xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 7a010c3c)
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 };
97 
98 u64 efi_setup;		/* efi setup_data physical address */
99 
100 static int add_efi_memmap __initdata;
101 static int __init setup_add_efi_memmap(char *arg)
102 {
103 	add_efi_memmap = 1;
104 	return 0;
105 }
106 early_param("add_efi_memmap", setup_add_efi_memmap);
107 
108 void __init efi_find_mirror(void)
109 {
110 	efi_memory_desc_t *md;
111 	u64 mirror_size = 0, total_size = 0;
112 
113 	if (!efi_enabled(EFI_MEMMAP))
114 		return;
115 
116 	for_each_efi_memory_desc(md) {
117 		unsigned long long start = md->phys_addr;
118 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
119 
120 		total_size += size;
121 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
122 			memblock_mark_mirror(start, size);
123 			mirror_size += size;
124 		}
125 	}
126 	if (mirror_size)
127 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
128 			mirror_size>>20, total_size>>20);
129 }
130 
131 /*
132  * Tell the kernel about the EFI memory map.  This might include
133  * more than the max 128 entries that can fit in the passed in e820
134  * legacy (zeropage) memory map, but the kernel's e820 table can hold
135  * E820_MAX_ENTRIES.
136  */
137 
138 static void __init do_add_efi_memmap(void)
139 {
140 	efi_memory_desc_t *md;
141 
142 	if (!efi_enabled(EFI_MEMMAP))
143 		return;
144 
145 	for_each_efi_memory_desc(md) {
146 		unsigned long long start = md->phys_addr;
147 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
148 		int e820_type;
149 
150 		switch (md->type) {
151 		case EFI_LOADER_CODE:
152 		case EFI_LOADER_DATA:
153 		case EFI_BOOT_SERVICES_CODE:
154 		case EFI_BOOT_SERVICES_DATA:
155 		case EFI_CONVENTIONAL_MEMORY:
156 			if (efi_soft_reserve_enabled()
157 			    && (md->attribute & EFI_MEMORY_SP))
158 				e820_type = E820_TYPE_SOFT_RESERVED;
159 			else if (md->attribute & EFI_MEMORY_WB)
160 				e820_type = E820_TYPE_RAM;
161 			else
162 				e820_type = E820_TYPE_RESERVED;
163 			break;
164 		case EFI_ACPI_RECLAIM_MEMORY:
165 			e820_type = E820_TYPE_ACPI;
166 			break;
167 		case EFI_ACPI_MEMORY_NVS:
168 			e820_type = E820_TYPE_NVS;
169 			break;
170 		case EFI_UNUSABLE_MEMORY:
171 			e820_type = E820_TYPE_UNUSABLE;
172 			break;
173 		case EFI_PERSISTENT_MEMORY:
174 			e820_type = E820_TYPE_PMEM;
175 			break;
176 		default:
177 			/*
178 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
179 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
180 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
181 			 */
182 			e820_type = E820_TYPE_RESERVED;
183 			break;
184 		}
185 
186 		e820__range_add(start, size, e820_type);
187 	}
188 	e820__update_table(e820_table);
189 }
190 
191 /*
192  * Given add_efi_memmap defaults to 0 and there there is no alternative
193  * e820 mechanism for soft-reserved memory, import the full EFI memory
194  * map if soft reservations are present and enabled. Otherwise, the
195  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
196  * the efi=nosoftreserve option.
