xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 747f7a29) 
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 };
100 
101 u64 efi_setup;		/* efi setup_data physical address */
102 
103 static int add_efi_memmap __initdata;
104 static int __init setup_add_efi_memmap(char *arg)
105 {
106 	add_efi_memmap = 1;
107 	return 0;
108 }
109 early_param("add_efi_memmap", setup_add_efi_memmap);
110 
111 /*
112  * Tell the kernel about the EFI memory map.  This might include
113  * more than the max 128 entries that can fit in the passed in e820
114  * legacy (zeropage) memory map, but the kernel's e820 table can hold
115  * E820_MAX_ENTRIES.
116  */
117 
118 static void __init do_add_efi_memmap(void)
119 {
120 	efi_memory_desc_t *md;
121 
122 	if (!efi_enabled(EFI_MEMMAP))
123 		return;
124 
125 	for_each_efi_memory_desc(md) {
126 		unsigned long long start = md->phys_addr;
127 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
128 		int e820_type;
129 
130 		switch (md->type) {
131 		case EFI_LOADER_CODE:
132 		case EFI_LOADER_DATA:
133 		case EFI_BOOT_SERVICES_CODE:
134 		case EFI_BOOT_SERVICES_DATA:
135 		case EFI_CONVENTIONAL_MEMORY:
136 			if (efi_soft_reserve_enabled()
137 			    && (md->attribute & EFI_MEMORY_SP))
138 				e820_type = E820_TYPE_SOFT_RESERVED;
139 			else if (md->attribute & EFI_MEMORY_WB)
140 				e820_type = E820_TYPE_RAM;
141 			else
142 				e820_type = E820_TYPE_RESERVED;
143 			break;
144 		case EFI_ACPI_RECLAIM_MEMORY:
145 			e820_type = E820_TYPE_ACPI;
146 			break;
147 		case EFI_ACPI_MEMORY_NVS:
148 			e820_type = E820_TYPE_NVS;
149 			break;
150 		case EFI_UNUSABLE_MEMORY:
151 			e820_type = E820_TYPE_UNUSABLE;
152 			break;
153 		case EFI_PERSISTENT_MEMORY:
154 			e820_type = E820_TYPE_PMEM;
155 			break;
156 		default:
157 			/*
158 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
159 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
160 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
161 			 */
162 			e820_type = E820_TYPE_RESERVED;
163 			break;
164 		}
165 
166 		e820__range_add(start, size, e820_type);
167 	}
168 	e820__update_table(e820_table);
169 }
170 
171 /*
172  * Given add_efi_memmap defaults to 0 and there there is no alternative
173  * e820 mechanism for soft-reserved memory, import the full EFI memory
174  * map if soft reservations are present and enabled. Otherwise, the
175  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
176  * the efi=nosoftreserve option.
