xref: /linux/arch/x86/platform/efi/efi.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
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 
56 static unsigned long efi_systab_phys __initdata;
57 static unsigned long prop_phys = EFI_INVALID_TABLE_ADDR;
58 static unsigned long uga_phys = EFI_INVALID_TABLE_ADDR;
59 static unsigned long efi_runtime, efi_nr_tables;
60 
61 unsigned long efi_fw_vendor, efi_config_table;
62 
63 static 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 
72 static 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 #ifdef CONFIG_UNACCEPTED_MEMORY
100 	&efi.unaccepted,
101 #endif
102 };
103 
104 u64 efi_setup;		/* efi setup_data physical address */
105 
106 static int add_efi_memmap __initdata;
setup_add_efi_memmap(char * arg)107 static int __init setup_add_efi_memmap(char *arg)
108 {
109 	add_efi_memmap = 1;
110 	return 0;
111 }
112 early_param("add_efi_memmap", setup_add_efi_memmap);
113 
114 /*
115  * Tell the kernel about the EFI memory map.  This might include
116  * more than the max 128 entries that can fit in the passed in e820
117  * legacy (zeropage) memory map, but the kernel's e820 table can hold
118  * E820_MAX_ENTRIES.
119  */
120 
do_add_efi_memmap(void)121 static void __init do_add_efi_memmap(void)
122 {
123 	efi_memory_desc_t *md;
124 
125 	if (!efi_enabled(EFI_MEMMAP))
126 		return;
127 
128 	for_each_efi_memory_desc(md) {
129 		unsigned long long start = md->phys_addr;
130 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
131 		int e820_type;
132 
133 		switch (md->type) {
134 		case EFI_LOADER_CODE:
135 		case EFI_LOADER_DATA:
136 		case EFI_BOOT_SERVICES_CODE:
137 		case EFI_BOOT_SERVICES_DATA:
138 		case EFI_CONVENTIONAL_MEMORY:
139 			if (efi_soft_reserve_enabled()
140 			    && (md->attribute & EFI_MEMORY_SP))
141 				e820_type = E820_TYPE_SOFT_RESERVED;
142 			else if (md->attribute & EFI_MEMORY_WB)
143 				e820_type = E820_TYPE_RAM;
144 			else
145 				e820_type = E820_TYPE_RESERVED;
146 			break;
147 		case EFI_ACPI_RECLAIM_MEMORY:
148 			e820_type = E820_TYPE_ACPI;
149 			break;
150 		case EFI_ACPI_MEMORY_NVS:
151 			e820_type = E820_TYPE_NVS;
152 			break;
153 		case EFI_UNUSABLE_MEMORY:
154 			e820_type = E820_TYPE_UNUSABLE;
155 			break;
156 		case EFI_PERSISTENT_MEMORY:
157 			e820_type = E820_TYPE_PMEM;
158 			break;
159 		default:
160 			/*
161 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
162 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
163 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
164 			 */
165 			e820_type = E820_TYPE_RESERVED;
166 			break;
167 		}
168 
169 		e820__range_add(start, size, e820_type);
170 	}
171 	e820__update_table(e820_table);
172 }
173 
174 /*
175  * Given add_efi_memmap defaults to 0 and there is no alternative
176  * e820 mechanism for soft-reserved memory, import the full EFI memory
177  * map if soft reservations are present and enabled. Otherwise, the
178  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
179  * the efi=nosoftreserve option.
