xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision f97769fd)
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 };
94 
95 u64 efi_setup;		/* efi setup_data physical address */
96 
97 static int add_efi_memmap __initdata;
98 static int __init setup_add_efi_memmap(char *arg)
99 {
100 	add_efi_memmap = 1;
101 	return 0;
102 }
103 early_param("add_efi_memmap", setup_add_efi_memmap);
104 
105 void __init efi_find_mirror(void)
106 {
107 	efi_memory_desc_t *md;
108 	u64 mirror_size = 0, total_size = 0;
109 
110 	if (!efi_enabled(EFI_MEMMAP))
111 		return;
112 
113 	for_each_efi_memory_desc(md) {
114 		unsigned long long start = md->phys_addr;
115 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
116 
117 		total_size += size;
118 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
119 			memblock_mark_mirror(start, size);
120 			mirror_size += size;
121 		}
122 	}
123 	if (mirror_size)
124 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
125 			mirror_size>>20, total_size>>20);
126 }
127 
128 /*
129  * Tell the kernel about the EFI memory map.  This might include
130  * more than the max 128 entries that can fit in the passed in e820
131  * legacy (zeropage) memory map, but the kernel's e820 table can hold
132  * E820_MAX_ENTRIES.
133  */
134 
135 static void __init do_add_efi_memmap(void)
136 {
137 	efi_memory_desc_t *md;
138 
139 	if (!efi_enabled(EFI_MEMMAP))
140 		return;
141 
142 	for_each_efi_memory_desc(md) {
143 		unsigned long long start = md->phys_addr;
144 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
145 		int e820_type;
146 
147 		switch (md->type) {
148 		case EFI_LOADER_CODE:
149 		case EFI_LOADER_DATA:
150 		case EFI_BOOT_SERVICES_CODE:
151 		case EFI_BOOT_SERVICES_DATA:
152 		case EFI_CONVENTIONAL_MEMORY:
153 			if (efi_soft_reserve_enabled()
154 			    && (md->attribute & EFI_MEMORY_SP))
155 				e820_type = E820_TYPE_SOFT_RESERVED;
156 			else if (md->attribute & EFI_MEMORY_WB)
157 				e820_type = E820_TYPE_RAM;
158 			else
159 				e820_type = E820_TYPE_RESERVED;
160 			break;
161 		case EFI_ACPI_RECLAIM_MEMORY:
162 			e820_type = E820_TYPE_ACPI;
163 			break;
164 		case EFI_ACPI_MEMORY_NVS:
165 			e820_type = E820_TYPE_NVS;
166 			break;
167 		case EFI_UNUSABLE_MEMORY:
168 			e820_type = E820_TYPE_UNUSABLE;
169 			break;
170 		case EFI_PERSISTENT_MEMORY:
171 			e820_type = E820_TYPE_PMEM;
172 			break;
173 		default:
174 			/*
175 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
176 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
177 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
178 			 */
179 			e820_type = E820_TYPE_RESERVED;
180 			break;
181 		}
182 
183 		e820__range_add(start, size, e820_type);
184 	}
185 	e820__update_table(e820_table);
186 }
187 
188 /*
189  * Given add_efi_memmap defaults to 0 and there there is no alternative
190  * e820 mechanism for soft-reserved memory, import the full EFI memory
191  * map if soft reservations are present and enabled. Otherwise, the
192  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
193  * the efi=nosoftreserve option.
