diff --git a/sys/x86/xen/hvm.c b/sys/x86/xen/hvm.c index 9dc9360d719c..85f274175ac6 100644 --- a/sys/x86/xen/hvm.c +++ b/sys/x86/xen/hvm.c @@ -1,546 +1,633 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2008, 2013 Citrix Systems, Inc. * Copyright (c) 2012 Spectra Logic Corporation * 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. */ #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 /*--------------------------- Forward Declarations ---------------------------*/ static void xen_hvm_cpu_init(void); /*-------------------------------- Global Data -------------------------------*/ #ifdef SMP struct cpu_ops xen_hvm_cpu_ops = { .cpu_init = xen_hvm_cpu_init, .cpu_resume = xen_hvm_cpu_init }; #endif static MALLOC_DEFINE(M_XENHVM, "xen_hvm", "Xen HVM PV Support"); /** * If non-zero, the hypervisor has been configured to use a direct * IDT event callback for interrupt injection. */ int xen_vector_callback_enabled; /** * Signal whether the vector injected for the event channel upcall requires to * be EOI'ed on the local APIC. */ bool xen_evtchn_needs_ack; /*------------------------------- Per-CPU Data -------------------------------*/ DPCPU_DECLARE(struct vcpu_info *, vcpu_info); /*------------------------------ Sysctl tunables -----------------------------*/ int xen_disable_pv_disks = 0; int xen_disable_pv_nics = 0; TUNABLE_INT("hw.xen.disable_pv_disks", &xen_disable_pv_disks); TUNABLE_INT("hw.xen.disable_pv_nics", &xen_disable_pv_nics); /*---------------------- XEN Hypervisor Probe and Setup ----------------------*/ void xen_emergency_print(const char *str, size_t size) { outsb(XEN_HVM_DEBUGCONS_IOPORT, str, size); } uint32_t xen_cpuid_base; static uint32_t xen_hvm_cpuid_base(void) { uint32_t base, regs[4]; for (base = 0x40000000; base < 0x40010000; base += 0x100) { do_cpuid(base, regs); if (!memcmp("XenVMMXenVMM", ®s[1], 12) && (regs[0] - base) >= 2) return (base); } return (0); } static void hypervisor_quirks(unsigned int major, unsigned int minor) { #ifdef SMP if (((major < 4) || (major == 4 && minor <= 5)) && msix_disable_migration == -1) { /* * Xen hypervisors prior to 4.6.0 do not properly * handle updates to enabled MSI-X table entries, * so disable MSI-X interrupt migration in that * case. */ if (bootverbose) printf( "Disabling MSI-X interrupt migration due to Xen hypervisor bug.\n" "Set machdep.msix_disable_migration=0 to forcefully enable it.\n"); msix_disable_migration = 1; } #endif } static void hypervisor_version(void) { uint32_t regs[4]; int major, minor; do_cpuid(xen_cpuid_base + 1, regs); major = regs[0] >> 16; minor = regs[0] & 0xffff; printf("XEN: Hypervisor version %d.%d detected.\n", major, minor); hypervisor_quirks(major, minor); } /* * Allocate and fill in the hypcall page. */ int xen_hvm_init_hypercall_stubs(enum xen_hvm_init_type init_type) { uint32_t regs[4]; if (xen_cpuid_base != 0) /* Already setup. */ goto out; xen_cpuid_base = xen_hvm_cpuid_base(); if (xen_cpuid_base == 0) return (ENXIO); /* * Find the hypercall pages. */ do_cpuid(xen_cpuid_base + 2, regs); if (regs[0] != 1) return (EINVAL); wrmsr(regs[1], (init_type == XEN_HVM_INIT_EARLY) ? (vm_paddr_t)((uintptr_t)&hypercall_page - KERNBASE) : vtophys(&hypercall_page)); out: hypervisor_version(); return (0); } /* * Translate linear to physical address when still