180  */
do_efi_soft_reserve(void)181 static bool do_efi_soft_reserve(void)
182 {
183 	efi_memory_desc_t *md;
184 
185 	if (!efi_enabled(EFI_MEMMAP))
186 		return false;
187 
188 	if (!efi_soft_reserve_enabled())
189 		return false;
190 
191 	for_each_efi_memory_desc(md)
192 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
193 		    (md->attribute & EFI_MEMORY_SP))
194 			return true;
195 	return false;
196 }
197 
efi_memblock_x86_reserve_range(void)198 int __init efi_memblock_x86_reserve_range(void)
199 {
200 	struct efi_info *e = &boot_params.efi_info;
201 	struct efi_memory_map_data data;
202 	phys_addr_t pmap;
203 	int rv;
204 
205 	if (efi_enabled(EFI_PARAVIRT))
206 		return 0;
207 
208 	/* Can't handle firmware tables above 4GB on i386 */
209 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
210 		pr_err("Memory map is above 4GB, disabling EFI.\n");
211 		return -EINVAL;
212 	}
213 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
214 
215 	data.phys_map		= pmap;
216 	data.size 		= e->efi_memmap_size;
217 	data.desc_size		= e->efi_memdesc_size;
218 	data.desc_version	= e->efi_memdesc_version;
219 
220 	if (!efi_enabled(EFI_PARAVIRT)) {
221 		rv = efi_memmap_init_early(&data);
222 		if (rv)
223 			return rv;
224 	}
225 
226 	if (add_efi_memmap || do_efi_soft_reserve())
227 		do_add_efi_memmap();
228 
229 	WARN(efi.memmap.desc_version != 1,
230 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
231 	     efi.memmap.desc_version);
232 
233 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
234 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
235 
236 	return 0;
237 }
238 
239 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
240 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
241 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
242 
efi_memmap_entry_valid(const efi_memory_desc_t * md,int i)243 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
244 {
245 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
246 	u64 end_hi = 0;
247 	char buf[64];
248 
249 	if (md->num_pages == 0) {
250 		end = 0;
251 	} else if (md->num_pages > EFI_PAGES_MAX ||
252 		   EFI_PAGES_MAX - md->num_pages <
253 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
254 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
255 			>> OVERFLOW_ADDR_SHIFT;
256 
257 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
258 			end_hi += 1;
259 	} else {
260 		return true;
261 	}
262 
263 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
264 
265 	if (end_hi) {
266 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
267 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
268 			md->phys_addr, end_hi, end);
269 	} else {
270 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
271 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
272 			md->phys_addr, end);
273 	}
274 	return false;
275 }
276 
efi_clean_memmap(void)277 static void __init efi_clean_memmap(void)
278 {
279 	efi_memory_desc_t *out = efi.memmap.map;
280 	const efi_memory_desc_t *in = out;
281 	const efi_memory_desc_t *end = efi.memmap.map_end;
282 	int i, n_removal;
283 
284 	for (i = n_removal = 0; in < end; i++) {
285 		if (efi_memmap_entry_valid(in, i)) {
286 			if (out != in)
287 				memcpy(out, in, efi.memmap.desc_size);
288 			out = (void *)out + efi.memmap.desc_size;
289 		} else {
290 			n_removal++;
291 		}
292 		in = (void *)in + efi.memmap.desc_size;
293 	}
294 
295 	if (n_removal > 0) {
296 		struct efi_memory_map_data data = {
297 			.phys_map	= efi.memmap.phys_map,
298 			.desc_version	= efi.memmap.desc_version,
299 			.desc_size	= efi.memmap.desc_size,
300 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
301 			.flags		= 0,
302 		};
303 
304 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
305 		efi_memmap_install(&data);
306 	}
307 }
308 
309 /*
310  * Firmware can use EfiMemoryMappedIO to request that MMIO regions be
311  * mapped by the OS so they can be accessed by EFI runtime services, but
312  * should have no other significance to the OS (UEFI r2.10, sec 7.2).
313  * However, most bootloaders and EFI stubs convert EfiMemoryMappedIO
314  * regions to E820_TYPE_RESERVED entries, which prevent Linux from
315  * allocating space from them (see remove_e820_regions()).
316  *
317  * Some platforms use EfiMemoryMappedIO entries for PCI MMCONFIG space and
318  * PCI host bridge windows, which means Linux can't allocate BAR space for
319  * hot-added devices.
