xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 45471cd98decae5fced8b38e46c223f54a924814)
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 /*
121  * Tell the kernel about the EFI memory map.  This might include
122  * more than the max 128 entries that can fit in the e820 legacy
123  * (zeropage) memory map.
124  */
125 
126 static void __init do_add_efi_memmap(void)
127 {
128 	void *p;
129 
130 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
131 		efi_memory_desc_t *md = p;
132 		unsigned long long start = md->phys_addr;
133 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
134 		int e820_type;
135 
136 		switch (md->type) {
137 		case EFI_LOADER_CODE:
138 		case EFI_LOADER_DATA:
139 		case EFI_BOOT_SERVICES_CODE:
140 		case EFI_BOOT_SERVICES_DATA:
141 		case EFI_CONVENTIONAL_MEMORY:
142 			if (md->attribute & EFI_MEMORY_WB)
143 				e820_type = E820_RAM;
144 			else
145 				e820_type = E820_RESERVED;
146 			break;
147 		case EFI_ACPI_RECLAIM_MEMORY:
148 			e820_type = E820_ACPI;
149 			break;
150 		case EFI_ACPI_MEMORY_NVS:
151 			e820_type = E820_NVS;
152 			break;
153 		case EFI_UNUSABLE_MEMORY:
154 			e820_type = E820_UNUSABLE;
155 			break;
156 		default:
157 			/*
158 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
159 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
160 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
161 			 */
162 			e820_type = E820_RESERVED;
163 			break;
164 		}
165 		e820_add_region(start, size, e820_type);
166 	}
167 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
168 }
169 
170 int __init efi_memblock_x86_reserve_range(void)
171 {
172 	struct efi_info *e = &boot_params.efi_info;
173 	unsigned long pmap;
174 
175 	if (efi_enabled(EFI_PARAVIRT))
176 		return 0;
177 
178 #ifdef CONFIG_X86_32
179 	/* Can't handle data above 4GB at this time */
180 	if (e->efi_memmap_hi) {
181 		pr_err("Memory map is above 4GB, disabling EFI.\n");
182 		return -EINVAL;
183 	}
184 	pmap =  e->efi_memmap;
185 #else
186 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
187 #endif
188 	memmap.phys_map		= (void *)pmap;
189 	memmap.nr_map		= e->efi_memmap_size /
190 				  e->efi_memdesc_size;
191 	memmap.desc_size	= e->efi_memdesc_size;
192 	memmap.desc_version	= e->efi_memdesc_version;
193 
194 	memblock_reserve(pmap, memmap.nr_map * memmap.desc_size);
195 
196 	efi.memmap = &memmap;
197 
198 	return 0;
199 }
200 
201 static void __init print_efi_memmap(void)
202 {
203 #ifdef EFI_DEBUG
204 	efi_memory_desc_t *md;
205 	void *p;
206 	int i;
207 
208 	for (p = memmap.map, i = 0;
209 	     p < memmap.map_end;
210 	     p += memmap.desc_size, i++) {
211 		char buf[64];
212 
213 		md = p;
214 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx) (%lluMB)\n",
215 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
216 			md->phys_addr,
217 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
218 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
219 	}
220 #endif  /*  EFI_DEBUG  */
221 }
222 
223 void __init efi_unmap_memmap(void)
224 {
225 	clear_bit(EFI_MEMMAP, &efi.flags);
226 	if (memmap.map) {
227 		early_memunmap(memmap.map, memmap.nr_map * memmap.desc_size);
228 		memmap.map = NULL;
229 	}
230 }
231 
232 static int __init efi_systab_init(void *phys)
233 {
234 	if (efi_enabled(EFI_64BIT)) {
235 		efi_system_table_64_t *systab64;
236 		struct efi_setup_data *data = NULL;
237 		u64 tmp = 0;
238 
239 		if (efi_setup) {
240 			data = early_memremap(efi_setup, sizeof(*data));
241 			if (!data)
242 				return -ENOMEM;
243 		}
244 		systab64 = early_memremap((unsigned long)phys,
245 					 sizeof(*systab64));
246 		if (systab64 == NULL) {
247 			pr_err("Couldn't map the system table!\n");
248 			if (data)
249 				early_memunmap(data, sizeof(*data));
250 			return -ENOMEM;
251 		}
252 
253 		efi_systab.hdr = systab64->hdr;
254 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
255 					      systab64->fw_vendor;
256 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
257 		efi_systab.fw_revision = systab64->fw_revision;
258 		efi_systab.con_in_handle = systab64->con_in_handle;
259 		tmp |= systab64->con_in_handle;
260 		efi_systab.con_in = systab64->con_in;
261 		tmp |= systab64->con_in;
262 		efi_systab.con_out_handle = systab64->con_out_handle;
263 		tmp |= systab64->con_out_handle;
264 		efi_systab.con_out = systab64->con_out;
265 		tmp |= systab64->con_out;
266 		efi_systab.stderr_handle = systab64->stderr_handle;
267 		tmp |= systab64->stderr_handle;
268 		efi_systab.stderr = systab64->stderr;
269 		tmp |= systab64->stderr;
270 		efi_systab.runtime = data ?