194  */
195 static bool do_efi_soft_reserve(void)
196 {
197 	efi_memory_desc_t *md;
198 
199 	if (!efi_enabled(EFI_MEMMAP))
200 		return false;
201 
202 	if (!efi_soft_reserve_enabled())
203 		return false;
204 
205 	for_each_efi_memory_desc(md)
206 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
207 		    (md->attribute & EFI_MEMORY_SP))
208 			return true;
209 	return false;
210 }
211 
212 int __init efi_memblock_x86_reserve_range(void)
213 {
214 	struct efi_info *e = &boot_params.efi_info;
215 	struct efi_memory_map_data data;
216 	phys_addr_t pmap;
217 	int rv;
218 
219 	if (efi_enabled(EFI_PARAVIRT))
220 		return 0;
221 
222 	/* Can't handle firmware tables above 4GB on i386 */
223 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
224 		pr_err("Memory map is above 4GB, disabling EFI.\n");
225 		return -EINVAL;
226 	}
227 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
228 
229 	data.phys_map		= pmap;
230 	data.size 		= e->efi_memmap_size;
231 	data.desc_size		= e->efi_memdesc_size;
232 	data.desc_version	= e->efi_memdesc_version;
233 
234 	rv = efi_memmap_init_early(&data);
235 	if (rv)
236 		return rv;
237 
238 	if (add_efi_memmap || do_efi_soft_reserve())
239 		do_add_efi_memmap();
240 
241 	efi_fake_memmap_early();
242 
243 	WARN(efi.memmap.desc_version != 1,
244 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
245 	     efi.memmap.desc_version);
246 
247 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
248 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
249 
250 	return 0;
251 }
252 
253 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
254 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
255 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
256 
257 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
258 {
259 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
260 	u64 end_hi = 0;
261 	char buf[64];
262 
263 	if (md->num_pages == 0) {
264 		end = 0;
265 	} else if (md->num_pages > EFI_PAGES_MAX ||
266 		   EFI_PAGES_MAX - md->num_pages <
267 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
268 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
269 			>> OVERFLOW_ADDR_SHIFT;
270 
271 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
272 			end_hi += 1;
273 	} else {
274 		return true;
275 	}
276 
277 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
278 
279 	if (end_hi) {
280 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
281 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
282 			md->phys_addr, end_hi, end);
283 	} else {
284 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
285 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
286 			md->phys_addr, end);
287 	}
288 	return false;
289 }
290 
291 static void __init efi_clean_memmap(void)
292 {
293 	efi_memory_desc_t *out = efi.memmap.map;
294 	const efi_memory_desc_t *in = out;
295 	const efi_memory_desc_t *end = efi.memmap.map_end;
296 	int i, n_removal;
297 
298 	for (i = n_removal = 0; in < end; i++) {
299 		if (efi_memmap_entry_valid(in, i)) {
300 			if (out != in)
301 				memcpy(out, in, efi.memmap.desc_size);
302 			out = (void *)out + efi.memmap.desc_size;
303 		} else {
304 			n_removal++;
305 		}
306 		in = (void *)in + efi.memmap.desc_size;
307 	}
308 
309 	if (n_removal > 0) {
310 		struct efi_memory_map_data data = {
311 			.phys_map	= efi.memmap.phys_map,
312 			.desc_version	= efi.memmap.desc_version,
313 			.desc_size	= efi.memmap.desc_size,
314 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
315 			.flags		= 0,
316 		};
317 
318 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
319 		efi_memmap_install(&data);
320 	}
321 }
322 
323 void __init efi_print_memmap(void)
324 {
325 	efi_memory_desc_t *md;
326 	int i = 0;
327 
328 	for_each_efi_memory_desc(md) {
329 		char buf[64];
330 
331 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
332 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
333 			md->phys_addr,
334 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
335 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
336 	}
337 }
338 
339 static int __init efi_systab_init(unsigned long phys)
340 {
341 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
342 					  : sizeof(efi_system_table_32_t);
343 	const efi_table_hdr_t *hdr;
344 	bool over4g = false;
345 	void *p;
346 	int ret;
347 
348 	hdr = p = early_memremap_ro(phys, size);
349 	if (p == NULL) {
350 		pr_err("Couldn't map the system table!\n");
351 		return -ENOMEM;
352 	}
353 
354 	ret = efi_systab_check_header(hdr, 1);
355 	if (ret) {
356 		early_memunmap(p, size);
357 		return ret;
358 	}
359 
360 	if (efi_enabled(EFI_64BIT)) {
361 		const efi_system_table_64_t *systab64 = p;