197  */
198 static bool do_efi_soft_reserve(void)
199 {
200 	efi_memory_desc_t *md;
201 
202 	if (!efi_enabled(EFI_MEMMAP))
203 		return false;
204 
205 	if (!efi_soft_reserve_enabled())
206 		return false;
207 
208 	for_each_efi_memory_desc(md)
209 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
210 		    (md->attribute & EFI_MEMORY_SP))
211 			return true;
212 	return false;
213 }
214 
215 int __init efi_memblock_x86_reserve_range(void)
216 {
217 	struct efi_info *e = &boot_params.efi_info;
218 	struct efi_memory_map_data data;
219 	phys_addr_t pmap;
220 	int rv;
221 
222 	if (efi_enabled(EFI_PARAVIRT))
223 		return 0;
224 
225 	/* Can't handle firmware tables above 4GB on i386 */
226 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
227 		pr_err("Memory map is above 4GB, disabling EFI.\n");
228 		return -EINVAL;
229 	}
230 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
231 
232 	data.phys_map		= pmap;
233 	data.size 		= e->efi_memmap_size;
234 	data.desc_size		= e->efi_memdesc_size;
235 	data.desc_version	= e->efi_memdesc_version;
236 
237 	rv = efi_memmap_init_early(&data);
238 	if (rv)
239 		return rv;
240 
241 	if (add_efi_memmap || do_efi_soft_reserve())
242 		do_add_efi_memmap();
243 
244 	efi_fake_memmap_early();
245 
246 	WARN(efi.memmap.desc_version != 1,
247 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
248 	     efi.memmap.desc_version);
249 
250 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
251 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
252 
253 	return 0;
254 }
255 
256 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
257 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
258 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
259 
260 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
261 {
262 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
263 	u64 end_hi = 0;
264 	char buf[64];
265 
266 	if (md->num_pages == 0) {
267 		end = 0;
268 	} else if (md->num_pages > EFI_PAGES_MAX ||
269 		   EFI_PAGES_MAX - md->num_pages <
270 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
271 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
272 			>> OVERFLOW_ADDR_SHIFT;
273 
274 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
275 			end_hi += 1;
276 	} else {
277 		return true;
278 	}
279 
280 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
281 
282 	if (end_hi) {
283 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
284 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
285 			md->phys_addr, end_hi, end);
286 	} else {
287 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
288 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
289 			md->phys_addr, end);
290 	}
291 	return false;
292 }
293 
294 static void __init efi_clean_memmap(void)
295 {
296 	efi_memory_desc_t *out = efi.memmap.map;
297 	const efi_memory_desc_t *in = out;
298 	const efi_memory_desc_t *end = efi.memmap.map_end;
299 	int i, n_removal;
300 
301 	for (i = n_removal = 0; in < end; i++) {
302 		if (efi_memmap_entry_valid(in, i)) {
303 			if (out != in)
304 				memcpy(out, in, efi.memmap.desc_size);
305 			out = (void *)out + efi.memmap.desc_size;
306 		} else {
307 			n_removal++;
308 		}
309 		in = (void *)in + efi.memmap.desc_size;
310 	}
311 
312 	if (n_removal > 0) {
313 		struct efi_memory_map_data data = {
314 			.phys_map	= efi.memmap.phys_map,
315 			.desc_version	= efi.memmap.desc_version,
316 			.desc_size	= efi.memmap.desc_size,
317 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
318 			.flags		= 0,
319 		};
320 
321 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
322 		efi_memmap_install(&data);
323 	}
324 }
325 
326 void __init efi_print_memmap(void)
327 {
328 	efi_memory_desc_t *md;
329 	int i = 0;
330 
331 	for_each_efi_memory_desc(md) {
332 		char buf[64];
333 
334 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
335 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
336 			md->phys_addr,
337 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
338 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
339 	}
340 }
341 
342 static int __init efi_systab_init(unsigned long phys)
343 {
344 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
345 					  : sizeof(efi_system_table_32_t);
346 	const efi_table_hdr_t *hdr;
347 	bool over4g = false;
348 	void *p;
349 	int ret;
350 
351 	hdr = p = early_memremap_ro(phys, size);
352 	if (p == NULL) {
353 		pr_err("Couldn't map the system table!\n");
354 		return -ENOMEM;
355 	}
356 
357 	ret = efi_systab_check_header(hdr, 1);
358 	if (ret) {
359 		early_memunmap(p, size);
360 		return ret;
361 	}
362 
363 	if (efi_enabled(EFI_64BIT)) {
364 		const efi_system_table_64_t *systab64 = p;
