xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 5e012745)
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 struct efi efi_phys __initdata;
58 static efi_system_table_t efi_systab __initdata;
59 
60 static efi_config_table_type_t arch_tables[] __initdata = {
61 #ifdef CONFIG_X86_UV
62 	{UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
63 #endif
64 	{NULL_GUID, NULL, NULL},
65 };
66 
67 u64 efi_setup;		/* efi setup_data physical address */
68 
69 static int add_efi_memmap __initdata;
70 static int __init setup_add_efi_memmap(char *arg)
71 {
72 	add_efi_memmap = 1;
73 	return 0;
74 }
75 early_param("add_efi_memmap", setup_add_efi_memmap);
76 
77 static efi_status_t __init phys_efi_set_virtual_address_map(
78 	unsigned long memory_map_size,
79 	unsigned long descriptor_size,
80 	u32 descriptor_version,
81 	efi_memory_desc_t *virtual_map)
82 {
83 	efi_status_t status;
84 	unsigned long flags;
85 	pgd_t *save_pgd;
86 
87 	save_pgd = efi_call_phys_prolog();
88 	if (!save_pgd)
89 		return EFI_ABORTED;
90 
91 	/* Disable interrupts around EFI calls: */
92 	local_irq_save(flags);
93 	status = efi_call_phys(efi_phys.set_virtual_address_map,
94 			       memory_map_size, descriptor_size,
95 			       descriptor_version, virtual_map);
96 	local_irq_restore(flags);
97 
98 	efi_call_phys_epilog(save_pgd);
99 
100 	return status;
101 }
102 
103 void __init efi_find_mirror(void)
104 {
105 	efi_memory_desc_t *md;
106 	u64 mirror_size = 0, total_size = 0;
107 
108 	for_each_efi_memory_desc(md) {
109 		unsigned long long start = md->phys_addr;
110 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
111 
112 		total_size += size;
113 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
114 			memblock_mark_mirror(start, size);
115 			mirror_size += size;
116 		}
117 	}
118 	if (mirror_size)
119 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
120 			mirror_size>>20, total_size>>20);
121 }
122 
123 /*
124  * Tell the kernel about the EFI memory map.  This might include
125  * more than the max 128 entries that can fit in the e820 legacy
126  * (zeropage) memory map.
127  */
128 
129 static void __init do_add_efi_memmap(void)
130 {
131 	efi_memory_desc_t *md;
132 
133 	for_each_efi_memory_desc(md) {
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_TYPE_RAM;
146 			else
147 				e820_type = E820_TYPE_RESERVED;
148 			break;
149 		case EFI_ACPI_RECLAIM_MEMORY:
150 			e820_type = E820_TYPE_ACPI;
151 			break;
152 		case EFI_ACPI_MEMORY_NVS:
153 			e820_type = E820_TYPE_NVS;
154 			break;
155 		case EFI_UNUSABLE_MEMORY:
156 			e820_type = E820_TYPE_UNUSABLE;
157 			break;
158 		case EFI_PERSISTENT_MEMORY:
159 			e820_type = E820_TYPE_PMEM;
160 			break;
161 		default:
162 			/*
163 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
164 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
165 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
166 			 */
167 			e820_type = E820_TYPE_RESERVED;
168 			break;
169 		}
170 		e820__range_add(start, size, e820_type);
171 	}
172 	e820__update_table(e820_table);
173 }
174 
175 int __init efi_memblock_x86_reserve_range(void)
176 {
177 	struct efi_info *e = &boot_params.efi_info;
178 	struct efi_memory_map_data data;
179 	phys_addr_t pmap;
180 	int rv;
181 
182 	if (efi_enabled(EFI_PARAVIRT))
183 		return 0;
184 
185 #ifdef CONFIG_X86_32
186 	/* Can't handle data above 4GB at this time */
187 	if (e->efi_memmap_hi) {
188 		pr_err("Memory map is above 4GB, disabling EFI.\n");
189 		return -EINVAL;
190 	}
191 	pmap =  e->efi_memmap;
192 #else
193 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
194 #endif
195 	data.phys_map		= pmap;
196 	data.size 		= e->efi_memmap_size;
197 	data.desc_size		= e->efi_memdesc_size;
198 	data.desc_version	= e->efi_memdesc_version;
199 
200 	rv = efi_memmap_init_early(&data);
201 	if (rv)
202 		return rv;
203 
204 	if (add_efi_memmap)
205 		do_add_efi_memmap();
206 
207 	WARN(efi.memmap.desc_version != 1,
