xref: /linux/kernel/crash_core.c (revision 778e73d2411abc8f3a2d60dbf038acaec218792e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * crash.c - kernel crash support code.
4  * Copyright (C) 2002-2004 Eric Biederman  <ebiederm@xmission.com>
5  */
6 
7 #include <linux/buildid.h>
8 #include <linux/init.h>
9 #include <linux/utsname.h>
10 #include <linux/vmalloc.h>
11 #include <linux/sizes.h>
12 #include <linux/kexec.h>
13 #include <linux/memory.h>
14 #include <linux/cpuhotplug.h>
15 #include <linux/memblock.h>
16 #include <linux/kmemleak.h>
17 
18 #include <asm/page.h>
19 #include <asm/sections.h>
20 
21 #include <crypto/sha1.h>
22 
23 #include "kallsyms_internal.h"
24 #include "kexec_internal.h"
25 
26 /* Per cpu memory for storing cpu states in case of system crash. */
27 note_buf_t __percpu *crash_notes;
28 
29 /* vmcoreinfo stuff */
30 unsigned char *vmcoreinfo_data;
31 size_t vmcoreinfo_size;
32 u32 *vmcoreinfo_note;
33 
34 /* trusted vmcoreinfo, e.g. we can make a copy in the crash memory */
35 static unsigned char *vmcoreinfo_data_safecopy;
36 
37 /* Location of the reserved area for the crash kernel */
38 struct resource crashk_res = {
39 	.name  = "Crash kernel",
40 	.start = 0,
41 	.end   = 0,
42 	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
43 	.desc  = IORES_DESC_CRASH_KERNEL
44 };
45 struct resource crashk_low_res = {
46 	.name  = "Crash kernel",
47 	.start = 0,
48 	.end   = 0,
49 	.flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
50 	.desc  = IORES_DESC_CRASH_KERNEL
51 };
52 
53 /*
54  * parsing the "crashkernel" commandline
55  *
56  * this code is intended to be called from architecture specific code
57  */
58 
59 
60 /*
61  * This function parses command lines in the format
62  *
63  *   crashkernel=ramsize-range:size[,...][@offset]
64  *
65  * The function returns 0 on success and -EINVAL on failure.
66  */
67 static int __init parse_crashkernel_mem(char *cmdline,
68 					unsigned long long system_ram,
69 					unsigned long long *crash_size,
70 					unsigned long long *crash_base)
71 {
72 	char *cur = cmdline, *tmp;
73 	unsigned long long total_mem = system_ram;
74 
75 	/*
76 	 * Firmware sometimes reserves some memory regions for its own use,
77 	 * so the system memory size is less than the actual physical memory
78 	 * size. Work around this by rounding up the total size to 128M,
79 	 * which is enough for most test cases.
80 	 */
81 	total_mem = roundup(total_mem, SZ_128M);
82 
83 	/* for each entry of the comma-separated list */
84 	do {
85 		unsigned long long start, end = ULLONG_MAX, size;
86 
87 		/* get the start of the range */
88 		start = memparse(cur, &tmp);
89 		if (cur == tmp) {
90 			pr_warn("crashkernel: Memory value expected\n");
91 			return -EINVAL;
92 		}
93 		cur = tmp;
94 		if (*cur != '-') {
95 			pr_warn("crashkernel: '-' expected\n");
96 			return -EINVAL;
97 		}
98 		cur++;
99 
100 		/* if no ':' is here, than we read the end */
101 		if (*cur != ':') {
102 			end = memparse(cur, &tmp);
103 			if (cur == tmp) {
104 				pr_warn("crashkernel: Memory value expected\n");
105 				return -EINVAL;
106 			}
107 			cur = tmp;
108 			if (end <= start) {
109 				pr_warn("crashkernel: end <= start\n");
110 				return -EINVAL;
111 			}
112 		}
113 
114 		if (*cur != ':') {
115 			pr_warn("crashkernel: ':' expected\n");
116 			return -EINVAL;
117 		}
118 		cur++;
119 
120 		size = memparse(cur, &tmp);
121 		if (cur == tmp) {
122 			pr_warn("Memory value expected\n");
123 			return -EINVAL;
124 		}
125 		cur = tmp;
126 		if (size >= total_mem) {
127 			pr_warn("crashkernel: invalid size\n");
128 			return -EINVAL;
129 		}
130 
131 		/* match ? */
132 		if (total_mem >= start && total_mem < end) {
133 			*crash_size = size;
134 			break;
135 		}
136 	} while (*cur++ == ',');
137 
138 	if (*crash_size > 0) {
139 		while (*cur && *cur != ' ' && *cur != '@')
140 			cur++;
141 		if (*cur == '@') {
142 			cur++;
143 			*crash_base = memparse(cur, &tmp);
144 			if (cur == tmp) {
145 				pr_warn("Memory value expected after '@'\n");
146 				return -EINVAL;
147 			}
148 		}
149 	} else
150 		pr_info("crashkernel size resulted in zero bytes\n");
151 
152 	return 0;
153 }
154 
155 /*
156  * That function parses "simple" (old) crashkernel command lines like
157  *
158  *	crashkernel=size[@offset]
159  *
160  * It returns 0 on success and -EINVAL on failure.
