->pv_va, tpte); } /* * We cannot remove wired pages from a process' mapping at this time */ if (tpte & PG_W) { allfree = 0; continue; } if (superpage) pa = tpte & PG_PS_FRAME; else pa = tpte & PG_FRAME; m = PHYS_TO_VM_PAGE(pa); KASSERT(m->phys_addr == pa, ("vm_page_t %p phys_addr mismatch %016jx %016jx", m, (uintmax_t)m->phys_addr, (uintmax_t)tpte)); KASSERT((m->flags & PG_FICTITIOUS) != 0 || m < &vm_page_array[vm_page_array_size], ("pmap_remove_pages: bad tpte %#jx", (uintmax_t)tpte)); pte_clear(pte); /* * Update the vm_page_t clean/reference bits. */ if ((tpte & (PG_M | PG_RW)) == (PG_M | PG_RW)) { if (superpage) { for (mt = m; mt < &m[NBPDR / PAGE_SIZE]; mt++) vm_page_dirty(mt); } else vm_page_dirty(m); } CHANGE_PV_LIST_LOCK_TO_VM_PAGE(&lock, m); /* Mark free */ pc->pc_map[field] |= bitmask; if (superpage) { pmap_resident_count_dec(pmap, NBPDR / PAGE_SIZE); pvh = pa_to_pvh(tpte & PG_PS_FRAME); TAILQ_REMOVE(&pvh->pv_list, pv, pv_next); pvh->pv_gen++; if (TAILQ_EMPTY(&pvh->pv_list)) { for (mt = m; mt < &m[NBPDR / PAGE_SIZE]; mt++) if ((mt->aflags & PGA_WRITEABLE) != 0 && TAILQ_EMPTY(&mt->md.pv_list)) vm_page_aflag_clear(mt, PGA_WRITEABLE); } mpte = pmap_lookup_pt_page(pmap, pv->pv_va); if (mpte != NULL) { pmap_remove_pt_page(pmap, mpte); pmap_resident_count_dec(pmap, 1); KASSERT(mpte->wire_count == NPTEPG, ("pmap_remove_pages: pte page wire count error")); mpte->wire_count = 0; pmap_add_delayed_free_list(mpte, &free, FALSE); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } } else { pmap_resident_count_dec(pmap, 1); TAILQ_REMOVE(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; if ((m->aflags & PGA_WRITEABLE) != 0 && TAILQ_EMPTY(&m->md.pv_list) && (m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); if (TAILQ_EMPTY(&pvh->pv_list)) vm_page_aflag_clear(m, PGA_WRITEABLE); } } pmap_unuse_pt(pmap, pv->pv_va, ptepde, &free); freed++; } } PV_STAT(atomic_add_long(&pv_entry_frees, freed)); PV_STAT(atomic_add_int(&pv_entry_spare, freed)); PV_STAT(atomic_subtract_long(&pv_entry_count, freed)); if (allfree) { TAILQ_REMOVE(&pmap->pm_pvchunk, pc, pc_list); free_pv_chunk(pc); } } if (lock != NULL) rw_wunlock(lock); pmap_invalidate_all(pmap); rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); pmap_free_zero_pages(&free); } static boolean_t pmap_page_test_mappings(vm_page_t m, boolean_t accessed, boolean_t modified) { struct rwlock *lock; pv_entry_t pv; struct md_page *pvh; pt_entry_t *pte, mask; pt_entry_t PG_A, PG_M, PG_RW, PG_V; pmap_t pmap; int md_gen, pvh_gen; boolean_t rv; rv = FALSE; rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); rw_rlock(lock); restart: TAILQ_FOREACH(pv, &m->md.pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { md_gen = m->md.pv_gen; rw_runlock(lock); PMAP_LOCK(pmap); rw_rlock(lock); if (md_gen != m->md.pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } pte = pmap_pte(pmap, pv->pv_va); mask = 0; if (modified) { PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); mask |= PG_RW | PG_M; } if (accessed) { PG_A = pmap_accessed_bit(pmap); PG_V = pmap_valid_bit(pmap); mask |= PG_V | PG_A; } rv = (*pte & mask) == mask; PMAP_UNLOCK(pmap); if (rv) goto out; } if ((m->flags & PG_FICTITIOUS) == 0) { pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); TAILQ_FOREACH(pv, &pvh->pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { md_gen = m->md.pv_gen; pvh_gen = pvh->pv_gen; rw_runlock(lock); PMAP_LOCK(pmap); rw_rlock(lock); if (md_gen != m->md.pv_gen || pvh_gen != pvh->pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } pte = pmap_pde(pmap, pv->pv_va); mask = 0; if (modified) { PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); mask |= PG_RW | PG_M; } if (accessed) { PG_A = pmap_accessed_bit(pmap); PG_V = pmap_valid_bit(pmap); mask |= PG_V | PG_A; } rv = (*pte & mask) == mask; PMAP_UNLOCK(pmap); if (rv) goto out; } } out: rw_runlock(lock); rw_runlock(&pvh_global_lock); return (rv); } /* * pmap_is_modified: * * Return whether or not the specified physical page was modified * in any physical maps. */ boolean_t pmap_is_modified(vm_page_t m) { KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_is_modified: page %p is not managed", m)); /* * If the page is not exclusive busied, then PGA_WRITEABLE cannot be * concurrently set while the object is locked. Thus, if PGA_WRITEABLE * is clear, no PTEs can have PG_M set. */ VM_OBJECT_ASSERT_WLOCKED(m->object); if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0) return (FALSE); return (pmap_page_test_mappings(m, FALSE, TRUE)); } /* * pmap_is_prefaultable: * * Return whether or not the specified virtual address is eligible * for prefault. */ boolean_t pmap_is_prefaultable(pmap_t pmap, vm_offset_t addr) { pd_entry_t *pde; pt_entry_t *pte, PG_V; boolean_t rv; PG_V = pmap_valid_bit(pmap); rv = FALSE; PMAP_LOCK(pmap); pde = pmap_pde(pmap, addr); if (pde != NULL && (*pde & (PG_PS | PG_V)) == PG_V) { pte = pmap_pde_to_pte(pde, addr); rv = (*pte & PG_V) == 0; } PMAP_UNLOCK(pmap); return (rv); } /* * pmap_is_referenced: * * Return whether or not the specified physical page was referenced * in any physical maps. */ boolean_t pmap_is_referenced(vm_page_t m) { KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_is_referenced: page %p is not managed", m)); return (pmap_page_test_mappings(m, TRUE, FALSE)); } /* * Clear the write and modified bits in each of the given page's mappings. */ void pmap_remove_write(vm_page_t m) { struct md_page *pvh; pmap_t pmap; struct rwlock *lock; pv_entry_t next_pv, pv; pd_entry_t *pde; pt_entry_t oldpte, *pte, PG_M, PG_RW; vm_offset_t va; int pvh_gen, md_gen; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_remove_write: page %p is not managed", m)); /* * If the page is not exclusive busied, then PGA_WRITEABLE cannot be * set by another thread while the object is locked. Thus, * if PGA_WRITEABLE is clear, no page table entries need updating. */ VM_OBJECT_ASSERT_WLOCKED(m->object); if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0) return; rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); retry_pv_loop: rw_wlock(lock); if ((m->flags & PG_FICTITIOUS) != 0) goto small_mappings; TAILQ_FOREACH_SAFE(pv, &pvh->pv_list, pv_next, next_pv) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { pvh_gen = pvh->pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen) { PMAP_UNLOCK(pmap); rw_wunlock(lock); goto retry_pv_loop; } } PG_RW = pmap_rw_bit(pmap); va = pv->pv_va; pde = pmap_pde(pmap, va); if ((*pde & PG_RW) != 0) (void)pmap_demote_pde_locked(pmap, pde, va, &lock); KASSERT(lock == VM_PAGE_TO_PV_LIST_LOCK(m), ("inconsistent pv lock %p %p for page %p", lock, VM_PAGE_TO_PV_LIST_LOCK(m), m)); PMAP_UNLOCK(pmap); } small_mappings: TAILQ_FOREACH(pv, &m->md.pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { pvh_gen = pvh->pv_gen; md_gen = m->md.pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen || md_gen != m->md.pv_gen) { PMAP_UNLOCK(pmap); rw_wunlock(lock); goto retry_pv_loop; } } PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); pde = pmap_pde(pmap, pv->pv_va); KASSERT((*pde & PG_PS) == 0, ("pmap_remove_write: found a 2mpage in page %p's pv list", m)); pte = pmap_pde_to_pte(pde, pv->pv_va); retry: oldpte = *pte; if (oldpte & PG_RW) { if (!atomic_cmpset_long(pte, oldpte, oldpte & ~(PG_RW | PG_M))) goto retry; if ((oldpte & PG_M) != 0) vm_page_dirty(m); pmap_invalidate_page(pmap, pv->pv_va); } PMAP_UNLOCK(pmap); } rw_wunlock(lock); vm_page_aflag_clear(m, PGA_WRITEABLE); rw_runlock(&pvh_global_lock); } static __inline boolean_t safe_to_clear_referenced(pmap_t pmap, pt_entry_t pte) { if (!pmap_emulate_ad_bits(pmap)) return (TRUE); KASSERT(pmap->pm_type == PT_EPT, ("invalid pm_type %d", pmap->pm_type)); /* * XWR = 010 or 110 will cause an unconditional EPT misconfiguration * so we don't let the referenced (aka EPT_PG_READ) bit to be cleared * if the EPT_PG_WRITE bit is set. */ if ((pte & EPT_PG_WRITE) != 0) return (FALSE); /* * XWR = 100 is allowed only if the PMAP_SUPPORTS_EXEC_ONLY is set. */ if ((pte & EPT_PG_EXECUTE) == 0 || ((pmap->pm_flags & PMAP_SUPPORTS_EXEC_ONLY) != 0)) return (TRUE); else return (FALSE); } #define PMAP_TS_REFERENCED_MAX 5 /* * pmap_ts_referenced: * * Return a count of reference bits for a page, clearing those bits. * It is not necessary for every reference bit to be cleared, but it * is necessary that 0 only be returned when there are truly no * reference bits set. * * XXX: The exact number of bits to check and clear is a matter that * should be tested and standardized at some point in the future for * optimal aging of shared pages. */ int pmap_ts_referenced(vm_page_t m) { struct md_page *pvh; pv_entry_t pv, pvf; pmap_t pmap; struct rwlock *lock; pd_entry_t oldpde, *pde; pt_entry_t *pte, PG_A; vm_offset_t va; vm_paddr_t pa; int cleared, md_gen, not_cleared, pvh_gen; struct spglist free; boolean_t demoted; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_ts_referenced: page %p is not managed", m)); SLIST_INIT(&free); cleared = 0; pa = VM_PAGE_TO_PHYS(m); lock = PHYS_TO_PV_LIST_LOCK(pa); pvh = pa_to_pvh(pa); rw_rlock(&pvh_global_lock); rw_wlock(lock); retry: not_cleared = 0; if ((m->flags & PG_FICTITIOUS) != 0 || (pvf = TAILQ_FIRST(&pvh->pv_list)) == NULL) goto small_mappings; pv = pvf; do { if (pvf == NULL) pvf = pv; pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { pvh_gen = pvh->pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen) { PMAP_UNLOCK(pmap); goto retry; } } PG_A = pmap_accessed_bit(pmap); va = pv->pv_va; pde = pmap_pde(pmap, pv->pv_va); oldpde = *pde; if ((*pde & PG_A) != 0) { /* * Since this reference bit is shared by 512 4KB * pages, it should not be cleared every time it is * tested. Apply a simple "hash" function on the * physical page number, the virtual superpage number, * and the pmap address to select one 4KB page out of * the 512 on which testing the reference bit will * result in clearing that reference bit. This * function is designed to avoid the selection of the * same 4KB page for every 2MB page mapping. * * On demotion, a mapping that hasn't been referenced * is simply destroyed. To avoid the possibility of a * subsequent page fault on a demoted wired mapping, * always leave its reference bit set. Moreover, * since the superpage is wired, the current state of * its reference bit won't affect page replacement. */ if ((((pa >> PAGE_SHIFT) ^ (pv->pv_va >> PDRSHIFT) ^ (uintptr_t)pmap) & (NPTEPG - 1)) == 0 && (*pde & PG_W) == 0) { if (safe_to_clear_referenced(pmap, oldpde)) { atomic_clear_long(pde, PG_A); pmap_invalidate_page(pmap, pv->pv_va); demoted = FALSE; } else if (pmap_demote_pde_locked(pmap, pde, pv->pv_va, &lock)) { /* * Remove the mapping to a single page * so that a subsequent access may * repromote. Since the underlying * page table page is fully populated, * this removal never frees a page * table page. */ demoted = TRUE; va += VM_PAGE_TO_PHYS(m) - (oldpde & PG_PS_FRAME); pte = pmap_pde_to_pte(pde, va); pmap_remove_pte(pmap, pte, va, *pde, NULL, &lock); pmap_invalidate_page(pmap, va); } else demoted = TRUE; if (demoted) { /* * The superpage mapping was removed * entirely and therefore 'pv' is no * longer valid. */ if (pvf == pv) pvf = NULL; pv = NULL; } cleared++; KASSERT(lock == VM_PAGE_TO_PV_LIST_LOCK(m), ("inconsistent pv lock %p %p for page %p", lock, VM_PAGE_TO_PV_LIST_LOCK(m), m)); } else not_cleared++; } PMAP_UNLOCK(pmap); /* Rotate the PV list if it has more than one entry. */ if (pv != NULL && TAILQ_NEXT(pv, pv_next) != NULL) { TAILQ_REMOVE(&pvh->pv_list, pv, pv_next); TAILQ_INSERT_TAIL(&pvh->pv_list, pv, pv_next); pvh->pv_gen++; } if (cleared + not_cleared >= PMAP_TS_REFERENCED_MAX) goto out; } while ((pv = TAILQ_FIRST(&pvh->pv_list)) != pvf); small_mappings: if ((pvf = TAILQ_FIRST(&m->md.pv_list)) == NULL) goto out; pv = pvf; do { if (pvf == NULL) pvf = pv; pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { pvh_gen = pvh->pv_gen; md_gen = m->md.pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen || md_gen != m->md.pv_gen) { PMAP_UNLOCK(pmap); goto retry; } } PG_A = pmap_accessed_bit(pmap); pde = pmap_pde(pmap, pv->pv_va); KASSERT((*pde & PG_PS) == 0, ("pmap_ts_referenced: found a 2mpage in page %p's pv list", m)); pte = pmap_pde_to_pte(pde, pv->pv_va); if ((*pte & PG_A) != 0) { if (safe_to_clear_referenced(pmap, *pte)) { atomic_clear_long(pte, PG_A); pmap_invalidate_page(pmap, pv->pv_va); cleared++; } else if ((*pte & PG_W) == 0) { /* * Wired pages cannot be paged out so * doing accessed bit emulation for * them is wasted effort. We do the * hard work for unwired pages only. */ pmap_remove_pte(pmap, pte, pv->pv_va, *pde, &free, &lock); pmap_invalidate_page(pmap, pv->pv_va); cleared++; if (pvf == pv) pvf = NULL; pv = NULL; KASSERT(lock == VM_PAGE_TO_PV_LIST_LOCK(m), ("inconsistent pv lock %p %p for page %p", lock, VM_PAGE_TO_PV_LIST_LOCK(m), m)); } else not_cleared++; } PMAP_UNLOCK(pmap); /* Rotate the PV list if it has more than one entry. */ if (pv != NULL && TAILQ_NEXT(pv, pv_next) != NULL) { TAILQ_REMOVE(&m->md.pv_list, pv, pv_next); TAILQ_INSERT_TAIL(&m->md.pv_list, pv, pv_next); m->md.pv_gen++; } } while ((pv = TAILQ_FIRST(&m->md.pv_list)) != pvf && cleared + not_cleared < PMAP_TS_REFERENCED_MAX); out: rw_wunlock(lock); rw_runlock(&pvh_global_lock); pmap_free_zero_pages(&free); return (cleared + not_cleared); } /* * Apply the given advice to the specified range of addresses within the * given pmap. Depending on the advice, clear the referenced and/or * modified flags in each mapping and set the mapped page's dirty field. */ void pmap_advise(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, int advice) { struct rwlock *lock; pml4_entry_t *pml4e; pdp_entry_t *pdpe; pd_entry_t oldpde, *pde; pt_entry_t *pte, PG_A, PG_G, PG_M, PG_RW, PG_V; vm_offset_t va_next; vm_page_t m; boolean_t anychanged, pv_lists_locked; if (advice != MADV_DONTNEED && advice != MADV_FREE) return; /* * A/D bit emulation requires an alternate code path when clearing * the modified and accessed bits below. Since this function is * advisory in nature we skip it entirely for pmaps that require * A/D bit emulation. */ if (pmap_emulate_ad_bits(pmap)) return; PG_A = pmap_accessed_bit(pmap); PG_G = pmap_global_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); pv_lists_locked = FALSE; resume: anychanged = FALSE; PMAP_LOCK(pmap); for (; sva < eva; sva = va_next) { pml4e = pmap_pml4e(pmap, sva); if ((*pml4e & PG_V) == 0) { va_next = (sva + NBPML4) & ~PML4MASK; if (va_next < sva) va_next = eva; continue; } pdpe = pmap_pml4e_to_pdpe(pml4e, sva); if ((*pdpe & PG_V) == 0) { va_next = (sva + NBPDP) & ~PDPMASK; if (va_next < sva) va_next = eva; continue; } va_next = (sva + NBPDR) & ~PDRMASK; if (va_next < sva) va_next = eva; pde = pmap_pdpe_to_pde(pdpe, sva); oldpde = *pde; if ((oldpde & PG_V) == 0) continue; else if ((oldpde & PG_PS) != 0) { if ((oldpde & PG_MANAGED) == 0) continue; if (!pv_lists_locked) { pv_lists_locked = TRUE; if (!rw_try_rlock(&pvh_global_lock)) { if (anychanged) pmap_invalidate_all(pmap); PMAP_UNLOCK(pmap); rw_rlock(&pvh_global_lock); goto resume; } } lock = NULL; if (!pmap_demote_pde_locked(pmap, pde, sva, &lock)) { if (lock != NULL) rw_wunlock(lock); /* * The large page mapping was destroyed. */ continue; } /* * Unless the page mappings are wired, remove the * mapping to a single page so that a subsequent * access may repromote. Since the underlying page * table page is fully populated, this removal never * frees a page table page. */ if ((oldpde & PG_W) == 0) { pte = pmap_pde_to_pte(pde, sva); KASSERT((*pte & PG_V) != 0, ("pmap_advise: invalid PTE")); pmap_remove_pte(pmap, pte, sva, *pde, NULL, &lock); anychanged = TRUE; } if (lock != NULL) rw_wunlock(lock); } if (va_next > eva) va_next = eva; for (pte = pmap_pde_to_pte(pde, sva); sva != va_next; pte++, sva += PAGE_SIZE) { if ((*pte & (PG_MANAGED | PG_V)) != (PG_MANAGED | PG_V)) continue; else if ((*pte & (PG_M | PG_RW)) == (PG_M | PG_RW)) { if (advice == MADV_DONTNEED) { /* * Future calls to pmap_is_modified() * can be avoided by making the page * dirty now. */ m = PHYS_TO_VM_PAGE(*pte & PG_FRAME); vm_page_dirty(m); } atomic_clear_long(pte, PG_M | PG_A); } else if ((*pte & PG_A) != 0) atomic_clear_long(pte, PG_A); else continue; if ((*pte & PG_G) != 0) pmap_invalidate_page(pmap, sva); else anychanged = TRUE; } } if (anychanged) pmap_invalidate_all(pmap); if (pv_lists_locked) rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); } /* * Clear the modify bits on the specified physical page. */ void pmap_clear_modify(vm_page_t m) { struct md_page *pvh; pmap_t pmap; pv_entry_t next_pv, pv; pd_entry_t oldpde, *pde; pt_entry_t oldpte, *pte, PG_M, PG_RW, PG_V; struct rwlock *lock; vm_offset_t va; int md_gen, pvh_gen; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("pmap_clear_modify: page %p is not managed", m)); VM_OBJECT_ASSERT_WLOCKED(m->object); KASSERT(!vm_page_xbusied(m), ("pmap_clear_modify: page %p is exclusive busied", m)); /* * If the page is not PGA_WRITEABLE, then no PTEs can have PG_M set. * If the object containing the page is locked and the page is not * exclusive busied, then PGA_WRITEABLE cannot be concurrently set. */ if ((m->aflags & PGA_WRITEABLE) == 0) return; pvh = pa_to_pvh(VM_PAGE_TO_PHYS(m)); rw_rlock(&pvh_global_lock); lock = VM_PAGE_TO_PV_LIST_LOCK(m); rw_wlock(lock); restart: if ((m->flags & PG_FICTITIOUS) != 0) goto small_mappings; TAILQ_FOREACH_SAFE(pv, &pvh->pv_list, pv_next, next_pv) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { pvh_gen = pvh->pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); va = pv->pv_va; pde = pmap_pde(pmap, va); oldpde = *pde; if ((oldpde & PG_RW) != 0) { if (pmap_demote_pde_locked(pmap, pde, va, &lock)) { if ((oldpde & PG_W) == 0) { /* * Write protect the mapping to a * single page so that a subsequent * write access may repromote. */ va += VM_PAGE_TO_PHYS(m) - (oldpde & PG_PS_FRAME); pte = pmap_pde_to_pte(pde, va); oldpte = *pte; if ((oldpte & PG_V) != 0) { while (!atomic_cmpset_long(pte, oldpte, oldpte & ~(PG_M | PG_RW))) oldpte = *pte; vm_page_dirty(m); pmap_invalidate_page(pmap, va); } } } } PMAP_UNLOCK(pmap); } small_mappings: TAILQ_FOREACH(pv, &m->md.pv_list, pv_next) { pmap = PV_PMAP(pv); if (!PMAP_TRYLOCK(pmap)) { md_gen = m->md.pv_gen; pvh_gen = pvh->pv_gen; rw_wunlock(lock); PMAP_LOCK(pmap); rw_wlock(lock); if (pvh_gen != pvh->pv_gen || md_gen != m->md.pv_gen) { PMAP_UNLOCK(pmap); goto restart; } } PG_M = pmap_modified_bit(pmap); PG_RW = pmap_rw_bit(pmap); pde = pmap_pde(pmap, pv->pv_va); KASSERT((*pde & PG_PS) == 0, ("pmap_clear_modify: found" " a 2mpage in page %p's pv list", m)); pte = pmap_pde_to_pte(pde, pv->pv_va); if ((*pte & (PG_M | PG_RW)) == (PG_M | PG_RW)) { atomic_clear_long(pte, PG_M); pmap_invalidate_page(pmap, pv->pv_va); } PMAP_UNLOCK(pmap); } rw_wunlock(lock); rw_runlock(&pvh_global_lock); } /* * Miscellaneous support routines follow */ /* Adjust the cache mode for a 4KB page mapped via a PTE. */ static __inline void pmap_pte_attr(pt_entry_t *pte, int cache_bits, int mask) { u_int opte, npte; /* * The cache mode bits are all in the low 32-bits of the * PTE, so we can just spin on updating the low 32-bits. */ do { opte = *(u_int *)pte; npte = opte & ~mask; npte |= cache_bits; } while (npte != opte && !atomic_cmpset_int((u_int *)pte, opte, npte)); } /* Adjust the cache mode for a 2MB page mapped via a PDE. */ static __inline void pmap_pde_attr(pd_entry_t *pde, int cache_bits, int mask) { u_int opde, npde; /* * The cache mode bits are all in the low 32-bits of the * PDE, so we can just spin on updating the low 32-bits. */ do { opde = *(u_int *)pde; npde = opde & ~mask; npde |= cache_bits; } while (npde != opde && !atomic_cmpset_int((u_int *)pde, opde, npde)); } /* * Map a set of physical memory pages into the kernel virtual * address space. Return a pointer to where it is mapped. This * routine is intended to be used for mapping device memory, * NOT real memory. */ void * pmap_mapdev_attr(vm_paddr_t pa, vm_size_t size, int mode) { struct pmap_preinit_mapping *ppim; vm_offset_t va, offset; vm_size_t tmpsize; int i; offset = pa & PAGE_MASK; size = round_page(offset + size); pa = trunc_page(pa); if (!pmap_initialized) { va = 0; for (i = 0; i < PMAP_PREINIT_MAPPING_COUNT; i++) { ppim = pmap_preinit_mapping + i; if (ppim->va == 0) { ppim->pa = pa; ppim->sz = size; ppim->mode = mode; ppim->va = virtual_avail; virtual_avail += size; va = ppim->va; break; } } if (va == 0) panic("%s: too many preinit mappings", __func__); } else { /* * If we have a preinit mapping, re-use it. */ for (i = 0; i < PMAP_PREINIT_MAPPING_COUNT; i++) { ppim = pmap_preinit_mapping + i; if (ppim->pa == pa && ppim->sz == size && ppim->mode == mode) return ((void *)(ppim->va + offset)); } /* * If the specified range of physical addresses fits within * the direct map window, use the direct map. */ if (pa < dmaplimit && pa + size < dmaplimit) { va = PHYS_TO_DMAP(pa); if (!pmap_change_attr(va, size, mode)) return ((void *)(va + offset)); } va = kva_alloc(size); if (va == 0) panic("%s: Couldn't allocate KVA", __func__); } for (tmpsize = 0; tmpsize < size; tmpsize += PAGE_SIZE) pmap_kenter_attr(va + tmpsize, pa + tmpsize, mode); pmap_invalidate_range(kernel_pmap, va, va + tmpsize); pmap_invalidate_cache_range(va, va + tmpsize, FALSE); return ((void *)(va + offset)); } void * pmap_mapdev(vm_paddr_t pa, vm_size_t size) { return (pmap_mapdev_attr(pa, size, PAT_UNCACHEABLE)); } void * pmap_mapbios(vm_paddr_t pa, vm_size_t size) { return (pmap_mapdev_attr(pa, size, PAT_WRITE_BACK)); } void pmap_unmapdev(vm_offset_t va, vm_size_t size) { struct pmap_preinit_mapping *ppim; vm_offset_t offset; int i; /* If we gave a direct map region in pmap_mapdev, do nothing */ if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) return; offset = va & PAGE_MASK; size = round_page(offset + size); va = trunc_page(va); for (i = 0; i < PMAP_PREINIT_MAPPING_COUNT; i++) { ppim = pmap_preinit_mapping + i; if (ppim->va == va && ppim->sz == size) { if (pmap_initialized) return; ppim->pa = 0; ppim->va = 0; ppim->sz = 0; ppim->mode = 0; if (va + size == virtual_avail) virtual_avail = va; return; } } if (pmap_initialized) kva_free(va, size); } /* * Tries to demote a 1GB page mapping. */ static boolean_t pmap_demote_pdpe(pmap_t pmap, pdp_entry_t *pdpe, vm_offset_t va) { pdp_entry_t newpdpe, oldpdpe; pd_entry_t *firstpde, newpde, *pde; pt_entry_t PG_A, PG_M, PG_RW, PG_V; vm_paddr_t mpdepa; vm_page_t mpde; PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); PMAP_LOCK_ASSERT(pmap, MA_OWNED); oldpdpe = *pdpe; KASSERT((oldpdpe & (PG_PS | PG_V)) == (PG_PS | PG_V), ("pmap_demote_pdpe: oldpdpe is missing PG_PS and/or PG_V")); if ((mpde = vm_page_alloc(NULL, va >> PDPSHIFT, VM_ALLOC_INTERRUPT | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) { CTR2(KTR_PMAP, "pmap_demote_pdpe: failure for va %#lx" " in pmap %p", va, pmap); return (FALSE); } mpdepa = VM_PAGE_TO_PHYS(mpde); firstpde = (pd_entry_t *)PHYS_TO_DMAP(mpdepa); newpdpe = mpdepa | PG_M | PG_A | (oldpdpe & PG_U) | PG_RW | PG_V; KASSERT((oldpdpe & PG_A) != 0, ("pmap_demote_pdpe: oldpdpe is missing PG_A")); KASSERT((oldpdpe & (PG_M | PG_RW)) != PG_RW, ("pmap_demote_pdpe: oldpdpe is missing PG_M")); newpde = oldpdpe; /* * Initialize the page directory page. */ for (pde = firstpde; pde < firstpde + NPDEPG; pde++) { *pde = newpde; newpde += NBPDR; } /* * Demote the mapping. */ *pdpe = newpdpe; /* * Invalidate a stale recursive mapping of the page directory page. */ pmap_invalidate_page(pmap, (vm_offset_t)vtopde(va)); pmap_pdpe_demotions++; CTR2(KTR_PMAP, "pmap_demote_pdpe: success for va %#lx" " in pmap %p", va, pmap); return (TRUE); } /* * Sets the memory attribute for the specified page. */ void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) { m->md.pat_mode = ma; /* * If "m" is a normal page, update its direct mapping. This update * can be relied upon to perform any cache operations that are * required for data coherence. */ if ((m->flags & PG_FICTITIOUS) == 0 && pmap_change_attr(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)), PAGE_SIZE, m->md.pat_mode)) panic("memory attribute change on the direct map failed"); } /* * Changes the specified virtual address range's memory type to that given by * the parameter "mode". The specified virtual address range must be * completely contained within either the direct map or the kernel map. If * the virtual address range is contained within the kernel map, then the * memory type for each of the corresponding ranges of the direct map is also * changed. (The corresponding ranges of the direct map are those ranges that * map the same physical pages as the specified virtual address range.) These * changes to the direct map are necessary because Intel describes the * behavior of their processors as "undefined" if two or more mappings to the * same physical page have different memory types. * * Returns zero if the change completed successfully, and either EINVAL or * ENOMEM if the change failed. Specifically, EINVAL is returned if some part * of the virtual address range was not mapped, and ENOMEM is returned if * there was insufficient memory available to complete the change. In the * latter case, the memory type may have been changed on some part of the * virtual address range or the direct map. */ int pmap_change_attr(vm_offset_t va, vm_size_t size, int mode) { int error; PMAP_LOCK(kernel_pmap); error = pmap_change_attr_locked(va, size, mode); PMAP_UNLOCK(kernel_pmap); return (error); } static int pmap_change_attr_locked(vm_offset_t va, vm_size_t size, int mode) { vm_offset_t base, offset, tmpva; vm_paddr_t pa_start, pa_end; pdp_entry_t *pdpe; pd_entry_t *pde; pt_entry_t *pte; int cache_bits_pte, cache_bits_pde, error; boolean_t changed; PMAP_LOCK_ASSERT(kernel_pmap, MA_OWNED); base = trunc_page(va); offset = va & PAGE_MASK; size = round_page(offset + size); /* * Only supported on kernel virtual addresses, including the direct * map but excluding the recursive map. */ if (base < DMAP_MIN_ADDRESS) return (EINVAL); cache_bits_pde = pmap_cache_bits(kernel_pmap, mode, 1); cache_bits_pte = pmap_cache_bits(kernel_pmap, mode, 0); changed = FALSE; /* * Pages that aren't mapped aren't supported. Also break down 2MB pages * into 4KB pages if required. */ for (tmpva = base; tmpva < base + size; ) { pdpe = pmap_pdpe(kernel_pmap, tmpva); if (pdpe == NULL || *pdpe == 0) return (EINVAL); if (*pdpe & PG_PS) { /* * If the current 1GB page already has the required * memory type, then we need not demote this page. Just * increment tmpva to the next 1GB page frame. */ if ((*pdpe & X86_PG_PDE_CACHE) == cache_bits_pde) { tmpva = trunc_1gpage(tmpva) + NBPDP; continue; } /* * If the current offset aligns with a 1GB page frame * and there is at least 1GB left within the range, then * we need not break down this page into 2MB pages. */ if ((tmpva & PDPMASK) == 0 && tmpva + PDPMASK < base + size) { tmpva += NBPDP; continue; } if (!pmap_demote_pdpe(kernel_pmap, pdpe, tmpva)) return (ENOMEM); } pde = pmap_pdpe_to_pde(pdpe, tmpva); if (*pde == 0) return (EINVAL); if (*pde & PG_PS) { /* * If the current 2MB page already has the required * memory type, then we need not demote this page. Just * increment tmpva to the next 2MB page frame. */ if ((*pde & X86_PG_PDE_CACHE) == cache_bits_pde) { tmpva = trunc_2mpage(tmpva) + NBPDR; continue; } /* * If the current offset aligns with a 2MB page frame * and there is at least 2MB left within the range, then * we need not break down this page into 4KB pages. */ if ((tmpva & PDRMASK) == 0 && tmpva + PDRMASK < base + size) { tmpva += NBPDR; continue; } if (!pmap_demote_pde(kernel_pmap, pde, tmpva)) return (ENOMEM); } pte = pmap_pde_to_pte(pde, tmpva); if (*pte == 0) return (EINVAL); tmpva += PAGE_SIZE; } error = 0; /* * Ok, all the pages exist, so run through them updating their * cache mode if required. */ pa_start = pa_end = 0; for (tmpva = base; tmpva < base + size; ) { pdpe = pmap_pdpe(kernel_pmap, tmpva); if (*pdpe & PG_PS) { if ((*pdpe & X86_PG_PDE_CACHE) != cache_bits_pde) { pmap_pde_attr(pdpe, cache_bits_pde, X86_PG_PDE_CACHE); changed = TRUE; } if (tmpva >= VM_MIN_KERNEL_ADDRESS && (*pdpe & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ pa_start = *pdpe & PG_PS_FRAME; pa_end = pa_start + NBPDP; } else if (pa_end == (*pdpe & PG_PS_FRAME)) pa_end += NBPDP; else { /* Run ended, update direct map. */ error = pmap_change_attr_locked( PHYS_TO_DMAP(pa_start), pa_end - pa_start, mode); if (error != 0) break; /* Start physical address run. */ pa_start = *pdpe & PG_PS_FRAME; pa_end = pa_start + NBPDP; } } tmpva = trunc_1gpage(tmpva) + NBPDP; continue; } pde = pmap_pdpe_to_pde(pdpe, tmpva); if (*pde & PG_PS) { if ((*pde & X86_PG_PDE_CACHE) != cache_bits_pde) { pmap_pde_attr(pde, cache_bits_pde, X86_PG_PDE_CACHE); changed = TRUE; } if (tmpva >= VM_MIN_KERNEL_ADDRESS && (*pde & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ pa_start = *pde & PG_PS_FRAME; pa_end = pa_start + NBPDR; } else if (pa_end == (*pde & PG_PS_FRAME)) pa_end += NBPDR; else { /* Run ended, update direct map. */ error = pmap_change_attr_locked( PHYS_TO_DMAP(pa_start), pa_end - pa_start, mode); if (error != 0) break; /* Start physical address run. */ pa_start = *pde & PG_PS_FRAME; pa_end = pa_start + NBPDR; } } tmpva = trunc_2mpage(tmpva) + NBPDR; } else { pte = pmap_pde_to_pte(pde, tmpva); if ((*pte & X86_PG_PTE_CACHE) != cache_bits_pte) { pmap_pte_attr(pte, cache_bits_pte, X86_PG_PTE_CACHE); changed = TRUE; } if (tmpva >= VM_MIN_KERNEL_ADDRESS && (*pte & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ pa_start = *pte & PG_FRAME; pa_end = pa_start + PAGE_SIZE; } else if (pa_end == (*pte & PG_FRAME)) pa_end += PAGE_SIZE; else { /* Run ended, update direct map. */ error = pmap_change_attr_locked( PHYS_TO_DMAP(pa_start), pa_end - pa_start, mode); if (error != 0) break; /* Start physical address run. */ pa_start = *pte & PG_FRAME; pa_end = pa_start + PAGE_SIZE; } } tmpva += PAGE_SIZE; } } if (error == 0 && pa_start != pa_end) error = pmap_change_attr_locked(PHYS_TO_DMAP(pa_start), pa_end - pa_start, mode); /* * Flush CPU caches if required to make sure any data isn't cached that * shouldn't be, etc. */ if (changed) { pmap_invalidate_range(kernel_pmap, base, tmpva); pmap_invalidate_cache_range(base, tmpva, FALSE); } return (error); } /* * Demotes any mapping within the direct map region that covers more than the * specified range of physical addresses. This range's size must be a power * of two and its starting address must be a multiple of its size. Since the * demotion does not change any attributes of the mapping, a TLB invalidation * is not mandatory. The caller may, however, request a TLB invalidation. */ void pmap_demote_DMAP(vm_paddr_t base, vm_size_t len, boolean_t invalidate) { pdp_entry_t *pdpe; pd_entry_t *pde; vm_offset_t va; boolean_t changed; if (len == 0) return; KASSERT(powerof2(len), ("pmap_demote_DMAP: len is not a power of 2")); KASSERT((base & (len - 1)) == 0, ("pmap_demote_DMAP: base is not a multiple of len")); if (len < NBPDP && base < dmaplimit) { va = PHYS_TO_DMAP(base); changed = FALSE; PMAP_LOCK(kernel_pmap); pdpe = pmap_pdpe(kernel_pmap, va); if ((*pdpe & X86_PG_V) == 0) panic("pmap_demote_DMAP: invalid PDPE"); if ((*pdpe & PG_PS) != 0) { if (!pmap_demote_pdpe(kernel_pmap, pdpe, va)) panic("pmap_demote_DMAP: PDPE failed"); changed = TRUE; } if (len < NBPDR) { pde = pmap_pdpe_to_pde(pdpe, va); if ((*pde & X86_PG_V) == 0) panic("pmap_demote_DMAP: invalid PDE"); if ((*pde & PG_PS) != 0) { if (!pmap_demote_pde(kernel_pmap, pde, va)) panic("pmap_demote_DMAP: PDE failed"); changed = TRUE; } } if (changed && invalidate) pmap_invalidate_page(kernel_pmap, va); PMAP_UNLOCK(kernel_pmap); } } /* * perform the pmap work for mincore */ int pmap_mincore(pmap_t pmap, vm_offset_t addr, vm_paddr_t *locked_pa) { pd_entry_t *pdep; pt_entry_t pte, PG_A, PG_M, PG_RW, PG_V; vm_paddr_t pa; int val; PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); PMAP_LOCK(pmap); retry: pdep = pmap_pde(pmap, addr); if (pdep != NULL && (*pdep & PG_V)) { if (*pdep & PG_PS) { pte = *pdep; /* Compute the physical address of the 4KB page. */ pa = ((*pdep & PG_PS_FRAME) | (addr & PDRMASK)) & PG_FRAME; val = MINCORE_SUPER; } else { pte = *pmap_pde_to_pte(pdep, addr); pa = pte & PG_FRAME; val = 0; } } else { pte = 0; pa = 0; val = 0; } if ((pte & PG_V) != 0) { val |= MINCORE_INCORE; if ((pte & (PG_M | PG_RW)) == (PG_M | PG_RW)) val |= MINCORE_MODIFIED | MINCORE_MODIFIED_OTHER; if ((pte & PG_A) != 0) val |= MINCORE_REFERENCED | MINCORE_REFERENCED_OTHER; } if ((val & (MINCORE_MODIFIED_OTHER | MINCORE_REFERENCED_OTHER)) != (MINCORE_MODIFIED_OTHER | MINCORE_REFERENCED_OTHER) && (pte & (PG_MANAGED | PG_V)) == (PG_MANAGED | PG_V)) { /* Ensure that "PHYS_TO_VM_PAGE(pa)->object" doesn't change. */ if (vm_page_pa_tryrelock(pmap, pa, locked_pa)) goto retry; } else PA_UNLOCK_COND(*locked_pa); PMAP_UNLOCK(pmap); return (val); } static uint64_t pmap_pcid_alloc(pmap_t pmap, u_int cpuid) { uint32_t gen, new_gen, pcid_next; CRITICAL_ASSERT(curthread); gen = PCPU_GET(pcid_gen); if (pmap->pm_pcids[cpuid].pm_pcid == PMAP_PCID_KERN || pmap->pm_pcids[cpuid].pm_gen == gen) return (CR3_PCID_SAVE); pcid_next = PCPU_GET(pcid_next); KASSERT(pcid_next <= PMAP_PCID_OVERMAX, ("cpu %d pcid_next %#x", cpuid, pcid_next)); if (pcid_next == PMAP_PCID_OVERMAX) { new_gen = gen + 1; if (new_gen == 0) new_gen = 1; PCPU_SET(pcid_gen, new_gen); pcid_next = PMAP_PCID_KERN + 1; } else { new_gen = gen; } pmap->pm_pcids[cpuid].pm_pcid = pcid_next; pmap->pm_pcids[cpuid].pm_gen = new_gen; PCPU_SET(pcid_next, pcid_next + 1); return (0); } void pmap_activate_sw(struct thread *td) { pmap_t oldpmap, pmap; uint64_t cached, cr3; u_int cpuid; oldpmap = PCPU_GET(curpmap); pmap = vmspace_pmap(td->td_proc->p_vmspace); if (oldpmap == pmap) return; cpuid = PCPU_GET(cpuid); #ifdef SMP CPU_SET_ATOMIC(cpuid, &pmap->pm_active); #else CPU_SET(cpuid, &pmap->pm_active); #endif cr3 = rcr3(); if (pmap_pcid_enabled) { cached = pmap_pcid_alloc(pmap, cpuid); KASSERT(pmap->pm_pcids[cpuid].pm_pcid >= 0 && pmap->pm_pcids[cpuid].pm_pcid < PMAP_PCID_OVERMAX, ("pmap %p cpu %d pcid %#x", pmap, cpuid, pmap->pm_pcids[cpuid].pm_pcid)); KASSERT(pmap->pm_pcids[cpuid].pm_pcid != PMAP_PCID_KERN || pmap == kernel_pmap, ("non-kernel pmap thread %p pmap %p cpu %d pcid %#x", td, pmap, cpuid, pmap->pm_pcids[cpuid].pm_pcid)); if (!cached || (cr3 & ~CR3_PCID_MASK) != pmap->pm_cr3) { load_cr3(pmap->pm_cr3 | pmap->pm_pcids[cpuid].pm_pcid | cached); if (cached) PCPU_INC(pm_save_cnt); } } else if (cr3 != pmap->pm_cr3) { load_cr3(pmap->pm_cr3); } PCPU_SET(curpmap, pmap); #ifdef SMP CPU_CLR_ATOMIC(cpuid, &oldpmap->pm_active); #else CPU_CLR(cpuid, &oldpmap->pm_active); #endif } void pmap_activate(struct thread *td) { critical_enter(); pmap_activate_sw(td); critical_exit(); } void pmap_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz) { } /* * Increase the starting virtual address of the given mapping if a * different alignment might result in more superpage mappings. */ void pmap_align_superpage(vm_object_t object, vm_ooffset_t offset, vm_offset_t *addr, vm_size_t size) { vm_offset_t superpage_offset; if (size < NBPDR) return; if (object != NULL && (object->flags & OBJ_COLORED) != 0) offset += ptoa(object->pg_color); superpage_offset = offset & PDRMASK; if (size - ((NBPDR - superpage_offset) & PDRMASK) < NBPDR || (*addr & PDRMASK) == superpage_offset) return; if ((*addr & PDRMASK) < superpage_offset) *addr = (*addr & ~PDRMASK) + superpage_offset; else *addr = ((*addr + PDRMASK) & ~PDRMASK) + superpage_offset; } #ifdef INVARIANTS static unsigned long num_dirty_emulations; SYSCTL_ULONG(_vm_pmap, OID_AUTO, num_dirty_emulations, CTLFLAG_RW, &num_dirty_emulations, 0, NULL); static unsigned long num_accessed_emulations; SYSCTL_ULONG(_vm_pmap, OID_AUTO, num_accessed_emulations, CTLFLAG_RW, &num_accessed_emulations, 0, NULL); static unsigned long num_superpage_accessed_emulations; SYSCTL_ULONG(_vm_pmap, OID_AUTO, num_superpage_accessed_emulations, CTLFLAG_RW, &num_superpage_accessed_emulations, 0, NULL); static unsigned long ad_emulation_superpage_promotions; SYSCTL_ULONG(_vm_pmap, OID_AUTO, ad_emulation_superpage_promotions, CTLFLAG_RW, &ad_emulation_superpage_promotions, 0, NULL); #endif /* INVARIANTS */ int pmap_emulate_accessed_dirty(pmap_t pmap, vm_offset_t va, int ftype) { int rv; struct rwlock *lock; vm_page_t m, mpte; pd_entry_t *pde; pt_entry_t *pte, PG_A, PG_M, PG_RW, PG_V; boolean_t pv_lists_locked; KASSERT(ftype == VM_PROT_READ || ftype == VM_PROT_WRITE, ("pmap_emulate_accessed_dirty: invalid fault type %d", ftype)); if (!pmap_emulate_ad_bits(pmap)) return (-1); PG_A = pmap_accessed_bit(pmap); PG_M = pmap_modified_bit(pmap); PG_V = pmap_valid_bit(pmap); PG_RW = pmap_rw_bit(pmap); rv = -1; lock = NULL; pv_lists_locked = FALSE; retry: PMAP_LOCK(pmap); pde = pmap_pde(pmap, va); if (pde == NULL || (*pde & PG_V) == 0) goto done; if ((*pde & PG_PS) != 0) { if (ftype == VM_PROT_READ) { #ifdef INVARIANTS atomic_add_long(&num_superpage_accessed_emulations, 1); #endif *pde |= PG_A; rv = 0; } goto done; } pte = pmap_pde_to_pte(pde, va); if ((*pte & PG_V) == 0) goto done; if (ftype == VM_PROT_WRITE) { if ((*pte & PG_RW) == 0) goto done; /* * Set the modified and accessed bits simultaneously. * * Intel EPT PTEs that do software emulation of A/D bits map * PG_A and PG_M to EPT_PG_READ and EPT_PG_WRITE respectively. * An EPT misconfiguration is triggered if the PTE is writable * but not readable (WR=10). This is avoided by setting PG_A * and PG_M simultaneously. */ *pte |= PG_M | PG_A; } else { *pte |= PG_A; } /* try to promote the mapping */ if (va < VM_MAXUSER_ADDRESS) mpte = PHYS_TO_VM_PAGE(*pde & PG_FRAME); else mpte = NULL; m = PHYS_TO_VM_PAGE(*pte & PG_FRAME); if ((mpte == NULL || mpte->wire_count == NPTEPG) && pmap_ps_enabled(pmap) && (m->flags & PG_FICTITIOUS) == 0 && vm_reserv_level_iffullpop(m) == 0) { if (!pv_lists_locked) { pv_lists_locked = TRUE; if (!rw_try_rlock(&pvh_global_lock)) { PMAP_UNLOCK(pmap); rw_rlock(&pvh_global_lock); goto retry; } } pmap_promote_pde(pmap, pde, va, &lock); #ifdef INVARIANTS atomic_add_long(&ad_emulation_superpage_promotions, 1); #endif } #ifdef INVARIANTS if (ftype == VM_PROT_WRITE) atomic_add_long(&num_dirty_emulations, 1); else atomic_add_long(&num_accessed_emulations, 1); #endif rv = 0; /* success */ done: if (lock != NULL) rw_wunlock(lock); if (pv_lists_locked) rw_runlock(&pvh_global_lock); PMAP_UNLOCK(pmap); return (rv); } void pmap_get_mapping(pmap_t pmap, vm_offset_t va, uint64_t *ptr, int *num) { pml4_entry_t *pml4; pdp_entry_t *pdp; pd_entry_t *pde; pt_entry_t *pte, PG_V; int idx; idx = 0; PG_V = pmap_valid_bit(pmap); PMAP_LOCK(pmap); pml4 = pmap_pml4e(pmap, va); ptr[idx++] = *pml4; if ((*pml4 & PG_V) == 0) goto done; pdp = pmap_pml4e_to_pdpe(pml4, va); ptr[idx++] = *pdp; if ((*pdp & PG_V) == 0 || (*pdp & PG_PS) != 0) goto done; pde = pmap_pdpe_to_pde(pdp, va); ptr[idx++] = *pde; if ((*pde & PG_V) == 0 || (*pde & PG_PS) != 0) goto done; pte = pmap_pde_to_pte(pde, va); ptr[idx++] = *pte; done: PMAP_UNLOCK(pmap); *num = idx; } /** * Get the kernel virtual address of a set of physical pages. If there are * physical addresses not covered by the DMAP perform a transient mapping * that will be removed when calling pmap_unmap_io_transient. * * \param page The pages the caller wishes to obtain the virtual * address on the kernel memory map. * \param vaddr On return contains the kernel virtual memory address * of the pages passed in the page parameter. * \param count Number of pages passed in. * \param can_fault TRUE if the thread using the mapped pages can take * page faults, FALSE otherwise. * * \returns TRUE if the caller must call pmap_unmap_io_transient when * finished or FALSE otherwise. * */ boolean_t pmap_map_io_transient(vm_page_t page[], vm_offset_t vaddr[], int count, boolean_t can_fault) { vm_paddr_t paddr; boolean_t needs_mapping; pt_entry_t *pte; int cache_bits, error, i; /* * Allocate any KVA space that we need, this is done in a separate * loop to prevent calling vmem_alloc while pinned. */ needs_mapping = FALSE; for (i = 0; i < count; i++) { paddr = VM_PAGE_TO_PHYS(page[i]); if (__predict_false(paddr >= dmaplimit)) { error = vmem_alloc(kernel_arena, PAGE_SIZE, M_BESTFIT | M_WAITOK, &vaddr[i]); KASSERT(error == 0, ("vmem_alloc failed: %d", error)); needs_mapping = TRUE; } else { vaddr[i] = PHYS_TO_DMAP(paddr); } } /* Exit early if everything is covered by the DMAP */ if (!needs_mapping) return (FALSE); /* * NB: The sequence of updating a page table followed by accesses * to the corresponding pages used in the !DMAP case is subject to * the situation described in the "AMD64 Architecture Programmer's * Manual Volume 2: System Programming" rev. 3.23, "7.3.1 Special * Coherency Considerations". Therefore, issuing the INVLPG right * after modifying the PTE bits is crucial. */ if (!can_fault) sched_pin(); for (i = 0; i < count; i++) { paddr = VM_PAGE_TO_PHYS(page[i]); if (paddr >= dmaplimit) { if (can_fault) { /* * Slow path, since we can get page faults * while mappings are active don't pin the * thread to the CPU and instead add a global * mapping visible to all CPUs. */ pmap_qenter(vaddr[i], &page[i], 1); } else { pte = vtopte(vaddr[i]); cache_bits = pmap_cache_bits(kernel_pmap, page[i]->md.pat_mode, 0); pte_store(pte, paddr | X86_PG_RW | X86_PG_V | cache_bits); invlpg(vaddr[i]); } } } return (needs_mapping); } void pmap_unmap_io_transient(vm_page_t page[], vm_offset_t vaddr[], int count, boolean_t can_fault) { vm_paddr_t paddr; int i; if (!can_fault) sched_unpin(); for (i = 0; i < count; i++) { paddr = VM_PAGE_TO_PHYS(page[i]); if (paddr >= dmaplimit) { if (can_fault) pmap_qremove(vaddr[i], 1); vmem_free(kernel_arena, vaddr[i], PAGE_SIZE); } } } vm_offset_t pmap_quick_enter_page(vm_page_t m) { vm_paddr_t paddr; paddr = VM_PAGE_TO_PHYS(m); if (paddr < dmaplimit) return (PHYS_TO_DMAP(paddr)); mtx_lock_spin(&qframe_mtx); KASSERT(*vtopte(qframe) == 0, ("qframe busy")); pte_store(vtopte(qframe), paddr | X86_PG_RW | X86_PG_V | X86_PG_A | X86_PG_M | pmap_cache_bits(kernel_pmap, m->md.pat_mode, 0)); return (qframe); } void pmap_quick_remove_page(vm_offset_t addr) { if (addr != qframe) return; pte_store(vtopte(qframe), 0); invlpg(qframe); mtx_unlock_spin(&qframe_mtx); } #include "opt_ddb.h" #ifdef DDB #include DB_SHOW_COMMAND(pte, pmap_print_pte) { pmap_t pmap; pml4_entry_t *pml4; pdp_entry_t *pdp; pd_entry_t *pde; pt_entry_t *pte, PG_V; vm_offset_t va; if (have_addr) { va = (vm_offset_t)addr; pmap = PCPU_GET(curpmap); /* XXX */ } else { db_printf("show pte addr\n"); return; } PG_V = pmap_valid_bit(pmap); pml4 = pmap_pml4e(pmap, va); db_printf("VA %#016lx pml4e %#016lx", va, *pml4); if ((*pml4 & PG_V) == 0) { db_printf("\n"); return; } pdp = pmap_pml4e_to_pdpe(pml4, va); db_printf(" pdpe %#016lx", *pdp); if ((*pdp & PG_V) == 0 || (*pdp & PG_PS) != 0) { db_printf("\n"); return; } pde = pmap_pdpe_to_pde(pdp, va); db_printf(" pde %#016lx", *pde); if ((*pde & PG_V) == 0 || (*pde & PG_PS) != 0) { db_printf("\n"); return; } pte = pmap_pde_to_pte(pde, va); db_printf(" pte %#016lx\n", *pte); } DB_SHOW_COMMAND(phys2dmap, pmap_phys2dmap) { vm_paddr_t a; if (have_addr) { a = (vm_paddr_t)addr; db_printf("0x%jx\n", (uintmax_t)PHYS_TO_DMAP(a)); } else { db_printf("show phys2dmap addr\n"); } } #endif Index: head/sys/amd64/linux/linux_sysvec.c =================================================================== --- head/sys/amd64/linux/linux_sysvec.c (revision 298481) +++ head/sys/amd64/linux/linux_sysvec.c (revision 298482) @@ -1,995 +1,995 @@ /*- * Copyright (c) 2013 Dmitry Chagin * Copyright (c) 2004 Tim J. Robbins * Copyright (c) 2003 Peter Wemm * Copyright (c) 2002 Doug Rabson * Copyright (c) 1998-1999 Andrew Gallatin * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #define __ELF_WORD_SIZE 64 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MODULE_VERSION(linux64, 1); #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif #if defined(DEBUG) SYSCTL_PROC(_compat_linux, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0, linux_sysctl_debug, "A", "Linux 64 debugging control"); #endif /* * Allow the this functions to use the ldebug() facility * even though they are not syscalls themselves. Map them * to syscall 0. This is slightly less bogus than using * ldebug(sigreturn). */ #define LINUX_SYS_linux_rt_sendsig 0 const char *linux_kplatform; static int linux_szsigcode; static vm_object_t linux_shared_page_obj; static char *linux_shared_page_mapping; extern char _binary_linux_locore_o_start; extern char _binary_linux_locore_o_end; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static register_t * linux_copyout_strings(struct image_params *imgp); static int elf_linux_fixup(register_t **stack_base, struct image_params *iparams); static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel); static void linux_vdso_install(void *param); static void linux_vdso_deinstall(void *param); static void linux_set_syscall_retval(struct thread *td, int error); static int linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa); static void linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack); static int linux_vsyscall(struct thread *td); /* * Linux syscalls return negative errno's, we do positive and map them * Reference: * FreeBSD: src/sys/sys/errno.h * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h * linux-2.6.17.8/include/asm-generic/errno.h */ static int bsd_to_linux_errno[ELAST + 1] = { -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, -6, -6, -43, -42, -75,-125, -84, -95, -16, -74, -72, -67, -71 }; #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)td_proc; frame = td->td_frame; sa->args[0] = frame->tf_rdi; sa->args[1] = frame->tf_rsi; sa->args[2] = frame->tf_rdx; sa->args[3] = frame->tf_rcx; sa->args[4] = frame->tf_r8; sa->args[5] = frame->tf_r9; sa->code = frame->tf_rax; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; sa->narg = sa->callp->sy_narg; td->td_retval[0] = 0; return (0); } static void linux_set_syscall_retval(struct thread *td, int error) { struct trapframe *frame = td->td_frame; /* * On Linux only %rcx and %r11 values are not preserved across * the syscall. * So, do not clobber %rdx and %r10 */ td->td_retval[1] = frame->tf_rdx; frame->tf_r10 = frame->tf_rcx; cpu_set_syscall_retval(td, error); /* Restore all registers. */ set_pcb_flags(td->td_pcb, PCB_FULL_IRET); } static int elf_linux_fixup(register_t **stack_base, struct image_params *imgp) { Elf_Auxargs *args; Elf_Addr *base; Elf_Addr *pos; struct ps_strings *arginfo; struct proc *p; int issetugid; p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; KASSERT(curthread->td_proc == imgp->proc, ("unsafe elf_linux_fixup(), should be curproc")); base = (Elf64_Addr *)*stack_base; args = (Elf64_Auxargs *)imgp->auxargs; pos = base + (imgp->args->argc + imgp->args->envc + 2); issetugid = p->p_flag & P_SUGID ? 1 : 0; AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, imgp->proc->p_sysent->sv_shared_page_base); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary); if (imgp->execpathp != 0) AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; base--; suword(base, (uint64_t)imgp->args->argc); *stack_base = (register_t *)base; return (0); } /* * Copy strings out to the new process address space, constructing new arg * and env vector tables. Return a pointer to the base so that it can be used * as the initial stack pointer. */ static register_t * linux_copyout_strings(struct image_params *imgp) { int argc, envc; char **vectp; char *stringp, *destp; register_t *stack_base; struct ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; struct proc *p; /* * Calculate string base and vector table pointers. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; p = imgp->proc; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (caddr_t)arginfo - SPARE_USRSPACE - roundup(sizeof(canary), sizeof(char *)) - roundup(execpath_len, sizeof(char *)) - - roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); + roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); if (execpath_len != 0) { imgp->execpathp = (uintptr_t)arginfo - execpath_len; copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); } /* * Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); imgp->canary = (uintptr_t)arginfo - roundup(execpath_len, sizeof(char *)) - roundup(sizeof(canary), sizeof(char *)); copyout(canary, (void *)imgp->canary, sizeof(canary)); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (LINUX_AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *)); } else { /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(char *)); } /* * vectp also becomes our initial stack base */ stack_base = (register_t *)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword(vectp++, 0); suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword(vectp, 0); return (stack_base); } /* * Reset registers to default values on exec. */ static void linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack) { struct trapframe *regs = td->td_frame; struct pcb *pcb = td->td_pcb; mtx_lock(&dt_lock); if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); else mtx_unlock(&dt_lock); pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; clear_pcb_flags(pcb, PCB_32BIT); pcb->pcb_initial_fpucw = __LINUX_NPXCW__; set_pcb_flags(pcb, PCB_FULL_IRET); bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = imgp->entry_addr; regs->tf_rsp = stack; regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); regs->tf_ss = _udatasel; regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; /* * Reset the hardware debug registers if they were in use. * They won't have any meaning for the newly exec'd process. */ if (pcb->pcb_flags & PCB_DBREGS) { pcb->pcb_dr0 = 0; pcb->pcb_dr1 = 0; pcb->pcb_dr2 = 0; pcb->pcb_dr3 = 0; pcb->pcb_dr6 = 0; pcb->pcb_dr7 = 0; if (pcb == curpcb) { /* * Clear the debug registers on the running * CPU, otherwise they will end up affecting * the next process we switch to. */ reset_dbregs(); } clear_pcb_flags(pcb, PCB_DBREGS); } /* * Drop the FP state if we hold it, so that the process gets a * clean FP state if it uses the FPU again. */ fpstate_drop(td); } /* * Copied from amd64/amd64/machdep.c * * XXX fpu state need? don't think so */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct proc *p; struct l_ucontext uc; struct l_sigcontext *context; struct trapframe *regs; unsigned long rflags; int error; ksiginfo_t ksi; regs = td->td_frame; error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc)); if (error != 0) return (error); p = td->td_proc; context = &uc.uc_mcontext; rflags = context->sc_rflags; /* * Don't allow users to change privileged or reserved flags. */ /* * XXX do allow users to change the privileged flag PSL_RF. * The cpu sets PSL_RF in tf_rflags for faults. Debuggers * should sometimes set it there too. tf_rflags is kept in * the signal context during signal handling and there is no * other place to remember it, so the PSL_RF bit may be * corrupted by the signal handler without us knowing. * Corruption of the PSL_RF bit at worst causes one more or * one less debugger trap, so allowing it is fairly harmless. */ #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) { printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags); return (EINVAL); } /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(context->sc_cs)) { printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs); ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return (EINVAL); } PROC_LOCK(p); linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask); SIG_CANTMASK(td->td_sigmask); signotify(td); PROC_UNLOCK(p); regs->tf_rdi = context->sc_rdi; regs->tf_rsi = context->sc_rsi; regs->tf_rdx = context->sc_rdx; regs->tf_rbp = context->sc_rbp; regs->tf_rbx = context->sc_rbx; regs->tf_rcx = context->sc_rcx; regs->tf_rax = context->sc_rax; regs->tf_rip = context->sc_rip; regs->tf_rsp = context->sc_rsp; regs->tf_r8 = context->sc_r8; regs->tf_r9 = context->sc_r9; regs->tf_r10 = context->sc_r10; regs->tf_r11 = context->sc_r11; regs->tf_r12 = context->sc_r12; regs->tf_r13 = context->sc_r13; regs->tf_r14 = context->sc_r14; regs->tf_r15 = context->sc_r15; regs->tf_cs = context->sc_cs; regs->tf_err = context->sc_err; regs->tf_rflags = rflags; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * copied from amd64/amd64/machdep.c * * Send an interrupt to process. */ static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct l_rt_sigframe sf, *sfp; struct proc *p; struct thread *td; struct sigacts *psp; caddr_t sp; struct trapframe *regs; int sig, code; int oonstack; td = curthread; p = td->td_proc; PROC_LOCK_ASSERT(p, MA_OWNED); sig = ksi->ksi_signo; psp = p->p_sigacts; code = ksi->ksi_code; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u", catcher, sig, mask, code); /* Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe); } else sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128; /* Align to 16 bytes. */ sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul); mtx_unlock(&psp->ps_mtx); /* Translate the signal. */ sig = bsd_to_linux_signal(sig); /* Save user context. */ bzero(&sf, sizeof(sf)); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask); bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask); sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; PROC_UNLOCK(p); sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi; sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi; sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx; sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp; sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx; sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx; sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax; sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip; sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp; sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8; sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9; sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10; sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11; sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12; sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13; sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14; sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15; sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags; sf.sf_sc.uc_mcontext.sc_err = regs->tf_err; sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; /* Build the argument list for the signal handler. */ regs->tf_rdi = sig; /* arg 1 in %rdi */ regs->tf_rax = 0; regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */ regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */ sf.sf_handler = catcher; /* Fill in POSIX parts */ ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig); /* * Copy the sigframe out to the user's stack. */ if (copyout(&sf, sfp, sizeof(*sfp)) != 0) { #ifdef DEBUG printf("process %ld has trashed its stack\n", (long)p->p_pid); #endif PROC_LOCK(p); sigexit(td, SIGILL); } regs->tf_rsp = (long)sfp; regs->tf_rip = linux_rt_sigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucodesel; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * If a linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a linux * binary is doing the exec, so we do not create an EXEC module for it. */ static int exec_linux_imgact_try(struct image_params *iparams); static int exec_linux_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1, len; /* * The interpreter for shell scripts run from a linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds * attempt to use the alternate path for the interpreter. * If an alternate path is found, use our stringspace * to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD); if (rpath != NULL) { len = strlen(rpath) + 1; if (len <= MAXSHELLCMDLEN) memcpy(imgp->interpreter_name, rpath, len); free(rpath, M_TEMP); } } } return(error); } #define LINUX_VSYSCALL_START (-10UL << 20) #define LINUX_VSYSCALL_SZ 1024 const unsigned long linux_vsyscall_vector[] = { LINUX_SYS_gettimeofday, LINUX_SYS_linux_time, /* getcpu not implemented */ }; static int linux_vsyscall(struct thread *td) { struct trapframe *frame; uint64_t retqaddr; int code, traced; int error; frame = td->td_frame; /* Check %rip for vsyscall area */ if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START)) return (EINVAL); if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0) return (EINVAL); code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ; if (code >= nitems(linux_vsyscall_vector)) return (EINVAL); /* * vsyscall called as callq *(%rax), so we must * use return address from %rsp and also fixup %rsp */ error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr)); if (error) return (error); frame->tf_rip = retqaddr; frame->tf_rax = linux_vsyscall_vector[code]; frame->tf_rsp += 8; traced = (frame->tf_flags & PSL_T); amd64_syscall(td, traced); return (0); } struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = elf_linux_fixup, .sv_sendsig = linux_rt_sendsig, .sv_sigcode = &_binary_linux_locore_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF64", .sv_coredump = elf64_coredump, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = linux_exec_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP, .sv_set_syscall_retval = linux_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .sv_shared_page_base = SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = linux_vsyscall, }; static void linux_vdso_install(void *param) { linux_szsigcode = (&_binary_linux_locore_o_end - &_binary_linux_locore_o_start); if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) panic("Linux invalid vdso size\n"); __elfN(linux_vdso_fixup)(&elf_linux_sysvec); linux_shared_page_obj = __elfN(linux_shared_page_init) (&linux_shared_page_mapping); __elfN(linux_vdso_reloc)(&elf_linux_sysvec, SHAREDPAGE); bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, linux_szsigcode); elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; linux_kplatform = linux_shared_page_mapping + (linux_platform - (caddr_t)SHAREDPAGE); } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, (sysinit_cfunc_t)linux_vdso_install, NULL); static void linux_vdso_deinstall(void *param) { __elfN(linux_shared_page_fini)(linux_shared_page_obj); }; SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)linux_vdso_deinstall, NULL); static char GNULINUX_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (FALSE); /* * For linux we encode osrel as follows (see linux_mib.c): * VVVMMMIII (version, major, minor), see linux_mib.c. */ *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; return (TRUE); } static Elf_Brandnote linux64_brandnote = { .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR), .hdr.n_descsz = 16, .hdr.n_type = 1, .vendor = GNULINUX_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux_trans_osrel }; static Elf64_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib64/ld-linux-x86-64.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf64_Brandinfo linux_glibc2brandshort = { .brand = ELFOSABI_LINUX, .machine = EM_X86_64, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib64/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux64_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; Elf64_Brandinfo *linux_brandlist[] = { &linux_glibc2brand, &linux_glibc2brandshort, NULL }; static int linux64_elf_modevent(module_t mod, int type, void *data) { Elf64_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; error = 0; switch(type) { case MOD_LOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_insert_brand_entry(*brandinfo) < 0) error = EINVAL; if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_register_handler(*lihp); LIST_INIT(&futex_list); mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux x86-64 ELF exec handler installed\n"); } else printf("cannot insert Linux x86-64 ELF brand handler\n"); break; case MOD_UNLOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_brand_inuse(*brandinfo)) error = EBUSY; if (error == 0) { for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf64_remove_brand_entry(*brandinfo) < 0) error = EINVAL; } if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_unregister_handler(*lihp); mtx_destroy(&futex_mtx); if (bootverbose) printf("Linux ELF exec handler removed\n"); } else printf("Could not deinstall ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux64_elf_mod = { "linux64elf", linux64_elf_modevent, 0 }; DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1); FEATURE(linux64, "Linux 64bit support"); Index: head/sys/amd64/linux32/linux32_sysvec.c =================================================================== --- head/sys/amd64/linux32/linux32_sysvec.c (revision 298481) +++ head/sys/amd64/linux32/linux32_sysvec.c (revision 298482) @@ -1,1208 +1,1208 @@ /*- * Copyright (c) 2004 Tim J. Robbins * Copyright (c) 2003 Peter Wemm * Copyright (c) 2002 Doug Rabson * Copyright (c) 1998-1999 Andrew Gallatin * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #ifndef COMPAT_FREEBSD32 #error "Unable to compile Linux-emulator due to missing COMPAT_FREEBSD32 option!" #endif #define __ELF_WORD_SIZE 32 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MODULE_VERSION(linux, 1); #define AUXARGS_ENTRY_32(pos, id, val) \ do { \ suword32(pos++, id); \ suword32(pos++, val); \ } while (0) #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif /* * Allow the sendsig functions to use the ldebug() facility * even though they are not syscalls themselves. Map them * to syscall 0. This is slightly less bogus than using * ldebug(sigreturn). */ #define LINUX32_SYS_linux_rt_sendsig 0 #define LINUX32_SYS_linux_sendsig 0 const char *linux_kplatform; static int linux_szsigcode; static vm_object_t linux_shared_page_obj; static char *linux_shared_page_mapping; extern char _binary_linux32_locore_o_start; extern char _binary_linux32_locore_o_end; extern struct sysent linux32_sysent[LINUX32_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int elf_linux_fixup(register_t **stack_base, struct image_params *iparams); static register_t *linux_copyout_strings(struct image_params *imgp); static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); static void exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack); static void linux32_fixlimit(struct rlimit *rl, int which); static boolean_t linux32_trans_osrel(const Elf_Note *note, int32_t *osrel); static void linux_vdso_install(void *param); static void linux_vdso_deinstall(void *param); /* * Linux syscalls return negative errno's, we do positive and map them * Reference: * FreeBSD: src/sys/sys/errno.h * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h * linux-2.6.17.8/include/asm-generic/errno.h */ static int bsd_to_linux_errno[ELAST + 1] = { -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, -6, -6, -43, -42, -75,-125, -84, -95, -16, -74, -72, -67, -71 }; #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)td_proc == imgp->proc, ("unsafe elf_linux_fixup(), should be curproc")); base = (Elf32_Addr *)*stack_base; args = (Elf32_Auxargs *)imgp->auxargs; pos = base + (imgp->args->argc + imgp->args->envc + 2); issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO_EHDR, imgp->proc->p_sysent->sv_shared_page_base); AUXARGS_ENTRY_32(pos, LINUX_AT_SYSINFO, linux32_vsyscall); AUXARGS_ENTRY_32(pos, LINUX_AT_HWCAP, cpu_feature); /* * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, * as it has appeared in the 2.4.0-rc7 first time. * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), * glibc falls back to the hard-coded CLK_TCK value when aux entry * is not present. * Also see linux_times() implementation. */ if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) AUXARGS_ENTRY_32(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY_32(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY_32(pos, AT_PHENT, args->phent); AUXARGS_ENTRY_32(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY_32(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY_32(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY_32(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY_32(pos, AT_BASE, args->base); AUXARGS_ENTRY_32(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY_32(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY_32(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY_32(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY_32(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY_32(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform)); AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, PTROUT(imgp->canary)); if (imgp->execpathp != 0) AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, PTROUT(imgp->execpathp)); if (args->execfd != -1) AUXARGS_ENTRY_32(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY_32(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; base--; suword32(base, (uint32_t)imgp->args->argc); *stack_base = (register_t *)base; return (0); } static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_rt_sigframe *fp, frame; int oonstack; int sig; int code; sig = ksi->ksi_signo; code = ksi->ksi_code; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); } else fp = (struct l_rt_sigframe *)regs->tf_rsp - 1; mtx_unlock(&psp->ps_mtx); /* * Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_handler = PTROUT(catcher); frame.sf_sig = sig; frame.sf_siginfo = PTROUT(&fp->sf_si); frame.sf_ucontext = PTROUT(&fp->sf_sc); /* Fill in POSIX parts */ ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); /* * Build the signal context to be used by sigreturn * and libgcc unwind. */ frame.sf_sc.uc_flags = 0; /* XXX ??? */ frame.sf_sc.uc_link = 0; /* XXX ??? */ frame.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp); frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; PROC_UNLOCK(p); bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; frame.sf_sc.uc_mcontext.sc_edi = regs->tf_rdi; frame.sf_sc.uc_mcontext.sc_esi = regs->tf_rsi; frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_rbp; frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_rbx; frame.sf_sc.uc_mcontext.sc_esp = regs->tf_rsp; frame.sf_sc.uc_mcontext.sc_edx = regs->tf_rdx; frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_rcx; frame.sf_sc.uc_mcontext.sc_eax = regs->tf_rax; frame.sf_sc.uc_mcontext.sc_eip = regs->tf_rip; frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; frame.sf_sc.uc_mcontext.sc_gs = regs->tf_gs; frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_rflags; frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_rsp; frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; frame.sf_sc.uc_mcontext.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); #endif if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), fp, oonstack); #endif PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = linux32_rt_sigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * Send an interrupt to process. * * Stack is set up to allow sigcode stored * in u. to call routine, followed by kcall * to sigreturn routine below. After sigreturn * resets the signal mask, the stack, and the * frame pointer, it returns to the user * specified pc, psl. */ static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_sigframe *fp, frame; l_sigset_t lmask; int oonstack; int sig, code; sig = ksi->ksi_signo; code = ksi->ksi_code; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ linux_rt_sendsig(catcher, ksi, mask); return; } regs = td->td_frame; oonstack = sigonstack(regs->tf_rsp); #ifdef DEBUG if (ldebug(sendsig)) printf(ARGS(sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_sigframe)); } else fp = (struct l_sigframe *)regs->tf_rsp - 1; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); /* * Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_handler = PTROUT(catcher); frame.sf_sig = sig; bsd_to_linux_sigset(mask, &lmask); /* * Build the signal context to be used by sigreturn. */ frame.sf_sc.sc_mask = lmask.__mask; frame.sf_sc.sc_gs = regs->tf_gs; frame.sf_sc.sc_fs = regs->tf_fs; frame.sf_sc.sc_es = regs->tf_es; frame.sf_sc.sc_ds = regs->tf_ds; frame.sf_sc.sc_edi = regs->tf_rdi; frame.sf_sc.sc_esi = regs->tf_rsi; frame.sf_sc.sc_ebp = regs->tf_rbp; frame.sf_sc.sc_ebx = regs->tf_rbx; frame.sf_sc.sc_esp = regs->tf_rsp; frame.sf_sc.sc_edx = regs->tf_rdx; frame.sf_sc.sc_ecx = regs->tf_rcx; frame.sf_sc.sc_eax = regs->tf_rax; frame.sf_sc.sc_eip = regs->tf_rip; frame.sf_sc.sc_cs = regs->tf_cs; frame.sf_sc.sc_eflags = regs->tf_rflags; frame.sf_sc.sc_esp_at_signal = regs->tf_rsp; frame.sf_sc.sc_ss = regs->tf_ss; frame.sf_sc.sc_err = regs->tf_err; frame.sf_sc.sc_cr2 = (u_int32_t)(uintptr_t)ksi->ksi_addr; frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(code); frame.sf_extramask[0] = lmask.__mask; if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_rsp = PTROUT(fp); regs->tf_rip = linux32_sigcode; regs->tf_rflags &= ~(PSL_T | PSL_D); regs->tf_cs = _ucode32sel; regs->tf_ss = _udatasel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _ufssel; regs->tf_gs = _ugssel; regs->tf_flags = TF_HASSEGS; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * psl to gain improper privileges or to cause * a machine fault. */ int linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) { struct l_sigframe frame; struct trapframe *regs; sigset_t bmask; l_sigset_t lmask; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(sigreturn)) printf(ARGS(sigreturn, "%p"), (void *)args->sfp); #endif /* * The trampoline code hands us the sigframe. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->sfp, &frame, sizeof(frame)) != 0) return (EFAULT); /* * Check for security violations. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = frame.sf_sc.sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) return(EINVAL); /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(frame.sf_sc.sc_cs)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return(EINVAL); } lmask.__mask = frame.sf_sc.sc_mask; lmask.__mask = frame.sf_extramask[0]; linux_to_bsd_sigset(&lmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* * Restore signal context. */ regs->tf_rdi = frame.sf_sc.sc_edi; regs->tf_rsi = frame.sf_sc.sc_esi; regs->tf_rbp = frame.sf_sc.sc_ebp; regs->tf_rbx = frame.sf_sc.sc_ebx; regs->tf_rdx = frame.sf_sc.sc_edx; regs->tf_rcx = frame.sf_sc.sc_ecx; regs->tf_rax = frame.sf_sc.sc_eax; regs->tf_rip = frame.sf_sc.sc_eip; regs->tf_cs = frame.sf_sc.sc_cs; regs->tf_ds = frame.sf_sc.sc_ds; regs->tf_es = frame.sf_sc.sc_es; regs->tf_fs = frame.sf_sc.sc_fs; regs->tf_gs = frame.sf_sc.sc_gs; regs->tf_rflags = eflags; regs->tf_rsp = frame.sf_sc.sc_esp_at_signal; regs->tf_ss = frame.sf_sc.sc_ss; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); return (EJUSTRETURN); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by rt_sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * psl to gain improper privileges or to cause * a machine fault. */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct l_ucontext uc; struct l_sigcontext *context; sigset_t bmask; l_stack_t *lss; stack_t ss; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); #endif /* * The trampoline code hands us the ucontext. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->ucp, &uc, sizeof(uc)) != 0) return (EFAULT); context = &uc.uc_mcontext; /* * Check for security violations. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = context->sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_rflags)) return(EINVAL); /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(context->sc_cs)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_rip; trapsignal(td, &ksi); return(EINVAL); } linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* * Restore signal context */ regs->tf_gs = context->sc_gs; regs->tf_fs = context->sc_fs; regs->tf_es = context->sc_es; regs->tf_ds = context->sc_ds; regs->tf_rdi = context->sc_edi; regs->tf_rsi = context->sc_esi; regs->tf_rbp = context->sc_ebp; regs->tf_rbx = context->sc_ebx; regs->tf_rdx = context->sc_edx; regs->tf_rcx = context->sc_ecx; regs->tf_rax = context->sc_eax; regs->tf_rip = context->sc_eip; regs->tf_cs = context->sc_cs; regs->tf_rflags = eflags; regs->tf_rsp = context->sc_esp_at_signal; regs->tf_ss = context->sc_ss; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); /* * call sigaltstack & ignore results.. */ lss = &uc.uc_stack; ss.ss_sp = PTRIN(lss->ss_sp); ss.ss_size = lss->ss_size; ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%lx, mask: 0x%x"), ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); #endif (void)kern_sigaltstack(td, &ss, NULL); return (EJUSTRETURN); } static int linux32_fetch_syscall_args(struct thread *td, struct syscall_args *sa) { struct proc *p; struct trapframe *frame; p = td->td_proc; frame = td->td_frame; sa->args[0] = frame->tf_rbx; sa->args[1] = frame->tf_rcx; sa->args[2] = frame->tf_rdx; sa->args[3] = frame->tf_rsi; sa->args[4] = frame->tf_rdi; sa->args[5] = frame->tf_rbp; /* Unconfirmed */ sa->code = frame->tf_rax; if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; sa->narg = sa->callp->sy_narg; td->td_retval[0] = 0; td->td_retval[1] = frame->tf_rdx; return (0); } /* * If a linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a linux * binary is doing the exec, so we do not create an EXEC module for it. */ static int exec_linux_imgact_try(struct image_params *iparams); static int exec_linux_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1; /* * The interpreter for shell scripts run from a linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds attempt * to use the alternate path for the interpreter. If an * alternate * path is found, use our stringspace to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD); if (rpath != NULL) imgp->args->fname_buf = imgp->interpreter_name = rpath; } } return (error); } /* * Clear registers on exec * XXX copied from ia32_signal.c. */ static void exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack) { struct trapframe *regs = td->td_frame; struct pcb *pcb = td->td_pcb; mtx_lock(&dt_lock); if (td->td_proc->p_md.md_ldt != NULL) user_ldt_free(td); else mtx_unlock(&dt_lock); critical_enter(); wrmsr(MSR_FSBASE, 0); wrmsr(MSR_KGSBASE, 0); /* User value while we're in the kernel */ pcb->pcb_fsbase = 0; pcb->pcb_gsbase = 0; critical_exit(); pcb->pcb_initial_fpucw = __LINUX_NPXCW__; bzero((char *)regs, sizeof(struct trapframe)); regs->tf_rip = imgp->entry_addr; regs->tf_rsp = stack; regs->tf_rflags = PSL_USER | (regs->tf_rflags & PSL_T); regs->tf_gs = _ugssel; regs->tf_fs = _ufssel; regs->tf_es = _udatasel; regs->tf_ds = _udatasel; regs->tf_ss = _udatasel; regs->tf_flags = TF_HASSEGS; regs->tf_cs = _ucode32sel; regs->tf_rbx = imgp->ps_strings; fpstate_drop(td); /* Do full restore on return so that we can change to a different %cs */ set_pcb_flags(pcb, PCB_32BIT | PCB_FULL_IRET); td->td_retval[1] = 0; } /* * XXX copied from ia32_sysvec.c. */ static register_t * linux_copyout_strings(struct image_params *imgp) { int argc, envc; u_int32_t *vectp; char *stringp, *destp; u_int32_t *stack_base; struct linux32_ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; /* * Calculate string base and vector table pointers. */ if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; arginfo = (struct linux32_ps_strings *)LINUX32_PS_STRINGS; destp = (caddr_t)arginfo - SPARE_USRSPACE - roundup(sizeof(canary), sizeof(char *)) - roundup(execpath_len, sizeof(char *)) - - roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); + roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); if (execpath_len != 0) { imgp->execpathp = (uintptr_t)arginfo - execpath_len; copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); } /* * Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); imgp->canary = (uintptr_t)arginfo - roundup(execpath_len, sizeof(char *)) - roundup(sizeof(canary), sizeof(char *)); copyout(canary, (void *)imgp->canary, sizeof(canary)); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (LINUX_AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (u_int32_t *) (destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(u_int32_t)); } else /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (u_int32_t *)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(u_int32_t)); /* * vectp also becomes our initial stack base */ stack_base = vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword32(&arginfo->ps_argvstr, (uint32_t)(intptr_t)vectp); suword32(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword32(vectp++, (uint32_t)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword32(vectp++, 0); suword32(&arginfo->ps_envstr, (uint32_t)(intptr_t)vectp); suword32(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword32(vectp++, (uint32_t)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword32(vectp, 0); return ((register_t *)stack_base); } static SYSCTL_NODE(_compat, OID_AUTO, linux32, CTLFLAG_RW, 0, "32-bit Linux emulation"); static u_long linux32_maxdsiz = LINUX32_MAXDSIZ; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxdsiz, CTLFLAG_RW, &linux32_maxdsiz, 0, ""); static u_long linux32_maxssiz = LINUX32_MAXSSIZ; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxssiz, CTLFLAG_RW, &linux32_maxssiz, 0, ""); static u_long linux32_maxvmem = LINUX32_MAXVMEM; SYSCTL_ULONG(_compat_linux32, OID_AUTO, maxvmem, CTLFLAG_RW, &linux32_maxvmem, 0, ""); #if defined(DEBUG) SYSCTL_PROC(_compat_linux32, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0, linux_sysctl_debug, "A", "Linux debugging control"); #endif static void linux32_fixlimit(struct rlimit *rl, int which) { switch (which) { case RLIMIT_DATA: if (linux32_maxdsiz != 0) { if (rl->rlim_cur > linux32_maxdsiz) rl->rlim_cur = linux32_maxdsiz; if (rl->rlim_max > linux32_maxdsiz) rl->rlim_max = linux32_maxdsiz; } break; case RLIMIT_STACK: if (linux32_maxssiz != 0) { if (rl->rlim_cur > linux32_maxssiz) rl->rlim_cur = linux32_maxssiz; if (rl->rlim_max > linux32_maxssiz) rl->rlim_max = linux32_maxssiz; } break; case RLIMIT_VMEM: if (linux32_maxvmem != 0) { if (rl->rlim_cur > linux32_maxvmem) rl->rlim_cur = linux32_maxvmem; if (rl->rlim_max > linux32_maxvmem) rl->rlim_max = linux32_maxvmem; } break; } } struct sysentvec elf_linux_sysvec = { .sv_size = LINUX32_SYS_MAXSYSCALL, .sv_table = linux32_sysent, .sv_mask = 0, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = elf_linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux32_locore_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF32", .sv_coredump = elf32_coredump, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = LINUX32_MAXUSER, .sv_usrstack = LINUX32_USRSTACK, .sv_psstrings = LINUX32_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = linux32_fixlimit, .sv_maxssiz = &linux32_maxssiz, .sv_flags = SV_ABI_LINUX | SV_ILP32 | SV_IA32 | SV_SHP, .sv_set_syscall_retval = cpu_set_syscall_retval, .sv_fetch_syscall_args = linux32_fetch_syscall_args, .sv_syscallnames = NULL, .sv_shared_page_base = LINUX32_SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, }; static void linux_vdso_install(void *param) { linux_szsigcode = (&_binary_linux32_locore_o_end - &_binary_linux32_locore_o_start); if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) panic("Linux invalid vdso size\n"); __elfN(linux_vdso_fixup)(&elf_linux_sysvec); linux_shared_page_obj = __elfN(linux_shared_page_init) (&linux_shared_page_mapping); __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX32_SHAREDPAGE); bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, linux_szsigcode); elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; linux_kplatform = linux_shared_page_mapping + (linux_platform - (caddr_t)LINUX32_SHAREDPAGE); } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, (sysinit_cfunc_t)linux_vdso_install, NULL); static void linux_vdso_deinstall(void *param) { __elfN(linux_shared_page_fini)(linux_shared_page_obj); }; SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)linux_vdso_deinstall, NULL); static char GNU_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static boolean_t linux32_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (FALSE); /* * For linux we encode osrel as follows (see linux_mib.c): * VVVMMMIII (version, major, minor), see linux_mib.c. */ *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; return (TRUE); } static Elf_Brandnote linux32_brandnote = { .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), .hdr.n_descsz = 16, /* XXX at least 16 */ .hdr.n_type = 1, .vendor = GNU_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux32_trans_osrel }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux32_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; Elf32_Brandinfo *linux_brandlist[] = { &linux_brand, &linux_glibc2brand, NULL }; static int linux_elf_modevent(module_t mod, int type, void *data) { Elf32_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; error = 0; switch(type) { case MOD_LOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_insert_brand_entry(*brandinfo) < 0) error = EINVAL; if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_register_handler(*lihp); LIST_INIT(&futex_list); mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux ELF exec handler installed\n"); } else printf("cannot insert Linux ELF brand handler\n"); break; case MOD_UNLOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_brand_inuse(*brandinfo)) error = EBUSY; if (error == 0) { for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_remove_brand_entry(*brandinfo) < 0) error = EINVAL; } if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_unregister_handler(*lihp); mtx_destroy(&futex_mtx); if (bootverbose) printf("Linux ELF exec handler removed\n"); } else printf("Could not deinstall ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux_elf_mod = { "linuxelf", linux_elf_modevent, 0 }; DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); MODULE_DEPEND(linuxelf, linux_common, 1, 1, 1); FEATURE(linux, "Linux 32bit support"); Index: head/sys/compat/linux/linux_file.c =================================================================== --- head/sys/compat/linux/linux_file.c (revision 298481) +++ head/sys/compat/linux/linux_file.c (revision 298482) @@ -1,1652 +1,1651 @@ /*- * Copyright (c) 1994-1995 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef COMPAT_LINUX32 #include #include #else #include #include #endif #include #include #include int linux_creat(struct thread *td, struct linux_creat_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(creat)) printf(ARGS(creat, "%s, %d"), path, args->mode); #endif error = kern_openat(td, AT_FDCWD, path, UIO_SYSSPACE, O_WRONLY | O_CREAT | O_TRUNC, args->mode); LFREEPATH(path); return (error); } static int linux_common_open(struct thread *td, int dirfd, char *path, int l_flags, int mode) { cap_rights_t rights; struct proc *p = td->td_proc; struct file *fp; int fd; int bsd_flags, error; bsd_flags = 0; switch (l_flags & LINUX_O_ACCMODE) { case LINUX_O_WRONLY: bsd_flags |= O_WRONLY; break; case LINUX_O_RDWR: bsd_flags |= O_RDWR; break; default: bsd_flags |= O_RDONLY; } if (l_flags & LINUX_O_NDELAY) bsd_flags |= O_NONBLOCK; if (l_flags & LINUX_O_APPEND) bsd_flags |= O_APPEND; if (l_flags & LINUX_O_SYNC) bsd_flags |= O_FSYNC; if (l_flags & LINUX_O_NONBLOCK) bsd_flags |= O_NONBLOCK; if (l_flags & LINUX_FASYNC) bsd_flags |= O_ASYNC; if (l_flags & LINUX_O_CREAT) bsd_flags |= O_CREAT; if (l_flags & LINUX_O_TRUNC) bsd_flags |= O_TRUNC; if (l_flags & LINUX_O_EXCL) bsd_flags |= O_EXCL; if (l_flags & LINUX_O_NOCTTY) bsd_flags |= O_NOCTTY; if (l_flags & LINUX_O_DIRECT) bsd_flags |= O_DIRECT; if (l_flags & LINUX_O_NOFOLLOW) bsd_flags |= O_NOFOLLOW; if (l_flags & LINUX_O_DIRECTORY) bsd_flags |= O_DIRECTORY; /* XXX LINUX_O_NOATIME: unable to be easily implemented. */ error = kern_openat(td, dirfd, path, UIO_SYSSPACE, bsd_flags, mode); if (error != 0) goto done; if (bsd_flags & O_NOCTTY) goto done; /* * XXX In between kern_open() and fget(), another process * having the same filedesc could use that fd without * checking below. */ fd = td->td_retval[0]; if (fget(td, fd, cap_rights_init(&rights, CAP_IOCTL), &fp) == 0) { if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); goto done; } sx_slock(&proctree_lock); PROC_LOCK(p); if (SESS_LEADER(p) && !(p->p_flag & P_CONTROLT)) { PROC_UNLOCK(p); sx_sunlock(&proctree_lock); /* XXXPJD: Verify if TIOCSCTTY is allowed. */ (void) fo_ioctl(fp, TIOCSCTTY, (caddr_t) 0, td->td_ucred, td); } else { PROC_UNLOCK(p); sx_sunlock(&proctree_lock); } fdrop(fp, td); } done: #ifdef DEBUG if (ldebug(open)) printf(LMSG("open returns error %d"), error); #endif LFREEPATH(path); return (error); } int linux_openat(struct thread *td, struct linux_openat_args *args) { char *path; int dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; if (args->flags & LINUX_O_CREAT) LCONVPATH_AT(td, args->filename, &path, 1, dfd); else LCONVPATH_AT(td, args->filename, &path, 0, dfd); #ifdef DEBUG if (ldebug(openat)) printf(ARGS(openat, "%i, %s, 0x%x, 0x%x"), args->dfd, path, args->flags, args->mode); #endif return (linux_common_open(td, dfd, path, args->flags, args->mode)); } int linux_open(struct thread *td, struct linux_open_args *args) { char *path; if (args->flags & LINUX_O_CREAT) LCONVPATHCREAT(td, args->path, &path); else LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(open)) printf(ARGS(open, "%s, 0x%x, 0x%x"), path, args->flags, args->mode); #endif return (linux_common_open(td, AT_FDCWD, path, args->flags, args->mode)); } int linux_lseek(struct thread *td, struct linux_lseek_args *args) { struct lseek_args /* { int fd; int pad; off_t offset; int whence; } */ tmp_args; int error; #ifdef DEBUG if (ldebug(lseek)) printf(ARGS(lseek, "%d, %ld, %d"), args->fdes, (long)args->off, args->whence); #endif tmp_args.fd = args->fdes; tmp_args.offset = (off_t)args->off; tmp_args.whence = args->whence; error = sys_lseek(td, &tmp_args); return error; } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_llseek(struct thread *td, struct linux_llseek_args *args) { struct lseek_args bsd_args; int error; off_t off; #ifdef DEBUG if (ldebug(llseek)) printf(ARGS(llseek, "%d, %d:%d, %d"), args->fd, args->ohigh, args->olow, args->whence); #endif off = (args->olow) | (((off_t) args->ohigh) << 32); bsd_args.fd = args->fd; bsd_args.offset = off; bsd_args.whence = args->whence; if ((error = sys_lseek(td, &bsd_args))) return error; if ((error = copyout(td->td_retval, args->res, sizeof (off_t)))) return error; td->td_retval[0] = 0; return 0; } int linux_readdir(struct thread *td, struct linux_readdir_args *args) { struct linux_getdents_args lda; lda.fd = args->fd; lda.dent = args->dent; lda.count = 1; return linux_getdents(td, &lda); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ /* * Note that linux_getdents(2) and linux_getdents64(2) have the same * arguments. They only differ in the definition of struct dirent they * operate on. We use this to common the code, with the exception of * accessing struct dirent. Note that linux_readdir(2) is implemented * by means of linux_getdents(2). In this case we never operate on * struct dirent64 and thus don't need to handle it... */ struct l_dirent { l_ulong d_ino; l_off_t d_off; l_ushort d_reclen; char d_name[LINUX_NAME_MAX + 1]; }; struct l_dirent64 { uint64_t d_ino; int64_t d_off; l_ushort d_reclen; u_char d_type; char d_name[LINUX_NAME_MAX + 1]; }; /* * Linux uses the last byte in the dirent buffer to store d_type, * at least glibc-2.7 requires it. That is why l_dirent is padded with 2 bytes. */ #define LINUX_RECLEN(namlen) \ - roundup((offsetof(struct l_dirent, d_name) + (namlen) + 2), \ - sizeof(l_ulong)) + roundup(offsetof(struct l_dirent, d_name) + (namlen) + 2, sizeof(l_ulong)) #define LINUX_RECLEN64(namlen) \ - roundup((offsetof(struct l_dirent64, d_name) + (namlen) + 1), \ + roundup(offsetof(struct l_dirent64, d_name) + (namlen) + 1, \ sizeof(uint64_t)) #define LINUX_MAXRECLEN max(LINUX_RECLEN(LINUX_NAME_MAX), \ LINUX_RECLEN64(LINUX_NAME_MAX)) #define LINUX_DIRBLKSIZ 512 static int getdents_common(struct thread *td, struct linux_getdents64_args *args, int is64bit) { struct dirent *bdp; struct vnode *vp; caddr_t inp, buf; /* BSD-format */ int len, reclen; /* BSD-format */ caddr_t outp; /* Linux-format */ int resid, linuxreclen=0; /* Linux-format */ caddr_t lbuf; /* Linux-format */ cap_rights_t rights; struct file *fp; struct uio auio; struct iovec aiov; off_t off; struct l_dirent *linux_dirent; struct l_dirent64 *linux_dirent64; int buflen, error, eofflag, nbytes, justone; u_long *cookies = NULL, *cookiep; int ncookies; nbytes = args->count; if (nbytes == 1) { /* readdir(2) case. Always struct dirent. */ if (is64bit) return (EINVAL); nbytes = sizeof(*linux_dirent); justone = 1; } else justone = 0; error = getvnode(td, args->fd, cap_rights_init(&rights, CAP_READ), &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } off = foffset_lock(fp, 0); vp = fp->f_vnode; if (vp->v_type != VDIR) { foffset_unlock(fp, off, 0); fdrop(fp, td); return (EINVAL); } buflen = max(LINUX_DIRBLKSIZ, nbytes); buflen = min(buflen, MAXBSIZE); buf = malloc(buflen, M_LINUX, M_WAITOK); lbuf = malloc(LINUX_MAXRECLEN, M_LINUX, M_WAITOK | M_ZERO); vn_lock(vp, LK_SHARED | LK_RETRY); aiov.iov_base = buf; aiov.iov_len = buflen; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_resid = buflen; auio.uio_offset = off; #ifdef MAC /* * Do directory search MAC check using non-cached credentials. */ if ((error = mac_vnode_check_readdir(td->td_ucred, vp))) goto out; #endif /* MAC */ if ((error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, &ncookies, &cookies))) goto out; inp = buf; outp = (caddr_t)args->dirent; resid = nbytes; if ((len = buflen - auio.uio_resid) <= 0) goto eof; cookiep = cookies; if (cookies) { /* * When using cookies, the vfs has the option of reading from * a different offset than that supplied (UFS truncates the * offset to a block boundary to make sure that it never reads * partway through a directory entry, even if the directory * has been compacted). */ while (len > 0 && ncookies > 0 && *cookiep <= off) { bdp = (struct dirent *) inp; len -= bdp->d_reclen; inp += bdp->d_reclen; cookiep++; ncookies--; } } while (len > 0) { if (cookiep && ncookies == 0) break; bdp = (struct dirent *) inp; reclen = bdp->d_reclen; if (reclen & 3) { error = EFAULT; goto out; } if (bdp->d_fileno == 0) { inp += reclen; if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; len -= reclen; continue; } linuxreclen = (is64bit) ? LINUX_RECLEN64(bdp->d_namlen) : LINUX_RECLEN(bdp->d_namlen); if (reclen > len || resid < linuxreclen) { outp++; break; } if (justone) { /* readdir(2) case. */ linux_dirent = (struct l_dirent*)lbuf; linux_dirent->d_ino = bdp->d_fileno; linux_dirent->d_off = (l_off_t)linuxreclen; linux_dirent->d_reclen = (l_ushort)bdp->d_namlen; strlcpy(linux_dirent->d_name, bdp->d_name, linuxreclen - offsetof(struct l_dirent, d_name)); error = copyout(linux_dirent, outp, linuxreclen); } if (is64bit) { linux_dirent64 = (struct l_dirent64*)lbuf; linux_dirent64->d_ino = bdp->d_fileno; linux_dirent64->d_off = (cookiep) ? (l_off_t)*cookiep : (l_off_t)(off + reclen); linux_dirent64->d_reclen = (l_ushort)linuxreclen; linux_dirent64->d_type = bdp->d_type; strlcpy(linux_dirent64->d_name, bdp->d_name, linuxreclen - offsetof(struct l_dirent64, d_name)); error = copyout(linux_dirent64, outp, linuxreclen); } else if (!justone) { linux_dirent = (struct l_dirent*)lbuf; linux_dirent->d_ino = bdp->d_fileno; linux_dirent->d_off = (cookiep) ? (l_off_t)*cookiep : (l_off_t)(off + reclen); linux_dirent->d_reclen = (l_ushort)linuxreclen; /* * Copy d_type to last byte of l_dirent buffer */ lbuf[linuxreclen-1] = bdp->d_type; strlcpy(linux_dirent->d_name, bdp->d_name, linuxreclen - offsetof(struct l_dirent, d_name)-1); error = copyout(linux_dirent, outp, linuxreclen); } if (error) goto out; inp += reclen; if (cookiep) { off = *cookiep++; ncookies--; } else off += reclen; outp += linuxreclen; resid -= linuxreclen; len -= reclen; if (justone) break; } if (outp == (caddr_t)args->dirent) { nbytes = resid; goto eof; } if (justone) nbytes = resid + linuxreclen; eof: td->td_retval[0] = nbytes - resid; out: free(cookies, M_TEMP); VOP_UNLOCK(vp, 0); foffset_unlock(fp, off, 0); fdrop(fp, td); free(buf, M_LINUX); free(lbuf, M_LINUX); return (error); } int linux_getdents(struct thread *td, struct linux_getdents_args *args) { #ifdef DEBUG if (ldebug(getdents)) printf(ARGS(getdents, "%d, *, %d"), args->fd, args->count); #endif return (getdents_common(td, (struct linux_getdents64_args*)args, 0)); } int linux_getdents64(struct thread *td, struct linux_getdents64_args *args) { #ifdef DEBUG if (ldebug(getdents64)) printf(ARGS(getdents64, "%d, *, %d"), args->fd, args->count); #endif return (getdents_common(td, args, 1)); } /* * These exist mainly for hooks for doing /compat/linux translation. */ int linux_access(struct thread *td, struct linux_access_args *args) { char *path; int error; /* linux convention */ if (args->amode & ~(F_OK | X_OK | W_OK | R_OK)) return (EINVAL); LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(access)) printf(ARGS(access, "%s, %d"), path, args->amode); #endif error = kern_accessat(td, AT_FDCWD, path, UIO_SYSSPACE, 0, args->amode); LFREEPATH(path); return (error); } int linux_faccessat(struct thread *td, struct linux_faccessat_args *args) { char *path; int error, dfd; /* linux convention */ if (args->amode & ~(F_OK | X_OK | W_OK | R_OK)) return (EINVAL); dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->filename, &path, dfd); #ifdef DEBUG if (ldebug(access)) printf(ARGS(access, "%s, %d"), path, args->amode); #endif error = kern_accessat(td, dfd, path, UIO_SYSSPACE, 0, args->amode); LFREEPATH(path); return (error); } int linux_unlink(struct thread *td, struct linux_unlink_args *args) { char *path; int error; struct stat st; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(unlink)) printf(ARGS(unlink, "%s"), path); #endif error = kern_unlinkat(td, AT_FDCWD, path, UIO_SYSSPACE, 0); if (error == EPERM) { /* Introduce POSIX noncompliant behaviour of Linux */ if (kern_statat(td, 0, AT_FDCWD, path, UIO_SYSSPACE, &st, NULL) == 0) { if (S_ISDIR(st.st_mode)) error = EISDIR; } } LFREEPATH(path); return (error); } int linux_unlinkat(struct thread *td, struct linux_unlinkat_args *args) { char *path; int error, dfd; struct stat st; if (args->flag & ~LINUX_AT_REMOVEDIR) return (EINVAL); dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->pathname, &path, dfd); #ifdef DEBUG if (ldebug(unlinkat)) printf(ARGS(unlinkat, "%s"), path); #endif if (args->flag & LINUX_AT_REMOVEDIR) error = kern_rmdirat(td, dfd, path, UIO_SYSSPACE); else error = kern_unlinkat(td, dfd, path, UIO_SYSSPACE, 0); if (error == EPERM && !