365 
366 		efi_runtime	= systab64->runtime;
367 		over4g		= systab64->runtime > U32_MAX;
368 
369 		if (efi_setup) {
370 			struct efi_setup_data *data;
371 
372 			data = early_memremap_ro(efi_setup, sizeof(*data));
373 			if (!data) {
374 				early_memunmap(p, size);
375 				return -ENOMEM;
376 			}
377 
378 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
379 			efi_config_table	= (unsigned long)data->tables;
380 
381 			over4g |= data->fw_vendor	> U32_MAX ||
382 				  data->tables		> U32_MAX;
383 
384 			early_memunmap(data, sizeof(*data));
385 		} else {
386 			efi_fw_vendor		= systab64->fw_vendor;
387 			efi_config_table	= systab64->tables;
388 
389 			over4g |= systab64->fw_vendor	> U32_MAX ||
390 				  systab64->tables	> U32_MAX;
391 		}
392 		efi_nr_tables = systab64->nr_tables;
393 	} else {
394 		const efi_system_table_32_t *systab32 = p;
395 
396 		efi_fw_vendor		= systab32->fw_vendor;
397 		efi_runtime		= systab32->runtime;
398 		efi_config_table	= systab32->tables;
399 		efi_nr_tables		= systab32->nr_tables;
400 	}
401 
402 	efi.runtime_version = hdr->revision;
403 
404 	efi_systab_report_header(hdr, efi_fw_vendor);
405 	early_memunmap(p, size);
406 
407 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
408 		pr_err("EFI data located above 4GB, disabling EFI.\n");
409 		return -EINVAL;
410 	}
411 
412 	return 0;
413 }
414 
415 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
416 {
417 	void *config_tables;
418 	int sz, ret;
419 
420 	if (efi_nr_tables == 0)
421 		return 0;
422 
423 	if (efi_enabled(EFI_64BIT))
424 		sz = sizeof(efi_config_table_64_t);
425 	else
426 		sz = sizeof(efi_config_table_32_t);
427 
428 	/*
429 	 * Let's see what config tables the firmware passed to us.
430 	 */
431 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
432 	if (config_tables == NULL) {
433 		pr_err("Could not map Configuration table!\n");
434 		return -ENOMEM;
435 	}
436 
437 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
438 				      arch_tables);
439 
440 	early_memunmap(config_tables, efi_nr_tables * sz);
441 	return ret;
442 }
443 
444 void __init efi_init(void)
445 {
446 	if (IS_ENABLED(CONFIG_X86_32) &&
447 	    (boot_params.efi_info.efi_systab_hi ||
448 	     boot_params.efi_info.efi_memmap_hi)) {
449 		pr_info("Table located above 4GB, disabling EFI.\n");
450 		return;
451 	}
452 
453 	efi_systab_phys = boot_params.efi_info.efi_systab |
454 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
455 
456 	if (efi_systab_init(efi_systab_phys))
457 		return;
458 
459 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
460 		return;
461 
462 	if (efi_config_init(arch_tables))
463 		return;
464 
465 	/*
466 	 * Note: We currently don't support runtime services on an EFI
467 	 * that doesn't match the kernel 32/64-bit mode.
468 	 */
469 
470 	if (!efi_runtime_supported())
471 		pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
472 
473 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
474 		efi_memmap_unmap();
475 		return;
476 	}
477 
478 	/* Parse the EFI Properties table if it exists */
479 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
480 		efi_properties_table_t *tbl;
481 
482 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
483 		if (tbl == NULL) {
484 			pr_err("Could not map Properties table!\n");
485 		} else {
486 			if (tbl->memory_protection_attribute &
487 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
488 				set_bit(EFI_NX_PE_DATA, &efi.flags);
489 
490 			early_memunmap(tbl, sizeof(*tbl));
491 		}
492 	}
493 
494 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
495 	efi_clean_memmap();
496 
497 	if (efi_enabled(EFI_DBG))
498 		efi_print_memmap();
499 }
500 
501 /* Merge contiguous regions of the same type and attribute */
502 static void __init efi_merge_regions(void)
503 {
504 	efi_memory_desc_t *md, *prev_md = NULL;
505 
506 	for_each_efi_memory_desc(md) {
507 		u64 prev_size;
508 
509 		if (!prev_md) {
510 			prev_md = md;
511 			continue;
512 		}
513 
514 		if (prev_md->type != md->type ||
515 		    prev_md->attribute != md->attribute) {
516 			prev_md = md;
517 			continue;
518 		}
519 
520 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
521 
522 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
523 			prev_md->num_pages += md->num_pages;
524 			md->type = EFI_RESERVED_TYPE;
525 			md->attribute = 0;
526 			continue;
527 		}
528 		prev_md = md;
529 	}
530 }
531 
532 static void *realloc_pages(void *old_memmap, int old_shift)
533 {
534 	void *ret;
535 
536 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
537 	if (!ret)
538 		goto out;
539 
540 	/*
541 	 * A first-time allocation doesn't have anything to copy.
