xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 09308012d8546dda75e96c02bed19e2ba1e875fd)
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 efi_system_table_t efi_systab __initdata;
58 static u64 efi_systab_phys __initdata;
59 
60 static unsigned long prop_phys = EFI_INVALID_TABLE_ADDR;
61 static unsigned long uga_phys = EFI_INVALID_TABLE_ADDR;
62 static unsigned long efi_runtime, efi_nr_tables;
63 
64 unsigned long efi_fw_vendor, efi_config_table;
65 
66 static const efi_config_table_type_t arch_tables[] __initconst = {
67 	{EFI_PROPERTIES_TABLE_GUID, "PROP", &prop_phys},
68 	{UGA_IO_PROTOCOL_GUID, "UGA", &uga_phys},
69 #ifdef CONFIG_X86_UV
70 	{UV_SYSTEM_TABLE_GUID, "UVsystab", &uv_systab_phys},
71 #endif
72 	{NULL_GUID, NULL, NULL},
73 };
74 
75 static const unsigned long * const efi_tables[] = {
76 	&efi.acpi,
77 	&efi.acpi20,
78 	&efi.smbios,
79 	&efi.smbios3,
80 	&uga_phys,
81 #ifdef CONFIG_X86_UV
82 	&uv_systab_phys,
83 #endif
84 	&efi_fw_vendor,
85 	&efi_runtime,
86 	&efi_config_table,
87 	&efi.esrt,
88 	&prop_phys,
89 	&efi_mem_attr_table,
90 #ifdef CONFIG_EFI_RCI2_TABLE
91 	&rci2_table_phys,
92 #endif
93 };
94 
95 u64 efi_setup;		/* efi setup_data physical address */
96 
97 static int add_efi_memmap __initdata;
98 static int __init setup_add_efi_memmap(char *arg)
99 {
100 	add_efi_memmap = 1;
101 	return 0;
102 }
103 early_param("add_efi_memmap", setup_add_efi_memmap);
104 
105 void __init efi_find_mirror(void)
106 {
107 	efi_memory_desc_t *md;
108 	u64 mirror_size = 0, total_size = 0;
109 
110 	if (!efi_enabled(EFI_MEMMAP))
111 		return;
112 
113 	for_each_efi_memory_desc(md) {
114 		unsigned long long start = md->phys_addr;
115 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
116 
117 		total_size += size;
118 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
119 			memblock_mark_mirror(start, size);
120 			mirror_size += size;
121 		}
122 	}
123 	if (mirror_size)
124 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
125 			mirror_size>>20, total_size>>20);
126 }
127 
128 /*
129  * Tell the kernel about the EFI memory map.  This might include
130  * more than the max 128 entries that can fit in the passed in e820
131  * legacy (zeropage) memory map, but the kernel's e820 table can hold
132  * E820_MAX_ENTRIES.
133  */
134 
135 static void __init do_add_efi_memmap(void)
136 {
137 	efi_memory_desc_t *md;
138 
139 	if (!efi_enabled(EFI_MEMMAP))
140 		return;
141 
142 	for_each_efi_memory_desc(md) {
143 		unsigned long long start = md->phys_addr;
144 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
145 		int e820_type;
146 
147 		switch (md->type) {
148 		case EFI_LOADER_CODE:
149 		case EFI_LOADER_DATA:
150 		case EFI_BOOT_SERVICES_CODE:
151 		case EFI_BOOT_SERVICES_DATA:
152 		case EFI_CONVENTIONAL_MEMORY:
153 			if (efi_soft_reserve_enabled()
154 			    && (md->attribute & EFI_MEMORY_SP))
155 				e820_type = E820_TYPE_SOFT_RESERVED;
156 			else if (md->attribute & EFI_MEMORY_WB)
157 				e820_type = E820_TYPE_RAM;
158 			else
159 				e820_type = E820_TYPE_RESERVED;
160 			break;
161 		case EFI_ACPI_RECLAIM_MEMORY:
162 			e820_type = E820_TYPE_ACPI;
163 			break;
164 		case EFI_ACPI_MEMORY_NVS:
165 			e820_type = E820_TYPE_NVS;
166 			break;
167 		case EFI_UNUSABLE_MEMORY:
168 			e820_type = E820_TYPE_UNUSABLE;
169 			break;
170 		case EFI_PERSISTENT_MEMORY:
171 			e820_type = E820_TYPE_PMEM;
172 			break;
173 		default:
174 			/*
175 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
176 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
177 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
178 			 */
179 			e820_type = E820_TYPE_RESERVED;
180 			break;
181 		}
182 
183 		e820__range_add(start, size, e820_type);
184 	}
185 	e820__update_table(e820_table);
186 }
187 
188 /*
189  * Given add_efi_memmap defaults to 0 and there there is no alternative
190  * e820 mechanism for soft-reserved memory, import the full EFI memory
191  * map if soft reservations are present and enabled. Otherwise, the
192  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
193  * the efi=nosoftreserve option.
