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 #ifdef CONFIG_UNACCEPTED_MEMORY
100 	&efi.unaccepted,
101 #endif
102 };
103 
104 u64 efi_setup;		/* efi setup_data physical address */
105 
106 static int add_efi_memmap __initdata;
setup_add_efi_memmap(char * arg)107 static int __init setup_add_efi_memmap(char *arg)
108 {
109 	add_efi_memmap = 1;
110 	return 0;
111 }
112 early_param("add_efi_memmap", setup_add_efi_memmap);
113 
114 /*
115  * Tell the kernel about the EFI memory map.  This might include
116  * more than the max 128 entries that can fit in the passed in e820
117  * legacy (zeropage) memory map, but the kernel's e820 table can hold
118  * E820_MAX_ENTRIES.
119  */
120 
do_add_efi_memmap(void)121 static void __init do_add_efi_memmap(void)
122 {
123 	efi_memory_desc_t *md;
124 
125 	if (!efi_enabled(EFI_MEMMAP))
126 		return;
127 
128 	for_each_efi_memory_desc(md) {
129 		unsigned long long start = md->phys_addr;
130 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
131 		int e820_type;
132 
133 		switch (md->type) {
134 		case EFI_LOADER_CODE:
135 		case EFI_LOADER_DATA:
136 		case EFI_BOOT_SERVICES_CODE:
137 		case EFI_BOOT_SERVICES_DATA:
138 		case EFI_CONVENTIONAL_MEMORY:
139 			if (efi_soft_reserve_enabled()
140 			    && (md->attribute & EFI_MEMORY_SP))
141 				e820_type = E820_TYPE_SOFT_RESERVED;
142 			else if (md->attribute & EFI_MEMORY_WB)
143 				e820_type = E820_TYPE_RAM;
144 			else
145 				e820_type = E820_TYPE_RESERVED;
146 			break;
147 		case EFI_ACPI_RECLAIM_MEMORY:
148 			e820_type = E820_TYPE_ACPI;
149 			break;
150 		case EFI_ACPI_MEMORY_NVS:
151 			e820_type = E820_TYPE_NVS;
152 			break;
153 		case EFI_UNUSABLE_MEMORY:
154 			e820_type = E820_TYPE_UNUSABLE;
155 			break;
156 		case EFI_PERSISTENT_MEMORY:
157 			e820_type = E820_TYPE_PMEM;
158 			break;
159 		default:
160 			/*
161 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
162 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
163 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
164 			 */
165 			e820_type = E820_TYPE_RESERVED;
166 			break;
167 		}
168 
169 		e820__range_add(start, size, e820_type);
170 	}
171 	e820__update_table(e820_table);
172 }
173 
174 /*
175  * Given add_efi_memmap defaults to 0 and there is no alternative
176  * e820 mechanism for soft-reserved memory, import the full EFI memory
177  * map if soft reservations are present and enabled. Otherwise, the
178  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
179  * the efi=nosoftreserve option.