161  */
162 static int __init parse_crashkernel_simple(char *cmdline,
163 					   unsigned long long *crash_size,
164 					   unsigned long long *crash_base)
165 {
166 	char *cur = cmdline;
167 
168 	*crash_size = memparse(cmdline, &cur);
169 	if (cmdline == cur) {
170 		pr_warn("crashkernel: memory value expected\n");
171 		return -EINVAL;
172 	}
173 
174 	if (*cur == '@')
175 		*crash_base = memparse(cur+1, &cur);
176 	else if (*cur != ' ' && *cur != '\0') {
177 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
178 		return -EINVAL;
179 	}
180 
181 	return 0;
182 }
183 
184 #define SUFFIX_HIGH 0
185 #define SUFFIX_LOW  1
186 #define SUFFIX_NULL 2
187 static __initdata char *suffix_tbl[] = {
188 	[SUFFIX_HIGH] = ",high",
189 	[SUFFIX_LOW]  = ",low",
190 	[SUFFIX_NULL] = NULL,
191 };
192 
193 /*
194  * That function parses "suffix"  crashkernel command lines like
195  *
196  *	crashkernel=size,[high|low]
197  *
198  * It returns 0 on success and -EINVAL on failure.
199  */
200 static int __init parse_crashkernel_suffix(char *cmdline,
201 					   unsigned long long *crash_size,
202 					   const char *suffix)
203 {
204 	char *cur = cmdline;
205 
206 	*crash_size = memparse(cmdline, &cur);
207 	if (cmdline == cur) {
208 		pr_warn("crashkernel: memory value expected\n");
209 		return -EINVAL;
210 	}
211 
212 	/* check with suffix */
213 	if (strncmp(cur, suffix, strlen(suffix))) {
214 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
215 		return -EINVAL;
216 	}
217 	cur += strlen(suffix);
218 	if (*cur != ' ' && *cur != '\0') {
219 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
220 		return -EINVAL;
221 	}
222 
223 	return 0;
224 }
225 
226 static __init char *get_last_crashkernel(char *cmdline,
227 			     const char *name,
228 			     const char *suffix)
229 {
230 	char *p = cmdline, *ck_cmdline = NULL;
231 
232 	/* find crashkernel and use the last one if there are more */
233 	p = strstr(p, name);
234 	while (p) {
235 		char *end_p = strchr(p, ' ');
236 		char *q;
237 
238 		if (!end_p)
239 			end_p = p + strlen(p);
240 
241 		if (!suffix) {
242 			int i;
243 
244 			/* skip the one with any known suffix */
245 			for (i = 0; suffix_tbl[i]; i++) {
246 				q = end_p - strlen(suffix_tbl[i]);
247 				if (!strncmp(q, suffix_tbl[i],
248 					     strlen(suffix_tbl[i])))
249 					goto next;
250 			}
251 			ck_cmdline = p;
252 		} else {
253 			q = end_p - strlen(suffix);
254 			if (!strncmp(q, suffix, strlen(suffix)))
255 				ck_cmdline = p;
256 		}
257 next:
258 		p = strstr(p+1, name);
259 	}
260 
261 	return ck_cmdline;
262 }
263 
264 static int __init __parse_crashkernel(char *cmdline,
265 			     unsigned long long system_ram,
266 			     unsigned long long *crash_size,
267 			     unsigned long long *crash_base,
268 			     const char *suffix)
269 {
270 	char *first_colon, *first_space;
271 	char *ck_cmdline;
272 	char *name = "crashkernel=";
273 
274 	BUG_ON(!crash_size || !crash_base);
275 	*crash_size = 0;
276 	*crash_base = 0;
277 
278 	ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
279 	if (!ck_cmdline)
280 		return -ENOENT;
281 
282 	ck_cmdline += strlen(name);
283 
284 	if (suffix)
285 		return parse_crashkernel_suffix(ck_cmdline, crash_size,
286 				suffix);
287 	/*
288 	 * if the commandline contains a ':', then that's the extended
289 	 * syntax -- if not, it must be the classic syntax
290 	 */
291 	first_colon = strchr(ck_cmdline, ':');
292 	first_space = strchr(ck_cmdline, ' ');
293 	if (first_colon && (!first_space || first_colon < first_space))
294 		return parse_crashkernel_mem(ck_cmdline, system_ram,
295 				crash_size, crash_base);
296 
297 	return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
298 }
299 
300 /*
301  * That function is the entry point for command line parsing and should be
302  * called from the arch-specific code.
