xref: /linux/arch/x86/platform/efi/efi.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * Common EFI (Extensible Firmware Interface) support functions
3  * Based on Extensible Firmware Interface Specification version 1.0
4  *
5  * Copyright (C) 1999 VA Linux Systems
6  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
7  * Copyright (C) 1999-2002 Hewlett-Packard Co.
8  *	David Mosberger-Tang <davidm@hpl.hp.com>
9  *	Stephane Eranian <eranian@hpl.hp.com>
10  * Copyright (C) 2005-2008 Intel Co.
11  *	Fenghua Yu <fenghua.yu@intel.com>
12  *	Bibo Mao <bibo.mao@intel.com>
13  *	Chandramouli Narayanan <mouli@linux.intel.com>
14  *	Huang Ying <ying.huang@intel.com>
15  * Copyright (C) 2013 SuSE Labs
16  *	Borislav Petkov <bp@suse.de> - runtime services VA mapping
17  *
18  * Copied from efi_32.c to eliminate the duplicated code between EFI
19  * 32/64 support code. --ying 2007-10-26
20  *
21  * All EFI Runtime Services are not implemented yet as EFI only
22  * supports physical mode addressing on SoftSDV. This is to be fixed
23  * in a future version.  --drummond 1999-07-20
24  *
25  * Implemented EFI runtime services and virtual mode calls.  --davidm
26  *
27  * Goutham Rao: <goutham.rao@intel.com>
28  *	Skip non-WB memory and ignore empty memory ranges.
29  */
30 
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32 
33 #include <linux/kernel.h>
34 #include <linux/init.h>
35 #include <linux/efi.h>
36 #include <linux/efi-bgrt.h>
37 #include <linux/export.h>
38 #include <linux/bootmem.h>
39 #include <linux/slab.h>
40 #include <linux/memblock.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/time.h>
51 #include <asm/cacheflush.h>
52 #include <asm/tlbflush.h>
53 #include <asm/x86_init.h>
54 #include <asm/rtc.h>
55 #include <asm/uv/uv.h>
56 
57 #define EFI_DEBUG
58 
59 struct efi_memory_map memmap;
60 
61 static struct efi efi_phys __initdata;
62 static efi_system_table_t efi_systab __initdata;
63 
64 static efi_config_table_type_t arch_tables[] __initdata = {
65 #ifdef CONFIG_X86_UV
66 	{UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
67 #endif
68 	{NULL_GUID, NULL, NULL},
69 };
70 
71 u64 efi_setup;		/* efi setup_data physical address */
72 
73 static bool disable_runtime __initdata = false;
74 static int __init setup_noefi(char *arg)
75 {
76 	disable_runtime = true;
77 	return 0;
78 }
79 early_param("noefi", setup_noefi);
80 
81 int add_efi_memmap;
82 EXPORT_SYMBOL(add_efi_memmap);
83 
84 static int __init setup_add_efi_memmap(char *arg)
85 {
86 	add_efi_memmap = 1;
87 	return 0;
88 }
89 early_param("add_efi_memmap", setup_add_efi_memmap);
90 
91 static efi_status_t __init phys_efi_set_virtual_address_map(
92 	unsigned long memory_map_size,
93 	unsigned long descriptor_size,
94 	u32 descriptor_version,
95 	efi_memory_desc_t *virtual_map)
96 {
97 	efi_status_t status;
98 
99 	efi_call_phys_prelog();
100 	status = efi_call_phys(efi_phys.set_virtual_address_map,
101 			       memory_map_size, descriptor_size,
102 			       descriptor_version, virtual_map);
103 	efi_call_phys_epilog();
104 	return status;
105 }
106 
107 void efi_get_time(struct timespec *now)
108 {
109 	efi_status_t status;
110 	efi_time_t eft;
111 	efi_time_cap_t cap;
112 
113 	status = efi.get_time(&eft, &cap);
114 	if (status != EFI_SUCCESS)
115 		pr_err("Oops: efitime: can't read time!\n");
116 
117 	now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
118 			     eft.minute, eft.second);
119 	now->tv_nsec = 0;
120 }
121 
122 /*
123  * Tell the kernel about the EFI memory map.  This might include
124  * more than the max 128 entries that can fit in the e820 legacy
125  * (zeropage) memory map.
