Merge tag 'v5.7-rc7' into efi/core, to refresh the branch and pick up fixes
authorIngo Molnar <mingo@kernel.org>
Mon, 25 May 2020 13:10:37 +0000 (15:10 +0200)
committerIngo Molnar <mingo@kernel.org>
Mon, 25 May 2020 13:10:37 +0000 (15:10 +0200)
Signed-off-by: Ingo Molnar <mingo@kernel.org>
1  2 
arch/arm/Kconfig
arch/arm64/Kconfig
drivers/firmware/efi/efi.c
drivers/firmware/efi/libstub/efi-stub.c
drivers/firmware/efi/libstub/efistub.h
drivers/firmware/efi/libstub/mem.c
drivers/firmware/efi/libstub/x86-stub.c
include/linux/efi.h

Simple merge
Simple merge
Simple merge
index ee225b323687e41d6c5efa74aa3fb8de5626855a,0000000000000000000000000000000000000000..60377e5ceab351196aa0e569a796c776262df185
mode 100644,000000..100644
--- /dev/null
@@@ -1,403 -1,0 +1,407 @@@
-               efi_setup_gop(si, &gop_proto, size);
 +// SPDX-License-Identifier: GPL-2.0-only
 +/*
 + * EFI stub implementation that is shared by arm and arm64 architectures.
 + * This should be #included by the EFI stub implementation files.
 + *
 + * Copyright (C) 2013,2014 Linaro Limited
 + *     Roy Franz <roy.franz@linaro.org
 + * Copyright (C) 2013 Red Hat, Inc.
 + *     Mark Salter <msalter@redhat.com>
 + */
 +
 +#include <linux/efi.h>
 +#include <linux/libfdt.h>
 +#include <asm/efi.h>
 +
 +#include "efistub.h"
 +
 +/*
 + * This is the base address at which to start allocating virtual memory ranges
 + * for UEFI Runtime Services. This is in the low TTBR0 range so that we can use
 + * any allocation we choose, and eliminate the risk of a conflict after kexec.
 + * The value chosen is the largest non-zero power of 2 suitable for this purpose
 + * both on 32-bit and 64-bit ARM CPUs, to maximize the likelihood that it can
 + * be mapped efficiently.
 + * Since 32-bit ARM could potentially execute with a 1G/3G user/kernel split,
 + * map everything below 1 GB. (512 MB is a reasonable upper bound for the
 + * entire footprint of the UEFI runtime services memory regions)
 + */
 +#define EFI_RT_VIRTUAL_BASE   SZ_512M
 +#define EFI_RT_VIRTUAL_SIZE   SZ_512M
 +
 +#ifdef CONFIG_ARM64
 +# define EFI_RT_VIRTUAL_LIMIT DEFAULT_MAP_WINDOW_64
 +#else
 +# define EFI_RT_VIRTUAL_LIMIT TASK_SIZE
 +#endif
 +
 +static u64 virtmap_base = EFI_RT_VIRTUAL_BASE;
 +static bool flat_va_mapping;
 +
 +const efi_system_table_t *efi_system_table;
 +
 +static struct screen_info *setup_graphics(void)
 +{
 +      efi_guid_t gop_proto = EFI_GRAPHICS_OUTPUT_PROTOCOL_GUID;
 +      efi_status_t status;
 +      unsigned long size;
 +      void **gop_handle = NULL;
 +      struct screen_info *si = NULL;
 +
 +      size = 0;
 +      status = efi_bs_call(locate_handle, EFI_LOCATE_BY_PROTOCOL,
 +                           &gop_proto, NULL, &size, gop_handle);
 +      if (status == EFI_BUFFER_TOO_SMALL) {
 +              si = alloc_screen_info();
 +              if (!si)
 +                      return NULL;
++              status = efi_setup_gop(si, &gop_proto, size);
++              if (status != EFI_SUCCESS) {
++                      free_screen_info(si);
++                      return NULL;
++              }
 +      }
 +      return si;
 +}
 +
 +static void install_memreserve_table(void)
 +{
 +      struct linux_efi_memreserve *rsv;
 +      efi_guid_t memreserve_table_guid = LINUX_EFI_MEMRESERVE_TABLE_GUID;
 +      efi_status_t status;
 +
 +      status = efi_bs_call(allocate_pool, EFI_LOADER_DATA, sizeof(*rsv),
 +                           (void **)&rsv);
 +      if (status != EFI_SUCCESS) {
 +              pr_efi_err("Failed to allocate memreserve entry!\n");
 +              return;
 +      }
 +
 +      rsv->next = 0;
 +      rsv->size = 0;
 +      atomic_set(&rsv->count, 0);
 +
 +      status = efi_bs_call(install_configuration_table,
 +                           &memreserve_table_guid, rsv);
 +      if (status != EFI_SUCCESS)
 +              pr_efi_err("Failed to install memreserve config table!\n");
 +}
 +
 +static unsigned long get_dram_base(void)
 +{
 +      efi_status_t status;
 +      unsigned long map_size, buff_size;
 +      unsigned long membase  = EFI_ERROR;
 +      struct efi_memory_map map;