running on the bootloader * created page-tables. */ static vm_paddr_t early_init_vtop(void *addr) { /* * Using a KASSERT won't print anything, as this is before console * initialization. */ if (__predict_false((uintptr_t)addr < KERNBASE)) { xc_printf("invalid linear address: %#lx\n", (uintptr_t)addr); halt(); } return ((uintptr_t)addr - KERNBASE #ifdef __amd64__ + kernphys - KERNLOAD #endif ); } static int map_shared_info(void) { /* * TODO shared info page should be mapped in an unpopulated (IOW: * non-RAM) address. But finding one at this point in boot is * complicated, hence re-use a RAM address for the time being. This * sadly causes super-page shattering in the second stage translation * page tables. */ static union { shared_info_t shared_info; uint8_t raw[PAGE_SIZE]; } shared_page __attribute__((aligned(PAGE_SIZE))); static struct xen_add_to_physmap xatp = { .domid = DOMID_SELF, .space = XENMAPSPACE_shared_info, }; int rc; _Static_assert(sizeof(shared_page) == PAGE_SIZE, "invalid Xen shared_info struct size"); if (xatp.gpfn == 0) xatp.gpfn = atop(early_init_vtop(&shared_page.shared_info)); rc = HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp); if (rc != 0) { xc_printf("cannot map shared info page: %d\n", rc); HYPERVISOR_shared_info = NULL; } else if (HYPERVISOR_shared_info == NULL) HYPERVISOR_shared_info = &shared_page.shared_info; return (rc); } +static void +fixup_console(void) +{ + struct xen_platform_op op = { + .cmd = XENPF_get_dom0_console, + }; + xenpf_dom0_console_t *console = &op.u.dom0_console; + union { + struct efi_fb efi; + struct vbe_fb vbe; + } *fb = NULL; + int size; + caddr_t kmdp; + + kmdp = preload_search_by_type("elf kernel"); + if (kmdp == NULL) + kmdp = preload_search_by_type("elf64 kernel"); + if (kmdp == NULL) { + xc_printf("Unable to find kernel metadata\n"); + return; + } + + size = HYPERVISOR_platform_op(&op); + if (size < 0) { + xc_printf("Failed to get video console info: %d\n", size); + return; + } + + switch (console->video_type) { + case XEN_VGATYPE_VESA_LFB: + fb = (__typeof__ (fb))preload_search_info(kmdp, + MODINFO_METADATA | MODINFOMD_VBE_FB); + + if (fb == NULL) { + xc_printf("No VBE FB in kernel metadata\n"); + return; + } + + _Static_assert(offsetof(struct vbe_fb, fb_bpp) == + offsetof(struct efi_fb, fb_mask_reserved) + + sizeof(fb->efi.fb_mask_reserved), + "Bad structure overlay\n"); + fb->vbe.fb_bpp = console->u.vesa_lfb.bits_per_pixel; + /* FALLTHROUGH */ + case XEN_VGATYPE_EFI_LFB: + if (fb == NULL) { + fb = (__typeof__ (fb))preload_search_info(kmdp, + MODINFO_METADATA | MODINFOMD_EFI_FB); + if (fb == NULL) { + xc_printf("No EFI FB in kernel metadata\n"); + return; + } + } + + fb->efi.fb_addr = console->u.vesa_lfb.lfb_base; + if (size > + offsetof(xenpf_dom0_console_t, u.vesa_lfb.ext_lfb_base)) + fb->efi.fb_addr |= + (uint64_t)console->u.vesa_lfb.ext_lfb_base << 32; + fb->efi.fb_size = console->u.vesa_lfb.lfb_size << 16; + fb->efi.fb_height = console->u.vesa_lfb.height; + fb->efi.fb_width = console->u.vesa_lfb.width; + fb->efi.fb_stride = (console->u.vesa_lfb.bytes_per_line << 3) / + console->u.vesa_lfb.bits_per_pixel; +#define FBMASK(c) \ + ((~0u << console->u.vesa_lfb.c ## _pos) & \ + (~0u >> (32 - console->u.vesa_lfb.c ## _pos - \ + console->u.vesa_lfb.c ## _size))) + fb->efi.fb_mask_red = FBMASK(red); + fb->efi.fb_mask_green = FBMASK(green); + fb->efi.fb_mask_blue = FBMASK(blue); + fb->efi.fb_mask_reserved = FBMASK(rsvd); +#undef FBMASK + break; + + default: + xc_printf("Video console type unsupported\n"); + return; + } +} + /* Early initialization when running as a Xen guest. */ void xen_early_init(void) { uint32_t regs[4]; int rc; xen_cpuid_base = xen_hvm_cpuid_base(); if (xen_cpuid_base == 0) return; /* Find the hypercall pages. */ do_cpuid(xen_cpuid_base + 2, regs); if (regs[0] != 1) { xc_printf("Invalid number of hypercall pages %u\n", regs[0]); vm_guest = VM_GUEST_VM; return; } wrmsr(regs[1], early_init_vtop(&hypercall_page)); rc = map_shared_info(); if (rc != 0) { vm_guest = VM_GUEST_VM; return; } + + if (xen_initial_domain()) + /* Fixup video console information in case Xen changed the mode. */ + fixup_console(); } static void xen_hvm_init_shared_info_page(void) { struct xen_add_to_physmap xatp; if (xen_pv_domain()) { /* * Already setup in the PV case, shared_info is passed inside * of the start_info struct at start of day. */ return; } if (HYPERVISOR_shared_info == NULL) { HYPERVISOR_shared_info = malloc(PAGE_SIZE, M_XENHVM, M_NOWAIT); if (HYPERVISOR_shared_info == NULL) panic("Unable to allocate Xen shared info page"); } xatp.domid = DOMID_SELF; xatp.idx = 0; xatp.space = XENMAPSPACE_shared_info; xatp.gpfn = vtophys(HYPERVISOR_shared_info) >> PAGE_SHIFT; if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) panic("HYPERVISOR_memory_op failed"); } static int set_percpu_callback(unsigned int vcpu) { struct xen_hvm_evtchn_upcall_vector vec; int error; vec.vcpu = vcpu; vec.vector = IDT_EVTCHN; error = HYPERVISOR_hvm_op(HVMOP_set_evtchn_upcall_vector, &vec); return (error != 0 ? xen_translate_error(error) : 0); } /* * Tell the hypervisor how to contact us for event channel callbacks. */ void xen_hvm_set_callback(device_t dev) { struct xen_hvm_param xhp; int irq; if (xen_vector_callback_enabled) return; xhp.domid = DOMID_SELF; xhp.index = HVM_PARAM_CALLBACK_IRQ; if (xen_feature(XENFEAT_hvm_callback_vector) != 0) { int error; error = set_percpu_callback(0); if (error == 0) { xen_evtchn_needs_ack = true; /* Trick toolstack to think we are enlightened */ xhp.value = 1; } else xhp.value = HVM_CALLBACK_VECTOR(IDT_EVTCHN); error = HYPERVISOR_hvm_op(HVMOP_set_param, &xhp); if (error == 0) { xen_vector_callback_enabled = 1; return; } else if (xen_evtchn_needs_ack) panic("Unable to setup fake HVM param: %d", error); printf("Xen HVM callback vector registration failed (%d). " "Falling back to emulated device interrupt\n", error); } xen_vector_callback_enabled = 0; if (dev == NULL) { /* * Called from early boot or resume. * xenpci will invoke us again later. */ return; } irq = pci_get_irq(dev); if (irq < 16) { xhp.value = HVM_CALLBACK_GSI(irq); } else { u_int slot; u_int pin; slot = pci_get_slot(dev); pin = pci_get_intpin(dev) - 1; xhp.value = HVM_CALLBACK_PCI_INTX(slot, pin); } if (HYPERVISOR_hvm_op(HVMOP_set_param, &xhp) != 0) panic("Can't set evtchn callback"); } #define XEN_MAGIC_IOPORT 0x10 enum { XMI_MAGIC = 0x49d2, XMI_UNPLUG_IDE_DISKS = 0x01, XMI_UNPLUG_NICS = 0x02, XMI_UNPLUG_IDE_EXCEPT_PRI_MASTER = 0x04 }; static void xen_hvm_disable_emulated_devices(void) { u_short disable_devs = 0; if (xen_pv_domain()) { /* * No emulated devices in the