xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision fb574682)
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/set_memory.h>
53 #include <asm/tlbflush.h>
54 #include <asm/x86_init.h>
55 #include <asm/uv/uv.h>
56 
57 static unsigned long efi_systab_phys __initdata;
58 static unsigned long prop_phys = EFI_INVALID_TABLE_ADDR;
59 static unsigned long uga_phys = EFI_INVALID_TABLE_ADDR;
60 static unsigned long efi_runtime, efi_nr_tables;
61 
62 unsigned long efi_fw_vendor, efi_config_table;
63 
64 static const efi_config_table_type_t arch_tables[] __initconst = {
65 	{EFI_PROPERTIES_TABLE_GUID,	&prop_phys,		"PROP"		},
66 	{UGA_IO_PROTOCOL_GUID,		&uga_phys,		"UGA"		},
67 #ifdef CONFIG_X86_UV
68 	{UV_SYSTEM_TABLE_GUID,		&uv_systab_phys,	"UVsystab"	},
69 #endif
70 	{},
71 };
72 
73 static const unsigned long * const efi_tables[] = {
74 	&efi.acpi,
75 	&efi.acpi20,
76 	&efi.smbios,
77 	&efi.smbios3,
78 	&uga_phys,
79 #ifdef CONFIG_X86_UV
80 	&uv_systab_phys,
81 #endif
82 	&efi_fw_vendor,
83 	&efi_runtime,
84 	&efi_config_table,
85 	&efi.esrt,
86 	&prop_phys,
87 	&efi_mem_attr_table,
88 #ifdef CONFIG_EFI_RCI2_TABLE
89 	&rci2_table_phys,
90 #endif
91 	&efi.tpm_log,
92 	&efi.tpm_final_log,
93 	&efi_rng_seed,
94 };
95 
96 u64 efi_setup;		/* efi setup_data physical address */
97 
98 static int add_efi_memmap __initdata;
99 static int __init setup_add_efi_memmap(char *arg)
100 {
101 	add_efi_memmap = 1;
102 	return 0;
103 }
104 early_param("add_efi_memmap", setup_add_efi_memmap);
105 
106 void __init efi_find_mirror(void)
107 {
108 	efi_memory_desc_t *md;
109 	u64 mirror_size = 0, total_size = 0;
110 
111 	if (!efi_enabled(EFI_MEMMAP))
112 		return;
113 
114 	for_each_efi_memory_desc(md) {
115 		unsigned long long start = md->phys_addr;
116 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
117 
118 		total_size += size;
119 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
120 			memblock_mark_mirror(start, size);
121 			mirror_size += size;
122 		}
123 	}
124 	if (mirror_size)
125 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
126 			mirror_size>>20, total_size>>20);
127 }
128 
129 /*
130  * Tell the kernel about the EFI memory map.  This might include
131  * more than the max 128 entries that can fit in the passed in e820
132  * legacy (zeropage) memory map, but the kernel's e820 table can hold
133  * E820_MAX_ENTRIES.
134  */
135 
136 static void __init do_add_efi_memmap(void)
137 {
138 	efi_memory_desc_t *md;
139 
140 	if (!efi_enabled(EFI_MEMMAP))
141 		return;
142 
143 	for_each_efi_memory_desc(md) {
144 		unsigned long long start = md->phys_addr;
145 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
146 		int e820_type;
147 
148 		switch (md->type) {
149 		case EFI_LOADER_CODE:
150 		case EFI_LOADER_DATA:
151 		case EFI_BOOT_SERVICES_CODE:
152 		case EFI_BOOT_SERVICES_DATA:
153 		case EFI_CONVENTIONAL_MEMORY:
154 			if (efi_soft_reserve_enabled()
155 			    && (md->attribute & EFI_MEMORY_SP))
156 				e820_type = E820_TYPE_SOFT_RESERVED;
157 			else if (md->attribute & EFI_MEMORY_WB)
158 				e820_type = E820_TYPE_RAM;
159 			else
160 				e820_type = E820_TYPE_RESERVED;
161 			break;
162 		case EFI_ACPI_RECLAIM_MEMORY:
163 			e820_type = E820_TYPE_ACPI;
164 			break;
165 		case EFI_ACPI_MEMORY_NVS:
166 			e820_type = E820_TYPE_NVS;
167 			break;
168 		case EFI_UNUSABLE_MEMORY:
169 			e820_type = E820_TYPE_UNUSABLE;
170 			break;
171 		case EFI_PERSISTENT_MEMORY:
172 			e820_type = E820_TYPE_PMEM;
173 			break;
174 		default:
175 			/*
176 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
177 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
178 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
179 			 */
180 			e820_type = E820_TYPE_RESERVED;
181 			break;
182 		}
183 
184 		e820__range_add(start, size, e820_type);
185 	}
186 	e820__update_table(e820_table);
187 }
188 
189 /*
190  * Given add_efi_memmap defaults to 0 and there there is no alternative
191  * e820 mechanism for soft-reserved memory, import the full EFI memory
192  * map if soft reservations are present and enabled. Otherwise, the
193  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
194  * the efi=nosoftreserve option.
