xref: /linux/kernel/crash_core.c (revision 561add0da6d3d07c9bccb0832fb6ed5619167d26)
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/crash_core.h>
9 #include <linux/init.h>
10 #include <linux/utsname.h>
11 #include <linux/vmalloc.h>
12 #include <linux/sizes.h>
13 #include <linux/kexec.h>
14 #include <linux/memory.h>
15 #include <linux/cpuhotplug.h>
16 
17 #include <asm/page.h>
18 #include <asm/sections.h>
19 
20 #include <crypto/sha1.h>
21 
22 #include "kallsyms_internal.h"
23 #include "kexec_internal.h"
24 
25 /* Per cpu memory for storing cpu states in case of system crash. */
26 note_buf_t __percpu *crash_notes;
27 
28 /* vmcoreinfo stuff */
29 unsigned char *vmcoreinfo_data;
30 size_t vmcoreinfo_size;
31 u32 *vmcoreinfo_note;
32 
33 /* trusted vmcoreinfo, e.g. we can make a copy in the crash memory */
34 static unsigned char *vmcoreinfo_data_safecopy;
35 
36 /*
37  * parsing the "crashkernel" commandline
38  *
39  * this code is intended to be called from architecture specific code
40  */
41 
42 
43 /*
44  * This function parses command lines in the format
45  *
46  *   crashkernel=ramsize-range:size[,...][@offset]
47  *
48  * The function returns 0 on success and -EINVAL on failure.
49  */
50 static int __init parse_crashkernel_mem(char *cmdline,
51 					unsigned long long system_ram,
52 					unsigned long long *crash_size,
53 					unsigned long long *crash_base)
54 {
55 	char *cur = cmdline, *tmp;
56 	unsigned long long total_mem = system_ram;
57 
58 	/*
59 	 * Firmware sometimes reserves some memory regions for its own use,
60 	 * so the system memory size is less than the actual physical memory
61 	 * size. Work around this by rounding up the total size to 128M,
62 	 * which is enough for most test cases.
63 	 */
64 	total_mem = roundup(total_mem, SZ_128M);
65 
66 	/* for each entry of the comma-separated list */
67 	do {
68 		unsigned long long start, end = ULLONG_MAX, size;
69 
70 		/* get the start of the range */
71 		start = memparse(cur, &tmp);
72 		if (cur == tmp) {
73 			pr_warn("crashkernel: Memory value expected\n");
74 			return -EINVAL;
75 		}
76 		cur = tmp;
77 		if (*cur != '-') {
78 			pr_warn("crashkernel: '-' expected\n");
79 			return -EINVAL;
80 		}
81 		cur++;
82 
83 		/* if no ':' is here, than we read the end */
84 		if (*cur != ':') {
85 			end = memparse(cur, &tmp);
86 			if (cur == tmp) {
87 				pr_warn("crashkernel: Memory value expected\n");
88 				return -EINVAL;
89 			}
90 			cur = tmp;
91 			if (end <= start) {
92 				pr_warn("crashkernel: end <= start\n");
93 				return -EINVAL;
94 			}
95 		}
96 
97 		if (*cur != ':') {
98 			pr_warn("crashkernel: ':' expected\n");
99 			return -EINVAL;
100 		}
101 		cur++;
102 
103 		size = memparse(cur, &tmp);
104 		if (cur == tmp) {
105 			pr_warn("Memory value expected\n");
106 			return -EINVAL;
107 		}
108 		cur = tmp;
109 		if (size >= total_mem) {
110 			pr_warn("crashkernel: invalid size\n");
111 			return -EINVAL;
112 		}
113 
114 		/* match ? */
115 		if (total_mem >= start && total_mem < end) {
116 			*crash_size = size;
117 			break;
118 		}
119 	} while (*cur++ == ',');
120 
121 	if (*crash_size > 0) {
122 		while (*cur && *cur != ' ' && *cur != '@')
123 			cur++;
124 		if (*cur == '@') {
125 			cur++;
126 			*crash_base = memparse(cur, &tmp);
127 			if (cur == tmp) {
128 				pr_warn("Memory value expected after '@'\n");
129 				return -EINVAL;
130 			}
131 		}
132 	} else
133 		pr_info("crashkernel size resulted in zero bytes\n");
134 
135 	return 0;
136 }
137 
138 /*
139  * That function parses "simple" (old) crashkernel command lines like
140  *
141  *	crashkernel=size[@offset]
142  *
143  * It returns 0 on success and -EINVAL on failure.
