2 * x86_64 specific EFI support functions
3 * Based on Extensible Firmware Interface Specification version 1.0
5 * Copyright (C) 2005-2008 Intel Co.
6 * Fenghua Yu <fenghua.yu@intel.com>
7 * Bibo Mao <bibo.mao@intel.com>
8 * Chandramouli Narayanan <mouli@linux.intel.com>
9 * Huang Ying <ying.huang@intel.com>
11 * Code to convert EFI to E820 map has been implemented in elilo bootloader
12 * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
13 * is setup appropriately for EFI runtime code.
18 #define pr_fmt(fmt) "efi: " fmt
20 #include <linux/kernel.h>
21 #include <linux/init.h>
23 #include <linux/types.h>
24 #include <linux/spinlock.h>
25 #include <linux/bootmem.h>
26 #include <linux/ioport.h>
27 #include <linux/init.h>
28 #include <linux/mc146818rtc.h>
29 #include <linux/efi.h>
30 #include <linux/uaccess.h>
32 #include <linux/reboot.h>
33 #include <linux/slab.h>
35 #include <asm/setup.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <asm/proto.h>
42 #include <asm/cacheflush.h>
43 #include <asm/fixmap.h>
44 #include <asm/realmode.h>
46 #include <asm/pgalloc.h>
49 * We allocate runtime services regions bottom-up, starting from -4G, i.e.
50 * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
52 static u64 efi_va = EFI_VA_START;
54 struct efi_scratch efi_scratch;
56 static void __init early_code_mapping_set_exec(int executable)
58 efi_memory_desc_t *md;
60 if (!(__supported_pte_mask & _PAGE_NX))
63 /* Make EFI service code area executable */
64 for_each_efi_memory_desc(md) {
65 if (md->type == EFI_RUNTIME_SERVICES_CODE ||
66 md->type == EFI_BOOT_SERVICES_CODE)
67 efi_set_executable(md, executable);
71 pgd_t * __init efi_call_phys_prolog(void)
73 unsigned long vaddress;
79 if (!efi_enabled(EFI_OLD_MEMMAP)) {
80 save_pgd = (pgd_t *)read_cr3();
81 write_cr3((unsigned long)efi_scratch.efi_pgt);
85 early_code_mapping_set_exec(1);
87 n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
88 save_pgd = kmalloc(n_pgds * sizeof(pgd_t), GFP_KERNEL);
90 for (pgd = 0; pgd < n_pgds; pgd++) {
91 save_pgd[pgd] = *pgd_offset_k(pgd * PGDIR_SIZE);
92 vaddress = (unsigned long)__va(pgd * PGDIR_SIZE);
93 set_pgd(pgd_offset_k(pgd * PGDIR_SIZE), *pgd_offset_k(vaddress));
101 void __init efi_call_phys_epilog(pgd_t *save_pgd)
104 * After the lock is released, the original page table is restored.
109 if (!efi_enabled(EFI_OLD_MEMMAP)) {
110 write_cr3((unsigned long)save_pgd);
115 nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
117 for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++)
118 set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
123 early_code_mapping_set_exec(0);
126 static pgd_t *efi_pgd;
129 * We need our own copy of the higher levels of the page tables
130 * because we want to avoid inserting EFI region mappings (EFI_VA_END
131 * to EFI_VA_START) into the standard kernel page tables. Everything
132 * else can be shared, see efi_sync_low_kernel_mappings().
134 int __init efi_alloc_page_tables(void)
140 if (efi_enabled(EFI_OLD_MEMMAP))
143 gfp_mask = GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO;
144 efi_pgd = (pgd_t *)__get_free_page(gfp_mask);
148 pgd = efi_pgd + pgd_index(EFI_VA_END);
150 pud = pud_alloc_one(NULL, 0);
152 free_page((unsigned long)efi_pgd);
156 pgd_populate(NULL, pgd, pud);
162 * Add low kernel mappings for passing arguments to EFI functions.
164 void efi_sync_low_kernel_mappings(void)
166 unsigned num_entries;
167 pgd_t *pgd_k, *pgd_efi;
168 pud_t *pud_k, *pud_efi;
170 if (efi_enabled(EFI_OLD_MEMMAP))
174 * We can share all PGD entries apart from the one entry that
175 * covers the EFI runtime mapping space.