208 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
209 	     efi.memmap.desc_version);
210 
211 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
212 
213 	return 0;
214 }
215 
216 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
217 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
218 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
219 
220 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
221 {
222 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
223 	u64 end_hi = 0;
224 	char buf[64];
225 
226 	if (md->num_pages == 0) {
227 		end = 0;
228 	} else if (md->num_pages > EFI_PAGES_MAX ||
229 		   EFI_PAGES_MAX - md->num_pages <
230 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
231 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
232 			>> OVERFLOW_ADDR_SHIFT;
233 
234 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
235 			end_hi += 1;
236 	} else {
237 		return true;
238 	}
239 
240 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
241 
242 	if (end_hi) {
243 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
244 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
245 			md->phys_addr, end_hi, end);
246 	} else {
247 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
248 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
249 			md->phys_addr, end);
250 	}
251 	return false;
252 }
253 
254 static void __init efi_clean_memmap(void)
255 {
256 	efi_memory_desc_t *out = efi.memmap.map;
257 	const efi_memory_desc_t *in = out;
258 	const efi_memory_desc_t *end = efi.memmap.map_end;
259 	int i, n_removal;
260 
261 	for (i = n_removal = 0; in < end; i++) {
262 		if (efi_memmap_entry_valid(in, i)) {
263 			if (out != in)
264 				memcpy(out, in, efi.memmap.desc_size);
265 			out = (void *)out + efi.memmap.desc_size;
266 		} else {
267 			n_removal++;
268 		}
269 		in = (void *)in + efi.memmap.desc_size;
270 	}
271 
272 	if (n_removal > 0) {
273 		u64 size = efi.memmap.nr_map - n_removal;
274 
275 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
276 		efi_memmap_install(efi.memmap.phys_map, size);
277 	}
278 }
279 
280 void __init efi_print_memmap(void)
281 {
282 	efi_memory_desc_t *md;
283 	int i = 0;
284 
285 	for_each_efi_memory_desc(md) {
286 		char buf[64];
287 
288 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
289 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
290 			md->phys_addr,
291 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
292 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
293 	}
294 }
295 
296 static int __init efi_systab_init(void *phys)
297 {
298 	if (efi_enabled(EFI_64BIT)) {
299 		efi_system_table_64_t *systab64;
300 		struct efi_setup_data *data = NULL;
301 		u64 tmp = 0;
302 
303 		if (efi_setup) {
304 			data = early_memremap(efi_setup, sizeof(*data));
305 			if (!data)
306 				return -ENOMEM;
307 		}
308 		systab64 = early_memremap((unsigned long)phys,
309 					 sizeof(*systab64));
310 		if (systab64 == NULL) {
311 			pr_err("Couldn't map the system table!\n");
312 			if (data)
313 				early_memunmap(data, sizeof(*data));
314 			return -ENOMEM;
315 		}
316 
317 		efi_systab.hdr = systab64->hdr;
318 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
319 					      systab64->fw_vendor;
320 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
321 		efi_systab.fw_revision = systab64->fw_revision;
322 		efi_systab.con_in_handle = systab64->con_in_handle;
323 		tmp |= systab64->con_in_handle;
324 		efi_systab.con_in = systab64->con_in;
325 		tmp |= systab64->con_in;
326 		efi_systab.con_out_handle = systab64->con_out_handle;
327 		tmp |= systab64->con_out_handle;
328 		efi_systab.con_out = systab64->con_out;
329 		tmp |= systab64->con_out;
330 		efi_systab.stderr_handle = systab64->stderr_handle;
331 		tmp |= systab64->stderr_handle;
332 		efi_systab.stderr = systab64->stderr;
333 		tmp |= systab64->stderr;
334 		efi_systab.runtime = data ?
