Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Common EFI (Extensible Firmware Interface) support functions
4 * Based on Extensible Firmware Interface Specification version 1.0
5 *
6 * Copyright (C) 1999 VA Linux Systems
7 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
8 * Copyright (C) 1999-2002 Hewlett-Packard Co.
9 * David Mosberger-Tang <davidm@hpl.hp.com>
10 * Stephane Eranian <eranian@hpl.hp.com>
11 * Copyright (C) 2005-2008 Intel Co.
12 * Fenghua Yu <fenghua.yu@intel.com>
13 * Bibo Mao <bibo.mao@intel.com>
14 * Chandramouli Narayanan <mouli@linux.intel.com>
15 * Huang Ying <ying.huang@intel.com>
16 * Copyright (C) 2013 SuSE Labs
17 * Borislav Petkov <bp@suse.de> - runtime services VA mapping
18 *
19 * Copied from efi_32.c to eliminate the duplicated code between EFI
20 * 32/64 support code. --ying 2007-10-26
21 *
22 * All EFI Runtime Services are not implemented yet as EFI only
23 * supports physical mode addressing on SoftSDV. This is to be fixed
24 * in a future version. --drummond 1999-07-20
25 *
26 * Implemented EFI runtime services and virtual mode calls. --davidm
27 *
28 * Goutham Rao: <goutham.rao@intel.com>
29 * Skip non-WB memory and ignore empty memory ranges.
30 */
31
32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33
34#include <linux/kernel.h>
35#include <linux/init.h>
36#include <linux/efi.h>
37#include <linux/efi-bgrt.h>
38#include <linux/export.h>
39#include <linux/memblock.h>
40#include <linux/slab.h>
41#include <linux/spinlock.h>
42#include <linux/uaccess.h>
43#include <linux/time.h>
44#include <linux/io.h>
45#include <linux/reboot.h>
46#include <linux/bcd.h>
47
48#include <asm/setup.h>
49#include <asm/efi.h>
50#include <asm/e820/api.h>
51#include <asm/time.h>
52#include <asm/tlbflush.h>
53#include <asm/x86_init.h>
54#include <asm/uv/uv.h>
55
56static unsigned long efi_systab_phys __initdata;
57static unsigned long prop_phys = EFI_INVALID_TABLE_ADDR;
58static unsigned long uga_phys = EFI_INVALID_TABLE_ADDR;
59static unsigned long efi_runtime, efi_nr_tables;
60
61unsigned long efi_fw_vendor, efi_config_table;
62
63static const efi_config_table_type_t arch_tables[] __initconst = {
64 {EFI_PROPERTIES_TABLE_GUID, &prop_phys, "PROP" },
65 {UGA_IO_PROTOCOL_GUID, &uga_phys, "UGA" },
66#ifdef CONFIG_X86_UV
67 {UV_SYSTEM_TABLE_GUID, &uv_systab_phys, "UVsystab" },
68#endif
69 {},
70};
71
72static const unsigned long * const efi_tables[] = {
73 &efi.acpi,
74 &efi.acpi20,
75 &efi.smbios,
76 &efi.smbios3,
77 &uga_phys,
78#ifdef CONFIG_X86_UV
79 &uv_systab_phys,
80#endif
81 &efi_fw_vendor,
82 &efi_runtime,
83 &efi_config_table,
84 &efi.esrt,
85 &prop_phys,
86 &efi_mem_attr_table,
87#ifdef CONFIG_EFI_RCI2_TABLE
88 &rci2_table_phys,
89#endif
90 &efi.tpm_log,
91 &efi.tpm_final_log,
92 &efi_rng_seed,
93#ifdef CONFIG_LOAD_UEFI_KEYS
94 &efi.mokvar_table,
95#endif
96#ifdef CONFIG_EFI_COCO_SECRET
97 &efi.coco_secret,
98#endif
99};
100
101u64 efi_setup; /* efi setup_data physical address */
102
103static int add_efi_memmap __initdata;
104static int __init setup_add_efi_memmap(char *arg)
105{
106 add_efi_memmap = 1;
107 return 0;
108}
109early_param("add_efi_memmap", setup_add_efi_memmap);
110
111/*
112 * Tell the kernel about the EFI memory map. This might include
113 * more than the max 128 entries that can fit in the passed in e820
114 * legacy (zeropage) memory map, but the kernel's e820 table can hold
115 * E820_MAX_ENTRIES.
116 */
117
118static void __init do_add_efi_memmap(void)
119{
120 efi_memory_desc_t *md;
121
122 if (!efi_enabled(EFI_MEMMAP))
123 return;
124
125 for_each_efi_memory_desc(md) {
126 unsigned long long start = md->phys_addr;
127 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
128 int e820_type;
129
130 switch (md->type) {
131 case EFI_LOADER_CODE:
132 case EFI_LOADER_DATA:
133 case EFI_BOOT_SERVICES_CODE:
134 case EFI_BOOT_SERVICES_DATA:
135 case EFI_CONVENTIONAL_MEMORY:
136 if (efi_soft_reserve_enabled()
137 && (md->attribute & EFI_MEMORY_SP))
138 e820_type = E820_TYPE_SOFT_RESERVED;
139 else if (md->attribute & EFI_MEMORY_WB)
140 e820_type = E820_TYPE_RAM;
141 else
142 e820_type = E820_TYPE_RESERVED;
143 break;
144 case EFI_ACPI_RECLAIM_MEMORY:
145 e820_type = E820_TYPE_ACPI;
146 break;
147 case EFI_ACPI_MEMORY_NVS:
148 e820_type = E820_TYPE_NVS;
149 break;
150 case EFI_UNUSABLE_MEMORY:
151 e820_type = E820_TYPE_UNUSABLE;
152 break;
153 case EFI_PERSISTENT_MEMORY:
154 e820_type = E820_TYPE_PMEM;
155 break;
156 default:
157 /*
158 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
159 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
160 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
161 */
162 e820_type = E820_TYPE_RESERVED;
163 break;
164 }
165
166 e820__range_add(start, size, e820_type);
167 }
168 e820__update_table(e820_table);
169}
170
171/*
172 * Given add_efi_memmap defaults to 0 and there is no alternative
173 * e820 mechanism for soft-reserved memory, import the full EFI memory
174 * map if soft reservations are present and enabled. Otherwise, the
175 * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
176 * the efi=nosoftreserve option.