PV case, so no need to unplug * anything. */ if (xen_disable_pv_disks != 0 || xen_disable_pv_nics != 0) printf("PV devices cannot be disabled in PV guests\n"); return; } if (inw(XEN_MAGIC_IOPORT) != XMI_MAGIC) return; if (xen_disable_pv_disks == 0) { if (bootverbose) printf("XEN: disabling emulated disks\n"); disable_devs |= XMI_UNPLUG_IDE_DISKS; } if (xen_disable_pv_nics == 0) { if (bootverbose) printf("XEN: disabling emulated nics\n"); disable_devs |= XMI_UNPLUG_NICS; } if (disable_devs != 0) outw(XEN_MAGIC_IOPORT, disable_devs); } static void xen_hvm_init(enum xen_hvm_init_type init_type) { int error; int i; if (!xen_domain() || init_type == XEN_HVM_INIT_CANCELLED_SUSPEND) return; error = xen_hvm_init_hypercall_stubs(init_type); switch (init_type) { case XEN_HVM_INIT_LATE: if (error != 0) return; setup_xen_features(); #ifdef SMP cpu_ops = xen_hvm_cpu_ops; #endif break; case XEN_HVM_INIT_RESUME: if (error != 0) panic("Unable to init Xen hypercall stubs on resume"); /* Clear stale vcpu_info. */ CPU_FOREACH(i) DPCPU_ID_SET(i, vcpu_info, NULL); break; default: panic("Unsupported HVM initialization type"); } xen_vector_callback_enabled = 0; xen_evtchn_needs_ack = false; xen_hvm_set_callback(NULL); /* * On (PV)HVM domains we need to request the hypervisor to * fill the shared info page, for PVH guest the shared_info page * is passed inside the start_info struct and is already set, so this * functions are no-ops. */ xen_hvm_init_shared_info_page(); xen_hvm_disable_emulated_devices(); } void xen_hvm_suspend(void) { } void xen_hvm_resume(bool suspend_cancelled) { xen_hvm_init(suspend_cancelled ? XEN_HVM_INIT_CANCELLED_SUSPEND : XEN_HVM_INIT_RESUME); /* Register vcpu_info area for CPU#0. */ xen_hvm_cpu_init(); } static void xen_hvm_sysinit(void *arg __unused) { xen_hvm_init(XEN_HVM_INIT_LATE); } SYSINIT(xen_hvm_init, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, xen_hvm_sysinit, NULL); static void xen_hvm_cpu_init(void) { uint32_t regs[4]; int rc; if (!xen_domain()) return; if (DPCPU_GET(vcpu_info) != NULL) { /* * vcpu_info is already set. We're resuming * from a failed migration and our pre-suspend * configuration is still valid. */ return; } /* * Set vCPU ID. If available fetch the ID from CPUID, if not just use * the ACPI ID. */ KASSERT(xen_cpuid_base != 0, ("Invalid base Xen CPUID leaf")); cpuid_count(xen_cpuid_base + 4, 0, regs); KASSERT((regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) || !xen_pv_domain(), ("Xen PV domain without vcpu_id in cpuid")); PCPU_SET(vcpu_id, (regs[0] & XEN_HVM_CPUID_VCPU_ID_PRESENT) ? regs[1] : PCPU_GET(acpi_id)); if (xen_evtchn_needs_ack && !IS_BSP()) { /* * Setup the per-vpcu event channel upcall vector. This is only * required when using the new HVMOP_set_evtchn_upcall_vector * hypercall, which allows using a different vector for each * vCPU. Note that FreeBSD uses the same vector for all vCPUs * because it's not dynamically allocated. */ rc = set_percpu_callback(PCPU_GET(vcpu_id)); if (rc != 0) panic("Event channel upcall vector setup failed: %d", rc); } xen_setup_vcpu_info(); } SYSINIT(xen_hvm_cpu_init, SI_SUB_INTR, SI_ORDER_FIRST, xen_hvm_cpu_init, NULL); bool xen_has_iommu_maps(void) { uint32_t regs[4]; KASSERT(xen_cpuid_base != 0, ("Invalid base Xen CPUID leaf")); cpuid_count(xen_cpuid_base + 4, 