Loading...
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/crash_core.h>
8#include <linux/utsname.h>
9#include <linux/vmalloc.h>
10
11#include <asm/page.h>
12#include <asm/sections.h>
13
14/* vmcoreinfo stuff */
15unsigned char *vmcoreinfo_data;
16size_t vmcoreinfo_size;
17u32 *vmcoreinfo_note;
18
19/* trusted vmcoreinfo, e.g. we can make a copy in the crash memory */
20static unsigned char *vmcoreinfo_data_safecopy;
21
22/*
23 * parsing the "crashkernel" commandline
24 *
25 * this code is intended to be called from architecture specific code
26 */
27
28
29/*
30 * This function parses command lines in the format
31 *
32 * crashkernel=ramsize-range:size[,...][@offset]
33 *
34 * The function returns 0 on success and -EINVAL on failure.
35 */
36static int __init parse_crashkernel_mem(char *cmdline,
37 unsigned long long system_ram,
38 unsigned long long *crash_size,
39 unsigned long long *crash_base)
40{
41 char *cur = cmdline, *tmp;
42
43 /* for each entry of the comma-separated list */
44 do {
45 unsigned long long start, end = ULLONG_MAX, size;
46
47 /* get the start of the range */
48 start = memparse(cur, &tmp);
49 if (cur == tmp) {
50 pr_warn("crashkernel: Memory value expected\n");
51 return -EINVAL;
52 }
53 cur = tmp;
54 if (*cur != '-') {
55 pr_warn("crashkernel: '-' expected\n");
56 return -EINVAL;
57 }
58 cur++;
59
60 /* if no ':' is here, than we read the end */
61 if (*cur != ':') {
62 end = memparse(cur, &tmp);
63 if (cur == tmp) {
64 pr_warn("crashkernel: Memory value expected\n");
65 return -EINVAL;
66 }
67 cur = tmp;
68 if (end <= start) {
69 pr_warn("crashkernel: end <= start\n");
70 return -EINVAL;
71 }
72 }
73
74 if (*cur != ':') {
75 pr_warn("crashkernel: ':' expected\n");
76 return -EINVAL;
77 }
78 cur++;
79
80 size = memparse(cur, &tmp);
81 if (cur == tmp) {
82 pr_warn("Memory value expected\n");
83 return -EINVAL;
84 }
85 cur = tmp;
86 if (size >= system_ram) {
87 pr_warn("crashkernel: invalid size\n");
88 return -EINVAL;
89 }
90
91 /* match ? */
92 if (system_ram >= start && system_ram < end) {
93 *crash_size = size;
94 break;
95 }
96 } while (*cur++ == ',');
97
98 if (*crash_size > 0) {
99 while (*cur && *cur != ' ' && *cur != '@')
100 cur++;
101 if (*cur == '@') {
102 cur++;
103 *crash_base = memparse(cur, &tmp);
104 if (cur == tmp) {
105 pr_warn("Memory value expected after '@'\n");
106 return -EINVAL;
107 }
108 }
109 } else
110 pr_info("crashkernel size resulted in zero bytes\n");
111
112 return 0;
113}
114
115/*
116 * That function parses "simple" (old) crashkernel command lines like
117 *
118 * crashkernel=size[@offset]
119 *
120 * It returns 0 on success and -EINVAL on failure.
121 */
122static int __init parse_crashkernel_simple(char *cmdline,
123 unsigned long long *crash_size,
124 unsigned long long *crash_base)
125{
126 char *cur = cmdline;
127
128 *crash_size = memparse(cmdline, &cur);
129 if (cmdline == cur) {
130 pr_warn("crashkernel: memory value expected\n");
131 return -EINVAL;
132 }
133
134 if (*cur == '@')
135 *crash_base = memparse(cur+1, &cur);
136 else if (*cur != ' ' && *cur != '\0') {
137 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
138 return -EINVAL;
139 }
140
141 return 0;
142}
143
144#define SUFFIX_HIGH 0
145#define SUFFIX_LOW 1
146#define SUFFIX_NULL 2
147static __initdata char *suffix_tbl[] = {
148 [SUFFIX_HIGH] = ",high",
149 [SUFFIX_LOW] = ",low",
150 [SUFFIX_NULL] = NULL,
151};
152
153/*
154 * That function parses "suffix" crashkernel command lines like
155 *
156 * crashkernel=size,[high|low]
157 *
158 * It returns 0 on success and -EINVAL on failure.