144  */
145 static int __init parse_crashkernel_simple(char *cmdline,
146 					   unsigned long long *crash_size,
147 					   unsigned long long *crash_base)
148 {
149 	char *cur = cmdline;
150 
151 	*crash_size = memparse(cmdline, &cur);
152 	if (cmdline == cur) {
153 		pr_warn("crashkernel: memory value expected\n");
154 		return -EINVAL;
155 	}
156 
157 	if (*cur == '@')
158 		*crash_base = memparse(cur+1, &cur);
159 	else if (*cur != ' ' && *cur != '\0') {
160 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
161 		return -EINVAL;
162 	}
163 
164 	return 0;
165 }
166 
167 #define SUFFIX_HIGH 0
168 #define SUFFIX_LOW  1
169 #define SUFFIX_NULL 2
170 static __initdata char *suffix_tbl[] = {
171 	[SUFFIX_HIGH] = ",high",
172 	[SUFFIX_LOW]  = ",low",
173 	[SUFFIX_NULL] = NULL,
174 };
175 
176 /*
177  * That function parses "suffix"  crashkernel command lines like
178  *
179  *	crashkernel=size,[high|low]
180  *
181  * It returns 0 on success and -EINVAL on failure.
182  */
183 static int __init parse_crashkernel_suffix(char *cmdline,
184 					   unsigned long long	*crash_size,
185 					   const char *suffix)
186 {
187 	char *cur = cmdline;
188 
189 	*crash_size = memparse(cmdline, &cur);
190 	if (cmdline == cur) {
191 		pr_warn("crashkernel: memory value expected\n");
192 		return -EINVAL;
193 	}
194 
195 	/* check with suffix */
196 	if (strncmp(cur, suffix, strlen(suffix))) {
197 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
198 		return -EINVAL;
199 	}
200 	cur += strlen(suffix);
201 	if (*cur != ' ' && *cur != '\0') {
202 		pr_warn("crashkernel: unrecognized char: %c\n", *cur);
203 		return -EINVAL;
204 	}
205 
206 	return 0;
207 }
208 
209 static __init char *get_last_crashkernel(char *cmdline,
210 			     const char *name,
211 			     const char *suffix)
212 {
213 	char *p = cmdline, *ck_cmdline = NULL;
214 
215 	/* find crashkernel and use the last one if there are more */
216 	p = strstr(p, name);
217 	while (p) {
218 		char *end_p = strchr(p, ' ');
219 		char *q;
220 
221 		if (!end_p)
222 			end_p = p + strlen(p);
223 
224 		if (!suffix) {
225 			int i;
226 
227 			/* skip the one with any known suffix */
228 			for (i = 0; suffix_tbl[i]; i++) {
229 				q = end_p - strlen(suffix_tbl[i]);
230 				if (!strncmp(q, suffix_tbl[i],
231 					     strlen(suffix_tbl[i])))
232 					goto next;
233 			}
234 			ck_cmdline = p;
235 		} else {
236 			q = end_p - strlen(suffix);
237 			if (!strncmp(q, suffix, strlen(suffix)))
238 				ck_cmdline = p;
239 		}
240 next:
241 		p = strstr(p+1, name);
242 	}
243 
244 	return ck_cmdline;
245 }
246 
247 static int __init __parse_crashkernel(char *cmdline,
248 			     unsigned long long system_ram,
249 			     unsigned long long *crash_size,
250 			     unsigned long long *crash_base,
251 			     const char *name,
252 			     const char *suffix)
253 {
254 	char	*first_colon, *first_space;
255 	char	*ck_cmdline;
256 
257 	BUG_ON(!crash_size || !crash_base);
258 	*crash_size = 0;
259 	*crash_base = 0;
260 
261 	ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
262 	if (!ck_cmdline)
263 		return -ENOENT;
264 
265 	ck_cmdline += strlen(name);
266 
267 	if (suffix)
268 		return parse_crashkernel_suffix(ck_cmdline, crash_size,
269 				suffix);
270 	/*
271 	 * if the commandline contains a ':', then that's the extended
272 	 * syntax -- if not, it must be the classic syntax
273 	 */
274 	first_colon = strchr(ck_cmdline, ':');
275 	first_space = strchr(ck_cmdline, ' ');
276 	if (first_colon && (!first_space || first_colon < first_space))
277 		return parse_crashkernel_mem(ck_cmdline, system_ram,
278 				crash_size, crash_base);
279 
280 	return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
281 }
282 
283 /*
284  * That function is the entry point for command line parsing and should be
285  * called from the arch-specific code.