320  *
321  * Remove large EfiMemoryMappedIO regions from the E820 map to avoid this
322  * problem.
323  *
324  * Retain small EfiMemoryMappedIO regions because on some platforms, these
325  * describe non-window space that's included in host bridge _CRS.  If we
326  * assign that space to PCI devices, they don't work.
327  */
efi_remove_e820_mmio(void)328 static void __init efi_remove_e820_mmio(void)
329 {
330 	efi_memory_desc_t *md;
331 	u64 size, start, end;
332 	int i = 0;
333 
334 	for_each_efi_memory_desc(md) {
335 		if (md->type == EFI_MEMORY_MAPPED_IO) {
336 			size = md->num_pages << EFI_PAGE_SHIFT;
337 			start = md->phys_addr;
338 			end = start + size - 1;
339 			if (size >= 256*1024) {
340 				pr_info("Remove mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluMB) from e820 map\n",
341 					i, start, end, size >> 20);
342 				e820__range_remove(start, size,
343 						   E820_TYPE_RESERVED, 1);
344 			} else {
345 				pr_info("Not removing mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluKB) from e820 map\n",
346 					i, start, end, size >> 10);
347 			}
348 		}
349 		i++;
350 	}
351 }
352 
efi_print_memmap(void)353 void __init efi_print_memmap(void)
354 {
355 	efi_memory_desc_t *md;
356 	int i = 0;
357 
358 	for_each_efi_memory_desc(md) {
359 		char buf[64];
360 
361 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
362 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
363 			md->phys_addr,
364 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
365 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
366 	}
367 }
368 
efi_systab_init(unsigned long phys)369 static int __init efi_systab_init(unsigned long phys)
370 {
371 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
372 					  : sizeof(efi_system_table_32_t);
373 	const efi_table_hdr_t *hdr;
374 	bool over4g = false;
375 	void *p;
376 	int ret;
377 
378 	hdr = p = early_memremap_ro(phys, size);
379 	if (p == NULL) {
380 		pr_err("Couldn't map the system table!\n");
381 		return -ENOMEM;
382 	}
383 
384 	ret = efi_systab_check_header(hdr);
385 	if (ret) {
386 		early_memunmap(p, size);
387 		return ret;
388 	}
389 
390 	if (efi_enabled(EFI_64BIT)) {
391 		const efi_system_table_64_t *systab64 = p;
392 
393 		efi_runtime	= systab64->runtime;
394 		over4g		= systab64->runtime > U32_MAX;
395 
396 		if (efi_setup) {
397 			struct efi_setup_data *data;
398 
399 			data = early_memremap_ro(efi_setup, sizeof(*data));
400 			if (!data) {
401 				early_memunmap(p, size);
402 				return -ENOMEM;
403 			}
404 
405 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
406 			efi_config_table	= (unsigned long)data->tables;
407 
408 			over4g |= data->fw_vendor	> U32_MAX ||
409 				  data->tables		> U32_MAX;
410 
411 			early_memunmap(data, sizeof(*data));
412 		} else {
413 			efi_fw_vendor		= systab64->fw_vendor;
414 			efi_config_table	= systab64->tables;
415 
416 			over4g |= systab64->fw_vendor	> U32_MAX ||
417 				  systab64->tables	> U32_MAX;
418 		}
419 		efi_nr_tables = systab64->nr_tables;
420 	} else {
421 		const efi_system_table_32_t *systab32 = p;
422 
423 		efi_fw_vendor		= systab32->fw_vendor;
424 		efi_runtime		= systab32->runtime;
425 		efi_config_table	= systab32->tables;
426 		efi_nr_tables		= systab32->nr_tables;
427 	}
428 
429 	efi.runtime_version = hdr->revision;
430 
431 	efi_systab_report_header(hdr, efi_fw_vendor);
432 	early_memunmap(p, size);
433 
434 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
435 		pr_err("EFI data located above 4GB, disabling EFI.\n");
436 		return -EINVAL;
437 	}
438 
439 	return 0;
440 }
441 
efi_config_init(const efi_config_table_type_t * arch_tables)442 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
443 {
444 	void *config_tables;
445 	int sz, ret;
446 
447 	if (efi_nr_tables == 0)
448 		return 0;
449 
450 	if (efi_enabled(EFI_64BIT))
451 		sz = sizeof(efi_config_table_64_t);
452 	else
453 		sz = sizeof(efi_config_table_32_t);
454 
455 	/*
456 	 * Let's see what config tables the firmware passed to us.