194  */
195 static bool do_efi_soft_reserve(void)
196 {
197 	efi_memory_desc_t *md;
198 
199 	if (!efi_enabled(EFI_MEMMAP))
200 		return false;
201 
202 	if (!efi_soft_reserve_enabled())
203 		return false;
204 
205 	for_each_efi_memory_desc(md)
206 		if (md->type == EFI_CONVENTIONAL_MEMORY &&
207 		    (md->attribute & EFI_MEMORY_SP))
208 			return true;
209 	return false;
210 }
211 
212 int __init efi_memblock_x86_reserve_range(void)
213 {
214 	struct efi_info *e = &boot_params.efi_info;
215 	struct efi_memory_map_data data;
216 	phys_addr_t pmap;
217 	int rv;
218 
219 	if (efi_enabled(EFI_PARAVIRT))
220 		return 0;
221 
222 	/* Can't handle firmware tables above 4GB on i386 */
223 	if (IS_ENABLED(CONFIG_X86_32) && e->efi_memmap_hi > 0) {
224 		pr_err("Memory map is above 4GB, disabling EFI.\n");
225 		return -EINVAL;
226 	}
227 	pmap = (phys_addr_t)(e->efi_memmap | ((u64)e->efi_memmap_hi << 32));
228 
229 	data.phys_map		= pmap;
230 	data.size 		= e->efi_memmap_size;
231 	data.desc_size		= e->efi_memdesc_size;
232 	data.desc_version	= e->efi_memdesc_version;
233 
234 	rv = efi_memmap_init_early(&data);
235 	if (rv)
236 		return rv;
237 
238 	if (add_efi_memmap || do_efi_soft_reserve())
239 		do_add_efi_memmap();
240 
241 	efi_fake_memmap_early();
242 
243 	WARN(efi.memmap.desc_version != 1,
244 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
245 	     efi.memmap.desc_version);
246 
247 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
248 
249 	return 0;
250 }
251 
252 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
253 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
254 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
255 
256 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
257 {
258 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
259 	u64 end_hi = 0;
260 	char buf[64];
261 
262 	if (md->num_pages == 0) {
263 		end = 0;
264 	} else if (md->num_pages > EFI_PAGES_MAX ||
265 		   EFI_PAGES_MAX - md->num_pages <
266 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
267 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
268 			>> OVERFLOW_ADDR_SHIFT;
269 
270 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
271 			end_hi += 1;
272 	} else {
273 		return true;
274 	}
275 
276 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
277 
278 	if (end_hi) {
279 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
280 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
281 			md->phys_addr, end_hi, end);
282 	} else {
283 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
284 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
285 			md->phys_addr, end);
286 	}
287 	return false;
288 }
289 
290 static void __init efi_clean_memmap(void)
291 {
292 	efi_memory_desc_t *out = efi.memmap.map;
293 	const efi_memory_desc_t *in = out;
294 	const efi_memory_desc_t *end = efi.memmap.map_end;
295 	int i, n_removal;
296 
297 	for (i = n_removal = 0; in < end; i++) {
298 		if (efi_memmap_entry_valid(in, i)) {
299 			if (out != in)
300 				memcpy(out, in, efi.memmap.desc_size);