159 */
160static int __init parse_crashkernel_suffix(char *cmdline,
161 unsigned long long *crash_size,
162 const char *suffix)
163{
164 char *cur = cmdline;
165
166 *crash_size = memparse(cmdline, &cur);
167 if (cmdline == cur) {
168 pr_warn("crashkernel: memory value expected\n");
169 return -EINVAL;
170 }
171
172 /* check with suffix */
173 if (strncmp(cur, suffix, strlen(suffix))) {
174 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
175 return -EINVAL;
176 }
177 cur += strlen(suffix);
178 if (*cur != ' ' && *cur != '\0') {
179 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
180 return -EINVAL;
181 }
182
183 return 0;
184}
185
186static __init char *get_last_crashkernel(char *cmdline,
187 const char *name,
188 const char *suffix)
189{
190 char *p = cmdline, *ck_cmdline = NULL;
191
192 /* find crashkernel and use the last one if there are more */
193 p = strstr(p, name);
194 while (p) {
195 char *end_p = strchr(p, ' ');
196 char *q;
197
198 if (!end_p)
199 end_p = p + strlen(p);
200
201 if (!suffix) {
202 int i;
203
204 /* skip the one with any known suffix */
205 for (i = 0; suffix_tbl[i]; i++) {
206 q = end_p - strlen(suffix_tbl[i]);
207 if (!strncmp(q, suffix_tbl[i],
208 strlen(suffix_tbl[i])))
209 goto next;
210 }
211 ck_cmdline = p;
212 } else {
213 q = end_p - strlen(suffix);
214 if (!strncmp(q, suffix, strlen(suffix)))
215 ck_cmdline = p;
216 }
217next:
218 p = strstr(p+1, name);
219 }
220
221 if (!ck_cmdline)
222 return NULL;
223
224 return ck_cmdline;
225}
226
227static int __init __parse_crashkernel(char *cmdline,
228 unsigned long long system_ram,
229 unsigned long long *crash_size,
230 unsigned long long *crash_base,
231 const char *name,
232 const char *suffix)
233{
234 char *first_colon, *first_space;
235 char *ck_cmdline;
236
237 BUG_ON(!crash_size || !crash_base);
238 *crash_size = 0;
239 *crash_base = 0;
240
241 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
242
243 if (!ck_cmdline)
244 return -EINVAL;
245
246 ck_cmdline += strlen(name);
247
248 if (suffix)
249 return parse_crashkernel_suffix(ck_cmdline, crash_size,
250 suffix);
251 /*
252 * if the commandline contains a ':', then that's the extended
253 * syntax -- if not, it must be the classic syntax
254 */
255 first_colon = strchr(ck_cmdline, ':');
256 first_space = strchr(ck_cmdline, ' ');
257 if (first_colon && (!first_space || first_colon < first_space))
258 return parse_crashkernel_mem(ck_cmdline, system_ram,
259 crash_size, crash_base);
260
261 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
262}
263
264/*
265 * That function is the entry point for command line parsing and should be
266 * called from the arch-specific code.