362 
363 		efi_runtime	= systab64->runtime;
364 		over4g		= systab64->runtime > U32_MAX;
365 
366 		if (efi_setup) {
367 			struct efi_setup_data *data;
368 
369 			data = early_memremap_ro(efi_setup, sizeof(*data));
370 			if (!data) {
371 				early_memunmap(p, size);
372 				return -ENOMEM;
373 			}
374 
375 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
376 			efi_config_table	= (unsigned long)data->tables;
377 
378 			over4g |= data->fw_vendor	> U32_MAX ||
379 				  data->tables		> U32_MAX;
380 
381 			early_memunmap(data, sizeof(*data));
382 		} else {
383 			efi_fw_vendor		= systab64->fw_vendor;
384 			efi_config_table	= systab64->tables;
385 
386 			over4g |= systab64->fw_vendor	> U32_MAX ||
387 				  systab64->tables	> U32_MAX;
388 		}
389 		efi_nr_tables = systab64->nr_tables;
390 	} else {
391 		const efi_system_table_32_t *systab32 = p;
392 
393 		efi_fw_vendor		= systab32->fw_vendor;
394 		efi_runtime		= systab32->runtime;
395 		efi_config_table	= systab32->tables;
396 		efi_nr_tables		= systab32->nr_tables;
397 	}
398 
399 	efi.runtime_version = hdr->revision;
400 
401 	efi_systab_report_header(hdr, efi_fw_vendor);
402 	early_memunmap(p, size);
403 
404 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
405 		pr_err("EFI data located above 4GB, disabling EFI.\n");
406 		return -EINVAL;
407 	}
408 
409 	return 0;
410 }
411 
412 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
413 {
414 	void *config_tables;
415 	int sz, ret;
416 
417 	if (efi_nr_tables == 0)
418 		return 0;
419 
420 	if (efi_enabled(EFI_64BIT))
421 		sz = sizeof(efi_config_table_64_t);
422 	else
423 		sz = sizeof(efi_config_table_32_t);
424 
425 	/*
426 	 * Let's see what config tables the firmware passed to us.
427 	 */
428 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
429 	if (config_tables == NULL) {
430 		pr_err("Could not map Configuration table!\n");
431 		return -ENOMEM;
432 	}
433 
434 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
435 				      arch_tables);
436 
437 	early_memunmap(config_tables, efi_nr_tables * sz);
438 	return ret;
439 }
440 
441 void __init efi_init(void)
442 {
443 	if (IS_ENABLED(CONFIG_X86_32) &&
444 	    (boot_params.efi_info.efi_systab_hi ||
445 	     boot_params.efi_info.efi_memmap_hi)) {
446 		pr_info("Table located above 4GB, disabling EFI.\n");
447 		return;
448 	}
449 
450 	efi_systab_phys = boot_params.efi_info.efi_systab |
451 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
452 
453 	if (efi_systab_init(efi_systab_phys))
454 		return;
455 
456 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
457 		return;
458 
459 	if (efi_config_init(arch_tables))
460 		return;
461 
462 	/*
463 	 * Note: We currently don't support runtime services on an EFI
464 	 * that doesn't match the kernel 32/64-bit mode.
465 	 */
466 
467 	if (!efi_runtime_supported())
468 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
469 
470 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
471 		efi_memmap_unmap();
472 		return;
473 	}
474 
475 	/* Parse the EFI Properties table if it exists */
476 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
477 		efi_properties_table_t *tbl;
478 
479 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
480 		if (tbl == NULL) {
481 			pr_err("Could not map Properties table!\n");
482 		} else {
483 			if (tbl->memory_protection_attribute &
484 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
485 				set_bit(EFI_NX_PE_DATA, &efi.flags);
486 
487 			early_memunmap(tbl, sizeof(*tbl));
488 		}
489 	}
490 
491 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
492 	efi_clean_memmap();
493 
494 	if (efi_enabled(EFI_DBG))
495 		efi_print_memmap();
496 }
497 
498 /* Merge contiguous regions of the same type and attribute */
499 static void __init efi_merge_regions(void)
500 {
501 	efi_memory_desc_t *md, *prev_md = NULL;
502 
503 	for_each_efi_memory_desc(md) {
504 		u64 prev_size;
505 
506 		if (!prev_md) {
507 			prev_md = md;
508 			continue;
509 		}
510 
511 		if (prev_md->type != md->type ||
512 		    prev_md->attribute != md->attribute) {
513 			prev_md = md;
514 			continue;
515 		}
516 
517 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
518 
519 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
520 			prev_md->num_pages += md->num_pages;
521 			md->type = EFI_RESERVED_TYPE;
522 			md->attribute = 0;
523 			continue;
524 		}
525 		prev_md = md;
526 	}
527 }
528 
529 static void *realloc_pages(void *old_memmap, int old_shift)
530 {
531 	void *ret;
532 
533 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
534 	if (!ret)
535 		goto out;
536 
537 	/*
538 	 * A first-time allocation doesn't have anything to copy.
