• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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", &uv_systab_phys},
63 #endif
64 	{NULL_GUID, NULL, NULL},
65 };
66 
67 static const unsigned long * const efi_tables[] = {
68 	&efi.mps,
69 	&efi.acpi,
70 	&efi.acpi20,
71 	&efi.smbios,
72 	&efi.smbios3,
73 	&efi.boot_info,
74 	&efi.hcdp,
75 	&efi.uga,
76 #ifdef CONFIG_X86_UV
77 	&uv_systab_phys,
78 #endif
79 	&efi.fw_vendor,
80 	&efi.runtime,
81 	&efi.config_table,
82 	&efi.esrt,
83 	&efi.properties_table,
84 	&efi.mem_attr_table,
85 #ifdef CONFIG_EFI_RCI2_TABLE
86 	&rci2_table_phys,
87 #endif
88 	&efi.tpm_log,
89 	&efi.tpm_final_log,
90 };
91 
92 u64 efi_setup;		/* efi setup_data physical address */
93 
94 static int add_efi_memmap __initdata;
setup_add_efi_memmap(char * arg)95 static int __init setup_add_efi_memmap(char *arg)
96 {
97 	add_efi_memmap = 1;
98 	return 0;
99 }
100 early_param("add_efi_memmap", setup_add_efi_memmap);
101 
phys_efi_set_virtual_address_map(unsigned long memory_map_size,unsigned long descriptor_size,u32 descriptor_version,efi_memory_desc_t * virtual_map)102 static efi_status_t __init phys_efi_set_virtual_address_map(
103 	unsigned long memory_map_size,
104 	unsigned long descriptor_size,
105 	u32 descriptor_version,
106 	efi_memory_desc_t *virtual_map)
107 {
108 	efi_status_t status;
109 	unsigned long flags;
110 	pgd_t *save_pgd;
111 
112 	save_pgd = efi_call_phys_prolog();
113 	if (!save_pgd)
114 		return EFI_ABORTED;
115 
116 	/* Disable interrupts around EFI calls: */
117 	local_irq_save(flags);
118 	status = efi_call_phys(efi_phys.set_virtual_address_map,
119 			       memory_map_size, descriptor_size,
120 			       descriptor_version, virtual_map);
121 	local_irq_restore(flags);
122 
123 	efi_call_phys_epilog(save_pgd);
124 
125 	return status;
126 }
127 
efi_find_mirror(void)128 void __init efi_find_mirror(void)
129 {
130 	efi_memory_desc_t *md;
131 	u64 mirror_size = 0, total_size = 0;
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 
137 		total_size += size;
138 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
139 			memblock_mark_mirror(start, size);
140 			mirror_size += size;
141 		}
142 	}
143 	if (mirror_size)
144 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
145 			mirror_size>>20, total_size>>20);
146 }
147 
148 /*
149  * Tell the kernel about the EFI memory map.  This might include
150  * more than the max 128 entries that can fit in the e820 legacy
151  * (zeropage) memory map.