195  */
196 static bool do_efi_soft_reserve(void)
197 {
198 	efi_memory_desc_t *md;
199 
200 	if (!efi_enabled(EFI_MEMMAP))
201 		return false;
202 
203 	if (!efi_soft_reserve_enabled())
204 		return false;
205 
206 	for_each_efi_memory_desc(md)
207 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
208 		    (md->attribute & EFI_MEMORY_SP))
209 			return true;
210 	return false;
211 }
212 
213 int __init efi_memblock_x86_reserve_range(void)
214 {
215 	struct efi_info *e = &boot_params.efi_info;
216 	struct efi_memory_map_data data;
217 	phys_addr_t pmap;
218 	int rv;
219 
220 	if (efi_enabled(EFI_PARAVIRT))
221 		return 0;
222 
223 	/* Can't handle firmware tables above 4GB on i386 */
224 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
225 		pr_err("Memory map is above 4GB, disabling EFI.\n");
226 		return -EINVAL;
227 	}
228 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
229 
230 	data.phys_map		= pmap;
231 	data.size 		= e->efi_memmap_size;
232 	data.desc_size		= e->efi_memdesc_size;
233 	data.desc_version	= e->efi_memdesc_version;
234 
235 	rv = efi_memmap_init_early(&data);
236 	if (rv)
237 		return rv;
238 
239 	if (add_efi_memmap || do_efi_soft_reserve())
240 		do_add_efi_memmap();
241 
242 	efi_fake_memmap_early();
243 
244 	WARN(efi.memmap.desc_version != 1,
245 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
246 	     efi.memmap.desc_version);
247 
248 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
249 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
250 
251 	return 0;
252 }
253 
254 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
255 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
256 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
257 
258 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
259 {
260 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
261 	u64 end_hi = 0;
262 	char buf[64];
263 
264 	if (md->num_pages == 0) {
265 		end = 0;
266 	} else if (md->num_pages > EFI_PAGES_MAX ||
267 		   EFI_PAGES_MAX - md->num_pages <
268 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
269 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
270 			>> OVERFLOW_ADDR_SHIFT;
271 
272 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
273 			end_hi += 1;
274 	} else {
275 		return true;
276 	}
277 
278 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
279 
280 	if (end_hi) {
281 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
282 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
283 			md->phys_addr, end_hi, end);
284 	} else {
285 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
286 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
287 			md->phys_addr, end);
288 	}
289 	return false;
290 }
291 
292 static void __init efi_clean_memmap(void)
293 {
294 	efi_memory_desc_t *out = efi.memmap.map;
295 	const efi_memory_desc_t *in = out;
296 	const efi_memory_desc_t *end = efi.memmap.map_end;
297 	int i, n_removal;
298 
299 	for (i = n_removal = 0; in < end; i++) {
300 		if (efi_memmap_entry_valid(in, i)) {
301 			if (out != in)
302 				memcpy(out, in, efi.memmap.desc_size);
303 			out = (void *)out + efi.memmap.desc_size;
304 		} else {
305 			n_removal++;
306 		}