539 	 */
540 	if (!old_memmap)
541 		return ret;
542 
543 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
544 
545 out:
546 	free_pages((unsigned long)old_memmap, old_shift);
547 	return ret;
548 }
549 
550 /*
551  * Iterate the EFI memory map in reverse order because the regions
552  * will be mapped top-down. The end result is the same as if we had
553  * mapped things forward, but doesn't require us to change the
554  * existing implementation of efi_map_region().
555  */
556 static inline void *efi_map_next_entry_reverse(void *entry)
557 {
558 	/* Initial call */
559 	if (!entry)
560 		return efi.memmap.map_end - efi.memmap.desc_size;
561 
562 	entry -= efi.memmap.desc_size;
563 	if (entry < efi.memmap.map)
564 		return NULL;
565 
566 	return entry;
567 }
568 
569 /*
570  * efi_map_next_entry - Return the next EFI memory map descriptor
571  * @entry: Previous EFI memory map descriptor
572  *
573  * This is a helper function to iterate over the EFI memory map, which
574  * we do in different orders depending on the current configuration.
575  *
576  * To begin traversing the memory map @entry must be %NULL.
577  *
578  * Returns %NULL when we reach the end of the memory map.
579  */
580 static void *efi_map_next_entry(void *entry)
581 {
582 	if (efi_enabled(EFI_64BIT)) {
583 		/*
584 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
585 		 * config table feature requires us to map all entries
586 		 * in the same order as they appear in the EFI memory
587 		 * map. That is to say, entry N must have a lower
588 		 * virtual address than entry N+1. This is because the
589 		 * firmware toolchain leaves relative references in
590 		 * the code/data sections, which are split and become
591 		 * separate EFI memory regions. Mapping things
592 		 * out-of-order leads to the firmware accessing
593 		 * unmapped addresses.
594 		 *
595 		 * Since we need to map things this way whether or not
596 		 * the kernel actually makes use of
597 		 * EFI_PROPERTIES_TABLE, let's just switch to this
598 		 * scheme by default for 64-bit.
599 		 */
600 		return efi_map_next_entry_reverse(entry);
601 	}
602 
603 	/* Initial call */
604 	if (!entry)
605 		return efi.memmap.map;
606 
607 	entry += efi.memmap.desc_size;
608 	if (entry >= efi.memmap.map_end)
609 		return NULL;
610 
611 	return entry;
612 }
613 
614 static bool should_map_region(efi_memory_desc_t *md)
615 {
616 	/*
617 	 * Runtime regions always require runtime mappings (obviously).
618 	 */
619 	if (md->attribute & EFI_MEMORY_RUNTIME)
620 		return true;
621 
622 	/*
623 	 * 32-bit EFI doesn't suffer from the bug that requires us to
624 	 * reserve boot services regions, and mixed mode support
625 	 * doesn't exist for 32-bit kernels.
626 	 */
627 	if (IS_ENABLED(CONFIG_X86_32))
628 		return false;
629 
630 	/*
631 	 * EFI specific purpose memory may be reserved by default
632 	 * depending on kernel config and boot options.
