As UEFI becomes popular, a few architectures support to boot a PE format
kernel image directly. And the internal of PE format may be different.
This introduces a new kexec_file_ops implementation, named pe_image_ops,
which prepares UEFI environment for the trampoline code 'efi emulator'.
The pe_image_ops considers efi emulator and its input parameters,
'efi_emulator_param' as two additional kexec_segment. And it constructs
efi_memory_desc_t[], encodes efi runtime service info inside the parameter buffer.
Finally, it asks architecture implement's page table routine to set up
identity map for all memory used in 'efi emulator'
To do: This is a POC version, at present, it aims for arm64, later, it
needs abstraction to cope with x86
Signed-off-by: Pingfan Liu <piliu@xxxxxxxxxx>
Cc: Baoquan He <bhe@xxxxxxxxxx>
Cc: Dave Young <dyoung@xxxxxxxxxx>
Cc: Eric Biederman <ebiederm@xxxxxxxxxxxx>
Cc: Ard Biesheuvel <ardb@xxxxxxxxxx>
Cc: linux-kernel@xxxxxxxxxxxxxxx
To: kexec@xxxxxxxxxxxxxxxxxxx
---
include/linux/kexec.h | 1 +
kernel/Makefile | 1 +
kernel/kexec_pe_image.c | 503 ++++++++++++++++++++++++++++++++++++++++
3 files changed, 505 insertions(+)
create mode 100644 kernel/kexec_pe_image.c
diff --git a/include/linux/kexec.h b/include/linux/kexec.h
index cff6b6869498b..57b98bcaa5228 100644
--- a/include/linux/kexec.h
+++ b/include/linux/kexec.h
@@ -385,6 +385,7 @@ static inline int machine_kexec_post_load(struct kimage *image) { return 0; }
extern struct kimage *kexec_image;
extern struct kimage *kexec_crash_image;
+extern const struct kexec_file_ops pe_image_ops;
bool kexec_load_permitted(int kexec_image_type);
diff --git a/kernel/Makefile b/kernel/Makefile
index 3c13240dfc9f0..f14d78b03fd0f 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -74,6 +74,7 @@ obj-$(CONFIG_KEXEC_CORE) += kexec_core.o
obj-$(CONFIG_CRASH_DUMP) += crash_core.o
obj-$(CONFIG_KEXEC) += kexec.o
obj-$(CONFIG_KEXEC_FILE) += kexec_file.o
+obj-$(CONFIG_ARCH_SELECTS_KEXEC_PEIMAGE) += kexec_pe_image.o
obj-$(CONFIG_KEXEC_ELF) += kexec_elf.o
obj-$(CONFIG_BACKTRACE_SELF_TEST) += backtracetest.o
obj-$(CONFIG_COMPAT) += compat.o
diff --git a/kernel/kexec_pe_image.c b/kernel/kexec_pe_image.c
new file mode 100644
index 0000000000000..d14b8a5f69a99
--- /dev/null
+++ b/kernel/kexec_pe_image.c
@@ -0,0 +1,503 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Kexec image loader
+
+ * Copyright (C) 2024 Red Hat, Inc
+ * Author: Pingfan Liu <piliu@xxxxxxxxxx>
+ */
+
+#define pr_fmt(fmt) "kexec_file(Image): " fmt
+
+#include <linux/err.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/kexec.h>
+#include <linux/pe.h>
+#include <linux/efi.h>
+#include <linux/efi_emulator.h>
+#include <linux/string.h>
+#include <asm/byteorder.h>
+#include <asm/cpufeature.h>
+#include <asm/image.h>
+#include <asm/memory.h>
+
+/*
+ * The UEFI Terse Executable (TE) image has MZ header.