(args->flag & LINUX_AT_REMOVEDIR)) { /* Introduce POSIX noncompliant behaviour of Linux */ if (kern_statat(td, AT_SYMLINK_NOFOLLOW, dfd, path, UIO_SYSSPACE, &st, NULL) == 0 && S_ISDIR(st.st_mode)) error = EISDIR; } LFREEPATH(path); return (error); } int linux_chdir(struct thread *td, struct linux_chdir_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(chdir)) printf(ARGS(chdir, "%s"), path); #endif error = kern_chdir(td, path, UIO_SYSSPACE); LFREEPATH(path); return (error); } int linux_chmod(struct thread *td, struct linux_chmod_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(chmod)) printf(ARGS(chmod, "%s, %d"), path, args->mode); #endif error = kern_fchmodat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode, 0); LFREEPATH(path); return (error); } int linux_fchmodat(struct thread *td, struct linux_fchmodat_args *args) { char *path; int error, dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->filename, &path, dfd); #ifdef DEBUG if (ldebug(fchmodat)) printf(ARGS(fchmodat, "%s, %d"), path, args->mode); #endif error = kern_fchmodat(td, dfd, path, UIO_SYSSPACE, args->mode, 0); LFREEPATH(path); return (error); } int linux_mkdir(struct thread *td, struct linux_mkdir_args *args) { char *path; int error; LCONVPATHCREAT(td, args->path, &path); #ifdef DEBUG if (ldebug(mkdir)) printf(ARGS(mkdir, "%s, %d"), path, args->mode); #endif error = kern_mkdirat(td, AT_FDCWD, path, UIO_SYSSPACE, args->mode); LFREEPATH(path); return (error); } int linux_mkdirat(struct thread *td, struct linux_mkdirat_args *args) { char *path; int error, dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHCREAT_AT(td, args->pathname, &path, dfd); #ifdef DEBUG if (ldebug(mkdirat)) printf(ARGS(mkdirat, "%s, %d"), path, args->mode); #endif error = kern_mkdirat(td, dfd, path, UIO_SYSSPACE, args->mode); LFREEPATH(path); return (error); } int linux_rmdir(struct thread *td, struct linux_rmdir_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(rmdir)) printf(ARGS(rmdir, "%s"), path); #endif error = kern_rmdirat(td, AT_FDCWD, path, UIO_SYSSPACE); LFREEPATH(path); return (error); } int linux_rename(struct thread *td, struct linux_rename_args *args) { char *from, *to; int error; LCONVPATHEXIST(td, args->from, &from); /* Expand LCONVPATHCREATE so that `from' can be freed on errors */ error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD); if (to == NULL) { LFREEPATH(from); return (error); } #ifdef DEBUG if (ldebug(rename)) printf(ARGS(rename, "%s, %s"), from, to); #endif error = kern_renameat(td, AT_FDCWD, from, AT_FDCWD, to, UIO_SYSSPACE); LFREEPATH(from); LFREEPATH(to); return (error); } int linux_renameat(struct thread *td, struct linux_renameat_args *args) { char *from, *to; int error, olddfd, newdfd; olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd; newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd; LCONVPATHEXIST_AT(td, args->oldname, &from, olddfd); /* Expand LCONVPATHCREATE so that `from' can be freed on errors */ error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd); if (to == NULL) { LFREEPATH(from); return (error); } #ifdef DEBUG if (ldebug(renameat)) printf(ARGS(renameat, "%s, %s"), from, to); #endif error = kern_renameat(td, olddfd, from, newdfd, to, UIO_SYSSPACE); LFREEPATH(from); LFREEPATH(to); return (error); } int linux_symlink(struct thread *td, struct linux_symlink_args *args) { char *path, *to; int error; LCONVPATHEXIST(td, args->path, &path); /* Expand LCONVPATHCREATE so that `path' can be freed on errors */ error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD); if (to == NULL) { LFREEPATH(path); return (error); } #ifdef DEBUG if (ldebug(symlink)) printf(ARGS(symlink, "%s, %s"), path, to); #endif error = kern_symlinkat(td, path, AT_FDCWD, to, UIO_SYSSPACE); LFREEPATH(path); LFREEPATH(to); return (error); } int linux_symlinkat(struct thread *td, struct linux_symlinkat_args *args) { char *path, *to; int error, dfd; dfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd; LCONVPATHEXIST_AT(td, args->oldname, &path, dfd); /* Expand LCONVPATHCREATE so that `path' can be freed on errors */ error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, dfd); if (to == NULL) { LFREEPATH(path); return (error); } #ifdef DEBUG if (ldebug(symlinkat)) printf(ARGS(symlinkat, "%s, %s"), path, to); #endif error = kern_symlinkat(td, path, dfd, to, UIO_SYSSPACE); LFREEPATH(path); LFREEPATH(to); return (error); } int linux_readlink(struct thread *td, struct linux_readlink_args *args) { char *name; int error; LCONVPATHEXIST(td, args->name, &name); #ifdef DEBUG if (ldebug(readlink)) printf(ARGS(readlink, "%s, %p, %d"), name, (void *)args->buf, args->count); #endif error = kern_readlinkat(td, AT_FDCWD, name, UIO_SYSSPACE, args->buf, UIO_USERSPACE, args->count); LFREEPATH(name); return (error); } int linux_readlinkat(struct thread *td, struct linux_readlinkat_args *args) { char *name; int error, dfd; dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->path, &name, dfd); #ifdef DEBUG if (ldebug(readlinkat)) printf(ARGS(readlinkat, "%s, %p, %d"), name, (void *)args->buf, args->bufsiz); #endif error = kern_readlinkat(td, dfd, name, UIO_SYSSPACE, args->buf, UIO_USERSPACE, args->bufsiz); LFREEPATH(name); return (error); } int linux_truncate(struct thread *td, struct linux_truncate_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(truncate)) printf(ARGS(truncate, "%s, %ld"), path, (long)args->length); #endif error = kern_truncate(td, path, UIO_SYSSPACE, args->length); LFREEPATH(path); return (error); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_truncate64(struct thread *td, struct linux_truncate64_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(truncate64)) printf(ARGS(truncate64, "%s, %jd"), path, args->length); #endif error = kern_truncate(td, path, UIO_SYSSPACE, args->length); LFREEPATH(path); return (error); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_ftruncate(struct thread *td, struct linux_ftruncate_args *args) { struct ftruncate_args /* { int fd; int pad; off_t length; } */ nuap; nuap.fd = args->fd; nuap.length = args->length; return (sys_ftruncate(td, &nuap)); } int linux_link(struct thread *td, struct linux_link_args *args) { char *path, *to; int error; LCONVPATHEXIST(td, args->path, &path); /* Expand LCONVPATHCREATE so that `path' can be freed on errors */ error = linux_emul_convpath(td, args->to, UIO_USERSPACE, &to, 1, AT_FDCWD); if (to == NULL) { LFREEPATH(path); return (error); } #ifdef DEBUG if (ldebug(link)) printf(ARGS(link, "%s, %s"), path, to); #endif error = kern_linkat(td, AT_FDCWD, AT_FDCWD, path, to, UIO_SYSSPACE, FOLLOW); LFREEPATH(path); LFREEPATH(to); return (error); } int linux_linkat(struct thread *td, struct linux_linkat_args *args) { char *path, *to; int error, olddfd, newdfd, follow; if (args->flag & ~LINUX_AT_SYMLINK_FOLLOW) return (EINVAL); olddfd = (args->olddfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->olddfd; newdfd = (args->newdfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->newdfd; LCONVPATHEXIST_AT(td, args->oldname, &path, olddfd); /* Expand LCONVPATHCREATE so that `path' can be freed on errors */ error = linux_emul_convpath(td, args->newname, UIO_USERSPACE, &to, 1, newdfd); if (to == NULL) { LFREEPATH(path); return (error); } #ifdef DEBUG if (ldebug(linkat)) printf(ARGS(linkat, "%i, %s, %i, %s, %i"), args->olddfd, path, args->newdfd, to, args->flag); #endif follow = (args->flag & LINUX_AT_SYMLINK_FOLLOW) == 0 ? NOFOLLOW : FOLLOW; error = kern_linkat(td, olddfd, newdfd, path, to, UIO_SYSSPACE, follow); LFREEPATH(path); LFREEPATH(to); return (error); } int linux_fdatasync(td, uap) struct thread *td; struct linux_fdatasync_args *uap; { struct fsync_args bsd; bsd.fd = uap->fd; return sys_fsync(td, &bsd); } int linux_pread(td, uap) struct thread *td; struct linux_pread_args *uap; { struct pread_args bsd; cap_rights_t rights; struct vnode *vp; int error; bsd.fd = uap->fd; bsd.buf = uap->buf; bsd.nbyte = uap->nbyte; bsd.offset = uap->offset; error = sys_pread(td, &bsd); if (error == 0) { /* This seems to violate POSIX but linux does it */ error = fgetvp(td, uap->fd, cap_rights_init(&rights, CAP_PREAD), &vp); if (error != 0) return (error); if (vp->v_type == VDIR) { vrele(vp); return (EISDIR); } vrele(vp); } return (error); } int linux_pwrite(td, uap) struct thread *td; struct linux_pwrite_args *uap; { struct pwrite_args bsd; bsd.fd = uap->fd; bsd.buf = uap->buf; bsd.nbyte = uap->nbyte; bsd.offset = uap->offset; return sys_pwrite(td, &bsd); } int linux_mount(struct thread *td, struct linux_mount_args *args) { char fstypename[MFSNAMELEN]; char mntonname[MNAMELEN], mntfromname[MNAMELEN]; int error; int fsflags; error = copyinstr(args->filesystemtype, fstypename, MFSNAMELEN - 1, NULL); if (error) return (error); error = copyinstr(args->specialfile, mntfromname, MNAMELEN - 1, NULL); if (error) return (error); error = copyinstr(args->dir, mntonname, MNAMELEN - 1, NULL); if (error) return (error); #ifdef DEBUG if (ldebug(mount)) printf(ARGS(mount, "%s, %s, %s"), fstypename, mntfromname, mntonname); #endif if (strcmp(fstypename, "ext2") == 0) { strcpy(fstypename, "ext2fs"); } else if (strcmp(fstypename, "proc") == 0) { strcpy(fstypename, "linprocfs"); } else if (strcmp(fstypename, "vfat") == 0) { strcpy(fstypename, "msdosfs"); } fsflags = 0; if ((args->rwflag & 0xffff0000) == 0xc0ed0000) { /* * Linux SYNC flag is not included; the closest equivalent * FreeBSD has is !ASYNC, which is our default. */ if (args->rwflag & LINUX_MS_RDONLY) fsflags |= MNT_RDONLY; if (args->rwflag & LINUX_MS_NOSUID) fsflags |= MNT_NOSUID; if (args->rwflag & LINUX_MS_NOEXEC) fsflags |= MNT_NOEXEC; if (args->rwflag & LINUX_MS_REMOUNT) fsflags |= MNT_UPDATE; } error = kernel_vmount(fsflags, "fstype", fstypename, "fspath", mntonname, "from", mntfromname, NULL); return (error); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_oldumount(struct thread *td, struct linux_oldumount_args *args) { struct linux_umount_args args2; args2.path = args->path; args2.flags = 0; return (linux_umount(td, &args2)); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_umount(struct thread *td, struct linux_umount_args *args) { struct unmount_args bsd; bsd.path = args->path; bsd.flags = args->flags; /* XXX correct? */ return (sys_unmount(td, &bsd)); } /* * fcntl family of syscalls */ struct l_flock { l_short l_type; l_short l_whence; l_off_t l_start; l_off_t l_len; l_pid_t l_pid; } #if defined(__amd64__) && defined(COMPAT_LINUX32) __packed #endif ; static void linux_to_bsd_flock(struct l_flock *linux_flock, struct flock *bsd_flock) { switch (linux_flock->l_type) { case LINUX_F_RDLCK: bsd_flock->l_type = F_RDLCK; break; case LINUX_F_WRLCK: bsd_flock->l_type = F_WRLCK; break; case LINUX_F_UNLCK: bsd_flock->l_type = F_UNLCK; break; default: bsd_flock->l_type = -1; break; } bsd_flock->l_whence = linux_flock->l_whence; bsd_flock->l_start = (off_t)linux_flock->l_start; bsd_flock->l_len = (off_t)linux_flock->l_len; bsd_flock->l_pid = (pid_t)linux_flock->l_pid; bsd_flock->l_sysid = 0; } static void bsd_to_linux_flock(struct flock *bsd_flock, struct l_flock *linux_flock) { switch (bsd_flock->l_type) { case F_RDLCK: linux_flock->l_type = LINUX_F_RDLCK; break; case F_WRLCK: linux_flock->l_type = LINUX_F_WRLCK; break; case F_UNLCK: linux_flock->l_type = LINUX_F_UNLCK; break; } linux_flock->l_whence = bsd_flock->l_whence; linux_flock->l_start = (l_off_t)bsd_flock->l_start; linux_flock->l_len = (l_off_t)bsd_flock->l_len; linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid; } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) struct l_flock64 { l_short l_type; l_short l_whence; l_loff_t l_start; l_loff_t l_len; l_pid_t l_pid; } #if defined(__amd64__) && defined(COMPAT_LINUX32) __packed #endif ; static void linux_to_bsd_flock64(struct l_flock64 *linux_flock, struct flock *bsd_flock) { switch (linux_flock->l_type) { case LINUX_F_RDLCK: bsd_flock->l_type = F_RDLCK; break; case LINUX_F_WRLCK: bsd_flock->l_type = F_WRLCK; break; case LINUX_F_UNLCK: bsd_flock->l_type = F_UNLCK; break; default: bsd_flock->l_type = -1; break; } bsd_flock->l_whence = linux_flock->l_whence; bsd_flock->l_start = (off_t)linux_flock->l_start; bsd_flock->l_len = (off_t)linux_flock->l_len; bsd_flock->l_pid = (pid_t)linux_flock->l_pid; bsd_flock->l_sysid = 0; } static void bsd_to_linux_flock64(struct flock *bsd_flock, struct l_flock64 *linux_flock) { switch (bsd_flock->l_type) { case F_RDLCK: linux_flock->l_type = LINUX_F_RDLCK; break; case F_WRLCK: linux_flock->l_type = LINUX_F_WRLCK; break; case F_UNLCK: linux_flock->l_type = LINUX_F_UNLCK; break; } linux_flock->l_whence = bsd_flock->l_whence; linux_flock->l_start = (l_loff_t)bsd_flock->l_start; linux_flock->l_len = (l_loff_t)bsd_flock->l_len; linux_flock->l_pid = (l_pid_t)bsd_flock->l_pid; } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ static int fcntl_common(struct thread *td, struct linux_fcntl_args *args) { struct l_flock linux_flock; struct flock bsd_flock; cap_rights_t rights; struct file *fp; long arg; int error, result; switch (args->cmd) { case LINUX_F_DUPFD: return (kern_fcntl(td, args->fd, F_DUPFD, args->arg)); case LINUX_F_GETFD: return (kern_fcntl(td, args->fd, F_GETFD, 0)); case LINUX_F_SETFD: return (kern_fcntl(td, args->fd, F_SETFD, args->arg)); case LINUX_F_GETFL: error = kern_fcntl(td, args->fd, F_GETFL, 0); result = td->td_retval[0]; td->td_retval[0] = 0; if (result & O_RDONLY) td->td_retval[0] |= LINUX_O_RDONLY; if (result & O_WRONLY) td->td_retval[0] |= LINUX_O_WRONLY; if (result & O_RDWR) td->td_retval[0] |= LINUX_O_RDWR; if (result & O_NDELAY) td->td_retval[0] |= LINUX_O_NONBLOCK; if (result & O_APPEND) td->td_retval[0] |= LINUX_O_APPEND; if (result & O_FSYNC) td->td_retval[0] |= LINUX_O_SYNC; if (result & O_ASYNC) td->td_retval[0] |= LINUX_FASYNC; #ifdef LINUX_O_NOFOLLOW if (result & O_NOFOLLOW) td->td_retval[0] |= LINUX_O_NOFOLLOW; #endif #ifdef LINUX_O_DIRECT if (result & O_DIRECT) td->td_retval[0] |= LINUX_O_DIRECT; #endif return (error); case LINUX_F_SETFL: arg = 0; if (args->arg & LINUX_O_NDELAY) arg |= O_NONBLOCK; if (args->arg & LINUX_O_APPEND) arg |= O_APPEND; if (args->arg & LINUX_O_SYNC) arg |= O_FSYNC; if (args->arg & LINUX_FASYNC) arg |= O_ASYNC; #ifdef LINUX_O_NOFOLLOW if (args->arg & LINUX_O_NOFOLLOW) arg |= O_NOFOLLOW; #endif #ifdef LINUX_O_DIRECT if (args->arg & LINUX_O_DIRECT) arg |= O_DIRECT; #endif return (kern_fcntl(td, args->fd, F_SETFL, arg)); case LINUX_F_GETLK: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock(&linux_flock, &bsd_flock); error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock); if (error) return (error); bsd_to_linux_flock(&bsd_flock, &linux_flock); return (copyout(&linux_flock, (void *)args->arg, sizeof(linux_flock))); case LINUX_F_SETLK: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock(&linux_flock, &bsd_flock); return (kern_fcntl(td, args->fd, F_SETLK, (intptr_t)&bsd_flock)); case LINUX_F_SETLKW: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock(&linux_flock, &bsd_flock); return (kern_fcntl(td, args->fd, F_SETLKW, (intptr_t)&bsd_flock)); case LINUX_F_GETOWN: return (kern_fcntl(td, args->fd, F_GETOWN, 0)); case LINUX_F_SETOWN: /* * XXX some Linux applications depend on F_SETOWN having no * significant effect for pipes (SIGIO is not delivered for * pipes under Linux-2.2.35 at least). */ error = fget(td, args->fd, cap_rights_init(&rights, CAP_FCNTL), &fp); if (error) return (error); if (fp->f_type == DTYPE_PIPE) { fdrop(fp, td); return (EINVAL); } fdrop(fp, td); return (kern_fcntl(td, args->fd, F_SETOWN, args->arg)); case LINUX_F_DUPFD_CLOEXEC: return (kern_fcntl(td, args->fd, F_DUPFD_CLOEXEC, args->arg)); } return (EINVAL); } int linux_fcntl(struct thread *td, struct linux_fcntl_args *args) { #ifdef DEBUG if (ldebug(fcntl)) printf(ARGS(fcntl, "%d, %08x, *"), args->fd, args->cmd); #endif return (fcntl_common(td, args)); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_fcntl64(struct thread *td, struct linux_fcntl64_args *args) { struct l_flock64 linux_flock; struct flock bsd_flock; struct linux_fcntl_args fcntl_args; int error; #ifdef DEBUG if (ldebug(fcntl64)) printf(ARGS(fcntl64, "%d, %08x, *"), args->fd, args->cmd); #endif switch (args->cmd) { case LINUX_F_GETLK64: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock64(&linux_flock, &bsd_flock); error = kern_fcntl(td, args->fd, F_GETLK, (intptr_t)&bsd_flock); if (error) return (error); bsd_to_linux_flock64(&bsd_flock, &linux_flock); return (copyout(&linux_flock, (void *)args->arg, sizeof(linux_flock))); case LINUX_F_SETLK64: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock64(&linux_flock, &bsd_flock); return (kern_fcntl(td, args->fd, F_SETLK, (intptr_t)&bsd_flock)); case LINUX_F_SETLKW64: error = copyin((void *)args->arg, &linux_flock, sizeof(linux_flock)); if (error) return (error); linux_to_bsd_flock64(&linux_flock, &bsd_flock); return (kern_fcntl(td, args->fd, F_SETLKW, (intptr_t)&bsd_flock)); } fcntl_args.fd = args->fd; fcntl_args.cmd = args->cmd; fcntl_args.arg = args->arg; return (fcntl_common(td, &fcntl_args)); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_chown(struct thread *td, struct linux_chown_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(chown)) printf(ARGS(chown, "%s, %d, %d"), path, args->uid, args->gid); #endif error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid, args->gid, 0); LFREEPATH(path); return (error); } int linux_fchownat(struct thread *td, struct linux_fchownat_args *args) { char *path; int error, dfd, flag; if (args->flag & ~LINUX_AT_SYMLINK_NOFOLLOW) return (EINVAL); dfd = (args->dfd == LINUX_AT_FDCWD) ? AT_FDCWD : args->dfd; LCONVPATHEXIST_AT(td, args->filename, &path, dfd); #ifdef DEBUG if (ldebug(fchownat)) printf(ARGS(fchownat, "%s, %d, %d"), path, args->uid, args->gid); #endif flag = (args->flag & LINUX_AT_SYMLINK_NOFOLLOW) == 0 ? 0 : AT_SYMLINK_NOFOLLOW; error = kern_fchownat(td, dfd, path, UIO_SYSSPACE, args->uid, args->gid, flag); LFREEPATH(path); return (error); } int linux_lchown(struct thread *td, struct linux_lchown_args *args) { char *path; int error; LCONVPATHEXIST(td, args->path, &path); #ifdef DEBUG if (ldebug(lchown)) printf(ARGS(lchown, "%s, %d, %d"), path, args->uid, args->gid); #endif error = kern_fchownat(td, AT_FDCWD, path, UIO_SYSSPACE, args->uid, args->gid, AT_SYMLINK_NOFOLLOW); LFREEPATH(path); return (error); } static int convert_fadvice(int advice) { switch (advice) { case LINUX_POSIX_FADV_NORMAL: return (POSIX_FADV_NORMAL); case LINUX_POSIX_FADV_RANDOM: return (POSIX_FADV_RANDOM); case LINUX_POSIX_FADV_SEQUENTIAL: return (POSIX_FADV_SEQUENTIAL); case LINUX_POSIX_FADV_WILLNEED: return (POSIX_FADV_WILLNEED); case LINUX_POSIX_FADV_DONTNEED: return (POSIX_FADV_DONTNEED); case LINUX_POSIX_FADV_NOREUSE: return (POSIX_FADV_NOREUSE); default: return (-1); } } int linux_fadvise64(struct thread *td, struct linux_fadvise64_args *args) { int advice; advice = convert_fadvice(args->advice); if (advice == -1) return (EINVAL); return (kern_posix_fadvise(td, args->fd, args->offset, args->len, advice)); } #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32)) int linux_fadvise64_64(struct thread *td, struct linux_fadvise64_64_args *args) { int advice; advice = convert_fadvice(args->advice); if (advice == -1) return (EINVAL); return (kern_posix_fadvise(td, args->fd, args->offset, args->len, advice)); } #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */ int linux_pipe(struct thread *td, struct linux_pipe_args *args) { int fildes[2]; int error; #ifdef DEBUG if (ldebug(pipe)) printf(ARGS(pipe, "*")); #endif error = kern_pipe(td, fildes, 0, NULL, NULL); if (error) return (error); /* XXX: Close descriptors on error. */ return (copyout(fildes, args->pipefds, sizeof(fildes))); } int linux_pipe2(struct thread *td, struct linux_pipe2_args *args) { int fildes[2]; int error, flags; #ifdef DEBUG if (ldebug(pipe2)) printf(ARGS(pipe2, "*, %d"), args->flags); #endif if ((args->flags & ~(LINUX_O_NONBLOCK | LINUX_O_CLOEXEC)) != 0) return (EINVAL); flags = 0; if ((args->flags & LINUX_O_NONBLOCK) != 0) flags |= O_NONBLOCK; if ((args->flags & LINUX_O_CLOEXEC) != 0) flags |= O_CLOEXEC; error = kern_pipe(td, fildes, flags, NULL, NULL); if (error) return (error); /* XXX: Close descriptors on error. */ return (copyout(fildes, args->pipefds, sizeof(fildes))); } int linux_dup3(struct thread *td, struct linux_dup3_args *args) { int cmd; intptr_t newfd; if (args->oldfd == args->newfd) return (EINVAL); if ((args->flags & ~LINUX_O_CLOEXEC) != 0) return (EINVAL); if (args->flags & LINUX_O_CLOEXEC) cmd = F_DUP2FD_CLOEXEC; else cmd = F_DUP2FD; newfd = args->newfd; return (kern_fcntl(td, args->oldfd, cmd, newfd)); } int linux_fallocate(struct thread *td, struct linux_fallocate_args *args) { /* * We emulate only posix_fallocate system call for which * mode should be 0. */ if (args->mode != 0) return (ENOSYS); return (kern_posix_fallocate(td, args->fd, args->offset, args->len)); } Index: head/sys/dev/cxgbe/tom/t4_tom.c =================================================================== --- head/sys/dev/cxgbe/tom/t4_tom.c (revision 298481) +++ head/sys/dev/cxgbe/tom/t4_tom.c (revision 298482) @@ -1,1203 +1,1203 @@ /*- * Copyright (c) 2012 Chelsio Communications, Inc. * All rights reserved. * Written by: Navdeep Parhar * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES #include #include #include #ifdef TCP_OFFLOAD #include "common/common.h" #include "common/t4_msg.h" #include "common/t4_regs.h" #include "common/t4_regs_values.h" #include "common/t4_tcb.h" #include "tom/t4_tom_l2t.h" #include "tom/t4_tom.h" static struct protosw ddp_protosw; static struct pr_usrreqs ddp_usrreqs; static struct protosw ddp6_protosw; static struct pr_usrreqs ddp6_usrreqs; /* Module ops */ static int t4_tom_mod_load(void); static int t4_tom_mod_unload(void); static int t4_tom_modevent(module_t, int, void *); /* ULD ops and helpers */ static int t4_tom_activate(struct adapter *); static int t4_tom_deactivate(struct adapter *); static struct uld_info tom_uld_info = { .uld_id = ULD_TOM, .activate = t4_tom_activate, .deactivate = t4_tom_deactivate, }; static void queue_tid_release(struct adapter *, int); static void release_offload_resources(struct toepcb *); static int alloc_tid_tabs(struct tid_info *); static void free_tid_tabs(struct tid_info *); static int add_lip(struct adapter *, struct in6_addr *); static int delete_lip(struct adapter *, struct in6_addr *); static struct clip_entry *search_lip(struct tom_data *, struct in6_addr *); static void init_clip_table(struct adapter *, struct tom_data *); static void update_clip(struct adapter *, void *); static void t4_clip_task(void *, int); static void update_clip_table(struct adapter *, struct tom_data *); static void destroy_clip_table(struct adapter *, struct tom_data *); static void free_tom_data(struct adapter *, struct tom_data *); static void reclaim_wr_resources(void *, int); static int in6_ifaddr_gen; static eventhandler_tag ifaddr_evhandler; static struct timeout_task clip_task; struct toepcb * alloc_toepcb(struct vi_info *vi, int txqid, int rxqid, int flags) { struct port_info *pi = vi->pi; struct adapter *sc = pi->adapter; struct toepcb *toep; int tx_credits, txsd_total, len; /* * The firmware counts tx work request credits in units of 16 bytes * each. Reserve room for an ABORT_REQ so the driver never has to worry * about tx credits if it wants to abort a connection. */ tx_credits = sc->params.ofldq_wr_cred; tx_credits -= howmany(sizeof(struct cpl_abort_req), 16); /* * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte * immediate payload, and firmware counts tx work request credits in * units of 16 byte. Calculate the maximum work requests possible. */ txsd_total = tx_credits / - howmany((sizeof(struct fw_ofld_tx_data_wr) + 1), 16); + howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16); if (txqid < 0) txqid = (arc4random() % vi->nofldtxq) + vi->first_ofld_txq; KASSERT(txqid >= vi->first_ofld_txq && txqid < vi->first_ofld_txq + vi->nofldtxq, ("%s: txqid %d for vi %p (first %d, n %d)", __func__, txqid, vi, vi->first_ofld_txq, vi->nofldtxq)); if (rxqid < 0) rxqid = (arc4random() % vi->nofldrxq) + vi->first_ofld_rxq; KASSERT(rxqid >= vi->first_ofld_rxq && rxqid < vi->first_ofld_rxq + vi->nofldrxq, ("%s: rxqid %d for vi %p (first %d, n %d)", __func__, rxqid, vi, vi->first_ofld_rxq, vi->nofldrxq)); len = offsetof(struct toepcb, txsd) + txsd_total * sizeof(struct ofld_tx_sdesc); toep = malloc(len, M_CXGBE, M_ZERO | flags); if (toep == NULL) return (NULL); toep->td = sc->tom_softc; toep->vi = vi; toep->tx_total = tx_credits; toep->tx_credits = tx_credits; toep->ofld_txq = &sc->sge.ofld_txq[txqid]; toep->ofld_rxq = &sc->sge.ofld_rxq[rxqid]; toep->ctrlq = &sc->sge.ctrlq[pi->port_id]; mbufq_init(&toep->ulp_pduq, INT_MAX); mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX); toep->txsd_total = txsd_total; toep->txsd_avail = txsd_total; toep->txsd_pidx = 0; toep->txsd_cidx = 0; return (toep); } void free_toepcb(struct toepcb *toep) { KASSERT(!