180  */
do_efi_soft_reserve(void)181 static bool do_efi_soft_reserve(void)
182 {
183 	efi_memory_desc_t *md;
184 
185 	if (!efi_enabled(EFI_MEMMAP))
186 		return false;
187 
188 	if (!efi_soft_reserve_enabled())
189 		return false;
190 
191 	for_each_efi_memory_desc(md)
192 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
193 		    (md->attribute & EFI_MEMORY_SP))
194 			return true;
195 	return false;
196 }
197 
efi_memblock_x86_reserve_range(void)198 int __init efi_memblock_x86_reserve_range(void)
199 {
200 	struct efi_info *e = &boot_params.efi_info;
201 	struct efi_memory_map_data data;
202 	phys_addr_t pmap;
203 	int rv;
204 
205 	if (efi_enabled(EFI_PARAVIRT))
206 		return 0;
207 
208 	/* Can't handle firmware tables above 4GB on i386 */
209 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
210 		pr_err("Memory map is above 4GB, disabling EFI.\n");
211 		return -EINVAL;
212 	}
213 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
214 
215 	data.phys_map		= pmap;
216 	data.size 		= e->efi_memmap_size;
217 	data.desc_size		= e->efi_memdesc_size;
218 	data.desc_version	= e->efi_memdesc_version;
219 
220 	if (!efi_enabled(EFI_PARAVIRT)) {
221 		rv = efi_memmap_init_early(&data);
222 		if (rv)
223 			return rv;
224 	}
225 
226 	if (add_efi_memmap || do_efi_soft_reserve())
227 		do_add_efi_memmap();
228 
229 	efi_fake_memmap_early();
230 
231 	WARN(efi.memmap.desc_version != 1,
232 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
233 	     efi.memmap.desc_version);
234 
235 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
236 	set_bit(EFI_PRESERVE_BS_REGIONS, &efi.flags);
237 
238 	return 0;
239 }
240 
241 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
242 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
243 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
244 
efi_memmap_entry_valid(const efi_memory_desc_t * md,int i)245 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
246 {
247 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
248 	u64 end_hi = 0;
249 	char buf[64];
250 
251 	if (md->num_pages == 0) {
252 		end = 0;
253 	} else if (md->num_pages > EFI_PAGES_MAX ||
254 		   EFI_PAGES_MAX - md->num_pages <
255 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
256 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
257 			>> OVERFLOW_ADDR_SHIFT;
258 
259 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
260 			end_hi += 1;
261 	} else {
262 		return true;
263 	}
264 
265 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
266 
267 	if (end_hi) {
268 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
269 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
270 			md->phys_addr, end_hi, end);
271 	} else {
272 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
273 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
274 			md->phys_addr, end);
275 	}
276 	return false;
277 }
278 
efi_clean_memmap(void)279 static void __init efi_clean_memmap(void)
280 {
281 	efi_memory_desc_t *out = efi.memmap.map;
282 	const efi_memory_desc_t *in = out;
283 	const efi_memory_desc_t *end = efi.memmap.map_end;
284 	int i, n_removal;
285 
286 	for (i = n_removal = 0; in < end; i++) {
287 		if (efi_memmap_entry_valid(in, i)) {
288 			if (out != in)
289 				memcpy(out, in, efi.memmap.desc_size);
290 			out = (void *)out + efi.memmap.desc_size;
291 		} else {
292 			n_removal++;
293 		}
294 		in = (void *)in + efi.memmap.desc_size;
295 	}
296 
297 	if (n_removal > 0) {
298 		struct efi_memory_map_data data = {
299 			.phys_map	= efi.memmap.phys_map,
300 			.desc_version	= efi.memmap.desc_version,
301 			.desc_size	= efi.memmap.desc_size,
302 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
303 			.flags		= 0,
304 		};
305 
306 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
307 		efi_memmap_install(&data);
308 	}
309 }
310 
311 /*
312  * Firmware can use EfiMemoryMappedIO to request that MMIO regions be
313  * mapped by the OS so they can be accessed by EFI runtime services, but
314  * should have no other significance to the OS (UEFI r2.10, sec 7.2).
315  * However, most bootloaders and EFI stubs convert EfiMemoryMappedIO
316  * regions to E820_TYPE_RESERVED entries, which prevent Linux from
317  * allocating space from them (see remove_e820_regions()).
318  *
319  * Some platforms use EfiMemoryMappedIO entries for PCI MMCONFIG space and
320  * PCI host bridge windows, which means Linux can't allocate BAR space for
321  * hot-added devices.
322  *
323  * Remove large EfiMemoryMappedIO regions from the E820 map to avoid this
324  * problem.
325  *
326  * Retain small EfiMemoryMappedIO regions because on some platforms, these
327  * describe non-window space that's included in host bridge _CRS.  If we
328  * assign that space to PCI devices, they don't work.
