xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 53a2a90d) 
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 void __init efi_find_mirror(void)
112 {
113 	efi_memory_desc_t *md;
114 	u64 mirror_size = 0, total_size = 0;
115 
116 	if (!efi_enabled(EFI_MEMMAP))
117 		return;
118 
119 	for_each_efi_memory_desc(md) {
120 		unsigned long long start = md->phys_addr;
121 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
122 
123 		total_size += size;
124 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
125 			memblock_mark_mirror(start, size);
126 			mirror_size += size;
127 		}
128 	}
129 	if (mirror_size)
130 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
131 			mirror_size>>20, total_size>>20);
132 }
133 
134 /*
135  * Tell the kernel about the EFI memory map.  This might include
136  * more than the max 128 entries that can fit in the passed in e820
137  * legacy (zeropage) memory map, but the kernel's e820 table can hold
138  * E820_MAX_ENTRIES.
139  */
140 
141 static void __init do_add_efi_memmap(void)
142 {
143 	efi_memory_desc_t *md;
144 
145 	if (!efi_enabled(EFI_MEMMAP))
146 		return;
147 
148 	for_each_efi_memory_desc(md) {
149 		unsigned long long start = md->phys_addr;
150 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
151 		int e820_type;
152 
153 		switch (md->type) {
154 		case EFI_LOADER_CODE:
155 		case EFI_LOADER_DATA:
156 		case EFI_BOOT_SERVICES_CODE:
157 		case EFI_BOOT_SERVICES_DATA:
158 		case EFI_CONVENTIONAL_MEMORY:
159 			if (efi_soft_reserve_enabled()
160 			    && (md->attribute & EFI_MEMORY_SP))
161 				e820_type = E820_TYPE_SOFT_RESERVED;
162 			else if (md->attribute & EFI_MEMORY_WB)
163 				e820_type = E820_TYPE_RAM;
164 			else
165 				e820_type = E820_TYPE_RESERVED;
166 			break;
167 		case EFI_ACPI_RECLAIM_MEMORY:
168 			e820_type = E820_TYPE_ACPI;
169 			break;
170 		case EFI_ACPI_MEMORY_NVS:
171 			e820_type = E820_TYPE_NVS;
172 			break;
173 		case EFI_UNUSABLE_MEMORY:
174 			e820_type = E820_TYPE_UNUSABLE;
175 			break;
176 		case EFI_PERSISTENT_MEMORY:
177 			e820_type = E820_TYPE_PMEM;
178 			break;
179 		default:
180 			/*
181 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
182 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
183 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
184 			 */
185 			e820_type = E820_TYPE_RESERVED;
186 			break;
187 		}
188 
189 		e820__range_add(start, size, e820_type);
190 	}
191 	e820__update_table(e820_table);
192 }
193 
194 /*
195  * Given add_efi_memmap defaults to 0 and there there is no alternative
196  * e820 mechanism for soft-reserved memory, import the full EFI memory
197  * map if soft reservations are present and enabled. Otherwise, the
198  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
199  * the efi=nosoftreserve option.
