xref: /openbmc/linux/arch/x86/platform/efi/efi.c (revision 5828efb95bc43ad6a59f05458d3aed9649dd5a63)
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 };
86 
87 u64 efi_setup;		/* efi setup_data physical address */
88 
89 static int add_efi_memmap __initdata;
90 static int __init setup_add_efi_memmap(char *arg)
91 {
92 	add_efi_memmap = 1;
93 	return 0;
94 }
95 early_param("add_efi_memmap", setup_add_efi_memmap);
96 
97 static efi_status_t __init phys_efi_set_virtual_address_map(
98 	unsigned long memory_map_size,
99 	unsigned long descriptor_size,
100 	u32 descriptor_version,
101 	efi_memory_desc_t *virtual_map)
102 {
103 	efi_status_t status;
104 	unsigned long flags;
105 	pgd_t *save_pgd;
106 
107 	save_pgd = efi_call_phys_prolog();
108 	if (!save_pgd)
109 		return EFI_ABORTED;
110 
111 	/* Disable interrupts around EFI calls: */
112 	local_irq_save(flags);
113 	status = efi_call_phys(efi_phys.set_virtual_address_map,
114 			       memory_map_size, descriptor_size,
115 			       descriptor_version, virtual_map);
116 	local_irq_restore(flags);
117 
118 	efi_call_phys_epilog(save_pgd);
119 
120 	return status;
121 }
122 
123 void __init efi_find_mirror(void)
124 {
125 	efi_memory_desc_t *md;
126 	u64 mirror_size = 0, total_size = 0;
127 
128 	for_each_efi_memory_desc(md) {
129 		unsigned long long start = md->phys_addr;
130 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
131 
132 		total_size += size;
133 		if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
134 			memblock_mark_mirror(start, size);
135 			mirror_size += size;
136 		}
137 	}
138 	if (mirror_size)
139 		pr_info("Memory: %lldM/%lldM mirrored memory\n",
140 			mirror_size>>20, total_size>>20);
141 }
142 
143 /*
144  * Tell the kernel about the EFI memory map.  This might include
145  * more than the max 128 entries that can fit in the e820 legacy
146  * (zeropage) memory map.
147  */
148 
149 static void __init do_add_efi_memmap(void)
150 {
151 	efi_memory_desc_t *md;
152 
153 	for_each_efi_memory_desc(md) {
154 		unsigned long long start = md->phys_addr;
155 		unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
156 		int e820_type;
157 
158 		switch (md->type) {
159 		case EFI_LOADER_CODE:
160 		case EFI_LOADER_DATA:
161 		case EFI_BOOT_SERVICES_CODE:
162 		case EFI_BOOT_SERVICES_DATA:
163 		case EFI_CONVENTIONAL_MEMORY:
164 			if (md->attribute & EFI_MEMORY_WB)
165 				e820_type = E820_TYPE_RAM;
166 			else
167 				e820_type = E820_TYPE_RESERVED;
168 			break;
169 		case EFI_ACPI_RECLAIM_MEMORY:
170 			e820_type = E820_TYPE_ACPI;
171 			break;
172 		case EFI_ACPI_MEMORY_NVS:
173 			e820_type = E820_TYPE_NVS;
174 			break;
175 		case EFI_UNUSABLE_MEMORY:
176 			e820_type = E820_TYPE_UNUSABLE;
177 			break;
178 		case EFI_PERSISTENT_MEMORY:
179 			e820_type = E820_TYPE_PMEM;
180 			break;
181 		default:
182 			/*
183 			 * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
184 			 * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
185 			 * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
186 			 */
187 			e820_type = E820_TYPE_RESERVED;
188 			break;
189 		}
190 		e820__range_add(start, size, e820_type);
191 	}
192 	e820__update_table(e820_table);
193 }
194 
195 int __init efi_memblock_x86_reserve_range(void)
196 {
197 	struct efi_info *e = &boot_params.efi_info;
198 	struct efi_memory_map_data data;
199 	phys_addr_t pmap;
200 	int rv;
201 
202 	if (efi_enabled(EFI_PARAVIRT))
203 		return 0;
204 
205 #ifdef CONFIG_X86_32
206 	/* Can't handle data above 4GB at this time */
207 	if (e->efi_memmap_hi) {
208 		pr_err("Memory map is above 4GB, disabling EFI.\n");
209 		return -EINVAL;
210 	}
211 	pmap =  e->efi_memmap;
212 #else
213 	pmap = (e->efi_memmap |	((__u64)e->efi_memmap_hi << 32));
214 #endif
215 	data.phys_map		= pmap;
216 	data.size 		= e->efi_memmap_size;
217 	data.desc_size		= e->efi_memdesc_size;
218 	data.desc_version	= e->efi_memdesc_version;
219 
220 	rv = efi_memmap_init_early(&data);
221 	if (rv)
222 		return rv;
223 
224 	if (add_efi_memmap)