301 			out = (void *)out + efi.memmap.desc_size;
302 		} else {
303 			n_removal++;
304 		}
305 		in = (void *)in + efi.memmap.desc_size;
306 	}
307 
308 	if (n_removal > 0) {
309 		struct efi_memory_map_data data = {
310 			.phys_map	= efi.memmap.phys_map,
311 			.desc_version	= efi.memmap.desc_version,
312 			.desc_size	= efi.memmap.desc_size,
313 			.size		= efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
314 			.flags		= 0,
315 		};
316 
317 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
318 		efi_memmap_install(&data);
319 	}
320 }
321 
322 void __init efi_print_memmap(void)
323 {
324 	efi_memory_desc_t *md;
325 	int i = 0;
326 
327 	for_each_efi_memory_desc(md) {
328 		char buf[64];
329 
330 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
331 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
332 			md->phys_addr,
333 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
334 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
335 	}
336 }
337 
338 static int __init efi_systab_init(u64 phys)
339 {
340 	int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
341 					  : sizeof(efi_system_table_32_t);
342 	const efi_table_hdr_t *hdr;
343 	bool over4g = false;
344 	void *p;
345 	int ret;
346 
347 	hdr = p = early_memremap_ro(phys, size);
348 	if (p == NULL) {
349 		pr_err("Couldn't map the system table!\n");
350 		return -ENOMEM;
351 	}
352 
353 	ret = efi_systab_check_header(hdr, 1);
354 	if (ret) {
355 		early_memunmap(p, size);
356 		return ret;
357 	}
358 
359 	if (efi_enabled(EFI_64BIT)) {
360 		const efi_system_table_64_t *systab64 = p;
361 
362 		efi_systab.hdr			= systab64->hdr;
363 		efi_systab.fw_vendor		= systab64->fw_vendor;
364 		efi_systab.fw_revision		= systab64->fw_revision;
365 		efi_systab.con_in_handle	= systab64->con_in_handle;
366 		efi_systab.con_in		= systab64->con_in;
367 		efi_systab.con_out_handle	= systab64->con_out_handle;
368 		efi_systab.con_out		= (void *)(unsigned long)systab64->con_out;
369 		efi_systab.stderr_handle	= systab64->stderr_handle;
370 		efi_systab.stderr		= systab64->stderr;
371 		efi_systab.runtime		= (void *)(unsigned long)systab64->runtime;
372 		efi_systab.boottime		= (void *)(unsigned long)systab64->boottime;
373 		efi_systab.nr_tables		= systab64->nr_tables;
374 		efi_systab.tables		= systab64->tables;
375 
376 		over4g = systab64->con_in_handle	> U32_MAX ||
377 			 systab64->con_in		> U32_MAX ||
378 			 systab64->con_out_handle	> U32_MAX ||
379 			 systab64->con_out		> U32_MAX ||
380 			 systab64->stderr_handle	> U32_MAX ||
381 			 systab64->stderr		> U32_MAX ||
382 			 systab64->runtime		> U32_MAX ||
383 			 systab64->boottime		> U32_MAX;
384 
385 		if (efi_setup) {
386 			struct efi_setup_data *data;
387 
388 			data = early_memremap_ro(efi_setup, sizeof(*data));
389 			if (!data) {
390 				early_memunmap(p, size);
391 				return -ENOMEM;
392 			}
393 
394 			efi_systab.fw_vendor	= (unsigned long)data->fw_vendor;
395 			efi_systab.tables	= (unsigned long)data->tables;
396 
397 			over4g |= data->fw_vendor	> U32_MAX ||