457 	 */
458 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
459 	if (config_tables == NULL) {
460 		pr_err("Could not map Configuration table!\n");
461 		return -ENOMEM;
462 	}
463 
464 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
465 				      arch_tables);
466 
467 	early_memunmap(config_tables, efi_nr_tables * sz);
468 	return ret;
469 }
470 
efi_init(void)471 void __init efi_init(void)
472 {
473 	if (IS_ENABLED(CONFIG_X86_32) &&
474 	    (boot_params.efi_info.efi_systab_hi ||
475 	     boot_params.efi_info.efi_memmap_hi)) {
476 		pr_info("Table located above 4GB, disabling EFI.\n");
477 		return;
478 	}
479 
480 	efi_systab_phys = boot_params.efi_info.efi_systab |
481 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
482 
483 	if (efi_systab_init(efi_systab_phys))
484 		return;
485 
486 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
487 		return;
488 
489 	if (efi_config_init(arch_tables))
490 		return;
491 
492 	/*
493 	 * Note: We currently don't support runtime services on an EFI
494 	 * that doesn't match the kernel 32/64-bit mode.
495 	 */
496 
497 	if (!efi_runtime_supported())
498 		pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
499 
500 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
501 		efi_memmap_unmap();
502 		return;
503 	}
504 
505 	/* Parse the EFI Properties table if it exists */
506 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
507 		efi_properties_table_t *tbl;
508 
509 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
510 		if (tbl == NULL) {
511 			pr_err("Could not map Properties table!\n");
512 		} else {
513 			if (tbl->memory_protection_attribute &
514 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
515 				set_bit(EFI_NX_PE_DATA, &efi.flags);
516 
517 			early_memunmap(tbl, sizeof(*tbl));
518 		}
519 	}
520 
521 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
522 	efi_clean_memmap();
523 
524 	efi_remove_e820_mmio();
525 
526 	if (efi_enabled(EFI_DBG))
527 		efi_print_memmap();
528 }
529 
530 /* Merge contiguous regions of the same type and attribute */
efi_merge_regions(void)531 static void __init efi_merge_regions(void)
532 {
533 	efi_memory_desc_t *md, *prev_md = NULL;
534 
535 	for_each_efi_memory_desc(md) {
536 		u64 prev_size;
537 
538 		if (!prev_md) {
539 			prev_md = md;
540 			continue;
541 		}
542 
543 		if (prev_md->type != md->type ||
544 		    prev_md->attribute != md->attribute) {
545 			prev_md = md;
546 			continue;
547 		}
548 
549 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
550 
551 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
552 			prev_md->num_pages += md->num_pages;
553 			md->type = EFI_RESERVED_TYPE;
554 			md->attribute = 0;
555 			continue;
556 		}
557 		prev_md = md;
558 	}
559 }
560 
realloc_pages(void * old_memmap,int old_shift)561 static void *realloc_pages(void *old_memmap, int old_shift)
562 {
563 	void *ret;
564 
565 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
566 	if (!ret)
567 		goto out;
568 
569 	/*
570 	 * A first-time allocation doesn't have anything to copy.
571 	 */
572 	if (!old_memmap)
573 		return ret;
574 
575 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
576 
577 out:
578 	free_pages((unsigned long)old_memmap, old_shift);
579 	return ret;
580 }
581 
582 /*
583  * Iterate the EFI memory map in reverse order because the regions
584  * will be mapped top-down. The end result is the same as if we had
585  * mapped things forward, but doesn't require us to change the
586  * existing implementation of efi_map_region().