225 		do_add_efi_memmap();
226 
227 	WARN(efi.memmap.desc_version != 1,
228 	     "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
229 	     efi.memmap.desc_version);
230 
231 	memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
232 
233 	return 0;
234 }
235 
236 #define OVERFLOW_ADDR_SHIFT	(64 - EFI_PAGE_SHIFT)
237 #define OVERFLOW_ADDR_MASK	(U64_MAX << OVERFLOW_ADDR_SHIFT)
238 #define U64_HIGH_BIT		(~(U64_MAX >> 1))
239 
240 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
241 {
242 	u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
243 	u64 end_hi = 0;
244 	char buf[64];
245 
246 	if (md->num_pages == 0) {
247 		end = 0;
248 	} else if (md->num_pages > EFI_PAGES_MAX ||
249 		   EFI_PAGES_MAX - md->num_pages <
250 		   (md->phys_addr >> EFI_PAGE_SHIFT)) {
251 		end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
252 			>> OVERFLOW_ADDR_SHIFT;
253 
254 		if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
255 			end_hi += 1;
256 	} else {
257 		return true;
258 	}
259 
260 	pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
261 
262 	if (end_hi) {
263 		pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
264 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
265 			md->phys_addr, end_hi, end);
266 	} else {
267 		pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
268 			i, efi_md_typeattr_format(buf, sizeof(buf), md),
269 			md->phys_addr, end);
270 	}
271 	return false;
272 }
273 
274 static void __init efi_clean_memmap(void)
275 {
276 	efi_memory_desc_t *out = efi.memmap.map;
277 	const efi_memory_desc_t *in = out;
278 	const efi_memory_desc_t *end = efi.memmap.map_end;
279 	int i, n_removal;
280 
281 	for (i = n_removal = 0; in < end; i++) {
282 		if (efi_memmap_entry_valid(in, i)) {
283 			if (out != in)
284 				memcpy(out, in, efi.memmap.desc_size);
285 			out = (void *)out + efi.memmap.desc_size;
286 		} else {
287 			n_removal++;
288 		}
289 		in = (void *)in + efi.memmap.desc_size;
290 	}
291 
292 	if (n_removal > 0) {
293 		u64 size = efi.memmap.nr_map - n_removal;
294 
295 		pr_warn("Removing %d invalid memory map entries.\n", n_removal);
296 		efi_memmap_install(efi.memmap.phys_map, size);
297 	}
298 }
299 
300 void __init efi_print_memmap(void)
301 {
302 	efi_memory_desc_t *md;
303 	int i = 0;
304 
305 	for_each_efi_memory_desc(md) {
306 		char buf[64];
307 
308 		pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
309 			i++, efi_md_typeattr_format(buf, sizeof(buf), md),
310 			md->phys_addr,
311 			md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
312 			(md->num_pages >> (20 - EFI_PAGE_SHIFT)));
313 	}
314 }
315 
316 static int __init efi_systab_init(void *phys)
317 {
318 	if (efi_enabled(EFI_64BIT)) {
319 		efi_system_table_64_t *systab64;
320 		struct efi_setup_data *data = NULL;
321 		u64 tmp = 0;
322 
323 		if (efi_setup) {
324 			data = early_memremap(efi_setup, sizeof(*data));
325 			if (!data)
326 				return -ENOMEM;
327 		}
328 		systab64 = early_memremap((unsigned long)phys,
329 					 sizeof(*systab64));
330 		if (systab64 == NULL) {
331 			pr_err("Couldn't map the system table!\n");
332 			if (data)
333 				early_memunmap(data, sizeof(*data));
334 			return -ENOMEM;
335 		}
336 
337 		efi_systab.hdr = systab64->hdr;
338 		efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
339 					      systab64->fw_vendor;
340 		tmp |= data ? data->fw_vendor : systab64->fw_vendor;
341 		efi_systab.fw_revision = systab64->fw_revision;
342 		efi_systab.con_in_handle = systab64->con_in_handle;
343 		tmp |= systab64->con_in_handle;
344 		efi_systab.con_in = systab64->con_in;
345 		tmp |= systab64->con_in;
346 		efi_systab.con_out_handle = systab64->con_out_handle;
347 		tmp |= systab64->con_out_handle;
348 		efi_systab.con_out = systab64->con_out;
349 		tmp |= systab64->con_out;
350 		efi_systab.stderr_handle = systab64->stderr_handle;
351 		tmp |= systab64->stderr_handle;
352 		efi_systab.stderr = systab64->stderr;
353 		tmp |= systab64->stderr;
354 		efi_systab.runtime = data ?