152  */
153 
do_add_efi_memmap(void)154 static void __init do_add_efi_memmap(void)
155 {
156 	efi_memory_desc_t *md;
157 
158 	for_each_efi_memory_desc(md) {
159 		unsigned long long start = md->phys_addr;
160 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
161 		int e820_type;
162 
163 		switch (md->type) {
164 		case EFI_LOADER_CODE:
165 		case EFI_LOADER_DATA:
166 		case EFI_BOOT_SERVICES_CODE:
167 		case EFI_BOOT_SERVICES_DATA:
168 		case EFI_CONVENTIONAL_MEMORY:
169 			if (md->attribute & EFI_MEMORY_WB)
170 				e820_type = E820_TYPE_RAM;
171 			else
172 				e820_type = E820_TYPE_RESERVED;
173 			break;
174 		case EFI_ACPI_RECLAIM_MEMORY:
175 			e820_type = E820_TYPE_ACPI;
176 			break;
177 		case EFI_ACPI_MEMORY_NVS:
178 			e820_type = E820_TYPE_NVS;
179 			break;
180 		case EFI_UNUSABLE_MEMORY:
181 			e820_type = E820_TYPE_UNUSABLE;
182 			break;
183 		case EFI_PERSISTENT_MEMORY:
184 			e820_type = E820_TYPE_PMEM;
185 			break;
186 		default:
187 			/*
188 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
189 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
190 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
191 			 */
192 			e820_type = E820_TYPE_RESERVED;
193 			break;
194 		}
195 		e820__range_add(start, size, e820_type);
196 	}
197 	e820__update_table(e820_table);
198 }
199 
efi_memblock_x86_reserve_range(void)200 int __init efi_memblock_x86_reserve_range(void)
201 {
202 	struct efi_info *e = &boot_params.efi_info;
203 	struct efi_memory_map_data data;
204 	phys_addr_t pmap;
205 	int rv;
206 
207 	if (efi_enabled(EFI_PARAVIRT))
208 		return 0;
209 
210 #ifdef CONFIG_X86_32
211 	/* Can't handle data above 4GB at this time */
212 	if (e->efi_memmap_hi) {
213 		pr_err("Memory map is above 4GB, disabling EFI.\n");
214 		return -EINVAL;
215 	}
216 	pmap =  e->efi_memmap;
217 #else
218 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
219 #endif
220 	data.phys_map		= pmap;
221 	data.size 		= e->efi_memmap_size;
222 	data.desc_size		= e->efi_memdesc_size;
223 	data.desc_version	= e->efi_memdesc_version;
224 
225 	rv = efi_memmap_init_early(&data);
226 	if (rv)
227 		return rv;
228 
229 	if (add_efi_memmap)
230 		do_add_efi_memmap();
231 
232 	WARN(efi.memmap.desc_version != 1,
233 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
234 	     efi.memmap.desc_version);
235 
236 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
237 
238 	return 0;
239 }
240 
241 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
242 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
243 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
244 
efi_memmap_entry_valid(const efi_memory_desc_t * md,int i)245 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
246 {
247 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
248 	u64 end_hi = 0;
249 	char buf[64];
250 
251 	if (md->num_pages == 0) {
252 		end = 0;
253 	} else if (md->num_pages > EFI_PAGES_MAX ||
254 		   EFI_PAGES_MAX - md->num_pages <
255 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
256 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
257 			>> OVERFLOW_ADDR_SHIFT;
258 
259 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
260 			end_hi += 1;
261 	} else {
262 		return true;
263 	}
264 
265 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
266 
267 	if (end_hi) {
268 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
269 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
270 			md->phys_addr, end_hi, end);
271 	} else {
272 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
273 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
274 			md->phys_addr, end);
275 	}
276 	return false;
277 }
278 
efi_clean_memmap(void)279 static void __init efi_clean_memmap(void)
280 {
281 	efi_memory_desc_t *out = efi.memmap.map;
282 	const efi_memory_desc_t *in = out;
283 	const efi_memory_desc_t *end = efi.memmap.map_end;
284 	int i, n_removal;
285 
286 	for (i = n_removal = 0; in < end; i++) {
287 		if (efi_memmap_entry_valid(in, i)) {
288 			if (out != in)
289 				memcpy(out, in, efi.memmap.desc_size);
290 			out = (void *)out + efi.memmap.desc_size;
291 		} else {
292 			n_removal++;
293 		}
294 		in = (void *)in + efi.memmap.desc_size;
295 	}
296 
297 	if (n_removal > 0) {
298 		u64 size = efi.memmap.nr_map - n_removal;
299 
300 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
301 		efi_memmap_install(efi.memmap.phys_map, size);
302 	}
303 }
304 
efi_print_memmap(void)305 void __init efi_print_memmap(void)
306 {
307 	efi_memory_desc_t *md;
308 	int i = 0;
309 
310 	for_each_efi_memory_desc(md) {
311 		char buf[64];
312 
313 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
314 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
315 			md->phys_addr,
316 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
317 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
318 	}
319 }
320 
efi_systab_init(void * phys)321 static int __init efi_systab_init(void *phys)
322 {
323 	if (efi_enabled(EFI_64BIT)) {
324 		efi_system_table_64_t *systab64;
325 		struct efi_setup_data *data = NULL;
326 		u64 tmp = 0;
327 
328 		if (efi_setup) {
329 			data = early_memremap(efi_setup, sizeof(*data));
330 			if (!data)
331 				return -ENOMEM;
332 		}
333 		systab64 = early_memremap((unsigned long)phys,
334 					 sizeof(*systab64));
335 		if (systab64 == NULL) {
336 			pr_err("Couldn't map the system table!\n");
337 			if (data)
338 				early_memunmap(data, sizeof(*data));
339 			return -ENOMEM;
340 		}
341 
342 		efi_systab.hdr = systab64->hdr;
343 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
344 					      systab64->fw_vendor;
345 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
346 		efi_systab.fw_revision = systab64->fw_revision;
347 		efi_systab.con_in_handle = systab64->con_in_handle;
348 		tmp |= systab64->con_in_handle;
349 		efi_systab.con_in = systab64->con_in;
350 		tmp |= systab64->con_in;
351 		efi_systab.con_out_handle = systab64->con_out_handle;
352 		tmp |= systab64->con_out_handle;
353 		efi_systab.con_out = systab64->con_out;
354 		tmp |= systab64->con_out;
355 		efi_systab.stderr_handle = systab64->stderr_handle;
356 		tmp |= systab64->stderr_handle;
357 		efi_systab.stderr = systab64->stderr;
358 		tmp |= systab64->stderr;
359 		efi_systab.runtime = data ?