633 	 */
634 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
635 	    efi_soft_reserve_enabled() &&
636 	    (md->attribute & EFI_MEMORY_SP))
637 		return false;
638 
639 	/*
640 	 * Map all of RAM so that we can access arguments in the 1:1
641 	 * mapping when making EFI runtime calls.
642 	 */
643 	if (efi_is_mixed()) {
644 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
645 		    md->type == EFI_LOADER_DATA ||
646 		    md->type == EFI_LOADER_CODE)
647 			return true;
648 	}
649 
650 	/*
651 	 * Map boot services regions as a workaround for buggy
652 	 * firmware that accesses them even when they shouldn't.
653 	 *
654 	 * See efi_{reserve,free}_boot_services().
655 	 */
656 	if (md->type == EFI_BOOT_SERVICES_CODE ||
657 	    md->type == EFI_BOOT_SERVICES_DATA)
658 		return true;
659 
660 	return false;
661 }
662 
663 /*
664  * Map the efi memory ranges of the runtime services and update new_mmap with
665  * virtual addresses.
666  */
667 static void * __init efi_map_regions(int *count, int *pg_shift)
668 {
669 	void *p, *new_memmap = NULL;
670 	unsigned long left = 0;
671 	unsigned long desc_size;
672 	efi_memory_desc_t *md;
673 
674 	desc_size = efi.memmap.desc_size;
675 
676 	p = NULL;
677 	while ((p = efi_map_next_entry(p))) {
678 		md = p;
679 
680 		if (!should_map_region(md))
681 			continue;
682 
683 		efi_map_region(md);
684 
685 		if (left < desc_size) {
686 			new_memmap = realloc_pages(new_memmap, *pg_shift);
687 			if (!new_memmap)
688 				return NULL;
689 
690 			left += PAGE_SIZE << *pg_shift;
691 			(*pg_shift)++;
692 		}
693 
694 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
695 
696 		left -= desc_size;
697 		(*count)++;
698 	}
699 
700 	return new_memmap;
701 }
702 
703 static void __init kexec_enter_virtual_mode(void)
704 {
705 #ifdef CONFIG_KEXEC_CORE
706 	efi_memory_desc_t *md;
707 	unsigned int num_pages;
708 
709 	/*
710 	 * We don't do virtual mode, since we don't do runtime services, on
711 	 * non-native EFI.
712 	 */
713 	if (efi_is_mixed()) {
714 		efi_memmap_unmap();
715 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
716 		return;
717 	}
718 
719 	if (efi_alloc_page_tables()) {
720 		pr_err("Failed to allocate EFI page tables\n");
721 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
722 		return;
723 	}
724 
725 	/*
726 	* Map efi regions which were passed via setup_data. The virt_addr is a
727 	* fixed addr which was used in first kernel of a kexec boot.
728 	*/
729 	for_each_efi_memory_desc(md)
730 		efi_map_region_fixed(md); /* FIXME: add error handling */
731 
732 	/*
733 	 * Unregister the early EFI memmap from efi_init() and install
734 	 * the new EFI memory map.
735 	 */
736 	efi_memmap_unmap();
737 
738 	if (efi_memmap_init_late(efi.memmap.phys_map,
739 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
740 		pr_err("Failed to remap late EFI memory map\n");
741 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
742 		return;
743 	}
744 
745 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
746 	num_pages >>= PAGE_SHIFT;
747 
748 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
749 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
750 		return;
751 	}
752 
753 	efi_sync_low_kernel_mappings();
754 	efi_native_runtime_setup();
755 #endif
756 }
757 
758 /*
759  * This function will switch the EFI runtime services to virtual mode.
760  * Essentially, we look through the EFI memmap and map every region that
761  * has the runtime attribute bit set in its memory descriptor into the
762  * efi_pgd page table.
763  *
764  * The new method does a pagetable switch in a preemption-safe manner
765  * so that we're in a different address space when calling a runtime
766  * function. For function arguments passing we do copy the PUDs of the
767  * kernel page table into efi_pgd prior to each call.
768  *
769  * Specially for kexec boot, efi runtime maps in previous kernel should
770  * be passed in via setup_data. In that case runtime ranges will be mapped
771  * to the same virtual addresses as the first kernel, see
772  * kexec_enter_virtual_mode().