355 				     (void *)(unsigned long)data->runtime :
356 				     (void *)(unsigned long)systab64->runtime;
357 		tmp |= data ? data->runtime : systab64->runtime;
358 		efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
359 		tmp |= systab64->boottime;
360 		efi_systab.nr_tables = systab64->nr_tables;
361 		efi_systab.tables = data ? (unsigned long)data->tables :
362 					   systab64->tables;
363 		tmp |= data ? data->tables : systab64->tables;
364 
365 		early_memunmap(systab64, sizeof(*systab64));
366 		if (data)
367 			early_memunmap(data, sizeof(*data));
368 #ifdef CONFIG_X86_32
369 		if (tmp >> 32) {
370 			pr_err("EFI data located above 4GB, disabling EFI.\n");
371 			return -EINVAL;
372 		}
373 #endif
374 	} else {
375 		efi_system_table_32_t *systab32;
376 
377 		systab32 = early_memremap((unsigned long)phys,
378 					 sizeof(*systab32));
379 		if (systab32 == NULL) {
380 			pr_err("Couldn't map the system table!\n");
381 			return -ENOMEM;
382 		}
383 
384 		efi_systab.hdr = systab32->hdr;
385 		efi_systab.fw_vendor = systab32->fw_vendor;
386 		efi_systab.fw_revision = systab32->fw_revision;
387 		efi_systab.con_in_handle = systab32->con_in_handle;
388 		efi_systab.con_in = systab32->con_in;
389 		efi_systab.con_out_handle = systab32->con_out_handle;
390 		efi_systab.con_out = systab32->con_out;
391 		efi_systab.stderr_handle = systab32->stderr_handle;
392 		efi_systab.stderr = systab32->stderr;
393 		efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
394 		efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
395 		efi_systab.nr_tables = systab32->nr_tables;
396 		efi_systab.tables = systab32->tables;
397 
398 		early_memunmap(systab32, sizeof(*systab32));
399 	}
400 
401 	efi.systab = &efi_systab;
402 
403 	/*
404 	 * Verify the EFI Table
405 	 */
406 	if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
407 		pr_err("System table signature incorrect!\n");
408 		return -EINVAL;
409 	}
410 	if ((efi.systab->hdr.revision >> 16) == 0)
411 		pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
412 		       efi.systab->hdr.revision >> 16,
413 		       efi.systab->hdr.revision & 0xffff);
414 
415 	return 0;
416 }
417 
418 static int __init efi_runtime_init32(void)
419 {
420 	efi_runtime_services_32_t *runtime;
421 
422 	runtime = early_memremap((unsigned long)efi.systab->runtime,
423 			sizeof(efi_runtime_services_32_t));
424 	if (!runtime) {
425 		pr_err("Could not map the runtime service table!\n");
426 		return -ENOMEM;
427 	}
428 
429 	/*
430 	 * We will only need *early* access to the SetVirtualAddressMap
431 	 * EFI runtime service. All other runtime services will be called
432 	 * via the virtual mapping.