200  */
201 static bool do_efi_soft_reserve(void)
202 {
203 	efi_memory_desc_t *md;
204 
205 	if (!efi_enabled(EFI_MEMMAP))
206 		return false;
207 
208 	if (!efi_soft_reserve_enabled())
209 		return false;
210 
211 	for_each_efi_memory_desc(md)
212 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
213 		    (md->attribute & EFI_MEMORY_SP))
214 			return true;
215 	return false;
216 }
217 
218 int __init efi_memblock_x86_reserve_range(void)
219 {
220 	struct efi_info *e = &boot_params.efi_info;
221 	struct efi_memory_map_data data;
222 	phys_addr_t pmap;
223 	int rv;
224 
225 	if (efi_enabled(EFI_PARAVIRT))
226 		return 0;
227 
228 	/* Can't handle firmware tables above 4GB on i386 */
229 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
230 		pr_err("Memory map is above 4GB, disabling EFI.\n");
231 		return -EINVAL;
232 	}
233 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
234 
235 	data.phys_map		= pmap;
236 	data.size 		= e->efi_memmap_size;
237 	data.desc_size		= e->efi_memdesc_size;
238 	data.desc_version	= e->efi_memdesc_version;
239 
240 	rv = efi_memmap_init_early(&data);
241 	if (rv)
242 		return rv;
243 
244 	if (add_efi_memmap || do_efi_soft_reserve())
245 		do_add_efi_memmap();
246 
247 	efi_fake_memmap_early();
248 
249 	WARN(efi.memmap.desc_version != 1,
250 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
251 	     efi.memmap.desc_version);
252 
253 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
254 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
255 
256 	return 0;
257 }
258 
259 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
260 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
261 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
262 
263 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
264 {
265 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
266 	u64 end_hi = 0;
267 	char buf[64];
268 
269 	if (md->num_pages == 0) {
270 		end = 0;
271 	} else if (md->num_pages > EFI_PAGES_MAX ||
272 		   EFI_PAGES_MAX - md->num_pages <
273 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
274 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
275 			>> OVERFLOW_ADDR_SHIFT;
276 
277 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
278 			end_hi += 1;
279 	} else {
280 		return true;
281 	}
282 
283 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
284 
285 	if (end_hi) {
286 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
287 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
288 			md->phys_addr, end_hi, end);
289 	} else {
290 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
291 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
292 			md->phys_addr, end);
293 	}
294 	return false;
295 }
296 
297 static void __init efi_clean_memmap(void)
298 {
299 	efi_memory_desc_t *out = efi.memmap.map;
300 	const efi_memory_desc_t *in = out;
301 	const efi_memory_desc_t *end = efi.memmap.map_end;
302 	int i, n_removal;
303 
304 	for (i = n_removal = 0; in < end; i++) {
305 		if (efi_memmap_entry_valid(in, i)) {
306 			if (out != in)
307 				memcpy(out, in, efi.memmap.desc_size);
308 			out = (void *)out + efi.memmap.desc_size;
309 		} else {
310 			n_removal++;
311 		}
312 		in = (void *)in + efi.memmap.desc_size;
313 	}
314 
315 	if (n_removal > 0) {
316 		struct efi_memory_map_data data = {
317 			.phys_map	= efi.memmap.phys_map,
318 			.desc_version	= efi.memmap.desc_version,
319 			.desc_size	= efi.memmap.desc_size,
320 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
321 			.flags		= 0,
322 		};
323 
324 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
325 		efi_memmap_install(&data);
326 	}
327 }
328 
329 void __init efi_print_memmap(void)
330 {
331 	efi_memory_desc_t *md;
332 	int i = 0;
333 
334 	for_each_efi_memory_desc(md) {
335 		char buf[64];
336 
337 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
338 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
339 			md->phys_addr,
340 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
341 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
342 	}
343 }
344 
345 static int __init efi_systab_init(unsigned long phys)
346 {
347 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
348 					  : sizeof(efi_system_table_32_t);
349 	const efi_table_hdr_t *hdr;
350 	bool over4g = false;
351 	void *p;
352 	int ret;
353 
354 	hdr = p = early_memremap_ro(phys, size);
355 	if (p == NULL) {
356 		pr_err("Couldn't map the system table!\n");
357 		return -ENOMEM;
358 	}
359 
360 	ret = efi_systab_check_header(hdr, 1);
361 	if (ret) {
362 		early_memunmap(p, size);
363 		return ret;
364 	}
365 
366 	if (efi_enabled(EFI_64BIT)) {
367 		const efi_system_table_64_t *systab64 = p;