0, regs); return (regs[0] & XEN_HVM_CPUID_IOMMU_MAPPINGS); } diff --git a/sys/x86/xen/pv.c b/sys/x86/xen/pv.c index 515e5c58d304..e33fa41c83d7 100644 --- a/sys/x86/xen/pv.c +++ b/sys/x86/xen/pv.c @@ -1,575 +1,492 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2004 Christian Limpach. * Copyright (c) 2004-2006,2008 Kip Macy * Copyright (c) 2008 The NetBSD Foundation, Inc. * Copyright (c) 2013 Roger Pau MonnĂ© * 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. */ #include #include "opt_ddb.h" #include "opt_kstack_pages.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 #include #ifdef DDB #include #endif /* Native initial function */ extern u_int64_t hammer_time(u_int64_t, u_int64_t); /* Xen initial function */ uint64_t hammer_time_xen(vm_paddr_t); #define MAX_E820_ENTRIES 128 /*--------------------------- Forward Declarations ---------------------------*/ static caddr_t xen_pvh_parse_preload_data(uint64_t); static void pvh_parse_memmap(caddr_t, vm_paddr_t *, int *); /*---------------------------- Extern Declarations ---------------------------*/ /* * Placed by the linker at the end of the bss section, which is the last * section loaded by Xen before loading the symtab and strtab. */ extern uint32_t end; /*-------------------------------- Global Data -------------------------------*/ struct init_ops xen_pvh_init_ops = { .parse_preload_data = xen_pvh_parse_preload_data, .early_clock_source_init = xen_clock_init, .early_delay = xen_delay, .parse_memmap = pvh_parse_memmap, }; static struct bios_smap xen_smap[MAX_E820_ENTRIES]; static struct hvm_start_info *start_info; /*-------------------------------- Xen PV init -------------------------------*/ static int isxen(void) { static int xen = -1; uint32_t base; u_int regs[4]; if (xen != -1) return (xen); /* * The full code for identifying which hypervisor we're running under * is in sys/x86/x86/identcpu.c and runs later in the boot process; * this is sufficient to distinguish Xen PVH booting from non-Xen PVH * and skip some very early Xen-specific code in the non-Xen case. */ xen = 0; for (base = 0x40000000; base < 0x40010000; base += 0x100) { do_cpuid(base, regs); if (regs[1] == XEN_CPUID_SIGNATURE_EBX && regs[2] == XEN_CPUID_SIGNATURE_ECX && regs[3] == XEN_CPUID_SIGNATURE_EDX) { xen = 1; break; } } return (xen); } #define CRASH(...) do { \ if (isxen()) { \ xc_printf(__VA_ARGS__); \ HYPERVISOR_shutdown(SHUTDOWN_crash); \ } else { \ halt(); \ } \ } while (0) uint64_t hammer_time_xen(vm_paddr_t start_info_paddr) { struct hvm_modlist_entry *mod; uint64_t physfree; char *kenv; if (isxen()) { vm_guest = VM_GUEST_XEN; xen_early_init(); if (xen_cpuid_base == 0) { xc_printf( "ERROR: failed to initialize hypercall page\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } } start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE); if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) { CRASH("Unknown magic value in start_info struct: %#x\n", start_info->magic); } /* * Select the higher address to use as physfree: either after * start_info, after the kernel, after the memory map or after any of * the modules. We assume enough memory to be available after the * selected address for the needs of very early memory allocations. */ physfree = roundup2(start_info_paddr + sizeof(struct hvm_start_info), PAGE_SIZE); physfree = MAX(roundup2((vm_paddr_t)_end - KERNBASE, PAGE_SIZE), physfree); if (start_info->memmap_paddr != 0) physfree = MAX(roundup2(start_info->memmap_paddr + start_info->memmap_entries * sizeof(struct hvm_memmap_table_entry), PAGE_SIZE), physfree); if (start_info->modlist_paddr != 0) { unsigned int i; if (start_info->nr_modules == 0) { CRASH( "ERROR: modlist_paddr != 0 but nr_modules == 0\n"); } mod = (struct hvm_modlist_entry *) (start_info->modlist_paddr + KERNBASE); for (i = 0; i < start_info->nr_modules; i++) physfree = MAX(roundup2(mod[i].paddr + mod[i].size, PAGE_SIZE), physfree); } /* * Init a static kenv using a free page. The contents will be filled * from the parse_preload_data hook. */ kenv = (void *)(physfree + KERNBASE); physfree += PAGE_SIZE; bzero_early(kenv, PAGE_SIZE); init_static_kenv(kenv, PAGE_SIZE); /* Set the hooks for early functions that diverge from bare metal */ init_ops = xen_pvh_init_ops; hvm_start_flags = start_info->flags; /* Now we can jump into the native init function */ return (hammer_time(0, physfree)); } /*-------------------------------- PV specific -------------------------------*/ /* * When booted as a PVH guest FreeBSD needs to avoid using the RSDP address * hint provided by the loader because it points to the native set of ACPI * tables instead of the ones crafted by Xen. The acpi.rsdp env variable is * removed from kenv if present, and a new acpi.rsdp is added to kenv that * points to the address of the Xen crafted RSDP. */ static bool reject_option(const char *option) { static const char *reject[] = { "acpi.rsdp", }; unsigned int i; for (i = 0; i < nitems(reject); i++) if (strncmp(option, reject[i], strlen(reject[i])) == 0) return (true); return (false); } static void xen_pvh_set_env(char *env, bool (*filter)(const char *)) { char *option; if (env == NULL) return; option = env; while (*option != 0) { char *value; if (filter != NULL && filter(option)) { option += strlen(option) + 1; continue; } value = option; option = strsep(&value, "="); if (kern_setenv(option, value) != 0 && isxen()) xc_printf("unable to add kenv %s=%s\n", option, value); option = value + strlen(value) + 1; } } #ifdef DDB /* * The way Xen loads the symtab is different from the native boot loader, * because it's tailored for NetBSD. So we have to adapt and use the same * method as NetBSD. Portions of the code below have been picked from NetBSD: * sys/kern/kern_ksyms.c CVS Revision 1.71. */ static void xen_pvh_parse_symtab(void) { Elf_Ehdr *ehdr; Elf_Shdr *shdr; int i, j; ehdr = (Elf_Ehdr *)(&end + 1); if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) || ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || ehdr->e_version > 1) { if (isxen()) xc_printf("Unable to load ELF symtab: invalid symbol table\n"); return; } shdr = (Elf_Shdr *)((uint8_t *)ehdr + ehdr->e_shoff); /* Find the symbol table and the corresponding string table. */ for (i = 1; i < ehdr->e_shnum; i++) { if (shdr[i].sh_type != SHT_SYMTAB) continue; if (shdr[i].sh_offset == 0) continue; ksymtab = (uintptr_t)((uint8_t *)ehdr + shdr[i].sh_offset); ksymtab_size = shdr[i].sh_size; j = shdr[i].sh_link; if (shdr[j].sh_offset == 0) continue; /* Can this happen? */ kstrtab = (uintptr_t)((uint8_t *)ehdr + shdr[j].sh_offset); break; } if ((ksymtab == 0 || kstrtab == 