303  *
304  * If crashkernel=,high|low is supported on architecture, non-NULL values
305  * should be passed to parameters 'low_size' and 'high'.
306  */
307 int __init parse_crashkernel(char *cmdline,
308 			     unsigned long long system_ram,
309 			     unsigned long long *crash_size,
310 			     unsigned long long *crash_base,
311 			     unsigned long long *low_size,
312 			     bool *high)
313 {
314 	int ret;
315 
316 	/* crashkernel=X[@offset] */
317 	ret = __parse_crashkernel(cmdline, system_ram, crash_size,
318 				crash_base, NULL);
319 #ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
320 	/*
321 	 * If non-NULL 'high' passed in and no normal crashkernel
322 	 * setting detected, try parsing crashkernel=,high|low.
323 	 */
324 	if (high && ret == -ENOENT) {
325 		ret = __parse_crashkernel(cmdline, 0, crash_size,
326 				crash_base, suffix_tbl[SUFFIX_HIGH]);
327 		if (ret || !*crash_size)
328 			return -EINVAL;
329 
330 		/*
331 		 * crashkernel=Y,low can be specified or not, but invalid value
332 		 * is not allowed.
333 		 */
334 		ret = __parse_crashkernel(cmdline, 0, low_size,
335 				crash_base, suffix_tbl[SUFFIX_LOW]);
336 		if (ret == -ENOENT) {
337 			*low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
338 			ret = 0;
339 		} else if (ret) {
340 			return ret;
341 		}
342 
343 		*high = true;
344 	}
345 #endif
346 	if (!*crash_size)
347 		ret = -EINVAL;
348 
349 	return ret;
350 }
351 
352 /*
353  * Add a dummy early_param handler to mark crashkernel= as a known command line
354  * parameter and suppress incorrect warnings in init/main.c.
355  */
356 static int __init parse_crashkernel_dummy(char *arg)
357 {
358 	return 0;
359 }
360 early_param("crashkernel", parse_crashkernel_dummy);
361 
362 #ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
363 static int __init reserve_crashkernel_low(unsigned long long low_size)
364 {
365 #ifdef CONFIG_64BIT
366 	unsigned long long low_base;
367 
368 	low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
369 	if (!low_base) {
370 		pr_err("cannot allocate crashkernel low memory (size:0x%llx).\n", low_size);
371 		return -ENOMEM;
372 	}
373 
374 	pr_info("crashkernel low memory reserved: 0x%08llx - 0x%08llx (%lld MB)\n",
375 		low_base, low_base + low_size, low_size >> 20);
376 
377 	crashk_low_res.start = low_base;
378 	crashk_low_res.end   = low_base + low_size - 1;
379 	insert_resource(&iomem_resource, &crashk_low_res);
380 #endif
381 	return 0;
382 }
383 
384 void __init reserve_crashkernel_generic(char *cmdline,
385 			     unsigned long long crash_size,
386 			     unsigned long long crash_base,
387 			     unsigned long long crash_low_size,
388 			     bool high)
389 {
390 	unsigned long long search_end = CRASH_ADDR_LOW_MAX, search_base = 0;
391 	bool fixed_base = false;
392 
393 	/* User specifies base address explicitly. */
394 	if (crash_base) {
395 		fixed_base = true;
396 		search_base = crash_base;
397 		search_end = crash_base + crash_size;
398 	} else if (high) {
399 		search_base = CRASH_ADDR_LOW_MAX;
400 		search_end = CRASH_ADDR_HIGH_MAX;
401 	}
402 
403 retry:
404 	crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN,
405 					       search_base, search_end);
406 	if (!crash_base) {
407 		/*
408 		 * For crashkernel=size[KMG]@offset[KMG], print out failure
409 		 * message if can't reserve the specified region.
410 		 */
411 		if (fixed_base) {
412 			pr_warn("crashkernel reservation failed - memory is in use.\n");
413 			return;
414 		}
415 
416 		/*
417 		 * For crashkernel=size[KMG], if the first attempt was for
418 		 * low memory, fall back to high memory, the minimum required
419 		 * low memory will be reserved later.