286  */
287 int __init parse_crashkernel(char *cmdline,
288 			     unsigned long long system_ram,
289 			     unsigned long long *crash_size,
290 			     unsigned long long *crash_base)
291 {
292 	return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
293 					"crashkernel=", NULL);
294 }
295 
296 int __init parse_crashkernel_high(char *cmdline,
297 			     unsigned long long system_ram,
298 			     unsigned long long *crash_size,
299 			     unsigned long long *crash_base)
300 {
301 	return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
302 				"crashkernel=", suffix_tbl[SUFFIX_HIGH]);
303 }
304 
305 int __init parse_crashkernel_low(char *cmdline,
306 			     unsigned long long system_ram,
307 			     unsigned long long *crash_size,
308 			     unsigned long long *crash_base)
309 {
310 	return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
311 				"crashkernel=", suffix_tbl[SUFFIX_LOW]);
312 }
313 
314 /*
315  * Add a dummy early_param handler to mark crashkernel= as a known command line
316  * parameter and suppress incorrect warnings in init/main.c.
317  */
318 static int __init parse_crashkernel_dummy(char *arg)
319 {
320 	return 0;
321 }
322 early_param("crashkernel", parse_crashkernel_dummy);
323 
324 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
325 			  void **addr, unsigned long *sz)
326 {
327 	Elf64_Ehdr *ehdr;
328 	Elf64_Phdr *phdr;
329 	unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
330 	unsigned char *buf;
331 	unsigned int cpu, i;
332 	unsigned long long notes_addr;
333 	unsigned long mstart, mend;
334 
335 	/* extra phdr for vmcoreinfo ELF note */
336 	nr_phdr = nr_cpus + 1;
337 	nr_phdr += mem->nr_ranges;
338 
339 	/*
340 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
341 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
342 	 * I think this is required by tools like gdb. So same physical
343 	 * memory will be mapped in two ELF headers. One will contain kernel
344 	 * text virtual addresses and other will have __va(physical) addresses.