177  */
178 static bool do_efi_soft_reserve(void)
179 {
180 	efi_memory_desc_t *md;
181 
182 	if (!efi_enabled(EFI_MEMMAP))
183 		return false;
184 
185 	if (!efi_soft_reserve_enabled())
186 		return false;
187 
188 	for_each_efi_memory_desc(md)
189 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
190 		    (md->attribute & EFI_MEMORY_SP))
191 			return true;
192 	return false;
193 }
194 
195 int __init efi_memblock_x86_reserve_range(void)
196 {
197 	struct efi_info *e = &boot_params.efi_info;
198 	struct efi_memory_map_data data;
199 	phys_addr_t pmap;
200 	int rv;
201 
202 	if (efi_enabled(EFI_PARAVIRT))
203 		return 0;
204 
205 	/* Can't handle firmware tables above 4GB on i386 */
206 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
207 		pr_err("Memory map is above 4GB, disabling EFI.\n");
208 		return -EINVAL;
209 	}
210 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
211 
212 	data.phys_map		= pmap;
213 	data.size 		= e->efi_memmap_size;
214 	data.desc_size		= e->efi_memdesc_size;
215 	data.desc_version	= e->efi_memdesc_version;
216 
217 	rv = efi_memmap_init_early(&data);
218 	if (rv)
219 		return rv;
220 
221 	if (add_efi_memmap || do_efi_soft_reserve())
222 		do_add_efi_memmap();
223 
224 	efi_fake_memmap_early();
225 
226 	WARN(efi.memmap.desc_version != 1,
227 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
228 	     efi.memmap.desc_version);
229 
230 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
231 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
232 
233 	return 0;
234 }
235 
236 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
237 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
238 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
239 
240 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
241 {
242 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
243 	u64 end_hi = 0;
244 	char buf[64];
245 
246 	if (md->num_pages == 0) {
247 		end = 0;
248 	} else if (md->num_pages > EFI_PAGES_MAX ||
249 		   EFI_PAGES_MAX - md->num_pages <
250 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
251 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
252 			>> OVERFLOW_ADDR_SHIFT;
253 
254 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
255 			end_hi += 1;
256 	} else {
257 		return true;
258 	}
259 
260 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
261 
262 	if (end_hi) {
263 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
264 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
265 			md->phys_addr, end_hi, end);
266 	} else {
267 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
268 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
269 			md->phys_addr, end);
270 	}
271 	return false;
272 }
273 
274 static void __init efi_clean_memmap(void)
275 {
276 	efi_memory_desc_t *out = efi.memmap.map;
277 	const efi_memory_desc_t *in = out;
278 	const efi_memory_desc_t *end = efi.memmap.map_end;
279 	int i, n_removal;
280 
281 	for (i = n_removal = 0; in < end; i++) {
282 		if (efi_memmap_entry_valid(in, i)) {
283 			if (out != in)
284 				memcpy(out, in, efi.memmap.desc_size);
285 			out = (void *)out + efi.memmap.desc_size;
286 		} else {
287 			n_removal++;
288 		}
289 		in = (void *)in + efi.memmap.desc_size;
290 	}
291 
292 	if (n_removal > 0) {
293 		struct efi_memory_map_data data = {
294 			.phys_map	= efi.memmap.phys_map,
295 			.desc_version	= efi.memmap.desc_version,
296 			.desc_size	= efi.memmap.desc_size,
297 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
298 			.flags		= 0,
299 		};
300 
301 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
302 		efi_memmap_install(&data);
303 	}
304 }
305 
306 void __init efi_print_memmap(void)
307 {
308 	efi_memory_desc_t *md;
309 	int i = 0;
310 
311 	for_each_efi_memory_desc(md) {
312 		char buf[64];
313 
314 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
315 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
316 			md->phys_addr,
317 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
318 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
319 	}
320 }
321 
322 static int __init efi_systab_init(unsigned long phys)
323 {
324 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
325 					  : sizeof(efi_system_table_32_t);
326 	const efi_table_hdr_t *hdr;
327 	bool over4g = false;
328 	void *p;
329 	int ret;
330 
331 	hdr = p = early_memremap_ro(phys, size);
332 	if (p == NULL) {
333 		pr_err("Couldn't map the system table!\n");
334 		return -ENOMEM;
335 	}
336 
337 	ret = efi_systab_check_header(hdr, 1);
338 	if (ret) {
339 		early_memunmap(p, size);
340 		return ret;
341 	}
342 
343 	if (efi_enabled(EFI_64BIT)) {
344 		const efi_system_table_64_t *systab64 = p;