420 		 */
421 		if (!high && search_end == CRASH_ADDR_LOW_MAX) {
422 			search_end = CRASH_ADDR_HIGH_MAX;
423 			search_base = CRASH_ADDR_LOW_MAX;
424 			crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
425 			goto retry;
426 		}
427 
428 		/*
429 		 * For crashkernel=size[KMG],high, if the first attempt was
430 		 * for high memory, fall back to low memory.
431 		 */
432 		if (high && search_end == CRASH_ADDR_HIGH_MAX) {
433 			search_end = CRASH_ADDR_LOW_MAX;
434 			search_base = 0;
435 			goto retry;
436 		}
437 		pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
438 			crash_size);
439 		return;
440 	}
441 
442 	if ((crash_base >= CRASH_ADDR_LOW_MAX) &&
443 	     crash_low_size && reserve_crashkernel_low(crash_low_size)) {
444 		memblock_phys_free(crash_base, crash_size);
445 		return;
446 	}
447 
448 	pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
449 		crash_base, crash_base + crash_size, crash_size >> 20);
450 
451 	/*
452 	 * The crashkernel memory will be removed from the kernel linear
453 	 * map. Inform kmemleak so that it won't try to access it.
454 	 */
455 	kmemleak_ignore_phys(crash_base);
456 	if (crashk_low_res.end)
457 		kmemleak_ignore_phys(crashk_low_res.start);
458 
459 	crashk_res.start = crash_base;
460 	crashk_res.end = crash_base + crash_size - 1;
461 	insert_resource(&iomem_resource, &crashk_res);
462 }
463 #endif
464 
465 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
466 			  void **addr, unsigned long *sz)
467 {
468 	Elf64_Ehdr *ehdr;
469 	Elf64_Phdr *phdr;
470 	unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
471 	unsigned char *buf;
472 	unsigned int cpu, i;
473 	unsigned long long notes_addr;
474 	unsigned long mstart, mend;
475 
476 	/* extra phdr for vmcoreinfo ELF note */
477 	nr_phdr = nr_cpus + 1;
478 	nr_phdr += mem->nr_ranges;
479 
480 	/*
481 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
482 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
483 	 * I think this is required by tools like gdb. So same physical
484 	 * memory will be mapped in two ELF headers. One will contain kernel
485 	 * text virtual addresses and other will have __va(physical) addresses.
486 	 */
487 
488 	nr_phdr++;
489 	elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
490 	elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
491 
492 	buf = vzalloc(elf_sz);
493 	if (!buf)
494 		return -ENOMEM;
495 
496 	ehdr = (Elf64_Ehdr *)buf;
497 	phdr = (Elf64_Phdr *)(ehdr + 1);
498 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
499 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
500 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
501 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
502 	ehdr->e_ident[EI_OSABI] = ELF_OSABI;
503 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
504 	ehdr->e_type = ET_CORE;
505 	ehdr->e_machine = ELF_ARCH;
506 	ehdr->e_version = EV_CURRENT;
507 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
508 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
509 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
510 
511 	/* Prepare one phdr of type PT_NOTE for each possible CPU */
512 	for_each_possible_cpu(cpu) {
513 		phdr->p_type = PT_NOTE;
514 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
515 		phdr->p_offset = phdr->p_paddr = notes_addr;
516 		phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
517 		(ehdr->e_phnum)++;
518 		phdr++;
519 	}
520 
521 	/* Prepare one PT_NOTE header for vmcoreinfo */
522 	phdr->p_type = PT_NOTE;
523 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
524 	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
525 	(ehdr->e_phnum)++;
526 	phdr++;
527 
528 	/* Prepare PT_LOAD type program header for kernel text region */
529 	if (need_kernel_map) {
530 		phdr->p_type = PT_LOAD;
531 		phdr->p_flags = PF_R|PF_W|PF_X;
532 		phdr->p_vaddr = (unsigned long) _text;
533 		phdr->p_filesz = phdr->p_memsz = _end - _text;
534 		phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
535 		ehdr->e_phnum++;
536 		phdr++;
537 	}
538 
539 	/* Go through all the ranges in mem->ranges[] and prepare phdr */
540 	for (i = 0; i < mem->nr_ranges; i++) {
541 		mstart = mem->ranges[i].start;
542 		mend = mem->ranges[i].end;
543 
544 		phdr->p_type = PT_LOAD;
545 		phdr->p_flags = PF_R|PF_W|PF_X;
546 		phdr->p_offset  = mstart;
547 
548 		phdr->p_paddr = mstart;
549 		phdr->p_vaddr = (unsigned long) __va(mstart);
550 		phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
551 		phdr->p_align = 0;
552 		ehdr->e_phnum++;
553 #ifdef CONFIG_KEXEC_FILE
554 		kexec_dprintk("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
555 			      phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
556 			      ehdr->e_phnum, phdr->p_offset);
557 #endif
558 		phdr++;
559 	}
560 
561 	*addr = buf;
562 	*sz = elf_sz;
563 	return 0;
564 }
565 
566 int crash_exclude_mem_range(struct crash_mem *mem,
567 			    unsigned long long mstart, unsigned long long mend)
568 {
569 	int i;
570 	unsigned long long start, end, p_start, p_end;
571 
572 	for (i = 0; i < mem->nr_ranges; i++) {
573 		start = mem->ranges[i].start;
574 		end = mem->ranges[i].end;
575 		p_start = mstart;
576 		p_end = mend;
577 
578 		if (p_start > end)
579 			continue;
580 
581 		/*
582 		 * Because the memory ranges in mem->ranges are stored in
583 		 * ascending order, when we detect `p_end < start`, we can
584 		 * immediately exit the for loop, as the subsequent memory
585 		 * ranges will definitely be outside the range we are looking
586 		 * for.