271 				     (void *)(unsigned long)data->runtime :
272 				     (void *)(unsigned long)systab64->runtime;
273 		tmp |= data ? data->runtime : systab64->runtime;
274 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
275 		tmp |= systab64->boottime;
276 		efi_systab.nr_tables = systab64->nr_tables;
277 		efi_systab.tables = data ? (unsigned long)data->tables :
278 					   systab64->tables;
279 		tmp |= data ? data->tables : systab64->tables;
280 
281 		early_memunmap(systab64, sizeof(*systab64));
282 		if (data)
283 			early_memunmap(data, sizeof(*data));
284 #ifdef CONFIG_X86_32
285 		if (tmp >> 32) {
286 			pr_err("EFI data located above 4GB, disabling EFI.\n");
287 			return -EINVAL;
288 		}
289 #endif
290 	} else {
291 		efi_system_table_32_t *systab32;
292 
293 		systab32 = early_memremap((unsigned long)phys,
294 					 sizeof(*systab32));
295 		if (systab32 == NULL) {
296 			pr_err("Couldn't map the system table!\n");
297 			return -ENOMEM;
298 		}
299 
300 		efi_systab.hdr = systab32->hdr;
301 		efi_systab.fw_vendor = systab32->fw_vendor;
302 		efi_systab.fw_revision = systab32->fw_revision;
303 		efi_systab.con_in_handle = systab32->con_in_handle;
304 		efi_systab.con_in = systab32->con_in;
305 		efi_systab.con_out_handle = systab32->con_out_handle;
306 		efi_systab.con_out = systab32->con_out;
307 		efi_systab.stderr_handle = systab32->stderr_handle;
308 		efi_systab.stderr = systab32->stderr;
309 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
310 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
311 		efi_systab.nr_tables = systab32->nr_tables;
312 		efi_systab.tables = systab32->tables;
313 
314 		early_memunmap(systab32, sizeof(*systab32));
315 	}
316 
317 	efi.systab = &efi_systab;
318 
319 	/*
320 	 * Verify the EFI Table
321 	 */
322 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
323 		pr_err("System table signature incorrect!\n");
324 		return -EINVAL;
325 	}
326 	if ((efi.systab->hdr.revision >> 16) == 0)
327 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
328 		       efi.systab->hdr.revision >> 16,
329 		       efi.systab->hdr.revision & 0xffff);
330 
331 	set_bit(EFI_SYSTEM_TABLES, &efi.flags);
332 
333 	return 0;
334 }
335 
336 static int __init efi_runtime_init32(void)
337 {
338 	efi_runtime_services_32_t *runtime;
339 
340 	runtime = early_memremap((unsigned long)efi.systab->runtime,
341 			sizeof(efi_runtime_services_32_t));
342 	if (!runtime) {
343 		pr_err("Could not map the runtime service table!\n");
344 		return -ENOMEM;
345 	}
346 
347 	/*
348 	 * We will only need *early* access to the SetVirtualAddressMap
349 	 * EFI runtime service. All other runtime services will be called
350 	 * via the virtual mapping.
351 	 */
352 	efi_phys.set_virtual_address_map =
353 			(efi_set_virtual_address_map_t *)
354 			(unsigned long)runtime->set_virtual_address_map;
355 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
356 
357 	return 0;
358 }
359 
360 static int __init efi_runtime_init64(void)
361 {
362 	efi_runtime_services_64_t *runtime;
363 
364 	runtime = early_memremap((unsigned long)efi.systab->runtime,
365 			sizeof(efi_runtime_services_64_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_64_t));
380 
381 	return 0;
382 }
383 
384 static int __init efi_runtime_init(void)
385 {
386 	int rv;
387 
388 	/*
389 	 * Check out the runtime services table. We need to map
390 	 * the runtime services table so that we can grab the physical
391 	 * address of several of the EFI runtime functions, needed to
392 	 * set the firmware into virtual mode.