587  */
efi_map_next_entry_reverse(void * entry)588 static inline void *efi_map_next_entry_reverse(void *entry)
589 {
590 	/* Initial call */
591 	if (!entry)
592 		return efi.memmap.map_end - efi.memmap.desc_size;
593 
594 	entry -= efi.memmap.desc_size;
595 	if (entry < efi.memmap.map)
596 		return NULL;
597 
598 	return entry;
599 }
600 
601 /*
602  * efi_map_next_entry - Return the next EFI memory map descriptor
603  * @entry: Previous EFI memory map descriptor
604  *
605  * This is a helper function to iterate over the EFI memory map, which
606  * we do in different orders depending on the current configuration.
607  *
608  * To begin traversing the memory map @entry must be %NULL.
609  *
610  * Returns %NULL when we reach the end of the memory map.
611  */
efi_map_next_entry(void * entry)612 static void *efi_map_next_entry(void *entry)
613 {
614 	if (efi_enabled(EFI_64BIT)) {
615 		/*
616 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
617 		 * config table feature requires us to map all entries
618 		 * in the same order as they appear in the EFI memory
619 		 * map. That is to say, entry N must have a lower
620 		 * virtual address than entry N+1. This is because the
621 		 * firmware toolchain leaves relative references in
622 		 * the code/data sections, which are split and become
623 		 * separate EFI memory regions. Mapping things
624 		 * out-of-order leads to the firmware accessing
625 		 * unmapped addresses.
626 		 *
627 		 * Since we need to map things this way whether or not
628 		 * the kernel actually makes use of
629 		 * EFI_PROPERTIES_TABLE, let's just switch to this
630 		 * scheme by default for 64-bit.
631 		 */
632 		return efi_map_next_entry_reverse(entry);
633 	}
634 
635 	/* Initial call */
636 	if (!entry)
637 		return efi.memmap.map;
638 
639 	entry += efi.memmap.desc_size;
640 	if (entry >= efi.memmap.map_end)
641 		return NULL;
642 
643 	return entry;
644 }
645 
should_map_region(efi_memory_desc_t * md)646 static bool should_map_region(efi_memory_desc_t *md)
647 {
648 	/*
649 	 * Runtime regions always require runtime mappings (obviously).
650 	 */
651 	if (md->attribute & EFI_MEMORY_RUNTIME)
652 		return true;
653 
654 	/*
655 	 * 32-bit EFI doesn't suffer from the bug that requires us to
656 	 * reserve boot services regions, and mixed mode support
657 	 * doesn't exist for 32-bit kernels.
658 	 */
659 	if (IS_ENABLED(CONFIG_X86_32))
660 		return false;
661 
662 	/*
663 	 * EFI specific purpose memory may be reserved by default
664 	 * depending on kernel config and boot options.
665 	 */
666 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
667 	    efi_soft_reserve_enabled() &&
668 	    (md->attribute & EFI_MEMORY_SP))
669 		return false;
670 
671 	/*
672 	 * Map all of RAM so that we can access arguments in the 1:1
673 	 * mapping when making EFI runtime calls.
674 	 */
675 	if (efi_is_mixed()) {
676 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
677 		    md->type == EFI_LOADER_DATA ||
678 		    md->type == EFI_LOADER_CODE)
679 			return true;
680 	}
681 
682 	/*
683 	 * Map boot services regions as a workaround for buggy
684 	 * firmware that accesses them even when they shouldn't.
685 	 *
686 	 * See efi_{reserve,free}_boot_services().
687 	 */
688 	if (md->type == EFI_BOOT_SERVICES_CODE ||
689 	    md->type == EFI_BOOT_SERVICES_DATA)
690 		return true;
691 
692 	return false;
693 }
694 
695 /*
696  * Map the efi memory ranges of the runtime services and update new_mmap with
697  * virtual addresses.