587 		 */
588 		if (p_end < start)
589 			break;
590 
591 		/* Truncate any area outside of range */
592 		if (p_start < start)
593 			p_start = start;
594 		if (p_end > end)
595 			p_end = end;
596 
597 		/* Found completely overlapping range */
598 		if (p_start == start && p_end == end) {
599 			memmove(&mem->ranges[i], &mem->ranges[i + 1],
600 				(mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
601 			i--;
602 			mem->nr_ranges--;
603 		} else if (p_start > start && p_end < end) {
604 			/* Split original range */
605 			if (mem->nr_ranges >= mem->max_nr_ranges)
606 				return -ENOMEM;
607 
608 			memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
609 				(mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
610 
611 			mem->ranges[i].end = p_start - 1;
612 			mem->ranges[i + 1].start = p_end + 1;
613 			mem->ranges[i + 1].end = end;
614 
615 			i++;
616 			mem->nr_ranges++;
617 		} else if (p_start != start)
618 			mem->ranges[i].end = p_start - 1;
619 		else
620 			mem->ranges[i].start = p_end + 1;
621 	}
622 
623 	return 0;
624 }
625 
626 Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type,
627 			  void *data, size_t data_len)
628 {
629 	struct elf_note *note = (struct elf_note *)buf;
630 
631 	note->n_namesz = strlen(name) + 1;
632 	note->n_descsz = data_len;
633 	note->n_type   = type;
634 	buf += DIV_ROUND_UP(sizeof(*note), sizeof(Elf_Word));
635 	memcpy(buf, name, note->n_namesz);
636 	buf += DIV_ROUND_UP(note->n_namesz, sizeof(Elf_Word));
637 	memcpy(buf, data, data_len);
638 	buf += DIV_ROUND_UP(data_len, sizeof(Elf_Word));
639 
640 	return buf;
641 }
642 
643 void final_note(Elf_Word *buf)
644 {
645 	memset(buf, 0, sizeof(struct elf_note));
646 }
647 
648 static void update_vmcoreinfo_note(void)
649 {
650 	u32 *buf = vmcoreinfo_note;
651 
652 	if (!vmcoreinfo_size)
653 		return;
654 	buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
655 			      vmcoreinfo_size);
656 	final_note(buf);
657 }
658 
659 void crash_update_vmcoreinfo_safecopy(void *ptr)
660 {
661 	if (ptr)
662 		memcpy(ptr, vmcoreinfo_data, vmcoreinfo_size);
663 
664 	vmcoreinfo_data_safecopy = ptr;
665 }
666 
667 void crash_save_vmcoreinfo(void)
668 {
669 	if (!vmcoreinfo_note)
670 		return;
671 
672 	/* Use the safe copy to generate vmcoreinfo note if have */
673 	if (vmcoreinfo_data_safecopy)
674 		vmcoreinfo_data = vmcoreinfo_data_safecopy;
675 
676 	vmcoreinfo_append_str("CRASHTIME=%lld\n", ktime_get_real_seconds());
677 	update_vmcoreinfo_note();
678 }
679 
680 void vmcoreinfo_append_str(const char *fmt, ...)