345 
346 		efi_runtime	= systab64->runtime;
347 		over4g		= systab64->runtime > U32_MAX;
348 
349 		if (efi_setup) {
350 			struct efi_setup_data *data;
351 
352 			data = early_memremap_ro(efi_setup, sizeof(*data));
353 			if (!data) {
354 				early_memunmap(p, size);
355 				return -ENOMEM;
356 			}
357 
358 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
359 			efi_config_table	= (unsigned long)data->tables;
360 
361 			over4g |= data->fw_vendor	> U32_MAX ||
362 				  data->tables		> U32_MAX;
363 
364 			early_memunmap(data, sizeof(*data));
365 		} else {
366 			efi_fw_vendor		= systab64->fw_vendor;
367 			efi_config_table	= systab64->tables;
368 
369 			over4g |= systab64->fw_vendor	> U32_MAX ||
370 				  systab64->tables	> U32_MAX;
371 		}
372 		efi_nr_tables = systab64->nr_tables;
373 	} else {
374 		const efi_system_table_32_t *systab32 = p;
375 
376 		efi_fw_vendor		= systab32->fw_vendor;
377 		efi_runtime		= systab32->runtime;
378 		efi_config_table	= systab32->tables;
379 		efi_nr_tables		= systab32->nr_tables;
380 	}
381 
382 	efi.runtime_version = hdr->revision;
383 
384 	efi_systab_report_header(hdr, efi_fw_vendor);
385 	early_memunmap(p, size);
386 
387 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
388 		pr_err("EFI data located above 4GB, disabling EFI.\n");
389 		return -EINVAL;
390 	}
391 
392 	return 0;
393 }
394 
395 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
396 {
397 	void *config_tables;
398 	int sz, ret;
399 
400 	if (efi_nr_tables == 0)
401 		return 0;
402 
403 	if (efi_enabled(EFI_64BIT))
404 		sz = sizeof(efi_config_table_64_t);
405 	else
406 		sz = sizeof(efi_config_table_32_t);
407 
408 	/*
409 	 * Let's see what config tables the firmware passed to us.
410 	 */
411 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
412 	if (config_tables == NULL) {
413 		pr_err("Could not map Configuration table!\n");
414 		return -ENOMEM;
415 	}
416 
417 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
418 				      arch_tables);
419 
420 	early_memunmap(config_tables, efi_nr_tables * sz);
421 	return ret;
422 }
423 
424 void __init efi_init(void)
425 {
426 	if (IS_ENABLED(CONFIG_X86_32) &&
427 	    (boot_params.efi_info.efi_systab_hi ||
428 	     boot_params.efi_info.efi_memmap_hi)) {
429 		pr_info("Table located above 4GB, disabling EFI.\n");
430 		return;
431 	}
432 
433 	efi_systab_phys = boot_params.efi_info.efi_systab |
434 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
435 
436 	if (efi_systab_init(efi_systab_phys))
437 		return;
438 
439 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
440 		return;
441 
442 	if (efi_config_init(arch_tables))
443 		return;
444 
445 	/*
446 	 * Note: We currently don't support runtime services on an EFI
447 	 * that doesn't match the kernel 32/64-bit mode.
448 	 */
449 
450 	if (!efi_runtime_supported())
451 		pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
452 
453 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
454 		efi_memmap_unmap();
455 		return;
456 	}
457 
458 	/* Parse the EFI Properties table if it exists */
459 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
460 		efi_properties_table_t *tbl;
461 
462 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
463 		if (tbl == NULL) {
464 			pr_err("Could not map Properties table!\n");
465 		} else {
466 			if (tbl->memory_protection_attribute &
467 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
468 				set_bit(EFI_NX_PE_DATA, &efi.flags);
469 
470 			early_memunmap(tbl, sizeof(*tbl));
471 		}
472 	}
473 
474 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
475 	efi_clean_memmap();
476 
477 	if (efi_enabled(EFI_DBG))
478 		efi_print_memmap();
479 }
480 
481 /* Merge contiguous regions of the same type and attribute */
482 static void __init efi_merge_regions(void)
483 {
484 	efi_memory_desc_t *md, *prev_md = NULL;
485 
486 	for_each_efi_memory_desc(md) {
487 		u64 prev_size;
488 
489 		if (!prev_md) {
490 			prev_md = md;
491 			continue;
492 		}
493 
494 		if (prev_md->type != md->type ||
495 		    prev_md->attribute != md->attribute) {
496 			prev_md = md;
497 			continue;
498 		}
499 
500 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
501 
502 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
503 			prev_md->num_pages += md->num_pages;
504 			md->type = EFI_RESERVED_TYPE;
505 			md->attribute = 0;
506 			continue;
507 		}
508 		prev_md = md;
509 	}
510 }
511 
512 static void *realloc_pages(void *old_memmap, int old_shift)
513 {
514 	void *ret;
515 
516 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
517 	if (!ret)
518 		goto out;
519 
520 	/*
521 	 * A first-time allocation doesn't have anything to copy.
