xref: /linux/arch/x86/platform/efi/efi.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 int add_efi_memmap __initdata;
74 static int __init setup_add_efi_memmap(char *arg)
75 {
76 	add_efi_memmap = 1;
77 	return 0;
78 }
79 early_param("add_efi_memmap", setup_add_efi_memmap);
80 
81 static efi_status_t __init phys_efi_set_virtual_address_map(
82 	unsigned long memory_map_size,
83 	unsigned long descriptor_size,
84 	u32 descriptor_version,
85 	efi_memory_desc_t *virtual_map)
86 {
87 	efi_status_t status;
88 	unsigned long flags;
89 	pgd_t *save_pgd;
90 
91 	save_pgd = efi_call_phys_prolog();
92 
93 	/* Disable interrupts around EFI calls: */
94 	local_irq_save(flags);
95 	status = efi_call_phys(efi_phys.set_virtual_address_map,
96 			       memory_map_size, descriptor_size,
97 			       descriptor_version, virtual_map);
98 	local_irq_restore(flags);
99 
100 	efi_call_phys_epilog(save_pgd);
101 
102 	return status;
103 }
104 
105 void efi_get_time(struct timespec *now)
106 {
107 	efi_status_t status;
108 	efi_time_t eft;
109 	efi_time_cap_t cap;
110 
111 	status = efi.get_time(&eft, &cap);
112 	if (status != EFI_SUCCESS)
113 		pr_err("Oops: efitime: can't read time!\n");
114 
115 	now->tv_sec = mktime(eft.year, eft.month, eft.day, eft.hour,
116 			     eft.minute, eft.second);
117 	now->tv_nsec = 0;
118 }
119 
120 void __init efi_find_mirror(void)
121 {
122 	void *p;
123 	u64 mirror_size = 0, total_size = 0;
124 
125 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
126 		efi_memory_desc_t *md = p;
127 		unsigned long long start = md->phys_addr;
128 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
129 
130 		total_size += size;
131 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
132 			memblock_mark_mirror(start, size);
133 			mirror_size += size;
134 		}
135 	}
136 	if (mirror_size)
137 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
138 			mirror_size>>20, total_size>>20);
139 }
140 
141 /*
142  * Tell the kernel about the EFI memory map.  This might include
143  * more than the max 128 entries that can fit in the e820 legacy
144  * (zeropage) memory map.
145  */
146 
147 static void __init do_add_efi_memmap(void)
148 {
149 	void *p;
150 
151 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
152 		efi_memory_desc_t *md = p;
153 		unsigned long long start = md->phys_addr;
154 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
155 		int e820_type;
156 
157 		switch (md->type) {
158 		case EFI_LOADER_CODE:
159 		case EFI_LOADER_DATA:
160 		case EFI_BOOT_SERVICES_CODE:
161 		case EFI_BOOT_SERVICES_DATA:
162 		case EFI_CONVENTIONAL_MEMORY:
163 			if (md->attribute & EFI_MEMORY_WB)
164 				e820_type = E820_RAM;
165 			else
166 				e820_type = E820_RESERVED;
167 			break;
168 		case EFI_ACPI_RECLAIM_MEMORY:
169 			e820_type = E820_ACPI;
170 			break;
171 		case EFI_ACPI_MEMORY_NVS:
172 			e820_type = E820_NVS;
173 			break;
174 		case EFI_UNUSABLE_MEMORY:
175 			e820_type = E820_UNUSABLE;
176 			break;
177 		case EFI_PERSISTENT_MEMORY:
178 			e820_type = E820_PMEM;
179 			break;
180 		default:
181 			/*
182 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
183 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
184 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
185 			 */
186 			e820_type = E820_RESERVED;
187 			break;
188 		}
189 		e820_add_region(start, size, e820_type);
190 	}
191 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
192 }
193 
194 int __init efi_memblock_x86_reserve_range(void)
195 {
196 	struct efi_info *e = &boot_params.efi_info;
197 	phys_addr_t pmap;
198 
199 	if (efi_enabled(EFI_PARAVIRT))