542 	 */
543 	if (!old_memmap)
544 		return ret;
545 
546 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
547 
548 out:
549 	free_pages((unsigned long)old_memmap, old_shift);
550 	return ret;
551 }
552 
553 /*
554  * Iterate the EFI memory map in reverse order because the regions
555  * will be mapped top-down. The end result is the same as if we had
556  * mapped things forward, but doesn't require us to change the
557  * existing implementation of efi_map_region().
558  */
559 static inline void *efi_map_next_entry_reverse(void *entry)
560 {
561 	/* Initial call */
562 	if (!entry)
563 		return efi.memmap.map_end - efi.memmap.desc_size;
564 
565 	entry -= efi.memmap.desc_size;
566 	if (entry < efi.memmap.map)
567 		return NULL;
568 
569 	return entry;
570 }
571 
572 /*
573  * efi_map_next_entry - Return the next EFI memory map descriptor
574  * @entry: Previous EFI memory map descriptor
575  *
576  * This is a helper function to iterate over the EFI memory map, which
577  * we do in different orders depending on the current configuration.
578  *
579  * To begin traversing the memory map @entry must be %NULL.
580  *
581  * Returns %NULL when we reach the end of the memory map.
582  */
583 static void *efi_map_next_entry(void *entry)
584 {
585 	if (efi_enabled(EFI_64BIT)) {
586 		/*
587 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
588 		 * config table feature requires us to map all entries
589 		 * in the same order as they appear in the EFI memory
590 		 * map. That is to say, entry N must have a lower
591 		 * virtual address than entry N+1. This is because the
592 		 * firmware toolchain leaves relative references in
593 		 * the code/data sections, which are split and become
594 		 * separate EFI memory regions. Mapping things
595 		 * out-of-order leads to the firmware accessing
596 		 * unmapped addresses.
597 		 *
598 		 * Since we need to map things this way whether or not
599 		 * the kernel actually makes use of
600 		 * EFI_PROPERTIES_TABLE, let's just switch to this
601 		 * scheme by default for 64-bit.
602 		 */
603 		return efi_map_next_entry_reverse(entry);
604 	}
605 
606 	/* Initial call */
607 	if (!entry)
608 		return efi.memmap.map;
609 
610 	entry += efi.memmap.desc_size;
611 	if (entry >= efi.memmap.map_end)
612 		return NULL;
613 
614 	return entry;
615 }
616 
617 static bool should_map_region(efi_memory_desc_t *md)
618 {
619 	/*
620 	 * Runtime regions always require runtime mappings (obviously).
621 	 */
622 	if (md->attribute & EFI_MEMORY_RUNTIME)
623 		return true;
624 
625 	/*
626 	 * 32-bit EFI doesn't suffer from the bug that requires us to
627 	 * reserve boot services regions, and mixed mode support
628 	 * doesn't exist for 32-bit kernels.
629 	 */
630 	if (IS_ENABLED(CONFIG_X86_32))
631 		return false;
632 
633 	/*
634 	 * EFI specific purpose memory may be reserved by default
635 	 * depending on kernel config and boot options.
