diff --git a/sys/vm/vm_extern.h b/sys/vm/vm_extern.h index b50abab7380c..8c9addda1196 100644 --- a/sys/vm/vm_extern.h +++ b/sys/vm/vm_extern.h @@ -1,170 +1,170 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1992, 1993 * The Regents of the University of California. 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. * 3. 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. */ #ifndef _VM_EXTERN_H_ #define _VM_EXTERN_H_ struct pmap; struct proc; struct vmspace; struct vnode; struct vmem; #ifdef _KERNEL #include struct cdev; struct cdevsw; struct domainset; /* These operate on kernel virtual addresses only. */ vm_offset_t kva_alloc(vm_size_t); vm_offset_t kva_alloc_aligned(vm_size_t, vm_size_t); void kva_free(vm_offset_t, vm_size_t); /* These operate on pageable virtual addresses. */ vm_offset_t kmap_alloc_wait(vm_map_t, vm_size_t); void kmap_free_wakeup(vm_map_t, vm_offset_t, vm_size_t); /* These operate on virtual addresses backed by memory. */ void *kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr); void *kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr); void *kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr); void *kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr); void *kmem_malloc(vm_size_t size, int flags); void *kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags); void kmem_free(void *addr, vm_size_t size); /* This provides memory for previously allocated address space. */ int kmem_back(vm_object_t, vm_offset_t, vm_size_t, int); int kmem_back_domain(int, vm_object_t, vm_offset_t, vm_size_t, int); void kmem_unback(vm_object_t, vm_offset_t, vm_size_t); /* Bootstrapping. */ void kmem_bootstrap_free(vm_offset_t, vm_size_t); void kmem_subinit(vm_map_t, vm_map_t, vm_offset_t *, vm_offset_t *, vm_size_t, bool); void kmem_init(vm_offset_t, vm_offset_t); void kmem_init_zero_region(void); void kmeminit(void); -int kernacc(void *, int, int); -int useracc(void *, int, int); +bool kernacc(void *, int, int); +bool useracc(void *, int, int); int vm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, int fault_flags, vm_page_t *m_hold); void vm_fault_copy_entry(vm_map_t, vm_map_t, vm_map_entry_t, vm_map_entry_t, vm_ooffset_t *); int vm_fault_disable_pagefaults(void); void vm_fault_enable_pagefaults(int save); int vm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len, vm_prot_t prot, vm_page_t *ma, int max_count); int vm_fault_trap(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type, int fault_flags, int *signo, int *ucode); int vm_forkproc(struct thread *, struct proc *, struct thread *, struct vmspace *, int); void vm_waitproc(struct proc *); int vm_mmap(vm_map_t, vm_offset_t *, vm_size_t, vm_prot_t, vm_prot_t, int, objtype_t, void *, vm_ooffset_t); int vm_mmap_object(vm_map_t, vm_offset_t *, vm_size_t, vm_prot_t, vm_prot_t, int, vm_object_t, vm_ooffset_t, boolean_t, struct thread *); int vm_mmap_to_errno(int rv); int vm_mmap_cdev(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *, int *, struct cdev *, struct cdevsw *, vm_ooffset_t *, vm_object_t *); int vm_mmap_vnode(struct thread *, vm_size_t, vm_prot_t, vm_prot_t *, int *, struct vnode *, vm_ooffset_t *, vm_object_t *, boolean_t *); void vm_set_page_size(void); void vm_sync_icache(vm_map_t, vm_offset_t, vm_size_t); typedef int (*pmap_pinit_t)(struct pmap *pmap); struct vmspace *vmspace_alloc(vm_offset_t, vm_offset_t, pmap_pinit_t); struct