200 		return 0;
201 
202 #ifdef CONFIG_X86_32
203 	/* Can't handle data above 4GB at this time */
204 	if (e->efi_memmap_hi) {
205 		pr_err("Memory map is above 4GB, disabling EFI.\n");
206 		return -EINVAL;
207 	}
208 	pmap =  e->efi_memmap;
209 #else
210 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
211 #endif
212 	memmap.phys_map		= pmap;
213 	memmap.nr_map		= e->efi_memmap_size /
214 				  e->efi_memdesc_size;
215 	memmap.desc_size	= e->efi_memdesc_size;
216 	memmap.desc_version	= e->efi_memdesc_version;
217 
218 	memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
219 
220 	efi.memmap = &memmap;
221 
222 	return 0;
223 }
224 
225 void __init efi_print_memmap(void)
226 {
227 #ifdef EFI_DEBUG
228 	efi_memory_desc_t *md;
229 	void *p;
230 	int i;
231 
232 	for (p = memmap.map, i = 0;
233 	     p < memmap.map_end;
234 	     p += memmap.desc_size, i++) {
235 		char buf[64];
236 
237 		md = p;
238 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx) (%lluMB)\n",
239 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
240 			md->phys_addr,
241 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
242 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
243 	}
244 #endif  /*  EFI_DEBUG  */
245 }
246 
247 void __init efi_unmap_memmap(void)
248 {
249 	clear_bit(EFI_MEMMAP, &efi.flags);
250 	if (memmap.map) {
251 		early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size);
252 		memmap.map = NULL;
253 	}
254 }
255 
256 static int __init efi_systab_init(void *phys)
257 {
258 	if (efi_enabled(EFI_64BIT)) {
259 		efi_system_table_64_t *systab64;
260 		struct efi_setup_data *data = NULL;
261 		u64 tmp = 0;
262 
263 		if (efi_setup) {
264 			data = early_memremap(efi_setup, sizeof(*data));
265 			if (!data)
266 				return -ENOMEM;
267 		}
268 		systab64 = early_memremap((unsigned long)phys,
269 					 sizeof(*systab64));
270 		if (systab64 == NULL) {
271 			pr_err("Couldn't map the system table!\n");
272 			if (data)
273 				early_memunmap(data, sizeof(*data));
274 			return -ENOMEM;
275 		}
276 
277 		efi_systab.hdr = systab64->hdr;
278 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
279 					      systab64->fw_vendor;
280 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
281 		efi_systab.fw_revision = systab64->fw_revision;
282 		efi_systab.con_in_handle = systab64->con_in_handle;
283 		tmp |= systab64->con_in_handle;
284 		efi_systab.con_in = systab64->con_in;
285 		tmp |= systab64->con_in;
286 		efi_systab.con_out_handle = systab64->con_out_handle;
287 		tmp |= systab64->con_out_handle;
288 		efi_systab.con_out = systab64->con_out;
289 		tmp |= systab64->con_out;
290 		efi_systab.stderr_handle = systab64->stderr_handle;
291 		tmp |= systab64->stderr_handle;
292 		efi_systab.stderr = systab64->stderr;
293 		tmp |= systab64->stderr;
294 		efi_systab.runtime = data ?
295 				     (void *)(unsigned long)data->runtime :
296 				     (void *)(unsigned long)systab64->runtime;
297 		tmp |= data ? data->runtime : systab64->runtime;
298 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
299 		tmp |= systab64->boottime;
300 		efi_systab.nr_tables = systab64->nr_tables;
301 		efi_systab.tables = data ? (unsigned long)data->tables :
302 					   systab64->tables;
303 		tmp |= data ? data->tables : systab64->tables;
304 
305 		early_memunmap(systab64, sizeof(*systab64));
306 		if (data)
307 			early_memunmap(data, sizeof(*data));
308 #ifdef CONFIG_X86_32
309 		if (tmp >> 32) {
310 			pr_err("EFI data located above 4GB, disabling EFI.\n");
311 			return -EINVAL;
312 		}
313 #endif
314 	} else {
315 		efi_system_table_32_t *systab32;
316 