177 * Make sure the EFI runtime region mappings are guaranteed to
178 * only span a single PGD entry and that the entry also maps
179 * other important kernel regions.
181 BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
182 BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
183 (EFI_VA_END & PGDIR_MASK));
185 pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
186 pgd_k = pgd_offset_k(PAGE_OFFSET);
188 num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
189 memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
192 * We share all the PUD entries apart from those that map the
193 * EFI regions. Copy around them.
195 BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
196 BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
198 pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
199 pud_efi = pud_offset(pgd_efi, 0);
201 pgd_k = pgd_offset_k(EFI_VA_END);
202 pud_k = pud_offset(pgd_k, 0);
204 num_entries = pud_index(EFI_VA_END);
205 memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
207 pud_efi = pud_offset(pgd_efi, EFI_VA_START);
208 pud_k = pud_offset(pgd_k, EFI_VA_START);
210 num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
211 memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
214 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
216 unsigned long pfn, text;
217 efi_memory_desc_t *md;
222 if (efi_enabled(EFI_OLD_MEMMAP))
225 efi_scratch.efi_pgt = (pgd_t *)__pa(efi_pgd);
229 * It can happen that the physical address of new_memmap lands in memory
230 * which is not mapped in the EFI page table. Therefore we need to go
231 * and ident-map those pages containing the map before calling
232 * phys_efi_set_virtual_address_map().
234 pfn = pa_memmap >> PAGE_SHIFT;
235 if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, _PAGE_NX | _PAGE_RW)) {
236 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
240 efi_scratch.use_pgd = true;
243 * When making calls to the firmware everything needs to be 1:1
244 * mapped and addressable with 32-bit pointers. Map the kernel
245 * text and allocate a new stack because we can't rely on the
246 * stack pointer being < 4GB.
248 if (!IS_ENABLED(CONFIG_EFI_MIXED))
252 * Map all of RAM so that we can access arguments in the 1:1
253 * mapping when making EFI runtime calls.
255 for_each_efi_memory_desc(md) {
256 if (md->type != EFI_CONVENTIONAL_MEMORY &&
257 md->type != EFI_LOADER_DATA &&
258 md->type != EFI_LOADER_CODE)
261 pfn = md->phys_addr >> PAGE_SHIFT;
262 npages = md->num_pages;
264 if (kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, npages, _PAGE_RW)) {
265 pr_err("Failed to map 1:1 memory\n");
270 page = alloc_page(GFP_KERNEL|__GFP_DMA32);
272 panic("Unable to allocate EFI runtime stack < 4GB\n");
274 efi_scratch.phys_stack = virt_to_phys(page_address(page));
275 efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
277 npages = (_etext - _text) >> PAGE_SHIFT;
279 pfn = text >> PAGE_SHIFT;
281 if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, _PAGE_RW)) {
282 pr_err("Failed to map kernel text 1:1\n");
289 static void __init __map_region(efi_memory_desc_t *md, u64 va)
291 unsigned long flags = _PAGE_RW;
293 pgd_t *pgd = efi_pgd;
295 if (!(md->attribute & EFI_MEMORY_WB))
298 pfn = md->phys_addr >> PAGE_SHIFT;
299 if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
300 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
304 void __init efi_map_region(efi_memory_desc_t *md)
306 unsigned long size = md->num_pages << PAGE_SHIFT;
307 u64 pa = md->phys_addr;
309 if (efi_enabled(EFI_OLD_MEMMAP))
310 return old_map_region(md);
313 * Make sure the 1:1 mappings are present as a catch-all for b0rked
314 * firmware which doesn't update all internal pointers after switching
315 * to virtual mode and would otherwise crap on us.
317 __map_region(md, md->phys_addr);
320 * Enforce the 1:1 mapping as the default virtual address when
321 * booting in EFI mixed mode, because even though we may be
322 * running a 64-bit kernel, the firmware may only be 32-bit.