267 */
268int __init parse_crashkernel(char *cmdline,
269 unsigned long long system_ram,
270 unsigned long long *crash_size,
271 unsigned long long *crash_base)
272{
273 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
274 "crashkernel=", NULL);
275}
276
277int __init parse_crashkernel_high(char *cmdline,
278 unsigned long long system_ram,
279 unsigned long long *crash_size,
280 unsigned long long *crash_base)
281{
282 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
283 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
284}
285
286int __init parse_crashkernel_low(char *cmdline,
287 unsigned long long system_ram,
288 unsigned long long *crash_size,
289 unsigned long long *crash_base)
290{
291 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
292 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
293}
294
295Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type,
296 void *data, size_t data_len)
297{
298 struct elf_note *note = (struct elf_note *)buf;
299
300 note->n_namesz = strlen(name) + 1;
301 note->n_descsz = data_len;
302 note->n_type = type;
303 buf += DIV_ROUND_UP(sizeof(*note), sizeof(Elf_Word));
304 memcpy(buf, name, note->n_namesz);
305 buf += DIV_ROUND_UP(note->n_namesz, sizeof(Elf_Word));
306 memcpy(buf, data, data_len);
307 buf += DIV_ROUND_UP(data_len, sizeof(Elf_Word));
308
309 return buf;
310}
311
312void final_note(Elf_Word *buf)
313{
314 memset(buf, 0, sizeof(struct elf_note));
315}
316
317static void update_vmcoreinfo_note(void)
318{
319 u32 *buf = vmcoreinfo_note;
320
321 if (!vmcoreinfo_size)
322 return;
323 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
324 vmcoreinfo_size);
325 final_note(buf);
326}
327
328void crash_update_vmcoreinfo_safecopy(void *ptr)
329{
330 if (ptr)
331 memcpy(ptr, vmcoreinfo_data, vmcoreinfo_size);
332
333 vmcoreinfo_data_safecopy = ptr;
334}
335
336void crash_save_vmcoreinfo(void)
337{
338 if (!vmcoreinfo_note)
339 return;
340
341 /* Use the safe copy to generate vmcoreinfo note if have */
342 if (vmcoreinfo_data_safecopy)
343 vmcoreinfo_data = vmcoreinfo_data_safecopy;
344
345 vmcoreinfo_append_str("CRASHTIME=%lld\n", ktime_get_real_seconds());
346 update_vmcoreinfo_note();
347}
348
349void vmcoreinfo_append_str(const char *fmt, ...)
350{
351 va_list args;
352 char buf[0x50];
353 size_t r;
354
355 va_start(args, fmt);
356 r = vscnprintf(buf, sizeof(buf), fmt, args);
357 va_end(args);
358
359 r = min(r, (size_t)VMCOREINFO_BYTES - vmcoreinfo_size);
360
361 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
362
363 vmcoreinfo_size += r;
364}
365
366/*
367 * provide an empty default implementation here -- architecture
368 * code may override this
369 */
370void __weak arch_crash_save_vmcoreinfo(void)
371{}
372
373phys_addr_t __weak paddr_vmcoreinfo_note(void)
374{
375 return __pa(vmcoreinfo_note);
376}
377EXPORT_SYMBOL(paddr_vmcoreinfo_note);
378
379static int __init crash_save_vmcoreinfo_init(void)
380{
381 vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
382 if (!vmcoreinfo_data) {
383 pr_warn("Memory allocation for vmcoreinfo_data failed\n");
384 return -ENOMEM;
385 }
386
387 vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
388 GFP_KERNEL | __GFP_ZERO);
389 if (!vmcoreinfo_note) {
390 free_page((unsigned long)vmcoreinfo_data);
391 vmcoreinfo_data = NULL;
392 pr_warn("Memory allocation for vmcoreinfo_note failed\n");
393 return -ENOMEM;
394 }
395
396 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
397 VMCOREINFO_PAGESIZE(PAGE_SIZE);
398
399 VMCOREINFO_SYMBOL(init_uts_ns);
400 VMCOREINFO_SYMBOL(node_online_map);
401#ifdef CONFIG_MMU
402 VMCOREINFO_SYMBOL_ARRAY(swapper_pg_dir);
403#endif
404 VMCOREINFO_SYMBOL(_stext);
405 VMCOREINFO_SYMBOL(vmap_area_list);
406
407#ifndef CONFIG_NEED_MULTIPLE_NODES
408 VMCOREINFO_SYMBOL(mem_map);
409 VMCOREINFO_SYMBOL(contig_page_data);
410#endif