317 		systab32 = early_memremap((unsigned long)phys,
318 					 sizeof(*systab32));
319 		if (systab32 == NULL) {
320 			pr_err("Couldn't map the system table!\n");
321 			return -ENOMEM;
322 		}
323 
324 		efi_systab.hdr = systab32->hdr;
325 		efi_systab.fw_vendor = systab32->fw_vendor;
326 		efi_systab.fw_revision = systab32->fw_revision;
327 		efi_systab.con_in_handle = systab32->con_in_handle;
328 		efi_systab.con_in = systab32->con_in;
329 		efi_systab.con_out_handle = systab32->con_out_handle;
330 		efi_systab.con_out = systab32->con_out;
331 		efi_systab.stderr_handle = systab32->stderr_handle;
332 		efi_systab.stderr = systab32->stderr;
333 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
334 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
335 		efi_systab.nr_tables = systab32->nr_tables;
336 		efi_systab.tables = systab32->tables;
337 
338 		early_memunmap(systab32, sizeof(*systab32));
339 	}
340 
341 	efi.systab = &efi_systab;
342 
343 	/*
344 	 * Verify the EFI Table
345 	 */
346 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
347 		pr_err("System table signature incorrect!\n");
348 		return -EINVAL;
349 	}
350 	if ((efi.systab->hdr.revision >> 16) == 0)
351 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
352 		       efi.systab->hdr.revision >> 16,
353 		       efi.systab->hdr.revision & 0xffff);
354 
355 	set_bit(EFI_SYSTEM_TABLES, &efi.flags);
356 
357 	return 0;
358 }
359 
360 static int __init efi_runtime_init32(void)
361 {
362 	efi_runtime_services_32_t *runtime;
363 
364 	runtime = early_memremap((unsigned long)efi.systab->runtime,
365 			sizeof(efi_runtime_services_32_t));
366 	if (!runtime) {
367 		pr_err("Could not map the runtime service table!\n");
368 		return -ENOMEM;
369 	}
370 
371 	/*
372 	 * We will only need *early* access to the SetVirtualAddressMap
373 	 * EFI runtime service. All other runtime services will be called
374 	 * via the virtual mapping.
375 	 */
376 	efi_phys.set_virtual_address_map =
377 			(efi_set_virtual_address_map_t *)
378 			(unsigned long)runtime->set_virtual_address_map;
379 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
380 
381 	return 0;
382 }
383 
384 static int __init efi_runtime_init64(void)
385 {
386 	efi_runtime_services_64_t *runtime;
387 
388 	runtime = early_memremap((unsigned long)efi.systab->runtime,
389 			sizeof(efi_runtime_services_64_t));
390 	if (!runtime) {
391 		pr_err("Could not map the runtime service table!\n");
392 		return -ENOMEM;
393 	}
394 
395 	/*
396 	 * We will only need *early* access to the SetVirtualAddressMap
397 	 * EFI runtime service. All other runtime services will be called
398 	 * via the virtual mapping.
399 	 */
400 	efi_phys.set_virtual_address_map =
401 			(efi_set_virtual_address_map_t *)
402 			(unsigned long)runtime->set_virtual_address_map;
403 	early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
404 
405 	return 0;
406 }
407 
408 static int __init efi_runtime_init(void)
409 {
410 	int rv;
411 
412 	/*
413 	 * Check out the runtime services table. We need to map
414 	 * the runtime services table so that we can grab the physical
415 	 * address of several of the EFI runtime functions, needed to
416 	 * set the firmware into virtual mode.
417 	 *
418 	 * When EFI_PARAVIRT is in force then we could not map runtime
419 	 * service memory region because we do not have direct access to it.
420 	 * However, runtime services are available through proxy functions
421 	 * (e.g. in case of Xen dom0 EFI implementation they call special
422 	 * hypercall which executes relevant EFI functions) and that is why
423 	 * they are always enabled.