433 	 */
434 	efi_phys.set_virtual_address_map =
435 			(efi_set_virtual_address_map_t *)
436 			(unsigned long)runtime->set_virtual_address_map;
437 	early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
438 
439 	return 0;
440 }
441 
442 static int __init efi_runtime_init64(void)
443 {
444 	efi_runtime_services_64_t *runtime;
445 
446 	runtime = early_memremap((unsigned long)efi.systab->runtime,
447 			sizeof(efi_runtime_services_64_t));
448 	if (!runtime) {
449 		pr_err("Could not map the runtime service table!\n");
450 		return -ENOMEM;
451 	}
452 
453 	/*
454 	 * We will only need *early* access to the SetVirtualAddressMap
455 	 * EFI runtime service. All other runtime services will be called
456 	 * via the virtual mapping.
457 	 */
458 	efi_phys.set_virtual_address_map =
459 			(efi_set_virtual_address_map_t *)
460 			(unsigned long)runtime->set_virtual_address_map;
461 	early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
462 
463 	return 0;
464 }
465 
466 static int __init efi_runtime_init(void)
467 {
468 	int rv;
469 
470 	/*
471 	 * Check out the runtime services table. We need to map
472 	 * the runtime services table so that we can grab the physical
473 	 * address of several of the EFI runtime functions, needed to
474 	 * set the firmware into virtual mode.
475 	 *
476 	 * When EFI_PARAVIRT is in force then we could not map runtime
477 	 * service memory region because we do not have direct access to it.
478 	 * However, runtime services are available through proxy functions
479 	 * (e.g. in case of Xen dom0 EFI implementation they call special
480 	 * hypercall which executes relevant EFI functions) and that is why
481 	 * they are always enabled.
482 	 */
483 
484 	if (!efi_enabled(EFI_PARAVIRT)) {
485 		if (efi_enabled(EFI_64BIT))
486 			rv = efi_runtime_init64();
487 		else
488 			rv = efi_runtime_init32();
489 
490 		if (rv)
491 			return rv;
492 	}
493 
494 	set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
495 
496 	return 0;
497 }
498 
499 void __init efi_init(void)
500 {
501 	efi_char16_t *c16;
502 	char vendor[100] = "unknown";
503 	int i = 0;
504 	void *tmp;
505 
506 #ifdef CONFIG_X86_32
507 	if (boot_params.efi_info.efi_systab_hi ||
508 	    boot_params.efi_info.efi_memmap_hi) {
509 		pr_info("Table located above 4GB, disabling EFI.\n");
510 		return;
511 	}
512 	efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
513 #else
514 	efi_phys.systab = (efi_system_table_t *)
515 			  (boot_params.efi_info.efi_systab |
516 			  ((__u64)boot_params.efi_info.efi_systab_hi<<32));
517 #endif
518 
519 	if (efi_systab_init(efi_phys.systab))
520 		return;
521 
522 	efi.config_table = (unsigned long)efi.systab->tables;
523 	efi.fw_vendor	 = (unsigned long)efi.systab->fw_vendor;
524 	efi.runtime	 = (unsigned long)efi.systab->runtime;
525 
526 	/*
527 	 * Show what we know for posterity
528 	 */
529 	c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
530 	if (c16) {
531 		for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
532 			vendor[i] = *c16++;
533 		vendor[i] = '\0';
534 	} else
535 		pr_err("Could not map the firmware vendor!\n");
536 	early_memunmap(tmp, 2);
537 
538 	pr_info("EFI v%u.%.02u by %s\n",
539 		efi.systab->hdr.revision >> 16,
540 		efi.systab->hdr.revision & 0xffff, vendor);
541 
542 	if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
543 		return;
544 
545 	if (efi_config_init(arch_tables))
546 		return;
547 
548 	/*
549 	 * Note: We currently don't support runtime services on an EFI
550 	 * that doesn't match the kernel 32/64-bit mode.