411#ifdef CONFIG_SPARSEMEM
412 VMCOREINFO_SYMBOL_ARRAY(mem_section);
413 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
414 VMCOREINFO_STRUCT_SIZE(mem_section);
415 VMCOREINFO_OFFSET(mem_section, section_mem_map);
416#endif
417 VMCOREINFO_STRUCT_SIZE(page);
418 VMCOREINFO_STRUCT_SIZE(pglist_data);
419 VMCOREINFO_STRUCT_SIZE(zone);
420 VMCOREINFO_STRUCT_SIZE(free_area);
421 VMCOREINFO_STRUCT_SIZE(list_head);
422 VMCOREINFO_SIZE(nodemask_t);
423 VMCOREINFO_OFFSET(page, flags);
424 VMCOREINFO_OFFSET(page, _refcount);
425 VMCOREINFO_OFFSET(page, mapping);
426 VMCOREINFO_OFFSET(page, lru);
427 VMCOREINFO_OFFSET(page, _mapcount);
428 VMCOREINFO_OFFSET(page, private);
429 VMCOREINFO_OFFSET(page, compound_dtor);
430 VMCOREINFO_OFFSET(page, compound_order);
431 VMCOREINFO_OFFSET(page, compound_head);
432 VMCOREINFO_OFFSET(pglist_data, node_zones);
433 VMCOREINFO_OFFSET(pglist_data, nr_zones);
434#ifdef CONFIG_FLAT_NODE_MEM_MAP
435 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
436#endif
437 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
438 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
439 VMCOREINFO_OFFSET(pglist_data, node_id);
440 VMCOREINFO_OFFSET(zone, free_area);
441 VMCOREINFO_OFFSET(zone, vm_stat);
442 VMCOREINFO_OFFSET(zone, spanned_pages);
443 VMCOREINFO_OFFSET(free_area, free_list);
444 VMCOREINFO_OFFSET(list_head, next);
445 VMCOREINFO_OFFSET(list_head, prev);
446 VMCOREINFO_OFFSET(vmap_area, va_start);
447 VMCOREINFO_OFFSET(vmap_area, list);
448 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
449 log_buf_vmcoreinfo_setup();
450 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
451 VMCOREINFO_NUMBER(NR_FREE_PAGES);
452 VMCOREINFO_NUMBER(PG_lru);
453 VMCOREINFO_NUMBER(PG_private);
454 VMCOREINFO_NUMBER(PG_swapcache);
455 VMCOREINFO_NUMBER(PG_swapbacked);
456 VMCOREINFO_NUMBER(PG_slab);
457#ifdef CONFIG_MEMORY_FAILURE
458 VMCOREINFO_NUMBER(PG_hwpoison);
459#endif
460 VMCOREINFO_NUMBER(PG_head_mask);
461#define PAGE_BUDDY_MAPCOUNT_VALUE (~PG_buddy)
462 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
463#ifdef CONFIG_HUGETLB_PAGE
464 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
465#define PAGE_OFFLINE_MAPCOUNT_VALUE (~PG_offline)
466 VMCOREINFO_NUMBER(PAGE_OFFLINE_MAPCOUNT_VALUE);
467#endif
468
469 arch_crash_save_vmcoreinfo();
470 update_vmcoreinfo_note();
471
472 return 0;
473}
474
475subsys_initcall(crash_save_vmcoreinfo_init);
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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/buildid.h>
10#include <linux/init.h>
11#include <linux/utsname.h>
12#include <linux/vmalloc.h>
13#include <linux/sizes.h>
14#include <linux/kexec.h>
15#include <linux/memory.h>
16#include <linux/mm.h>
17#include <linux/cpuhotplug.h>
18#include <linux/memblock.h>
19#include <linux/kmemleak.h>
20#include <linux/crash_core.h>
21#include <linux/reboot.h>
22#include <linux/btf.h>
23#include <linux/objtool.h>
24
25#include <asm/page.h>
26#include <asm/sections.h>
27
28#include <crypto/sha1.h>
29
30#include "kallsyms_internal.h"
31#include "kexec_internal.h"
32
33/* Per cpu memory for storing cpu states in case of system crash. */
34note_buf_t __percpu *crash_notes;
35
36#ifdef CONFIG_CRASH_DUMP
37
38int kimage_crash_copy_vmcoreinfo(struct kimage *image)
39{
40 struct page *vmcoreinfo_page;
41 void *safecopy;
42
43 if (!IS_ENABLED(CONFIG_CRASH_DUMP))
44 return 0;
45 if (image->type != KEXEC_TYPE_CRASH)
46 return 0;
47
48 /*
49 * For kdump, allocate one vmcoreinfo safe copy from the
50 * crash memory. as we have arch_kexec_protect_crashkres()
51 * after kexec syscall, we naturally protect it from write
52 * (even read) access under kernel direct mapping. But on
53 * the other hand, we still need to operate it when crash
54 * happens to generate vmcoreinfo note, hereby we rely on
55 * vmap for this purpose.