vmspace *vmspace_fork(struct vmspace *, vm_ooffset_t *); int vmspace_exec(struct proc *, vm_offset_t, vm_offset_t); int vmspace_unshare(struct proc *); void vmspace_exit(struct thread *); struct vmspace *vmspace_acquire_ref(struct proc *); void vmspace_free(struct vmspace *); void vmspace_exitfree(struct proc *); void vmspace_switch_aio(struct vmspace *); void vnode_pager_setsize(struct vnode *, vm_ooffset_t); void vnode_pager_purge_range(struct vnode *, vm_ooffset_t, vm_ooffset_t); int vslock(void *, size_t); void vsunlock(void *, size_t); struct sf_buf *vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset); void vm_imgact_unmap_page(struct sf_buf *sf); void vm_thread_dispose(struct thread *td); int vm_thread_new(struct thread *td, int pages); vm_pindex_t vm_kstack_pindex(vm_offset_t ks, int npages); vm_object_t vm_thread_kstack_size_to_obj(int npages); int vm_thread_stack_back(vm_offset_t kaddr, vm_page_t ma[], int npages, int req_class, int domain); u_int vm_active_count(void); u_int vm_inactive_count(void); u_int vm_laundry_count(void); u_int vm_wait_count(void); /* * Is pa a multiple of alignment, which is a power-of-two? */ static inline bool vm_addr_align_ok(vm_paddr_t pa, u_long alignment) { KASSERT(powerof2(alignment), ("%s: alignment is not a power of 2: %#lx", __func__, alignment)); return ((pa & (alignment - 1)) == 0); } /* * Do the first and last addresses of a range match in all bits except the ones * in -boundary (a power-of-two)? For boundary == 0, all addresses match. */ static inline bool vm_addr_bound_ok(vm_paddr_t pa, vm_paddr_t size, vm_paddr_t boundary) { KASSERT(powerof2(boundary), ("%s: boundary is not a power of 2: %#jx", __func__, (uintmax_t)boundary)); return (((pa ^ (pa + size - 1)) & -boundary) == 0); } static inline bool vm_addr_ok(vm_paddr_t pa, vm_paddr_t size, u_long alignment, vm_paddr_t boundary) { return (vm_addr_align_ok(pa, alignment) && vm_addr_bound_ok(pa, size, boundary)); } #endif /* _KERNEL */ #endif /* !_VM_EXTERN_H_ */ diff --git a/sys/vm/vm_glue.c b/sys/vm/vm_glue.c index 979a0fc27120..641800413e68 100644 --- a/sys/vm/vm_glue.c +++ b/sys/vm/vm_glue.c @@ -1,839 +1,839 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * 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. 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. * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ #include #include "opt_vm.h" #include "opt_kstack_pages.h" #include "opt_kstack_max_pages.h" #include "opt_kstack_usage_prof.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 #include #include #include #if VM_NRESERVLEVEL > 0 #define KVA_KSTACK_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT) #else #define KVA_KSTACK_QUANTUM_SHIFT (8 + PAGE_SHIFT) #endif #define KVA_KSTACK_QUANTUM (1ul << KVA_KSTACK_QUANTUM_SHIFT) /* * MPSAFE * * WARNING! This code calls vm_map_check_protection() which only checks * the associated vm_map_entry range. It does not determine whether the * contents of the memory is actually readable or writable. In most cases * just checking the vm_map_entry is sufficient within the kernel's address * space. */ -int +bool kernacc(void *addr, int len, int rw) { boolean_t rv; vm_offset_t saddr, eaddr; vm_prot_t prot; KASSERT((rw & ~VM_PROT_ALL) == 0, ("illegal ``rw'' argument to kernacc (%x)\n", rw)); if ((vm_offset_t)addr + len > vm_map_max(kernel_map) || (vm_offset_t)addr + len < (vm_offset_t)addr) - return (FALSE); + return (false); prot = rw; saddr = trunc_page((vm_offset_t)addr); eaddr = round_page((vm_offset_t)addr + len); vm_map_lock_read(kernel_map); rv = vm_map_check_protection(kernel_map, saddr, eaddr, prot); vm_map_unlock_read(kernel_map); return (rv == TRUE); } /* * MPSAFE * * WARNING! This code calls vm_map_check_protection() which only checks * the associated vm_map_entry range. It does not determine whether the * contents of the memory is actually readable or writable. vmapbuf(), * vm_fault_quick(), or copyin()/copout()/su*()/fu*() functions should be * used in conjunction with this call. */ -int +bool useracc(void *addr, int len, int rw) { boolean_t rv; vm_prot_t prot; vm_map_t map; KASSERT((rw & ~VM_PROT_ALL) == 0, ("illegal ``rw'' argument to useracc (%x)\n", rw)); prot = rw; map = &curproc->p_vmspace->vm_map; if ((vm_offset_t)addr + len > vm_map_max(map) || (vm_offset_t)addr + len < (vm_offset_t)addr) { - return (FALSE); + return (false); } vm_map_lock_read(map); rv = vm_map_check_protection(map, trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), prot); vm_map_unlock_read(map); return (rv == TRUE); } int vslock(void *addr, size_t len) { vm_offset_t end, last, start; vm_size_t npages; int error; last = (vm_offset_t)addr + len; start = trunc_page((vm_offset_t)addr); end = round_page(last); if (last < (vm_offset_t)addr || end < (vm_offset_t)addr) return (EINVAL); npages = atop(end - start); if (npages > vm_page_max_user_wired) return (ENOMEM); error = vm_map_wire(&curproc->p_vmspace->vm_map, start, end, VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); if (error == KERN_SUCCESS) { curthread->td_vslock_sz += len; return (0); } /* * Return EFAULT on error to match copy{in,out}() behaviour * rather than returning ENOMEM like mlock() would. */ return (EFAULT); } void vsunlock(void *addr, size_t len) { /* Rely on the parameter sanity checks performed by vslock(). */ MPASS(curthread->td_vslock_sz >= len); curthread->td_vslock_sz -= len; (void)vm_map_unwire(&curproc->p_vmspace->vm_map, trunc_page((vm_offset_t)addr), round_page((vm_offset_t)addr + len), VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); } /* * Pin the page contained within the given object at the given offset. If the * page is not resident, allocate and load it using the given object's pager. * Return the pinned page if successful; otherwise, return NULL. */ static vm_page_t vm_imgact_hold_page(vm_object_t object, vm_ooffset_t offset) { vm_page_t m; vm_pindex_t pindex; pindex = OFF_TO_IDX(offset); (void)vm_page_grab_valid_unlocked(&m, object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); return (m); } /* * Return a CPU private mapping to the page at the given offset within the * given object. The page is pinned before it is mapped. */ struct sf_buf * vm_imgact_map_page(vm_object_t object, vm_ooffset_t offset) { vm_page_t m; m = vm_imgact_hold_page(object, offset); if (m == NULL) return (NULL); sched_pin(); return (sf_buf_alloc(m, SFB_CPUPRIVATE)); } /* * Destroy the given CPU private mapping and unpin the page that it mapped. */ void vm_imgact_unmap_page(struct sf_buf *sf) { vm_page_t m; m = sf_buf_page(sf); sf_buf_free(sf); sched_unpin(); vm_page_unwire(m, PQ_ACTIVE); } void vm_sync_icache(vm_map_t map, vm_offset_t va, vm_offset_t sz) { pmap_sync_icache(map->pmap, va, sz); } static vm_object_t kstack_object; static