diff --git a/stand/i386/libi386/multiboot.c b/stand/i386/libi386/multiboot.c index a29696e4baec..71fd63bb4c88 100644 --- a/stand/i386/libi386/multiboot.c +++ b/stand/i386/libi386/multiboot.c @@ -1,467 +1,389 @@ /*- * Copyright (c) 2014 Roger Pau Monné * 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. */ /* * This multiboot implementation only implements a subset of the full * multiboot specification in order to be able to boot Xen and a * FreeBSD Dom0. Trying to use it to boot other multiboot compliant * kernels will most surely fail. * * The full multiboot specification can be found here: * http://www.gnu.org/software/grub/manual/multiboot/multiboot.html */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #define _MACHINE_ELF_WANT_32BIT #include +#include #include #include #include "bootstrap.h" #include "multiboot.h" #include "libi386.h" #include #define MULTIBOOT_SUPPORTED_FLAGS \ (MULTIBOOT_PAGE_ALIGN|MULTIBOOT_MEMORY_INFO) #define NUM_MODULES 2 -#define METADATA_FIXED_SIZE (PAGE_SIZE*4) -#define METADATA_MODULE_SIZE PAGE_SIZE - -#define METADATA_RESV_SIZE(mod_num) \ - roundup(METADATA_FIXED_SIZE + METADATA_MODULE_SIZE * mod_num, PAGE_SIZE) extern int elf32_loadfile_raw(char *filename, uint64_t dest, struct preloaded_file **result, int multiboot); extern int elf64_load_modmetadata(struct preloaded_file *fp, uint64_t dest); extern int elf64_obj_loadfile(char *filename, uint64_t dest, struct preloaded_file **result); static int multiboot_loadfile(char *, uint64_t, struct preloaded_file **); static int multiboot_exec(struct preloaded_file *); static int multiboot_obj_loadfile(char *, uint64_t, struct preloaded_file **); static int multiboot_obj_exec(struct preloaded_file *fp); struct file_format multiboot = { multiboot_loadfile, multiboot_exec }; struct file_format multiboot_obj = { multiboot_obj_loadfile, multiboot_obj_exec }; extern void multiboot_tramp(); static const char mbl_name[] = "FreeBSD Loader"; -static int -num_modules(struct preloaded_file *kfp) -{ - struct kernel_module *kmp; - int mod_num = 0; - - for (kmp = kfp->f_modules; kmp != NULL; kmp = kmp->m_next) - mod_num++; - - return (mod_num); -} - -static vm_offset_t -max_addr(void) -{ - struct preloaded_file *fp; - vm_offset_t addr = 0; - - for (fp = file_findfile(NULL, NULL); fp != NULL; fp = fp->f_next) { - if (addr < (fp->f_addr + fp->f_size)) - addr = fp->f_addr + fp->f_size; - } - - return (addr); -} - static int multiboot_loadfile(char *filename, uint64_t dest, struct preloaded_file **result) { uint32_t *magic; int i, error; caddr_t header_search; ssize_t search_size; int fd; struct multiboot_header *header; char *cmdline; /* * Read MULTIBOOT_SEARCH size in order to search for the * multiboot magic header. */ if (filename == NULL) return (EFTYPE); if ((fd = open(filename, O_RDONLY)) == -1) return (errno); header_search = malloc(MULTIBOOT_SEARCH); if (header_search == NULL) { close(fd); return (ENOMEM); } search_size = read(fd, header_search, MULTIBOOT_SEARCH); magic = (uint32_t *)header_search; header = NULL; for (i = 0; i < (search_size / sizeof(uint32_t)); i++) { if (magic[i] == MULTIBOOT_HEADER_MAGIC) { header = (struct multiboot_header *)&magic[i]; break; } } if (header == NULL) { error = EFTYPE; goto out; } /* Valid multiboot header has been found, validate checksum */ if (header->magic + header->flags + header->checksum != 0) { printf( "Multiboot checksum failed, magic: 0x%x flags: 0x%x checksum: 0x%x\n", header->magic, header->flags, header->checksum); error = EFTYPE; goto out; } if ((header->flags & ~MULTIBOOT_SUPPORTED_FLAGS) != 0) { printf("Unsupported multiboot flags found: 0x%x\n", header->flags); error = EFTYPE; goto