335 				     (void *)(unsigned long)data->runtime :
336 				     (void *)(unsigned long)systab64->runtime;
337 		tmp |= data ? data->runtime : systab64->runtime;
338 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
339 		tmp |= systab64->boottime;
340 		efi_systab.nr_tables = systab64->nr_tables;
341 		efi_systab.tables = data ? (unsigned long)data->tables :
342 					   systab64->tables;
343 		tmp |= data ? data->tables : systab64->tables;
344 
345 		early_memunmap(systab64, sizeof(*systab64));
346 		if (data)
347 			early_memunmap(data, sizeof(*data));
348 #ifdef CONFIG_X86_32
349 		if (tmp >> 32) {
350 			pr_err("EFI data located above 4GB, disabling EFI.\n");
351 			return -EINVAL;
352 		}
353 #endif
354 	} else {
355 		efi_system_table_32_t *systab32;
356 
357 		systab32 = early_memremap((unsigned long)phys,
358 					 sizeof(*systab32));
359 		if (systab32 == NULL) {
360 			pr_err("Couldn't map the system table!\n");
361 			return -ENOMEM;
362 		}
363 
364 		efi_systab.hdr = systab32->hdr;
365 		efi_systab.fw_vendor = systab32->fw_vendor;
366 		efi_systab.fw_revision = systab32->fw_revision;
367 		efi_systab.con_in_handle = systab32->con_in_handle;
368 		efi_systab.con_in = systab32->con_in;
369 		efi_systab.con_out_handle = systab32->con_out_handle;
370 		efi_systab.con_out = systab32->con_out;
371 		efi_systab.stderr_handle = systab32->stderr_handle;
372 		efi_systab.stderr = systab32->stderr;
373 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
374 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
375 		efi_systab.nr_tables = systab32->nr_tables;
376 		efi_systab.tables = systab32->tables;
377 
378 		early_memunmap(systab32, sizeof(*systab32));
379 	}
380 
381 	efi.systab = &efi_systab;
382 
383 	/*
384 	 * Verify the EFI Table
385 	 */
386 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
387 		pr_err("System table signature incorrect!\n");
388 		return -EINVAL;
389 	}
390 	if ((efi.systab->hdr.revision >> 16) == 0)
391 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
392 		       efi.systab->hdr.revision >> 16,
393 		       efi.systab->hdr.revision & 0xffff);
394 
395 	return 0;
396 }
397 
398 static int __init efi_runtime_init32(void)
399 {
400 	efi_runtime_services_32_t *runtime;
401 
402 	runtime = early_memremap((unsigned long)efi.systab->runtime,
403 			sizeof(efi_runtime_services_32_t));
404 	if (!runtime) {
405 		pr_err("Could not map the runtime service table!\n");
406 		return -ENOMEM;
407 	}
408 
409 	/*
410 	 * We will only need *early* access to the SetVirtualAddressMap
411 	 * EFI runtime service. All other runtime services will be called
412 	 * via the virtual mapping.