 +      efi_memory_desc_t *md;
 +      struct efi_boot_memmap boot_map;
 +
 +      boot_map.map            = (efi_memory_desc_t **)&map.map;
 +      boot_map.map_size       = &map_size;
 +      boot_map.desc_size      = &map.desc_size;
 +      boot_map.desc_ver       = NULL;
 +      boot_map.key_ptr        = NULL;
 +      boot_map.buff_size      = &buff_size;
 +
 +      status = efi_get_memory_map(&boot_map);
 +      if (status != EFI_SUCCESS)
 +              return membase;
 +
 +      map.map_end = map.map + map_size;
 +
 +      for_each_efi_memory_desc_in_map(&map, md) {
 +              if (md->attribute & EFI_MEMORY_WB) {
 +                      if (membase > md->phys_addr)
 +                              membase = md->phys_addr;
 +              }
 +      }
 +
 +      efi_bs_call(free_pool, map.map);
 +
 +      return membase;
 +}
 +
 +/*
 + * This function handles the architcture specific differences between arm and
 + * arm64 regarding where the kernel image must be loaded and any memory that
 + * must be reserved. On failure it is required to free all
 + * all allocations it has made.
 + */
 +efi_status_t handle_kernel_image(unsigned long *image_addr,
 +                               unsigned long *image_size,
 +                               unsigned long *reserve_addr,
 +                               unsigned long *reserve_size,
 +                               unsigned long dram_base,
 +                               efi_loaded_image_t *image);
 +
 +asmlinkage void __noreturn efi_enter_kernel(unsigned long entrypoint,
 +                                          unsigned long fdt_addr,
 +                                          unsigned long fdt_size);
 +
 +/*
 + * EFI entry point for the arm/arm64 EFI stubs.  This is the entrypoint
 + * that is described in the PE/COFF header.  Most of the code is the same
 + * for both archictectures, with the arch-specific code provided in the
 + * handle_kernel_image() function.
 + */
 +efi_status_t efi_entry(efi_handle_t handle, efi_system_table_t *sys_table_arg)
 +{
 +      efi_loaded_image_t *image;
 +      efi_status_t status;
 +      unsigned long image_addr;
 +      unsigned long image_size = 0;
 +      unsigned long dram_base;
 +      /* addr/point and size pairs for memory management*/
 +      unsigned long initrd_addr = 0;
 +      unsigned long initrd_size = 0;
 +      unsigned long fdt_addr = 0;  /* Original DTB */
 +      unsigned long fdt_size = 0;
 +      char *cmdline_ptr = NULL;
 +      int cmdline_size = 0;
 +      efi_guid_t loaded_image_proto = LOADED_IMAGE_PROTOCOL_GUID;
 +      unsigned long reserve_addr = 0;
 +      unsigned long reserve_size = 0;
 +      enum efi_secureboot_mode secure_boot;
 +      struct screen_info *si;
 +      efi_properties_table_t *prop_tbl;
 +      unsigned long max_addr;
 +
 +      efi_system_table = sys_table_arg;
 +
 +      /* Check if we were booted by the EFI firmware */
 +      if (efi_system_table->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 +              status = EFI_INVALID_PARAMETER;
 +              goto fail;
 +      }
 +
 +      status = check_platform_features();
 +      if (status != EFI_SUCCESS)
 +              goto fail;
 +
 +      /*
 +       * Get a handle to the loaded image protocol.  This is used to get
 +       * information about the running image, such as size and the command
 +       * line.
 +       */
 +      status = efi_system_table->boottime->handle_protocol(handle,
 +                                      &loaded_image_proto, (void *)&image);
 +      if (status != EFI_SUCCESS) {
 +              pr_efi_err("Failed to get loaded image protocol\n");
 +              goto fail;
 +      }
 +
 +      dram_base = get_dram_base();
 +      if (dram_base == EFI_ERROR) {
 +              pr_efi_err("Failed to find DRAM base\n");
 +              status = EFI_LOAD_ERROR;
 +              goto fail;
 +      }
 +
 +      /*
 +       * Get the command line from EFI, using the LOADED_IMAGE
 +       * protocol. We are going to copy the command line into the
 +       * device tree, so this can be allocated anywhere.