56 */
57 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
58 if (!vmcoreinfo_page) {
59 pr_warn("Could not allocate vmcoreinfo buffer\n");
60 return -ENOMEM;
61 }
62 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
63 if (!safecopy) {
64 pr_warn("Could not vmap vmcoreinfo buffer\n");
65 return -ENOMEM;
66 }
67
68 image->vmcoreinfo_data_copy = safecopy;
69 crash_update_vmcoreinfo_safecopy(safecopy);
70
71 return 0;
72}
73
74
75
76int kexec_should_crash(struct task_struct *p)
77{
78 /*
79 * If crash_kexec_post_notifiers is enabled, don't run
80 * crash_kexec() here yet, which must be run after panic
81 * notifiers in panic().
82 */
83 if (crash_kexec_post_notifiers)
84 return 0;
85 /*
86 * There are 4 panic() calls in make_task_dead() path, each of which
87 * corresponds to each of these 4 conditions.
88 */
89 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
90 return 1;
91 return 0;
92}
93
94int kexec_crash_loaded(void)
95{
96 return !!kexec_crash_image;
97}
98EXPORT_SYMBOL_GPL(kexec_crash_loaded);
99
100/*
101 * No panic_cpu check version of crash_kexec(). This function is called
102 * only when panic_cpu holds the current CPU number; this is the only CPU
103 * which processes crash_kexec routines.
104 */
105void __noclone __crash_kexec(struct pt_regs *regs)
106{
107 /* Take the kexec_lock here to prevent sys_kexec_load
108 * running on one cpu from replacing the crash kernel
109 * we are using after a panic on a different cpu.
110 *
111 * If the crash kernel was not located in a fixed area
112 * of memory the xchg(&kexec_crash_image) would be
113 * sufficient. But since I reuse the memory...
114 */
115 if (kexec_trylock()) {
116 if (kexec_crash_image) {
117 struct pt_regs fixed_regs;
118
119 crash_setup_regs(&fixed_regs, regs);
120 crash_save_vmcoreinfo();
121 machine_crash_shutdown(&fixed_regs);
122 machine_kexec(kexec_crash_image);
123 }
124 kexec_unlock();
125 }
126}
127STACK_FRAME_NON_STANDARD(__crash_kexec);
128
129__bpf_kfunc void crash_kexec(struct pt_regs *regs)
130{
131 int old_cpu, this_cpu;
132
133 /*
134 * Only one CPU is allowed to execute the crash_kexec() code as with
135 * panic(). Otherwise parallel calls of panic() and crash_kexec()
136 * may stop each other. To exclude them, we use panic_cpu here too.
137 */
138 old_cpu = PANIC_CPU_INVALID;
139 this_cpu = raw_smp_processor_id();
140
141 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
142 /* This is the 1st CPU which comes here, so go ahead. */
143 __crash_kexec(regs);
144
145 /*
146 * Reset panic_cpu to allow another panic()/crash_kexec()
147 * call.