177 */
178static bool do_efi_soft_reserve(void)
179{
180 efi_memory_desc_t *md;
181
182 if (!efi_enabled(EFI_MEMMAP))
183 return false;
184
185 if (!efi_soft_reserve_enabled())
186 return false;
187
188 for_each_efi_memory_desc(md)
189 if (md->type == EFI_CONVENTIONAL_MEMORY &&
190 (md->attribute & EFI_MEMORY_SP))
191 return true;
192 return false;
193}
194
195int __init efi_memblock_x86_reserve_range(void)
196{
197 struct efi_info *e = &boot_params.efi_info;
198 struct efi_memory_map_data data;
199 phys_addr_t pmap;
200 int rv;
201
202 if (efi_enabled(EFI_PARAVIRT))
203 return 0;
204
205 /* Can't handle firmware tables above 4GB on i386 */
206 if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
207 pr_err("Memory map is above 4GB, disabling EFI.\n");
208 return -EINVAL;
209 }
210 pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
211
212 data.phys_map = pmap;
213 data.size = e->efi_memmap_size;
214 data.desc_size = e->efi_memdesc_size;
215 data.desc_version = e->efi_memdesc_version;
216
217 if (!efi_enabled(EFI_PARAVIRT)) {
218 rv = efi_memmap_init_early(&data);
219 if (rv)
220 return rv;
221 }
222
223 if (add_efi_memmap || do_efi_soft_reserve())
224 do_add_efi_memmap();
225
226 efi_fake_memmap_early();
227
228 WARN(efi.memmap.desc_version != 1,
229 "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
230 efi.memmap.desc_version);
231
232 memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
233 set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
234
235 return 0;
236}
237
238#define OVERFLOW_ADDR_SHIFT (64 - EFI_PAGE_SHIFT)
239#define OVERFLOW_ADDR_MASK (U64_MAX << OVERFLOW_ADDR_SHIFT)
240#define U64_HIGH_BIT (~(U64_MAX >> 1))
241
242static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
243{
244 u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
245 u64 end_hi = 0;
246 char buf[64];
247
248 if (md->num_pages == 0) {
249 end = 0;
250 } else if (md->num_pages > EFI_PAGES_MAX ||
251 EFI_PAGES_MAX - md->num_pages <
252 (md->phys_addr >> EFI_PAGE_SHIFT)) {
253 end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
254 >> OVERFLOW_ADDR_SHIFT;
255
256 if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
257 end_hi += 1;
258 } else {
259 return true;
260 }
261
262 pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
263
264 if (end_hi) {
265 pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
266 i, efi_md_typeattr_format(buf, sizeof(buf), md),
267 md->phys_addr, end_hi, end);
268 } else {
269 pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
270 i, efi_md_typeattr_format(buf, sizeof(buf), md),
271 md->phys_addr, end);
272 }
273 return false;
274}
275
276static void __init efi_clean_memmap(void)
277{
278 efi_memory_desc_t *out = efi.memmap.map;
279 const efi_memory_desc_t *in = out;
280 const efi_memory_desc_t *end = efi.memmap.map_end;
281 int i, n_removal;
282
283 for (i = n_removal = 0; in < end; i++) {
284 if (efi_memmap_entry_valid(in, i)) {
285 if (out != in)
286 memcpy(out, in, efi.memmap.desc_size);
287 out = (void *)out + efi.memmap.desc_size;
288 } else {
289 n_removal++;
290 }
291 in = (void *)in + efi.memmap.desc_size;
292 }
293
294 if (n_removal > 0) {
295 struct efi_memory_map_data data = {
296 .phys_map = efi.memmap.phys_map,
297 .desc_version = efi.memmap.desc_version,
298 .desc_size = efi.memmap.desc_size,
299 .size = efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
300 .flags = 0,
301 };
302
303 pr_warn("Removing %d invalid memory map entries.\n", n_removal);
304 efi_memmap_install(&data);
305 }
306}
307
308/*
309 * Firmware can use EfiMemoryMappedIO to request that MMIO regions be
310 * mapped by the OS so they can be accessed by EFI runtime services, but
311 * should have no other significance to the OS (UEFI r2.10, sec 7.2).
312 * However, most bootloaders and EFI stubs convert EfiMemoryMappedIO
313 * regions to E820_TYPE_RESERVED entries, which prevent Linux from
314 * allocating space from them (see remove_e820_regions()).
315 *
316 * Some platforms use EfiMemoryMappedIO entries for PCI MMCONFIG space and
317 * PCI host bridge windows, which means Linux can't allocate BAR space for
318 * hot-added devices.
319 *
320 * Remove large EfiMemoryMappedIO regions from the E820 map to avoid this
321 * problem.
322 *
323 * Retain small EfiMemoryMappedIO regions because on some platforms, these
324 * describe non-window space that's included in host bridge _CRS. If we
325 * assign that space to PCI devices, they don't work.
326 */
327static void __init efi_remove_e820_mmio(void)
328{
329 efi_memory_desc_t *md;
330 u64 size, start, end;
331 int i = 0;
332
333 for_each_efi_memory_desc(md) {
334 if (md->type == EFI_MEMORY_MAPPED_IO) {
335 size = md->num_pages << EFI_PAGE_SHIFT;
336 start = md->phys_addr;
337 end = start + size - 1;
338 if (size >= 256*1024) {
339 pr_info("Remove mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluMB) from e820 map\n",
340 i, start, end, size >> 20);
341 e820__range_remove(start, size,
342 E820_TYPE_RESERVED, 1);
343 } else {
344 pr_info("Not removing mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluKB) from e820 map\n",
345 i, start, end, size >> 10);
346 }
347 }
348 i++;
349 }
350}
351
352void __init efi_print_memmap(void)
353{
354 efi_memory_desc_t *md;
355 int i = 0;
356
357 for_each_efi_memory_desc(md) {
358 char buf[64];
359
360 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
361 i++, efi_md_typeattr_format(buf, sizeof(buf), md),
362 md->phys_addr,
363 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
364 (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
365 }
366}
367
368static int __init efi_systab_init(unsigned long phys)
369{
370 int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
371 : sizeof(efi_system_table_32_t);
372 const efi_table_hdr_t *hdr;
373 bool over4g = false;
374 void *p;
375 int ret;
376
377 hdr = p = early_memremap_ro(phys, size);
378 if (p == NULL) {
379 pr_err("Couldn't map the system table!\n");
380 return -ENOMEM;
381 }
382
383 ret = efi_systab_check_header(hdr, 1);
384 if (ret) {
385 early_memunmap(p, size);
386 return ret;
387 }
388
389 if (efi_enabled(EFI_64BIT)) {
390 const efi_system_table_64_t *systab64 = p;
391
392 efi_runtime = systab64->runtime;
393 over4g = systab64->runtime > U32_MAX;
394
395 if (efi_setup) {
396 struct efi_setup_data *data;
397
398 data = early_memremap_ro(efi_setup, sizeof(*data));
399 if (!data) {
400 early_memunmap(p, size);
401 return -ENOMEM;
402 }
403
404 efi_fw_vendor = (unsigned long)data->fw_vendor;
405 efi_config_table = (unsigned long)data->tables;
406
407 over4g |= data->fw_vendor > U32_MAX ||
408 data->tables > U32_MAX;
409
410 early_memunmap(data, sizeof(*data));
411 } else {
412 efi_fw_vendor = systab64->fw_vendor;
413 efi_config_table = systab64->tables;
414
415 over4g |= systab64->fw_vendor > U32_MAX ||
416 systab64->tables > U32_MAX;
417 }
418 efi_nr_tables = systab64->nr_tables;
419 } else {
420 const efi_system_table_32_t *systab32 = p;
421
422 efi_fw_vendor = systab32->fw_vendor;
423 efi_runtime = systab32->runtime;
424 efi_config_table = systab32->tables;
425 efi_nr_tables = systab32->nr_tables;
426 }
427
428 efi.runtime_version = hdr->revision;
429
430 efi_systab_report_header(hdr, efi_fw_vendor);
431 early_memunmap(p, size);
432
433 if (IS_ENABLED(CONFIG_X86_32) && over4g) {
434 pr_err("EFI data located above 4GB, disabling EFI.\n");
435 return -EINVAL;
436 }
437
438 return 0;
439}
440
441static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
442{
443 void *config_tables;
444 int sz, ret;
445
446 if (efi_nr_tables == 0)
447 return 0;
448
449 if (efi_enabled(EFI_64BIT))
450 sz = sizeof(efi_config_table_64_t);
451 else
452 sz = sizeof(efi_config_table_32_t);
453
454 /*
455 * Let's see what config tables the firmware passed to us.