393 	 *
394 	 * When EFI_PARAVIRT is in force then we could not map runtime
395 	 * service memory region because we do not have direct access to it.
396 	 * However, runtime services are available through proxy functions
397 	 * (e.g. in case of Xen dom0 EFI implementation they call special
398 	 * hypercall which executes relevant EFI functions) and that is why
399 	 * they are always enabled.
400 	 */
401 
402 	if (!efi_enabled(EFI_PARAVIRT)) {
403 		if (efi_enabled(EFI_64BIT))
404 			rv = efi_runtime_init64();
405 		else
406 			rv = efi_runtime_init32();
407 
408 		if (rv)
409 			return rv;
410 	}
411 
412 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
413 
414 	return 0;
415 }
416 
417 static int __init efi_memmap_init(void)
418 {
419 	if (efi_enabled(EFI_PARAVIRT))
420 		return 0;
421 
422 	/* Map the EFI memory map */
423 	memmap.map = early_memremap((unsigned long)memmap.phys_map,
424 				   memmap.nr_map * memmap.desc_size);
425 	if (memmap.map == NULL) {
426 		pr_err("Could not map the memory map!\n");
427 		return -ENOMEM;
428 	}
429 	memmap.map_end = memmap.map + (memmap.nr_map * memmap.desc_size);
430 
431 	if (add_efi_memmap)
432 		do_add_efi_memmap();
433 
434 	set_bit(EFI_MEMMAP, &efi.flags);
435 
436 	return 0;
437 }
438 
439 void __init efi_init(void)
440 {
441 	efi_char16_t *c16;
442 	char vendor[100] = "unknown";
443 	int i = 0;
444 	void *tmp;
445 
446 #ifdef CONFIG_X86_32
447 	if (boot_params.efi_info.efi_systab_hi ||
448 	    boot_params.efi_info.efi_memmap_hi) {
449 		pr_info("Table located above 4GB, disabling EFI.\n");
450 		return;
451 	}
452 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
453 #else
454 	efi_phys.systab = (efi_system_table_t *)
455 			  (boot_params.efi_info.efi_systab |
456 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
457 #endif
458 
459 	if (efi_systab_init(efi_phys.systab))
460 		return;
461 
462 	efi.config_table = (unsigned long)efi.systab->tables;
463 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
464 	efi.runtime	 = (unsigned long)efi.systab->runtime;
465 
466 	/*
467 	 * Show what we know for posterity
468 	 */
469 	c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
470 	if (c16) {
471 		for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
472 			vendor[i] = *c16++;
473 		vendor[i] = '\0';
474 	} else
475 		pr_err("Could not map the firmware vendor!\n");
476 	early_memunmap(tmp, 2);
477 
478 	pr_info("EFI v%u.%.02u by %s\n",
479 		efi.systab->hdr.revision >> 16,
480 		efi.systab->hdr.revision & 0xffff, vendor);
481 
482 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
483 		return;
484 
485 	if (efi_config_init(arch_tables))
486 		return;
487 
488 	/*
489 	 * Note: We currently don't support runtime services on an EFI
490 	 * that doesn't match the kernel 32/64-bit mode.