126  */
127 
128 static void __init do_add_efi_memmap(void)
129 {
130 	void *p;
131 
132 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
133 		efi_memory_desc_t *md = p;
134 		unsigned long long start = md->phys_addr;
135 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
136 		int e820_type;
137 
138 		switch (md->type) {
139 		case EFI_LOADER_CODE:
140 		case EFI_LOADER_DATA:
141 		case EFI_BOOT_SERVICES_CODE:
142 		case EFI_BOOT_SERVICES_DATA:
143 		case EFI_CONVENTIONAL_MEMORY:
144 			if (md->attribute & EFI_MEMORY_WB)
145 				e820_type = E820_RAM;
146 			else
147 				e820_type = E820_RESERVED;
148 			break;
149 		case EFI_ACPI_RECLAIM_MEMORY:
150 			e820_type = E820_ACPI;
151 			break;
152 		case EFI_ACPI_MEMORY_NVS:
153 			e820_type = E820_NVS;
154 			break;
155 		case EFI_UNUSABLE_MEMORY:
156 			e820_type = E820_UNUSABLE;
157 			break;
158 		default:
159 			/*
160 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
161 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
162 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
163 			 */
164 			e820_type = E820_RESERVED;
165 			break;
166 		}
167 		e820_add_region(start, size, e820_type);
168 	}
169 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
170 }
171 
172 int __init efi_memblock_x86_reserve_range(void)
173 {
174 	struct efi_info *e = &boot_params.efi_info;
175 	unsigned long pmap;
176 
177 	if (efi_enabled(EFI_PARAVIRT))
178 		return 0;
179 
180 #ifdef CONFIG_X86_32
181 	/* Can't handle data above 4GB at this time */
182 	if (e->efi_memmap_hi) {
183 		pr_err("Memory map is above 4GB, disabling EFI.\n");
184 		return -EINVAL;
185 	}
186 	pmap =  e->efi_memmap;
187 #else
188 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
189 #endif
190 	memmap.phys_map		= (void *)pmap;
191 	memmap.nr_map		= e->efi_memmap_size /
192 				  e->efi_memdesc_size;
193 	memmap.desc_size	= e->efi_memdesc_size;
194 	memmap.desc_version	= e->efi_memdesc_version;
195 
196 	memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
197 
198 	efi.memmap = &memmap;
199 
200 	return 0;
201 }
202 
203 static void __init print_efi_memmap(void)
204 {
205 #ifdef EFI_DEBUG
206 	efi_memory_desc_t *md;
207 	void *p;
208 	int i;
209 
210 	for (p = memmap.map, i = 0;
211 	     p < memmap.map_end;
212 	     p += memmap.desc_size, i++) {
213 		md = p;
214 		pr_info("mem%02u: type=%u, attr=0x%llx, range=[0x%016llx-0x%016llx) (%lluMB)\n",
215 			i, md->type, md->attribute, md->phys_addr,
216 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
217 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
218 	}
219 #endif  /*  EFI_DEBUG  */
220 }
221 
222 void __init efi_unmap_memmap(void)
223 {
224 	clear_bit(EFI_MEMMAP, &efi.flags);
225 	if (memmap.map) {
226 		early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size);
227 		memmap.map = NULL;
228 	}
229 }
230 
231 static int __init efi_systab_init(void *phys)
232 {
233 	if (efi_enabled(EFI_64BIT)) {
234 		efi_system_table_64_t *systab64;
235 		struct efi_setup_data *data = NULL;
236 		u64 tmp = 0;
237 
238 		if (efi_setup) {
239 			data = early_memremap(efi_setup, sizeof(*data));
240 			if (!data)
241 				return -ENOMEM;
242 		}
243 		systab64 = early_memremap((unsigned long)phys,
244 					 sizeof(*systab64));
245 		if (systab64 == NULL) {
246 			pr_err("Couldn't map the system table!\n");
247 			if (data)
248 				early_memunmap(data, sizeof(*data));
249 			return -ENOMEM;
250 		}
251 
252 		efi_systab.hdr = systab64->hdr;
253 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
254 					      systab64->fw_vendor;
255 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
256 		efi_systab.fw_revision = systab64->fw_revision;
257 		efi_systab.con_in_handle = systab64->con_in_handle;
258 		tmp |= systab64->con_in_handle;
259 		efi_systab.con_in = systab64->con_in;
260 		tmp |= systab64->con_in;
261 		efi_systab.con_out_handle = systab64->con_out_handle;
262 		tmp |= systab64->con_out_handle;
263 		efi_systab.con_out = systab64->con_out;
264 		tmp |= systab64->con_out;
265 		efi_systab.stderr_handle = systab64->stderr_handle;
266 		tmp |= systab64->stderr_handle;
267 		efi_systab.stderr = systab64->stderr;
268 		tmp |= systab64->stderr;
269 		efi_systab.runtime = data ?