360 				     (void *)(unsigned long)data->runtime :
361 				     (void *)(unsigned long)systab64->runtime;
362 		tmp |= data ? data->runtime : systab64->runtime;
363 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
364 		tmp |= systab64->boottime;
365 		efi_systab.nr_tables = systab64->nr_tables;
366 		efi_systab.tables = data ? (unsigned long)data->tables :
367 					   systab64->tables;
368 		tmp |= data ? data->tables : systab64->tables;
369 
370 		early_memunmap(systab64, sizeof(*systab64));
371 		if (data)
372 			early_memunmap(data, sizeof(*data));
373 #ifdef CONFIG_X86_32
374 		if (tmp >> 32) {
375 			pr_err("EFI data located above 4GB, disabling EFI.\n");
376 			return -EINVAL;
377 		}
378 #endif
379 	} else {
380 		efi_system_table_32_t *systab32;
381 
382 		systab32 = early_memremap((unsigned long)phys,
383 					 sizeof(*systab32));
384 		if (systab32 == NULL) {
385 			pr_err("Couldn't map the system table!\n");
386 			return -ENOMEM;
387 		}
388 
389 		efi_systab.hdr = systab32->hdr;
390 		efi_systab.fw_vendor = systab32->fw_vendor;
391 		efi_systab.fw_revision = systab32->fw_revision;
392 		efi_systab.con_in_handle = systab32->con_in_handle;
393 		efi_systab.con_in = systab32->con_in;
394 		efi_systab.con_out_handle = systab32->con_out_handle;
395 		efi_systab.con_out = systab32->con_out;
396 		efi_systab.stderr_handle = systab32->stderr_handle;
397 		efi_systab.stderr = systab32->stderr;
398 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
399 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
400 		efi_systab.nr_tables = systab32->nr_tables;
401 		efi_systab.tables = systab32->tables;
402 
403 		early_memunmap(systab32, sizeof(*systab32));
404 	}
405 
406 	efi.systab = &efi_systab;
407 
408 	/*
409 	 * Verify the EFI Table
410 	 */
411 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
412 		pr_err("System table signature incorrect!\n");
413 		return -EINVAL;
414 	}
415 	if ((efi.systab->hdr.revision >> 16) == 0)
416 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
417 		       efi.systab->hdr.revision >> 16,
418 		       efi.systab->hdr.revision & 0xffff);
419 
420 	return 0;
421 }
422 
efi_runtime_init32(void)423 static int __init efi_runtime_init32(void)
424 {
425 	efi_runtime_services_32_t *runtime;
426 
427 	runtime = early_memremap((unsigned long)efi.systab->runtime,
428 			sizeof(efi_runtime_services_32_t));
429 	if (!runtime) {
430 		pr_err("Could not map the runtime service table!\n");
431 		return -ENOMEM;
432 	}
433 
434 	/*
435 	 * We will only need *early* access to the SetVirtualAddressMap
436 	 * EFI runtime service. All other runtime services will be called
437 	 * via the virtual mapping.
