Loading...
Note: File does not exist in v3.15.
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. */
27note_buf_t __percpu *crash_notes;
28
29/* vmcoreinfo stuff */
30unsigned char *vmcoreinfo_data;
31size_t vmcoreinfo_size;
32u32 *vmcoreinfo_note;
33
34/* trusted vmcoreinfo, e.g. we can make a copy in the crash memory */
35static unsigned char *vmcoreinfo_data_safecopy;
36
37/* Location of the reserved area for the crash kernel */
38struct 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};
45struct 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 */
67static 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 */
162static 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
187static __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 */
200static 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
226static __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 }
257next:
258 p = strstr(p+1, name);
259 }
260
261 return ck_cmdline;
262}
263
264static 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 */
307int __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 */
356static int __init parse_crashkernel_dummy(char *arg)
357{
358 return 0;
359}
360early_param("crashkernel", parse_crashkernel_dummy);
361
362#ifdef CONFIG_ARCH_HAS_GENERIC_CRASHKERNEL_RESERVATION
363static 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#endif
380 return 0;
381}
382
383void __init reserve_crashkernel_generic(char *cmdline,
384 unsigned long long crash_size,
385 unsigned long long crash_base,
386 unsigned long long crash_low_size,
387 bool high)
388{
389 unsigned long long search_end = CRASH_ADDR_LOW_MAX, search_base = 0;
390 bool fixed_base = false;
391
392 /* User specifies base address explicitly. */
393 if (crash_base) {
394 fixed_base = true;
395 search_base = crash_base;
396 search_end = crash_base + crash_size;
397 } else if (high) {
398 search_base = CRASH_ADDR_LOW_MAX;
399 search_end = CRASH_ADDR_HIGH_MAX;
400 }
401
402retry:
403 crash_base = memblock_phys_alloc_range(crash_size, CRASH_ALIGN,
404 search_base, search_end);
405 if (!crash_base) {
406 /*
407 * For crashkernel=size[KMG]@offset[KMG], print out failure
408 * message if can't reserve the specified region.
409 */
410 if (fixed_base) {
411 pr_warn("crashkernel reservation failed - memory is in use.\n");
412 return;
413 }
414
415 /*
416 * For crashkernel=size[KMG], if the first attempt was for
417 * low memory, fall back to high memory, the minimum required
418 * low memory will be reserved later.
419 */
420 if (!high && search_end == CRASH_ADDR_LOW_MAX) {
421 search_end = CRASH_ADDR_HIGH_MAX;
422 search_base = CRASH_ADDR_LOW_MAX;
423 crash_low_size = DEFAULT_CRASH_KERNEL_LOW_SIZE;
424 goto retry;
425 }
426
427 /*
428 * For crashkernel=size[KMG],high, if the first attempt was
429 * for high memory, fall back to low memory.
430 */
431 if (high && search_end == CRASH_ADDR_HIGH_MAX) {
432 search_end = CRASH_ADDR_LOW_MAX;
433 search_base = 0;
434 goto retry;
435 }
436 pr_warn("cannot allocate crashkernel (size:0x%llx)\n",
437 crash_size);
438 return;
439 }
440
441 if ((crash_base >= CRASH_ADDR_LOW_MAX) &&
442 crash_low_size && reserve_crashkernel_low(crash_low_size)) {
443 memblock_phys_free(crash_base, crash_size);
444 return;
445 }
446
447 pr_info("crashkernel reserved: 0x%016llx - 0x%016llx (%lld MB)\n",
448 crash_base, crash_base + crash_size, crash_size >> 20);
449
450 /*
451 * The crashkernel memory will be removed from the kernel linear
452 * map. Inform kmemleak so that it won't try to access it.