681 {
682 	va_list args;
683 	char buf[0x50];
684 	size_t r;
685 
686 	va_start(args, fmt);
687 	r = vscnprintf(buf, sizeof(buf), fmt, args);
688 	va_end(args);
689 
690 	r = min(r, (size_t)VMCOREINFO_BYTES - vmcoreinfo_size);
691 
692 	memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
693 
694 	vmcoreinfo_size += r;
695 
696 	WARN_ONCE(vmcoreinfo_size == VMCOREINFO_BYTES,
697 		  "vmcoreinfo data exceeds allocated size, truncating");
698 }
699 
700 /*
701  * provide an empty default implementation here -- architecture
702  * code may override this
703  */
704 void __weak arch_crash_save_vmcoreinfo(void)
705 {}
706 
707 phys_addr_t __weak paddr_vmcoreinfo_note(void)
708 {
709 	return __pa(vmcoreinfo_note);
710 }
711 EXPORT_SYMBOL(paddr_vmcoreinfo_note);
712 
713 static int __init crash_save_vmcoreinfo_init(void)
714 {
715 	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
716 	if (!vmcoreinfo_data) {
717 		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
718 		return -ENOMEM;
719 	}
720 
721 	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
722 						GFP_KERNEL | __GFP_ZERO);
723 	if (!vmcoreinfo_note) {
724 		free_page((unsigned long)vmcoreinfo_data);
725 		vmcoreinfo_data = NULL;
726 		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
727 		return -ENOMEM;
728 	}
729 
730 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
731 	VMCOREINFO_BUILD_ID();
732 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
733 
734 	VMCOREINFO_SYMBOL(init_uts_ns);
735 	VMCOREINFO_OFFSET(uts_namespace, name);
736 	VMCOREINFO_SYMBOL(node_online_map);
737 #ifdef CONFIG_MMU
738 	VMCOREINFO_SYMBOL_ARRAY(swapper_pg_dir);
739 #endif
740 	VMCOREINFO_SYMBOL(_stext);
741 	VMCOREINFO_SYMBOL(vmap_area_list);
742 
743 #ifndef CONFIG_NUMA
744 	VMCOREINFO_SYMBOL(mem_map);
745 	VMCOREINFO_SYMBOL(contig_page_data);
746 #endif
747 #ifdef CONFIG_SPARSEMEM
748 	VMCOREINFO_SYMBOL_ARRAY(mem_section);
749 	VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
750 	VMCOREINFO_STRUCT_SIZE(mem_section);
751 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
752 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
753 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
754 #endif
755 	VMCOREINFO_STRUCT_SIZE(page);
756 	VMCOREINFO_STRUCT_SIZE(pglist_data);
757 	VMCOREINFO_STRUCT_SIZE(zone);
758 	VMCOREINFO_STRUCT_SIZE(free_area);
759 	VMCOREINFO_STRUCT_SIZE(list_head);
760 	VMCOREINFO_SIZE(nodemask_t);
761 	VMCOREINFO_OFFSET(page, flags);
762 	VMCOREINFO_OFFSET(page, _refcount);
763 	VMCOREINFO_OFFSET(page, mapping);
764 	VMCOREINFO_OFFSET(page, lru);
765 	VMCOREINFO_OFFSET(page, _mapcount);
766 	VMCOREINFO_OFFSET(page, private);
767 	VMCOREINFO_OFFSET(page, compound_head);
768 	VMCOREINFO_OFFSET(pglist_data, node_zones);
769 	VMCOREINFO_OFFSET(pglist_data, nr_zones);
770 #ifdef CONFIG_FLATMEM
771 	VMCOREINFO_OFFSET(pglist_data, node_mem_map);
772 #endif
773 	VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
774 	VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
775 	VMCOREINFO_OFFSET(pglist_data, node_id);
776 	VMCOREINFO_OFFSET(zone, free_area);
777 	VMCOREINFO_OFFSET(zone, vm_stat);
778 	VMCOREINFO_OFFSET(zone, spanned_pages);
779 	VMCOREINFO_OFFSET(free_area, free_list);
780 	VMCOREINFO_OFFSET(list_head, next);
781 	VMCOREINFO_OFFSET(list_head, prev);
782 	VMCOREINFO_OFFSET(vmap_area, va_start);
783 	VMCOREINFO_OFFSET(vmap_area, list);
784 	VMCOREINFO_LENGTH(zone.free_area, NR_PAGE_ORDERS);
785 	log_buf_vmcoreinfo_setup();
786 	VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
787 	VMCOREINFO_NUMBER(NR_FREE_PAGES);
788 	VMCOREINFO_NUMBER(PG_lru);
789 	VMCOREINFO_NUMBER(PG_private);
790 	VMCOREINFO_NUMBER(PG_swapcache);
791 	VMCOREINFO_NUMBER(PG_swapbacked);
792 	VMCOREINFO_NUMBER(PG_slab);
793 #ifdef CONFIG_MEMORY_FAILURE
794 	VMCOREINFO_NUMBER(PG_hwpoison);
795 #endif
796 	VMCOREINFO_NUMBER(PG_head_mask);