vm_object_t kstack_alt_object; static uma_zone_t kstack_cache; static int kstack_cache_size; static vmem_t *vmd_kstack_arena[MAXMEMDOM]; static int sysctl_kstack_cache_size(SYSCTL_HANDLER_ARGS) { int error, oldsize; oldsize = kstack_cache_size; error = sysctl_handle_int(oidp, arg1, arg2, req); if (error == 0 && req->newptr && oldsize != kstack_cache_size) uma_zone_set_maxcache(kstack_cache, kstack_cache_size); return (error); } SYSCTL_PROC(_vm, OID_AUTO, kstack_cache_size, CTLTYPE_INT|CTLFLAG_MPSAFE|CTLFLAG_RW, &kstack_cache_size, 0, sysctl_kstack_cache_size, "IU", "Maximum number of cached kernel stacks"); /* * Allocate a virtual address range from a domain kstack arena, following * the specified NUMA policy. */ static vm_offset_t vm_thread_alloc_kstack_kva(vm_size_t size, int domain) { #ifndef __ILP32__ int rv; vmem_t *arena; vm_offset_t addr = 0; size = round_page(size); /* Allocate from the kernel arena for non-standard kstack sizes. */ if (size != ptoa(kstack_pages + KSTACK_GUARD_PAGES)) { arena = vm_dom[domain].vmd_kernel_arena; } else { arena = vmd_kstack_arena[domain]; } rv = vmem_alloc(arena, size, M_BESTFIT | M_NOWAIT, &addr); if (rv == ENOMEM) return (0); KASSERT(atop(addr - VM_MIN_KERNEL_ADDRESS) % (kstack_pages + KSTACK_GUARD_PAGES) == 0, ("%s: allocated kstack KVA not aligned to multiple of kstack size", __func__)); return (addr); #else return (kva_alloc(size)); #endif } /* * Release a region of kernel virtual memory * allocated from the kstack arena. */ static __noinline void vm_thread_free_kstack_kva(vm_offset_t addr, vm_size_t size, int domain) { vmem_t *arena; size = round_page(size); #ifdef __ILP32__ arena = kernel_arena; #else arena = vmd_kstack_arena[domain]; if (size != ptoa(kstack_pages + KSTACK_GUARD_PAGES)) { arena = vm_dom[domain].vmd_kernel_arena; } #endif vmem_free(arena, addr, size); } static vmem_size_t vm_thread_kstack_import_quantum(void) { #ifndef __ILP32__ /* * The kstack_quantum is larger than KVA_QUANTUM to account * for holes induced by guard pages. */ return (KVA_KSTACK_QUANTUM * (kstack_pages + KSTACK_GUARD_PAGES)); #else return (KVA_KSTACK_QUANTUM); #endif } /* * Import KVA from a parent arena into the kstack arena. Imports must be * a multiple of kernel stack pages + guard pages in size. * * Kstack VA allocations need to be aligned so that the linear KVA pindex * is divisible by the total number of kstack VA pages. This is necessary to * make vm_kstack_pindex work properly. * * We import a multiple of KVA_KSTACK_QUANTUM-sized region from the parent * arena. The actual size used by the kstack arena is one kstack smaller to * allow for the necessary alignment adjustments to be made. */ static int vm_thread_kstack_arena_import(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp) { int error, rem; size_t kpages = kstack_pages + KSTACK_GUARD_PAGES; KASSERT(atop(size) % kpages == 0, ("%s: Size %jd is not a multiple of kstack pages (%d)", __func__, (intmax_t)size, (int)kpages)); error = vmem_xalloc(arena, vm_thread_kstack_import_quantum(), KVA_KSTACK_QUANTUM, 0, 0, VMEM_ADDR_MIN, VMEM_ADDR_MAX, flags, addrp); if (error) { return (error); } rem = atop(*addrp - VM_MIN_KERNEL_ADDRESS) % kpages; if (rem != 0) { /* Bump addr to next aligned address */ *addrp = *addrp + (kpages - rem) * PAGE_SIZE; } return (0); } /* * Release KVA from a parent arena into the kstack arena. Released imports must * be a multiple of kernel