+*/
+static int pe_image_probe(const char *kernel_buf, unsigned long kernel_len)
+{
+ struct mz_hdr *mz = (struct mz_hdr *)kernel_buf;
+ struct pe_hdr *pe;
+
+ if (mz->magic != MZ_MAGIC)
+ return -1;
+ pe = (struct pe_hdr *)(kernel_buf + mz->peaddr);
+ if (pe->magic != PE_MAGIC)
+ return -1;
+
+ return 0;
+}
+
+/*
+ * Efi runtime code or data are marked as EFI_RUNTIME_SERVICES_CODE
+ * or EFI_RUNTIME_SERVICES_DATA memory descriptor. When building mapping for the
+ * emulator, they should be mapped. Here just recording the entries to those regions.
+ */
+static void build_rt_info(struct efi_rt_info *rt)
+{
+ unsigned int desc_size = efi.memmap.desc_size;
+ int i, cnt;
+
+ i = cnt = 0;
+ if (desc_size) {
+ cnt = (efi.memmap.map_end - efi.memmap.map) / desc_size;
+ /* The emulator heap memory split the continous memory into three chunk */
+ cnt += 2;
+ }
+
+ /*
+ * This virtual address is in UEFI address space, and the mapping is recorded
+ * in efi_memory_desc_t[] and passed to reboot kernel.
+ *
+ * EFI stub may use runtime service, so its mapping should be set up
+ */
+ rt->runtime = efi.runtime;
+ rt->runtime_version = efi.runtime_version;
+ rt->runtime_supported_mask = efi.runtime_supported_mask;
+ /*
+ * Reconstruct the persistent systab's tables, which are recorded in efi.
+ */
+ if (efi.acpi != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.acpi;
+ rt->systab_tables[i].guid = ACPI_TABLE_GUID;
+ i++;
+ }
+ if (efi.acpi20 != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.acpi20;
+ rt->systab_tables[i].guid = ACPI_20_TABLE_GUID;
+ i++;
+ }
+ if (efi.smbios != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.smbios;
+ rt->systab_tables[i].guid = SMBIOS_TABLE_GUID;
+ i++;
+ }
+ if (efi.smbios3 != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.smbios3;
+ rt->systab_tables[i].guid = SMBIOS3_TABLE_GUID;
+ i++;
+ }
+ if (efi.esrt != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.esrt;
+ rt->systab_tables[i].guid = EFI_SYSTEM_RESOURCE_TABLE_GUID;
+ i++;
+ }
+ if (efi.tpm_log != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.tpm_log;
+ rt->systab_tables[i].guid = LINUX_EFI_TPM_EVENT_LOG_GUID;
+ i++;
+ }
+ if (efi.tpm_final_log != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.tpm_final_log;
+ rt->systab_tables[i].guid = EFI_TCG2_FINAL_EVENTS_TABLE_GUID;
+ i++;
+ }
+#ifdef CONFIG_LOAD_UEFI_KEYS
+ if (efi.mokvar_table != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.mokvar_table;
+ rt->systab_tables[i].guid = LINUX_EFI_MOK_VARIABLE_TABLE_GUID;
+ i++;
+ }
+#endif
+#ifdef CONFIG_EFI_COCO_SECRET
+ if (efi.coco_secret != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.coco_secret;
+ rt->systab_tables[i].guid = LINUX_EFI_COCO_SECRET_AREA_GUID;
+ i++;
+ }
+#endif
+#ifdef CONFIG_UNACCEPTED_MEMORY
+ if (efi.unaccepted != EFI_INVALID_TABLE_ADDR) {
+ rt->systab_tables[i].table = (void *)efi.unaccepted;
+ rt->systab_tables[i].guid = LINUX_EFI_UNACCEPTED_MEM_TABLE_GUID;
+ i++;
+ }
+#endif
+ if (!desc_size) {
+ //todo, if noefi boot, LINUX_EFI_MEMRESERVE_TABLE_GUID should be used to pass IMA
+ }
+
+ rt->systab_nr_tables = i;
+
+ rt->memmap.map_size = desc_size * cnt;