368 
369 		efi_runtime	= systab64->runtime;
370 		over4g		= systab64->runtime > U32_MAX;
371 
372 		if (efi_setup) {
373 			struct efi_setup_data *data;
374 
375 			data = early_memremap_ro(efi_setup, sizeof(*data));
376 			if (!data) {
377 				early_memunmap(p, size);
378 				return -ENOMEM;
379 			}
380 
381 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
382 			efi_config_table	= (unsigned long)data->tables;
383 
384 			over4g |= data->fw_vendor	> U32_MAX ||
385 				  data->tables		> U32_MAX;
386 
387 			early_memunmap(data, sizeof(*data));
388 		} else {
389 			efi_fw_vendor		= systab64->fw_vendor;
390 			efi_config_table	= systab64->tables;
391 
392 			over4g |= systab64->fw_vendor	> U32_MAX ||
393 				  systab64->tables	> U32_MAX;
394 		}
395 		efi_nr_tables = systab64->nr_tables;
396 	} else {
397 		const efi_system_table_32_t *systab32 = p;
398 
399 		efi_fw_vendor		= systab32->fw_vendor;
400 		efi_runtime		= systab32->runtime;
401 		efi_config_table	= systab32->tables;
402 		efi_nr_tables		= systab32->nr_tables;
403 	}
404 
405 	efi.runtime_version = hdr->revision;
406 
407 	efi_systab_report_header(hdr, efi_fw_vendor);
408 	early_memunmap(p, size);
409 
410 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
411 		pr_err("EFI data located above 4GB, disabling EFI.\n");
412 		return -EINVAL;
413 	}
414 
415 	return 0;
416 }
417 
418 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
419 {
420 	void *config_tables;
421 	int sz, ret;
422 
423 	if (efi_nr_tables == 0)
424 		return 0;
425 
426 	if (efi_enabled(EFI_64BIT))
427 		sz = sizeof(efi_config_table_64_t);
428 	else
429 		sz = sizeof(efi_config_table_32_t);
430 
431 	/*
432 	 * Let's see what config tables the firmware passed to us.
433 	 */
434 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
435 	if (config_tables == NULL) {
436 		pr_err("Could not map Configuration table!\n");
437 		return -ENOMEM;
438 	}
439 
440 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
441 				      arch_tables);
442 
443 	early_memunmap(config_tables, efi_nr_tables * sz);
444 	return ret;
445 }
446 
447 void __init efi_init(void)
448 {
449 	if (IS_ENABLED(CONFIG_X86_32) &&
450 	    (boot_params.efi_info.efi_systab_hi ||
451 	     boot_params.efi_info.efi_memmap_hi)) {
452 		pr_info("Table located above 4GB, disabling EFI.\n");
453 		return;
454 	}
455 
456 	efi_systab_phys = boot_params.efi_info.efi_systab |
457 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
458 
459 	if (efi_systab_init(efi_systab_phys))
460 		return;
461 
462 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
463 		return;
464 
465 	if (efi_config_init(arch_tables))
466 		return;
467 
468 	/*
469 	 * Note: We currently don't support runtime services on an EFI
470 	 * that doesn't match the kernel 32/64-bit mode.
471 	 */
472 
473 	if (!efi_runtime_supported())
474 		pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
475 
476 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
477 		efi_memmap_unmap();
478 		return;
479 	}
480 
481 	/* Parse the EFI Properties table if it exists */
482 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
483 		efi_properties_table_t *tbl;
484 
485 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
486 		if (tbl == NULL) {
487 			pr_err("Could not map Properties table!\n");
488 		} else {
489 			if (tbl->memory_protection_attribute &
490 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
491 				set_bit(EFI_NX_PE_DATA, &efi.flags);
492 
493 			early_memunmap(tbl, sizeof(*tbl));
494 		}
495 	}
496 
497 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
498 	efi_clean_memmap();
499 
500 	if (efi_enabled(EFI_DBG))
501 		efi_print_memmap();
502 }
503 
504 /* Merge contiguous regions of the same type and attribute */
505 static void __init efi_merge_regions(void)
506 {
507 	efi_memory_desc_t *md, *prev_md = NULL;
508 
509 	for_each_efi_memory_desc(md) {
510 		u64 prev_size;
511 
512 		if (!prev_md) {
513 			prev_md = md;
514 			continue;
515 		}
516 
517 		if (prev_md->type != md->type ||
518 		    prev_md->attribute != md->attribute) {
519 			prev_md = md;
520 			continue;
521 		}
522 
523 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
524 
525 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
526 			prev_md->num_pages += md->num_pages;
527 			md->type = EFI_RESERVED_TYPE;
528 			md->attribute = 0;
529 			continue;
530 		}
531 		prev_md = md;
532 	}
533 }
534 
535 static void *realloc_pages(void *old_memmap, int old_shift)
536 {
537 	void *ret;
538 
539 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
540 	if (!ret)
541 		goto out;
542 
543 	/*
544 	 * A first-time allocation doesn't have anything to copy.