0) && isxen()) xc_printf( "Unable to load ELF symtab: could not find symtab or strtab\n"); } #endif -static void -fixup_console(void) -{ - struct xen_platform_op op = { - .cmd = XENPF_get_dom0_console, - }; - xenpf_dom0_console_t *console = &op.u.dom0_console; - union { - struct efi_fb efi; - struct vbe_fb vbe; - } *fb = NULL; - int size; - caddr_t kmdp; - - kmdp = preload_search_by_type("elf kernel"); - if (kmdp == NULL) - kmdp = preload_search_by_type("elf64 kernel"); - if (kmdp == NULL) { - xc_printf("Unable to find kernel metadata\n"); - return; - } - - size = HYPERVISOR_platform_op(&op); - if (size < 0) { - xc_printf("Failed to get video console info: %d\n", size); - return; - } - - switch (console->video_type) { - case XEN_VGATYPE_VESA_LFB: - fb = (__typeof__ (fb))preload_search_info(kmdp, - MODINFO_METADATA | MODINFOMD_VBE_FB); - - if (fb == NULL) { - xc_printf("No VBE FB in kernel metadata\n"); - return; - } - - _Static_assert(offsetof(struct vbe_fb, fb_bpp) == - offsetof(struct efi_fb, fb_mask_reserved) + - sizeof(fb->efi.fb_mask_reserved), - "Bad structure overlay\n"); - fb->vbe.fb_bpp = console->u.vesa_lfb.bits_per_pixel; - /* FALLTHROUGH */ - case XEN_VGATYPE_EFI_LFB: - if (fb == NULL) { - fb = (__typeof__ (fb))preload_search_info(kmdp, - MODINFO_METADATA | MODINFOMD_EFI_FB); - if (fb == NULL) { - xc_printf("No EFI FB in kernel metadata\n"); - return; - } - } - - fb->efi.fb_addr = console->u.vesa_lfb.lfb_base; - if (size > - offsetof(xenpf_dom0_console_t, u.vesa_lfb.ext_lfb_base)) - fb->efi.fb_addr |= - (uint64_t)console->u.vesa_lfb.ext_lfb_base << 32; - fb->efi.fb_size = console->u.vesa_lfb.lfb_size << 16; - fb->efi.fb_height = console->u.vesa_lfb.height; - fb->efi.fb_width = console->u.vesa_lfb.width; - fb->efi.fb_stride = (console->u.vesa_lfb.bytes_per_line << 3) / - console->u.vesa_lfb.bits_per_pixel; -#define FBMASK(c) \ - ((~0u << console->u.vesa_lfb.c ## _pos) & \ - (~0u >> (32 - console->u.vesa_lfb.c ## _pos - \ - console->u.vesa_lfb.c ## _size))) - fb->efi.fb_mask_red = FBMASK(red); - fb->efi.fb_mask_green = FBMASK(green); - fb->efi.fb_mask_blue = FBMASK(blue); - fb->efi.fb_mask_reserved = FBMASK(rsvd); -#undef FBMASK - break; - - default: - xc_printf("Video console type unsupported\n"); - return; - } -} - static caddr_t xen_pvh_parse_preload_data(uint64_t modulep) { caddr_t kmdp; vm_ooffset_t off; vm_paddr_t metadata; char *envp; char acpi_rsdp[19]; TSENTER(); if (start_info->modlist_paddr != 0) { struct hvm_modlist_entry *mod; const char *cmdline; mod = (struct hvm_modlist_entry *) (start_info->modlist_paddr + KERNBASE); cmdline = mod[0].cmdline_paddr ? (const char *)(mod[0].cmdline_paddr + KERNBASE) : NULL; if (strcmp(cmdline, "header") == 0) { struct xen_header *header; header = (struct xen_header *)(mod[0].paddr + KERNBASE); if ((header->flags & XENHEADER_HAS_MODULEP_OFFSET) != XENHEADER_HAS_MODULEP_OFFSET) { xc_printf("Unable to load module metadata\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } preload_metadata = (caddr_t)(mod[0].paddr + header->modulep_offset + KERNBASE); kmdp = preload_search_by_type("elf kernel"); if (kmdp == NULL) kmdp = preload_search_by_type("elf64 kernel"); if (kmdp == NULL) { xc_printf("Unable to find kernel\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } /* * Xen has relocated the metadata