345 	 */
346 
347 	nr_phdr++;
348 	elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
349 	elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
350 
351 	buf = vzalloc(elf_sz);
352 	if (!buf)
353 		return -ENOMEM;
354 
355 	ehdr = (Elf64_Ehdr *)buf;
356 	phdr = (Elf64_Phdr *)(ehdr + 1);
357 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
358 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
359 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
360 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
361 	ehdr->e_ident[EI_OSABI] = ELF_OSABI;
362 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
363 	ehdr->e_type = ET_CORE;
364 	ehdr->e_machine = ELF_ARCH;
365 	ehdr->e_version = EV_CURRENT;
366 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
367 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
368 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
369 
370 	/* Prepare one phdr of type PT_NOTE for each possible CPU */
371 	for_each_possible_cpu(cpu) {
372 		phdr->p_type = PT_NOTE;
373 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
374 		phdr->p_offset = phdr->p_paddr = notes_addr;
375 		phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
376 		(ehdr->e_phnum)++;
377 		phdr++;
378 	}
379 
380 	/* Prepare one PT_NOTE header for vmcoreinfo */
381 	phdr->p_type = PT_NOTE;
382 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
383 	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
384 	(ehdr->e_phnum)++;
385 	phdr++;
386 
387 	/* Prepare PT_LOAD type program header for kernel text region */
388 	if (need_kernel_map) {
389 		phdr->p_type = PT_LOAD;
390 		phdr->p_flags = PF_R|PF_W|PF_X;
391 		phdr->p_vaddr = (unsigned long) _text;
392 		phdr->p_filesz = phdr->p_memsz = _end - _text;
393 		phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
394 		ehdr->e_phnum++;
395 		phdr++;
396 	}
397 
398 	/* Go through all the ranges in mem->ranges[] and prepare phdr */
399 	for (i = 0; i < mem->nr_ranges; i++) {
400 		mstart = mem->ranges[i].start;
401 		mend = mem->ranges[i].end;
402 
403 		phdr->p_type = PT_LOAD;
404 		phdr->p_flags = PF_R|PF_W|PF_X;
405 		phdr->p_offset  = mstart;
406 
407 		phdr->p_paddr = mstart;
408 		phdr->p_vaddr = (unsigned long) __va(mstart);
409 		phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
410 		phdr->p_align = 0;
411 		ehdr->e_phnum++;
412 		pr_debug("Crash PT_LOAD ELF header. phdr=%p vaddr=0x%llx, paddr=0x%llx, sz=0x%llx e_phnum=%d p_offset=0x%llx\n",
413 			phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
414 			ehdr->e_phnum, phdr->p_offset);
415 		phdr++;
416 	}
417 
418 	*addr = buf;
419 	*sz = elf_sz;
420 	return 0;
421 }
422 
423 int crash_exclude_mem_range(struct crash_mem *mem,
424 			    unsigned long long mstart, unsigned long long mend)
425 {
426 	int i, j;
427 	unsigned long long start, end, p_start, p_end;
428 	struct range temp_range = {0, 0};
429 
430 	for (i = 0; i < mem->nr_ranges; i++) {
431 		start = mem->ranges[i].start;
432 		end = mem->ranges[i].end;
433 		p_start = mstart;
434 		p_end = mend;
435 
436 		if (mstart > end || mend < start)
437 			continue;
438 
439 		/* Truncate any area outside of range */
440 		if (mstart < start)
441 			p_start = start;
442 		if (mend > end)
443 			p_end = end;
444 
445 		/* Found completely overlapping range */
446 		if (p_start == start && p_end == end) {
447 			mem->ranges[i].start = 0;
448 			mem->ranges[i].end = 0;
449 			if (i < mem->nr_ranges - 1) {
450 				/* Shift rest of the ranges to left */
451 				for (j = i; j < mem->nr_ranges - 1; j++) {
452 					mem->ranges[j].start =
453 						mem->ranges[j+1].start;
454 					mem->ranges[j].end =
455 							mem->ranges[j+1].end;
456 				}
457 
458 				/*
459 				 * Continue to check if there are another overlapping ranges
460 				 * from the current position because of shifting the above
461 				 * mem ranges.