out; } error = elf32_loadfile_raw(filename, dest, result, 1); if (error != 0) { printf( "elf32_loadfile_raw failed: %d unable to load multiboot kernel\n", error); goto out; } /* * f_addr is already aligned to PAGE_SIZE, make sure * f_size it's also aligned so when the modules are loaded * they are aligned to PAGE_SIZE. */ (*result)->f_size = roundup((*result)->f_size, PAGE_SIZE); out: free(header_search); close(fd); return (error); } static int multiboot_exec(struct preloaded_file *fp) { - vm_offset_t module_start, last_addr, metadata_size; vm_offset_t modulep, kernend, entry; struct file_metadata *md; Elf_Ehdr *ehdr; struct multiboot_info *mb_info = NULL; struct multiboot_mod_list *mb_mod = NULL; char *cmdline = NULL; size_t len; int error, mod_num; + struct xen_header header; + + CTASSERT(sizeof(header) <= PAGE_SIZE); /* * Don't pass the memory size found by the bootloader, the memory * available to Dom0 will be lower than that. */ unsetenv("smbios.memory.enabled"); /* Allocate the multiboot struct and fill the basic details. */ mb_info = malloc(sizeof(struct multiboot_info)); if (mb_info == NULL) { error = ENOMEM; goto error; } bzero(mb_info, sizeof(struct multiboot_info)); mb_info->flags = MULTIBOOT_INFO_MEMORY|MULTIBOOT_INFO_BOOT_LOADER_NAME; mb_info->mem_lower = bios_basemem / 1024; mb_info->mem_upper = bios_extmem / 1024; mb_info->boot_loader_name = VTOP(mbl_name); /* Set the Xen command line. */ if (fp->f_args == NULL) { /* Add the Xen command line if it is set. */ cmdline = getenv("xen_cmdline"); if (cmdline != NULL) { fp->f_args = strdup(cmdline); if (fp->f_args == NULL) { error = ENOMEM; goto error; } } } if (fp->f_args != NULL) { len = strlen(fp->f_name) + 1 + strlen(fp->f_args) + 1; cmdline = malloc(len); if (cmdline == NULL) { error = ENOMEM; goto error; } snprintf(cmdline, len, "%s %s", fp->f_name, fp->f_args); mb_info->cmdline = VTOP(cmdline); mb_info->flags |= MULTIBOOT_INFO_CMDLINE; } /* Find the entry point of the Xen kernel and save it for later */ if ((md = file_findmetadata(fp, MODINFOMD_ELFHDR)) == NULL) { printf("Unable to find %s entry point\n", fp->f_name); error = EINVAL; goto error; } ehdr = (Elf_Ehdr *)&(md->md_data); entry = ehdr->e_entry & 0xffffff; /* * Prepare the multiboot module list, Xen assumes the first * module is the Dom0 kernel, and the second one is the initramfs. * This is not optimal for FreeBSD, that doesn't have a initramfs * but instead loads modules dynamically and creates the metadata * info on-the-fly. * * As expected, the first multiboot module is going to be the * FreeBSD kernel loaded as a raw file. The second module is going * to contain the metadata info and the loaded modules. * - * On native FreeBSD loads all the modules and then places the - * metadata info at the end, but this is painful when running on Xen, - * because it relocates the second multiboot module wherever it - * likes. In order to workaround this limitation the metadata - * information is placed at the start of the second module and - * the original modulep value is saved together with the other - * metadata, so we can relocate everything. + * There's a small header prefixed in the second module that contains + * some information required to calculate the relocated address of + * modulep based on the original offset of modulep from the start of + * the module address. Note other fields might be added to this header + * if required. * * Native layout: * fp->f_addr + fp->f_size * +---------+----------------+------------+ * | | | | * | Kernel | Modules | Metadata | * | | | | * +---------+----------------+------------+ * fp->f_addr modulep kernend * - * Xen layout: - * - * Initial: - * fp->f_addr + fp->f_size - * +---------+----------+----------------+------------+ - * | | | | | - * | Kernel | Reserved | Modules | Metadata | - * | | | | dry run | - * +---------+----------+----------------+------------+ - * fp->f_addr - * - * After metadata polacement (ie: final): - * fp->f_addr + fp->f_size - * +-----------+---------+----------+----------------+ - * | | | | | - * | Kernel | Free | Metadata | Modules | - * | | | | | - * +-----------+---------+----------+----------------+ - * fp->f_addr modulep kernend - * \__________/ \__________________________/ - * Multiboot module 0 Multiboot module 1 + * Xen dom0 layout: + * fp->f_addr fp->f_addr + fp->f_size + * +---------+------------+----------------+------------+ + * | | | | | + * | Kernel | xen_header | Modules | Metadata | + * | | | | | + * +---------+------------+----------------+------------+ + * modulep kernend + * \________/\__________________________________________/ + * module 0 module 1 */ fp = file_findfile(NULL, "elf kernel"); if (fp == NULL) { printf("No FreeBSD kernel provided, aborting\n"); error = EINVAL; goto error; } - if (fp->f_metadata != NULL) { - printf("FreeBSD kernel already contains metadata, aborting\n"); - error = EINVAL; - goto error; - } - - mb_mod = malloc(sizeof(struct multiboot_mod_list) * NUM_MODULES); if (mb_mod == NULL) { error = ENOMEM; goto error; } bzero(mb_mod, sizeof(struct multiboot_mod_list) * NUM_MODULES); - /* - * Calculate how much memory is needed for the metatdata. We did - * an approximation of the maximum size when loading the kernel, - * but now we know the exact size, so we can release some of this - * preallocated memory if not needed. - */ - last_addr = roundup(max_addr(), PAGE_SIZE); - mod_num = num_modules(fp); - - /* - * Place the metadata after the last used address in order to - * calculate it's size, this will not be used. - */ - error = bi_load64(fp->f_args, last_addr, &modulep, &kernend, 0); + error = bi_load64(fp->f_args, 0, &modulep, &kernend, 0); if (error != 0) { printf("bi_load64 failed: %d\n", error); goto error; } - metadata_size = roundup(kernend - last_addr, PAGE_SIZE); - /* Check that the size is not greater than what we have reserved */ - if (metadata_size > METADATA_RESV_SIZE(mod_num)) { - printf("Required memory for metadata is greater than reserved " - "space, please increase METADATA_FIXED_SIZE and " - "METADATA_MODULE_SIZE and rebuild the loader\n"); + mb_mod[0].mod_start = fp->f_addr; + mb_mod[0].mod_end = fp->f_addr + fp->f_size - PAGE_SIZE; + + mb_mod[1].mod_start = mb_mod[0].mod_end; + mb_mod[1].mod_end = kernend; + /* Indicate the kernel metadata is prefixed with a xen_header. */ + cmdline = strdup("header"); + if (cmdline == NULL) { + printf("Out of memory, aborting\n"); error = ENOMEM; goto error; } - - /* Clean the metadata added to the kernel in the bi_load64 dry run */ - file_removemetadata(fp); - - /* - * This is the position where the second multiboot module - * will be placed. - */ - module_start = fp->f_addr + fp->f_size - metadata_size; - - error = bi_load64(fp->f_args, module_start, &modulep, &kernend, 0); - if (error != 0) { - printf("bi_load64 failed: %d\n", error); - goto error; - } - - mb_mod[0].mod_start = fp->f_addr; - mb_mod[0].mod_end = fp->f_addr + fp->f_size; - mb_mod[0].mod_end -= METADATA_RESV_SIZE(mod_num); - - mb_mod[1].mod_start = module_start; - mb_mod[1].mod_end = last_addr; + mb_mod[1].cmdline = VTOP(cmdline); mb_info->mods_count = NUM_MODULES; mb_info->mods_addr = VTOP(mb_mod); mb_info->flags |= MULTIBOOT_INFO_MODS; + header.flags = XENHEADER_HAS_MODULEP_OFFSET; + header.modulep_offset = modulep - mb_mod[1].mod_start; + archsw.arch_copyin(&header, mb_mod[1].mod_start, sizeof(header)); + dev_cleanup(); __exec((void *)VTOP(multiboot_tramp), (void *)entry, (void *)VTOP(mb_info)); panic("exec returned"); error: if (mb_mod) free(mb_mod); if (mb_info) free(mb_info); if (cmdline) free(cmdline); return (error); } static int multiboot_obj_loadfile(char *filename, uint64_t dest, struct preloaded_file **result) { struct preloaded_file *mfp, *kfp, *rfp; struct kernel_module *kmp; int error, mod_num; /* See if there's a multiboot kernel loaded */ mfp = file_findfile(NULL, "elf multiboot kernel"); if (mfp == NULL) return (EFTYPE); /* * We have a multiboot kernel loaded, see if there's a FreeBSD * kernel loaded also. */ kfp = file_findfile(NULL, "elf kernel"); if (kfp == NULL) { /* * No kernel loaded, this must be it. The kernel has to * be loaded as a raw file, it will be processed by * Xen and correctly loaded as an ELF file. */ rfp = file_loadraw(filename, "elf kernel", 0); if (rfp == NULL) { printf( "Unable to load %s as a multiboot payload kernel\n", filename); return (EINVAL); } /* Load kernel metadata... */ setenv("kernelname", filename, 1); error = elf64_load_modmetadata(rfp, rfp->f_addr + rfp->f_size); if (error) { printf("Unable to load kernel %s metadata error: %d\n", rfp->f_name, error); return (EINVAL); } + /* - * Save space at the end of the kernel in order to place - * the metadata information. We do an approximation of the - * max metadata size, this is not optimal but it's probably - * the best we can do at this point. Once all modules are - * loaded and the size of the metadata is known this - * space will be recovered if not used. + * Reserve one page at the end of the kernel to place some + * metadata in order to cope for Xen relocating the modules and + * the metadata information. */ - mod_num = num_modules(rfp); rfp->f_size = roundup(rfp->f_size, PAGE_SIZE); - rfp->f_size += METADATA_RESV_SIZE(mod_num); + rfp->f_size += PAGE_SIZE; *result = rfp; } else { /* The rest should be loaded as regular modules */ error = elf64_obj_loadfile(filename, dest, result); if (error != 0) { printf("Unable to load %s as an object file, error: %d", filename, error); return (error); } } return (0); } static int multiboot_obj_exec(struct preloaded_file *fp) { return (EFTYPE); } diff --git a/sys/x86/include/metadata.h b/sys/x86/include/metadata.h index d157df0307fd..be0eae43d34c 100644 --- a/sys/x86/include/metadata.h +++ b/sys/x86/include/metadata.h @@ -1,71 +1,92 @@ /*- * Copyright (c) 2003 Peter Wemm * 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. * * $FreeBSD$ */ #ifndef _MACHINE_METADATA_H_ #define _MACHINE_METADATA_H_ #define MODINFOMD_SMAP 0x1001 #define MODINFOMD_SMAP_XATTR 0x1002 #define MODINFOMD_DTBP 0x1003 #define MODINFOMD_EFI_MAP 0x1004 #define MODINFOMD_EFI_FB 0x1005 #define MODINFOMD_MODULEP 0x1006 #define MODINFOMD_VBE_FB 0x1007 struct efi_map_header { uint64_t memory_size; uint64_t descriptor_size; uint32_t descriptor_version; }; struct efi_fb { uint64_t fb_addr; uint64_t fb_size; uint32_t fb_height; uint32_t fb_width; uint32_t fb_stride; uint32_t fb_mask_red; uint32_t fb_mask_green; uint32_t fb_mask_blue; uint32_t fb_mask_reserved; }; struct vbe_fb { uint64_t fb_addr; uint64_t fb_size; uint32_t fb_height; uint32_t fb_width; uint32_t fb_stride; uint32_t fb_mask_red; uint32_t fb_mask_green; uint32_t fb_mask_blue; uint32_t fb_mask_reserved; uint32_t fb_bpp; }; +/* + * The structure below is used when FreeBSD kernel is booted as a dom0 kernel + * from Xen. In such scenario we need to accommodate the modules and the + * metadata as a contiguous memory region, so it can be passed as a multiboot + * module, and some extra information is required which is conveyed from the + * loader to the kernel using the xen_header structure below. + * + * See the comment in multiboot.c about how the structure below is packaged + * together with the rest of the kernel payload data. + */ +struct xen_header { + uint64_t flags; +#define XENHEADER_HAS_MODULEP_OFFSET (1ull << 0) + + /* + * Offset of the modulep location from the start of the multiboot + * module blob. + */ + uint64_t modulep_offset; +}; + #endif /* !