329  */
efi_remove_e820_mmio(void)330 static void __init efi_remove_e820_mmio(void)
331 {
332 	efi_memory_desc_t *md;
333 	u64 size, start, end;
334 	int i = 0;
335 
336 	for_each_efi_memory_desc(md) {
337 		if (md->type == EFI_MEMORY_MAPPED_IO) {
338 			size = md->num_pages << EFI_PAGE_SHIFT;
339 			start = md->phys_addr;
340 			end = start + size - 1;
341 			if (size >= 256*1024) {
342 				pr_info("Remove mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluMB) from e820 map\n",
343 					i, start, end, size >> 20);
344 				e820__range_remove(start, size,
345 						   E820_TYPE_RESERVED, 1);
346 			} else {
347 				pr_info("Not removing mem%02u: MMIO range=[0x%08llx-0x%08llx] (%lluKB) from e820 map\n",
348 					i, start, end, size >> 10);
349 			}
350 		}
351 		i++;
352 	}
353 }
354 
efi_print_memmap(void)355 void __init efi_print_memmap(void)
356 {
357 	efi_memory_desc_t *md;
358 	int i = 0;
359 
360 	for_each_efi_memory_desc(md) {
361 		char buf[64];
362 
363 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
364 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
365 			md->phys_addr,
366 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
367 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
368 	}
369 }
370 
efi_systab_init(unsigned long phys)371 static int __init efi_systab_init(unsigned long phys)
372 {
373 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
374 					  : sizeof(efi_system_table_32_t);
375 	const efi_table_hdr_t *hdr;
376 	bool over4g = false;
377 	void *p;
378 	int ret;
379 
380 	hdr = p = early_memremap_ro(phys, size);
381 	if (p == NULL) {
382 		pr_err("Couldn't map the system table!\n");
383 		return -ENOMEM;
384 	}
385 
386 	ret = efi_systab_check_header(hdr);
387 	if (ret) {
388 		early_memunmap(p, size);
389 		return ret;
390 	}
391 
392 	if (efi_enabled(EFI_64BIT)) {
393 		const efi_system_table_64_t *systab64 = p;
394 
395 		efi_runtime	= systab64->runtime;
396 		over4g		= systab64->runtime > U32_MAX;
397 
398 		if (efi_setup) {
399 			struct efi_setup_data *data;
400 
401 			data = early_memremap_ro(efi_setup, sizeof(*data));
402 			if (!data) {
403 				early_memunmap(p, size);
404 				return -ENOMEM;
405 			}
406 
407 			efi_fw_vendor		= (unsigned long)data->fw_vendor;
408 			efi_config_table	= (unsigned long)data->tables;
409 
410 			over4g |= data->fw_vendor	> U32_MAX ||
411 				  data->tables		> U32_MAX;
412 
413 			early_memunmap(data, sizeof(*data));
414 		} else {
415 			efi_fw_vendor		= systab64->fw_vendor;
416 			efi_config_table	= systab64->tables;
417 
418 			over4g |= systab64->fw_vendor	> U32_MAX ||
419 				  systab64->tables	> U32_MAX;
420 		}
421 		efi_nr_tables = systab64->nr_tables;
422 	} else {
423 		const efi_system_table_32_t *systab32 = p;
424 
425 		efi_fw_vendor		= systab32->fw_vendor;
426 		efi_runtime		= systab32->runtime;
427 		efi_config_table	= systab32->tables;
428 		efi_nr_tables		= systab32->nr_tables;
429 	}
430 
431 	efi.runtime_version = hdr->revision;
432 
433 	efi_systab_report_header(hdr, efi_fw_vendor);
434 	early_memunmap(p, size);
435 
436 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
437 		pr_err("EFI data located above 4GB, disabling EFI.\n");
438 		return -EINVAL;
439 	}
440 
441 	return 0;
442 }
443 
efi_config_init(const efi_config_table_type_t * arch_tables)444 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
445 {
446 	void *config_tables;
447 	int sz, ret;
448 
449 	if (efi_nr_tables == 0)
450 		return 0;
451 
452 	if (efi_enabled(EFI_64BIT))
453 		sz = sizeof(efi_config_table_64_t);
454 	else
455 		sz = sizeof(efi_config_table_32_t);
456 
457 	/*
458 	 * Let's see what config tables the firmware passed to us.