and the modules, so * we need to recalculate it's position. This is done * by saving the original modulep address and then * calculating the offset from the real modulep * position. */ metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t); off = mod[0].paddr + header->modulep_offset - metadata + KERNBASE; } else { preload_metadata = (caddr_t)(mod[0].paddr + KERNBASE); kmdp = preload_search_by_type("elf kernel"); if (kmdp == NULL) kmdp = preload_search_by_type("elf64 kernel"); if (kmdp == NULL) { xc_printf("Unable to find kernel\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t); off = mod[0].paddr + KERNBASE - metadata; } preload_bootstrap_relocate(off); boothowto = MD_FETCH(kmdp, MODINFOMD_HOWTO, int); envp = MD_FETCH(kmdp, MODINFOMD_ENVP, char *); if (envp != NULL) envp += off; xen_pvh_set_env(envp, reject_option); if (MD_FETCH(kmdp, MODINFOMD_EFI_MAP, void *) != NULL) strlcpy(bootmethod, "UEFI", sizeof(bootmethod)); else strlcpy(bootmethod, "BIOS", sizeof(bootmethod)); - - fixup_console(); } else { /* Parse the extra boot information given by Xen */ if (start_info->cmdline_paddr != 0) boot_parse_cmdline_delim( (char *)(start_info->cmdline_paddr + KERNBASE), ", \t\n"); kmdp = NULL; strlcpy(bootmethod, "PVH", sizeof(bootmethod)); } boothowto |= boot_env_to_howto(); snprintf(acpi_rsdp, sizeof(acpi_rsdp), "%#" PRIx64, start_info->rsdp_paddr); kern_setenv("acpi.rsdp", acpi_rsdp); #ifdef DDB xen_pvh_parse_symtab(); #endif TSEXIT(); return (kmdp); } static void pvh_parse_memmap_start_info(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx) { const struct hvm_memmap_table_entry * entries; size_t nentries; size_t i; /* Extract from HVM start_info. */ entries = (struct hvm_memmap_table_entry *)(start_info->memmap_paddr + KERNBASE); nentries = start_info->memmap_entries; /* Convert into E820 format and handle one by one. */ for (i = 0; i < nentries; i++) { struct bios_smap entry; entry.base = entries[i].addr; entry.length = entries[i].size; /* * Luckily for us, the XEN_HVM_MEMMAP_TYPE_* values exactly * match the SMAP_TYPE_* values so we don't need to translate * anything here. */ entry.type = entries[i].type; bios_add_smap_entries(&entry, 1, physmap, physmap_idx); } } static void xen_pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx) { struct xen_memory_map memmap; u_int32_t size; int rc; /* We should only reach here if we're running under Xen. */ KASSERT(isxen(), ("xen_pvh_parse_memmap reached when !Xen")); /* Fetch the E820 map from Xen */ memmap.nr_entries = MAX_E820_ENTRIES; set_xen_guest_handle(memmap.buffer, xen_smap); rc = HYPERVISOR_memory_op(XENMEM_memory_map, &memmap); if (rc) { xc_printf("ERROR: unable to fetch Xen E820 memory map: %d\n", rc); HYPERVISOR_shutdown(SHUTDOWN_crash); } size = memmap.nr_entries * sizeof(xen_smap[0]); bios_add_smap_entries(xen_smap, size, physmap, physmap_idx); } static void pvh_parse_memmap(caddr_t kmdp, vm_paddr_t *physmap, int *physmap_idx) { /* * If version >= 1 and memmap_paddr != 0, use the memory map provided * in the start_info structure; if not, we're running under legacy * Xen and need to use the Xen hypercall. */ if ((start_info->version >= 1) && (start_info->memmap_paddr != 0)) pvh_parse_memmap_start_info(kmdp, physmap, physmap_idx); else xen_pvh_parse_memmap(kmdp, physmap, physmap_idx); }