(toep->flags & TPF_ATTACHED), ("%s: attached to an inpcb", __func__)); KASSERT(!(toep->flags & TPF_CPL_PENDING), ("%s: CPL pending", __func__)); free(toep, M_CXGBE); } /* * Set up the socket for TCP offload. */ void offload_socket(struct socket *so, struct toepcb *toep) { struct tom_data *td = toep->td; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct sockbuf *sb; INP_WLOCK_ASSERT(inp); /* Update socket */ sb = &so->so_snd; SOCKBUF_LOCK(sb); sb->sb_flags |= SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); sb = &so->so_rcv; SOCKBUF_LOCK(sb); sb->sb_flags |= SB_NOCOALESCE; if (toep->ulp_mode == ULP_MODE_TCPDDP) { if (inp->inp_vflag & INP_IPV6) so->so_proto = &ddp6_protosw; else so->so_proto = &ddp_protosw; } SOCKBUF_UNLOCK(sb); /* Update TCP PCB */ tp->tod = &td->tod; tp->t_toe = toep; tp->t_flags |= TF_TOE; /* Install an extra hold on inp */ toep->inp = inp; toep->flags |= TPF_ATTACHED; in_pcbref(inp); /* Add the TOE PCB to the active list */ mtx_lock(&td->toep_list_lock); TAILQ_INSERT_HEAD(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); } /* This is _not_ the normal way to "unoffload" a socket. */ void undo_offload_socket(struct socket *so) { struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct toepcb *toep = tp->t_toe; struct tom_data *td = toep->td; struct sockbuf *sb; INP_WLOCK_ASSERT(inp); sb = &so->so_snd; SOCKBUF_LOCK(sb); sb->sb_flags &= ~SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); sb = &so->so_rcv; SOCKBUF_LOCK(sb); sb->sb_flags &= ~SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); tp->tod = NULL; tp->t_toe = NULL; tp->t_flags &= ~TF_TOE; toep->inp = NULL; toep->flags &= ~TPF_ATTACHED; if (in_pcbrele_wlocked(inp)) panic("%s: inp freed.", __func__); mtx_lock(&td->toep_list_lock); TAILQ_REMOVE(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); } static void release_offload_resources(struct toepcb *toep) { struct tom_data *td = toep->td; struct adapter *sc = td_adapter(td); int tid = toep->tid; KASSERT(!(toep->flags & TPF_CPL_PENDING), ("%s: %p has CPL pending.", __func__, toep)); KASSERT(!(toep->flags & TPF_ATTACHED), ("%s: %p is still attached.", __func__, toep)); CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)", __func__, toep, tid, toep->l2te, toep->ce); /* * These queues should have been emptied at approximately the same time * that a normal connection's socket's so_snd would have been purged or * drained. Do _not_ clean up here. */ MPASS(mbufq_len(&toep->ulp_pduq) == 0); MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0); if (toep->ulp_mode == ULP_MODE_TCPDDP) release_ddp_resources(toep); if (toep->l2te) t4_l2t_release(toep->l2te); if (tid >= 0) { remove_tid(sc, tid); release_tid(sc, tid, toep->ctrlq); } if (toep->ce) release_lip(td, toep->ce); mtx_lock(&td->toep_list_lock); TAILQ_REMOVE(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); free_toepcb(toep); } /* * The kernel is done with the TCP PCB and this is our opportunity to unhook the * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no * pending CPL) then it is time to release all resources tied to the toepcb. * * Also gets called when an offloaded active open fails and the TOM wants the * kernel to take the TCP PCB back. */ static void t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp) { #if defined(KTR) || defined(INVARIANTS) struct inpcb *inp = tp->t_inpcb; #endif struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(inp); KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); KASSERT(toep->flags & TPF_ATTACHED, ("%s: not attached", __func__)); #ifdef KTR if (tp->t_state == TCPS_SYN_SENT) { CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)", __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); } else { CTR6(KTR_CXGBE, "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)", toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp, inp->inp_flags); } #endif tp->t_toe = NULL; tp->t_flags &= ~TF_TOE; toep->flags &= ~TPF_ATTACHED; if (!(toep->flags & TPF_CPL_PENDING)) release_offload_resources(toep); } /* * setsockopt handler. */ static void t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name) { struct adapter *sc = tod->tod_softc; struct toepcb *toep = tp->t_toe; if (dir == SOPT_GET) return; CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name); switch (name) { case TCP_NODELAY: t4_set_tcb_field(sc, toep, 1, W_TCB_T_FLAGS, V_TF_NAGLE(1), V_TF_NAGLE(tp->t_flags & TF_NODELAY ? 0 : 1)); break; default: break; } } /* * The TOE driver will not receive any more CPLs for the tid associated with the * toepcb; release the hold on the inpcb. */ void final_cpl_received(struct toepcb *toep) { struct inpcb *inp = toep->inp; KASSERT(inp != NULL, ("%s: inp is NULL", __func__)); INP_WLOCK_ASSERT(inp); KASSERT(toep->flags & TPF_CPL_PENDING, ("%s: CPL not pending already?", __func__)); CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)", __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); toep->inp = NULL; toep->flags &= ~TPF_CPL_PENDING; mbufq_drain(&toep->ulp_pdu_reclaimq); if (!(toep->flags & TPF_ATTACHED)) release_offload_resources(toep); if (!in_pcbrele_wlocked(inp)) INP_WUNLOCK(inp); } void insert_tid(struct adapter *sc, int tid, void *ctx) { struct tid_info *t = &sc->tids; t->tid_tab[tid] = ctx; atomic_add_int(&t->tids_in_use, 1); } void * lookup_tid(struct adapter *sc, int tid) { struct tid_info *t = &sc->tids; return (t->tid_tab[tid]); } void update_tid(struct adapter *sc, int tid, void *ctx) { struct tid_info *t = &sc->tids; t->tid_tab[tid] = ctx; } void remove_tid(struct adapter *sc, int tid) { struct tid_info *t = &sc->tids; t->tid_tab[tid] = NULL; atomic_subtract_int(&t->tids_in_use, 1); } void release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq) { struct wrqe *wr; struct cpl_tid_release *req; wr = alloc_wrqe(sizeof(*req), ctrlq); if (wr == NULL) { queue_tid_release(sc, tid); /* defer */ return; } req = wrtod(wr); INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid); t4_wrq_tx(sc, wr); } static void queue_tid_release(struct adapter *sc, int tid) { CXGBE_UNIMPLEMENTED("deferred tid release"); } /* * What mtu_idx to use, given a 4-tuple and/or an MSS cap */ int find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, int pmss) { unsigned short *mtus = &sc->params.mtus[0]; int i, mss, n; KASSERT(inc != NULL || pmss > 0, ("%s: at least one of inc/pmss must be specified", __func__)); mss = inc ? tcp_mssopt(inc) : pmss; if (pmss > 0 && mss > pmss) mss = pmss; if (inc->inc_flags & INC_ISIPV6) n = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); else n = sizeof(struct ip) + sizeof(struct tcphdr); for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mss + n; i++) continue; return (i); } /* * Determine the receive window size for a socket. */ u_long select_rcv_wnd(struct socket *so) { unsigned long wnd; SOCKBUF_LOCK_ASSERT(&so->so_rcv); wnd = sbspace(&so->so_rcv); if (wnd < MIN_RCV_WND) wnd = MIN_RCV_WND; return min(wnd, MAX_RCV_WND); } int select_rcv_wscale(void) { int wscale = 0; unsigned long space = sb_max; if (space > MAX_RCV_WND) space = MAX_RCV_WND; while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space) wscale++; return (wscale); } extern int always_keepalive; #define VIID_SMACIDX(v) (((unsigned int)(v) & 0x7f) << 1) /* * socket so could be a listening socket too. */ uint64_t calc_opt0(struct socket *so, struct vi_info *vi, struct l2t_entry *e, int mtu_idx, int rscale, int rx_credits, int ulp_mode) { uint64_t opt0; KASSERT(rx_credits <= M_RCV_BUFSIZ, ("%s: rcv_bufsiz too high", __func__)); opt0 = F_TCAM_BYPASS | V_WND_SCALE(rscale) | V_MSS_IDX(mtu_idx) | V_ULP_MODE(ulp_mode) | V_RCV_BUFSIZ(rx_credits); if (so != NULL) { struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); int keepalive = always_keepalive || so_options_get(so) & SO_KEEPALIVE; opt0 |= V_NAGLE((tp->t_flags & TF_NODELAY) == 0); opt0 |= V_KEEP_ALIVE(keepalive != 0); } if (e != NULL) opt0 |= V_L2T_IDX(e->idx); if (vi != NULL) { opt0 |= V_SMAC_SEL(VIID_SMACIDX(vi->viid)); opt0 |= V_TX_CHAN(vi->pi->tx_chan); } return htobe64(opt0); } uint64_t select_ntuple(struct vi_info *vi, struct l2t_entry *e) { struct adapter *sc = vi->pi->adapter; struct tp_params *tp = &sc->params.tp; uint16_t viid = vi->viid; uint64_t ntuple = 0; /* * Initialize each of the fields which we care about which are present * in the Compressed Filter Tuple. */ if (tp->vlan_shift >= 0 && e->vlan != CPL_L2T_VLAN_NONE) ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift; if (tp->port_shift >= 0) ntuple |= (uint64_t)e->lport << tp->port_shift; if (tp->protocol_shift >= 0) ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift; if (tp->vnic_shift >= 0) { uint32_t vf = G_FW_VIID_VIN(viid); uint32_t pf = G_FW_VIID_PFN(viid); uint32_t vld = G_FW_VIID_VIVLD(viid); ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vf) | V_FT_VNID_ID_PF(pf) | V_FT_VNID_ID_VLD(vld)) << tp->vnic_shift; } if (is_t4(sc)) return (htobe32((uint32_t)ntuple)); else return (htobe64(V_FILTER_TUPLE(ntuple))); } void set_tcpddp_ulp_mode(struct toepcb *toep) { toep->ulp_mode = ULP_MODE_TCPDDP; toep->ddp_flags = DDP_OK; toep->ddp_score = DDP_LOW_SCORE; } int negative_advice(int status) { return (status == CPL_ERR_RTX_NEG_ADVICE || status == CPL_ERR_PERSIST_NEG_ADVICE || status == CPL_ERR_KEEPALV_NEG_ADVICE); } static int alloc_tid_tabs(struct tid_info *t) { size_t size; unsigned int i; size = t->ntids * sizeof(*t->tid_tab) + t->natids * sizeof(*t->atid_tab) + t->nstids * sizeof(*t->stid_tab); t->tid_tab = malloc(size, M_CXGBE, M_ZERO | M_NOWAIT); if (t->tid_tab == NULL) return (ENOMEM); mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF); t->atid_tab = (union aopen_entry *)&t->tid_tab[t->ntids]; t->afree = t->atid_tab; t->atids_in_use = 0; for (i = 1; i < t->natids; i++) t->atid_tab[i - 1].next = &t->atid_tab[i]; t->atid_tab[t->natids - 1].next = NULL; mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF); t->stid_tab = (struct listen_ctx **)&t->atid_tab[t->natids]; t->stids_in_use = 0; TAILQ_INIT(&t->stids); t->nstids_free_head = t->nstids; atomic_store_rel_int(&t->tids_in_use, 0); return (0); } static void free_tid_tabs(struct tid_info *t) { KASSERT(t->tids_in_use == 0, ("%s: %d tids still in use.", __func__, t->tids_in_use)); KASSERT(t->atids_in_use == 0, ("%s: %d atids still in use.", __func__, t->atids_in_use)); KASSERT(t->stids_in_use == 0, ("%s: %d tids still in use.", __func__, t->stids_in_use)); free(t->tid_tab, M_CXGBE); t->tid_tab = NULL; if (mtx_initialized(&t->atid_lock)) mtx_destroy(&t->atid_lock); if (mtx_initialized(&t->stid_lock)) mtx_destroy(&t->stid_lock); } static int add_lip(struct adapter *sc, struct in6_addr *lip) { struct fw_clip_cmd c; ASSERT_SYNCHRONIZED_OP(sc); /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ memset(&c, 0, sizeof(c)); c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | F_FW_CMD_WRITE); c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_ALLOC | FW_LEN16(c)); c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); } static int delete_lip(struct adapter *sc, struct in6_addr *lip) { struct fw_clip_cmd c; ASSERT_SYNCHRONIZED_OP(sc); /* mtx_assert(&td->clip_table_lock, MA_OWNED); */ memset(&c, 0, sizeof(c)); c.op_to_write = htonl(V_FW_CMD_OP(FW_CLIP_CMD) | F_FW_CMD_REQUEST | F_FW_CMD_READ); c.alloc_to_len16 = htonl(F_FW_CLIP_CMD_FREE | FW_LEN16(c)); c.ip_hi = *(uint64_t *)&lip->s6_addr[0]; c.ip_lo = *(uint64_t *)&lip->s6_addr[8]; return (-t4_wr_mbox_ns(sc, sc->mbox, &c, sizeof(c), &c)); } static struct clip_entry * search_lip(struct tom_data *td, struct in6_addr *lip) { struct clip_entry *ce; mtx_assert(&td->clip_table_lock, MA_OWNED); TAILQ_FOREACH(ce, &td->clip_table, link) { if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) return (ce); } return (NULL); } struct clip_entry * hold_lip(struct tom_data *td, struct in6_addr *lip) { struct clip_entry *ce; mtx_lock(&td->clip_table_lock); ce = search_lip(td, lip); if (ce != NULL) ce->refcount++; mtx_unlock(&td->clip_table_lock); return (ce); } void release_lip(struct tom_data *td, struct clip_entry *ce) { mtx_lock(&td->clip_table_lock); KASSERT(search_lip(td, &ce->lip) == ce, ("%s: CLIP entry %p p not in CLIP table.", __func__, ce)); KASSERT(ce->refcount > 0, ("%s: CLIP entry %p has refcount 0", __func__, ce)); --ce->refcount; mtx_unlock(&td->clip_table_lock); } static void init_clip_table(struct adapter *sc, struct tom_data *td) { ASSERT_SYNCHRONIZED_OP(sc); mtx_init(&td->clip_table_lock, "CLIP table lock", NULL, MTX_DEF); TAILQ_INIT(&td->clip_table); td->clip_gen = -1; update_clip_table(sc, td); } static void update_clip(struct adapter *sc, void *arg __unused) { if (begin_synchronized_op(sc, NULL, HOLD_LOCK, "t4tomuc")) return; if (uld_active(sc, ULD_TOM)) update_clip_table(sc, sc->tom_softc); end_synchronized_op(sc, LOCK_HELD); } static void t4_clip_task(void *arg, int count) { t4_iterate(update_clip, NULL); } static void update_clip_table(struct adapter *sc, struct tom_data *td) { struct rm_priotracker in6_ifa_tracker; struct in6_ifaddr *ia; struct in6_addr *lip, tlip; struct clip_head stale; struct clip_entry *ce, *ce_temp; int rc, gen = atomic_load_acq_int(&in6_ifaddr_gen); ASSERT_SYNCHRONIZED_OP(sc); IN6_IFADDR_RLOCK(&in6_ifa_tracker); mtx_lock(&td->clip_table_lock); if (gen == td->clip_gen) goto done; TAILQ_INIT(&stale); TAILQ_CONCAT(&stale, &td->clip_table, link); TAILQ_FOREACH(ia, &V_in6_ifaddrhead, ia_link) { lip = &ia->ia_addr.sin6_addr; KASSERT(!IN6_IS_ADDR_MULTICAST(lip), ("%s: mcast address in in6_ifaddr list", __func__)); if (IN6_IS_ADDR_LOOPBACK(lip)) continue; if (IN6_IS_SCOPE_EMBED(lip)) { /* Remove the embedded scope */ tlip = *lip; lip = &tlip; in6_clearscope(lip); } /* * XXX: how to weed out the link local address for the loopback * interface? It's fe80::1 usually (always?). */ /* * If it's in the main list then we already know it's not stale. */ TAILQ_FOREACH(ce, &td->clip_table, link) { if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) goto next; } /* * If it's in the stale list we should move it to the main list. */ TAILQ_FOREACH(ce, &stale, link) { if (IN6_ARE_ADDR_EQUAL(&ce->lip, lip)) { TAILQ_REMOVE(&stale, ce, link); TAILQ_INSERT_TAIL(&td->clip_table, ce, link); goto next; } } /* A new IP6 address; add it to the CLIP table */ ce = malloc(sizeof(*ce), M_CXGBE, M_NOWAIT); memcpy(&ce->lip, lip, sizeof(ce->lip)); ce->refcount = 0; rc = add_lip(sc, lip); if (rc == 0) TAILQ_INSERT_TAIL(&td->clip_table, ce, link); else { char ip[INET6_ADDRSTRLEN]; inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip)); log(LOG_ERR, "%s: could not add %s (%d)\n", __func__, ip, rc); free(ce, M_CXGBE); } next: continue; } /* * Remove stale addresses (those no longer in V_in6_ifaddrhead) that are * no longer referenced by the driver. */ TAILQ_FOREACH_SAFE(ce, &stale, link, ce_temp) { if (ce->refcount == 0) { rc = delete_lip(sc, &ce->lip); if (rc == 0) { TAILQ_REMOVE(&stale, ce, link); free(ce, M_CXGBE); } else { char ip[INET6_ADDRSTRLEN]; inet_ntop(AF_INET6, &ce->lip, &ip[0], sizeof(ip)); log(LOG_ERR, "%s: could not delete %s (%d)\n", __func__, ip, rc); } } } /* The ones that are still referenced need to stay in the CLIP table */ TAILQ_CONCAT(&td->clip_table, &stale, link); td->clip_gen = gen; done: mtx_unlock(&td->clip_table_lock); IN6_IFADDR_RUNLOCK(&in6_ifa_tracker); } static void destroy_clip_table(struct adapter *sc, struct tom_data *td) { struct clip_entry *ce, *ce_temp; if (mtx_initialized(&td->clip_table_lock)) { mtx_lock(&td->clip_table_lock); TAILQ_FOREACH_SAFE(ce, &td->clip_table, link, ce_temp) { KASSERT(ce->refcount == 0, ("%s: CLIP entry %p still in use (%d)", __func__, ce, ce->refcount)); TAILQ_REMOVE(&td->clip_table, ce, link); delete_lip(sc, &ce->lip); free(ce, M_CXGBE); } mtx_unlock(&td->clip_table_lock); mtx_destroy(&td->clip_table_lock); } } static void free_tom_data(struct adapter *sc, struct tom_data *td) { ASSERT_SYNCHRONIZED_OP(sc); KASSERT(TAILQ_EMPTY(&td->toep_list), ("%s: TOE PCB list is not empty.", __func__)); KASSERT(td->lctx_count == 0, ("%s: lctx hash table is not empty.", __func__)); t4_uninit_l2t_cpl_handlers(sc); t4_uninit_cpl_io_handlers(sc); t4_uninit_ddp(sc, td); destroy_clip_table(sc, td); if (td->listen_mask != 0) hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask); if (mtx_initialized(&td->unsent_wr_lock)) mtx_destroy(&td->unsent_wr_lock); if (mtx_initialized(&td->lctx_hash_lock)) mtx_destroy(&td->lctx_hash_lock); if (mtx_initialized(&td->toep_list_lock)) mtx_destroy(&td->toep_list_lock); free_tid_tabs(&sc->tids); free(td, M_CXGBE); } static void reclaim_wr_resources(void *arg, int count) { struct tom_data *td = arg; STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list); struct cpl_act_open_req *cpl; u_int opcode, atid; struct wrqe *wr; struct adapter *sc; mtx_lock(&td->unsent_wr_lock); STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe); mtx_unlock(&td->unsent_wr_lock); while ((wr = STAILQ_FIRST(&twr_list)) != NULL) { STAILQ_REMOVE_HEAD(&twr_list, link); cpl = wrtod(wr); opcode = GET_OPCODE(cpl); switch (opcode) { case CPL_ACT_OPEN_REQ: case CPL_ACT_OPEN_REQ6: atid = G_TID_TID(be32toh(OPCODE_TID(cpl))); sc = td_adapter(td); CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid); act_open_failure_cleanup(sc, atid, EHOSTUNREACH); free(wr, M_CXGBE); break; default: log(LOG_ERR, "%s: leaked work request %p, wr_len %d, " "opcode %x\n", __func__, wr, wr->wr_len, opcode); /* WR not freed here; go look at it with a debugger. */ } } } /* * Ground control to Major TOM * Commencing countdown, engines on */ static int t4_tom_activate(struct adapter *sc) { struct tom_data *td; struct toedev *tod; struct vi_info *vi; int i, rc, v; ASSERT_SYNCHRONIZED_OP(sc); /* per-adapter softc for TOM */ td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT); if (td == NULL) return (ENOMEM); /* List of TOE PCBs and associated lock */ mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF); TAILQ_INIT(&td->toep_list); /* Listen context */ mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF); td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE, &td->listen_mask, HASH_NOWAIT); /* List of WRs for which L2 resolution failed */ mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF); STAILQ_INIT(&td->unsent_wr_list); TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td); /* TID tables */ rc = alloc_tid_tabs(&sc->tids); if (rc != 0) goto done; /* DDP page pods and CPL handlers */ t4_init_ddp(sc, td); /* CLIP table for IPv6 offload */ init_clip_table(sc, td); /* CPL handlers */ t4_init_connect_cpl_handlers(sc); t4_init_l2t_cpl_handlers(sc); t4_init_listen_cpl_handlers(sc); t4_init_cpl_io_handlers(sc); /* toedev ops */ tod = &td->tod; init_toedev(tod); tod->tod_softc = sc; tod->tod_connect = t4_connect; tod->tod_listen_start = t4_listen_start; tod->tod_listen_stop = t4_listen_stop; tod->tod_rcvd = t4_rcvd; tod->tod_output = t4_tod_output; tod->tod_send_rst = t4_send_rst; tod->tod_send_fin = t4_send_fin; tod->tod_pcb_detach = t4_pcb_detach; tod->tod_l2_update = t4_l2_update; tod->tod_syncache_added = t4_syncache_added; tod->tod_syncache_removed = t4_syncache_removed; tod->tod_syncache_respond = t4_syncache_respond; tod->tod_offload_socket = t4_offload_socket; tod->tod_ctloutput = t4_ctloutput; for_each_port(sc, i) { for_each_vi(sc->port[i], v, vi) { TOEDEV(vi->ifp) = &td->tod; } } sc->tom_softc = td; register_toedev(sc->tom_softc); done: if (rc != 0) free_tom_data(sc, td); return (rc); } static int t4_tom_deactivate(struct adapter *sc) { int rc = 0; struct tom_data *td = sc->tom_softc; ASSERT_SYNCHRONIZED_OP(sc); if (td == NULL) return (0); /* XXX. KASSERT? */ if (sc->offload_map != 0) return (EBUSY); /* at least one port has IFCAP_TOE enabled */ if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI)) return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */ mtx_lock(&td->toep_list_lock); if (!TAILQ_EMPTY(&td->toep_list)) rc = EBUSY; mtx_unlock(&td->toep_list_lock); mtx_lock(&td->lctx_hash_lock); if (td->lctx_count > 0) rc = EBUSY; mtx_unlock(&td->lctx_hash_lock); taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources); mtx_lock(&td->unsent_wr_lock); if (!STAILQ_EMPTY(&td->unsent_wr_list)) rc = EBUSY; mtx_unlock(&td->unsent_wr_lock); if (rc == 0) { unregister_toedev(sc->tom_softc); free_tom_data(sc, td); sc->tom_softc = NULL; } return (rc); } static void t4_tom_ifaddr_event(void *arg __unused, struct ifnet *ifp) { atomic_add_rel_int(&in6_ifaddr_gen, 1); taskqueue_enqueue_timeout(taskqueue_thread, &clip_task, -hz / 4); } static int t4_tom_mod_load(void) { int rc; struct protosw *tcp_protosw, *tcp6_protosw; tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM); if (tcp_protosw == NULL) return (ENOPROTOOPT); bcopy(tcp_protosw, &ddp_protosw, sizeof(ddp_protosw)); bcopy(tcp_protosw->pr_usrreqs, &ddp_usrreqs, sizeof(ddp_usrreqs)); ddp_usrreqs.pru_soreceive = t4_soreceive_ddp; ddp_protosw.pr_usrreqs = &ddp_usrreqs; tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM); if (tcp6_protosw == NULL) return (ENOPROTOOPT); bcopy(tcp6_protosw, &ddp6_protosw, sizeof(ddp6_protosw)); bcopy(tcp6_protosw->pr_usrreqs, &ddp6_usrreqs, sizeof(ddp6_usrreqs)); ddp6_usrreqs.pru_soreceive = t4_soreceive_ddp; ddp6_protosw.pr_usrreqs = &ddp6_usrreqs; TIMEOUT_TASK_INIT(taskqueue_thread, &clip_task, 0, t4_clip_task, NULL); ifaddr_evhandler = EVENTHANDLER_REGISTER(ifaddr_event, t4_tom_ifaddr_event, NULL, EVENTHANDLER_PRI_ANY); rc = t4_register_uld(&tom_uld_info); if (rc != 0) t4_tom_mod_unload(); return (rc); } static void tom_uninit(struct adapter *sc, void *arg __unused) { if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun")) return; /* Try to free resources (works only if no port has IFCAP_TOE) */ if (uld_active(sc, ULD_TOM)) t4_deactivate_uld(sc, ULD_TOM); end_synchronized_op(sc, 0); } static int t4_tom_mod_unload(void) { t4_iterate(tom_uninit, NULL); if (t4_unregister_uld(&tom_uld_info) == EBUSY) return (EBUSY); if (ifaddr_evhandler) { EVENTHANDLER_DEREGISTER(ifaddr_event, ifaddr_evhandler); taskqueue_cancel_timeout(taskqueue_thread, &clip_task, NULL); } return (0); } #endif /* TCP_OFFLOAD */ static int t4_tom_modevent(module_t mod, int cmd, void *arg) { int rc = 0; #ifdef TCP_OFFLOAD switch (cmd) { case MOD_LOAD: rc = t4_tom_mod_load(); break; case MOD_UNLOAD: rc = t4_tom_mod_unload(); break; default: rc = EINVAL; } #else printf("t4_tom: compiled without TCP_OFFLOAD support.