413 	 */
414 	efi_phys.set_virtual_address_map =
415 			(efi_set_virtual_address_map_t *)
416 			(unsigned long)runtime->set_virtual_address_map;
417 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
418 
419 	return 0;
420 }
421 
422 static int __init efi_runtime_init64(void)
423 {
424 	efi_runtime_services_64_t *runtime;
425 
426 	runtime = early_memremap((unsigned long)efi.systab->runtime,
427 			sizeof(efi_runtime_services_64_t));
428 	if (!runtime) {
429 		pr_err("Could not map the runtime service table!\n");
430 		return -ENOMEM;
431 	}
432 
433 	/*
434 	 * We will only need *early* access to the SetVirtualAddressMap
435 	 * EFI runtime service. All other runtime services will be called
436 	 * via the virtual mapping.
437 	 */
438 	efi_phys.set_virtual_address_map =
439 			(efi_set_virtual_address_map_t *)
440 			(unsigned long)runtime->set_virtual_address_map;
441 	early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
442 
443 	return 0;
444 }
445 
446 static int __init efi_runtime_init(void)
447 {
448 	int rv;
449 
450 	/*
451 	 * Check out the runtime services table. We need to map
452 	 * the runtime services table so that we can grab the physical
453 	 * address of several of the EFI runtime functions, needed to
454 	 * set the firmware into virtual mode.
455 	 *
456 	 * When EFI_PARAVIRT is in force then we could not map runtime
457 	 * service memory region because we do not have direct access to it.
458 	 * However, runtime services are available through proxy functions
459 	 * (e.g. in case of Xen dom0 EFI implementation they call special
460 	 * hypercall which executes relevant EFI functions) and that is why
461 	 * they are always enabled.
462 	 */
463 
464 	if (!efi_enabled(EFI_PARAVIRT)) {
465 		if (efi_enabled(EFI_64BIT))
466 			rv = efi_runtime_init64();
467 		else
468 			rv = efi_runtime_init32();
469 
470 		if (rv)
471 			return rv;
472 	}
473 
474 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
475 
476 	return 0;
477 }
478 
479 void __init efi_init(void)
480 {
481 	efi_char16_t *c16;
482 	char vendor[100] = "unknown";
483 	int i = 0;
484 	void *tmp;
485 
486 #ifdef CONFIG_X86_32
487 	if (boot_params.efi_info.efi_systab_hi ||
488 	    boot_params.efi_info.efi_memmap_hi) {
489 		pr_info("Table located above 4GB, disabling EFI.\n");
490 		return;
491 	}
492 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
493 #else
494 	efi_phys.systab = (efi_system_table_t *)
495 			  (boot_params.efi_info.efi_systab |
496 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
497 #endif
498 
499 	if (efi_systab_init(efi_phys.systab))
500 		return;
501 
502 	efi.config_table = (unsigned long)efi.systab->tables;
503 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
504 	efi.runtime	 = (unsigned long)efi.systab->runtime;
505 
506 	/*
507 	 * Show what we know for posterity
508 	 */
509 	c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
510 	if (c16) {
511 		for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
512 			vendor[i] = *c16++;
513 		vendor[i] = '\0';
514 	} else
515 		pr_err("Could not map the firmware vendor!\n");
516 	early_memunmap(tmp, 2);
517 
518 	pr_info("EFI v%u.%.02u by %s\n",
519 		efi.systab->hdr.revision >> 16,
520 		efi.systab->hdr.revision & 0xffff, vendor);
521 
522 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
523 		return;
524 
525 	if (efi_config_init(arch_tables))
526 		return;
527 
528 	/*
529 	 * Note: We currently don't support runtime services on an EFI
530 	 * that doesn't match the kernel 32/64-bit mode.