stack pages + guard pages in size. */ static void vm_thread_kstack_arena_release(void *arena, vmem_addr_t addr, vmem_size_t size) { int rem; size_t kpages __diagused = kstack_pages + KSTACK_GUARD_PAGES; KASSERT(size % kpages == 0, ("%s: Size %jd is not a multiple of kstack pages (%d)", __func__, (intmax_t)size, (int)kpages)); KASSERT((addr - VM_MIN_KERNEL_ADDRESS) % kpages == 0, ("%s: Address %p is not properly aligned (%p)", __func__, (void *)addr, (void *)VM_MIN_KERNEL_ADDRESS)); /* * If the address is not KVA_KSTACK_QUANTUM-aligned we have to decrement * it to account for the shift in kva_import_kstack. */ rem = addr % KVA_KSTACK_QUANTUM; if (rem) { KASSERT(rem <= ptoa(kpages), ("%s: rem > kpages (%d), (%d)", __func__, rem, (int)kpages)); addr -= rem; } vmem_xfree(arena, addr, vm_thread_kstack_import_quantum()); } /* * Create the kernel stack for a new thread. */ static vm_offset_t vm_thread_stack_create(struct domainset *ds, int pages) { vm_page_t ma[KSTACK_MAX_PAGES]; struct vm_domainset_iter di; int req = VM_ALLOC_NORMAL; vm_object_t obj; vm_offset_t ks; int domain, i; obj = vm_thread_kstack_size_to_obj(pages); if (vm_ndomains > 1) obj->domain.dr_policy = ds; vm_domainset_iter_page_init(&di, obj, 0, &domain, &req); do { /* * Get a kernel virtual address for this thread's kstack. */ ks = vm_thread_alloc_kstack_kva(ptoa(pages + KSTACK_GUARD_PAGES), domain); if (ks == 0) continue; ks += ptoa(KSTACK_GUARD_PAGES); /* * Allocate physical pages to back the stack. */ if (vm_thread_stack_back(ks, ma, pages, req, domain) != 0) { vm_thread_free_kstack_kva(ks - ptoa(KSTACK_GUARD_PAGES), ptoa(pages + KSTACK_GUARD_PAGES), domain); continue; } if (KSTACK_GUARD_PAGES != 0) { pmap_qremove(ks - ptoa(KSTACK_GUARD_PAGES), KSTACK_GUARD_PAGES); } for (i = 0; i < pages; i++) vm_page_valid(ma[i]); pmap_qenter(ks, ma, pages); return (ks); } while (vm_domainset_iter_page(&di, obj, &domain) == 0); return (0); } static __noinline void vm_thread_stack_dispose(vm_offset_t ks, int pages) { vm_page_t m; vm_pindex_t pindex; int i, domain; vm_object_t obj = vm_thread_kstack_size_to_obj(pages); pindex = vm_kstack_pindex(ks, pages); domain = vm_phys_domain(vtophys(ks)); pmap_qremove(ks, pages); VM_OBJECT_WLOCK(obj); for (i = 0; i < pages; i++) { m = vm_page_lookup(obj, pindex + i); if (m == NULL) panic("%s: kstack already missing?", __func__); KASSERT(vm_page_domain(m) == domain, ("%s: page %p domain mismatch, expected %d got %d", __func__, m, domain, vm_page_domain(m))); vm_page_xbusy_claim(m); vm_page_unwire_noq(m); vm_page_free(m); } VM_OBJECT_WUNLOCK(obj); kasan_mark((void *)ks, ptoa(pages), ptoa(pages), 0); vm_thread_free_kstack_kva(ks - (KSTACK_GUARD_PAGES * PAGE_SIZE), ptoa(pages + KSTACK_GUARD_PAGES), domain); } /* * Allocate the kernel stack for a new thread. */ int vm_thread_new(struct thread *td, int pages) { vm_offset_t ks; u_short ks_domain; /* Bounds check */ if (pages <= 1) pages = kstack_pages; else if (pages > KSTACK_MAX_PAGES) pages = KSTACK_MAX_PAGES; ks = 0; if (pages == kstack_pages && kstack_cache != NULL) ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); /* * Ensure that kstack objects can draw pages from any memory * domain. Otherwise a local memory shortage can block a process * swap-in. */ if (ks == 0) ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), pages); if (ks == 0) return (0); ks_domain = vm_phys_domain(vtophys(ks)); KASSERT(ks_domain >= 0 && ks_domain < vm_ndomains, ("%s: invalid domain for kstack %p", __func__, (void *)ks)); td->td_kstack = ks; td->td_kstack_pages = pages; td->td_kstack_domain = ks_domain; return (1); } /* * Dispose of a thread's kernel stack. */ void vm_thread_dispose(struct thread *td) { vm_offset_t ks; int pages; pages = td->td_kstack_pages; ks = td->td_kstack; td->td_kstack = 0; td->td_kstack_pages = 0; td->td_kstack_domain = MAXMEMDOM; if (pages == kstack_pages) { kasan_mark((void *)ks, 0, ptoa(pages), KASAN_KSTACK_FREED); uma_zfree(kstack_cache, (void *)ks); } else { vm_thread_stack_dispose(ks, pages); } } /* * Calculate kstack pindex. * * Uses a non-identity mapping if guard pages are * active to avoid pindex holes in the kstack object. */ vm_pindex_t vm_kstack_pindex(vm_offset_t ks, int kpages) { vm_pindex_t pindex = atop(ks - VM_MIN_KERNEL_ADDRESS); #ifdef __ILP32__ return (pindex); #else /* * Return the linear pindex if guard pages aren't active or if we are * allocating a non-standard kstack size. */ if (KSTACK_GUARD_PAGES == 0 || kpages != kstack_pages) { return (pindex); } KASSERT(pindex % (kpages + KSTACK_GUARD_PAGES) >= KSTACK_GUARD_PAGES, ("%s: Attempting to calculate kstack guard page pindex", __func__)); return (pindex - (pindex / (kpages + KSTACK_GUARD_PAGES) + 1) * KSTACK_GUARD_PAGES); #endif } /* * Allocate physical pages, following the specified NUMA policy, to back a * kernel stack. */ int vm_thread_stack_back(vm_offset_t ks, vm_page_t ma[], int npages, int req_class, int domain) { vm_object_t obj = vm_thread_kstack_size_to_obj(npages); vm_pindex_t pindex; vm_page_t m; int n; pindex = vm_kstack_pindex(ks, npages); VM_OBJECT_WLOCK(obj); for (n = 0; n < npages;) { m = vm_page_grab(obj, pindex + n, VM_ALLOC_NOCREAT | VM_ALLOC_WIRED); if (m == NULL) { m = vm_page_alloc_domain(obj, pindex + n, domain, req_class | VM_ALLOC_WIRED); } if (m == NULL) break; ma[n++] = m; } if (n < npages) goto cleanup; VM_OBJECT_WUNLOCK(obj); return (0); cleanup: for (int i = 0; i < n; i++) { m = ma[i]; (void)vm_page_unwire_noq(m); vm_page_free(m); } VM_OBJECT_WUNLOCK(obj); return (ENOMEM); } vm_object_t vm_thread_kstack_size_to_obj(int npages) { return (npages == kstack_pages ? kstack_object : kstack_alt_object); } static int kstack_import(void *arg, void **store, int cnt, int domain, int flags) { struct domainset *ds; int i; if (domain == UMA_ANYDOMAIN) ds = DOMAINSET_RR(); else ds = DOMAINSET_PREF(domain); for (i = 0; i < cnt; i++) { store[i] = (void *)vm_thread_stack_create(ds, kstack_pages); if (store[i] == NULL) break; } return (i); } static void kstack_release(void *arg, void **store, int cnt) { vm_offset_t ks; int i; for (i = 0; i < cnt; i++) { ks = (vm_offset_t)store[i]; vm_thread_stack_dispose(ks, kstack_pages); } } static void kstack_cache_init(void *null) { vm_size_t kstack_quantum; int domain; kstack_object = vm_object_allocate(OBJT_SWAP, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); kstack_cache = uma_zcache_create("kstack_cache", kstack_pages * PAGE_SIZE, NULL, NULL, NULL, NULL, kstack_import, kstack_release, NULL, UMA_ZONE_FIRSTTOUCH); kstack_cache_size = imax(128, mp_ncpus * 4); uma_zone_set_maxcache(kstack_cache, kstack_cache_size); kstack_alt_object = vm_object_allocate(OBJT_SWAP, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS)); kstack_quantum = vm_thread_kstack_import_quantum(); /* * Reduce size used by the