\n"); rc = EOPNOTSUPP; #endif return (rc); } static moduledata_t t4_tom_moddata= { "t4_tom", t4_tom_modevent, 0 }; MODULE_VERSION(t4_tom, 1); MODULE_DEPEND(t4_tom, toecore, 1, 1, 1); MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1); DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY); Index: head/sys/i386/linux/linux_sysvec.c =================================================================== --- head/sys/i386/linux/linux_sysvec.c (revision 298481) +++ head/sys/i386/linux/linux_sysvec.c (revision 298482) @@ -1,1200 +1,1200 @@ /*- * Copyright (c) 1994-1996 Søren Schmidt * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer * in this position and unchanged. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include MODULE_VERSION(linux, 1); #if BYTE_ORDER == LITTLE_ENDIAN #define SHELLMAGIC 0x2123 /* #! */ #else #define SHELLMAGIC 0x2321 #endif #if defined(DEBUG) SYSCTL_PROC(_compat_linux, OID_AUTO, debug, CTLTYPE_STRING | CTLFLAG_RW, 0, 0, linux_sysctl_debug, "A", "Linux debugging control"); #endif /* * Allow the sendsig functions to use the ldebug() facility * even though they are not syscalls themselves. Map them * to syscall 0. This is slightly less bogus than using * ldebug(sigreturn). */ #define LINUX_SYS_linux_rt_sendsig 0 #define LINUX_SYS_linux_sendsig 0 #define LINUX_PS_STRINGS (LINUX_USRSTACK - sizeof(struct ps_strings)) static int linux_szsigcode; static vm_object_t linux_shared_page_obj; static char *linux_shared_page_mapping; extern char _binary_linux_locore_o_start; extern char _binary_linux_locore_o_end; extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL]; SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler); static int linux_fixup(register_t **stack_base, struct image_params *iparams); static int elf_linux_fixup(register_t **stack_base, struct image_params *iparams); static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask); static void exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack); static register_t *linux_copyout_strings(struct image_params *imgp); static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel); static void linux_vdso_install(void *param); static void linux_vdso_deinstall(void *param); static int linux_szplatform; const char *linux_kplatform; static eventhandler_tag linux_exit_tag; static eventhandler_tag linux_exec_tag; static eventhandler_tag linux_thread_dtor_tag; /* * Linux syscalls return negative errno's, we do positive and map them * Reference: * FreeBSD: src/sys/sys/errno.h * Linux: linux-2.6.17.8/include/asm-generic/errno-base.h * linux-2.6.17.8/include/asm-generic/errno.h */ static int bsd_to_linux_errno[ELAST + 1] = { -0, -1, -2, -3, -4, -5, -6, -7, -8, -9, -10, -35, -12, -13, -14, -15, -16, -17, -18, -19, -20, -21, -22, -23, -24, -25, -26, -27, -28, -29, -30, -31, -32, -33, -34, -11,-115,-114, -88, -89, -90, -91, -92, -93, -94, -95, -96, -97, -98, -99, -100,-101,-102,-103,-104,-105,-106,-107,-108,-109, -110,-111, -40, -36,-112,-113, -39, -11, -87,-122, -116, -66, -6, -6, -6, -6, -6, -37, -38, -9, -6, -6, -43, -42, -75,-125, -84, -95, -16, -74, -72, -67, -71 }; #define LINUX_T_UNKNOWN 255 static int _bsd_to_linux_trapcode[] = { LINUX_T_UNKNOWN, /* 0 */ 6, /* 1 T_PRIVINFLT */ LINUX_T_UNKNOWN, /* 2 */ 3, /* 3 T_BPTFLT */ LINUX_T_UNKNOWN, /* 4 */ LINUX_T_UNKNOWN, /* 5 */ 16, /* 6 T_ARITHTRAP */ 254, /* 7 T_ASTFLT */ LINUX_T_UNKNOWN, /* 8 */ 13, /* 9 T_PROTFLT */ 1, /* 10 T_TRCTRAP */ LINUX_T_UNKNOWN, /* 11 */ 14, /* 12 T_PAGEFLT */ LINUX_T_UNKNOWN, /* 13 */ 17, /* 14 T_ALIGNFLT */ LINUX_T_UNKNOWN, /* 15 */ LINUX_T_UNKNOWN, /* 16 */ LINUX_T_UNKNOWN, /* 17 */ 0, /* 18 T_DIVIDE */ 2, /* 19 T_NMI */ 4, /* 20 T_OFLOW */ 5, /* 21 T_BOUND */ 7, /* 22 T_DNA */ 8, /* 23 T_DOUBLEFLT */ 9, /* 24 T_FPOPFLT */ 10, /* 25 T_TSSFLT */ 11, /* 26 T_SEGNPFLT */ 12, /* 27 T_STKFLT */ 18, /* 28 T_MCHK */ 19, /* 29 T_XMMFLT */ 15 /* 30 T_RESERVED */ }; #define bsd_to_linux_trapcode(code) \ ((code)args->argc + 1); (*stack_base)--; suword(*stack_base, (intptr_t)(void *)envp); (*stack_base)--; suword(*stack_base, (intptr_t)(void *)argv); (*stack_base)--; suword(*stack_base, imgp->args->argc); return (0); } static int elf_linux_fixup(register_t **stack_base, struct image_params *imgp) { struct proc *p; Elf32_Auxargs *args; Elf32_Addr *uplatform; struct ps_strings *arginfo; register_t *pos; int issetugid; KASSERT(curthread->td_proc == imgp->proc, ("unsafe elf_linux_fixup(), should be curproc")); p = imgp->proc; issetugid = imgp->proc->p_flag & P_SUGID ? 1 : 0; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; uplatform = (Elf32_Addr *)((caddr_t)arginfo - linux_szplatform); args = (Elf32_Auxargs *)imgp->auxargs; pos = *stack_base + (imgp->args->argc + imgp->args->envc + 2); AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, imgp->proc->p_sysent->sv_shared_page_base); AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO, linux_vsyscall); AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature); /* * Do not export AT_CLKTCK when emulating Linux kernel prior to 2.4.0, * as it has appeared in the 2.4.0-rc7 first time. * Being exported, AT_CLKTCK is returned by sysconf(_SC_CLK_TCK), * glibc falls back to the hard-coded CLK_TCK value when aux entry * is not present. * Also see linux_times() implementation. */ if (linux_kernver(curthread) >= LINUX_KERNVER_2004000) AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz); AUXARGS_ENTRY(pos, AT_PHDR, args->phdr); AUXARGS_ENTRY(pos, AT_PHENT, args->phent); AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum); AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz); AUXARGS_ENTRY(pos, AT_FLAGS, args->flags); AUXARGS_ENTRY(pos, AT_ENTRY, args->entry); AUXARGS_ENTRY(pos, AT_BASE, args->base); AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid); AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid); AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid); AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid); AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid); AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(uplatform)); AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary); if (imgp->execpathp != 0) AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp); if (args->execfd != -1) AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd); AUXARGS_ENTRY(pos, AT_NULL, 0); free(imgp->auxargs, M_TEMP); imgp->auxargs = NULL; (*stack_base)--; suword(*stack_base, (register_t)imgp->args->argc); return (0); } /* * Copied from kern/kern_exec.c */ static register_t * linux_copyout_strings(struct image_params *imgp) { int argc, envc; char **vectp; char *stringp, *destp; register_t *stack_base; struct ps_strings *arginfo; char canary[LINUX_AT_RANDOM_LEN]; size_t execpath_len; struct proc *p; /* * Calculate string base and vector table pointers. */ p = imgp->proc; if (imgp->execpath != NULL && imgp->auxargs != NULL) execpath_len = strlen(imgp->execpath) + 1; else execpath_len = 0; arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; destp = (caddr_t)arginfo - SPARE_USRSPACE - linux_szplatform - roundup(sizeof(canary), sizeof(char *)) - roundup(execpath_len, sizeof(char *)) - - roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); + roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *)); /* * install LINUX_PLATFORM */ copyout(linux_kplatform, ((caddr_t)arginfo - linux_szplatform), linux_szplatform); if (execpath_len != 0) { imgp->execpathp = (uintptr_t)arginfo - linux_szplatform - execpath_len; copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); } /* * Prepare the canary for SSP. */ arc4rand(canary, sizeof(canary), 0); imgp->canary = (uintptr_t)arginfo - linux_szplatform - roundup(execpath_len, sizeof(char *)) - roundup(sizeof(canary), sizeof(char *)); copyout(canary, (void *)imgp->canary, sizeof(canary)); /* * If we have a valid auxargs ptr, prepare some room * on the stack. */ if (imgp->auxargs) { /* * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for * lower compatibility. */ imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : (LINUX_AT_COUNT * 2); /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets,and imgp->auxarg_size is room * for argument of Runtime loader. */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2 + imgp->auxarg_size) * sizeof(char *)); } else { /* * The '+ 2' is for the null pointers at the end of each of * the arg and env vector sets */ vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * sizeof(char *)); } /* * vectp also becomes our initial stack base */ stack_base = (register_t *)vectp; stringp = imgp->args->begin_argv; argc = imgp->args->argc; envc = imgp->args->envc; /* * Copy out strings - arguments and environment. */ copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); /* * Fill in "ps_strings" struct for ps, w, etc. */ suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nargvstr, argc); /* * Fill in argument portion of vector table. */ for (; argc > 0; --argc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* a null vector table pointer separates the argp's from the envp's */ suword(vectp++, 0); suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); suword(&arginfo->ps_nenvstr, envc); /* * Fill in environment portion of vector table. */ for (; envc > 0; --envc) { suword(vectp++, (long)(intptr_t)destp); while (*stringp++ != 0) destp++; destp++; } /* end of vector table is a null pointer */ suword(vectp, 0); return (stack_base); } static void linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_rt_sigframe *fp, frame; int sig, code; int oonstack; sig = ksi->ksi_signo; code = ksi->ksi_code; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; mtx_assert(&psp->ps_mtx, MA_OWNED); regs = td->td_frame; oonstack = sigonstack(regs->tf_esp); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(ARGS(rt_sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_rt_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_rt_sigframe)); } else fp = (struct l_rt_sigframe *)regs->tf_esp - 1; mtx_unlock(&psp->ps_mtx); /* * Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_handler = catcher; frame.sf_sig = sig; frame.sf_siginfo = &fp->sf_si; frame.sf_ucontext = &fp->sf_sc; /* Fill in POSIX parts */ ksiginfo_to_lsiginfo(ksi, &frame.sf_si, sig); /* * Build the signal context to be used by sigreturn. */ frame.sf_sc.uc_flags = 0; /* XXX ??? */ frame.sf_sc.uc_link = NULL; /* XXX ??? */ frame.sf_sc.uc_stack.ss_sp = td->td_sigstk.ss_sp; frame.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size; frame.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK) ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE; PROC_UNLOCK(p); bsd_to_linux_sigset(mask, &frame.sf_sc.uc_sigmask); frame.sf_sc.uc_mcontext.sc_mask = frame.sf_sc.uc_sigmask.__mask; frame.sf_sc.uc_mcontext.sc_gs = rgs(); frame.sf_sc.uc_mcontext.sc_fs = regs->tf_fs; frame.sf_sc.uc_mcontext.sc_es = regs->tf_es; frame.sf_sc.uc_mcontext.sc_ds = regs->tf_ds; frame.sf_sc.uc_mcontext.sc_edi = regs->tf_edi; frame.sf_sc.uc_mcontext.sc_esi = regs->tf_esi; frame.sf_sc.uc_mcontext.sc_ebp = regs->tf_ebp; frame.sf_sc.uc_mcontext.sc_ebx = regs->tf_ebx; frame.sf_sc.uc_mcontext.sc_esp = regs->tf_esp; frame.sf_sc.uc_mcontext.sc_edx = regs->tf_edx; frame.sf_sc.uc_mcontext.sc_ecx = regs->tf_ecx; frame.sf_sc.uc_mcontext.sc_eax = regs->tf_eax; frame.sf_sc.uc_mcontext.sc_eip = regs->tf_eip; frame.sf_sc.uc_mcontext.sc_cs = regs->tf_cs; frame.sf_sc.uc_mcontext.sc_eflags = regs->tf_eflags; frame.sf_sc.uc_mcontext.sc_esp_at_signal = regs->tf_esp; frame.sf_sc.uc_mcontext.sc_ss = regs->tf_ss; frame.sf_sc.uc_mcontext.sc_err = regs->tf_err; frame.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr; frame.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code); #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"), frame.sf_sc.uc_stack.ss_flags, td->td_sigstk.ss_sp, td->td_sigstk.ss_size, frame.sf_sc.uc_mcontext.sc_mask); #endif if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ #ifdef DEBUG if (ldebug(rt_sendsig)) printf(LMSG("rt_sendsig: bad stack %p, oonstack=%x"), fp, oonstack); #endif PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_esp = (int)fp; regs->tf_eip = linux_rt_sigcode; regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _udatasel; regs->tf_ss = _udatasel; PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * Send an interrupt to process. * * Stack is set up to allow sigcode stored * in u. to call routine, followed by kcall * to sigreturn routine below. After sigreturn * resets the signal mask, the stack, and the * frame pointer, it returns to the user * specified pc, psl. */ static void linux_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) { struct thread *td = curthread; struct proc *p = td->td_proc; struct sigacts *psp; struct trapframe *regs; struct l_sigframe *fp, frame; l_sigset_t lmask; int sig, code; int oonstack; PROC_LOCK_ASSERT(p, MA_OWNED); psp = p->p_sigacts; sig = ksi->ksi_signo; code = ksi->ksi_code; mtx_assert(&psp->ps_mtx, MA_OWNED); if (SIGISMEMBER(psp->ps_siginfo, sig)) { /* Signal handler installed with SA_SIGINFO. */ linux_rt_sendsig(catcher, ksi, mask); return; } regs = td->td_frame; oonstack = sigonstack(regs->tf_esp); #ifdef DEBUG if (ldebug(sendsig)) printf(ARGS(sendsig, "%p, %d, %p, %u"), catcher, sig, (void*)mask, code); #endif /* * Allocate space for the signal handler context. */ if ((td->td_pflags & TDP_ALTSTACK) && !oonstack && SIGISMEMBER(psp->ps_sigonstack, sig)) { fp = (struct l_sigframe *)((uintptr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size - sizeof(struct l_sigframe)); } else fp = (struct l_sigframe *)regs->tf_esp - 1; mtx_unlock(&psp->ps_mtx); PROC_UNLOCK(p); /* * Build the argument list for the signal handler. */ sig = bsd_to_linux_signal(sig); bzero(&frame, sizeof(frame)); frame.sf_handler = catcher; frame.sf_sig = sig; bsd_to_linux_sigset(mask, &lmask); /* * Build the signal context to be used by sigreturn. */ frame.sf_sc.sc_mask = lmask.__mask; frame.sf_sc.sc_gs = rgs(); frame.sf_sc.sc_fs = regs->tf_fs; frame.sf_sc.sc_es = regs->tf_es; frame.sf_sc.sc_ds = regs->tf_ds; frame.sf_sc.sc_edi = regs->tf_edi; frame.sf_sc.sc_esi = regs->tf_esi; frame.sf_sc.sc_ebp = regs->tf_ebp; frame.sf_sc.sc_ebx = regs->tf_ebx; frame.sf_sc.sc_esp = regs->tf_esp; frame.sf_sc.sc_edx = regs->tf_edx; frame.sf_sc.sc_ecx = regs->tf_ecx; frame.sf_sc.sc_eax = regs->tf_eax; frame.sf_sc.sc_eip = regs->tf_eip; frame.sf_sc.sc_cs = regs->tf_cs; frame.sf_sc.sc_eflags = regs->tf_eflags; frame.sf_sc.sc_esp_at_signal = regs->tf_esp; frame.sf_sc.sc_ss = regs->tf_ss; frame.sf_sc.sc_err = regs->tf_err; frame.sf_sc.sc_cr2 = (register_t)ksi->ksi_addr; frame.sf_sc.sc_trapno = bsd_to_linux_trapcode(ksi->ksi_trapno); frame.sf_extramask[0] = lmask.__mask; if (copyout(&frame, fp, sizeof(frame)) != 0) { /* * Process has trashed its stack; give it an illegal * instruction to halt it in its tracks. */ PROC_LOCK(p); sigexit(td, SIGILL); } /* * Build context to run handler in. */ regs->tf_esp = (int)fp; regs->tf_eip = linux_sigcode; regs->tf_eflags &= ~(PSL_T | PSL_VM | PSL_D); regs->tf_cs = _ucodesel; regs->tf_ds = _udatasel; regs->tf_es = _udatasel; regs->tf_fs = _udatasel; regs->tf_ss = _udatasel; PROC_LOCK(p); mtx_lock(&psp->ps_mtx); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * psl to gain improper privileges or to cause * a machine fault. */ int linux_sigreturn(struct thread *td, struct linux_sigreturn_args *args) { struct l_sigframe frame; struct trapframe *regs; l_sigset_t lmask; sigset_t bmask; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(sigreturn)) printf(ARGS(sigreturn, "%p"), (void *)args->sfp); #endif /* * The trampoline code hands us the sigframe. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->sfp, &frame, sizeof(frame)) != 0) return (EFAULT); /* * Check for security violations. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = frame.sf_sc.sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_eflags)) return (EINVAL); /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(frame.sf_sc.sc_cs)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_eip; trapsignal(td, &ksi); return (EINVAL); } lmask.__mask = frame.sf_sc.sc_mask; linux_to_bsd_sigset(&lmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* * Restore signal context. */ /* %gs was restored by the trampoline. */ regs->tf_fs = frame.sf_sc.sc_fs; regs->tf_es = frame.sf_sc.sc_es; regs->tf_ds = frame.sf_sc.sc_ds; regs->tf_edi = frame.sf_sc.sc_edi; regs->tf_esi = frame.sf_sc.sc_esi; regs->tf_ebp = frame.sf_sc.sc_ebp; regs->tf_ebx = frame.sf_sc.sc_ebx; regs->tf_edx = frame.sf_sc.sc_edx; regs->tf_ecx = frame.sf_sc.sc_ecx; regs->tf_eax = frame.sf_sc.sc_eax; regs->tf_eip = frame.sf_sc.sc_eip; regs->tf_cs = frame.sf_sc.sc_cs; regs->tf_eflags = eflags; regs->tf_esp = frame.sf_sc.sc_esp_at_signal; regs->tf_ss = frame.sf_sc.sc_ss; return (EJUSTRETURN); } /* * System call to cleanup state after a signal * has been taken. Reset signal mask and * stack state from context left by rt_sendsig (above). * Return to previous pc and psl as specified by * context left by sendsig. Check carefully to * make sure that the user has not modified the * psl to gain improper privileges or to cause * a machine fault. */ int linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args) { struct l_ucontext uc; struct l_sigcontext *context; sigset_t bmask; l_stack_t *lss; stack_t ss; struct trapframe *regs; int eflags; ksiginfo_t ksi; regs = td->td_frame; #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(ARGS(rt_sigreturn, "%p"), (void *)args->ucp); #endif /* * The trampoline code hands us the ucontext. * It is unsafe to keep track of it ourselves, in the event that a * program jumps out of a signal handler. */ if (copyin(args->ucp, &uc, sizeof(uc)) != 0) return (EFAULT); context = &uc.uc_mcontext; /* * Check for security violations. */ #define EFLAGS_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0) eflags = context->sc_eflags; if (!EFLAGS_SECURE(eflags, regs->tf_eflags)) return (EINVAL); /* * Don't allow users to load a valid privileged %cs. Let the * hardware check for invalid selectors, excess privilege in * other selectors, invalid %eip's and invalid %esp's. */ #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL) if (!CS_SECURE(context->sc_cs)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGBUS; ksi.ksi_code = BUS_OBJERR; ksi.ksi_trapno = T_PROTFLT; ksi.ksi_addr = (void *)regs->tf_eip; trapsignal(td, &ksi); return (EINVAL); } linux_to_bsd_sigset(&uc.uc_sigmask, &bmask); kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0); /* * Restore signal context */ /* %gs was restored by the trampoline. */ regs->tf_fs = context->sc_fs; regs->tf_es = context->sc_es; regs->tf_ds = context->sc_ds; regs->tf_edi = context->sc_edi; regs->tf_esi = context->sc_esi; regs->tf_ebp = context->sc_ebp; regs->tf_ebx = context->sc_ebx; regs->tf_edx = context->sc_edx; regs->tf_ecx = context->sc_ecx; regs->tf_eax = context->sc_eax; regs->tf_eip = context->sc_eip; regs->tf_cs = context->sc_cs; regs->tf_eflags = eflags; regs->tf_esp = context->sc_esp_at_signal; regs->tf_ss = context->sc_ss; /* * call sigaltstack & ignore results.. */ lss = &uc.uc_stack; ss.ss_sp = lss->ss_sp; ss.ss_size = lss->ss_size; ss.ss_flags = linux_to_bsd_sigaltstack(lss->ss_flags); #ifdef DEBUG if (ldebug(rt_sigreturn)) printf(LMSG("rt_sigret flags: 0x%x, sp: %p, ss: 0x%x, mask: 0x%x"), ss.ss_flags, ss.ss_sp, ss.ss_size, context->sc_mask); #endif (void)kern_sigaltstack(td, &ss, NULL); return (EJUSTRETURN); } static int linux_fetch_syscall_args(struct thread *td, struct syscall_args *sa) { struct proc *p; struct trapframe *frame; p = td->td_proc; frame = td->td_frame; sa->code = frame->tf_eax; sa->args[0] = frame->tf_ebx; sa->args[1] = frame->tf_ecx; sa->args[2] = frame->tf_edx; sa->args[3] = frame->tf_esi; sa->args[4] = frame->tf_edi; sa->args[5] = frame->tf_ebp; /* Unconfirmed */ if (sa->code >= p->p_sysent->sv_size) /* nosys */ sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1]; else sa->callp = &p->p_sysent->sv_table[sa->code]; sa->narg = sa->callp->sy_narg; td->td_retval[0] = 0; td->td_retval[1] = frame->tf_edx; return (0); } /* * If a linux binary is exec'ing something, try this image activator * first. We override standard shell script execution in order to * be able to modify the interpreter path. We only do this if a linux * binary is doing the exec, so we do not create an EXEC module for it. */ static int exec_linux_imgact_try(struct image_params *iparams); static int exec_linux_imgact_try(struct image_params *imgp) { const char *head = (const char *)imgp->image_header; char *rpath; int error = -1; /* * The interpreter for shell scripts run from a linux binary needs * to be located in /compat/linux if possible in order to recursively * maintain linux path emulation. */ if (((const short *)head)[0] == SHELLMAGIC) { /* * Run our normal shell image activator. If it succeeds attempt * to use the alternate path for the interpreter. If an alternate * path is found, use our stringspace to store it. */ if ((error = exec_shell_imgact(imgp)) == 0) { linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc), imgp->interpreter_name, UIO_SYSSPACE, &rpath, 0, AT_FDCWD); if (rpath != NULL) imgp->args->fname_buf = imgp->interpreter_name = rpath; } } return (error); } /* * exec_setregs may initialize some registers differently than Linux * does, thus potentially confusing Linux binaries. If necessary, we * override the exec_setregs default(s) here. */ static void exec_linux_setregs(struct thread *td, struct image_params *imgp, u_long stack) { struct pcb *pcb = td->td_pcb; exec_setregs(td, imgp, stack); /* Linux sets %gs to 0, we default to _udatasel */ pcb->pcb_gs = 0; load_gs(0); pcb->pcb_initial_npxcw = __LINUX_NPXCW__; } static void linux_get_machine(const char **dst) { switch (cpu_class) { case CPUCLASS_686: *dst = "i686"; break; case CPUCLASS_586: *dst = "i586"; break; case CPUCLASS_486: *dst = "i486"; break; default: *dst = "i386"; } } struct sysentvec linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux_locore_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux a.out", .sv_coredump = NULL, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = LINUX_USRSTACK, .sv_psstrings = PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = exec_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_AOUT | SV_IA32 | SV_ILP32, .sv_set_syscall_retval = cpu_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .sv_shared_page_base = LINUX_SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, }; INIT_SYSENTVEC(aout_sysvec, &linux_sysvec); struct sysentvec elf_linux_sysvec = { .sv_size = LINUX_SYS_MAXSYSCALL, .sv_table = linux_sysent, .sv_mask = 0, .sv_errsize = ELAST + 1, .sv_errtbl = bsd_to_linux_errno, .sv_transtrap = translate_traps, .sv_fixup = elf_linux_fixup, .sv_sendsig = linux_sendsig, .sv_sigcode = &_binary_linux_locore_o_start, .sv_szsigcode = &linux_szsigcode, .sv_name = "Linux ELF", .sv_coredump = elf32_coredump, .sv_imgact_try = exec_linux_imgact_try, .sv_minsigstksz = LINUX_MINSIGSTKSZ, .sv_pagesize = PAGE_SIZE, .sv_minuser = VM_MIN_ADDRESS, .sv_maxuser = VM_MAXUSER_ADDRESS, .sv_usrstack = LINUX_USRSTACK, .sv_psstrings = LINUX_PS_STRINGS, .sv_stackprot = VM_PROT_ALL, .sv_copyout_strings = linux_copyout_strings, .sv_setregs = exec_linux_setregs, .sv_fixlimit = NULL, .sv_maxssiz = NULL, .sv_flags = SV_ABI_LINUX | SV_IA32 | SV_ILP32 | SV_SHP, .sv_set_syscall_retval = cpu_set_syscall_retval, .sv_fetch_syscall_args = linux_fetch_syscall_args, .sv_syscallnames = NULL, .sv_shared_page_base = LINUX_SHAREDPAGE, .sv_shared_page_len = PAGE_SIZE, .sv_schedtail = linux_schedtail, .sv_thread_detach = linux_thread_detach, .sv_trap = NULL, }; static void linux_vdso_install(void *param) { linux_szsigcode = (&_binary_linux_locore_o_end - &_binary_linux_locore_o_start); if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len) panic("Linux invalid vdso size\n"); __elfN(linux_vdso_fixup)(&elf_linux_sysvec); linux_shared_page_obj = __elfN(linux_shared_page_init) (&linux_shared_page_mapping); __elfN(linux_vdso_reloc)(&elf_linux_sysvec, LINUX_SHAREDPAGE); bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping, linux_szsigcode); elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj; } SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY, (sysinit_cfunc_t)linux_vdso_install, NULL); static void linux_vdso_deinstall(void *param) { __elfN(linux_shared_page_fini)(linux_shared_page_obj); }; SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)linux_vdso_deinstall, NULL); static char GNU_ABI_VENDOR[] = "GNU"; static int GNULINUX_ABI_DESC = 0; static boolean_t linux_trans_osrel(const Elf_Note *note, int32_t *osrel) { const Elf32_Word *desc; uintptr_t p; p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); desc = (const Elf32_Word *)p; if (desc[0] != GNULINUX_ABI_DESC) return (FALSE); /* * For linux we encode osrel as follows (see linux_mib.c): * VVVMMMIII (version, major, minor), see linux_mib.c. */ *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3]; return (TRUE); } static Elf_Brandnote linux_brandnote = { .hdr.n_namesz = sizeof(GNU_ABI_VENDOR), .hdr.n_descsz = 16, /* XXX at least 16 */ .hdr.n_type = 1, .vendor = GNU_ABI_VENDOR, .flags = BN_TRANSLATE_OSREL, .trans_osrel = linux_trans_osrel }; static Elf32_Brandinfo linux_brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.1", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; static Elf32_Brandinfo linux_glibc2brand = { .brand = ELFOSABI_LINUX, .machine = EM_386, .compat_3_brand = "Linux", .emul_path = "/compat/linux", .interp_path = "/lib/ld-linux.so.2", .sysvec = &elf_linux_sysvec, .interp_newpath = NULL, .brand_note = &linux_brandnote, .