456 */
457 config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
458 if (config_tables == NULL) {
459 pr_err("Could not map Configuration table!\n");
460 return -ENOMEM;
461 }
462
463 ret = efi_config_parse_tables(config_tables, efi_nr_tables,
464 arch_tables);
465
466 early_memunmap(config_tables, efi_nr_tables * sz);
467 return ret;
468}
469
470void __init efi_init(void)
471{
472 if (IS_ENABLED(CONFIG_X86_32) &&
473 (boot_params.efi_info.efi_systab_hi ||
474 boot_params.efi_info.efi_memmap_hi)) {
475 pr_info("Table located above 4GB, disabling EFI.\n");
476 return;
477 }
478
479 efi_systab_phys = boot_params.efi_info.efi_systab |
480 ((__u64)boot_params.efi_info.efi_systab_hi << 32);
481
482 if (efi_systab_init(efi_systab_phys))
483 return;
484
485 if (efi_reuse_config(efi_config_table, efi_nr_tables))
486 return;
487
488 if (efi_config_init(arch_tables))
489 return;
490
491 /*
492 * Note: We currently don't support runtime services on an EFI
493 * that doesn't match the kernel 32/64-bit mode.
494 */
495
496 if (!efi_runtime_supported())
497 pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
498
499 if (!efi_runtime_supported() || efi_runtime_disabled()) {
500 efi_memmap_unmap();
501 return;
502 }
503
504 /* Parse the EFI Properties table if it exists */
505 if (prop_phys != EFI_INVALID_TABLE_ADDR) {
506 efi_properties_table_t *tbl;
507
508 tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
509 if (tbl == NULL) {
510 pr_err("Could not map Properties table!\n");
511 } else {
512 if (tbl->memory_protection_attribute &
513 EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
514 set_bit(EFI_NX_PE_DATA, &efi.flags);
515
516 early_memunmap(tbl, sizeof(*tbl));
517 }
518 }
519
520 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
521 efi_clean_memmap();
522
523 efi_remove_e820_mmio();
524
525 if (efi_enabled(EFI_DBG))
526 efi_print_memmap();
527}
528
529/* Merge contiguous regions of the same type and attribute */
530static void __init efi_merge_regions(void)
531{
532 efi_memory_desc_t *md, *prev_md = NULL;
533
534 for_each_efi_memory_desc(md) {
535 u64 prev_size;
536
537 if (!prev_md) {
538 prev_md = md;
539 continue;
540 }
541
542 if (prev_md->type != md->type ||
543 prev_md->attribute != md->attribute) {
544 prev_md = md;
545 continue;
546 }
547
548 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
549
550 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
551 prev_md->num_pages += md->num_pages;
552 md->type = EFI_RESERVED_TYPE;
553 md->attribute = 0;
554 continue;
555 }
556 prev_md = md;
557 }
558}
559
560static void *realloc_pages(void *old_memmap, int old_shift)
561{
562 void *ret;
563
564 ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
565 if (!ret)
566 goto out;
567
568 /*
569 * A first-time allocation doesn't have anything to copy.
570 */
571 if (!old_memmap)
572 return ret;
573
574 memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
575
576out:
577 free_pages((unsigned long)old_memmap, old_shift);
578 return ret;
579}
580
581/*
582 * Iterate the EFI memory map in reverse order because the regions
583 * will be mapped top-down. The end result is the same as if we had
584 * mapped things forward, but doesn't require us to change the
585 * existing implementation of efi_map_region().
586 */
587static inline void *efi_map_next_entry_reverse(void *entry)
588{
589 /* Initial call */
590 if (!entry)
591 return efi.memmap.map_end - efi.memmap.desc_size;
592
593 entry -= efi.memmap.desc_size;
594 if (entry < efi.memmap.map)
595 return NULL;
596
597 return entry;
598}
599
600/*
601 * efi_map_next_entry - Return the next EFI memory map descriptor
602 * @entry: Previous EFI memory map descriptor
603 *
604 * This is a helper function to iterate over the EFI memory map, which
605 * we do in different orders depending on the current configuration.
606 *
607 * To begin traversing the memory map @entry must be %NULL.
608 *
609 * Returns %NULL when we reach the end of the memory map.
610 */
611static void *efi_map_next_entry(void *entry)
612{
613 if (efi_enabled(EFI_64BIT)) {
614 /*
615 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
616 * config table feature requires us to map all entries
617 * in the same order as they appear in the EFI memory
618 * map. That is to say, entry N must have a lower
619 * virtual address than entry N+1. This is because the
620 * firmware toolchain leaves relative references in
621 * the code/data sections, which are split and become
622 * separate EFI memory regions. Mapping things
623 * out-of-order leads to the firmware accessing
624 * unmapped addresses.
625 *
626 * Since we need to map things this way whether or not
627 * the kernel actually makes use of
628 * EFI_PROPERTIES_TABLE, let's just switch to this
629 * scheme by default for 64-bit.
630 */
631 return efi_map_next_entry_reverse(entry);
632 }
633
634 /* Initial call */
635 if (!entry)
636 return efi.memmap.map;
637
638 entry += efi.memmap.desc_size;
639 if (entry >= efi.memmap.map_end)
640 return NULL;
641
642 return entry;
643}
644
645static bool should_map_region(efi_memory_desc_t *md)
646{
647 /*
648 * Runtime regions always require runtime mappings (obviously).
649 */
650 if (md->attribute & EFI_MEMORY_RUNTIME)
651 return true;
652
653 /*
654 * 32-bit EFI doesn't suffer from the bug that requires us to
655 * reserve boot services regions, and mixed mode support
656 * doesn't exist for 32-bit kernels.
657 */
658 if (IS_ENABLED(CONFIG_X86_32))
659 return false;
660
661 /*
662 * EFI specific purpose memory may be reserved by default
663 * depending on kernel config and boot options.
664 */
665 if (md->type == EFI_CONVENTIONAL_MEMORY &&
666 efi_soft_reserve_enabled() &&
667 (md->attribute & EFI_MEMORY_SP))
668 return false;
669
670 /*
671 * Map all of RAM so that we can access arguments in the 1:1
672 * mapping when making EFI runtime calls.
673 */
674 if (efi_is_mixed()) {
675 if (md->type == EFI_CONVENTIONAL_MEMORY ||
676 md->type == EFI_LOADER_DATA ||
677 md->type == EFI_LOADER_CODE)
678 return true;
679 }
680
681 /*
682 * Map boot services regions as a workaround for buggy
683 * firmware that accesses them even when they shouldn't.
684 *
685 * See efi_{reserve,free}_boot_services().
686 */
687 if (md->type == EFI_BOOT_SERVICES_CODE ||
688 md->type == EFI_BOOT_SERVICES_DATA)
689 return true;
690
691 return false;
692}
693
694/*
695 * Map the efi memory ranges of the runtime services and update new_mmap with
696 * virtual addresses.