270 				     (void *)(unsigned long)data->runtime :
271 				     (void *)(unsigned long)systab64->runtime;
272 		tmp |= data ? data->runtime : systab64->runtime;
273 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
274 		tmp |= systab64->boottime;
275 		efi_systab.nr_tables = systab64->nr_tables;
276 		efi_systab.tables = data ? (unsigned long)data->tables :
277 					   systab64->tables;
278 		tmp |= data ? data->tables : systab64->tables;
279 
280 		early_memunmap(systab64, sizeof(*systab64));
281 		if (data)
282 			early_memunmap(data, sizeof(*data));
283 #ifdef CONFIG_X86_32
284 		if (tmp >> 32) {
285 			pr_err("EFI data located above 4GB, disabling EFI.\n");
286 			return -EINVAL;
287 		}
288 #endif
289 	} else {
290 		efi_system_table_32_t *systab32;
291 
292 		systab32 = early_memremap((unsigned long)phys,
293 					 sizeof(*systab32));
294 		if (systab32 == NULL) {
295 			pr_err("Couldn't map the system table!\n");
296 			return -ENOMEM;
297 		}
298 
299 		efi_systab.hdr = systab32->hdr;
300 		efi_systab.fw_vendor = systab32->fw_vendor;
301 		efi_systab.fw_revision = systab32->fw_revision;
302 		efi_systab.con_in_handle = systab32->con_in_handle;
303 		efi_systab.con_in = systab32->con_in;
304 		efi_systab.con_out_handle = systab32->con_out_handle;
305 		efi_systab.con_out = systab32->con_out;
306 		efi_systab.stderr_handle = systab32->stderr_handle;
307 		efi_systab.stderr = systab32->stderr;
308 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
309 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
310 		efi_systab.nr_tables = systab32->nr_tables;
311 		efi_systab.tables = systab32->tables;
312 
313 		early_memunmap(systab32, sizeof(*systab32));
314 	}
315 
316 	efi.systab = &efi_systab;
317 
318 	/*
319 	 * Verify the EFI Table
320 	 */
321 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
322 		pr_err("System table signature incorrect!\n");
323 		return -EINVAL;
324 	}
325 	if ((efi.systab->hdr.revision >> 16) == 0)
326 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
327 		       efi.systab->hdr.revision >> 16,
328 		       efi.systab->hdr.revision & 0xffff);
329 
330 	set_bit(EFI_SYSTEM_TABLES, &efi.flags);
331 
332 	return 0;
333 }
334 
335 static int __init efi_runtime_init32(void)
336 {
337 	efi_runtime_services_32_t *runtime;
338 
339 	runtime = early_memremap((unsigned long)efi.systab->runtime,
340 			sizeof(efi_runtime_services_32_t));
341 	if (!runtime) {
342 		pr_err("Could not map the runtime service table!\n");
343 		return -ENOMEM;
344 	}
345 
346 	/*
347 	 * We will only need *early* access to the following two
348 	 * EFI runtime services before set_virtual_address_map
349 	 * is invoked.
350 	 */
351 	efi_phys.set_virtual_address_map =
352 			(efi_set_virtual_address_map_t *)
353 			(unsigned long)runtime->set_virtual_address_map;
354 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
355 
356 	return 0;
357 }
358 
359 static int __init efi_runtime_init64(void)
360 {
361 	efi_runtime_services_64_t *runtime;
362 
363 	runtime = early_memremap((unsigned long)efi.systab->runtime,
364 			sizeof(efi_runtime_services_64_t));
365 	if (!runtime) {
366 		pr_err("Could not map the runtime service table!\n");
367 		return -ENOMEM;
368 	}
369 
370 	/*
371 	 * We will only need *early* access to the following two
372 	 * EFI runtime services before set_virtual_address_map
373 	 * is invoked.