462 				 */
463 				i--;
464 				mem->nr_ranges--;
465 				continue;
466 			}
467 			mem->nr_ranges--;
468 			return 0;
469 		}
470 
471 		if (p_start > start && p_end < end) {
472 			/* Split original range */
473 			mem->ranges[i].end = p_start - 1;
474 			temp_range.start = p_end + 1;
475 			temp_range.end = end;
476 		} else if (p_start != start)
477 			mem->ranges[i].end = p_start - 1;
478 		else
479 			mem->ranges[i].start = p_end + 1;
480 		break;
481 	}
482 
483 	/* If a split happened, add the split to array */
484 	if (!temp_range.end)
485 		return 0;
486 
487 	/* Split happened */
488 	if (i == mem->max_nr_ranges - 1)
489 		return -ENOMEM;
490 
491 	/* Location where new range should go */
492 	j = i + 1;
493 	if (j < mem->nr_ranges) {
494 		/* Move over all ranges one slot towards the end */
495 		for (i = mem->nr_ranges - 1; i >= j; i--)
496 			mem->ranges[i + 1] = mem->ranges[i];
497 	}
498 
499 	mem->ranges[j].start = temp_range.start;
500 	mem->ranges[j].end = temp_range.end;
501 	mem->nr_ranges++;
502 	return 0;
503 }
504 
505 Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type,
506 			  void *data, size_t data_len)
507 {
508 	struct elf_note *note = (struct elf_note *)buf;
509 
510 	note->n_namesz = strlen(name) + 1;
511 	note->n_descsz = data_len;
512 	note->n_type   = type;
513 	buf += DIV_ROUND_UP(sizeof(*note), sizeof(Elf_Word));
514 	memcpy(buf, name, note->n_namesz);
515 	buf += DIV_ROUND_UP(note->n_namesz, sizeof(Elf_Word));
516 	memcpy(buf, data, data_len);
517 	buf += DIV_ROUND_UP(data_len, sizeof(Elf_Word));
518 
519 	return buf;
520 }
521 
522 void final_note(Elf_Word *buf)
523 {
524 	memset(buf, 0, sizeof(struct elf_note));
525 }
526 
527 static void update_vmcoreinfo_note(void)
528 {
529 	u32 *buf = vmcoreinfo_note;
530 
531 	if (!vmcoreinfo_size)
532 		return;
533 	buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
534 			      vmcoreinfo_size);
535 	final_note(buf);
536 }
537 
538 void crash_update_vmcoreinfo_safecopy(void *ptr)
539 {
540 	if (ptr)
541 		memcpy(ptr, vmcoreinfo_data, vmcoreinfo_size);
542 
543 	vmcoreinfo_data_safecopy = ptr;
544 }
545 
546 void crash_save_vmcoreinfo(void)
547 {
548 	if (!vmcoreinfo_note)
549 		return;
550 
551 	/* Use the safe copy to generate vmcoreinfo note if have */
552 	if (vmcoreinfo_data_safecopy)
553 		vmcoreinfo_data = vmcoreinfo_data_safecopy;
554 
555 	vmcoreinfo_append_str("CRASHTIME=%lld\n", ktime_get_real_seconds());
556 	update_vmcoreinfo_note();
557 }
558 
559 void vmcoreinfo_append_str(const char *fmt, ...)
560 {
561 	va_list args;
562 	char buf[0x50];
563 	size_t r;
564 
565 	va_start(args, fmt);
566 	r = vscnprintf(buf, sizeof(buf), fmt, args);
567 	va_end(args);
568 
569 	r = min(r, (size_t)VMCOREINFO_BYTES - vmcoreinfo_size);
570 
571 	memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
572 
573 	vmcoreinfo_size += r;
574 
575 	WARN_ONCE(vmcoreinfo_size == VMCOREINFO_BYTES,
576 		  "vmcoreinfo data exceeds allocated size, truncating");
577 }
578 
579 /*
580  * provide an empty default implementation here -- architecture
581  * code may override this
582  */
583 void __weak arch_crash_save_vmcoreinfo(void)
584 {}
585 
586 phys_addr_t __weak paddr_vmcoreinfo_note(void)