424 	 */
425 
426 	if (!efi_enabled(EFI_PARAVIRT)) {
427 		if (efi_enabled(EFI_64BIT))
428 			rv = efi_runtime_init64();
429 		else
430 			rv = efi_runtime_init32();
431 
432 		if (rv)
433 			return rv;
434 	}
435 
436 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
437 
438 	return 0;
439 }
440 
441 static int __init efi_memmap_init(void)
442 {
443 	if (efi_enabled(EFI_PARAVIRT))
444 		return 0;
445 
446 	/* Map the EFI memory map */
447 	memmap.map = early_memremap((unsigned long)memmap.phys_map,
448 				   memmap.nr_map * memmap.desc_size);
449 	if (memmap.map == NULL) {
450 		pr_err("Could not map the memory map!\n");
451 		return -ENOMEM;
452 	}
453 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
454 
455 	if (add_efi_memmap)
456 		do_add_efi_memmap();
457 
458 	set_bit(EFI_MEMMAP, &efi.flags);
459 
460 	return 0;
461 }
462 
463 void __init efi_init(void)
464 {
465 	efi_char16_t *c16;
466 	char vendor[100] = "unknown";
467 	int i = 0;
468 	void *tmp;
469 
470 #ifdef CONFIG_X86_32
471 	if (boot_params.efi_info.efi_systab_hi ||
472 	    boot_params.efi_info.efi_memmap_hi) {
473 		pr_info("Table located above 4GB, disabling EFI.\n");
474 		return;
475 	}
476 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
477 #else
478 	efi_phys.systab = (efi_system_table_t *)
479 			  (boot_params.efi_info.efi_systab |
480 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
481 #endif
482 
483 	if (efi_systab_init(efi_phys.systab))
484 		return;
485 
486 	efi.config_table = (unsigned long)efi.systab->tables;
487 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
488 	efi.runtime	 = (unsigned long)efi.systab->runtime;
489 
490 	/*
491 	 * Show what we know for posterity
492 	 */
493 	c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
494 	if (c16) {
495 		for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
496 			vendor[i] = *c16++;
497 		vendor[i] = '\0';
498 	} else
499 		pr_err("Could not map the firmware vendor!\n");
500 	early_memunmap(tmp, 2);
501 
502 	pr_info("EFI v%u.%.02u by %s\n",
503 		efi.systab->hdr.revision >> 16,
504 		efi.systab->hdr.revision & 0xffff, vendor);
505 
506 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
507 		return;
508 
509 	if (efi_config_init(arch_tables))
510 		return;
511 
512 	/*
513 	 * Note: We currently don't support runtime services on an EFI
514 	 * that doesn't match the kernel 32/64-bit mode.
515 	 */
516 
517 	if (!efi_runtime_supported())
518 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
519 	else {
520 		if (efi_runtime_disabled() || efi_runtime_init())
521 			return;
522 	}
523 	if (efi_memmap_init())
524 		return;
525 
526 	if (efi_enabled(EFI_DBG))
527 		efi_print_memmap();
528 
529 	efi_esrt_init();
530 }
531 
532 void __init efi_late_init(void)
533 {
534 	efi_bgrt_init();
535 }
536 
537 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
538 {
539 	u64 addr, npages;
540 
541 	addr = md->virt_addr;
542 	npages = md->num_pages;
543 
544 	memrange_efi_to_native(&addr, &npages);
545 
546 	if (executable)
547 		set_memory_x(addr, npages);
548 	else
549 		set_memory_nx(addr, npages);
550 }
551 
552 void __init runtime_code_page_mkexec(void)
553 {
554 	efi_memory_desc_t *md;
555 	void *p;
556 
557 	/* Make EFI runtime service code area executable */
558 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
559 		md = p;
560 
561 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
562 			continue;
563 
564 		efi_set_executable(md, true);
565 	}
566 }
567 
568 void __init efi_memory_uc(u64 addr, unsigned long size)
569 {
570 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
571 	u64 npages;
572 
573 	npages = round_up(size, page_shift) / page_shift;
574 	memrange_efi_to_native(&addr, &npages);
575 	set_memory_uc(addr, npages);
576 }
577 
578 void __init old_map_region(efi_memory_desc_t *md)
579 {
580 	u64 start_pfn, end_pfn, end;
581 	unsigned long size;
582 	void *va;
583 
584 	start_pfn = PFN_DOWN(md->phys_addr);
585 	size	  = md->num_pages << PAGE_SHIFT;
586 	end	  = md->phys_addr + size;
587 	end_pfn   = PFN_UP(end);
588 
589 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
590 		va = __va(md->phys_addr);
591 
592 		if (!(md->attribute & EFI_MEMORY_WB))
593 			efi_memory_uc((u64)(unsigned long)va, size);
594 	} else
595 		va = efi_ioremap(md->phys_addr, size,
596 				 md->type, md->attribute);
597 
598 	md->virt_addr = (u64) (unsigned long) va;
599 	if (!va)
600 		pr_err("ioremap of 0x%llX failed!\n",