307 		in = (void *)in + efi.memmap.desc_size;
308 	}
309 
310 	if (n_removal > 0) {
311 		struct efi_memory_map_data data = {
312 			.phys_map	= efi.memmap.phys_map,
313 			.desc_version	= efi.memmap.desc_version,
314 			.desc_size	= efi.memmap.desc_size,
315 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
316 			.flags		= 0,
317 		};
318 
319 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
320 		efi_memmap_install(&data);
321 	}
322 }
323 
324 void __init efi_print_memmap(void)
325 {
326 	efi_memory_desc_t *md;
327 	int i = 0;
328 
329 	for_each_efi_memory_desc(md) {
330 		char buf[64];
331 
332 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
333 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
334 			md->phys_addr,
335 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
336 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
337 	}
338 }
339 
340 static int __init efi_systab_init(unsigned long phys)
341 {
342 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
343 					  : sizeof(efi_system_table_32_t);
344 	const efi_table_hdr_t *hdr;
345 	bool over4g = false;
346 	void *p;
347 	int ret;
348 
349 	hdr = p = early_memremap_ro(phys, size);
350 	if (p == NULL) {
351 		pr_err("Couldn't map the system table!\n");
352 		return -ENOMEM;
353 	}
354 
355 	ret = efi_systab_check_header(hdr, 1);
356 	if (ret) {
357 		early_memunmap(p, size);
358 		return ret;
359 	}
360 
361 	if (efi_enabled(EFI_64BIT)) {
362 		const efi_system_table_64_t *systab64 = p;
363 
364 		efi_runtime	= systab64->runtime;
365 		over4g		= systab64->runtime > U32_MAX;
366 
367 		if (efi_setup) {
368 			struct efi_setup_data *data;
369 
370 			data = early_memremap_ro(efi_setup, sizeof(*data));
371 			if (!data) {
372 				early_memunmap(p, size);
373 				return -ENOMEM;
374 			}
375 
376 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
377 			efi_config_table	= (unsigned long)data->tables;
378 
379 			over4g |= data->fw_vendor	> U32_MAX ||
380 				  data->tables		> U32_MAX;
381 
382 			early_memunmap(data, sizeof(*data));
383 		} else {
384 			efi_fw_vendor		= systab64->fw_vendor;
385 			efi_config_table	= systab64->tables;
386 
387 			over4g |= systab64->fw_vendor	> U32_MAX ||
388 				  systab64->tables	> U32_MAX;
389 		}
390 		efi_nr_tables = systab64->nr_tables;
391 	} else {
392 		const efi_system_table_32_t *systab32 = p;
393 
394 		efi_fw_vendor		= systab32->fw_vendor;
395 		efi_runtime		= systab32->runtime;
396 		efi_config_table	= systab32->tables;
397 		efi_nr_tables		= systab32->nr_tables;
398 	}
399 
400 	efi.runtime_version = hdr->revision;
401 
402 	efi_systab_report_header(hdr, efi_fw_vendor);
403 	early_memunmap(p, size);
404 
405 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
406 		pr_err("EFI data located above 4GB, disabling EFI.\n");
407 		return -EINVAL;
408 	}
409 
410 	return 0;
411 }
412 
413 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
414 {
415 	void *config_tables;
416 	int sz, ret;
417 
418 	if (efi_nr_tables == 0)
419 		return 0;
420 
421 	if (efi_enabled(EFI_64BIT))
422 		sz = sizeof(efi_config_table_64_t);
423 	else
424 		sz = sizeof(efi_config_table_32_t);
425 
426 	/*
427 	 * Let's see what config tables the firmware passed to us.