587 {
588 	return __pa(vmcoreinfo_note);
589 }
590 EXPORT_SYMBOL(paddr_vmcoreinfo_note);
591 
592 static int __init crash_save_vmcoreinfo_init(void)
593 {
594 	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
595 	if (!vmcoreinfo_data) {
596 		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
597 		return -ENOMEM;
598 	}
599 
600 	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
601 						GFP_KERNEL | __GFP_ZERO);
602 	if (!vmcoreinfo_note) {
603 		free_page((unsigned long)vmcoreinfo_data);
604 		vmcoreinfo_data = NULL;
605 		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
606 		return -ENOMEM;
607 	}
608 
609 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
610 	VMCOREINFO_BUILD_ID();
611 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
612 
613 	VMCOREINFO_SYMBOL(init_uts_ns);
614 	VMCOREINFO_OFFSET(uts_namespace, name);
615 	VMCOREINFO_SYMBOL(node_online_map);
616 #ifdef CONFIG_MMU
617 	VMCOREINFO_SYMBOL_ARRAY(swapper_pg_dir);
618 #endif
619 	VMCOREINFO_SYMBOL(_stext);
620 	VMCOREINFO_SYMBOL(vmap_area_list);
621 
622 #ifndef CONFIG_NUMA
623 	VMCOREINFO_SYMBOL(mem_map);
624 	VMCOREINFO_SYMBOL(contig_page_data);
625 #endif
626 #ifdef CONFIG_SPARSEMEM
627 	VMCOREINFO_SYMBOL_ARRAY(mem_section);
628 	VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
629 	VMCOREINFO_STRUCT_SIZE(mem_section);
630 	VMCOREINFO_OFFSET(mem_section, section_mem_map);
631 	VMCOREINFO_NUMBER(SECTION_SIZE_BITS);
632 	VMCOREINFO_NUMBER(MAX_PHYSMEM_BITS);
633 #endif
634 	VMCOREINFO_STRUCT_SIZE(page);
635 	VMCOREINFO_STRUCT_SIZE(pglist_data);
636 	VMCOREINFO_STRUCT_SIZE(zone);
637 	VMCOREINFO_STRUCT_SIZE(free_area);
638 	VMCOREINFO_STRUCT_SIZE(list_head);
639 	VMCOREINFO_SIZE(nodemask_t);
640 	VMCOREINFO_OFFSET(page, flags);
641 	VMCOREINFO_OFFSET(page, _refcount);
642 	VMCOREINFO_OFFSET(page, mapping);
643 	VMCOREINFO_OFFSET(page, lru);
644 	VMCOREINFO_OFFSET(page, _mapcount);
645 	VMCOREINFO_OFFSET(page, private);
646 	VMCOREINFO_OFFSET(page, compound_head);
647 	VMCOREINFO_OFFSET(pglist_data, node_zones);
648 	VMCOREINFO_OFFSET(pglist_data, nr_zones);
649 #ifdef CONFIG_FLATMEM
650 	VMCOREINFO_OFFSET(pglist_data, node_mem_map);
651 #endif
652 	VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
653 	VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
654 	VMCOREINFO_OFFSET(pglist_data, node_id);
655 	VMCOREINFO_OFFSET(zone, free_area);
656 	VMCOREINFO_OFFSET(zone, vm_stat);
657 	VMCOREINFO_OFFSET(zone, spanned_pages);
658 	VMCOREINFO_OFFSET(free_area, free_list);
659 	VMCOREINFO_OFFSET(list_head, next);
660 	VMCOREINFO_OFFSET(list_head, prev);
661 	VMCOREINFO_OFFSET(vmap_area, va_start);
662 	VMCOREINFO_OFFSET(vmap_area, list);
663 	VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER + 1);
664 	log_buf_vmcoreinfo_setup();
665 	VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
666 	VMCOREINFO_NUMBER(NR_FREE_PAGES);
667 	VMCOREINFO_NUMBER(PG_lru);
668 	VMCOREINFO_NUMBER(PG_private);
669 	VMCOREINFO_NUMBER(PG_swapcache);
670 	VMCOREINFO_NUMBER(PG_swapbacked);
671 	VMCOREINFO_NUMBER(PG_slab);
672 #ifdef CONFIG_MEMORY_FAILURE
673 	VMCOREINFO_NUMBER(PG_hwpoison);
674 #endif
675 	VMCOREINFO_NUMBER(PG_head_mask);
676 #define PAGE_BUDDY_MAPCOUNT_VALUE	(~PG_buddy)