374 	 */
375 	efi_phys.set_virtual_address_map =
376 			(efi_set_virtual_address_map_t *)
377 			(unsigned long)runtime->set_virtual_address_map;
378 	early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
379 
380 	return 0;
381 }
382 
383 static int __init efi_runtime_init(void)
384 {
385 	int rv;
386 
387 	/*
388 	 * Check out the runtime services table. We need to map
389 	 * the runtime services table so that we can grab the physical
390 	 * address of several of the EFI runtime functions, needed to
391 	 * set the firmware into virtual mode.
392 	 *
393 	 * When EFI_PARAVIRT is in force then we could not map runtime
394 	 * service memory region because we do not have direct access to it.
395 	 * However, runtime services are available through proxy functions
396 	 * (e.g. in case of Xen dom0 EFI implementation they call special
397 	 * hypercall which executes relevant EFI functions) and that is why
398 	 * they are always enabled.
399 	 */
400 
401 	if (!efi_enabled(EFI_PARAVIRT)) {
402 		if (efi_enabled(EFI_64BIT))
403 			rv = efi_runtime_init64();
404 		else
405 			rv = efi_runtime_init32();
406 
407 		if (rv)
408 			return rv;
409 	}
410 
411 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
412 
413 	return 0;
414 }
415 
416 static int __init efi_memmap_init(void)
417 {
418 	if (efi_enabled(EFI_PARAVIRT))
419 		return 0;
420 
421 	/* Map the EFI memory map */
422 	memmap.map = early_memremap((unsigned long)memmap.phys_map,
423 				   memmap.nr_map * memmap.desc_size);
424 	if (memmap.map == NULL) {
425 		pr_err("Could not map the memory map!\n");
426 		return -ENOMEM;
427 	}
428 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
429 
430 	if (add_efi_memmap)
431 		do_add_efi_memmap();
432 
433 	set_bit(EFI_MEMMAP, &efi.flags);
434 
435 	return 0;
436 }
437 
438 void __init efi_init(void)
439 {
440 	efi_char16_t *c16;
441 	char vendor[100] = "unknown";
442 	int i = 0;
443 	void *tmp;
444 
445 #ifdef CONFIG_X86_32
446 	if (boot_params.efi_info.efi_systab_hi ||
447 	    boot_params.efi_info.efi_memmap_hi) {
448 		pr_info("Table located above 4GB, disabling EFI.\n");
449 		return;
450 	}
451 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
452 #else
453 	efi_phys.systab = (efi_system_table_t *)
454 			  (boot_params.efi_info.efi_systab |
455 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
456 #endif
457 
458 	if (efi_systab_init(efi_phys.systab))
459 		return;
460 
461 	efi.config_table = (unsigned long)efi.systab->tables;
462 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
463 	efi.runtime	 = (unsigned long)efi.systab->runtime;
464 
465 	/*
466 	 * Show what we know for posterity
467 	 */
468 	c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
469 	if (c16) {
470 		for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
471 			vendor[i] = *c16++;
472 		vendor[i] = '\0';
473 	} else
474 		pr_err("Could not map the firmware vendor!\n");
475 	early_memunmap(tmp, 2);
476 
477 	pr_info("EFI v%u.%.02u by %s\n",
478 		efi.systab->hdr.revision >> 16,
479 		efi.systab->hdr.revision & 0xffff, vendor);
480 
481 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
482 		return;
483 
484 	if (efi_config_init(arch_tables))
485 		return;
486 
487 	/*
488 	 * Note: We currently don't support runtime services on an EFI
489 	 * that doesn't match the kernel 32/64-bit mode.