601 		       (unsigned long long)md->phys_addr);
602 }
603 
604 /* Merge contiguous regions of the same type and attribute */
605 static void __init efi_merge_regions(void)
606 {
607 	void *p;
608 	efi_memory_desc_t *md, *prev_md = NULL;
609 
610 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
611 		u64 prev_size;
612 		md = p;
613 
614 		if (!prev_md) {
615 			prev_md = md;
616 			continue;
617 		}
618 
619 		if (prev_md->type != md->type ||
620 		    prev_md->attribute != md->attribute) {
621 			prev_md = md;
622 			continue;
623 		}
624 
625 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
626 
627 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
628 			prev_md->num_pages += md->num_pages;
629 			md->type = EFI_RESERVED_TYPE;
630 			md->attribute = 0;
631 			continue;
632 		}
633 		prev_md = md;
634 	}
635 }
636 
637 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
638 {
639 	unsigned long size;
640 	u64 end, systab;
641 
642 	size = md->num_pages << EFI_PAGE_SHIFT;
643 	end = md->phys_addr + size;
644 	systab = (u64)(unsigned long)efi_phys.systab;
645 	if (md->phys_addr <= systab && systab < end) {
646 		systab += md->virt_addr - md->phys_addr;
647 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
648 	}
649 }
650 
651 static void __init save_runtime_map(void)
652 {
653 #ifdef CONFIG_KEXEC_CORE
654 	efi_memory_desc_t *md;
655 	void *tmp, *p, *q = NULL;
656 	int count = 0;
657 
658 	if (efi_enabled(EFI_OLD_MEMMAP))
659 		return;
660 
661 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
662 		md = p;
663 
664 		if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
665 		    (md->type == EFI_BOOT_SERVICES_CODE) ||
666 		    (md->type == EFI_BOOT_SERVICES_DATA))
667 			continue;
668 		tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
669 		if (!tmp)
670 			goto out;
671 		q = tmp;
672 
673 		memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
674 		count++;
675 	}
676 
677 	efi_runtime_map_setup(q, count, memmap.desc_size);
678 	return;
679 
680 out:
681 	kfree(q);
682 	pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
683 #endif
684 }
685 
686 static void *realloc_pages(void *old_memmap, int old_shift)
687 {
688 	void *ret;
689 
690 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
691 	if (!ret)
692 		goto out;
693 
694 	/*
695 	 * A first-time allocation doesn't have anything to copy.
696 	 */
697 	if (!old_memmap)
698 		return ret;
699 
700 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
701 
702 out:
703 	free_pages((unsigned long)old_memmap, old_shift);
704 	return ret;
705 }
706 
707 /*
708  * Iterate the EFI memory map in reverse order because the regions
709  * will be mapped top-down. The end result is the same as if we had
710  * mapped things forward, but doesn't require us to change the
711  * existing implementation of efi_map_region().
712  */
713 static inline void *efi_map_next_entry_reverse(void *entry)
714 {
715 	/* Initial call */
716 	if (!entry)
717 		return memmap.map_end - memmap.desc_size;
718 
719 	entry -= memmap.desc_size;
720 	if (entry < memmap.map)
721 		return NULL;
722 
723 	return entry;
724 }
725 
726 /*
727  * efi_map_next_entry - Return the next EFI memory map descriptor
728  * @entry: Previous EFI memory map descriptor
729  *
730  * This is a helper function to iterate over the EFI memory map, which
731  * we do in different orders depending on the current configuration.
732  *
733  * To begin traversing the memory map @entry must be %NULL.
734  *
735  * Returns %NULL when we reach the end of the memory map.
736  */
737 static void *efi_map_next_entry(void *entry)
738 {
739 	if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
740 		/*
741 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
742 		 * config table feature requires us to map all entries
743 		 * in the same order as they appear in the EFI memory
744 		 * map. That is to say, entry N must have a lower
745 		 * virtual address than entry N+1. This is because the
746 		 * firmware toolchain leaves relative references in
747 		 * the code/data sections, which are split and become
748 		 * separate EFI memory regions. Mapping things
749 		 * out-of-order leads to the firmware accessing
750 		 * unmapped addresses.
751 		 *
752 		 * Since we need to map things this way whether or not
753 		 * the kernel actually makes use of
754 		 * EFI_PROPERTIES_TABLE, let's just switch to this
755 		 * scheme by default for 64-bit.