398 				  data->tables		> U32_MAX;
399 
400 			early_memunmap(data, sizeof(*data));
401 		} else {
402 			over4g |= systab64->fw_vendor	> U32_MAX ||
403 				  systab64->tables	> U32_MAX;
404 		}
405 	} else {
406 		const efi_system_table_32_t *systab32 = p;
407 
408 		efi_systab.hdr			= systab32->hdr;
409 		efi_systab.fw_vendor		= systab32->fw_vendor;
410 		efi_systab.fw_revision		= systab32->fw_revision;
411 		efi_systab.con_in_handle	= systab32->con_in_handle;
412 		efi_systab.con_in		= systab32->con_in;
413 		efi_systab.con_out_handle	= systab32->con_out_handle;
414 		efi_systab.con_out		= (void *)(unsigned long)systab32->con_out;
415 		efi_systab.stderr_handle	= systab32->stderr_handle;
416 		efi_systab.stderr		= systab32->stderr;
417 		efi_systab.runtime		= (void *)(unsigned long)systab32->runtime;
418 		efi_systab.boottime		= (void *)(unsigned long)systab32->boottime;
419 		efi_systab.nr_tables		= systab32->nr_tables;
420 		efi_systab.tables		= systab32->tables;
421 	}
422 
423 	efi.runtime_version = hdr->revision;
424 
425 	efi_systab_report_header(hdr, efi_systab.fw_vendor);
426 	early_memunmap(p, size);
427 
428 	if (IS_ENABLED(CONFIG_X86_32) && over4g) {
429 		pr_err("EFI data located above 4GB, disabling EFI.\n");
430 		return -EINVAL;
431 	}
432 
433 	efi.systab = &efi_systab;
434 	return 0;
435 }
436 
437 static int __init efi_config_init(const efi_config_table_type_t *arch_tables)
438 {
439 	void *config_tables;
440 	int sz, ret;
441 
442 	if (efi_nr_tables == 0)
443 		return 0;
444 
445 	if (efi_enabled(EFI_64BIT))
446 		sz = sizeof(efi_config_table_64_t);
447 	else
448 		sz = sizeof(efi_config_table_32_t);
449 
450 	/*
451 	 * Let's see what config tables the firmware passed to us.
452 	 */
453 	config_tables = early_memremap(efi_config_table, efi_nr_tables * sz);
454 	if (config_tables == NULL) {
455 		pr_err("Could not map Configuration table!\n");
456 		return -ENOMEM;
457 	}
458 
459 	ret = efi_config_parse_tables(config_tables, efi_nr_tables,
460 				      arch_tables);
461 
462 	early_memunmap(config_tables, efi_nr_tables * sz);
463 	return ret;
464 }
465 
466 void __init efi_init(void)
467 {
468 	if (IS_ENABLED(CONFIG_X86_32) &&
469 	    (boot_params.efi_info.efi_systab_hi ||
470 	     boot_params.efi_info.efi_memmap_hi)) {
471 		pr_info("Table located above 4GB, disabling EFI.\n");
472 		return;
473 	}
474 
475 	efi_systab_phys = boot_params.efi_info.efi_systab |
476 			  ((__u64)boot_params.efi_info.efi_systab_hi << 32);
477 
478 	if (efi_systab_init(efi_systab_phys))
479 		return;
480 
481 	efi_config_table = (unsigned long)efi.systab->tables;
482 	efi_nr_tables    = efi.systab->nr_tables;
483 	efi_fw_vendor    = (unsigned long)efi.systab->fw_vendor;
484 	efi_runtime      = (unsigned long)efi.systab->runtime;
485 
486 	if (efi_reuse_config(efi_config_table, efi_nr_tables))
487 		return;
488 
489 	if (efi_config_init(arch_tables))
490 		return;
491 
492 	/*
493 	 * Note: We currently don't support runtime services on an EFI
494 	 * that doesn't match the kernel 32/64-bit mode.