490 	 */
491 
492 	if (!efi_runtime_supported())
493 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
494 	else {
495 		if (disable_runtime || efi_runtime_init())
496 			return;
497 	}
498 	if (efi_memmap_init())
499 		return;
500 
501 	print_efi_memmap();
502 }
503 
504 void __init efi_late_init(void)
505 {
506 	efi_bgrt_init();
507 }
508 
509 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
510 {
511 	u64 addr, npages;
512 
513 	addr = md->virt_addr;
514 	npages = md->num_pages;
515 
516 	memrange_efi_to_native(&addr, &npages);
517 
518 	if (executable)
519 		set_memory_x(addr, npages);
520 	else
521 		set_memory_nx(addr, npages);
522 }
523 
524 void __init runtime_code_page_mkexec(void)
525 {
526 	efi_memory_desc_t *md;
527 	void *p;
528 
529 	/* Make EFI runtime service code area executable */
530 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
531 		md = p;
532 
533 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
534 			continue;
535 
536 		efi_set_executable(md, true);
537 	}
538 }
539 
540 void efi_memory_uc(u64 addr, unsigned long size)
541 {
542 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
543 	u64 npages;
544 
545 	npages = round_up(size, page_shift) / page_shift;
546 	memrange_efi_to_native(&addr, &npages);
547 	set_memory_uc(addr, npages);
548 }
549 
550 void __init old_map_region(efi_memory_desc_t *md)
551 {
552 	u64 start_pfn, end_pfn, end;
553 	unsigned long size;
554 	void *va;
555 
556 	start_pfn = PFN_DOWN(md->phys_addr);
557 	size	  = md->num_pages << PAGE_SHIFT;
558 	end	  = md->phys_addr + size;
559 	end_pfn   = PFN_UP(end);
560 
561 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
562 		va = __va(md->phys_addr);
563 
564 		if (!(md->attribute & EFI_MEMORY_WB))
565 			efi_memory_uc((u64)(unsigned long)va, size);
566 	} else
567 		va = efi_ioremap(md->phys_addr, size,
568 				 md->type, md->attribute);
569 
570 	md->virt_addr = (u64) (unsigned long) va;
571 	if (!va)
572 		pr_err("ioremap of 0x%llX failed!\n",
573 		       (unsigned long long)md->phys_addr);
574 }
575 
576 /* Merge contiguous regions of the same type and attribute */
577 static void __init efi_merge_regions(void)
578 {
579 	void *p;
580 	efi_memory_desc_t *md, *prev_md = NULL;
581 
582 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
583 		u64 prev_size;
584 		md = p;
585 
586 		if (!prev_md) {
587 			prev_md = md;
588 			continue;
589 		}
590 
591 		if (prev_md->type != md->type ||
592 		    prev_md->attribute != md->attribute) {
593 			prev_md = md;
594 			continue;
595 		}
596 
597 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
598 
599 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
600 			prev_md->num_pages += md->num_pages;
601 			md->type = EFI_RESERVED_TYPE;
602 			md->attribute = 0;
603 			continue;
604 		}
605 		prev_md = md;
606 	}
607 }
608 
609 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
610 {
611 	unsigned long size;
612 	u64 end, systab;
613 
614 	size = md->num_pages << EFI_PAGE_SHIFT;
615 	end = md->phys_addr + size;
616 	systab = (u64)(unsigned long)efi_phys.systab;
617 	if (md->phys_addr <= systab && systab < end) {
618 		systab += md->virt_addr - md->phys_addr;
619 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
620 	}
621 }
622 
623 static void __init save_runtime_map(void)
624 {
625 #ifdef CONFIG_KEXEC
626 	efi_memory_desc_t *md;
627 	void *tmp, *p, *q = NULL;
628 	int count = 0;
629 
630 	if (efi_enabled(EFI_OLD_MEMMAP))
631 		return;
632 
633 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
634 		md = p;
635 
636 		if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
637 		    (md->type == EFI_BOOT_SERVICES_CODE) ||
638 		    (md->type == EFI_BOOT_SERVICES_DATA))
639 			continue;
640 		tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
641 		if (!tmp)
642 			goto out;
643 		q = tmp;
644 
645 		memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
646 		count++;
647 	}
648 
649 	efi_runtime_map_setup(q, count, memmap.desc_size);
650 	return;
651 
652 out:
653 	kfree(q);
654 	pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
655 #endif
656 }
657 
658 static void *realloc_pages(void *old_memmap, int old_shift)
659 {
660 	void *ret;
661 
662 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
663 	if (!ret)
664 		goto out;
665 
666 	/*
667 	 * A first-time allocation doesn't have anything to copy.
668 	 */
669 	if (!old_memmap)
670 		return ret;
671 
672 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
673 
674 out:
675 	free_pages((unsigned long)old_memmap, old_shift);
676 	return ret;
677 }
678 
679 /*
680  * Map the efi memory ranges of the runtime services and update new_mmap with
681  * virtual addresses.