636 	 */
637 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
638 	    efi_soft_reserve_enabled() &&
639 	    (md->attribute & EFI_MEMORY_SP))
640 		return false;
641 
642 	/*
643 	 * Map all of RAM so that we can access arguments in the 1:1
644 	 * mapping when making EFI runtime calls.
645 	 */
646 	if (efi_is_mixed()) {
647 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
648 		    md->type == EFI_LOADER_DATA ||
649 		    md->type == EFI_LOADER_CODE)
650 			return true;
651 	}
652 
653 	/*
654 	 * Map boot services regions as a workaround for buggy
655 	 * firmware that accesses them even when they shouldn't.
656 	 *
657 	 * See efi_{reserve,free}_boot_services().
658 	 */
659 	if (md->type == EFI_BOOT_SERVICES_CODE ||
660 	    md->type == EFI_BOOT_SERVICES_DATA)
661 		return true;
662 
663 	return false;
664 }
665 
666 /*
667  * Map the efi memory ranges of the runtime services and update new_mmap with
668  * virtual addresses.
669  */
670 static void * __init efi_map_regions(int *count, int *pg_shift)
671 {
672 	void *p, *new_memmap = NULL;
673 	unsigned long left = 0;
674 	unsigned long desc_size;
675 	efi_memory_desc_t *md;
676 
677 	desc_size = efi.memmap.desc_size;
678 
679 	p = NULL;
680 	while ((p = efi_map_next_entry(p))) {
681 		md = p;
682 
683 		if (!should_map_region(md))
684 			continue;
685 
686 		efi_map_region(md);
687 
688 		if (left < desc_size) {
689 			new_memmap = realloc_pages(new_memmap, *pg_shift);
690 			if (!new_memmap)
691 				return NULL;
692 
693 			left += PAGE_SIZE << *pg_shift;
694 			(*pg_shift)++;
695 		}
696 
697 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
698 
699 		left -= desc_size;
700 		(*count)++;
701 	}
702 
703 	return new_memmap;
704 }
705 
706 static void __init kexec_enter_virtual_mode(void)
707 {
708 #ifdef CONFIG_KEXEC_CORE
709 	efi_memory_desc_t *md;
710 	unsigned int num_pages;
711 
712 	/*
713 	 * We don't do virtual mode, since we don't do runtime services, on
714 	 * non-native EFI.
715 	 */
716 	if (efi_is_mixed()) {
717 		efi_memmap_unmap();
718 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
719 		return;
720 	}
721 
722 	if (efi_alloc_page_tables()) {
723 		pr_err("Failed to allocate EFI page tables\n");
724 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
725 		return;
726 	}
727 
728 	/*
729 	* Map efi regions which were passed via setup_data. The virt_addr is a
730 	* fixed addr which was used in first kernel of a kexec boot.
731 	*/
732 	for_each_efi_memory_desc(md)
733 		efi_map_region_fixed(md); /* FIXME: add error handling */
734 
735 	/*
736 	 * Unregister the early EFI memmap from efi_init() and install
737 	 * the new EFI memory map.
738 	 */
739 	efi_memmap_unmap();
740 
741 	if (efi_memmap_init_late(efi.memmap.phys_map,
742 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
743 		pr_err("Failed to remap late EFI memory map\n");
744 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
745 		return;
746 	}
747 
748 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
749 	num_pages >>= PAGE_SHIFT;
750 
751 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
752 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
753 		return;
754 	}
755 
756 	efi_sync_low_kernel_mappings();
757 	efi_native_runtime_setup();
758 #endif
759 }
760 
761 /*
762  * This function will switch the EFI runtime services to virtual mode.
763  * Essentially, we look through the EFI memmap and map every region that
764  * has the runtime attribute bit set in its memory descriptor into the
765  * efi_pgd page table.
766  *
767  * The new method does a pagetable switch in a preemption-safe manner
768  * so that we're in a different address space when calling a runtime
769  * function. For function arguments passing we do copy the PUDs of the
770  * kernel page table into efi_pgd prior to each call.
771  *
772  * Specially for kexec boot, efi runtime maps in previous kernel should
773  * be passed in via setup_data. In that case runtime ranges will be mapped
774  * to the same virtual addresses as the first kernel, see
775  * kexec_enter_virtual_mode().