491 	 */
492 
493 	if (!efi_runtime_supported())
494 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
495 	else {
496 		if (efi_runtime_disabled() || efi_runtime_init())
497 			return;
498 	}
499 	if (efi_memmap_init())
500 		return;
501 
502 	if (efi_enabled(EFI_DBG))
503 		print_efi_memmap();
504 
505 	efi_esrt_init();
506 }
507 
508 void __init efi_late_init(void)
509 {
510 	efi_bgrt_init();
511 }
512 
513 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
514 {
515 	u64 addr, npages;
516 
517 	addr = md->virt_addr;
518 	npages = md->num_pages;
519 
520 	memrange_efi_to_native(&addr, &npages);
521 
522 	if (executable)
523 		set_memory_x(addr, npages);
524 	else
525 		set_memory_nx(addr, npages);
526 }
527 
528 void __init runtime_code_page_mkexec(void)
529 {
530 	efi_memory_desc_t *md;
531 	void *p;
532 
533 	/* Make EFI runtime service code area executable */
534 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
535 		md = p;
536 
537 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
538 			continue;
539 
540 		efi_set_executable(md, true);
541 	}
542 }
543 
544 void __init efi_memory_uc(u64 addr, unsigned long size)
545 {
546 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
547 	u64 npages;
548 
549 	npages = round_up(size, page_shift) / page_shift;
550 	memrange_efi_to_native(&addr, &npages);
551 	set_memory_uc(addr, npages);
552 }
553 
554 void __init old_map_region(efi_memory_desc_t *md)
555 {
556 	u64 start_pfn, end_pfn, end;
557 	unsigned long size;
558 	void *va;
559 
560 	start_pfn = PFN_DOWN(md->phys_addr);
561 	size	  = md->num_pages << PAGE_SHIFT;
562 	end	  = md->phys_addr + size;
563 	end_pfn   = PFN_UP(end);
564 
565 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
566 		va = __va(md->phys_addr);
567 
568 		if (!(md->attribute & EFI_MEMORY_WB))
569 			efi_memory_uc((u64)(unsigned long)va, size);
570 	} else
571 		va = efi_ioremap(md->phys_addr, size,
572 				 md->type, md->attribute);
573 
574 	md->virt_addr = (u64) (unsigned long) va;
575 	if (!va)
576 		pr_err("ioremap of 0x%llX failed!\n",
577 		       (unsigned long long)md->phys_addr);
578 }
579 
580 /* Merge contiguous regions of the same type and attribute */
581 static void __init efi_merge_regions(void)
582 {
583 	void *p;
584 	efi_memory_desc_t *md, *prev_md = NULL;
585 
586 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
587 		u64 prev_size;
588 		md = p;
589 
590 		if (!prev_md) {
591 			prev_md = md;
592 			continue;
593 		}
594 
595 		if (prev_md->type != md->type ||
596 		    prev_md->attribute != md->attribute) {
597 			prev_md = md;
598 			continue;
599 		}
600 
601 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
602 
603 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
604 			prev_md->num_pages += md->num_pages;
605 			md->type = EFI_RESERVED_TYPE;
606 			md->attribute = 0;
607 			continue;
608 		}
609 		prev_md = md;
610 	}
611 }
612 
613 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
614 {
615 	unsigned long size;
616 	u64 end, systab;
617 
618 	size = md->num_pages << EFI_PAGE_SHIFT;
619 	end = md->phys_addr + size;
620 	systab = (u64)(unsigned long)efi_phys.systab;
621 	if (md->phys_addr <= systab && systab < end) {
622 		systab += md->virt_addr - md->phys_addr;
623 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
624 	}
625 }
626 
627 static void __init save_runtime_map(void)
628 {
629 #ifdef CONFIG_KEXEC
630 	efi_memory_desc_t *md;
631 	void *tmp, *p, *q = NULL;
632 	int count = 0;
633 
634 	if (efi_enabled(EFI_OLD_MEMMAP))
635 		return;
636 
637 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
638 		md = p;
639 
640 		if (!(md->attribute & EFI_MEMORY_RUNTIME) ||
641 		    (md->type == EFI_BOOT_SERVICES_CODE) ||
642 		    (md->type == EFI_BOOT_SERVICES_DATA))
643 			continue;
644 		tmp = krealloc(q, (count + 1) * memmap.desc_size, GFP_KERNEL);
645 		if (!tmp)
646 			goto out;
647 		q = tmp;
648 
649 		memcpy(q + count * memmap.desc_size, md, memmap.desc_size);
650 		count++;
651 	}
652 
653 	efi_runtime_map_setup(q, count, memmap.desc_size);
654 	return;
655 
656 out:
657 	kfree(q);
658 	pr_err("Error saving runtime map, efi runtime on kexec non-functional!!\n");
659 #endif
660 }
661 
662 static void *realloc_pages(void *old_memmap, int old_shift)
663 {
664 	void *ret;
665 
666 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
667 	if (!ret)
668 		goto out;
669 
670 	/*
671 	 * A first-time allocation doesn't have anything to copy.
672 	 */
673 	if (!old_memmap)
674 		return ret;
675 
676 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
677 
678 out:
679 	free_pages((unsigned long)old_memmap, old_shift);
680 	return ret;
681 }
682 
683 /*
684  * Map the efi memory ranges of the runtime services and update new_mmap with
685  * virtual addresses.
