arch/x86/platform/efi/quirks.c

   1 #define pr_fmt(fmt) "efi: " fmt
   2
   3 #include <linux/init.h>
   4 #include <linux/kernel.h>
   5 #include <linux/string.h>
   6 #include <linux/time.h>
   7 #include <linux/types.h>
   8 #include <linux/efi.h>
   9 #include <linux/slab.h>
  10 #include <linux/memblock.h>
  11 #include <linux/bootmem.h>
  12 #include <linux/acpi.h>
  13 #include <linux/dmi.h>
  14 #include <asm/efi.h>
  15 #include <asm/uv/uv.h>
  16
  17 #define EFI_MIN_RESERVE 5120
  18
  19 #define EFI_DUMMY_GUID \
  20         EFI_GUID(0x4424ac57, 0xbe4b, 0x47dd, 0x9e, 0x97, 0xed, 0x50, 0xf0, 0x9f, 0x92, 0xa9)
  21
  22 static efi_char16_t efi_dummy_name[6] = { 'D', 'U', 'M', 'M', 'Y', 0 };
  23
  24 static bool efi_no_storage_paranoia;
  25
  26 /*
  27  * Some firmware implementations refuse to boot if there's insufficient
  28  * space in the variable store. The implementation of garbage collection
  29  * in some FW versions causes stale (deleted) variables to take up space
  30  * longer than intended and space is only freed once the store becomes
  31  * almost completely full.
  32  *
  33  * Enabling this option disables the space checks in
  34  * efi_query_variable_store() and forces garbage collection.
  35  *
  36  * Only enable this option if deleting EFI variables does not free up
  37  * space in your variable store, e.g. if despite deleting variables
  38  * you're unable to create new ones.
  39  */
  40 static int __init setup_storage_paranoia(char *arg)
  41 {
  42         efi_no_storage_paranoia = true;
  43         return 0;
  44 }
  45 early_param("efi_no_storage_paranoia", setup_storage_paranoia);
  46
  47 /*
  48  * Deleting the dummy variable which kicks off garbage collection
  49 */
  50 void efi_delete_dummy_variable(void)
  51 {
  52         efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
  53                          EFI_VARIABLE_NON_VOLATILE |
  54                          EFI_VARIABLE_BOOTSERVICE_ACCESS |
  55                          EFI_VARIABLE_RUNTIME_ACCESS,
  56                          0, NULL);
  57 }
  58
  59 /*
  60  * In the nonblocking case we do not attempt to perform garbage
  61  * collection if we do not have enough free space. Rather, we do the
  62  * bare minimum check and give up immediately if the available space
  63  * is below EFI_MIN_RESERVE.
  64  *
  65  * This function is intended to be small and simple because it is
  66  * invoked from crash handler paths.
  67  */
  68 static efi_status_t
  69 query_variable_store_nonblocking(u32 attributes, unsigned long size)
  70 {
  71         efi_status_t status;
  72         u64 storage_size, remaining_size, max_size;
  73
  74         status = efi.query_variable_info_nonblocking(attributes, &storage_size,
  75                                                      &remaining_size,
  76                                                      &max_size);
  77         if (status != EFI_SUCCESS)
  78                 return status;
  79
  80         if (remaining_size - size < EFI_MIN_RESERVE)
  81                 return EFI_OUT_OF_RESOURCES;
  82
  83         return EFI_SUCCESS;
  84 }
  85
  86 /*
  87  * Some firmware implementations refuse to boot if there's insufficient space
  88  * in the variable store. Ensure that we never use more than a safe limit.
  89  *
  90  * Return EFI_SUCCESS if it is safe to write 'size' bytes to the variable
  91  * store.
  92  */
  93 efi_status_t efi_query_variable_store(u32 attributes, unsigned long size,
  94                                       bool nonblocking)
  95 {
  96         efi_status_t status;
  97         u64 storage_size, remaining_size, max_size;
  98
  99         if (!(attributes & EFI_VARIABLE_NON_VOLATILE))
 100                 return 0;
 101
 102         if (nonblocking)
 103                 return query_variable_store_nonblocking(attributes, size);
 104
 105         status = efi.query_variable_info(attributes, &storage_size,
 106                                          &remaining_size, &max_size);
 107         if (status != EFI_SUCCESS)
 108                 return status;
 109
 110         /*
 111          * We account for that by refusing the write if permitting it would
 112          * reduce the available space to under 5KB. This figure was provided by
 113          * Samsung, so should be safe.
 114          */
 115         if ((remaining_size - size < EFI_MIN_RESERVE) &&
 116                 !efi_no_storage_paranoia) {
 117
 118                 /*
 119                  * Triggering garbage collection may require that the firmware
 120                  * generate a real EFI_OUT_OF_RESOURCES error. We can force
 121                  * that by attempting to use more space than is available.