148 */
149 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
150 }
151}
152
153static inline resource_size_t crash_resource_size(const struct resource *res)
154{
155 return !res->end ? 0 : resource_size(res);
156}
157
158
159
160
161int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
162 void **addr, unsigned long *sz)
163{
164 Elf64_Ehdr *ehdr;
165 Elf64_Phdr *phdr;
166 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
167 unsigned char *buf;
168 unsigned int cpu, i;
169 unsigned long long notes_addr;
170 unsigned long mstart, mend;
171
172 /* extra phdr for vmcoreinfo ELF note */
173 nr_phdr = nr_cpus + 1;
174 nr_phdr += mem->nr_ranges;
175
176 /*
177 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
178 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
179 * I think this is required by tools like gdb. So same physical
180 * memory will be mapped in two ELF headers. One will contain kernel
181 * text virtual addresses and other will have __va(physical) addresses.
182 */
183
184 nr_phdr++;
185 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
186 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
187
188 buf = vzalloc(elf_sz);
189 if (!buf)
190 return -ENOMEM;
191
192 ehdr = (Elf64_Ehdr *)buf;
193 phdr = (Elf64_Phdr *)(ehdr + 1);
194 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
195 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
196 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
197 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
198 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
199 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
200 ehdr->e_type = ET_CORE;
201 ehdr->e_machine = ELF_ARCH;
202 ehdr->e_version = EV_CURRENT;
203 ehdr->e_phoff = sizeof(Elf64_Ehdr);
204 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
205 ehdr->e_phentsize = sizeof(Elf64_Phdr);
206
207 /* Prepare one phdr of type PT_NOTE for each possible CPU */
208 for_each_possible_cpu(cpu) {
209 phdr->p_type = PT_NOTE;
210 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
211 phdr->p_offset = phdr->p_paddr = notes_addr;
212 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
213 (ehdr->e_phnum)++;
214 phdr++;
215 }
216
217 /* Prepare one PT_NOTE header for vmcoreinfo */
218 phdr->p_type = PT_NOTE;
219 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
220 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
221 (ehdr->e_phnum)++;
222 phdr++;
223
224 /* Prepare PT_LOAD type program header for kernel text region */
225 if (need_kernel_map) {
226 phdr->p_type = PT_LOAD;
227 phdr->p_flags = PF_R|PF_W|PF_X;
228 phdr->p_vaddr = (unsigned long) _text;
229 phdr->p_filesz = phdr->p_memsz = _end - _text;
230 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
231 ehdr->e_phnum++;
232 phdr++;
233 }
234
235 /* Go through all the ranges in mem->ranges[] and prepare phdr */
236 for (i = 0; i < mem->nr_ranges; i++) {
237 mstart = mem->ranges[i].start;
238 mend = mem->ranges[i].end;
239
240 phdr->p_type = PT_LOAD;
241 phdr->p_flags = PF_R|PF_W|PF_X;
242 phdr->p_offset = mstart;
243
244 phdr->p_paddr = mstart;
245 phdr->p_vaddr = (unsigned long) __va(mstart);
246 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
247 phdr->p_align = 0;
248 ehdr->e_phnum++;
249#ifdef CONFIG_KEXEC_FILE
250 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",
251 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
252 ehdr->e_phnum, phdr->p_offset);
253#endif
254 phdr++;
255 }
256
257 *addr = buf;
258 *sz = elf_sz;
259 return 0;
260}
261
262int crash_exclude_mem_range(struct crash_mem *mem,
263 unsigned long long mstart, unsigned long long mend)
264{
265 int i;
266 unsigned long long start, end, p_start, p_end;
267
268 for (i = 0; i < mem->nr_ranges; i++) {
269 start = mem->ranges[i].start;
270 end = mem->ranges[i].end;
271 p_start = mstart;
272 p_end = mend;
273
274 if (p_start > end)
275 continue;
276
277 /*
278 * Because the memory ranges in mem->ranges are stored in
279 * ascending order, when we detect `p_end < start`, we can
280 * immediately exit the for loop, as the subsequent memory
281 * ranges will definitely be outside the range we are looking
282 * for.