551 	 */
552 
553 	if (!efi_runtime_supported())
554 		pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
555 	else {
556 		if (efi_runtime_disabled() || efi_runtime_init()) {
557 			efi_memmap_unmap();
558 			return;
559 		}
560 	}
561 
562 	efi_clean_memmap();
563 
564 	if (efi_enabled(EFI_DBG))
565 		efi_print_memmap();
566 }
567 
568 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
569 {
570 	u64 addr, npages;
571 
572 	addr = md->virt_addr;
573 	npages = md->num_pages;
574 
575 	memrange_efi_to_native(&addr, &npages);
576 
577 	if (executable)
578 		set_memory_x(addr, npages);
579 	else
580 		set_memory_nx(addr, npages);
581 }
582 
583 void __init runtime_code_page_mkexec(void)
584 {
585 	efi_memory_desc_t *md;
586 
587 	/* Make EFI runtime service code area executable */
588 	for_each_efi_memory_desc(md) {
589 		if (md->type != EFI_RUNTIME_SERVICES_CODE)
590 			continue;
591 
592 		efi_set_executable(md, true);
593 	}
594 }
595 
596 void __init efi_memory_uc(u64 addr, unsigned long size)
597 {
598 	unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
599 	u64 npages;
600 
601 	npages = round_up(size, page_shift) / page_shift;
602 	memrange_efi_to_native(&addr, &npages);
603 	set_memory_uc(addr, npages);
604 }
605 
606 void __init old_map_region(efi_memory_desc_t *md)
607 {
608 	u64 start_pfn, end_pfn, end;
609 	unsigned long size;
610 	void *va;
611 
612 	start_pfn = PFN_DOWN(md->phys_addr);
613 	size	  = md->num_pages << PAGE_SHIFT;
614 	end	  = md->phys_addr + size;
615 	end_pfn   = PFN_UP(end);
616 
617 	if (pfn_range_is_mapped(start_pfn, end_pfn)) {
618 		va = __va(md->phys_addr);
619 
620 		if (!(md->attribute & EFI_MEMORY_WB))
621 			efi_memory_uc((u64)(unsigned long)va, size);
622 	} else
623 		va = efi_ioremap(md->phys_addr, size,
624 				 md->type, md->attribute);
625 
626 	md->virt_addr = (u64) (unsigned long) va;
627 	if (!va)
628 		pr_err("ioremap of 0x%llX failed!\n",
629 		       (unsigned long long)md->phys_addr);
630 }
631 
632 /* Merge contiguous regions of the same type and attribute */
633 static void __init efi_merge_regions(void)
634 {
635 	efi_memory_desc_t *md, *prev_md = NULL;
636 
637 	for_each_efi_memory_desc(md) {
638 		u64 prev_size;
639 
640 		if (!prev_md) {
641 			prev_md = md;
642 			continue;
643 		}
644 
645 		if (prev_md->type != md->type ||
646 		    prev_md->attribute != md->attribute) {
647 			prev_md = md;
648 			continue;
649 		}
650 
651 		prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
652 
653 		if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
654 			prev_md->num_pages += md->num_pages;
655 			md->type = EFI_RESERVED_TYPE;
656 			md->attribute = 0;
657 			continue;
658 		}
659 		prev_md = md;
660 	}
661 }
662 
663 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
664 {
665 	unsigned long size;
666 	u64 end, systab;
667 
668 	size = md->num_pages << EFI_PAGE_SHIFT;
669 	end = md->phys_addr + size;
670 	systab = (u64)(unsigned long)efi_phys.systab;
671 	if (md->phys_addr <= systab && systab < end) {
672 		systab += md->virt_addr - md->phys_addr;
673 		efi.systab = (efi_system_table_t *)(unsigned long)systab;
674 	}
675 }
676 
677 static void *realloc_pages(void *old_memmap, int old_shift)
678 {
679 	void *ret;
680 
681 	ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
682 	if (!ret)
683 		goto out;
684 
685 	/*
686 	 * A first-time allocation doesn't have anything to copy.
687 	 */
688 	if (!old_memmap)
689 		return ret;
690 
691 	memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
692 
693 out:
694 	free_pages((unsigned long)old_memmap, old_shift);
695 	return ret;
696 }
697 
698 /*
699  * Iterate the EFI memory map in reverse order because the regions
700  * will be mapped top-down. The end result is the same as if we had
701  * mapped things forward, but doesn't require us to change the
702  * existing implementation of efi_map_region().