682  */
683 static void * __init efi_map_regions(int *count, int *pg_shift)
684 {
685 	void *p, *new_memmap = NULL;
686 	unsigned long left = 0;
687 	efi_memory_desc_t *md;
688 
689 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
690 		md = p;
691 		if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
692 #ifdef CONFIG_X86_64
693 			if (md->type != EFI_BOOT_SERVICES_CODE &&
694 			    md->type != EFI_BOOT_SERVICES_DATA)
695 #endif
696 				continue;
697 		}
698 
699 		efi_map_region(md);
700 		get_systab_virt_addr(md);
701 
702 		if (left < memmap.desc_size) {
703 			new_memmap = realloc_pages(new_memmap, *pg_shift);
704 			if (!new_memmap)
705 				return NULL;
706 
707 			left += PAGE_SIZE << *pg_shift;
708 			(*pg_shift)++;
709 		}
710 
711 		memcpy(new_memmap + (*count * memmap.desc_size), md,
712 		       memmap.desc_size);
713 
714 		left -= memmap.desc_size;
715 		(*count)++;
716 	}
717 
718 	return new_memmap;
719 }
720 
721 static void __init kexec_enter_virtual_mode(void)
722 {
723 #ifdef CONFIG_KEXEC
724 	efi_memory_desc_t *md;
725 	void *p;
726 
727 	efi.systab = NULL;
728 
729 	/*
730 	 * We don't do virtual mode, since we don't do runtime services, on
731 	 * non-native EFI
732 	 */
733 	if (!efi_is_native()) {
734 		efi_unmap_memmap();
735 		return;
736 	}
737 
738 	/*
739 	* Map efi regions which were passed via setup_data. The virt_addr is a
740 	* fixed addr which was used in first kernel of a kexec boot.
741 	*/
742 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
743 		md = p;
744 		efi_map_region_fixed(md); /* FIXME: add error handling */
745 		get_systab_virt_addr(md);
746 	}
747 
748 	save_runtime_map();
749 
750 	BUG_ON(!efi.systab);
751 
752 	efi_sync_low_kernel_mappings();
753 
754 	/*
755 	 * Now that EFI is in virtual mode, update the function
756 	 * pointers in the runtime service table to the new virtual addresses.
757 	 *
758 	 * Call EFI services through wrapper functions.
759 	 */
760 	efi.runtime_version = efi_systab.hdr.revision;
761 
762 	efi_native_runtime_setup();
763 
764 	efi.set_virtual_address_map = NULL;
765 
766 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
767 		runtime_code_page_mkexec();
768 
769 	/* clean DUMMY object */
770 	efi_delete_dummy_variable();
771 #endif
772 }
773 
774 /*
775  * This function will switch the EFI runtime services to virtual mode.
776  * Essentially, we look through the EFI memmap and map every region that
777  * has the runtime attribute bit set in its memory descriptor into the
778  * ->trampoline_pgd page table using a top-down VA allocation scheme.
779  *
780  * The old method which used to update that memory descriptor with the
781  * virtual address obtained from ioremap() is still supported when the
782  * kernel is booted with efi=old_map on its command line. Same old
783  * method enabled the runtime services to be called without having to
784  * thunk back into physical mode for every invocation.
785  *
786  * The new method does a pagetable switch in a preemption-safe manner
787  * so that we're in a different address space when calling a runtime
788  * function. For function arguments passing we do copy the PGDs of the
789  * kernel page table into ->trampoline_pgd prior to each call.
790  *
791  * Specially for kexec boot, efi runtime maps in previous kernel should
792  * be passed in via setup_data. In that case runtime ranges will be mapped
793  * to the same virtual addresses as the first kernel, see
794  * kexec_enter_virtual_mode().