459 	 */
460 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
461 	if (config_tables == NULL) {
462 		pr_err("Could not map Configuration table!\n");
463 		return -ENOMEM;
464 	}
465 
466 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
467 				      arch_tables);
468 
469 	early_memunmap(config_tables, efi_nr_tables * sz);
470 	return ret;
471 }
472 
efi_init(void)473 void __init efi_init(void)
474 {
475 	if (IS_ENABLED(CONFIG_X86_32) &&
476 	    (boot_params.efi_info.efi_systab_hi ||
477 	     boot_params.efi_info.efi_memmap_hi)) {
478 		pr_info("Table located above 4GB, disabling EFI.\n");
479 		return;
480 	}
481 
482 	efi_systab_phys = boot_params.efi_info.efi_systab |
483 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
484 
485 	if (efi_systab_init(efi_systab_phys))
486 		return;
487 
488 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
489 		return;
490 
491 	if (efi_config_init(arch_tables))
492 		return;
493 
494 	/*
495 	 * Note: We currently don't support runtime services on an EFI
496 	 * that doesn't match the kernel 32/64-bit mode.
497 	 */
498 
499 	if (!efi_runtime_supported())
500 		pr_err("No EFI runtime due to 32/64-bit mismatch with kernel\n");
501 
502 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
503 		efi_memmap_unmap();
504 		return;
505 	}
506 
507 	/* Parse the EFI Properties table if it exists */
508 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
509 		efi_properties_table_t *tbl;
510 
511 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
512 		if (tbl == NULL) {
513 			pr_err("Could not map Properties table!\n");
514 		} else {
515 			if (tbl->memory_protection_attribute &
516 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
517 				set_bit(EFI_NX_PE_DATA, &efi.flags);
518 
519 			early_memunmap(tbl, sizeof(*tbl));
520 		}
521 	}
522 
523 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
524 	efi_clean_memmap();
525 
526 	efi_remove_e820_mmio();
527 
528 	if (efi_enabled(EFI_DBG))
529 		efi_print_memmap();
530 }
531 
532 /* Merge contiguous regions of the same type and attribute */
efi_merge_regions(void)533 static void __init efi_merge_regions(void)
534 {
535 	efi_memory_desc_t *md, *prev_md = NULL;
536 
537 	for_each_efi_memory_desc(md) {
538 		u64 prev_size;
539 
540 		if (!prev_md) {
541 			prev_md = md;
542 			continue;
543 		}
544 
545 		if (prev_md->type != md->type ||
546 		    prev_md->attribute != md->attribute) {
547 			prev_md = md;
548 			continue;
549 		}
550 
551 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
552 
553 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
554 			prev_md->num_pages += md->num_pages;
555 			md->type = EFI_RESERVED_TYPE;
556 			md->attribute = 0;
557 			continue;
558 		}
559 		prev_md = md;
560 	}
561 }
562 
realloc_pages(void * old_memmap,int old_shift)563 static void *realloc_pages(void *old_memmap, int old_shift)
564 {
565 	void *ret;
566 
567 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
568 	if (!ret)
569 		goto out;
570 
571 	/*
572 	 * A first-time allocation doesn't have anything to copy.