_MACHINE_METADATA_H_ */ diff --git a/sys/x86/xen/pv.c b/sys/x86/xen/pv.c index 34fcbd990491..2fd698772f9d 100644 --- a/sys/x86/xen/pv.c +++ b/sys/x86/xen/pv.c @@ -1,639 +1,674 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-NetBSD * * Copyright (c) 2004 Christian Limpach. * Copyright (c) 2004-2006,2008 Kip Macy * Copyright (c) 2008 The NetBSD Foundation, Inc. * Copyright (c) 2013 Roger Pau Monné * 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 __FBSDID("$FreeBSD$"); #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 #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_legacy(start_info_t *, uint64_t); uint64_t hammer_time_xen(vm_paddr_t); #define MAX_E820_ENTRIES 128 /*--------------------------- Forward Declarations ---------------------------*/ static caddr_t xen_legacy_pvh_parse_preload_data(uint64_t); static caddr_t xen_pvh_parse_preload_data(uint64_t); static void xen_pvh_parse_memmap(caddr_t, vm_paddr_t *, int *); #ifdef SMP static int xen_pv_start_all_aps(void); #endif /*---------------------------- Extern Declarations ---------------------------*/ #ifdef SMP /* Variables used by amd64 mp_machdep to start APs */ extern char *doublefault_stack; extern char *mce_stack; extern char *nmi_stack; extern char *dbg_stack; #endif /* * 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 -------------------------------*/ /* Xen init_ops implementation. */ struct init_ops xen_legacy_init_ops = { .parse_preload_data = xen_legacy_pvh_parse_preload_data, .early_clock_source_init = xen_clock_init, .early_delay = xen_delay, .parse_memmap = xen_pvh_parse_memmap, #ifdef SMP .start_all_aps = xen_pv_start_all_aps, #endif .msi_init = xen_msi_init, }; 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 = xen_pvh_parse_memmap, #ifdef SMP .mp_bootaddress = mp_bootaddress, .start_all_aps = native_start_all_aps, #endif .msi_init = msi_init, }; static struct bios_smap xen_smap[MAX_E820_ENTRIES]; static start_info_t *legacy_start_info; static struct hvm_start_info *start_info; /*----------------------- Legacy PVH start_info accessors --------------------*/ static vm_paddr_t legacy_get_xenstore_mfn(void) { return (legacy_start_info->store_mfn); } static evtchn_port_t legacy_get_xenstore_evtchn(void) { return (legacy_start_info->store_evtchn); } static vm_paddr_t legacy_get_console_mfn(void) { return (legacy_start_info->console.domU.mfn); } static evtchn_port_t legacy_get_console_evtchn(void) { return (legacy_start_info->console.domU.evtchn); } static uint32_t legacy_get_start_flags(void) { return (legacy_start_info->flags); } struct hypervisor_info legacy_info = { .get_xenstore_mfn = legacy_get_xenstore_mfn, .get_xenstore_evtchn = legacy_get_xenstore_evtchn, .get_console_mfn = legacy_get_console_mfn, .get_console_evtchn = legacy_get_console_evtchn, .get_start_flags = legacy_get_start_flags, }; /*-------------------------------- Xen PV init -------------------------------*/ /* * First function called by the Xen legacy PVH boot sequence. * * Set some Xen global variables and prepare the environment so it is * as similar as possible to what native FreeBSD init function expects. */ uint64_t hammer_time_xen_legacy(start_info_t *si, uint64_t xenstack) { uint64_t