698  */
efi_map_regions(int * count,int * pg_shift)699 static void * __init efi_map_regions(int *count, int *pg_shift)
700 {
701 	void *p, *new_memmap = NULL;
702 	unsigned long left = 0;
703 	unsigned long desc_size;
704 	efi_memory_desc_t *md;
705 
706 	desc_size = efi.memmap.desc_size;
707 
708 	p = NULL;
709 	while ((p = efi_map_next_entry(p))) {
710 		md = p;
711 
712 		if (!should_map_region(md))
713 			continue;
714 
715 		efi_map_region(md);
716 
717 		if (left < desc_size) {
718 			new_memmap = realloc_pages(new_memmap, *pg_shift);
719 			if (!new_memmap)
720 				return NULL;
721 
722 			left += PAGE_SIZE << *pg_shift;
723 			(*pg_shift)++;
724 		}
725 
726 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
727 
728 		left -= desc_size;
729 		(*count)++;
730 	}
731 
732 	return new_memmap;
733 }
734 
kexec_enter_virtual_mode(void)735 static void __init kexec_enter_virtual_mode(void)
736 {
737 #ifdef CONFIG_KEXEC_CORE
738 	efi_memory_desc_t *md;
739 	unsigned int num_pages;
740 
741 	/*
742 	 * We don't do virtual mode, since we don't do runtime services, on
743 	 * non-native EFI.
744 	 */
745 	if (efi_is_mixed()) {
746 		efi_memmap_unmap();
747 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
748 		return;
749 	}
750 
751 	if (efi_alloc_page_tables()) {
752 		pr_err("Failed to allocate EFI page tables\n");
753 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
754 		return;
755 	}
756 
757 	/*
758 	* Map efi regions which were passed via setup_data. The virt_addr is a
759 	* fixed addr which was used in first kernel of a kexec boot.
760 	*/
761 	for_each_efi_memory_desc(md)
762 		efi_map_region_fixed(md); /* FIXME: add error handling */
763 
764 	/*
765 	 * Unregister the early EFI memmap from efi_init() and install
766 	 * the new EFI memory map.
767 	 */
768 	efi_memmap_unmap();
769 
770 	if (efi_memmap_init_late(efi.memmap.phys_map,
771 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
772 		pr_err("Failed to remap late EFI memory map\n");
773 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
774 		return;
775 	}
776 
777 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
778 	num_pages >>= PAGE_SHIFT;
779 
780 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
781 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
782 		return;
783 	}
784 
785 	efi_sync_low_kernel_mappings();
786 	efi_native_runtime_setup();
787 #endif
788 }
789 
790 /*
791  * This function will switch the EFI runtime services to virtual mode.
792  * Essentially, we look through the EFI memmap and map every region that
793  * has the runtime attribute bit set in its memory descriptor into the
794  * efi_pgd page table.
795  *
796  * The new method does a pagetable switch in a preemption-safe manner
797  * so that we're in a different address space when calling a runtime
798  * function. For function arguments passing we do copy the PUDs of the
799  * kernel page table into efi_pgd prior to each call.
800  *
801  * Specially for kexec boot, efi runtime maps in previous kernel should
802  * be passed in via setup_data. In that case runtime ranges will be mapped
803  * to the same virtual addresses as the first kernel, see
804  * kexec_enter_virtual_mode().
805  */
__efi_enter_virtual_mode(void)806 static void __init __efi_enter_virtual_mode(void)
807 {
808 	int count = 0, pg_shift = 0;
809 	void *new_memmap = NULL;
810 	efi_status_t status;
811 	unsigned long pa;
812 
813 	if (efi_alloc_page_tables()) {
814 		pr_err("Failed to allocate EFI page tables\n");
815 		goto err;
816 	}
817 
818 	efi_merge_regions();
819 	new_memmap = efi_map_regions(&count, &pg_shift);
820 	if (!new_memmap) {
821 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
822 		goto err;
823 	}
824 
825 	pa = __pa(new_memmap);
826 
827 	/*
828 	 * Unregister the early EFI memmap from efi_init() and install
829 	 * the new EFI memory map that we are about to pass to the
830 	 * firmware via SetVirtualAddressMap().