522 	 */
523 	if (!old_memmap)
524 		return ret;
525 
526 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
527 
528 out:
529 	free_pages((unsigned long)old_memmap, old_shift);
530 	return ret;
531 }
532 
533 /*
534  * Iterate the EFI memory map in reverse order because the regions
535  * will be mapped top-down. The end result is the same as if we had
536  * mapped things forward, but doesn't require us to change the
537  * existing implementation of efi_map_region().
538  */
539 static inline void *efi_map_next_entry_reverse(void *entry)
540 {
541 	/* Initial call */
542 	if (!entry)
543 		return efi.memmap.map_end - efi.memmap.desc_size;
544 
545 	entry -= efi.memmap.desc_size;
546 	if (entry < efi.memmap.map)
547 		return NULL;
548 
549 	return entry;
550 }
551 
552 /*
553  * efi_map_next_entry - Return the next EFI memory map descriptor
554  * @entry: Previous EFI memory map descriptor
555  *
556  * This is a helper function to iterate over the EFI memory map, which
557  * we do in different orders depending on the current configuration.
558  *
559  * To begin traversing the memory map @entry must be %NULL.
560  *
561  * Returns %NULL when we reach the end of the memory map.
562  */
563 static void *efi_map_next_entry(void *entry)
564 {
565 	if (efi_enabled(EFI_64BIT)) {
566 		/*
567 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
568 		 * config table feature requires us to map all entries
569 		 * in the same order as they appear in the EFI memory
570 		 * map. That is to say, entry N must have a lower
571 		 * virtual address than entry N+1. This is because the
572 		 * firmware toolchain leaves relative references in
573 		 * the code/data sections, which are split and become
574 		 * separate EFI memory regions. Mapping things
575 		 * out-of-order leads to the firmware accessing
576 		 * unmapped addresses.
577 		 *
578 		 * Since we need to map things this way whether or not
579 		 * the kernel actually makes use of
580 		 * EFI_PROPERTIES_TABLE, let's just switch to this
581 		 * scheme by default for 64-bit.
582 		 */
583 		return efi_map_next_entry_reverse(entry);
584 	}
585 
586 	/* Initial call */
587 	if (!entry)
588 		return efi.memmap.map;
589 
590 	entry += efi.memmap.desc_size;
591 	if (entry >= efi.memmap.map_end)
592 		return NULL;
593 
594 	return entry;
595 }
596 
597 static bool should_map_region(efi_memory_desc_t *md)
598 {
599 	/*
600 	 * Runtime regions always require runtime mappings (obviously).
601 	 */
602 	if (md->attribute & EFI_MEMORY_RUNTIME)
603 		return true;
604 
605 	/*
606 	 * 32-bit EFI doesn't suffer from the bug that requires us to
607 	 * reserve boot services regions, and mixed mode support
608 	 * doesn't exist for 32-bit kernels.
609 	 */
610 	if (IS_ENABLED(CONFIG_X86_32))
611 		return false;
612 
613 	/*
614 	 * EFI specific purpose memory may be reserved by default
615 	 * depending on kernel config and boot options.
616 	 */
617 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
618 	    efi_soft_reserve_enabled() &&
619 	    (md->attribute & EFI_MEMORY_SP))
620 		return false;
621 
622 	/*
623 	 * Map all of RAM so that we can access arguments in the 1:1
624 	 * mapping when making EFI runtime calls.