453 */
454 kmemleak_ignore_phys(crash_base);
455 if (crashk_low_res.end)
456 kmemleak_ignore_phys(crashk_low_res.start);
457
458 crashk_res.start = crash_base;
459 crashk_res.end = crash_base + crash_size - 1;
460}
461
462static __init int insert_crashkernel_resources(void)
463{
464 if (crashk_res.start < crashk_res.end)
465 insert_resource(&iomem_resource, &crashk_res);
466
467 if (crashk_low_res.start < crashk_low_res.end)
468 insert_resource(&iomem_resource, &crashk_low_res);
469
470 return 0;
471}
472early_initcall(insert_crashkernel_resources);
473#endif
474
475int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
476 void **addr, unsigned long *sz)
477{
478 Elf64_Ehdr *ehdr;
479 Elf64_Phdr *phdr;
480 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
481 unsigned char *buf;
482 unsigned int cpu, i;
483 unsigned long long notes_addr;
484 unsigned long mstart, mend;
485
486 /* extra phdr for vmcoreinfo ELF note */
487 nr_phdr = nr_cpus + 1;
488 nr_phdr += mem->nr_ranges;
489
490 /*
491 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
492 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
493 * I think this is required by tools like gdb. So same physical
494 * memory will be mapped in two ELF headers. One will contain kernel
495 * text virtual addresses and other will have __va(physical) addresses.
496 */
497
498 nr_phdr++;
499 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
500 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
501
502 buf = vzalloc(elf_sz);
503 if (!buf)
504 return -ENOMEM;
505
506 ehdr = (Elf64_Ehdr *)buf;
507 phdr = (Elf64_Phdr *)(ehdr + 1);
508 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
509 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
510 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
511 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
512 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
513 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
514 ehdr->e_type = ET_CORE;
515 ehdr->e_machine = ELF_ARCH;
516 ehdr->e_version = EV_CURRENT;
517 ehdr->e_phoff = sizeof(Elf64_Ehdr);
518 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
519 ehdr->e_phentsize = sizeof(Elf64_Phdr);
520
521 /* Prepare one phdr of type PT_NOTE for each possible CPU */
522 for_each_possible_cpu(cpu) {
523 phdr->p_type = PT_NOTE;
524 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
525 phdr->p_offset = phdr->p_paddr = notes_addr;
526 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
527 (ehdr->e_phnum)++;
528 phdr++;
529 }
530
531 /* Prepare one PT_NOTE header for vmcoreinfo */
532 phdr->p_type = PT_NOTE;
533 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
534 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
535 (ehdr->e_phnum)++;
536 phdr++;
537
538 /* Prepare PT_LOAD type program header for kernel text region */
539 if (need_kernel_map) {
540 phdr->p_type = PT_LOAD;
541 phdr->p_flags = PF_R|PF_W|PF_X;
542 phdr->p_vaddr = (unsigned long) _text;
543 phdr->p_filesz = phdr->p_memsz = _end - _text;
544 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
545 ehdr->e_phnum++;
546 phdr++;
547 }
548
549 /* Go through all the ranges in mem->ranges[] and prepare phdr */
550 for (i = 0; i < mem->nr_ranges; i++) {
551 mstart = mem->ranges[i].start;
552 mend = mem->ranges[i].end;
553
554 phdr->p_type = PT_LOAD;
555 phdr->p_flags = PF_R|PF_W|PF_X;
556 phdr->p_offset = mstart;
557
558 phdr->p_paddr = mstart;
559 phdr->p_vaddr = (unsigned long) __va(mstart);
560 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
561 phdr->p_align = 0;
562 ehdr->e_phnum++;
563#ifdef CONFIG_KEXEC_FILE
564 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",
565 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
566 ehdr->e_phnum, phdr->p_offset);
567#endif
568 phdr++;
569 }
570
571 *addr = buf;
572 *sz = elf_sz;
573 return 0;
574}
575
576int crash_exclude_mem_range(struct crash_mem *mem,
577 unsigned long long mstart, unsigned long long mend)
578{
579 int i;
580 unsigned long long start, end, p_start, p_end;
581
582 for (i = 0; i < mem->nr_ranges; i++) {
583 start = mem->ranges[i].start;
584 end = mem->ranges[i].end;
585 p_start = mstart;
586 p_end = mend;
587
588 if (p_start > end)
589 continue;
590
591 /*
592 * Because the memory ranges in mem->ranges are stored in
593 * ascending order, when we detect `p_end < start`, we can
594 * immediately exit the for loop, as the subsequent memory
595 * ranges will definitely be outside the range we are looking
596 * for.