+ /* In case of non-EFI booting, these two fields will be faked later */
+ rt->memmap.desc_size = efi.memmap.desc_size;
+ rt->memmap.desc_ver = efi.memmap.desc_version;
+}
+
+static int create_md_from_res(struct resource *res, void *data)
+{
+ struct efi_emulator_param *param = (struct efi_emulator_param *)data;
+ struct efi_boot_memmap *memmap = ¶m->rt_info.memmap;
+ efi_memory_desc_t *dst_md;
+
+ dst_md = (void *)memmap->map + memmap->map_size;
+
+ /* Split res into three chunk */
+ if ((res->start <= param->mempool_start) &&
+ res->end > (param->mempool_start + param->mempool_sz)) {
+
+ dst_md->phys_addr = res->start;
+ dst_md->num_pages = (param->mempool_start - res->start) >> EFI_PAGE_SHIFT;
+ /* Pretend that it is occupied */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ dst_md->attribute = EFI_MEMORY_WB;
+ dst_md = (void *)dst_md + memmap->desc_size;
+
+ dst_md->phys_addr = param->mempool_start;
+ dst_md->num_pages = param->mempool_sz >> EFI_PAGE_SHIFT;
+ /* Confine memory footprint inside this region before exit boot service */
+ dst_md->type = EFI_CONVENTIONAL_MEMORY;
+ dst_md->attribute = EFI_MEMORY_WB;
+ dst_md = (void *)dst_md + memmap->desc_size;
+
+ dst_md->phys_addr = param->mempool_start + param->mempool_sz;
+ dst_md->num_pages = (res->end - dst_md->phys_addr) >> EFI_PAGE_SHIFT;
+ /* Pretend that it is occupied */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ dst_md->attribute = EFI_MEMORY_WB;
+
+ memmap->map_size += 3 * (memmap->desc_size);
+ } else {
+ dst_md->phys_addr = res->start;
+ dst_md->num_pages = (res->end - res->start) >> EFI_PAGE_SHIFT;
+ /* Pretend that it is occupied */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ dst_md->attribute = EFI_MEMORY_WB;
+
+ memmap->map_size += memmap->desc_size;
+ }
+
+ return 0;
+}
+
+static efi_memory_desc_t *conclude_md(efi_memory_desc_t *dst_md,
+ unsigned int desc_size, unsigned long pool_start, unsigned long num_pages)
+{
+ if (dst_md->num_pages == 0)
+ return dst_md;
+
+ /* Split into three sections */
+ if (dst_md->phys_addr <= pool_start && dst_md->num_pages >= num_pages ) {
+ u64 virt_base, phys_base, left_pages, attribute;
+ u32 type;
+
+ type = dst_md->type;
+ attribute = dst_md->attribute;
+ phys_base = dst_md->phys_addr;
+ /*
+ * After SetVirtualAddressMap, the mapping is installed into
+ * firmware and can not be changed. The second kernel should
+ * be aware of this info
+ */
+ virt_base = dst_md->virt_addr;
+ left_pages = dst_md->num_pages;
+
+ dst_md->num_pages = (pool_start - phys_base) >> EFI_PAGE_SHIFT;
+ /* Pretend that it is occupied until exit boot service */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ left_pages -= dst_md->num_pages;
+
+ dst_md = (void *)dst_md + desc_size;
+ dst_md->phys_addr = pool_start;
+ dst_md->virt_addr = virt_base + (dst_md->phys_addr - phys_base);
+ dst_md->num_pages = num_pages;
+ dst_md->attribute = attribute;
+ /* Confine the memory footprint on it, which has page table mapping */
+ dst_md->type = EFI_CONVENTIONAL_MEMORY;
+ left_pages -= dst_md->num_pages;
+
+ dst_md = (void *)dst_md + desc_size;
+ dst_md->phys_addr = pool_start + (num_pages << EFI_PAGE_SHIFT);
+ dst_md->virt_addr = virt_base + (dst_md->phys_addr - phys_base);