773  */
774 static void __init __efi_enter_virtual_mode(void)
775 {
776 	int count = 0, pg_shift = 0;
777 	void *new_memmap = NULL;
778 	efi_status_t status;
779 	unsigned long pa;
780 
781 	if (efi_alloc_page_tables()) {
782 		pr_err("Failed to allocate EFI page tables\n");
783 		goto err;
784 	}
785 
786 	efi_merge_regions();
787 	new_memmap = efi_map_regions(&count, &pg_shift);
788 	if (!new_memmap) {
789 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
790 		goto err;
791 	}
792 
793 	pa = __pa(new_memmap);
794 
795 	/*
796 	 * Unregister the early EFI memmap from efi_init() and install
797 	 * the new EFI memory map that we are about to pass to the
798 	 * firmware via SetVirtualAddressMap().
799 	 */
800 	efi_memmap_unmap();
801 
802 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
803 		pr_err("Failed to remap late EFI memory map\n");
804 		goto err;
805 	}
806 
807 	if (efi_enabled(EFI_DBG)) {
808 		pr_info("EFI runtime memory map:\n");
809 		efi_print_memmap();
810 	}
811 
812 	if (efi_setup_page_tables(pa, 1 << pg_shift))
813 		goto err;
814 
815 	efi_sync_low_kernel_mappings();
816 
817 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
818 					     efi.memmap.desc_size,
819 					     efi.memmap.desc_version,
820 					     (efi_memory_desc_t *)pa,
821 					     efi_systab_phys);
822 	if (status != EFI_SUCCESS) {
823 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
824 		       status);
825 		goto err;
826 	}
827 
828 	efi_check_for_embedded_firmwares();
829 	efi_free_boot_services();
830 
831 	if (!efi_is_mixed())
832 		efi_native_runtime_setup();
833 	else
834 		efi_thunk_runtime_setup();
835 
836 	/*
837 	 * Apply more restrictive page table mapping attributes now that
838 	 * SVAM() has been called and the firmware has performed all
839 	 * necessary relocation fixups for the new virtual addresses.
840 	 */
841 	efi_runtime_update_mappings();
842 
843 	/* clean DUMMY object */
844 	efi_delete_dummy_variable();
845 	return;
846 
847 err:
848 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
849 }
850 
851 void __init efi_enter_virtual_mode(void)
852 {
853 	if (efi_enabled(EFI_PARAVIRT))
854 		return;
855 
856 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
857 
858 	if (efi_setup)
859 		kexec_enter_virtual_mode();
860 	else
861 		__efi_enter_virtual_mode();
862 
863 	efi_dump_pagetable();
864 }
865 
866 bool efi_is_table_address(unsigned long phys_addr)
867 {
868 	unsigned int i;
869 
870 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
871 		return false;
872 
873 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
874 		if (*(efi_tables[i]) == phys_addr)
875 			return true;
876 
877 	return false;
878 }
879 
880 char *efi_systab_show_arch(char *str)
881 {
882 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
883 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
884 	return str;
885 }
886 
887 #define EFI_FIELD(var) efi_ ## var
888 
889 #define EFI_ATTR_SHOW(name) \
890 static ssize_t name##_show(struct kobject *kobj, \
891 				struct kobj_attribute *attr, char *buf) \
892 { \
893 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
894 }
895 
896 EFI_ATTR_SHOW(fw_vendor);
897 EFI_ATTR_SHOW(runtime);
898 EFI_ATTR_SHOW(config_table);
899 
900 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
901 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
902 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
903 
904 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
905 {
906 	if (attr == &efi_attr_fw_vendor.attr) {
907 		if (efi_enabled(EFI_PARAVIRT) ||
908 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
909 			return 0;
910 	} else if (attr == &efi_attr_runtime.attr) {
911 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
912 			return 0;
913 	} else if (attr == &efi_attr_config_table.attr) {
914 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
915 			return 0;
916 	}
917 	return attr->mode;
918 }
919