531 	 */
532 
533 	if (!efi_runtime_supported())
534 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
535 	else {
536 		if (efi_runtime_disabled() || efi_runtime_init()) {
537 			efi_memmap_unmap();
538 			return;
539 		}
540 	}
541 
542 	efi_clean_memmap();
543 
544 	if (efi_enabled(EFI_DBG))
545 		efi_print_memmap();
546 }
547 
548 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
549 {
550 	u64 addr, npages;
551 
552 	addr = md->virt_addr;
553 	npages = md->num_pages;
554 
555 	memrange_efi_to_native(&addr, &npages);
556 
557 	if (executable)
558 		set_memory_x(addr, npages);
559 	else
560 		set_memory_nx(addr, npages);
561 }
562 
563 void __init runtime_code_page_mkexec(void)
564 {
565 	efi_memory_desc_t *md;
566 
567 	/* Make EFI runtime service code area executable */
568 	for_each_efi_memory_desc(md) {
569 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
570 			continue;
571 
572 		efi_set_executable(md, true);
573 	}
574 }
575 
576 void __init efi_memory_uc(u64 addr, unsigned long size)
577 {
578 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
579 	u64 npages;
580 
581 	npages = round_up(size, page_shift) / page_shift;
582 	memrange_efi_to_native(&addr, &npages);
583 	set_memory_uc(addr, npages);
584 }
585 
586 void __init old_map_region(efi_memory_desc_t *md)
587 {
588 	u64 start_pfn, end_pfn, end;
589 	unsigned long size;
590 	void *va;
591 
592 	start_pfn = PFN_DOWN(md->phys_addr);
593 	size	  = md->num_pages << PAGE_SHIFT;
594 	end	  = md->phys_addr + size;
595 	end_pfn   = PFN_UP(end);
596 
597 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
598 		va = __va(md->phys_addr);
599 
600 		if (!(md->attribute & EFI_MEMORY_WB))
601 			efi_memory_uc((u64)(unsigned long)va, size);
602 	} else
603 		va = efi_ioremap(md->phys_addr, size,
604 				 md->type, md->attribute);
605 
606 	md->virt_addr = (u64) (unsigned long) va;
607 	if (!va)
608 		pr_err("ioremap of 0x%llX failed!\n",
609 		       (unsigned long long)md->phys_addr);
610 }
611 
612 /* Merge contiguous regions of the same type and attribute */
613 static void __init efi_merge_regions(void)
614 {
615 	efi_memory_desc_t *md, *prev_md = NULL;
616 
617 	for_each_efi_memory_desc(md) {
618 		u64 prev_size;
619 
620 		if (!prev_md) {
621 			prev_md = md;
622 			continue;
623 		}
624 
625 		if (prev_md->type != md->type ||
626 		    prev_md->attribute != md->attribute) {
627 			prev_md = md;
628 			continue;
629 		}
630 
631 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
632 
633 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
634 			prev_md->num_pages += md->num_pages;
635 			md->type = EFI_RESERVED_TYPE;
636 			md->attribute = 0;
637 			continue;
638 		}
639 		prev_md = md;
640 	}
641 }
642 
643 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
644 {
645 	unsigned long size;
646 	u64 end, systab;
647 
648 	size = md->num_pages << EFI_PAGE_SHIFT;
649 	end = md->phys_addr + size;
650 	systab = (u64)(unsigned long)efi_phys.systab;
651 	if (md->phys_addr <= systab && systab < end) {
652 		systab += md->virt_addr - md->phys_addr;
653 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
654 	}
655 }
656 
657 static void *realloc_pages(void *old_memmap, int old_shift)
658 {
659 	void *ret;
660 
661 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
662 	if (!ret)
663 		goto out;
664 
665 	/*
666 	 * A first-time allocation doesn't have anything to copy.
667 	 */
668 	if (!old_memmap)
669 		return ret;
670 
671 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
672 
673 out:
674 	free_pages((unsigned long)old_memmap, old_shift);
675 	return ret;
676 }
677 
678 /*
679  * Iterate the EFI memory map in reverse order because the regions
680  * will be mapped top-down. The end result is the same as if we had
681  * mapped things forward, but doesn't require us to change the
682  * existing implementation of efi_map_region().