697 */
698static void * __init efi_map_regions(int *count, int *pg_shift)
699{
700 void *p, *new_memmap = NULL;
701 unsigned long left = 0;
702 unsigned long desc_size;
703 efi_memory_desc_t *md;
704
705 desc_size = efi.memmap.desc_size;
706
707 p = NULL;
708 while ((p = efi_map_next_entry(p))) {
709 md = p;
710
711 if (!should_map_region(md))
712 continue;
713
714 efi_map_region(md);
715
716 if (left < desc_size) {
717 new_memmap = realloc_pages(new_memmap, *pg_shift);
718 if (!new_memmap)
719 return NULL;
720
721 left += PAGE_SIZE << *pg_shift;
722 (*pg_shift)++;
723 }
724
725 memcpy(new_memmap + (*count * desc_size), md, desc_size);
726
727 left -= desc_size;
728 (*count)++;
729 }
730
731 return new_memmap;
732}
733
734static void __init kexec_enter_virtual_mode(void)
735{
736#ifdef CONFIG_KEXEC_CORE
737 efi_memory_desc_t *md;
738 unsigned int num_pages;
739
740 /*
741 * We don't do virtual mode, since we don't do runtime services, on
742 * non-native EFI.
743 */
744 if (efi_is_mixed()) {
745 efi_memmap_unmap();
746 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
747 return;
748 }
749
750 if (efi_alloc_page_tables()) {
751 pr_err("Failed to allocate EFI page tables\n");
752 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
753 return;
754 }
755
756 /*
757 * Map efi regions which were passed via setup_data. The virt_addr is a
758 * fixed addr which was used in first kernel of a kexec boot.
759 */
760 for_each_efi_memory_desc(md)
761 efi_map_region_fixed(md); /* FIXME: add error handling */
762
763 /*
764 * Unregister the early EFI memmap from efi_init() and install
765 * the new EFI memory map.
766 */
767 efi_memmap_unmap();
768
769 if (efi_memmap_init_late(efi.memmap.phys_map,
770 efi.memmap.desc_size * efi.memmap.nr_map)) {
771 pr_err("Failed to remap late EFI memory map\n");
772 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
773 return;
774 }
775
776 num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
777 num_pages >>= PAGE_SHIFT;
778
779 if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
780 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
781 return;
782 }
783
784 efi_sync_low_kernel_mappings();
785 efi_native_runtime_setup();
786#endif
787}
788
789/*
790 * This function will switch the EFI runtime services to virtual mode.
791 * Essentially, we look through the EFI memmap and map every region that
792 * has the runtime attribute bit set in its memory descriptor into the
793 * efi_pgd page table.
794 *
795 * The new method does a pagetable switch in a preemption-safe manner
796 * so that we're in a different address space when calling a runtime
797 * function. For function arguments passing we do copy the PUDs of the
798 * kernel page table into efi_pgd prior to each call.
799 *
800 * Specially for kexec boot, efi runtime maps in previous kernel should
801 * be passed in via setup_data. In that case runtime ranges will be mapped
802 * to the same virtual addresses as the first kernel, see
803 * kexec_enter_virtual_mode().
804 */
805static void __init __efi_enter_virtual_mode(void)
806{
807 int count = 0, pg_shift = 0;
808 void *new_memmap = NULL;
809 efi_status_t status;
810 unsigned long pa;
811
812 if (efi_alloc_page_tables()) {
813 pr_err("Failed to allocate EFI page tables\n");
814 goto err;
815 }
816
817 efi_merge_regions();
818 new_memmap = efi_map_regions(&count, &pg_shift);
819 if (!new_memmap) {
820 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
821 goto err;
822 }
823
824 pa = __pa(new_memmap);
825
826 /*
827 * Unregister the early EFI memmap from efi_init() and install
828 * the new EFI memory map that we are about to pass to the
829 * firmware via SetVirtualAddressMap().
830 */
831 efi_memmap_unmap();
832
833 if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
834 pr_err("Failed to remap late EFI memory map\n");
835 goto err;
836 }
837
838 if (efi_enabled(EFI_DBG)) {
839 pr_info("EFI runtime memory map:\n");
840 efi_print_memmap();
841 }
842
843 if (efi_setup_page_tables(pa, 1 << pg_shift))
844 goto err;
845
846 efi_sync_low_kernel_mappings();
847
848 status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
849 efi.memmap.desc_size,
850 efi.memmap.desc_version,
851 (efi_memory_desc_t *)pa,
852 efi_systab_phys);
853 if (status != EFI_SUCCESS) {
854 pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
855 status);
856 goto err;
857 }
858
859 efi_check_for_embedded_firmwares();
860 efi_free_boot_services();
861
862 if (!efi_is_mixed())
863 efi_native_runtime_setup();
864 else
865 efi_thunk_runtime_setup();
866
867 /*
868 * Apply more restrictive page table mapping attributes now that
869 * SVAM() has been called and the firmware has performed all
870 * necessary relocation fixups for the new virtual addresses.
871 */
872 efi_runtime_update_mappings();
873
874 /* clean DUMMY object */
875 efi_delete_dummy_variable();
876 return;
877
878err:
879 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
880}
881
882void __init efi_enter_virtual_mode(void)
883{
884 if (efi_enabled(EFI_PARAVIRT))
885 return;
886
887 efi.runtime = (efi_runtime_services_t *)efi_runtime;
888
889 if (efi_setup)
890 kexec_enter_virtual_mode();
891 else
892 __efi_enter_virtual_mode();
893
894 efi_dump_pagetable();
895}
896
897bool efi_is_table_address(unsigned long phys_addr)
898{
899 unsigned int i;
900
901 if (phys_addr == EFI_INVALID_TABLE_ADDR)
902 return false;
903
904 for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
905 if (*(efi_tables[i]) == phys_addr)
906 return true;
907
908 return false;
909}
910
911char *efi_systab_show_arch(char *str)
912{
913 if (uga_phys != EFI_INVALID_TABLE_ADDR)
914 str += sprintf(str, "UGA=0x%lx\n", uga_phys);
915 return str;
916}
917
918#define EFI_FIELD(var) efi_ ## var
919
920#define EFI_ATTR_SHOW(name) \
921static ssize_t name##_show(struct kobject *kobj, \
922 struct kobj_attribute *attr, char *buf) \
923{ \
924 return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
925}
926
927EFI_ATTR_SHOW(fw_vendor);
928EFI_ATTR_SHOW(runtime);
929EFI_ATTR_SHOW(config_table);
930
931struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
932struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
933struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
934
935umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
936{
937 if (attr == &efi_attr_fw_vendor.attr) {
938 if (efi_enabled(EFI_PARAVIRT) ||
939 efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
940 return 0;
941 } else if (attr == &efi_attr_runtime.attr) {
942 if (efi_runtime == EFI_INVALID_TABLE_ADDR)
943 return 0;
944 } else if (attr == &efi_attr_config_table.attr) {
945 if (efi_config_table == EFI_INVALID_TABLE_ADDR)
946 return 0;
947 }
948 return attr->mode;
949}
1/*
2 * Common EFI (Extensible Firmware Interface) support functions
3 * Based on Extensible Firmware Interface Specification version 1.0
4 *
5 * Copyright (C) 1999 VA Linux Systems
6 * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
7 * Copyright (C) 1999-2002 Hewlett-Packard Co.
8 * David Mosberger-Tang <davidm@hpl.hp.com>
9 * Stephane Eranian <eranian@hpl.hp.com>
10 * Copyright (C) 2005-2008 Intel Co.