597 */
598 if (p_end < start)
599 break;
600
601 /* Truncate any area outside of range */
602 if (p_start < start)
603 p_start = start;
604 if (p_end > end)
605 p_end = end;
606
607 /* Found completely overlapping range */
608 if (p_start == start && p_end == end) {
609 memmove(&mem->ranges[i], &mem->ranges[i + 1],
610 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
611 i--;
612 mem->nr_ranges--;
613 } else if (p_start > start && p_end < end) {
614 /* Split original range */
615 if (mem->nr_ranges >= mem->max_nr_ranges)
616 return -ENOMEM;
617
618 memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
619 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
620
621 mem->ranges[i].end = p_start - 1;
622 mem->ranges[i + 1].start = p_end + 1;
623 mem->ranges[i + 1].end = end;
624
625 i++;
626 mem->nr_ranges++;
627 } else if (p_start != start)
628 mem->ranges[i].end = p_start - 1;
629 else
630 mem->ranges[i].start = p_end + 1;
631 }
632
633 return 0;
634}
635
636Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type,
637 void *data, size_t data_len)
638{
639 struct elf_note *note = (struct elf_note *)buf;
640
641 note->n_namesz = strlen(name) + 1;
642 note->n_descsz = data_len;
643 note->n_type = type;
644 buf += DIV_ROUND_UP(sizeof(*note), sizeof(Elf_Word));
645 memcpy(buf, name, note->n_namesz);
646 buf += DIV_ROUND_UP(note->n_namesz, sizeof(Elf_Word));
647 memcpy(buf, data, data_len);
648 buf += DIV_ROUND_UP(data_len, sizeof(Elf_Word));
649
650 return buf;
651}
652
653void final_note(Elf_Word *buf)
654{
655 memset(buf, 0, sizeof(struct elf_note));
656}
657
658static void update_vmcoreinfo_note(void)
659{
660 u32 *buf = vmcoreinfo_note;
661
662 if (!vmcoreinfo_size)
663 return;
664 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
665 vmcoreinfo_size);
666 final_note(buf);
667}
668
669void crash_update_vmcoreinfo_safecopy(void *ptr)
670{
671 if (ptr)
672 memcpy(ptr, vmcoreinfo_data, vmcoreinfo_size);
673
674 vmcoreinfo_data_safecopy = ptr;
675}
676
677void crash_save_vmcoreinfo(void)
678{
679 if (!vmcoreinfo_note)
680 return;
681
682 /* Use the safe copy to generate vmcoreinfo note if have */
683 if (vmcoreinfo_data_safecopy)
684 vmcoreinfo_data = vmcoreinfo_data_safecopy;
685
686 vmcoreinfo_append_str("CRASHTIME=%lld\n", ktime_get_real_seconds());
687 update_vmcoreinfo_note();
688}
689
690void vmcoreinfo_append_str(const char *fmt, ...)