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE }; Elf32_Brandinfo *linux_brandlist[] = { &linux_brand, &linux_glibc2brand, NULL }; static int linux_elf_modevent(module_t mod, int type, void *data) { Elf32_Brandinfo **brandinfo; int error; struct linux_ioctl_handler **lihp; error = 0; switch(type) { case MOD_LOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_insert_brand_entry(*brandinfo) < 0) error = EINVAL; if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_register_handler(*lihp); LIST_INIT(&futex_list); mtx_init(&futex_mtx, "ftllk", NULL, MTX_DEF); linux_exit_tag = EVENTHANDLER_REGISTER(process_exit, linux_proc_exit, NULL, 1000); linux_exec_tag = EVENTHANDLER_REGISTER(process_exec, linux_proc_exec, NULL, 1000); linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor, linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY); linux_get_machine(&linux_kplatform); linux_szplatform = roundup(strlen(linux_kplatform) + 1, sizeof(char *)); linux_osd_jail_register(); stclohz = (stathz ? stathz : hz); if (bootverbose) printf("Linux ELF exec handler installed\n"); } else printf("cannot insert Linux ELF brand handler\n"); break; case MOD_UNLOAD: for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_brand_inuse(*brandinfo)) error = EBUSY; if (error == 0) { for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL; ++brandinfo) if (elf32_remove_brand_entry(*brandinfo) < 0) error = EINVAL; } if (error == 0) { SET_FOREACH(lihp, linux_ioctl_handler_set) linux_ioctl_unregister_handler(*lihp); mtx_destroy(&futex_mtx); EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag); EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag); EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag); linux_osd_jail_deregister(); if (bootverbose) printf("Linux ELF exec handler removed\n"); } else printf("Could not deinstall ELF interpreter entry\n"); break; default: return (EOPNOTSUPP); } return (error); } static moduledata_t linux_elf_mod = { "linuxelf", linux_elf_modevent, 0 }; DECLARE_MODULE_TIED(linuxelf, linux_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY); FEATURE(linux, "Linux 32bit support"); Index: head/sys/vm/vm_radix.c =================================================================== --- head/sys/vm/vm_radix.c (revision 298481) +++ head/sys/vm/vm_radix.c (revision 298482) @@ -1,856 +1,856 @@ /* * Copyright (c) 2013 EMC Corp. * Copyright (c) 2011 Jeffrey Roberson * Copyright (c) 2008 Mayur Shardul * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ /* * Path-compressed radix trie implementation. * The following code is not generalized into a general purpose library * because there are way too many parameters embedded that should really * be decided by the library consumers. At the same time, consumers * of this code must achieve highest possible performance. * * The implementation takes into account the following rationale: * - Size of the nodes should be as small as possible but still big enough * to avoid a large maximum depth for the trie. This is a balance * between the necessity to not wire too much physical memory for the nodes * and the necessity to avoid too much cache pollution during the trie * operations. * - There is not a huge bias toward the number of lookup operations over * the number of insert and remove operations. This basically implies * that optimizations supposedly helping one operation but hurting the * other might be carefully evaluated. * - On average not many nodes are expected to be fully populated, hence * level compression may just complicate things. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif /* * These widths should allow the pointers to a node's children to fit within * a single cache line. The extra levels from a narrow width should not be * a problem thanks to path compression. */ #ifdef __LP64__ #define VM_RADIX_WIDTH 4 #else #define VM_RADIX_WIDTH 3 #endif #define VM_RADIX_COUNT (1 << VM_RADIX_WIDTH) #define VM_RADIX_MASK (VM_RADIX_COUNT - 1) #define VM_RADIX_LIMIT \ - (howmany((sizeof(vm_pindex_t) * NBBY), VM_RADIX_WIDTH) - 1) + (howmany(sizeof(vm_pindex_t) * NBBY, VM_RADIX_WIDTH) - 1) /* Flag bits stored in node pointers. */ #define VM_RADIX_ISLEAF 0x1 #define VM_RADIX_FLAGS 0x1 #define VM_RADIX_PAD VM_RADIX_FLAGS /* Returns one unit associated with specified level. */ #define VM_RADIX_UNITLEVEL(lev) \ ((vm_pindex_t)1 << ((lev) * VM_RADIX_WIDTH)) struct vm_radix_node { vm_pindex_t rn_owner; /* Owner of record. */ uint16_t rn_count; /* Valid children. */ uint16_t rn_clev; /* Current level. */ void *rn_child[VM_RADIX_COUNT]; /* Child nodes. */ }; static uma_zone_t vm_radix_node_zone; /* * Allocate a radix node. */ static __inline struct vm_radix_node * vm_radix_node_get(vm_pindex_t owner, uint16_t count, uint16_t clevel) { struct vm_radix_node *rnode; rnode = uma_zalloc(vm_radix_node_zone, M_NOWAIT | M_ZERO); if (rnode == NULL) return (NULL); rnode->rn_owner = owner; rnode->rn_count = count; rnode->rn_clev = clevel; return (rnode); } /* * Free radix node. */ static __inline void vm_radix_node_put(struct vm_radix_node *rnode) { uma_zfree(vm_radix_node_zone, rnode); } /* * Return the position in the array for a given level. */ static __inline int vm_radix_slot(vm_pindex_t index, uint16_t level) { return ((index >> (level * VM_RADIX_WIDTH)) & VM_RADIX_MASK); } /* Trims the key after the specified level. */ static __inline vm_pindex_t vm_radix_trimkey(vm_pindex_t index, uint16_t level) { vm_pindex_t ret; ret = index; if (level > 0) { ret >>= level * VM_RADIX_WIDTH; ret <<= level * VM_RADIX_WIDTH; } return (ret); } /* * Get the root node for a radix tree. */ static __inline struct vm_radix_node * vm_radix_getroot(struct vm_radix *rtree) { return ((struct vm_radix_node *)rtree->rt_root); } /* * Set the root node for a radix tree. */ static __inline void vm_radix_setroot(struct vm_radix *rtree, struct vm_radix_node *rnode) { rtree->rt_root = (uintptr_t)rnode; } /* * Returns TRUE if the specified radix node is a leaf and FALSE otherwise. */ static __inline boolean_t vm_radix_isleaf(struct vm_radix_node *rnode) { return (((uintptr_t)rnode & VM_RADIX_ISLEAF) != 0); } /* * Returns the associated page extracted from rnode. */ static __inline vm_page_t vm_radix_topage(struct vm_radix_node *rnode) { return ((vm_page_t)((uintptr_t)rnode & ~VM_RADIX_FLAGS)); } /* * Adds the page as a child of the provided node. */ static __inline void vm_radix_addpage(struct vm_radix_node *rnode, vm_pindex_t index, uint16_t clev, vm_page_t page) { int slot; slot = vm_radix_slot(index, clev); rnode->rn_child[slot] = (void *)((uintptr_t)page | VM_RADIX_ISLEAF); } /* * Returns the slot where two keys differ. * It cannot accept 2 equal keys. */ static __inline uint16_t vm_radix_keydiff(vm_pindex_t index1, vm_pindex_t index2) { uint16_t clev; KASSERT(index1 != index2, ("%s: passing the same key value %jx", __func__, (uintmax_t)index1)); index1 ^= index2; for (clev = VM_RADIX_LIMIT;; clev--) if (vm_radix_slot(index1, clev) != 0) return (clev); } /* * Returns TRUE if it can be determined that key does not belong to the * specified rnode. Otherwise, returns FALSE. */ static __inline boolean_t vm_radix_keybarr(struct vm_radix_node *rnode, vm_pindex_t idx) { if (rnode->rn_clev < VM_RADIX_LIMIT) { idx = vm_radix_trimkey(idx, rnode->rn_clev + 1); return (idx != rnode->rn_owner); } return (FALSE); } /* * Internal helper for vm_radix_reclaim_allnodes(). * This function is recursive. */ static void vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode) { int slot; KASSERT(rnode->rn_count <= VM_RADIX_COUNT, ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode)); for (slot = 0; rnode->rn_count != 0; slot++) { if (rnode->rn_child[slot] == NULL) continue; if (!vm_radix_isleaf(rnode->rn_child[slot])) vm_radix_reclaim_allnodes_int(rnode->rn_child[slot]); rnode->rn_child[slot] = NULL; rnode->rn_count--; } vm_radix_node_put(rnode); } #ifdef INVARIANTS /* * Radix node zone destructor. */ static void vm_radix_node_zone_dtor(void *mem, int size __unused, void *arg __unused) { struct vm_radix_node *rnode; int slot; rnode = mem; KASSERT(rnode->rn_count == 0, ("vm_radix_node_put: rnode %p has %d children", rnode, rnode->rn_count)); for (slot = 0; slot < VM_RADIX_COUNT; slot++) KASSERT(rnode->rn_child[slot] == NULL, ("vm_radix_node_put: rnode %p has a child", rnode)); } #endif #ifndef UMA_MD_SMALL_ALLOC /* * Reserve the KVA necessary to satisfy the node allocation. * This is mandatory in architectures not supporting direct * mapping as they will need otherwise to carve into the kernel maps for * every node allocation, resulting into deadlocks for consumers already * working with kernel maps. */ static void vm_radix_reserve_kva(void *arg __unused) { /* * Calculate the number of reserved nodes, discounting the pages that * are needed to store them. */ if (!uma_zone_reserve_kva(vm_radix_node_zone, ((vm_paddr_t)vm_cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE + sizeof(struct vm_radix_node)))) panic("%s: unable to reserve KVA", __func__); } SYSINIT(vm_radix_reserve_kva, SI_SUB_KMEM, SI_ORDER_THIRD, vm_radix_reserve_kva, NULL); #endif /* * Initialize the UMA slab zone. * Until vm_radix_prealloc() is called, the zone will be served by the * UMA boot-time pre-allocated pool of pages. */ void vm_radix_init(void) { vm_radix_node_zone = uma_zcreate("RADIX NODE", sizeof(struct vm_radix_node), NULL, #ifdef INVARIANTS vm_radix_node_zone_dtor, #else NULL, #endif NULL, NULL, VM_RADIX_PAD, UMA_ZONE_VM); } /* * Inserts the key-value pair into the trie. * Panics if the key already exists. */ int vm_radix_insert(struct vm_radix *rtree, vm_page_t page) { vm_pindex_t index, newind; void **parentp; struct vm_radix_node *rnode, *tmp; vm_page_t m; int slot; uint16_t clev; index = page->pindex; restart: /* * The owner of record for root is not really important because it * will never be used. */ rnode = vm_radix_getroot(rtree); if (rnode == NULL) { rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF; return (0); } parentp = (void **)&rtree->rt_root; for (;;) { if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex == index) panic("%s: key %jx is already present", __func__, (uintmax_t)index); clev = vm_radix_keydiff(m->pindex, index); /* * During node allocation the trie that is being * walked can be modified because of recursing radix * trie operations. * If this is the case, the recursing functions signal * such situation and the insert operation must * start from scratch again. * The freed radix node will then be in the UMA * caches very likely to avoid the same situation * to happen. */ rtree->rt_flags |= RT_INSERT_INPROG; tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev); rtree->rt_flags &= ~RT_INSERT_INPROG; if (tmp == NULL) { rtree->rt_flags &= ~RT_TRIE_MODIFIED; return (ENOMEM); } if ((rtree->rt_flags & RT_TRIE_MODIFIED) != 0) { rtree->rt_flags &= ~RT_TRIE_MODIFIED; tmp->rn_count = 0; vm_radix_node_put(tmp); goto restart; } *parentp = tmp; vm_radix_addpage(tmp, index, clev, page); vm_radix_addpage(tmp, m->pindex, clev, m); return (0); } else if (vm_radix_keybarr(rnode, index)) break; slot = vm_radix_slot(index, rnode->rn_clev); if (rnode->rn_child[slot] == NULL) { rnode->rn_count++; vm_radix_addpage(rnode, index, rnode->rn_clev, page); return (0); } parentp = &rnode->rn_child[slot]; rnode = rnode->rn_child[slot]; } /* * A new node is needed because the right insertion level is reached. * Setup the new intermediate node and add the 2 children: the * new object and the older edge. */ newind = rnode->rn_owner; clev = vm_radix_keydiff(newind, index); /* See the comments above. */ rtree->rt_flags |= RT_INSERT_INPROG; tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev); rtree->rt_flags &= ~RT_INSERT_INPROG; if (tmp == NULL) { rtree->rt_flags &= ~RT_TRIE_MODIFIED; return (ENOMEM); } if ((rtree->rt_flags & RT_TRIE_MODIFIED) != 0) { rtree->rt_flags &= ~RT_TRIE_MODIFIED; tmp->rn_count = 0; vm_radix_node_put(tmp); goto restart; } *parentp = tmp; vm_radix_addpage(tmp, index, clev, page); slot = vm_radix_slot(newind, clev); tmp->rn_child[slot] = rnode; return (0); } /* * Returns TRUE if the specified radix tree contains a single leaf and FALSE * otherwise. */ boolean_t vm_radix_is_singleton(struct vm_radix *rtree) { struct vm_radix_node *rnode; rnode = vm_radix_getroot(rtree); if (rnode == NULL) return (FALSE); return (vm_radix_isleaf(rnode)); } /* * Returns the value stored at the index. If the index is not present, * NULL is returned. */ vm_page_t vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *rnode; vm_page_t m; int slot; rnode = vm_radix_getroot(rtree); while (rnode != NULL) { if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex == index) return (m); else break; } else if (vm_radix_keybarr(rnode, index)) break; slot = vm_radix_slot(index, rnode->rn_clev); rnode = rnode->rn_child[slot]; } return (NULL); } /* * Look up the nearest entry at a position bigger than or equal to index. */ vm_page_t vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *stack[VM_RADIX_LIMIT]; vm_pindex_t inc; vm_page_t m; struct vm_radix_node *child, *rnode; #ifdef INVARIANTS int loops = 0; #endif int slot, tos; rnode = vm_radix_getroot(rtree); if (rnode == NULL) return (NULL); else if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex >= index) return (m); else return (NULL); } tos = 0; for (;;) { /* * If the keys differ before the current bisection node, * then the search key might rollback to the earliest * available bisection node or to the smallest key * in the current node (if the owner is bigger than the * search key). */ if (vm_radix_keybarr(rnode, index)) { if (index > rnode->rn_owner) { ascend: KASSERT(++loops < 1000, ("vm_radix_lookup_ge: too many loops")); /* * Pop nodes from the stack until either the * stack is empty or a node that could have a * matching descendant is found. */ do { if (tos == 0) return (NULL); rnode = stack[--tos]; } while (vm_radix_slot(index, rnode->rn_clev) == (VM_RADIX_COUNT - 1)); /* * The following computation cannot overflow * because index's slot at the current level * is less than VM_RADIX_COUNT - 1. */ index = vm_radix_trimkey(index, rnode->rn_clev); index += VM_RADIX_UNITLEVEL(rnode->rn_clev); } else index = rnode->rn_owner; KASSERT(!vm_radix_keybarr(rnode, index), ("vm_radix_lookup_ge: keybarr failed")); } slot = vm_radix_slot(index, rnode->rn_clev); child = rnode->rn_child[slot]; if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex >= index) return (m); } else if (child != NULL) goto descend; /* * Look for an available edge or page within the current * bisection node. */ if (slot < (VM_RADIX_COUNT - 1)) { inc = VM_RADIX_UNITLEVEL(rnode->rn_clev); index = vm_radix_trimkey(index, rnode->rn_clev); do { index += inc; slot++; child = rnode->rn_child[slot]; if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex >= index) return (m); } else if (child != NULL) goto descend; } while (slot < (VM_RADIX_COUNT - 1)); } KASSERT(child == NULL || vm_radix_isleaf(child), ("vm_radix_lookup_ge: child is radix node")); /* * If a page or edge bigger than the search slot is not found * in the current node, ascend to the next higher-level node. */ goto ascend; descend: KASSERT(rnode->rn_clev > 0, ("vm_radix_lookup_ge: pushing leaf's parent")); KASSERT(tos < VM_RADIX_LIMIT, ("vm_radix_lookup_ge: stack overflow")); stack[tos++] = rnode; rnode = child; } } /* * Look up the nearest entry at a position less than or equal to index. */ vm_page_t vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *stack[VM_RADIX_LIMIT]; vm_pindex_t inc; vm_page_t m; struct vm_radix_node *child, *rnode; #ifdef INVARIANTS int loops = 0; #endif int slot, tos; rnode = vm_radix_getroot(rtree); if (rnode == NULL) return (NULL); else if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex <= index) return (m); else return (NULL); } tos = 0; for (;;) { /* * If the keys differ before the current bisection node, * then the search key might rollback to the earliest * available bisection node or to the largest key * in the current node (if the owner is smaller than the * search key). */ if (vm_radix_keybarr(rnode, index)) { if (index > rnode->rn_owner) { index = rnode->rn_owner + VM_RADIX_COUNT * VM_RADIX_UNITLEVEL(rnode->rn_clev); } else { ascend: KASSERT(++loops < 1000, ("vm_radix_lookup_le: too many loops")); /* * Pop nodes from the stack until either the * stack is empty or a node that could have a * matching descendant is found. */ do { if (tos == 0) return (NULL); rnode = stack[--tos]; } while (vm_radix_slot(index, rnode->rn_clev) == 0); /* * The following computation cannot overflow * because index's slot at the current level * is greater than 0. */ index = vm_radix_trimkey(index, rnode->rn_clev); } index--; KASSERT(!vm_radix_keybarr(rnode, index), ("vm_radix_lookup_le: keybarr failed")); } slot = vm_radix_slot(index, rnode->rn_clev); child = rnode->rn_child[slot]; if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex <= index) return (m); } else if (child != NULL) goto descend; /* * Look for an available edge or page within the current * bisection node. */ if (slot > 0) { inc = VM_RADIX_UNITLEVEL(rnode->rn_clev); index |= inc - 1; do { index -= inc; slot--; child = rnode->rn_child[slot]; if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex <= index) return (m); } else if (child != NULL) goto descend; } while (slot > 0); } KASSERT(child == NULL || vm_radix_isleaf(child), ("vm_radix_lookup_le: child is radix node")); /* * If a page or edge smaller than the search slot is not found * in the current node, ascend to the next higher-level node. */ goto ascend; descend: KASSERT(rnode->rn_clev > 0, ("vm_radix_lookup_le: pushing leaf's parent")); KASSERT(tos < VM_RADIX_LIMIT, ("vm_radix_lookup_le: stack overflow")); stack[tos++] = rnode; rnode = child; } } /* * Remove the specified index from the tree. * Panics if the key is not present. */ void vm_radix_remove(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *rnode, *parent; vm_page_t m; int i, slot; /* * Detect if a page is going to be removed from a trie which is * already undergoing another trie operation. * Right now this is only possible for vm_radix_remove() recursing * into vm_radix_insert(). * If this is the case, the caller must be notified about this * situation. It will also takecare to update the RT_TRIE_MODIFIED * accordingly. * The RT_TRIE_MODIFIED bit is set here because the remove operation * will always succeed. */ if ((rtree->rt_flags & RT_INSERT_INPROG) != 0) rtree->rt_flags |= RT_TRIE_MODIFIED; rnode = vm_radix_getroot(rtree); if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex != index) panic("%s: invalid key found", __func__); vm_radix_setroot(rtree, NULL); return; } parent = NULL; for (;;) { if (rnode == NULL) panic("vm_radix_remove: impossible to locate the key"); slot = vm_radix_slot(index, rnode->rn_clev); if (vm_radix_isleaf(rnode->rn_child[slot])) { m = vm_radix_topage(rnode->rn_child[slot]); if (m->pindex != index) panic("%s: invalid key found", __func__); rnode->rn_child[slot] = NULL; rnode->rn_count--; if (rnode->rn_count > 1) break; for (i = 0; i < VM_RADIX_COUNT; i++) if (rnode->rn_child[i] != NULL) break; KASSERT(i != VM_RADIX_COUNT, ("%s: invalid node configuration", __func__)); if (parent == NULL) vm_radix_setroot(rtree, rnode->rn_child[i]); else { slot = vm_radix_slot(index, parent->rn_clev); KASSERT(parent->rn_child[slot] == rnode, ("%s: invalid child value", __func__)); parent->rn_child[slot] = rnode->rn_child[i]; } rnode->rn_count--; rnode->rn_child[i] = NULL; vm_radix_node_put(rnode); break; } parent = rnode; rnode = rnode->rn_child[slot]; } } /* * Remove and free all the nodes from the radix tree. * This function is recursive but there is a tight control on it as the * maximum depth of the tree is fixed. */ void vm_radix_reclaim_allnodes(struct vm_radix *rtree) { struct vm_radix_node *root; KASSERT((rtree->rt_flags & RT_INSERT_INPROG) == 0, ("vm_radix_reclaim_allnodes: unexpected trie recursion")); root = vm_radix_getroot(rtree); if (root == NULL) return; vm_radix_setroot(rtree, NULL); if (!vm_radix_isleaf(root)) vm_radix_reclaim_allnodes_int(root); } /* * Replace an existing page in the trie with another one. * Panics if there is not an old page in the trie at the new page's index. */ vm_page_t vm_radix_replace(struct vm_radix *rtree, vm_page_t newpage) { struct vm_radix_node *rnode; vm_page_t m; vm_pindex_t index; int slot; index = newpage->pindex; rnode = vm_radix_getroot(rtree); if (rnode == NULL) panic("%s: replacing page on an empty trie", __func__); if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex != index) panic("%s: original replacing root key not found", __func__); rtree->rt_root = (uintptr_t)newpage | VM_RADIX_ISLEAF; return (m); } for (;;) { slot = vm_radix_slot(index, rnode->rn_clev); if (vm_radix_isleaf(rnode->rn_child[slot])) { m = vm_radix_topage(rnode->rn_child[slot]); if (m->pindex == index) { rnode->rn_child[slot] = (void *)((uintptr_t)newpage | VM_RADIX_ISLEAF); return (m); } else break; } else if (rnode->rn_child[slot] == NULL || vm_radix_keybarr(rnode->rn_child[slot], index)) break; rnode = rnode->rn_child[slot]; } panic("%s: original replacing page not found", __func__); } #ifdef DDB /* * Show details about the given radix node. */ DB_SHOW_COMMAND(radixnode, db_show_radixnode) { struct vm_radix_node *rnode; int i; if (!have_addr) return; rnode = (struct vm_radix_node *)addr; db_printf("radixnode %p, owner %jx, children count %u, level %u:\n", (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count, rnode->rn_clev); for (i = 0; i < VM_RADIX_COUNT; i++) if (rnode->rn_child[i] != NULL) db_printf("slot: %d, val: %p, page: %p, clev: %d\n", i, (void *)rnode->rn_child[i], vm_radix_isleaf(rnode->rn_child[i]) ? vm_radix_topage(rnode->rn_child[i]) : NULL, rnode->rn_clev); } #endif /* DDB */