438 	 */
439 	efi_phys.set_virtual_address_map =
440 			(efi_set_virtual_address_map_t *)
441 			(unsigned long)runtime->set_virtual_address_map;
442 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
443 
444 	return 0;
445 }
446 
efi_runtime_init64(void)447 static int __init efi_runtime_init64(void)
448 {
449 	efi_runtime_services_64_t *runtime;
450 
451 	runtime = early_memremap((unsigned long)efi.systab->runtime,
452 			sizeof(efi_runtime_services_64_t));
453 	if (!runtime) {
454 		pr_err("Could not map the runtime service table!\n");
455 		return -ENOMEM;
456 	}
457 
458 	/*
459 	 * We will only need *early* access to the SetVirtualAddressMap
460 	 * EFI runtime service. All other runtime services will be called
461 	 * via the virtual mapping.
462 	 */
463 	efi_phys.set_virtual_address_map =
464 			(efi_set_virtual_address_map_t *)
465 			(unsigned long)runtime->set_virtual_address_map;
466 	early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
467 
468 	return 0;
469 }
470 
efi_runtime_init(void)471 static int __init efi_runtime_init(void)
472 {
473 	int rv;
474 
475 	/*
476 	 * Check out the runtime services table. We need to map
477 	 * the runtime services table so that we can grab the physical
478 	 * address of several of the EFI runtime functions, needed to
479 	 * set the firmware into virtual mode.
480 	 *
481 	 * When EFI_PARAVIRT is in force then we could not map runtime
482 	 * service memory region because we do not have direct access to it.
483 	 * However, runtime services are available through proxy functions
484 	 * (e.g. in case of Xen dom0 EFI implementation they call special
485 	 * hypercall which executes relevant EFI functions) and that is why
486 	 * they are always enabled.
487 	 */
488 
489 	if (!efi_enabled(EFI_PARAVIRT)) {
490 		if (efi_enabled(EFI_64BIT))
491 			rv = efi_runtime_init64();
492 		else
493 			rv = efi_runtime_init32();
494 
495 		if (rv)
496 			return rv;
497 	}
498 
499 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
500 
501 	return 0;
502 }
503 
efi_init(void)504 void __init efi_init(void)
505 {
506 	efi_char16_t *c16;
507 	char vendor[100] = "unknown";
508 	int i = 0;
509 
510 #ifdef CONFIG_X86_32
511 	if (boot_params.efi_info.efi_systab_hi ||
512 	    boot_params.efi_info.efi_memmap_hi) {
513 		pr_info("Table located above 4GB, disabling EFI.\n");
514 		return;
515 	}
516 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
517 #else
518 	efi_phys.systab = (efi_system_table_t *)
519 			  (boot_params.efi_info.efi_systab |
520 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
521 #endif
522 
523 	if (efi_systab_init(efi_phys.systab))
524 		return;
525 
526 	efi.config_table = (unsigned long)efi.systab->tables;
527 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
528 	efi.runtime	 = (unsigned long)efi.systab->runtime;
529 
530 	/*
531 	 * Show what we know for posterity
532 	 */
533 	c16 = early_memremap_ro(efi.systab->fw_vendor,
534 				sizeof(vendor) * sizeof(efi_char16_t));
535 	if (c16) {
536 		for (i = 0; i < sizeof(vendor) - 1 && c16[i]; ++i)
537 			vendor[i] = c16[i];
538 		vendor[i] = '\0';
539 		early_memunmap(c16, sizeof(vendor) * sizeof(efi_char16_t));
540 	} else {
541 		pr_err("Could not map the firmware vendor!\n");
542 	}
543 
544 	pr_info("EFI v%u.%.02u by %s\n",
545 		efi.systab->hdr.revision >> 16,
546 		efi.systab->hdr.revision & 0xffff, vendor);
547 
548 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
549 		return;
550 
551 	if (efi_config_init(arch_tables))
552 		return;
553 
554 	/*
555 	 * Note: We currently don't support runtime services on an EFI
556 	 * that doesn't match the kernel 32/64-bit mode.