797 #define PAGE_BUDDY_MAPCOUNT_VALUE	(~PG_buddy)
798 	VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
799 #ifdef CONFIG_HUGETLB_PAGE
800 	VMCOREINFO_NUMBER(PG_hugetlb);
801 #define PAGE_OFFLINE_MAPCOUNT_VALUE	(~PG_offline)
802 	VMCOREINFO_NUMBER(PAGE_OFFLINE_MAPCOUNT_VALUE);
803 #endif
804 
805 #ifdef CONFIG_KALLSYMS
806 	VMCOREINFO_SYMBOL(kallsyms_names);
807 	VMCOREINFO_SYMBOL(kallsyms_num_syms);
808 	VMCOREINFO_SYMBOL(kallsyms_token_table);
809 	VMCOREINFO_SYMBOL(kallsyms_token_index);
810 #ifdef CONFIG_KALLSYMS_BASE_RELATIVE
811 	VMCOREINFO_SYMBOL(kallsyms_offsets);
812 	VMCOREINFO_SYMBOL(kallsyms_relative_base);
813 #else
814 	VMCOREINFO_SYMBOL(kallsyms_addresses);
815 #endif /* CONFIG_KALLSYMS_BASE_RELATIVE */
816 #endif /* CONFIG_KALLSYMS */
817 
818 	arch_crash_save_vmcoreinfo();
819 	update_vmcoreinfo_note();
820 
821 	return 0;
822 }
823 
824 subsys_initcall(crash_save_vmcoreinfo_init);
825 
826 static int __init crash_notes_memory_init(void)
827 {
828 	/* Allocate memory for saving cpu registers. */
829 	size_t size, align;
830 
831 	/*
832 	 * crash_notes could be allocated across 2 vmalloc pages when percpu
833 	 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
834 	 * pages are also on 2 continuous physical pages. In this case the
835 	 * 2nd part of crash_notes in 2nd page could be lost since only the
836 	 * starting address and size of crash_notes are exported through sysfs.
837 	 * Here round up the size of crash_notes to the nearest power of two
838 	 * and pass it to __alloc_percpu as align value. This can make sure
839 	 * crash_notes is allocated inside one physical page.
840 	 */
841 	size = sizeof(note_buf_t);
842 	align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
843 
844 	/*
845 	 * Break compile if size is bigger than PAGE_SIZE since crash_notes
846 	 * definitely will be in 2 pages with that.
847 	 */
848 	BUILD_BUG_ON(size > PAGE_SIZE);
849 
850 	crash_notes = __alloc_percpu(size, align);
851 	if (!crash_notes) {
852 		pr_warn("Memory allocation for saving cpu register states failed\n");
853 		return -ENOMEM;
854 	}
855 	return 0;
856 }
857 subsys_initcall(crash_notes_memory_init);
858 
859 #ifdef CONFIG_CRASH_HOTPLUG
860 #undef pr_fmt
861 #define pr_fmt(fmt) "crash hp: " fmt
862 
863 /*
864  * Different than kexec/kdump loading/unloading/jumping/shrinking which
865  * usually rarely happen, there will be many crash hotplug events notified
866  * during one short period, e.g one memory board is hot added and memory
867  * regions are online. So mutex lock  __crash_hotplug_lock is used to
868  * serialize the crash hotplug handling specifically.
869  */
870 DEFINE_MUTEX(__crash_hotplug_lock);
871 #define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
872 #define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
873 
874 /*
875  * This routine utilized when the crash_hotplug sysfs node is read.
876  * It reflects the kernel's ability/permission to update the crash
877  * elfcorehdr directly.
878  */
879 int crash_check_update_elfcorehdr(void)
880 {
881 	int rc = 0;
882 
883 	crash_hotplug_lock();
884 	/* Obtain lock while reading crash information */
885 	if (!kexec_trylock()) {
886 		pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
887 		crash_hotplug_unlock();
888 		return 0;
889 	}
890 	if (kexec_crash_image) {
891 		if (kexec_crash_image->file_mode)
892 			rc = 1;
893 		else
894 			rc = kexec_crash_image->update_elfcorehdr;
895 	}
896 	/* Release lock now that update complete */
897 	kexec_unlock();
898 	crash_hotplug_unlock();
899 
900 	return rc;
901 }
902 
903 /*
904  * To accurately reflect hot un/plug changes of cpu and memory resources
905  * (including onling and offlining of those resources), the elfcorehdr
906  * (which is passed to the crash kernel via the elfcorehdr= parameter)
907  * must be updated with the new list of CPUs and memories.