 122                  */
 123                 unsigned long dummy_size = remaining_size + 1024;
 124                 void *dummy = kzalloc(dummy_size, GFP_ATOMIC);
 125
 126                 if (!dummy)
 127                         return EFI_OUT_OF_RESOURCES;
 128
 129                 status = efi.set_variable(efi_dummy_name, &EFI_DUMMY_GUID,
 130                                           EFI_VARIABLE_NON_VOLATILE |
 131                                           EFI_VARIABLE_BOOTSERVICE_ACCESS |
 132                                           EFI_VARIABLE_RUNTIME_ACCESS,
 133                                           dummy_size, dummy);
 134
 135                 if (status == EFI_SUCCESS) {
 136                         /*
 137                          * This should have failed, so if it didn't make sure
 138                          * that we delete it...
 139                          */
 140                         efi_delete_dummy_variable();
 141                 }
 142
 143                 kfree(dummy);
 144
 145                 /*
 146                  * The runtime code may now have triggered a garbage collection
 147                  * run, so check the variable info again
 148                  */
 149                 status = efi.query_variable_info(attributes, &storage_size,
 150                                                  &remaining_size, &max_size);
 151
 152                 if (status != EFI_SUCCESS)
 153                         return status;
 154
 155                 /*
 156                  * There still isn't enough room, so return an error
 157                  */
 158                 if (remaining_size - size < EFI_MIN_RESERVE)
 159                         return EFI_OUT_OF_RESOURCES;
 160         }
 161
 162         return EFI_SUCCESS;
 163 }
 164 EXPORT_SYMBOL_GPL(efi_query_variable_store);
 165
 166 /*
 167  * Helper function for efi_reserve_boot_services() to figure out if we
 168  * can free regions in efi_free_boot_services().
 169  *
 170  * Use this function to ensure we do not free regions owned by somebody
 171  * else. We must only reserve (and then free) regions:
 172  *
 173  * - Not within any part of the kernel
 174  * - Not the BIOS reserved area (E820_RESERVED, E820_NVS, etc)
 175  */
 176 static bool can_free_region(u64 start, u64 size)
 177 {
 178         if (start + size > __pa_symbol(_text) && start <= __pa_symbol(_end))
 179                 return false;
 180
 181         if (!e820_all_mapped(start, start+size, E820_RAM))
 182                 return false;
 183
 184         return true;
 185 }
 186
 187 /*
 188  * The UEFI specification makes it clear that the operating system is free to do
 189  * whatever it wants with boot services code after ExitBootServices() has been
 190  * called. Ignoring this recommendation a significant bunch of EFI implementations
 191  * continue calling into boot services code (SetVirtualAddressMap). In order to
 192  * work around such buggy implementations we reserve boot services region during
 193  * EFI init and make sure it stays executable. Then, after SetVirtualAddressMap(), it
 194 * is discarded.
 195 */
 196 void __init efi_reserve_boot_services(void)
 197 {
 198         void *p;
 199
 200         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 201                 efi_memory_desc_t *md = p;
 202                 u64 start = md->phys_addr;
 203                 u64 size = md->num_pages << EFI_PAGE_SHIFT;
 204                 bool already_reserved;
 205
 206                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 207                     md->type != EFI_BOOT_SERVICES_DATA)
 208                         continue;
 209
 210                 already_reserved = memblock_is_region_reserved(start, size);
 211
 212                 /*
 213                  * Because the following memblock_reserve() is paired
 214                  * with free_bootmem_late() for this region in
 215                  * efi_free_boot_services(), we must be extremely
 216                  * careful not to reserve, and subsequently free,
 217                  * critical regions of memory (like the kernel image) or
 218                  * those regions that somebody else has already
 219                  * reserved.
 220                  *
 221                  * A good example of a critical region that must not be
 222                  * freed is page zero (first 4Kb of memory), which may
 223                  * contain boot services code/data but is marked
 224                  * E820_RESERVED by trim_bios_range().
 225                  */
 226                 if (!already_reserved) {
 227                         memblock_reserve(start, size);
 228
 229                         /*
 230                          * If we are the first to reserve the region, no
 231                          * one else cares about it. We own it and can
 232                          * free it later.
 233                          */
 234                         if (can_free_region(start, size))
 235                                 continue;
 236                 }
 237
 238                 /*
 239                  * We don't own the region. We must not free it.
 240                  *
 241                  * Setting this bit for a boot services region really
 242                  * doesn't make sense as far as the firmware is
 243                  * concerned, but it does provide us with a way to tag
 244                  * those regions that must not be paired with
 245                  * free_bootmem_late().