11 * Fenghua Yu <fenghua.yu@intel.com>
12 * Bibo Mao <bibo.mao@intel.com>
13 * Chandramouli Narayanan <mouli@linux.intel.com>
14 * Huang Ying <ying.huang@intel.com>
15 * Copyright (C) 2013 SuSE Labs
16 * Borislav Petkov <bp@suse.de> - runtime services VA mapping
17 *
18 * Copied from efi_32.c to eliminate the duplicated code between EFI
19 * 32/64 support code. --ying 2007-10-26
20 *
21 * All EFI Runtime Services are not implemented yet as EFI only
22 * supports physical mode addressing on SoftSDV. This is to be fixed
23 * in a future version. --drummond 1999-07-20
24 *
25 * Implemented EFI runtime services and virtual mode calls. --davidm
26 *
27 * Goutham Rao: <goutham.rao@intel.com>
28 * Skip non-WB memory and ignore empty memory ranges.
29 */
30
31#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32
33#include <linux/kernel.h>
34#include <linux/init.h>
35#include <linux/efi.h>
36#include <linux/efi-bgrt.h>
37#include <linux/export.h>
38#include <linux/bootmem.h>
39#include <linux/slab.h>
40#include <linux/memblock.h>
41#include <linux/spinlock.h>
42#include <linux/uaccess.h>
43#include <linux/time.h>
44#include <linux/io.h>
45#include <linux/reboot.h>
46#include <linux/bcd.h>
47
48#include <asm/setup.h>
49#include <asm/efi.h>
50#include <asm/time.h>
51#include <asm/cacheflush.h>
52#include <asm/tlbflush.h>
53#include <asm/x86_init.h>
54#include <asm/rtc.h>
55#include <asm/uv/uv.h>
56
57#define EFI_DEBUG
58
59struct efi_memory_map memmap;
60
61static struct efi efi_phys __initdata;
62static efi_system_table_t efi_systab __initdata;
63
64static efi_config_table_type_t arch_tables[] __initdata = {
65#ifdef CONFIG_X86_UV
66 {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
67#endif
68 {NULL_GUID, NULL, NULL},
69};
70
71u64 efi_setup; /* efi setup_data physical address */
72
73static int add_efi_memmap __initdata;
74static int __init setup_add_efi_memmap(char *arg)
75{
76 add_efi_memmap = 1;
77 return 0;
78}
79early_param("add_efi_memmap", setup_add_efi_memmap);
80
81static efi_status_t __init phys_efi_set_virtual_address_map(
82 unsigned long memory_map_size,
83 unsigned long descriptor_size,
84 u32 descriptor_version,
85 efi_memory_desc_t *virtual_map)
86{
87 efi_status_t status;
88 unsigned long flags;
89 pgd_t *save_pgd;
90
91 save_pgd = efi_call_phys_prolog();
92
93 /* Disable interrupts around EFI calls: */
94 local_irq_save(flags);
95 status = efi_call_phys(efi_phys.set_virtual_address_map,
96 memory_map_size, descriptor_size,
97 descriptor_version, virtual_map);
98 local_irq_restore(flags);
99
100 efi_call_phys_epilog(save_pgd);
101
102 return status;
103}
104
105void efi_get_time(struct timespec *now)
106{
107 efi_status_t status;
108 efi_time_t eft;
109 efi_time_cap_t cap;
110
111 status = efi.get_time(&eft, &cap);
112 if (status != EFI_SUCCESS)
113 pr_err("Oops: efitime: can't read time!\n");
114
115 now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
116 eft.minute, eft.second);
117 now->tv_nsec = 0;
118}
119
120void __init efi_find_mirror(void)
121{
122 void *p;
123 u64 mirror_size = 0, total_size = 0;
124
125 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
126 efi_memory_desc_t *md = p;
127 unsigned long long start = md->phys_addr;
128 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
129
130 total_size += size;
131 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
132 memblock_mark_mirror(start, size);
133 mirror_size += size;
134 }
135 }
136 if (mirror_size)
137 pr_info("Memory: %lldM/%lldM mirrored memory\n",
138 mirror_size>>20, total_size>>20);
139}
140
141/*
142 * Tell the kernel about the EFI memory map. This might include
143 * more than the max 128 entries that can fit in the e820 legacy
144 * (zeropage) memory map.
145 */
146
147static void __init do_add_efi_memmap(void)
148{
149 void *p;
150
151 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
152 efi_memory_desc_t *md = p;
153 unsigned long long start = md->phys_addr;
154 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
155 int e820_type;
156
157 switch (md->type) {
158 case EFI_LOADER_CODE:
159 case EFI_LOADER_DATA:
160 case EFI_BOOT_SERVICES_CODE:
161 case EFI_BOOT_SERVICES_DATA:
162 case EFI_CONVENTIONAL_MEMORY:
163 if (md->attribute & EFI_MEMORY_WB)
164 e820_type = E820_RAM;
165 else
166 e820_type = E820_RESERVED;
167 break;
168 case EFI_ACPI_RECLAIM_MEMORY:
169 e820_type = E820_ACPI;
170 break;
171 case EFI_ACPI_MEMORY_NVS:
172 e820_type = E820_NVS;
173 break;
174 case EFI_UNUSABLE_MEMORY:
175 e820_type = E820_UNUSABLE;
176 break;
177 case EFI_PERSISTENT_MEMORY:
178 e820_type = E820_PMEM;
179 break;
180 default:
181 /*
182 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
183 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
184 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
185 */
186 e820_type = E820_RESERVED;
187 break;
188 }
189 e820_add_region(start, size, e820_type);
190 }
191 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
192}
193
194int __init efi_memblock_x86_reserve_range(void)
195{
196 struct efi_info *e = &boot_params.efi_info;
197 phys_addr_t pmap;
198
199 if (efi_enabled(EFI_PARAVIRT))
200 return 0;
201
202#ifdef CONFIG_X86_32
203 /* Can't handle data above 4GB at this time */
204 if (e->efi_memmap_hi) {
205 pr_err("Memory map is above 4GB, disabling EFI.\n");
206 return -EINVAL;
207 }
208 pmap = e->efi_memmap;
209#else
210 pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
211#endif
212 memmap.phys_map = pmap;
213 memmap.nr_map = e->efi_memmap_size /
214 e->efi_memdesc_size;
215 memmap.desc_size = e->efi_memdesc_size;
216 memmap.desc_version = e->efi_memdesc_version;
217
218 memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
219
220 efi.memmap = &memmap;
221
222 return 0;
223}
224
225void __init efi_print_memmap(void)
226{
227#ifdef EFI_DEBUG
228 efi_memory_desc_t *md;
229 void *p;
230 int i;
231
232 for (p = memmap.map, i = 0;
233 p < memmap.map_end;
234 p += memmap.desc_size, i++) {
235 char buf[64];
236
237 md = p;
238 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
239 i, efi_md_typeattr_format(buf, sizeof(buf), md),
240 md->phys_addr,
241 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
242 (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
243 }
244#endif /* EFI_DEBUG */
245}
246
247void __init efi_unmap_memmap(void)
248{
249 clear_bit(EFI_MEMMAP, &efi.flags);
250 if (memmap.map) {
251 early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size);
252 memmap.map = NULL;
253 }
254}
255
256static int __init efi_systab_init(void *phys)
257{
258 if (efi_enabled(EFI_64BIT)) {
259 efi_system_table_64_t *systab64;
260 struct efi_setup_data *data = NULL;
261 u64 tmp = 0;
262
263 if (efi_setup) {
264 data = early_memremap(efi_setup, sizeof(*data));
265 if (!data)
266 return -ENOMEM;
267 }
268 systab64 = early_memremap((unsigned long)phys,
269 sizeof(*systab64));
270 if (systab64 == NULL) {
271 pr_err("Couldn't map the system table!\n");
272 if (data)
273 early_memunmap(data, sizeof(*data));
274 return -ENOMEM;
275 }
276
277 efi_systab.hdr = systab64->hdr;
278 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
279 systab64->fw_vendor;
280 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
281 efi_systab.fw_revision = systab64->fw_revision;
282 efi_systab.con_in_handle = systab64->con_in_handle;
283 tmp |= systab64->con_in_handle;
284 efi_systab.con_in = systab64->con_in;
285 tmp |= systab64->con_in;
286 efi_systab.con_out_handle = systab64->con_out_handle;
287 tmp |= systab64->con_out_handle;
288 efi_systab.con_out = systab64->con_out;
289 tmp |= systab64->con_out;
290 efi_systab.stderr_handle = systab64->stderr_handle;
291 tmp |= systab64->stderr_handle;
292 efi_systab.stderr = systab64->stderr;
293 tmp |= systab64->stderr;
294 efi_systab.runtime = data ?