691{
692 va_list args;
693 char buf[0x50];
694 size_t r;
695
696 va_start(args, fmt);
697 r = vscnprintf(buf, sizeof(buf), fmt, args);
698 va_end(args);
699
700 r = min(r, (size_t)VMCOREINFO_BYTES - vmcoreinfo_size);
701
702 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
703
704 vmcoreinfo_size += r;
705
706 WARN_ONCE(vmcoreinfo_size == VMCOREINFO_BYTES,
707 "vmcoreinfo data exceeds allocated size, truncating");
708}
709
710/*
711 * provide an empty default implementation here -- architecture
712 * code may override this
713 */
714void __weak arch_crash_save_vmcoreinfo(void)
715{}
716
717phys_addr_t __weak paddr_vmcoreinfo_note(void)
718{
719 return __pa(vmcoreinfo_note);
720}
721EXPORT_SYMBOL(paddr_vmcoreinfo_note);
722
723static int __init crash_save_vmcoreinfo_init(void)
724{
725 vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
726 if (!vmcoreinfo_data) {
727 pr_warn("Memory allocation for vmcoreinfo_data failed\n");
728 return -ENOMEM;
729 }
730
731 vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
732 GFP_KERNEL | __GFP_ZERO);
733 if (!vmcoreinfo_note) {
734 free_page((unsigned long)vmcoreinfo_data);
735 vmcoreinfo_data = NULL;
736 pr_warn("Memory allocation for vmcoreinfo_note failed\n");
737 return -ENOMEM;
738 }
739
740 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
741 VMCOREINFO_BUILD_ID();
742 VMCOREINFO_PAGESIZE(PAGE_SIZE);
743
744 VMCOREINFO_SYMBOL(init_uts_ns);
745 VMCOREINFO_OFFSET(uts_namespace, name);
746 VMCOREINFO_SYMBOL(node_online_map);
747#ifdef CONFIG_MMU
748 VMCOREINFO_SYMBOL_ARRAY(swapper_pg_dir);
749#endif
750 VMCOREINFO_SYMBOL(_stext);
751 VMCOREINFO_SYMBOL(vmap_area_list);
752
753#ifndef CONFIG_NUMA
754 VMCOREINFO_SYMBOL(mem_map);
755 VMCOREINFO_SYMBOL(contig_page_data);
756#endif
757#ifdef CONFIG_SPARSEMEM
758 VMCOREINFO_SYMBOL_ARRAY(mem_section);
759 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
760 VMCOREINFO_STRUCT_SIZE(mem_section);
761 VMCOREINFO_OFFSET(mem_section, section_mem_map);
762 VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
763 VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
764#endif
765 VMCOREINFO_STRUCT_SIZE(page);
766 VMCOREINFO_STRUCT_SIZE(pglist_data);
767 VMCOREINFO_STRUCT_SIZE(zone);
768 VMCOREINFO_STRUCT_SIZE(free_area);
769 VMCOREINFO_STRUCT_SIZE(list_head);
770 VMCOREINFO_SIZE(nodemask_t);
771 VMCOREINFO_OFFSET(page, flags);
772 VMCOREINFO_OFFSET(page, _refcount);
773 VMCOREINFO_OFFSET(page, mapping);
774 VMCOREINFO_OFFSET(page, lru);
775 VMCOREINFO_OFFSET(page, _mapcount);
776 VMCOREINFO_OFFSET(page, private);
777 VMCOREINFO_OFFSET(page, compound_head);
778 VMCOREINFO_OFFSET(pglist_data, node_zones);
779 VMCOREINFO_OFFSET(pglist_data, nr_zones);
780#ifdef CONFIG_FLATMEM
781 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
782#endif
783 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
784 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
785 VMCOREINFO_OFFSET(pglist_data, node_id);
786 VMCOREINFO_OFFSET(zone, free_area);
787 VMCOREINFO_OFFSET(zone, vm_stat);
788 VMCOREINFO_OFFSET(zone, spanned_pages);
789 VMCOREINFO_OFFSET(free_area, free_list);
790 VMCOREINFO_OFFSET(list_head, next);
791 VMCOREINFO_OFFSET(list_head, prev);
792 VMCOREINFO_OFFSET(vmap_area, va_start);
793 VMCOREINFO_OFFSET(vmap_area, list);
794 VMCOREINFO_LENGTH(zone.free_area, NR_PAGE_ORDERS);
795 log_buf_vmcoreinfo_setup();
796 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
797 VMCOREINFO_NUMBER(NR_FREE_PAGES);
798 VMCOREINFO_NUMBER(PG_lru);
799 VMCOREINFO_NUMBER(PG_private);
800 VMCOREINFO_NUMBER(PG_swapcache);
801 VMCOREINFO_NUMBER(PG_swapbacked);
802 VMCOREINFO_NUMBER(PG_slab);
803#ifdef CONFIG_MEMORY_FAILURE
804 VMCOREINFO_NUMBER(PG_hwpoison);
805#endif
806 VMCOREINFO_NUMBER(PG_head_mask);