756 		 */
757 		return efi_map_next_entry_reverse(entry);
758 	}
759 
760 	/* Initial call */
761 	if (!entry)
762 		return memmap.map;
763 
764 	entry += memmap.desc_size;
765 	if (entry >= memmap.map_end)
766 		return NULL;
767 
768 	return entry;
769 }
770 
771 /*
772  * Map the efi memory ranges of the runtime services and update new_mmap with
773  * virtual addresses.
774  */
775 static void * __init efi_map_regions(int *count, int *pg_shift)
776 {
777 	void *p, *new_memmap = NULL;
778 	unsigned long left = 0;
779 	efi_memory_desc_t *md;
780 
781 	p = NULL;
782 	while ((p = efi_map_next_entry(p))) {
783 		md = p;
784 		if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
785 #ifdef CONFIG_X86_64
786 			if (md->type != EFI_BOOT_SERVICES_CODE &&
787 			    md->type != EFI_BOOT_SERVICES_DATA)
788 #endif
789 				continue;
790 		}
791 
792 		efi_map_region(md);
793 		get_systab_virt_addr(md);
794 
795 		if (left < memmap.desc_size) {
796 			new_memmap = realloc_pages(new_memmap, *pg_shift);
797 			if (!new_memmap)
798 				return NULL;
799 
800 			left += PAGE_SIZE << *pg_shift;
801 			(*pg_shift)++;
802 		}
803 
804 		memcpy(new_memmap + (*count * memmap.desc_size), md,
805 		       memmap.desc_size);
806 
807 		left -= memmap.desc_size;
808 		(*count)++;
809 	}
810 
811 	return new_memmap;
812 }
813 
814 static void __init kexec_enter_virtual_mode(void)
815 {
816 #ifdef CONFIG_KEXEC_CORE
817 	efi_memory_desc_t *md;
818 	void *p;
819 
820 	efi.systab = NULL;
821 
822 	/*
823 	 * We don't do virtual mode, since we don't do runtime services, on
824 	 * non-native EFI
825 	 */
826 	if (!efi_is_native()) {
827 		efi_unmap_memmap();
828 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
829 		return;
830 	}
831 
832 	/*
833 	* Map efi regions which were passed via setup_data. The virt_addr is a
834 	* fixed addr which was used in first kernel of a kexec boot.
835 	*/
836 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
837 		md = p;
838 		efi_map_region_fixed(md); /* FIXME: add error handling */
839 		get_systab_virt_addr(md);
840 	}
841 
842 	save_runtime_map();
843 
844 	BUG_ON(!efi.systab);
845 
846 	efi_sync_low_kernel_mappings();
847 
848 	/*
849 	 * Now that EFI is in virtual mode, update the function
850 	 * pointers in the runtime service table to the new virtual addresses.
851 	 *
852 	 * Call EFI services through wrapper functions.
853 	 */
854 	efi.runtime_version = efi_systab.hdr.revision;
855 
856 	efi_native_runtime_setup();
857 
858 	efi.set_virtual_address_map = NULL;
859 
860 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
861 		runtime_code_page_mkexec();
862 
863 	/* clean DUMMY object */
864 	efi_delete_dummy_variable();
865 #endif
866 }
867 
868 /*
869  * This function will switch the EFI runtime services to virtual mode.
870  * Essentially, we look through the EFI memmap and map every region that
871  * has the runtime attribute bit set in its memory descriptor into the
872  * ->trampoline_pgd page table using a top-down VA allocation scheme.
873  *
874  * The old method which used to update that memory descriptor with the
875  * virtual address obtained from ioremap() is still supported when the
876  * kernel is booted with efi=old_map on its command line. Same old
877  * method enabled the runtime services to be called without having to
878  * thunk back into physical mode for every invocation.
879  *
880  * The new method does a pagetable switch in a preemption-safe manner
881  * so that we're in a different address space when calling a runtime
882  * function. For function arguments passing we do copy the PGDs of the
883  * kernel page table into ->trampoline_pgd prior to each call.
884  *
885  * Specially for kexec boot, efi runtime maps in previous kernel should
886  * be passed in via setup_data. In that case runtime ranges will be mapped
887  * to the same virtual addresses as the first kernel, see
888  * kexec_enter_virtual_mode().