295 (void *)(unsigned long)data->runtime :
296 (void *)(unsigned long)systab64->runtime;
297 tmp |= data ? data->runtime : systab64->runtime;
298 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
299 tmp |= systab64->boottime;
300 efi_systab.nr_tables = systab64->nr_tables;
301 efi_systab.tables = data ? (unsigned long)data->tables :
302 systab64->tables;
303 tmp |= data ? data->tables : systab64->tables;
304
305 early_memunmap(systab64, sizeof(*systab64));
306 if (data)
307 early_memunmap(data, sizeof(*data));
308#ifdef CONFIG_X86_32
309 if (tmp >> 32) {
310 pr_err("EFI data located above 4GB, disabling EFI.\n");
311 return -EINVAL;
312 }
313#endif
314 } else {
315 efi_system_table_32_t *systab32;
316
317 systab32 = early_memremap((unsigned long)phys,
318 sizeof(*systab32));
319 if (systab32 == NULL) {
320 pr_err("Couldn't map the system table!\n");
321 return -ENOMEM;
322 }
323
324 efi_systab.hdr = systab32->hdr;
325 efi_systab.fw_vendor = systab32->fw_vendor;
326 efi_systab.fw_revision = systab32->fw_revision;
327 efi_systab.con_in_handle = systab32->con_in_handle;
328 efi_systab.con_in = systab32->con_in;
329 efi_systab.con_out_handle = systab32->con_out_handle;
330 efi_systab.con_out = systab32->con_out;
331 efi_systab.stderr_handle = systab32->stderr_handle;
332 efi_systab.stderr = systab32->stderr;
333 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
334 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
335 efi_systab.nr_tables = systab32->nr_tables;
336 efi_systab.tables = systab32->tables;
337
338 early_memunmap(systab32, sizeof(*systab32));
339 }
340
341 efi.systab = &efi_systab;
342
343 /*
344 * Verify the EFI Table
345 */
346 if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
347 pr_err("System table signature incorrect!\n");
348 return -EINVAL;
349 }
350 if ((efi.systab->hdr.revision >> 16) == 0)
351 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
352 efi.systab->hdr.revision >> 16,
353 efi.systab->hdr.revision & 0xffff);
354
355 set_bit(EFI_SYSTEM_TABLES, &efi.flags);
356
357 return 0;
358}
359
360static int __init efi_runtime_init32(void)
361{
362 efi_runtime_services_32_t *runtime;
363
364 runtime = early_memremap((unsigned long)efi.systab->runtime,
365 sizeof(efi_runtime_services_32_t));
366 if (!runtime) {
367 pr_err("Could not map the runtime service table!\n");
368 return -ENOMEM;
369 }
370
371 /*
372 * We will only need *early* access to the SetVirtualAddressMap
373 * EFI runtime service. All other runtime services will be called
374 * via the virtual mapping.
375 */
376 efi_phys.set_virtual_address_map =
377 (efi_set_virtual_address_map_t *)
378 (unsigned long)runtime->set_virtual_address_map;
379 early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
380
381 return 0;
382}
383
384static int __init efi_runtime_init64(void)
385{
386 efi_runtime_services_64_t *runtime;
387
388 runtime = early_memremap((unsigned long)efi.systab->runtime,
389 sizeof(efi_runtime_services_64_t));
390 if (!runtime) {
391 pr_err("Could not map the runtime service table!\n");
392 return -ENOMEM;
393 }
394
395 /*
396 * We will only need *early* access to the SetVirtualAddressMap
397 * EFI runtime service. All other runtime services will be called
398 * via the virtual mapping.
399 */
400 efi_phys.set_virtual_address_map =
401 (efi_set_virtual_address_map_t *)
402 (unsigned long)runtime->set_virtual_address_map;
403 early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
404
405 return 0;
406}
407
408static int __init efi_runtime_init(void)
409{
410 int rv;
411
412 /*
413 * Check out the runtime services table. We need to map
414 * the runtime services table so that we can grab the physical
415 * address of several of the EFI runtime functions, needed to
416 * set the firmware into virtual mode.
417 *
418 * When EFI_PARAVIRT is in force then we could not map runtime
419 * service memory region because we do not have direct access to it.
420 * However, runtime services are available through proxy functions
421 * (e.g. in case of Xen dom0 EFI implementation they call special
422 * hypercall which executes relevant EFI functions) and that is why
423 * they are always enabled.
424 */
425
426 if (!efi_enabled(EFI_PARAVIRT)) {
427 if (efi_enabled(EFI_64BIT))
428 rv = efi_runtime_init64();
429 else
430 rv = efi_runtime_init32();
431
432 if (rv)
433 return rv;
434 }
435
436 set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
437
438 return 0;
439}
440
441static int __init efi_memmap_init(void)
442{
443 if (efi_enabled(EFI_PARAVIRT))
444 return 0;
445
446 /* Map the EFI memory map */
447 memmap.map = early_memremap((unsigned long)memmap.phys_map,
448 memmap.nr_map * memmap.desc_size);
449 if (memmap.map == NULL) {
450 pr_err("Could not map the memory map!\n");
451 return -ENOMEM;
452 }
453 memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
454
455 if (add_efi_memmap)
456 do_add_efi_memmap();
457
458 set_bit(EFI_MEMMAP, &efi.flags);
459
460 return 0;
461}
462
463void __init efi_init(void)
464{
465 efi_char16_t *c16;
466 char vendor[100] = "unknown";
467 int i = 0;
468 void *tmp;
469
470#ifdef CONFIG_X86_32
471 if (boot_params.efi_info.efi_systab_hi ||
472 boot_params.efi_info.efi_memmap_hi) {
473 pr_info("Table located above 4GB, disabling EFI.\n");
474 return;
475 }
476 efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
477#else
478 efi_phys.systab = (efi_system_table_t *)
479 (boot_params.efi_info.efi_systab |
480 ((__u64)boot_params.efi_info.efi_systab_hi<<32));
481#endif
482
483 if (efi_systab_init(efi_phys.systab))
484 return;
485
486 efi.config_table = (unsigned long)efi.systab->tables;
487 efi.fw_vendor = (unsigned long)efi.systab->fw_vendor;
488 efi.runtime = (unsigned long)efi.systab->runtime;
489
490 /*
491 * Show what we know for posterity
492 */
493 c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
494 if (c16) {
495 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
496 vendor[i] = *c16++;
497 vendor[i] = '\0';
498 } else
499 pr_err("Could not map the firmware vendor!\n");
500 early_memunmap(tmp, 2);
501
502 pr_info("EFI v%u.%.02u by %s\n",
503 efi.systab->hdr.revision >> 16,
504 efi.systab->hdr.revision & 0xffff, vendor);
505
506 if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
507 return;
508
509 if (efi_config_init(arch_tables))
510 return;
511
512 /*
513 * Note: We currently don't support runtime services on an EFI
514 * that doesn't match the kernel 32/64-bit mode.