625 	 */
626 	if (efi_is_mixed()) {
627 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
628 		    md->type == EFI_LOADER_DATA ||
629 		    md->type == EFI_LOADER_CODE)
630 			return true;
631 	}
632 
633 	/*
634 	 * Map boot services regions as a workaround for buggy
635 	 * firmware that accesses them even when they shouldn't.
636 	 *
637 	 * See efi_{reserve,free}_boot_services().
638 	 */
639 	if (md->type == EFI_BOOT_SERVICES_CODE ||
640 	    md->type == EFI_BOOT_SERVICES_DATA)
641 		return true;
642 
643 	return false;
644 }
645 
646 /*
647  * Map the efi memory ranges of the runtime services and update new_mmap with
648  * virtual addresses.
649  */
650 static void * __init efi_map_regions(int *count, int *pg_shift)
651 {
652 	void *p, *new_memmap = NULL;
653 	unsigned long left = 0;
654 	unsigned long desc_size;
655 	efi_memory_desc_t *md;
656 
657 	desc_size = efi.memmap.desc_size;
658 
659 	p = NULL;
660 	while ((p = efi_map_next_entry(p))) {
661 		md = p;
662 
663 		if (!should_map_region(md))
664 			continue;
665 
666 		efi_map_region(md);
667 
668 		if (left < desc_size) {
669 			new_memmap = realloc_pages(new_memmap, *pg_shift);
670 			if (!new_memmap)
671 				return NULL;
672 
673 			left += PAGE_SIZE << *pg_shift;
674 			(*pg_shift)++;
675 		}
676 
677 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
678 
679 		left -= desc_size;
680 		(*count)++;
681 	}
682 
683 	return new_memmap;
684 }
685 
686 static void __init kexec_enter_virtual_mode(void)
687 {
688 #ifdef CONFIG_KEXEC_CORE
689 	efi_memory_desc_t *md;
690 	unsigned int num_pages;
691 
692 	/*
693 	 * We don't do virtual mode, since we don't do runtime services, on
694 	 * non-native EFI.
695 	 */
696 	if (efi_is_mixed()) {
697 		efi_memmap_unmap();
698 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
699 		return;
700 	}
701 
702 	if (efi_alloc_page_tables()) {
703 		pr_err("Failed to allocate EFI page tables\n");
704 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
705 		return;
706 	}
707 
708 	/*
709 	* Map efi regions which were passed via setup_data. The virt_addr is a
710 	* fixed addr which was used in first kernel of a kexec boot.
711 	*/
712 	for_each_efi_memory_desc(md)
713 		efi_map_region_fixed(md); /* FIXME: add error handling */
714 
715 	/*
716 	 * Unregister the early EFI memmap from efi_init() and install
717 	 * the new EFI memory map.
718 	 */
719 	efi_memmap_unmap();
720 
721 	if (efi_memmap_init_late(efi.memmap.phys_map,
722 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
723 		pr_err("Failed to remap late EFI memory map\n");
724 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
725 		return;
726 	}
727 
728 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
729 	num_pages >>= PAGE_SHIFT;
730 
731 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
732 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
733 		return;
734 	}
735 
736 	efi_sync_low_kernel_mappings();
737 	efi_native_runtime_setup();
738 #endif
739 }
740 
741 /*
742  * This function will switch the EFI runtime services to virtual mode.
743  * Essentially, we look through the EFI memmap and map every region that
744  * has the runtime attribute bit set in its memory descriptor into the
745  * efi_pgd page table.
746  *
747  * The new method does a pagetable switch in a preemption-safe manner
748  * so that we're in a different address space when calling a runtime
749  * function. For function arguments passing we do copy the PUDs of the
750  * kernel page table into efi_pgd prior to each call.
751  *
752  * Specially for kexec boot, efi runtime maps in previous kernel should
753  * be passed in via setup_data. In that case runtime ranges will be mapped
754  * to the same virtual addresses as the first kernel, see
755  * kexec_enter_virtual_mode().