+ dst_md->num_pages = left_pages;
+ dst_md->attribute = attribute;
+ /* Pretend that it is occupied until exit boot service */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ }
+
+ dst_md = (void *)dst_md + desc_size;
+
+ return dst_md;
+}
+
+static void create_md_from_efi(efi_memory_desc_t *dst, unsigned int desc_sz,
+ struct efi_emulator_param *param)
+{
+ efi_memory_desc_t *md, *dst_md, *prev_md = NULL;
+ unsigned long pool_start, num_pages;
+
+ dst_md = dst;
+ pool_start = param->mempool_start;
+ num_pages = param->mempool_sz >> EFI_PAGE_SHIFT;
+ for_each_efi_memory_desc(md) {
+ switch (md->type) {
+ // todo, how to ensure kexec dst avoid the EFI_RUNTIME_SERVICES_CODE etc
+ case EFI_RUNTIME_SERVICES_CODE:
+ case EFI_RUNTIME_SERVICES_DATA:
+ case EFI_RESERVED_TYPE:
+ case EFI_UNUSABLE_MEMORY:
+ case EFI_ACPI_RECLAIM_MEMORY:
+ //test whether the current dst_md covers the heap, if yes, split it into three
+ dst_md = conclude_md(dst_md, desc_sz, pool_start, num_pages);
+ dst_md->phys_addr = md->phys_addr;
+ dst_md->virt_addr = md->virt_addr;
+ dst_md->num_pages = md->num_pages;
+ dst_md->type = md->type;
+ dst_md->attribute = md->attribute;
+ dst_md = conclude_md(dst_md, desc_sz, pool_start, num_pages);
+ break;
+
+ default:
+ if (dst_md->num_pages == 0) {
+ dst_md->phys_addr = md->phys_addr;
+ dst_md->virt_addr = md->virt_addr;
+ dst_md->num_pages = md->num_pages;
+ /*
+ * Pretend as boot service data to prevent allocation
+ * from it before efi_exit_boot_services.
+ */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ dst_md->attribute = md->attribute;
+ } else {
+ /* merge */
+ if (prev_md && prev_md->attribute == md->attribute &&
+ (md->phys_addr - prev_md->phys_addr) >> EFI_PAGE_SHIFT == prev_md->num_pages) {
+ dst_md->num_pages += md->num_pages;
+ } else {
+ dst_md = conclude_md(dst_md, desc_sz, pool_start, num_pages);
+ dst_md->phys_addr = md->phys_addr;
+ dst_md->virt_addr = md->virt_addr;
+ dst_md->num_pages = md->num_pages;
+ /* Pretend as boot service data */
+ dst_md->type = EFI_BOOT_SERVICES_DATA;
+ dst_md->attribute = md->attribute;
+ }
+ }
+ break;
+ }
+
+ prev_md = md;
+ }
+ dst_md = conclude_md(dst_md, desc_sz, pool_start, num_pages);
+ param->rt_info.memmap.map_size = (void *)dst_md - (void *)dst;
+}
+
+/*
+ * All efi runtime information should be passed to kexec reboot kernel. Record
+ * them in scratch. These information should be in EFI_RUNTIME_SERVICES md, so
+ * copying index is enough. Later the mapping for that md will be set up in
+ * arch_emulator_prepare_pgtable().
+ *
+ * If the system is not booted by efi, fake one.
+ */
+static void encode_efi_runtime_info(struct efi_emulator_param *param, struct kimage *image)
+{
+ struct efi_rt_info *rt = ¶m->rt_info;
+ efi_memory_desc_t *dst_md;
+ unsigned int sz;
+
+ build_rt_info(rt);
+ sz = efi.memmap.map_end - efi.memmap.map;
+ /* booted by efi firmware */
+ if (sz) {
+ unsigned long desc_size = efi.memmap.desc_size;
+
+ dst_md = rt->memmap.map;
+ /*
+ * It has no point to pass efi.memmap directly to the reboot kernel since
+ * EFI_BOOT_SERVICES_DATA etc has changed. But EFI_RUNTIME_SERVICES_DATA
+ * etc should be paid attention to.