 +       */
 +      cmdline_ptr = efi_convert_cmdline(image, &cmdline_size, ULONG_MAX);
 +      if (!cmdline_ptr) {
 +              pr_efi_err("getting command line via LOADED_IMAGE_PROTOCOL\n");
 +              status = EFI_OUT_OF_RESOURCES;
 +              goto fail;
 +      }
 +
 +      if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) ||
 +          IS_ENABLED(CONFIG_CMDLINE_FORCE) ||
 +          cmdline_size == 0)
 +              efi_parse_options(CONFIG_CMDLINE);
 +
 +      if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && cmdline_size > 0)
 +              efi_parse_options(cmdline_ptr);
 +
 +      pr_efi("Booting Linux Kernel...\n");
 +
 +      si = setup_graphics();
 +
 +      status = handle_kernel_image(&image_addr, &image_size,
 +                                   &reserve_addr,
 +                                   &reserve_size,
 +                                   dram_base, image);
 +      if (status != EFI_SUCCESS) {
 +              pr_efi_err("Failed to relocate kernel\n");
 +              goto fail_free_cmdline;
 +      }
 +
 +      efi_retrieve_tpm2_eventlog();
 +
 +      /* Ask the firmware to clear memory on unclean shutdown */
 +      efi_enable_reset_attack_mitigation();
 +
 +      secure_boot = efi_get_secureboot();
 +
 +      /*
 +       * Unauthenticated device tree data is a security hazard, so ignore
 +       * 'dtb=' unless UEFI Secure Boot is disabled.  We assume that secure
 +       * boot is enabled if we can't determine its state.
 +       */
 +      if (!IS_ENABLED(CONFIG_EFI_ARMSTUB_DTB_LOADER) ||
 +           secure_boot != efi_secureboot_mode_disabled) {
 +              if (strstr(cmdline_ptr, "dtb="))
 +                      pr_efi("Ignoring DTB from command line.\n");
 +      } else {
 +              status = efi_load_dtb(image, &fdt_addr, &fdt_size);
 +
 +              if (status != EFI_SUCCESS) {
 +                      pr_efi_err("Failed to load device tree!\n");
 +                      goto fail_free_image;
 +              }
 +      }
 +
 +      if (fdt_addr) {
 +              pr_efi("Using DTB from command line\n");
 +      } else {
 +              /* Look for a device tree configuration table entry. */
 +              fdt_addr = (uintptr_t)get_fdt(&fdt_size);
 +              if (fdt_addr)
 +                      pr_efi("Using DTB from configuration table\n");
 +      }
 +
 +      if (!fdt_addr)
 +              pr_efi("Generating empty DTB\n");
 +
 +      if (!efi_noinitrd) {
 +              max_addr = efi_get_max_initrd_addr(dram_base, image_addr);
 +              status = efi_load_initrd_dev_path(&initrd_addr, &initrd_size,
 +                                                max_addr);
 +              if (status == EFI_SUCCESS) {
 +                      pr_efi("Loaded initrd from LINUX_EFI_INITRD_MEDIA_GUID device path\n");
 +              } else if (status == EFI_NOT_FOUND) {
 +                      status = efi_load_initrd(image, &initrd_addr, &initrd_size,
 +                                               ULONG_MAX, max_addr);
 +                      if (status == EFI_SUCCESS && initrd_size > 0)
 +                              pr_efi("Loaded initrd from command line option\n");
 +              }
 +              if (status != EFI_SUCCESS)
 +                      pr_efi_err("Failed to load initrd!\n");
 +      }
 +
 +      efi_random_get_seed();
 +
 +      /*
 +       * If the NX PE data feature is enabled in the properties table, we
 +       * should take care not to create a virtual mapping that changes the
 +       * relative placement of runtime services code and data regions, as
 +       * they may belong to the same PE/COFF executable image in memory.
 +       * The easiest way to achieve that is to simply use a 1:1 mapping.
 +       */
 +      prop_tbl = get_efi_config_table(EFI_PROPERTIES_TABLE_GUID);
 +      flat_va_mapping = prop_tbl &&
 +                        (prop_tbl->memory_protection_attribute &
 +                         EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA);
 +
 +      /* hibernation expects the runtime regions to stay in the same place */
 +      if (!IS_ENABLED(CONFIG_HIBERNATION) && !efi_nokaslr && !flat_va_mapping) {
 +              /*
 +               * Randomize the base of the UEFI runtime services region.