kstack arena to allow for * alignment adjustments in vm_thread_kstack_arena_import. */ kstack_quantum -= (kstack_pages + KSTACK_GUARD_PAGES) * PAGE_SIZE; /* * Create the kstack_arena for each domain and set kernel_arena as * parent. */ for (domain = 0; domain < vm_ndomains; domain++) { vmd_kstack_arena[domain] = vmem_create("kstack arena", 0, 0, PAGE_SIZE, 0, M_WAITOK); KASSERT(vmd_kstack_arena[domain] != NULL, ("%s: failed to create domain %d kstack_arena", __func__, domain)); vmem_set_import(vmd_kstack_arena[domain], vm_thread_kstack_arena_import, vm_thread_kstack_arena_release, vm_dom[domain].vmd_kernel_arena, kstack_quantum); } } SYSINIT(vm_kstacks, SI_SUB_KMEM, SI_ORDER_ANY, kstack_cache_init, NULL); #ifdef KSTACK_USAGE_PROF /* * Track maximum stack used by a thread in kernel. */ static int max_kstack_used; SYSCTL_INT(_debug, OID_AUTO, max_kstack_used, CTLFLAG_RD, &max_kstack_used, 0, "Maximum stack depth used by a thread in kernel"); void intr_prof_stack_use(struct thread *td, struct trapframe *frame) { vm_offset_t stack_top; vm_offset_t current; int used, prev_used; /* * Testing for interrupted kernel mode isn't strictly * needed. It optimizes the execution, since interrupts from * usermode will have only the trap frame on the stack. */ if (TRAPF_USERMODE(frame)) return; stack_top = td->td_kstack + td->td_kstack_pages * PAGE_SIZE; current = (vm_offset_t)(uintptr_t)&stack_top; /* * Try to detect if interrupt is using kernel thread stack. * Hardware could use a dedicated stack for interrupt handling. */ if (stack_top <= current || current < td->td_kstack) return; used = stack_top - current; for (;;) { prev_used = max_kstack_used; if (prev_used >= used) break; if (atomic_cmpset_int(&max_kstack_used, prev_used, used)) break; } } #endif /* KSTACK_USAGE_PROF */ /* * Implement fork's actions on an address space. * Here we arrange for the address space to be copied or referenced, * allocate a user struct (pcb and kernel stack), then call the * machine-dependent layer to fill those in and make the new process * ready to run. The new process is set up so that it returns directly * to user mode to avoid stack copying and relocation problems. */ int vm_forkproc(struct thread *td, struct proc *p2, struct thread *td2, struct vmspace *vm2, int flags) { struct proc *p1 = td->td_proc; struct domainset *dset; int error; if ((flags & RFPROC) == 0) { /* * Divorce the memory, if it is shared, essentially * this changes shared memory amongst threads, into * COW locally. */ if ((flags & RFMEM) == 0) { error = vmspace_unshare(p1); if (error) return (error); } cpu_fork(td, p2, td2, flags); return (0); } if (flags & RFMEM) { p2->p_vmspace = p1->p_vmspace; refcount_acquire(&p1->p_vmspace->vm_refcnt); } dset = td2->td_domain.dr_policy; while (vm_page_count_severe_set(&dset->ds_mask)) { vm_wait_doms(&dset->ds_mask, 0); } if ((flags & RFMEM) == 0) { p2->p_vmspace = vm2; if (p1->p_vmspace->vm_shm) shmfork(p1, p2); } /* * cpu_fork will copy and update the pcb, set up the kernel stack, * and make the child ready to run. */ cpu_fork(td, p2, td2, flags); return (0); } /* * Called after process has been wait(2)'ed upon and is being reaped. * The idea is to reclaim resources that we could not reclaim while * the process was still executing. */ void vm_waitproc(struct proc *p) { vmspace_exitfree(p); /* and clean-out the vmspace */ } void kick_proc0(void) { wakeup(&proc0); }