703  */
704 static inline void *efi_map_next_entry_reverse(void *entry)
705 {
706 	/* Initial call */
707 	if (!entry)
708 		return efi.memmap.map_end - efi.memmap.desc_size;
709 
710 	entry -= efi.memmap.desc_size;
711 	if (entry < efi.memmap.map)
712 		return NULL;
713 
714 	return entry;
715 }
716 
717 /*
718  * efi_map_next_entry - Return the next EFI memory map descriptor
719  * @entry: Previous EFI memory map descriptor
720  *
721  * This is a helper function to iterate over the EFI memory map, which
722  * we do in different orders depending on the current configuration.
723  *
724  * To begin traversing the memory map @entry must be %NULL.
725  *
726  * Returns %NULL when we reach the end of the memory map.
727  */
728 static void *efi_map_next_entry(void *entry)
729 {
730 	if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
731 		/*
732 		 * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
733 		 * config table feature requires us to map all entries
734 		 * in the same order as they appear in the EFI memory
735 		 * map. That is to say, entry N must have a lower
736 		 * virtual address than entry N+1. This is because the
737 		 * firmware toolchain leaves relative references in
738 		 * the code/data sections, which are split and become
739 		 * separate EFI memory regions. Mapping things
740 		 * out-of-order leads to the firmware accessing
741 		 * unmapped addresses.
742 		 *
743 		 * Since we need to map things this way whether or not
744 		 * the kernel actually makes use of
745 		 * EFI_PROPERTIES_TABLE, let's just switch to this
746 		 * scheme by default for 64-bit.
747 		 */
748 		return efi_map_next_entry_reverse(entry);
749 	}
750 
751 	/* Initial call */
752 	if (!entry)
753 		return efi.memmap.map;
754 
755 	entry += efi.memmap.desc_size;
756 	if (entry >= efi.memmap.map_end)
757 		return NULL;
758 
759 	return entry;
760 }
761 
762 static bool should_map_region(efi_memory_desc_t *md)
763 {
764 	/*
765 	 * Runtime regions always require runtime mappings (obviously).
766 	 */
767 	if (md->attribute & EFI_MEMORY_RUNTIME)
768 		return true;
769 
770 	/*
771 	 * 32-bit EFI doesn't suffer from the bug that requires us to
772 	 * reserve boot services regions, and mixed mode support
773 	 * doesn't exist for 32-bit kernels.
774 	 */
775 	if (IS_ENABLED(CONFIG_X86_32))
776 		return false;
777 
778 	/*
779 	 * Map all of RAM so that we can access arguments in the 1:1
780 	 * mapping when making EFI runtime calls.
781 	 */
782 	if (IS_ENABLED(CONFIG_EFI_MIXED) && !efi_is_native()) {
783 		if (md->type == EFI_CONVENTIONAL_MEMORY ||
784 		    md->type == EFI_LOADER_DATA ||
785 		    md->type == EFI_LOADER_CODE)
786 			return true;
787 	}
788 
789 	/*
790 	 * Map boot services regions as a workaround for buggy
791 	 * firmware that accesses them even when they shouldn't.
792 	 *
793 	 * See efi_{reserve,free}_boot_services().
794 	 */
795 	if (md->type == EFI_BOOT_SERVICES_CODE ||
796 	    md->type == EFI_BOOT_SERVICES_DATA)
797 		return true;
798 
799 	return false;
800 }
801 
802 /*
803  * Map the efi memory ranges of the runtime services and update new_mmap with
804  * virtual addresses.