776  */
777 static void __init __efi_enter_virtual_mode(void)
778 {
779 	int count = 0, pg_shift = 0;
780 	void *new_memmap = NULL;
781 	efi_status_t status;
782 	unsigned long pa;
783 
784 	if (efi_alloc_page_tables()) {
785 		pr_err("Failed to allocate EFI page tables\n");
786 		goto err;
787 	}
788 
789 	efi_merge_regions();
790 	new_memmap = efi_map_regions(&count, &pg_shift);
791 	if (!new_memmap) {
792 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
793 		goto err;
794 	}
795 
796 	pa = __pa(new_memmap);
797 
798 	/*
799 	 * Unregister the early EFI memmap from efi_init() and install
800 	 * the new EFI memory map that we are about to pass to the
801 	 * firmware via SetVirtualAddressMap().
802 	 */
803 	efi_memmap_unmap();
804 
805 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
806 		pr_err("Failed to remap late EFI memory map\n");
807 		goto err;
808 	}
809 
810 	if (efi_enabled(EFI_DBG)) {
811 		pr_info("EFI runtime memory map:\n");
812 		efi_print_memmap();
813 	}
814 
815 	if (efi_setup_page_tables(pa, 1 << pg_shift))
816 		goto err;
817 
818 	efi_sync_low_kernel_mappings();
819 
820 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
821 					     efi.memmap.desc_size,
822 					     efi.memmap.desc_version,
823 					     (efi_memory_desc_t *)pa,
824 					     efi_systab_phys);
825 	if (status != EFI_SUCCESS) {
826 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
827 		       status);
828 		goto err;
829 	}
830 
831 	efi_check_for_embedded_firmwares();
832 	efi_free_boot_services();
833 
834 	if (!efi_is_mixed())
835 		efi_native_runtime_setup();
836 	else
837 		efi_thunk_runtime_setup();
838 
839 	/*
840 	 * Apply more restrictive page table mapping attributes now that
841 	 * SVAM() has been called and the firmware has performed all
842 	 * necessary relocation fixups for the new virtual addresses.
843 	 */
844 	efi_runtime_update_mappings();
845 
846 	/* clean DUMMY object */
847 	efi_delete_dummy_variable();
848 	return;
849 
850 err:
851 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
852 }
853 
854 void __init efi_enter_virtual_mode(void)
855 {
856 	if (efi_enabled(EFI_PARAVIRT))
857 		return;
858 
859 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
860 
861 	if (efi_setup)
862 		kexec_enter_virtual_mode();
863 	else
864 		__efi_enter_virtual_mode();
865 
866 	efi_dump_pagetable();
867 }
868 
869 bool efi_is_table_address(unsigned long phys_addr)
870 {
871 	unsigned int i;
872 
873 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
874 		return false;
875 
876 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
877 		if (*(efi_tables[i]) == phys_addr)
878 			return true;
879 
880 	return false;
881 }
882 
883 char *efi_systab_show_arch(char *str)
884 {
885 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
886 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
887 	return str;
888 }
889 
890 #define EFI_FIELD(var) efi_ ## var
891 
892 #define EFI_ATTR_SHOW(name) \
893 static ssize_t name##_show(struct kobject *kobj, \
894 				struct kobj_attribute *attr, char *buf) \
895 { \
896 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
897 }
898 
899 EFI_ATTR_SHOW(fw_vendor);
900 EFI_ATTR_SHOW(runtime);
901 EFI_ATTR_SHOW(config_table);
902 
903 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
904 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
905 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
906 
907 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
908 {
909 	if (attr == &efi_attr_fw_vendor.attr) {
910 		if (efi_enabled(EFI_PARAVIRT) ||
911 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
912 			return 0;
913 	} else if (attr == &efi_attr_runtime.attr) {
914 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
915 			return 0;
916 	} else if (attr == &efi_attr_config_table.attr) {
917 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
918 			return 0;
919 	}
920 	return attr->mode;
921 }
922