283 */
284 if (p_end < start)
285 break;
286
287 /* Truncate any area outside of range */
288 if (p_start < start)
289 p_start = start;
290 if (p_end > end)
291 p_end = end;
292
293 /* Found completely overlapping range */
294 if (p_start == start && p_end == end) {
295 memmove(&mem->ranges[i], &mem->ranges[i + 1],
296 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
297 i--;
298 mem->nr_ranges--;
299 } else if (p_start > start && p_end < end) {
300 /* Split original range */
301 if (mem->nr_ranges >= mem->max_nr_ranges)
302 return -ENOMEM;
303
304 memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
305 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
306
307 mem->ranges[i].end = p_start - 1;
308 mem->ranges[i + 1].start = p_end + 1;
309 mem->ranges[i + 1].end = end;
310
311 i++;
312 mem->nr_ranges++;
313 } else if (p_start != start)
314 mem->ranges[i].end = p_start - 1;
315 else
316 mem->ranges[i].start = p_end + 1;
317 }
318
319 return 0;
320}
321
322ssize_t crash_get_memory_size(void)
323{
324 ssize_t size = 0;
325
326 if (!kexec_trylock())
327 return -EBUSY;
328
329 size += crash_resource_size(&crashk_res);
330 size += crash_resource_size(&crashk_low_res);
331
332 kexec_unlock();
333 return size;
334}
335
336static int __crash_shrink_memory(struct resource *old_res,
337 unsigned long new_size)
338{
339 struct resource *ram_res;
340
341 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
342 if (!ram_res)
343 return -ENOMEM;
344
345 ram_res->start = old_res->start + new_size;
346 ram_res->end = old_res->end;
347 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
348 ram_res->name = "System RAM";
349
350 if (!new_size) {
351 release_resource(old_res);
352 old_res->start = 0;
353 old_res->end = 0;
354 } else {
355 crashk_res.end = ram_res->start - 1;
356 }
357
358 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
359 insert_resource(&iomem_resource, ram_res);
360
361 return 0;
362}
363
364int crash_shrink_memory(unsigned long new_size)
365{
366 int ret = 0;
367 unsigned long old_size, low_size;
368
369 if (!kexec_trylock())
370 return -EBUSY;
371
372 if (kexec_crash_image) {
373 ret = -ENOENT;
374 goto unlock;
375 }
376
377 low_size = crash_resource_size(&crashk_low_res);
378 old_size = crash_resource_size(&crashk_res) + low_size;
379 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
380 if (new_size >= old_size) {
381 ret = (new_size == old_size) ? 0 : -EINVAL;
382 goto unlock;
383 }
384
385 /*
386 * (low_size > new_size) implies that low_size is greater than zero.
387 * This also means that if low_size is zero, the else branch is taken.
388 *
389 * If low_size is greater than 0, (low_size > new_size) indicates that
390 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
391 * needs to be shrunken.
392 */
393 if (low_size > new_size) {
394 ret = __crash_shrink_memory(&crashk_res, 0);
395 if (ret)
396 goto unlock;
397
398 ret = __crash_shrink_memory(&crashk_low_res, new_size);
399 } else {
400 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
401 }
402
403 /* Swap crashk_res and crashk_low_res if needed */
404 if (!crashk_res.end && crashk_low_res.end) {
405 crashk_res.start = crashk_low_res.start;
406 crashk_res.end = crashk_low_res.end;
407 release_resource(&crashk_low_res);
408 crashk_low_res.start = 0;
409 crashk_low_res.end = 0;
410 insert_resource(&iomem_resource, &crashk_res);
411 }
412
413unlock:
414 kexec_unlock();
415 return ret;
416}
417
418void crash_save_cpu(struct pt_regs *regs, int cpu)
419{
420 struct elf_prstatus prstatus;
421 u32 *buf;
422
423 if ((cpu < 0) || (cpu >= nr_cpu_ids))
424 return;
425
426 /* Using ELF notes here is opportunistic.
427 * I need a well defined structure format
428 * for the data I pass, and I need tags
429 * on the data to indicate what information I have
430 * squirrelled away. ELF notes happen to provide
431 * all of that, so there is no need to invent something new.