805  */
806 static void * __init efi_map_regions(int *count, int *pg_shift)
807 {
808 	void *p, *new_memmap = NULL;
809 	unsigned long left = 0;
810 	unsigned long desc_size;
811 	efi_memory_desc_t *md;
812 
813 	desc_size = efi.memmap.desc_size;
814 
815 	p = NULL;
816 	while ((p = efi_map_next_entry(p))) {
817 		md = p;
818 
819 		if (!should_map_region(md))
820 			continue;
821 
822 		efi_map_region(md);
823 		get_systab_virt_addr(md);
824 
825 		if (left < desc_size) {
826 			new_memmap = realloc_pages(new_memmap, *pg_shift);
827 			if (!new_memmap)
828 				return NULL;
829 
830 			left += PAGE_SIZE << *pg_shift;
831 			(*pg_shift)++;
832 		}
833 
834 		memcpy(new_memmap + (*count * desc_size), md, desc_size);
835 
836 		left -= desc_size;
837 		(*count)++;
838 	}
839 
840 	return new_memmap;
841 }
842 
843 static void __init kexec_enter_virtual_mode(void)
844 {
845 #ifdef CONFIG_KEXEC_CORE
846 	efi_memory_desc_t *md;
847 	unsigned int num_pages;
848 
849 	efi.systab = NULL;
850 
851 	/*
852 	 * We don't do virtual mode, since we don't do runtime services, on
853 	 * non-native EFI. With efi=old_map, we don't do runtime services in
854 	 * kexec kernel because in the initial boot something else might
855 	 * have been mapped at these virtual addresses.
856 	 */
857 	if (!efi_is_native() || efi_enabled(EFI_OLD_MEMMAP)) {
858 		efi_memmap_unmap();
859 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
860 		return;
861 	}
862 
863 	if (efi_alloc_page_tables()) {
864 		pr_err("Failed to allocate EFI page tables\n");
865 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
866 		return;
867 	}
868 
869 	/*
870 	* Map efi regions which were passed via setup_data. The virt_addr is a
871 	* fixed addr which was used in first kernel of a kexec boot.
872 	*/
873 	for_each_efi_memory_desc(md) {
874 		efi_map_region_fixed(md); /* FIXME: add error handling */
875 		get_systab_virt_addr(md);
876 	}
877 
878 	/*
879 	 * Unregister the early EFI memmap from efi_init() and install
880 	 * the new EFI memory map.
881 	 */
882 	efi_memmap_unmap();
883 
884 	if (efi_memmap_init_late(efi.memmap.phys_map,
885 				 efi.memmap.desc_size * efi.memmap.nr_map)) {
886 		pr_err("Failed to remap late EFI memory map\n");
887 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
888 		return;
889 	}
890 
891 	BUG_ON(!efi.systab);
892 
893 	num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
894 	num_pages >>= PAGE_SHIFT;
895 
896 	if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
897 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
898 		return;
899 	}
900 
901 	efi_sync_low_kernel_mappings();
902 
903 	/*
904 	 * Now that EFI is in virtual mode, update the function
905 	 * pointers in the runtime service table to the new virtual addresses.
906 	 *
907 	 * Call EFI services through wrapper functions.
908 	 */
909 	efi.runtime_version = efi_systab.hdr.revision;
910 
911 	efi_native_runtime_setup();
912 
913 	efi.set_virtual_address_map = NULL;
914 
915 	if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
916 		runtime_code_page_mkexec();
917 
918 	/* clean DUMMY object */
919 	efi_delete_dummy_variable();
920 #endif
921 }
922 
923 /*
924  * This function will switch the EFI runtime services to virtual mode.
925  * Essentially, we look through the EFI memmap and map every region that
926  * has the runtime attribute bit set in its memory descriptor into the
927  * efi_pgd page table.
928  *
929  * The old method which used to update that memory descriptor with the
930  * virtual address obtained from ioremap() is still supported when the
931  * kernel is booted with efi=old_map on its command line. Same old
932  * method enabled the runtime services to be called without having to
933  * thunk back into physical mode for every invocation.
934  *
935  * The new method does a pagetable switch in a preemption-safe manner
936  * so that we're in a different address space when calling a runtime
937  * function. For function arguments passing we do copy the PUDs of the
938  * kernel page table into efi_pgd prior to each call.
939  *
940  * Specially for kexec boot, efi runtime maps in previous kernel should
941  * be passed in via setup_data. In that case runtime ranges will be mapped
942  * to the same virtual addresses as the first kernel, see
943  * kexec_enter_virtual_mode().