515 */
516
517 if (!efi_runtime_supported())
518 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
519 else {
520 if (efi_runtime_disabled() || efi_runtime_init())
521 return;
522 }
523 if (efi_memmap_init())
524 return;
525
526 if (efi_enabled(EFI_DBG))
527 efi_print_memmap();
528
529 efi_esrt_init();
530}
531
532void __init efi_late_init(void)
533{
534 efi_bgrt_init();
535}
536
537void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
538{
539 u64 addr, npages;
540
541 addr = md->virt_addr;
542 npages = md->num_pages;
543
544 memrange_efi_to_native(&addr, &npages);
545
546 if (executable)
547 set_memory_x(addr, npages);
548 else
549 set_memory_nx(addr, npages);
550}
551
552void __init runtime_code_page_mkexec(void)
553{
554 efi_memory_desc_t *md;
555 void *p;
556
557 /* Make EFI runtime service code area executable */
558 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
559 md = p;
560
561 if (md->type != EFI_RUNTIME_SERVICES_CODE)
562 continue;
563
564 efi_set_executable(md, true);
565 }
566}
567
568void __init efi_memory_uc(u64 addr, unsigned long size)
569{
570 unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
571 u64 npages;
572
573 npages = round_up(size, page_shift) / page_shift;
574 memrange_efi_to_native(&addr, &npages);
575 set_memory_uc(addr, npages);
576}
577
578void __init old_map_region(efi_memory_desc_t *md)
579{
580 u64 start_pfn, end_pfn, end;
581 unsigned long size;
582 void *va;
583
584 start_pfn = PFN_DOWN(md->phys_addr);
585 size = md->num_pages << PAGE_SHIFT;
586 end = md->phys_addr + size;
587 end_pfn = PFN_UP(end);
588
589 if (pfn_range_is_mapped(start_pfn, end_pfn)) {
590 va = __va(md->phys_addr);
591
592 if (!(md->attribute & EFI_MEMORY_WB))
593 efi_memory_uc((u64)(unsigned long)va, size);
594 } else
595 va = efi_ioremap(md->phys_addr, size,
596 md->type, md->attribute);
597
598 md->virt_addr = (u64) (unsigned long) va;
599 if (!va)
600 pr_err("ioremap of 0x%llX failed!\n",
601 (unsigned long long)md->phys_addr);
602}
603
604/* Merge contiguous regions of the same type and attribute */
605static void __init efi_merge_regions(void)
606{
607 void *p;
608 efi_memory_desc_t *md, *prev_md = NULL;
609
610 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
611 u64 prev_size;
612 md = p;
613
614 if (!prev_md) {
615 prev_md = md;
616 continue;
617 }
618
619 if (prev_md->type != md->type ||
620 prev_md->attribute != md->attribute) {
621 prev_md = md;
622 continue;
623 }
624
625 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
626
627 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
628 prev_md->num_pages += md->num_pages;
629 md->type = EFI_RESERVED_TYPE;
630 md->attribute = 0;
631 continue;
632 }
633 prev_md = md;
634 }
635}
636
637static void __init get_systab_virt_addr(efi_memory_desc_t *md)
638{
639 unsigned long size;
640 u64 end, systab;
641
642 size = md->num_pages << EFI_PAGE_SHIFT;
643 end = md->phys_addr + size;
644 systab = (u64)(unsigned long)efi_phys.systab;
645 if (md->phys_addr <= systab && systab < end) {
646 systab += md->virt_addr - md->phys_addr;
647 efi.systab = (efi_system_table_t *)(unsigned long)systab;
648 }
649}
650
651static void __init save_runtime_map(void)
652{
653#ifdef CONFIG_KEXEC_CORE
654 efi_memory_desc_t *md;
655 void *tmp, *p, *q = NULL;
656 int count = 0;
657
658 if (efi_enabled(EFI_OLD_MEMMAP))
659 return;
660
661 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
662 md = p;
663
664 if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
665 (md->type == EFI_BOOT_SERVICES_CODE) ||
666 (md->type == EFI_BOOT_SERVICES_DATA))
667 continue;
668 tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
669 if (!tmp)
670 goto out;
671 q = tmp;
672
673 memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
674 count++;
675 }
676
677 efi_runtime_map_setup(q, count, memmap.desc_size);
678 return;
679
680out:
681 kfree(q);
682 pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
683#endif
684}
685
686static void *realloc_pages(void *old_memmap, int old_shift)
687{
688 void *ret;
689
690 ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
691 if (!ret)
692 goto out;
693
694 /*
695 * A first-time allocation doesn't have anything to copy.
696 */
697 if (!old_memmap)
698 return ret;
699
700 memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
701
702out:
703 free_pages((unsigned long)old_memmap, old_shift);
704 return ret;
705}
706
707/*
708 * Iterate the EFI memory map in reverse order because the regions
709 * will be mapped top-down. The end result is the same as if we had
710 * mapped things forward, but doesn't require us to change the
711 * existing implementation of efi_map_region().
712 */
713static inline void *efi_map_next_entry_reverse(void *entry)
714{
715 /* Initial call */
716 if (!entry)
717 return memmap.map_end - memmap.desc_size;
718
719 entry -= memmap.desc_size;
720 if (entry < memmap.map)
721 return NULL;
722
723 return entry;
724}
725
726/*
727 * efi_map_next_entry - Return the next EFI memory map descriptor
728 * @entry: Previous EFI memory map descriptor
729 *
730 * This is a helper function to iterate over the EFI memory map, which
731 * we do in different orders depending on the current configuration.
732 *
733 * To begin traversing the memory map @entry must be %NULL.
734 *
735 * Returns %NULL when we reach the end of the memory map.
736 */
737static void *efi_map_next_entry(void *entry)
738{
739 if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
740 /*
741 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
742 * config table feature requires us to map all entries
743 * in the same order as they appear in the EFI memory
744 * map. That is to say, entry N must have a lower
745 * virtual address than entry N+1. This is because the
746 * firmware toolchain leaves relative references in
747 * the code/data sections, which are split and become
748 * separate EFI memory regions. Mapping things
749 * out-of-order leads to the firmware accessing
750 * unmapped addresses.
751 *
752 * Since we need to map things this way whether or not
753 * the kernel actually makes use of
754 * EFI_PROPERTIES_TABLE, let's just switch to this
755 * scheme by default for 64-bit.
756 */
757 return efi_map_next_entry_reverse(entry);
758 }
759
760 /* Initial call */
761 if (!entry)
762 return memmap.map;
763
764 entry += memmap.desc_size;
765 if (entry >= memmap.map_end)
766 return NULL;
767
768 return entry;
769}
770
771/*
772 * Map the efi memory ranges of the runtime services and update new_mmap with
773 * virtual addresses.