physfree; uint64_t *PT4 = (u_int64_t *)xenstack; uint64_t *PT3 = (u_int64_t *)(xenstack + PAGE_SIZE); uint64_t *PT2 = (u_int64_t *)(xenstack + 2 * PAGE_SIZE); int i; char *kenv; xen_domain_type = XEN_PV_DOMAIN; vm_guest = VM_GUEST_XEN; if ((si == NULL) || (xenstack == 0)) { xc_printf("ERROR: invalid start_info or xen stack, halting\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } xc_printf("FreeBSD PVH running on %s\n", si->magic); /* We use 3 pages of xen stack for the boot pagetables */ physfree = xenstack + 3 * PAGE_SIZE - KERNBASE; /* Setup Xen global variables */ legacy_start_info = si; HYPERVISOR_shared_info = (shared_info_t *)(si->shared_info + KERNBASE); /* * Use the stack Xen gives us to build the page tables * as native FreeBSD expects to find them (created * by the boot trampoline). */ for (i = 0; i < (PAGE_SIZE / sizeof(uint64_t)); i++) { /* * Each slot of the level 4 pages points * to the same level 3 page */ PT4[i] = ((uint64_t)&PT3[0]) - KERNBASE; PT4[i] |= PG_V | PG_RW | PG_U; /* * Each slot of the level 3 pages points * to the same level 2 page */ PT3[i] = ((uint64_t)&PT2[0]) - KERNBASE; PT3[i] |= PG_V | PG_RW | PG_U; /* * The level 2 page slots are mapped with * 2MB pages for 1GB. */ PT2[i] = i * (2 * 1024 * 1024); PT2[i] |= PG_V | PG_RW | PG_PS | PG_U; } load_cr3(((uint64_t)&PT4[0]) - KERNBASE); /* * Init an empty 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_legacy_init_ops; apic_ops = xen_apic_ops; hypervisor_info = legacy_info; /* Now we can jump into the native init function */ return (hammer_time(0, physfree)); } uint64_t hammer_time_xen(vm_paddr_t start_info_paddr) { struct hvm_modlist_entry *mod; struct xen_add_to_physmap xatp; uint64_t physfree; char *kenv; int rc; xen_domain_type = XEN_HVM_DOMAIN; vm_guest = VM_GUEST_XEN; rc = xen_hvm_init_hypercall_stubs(XEN_HVM_INIT_EARLY); if (rc) { xc_printf("ERROR: failed to initialize hypercall page: %d\n", rc); HYPERVISOR_shutdown(SHUTDOWN_crash); } start_info = (struct hvm_start_info *)(start_info_paddr + KERNBASE); if (start_info->magic != XEN_HVM_START_MAGIC_VALUE) { xc_printf("Unknown magic value in start_info struct: %#x\n", start_info->magic); HYPERVISOR_shutdown(SHUTDOWN_crash); } /* * The hvm_start_into structure is always appended after loading * the kernel and modules. */ physfree = roundup2(start_info_paddr + PAGE_SIZE, PAGE_SIZE); xatp.domid = DOMID_SELF; xatp.idx = 0; xatp.space = XENMAPSPACE_shared_info; xatp.gpfn = atop(physfree); if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp)) { xc_printf("ERROR: failed to setup shared_info page\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } HYPERVISOR_shared_info = (shared_info_t *)(physfree + KERNBASE); physfree += PAGE_SIZE; /* * 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); if (start_info->modlist_paddr != 0) { if (start_info->modlist_paddr >= physfree) { xc_printf( "ERROR: unexpected module list memory address\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } if (start_info->nr_modules == 0) { xc_printf( "ERROR: modlist_paddr != 0 but nr_modules == 0\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } mod = (struct hvm_modlist_entry *) (vm_paddr_t)start_info->modlist_paddr + KERNBASE; if (mod[0].paddr >= physfree) { xc_printf("ERROR: unexpected module memory address\n"); HYPERVISOR_shutdown(SHUTDOWN_crash); } } /* 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 -------------------------------*/ #ifdef SMP static bool start_xen_ap(int cpu) { struct vcpu_guest_context *ctxt; int ms, cpus = mp_naps; const size_t stacksize = kstack_pages * PAGE_SIZE; /* allocate and set up an idle stack data page */ bootstacks[cpu] = (void *)kmem_malloc(stacksize, M_WAITOK | M_ZERO); doublefault_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO); mce_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO); nmi_stack = (char *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO); dbg_stack = (void *)kmem_malloc(PAGE_SIZE, M_WAITOK | M_ZERO); dpcpu = (void *)kmem_malloc(DPCPU_SIZE, M_WAITOK | M_ZERO); bootSTK = (char *)bootstacks[cpu] + kstack_pages * PAGE_SIZE - 8; bootAP = cpu; ctxt = malloc(sizeof(*ctxt), M_TEMP, M_WAITOK | M_ZERO); ctxt->flags = VGCF_IN_KERNEL; ctxt->user_regs.rip = (unsigned long) init_secondary; ctxt->user_regs.rsp = (unsigned long) bootSTK; /* Set the AP to use the same page tables */ ctxt->ctrlreg[3] = KPML4phys; if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) panic("unable to initialize AP#%d", cpu); free(ctxt, M_TEMP); /* Launch the vCPU */ if (HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL)) panic("unable to start AP#%d", cpu); /* Wait up to 5 seconds for it to start. */ for (ms = 0; ms < 5000; ms++) { if (mp_naps > cpus) return (true); DELAY(1000); } return (false); } static int xen_pv_start_all_aps(void) { int cpu; mtx_init(&ap_boot_mtx, "ap boot", NULL, MTX_SPIN); for (cpu = 1; cpu < mp_ncpus; cpu++) { /* attempt to start the Application Processor */ if (!start_xen_ap(cpu)) panic("AP #%d failed to start!", cpu); CPU_SET(cpu, &all_cpus); /* record AP in CPU map */ } return (mp_naps); } #endif /* SMP */ /* * 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) 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; uint32_t size; int i, j; size = end; ehdr = (Elf_Ehdr *)(&end + 1); if (memcmp(ehdr->e_ident, ELFMAG, SELFMAG) || ehdr->e_ident[EI_CLASS] != ELF_TARG_CLASS || ehdr->e_version > 1) { 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) xc_printf( "Unable to load ELF symtab: could not find symtab or strtab\n"); } #endif static caddr_t xen_legacy_pvh_parse_preload_data(uint64_t modulep) { caddr_t kmdp; vm_ooffset_t off; vm_paddr_t metadata; char *envp; if (legacy_start_info->mod_start != 0) { preload_metadata = (caddr_t)legacy_start_info->mod_start; kmdp = preload_search_by_type("elf kernel"); if (kmdp == NULL) kmdp = preload_search_by_type("elf64 kernel"); KASSERT(kmdp != NULL, ("unable to find kernel")); /* * 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 with mod_start, * which contains the relocated modulep address. */ metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t); off = legacy_start_info->mod_start - 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, NULL); } else { /* Parse the extra boot information given by Xen */ boot_parse_cmdline_delim(legacy_start_info->cmd_line, ","); kmdp = NULL; } boothowto |= boot_env_to_howto(); #ifdef DDB xen_pvh_parse_symtab(); #endif return (kmdp); } 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]; if (start_info->modlist_paddr != 0) { struct hvm_modlist_entry *mod; + const char *cmdline; mod = (struct hvm_modlist_entry *) (start_info->modlist_paddr + KERNBASE); - 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"); - KASSERT(kmdp != NULL, ("unable to find kernel")); - - /* - * 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 with mod_start, - * which contains the relocated modulep address. - */ - metadata = MD_FETCH(kmdp, MODINFOMD_MODULEP, vm_paddr_t); - off = mod[0].paddr + KERNBASE - metadata; + 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); } 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), ","); kmdp = NULL; } 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 return (kmdp); } 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; /* 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); }