795  */
796 static void __init __efi_enter_virtual_mode(void)
797 {
798 	int count = 0, pg_shift = 0;
799 	void *new_memmap = NULL;
800 	efi_status_t status;
801 
802 	efi.systab = NULL;
803 
804 	efi_merge_regions();
805 	new_memmap = efi_map_regions(&count, &pg_shift);
806 	if (!new_memmap) {
807 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
808 		return;
809 	}
810 
811 	save_runtime_map();
812 
813 	BUG_ON(!efi.systab);
814 
815 	if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift))
816 		return;
817 
818 	efi_sync_low_kernel_mappings();
819 	efi_dump_pagetable();
820 
821 	if (efi_is_native()) {
822 		status = phys_efi_set_virtual_address_map(
823 				memmap.desc_size * count,
824 				memmap.desc_size,
825 				memmap.desc_version,
826 				(efi_memory_desc_t *)__pa(new_memmap));
827 	} else {
828 		status = efi_thunk_set_virtual_address_map(
829 				efi_phys.set_virtual_address_map,
830 				memmap.desc_size * count,
831 				memmap.desc_size,
832 				memmap.desc_version,
833 				(efi_memory_desc_t *)__pa(new_memmap));
834 	}
835 
836 	if (status != EFI_SUCCESS) {
837 		pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
838 			 status);
839 		panic("EFI call to SetVirtualAddressMap() failed!");
840 	}
841 
842 	/*
843 	 * Now that EFI is in virtual mode, update the function
844 	 * pointers in the runtime service table to the new virtual addresses.
845 	 *
846 	 * Call EFI services through wrapper functions.
847 	 */
848 	efi.runtime_version = efi_systab.hdr.revision;
849 
850 	if (efi_is_native())
851 		efi_native_runtime_setup();
852 	else
853 		efi_thunk_runtime_setup();
854 
855 	efi.set_virtual_address_map = NULL;
856 
857 	efi_runtime_mkexec();
858 
859 	/*
860 	 * We mapped the descriptor array into the EFI pagetable above but we're
861 	 * not unmapping it here. Here's why:
862 	 *
863 	 * We're copying select PGDs from the kernel page table to the EFI page
864 	 * table and when we do so and make changes to those PGDs like unmapping
865 	 * stuff from them, those changes appear in the kernel page table and we
866 	 * go boom.
867 	 *
868 	 * From setup_real_mode():
869 	 *
870 	 * ...
871 	 * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
872 	 *
873 	 * In this particular case, our allocation is in PGD 0 of the EFI page
874 	 * table but we've copied that PGD from PGD[272] of the EFI page table:
875 	 *
876 	 *	pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
877 	 *
878 	 * where the direct memory mapping in kernel space is.
879 	 *
880 	 * new_memmap's VA comes from that direct mapping and thus clearing it,
881 	 * it would get cleared in the kernel page table too.
882 	 *
883 	 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
884 	 */
885 	free_pages((unsigned long)new_memmap, pg_shift);
886 
887 	/* clean DUMMY object */
888 	efi_delete_dummy_variable();
889 }
890 
891 void __init efi_enter_virtual_mode(void)
892 {
893 	if (efi_enabled(EFI_PARAVIRT))
894 		return;
895 
896 	if (efi_setup)
897 		kexec_enter_virtual_mode();
898 	else
899 		__efi_enter_virtual_mode();
900 }
901 
902 /*
903  * Convenience functions to obtain memory types and attributes
904  */
905 u32 efi_mem_type(unsigned long phys_addr)
906 {
907 	efi_memory_desc_t *md;
908 	void *p;
909 
910 	if (!efi_enabled(EFI_MEMMAP))
911 		return 0;
912 
913 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
914 		md = p;
915 		if ((md->phys_addr <= phys_addr) &&
916 		    (phys_addr < (md->phys_addr +
917 				  (md->num_pages << EFI_PAGE_SHIFT))))
918 			return md->type;
919 	}
920 	return 0;
921 }
922 
923 u64 efi_mem_attributes(unsigned long phys_addr)
924 {
925 	efi_memory_desc_t *md;
926 	void *p;
927 
928 	if (!efi_enabled(EFI_MEMMAP))
929 		return 0;
930 
931 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
932 		md = p;
933 		if ((md->phys_addr <= phys_addr) &&
934 		    (phys_addr < (md->phys_addr +
935 				  (md->num_pages << EFI_PAGE_SHIFT))))
936 			return md->attribute;
937 	}
938 	return 0;
939 }
940 
941 static int __init parse_efi_cmdline(char *str)
942 {
943 	if (*str == '=')
944 		str++;
945 
946 	if (!strncmp(str, "old_map", 7))
947 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
948 
949 	return 0;
950 }
951 early_param("efi", parse_efi_cmdline);
952