495 	 */
496 
497 	if (!efi_runtime_supported())
498 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
499 
500 	if (!efi_runtime_supported() || efi_runtime_disabled()) {
501 		efi_memmap_unmap();
502 		return;
503 	}
504 
505 	/* Parse the EFI Properties table if it exists */
506 	if (prop_phys != EFI_INVALID_TABLE_ADDR) {
507 		efi_properties_table_t *tbl;
508 
509 		tbl = early_memremap_ro(prop_phys, sizeof(*tbl));
510 		if (tbl == NULL) {
511 			pr_err("Could not map Properties table!\n");
512 		} else {
513 			if (tbl->memory_protection_attribute &
514 			    EFI_PROPERTIES_RUNTIME_MEMORY_PROTECTION_NON_EXECUTABLE_PE_DATA)
515 				set_bit(EFI_NX_PE_DATA, &efi.flags);
516 
517 			early_memunmap(tbl, sizeof(*tbl));
518 		}
519 	}
520 
521 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
522 	efi_clean_memmap();
523 
524 	if (efi_enabled(EFI_DBG))
525 		efi_print_memmap();
526 }
527 
528 #if defined(CONFIG_X86_32) || defined(CONFIG_X86_UV)
529 
530 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
531 {
532 	u64 addr, npages;
533 
534 	addr = md->virt_addr;
535 	npages = md->num_pages;
536 
537 	memrange_efi_to_native(&addr, &npages);
538 
539 	if (executable)
540 		set_memory_x(addr, npages);
541 	else
542 		set_memory_nx(addr, npages);
543 }
544 
545 void __init runtime_code_page_mkexec(void)
546 {
547 	efi_memory_desc_t *md;
548 
549 	/* Make EFI runtime service code area executable */
550 	for_each_efi_memory_desc(md) {
551 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
552 			continue;
553 
554 		efi_set_executable(md, true);
555 	}
556 }
557 
558 void __init efi_memory_uc(u64 addr, unsigned long size)
559 {
560 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
561 	u64 npages;
562 
563 	npages = round_up(size, page_shift) / page_shift;
564 	memrange_efi_to_native(&addr, &npages);
565 	set_memory_uc(addr, npages);
566 }
567 
568 void __init old_map_region(efi_memory_desc_t *md)
569 {
570 	u64 start_pfn, end_pfn, end;
571 	unsigned long size;
572 	void *va;
573 
574 	start_pfn = PFN_DOWN(md->phys_addr);
575 	size	  = md->num_pages << PAGE_SHIFT;
576 	end	  = md->phys_addr + size;
577 	end_pfn   = PFN_UP(end);
578 
579 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
580 		va = __va(md->phys_addr);
581 
582 		if (!(md->attribute & EFI_MEMORY_WB))
583 			efi_memory_uc((u64)(unsigned long)va, size);
584 	} else
585 		va = efi_ioremap(md->phys_addr, size,
586 				 md->type, md->attribute);
587 
588 	md->virt_addr = (u64) (unsigned long) va;
589 	if (!va)
590 		pr_err("ioremap of 0x%llX failed!\n",
591 		       (unsigned long long)md->phys_addr);
592 }
593 
594 #endif
595 
596 /* Merge contiguous regions of the same type and attribute */
597 static void __init efi_merge_regions(void)
598 {
599 	efi_memory_desc_t *md, *prev_md = NULL;
600 
601 	for_each_efi_memory_desc(md) {
602 		u64 prev_size;
603 
604 		if (!prev_md) {
605 			prev_md = md;
606 			continue;
607 		}
608 
609 		if (prev_md->type != md->type ||
610 		    prev_md->attribute != md->attribute) {
611 			prev_md = md;
612 			continue;
613 		}
614 
615 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
616 
617 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
618 			prev_md->num_pages += md->num_pages;
619 			md->type = EFI_RESERVED_TYPE;
620 			md->attribute = 0;
621 			continue;
622 		}
623 		prev_md = md;
624 	}
625 }
626 
627 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
628 {
629 	unsigned long size;
630 	u64 end, systab;
631 
632 	size = md->num_pages << EFI_PAGE_SHIFT;
633 	end = md->phys_addr + size;
634 	systab = efi_systab_phys;
635 	if (md->phys_addr <= systab && systab < end) {
636 		systab += md->virt_addr - md->phys_addr;
637 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
638 	}
639 }
640 
641 static void *realloc_pages(void *old_memmap, int old_shift)
642 {
643 	void *ret;
644 
645 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
646 	if (!ret)
647 		goto out;
648 
649 	/*
650 	 * A first-time allocation doesn't have anything to copy.
651 	 */
652 	if (!old_memmap)
653 		return ret;
654 
655 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
656 
657 out:
658 	free_pages((unsigned long)old_memmap, old_shift);
659 	return ret;
660 }
661 
662 /*
663  * Iterate the EFI memory map in reverse order because the regions
664  * will be mapped top-down. The end result is the same as if we had
665  * mapped things forward, but doesn't require us to change the
666  * existing implementation of efi_map_region().