+ */
+ create_md_from_efi(dst_md, desc_size, param);
+ } else {
+ param->noefi_boot = true;
+ /*
+ * Kernel is booted by non-EFI loader, which parses the PE format. In
+ * kexec case, faking memory descriptor so that efi-stub can self-parse.
+ * But there is no efi runtime service, and the kernel cmdline should
+ * have 'noefi'
+ */
+ rt->memmap.desc_size = sizeof(efi_memory_desc_t);
+ rt->memmap.desc_ver = EFI_MEMORY_DESCRIPTOR_VERSION;
+ walk_system_ram_res(0, ULONG_MAX, param, create_md_from_res);
+
+ /*
+ * Besides, efi stub removes two dt node '/memreserve', one for initrd
+ * and the other for IMA. EFI_RESERVED_TYPE can not serve that purpose.
+ * That should be handled by LINUX_EFI_MEMRESERVE_TABLE_GUID
+ */
+ }
+}
+
+extern phys_addr_t arch_emulator_prepare_pgtable(struct kimage *kimage,
+ struct efi_emulator_param *param);
+
+static void detect_earlycon(struct efi_emulator_param *param)
+{
+ //todo, autodetect the console register base and size
+ // aarch64 MACHINE_VIRT_UART_BASE
+ size_t sz = 15 < strlen("amba-pl011") ? 15 : strlen("amba-pl011");
+
+ memcpy(param->earlycon_name, "amba-pl011", sz);
+ param->earlycon_reg_base = 0x9000000;
+ param->earlycon_reg_sz = PAGE_SIZE;
+ param->print_enabled = true;
+}
+
+static phys_addr_t emulator_prepare_pgtable(struct kimage *kimage,
+ struct efi_emulator_param *param)
+{
+
+ detect_earlycon(param);
+
+ return arch_emulator_prepare_pgtable(kimage, param);
+}
+
+/* param, stack, and basic heap */
+#define EMULATOR_STACK_SIZE (1 << 17)
+#define EMULATOR_PARAM_SIZE (1 << 20)
+
+static void *pe_image_load(struct kimage *image,
+ char *kernel, unsigned long kernel_len,
+ char *initrd, unsigned long initrd_len,
+ char *cmdline, unsigned long cmdline_len)
+{
+ struct kexec_segment *emulator_seg, *param_seg, *kernel_segment;
+ struct efi_emulator_param *param;
+ struct page *param_pages;
+ struct kexec_buf kbuf;
+ unsigned long kseg_num;
+ int ret;
+
+ image->is_pe = true;
+ /* Do not parse image format, just load it. */
+ kbuf.image = image;
+ kbuf.buf_min = 0;
+ kbuf.buf_max = ULONG_MAX;
+ kbuf.top_down = false;
+ kbuf.buffer = kernel;
+ kbuf.bufsz = kernel_len;
+ kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
+ kbuf.memsz = kernel_len;
+ kbuf.buf_align = MIN_KIMG_ALIGN;
+
+ kseg_num = image->nr_segments;
+ /*
+ * The location of the kernel segment may make it impossible to satisfy
+ * the other segment requirements, so we try repeatedly to find a
+ * location that will work.
+ */
+ while ((ret = kexec_add_buffer(&kbuf)) == 0) {
+ /* Try to load additional data */
+ kernel_segment = &image->segment[kseg_num];
+ ret = load_other_segments(image, kernel_segment->mem,
+ kernel_segment->memsz, initrd,
+ initrd_len, cmdline);
+ if (!ret)
+ break;
+
+ /*
+ * We couldn't find space for the other segments; erase the
+ * kernel segment and try the next available hole.