683  */
684 static inline void *efi_map_next_entry_reverse(void *entry)
685 {
686 	/* Initial call */
687 	if (!entry)
688 		return efi.memmap.map_end - efi.memmap.desc_size;
689 
690 	entry -= efi.memmap.desc_size;
691 	if (entry < efi.memmap.map)
692 		return NULL;
693 
694 	return entry;
695 }
696 
697 /*
698  * efi_map_next_entry - Return the next EFI memory map descriptor
699  * @entry: Previous EFI memory map descriptor
700  *
701  * This is a helper function to iterate over the EFI memory map, which
702  * we do in different orders depending on the current configuration.
703  *
704  * To begin traversing the memory map @entry must be %NULL.
705  *
706  * Returns %NULL when we reach the end of the memory map.
707  */
708 static void *efi_map_next_entry(void *entry)
709 {
710 	if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
711 		/*
712 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
713 		 * config table feature requires us to map all entries
714 		 * in the same order as they appear in the EFI memory
715 		 * map. That is to say, entry N must have a lower
716 		 * virtual address than entry N+1. This is because the
717 		 * firmware toolchain leaves relative references in
718 		 * the code/data sections, which are split and become
719 		 * separate EFI memory regions. Mapping things
720 		 * out-of-order leads to the firmware accessing
721 		 * unmapped addresses.
722 		 *
723 		 * Since we need to map things this way whether or not
724 		 * the kernel actually makes use of
725 		 * EFI_PROPERTIES_TABLE, let's just switch to this
726 		 * scheme by default for 64-bit.
727 		 */
728 		return efi_map_next_entry_reverse(entry);
729 	}
730 
731 	/* Initial call */
732 	if (!entry)
733 		return efi.memmap.map;
734 
735 	entry += efi.memmap.desc_size;
736 	if (entry >= efi.memmap.map_end)
737 		return NULL;
738 
739 	return entry;
740 }
741 
742 static bool should_map_region(efi_memory_desc_t *md)
743 {
744 	/*
745 	 * Runtime regions always require runtime mappings (obviously).
746 	 */
747 	if (md->attribute & EFI_MEMORY_RUNTIME)
748 		return true;
749 
750 	/*
751 	 * 32-bit EFI doesn't suffer from the bug that requires us to
752 	 * reserve boot services regions, and mixed mode support
753 	 * doesn't exist for 32-bit kernels.
754 	 */
755 	if (IS_ENABLED(CONFIG_X86_32))
756 		return false;
757 
758 	/*
759 	 * Map all of RAM so that we can access arguments in the 1:1
760 	 * mapping when making EFI runtime calls.
761 	 */
762 	if (IS_ENABLED(CONFIG_EFI_MIXED) && !efi_is_native()) {
763 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
764 		    md->type == EFI_LOADER_DATA ||
765 		    md->type == EFI_LOADER_CODE)
766 			return true;
767 	}
768 
769 	/*
770 	 * Map boot services regions as a workaround for buggy
771 	 * firmware that accesses them even when they shouldn't.
772 	 *
773 	 * See efi_{reserve,free}_boot_services().
774 	 */
775 	if (md->type == EFI_BOOT_SERVICES_CODE ||
776 	    md->type == EFI_BOOT_SERVICES_DATA)
777 		return true;
778 
779 	return false;
780 }
781 
782 /*
783  * Map the efi memory ranges of the runtime services and update new_mmap with
784  * virtual addresses.