686  */
687 static void * __init efi_map_regions(int *count, int *pg_shift)
688 {
689 	void *p, *new_memmap = NULL;
690 	unsigned long left = 0;
691 	efi_memory_desc_t *md;
692 
693 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
694 		md = p;
695 		if (!(md->attribute & EFI_MEMORY_RUNTIME)) {
696 #ifdef CONFIG_X86_64
697 			if (md->type != EFI_BOOT_SERVICES_CODE &&
698 			    md->type != EFI_BOOT_SERVICES_DATA)
699 #endif
700 				continue;
701 		}
702 
703 		efi_map_region(md);
704 		get_systab_virt_addr(md);
705 
706 		if (left < memmap.desc_size) {
707 			new_memmap = realloc_pages(new_memmap, *pg_shift);
708 			if (!new_memmap)
709 				return NULL;
710 
711 			left += PAGE_SIZE << *pg_shift;
712 			(*pg_shift)++;
713 		}
714 
715 		memcpy(new_memmap + (*count * memmap.desc_size), md,
716 		       memmap.desc_size);
717 
718 		left -= memmap.desc_size;
719 		(*count)++;
720 	}
721 
722 	return new_memmap;
723 }
724 
725 static void __init kexec_enter_virtual_mode(void)
726 {
727 #ifdef CONFIG_KEXEC
728 	efi_memory_desc_t *md;
729 	void *p;
730 
731 	efi.systab = NULL;
732 
733 	/*
734 	 * We don't do virtual mode, since we don't do runtime services, on
735 	 * non-native EFI
736 	 */
737 	if (!efi_is_native()) {
738 		efi_unmap_memmap();
739 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
740 		return;
741 	}
742 
743 	/*
744 	* Map efi regions which were passed via setup_data. The virt_addr is a
745 	* fixed addr which was used in first kernel of a kexec boot.
746 	*/
747 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
748 		md = p;
749 		efi_map_region_fixed(md); /* FIXME: add error handling */
750 		get_systab_virt_addr(md);
751 	}
752 
753 	save_runtime_map();
754 
755 	BUG_ON(!efi.systab);
756 
757 	efi_sync_low_kernel_mappings();
758 
759 	/*
760 	 * Now that EFI is in virtual mode, update the function
761 	 * pointers in the runtime service table to the new virtual addresses.
762 	 *
763 	 * Call EFI services through wrapper functions.
764 	 */
765 	efi.runtime_version = efi_systab.hdr.revision;
766 
767 	efi_native_runtime_setup();
768 
769 	efi.set_virtual_address_map = NULL;
770 
771 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
772 		runtime_code_page_mkexec();
773 
774 	/* clean DUMMY object */
775 	efi_delete_dummy_variable();
776 #endif
777 }
778 
779 /*
780  * This function will switch the EFI runtime services to virtual mode.
781  * Essentially, we look through the EFI memmap and map every region that
782  * has the runtime attribute bit set in its memory descriptor into the
783  * ->trampoline_pgd page table using a top-down VA allocation scheme.
784  *
785  * The old method which used to update that memory descriptor with the
786  * virtual address obtained from ioremap() is still supported when the
787  * kernel is booted with efi=old_map on its command line. Same old
788  * method enabled the runtime services to be called without having to
789  * thunk back into physical mode for every invocation.
790  *
791  * The new method does a pagetable switch in a preemption-safe manner
792  * so that we're in a different address space when calling a runtime
793  * function. For function arguments passing we do copy the PGDs of the
794  * kernel page table into ->trampoline_pgd prior to each call.
795  *
796  * Specially for kexec boot, efi runtime maps in previous kernel should
797  * be passed in via setup_data. In that case runtime ranges will be mapped
798  * to the same virtual addresses as the first kernel, see
799  * kexec_enter_virtual_mode().