+ */
+ image->nr_segments -= 1;
+ kbuf.buf_min = kernel_segment->mem + kernel_segment->memsz;
+ kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
+ }
+
+ if (ret) {
+ pr_err("Could not find any suitable kernel location!");
+ return ERR_PTR(ret);
+ }
+ kernel_segment = &image->segment[kseg_num];
+
+ /* Load EFI emulator */
+ emulator_seg = &image->segment[image->nr_segments];
+ kbuf.buffer = _efi_emulator_start;
+ kbuf.bufsz = _efi_emulator_end - _efi_emulator_start;
+ kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
+ kbuf.memsz = kbuf.bufsz;
+ kbuf.buf_align = PAGE_SIZE;
+ ret = kexec_add_buffer(&kbuf);
+ if (ret)
+ return ERR_PTR(ret);
+
+ /*
+ * Prepare param and memory for emulator. One page for param,
+ * rear page for stack and the rest for runtime heap.
+ */
+ param_seg = &image->segment[image->nr_segments];
+ //to do, zero page is not required in kimage_entry_t
+ // and free them
+ param_pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, 2);
+ param = page_to_virt(param_pages);
+
+ /* These chunk of information will be copied to the KEXEC SOURCE PAGE */
+ /* emulator loaded address */
+ param->load_address = emulator_seg->mem;
+ /* PE file payload at the beginning of this RAM range */
+ param->kernel_img_start = kernel_segment->mem;
+ param->kernel_img_sz = kernel_segment->memsz;
+ /* uefi protocol need it */
+ param->dtb = image->arch.dtb_mem;
+ param->sz_in_byte = utf8s_to_utf16s(cmdline, cmdline_len,
+ UTF16_LITTLE_ENDIAN, param->cmdline, 512);
+ param->sz_in_byte *= sizeof(wchar_t);
+ kbuf.buffer = param;
+ /*
+ * One page for param, one page for stack, the rest for heap, which should
+ * also has enough room for kernel and its decompression (256MB)
+ */
+ kbuf.bufsz = EMULATOR_PARAM_SIZE + kernel_segment->memsz + (1 << 28);
+ kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
+ kbuf.memsz = kbuf.bufsz;
+ kbuf.buf_align = PAGE_SIZE;
+ ret = kexec_add_buffer(&kbuf);
+ if (ret)
+ return ERR_PTR(ret);
+
+ /* These info is formed after param segment is built */
+ /* stack size equals 64K -sizeof(*param) */
+ param->sp = param_seg->mem + EMULATOR_STACK_SIZE;
+ /* heap information */
+ param->mempool_start = param->sp;
+ param->mempool_sz = kbuf.bufsz - EMULATOR_STACK_SIZE;
+
+ param->pgd_root = emulator_prepare_pgtable(image, param);
+ /* For the time being, reset routine will turn off mmu */
+ param->mmu_on = false;
+ encode_efi_runtime_info(param, image);
+
+ /* relocate.s jumps to it */
+ image->start = emulator_seg->mem;
+ /* the 3rd input param for restart() */
+ image->arch.param_mem = param_seg->mem;
+ image->arch.dtb_mem = image->arch.param_mem;
+
+ kexec_dprintk("Loaded emulator at 0x%lx bufsz=0x%lx\n",
+ emulator_seg->mem, emulator_seg->memsz);
+ kexec_dprintk("Loaded param blob at 0x%lx bufsz=0x%lx, sp=0x%lx\n",
+ param_seg->mem, param_seg->memsz, param->sp);
+ kexec_dprintk("pgd_root:0x%llx\n", param->pgd_root);
+
+ return NULL;
+}
+
+const struct kexec_file_ops pe_image_ops = {
+ .probe = pe_image_probe,
+ .load = pe_image_load,
+#ifdef CONFIG_KEXEC_IMAGE_VERIFY_SIG
+ .verify_sig = kexec_kernel_verify_pe_sig,
+#endif
+};
--
2.41.0