774 */
775static void * __init efi_map_regions(int *count, int *pg_shift)
776{
777 void *p, *new_memmap = NULL;
778 unsigned long left = 0;
779 efi_memory_desc_t *md;
780
781 p = NULL;
782 while ((p = efi_map_next_entry(p))) {
783 md = p;
784 if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
785#ifdef CONFIG_X86_64
786 if (md->type != EFI_BOOT_SERVICES_CODE &&
787 md->type != EFI_BOOT_SERVICES_DATA)
788#endif
789 continue;
790 }
791
792 efi_map_region(md);
793 get_systab_virt_addr(md);
794
795 if (left < memmap.desc_size) {
796 new_memmap = realloc_pages(new_memmap, *pg_shift);
797 if (!new_memmap)
798 return NULL;
799
800 left += PAGE_SIZE << *pg_shift;
801 (*pg_shift)++;
802 }
803
804 memcpy(new_memmap + (*count * memmap.desc_size), md,
805 memmap.desc_size);
806
807 left -= memmap.desc_size;
808 (*count)++;
809 }
810
811 return new_memmap;
812}
813
814static void __init kexec_enter_virtual_mode(void)
815{
816#ifdef CONFIG_KEXEC_CORE
817 efi_memory_desc_t *md;
818 unsigned int num_pages;
819 void *p;
820
821 efi.systab = NULL;
822
823 /*
824 * We don't do virtual mode, since we don't do runtime services, on
825 * non-native EFI
826 */
827 if (!efi_is_native()) {
828 efi_unmap_memmap();
829 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
830 return;
831 }
832
833 if (efi_alloc_page_tables()) {
834 pr_err("Failed to allocate EFI page tables\n");
835 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
836 return;
837 }
838
839 /*
840 * Map efi regions which were passed via setup_data. The virt_addr is a
841 * fixed addr which was used in first kernel of a kexec boot.
842 */
843 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
844 md = p;
845 efi_map_region_fixed(md); /* FIXME: add error handling */
846 get_systab_virt_addr(md);
847 }
848
849 save_runtime_map();
850
851 BUG_ON(!efi.systab);
852
853 num_pages = ALIGN(memmap.nr_map * memmap.desc_size, PAGE_SIZE);
854 num_pages >>= PAGE_SHIFT;
855
856 if (efi_setup_page_tables(memmap.phys_map, num_pages)) {
857 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
858 return;
859 }
860
861 efi_sync_low_kernel_mappings();
862
863 /*
864 * Now that EFI is in virtual mode, update the function
865 * pointers in the runtime service table to the new virtual addresses.
866 *
867 * Call EFI services through wrapper functions.
868 */
869 efi.runtime_version = efi_systab.hdr.revision;
870
871 efi_native_runtime_setup();
872
873 efi.set_virtual_address_map = NULL;
874
875 if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
876 runtime_code_page_mkexec();
877
878 /* clean DUMMY object */
879 efi_delete_dummy_variable();
880#endif
881}
882
883/*
884 * This function will switch the EFI runtime services to virtual mode.
885 * Essentially, we look through the EFI memmap and map every region that
886 * has the runtime attribute bit set in its memory descriptor into the
887 * efi_pgd page table.
888 *
889 * The old method which used to update that memory descriptor with the
890 * virtual address obtained from ioremap() is still supported when the
891 * kernel is booted with efi=old_map on its command line. Same old
892 * method enabled the runtime services to be called without having to
893 * thunk back into physical mode for every invocation.
894 *
895 * The new method does a pagetable switch in a preemption-safe manner
896 * so that we're in a different address space when calling a runtime
897 * function. For function arguments passing we do copy the PUDs of the
898 * kernel page table into efi_pgd prior to each call.
899 *
900 * Specially for kexec boot, efi runtime maps in previous kernel should
901 * be passed in via setup_data. In that case runtime ranges will be mapped
902 * to the same virtual addresses as the first kernel, see
903 * kexec_enter_virtual_mode().
904 */
905static void __init __efi_enter_virtual_mode(void)
906{
907 int count = 0, pg_shift = 0;
908 void *new_memmap = NULL;
909 efi_status_t status;
910
911 efi.systab = NULL;
912
913 if (efi_alloc_page_tables()) {
914 pr_err("Failed to allocate EFI page tables\n");
915 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
916 return;
917 }
918
919 efi_merge_regions();
920 new_memmap = efi_map_regions(&count, &pg_shift);
921 if (!new_memmap) {
922 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
923 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
924 return;
925 }
926
927 save_runtime_map();
928
929 BUG_ON(!efi.systab);
930
931 if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) {
932 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
933 return;
934 }
935
936 efi_sync_low_kernel_mappings();
937
938 if (efi_is_native()) {
939 status = phys_efi_set_virtual_address_map(
940 memmap.desc_size * count,
941 memmap.desc_size,
942 memmap.desc_version,
943 (efi_memory_desc_t *)__pa(new_memmap));
944 } else {
945 status = efi_thunk_set_virtual_address_map(
946 efi_phys.set_virtual_address_map,
947 memmap.desc_size * count,
948 memmap.desc_size,
949 memmap.desc_version,
950 (efi_memory_desc_t *)__pa(new_memmap));
951 }
952
953 if (status != EFI_SUCCESS) {
954 pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
955 status);
956 panic("EFI call to SetVirtualAddressMap() failed!");
957 }
958
959 /*
960 * Now that EFI is in virtual mode, update the function
961 * pointers in the runtime service table to the new virtual addresses.
962 *
963 * Call EFI services through wrapper functions.
964 */
965 efi.runtime_version = efi_systab.hdr.revision;
966
967 if (efi_is_native())
968 efi_native_runtime_setup();
969 else
970 efi_thunk_runtime_setup();
971
972 efi.set_virtual_address_map = NULL;
973
974 /*
975 * Apply more restrictive page table mapping attributes now that
976 * SVAM() has been called and the firmware has performed all
977 * necessary relocation fixups for the new virtual addresses.
978 */
979 efi_runtime_update_mappings();
980 efi_dump_pagetable();
981
982 /*
983 * We mapped the descriptor array into the EFI pagetable above
984 * but we're not unmapping it here because if we're running in
985 * EFI mixed mode we need all of memory to be accessible when
986 * we pass parameters to the EFI runtime services in the
987 * thunking code.
988 *
989 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
990 */
991 free_pages((unsigned long)new_memmap, pg_shift);
992
993 /* clean DUMMY object */
994 efi_delete_dummy_variable();
995}
996
997void __init efi_enter_virtual_mode(void)
998{
999 if (efi_enabled(EFI_PARAVIRT))
1000 return;
1001
1002 if (efi_setup)
1003 kexec_enter_virtual_mode();
1004 else
1005 __efi_enter_virtual_mode();
1006}
1007
1008/*
1009 * Convenience functions to obtain memory types and attributes
1010 */
1011u32 efi_mem_type(unsigned long phys_addr)
1012{
1013 efi_memory_desc_t *md;
1014 void *p;
1015
1016 if (!efi_enabled(EFI_MEMMAP))
1017 return 0;
1018
1019 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1020 md = p;
1021 if ((md->phys_addr <= phys_addr) &&
1022 (phys_addr < (md->phys_addr +
1023 (md->num_pages << EFI_PAGE_SHIFT))))
1024 return md->type;
1025 }
1026 return 0;
1027}
1028
1029static int __init arch_parse_efi_cmdline(char *str)
1030{
1031 if (!str) {
1032 pr_warn("need at least one option\n");
1033 return -EINVAL;
1034 }
1035
1036 if (parse_option_str(str, "old_map"))
1037 set_bit(EFI_OLD_MEMMAP, &efi.flags);
1038
1039 return 0;
1040}
1041early_param("efi", arch_parse_efi_cmdline);