573 	 */
574 	if (!old_memmap)
575 		return ret;
576 
577 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
578 
579 out:
580 	free_pages((unsigned long)old_memmap, old_shift);
581 	return ret;
582 }
583 
584 /*
585  * Iterate the EFI memory map in reverse order because the regions
586  * will be mapped top-down. The end result is the same as if we had
587  * mapped things forward, but doesn't require us to change the
588  * existing implementation of efi_map_region().
589  */
efi_map_next_entry_reverse(void * entry)590 static inline void *efi_map_next_entry_reverse(void *entry)
591 {
592 	/* Initial call */
593 	if (!entry)
594 		return efi.memmap.map_end - efi.memmap.desc_size;
595 
596 	entry -= efi.memmap.desc_size;
597 	if (entry < efi.memmap.map)
598 		return NULL;
599 
600 	return entry;
601 }
602 
603 /*
604  * efi_map_next_entry - Return the next EFI memory map descriptor
605  * @entry: Previous EFI memory map descriptor
606  *
607  * This is a helper function to iterate over the EFI memory map, which
608  * we do in different orders depending on the current configuration.
609  *
610  * To begin traversing the memory map @entry must be %NULL.
611  *
612  * Returns %NULL when we reach the end of the memory map.
613  */
efi_map_next_entry(void * entry)614 static void *efi_map_next_entry(void *entry)
615 {
616 	if (efi_enabled(EFI_64BIT)) {
617 		/*
618 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
619 		 * config table feature requires us to map all entries
620 		 * in the same order as they appear in the EFI memory
621 		 * map. That is to say, entry N must have a lower
622 		 * virtual address than entry N+1. This is because the
623 		 * firmware toolchain leaves relative references in
624 		 * the code/data sections, which are split and become
625 		 * separate EFI memory regions. Mapping things
626 		 * out-of-order leads to the firmware accessing
627 		 * unmapped addresses.
628 		 *
629 		 * Since we need to map things this way whether or not
630 		 * the kernel actually makes use of
631 		 * EFI_PROPERTIES_TABLE, let's just switch to this
632 		 * scheme by default for 64-bit.
633 		 */
634 		return efi_map_next_entry_reverse(entry);
635 	}
636 
637 	/* Initial call */
638 	if (!entry)
639 		return efi.memmap.map;
640 
641 	entry += efi.memmap.desc_size;
642 	if (entry >= efi.memmap.map_end)
643 		return NULL;
644 
645 	return entry;
646 }
647 
should_map_region(efi_memory_desc_t * md)648 static bool should_map_region(efi_memory_desc_t *md)
649 {
650 	/*
651 	 * Runtime regions always require runtime mappings (obviously).
652 	 */
653 	if (md->attribute & EFI_MEMORY_RUNTIME)
654 		return true;
655 
656 	/*
657 	 * 32-bit EFI doesn't suffer from the bug that requires us to
658 	 * reserve boot services regions, and mixed mode support
659 	 * doesn't exist for 32-bit kernels.
660 	 */
661 	if (IS_ENABLED(CONFIG_X86_32))
662 		return false;
663 
664 	/*
665 	 * EFI specific purpose memory may be reserved by default
666 	 * depending on kernel config and boot options.
667 	 */
668 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
669 	    efi_soft_reserve_enabled() &&
670 	    (md->attribute & EFI_MEMORY_SP))
671 		return false;
672 
673 	/*
674 	 * Map all of RAM so that we can access arguments in the 1:1
675 	 * mapping when making EFI runtime calls.
676 	 */
677 	if (efi_is_mixed()) {
678 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
679 		    md->type == EFI_LOADER_DATA ||
680 		    md->type == EFI_LOADER_CODE)
681 			return true;
682 	}
683 
684 	/*
685 	 * Map boot services regions as a workaround for buggy
686 	 * firmware that accesses them even when they shouldn't.
687 	 *
688 	 * See efi_{reserve,free}_boot_services().