889  */
890 static void __init __efi_enter_virtual_mode(void)
891 {
892 	int count = 0, pg_shift = 0;
893 	void *new_memmap = NULL;
894 	efi_status_t status;
895 
896 	efi.systab = NULL;
897 
898 	efi_merge_regions();
899 	new_memmap = efi_map_regions(&count, &pg_shift);
900 	if (!new_memmap) {
901 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
902 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
903 		return;
904 	}
905 
906 	save_runtime_map();
907 
908 	BUG_ON(!efi.systab);
909 
910 	if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) {
911 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
912 		return;
913 	}
914 
915 	efi_sync_low_kernel_mappings();
916 	efi_dump_pagetable();
917 
918 	if (efi_is_native()) {
919 		status = phys_efi_set_virtual_address_map(
920 				memmap.desc_size * count,
921 				memmap.desc_size,
922 				memmap.desc_version,
923 				(efi_memory_desc_t *)__pa(new_memmap));
924 	} else {
925 		status = efi_thunk_set_virtual_address_map(
926 				efi_phys.set_virtual_address_map,
927 				memmap.desc_size * count,
928 				memmap.desc_size,
929 				memmap.desc_version,
930 				(efi_memory_desc_t *)__pa(new_memmap));
931 	}
932 
933 	if (status != EFI_SUCCESS) {
934 		pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
935 			 status);
936 		panic("EFI call to SetVirtualAddressMap() failed!");
937 	}
938 
939 	/*
940 	 * Now that EFI is in virtual mode, update the function
941 	 * pointers in the runtime service table to the new virtual addresses.
942 	 *
943 	 * Call EFI services through wrapper functions.
944 	 */
945 	efi.runtime_version = efi_systab.hdr.revision;
946 
947 	if (efi_is_native())
948 		efi_native_runtime_setup();
949 	else
950 		efi_thunk_runtime_setup();
951 
952 	efi.set_virtual_address_map = NULL;
953 
954 	efi_runtime_mkexec();
955 
956 	/*
957 	 * We mapped the descriptor array into the EFI pagetable above but we're
958 	 * not unmapping it here. Here's why:
959 	 *
960 	 * We're copying select PGDs from the kernel page table to the EFI page
961 	 * table and when we do so and make changes to those PGDs like unmapping
962 	 * stuff from them, those changes appear in the kernel page table and we
963 	 * go boom.
964 	 *
965 	 * From setup_real_mode():
966 	 *
967 	 * ...
968 	 * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
969 	 *
970 	 * In this particular case, our allocation is in PGD 0 of the EFI page
971 	 * table but we've copied that PGD from PGD[272] of the EFI page table:
972 	 *
973 	 *	pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
974 	 *
975 	 * where the direct memory mapping in kernel space is.
976 	 *
977 	 * new_memmap's VA comes from that direct mapping and thus clearing it,
978 	 * it would get cleared in the kernel page table too.
979 	 *
980 	 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
981 	 */
982 	free_pages((unsigned long)new_memmap, pg_shift);
983 
984 	/* clean DUMMY object */
985 	efi_delete_dummy_variable();
986 }
987 
988 void __init efi_enter_virtual_mode(void)
989 {
990 	if (efi_enabled(EFI_PARAVIRT))
991 		return;
992 
993 	if (efi_setup)
994 		kexec_enter_virtual_mode();
995 	else
996 		__efi_enter_virtual_mode();
997 }
998 
999 /*
1000  * Convenience functions to obtain memory types and attributes
1001  */
1002 u32 efi_mem_type(unsigned long phys_addr)
1003 {
1004 	efi_memory_desc_t *md;
1005 	void *p;
1006 
1007 	if (!efi_enabled(EFI_MEMMAP))
1008 		return 0;
1009 
1010 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
1011 		md = p;
1012 		if ((md->phys_addr <= phys_addr) &&
1013 		    (phys_addr < (md->phys_addr +
1014 				  (md->num_pages << EFI_PAGE_SHIFT))))
1015 			return md->type;
1016 	}
1017 	return 0;
1018 }
1019 
1020 static int __init arch_parse_efi_cmdline(char *str)
1021 {
1022 	if (!str) {
1023 		pr_warn("need at least one option\n");
1024 		return -EINVAL;
1025 	}
1026 
1027 	if (parse_option_str(str, "old_map"))
1028 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
1029 
1030 	return 0;
1031 }
1032 early_param("efi", arch_parse_efi_cmdline);
1033