545 	 */
546 	if (!old_memmap)
547 		return ret;
548 
549 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
550 
551 out:
552 	free_pages((unsigned long)old_memmap, old_shift);
553 	return ret;
554 }
555 
556 /*
557  * Iterate the EFI memory map in reverse order because the regions
558  * will be mapped top-down. The end result is the same as if we had
559  * mapped things forward, but doesn't require us to change the
560  * existing implementation of efi_map_region().
561  */
562 static inline void *efi_map_next_entry_reverse(void *entry)
563 {
564 	/* Initial call */
565 	if (!entry)
566 		return efi.memmap.map_end - efi.memmap.desc_size;
567 
568 	entry -= efi.memmap.desc_size;
569 	if (entry < efi.memmap.map)
570 		return NULL;
571 
572 	return entry;
573 }
574 
575 /*
576  * efi_map_next_entry - Return the next EFI memory map descriptor
577  * @entry: Previous EFI memory map descriptor
578  *
579  * This is a helper function to iterate over the EFI memory map, which
580  * we do in different orders depending on the current configuration.
581  *
582  * To begin traversing the memory map @entry must be %NULL.
583  *
584  * Returns %NULL when we reach the end of the memory map.
585  */
586 static void *efi_map_next_entry(void *entry)
587 {
588 	if (efi_enabled(EFI_64BIT)) {
589 		/*
590 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
591 		 * config table feature requires us to map all entries
592 		 * in the same order as they appear in the EFI memory
593 		 * map. That is to say, entry N must have a lower
594 		 * virtual address than entry N+1. This is because the
595 		 * firmware toolchain leaves relative references in
596 		 * the code/data sections, which are split and become
597 		 * separate EFI memory regions. Mapping things
598 		 * out-of-order leads to the firmware accessing
599 		 * unmapped addresses.
600 		 *
601 		 * Since we need to map things this way whether or not
602 		 * the kernel actually makes use of
603 		 * EFI_PROPERTIES_TABLE, let's just switch to this
604 		 * scheme by default for 64-bit.
605 		 */
606 		return efi_map_next_entry_reverse(entry);
607 	}
608 
609 	/* Initial call */
610 	if (!entry)
611 		return efi.memmap.map;
612 
613 	entry += efi.memmap.desc_size;
614 	if (entry >= efi.memmap.map_end)
615 		return NULL;
616 
617 	return entry;
618 }
619 
620 static bool should_map_region(efi_memory_desc_t *md)
621 {
622 	/*
623 	 * Runtime regions always require runtime mappings (obviously).
624 	 */
625 	if (md->attribute & EFI_MEMORY_RUNTIME)
626 		return true;
627 
628 	/*
629 	 * 32-bit EFI doesn't suffer from the bug that requires us to
630 	 * reserve boot services regions, and mixed mode support
631 	 * doesn't exist for 32-bit kernels.
632 	 */
633 	if (IS_ENABLED(CONFIG_X86_32))
634 		return false;
635 
636 	/*
637 	 * EFI specific purpose memory may be reserved by default
638 	 * depending on kernel config and boot options.
639 	 */
640 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
641 	    efi_soft_reserve_enabled() &&
642 	    (md->attribute & EFI_MEMORY_SP))
643 		return false;
644 
645 	/*
646 	 * Map all of RAM so that we can access arguments in the 1:1
647 	 * mapping when making EFI runtime calls.