689 	 */
690 	if (md->type == EFI_BOOT_SERVICES_CODE ||
691 	    md->type == EFI_BOOT_SERVICES_DATA)
692 		return true;
693 
694 	return false;
695 }
696 
697 /*
698  * Map the efi memory ranges of the runtime services and update new_mmap with
699  * virtual addresses.
700  */
efi_map_regions(int * count,int * pg_shift)701 static void * __init efi_map_regions(int *count, int *pg_shift)
702 {
703 	void *p, *new_memmap = NULL;
704 	unsigned long left = 0;
705 	unsigned long desc_size;
706 	efi_memory_desc_t *md;
707 
708 	desc_size = efi.memmap.desc_size;
709 
710 	p = NULL;
711 	while ((p = efi_map_next_entry(p))) {
712 		md = p;
713 
714 		if (!should_map_region(md))
715 			continue;
716 
717 		efi_map_region(md);
718 
719 		if (left < desc_size) {
720 			new_memmap = realloc_pages(new_memmap, *pg_shift);
721 			if (!new_memmap)
722 				return NULL;
723 
724 			left += PAGE_SIZE << *pg_shift;
725 			(*pg_shift)++;
726 		}
727 
728 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
729 
730 		left -= desc_size;
731 		(*count)++;
732 	}
733 
734 	return new_memmap;
735 }
736 
kexec_enter_virtual_mode(void)737 static void __init kexec_enter_virtual_mode(void)
738 {
739 #ifdef CONFIG_KEXEC_CORE
740 	efi_memory_desc_t *md;
741 	unsigned int num_pages;
742 
743 	/*
744 	 * We don't do virtual mode, since we don't do runtime services, on
745 	 * non-native EFI.
746 	 */
747 	if (efi_is_mixed()) {
748 		efi_memmap_unmap();
749 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
750 		return;
751 	}
752 
753 	if (efi_alloc_page_tables()) {
754 		pr_err("Failed to allocate EFI page tables\n");
755 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
756 		return;
757 	}
758 
759 	/*
760 	* Map efi regions which were passed via setup_data. The virt_addr is a
761 	* fixed addr which was used in first kernel of a kexec boot.
762 	*/
763 	for_each_efi_memory_desc(md)
764 		efi_map_region_fixed(md); /* FIXME: add error handling */
765 
766 	/*
767 	 * Unregister the early EFI memmap from efi_init() and install
768 	 * the new EFI memory map.
769 	 */
770 	efi_memmap_unmap();
771 
772 	if (efi_memmap_init_late(efi.memmap.phys_map,
773 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
774 		pr_err("Failed to remap late EFI memory map\n");
775 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
776 		return;
777 	}
778 
779 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
780 	num_pages >>= PAGE_SHIFT;
781 
782 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
783 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
784 		return;
785 	}
786 
787 	efi_sync_low_kernel_mappings();
788 	efi_native_runtime_setup();
789 #endif
790 }
791 
792 /*
793  * This function will switch the EFI runtime services to virtual mode.
794  * Essentially, we look through the EFI memmap and map every region that
795  * has the runtime attribute bit set in its memory descriptor into the
796  * efi_pgd page table.
797  *
798  * The new method does a pagetable switch in a preemption-safe manner
799  * so that we're in a different address space when calling a runtime
800  * function. For function arguments passing we do copy the PUDs of the
801  * kernel page table into efi_pgd prior to each call.
802  *
803  * Specially for kexec boot, efi runtime maps in previous kernel should
804  * be passed in via setup_data. In that case runtime ranges will be mapped
805  * to the same virtual addresses as the first kernel, see
806  * kexec_enter_virtual_mode().