807#define PAGE_BUDDY_MAPCOUNT_VALUE (~PG_buddy)
808 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
809#ifdef CONFIG_HUGETLB_PAGE
810 VMCOREINFO_NUMBER(PG_hugetlb);
811#define PAGE_OFFLINE_MAPCOUNT_VALUE (~PG_offline)
812 VMCOREINFO_NUMBER(PAGE_OFFLINE_MAPCOUNT_VALUE);
813#endif
814
815#ifdef CONFIG_KALLSYMS
816 VMCOREINFO_SYMBOL(kallsyms_names);
817 VMCOREINFO_SYMBOL(kallsyms_num_syms);
818 VMCOREINFO_SYMBOL(kallsyms_token_table);
819 VMCOREINFO_SYMBOL(kallsyms_token_index);
820#ifdef CONFIG_KALLSYMS_BASE_RELATIVE
821 VMCOREINFO_SYMBOL(kallsyms_offsets);
822 VMCOREINFO_SYMBOL(kallsyms_relative_base);
823#else
824 VMCOREINFO_SYMBOL(kallsyms_addresses);
825#endif /* CONFIG_KALLSYMS_BASE_RELATIVE */
826#endif /* CONFIG_KALLSYMS */
827
828 arch_crash_save_vmcoreinfo();
829 update_vmcoreinfo_note();
830
831 return 0;
832}
833
834subsys_initcall(crash_save_vmcoreinfo_init);
835
836static int __init crash_notes_memory_init(void)
837{
838 /* Allocate memory for saving cpu registers. */
839 size_t size, align;
840
841 /*
842 * crash_notes could be allocated across 2 vmalloc pages when percpu
843 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
844 * pages are also on 2 continuous physical pages. In this case the
845 * 2nd part of crash_notes in 2nd page could be lost since only the
846 * starting address and size of crash_notes are exported through sysfs.
847 * Here round up the size of crash_notes to the nearest power of two
848 * and pass it to __alloc_percpu as align value. This can make sure
849 * crash_notes is allocated inside one physical page.
850 */
851 size = sizeof(note_buf_t);
852 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
853
854 /*
855 * Break compile if size is bigger than PAGE_SIZE since crash_notes
856 * definitely will be in 2 pages with that.
857 */
858 BUILD_BUG_ON(size > PAGE_SIZE);
859
860 crash_notes = __alloc_percpu(size, align);
861 if (!crash_notes) {
862 pr_warn("Memory allocation for saving cpu register states failed\n");
863 return -ENOMEM;
864 }
865 return 0;
866}
867subsys_initcall(crash_notes_memory_init);
868
869#ifdef CONFIG_CRASH_HOTPLUG
870#undef pr_fmt
871#define pr_fmt(fmt) "crash hp: " fmt
872
873/*
874 * Different than kexec/kdump loading/unloading/jumping/shrinking which
875 * usually rarely happen, there will be many crash hotplug events notified
876 * during one short period, e.g one memory board is hot added and memory
877 * regions are online. So mutex lock __crash_hotplug_lock is used to
878 * serialize the crash hotplug handling specifically.
879 */
880static DEFINE_MUTEX(__crash_hotplug_lock);
881#define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
882#define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
883
884/*
885 * This routine utilized when the crash_hotplug sysfs node is read.
886 * It reflects the kernel's ability/permission to update the crash
887 * elfcorehdr directly.
888 */
889int crash_check_update_elfcorehdr(void)
890{
891 int rc = 0;
892
893 crash_hotplug_lock();
894 /* Obtain lock while reading crash information */
895 if (!kexec_trylock()) {
896 pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
897 crash_hotplug_unlock();
898 return 0;
899 }
900 if (kexec_crash_image) {
901 if (kexec_crash_image->file_mode)
902 rc = 1;
903 else
904 rc = kexec_crash_image->update_elfcorehdr;
905 }
906 /* Release lock now that update complete */
907 kexec_unlock();
908 crash_hotplug_unlock();
909
910 return rc;
911}
912
913/*
914 * To accurately reflect hot un/plug changes of cpu and memory resources
915 * (including onling and offlining of those resources), the elfcorehdr
916 * (which is passed to the crash kernel via the elfcorehdr= parameter)
917 * must be updated with the new list of CPUs and memories.