 246                  */
 247                 md->attribute |= EFI_MEMORY_RUNTIME;
 248         }
 249 }
 250
 251 void __init efi_free_boot_services(void)
 252 {
 253         void *p;
 254
 255         for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 256                 efi_memory_desc_t *md = p;
 257                 unsigned long long start = md->phys_addr;
 258                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 259
 260                 if (md->type != EFI_BOOT_SERVICES_CODE &&
 261                     md->type != EFI_BOOT_SERVICES_DATA)
 262                         continue;
 263
 264                 /* Do not free, someone else owns it: */
 265                 if (md->attribute & EFI_MEMORY_RUNTIME)
 266                         continue;
 267
 268                 free_bootmem_late(start, size);
 269         }
 270
 271         efi_unmap_memmap();
 272 }
 273
 274 /*
 275  * A number of config table entries get remapped to virtual addresses
 276  * after entering EFI virtual mode. However, the kexec kernel requires
 277  * their physical addresses therefore we pass them via setup_data and
 278  * correct those entries to their respective physical addresses here.
 279  *
 280  * Currently only handles smbios which is necessary for some firmware
 281  * implementation.
 282  */
 283 int __init efi_reuse_config(u64 tables, int nr_tables)
 284 {
 285         int i, sz, ret = 0;
 286         void *p, *tablep;
 287         struct efi_setup_data *data;
 288
 289         if (!efi_setup)
 290                 return 0;
 291
 292         if (!efi_enabled(EFI_64BIT))
 293                 return 0;
 294
 295         data = early_memremap(efi_setup, sizeof(*data));
 296         if (!data) {
 297                 ret = -ENOMEM;
 298                 goto out;
 299         }
 300
 301         if (!data->smbios)
 302                 goto out_memremap;
 303
 304         sz = sizeof(efi_config_table_64_t);
 305
 306         p = tablep = early_memremap(tables, nr_tables * sz);
 307         if (!p) {
 308                 pr_err("Could not map Configuration table!\n");
 309                 ret = -ENOMEM;
 310                 goto out_memremap;
 311         }
 312
 313         for (i = 0; i < efi.systab->nr_tables; i++) {
 314                 efi_guid_t guid;
 315
 316                 guid = ((efi_config_table_64_t *)p)->guid;
 317
 318                 if (!efi_guidcmp(guid, SMBIOS_TABLE_GUID))
 319                         ((efi_config_table_64_t *)p)->table = data->smbios;
 320                 p += sz;
 321         }
 322         early_memunmap(tablep, nr_tables * sz);
 323
 324 out_memremap:
 325         early_memunmap(data, sizeof(*data));
 326 out:
 327         return ret;
 328 }
 329
 330 static const struct dmi_system_id sgi_uv1_dmi[] = {
 331         { NULL, "SGI UV1",
 332                 {       DMI_MATCH(DMI_PRODUCT_NAME,     "Stoutland Platform"),
 333                         DMI_MATCH(DMI_PRODUCT_VERSION,  "1.0"),
 334                         DMI_MATCH(DMI_BIOS_VENDOR,      "SGI.COM"),
 335                 }
 336         },
 337         { } /* NULL entry stops DMI scanning */
 338 };
 339
 340 void __init efi_apply_memmap_quirks(void)
 341 {
 342         /*
 343          * Once setup is done earlier, unmap the EFI memory map on mismatched
 344          * firmware/kernel architectures since there is no support for runtime
 345          * services.
 346          */
 347         if (!efi_runtime_supported()) {
 348                 pr_info("Setup done, disabling due to 32/64-bit mismatch\n");
 349                 efi_unmap_memmap();
 350         }
 351
 352         /* UV2+ BIOS has a fix for this issue.  UV1 still needs the quirk. */
 353         if (dmi_check_system(sgi_uv1_dmi))
 354                 set_bit(EFI_OLD_MEMMAP, &efi.flags);
 355 }
 356
 357 /*
 358  * For most modern platforms the preferred method of powering off is via
 359  * ACPI. However, there are some that are known to require the use of
 360  * EFI runtime services and for which ACPI does not work at all.
 361  *
 362  * Using EFI is a last resort, to be used only if no other option
 363  * exists.
 364  */
 365 bool efi_reboot_required(void)
 366 {
 367         if (!acpi_gbl_reduced_hardware)
 368                 return false;
 369
 370         efi_reboot_quirk_mode = EFI_RESET_WARM;
 371         return true;
 372 }
 373
 374 bool efi_poweroff_required(void)
 375 {
 376         return !!acpi_gbl_reduced_hardware;
 377 }