428 	 */
429 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
430 	if (config_tables == NULL) {
431 		pr_err("Could not map Configuration table!\n");
432 		return -ENOMEM;
433 	}
434 
435 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
436 				      arch_tables);
437 
438 	early_memunmap(config_tables, efi_nr_tables * sz);
439 	return ret;
440 }
441 
442 void __init efi_init(void)
443 {
444 	if (IS_ENABLED(CONFIG_X86_32) &&
445 	    (boot_params.efi_info.efi_systab_hi ||
446 	     boot_params.efi_info.efi_memmap_hi)) {
447 		pr_info("Table located above 4GB, disabling EFI.\n");
448 		return;
449 	}
450 
451 	efi_systab_phys = boot_params.efi_info.efi_systab |
452 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
453 
454 	if (efi_systab_init(efi_systab_phys))
455 		return;
456 
457 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
458 		return;
459 
460 	if (efi_config_init(arch_tables))
461 		return;
462 
463 	/*
464 	 * Note: We currently don't support runtime services on an EFI
465 	 * that doesn't match the kernel 32/64-bit mode.
466 	 */
467 
468 	if (!efi_runtime_supported())
469 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
470 
471 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
472 		efi_memmap_unmap();
473 		return;
474 	}
475 
476 	/* Parse the EFI Properties table if it exists */
477 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
478 		efi_properties_table_t *tbl;
479 
480 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
481 		if (tbl == NULL) {
482 			pr_err("Could not map Properties table!\n");
483 		} else {
484 			if (tbl->memory_protection_attribute &
485 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
486 				set_bit(EFI_NX_PE_DATA, &efi.flags);
487 
488 			early_memunmap(tbl, sizeof(*tbl));
489 		}
490 	}
491 
492 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
493 	efi_clean_memmap();
494 
495 	if (efi_enabled(EFI_DBG))
496 		efi_print_memmap();
497 }
498 
499 #if defined(CONFIG_X86_32) || defined(CONFIG_X86_UV)
500 
501 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
502 {
503 	u64 addr, npages;
504 
505 	addr = md->virt_addr;
506 	npages = md->num_pages;
507 
508 	memrange_efi_to_native(&addr, &npages);
509 
510 	if (executable)
511 		set_memory_x(addr, npages);
512 	else
513 		set_memory_nx(addr, npages);
514 }
515 
516 void __init runtime_code_page_mkexec(void)
517 {
518 	efi_memory_desc_t *md;
519 
520 	/* Make EFI runtime service code area executable */
521 	for_each_efi_memory_desc(md) {
522 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
523 			continue;
524 
525 		efi_set_executable(md, true);
526 	}
527 }
528 
529 void __init efi_memory_uc(u64 addr, unsigned long size)
530 {
531 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
532 	u64 npages;
533 
534 	npages = round_up(size, page_shift) / page_shift;
535 	memrange_efi_to_native(&addr, &npages);
536 	set_memory_uc(addr, npages);
537 }
538 
539 void __init old_map_region(efi_memory_desc_t *md)
540 {
541 	u64 start_pfn, end_pfn, end;
542 	unsigned long size;
543 	void *va;
544 
545 	start_pfn = PFN_DOWN(md->phys_addr);
546 	size	  = md->num_pages << PAGE_SHIFT;
547 	end	  = md->phys_addr + size;
548 	end_pfn   = PFN_UP(end);
549 
550 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
551 		va = __va(md->phys_addr);
552 
553 		if (!(md->attribute & EFI_MEMORY_WB))
554 			efi_memory_uc((u64)(unsigned long)va, size);
555 	} else
556 		va = efi_ioremap(md->phys_addr, size,
557 				 md->type, md->attribute);
558 
559 	md->virt_addr = (u64) (unsigned long) va;
560 	if (!va)
561 		pr_err("ioremap of 0x%llX failed!\n",
562 		       (unsigned long long)md->phys_addr);
563 }
564 
565 #endif
566 
567 /* Merge contiguous regions of the same type and attribute */
568 static void __init efi_merge_regions(void)
569 {
570 	efi_memory_desc_t *md, *prev_md = NULL;
571 
572 	for_each_efi_memory_desc(md) {
573 		u64 prev_size;
574 
575 		if (!prev_md) {
576 			prev_md = md;
577 			continue;
578 		}
579 
580 		if (prev_md->type != md->type ||
581 		    prev_md->attribute != md->attribute) {
582 			prev_md = md;
583 			continue;
584 		}
585 
586 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
587 
588 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
589 			prev_md->num_pages += md->num_pages;
590 			md->type = EFI_RESERVED_TYPE;
591 			md->attribute = 0;
592 			continue;
593 		}
594 		prev_md = md;
595 	}
596 }
597 
598 static void *realloc_pages(void *old_memmap, int old_shift)
599 {
600 	void *ret;
601 
602 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
603 	if (!ret)
604 		goto out;
605 
606 	/*
607 	 * A first-time allocation doesn't have anything to copy.