831 	 */
832 	efi_memmap_unmap();
833 
834 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
835 		pr_err("Failed to remap late EFI memory map\n");
836 		goto err;
837 	}
838 
839 	if (efi_enabled(EFI_DBG)) {
840 		pr_info("EFI runtime memory map:\n");
841 		efi_print_memmap();
842 	}
843 
844 	if (efi_setup_page_tables(pa, 1 << pg_shift))
845 		goto err;
846 
847 	efi_sync_low_kernel_mappings();
848 
849 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
850 					     efi.memmap.desc_size,
851 					     efi.memmap.desc_version,
852 					     (efi_memory_desc_t *)pa,
853 					     efi_systab_phys);
854 	if (status != EFI_SUCCESS) {
855 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
856 		       status);
857 		goto err;
858 	}
859 
860 	efi_check_for_embedded_firmwares();
861 	efi_free_boot_services();
862 
863 	if (!efi_is_mixed())
864 		efi_native_runtime_setup();
865 	else
866 		efi_thunk_runtime_setup();
867 
868 	/*
869 	 * Apply more restrictive page table mapping attributes now that
870 	 * SVAM() has been called and the firmware has performed all
871 	 * necessary relocation fixups for the new virtual addresses.
872 	 */
873 	efi_runtime_update_mappings();
874 
875 	/* clean DUMMY object */
876 	efi_delete_dummy_variable();
877 	return;
878 
879 err:
880 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
881 }
882 
efi_enter_virtual_mode(void)883 void __init efi_enter_virtual_mode(void)
884 {
885 	if (efi_enabled(EFI_PARAVIRT))
886 		return;
887 
888 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
889 
890 	if (efi_setup)
891 		kexec_enter_virtual_mode();
892 	else
893 		__efi_enter_virtual_mode();
894 
895 	efi_dump_pagetable();
896 }
897 
efi_is_table_address(unsigned long phys_addr)898 bool efi_is_table_address(unsigned long phys_addr)
899 {
900 	unsigned int i;
901 
902 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
903 		return false;
904 
905 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
906 		if (*(efi_tables[i]) == phys_addr)
907 			return true;
908 
909 	return false;
910 }
911 
efi_systab_show_arch(char * str)912 char *efi_systab_show_arch(char *str)
913 {
914 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
915 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
916 	return str;
917 }
918 
919 #define EFI_FIELD(var) efi_ ## var
920 
921 #define EFI_ATTR_SHOW(name) \
922 static ssize_t name##_show(struct kobject *kobj, \
923 				struct kobj_attribute *attr, char *buf) \
924 { \
925 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
926 }
927 
928 EFI_ATTR_SHOW(fw_vendor);
929 EFI_ATTR_SHOW(runtime);
930 EFI_ATTR_SHOW(config_table);
931 
932 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
933 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
934 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
935 
efi_attr_is_visible(struct kobject * kobj,struct attribute * attr,int n)936 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
937 {
938 	if (attr == &efi_attr_fw_vendor.attr) {
939 		if (efi_enabled(EFI_PARAVIRT) ||
940 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
941 			return 0;
942 	} else if (attr == &efi_attr_runtime.attr) {
943 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
944 			return 0;
945 	} else if (attr == &efi_attr_config_table.attr) {
946 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
947 			return 0;
948 	}
949 	return attr->mode;
950 }
951 
__x86_ima_efi_boot_mode(void)952 enum efi_secureboot_mode __x86_ima_efi_boot_mode(void)
953 {
954 	return boot_params.secure_boot;
955 }
956