785  */
786 static void * __init efi_map_regions(int *count, int *pg_shift)
787 {
788 	void *p, *new_memmap = NULL;
789 	unsigned long left = 0;
790 	unsigned long desc_size;
791 	efi_memory_desc_t *md;
792 
793 	desc_size = efi.memmap.desc_size;
794 
795 	p = NULL;
796 	while ((p = efi_map_next_entry(p))) {
797 		md = p;
798 
799 		if (!should_map_region(md))
800 			continue;
801 
802 		efi_map_region(md);
803 		get_systab_virt_addr(md);
804 
805 		if (left < desc_size) {
806 			new_memmap = realloc_pages(new_memmap, *pg_shift);
807 			if (!new_memmap)
808 				return NULL;
809 
810 			left += PAGE_SIZE << *pg_shift;
811 			(*pg_shift)++;
812 		}
813 
814 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
815 
816 		left -= desc_size;
817 		(*count)++;
818 	}
819 
820 	return new_memmap;
821 }
822 
823 static void __init kexec_enter_virtual_mode(void)
824 {
825 #ifdef CONFIG_KEXEC_CORE
826 	efi_memory_desc_t *md;
827 	unsigned int num_pages;
828 
829 	efi.systab = NULL;
830 
831 	/*
832 	 * We don't do virtual mode, since we don't do runtime services, on
833 	 * non-native EFI. With efi=old_map, we don't do runtime services in
834 	 * kexec kernel because in the initial boot something else might
835 	 * have been mapped at these virtual addresses.
836 	 */
837 	if (!efi_is_native() || efi_enabled(EFI_OLD_MEMMAP)) {
838 		efi_memmap_unmap();
839 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
840 		return;
841 	}
842 
843 	if (efi_alloc_page_tables()) {
844 		pr_err("Failed to allocate EFI page tables\n");
845 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
846 		return;
847 	}
848 
849 	/*
850 	* Map efi regions which were passed via setup_data. The virt_addr is a
851 	* fixed addr which was used in first kernel of a kexec boot.
852 	*/
853 	for_each_efi_memory_desc(md) {
854 		efi_map_region_fixed(md); /* FIXME: add error handling */
855 		get_systab_virt_addr(md);
856 	}
857 
858 	/*
859 	 * Unregister the early EFI memmap from efi_init() and install
860 	 * the new EFI memory map.
861 	 */
862 	efi_memmap_unmap();
863 
864 	if (efi_memmap_init_late(efi.memmap.phys_map,
865 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
866 		pr_err("Failed to remap late EFI memory map\n");
867 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
868 		return;
869 	}
870 
871 	BUG_ON(!efi.systab);
872 
873 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
874 	num_pages >>= PAGE_SHIFT;
875 
876 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
877 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
878 		return;
879 	}
880 
881 	efi_sync_low_kernel_mappings();
882 
883 	/*
884 	 * Now that EFI is in virtual mode, update the function
885 	 * pointers in the runtime service table to the new virtual addresses.
886 	 *
887 	 * Call EFI services through wrapper functions.
888 	 */
889 	efi.runtime_version = efi_systab.hdr.revision;
890 
891 	efi_native_runtime_setup();
892 
893 	efi.set_virtual_address_map = NULL;
894 
895 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
896 		runtime_code_page_mkexec();
897 
898 	/* clean DUMMY object */
899 	efi_delete_dummy_variable();
900 #endif
901 }
902 
903 /*
904  * This function will switch the EFI runtime services to virtual mode.
905  * Essentially, we look through the EFI memmap and map every region that
906  * has the runtime attribute bit set in its memory descriptor into the
907  * efi_pgd page table.
908  *
909  * The old method which used to update that memory descriptor with the
910  * virtual address obtained from ioremap() is still supported when the
911  * kernel is booted with efi=old_map on its command line. Same old
912  * method enabled the runtime services to be called without having to
913  * thunk back into physical mode for every invocation.
914  *
915  * The new method does a pagetable switch in a preemption-safe manner
916  * so that we're in a different address space when calling a runtime
917  * function. For function arguments passing we do copy the PUDs of the
918  * kernel page table into efi_pgd prior to each call.
919  *
920  * Specially for kexec boot, efi runtime maps in previous kernel should
921  * be passed in via setup_data. In that case runtime ranges will be mapped
922  * to the same virtual addresses as the first kernel, see
923  * kexec_enter_virtual_mode().