608 	 */
609 	if (!old_memmap)
610 		return ret;
611 
612 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
613 
614 out:
615 	free_pages((unsigned long)old_memmap, old_shift);
616 	return ret;
617 }
618 
619 /*
620  * Iterate the EFI memory map in reverse order because the regions
621  * will be mapped top-down. The end result is the same as if we had
622  * mapped things forward, but doesn't require us to change the
623  * existing implementation of efi_map_region().
624  */
625 static inline void *efi_map_next_entry_reverse(void *entry)
626 {
627 	/* Initial call */
628 	if (!entry)
629 		return efi.memmap.map_end - efi.memmap.desc_size;
630 
631 	entry -= efi.memmap.desc_size;
632 	if (entry < efi.memmap.map)
633 		return NULL;
634 
635 	return entry;
636 }
637 
638 /*
639  * efi_map_next_entry - Return the next EFI memory map descriptor
640  * @entry: Previous EFI memory map descriptor
641  *
642  * This is a helper function to iterate over the EFI memory map, which
643  * we do in different orders depending on the current configuration.
644  *
645  * To begin traversing the memory map @entry must be %NULL.
646  *
647  * Returns %NULL when we reach the end of the memory map.
648  */
649 static void *efi_map_next_entry(void *entry)
650 {
651 	if (!efi_have_uv1_memmap() && efi_enabled(EFI_64BIT)) {
652 		/*
653 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
654 		 * config table feature requires us to map all entries
655 		 * in the same order as they appear in the EFI memory
656 		 * map. That is to say, entry N must have a lower
657 		 * virtual address than entry N+1. This is because the
658 		 * firmware toolchain leaves relative references in
659 		 * the code/data sections, which are split and become
660 		 * separate EFI memory regions. Mapping things
661 		 * out-of-order leads to the firmware accessing
662 		 * unmapped addresses.
663 		 *
664 		 * Since we need to map things this way whether or not
665 		 * the kernel actually makes use of
666 		 * EFI_PROPERTIES_TABLE, let's just switch to this
667 		 * scheme by default for 64-bit.
668 		 */
669 		return efi_map_next_entry_reverse(entry);
670 	}
671 
672 	/* Initial call */
673 	if (!entry)
674 		return efi.memmap.map;
675 
676 	entry += efi.memmap.desc_size;
677 	if (entry >= efi.memmap.map_end)
678 		return NULL;
679 
680 	return entry;
681 }
682 
683 static bool should_map_region(efi_memory_desc_t *md)
684 {
685 	/*
686 	 * Runtime regions always require runtime mappings (obviously).
687 	 */
688 	if (md->attribute & EFI_MEMORY_RUNTIME)
689 		return true;
690 
691 	/*
692 	 * 32-bit EFI doesn't suffer from the bug that requires us to
693 	 * reserve boot services regions, and mixed mode support
694 	 * doesn't exist for 32-bit kernels.
695 	 */
696 	if (IS_ENABLED(CONFIG_X86_32))
697 		return false;
698 
699 	/*
700 	 * EFI specific purpose memory may be reserved by default
701 	 * depending on kernel config and boot options.
702 	 */
703 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
704 	    efi_soft_reserve_enabled() &&
705 	    (md->attribute & EFI_MEMORY_SP))
706 		return false;
707 
708 	/*
709 	 * Map all of RAM so that we can access arguments in the 1:1
710 	 * mapping when making EFI runtime calls.
711 	 */
712 	if (efi_is_mixed()) {
713 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
714 		    md->type == EFI_LOADER_DATA ||
715 		    md->type == EFI_LOADER_CODE)
716 			return true;
717 	}
718 
719 	/*
720 	 * Map boot services regions as a workaround for buggy
721 	 * firmware that accesses them even when they shouldn't.