648 	 */
649 	if (efi_is_mixed()) {
650 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
651 		    md->type == EFI_LOADER_DATA ||
652 		    md->type == EFI_LOADER_CODE)
653 			return true;
654 	}
655 
656 	/*
657 	 * Map boot services regions as a workaround for buggy
658 	 * firmware that accesses them even when they shouldn't.
659 	 *
660 	 * See efi_{reserve,free}_boot_services().
661 	 */
662 	if (md->type == EFI_BOOT_SERVICES_CODE ||
663 	    md->type == EFI_BOOT_SERVICES_DATA)
664 		return true;
665 
666 	return false;
667 }
668 
669 /*
670  * Map the efi memory ranges of the runtime services and update new_mmap with
671  * virtual addresses.
672  */
673 static void * __init efi_map_regions(int *count, int *pg_shift)
674 {
675 	void *p, *new_memmap = NULL;
676 	unsigned long left = 0;
677 	unsigned long desc_size;
678 	efi_memory_desc_t *md;
679 
680 	desc_size = efi.memmap.desc_size;
681 
682 	p = NULL;
683 	while ((p = efi_map_next_entry(p))) {
684 		md = p;
685 
686 		if (!should_map_region(md))
687 			continue;
688 
689 		efi_map_region(md);
690 
691 		if (left < desc_size) {
692 			new_memmap = realloc_pages(new_memmap, *pg_shift);
693 			if (!new_memmap)
694 				return NULL;
695 
696 			left += PAGE_SIZE << *pg_shift;
697 			(*pg_shift)++;
698 		}
699 
700 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
701 
702 		left -= desc_size;
703 		(*count)++;
704 	}
705 
706 	return new_memmap;
707 }
708 
709 static void __init kexec_enter_virtual_mode(void)
710 {
711 #ifdef CONFIG_KEXEC_CORE
712 	efi_memory_desc_t *md;
713 	unsigned int num_pages;
714 
715 	/*
716 	 * We don't do virtual mode, since we don't do runtime services, on
717 	 * non-native EFI.
718 	 */
719 	if (efi_is_mixed()) {
720 		efi_memmap_unmap();
721 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
722 		return;
723 	}
724 
725 	if (efi_alloc_page_tables()) {
726 		pr_err("Failed to allocate EFI page tables\n");
727 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
728 		return;
729 	}
730 
731 	/*
732 	* Map efi regions which were passed via setup_data. The virt_addr is a
733 	* fixed addr which was used in first kernel of a kexec boot.
734 	*/
735 	for_each_efi_memory_desc(md)
736 		efi_map_region_fixed(md); /* FIXME: add error handling */
737 
738 	/*
739 	 * Unregister the early EFI memmap from efi_init() and install
740 	 * the new EFI memory map.
741 	 */
742 	efi_memmap_unmap();
743 
744 	if (efi_memmap_init_late(efi.memmap.phys_map,
745 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
746 		pr_err("Failed to remap late EFI memory map\n");
747 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
748 		return;
749 	}
750 
751 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
752 	num_pages >>= PAGE_SHIFT;
753 
754 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
755 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
756 		return;
757 	}
758 
759 	efi_sync_low_kernel_mappings();
760 	efi_native_runtime_setup();
761 #endif
762 }
763 
764 /*
765  * This function will switch the EFI runtime services to virtual mode.
766  * Essentially, we look through the EFI memmap and map every region that
767  * has the runtime attribute bit set in its memory descriptor into the
768  * efi_pgd page table.
769  *
770  * The new method does a pagetable switch in a preemption-safe manner
771  * so that we're in a different address space when calling a runtime
772  * function. For function arguments passing we do copy the PUDs of the
773  * kernel page table into efi_pgd prior to each call.
774  *
775  * Specially for kexec boot, efi runtime maps in previous kernel should
776  * be passed in via setup_data. In that case runtime ranges will be mapped
777  * to the same virtual addresses as the first kernel, see
778  * kexec_enter_virtual_mode().