944  */
945 static void __init __efi_enter_virtual_mode(void)
946 {
947 	int count = 0, pg_shift = 0;
948 	void *new_memmap = NULL;
949 	efi_status_t status;
950 	unsigned long pa;
951 
952 	efi.systab = NULL;
953 
954 	if (efi_alloc_page_tables()) {
955 		pr_err("Failed to allocate EFI page tables\n");
956 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
957 		return;
958 	}
959 
960 	efi_merge_regions();
961 	new_memmap = efi_map_regions(&count, &pg_shift);
962 	if (!new_memmap) {
963 		pr_err("Error reallocating memory, EFI runtime non-functional!\n");
964 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
965 		return;
966 	}
967 
968 	pa = __pa(new_memmap);
969 
970 	/*
971 	 * Unregister the early EFI memmap from efi_init() and install
972 	 * the new EFI memory map that we are about to pass to the
973 	 * firmware via SetVirtualAddressMap().
974 	 */
975 	efi_memmap_unmap();
976 
977 	if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
978 		pr_err("Failed to remap late EFI memory map\n");
979 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
980 		return;
981 	}
982 
983 	if (efi_enabled(EFI_DBG)) {
984 		pr_info("EFI runtime memory map:\n");
985 		efi_print_memmap();
986 	}
987 
988 	BUG_ON(!efi.systab);
989 
990 	if (efi_setup_page_tables(pa, 1 << pg_shift)) {
991 		clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
992 		return;
993 	}
994 
995 	efi_sync_low_kernel_mappings();
996 
997 	if (efi_is_native()) {
998 		status = phys_efi_set_virtual_address_map(
999 				efi.memmap.desc_size * count,
1000 				efi.memmap.desc_size,
1001 				efi.memmap.desc_version,
1002 				(efi_memory_desc_t *)pa);
1003 	} else {
1004 		status = efi_thunk_set_virtual_address_map(
1005 				efi_phys.set_virtual_address_map,
1006 				efi.memmap.desc_size * count,
1007 				efi.memmap.desc_size,
1008 				efi.memmap.desc_version,
1009 				(efi_memory_desc_t *)pa);
1010 	}
1011 
1012 	if (status != EFI_SUCCESS) {
1013 		pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
1014 			 status);
1015 		panic("EFI call to SetVirtualAddressMap() failed!");
1016 	}
1017 
1018 	efi_free_boot_services();
1019 
1020 	/*
1021 	 * Now that EFI is in virtual mode, update the function
1022 	 * pointers in the runtime service table to the new virtual addresses.
1023 	 *
1024 	 * Call EFI services through wrapper functions.
1025 	 */
1026 	efi.runtime_version = efi_systab.hdr.revision;
1027 
1028 	if (efi_is_native())
1029 		efi_native_runtime_setup();
1030 	else
1031 		efi_thunk_runtime_setup();
1032 
1033 	efi.set_virtual_address_map = NULL;
1034 
1035 	/*
1036 	 * Apply more restrictive page table mapping attributes now that
1037 	 * SVAM() has been called and the firmware has performed all
1038 	 * necessary relocation fixups for the new virtual addresses.
1039 	 */
1040 	efi_runtime_update_mappings();
1041 
1042 	/* clean DUMMY object */
1043 	efi_delete_dummy_variable();
1044 }
1045 
1046 void __init efi_enter_virtual_mode(void)
1047 {
1048 	if (efi_enabled(EFI_PARAVIRT))
1049 		return;
1050 
1051 	if (efi_setup)
1052 		kexec_enter_virtual_mode();
1053 	else
1054 		__efi_enter_virtual_mode();
1055 
1056 	efi_dump_pagetable();
1057 }
1058 
1059 static int __init arch_parse_efi_cmdline(char *str)
1060 {
1061 	if (!str) {
1062 		pr_warn("need at least one option\n");
1063 		return -EINVAL;
1064 	}
1065 
1066 	if (parse_option_str(str, "old_map"))
1067 		set_bit(EFI_OLD_MEMMAP, &efi.flags);
1068 
1069 	return 0;
1070 }
1071 early_param("efi", arch_parse_efi_cmdline);
1072 
1073 bool efi_is_table_address(unsigned long phys_addr)
1074 {
1075 	unsigned int i;
1076 
1077 	if (phys_addr == EFI_INVALID_TABLE_ADDR)
1078 		return false;
1079 
1080 	for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
1081 		if (*(efi_tables[i]) == phys_addr)
1082 			return true;
1083 
1084 	return false;
1085 }
1086