722 	 *
723 	 * See efi_{reserve,free}_boot_services().
724 	 */
725 	if (md->type == EFI_BOOT_SERVICES_CODE ||
726 	    md->type == EFI_BOOT_SERVICES_DATA)
727 		return true;
728 
729 	return false;
730 }
731 
732 /*
733  * Map the efi memory ranges of the runtime services and update new_mmap with
734  * virtual addresses.
735  */
736 static void * __init efi_map_regions(int *count, int *pg_shift)
737 {
738 	void *p, *new_memmap = NULL;
739 	unsigned long left = 0;
740 	unsigned long desc_size;
741 	efi_memory_desc_t *md;
742 
743 	desc_size = efi.memmap.desc_size;
744 
745 	p = NULL;
746 	while ((p = efi_map_next_entry(p))) {
747 		md = p;
748 
749 		if (!should_map_region(md))
750 			continue;
751 
752 		efi_map_region(md);
753 
754 		if (left < desc_size) {
755 			new_memmap = realloc_pages(new_memmap, *pg_shift);
756 			if (!new_memmap)
757 				return NULL;
758 
759 			left += PAGE_SIZE << *pg_shift;
760 			(*pg_shift)++;
761 		}
762 
763 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
764 
765 		left -= desc_size;
766 		(*count)++;
767 	}
768 
769 	return new_memmap;
770 }
771 
772 static void __init kexec_enter_virtual_mode(void)
773 {
774 #ifdef CONFIG_KEXEC_CORE
775 	efi_memory_desc_t *md;
776 	unsigned int num_pages;
777 
778 	/*
779 	 * We don't do virtual mode, since we don't do runtime services, on
780 	 * non-native EFI. With the UV1 memmap, we don't do runtime services in
781 	 * kexec kernel because in the initial boot something else might
782 	 * have been mapped at these virtual addresses.
783 	 */
784 	if (efi_is_mixed() || efi_have_uv1_memmap()) {
785 		efi_memmap_unmap();
786 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
787 		return;
788 	}
789 
790 	if (efi_alloc_page_tables()) {
791 		pr_err("Failed to allocate EFI page tables\n");
792 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
793 		return;
794 	}
795 
796 	/*
797 	* Map efi regions which were passed via setup_data. The virt_addr is a
798 	* fixed addr which was used in first kernel of a kexec boot.
799 	*/
800 	for_each_efi_memory_desc(md)
801 		efi_map_region_fixed(md); /* FIXME: add error handling */
802 
803 	/*
804 	 * Unregister the early EFI memmap from efi_init() and install
805 	 * the new EFI memory map.
806 	 */
807 	efi_memmap_unmap();
808 
809 	if (efi_memmap_init_late(efi.memmap.phys_map,
810 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
811 		pr_err("Failed to remap late EFI memory map\n");
812 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
813 		return;
814 	}
815 
816 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
817 	num_pages >>= PAGE_SHIFT;
818 
819 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
820 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
821 		return;
822 	}
823 
824 	efi_sync_low_kernel_mappings();
825 	efi_native_runtime_setup();
826 #endif
827 }
828 
829 /*
830  * This function will switch the EFI runtime services to virtual mode.
831  * Essentially, we look through the EFI memmap and map every region that
832  * has the runtime attribute bit set in its memory descriptor into the
833  * efi_pgd page table.
834  *
835  * The old method which used to update that memory descriptor with the
836  * virtual address obtained from ioremap() is still supported when the
837  * kernel is booted on SG1 UV1 hardware. Same old method enabled the
838  * runtime services to be called without having to thunk back into
839  * physical mode for every invocation.
840  *
841  * The new method does a pagetable switch in a preemption-safe manner
842  * so that we're in a different address space when calling a runtime
843  * function. For function arguments passing we do copy the PUDs of the
844  * kernel page table into efi_pgd prior to each call.