324 if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
325 md->virt_addr = md->phys_addr;
331 /* Is PA 2M-aligned? */
332 if (!(pa & (PMD_SIZE - 1))) {
335 u64 pa_offset = pa & (PMD_SIZE - 1);
336 u64 prev_va = efi_va;
338 /* get us the same offset within this 2M page */
339 efi_va = (efi_va & PMD_MASK) + pa_offset;
341 if (efi_va > prev_va)
345 if (efi_va < EFI_VA_END) {
346 pr_warn(FW_WARN "VA address range overflow!\n");
351 __map_region(md, efi_va);
352 md->virt_addr = efi_va;
356 * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
357 * md->virt_addr is the original virtual address which had been mapped in kexec
360 void __init efi_map_region_fixed(efi_memory_desc_t *md)
362 __map_region(md, md->virt_addr);
365 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
366 u32 type, u64 attribute)
368 unsigned long last_map_pfn;
370 if (type == EFI_MEMORY_MAPPED_IO)
371 return ioremap(phys_addr, size);
373 last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
374 if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
375 unsigned long top = last_map_pfn << PAGE_SHIFT;
376 efi_ioremap(top, size - (top - phys_addr), type, attribute);
379 if (!(attribute & EFI_MEMORY_WB))
380 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
382 return (void __iomem *)__va(phys_addr);
385 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
387 efi_setup = phys_addr + sizeof(struct setup_data);
390 void __init efi_runtime_update_mappings(void)
393 pgd_t *pgd = efi_pgd;
394 efi_memory_desc_t *md;
396 if (efi_enabled(EFI_OLD_MEMMAP)) {
397 if (__supported_pte_mask & _PAGE_NX)
398 runtime_code_page_mkexec();
402 if (!efi_enabled(EFI_NX_PE_DATA))
405 for_each_efi_memory_desc(md) {
406 unsigned long pf = 0;
408 if (!(md->attribute & EFI_MEMORY_RUNTIME))
411 if (!(md->attribute & EFI_MEMORY_WB))
414 if ((md->attribute & EFI_MEMORY_XP) ||
415 (md->type == EFI_RUNTIME_SERVICES_DATA))
418 if (!(md->attribute & EFI_MEMORY_RO) &&
419 (md->type != EFI_RUNTIME_SERVICES_CODE))
422 /* Update the 1:1 mapping */
423 pfn = md->phys_addr >> PAGE_SHIFT;
424 if (kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf))
425 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
426 md->phys_addr, md->virt_addr);
428 if (kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf))
429 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
430 md->phys_addr, md->virt_addr);
434 void __init efi_dump_pagetable(void)
436 #ifdef CONFIG_EFI_PGT_DUMP
437 ptdump_walk_pgd_level(NULL, efi_pgd);
441 #ifdef CONFIG_EFI_MIXED
442 extern efi_status_t efi64_thunk(u32, ...);
444 #define runtime_service32(func) \
446 u32 table = (u32)(unsigned long)efi.systab; \
449 rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime)); \
450 ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
455 * Switch to the EFI page tables early so that we can access the 1:1
456 * runtime services mappings which are not mapped in any other page
457 * tables. This function must be called before runtime_service32().
459 * Also, disable interrupts because the IDT points to 64-bit handlers,
460 * which aren't going to function correctly when we switch to 32-bit.
462 #define efi_thunk(f, ...) \
465 unsigned long __flags; \
468 local_irq_save(__flags); \
469 arch_efi_call_virt_setup(); \
471 __func = runtime_service32(f); \
472 __s = efi64_thunk(__func, __VA_ARGS__); \
474 arch_efi_call_virt_teardown(); \
475 local_irq_restore(__flags); \
480 efi_status_t efi_thunk_set_virtual_address_map(
481 void *phys_set_virtual_address_map,
482 unsigned long memory_map_size,
483 unsigned long descriptor_size,
484 u32 descriptor_version,
485 efi_memory_desc_t *virtual_map)
491 efi_sync_low_kernel_mappings();
492 local_irq_save(flags);
494 efi_scratch.prev_cr3 = read_cr3();
495 write_cr3((unsigned long)efi_scratch.efi_pgt);
498 func = (u32)(unsigned long)phys_set_virtual_address_map;
499 status = efi64_thunk(func, memory_map_size, descriptor_size,
500 descriptor_version, virtual_map);
502 write_cr3(efi_scratch.prev_cr3);
504 local_irq_restore(flags);
509 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
512 u32 phys_tm, phys_tc;
514 spin_lock(&rtc_lock);