667  */
668 static inline void *efi_map_next_entry_reverse(void *entry)
669 {
670 	/* Initial call */
671 	if (!entry)
672 		return efi.memmap.map_end - efi.memmap.desc_size;
673 
674 	entry -= efi.memmap.desc_size;
675 	if (entry < efi.memmap.map)
676 		return NULL;
677 
678 	return entry;
679 }
680 
681 /*
682  * efi_map_next_entry - Return the next EFI memory map descriptor
683  * @entry: Previous EFI memory map descriptor
684  *
685  * This is a helper function to iterate over the EFI memory map, which
686  * we do in different orders depending on the current configuration.
687  *
688  * To begin traversing the memory map @entry must be %NULL.
689  *
690  * Returns %NULL when we reach the end of the memory map.
691  */
692 static void *efi_map_next_entry(void *entry)
693 {
694 	if (!efi_have_uv1_memmap() && efi_enabled(EFI_64BIT)) {
695 		/*
696 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
697 		 * config table feature requires us to map all entries
698 		 * in the same order as they appear in the EFI memory
699 		 * map. That is to say, entry N must have a lower
700 		 * virtual address than entry N+1. This is because the
701 		 * firmware toolchain leaves relative references in
702 		 * the code/data sections, which are split and become
703 		 * separate EFI memory regions. Mapping things
704 		 * out-of-order leads to the firmware accessing
705 		 * unmapped addresses.
706 		 *
707 		 * Since we need to map things this way whether or not
708 		 * the kernel actually makes use of
709 		 * EFI_PROPERTIES_TABLE, let's just switch to this
710 		 * scheme by default for 64-bit.
711 		 */
712 		return efi_map_next_entry_reverse(entry);
713 	}
714 
715 	/* Initial call */
716 	if (!entry)
717 		return efi.memmap.map;
718 
719 	entry += efi.memmap.desc_size;
720 	if (entry >= efi.memmap.map_end)
721 		return NULL;
722 
723 	return entry;
724 }
725 
726 static bool should_map_region(efi_memory_desc_t *md)
727 {
728 	/*
729 	 * Runtime regions always require runtime mappings (obviously).
730 	 */
731 	if (md->attribute & EFI_MEMORY_RUNTIME)
732 		return true;
733 
734 	/*
735 	 * 32-bit EFI doesn't suffer from the bug that requires us to
736 	 * reserve boot services regions, and mixed mode support
737 	 * doesn't exist for 32-bit kernels.
738 	 */
739 	if (IS_ENABLED(CONFIG_X86_32))
740 		return false;
741 
742 	/*
743 	 * EFI specific purpose memory may be reserved by default
744 	 * depending on kernel config and boot options.
745 	 */
746 	if (md->type == EFI_CONVENTIONAL_MEMORY &&
747 	    efi_soft_reserve_enabled() &&
748 	    (md->attribute & EFI_MEMORY_SP))
749 		return false;
750 
751 	/*
752 	 * Map all of RAM so that we can access arguments in the 1:1
753 	 * mapping when making EFI runtime calls.
754 	 */
755 	if (efi_is_mixed()) {
756 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
757 		    md->type == EFI_LOADER_DATA ||
758 		    md->type == EFI_LOADER_CODE)
759 			return true;
760 	}
761 
762 	/*
763 	 * Map boot services regions as a workaround for buggy
764 	 * firmware that accesses them even when they shouldn't.
765 	 *
766 	 * See efi_{reserve,free}_boot_services().
767 	 */
768 	if (md->type == EFI_BOOT_SERVICES_CODE ||
769 	    md->type == EFI_BOOT_SERVICES_DATA)
770 		return true;
771 
772 	return false;
773 }
774 
775 /*
776  * Map the efi memory ranges of the runtime services and update new_mmap with
777  * virtual addresses.