779  */
780 static void __init __efi_enter_virtual_mode(void)
781 {
782 	int count = 0, pg_shift = 0;
783 	void *new_memmap = NULL;
784 	efi_status_t status;
785 	unsigned long pa;
786 
787 	if (efi_alloc_page_tables()) {
788 		pr_err("Failed to allocate EFI page tables\n");
789 		goto err;
790 	}
791 
792 	efi_merge_regions();
793 	new_memmap = efi_map_regions(&count, &pg_shift);
794 	if (!new_memmap) {
795 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
796 		goto err;
797 	}
798 
799 	pa = __pa(new_memmap);
800 
801 	/*
802 	 * Unregister the early EFI memmap from efi_init() and install
803 	 * the new EFI memory map that we are about to pass to the
804 	 * firmware via SetVirtualAddressMap().
805 	 */
806 	efi_memmap_unmap();
807 
808 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
809 		pr_err("Failed to remap late EFI memory map\n");
810 		goto err;
811 	}
812 
813 	if (efi_enabled(EFI_DBG)) {
814 		pr_info("EFI runtime memory map:\n");
815 		efi_print_memmap();
816 	}
817 
818 	if (efi_setup_page_tables(pa, 1 << pg_shift))
819 		goto err;
820 
821 	efi_sync_low_kernel_mappings();
822 
823 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
824 					     efi.memmap.desc_size,
825 					     efi.memmap.desc_version,
826 					     (efi_memory_desc_t *)pa,
827 					     efi_systab_phys);
828 	if (status != EFI_SUCCESS) {
829 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
830 		       status);
831 		goto err;
832 	}
833 
834 	efi_check_for_embedded_firmwares();
835 	efi_free_boot_services();
836 
837 	if (!efi_is_mixed())
838 		efi_native_runtime_setup();
839 	else
840 		efi_thunk_runtime_setup();
841 
842 	/*
843 	 * Apply more restrictive page table mapping attributes now that
844 	 * SVAM() has been called and the firmware has performed all
845 	 * necessary relocation fixups for the new virtual addresses.
846 	 */
847 	efi_runtime_update_mappings();
848 
849 	/* clean DUMMY object */
850 	efi_delete_dummy_variable();
851 	return;
852 
853 err:
854 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
855 }
856 
857 void __init efi_enter_virtual_mode(void)
858 {
859 	if (efi_enabled(EFI_PARAVIRT))
860 		return;
861 
862 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
863 
864 	if (efi_setup)
865 		kexec_enter_virtual_mode();
866 	else
867 		__efi_enter_virtual_mode();
868 
869 	efi_dump_pagetable();
870 }
871 
872 bool efi_is_table_address(unsigned long phys_addr)
873 {
874 	unsigned int i;
875 
876 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
877 		return false;
878 
879 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
880 		if (*(efi_tables[i]) == phys_addr)
881 			return true;
882 
883 	return false;
884 }
885 
886 char *efi_systab_show_arch(char *str)
887 {
888 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
889 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
890 	return str;
891 }
892 
893 #define EFI_FIELD(var) efi_ ## var
894 
895 #define EFI_ATTR_SHOW(name) \
896 static ssize_t name##_show(struct kobject *kobj, \
897 				struct kobj_attribute *attr, char *buf) \
898 { \
899 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
900 }
901 
902 EFI_ATTR_SHOW(fw_vendor);
903 EFI_ATTR_SHOW(runtime);
904 EFI_ATTR_SHOW(config_table);
905 
906 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
907 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
908 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
909 
910 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
911 {
912 	if (attr == &efi_attr_fw_vendor.attr) {
913 		if (efi_enabled(EFI_PARAVIRT) ||
914 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
915 			return 0;
916 	} else if (attr == &efi_attr_runtime.attr) {
917 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
918 			return 0;
919 	} else if (attr == &efi_attr_config_table.attr) {
920 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
921 			return 0;
922 	}
923 	return attr->mode;
924 }
925