845  *
846  * Specially for kexec boot, efi runtime maps in previous kernel should
847  * be passed in via setup_data. In that case runtime ranges will be mapped
848  * to the same virtual addresses as the first kernel, see
849  * kexec_enter_virtual_mode().
850  */
851 static void __init __efi_enter_virtual_mode(void)
852 {
853 	int count = 0, pg_shift = 0;
854 	void *new_memmap = NULL;
855 	efi_status_t status;
856 	unsigned long pa;
857 
858 	if (efi_alloc_page_tables()) {
859 		pr_err("Failed to allocate EFI page tables\n");
860 		goto err;
861 	}
862 
863 	efi_merge_regions();
864 	new_memmap = efi_map_regions(&count, &pg_shift);
865 	if (!new_memmap) {
866 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
867 		goto err;
868 	}
869 
870 	pa = __pa(new_memmap);
871 
872 	/*
873 	 * Unregister the early EFI memmap from efi_init() and install
874 	 * the new EFI memory map that we are about to pass to the
875 	 * firmware via SetVirtualAddressMap().
876 	 */
877 	efi_memmap_unmap();
878 
879 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
880 		pr_err("Failed to remap late EFI memory map\n");
881 		goto err;
882 	}
883 
884 	if (efi_enabled(EFI_DBG)) {
885 		pr_info("EFI runtime memory map:\n");
886 		efi_print_memmap();
887 	}
888 
889 	if (efi_setup_page_tables(pa, 1 << pg_shift))
890 		goto err;
891 
892 	efi_sync_low_kernel_mappings();
893 
894 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
895 					     efi.memmap.desc_size,
896 					     efi.memmap.desc_version,
897 					     (efi_memory_desc_t *)pa,
898 					     efi_systab_phys);
899 	if (status != EFI_SUCCESS) {
900 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
901 		       status);
902 		goto err;
903 	}
904 
905 	efi_check_for_embedded_firmwares();
906 	efi_free_boot_services();
907 
908 	if (!efi_is_mixed())
909 		efi_native_runtime_setup();
910 	else
911 		efi_thunk_runtime_setup();
912 
913 	/*
914 	 * Apply more restrictive page table mapping attributes now that
915 	 * SVAM() has been called and the firmware has performed all
916 	 * necessary relocation fixups for the new virtual addresses.
917 	 */
918 	efi_runtime_update_mappings();
919 
920 	/* clean DUMMY object */
921 	efi_delete_dummy_variable();
922 	return;
923 
924 err:
925 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
926 }
927 
928 void __init efi_enter_virtual_mode(void)
929 {
930 	if (efi_enabled(EFI_PARAVIRT))
931 		return;
932 
933 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
934 
935 	if (efi_setup)
936 		kexec_enter_virtual_mode();
937 	else
938 		__efi_enter_virtual_mode();
939 
940 	efi_dump_pagetable();
941 }
942 
943 bool efi_is_table_address(unsigned long phys_addr)
944 {
945 	unsigned int i;
946 
947 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
948 		return false;
949 
950 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
951 		if (*(efi_tables[i]) == phys_addr)
952 			return true;
953 
954 	return false;
955 }
956 
957 char *efi_systab_show_arch(char *str)
958 {
959 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
960 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
961 	return str;
962 }
963 
964 #define EFI_FIELD(var) efi_ ## var
965 
966 #define EFI_ATTR_SHOW(name) \
967 static ssize_t name##_show(struct kobject *kobj, \
968 				struct kobj_attribute *attr, char *buf) \
969 { \
970 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
971 }
972 
973 EFI_ATTR_SHOW(fw_vendor);
974 EFI_ATTR_SHOW(runtime);
975 EFI_ATTR_SHOW(config_table);
976 
977 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
978 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
979 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
980 
981 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
982 {
983 	if (attr == &efi_attr_fw_vendor.attr) {
984 		if (efi_enabled(EFI_PARAVIRT) ||
985 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
986 			return 0;
987 	} else if (attr == &efi_attr_runtime.attr) {
988 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
989 			return 0;
990 	} else if (attr == &efi_attr_config_table.attr) {
991 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
992 			return 0;
993 	}
994 	return attr->mode;
995 }
996