807  */
__efi_enter_virtual_mode(void)808 static void __init __efi_enter_virtual_mode(void)
809 {
810 	int count = 0, pg_shift = 0;
811 	void *new_memmap = NULL;
812 	efi_status_t status;
813 	unsigned long pa;
814 
815 	if (efi_alloc_page_tables()) {
816 		pr_err("Failed to allocate EFI page tables\n");
817 		goto err;
818 	}
819 
820 	efi_merge_regions();
821 	new_memmap = efi_map_regions(&count, &pg_shift);
822 	if (!new_memmap) {
823 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
824 		goto err;
825 	}
826 
827 	pa = __pa(new_memmap);
828 
829 	/*
830 	 * Unregister the early EFI memmap from efi_init() and install
831 	 * the new EFI memory map that we are about to pass to the
832 	 * firmware via SetVirtualAddressMap().
833 	 */
834 	efi_memmap_unmap();
835 
836 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
837 		pr_err("Failed to remap late EFI memory map\n");
838 		goto err;
839 	}
840 
841 	if (efi_enabled(EFI_DBG)) {
842 		pr_info("EFI runtime memory map:\n");
843 		efi_print_memmap();
844 	}
845 
846 	if (efi_setup_page_tables(pa, 1 << pg_shift))
847 		goto err;
848 
849 	efi_sync_low_kernel_mappings();
850 
851 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
852 					     efi.memmap.desc_size,
853 					     efi.memmap.desc_version,
854 					     (efi_memory_desc_t *)pa,
855 					     efi_systab_phys);
856 	if (status != EFI_SUCCESS) {
857 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
858 		       status);
859 		goto err;
860 	}
861 
862 	efi_check_for_embedded_firmwares();
863 	efi_free_boot_services();
864 
865 	if (!efi_is_mixed())
866 		efi_native_runtime_setup();
867 	else
868 		efi_thunk_runtime_setup();
869 
870 	/*
871 	 * Apply more restrictive page table mapping attributes now that
872 	 * SVAM() has been called and the firmware has performed all
873 	 * necessary relocation fixups for the new virtual addresses.
874 	 */
875 	efi_runtime_update_mappings();
876 
877 	/* clean DUMMY object */
878 	efi_delete_dummy_variable();
879 	return;
880 
881 err:
882 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
883 }
884 
efi_enter_virtual_mode(void)885 void __init efi_enter_virtual_mode(void)
886 {
887 	if (efi_enabled(EFI_PARAVIRT))
888 		return;
889 
890 	efi.runtime = (efi_runtime_services_t *)efi_runtime;
891 
892 	if (efi_setup)
893 		kexec_enter_virtual_mode();
894 	else
895 		__efi_enter_virtual_mode();
896 
897 	efi_dump_pagetable();
898 }
899 
efi_is_table_address(unsigned long phys_addr)900 bool efi_is_table_address(unsigned long phys_addr)
901 {
902 	unsigned int i;
903 
904 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
905 		return false;
906 
907 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
908 		if (*(efi_tables[i]) == phys_addr)
909 			return true;
910 
911 	return false;
912 }
913 
efi_systab_show_arch(char * str)914 char *efi_systab_show_arch(char *str)
915 {
916 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
917 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
918 	return str;
919 }
920 
921 #define EFI_FIELD(var) efi_ ## var
922 
923 #define EFI_ATTR_SHOW(name) \
924 static ssize_t name##_show(struct kobject *kobj, \
925 				struct kobj_attribute *attr, char *buf) \
926 { \
927 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
928 }
929 
930 EFI_ATTR_SHOW(fw_vendor);
931 EFI_ATTR_SHOW(runtime);
932 EFI_ATTR_SHOW(config_table);
933 
934 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
935 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
936 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
937 
efi_attr_is_visible(struct kobject * kobj,struct attribute * attr,int n)938 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
939 {
940 	if (attr == &efi_attr_fw_vendor.attr) {
941 		if (efi_enabled(EFI_PARAVIRT) ||
942 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
943 			return 0;
944 	} else if (attr == &efi_attr_runtime.attr) {
945 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
946 			return 0;
947 	} else if (attr == &efi_attr_config_table.attr) {
948 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
949 			return 0;
950 	}
951 	return attr->mode;
952 }
953