677 	VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
678 #ifdef CONFIG_HUGETLB_PAGE
679 	VMCOREINFO_NUMBER(PG_hugetlb);
680 #define PAGE_OFFLINE_MAPCOUNT_VALUE	(~PG_offline)
681 	VMCOREINFO_NUMBER(PAGE_OFFLINE_MAPCOUNT_VALUE);
682 #endif
683 
684 #ifdef CONFIG_KALLSYMS
685 	VMCOREINFO_SYMBOL(kallsyms_names);
686 	VMCOREINFO_SYMBOL(kallsyms_num_syms);
687 	VMCOREINFO_SYMBOL(kallsyms_token_table);
688 	VMCOREINFO_SYMBOL(kallsyms_token_index);
689 #ifdef CONFIG_KALLSYMS_BASE_RELATIVE
690 	VMCOREINFO_SYMBOL(kallsyms_offsets);
691 	VMCOREINFO_SYMBOL(kallsyms_relative_base);
692 #else
693 	VMCOREINFO_SYMBOL(kallsyms_addresses);
694 #endif /* CONFIG_KALLSYMS_BASE_RELATIVE */
695 #endif /* CONFIG_KALLSYMS */
696 
697 	arch_crash_save_vmcoreinfo();
698 	update_vmcoreinfo_note();
699 
700 	return 0;
701 }
702 
703 subsys_initcall(crash_save_vmcoreinfo_init);
704 
705 static int __init crash_notes_memory_init(void)
706 {
707 	/* Allocate memory for saving cpu registers. */
708 	size_t size, align;
709 
710 	/*
711 	 * crash_notes could be allocated across 2 vmalloc pages when percpu
712 	 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
713 	 * pages are also on 2 continuous physical pages. In this case the
714 	 * 2nd part of crash_notes in 2nd page could be lost since only the
715 	 * starting address and size of crash_notes are exported through sysfs.
716 	 * Here round up the size of crash_notes to the nearest power of two
717 	 * and pass it to __alloc_percpu as align value. This can make sure
718 	 * crash_notes is allocated inside one physical page.
719 	 */
720 	size = sizeof(note_buf_t);
721 	align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
722 
723 	/*
724 	 * Break compile if size is bigger than PAGE_SIZE since crash_notes
725 	 * definitely will be in 2 pages with that.
726 	 */
727 	BUILD_BUG_ON(size > PAGE_SIZE);
728 
729 	crash_notes = __alloc_percpu(size, align);
730 	if (!crash_notes) {
731 		pr_warn("Memory allocation for saving cpu register states failed\n");
732 		return -ENOMEM;
733 	}
734 	return 0;
735 }
736 subsys_initcall(crash_notes_memory_init);
737 
738 #ifdef CONFIG_CRASH_HOTPLUG
739 #undef pr_fmt
740 #define pr_fmt(fmt) "crash hp: " fmt
741 
742 /*
743  * This routine utilized when the crash_hotplug sysfs node is read.
744  * It reflects the kernel's ability/permission to update the crash
745  * elfcorehdr directly.
746  */
747 int crash_check_update_elfcorehdr(void)
748 {
749 	int rc = 0;
750 
751 	/* Obtain lock while reading crash information */
752 	if (!kexec_trylock()) {
753 		pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
754 		return 0;
755 	}
756 	if (kexec_crash_image) {
757 		if (kexec_crash_image->file_mode)
758 			rc = 1;
759 		else
760 			rc = kexec_crash_image->update_elfcorehdr;
761 	}
762 	/* Release lock now that update complete */
763 	kexec_unlock();
764 
765 	return rc;
766 }
767 
768 /*
769  * To accurately reflect hot un/plug changes of cpu and memory resources
770  * (including onling and offlining of those resources), the elfcorehdr
771  * (which is passed to the crash kernel via the elfcorehdr= parameter)
772  * must be updated with the new list of CPUs and memories.