924  */
925 static void __init __efi_enter_virtual_mode(void)
926 {
927 	int count = 0, pg_shift = 0;
928 	void *new_memmap = NULL;
929 	efi_status_t status;
930 	unsigned long pa;
931 
932 	efi.systab = NULL;
933 
934 	if (efi_alloc_page_tables()) {
935 		pr_err("Failed to allocate EFI page tables\n");
936 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
937 		return;
938 	}
939 
940 	efi_merge_regions();
941 	new_memmap = efi_map_regions(&count, &pg_shift);
942 	if (!new_memmap) {
943 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
944 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
945 		return;
946 	}
947 
948 	pa = __pa(new_memmap);
949 
950 	/*
951 	 * Unregister the early EFI memmap from efi_init() and install
952 	 * the new EFI memory map that we are about to pass to the
953 	 * firmware via SetVirtualAddressMap().
954 	 */
955 	efi_memmap_unmap();
956 
957 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
958 		pr_err("Failed to remap late EFI memory map\n");
959 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
960 		return;
961 	}
962 
963 	if (efi_enabled(EFI_DBG)) {
964 		pr_info("EFI runtime memory map:\n");
965 		efi_print_memmap();
966 	}
967 
968 	BUG_ON(!efi.systab);
969 
970 	if (efi_setup_page_tables(pa, 1 << pg_shift)) {
971 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
972 		return;
973 	}
974 
975 	efi_sync_low_kernel_mappings();
976 
977 	if (efi_is_native()) {
978 		status = phys_efi_set_virtual_address_map(
979 				efi.memmap.desc_size * count,
980 				efi.memmap.desc_size,
981 				efi.memmap.desc_version,
982 				(efi_memory_desc_t *)pa);
983 	} else {
984 		status = efi_thunk_set_virtual_address_map(
985 				efi_phys.set_virtual_address_map,
986 				efi.memmap.desc_size * count,
987 				efi.memmap.desc_size,
988 				efi.memmap.desc_version,
989 				(efi_memory_desc_t *)pa);
990 	}
991 
992 	if (status != EFI_SUCCESS) {
993 		pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
994 			 status);
995 		panic("EFI call to SetVirtualAddressMap() failed!");
996 	}
997 
998 	efi_free_boot_services();
999 
1000 	/*
1001 	 * Now that EFI is in virtual mode, update the function
1002 	 * pointers in the runtime service table to the new virtual addresses.
1003 	 *
1004 	 * Call EFI services through wrapper functions.
1005 	 */
1006 	efi.runtime_version = efi_systab.hdr.revision;
1007 
1008 	if (efi_is_native())
1009 		efi_native_runtime_setup();
1010 	else
1011 		efi_thunk_runtime_setup();
1012 
1013 	efi.set_virtual_address_map = NULL;
1014 
1015 	/*
1016 	 * Apply more restrictive page table mapping attributes now that
1017 	 * SVAM() has been called and the firmware has performed all
1018 	 * necessary relocation fixups for the new virtual addresses.
1019 	 */
1020 	efi_runtime_update_mappings();
1021 
1022 	/* clean DUMMY object */
1023 	efi_delete_dummy_variable();
1024 }
1025 
1026 void __init efi_enter_virtual_mode(void)
1027 {
1028 	if (efi_enabled(EFI_PARAVIRT))
1029 		return;
1030 
1031 	if (efi_setup)
1032 		kexec_enter_virtual_mode();
1033 	else
1034 		__efi_enter_virtual_mode();
1035 
1036 	efi_dump_pagetable();
1037 }
1038 
1039 static int __init arch_parse_efi_cmdline(char *str)
1040 {
1041 	if (!str) {
1042 		pr_warn("need at least one option\n");
1043 		return -EINVAL;
1044 	}
1045 
1046 	if (parse_option_str(str, "old_map"))
1047 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
1048 
1049 	return 0;
1050 }
1051 early_param("efi", arch_parse_efi_cmdline);
1052