516 phys_tm = virt_to_phys(tm);
517 phys_tc = virt_to_phys(tc);
519 status = efi_thunk(get_time, phys_tm, phys_tc);
521 spin_unlock(&rtc_lock);
526 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
531 spin_lock(&rtc_lock);
533 phys_tm = virt_to_phys(tm);
535 status = efi_thunk(set_time, phys_tm);
537 spin_unlock(&rtc_lock);
543 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
547 u32 phys_enabled, phys_pending, phys_tm;
549 spin_lock(&rtc_lock);
551 phys_enabled = virt_to_phys(enabled);
552 phys_pending = virt_to_phys(pending);
553 phys_tm = virt_to_phys(tm);
555 status = efi_thunk(get_wakeup_time, phys_enabled,
556 phys_pending, phys_tm);
558 spin_unlock(&rtc_lock);
564 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
569 spin_lock(&rtc_lock);
571 phys_tm = virt_to_phys(tm);
573 status = efi_thunk(set_wakeup_time, enabled, phys_tm);
575 spin_unlock(&rtc_lock);
582 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
583 u32 *attr, unsigned long *data_size, void *data)
586 u32 phys_name, phys_vendor, phys_attr;
587 u32 phys_data_size, phys_data;
589 phys_data_size = virt_to_phys(data_size);
590 phys_vendor = virt_to_phys(vendor);
591 phys_name = virt_to_phys(name);
592 phys_attr = virt_to_phys(attr);
593 phys_data = virt_to_phys(data);
595 status = efi_thunk(get_variable, phys_name, phys_vendor,
596 phys_attr, phys_data_size, phys_data);
602 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
603 u32 attr, unsigned long data_size, void *data)
605 u32 phys_name, phys_vendor, phys_data;
608 phys_name = virt_to_phys(name);
609 phys_vendor = virt_to_phys(vendor);
610 phys_data = virt_to_phys(data);
612 /* If data_size is > sizeof(u32) we've got problems */
613 status = efi_thunk(set_variable, phys_name, phys_vendor,
614 attr, data_size, phys_data);
620 efi_thunk_get_next_variable(unsigned long *name_size,
625 u32 phys_name_size, phys_name, phys_vendor;
627 phys_name_size = virt_to_phys(name_size);
628 phys_vendor = virt_to_phys(vendor);
629 phys_name = virt_to_phys(name);
631 status = efi_thunk(get_next_variable, phys_name_size,
632 phys_name, phys_vendor);
638 efi_thunk_get_next_high_mono_count(u32 *count)
643 phys_count = virt_to_phys(count);
644 status = efi_thunk(get_next_high_mono_count, phys_count);
650 efi_thunk_reset_system(int reset_type, efi_status_t status,
651 unsigned long data_size, efi_char16_t *data)
655 phys_data = virt_to_phys(data);
657 efi_thunk(reset_system, reset_type, status, data_size, phys_data);
661 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
662 unsigned long count, unsigned long sg_list)
665 * To properly support this function we would need to repackage
666 * 'capsules' because the firmware doesn't understand 64-bit
669 return EFI_UNSUPPORTED;
673 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
674 u64 *remaining_space,
675 u64 *max_variable_size)
678 u32 phys_storage, phys_remaining, phys_max;
680 if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
681 return EFI_UNSUPPORTED;
683 phys_storage = virt_to_phys(storage_space);
684 phys_remaining = virt_to_phys(remaining_space);
685 phys_max = virt_to_phys(max_variable_size);
687 status = efi_thunk(query_variable_info, attr, phys_storage,
688 phys_remaining, phys_max);
694 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
695 unsigned long count, u64 *max_size,
699 * To properly support this function we would need to repackage
700 * 'capsules' because the firmware doesn't understand 64-bit
703 return EFI_UNSUPPORTED;
706 void efi_thunk_runtime_setup(void)
708 efi.get_time = efi_thunk_get_time;
709 efi.set_time = efi_thunk_set_time;
710 efi.get_wakeup_time = efi_thunk_get_wakeup_time;
711 efi.set_wakeup_time = efi_thunk_set_wakeup_time;
712 efi.get_variable = efi_thunk_get_variable;
713 efi.get_next_variable = efi_thunk_get_next_variable;
714 efi.set_variable = efi_thunk_set_variable;
715 efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
716 efi.reset_system = efi_thunk_reset_system;
717 efi.query_variable_info = efi_thunk_query_variable_info;
718 efi.update_capsule = efi_thunk_update_capsule;
719 efi.query_capsule_caps = efi_thunk_query_capsule_caps;
721 #endif /* CONFIG_EFI_MIXED */