773  *
774  * In order to make changes to elfcorehdr, two conditions are needed:
775  * First, the segment containing the elfcorehdr must be large enough
776  * to permit a growing number of resources; the elfcorehdr memory size
777  * is based on NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES.
778  * Second, purgatory must explicitly exclude the elfcorehdr from the
779  * list of segments it checks (since the elfcorehdr changes and thus
780  * would require an update to purgatory itself to update the digest).
781  */
782 static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu)
783 {
784 	struct kimage *image;
785 
786 	/* Obtain lock while changing crash information */
787 	if (!kexec_trylock()) {
788 		pr_info("kexec_trylock() failed, elfcorehdr may be inaccurate\n");
789 		return;
790 	}
791 
792 	/* Check kdump is not loaded */
793 	if (!kexec_crash_image)
794 		goto out;
795 
796 	image = kexec_crash_image;
797 
798 	/* Check that updating elfcorehdr is permitted */
799 	if (!(image->file_mode || image->update_elfcorehdr))
800 		goto out;
801 
802 	if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
803 		hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
804 		pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
805 	else
806 		pr_debug("hp_action %u\n", hp_action);
807 
808 	/*
809 	 * The elfcorehdr_index is set to -1 when the struct kimage
810 	 * is allocated. Find the segment containing the elfcorehdr,
811 	 * if not already found.
812 	 */
813 	if (image->elfcorehdr_index < 0) {
814 		unsigned long mem;
815 		unsigned char *ptr;
816 		unsigned int n;
817 
818 		for (n = 0; n < image->nr_segments; n++) {
819 			mem = image->segment[n].mem;
820 			ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
821 			if (ptr) {
822 				/* The segment containing elfcorehdr */
823 				if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
824 					image->elfcorehdr_index = (int)n;
825 				kunmap_local(ptr);
826 			}
827 		}
828 	}
829 
830 	if (image->elfcorehdr_index < 0) {
831 		pr_err("unable to locate elfcorehdr segment");
832 		goto out;
833 	}
834 
835 	/* Needed in order for the segments to be updated */
836 	arch_kexec_unprotect_crashkres();
837 
838 	/* Differentiate between normal load and hotplug update */
839 	image->hp_action = hp_action;
840 
841 	/* Now invoke arch-specific update handler */
842 	arch_crash_handle_hotplug_event(image);
843 
844 	/* No longer handling a hotplug event */
845 	image->hp_action = KEXEC_CRASH_HP_NONE;
846 	image->elfcorehdr_updated = true;
847 
848 	/* Change back to read-only */
849 	arch_kexec_protect_crashkres();
850 
851 	/* Errors in the callback is not a reason to rollback state */
852 out:
853 	/* Release lock now that update complete */
854 	kexec_unlock();
855 }
856 
857 static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *v)
858 {
859 	switch (val) {
860 	case MEM_ONLINE:
861 		crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
862 			KEXEC_CRASH_HP_INVALID_CPU);
863 		break;
864 
865 	case MEM_OFFLINE:
866 		crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
867 			KEXEC_CRASH_HP_INVALID_CPU);
868 		break;
869 	}
870 	return NOTIFY_OK;
871 }
872 
873 static struct notifier_block crash_memhp_nb = {
874 	.notifier_call = crash_memhp_notifier,
875 	.priority = 0
876 };
877 
878 static int crash_cpuhp_online(unsigned int cpu)
879 {
880 	crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu);
881 	return 0;
882 }
883 
884 static int crash_cpuhp_offline(unsigned int cpu)
885 {
886 	crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu);
887 	return 0;
888 }
889 
890 static int __init crash_hotplug_init(void)
891 {
892 	int result = 0;
893 
894 	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
895 		register_memory_notifier(&crash_memhp_nb);
896 
897 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
898 		result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
899 			"crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
900 	}
901 
902 	return result;
903 }
904 
905 subsys_initcall(crash_hotplug_init);
906 #endif
907