756  */
757 static void __init __efi_enter_virtual_mode(void)
758 {
759 	int count = 0, pg_shift = 0;
760 	void *new_memmap = NULL;
761 	efi_status_t status;
762 	unsigned long pa;
763 
764 	if (efi_alloc_page_tables()) {
765 		pr_err("Failed to allocate EFI page tables\n");
766 		goto err;
767 	}
768 
769 	efi_merge_regions();
770 	new_memmap = efi_map_regions(&count, &pg_shift);
771 	if (!new_memmap) {
772 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
773 		goto err;
774 	}
775 
776 	pa = __pa(new_memmap);
777 
778 	/*
779 	 * Unregister the early EFI memmap from efi_init() and install
780 	 * the new EFI memory map that we are about to pass to the
781 	 * firmware via SetVirtualAddressMap().
782 	 */
783 	efi_memmap_unmap();
784 
785 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
786 		pr_err("Failed to remap late EFI memory map\n");
787 		goto err;
788 	}
789 
790 	if (efi_enabled(EFI_DBG)) {
791 		pr_info("EFI runtime memory map:\n");
792 		efi_print_memmap();
793 	}
794 
795 	if (efi_setup_page_tables(pa, 1 << pg_shift))
796 		goto err;
797 
798 	efi_sync_low_kernel_mappings();
799 
800 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
801 					     efi.memmap.desc_size,
802 					     efi.memmap.desc_version,
803 					     (efi_memory_desc_t *)pa,
804 					     efi_systab_phys);
805 	if (status != EFI_SUCCESS) {
806 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
807 		       status);
808 		goto err;
809 	}
810 
811 	efi_check_for_embedded_firmwares();
812 	efi_free_boot_services();
813 
814 	if (!efi_is_mixed())
815 		efi_native_runtime_setup();
816 	else
817 		efi_thunk_runtime_setup();
818 
819 	/*
820 	 * Apply more restrictive page table mapping attributes now that
821 	 * SVAM() has been called and the firmware has performed all
822 	 * necessary relocation fixups for the new virtual addresses.
823 	 */
824 	efi_runtime_update_mappings();
825 
826 	/* clean DUMMY object */
827 	efi_delete_dummy_variable();
828 	return;
829 
830 err:
831 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
832 }
833 
834 void __init efi_enter_virtual_mode(void)
835 {
836 	if (efi_enabled(EFI_PARAVIRT))
837 		return;
838 
839 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
840 
841 	if (efi_setup)
842 		kexec_enter_virtual_mode();
843 	else
844 		__efi_enter_virtual_mode();
845 
846 	efi_dump_pagetable();
847 }
848 
849 bool efi_is_table_address(unsigned long phys_addr)
850 {
851 	unsigned int i;
852 
853 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
854 		return false;
855 
856 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
857 		if (*(efi_tables[i]) == phys_addr)
858 			return true;
859 
860 	return false;
861 }
862 
863 char *efi_systab_show_arch(char *str)
864 {
865 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
866 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
867 	return str;
868 }
869 
870 #define EFI_FIELD(var) efi_ ## var
871 
872 #define EFI_ATTR_SHOW(name) \
873 static ssize_t name##_show(struct kobject *kobj, \
874 				struct kobj_attribute *attr, char *buf) \
875 { \
876 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
877 }
878 
879 EFI_ATTR_SHOW(fw_vendor);
880 EFI_ATTR_SHOW(runtime);
881 EFI_ATTR_SHOW(config_table);
882 
883 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
884 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
885 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
886 
887 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
888 {
889 	if (attr == &efi_attr_fw_vendor.attr) {
890 		if (efi_enabled(EFI_PARAVIRT) ||
891 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
892 			return 0;
893 	} else if (attr == &efi_attr_runtime.attr) {
894 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
895 			return 0;
896 	} else if (attr == &efi_attr_config_table.attr) {
897 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
898 			return 0;
899 	}
900 	return attr->mode;
901 }
902