557 	 */
558 
559 	if (!efi_runtime_supported())
560 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
561 	else {
562 		if (efi_runtime_disabled() || efi_runtime_init()) {
563 			efi_memmap_unmap();
564 			return;
565 		}
566 	}
567 
568 	efi_clean_memmap();
569 
570 	if (efi_enabled(EFI_DBG))
571 		efi_print_memmap();
572 }
573 
efi_set_executable(efi_memory_desc_t * md,bool executable)574 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
575 {
576 	u64 addr, npages;
577 
578 	addr = md->virt_addr;
579 	npages = md->num_pages;
580 
581 	memrange_efi_to_native(&addr, &npages);
582 
583 	if (executable)
584 		set_memory_x(addr, npages);
585 	else
586 		set_memory_nx(addr, npages);
587 }
588 
runtime_code_page_mkexec(void)589 void __init runtime_code_page_mkexec(void)
590 {
591 	efi_memory_desc_t *md;
592 
593 	/* Make EFI runtime service code area executable */
594 	for_each_efi_memory_desc(md) {
595 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
596 			continue;
597 
598 		efi_set_executable(md, true);
599 	}
600 }
601 
efi_memory_uc(u64 addr,unsigned long size)602 void __init efi_memory_uc(u64 addr, unsigned long size)
603 {
604 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
605 	u64 npages;
606 
607 	npages = round_up(size, page_shift) / page_shift;
608 	memrange_efi_to_native(&addr, &npages);
609 	set_memory_uc(addr, npages);
610 }
611 
old_map_region(efi_memory_desc_t * md)612 void __init old_map_region(efi_memory_desc_t *md)
613 {
614 	u64 start_pfn, end_pfn, end;
615 	unsigned long size;
616 	void *va;
617 
618 	start_pfn = PFN_DOWN(md->phys_addr);
619 	size	  = md->num_pages << PAGE_SHIFT;
620 	end	  = md->phys_addr + size;
621 	end_pfn   = PFN_UP(end);
622 
623 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
624 		va = __va(md->phys_addr);
625 
626 		if (!(md->attribute & EFI_MEMORY_WB))
627 			efi_memory_uc((u64)(unsigned long)va, size);
628 	} else
629 		va = efi_ioremap(md->phys_addr, size,
630 				 md->type, md->attribute);
631 
632 	md->virt_addr = (u64) (unsigned long) va;
633 	if (!va)
634 		pr_err("ioremap of 0x%llX failed!\n",
635 		       (unsigned long long)md->phys_addr);
636 }
637 
638 /* Merge contiguous regions of the same type and attribute */
efi_merge_regions(void)639 static void __init efi_merge_regions(void)
640 {
641 	efi_memory_desc_t *md, *prev_md = NULL;
642 
643 	for_each_efi_memory_desc(md) {
644 		u64 prev_size;
645 
646 		if (!prev_md) {
647 			prev_md = md;
648 			continue;
649 		}
650 
651 		if (prev_md->type != md->type ||
652 		    prev_md->attribute != md->attribute) {
653 			prev_md = md;
654 			continue;
655 		}
656 
657 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
658 
659 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
660 			prev_md->num_pages += md->num_pages;
661 			md->type = EFI_RESERVED_TYPE;
662 			md->attribute = 0;
663 			continue;
664 		}
665 		prev_md = md;
666 	}
667 }
668 
get_systab_virt_addr(efi_memory_desc_t * md)669 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
670 {
671 	unsigned long size;
672 	u64 end, systab;
673 
674 	size = md->num_pages << EFI_PAGE_SHIFT;
675 	end = md->phys_addr + size;
676 	systab = (u64)(unsigned long)efi_phys.systab;
677 	if (md->phys_addr <= systab && systab < end) {
678 		systab += md->virt_addr - md->phys_addr;
679 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
680 	}
681 }
682 
realloc_pages(void * old_memmap,int old_shift)683 static void *realloc_pages(void *old_memmap, int old_shift)
684 {
685 	void *ret;
686 
687 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
688 	if (!ret)
689 		goto out;
690 
691 	/*
692 	 * A first-time allocation doesn't have anything to copy.
693 	 */
694 	if (!old_memmap)
695 		return ret;
696 
697 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
698 
699 out:
700 	free_pages((unsigned long)old_memmap, old_shift);
701 	return ret;
702 }
703 
704 /*
705  * Iterate the EFI memory map in reverse order because the regions
706  * will be mapped top-down. The end result is the same as if we had
707  * mapped things forward, but doesn't require us to change the
708  * existing implementation of efi_map_region().