778  */
779 static void * __init efi_map_regions(int *count, int *pg_shift)
780 {
781 	void *p, *new_memmap = NULL;
782 	unsigned long left = 0;
783 	unsigned long desc_size;
784 	efi_memory_desc_t *md;
785 
786 	desc_size = efi.memmap.desc_size;
787 
788 	p = NULL;
789 	while ((p = efi_map_next_entry(p))) {
790 		md = p;
791 
792 		if (!should_map_region(md))
793 			continue;
794 
795 		efi_map_region(md);
796 		get_systab_virt_addr(md);
797 
798 		if (left < desc_size) {
799 			new_memmap = realloc_pages(new_memmap, *pg_shift);
800 			if (!new_memmap)
801 				return NULL;
802 
803 			left += PAGE_SIZE << *pg_shift;
804 			(*pg_shift)++;
805 		}
806 
807 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
808 
809 		left -= desc_size;
810 		(*count)++;
811 	}
812 
813 	return new_memmap;
814 }
815 
816 static void __init kexec_enter_virtual_mode(void)
817 {
818 #ifdef CONFIG_KEXEC_CORE
819 	efi_memory_desc_t *md;
820 	unsigned int num_pages;
821 
822 	efi.systab = NULL;
823 
824 	/*
825 	 * We don't do virtual mode, since we don't do runtime services, on
826 	 * non-native EFI. With the UV1 memmap, we don't do runtime services in
827 	 * kexec kernel because in the initial boot something else might
828 	 * have been mapped at these virtual addresses.
829 	 */
830 	if (efi_is_mixed() || efi_have_uv1_memmap()) {
831 		efi_memmap_unmap();
832 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
833 		return;
834 	}
835 
836 	if (efi_alloc_page_tables()) {
837 		pr_err("Failed to allocate EFI page tables\n");
838 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
839 		return;
840 	}
841 
842 	/*
843 	* Map efi regions which were passed via setup_data. The virt_addr is a
844 	* fixed addr which was used in first kernel of a kexec boot.
845 	*/
846 	for_each_efi_memory_desc(md) {
847 		efi_map_region_fixed(md); /* FIXME: add error handling */
848 		get_systab_virt_addr(md);
849 	}
850 
851 	/*
852 	 * Unregister the early EFI memmap from efi_init() and install
853 	 * the new EFI memory map.
854 	 */
855 	efi_memmap_unmap();
856 
857 	if (efi_memmap_init_late(efi.memmap.phys_map,
858 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
859 		pr_err("Failed to remap late EFI memory map\n");
860 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
861 		return;
862 	}
863 
864 	BUG_ON(!efi.systab);
865 
866 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
867 	num_pages >>= PAGE_SHIFT;
868 
869 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
870 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
871 		return;
872 	}
873 
874 	efi_sync_low_kernel_mappings();
875 	efi_native_runtime_setup();
876 #endif
877 }
878 
879 /*
880  * This function will switch the EFI runtime services to virtual mode.
881  * Essentially, we look through the EFI memmap and map every region that
882  * has the runtime attribute bit set in its memory descriptor into the
883  * efi_pgd page table.
884  *
885  * The old method which used to update that memory descriptor with the
886  * virtual address obtained from ioremap() is still supported when the
887  * kernel is booted on SG1 UV1 hardware. Same old method enabled the
888  * runtime services to be called without having to thunk back into
889  * physical mode for every invocation.
890  *
891  * The new method does a pagetable switch in a preemption-safe manner
892  * so that we're in a different address space when calling a runtime
893  * function. For function arguments passing we do copy the PUDs of the
894  * kernel page table into efi_pgd prior to each call.
895  *
896  * Specially for kexec boot, efi runtime maps in previous kernel should
897  * be passed in via setup_data. In that case runtime ranges will be mapped
898  * to the same virtual addresses as the first kernel, see
899  * kexec_enter_virtual_mode().
900  */
901 static void __init __efi_enter_virtual_mode(void)
902 {
903 	int count = 0, pg_shift = 0;
904 	void *new_memmap = NULL;
905 	efi_status_t status;
906 	unsigned long pa;
907 
908 	efi.systab = NULL;
909 
910 	if (efi_alloc_page_tables()) {
911 		pr_err("Failed to allocate EFI page tables\n");
912 		goto err;
913 	}
914 
915 	efi_merge_regions();
916 	new_memmap = efi_map_regions(&count, &pg_shift);
917 	if (!new_memmap) {
918 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
919 		goto err;
920 	}
921 
922 	pa = __pa(new_memmap);
923 
924 	/*
925 	 * Unregister the early EFI memmap from efi_init() and install
926 	 * the new EFI memory map that we are about to pass to the
927 	 * firmware via SetVirtualAddressMap().