800  */
801 static void __init __efi_enter_virtual_mode(void)
802 {
803 	int count = 0, pg_shift = 0;
804 	void *new_memmap = NULL;
805 	efi_status_t status;
806 
807 	efi.systab = NULL;
808 
809 	efi_merge_regions();
810 	new_memmap = efi_map_regions(&count, &pg_shift);
811 	if (!new_memmap) {
812 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
813 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
814 		return;
815 	}
816 
817 	save_runtime_map();
818 
819 	BUG_ON(!efi.systab);
820 
821 	if (efi_setup_page_tables(__pa(new_memmap), 1 << pg_shift)) {
822 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
823 		return;
824 	}
825 
826 	efi_sync_low_kernel_mappings();
827 	efi_dump_pagetable();
828 
829 	if (efi_is_native()) {
830 		status = phys_efi_set_virtual_address_map(
831 				memmap.desc_size * count,
832 				memmap.desc_size,
833 				memmap.desc_version,
834 				(efi_memory_desc_t *)__pa(new_memmap));
835 	} else {
836 		status = efi_thunk_set_virtual_address_map(
837 				efi_phys.set_virtual_address_map,
838 				memmap.desc_size * count,
839 				memmap.desc_size,
840 				memmap.desc_version,
841 				(efi_memory_desc_t *)__pa(new_memmap));
842 	}
843 
844 	if (status != EFI_SUCCESS) {
845 		pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
846 			 status);
847 		panic("EFI call to SetVirtualAddressMap() failed!");
848 	}
849 
850 	/*
851 	 * Now that EFI is in virtual mode, update the function
852 	 * pointers in the runtime service table to the new virtual addresses.
853 	 *
854 	 * Call EFI services through wrapper functions.
855 	 */
856 	efi.runtime_version = efi_systab.hdr.revision;
857 
858 	if (efi_is_native())
859 		efi_native_runtime_setup();
860 	else
861 		efi_thunk_runtime_setup();
862 
863 	efi.set_virtual_address_map = NULL;
864 
865 	efi_runtime_mkexec();
866 
867 	/*
868 	 * We mapped the descriptor array into the EFI pagetable above but we're
869 	 * not unmapping it here. Here's why:
870 	 *
871 	 * We're copying select PGDs from the kernel page table to the EFI page
872 	 * table and when we do so and make changes to those PGDs like unmapping
873 	 * stuff from them, those changes appear in the kernel page table and we
874 	 * go boom.
875 	 *
876 	 * From setup_real_mode():
877 	 *
878 	 * ...
879 	 * trampoline_pgd[0] = init_level4_pgt[pgd_index(__PAGE_OFFSET)].pgd;
880 	 *
881 	 * In this particular case, our allocation is in PGD 0 of the EFI page
882 	 * table but we've copied that PGD from PGD[272] of the EFI page table:
883 	 *
884 	 *	pgd_index(__PAGE_OFFSET = 0xffff880000000000) = 272
885 	 *
886 	 * where the direct memory mapping in kernel space is.
887 	 *
888 	 * new_memmap's VA comes from that direct mapping and thus clearing it,
889 	 * it would get cleared in the kernel page table too.
890 	 *
891 	 * efi_cleanup_page_tables(__pa(new_memmap), 1 << pg_shift);
892 	 */
893 	free_pages((unsigned long)new_memmap, pg_shift);
894 
895 	/* clean DUMMY object */
896 	efi_delete_dummy_variable();
897 }
898 
899 void __init efi_enter_virtual_mode(void)
900 {
901 	if (efi_enabled(EFI_PARAVIRT))
902 		return;
903 
904 	if (efi_setup)
905 		kexec_enter_virtual_mode();
906 	else
907 		__efi_enter_virtual_mode();
908 }
909 
910 /*
911  * Convenience functions to obtain memory types and attributes
912  */
913 u32 efi_mem_type(unsigned long phys_addr)
914 {
915 	efi_memory_desc_t *md;
916 	void *p;
917 
918 	if (!efi_enabled(EFI_MEMMAP))
919 		return 0;
920 
921 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
922 		md = p;
923 		if ((md->phys_addr <= phys_addr) &&
924 		    (phys_addr < (md->phys_addr +
925 				  (md->num_pages << EFI_PAGE_SHIFT))))
926 			return md->type;
927 	}
928 	return 0;
929 }
930 
931 u64 efi_mem_attributes(unsigned long phys_addr)
932 {
933 	efi_memory_desc_t *md;
934 	void *p;
935 
936 	if (!efi_enabled(EFI_MEMMAP))
937 		return 0;
938 
939 	for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
940 		md = p;
941 		if ((md->phys_addr <= phys_addr) &&
942 		    (phys_addr < (md->phys_addr +
943 				  (md->num_pages << EFI_PAGE_SHIFT))))
944 			return md->attribute;
945 	}
946 	return 0;
947 }
948 
949 static int __init arch_parse_efi_cmdline(char *str)
950 {
951 	if (parse_option_str(str, "old_map"))
952 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
953 	if (parse_option_str(str, "debug"))
954 		set_bit(EFI_DBG, &efi.flags);
955 
956 	return 0;
957 }
958 early_param("efi", arch_parse_efi_cmdline);
959