709  */
efi_map_next_entry_reverse(void * entry)710 static inline void *efi_map_next_entry_reverse(void *entry)
711 {
712 	/* Initial call */
713 	if (!entry)
714 		return efi.memmap.map_end - efi.memmap.desc_size;
715 
716 	entry -= efi.memmap.desc_size;
717 	if (entry < efi.memmap.map)
718 		return NULL;
719 
720 	return entry;
721 }
722 
723 /*
724  * efi_map_next_entry - Return the next EFI memory map descriptor
725  * @entry: Previous EFI memory map descriptor
726  *
727  * This is a helper function to iterate over the EFI memory map, which
728  * we do in different orders depending on the current configuration.
729  *
730  * To begin traversing the memory map @entry must be %NULL.
731  *
732  * Returns %NULL when we reach the end of the memory map.
733  */
efi_map_next_entry(void * entry)734 static void *efi_map_next_entry(void *entry)
735 {
736 	if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
737 		/*
738 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
739 		 * config table feature requires us to map all entries
740 		 * in the same order as they appear in the EFI memory
741 		 * map. That is to say, entry N must have a lower
742 		 * virtual address than entry N+1. This is because the
743 		 * firmware toolchain leaves relative references in
744 		 * the code/data sections, which are split and become
745 		 * separate EFI memory regions. Mapping things
746 		 * out-of-order leads to the firmware accessing
747 		 * unmapped addresses.
748 		 *
749 		 * Since we need to map things this way whether or not
750 		 * the kernel actually makes use of
751 		 * EFI_PROPERTIES_TABLE, let's just switch to this
752 		 * scheme by default for 64-bit.
753 		 */
754 		return efi_map_next_entry_reverse(entry);
755 	}
756 
757 	/* Initial call */
758 	if (!entry)
759 		return efi.memmap.map;
760 
761 	entry += efi.memmap.desc_size;
762 	if (entry >= efi.memmap.map_end)
763 		return NULL;
764 
765 	return entry;
766 }
767 
should_map_region(efi_memory_desc_t * md)768 static bool should_map_region(efi_memory_desc_t *md)
769 {
770 	/*
771 	 * Runtime regions always require runtime mappings (obviously).
772 	 */
773 	if (md->attribute & EFI_MEMORY_RUNTIME)
774 		return true;
775 
776 	/*
777 	 * 32-bit EFI doesn't suffer from the bug that requires us to
778 	 * reserve boot services regions, and mixed mode support
779 	 * doesn't exist for 32-bit kernels.
780 	 */
781 	if (IS_ENABLED(CONFIG_X86_32))
782 		return false;
783 
784 	/*
785 	 * Map all of RAM so that we can access arguments in the 1:1
786 	 * mapping when making EFI runtime calls.
787 	 */
788 	if (IS_ENABLED(CONFIG_EFI_MIXED) && !efi_is_native()) {
789 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
790 		    md->type == EFI_LOADER_DATA ||
791 		    md->type == EFI_LOADER_CODE)
792 			return true;
793 	}
794 
795 	/*
796 	 * Map boot services regions as a workaround for buggy
797 	 * firmware that accesses them even when they shouldn't.
798 	 *
799 	 * See efi_{reserve,free}_boot_services().
800 	 */
801 	if (md->type == EFI_BOOT_SERVICES_CODE ||
802 	    md->type == EFI_BOOT_SERVICES_DATA)
803 		return true;
804 
805 	return false;
806 }
807 
808 /*
809  * Map the efi memory ranges of the runtime services and update new_mmap with
810  * virtual addresses.