918 *
919 * In order to make changes to elfcorehdr, two conditions are needed:
920 * First, the segment containing the elfcorehdr must be large enough
921 * to permit a growing number of resources; the elfcorehdr memory size
922 * is based on NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES.
923 * Second, purgatory must explicitly exclude the elfcorehdr from the
924 * list of segments it checks (since the elfcorehdr changes and thus
925 * would require an update to purgatory itself to update the digest).
926 */
927static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu)
928{
929 struct kimage *image;
930
931 crash_hotplug_lock();
932 /* Obtain lock while changing crash information */
933 if (!kexec_trylock()) {
934 pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
935 crash_hotplug_unlock();
936 return;
937 }
938
939 /* Check kdump is not loaded */
940 if (!kexec_crash_image)
941 goto out;
942
943 image = kexec_crash_image;
944
945 /* Check that updating elfcorehdr is permitted */
946 if (!(image->file_mode || image->update_elfcorehdr))
947 goto out;
948
949 if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
950 hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
951 pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
952 else
953 pr_debug("hp_action %u\n", hp_action);
954
955 /*
956 * The elfcorehdr_index is set to -1 when the struct kimage
957 * is allocated. Find the segment containing the elfcorehdr,
958 * if not already found.
959 */
960 if (image->elfcorehdr_index < 0) {
961 unsigned long mem;
962 unsigned char *ptr;
963 unsigned int n;
964
965 for (n = 0; n < image->nr_segments; n++) {
966 mem = image->segment[n].mem;
967 ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
968 if (ptr) {
969 /* The segment containing elfcorehdr */
970 if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
971 image->elfcorehdr_index = (int)n;
972 kunmap_local(ptr);
973 }
974 }
975 }
976
977 if (image->elfcorehdr_index < 0) {
978 pr_err("unable to locate elfcorehdr segment");
979 goto out;
980 }
981
982 /* Needed in order for the segments to be updated */
983 arch_kexec_unprotect_crashkres();
984
985 /* Differentiate between normal load and hotplug update */
986 image->hp_action = hp_action;
987
988 /* Now invoke arch-specific update handler */
989 arch_crash_handle_hotplug_event(image);
990
991 /* No longer handling a hotplug event */
992 image->hp_action = KEXEC_CRASH_HP_NONE;
993 image->elfcorehdr_updated = true;
994
995 /* Change back to read-only */
996 arch_kexec_protect_crashkres();
997
998 /* Errors in the callback is not a reason to rollback state */
999out:
1000 /* Release lock now that update complete */
1001 kexec_unlock();
1002 crash_hotplug_unlock();
1003}
1004
1005static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *v)
1006{
1007 switch (val) {
1008 case MEM_ONLINE:
1009 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
1010 KEXEC_CRASH_HP_INVALID_CPU);
1011 break;
1012
1013 case MEM_OFFLINE:
1014 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
1015 KEXEC_CRASH_HP_INVALID_CPU);
1016 break;
1017 }
1018 return NOTIFY_OK;
1019}
1020
1021static struct notifier_block crash_memhp_nb = {
1022 .notifier_call = crash_memhp_notifier,
1023 .priority = 0
1024};
1025
1026static int crash_cpuhp_online(unsigned int cpu)
1027{
1028 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu);
1029 return 0;
1030}
1031
1032static int crash_cpuhp_offline(unsigned int cpu)
1033{
1034 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu);
1035 return 0;
1036}
1037
1038static int __init crash_hotplug_init(void)
1039{
1040 int result = 0;
1041
1042 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
1043 register_memory_notifier(&crash_memhp_nb);
1044
1045 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
1046 result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
1047 "crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
1048 }
1049
1050 return result;
1051}
1052
1053subsys_initcall(crash_hotplug_init);
1054#endif