 +               * Preserve the 2 MB alignment of the region by taking a
 +               * shift of 21 bit positions into account when scaling
 +               * the headroom value using a 32-bit random value.
 +               */
 +              static const u64 headroom = EFI_RT_VIRTUAL_LIMIT -
 +                                          EFI_RT_VIRTUAL_BASE -
 +                                          EFI_RT_VIRTUAL_SIZE;
 +              u32 rnd;
 +
 +              status = efi_get_random_bytes(sizeof(rnd), (u8 *)&rnd);
 +              if (status == EFI_SUCCESS) {
 +                      virtmap_base = EFI_RT_VIRTUAL_BASE +
 +                                     (((headroom >> 21) * rnd) >> (32 - 21));
 +              }
 +      }
 +
 +      install_memreserve_table();
 +
 +      status = allocate_new_fdt_and_exit_boot(handle, &fdt_addr,
 +                                              efi_get_max_fdt_addr(dram_base),
 +                                              initrd_addr, initrd_size,
 +                                              cmdline_ptr, fdt_addr, fdt_size);
 +      if (status != EFI_SUCCESS)
 +              goto fail_free_initrd;
 +
 +      efi_enter_kernel(image_addr, fdt_addr, fdt_totalsize((void *)fdt_addr));
 +      /* not reached */
 +
 +fail_free_initrd:
 +      pr_efi_err("Failed to update FDT and exit boot services\n");
 +
 +      efi_free(initrd_size, initrd_addr);
 +      efi_free(fdt_size, fdt_addr);
 +
 +fail_free_image:
 +      efi_free(image_size, image_addr);
 +      efi_free(reserve_size, reserve_addr);
 +fail_free_cmdline:
 +      free_screen_info(si);
 +      efi_free(cmdline_size, (unsigned long)cmdline_ptr);
 +fail:
 +      return status;
 +}
 +
 +/*
 + * efi_get_virtmap() - create a virtual mapping for the EFI memory map
 + *
 + * This function populates the virt_addr fields of all memory region descriptors
 + * in @memory_map whose EFI_MEMORY_RUNTIME attribute is set. Those descriptors
 + * are also copied to @runtime_map, and their total count is returned in @count.
 + */
 +void efi_get_virtmap(efi_memory_desc_t *memory_map, unsigned long map_size,
 +                   unsigned long desc_size, efi_memory_desc_t *runtime_map,
 +                   int *count)
 +{
 +      u64 efi_virt_base = virtmap_base;
 +      efi_memory_desc_t *in, *out = runtime_map;
 +      int l;
 +
 +      for (l = 0; l < map_size; l += desc_size) {
 +              u64 paddr, size;
 +
 +              in = (void *)memory_map + l;
 +              if (!(in->attribute & EFI_MEMORY_RUNTIME))
 +                      continue;
 +
 +              paddr = in->phys_addr;
 +              size = in->num_pages * EFI_PAGE_SIZE;
 +
 +              in->virt_addr = in->phys_addr;
 +              if (efi_novamap) {
 +                      continue;
 +              }
 +
 +              /*
 +               * Make the mapping compatible with 64k pages: this allows
 +               * a 4k page size kernel to kexec a 64k page size kernel and
 +               * vice versa.
 +               */
 +              if (!flat_va_mapping) {
 +
 +                      paddr = round_down(in->phys_addr, SZ_64K);
 +                      size += in->phys_addr - paddr;
 +
 +                      /*
 +                       * Avoid wasting memory on PTEs by choosing a virtual
 +                       * base that is compatible with section mappings if this
 +                       * region has the appropriate size and physical
 +                       * alignment. (Sections are 2 MB on 4k granule kernels)
 +                       */
 +                      if (IS_ALIGNED(in->phys_addr, SZ_2M) && size >= SZ_2M)
 +                              efi_virt_base = round_up(efi_virt_base, SZ_2M);
 +                      else
 +                              efi_virt_base = round_up(efi_virt_base, SZ_64K);
 +
 +                      in->virt_addr += efi_virt_base - paddr;
 +                      efi_virt_base += size;
 +              }
 +
 +              memcpy(out, in, desc_size);
 +              out = (void *)out + desc_size;
 +              ++*count;
 +      }
 +}
Simple merge
Simple merge