811  */
efi_map_regions(int * count,int * pg_shift)812 static void * __init efi_map_regions(int *count, int *pg_shift)
813 {
814 	void *p, *new_memmap = NULL;
815 	unsigned long left = 0;
816 	unsigned long desc_size;
817 	efi_memory_desc_t *md;
818 
819 	desc_size = efi.memmap.desc_size;
820 
821 	p = NULL;
822 	while ((p = efi_map_next_entry(p))) {
823 		md = p;
824 
825 		if (!should_map_region(md))
826 			continue;
827 
828 		efi_map_region(md);
829 		get_systab_virt_addr(md);
830 
831 		if (left < desc_size) {
832 			new_memmap = realloc_pages(new_memmap, *pg_shift);
833 			if (!new_memmap)
834 				return NULL;
835 
836 			left += PAGE_SIZE << *pg_shift;
837 			(*pg_shift)++;
838 		}
839 
840 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
841 
842 		left -= desc_size;
843 		(*count)++;
844 	}
845 
846 	return new_memmap;
847 }
848 
kexec_enter_virtual_mode(void)849 static void __init kexec_enter_virtual_mode(void)
850 {
851 #ifdef CONFIG_KEXEC_CORE
852 	efi_memory_desc_t *md;
853 	unsigned int num_pages;
854 
855 	efi.systab = NULL;
856 
857 	/*
858 	 * We don't do virtual mode, since we don't do runtime services, on
859 	 * non-native EFI. With efi=old_map, we don't do runtime services in
860 	 * kexec kernel because in the initial boot something else might
861 	 * have been mapped at these virtual addresses.
862 	 */
863 	if (!efi_is_native() || efi_enabled(EFI_OLD_MEMMAP)) {
864 		efi_memmap_unmap();
865 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
866 		return;
867 	}
868 
869 	if (efi_alloc_page_tables()) {
870 		pr_err("Failed to allocate EFI page tables\n");
871 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
872 		return;
873 	}
874 
875 	/*
876 	* Map efi regions which were passed via setup_data. The virt_addr is a
877 	* fixed addr which was used in first kernel of a kexec boot.
878 	*/
879 	for_each_efi_memory_desc(md) {
880 		efi_map_region_fixed(md); /* FIXME: add error handling */
881 		get_systab_virt_addr(md);
882 	}
883 
884 	/*
885 	 * Unregister the early EFI memmap from efi_init() and install
886 	 * the new EFI memory map.
887 	 */
888 	efi_memmap_unmap();
889 
890 	if (efi_memmap_init_late(efi.memmap.phys_map,
891 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
892 		pr_err("Failed to remap late EFI memory map\n");
893 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
894 		return;
895 	}
896 
897 	BUG_ON(!efi.systab);
898 
899 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
900 	num_pages >>= PAGE_SHIFT;
901 
902 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
903 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
904 		return;
905 	}
906 
907 	efi_sync_low_kernel_mappings();
908 
909 	/*
910 	 * Now that EFI is in virtual mode, update the function
911 	 * pointers in the runtime service table to the new virtual addresses.
912 	 *
913 	 * Call EFI services through wrapper functions.
914 	 */
915 	efi.runtime_version = efi_systab.hdr.revision;
916 
917 	efi_native_runtime_setup();
918 
919 	efi.set_virtual_address_map = NULL;
920 
921 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
922 		runtime_code_page_mkexec();
923 #endif
924 }
925 
926 /*
927  * This function will switch the EFI runtime services to virtual mode.
928  * Essentially, we look through the EFI memmap and map every region that
929  * has the runtime attribute bit set in its memory descriptor into the
930  * efi_pgd page table.
931  *
932  * The old method which used to update that memory descriptor with the
933  * virtual address obtained from ioremap() is still supported when the
934  * kernel is booted with efi=old_map on its command line. Same old
935  * method enabled the runtime services to be called without having to
936  * thunk back into physical mode for every invocation.
937  *
938  * The new method does a pagetable switch in a preemption-safe manner
939  * so that we're in a different address space when calling a runtime
940  * function. For function arguments passing we do copy the PUDs of the
941  * kernel page table into efi_pgd prior to each call.
942  *
943  * Specially for kexec boot, efi runtime maps in previous kernel should
944  * be passed in via setup_data. In that case runtime ranges will be mapped
945  * to the same virtual addresses as the first kernel, see
946  * kexec_enter_virtual_mode().