928 	 */
929 	efi_memmap_unmap();
930 
931 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
932 		pr_err("Failed to remap late EFI memory map\n");
933 		goto err;
934 	}
935 
936 	if (efi_enabled(EFI_DBG)) {
937 		pr_info("EFI runtime memory map:\n");
938 		efi_print_memmap();
939 	}
940 
941 	if (WARN_ON(!efi.systab))
942 		goto err;
943 
944 	if (efi_setup_page_tables(pa, 1 << pg_shift))
945 		goto err;
946 
947 	efi_sync_low_kernel_mappings();
948 
949 	status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
950 					     efi.memmap.desc_size,
951 					     efi.memmap.desc_version,
952 					     (efi_memory_desc_t *)pa);
953 	if (status != EFI_SUCCESS) {
954 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
955 		       status);
956 		goto err;
957 	}
958 
959 	efi_free_boot_services();
960 
961 	if (!efi_is_mixed())
962 		efi_native_runtime_setup();
963 	else
964 		efi_thunk_runtime_setup();
965 
966 	/*
967 	 * Apply more restrictive page table mapping attributes now that
968 	 * SVAM() has been called and the firmware has performed all
969 	 * necessary relocation fixups for the new virtual addresses.
970 	 */
971 	efi_runtime_update_mappings();
972 
973 	/* clean DUMMY object */
974 	efi_delete_dummy_variable();
975 	return;
976 
977 err:
978 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
979 }
980 
981 void __init efi_enter_virtual_mode(void)
982 {
983 	if (efi_enabled(EFI_PARAVIRT))
984 		return;
985 
986 	if (efi_setup)
987 		kexec_enter_virtual_mode();
988 	else
989 		__efi_enter_virtual_mode();
990 
991 	efi_dump_pagetable();
992 }
993 
994 bool efi_is_table_address(unsigned long phys_addr)
995 {
996 	unsigned int i;
997 
998 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
999 		return false;
1000 
1001 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
1002 		if (*(efi_tables[i]) == phys_addr)
1003 			return true;
1004 
1005 	return false;
1006 }
1007 
1008 char *efi_systab_show_arch(char *str)
1009 {
1010 	if (uga_phys != EFI_INVALID_TABLE_ADDR)
1011 		str += sprintf(str, "UGA=0x%lx\n", uga_phys);
1012 	return str;
1013 }
1014 
1015 #define EFI_FIELD(var) efi_ ## var
1016 
1017 #define EFI_ATTR_SHOW(name) \
1018 static ssize_t name##_show(struct kobject *kobj, \
1019 				struct kobj_attribute *attr, char *buf) \
1020 { \
1021 	return sprintf(buf, "0x%lx\n", EFI_FIELD(name)); \
1022 }
1023 
1024 EFI_ATTR_SHOW(fw_vendor);
1025 EFI_ATTR_SHOW(runtime);
1026 EFI_ATTR_SHOW(config_table);
1027 
1028 struct kobj_attribute efi_attr_fw_vendor = __ATTR_RO(fw_vendor);
1029 struct kobj_attribute efi_attr_runtime = __ATTR_RO(runtime);
1030 struct kobj_attribute efi_attr_config_table = __ATTR_RO(config_table);
1031 
1032 umode_t efi_attr_is_visible(struct kobject *kobj, struct attribute *attr, int n)
1033 {
1034 	if (attr == &efi_attr_fw_vendor.attr) {
1035 		if (efi_enabled(EFI_PARAVIRT) ||
1036 				efi_fw_vendor == EFI_INVALID_TABLE_ADDR)
1037 			return 0;
1038 	} else if (attr == &efi_attr_runtime.attr) {
1039 		if (efi_runtime == EFI_INVALID_TABLE_ADDR)
1040 			return 0;
1041 	} else if (attr == &efi_attr_config_table.attr) {
1042 		if (efi_config_table == EFI_INVALID_TABLE_ADDR)
1043 			return 0;
1044 	}
1045 	return attr->mode;
1046 }
1047