908  *
909  * In order to make changes to elfcorehdr, two conditions are needed:
910  * First, the segment containing the elfcorehdr must be large enough
911  * to permit a growing number of resources; the elfcorehdr memory size
912  * is based on NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES.
913  * Second, purgatory must explicitly exclude the elfcorehdr from the
914  * list of segments it checks (since the elfcorehdr changes and thus
915  * would require an update to purgatory itself to update the digest).
916  */
917 static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu)
918 {
919 	struct kimage *image;
920 
921 	crash_hotplug_lock();
922 	/* Obtain lock while changing crash information */
923 	if (!kexec_trylock()) {
924 		pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
925 		crash_hotplug_unlock();
926 		return;
927 	}
928 
929 	/* Check kdump is not loaded */
930 	if (!kexec_crash_image)
931 		goto out;
932 
933 	image = kexec_crash_image;
934 
935 	/* Check that updating elfcorehdr is permitted */
936 	if (!(image->file_mode || image->update_elfcorehdr))
937 		goto out;
938 
939 	if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
940 		hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
941 		pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
942 	else
943 		pr_debug("hp_action %u\n", hp_action);
944 
945 	/*
946 	 * The elfcorehdr_index is set to -1 when the struct kimage
947 	 * is allocated. Find the segment containing the elfcorehdr,
948 	 * if not already found.
949 	 */
950 	if (image->elfcorehdr_index < 0) {
951 		unsigned long mem;
952 		unsigned char *ptr;
953 		unsigned int n;
954 
955 		for (n = 0; n < image->nr_segments; n++) {
956 			mem = image->segment[n].mem;
957 			ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
958 			if (ptr) {
959 				/* The segment containing elfcorehdr */
960 				if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
961 					image->elfcorehdr_index = (int)n;
962 				kunmap_local(ptr);
963 			}
964 		}
965 	}
966 
967 	if (image->elfcorehdr_index < 0) {
968 		pr_err("unable to locate elfcorehdr segment");
969 		goto out;
970 	}
971 
972 	/* Needed in order for the segments to be updated */
973 	arch_kexec_unprotect_crashkres();
974 
975 	/* Differentiate between normal load and hotplug update */
976 	image->hp_action = hp_action;
977 
978 	/* Now invoke arch-specific update handler */
979 	arch_crash_handle_hotplug_event(image);
980 
981 	/* No longer handling a hotplug event */
982 	image->hp_action = KEXEC_CRASH_HP_NONE;
983 	image->elfcorehdr_updated = true;
984 
985 	/* Change back to read-only */
986 	arch_kexec_protect_crashkres();
987 
988 	/* Errors in the callback is not a reason to rollback state */
989 out:
990 	/* Release lock now that update complete */
991 	kexec_unlock();
992 	crash_hotplug_unlock();
993 }
994 
995 static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *v)
996 {
997 	switch (val) {
998 	case MEM_ONLINE:
999 		crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
1000 			KEXEC_CRASH_HP_INVALID_CPU);
1001 		break;
1002 
1003 	case MEM_OFFLINE:
1004 		crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
1005 			KEXEC_CRASH_HP_INVALID_CPU);
1006 		break;
1007 	}
1008 	return NOTIFY_OK;
1009 }
1010 
1011 static struct notifier_block crash_memhp_nb = {
1012 	.notifier_call = crash_memhp_notifier,
1013 	.priority = 0
1014 };
1015 
1016 static int crash_cpuhp_online(unsigned int cpu)
1017 {
1018 	crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu);
1019 	return 0;
1020 }
1021 
1022 static int crash_cpuhp_offline(unsigned int cpu)
1023 {
1024 	crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu);
1025 	return 0;
1026 }
1027 
1028 static int __init crash_hotplug_init(void)
1029 {
1030 	int result = 0;
1031 
1032 	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
1033 		register_memory_notifier(&crash_memhp_nb);
1034 
1035 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
1036 		result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
1037 			"crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
1038 	}
1039 
1040 	return result;
1041 }
1042 
1043 subsys_initcall(crash_hotplug_init);
1044 #endif
1045