947  */
__efi_enter_virtual_mode(void)948 static void __init __efi_enter_virtual_mode(void)
949 {
950 	int count = 0, pg_shift = 0;
951 	void *new_memmap = NULL;
952 	efi_status_t status;
953 	unsigned long pa;
954 
955 	efi.systab = NULL;
956 
957 	if (efi_alloc_page_tables()) {
958 		pr_err("Failed to allocate EFI page tables\n");
959 		goto err;
960 	}
961 
962 	efi_merge_regions();
963 	new_memmap = efi_map_regions(&count, &pg_shift);
964 	if (!new_memmap) {
965 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
966 		goto err;
967 	}
968 
969 	pa = __pa(new_memmap);
970 
971 	/*
972 	 * Unregister the early EFI memmap from efi_init() and install
973 	 * the new EFI memory map that we are about to pass to the
974 	 * firmware via SetVirtualAddressMap().
975 	 */
976 	efi_memmap_unmap();
977 
978 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
979 		pr_err("Failed to remap late EFI memory map\n");
980 		goto err;
981 	}
982 
983 	if (efi_enabled(EFI_DBG)) {
984 		pr_info("EFI runtime memory map:\n");
985 		efi_print_memmap();
986 	}
987 
988 	if (WARN_ON(!efi.systab))
989 		goto err;
990 
991 	if (efi_setup_page_tables(pa, 1 << pg_shift))
992 		goto err;
993 
994 	efi_sync_low_kernel_mappings();
995 
996 	if (efi_is_native()) {
997 		status = phys_efi_set_virtual_address_map(
998 				efi.memmap.desc_size * count,
999 				efi.memmap.desc_size,
1000 				efi.memmap.desc_version,
1001 				(efi_memory_desc_t *)pa);
1002 	} else {
1003 		status = efi_thunk_set_virtual_address_map(
1004 				efi_phys.set_virtual_address_map,
1005 				efi.memmap.desc_size * count,
1006 				efi.memmap.desc_size,
1007 				efi.memmap.desc_version,
1008 				(efi_memory_desc_t *)pa);
1009 	}
1010 
1011 	if (status != EFI_SUCCESS) {
1012 		pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
1013 		       status);
1014 		goto err;
1015 	}
1016 
1017 	efi_free_boot_services();
1018 
1019 	/*
1020 	 * Now that EFI is in virtual mode, update the function
1021 	 * pointers in the runtime service table to the new virtual addresses.
1022 	 *
1023 	 * Call EFI services through wrapper functions.
1024 	 */
1025 	efi.runtime_version = efi_systab.hdr.revision;
1026 
1027 	if (efi_is_native())
1028 		efi_native_runtime_setup();
1029 	else
1030 		efi_thunk_runtime_setup();
1031 
1032 	efi.set_virtual_address_map = NULL;
1033 
1034 	/*
1035 	 * Apply more restrictive page table mapping attributes now that
1036 	 * SVAM() has been called and the firmware has performed all
1037 	 * necessary relocation fixups for the new virtual addresses.
1038 	 */
1039 	efi_runtime_update_mappings();
1040 
1041 	/* clean DUMMY object */
1042 	efi_delete_dummy_variable();
1043 	return;
1044 
1045 err:
1046 	clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
1047 }
1048 
efi_enter_virtual_mode(void)1049 void __init efi_enter_virtual_mode(void)
1050 {
1051 	if (efi_enabled(EFI_PARAVIRT))
1052 		return;
1053 
1054 	if (efi_setup)
1055 		kexec_enter_virtual_mode();
1056 	else
1057 		__efi_enter_virtual_mode();
1058 
1059 	efi_dump_pagetable();
1060 }
1061 
arch_parse_efi_cmdline(char * str)1062 static int __init arch_parse_efi_cmdline(char *str)
1063 {
1064 	if (!str) {
1065 		pr_warn("need at least one option\n");
1066 		return -EINVAL;
1067 	}
1068 
1069 	if (parse_option_str(str, "old_map"))
1070 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
1071 
1072 	return 0;
1073 }
1074 early_param("efi", arch_parse_efi_cmdline);
1075 
efi_is_table_address(unsigned long phys_addr)1076 bool efi_is_table_address(unsigned long phys_addr)
1077 {
1078 	unsigned int i;
1079 
1080 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
1081 		return false;
1082 
1083 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
1084 		if (*(efi_tables[i]) == phys_addr)
1085 			return true;
1086 
1087 	return false;
1088 }
1089