432 */
433 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
434 if (!buf)
435 return;
436 memset(&prstatus, 0, sizeof(prstatus));
437 prstatus.common.pr_pid = current->pid;
438 elf_core_copy_regs(&prstatus.pr_reg, regs);
439 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
440 &prstatus, sizeof(prstatus));
441 final_note(buf);
442}
443
444
445
446static int __init crash_notes_memory_init(void)
447{
448 /* Allocate memory for saving cpu registers. */
449 size_t size, align;
450
451 /*
452 * crash_notes could be allocated across 2 vmalloc pages when percpu
453 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
454 * pages are also on 2 continuous physical pages. In this case the
455 * 2nd part of crash_notes in 2nd page could be lost since only the
456 * starting address and size of crash_notes are exported through sysfs.
457 * Here round up the size of crash_notes to the nearest power of two
458 * and pass it to __alloc_percpu as align value. This can make sure
459 * crash_notes is allocated inside one physical page.
460 */
461 size = sizeof(note_buf_t);
462 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
463
464 /*
465 * Break compile if size is bigger than PAGE_SIZE since crash_notes
466 * definitely will be in 2 pages with that.
467 */
468 BUILD_BUG_ON(size > PAGE_SIZE);
469
470 crash_notes = __alloc_percpu(size, align);
471 if (!crash_notes) {
472 pr_warn("Memory allocation for saving cpu register states failed\n");
473 return -ENOMEM;
474 }
475 return 0;
476}
477subsys_initcall(crash_notes_memory_init);
478
479#endif /*CONFIG_CRASH_DUMP*/
480
481#ifdef CONFIG_CRASH_HOTPLUG
482#undef pr_fmt
483#define pr_fmt(fmt) "crash hp: " fmt
484
485/*
486 * Different than kexec/kdump loading/unloading/jumping/shrinking which
487 * usually rarely happen, there will be many crash hotplug events notified
488 * during one short period, e.g one memory board is hot added and memory
489 * regions are online. So mutex lock __crash_hotplug_lock is used to
490 * serialize the crash hotplug handling specifically.
491 */
492static DEFINE_MUTEX(__crash_hotplug_lock);
493#define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
494#define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
495
496/*
497 * This routine utilized when the crash_hotplug sysfs node is read.
498 * It reflects the kernel's ability/permission to update the kdump
499 * image directly.
500 */
501int crash_check_hotplug_support(void)
502{
503 int rc = 0;
504
505 crash_hotplug_lock();
506 /* Obtain lock while reading crash information */
507 if (!kexec_trylock()) {
508 if (!kexec_in_progress)
509 pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
510 crash_hotplug_unlock();
511 return 0;
512 }
513 if (kexec_crash_image) {
514 rc = kexec_crash_image->hotplug_support;
515 }
516 /* Release lock now that update complete */
517 kexec_unlock();
518 crash_hotplug_unlock();
519
520 return rc;
521}
522
523/*
524 * To accurately reflect hot un/plug changes of CPU and Memory resources
525 * (including onling and offlining of those resources), the relevant
526 * kexec segments must be updated with latest CPU and Memory resources.
527 *
528 * Architectures must ensure two things for all segments that need
529 * updating during hotplug events:
530 *
531 * 1. Segments must be large enough to accommodate a growing number of
532 * resources.
533 * 2. Exclude the segments from SHA verification.
534 *
535 * For example, on most architectures, the elfcorehdr (which is passed
536 * to the crash kernel via the elfcorehdr= parameter) must include the
537 * new list of CPUs and memory. To make changes to the elfcorehdr, it
538 * should be large enough to permit a growing number of CPU and Memory
539 * resources. One can estimate the elfcorehdr memory size based on
540 * NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES. The elfcorehdr is
541 * excluded from SHA verification by default if the architecture
542 * supports crash hotplug.
543 */
544static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu, void *arg)
545{
546 struct kimage *image;
547
548 crash_hotplug_lock();
549 /* Obtain lock while changing crash information */
550 if (!kexec_trylock()) {
551 if (!kexec_in_progress)
552 pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
553 crash_hotplug_unlock();
554 return;
555 }
556
557 /* Check kdump is not loaded */
558 if (!kexec_crash_image)
559 goto out;
560
561 image = kexec_crash_image;
562
563 /* Check that kexec segments update is permitted */
564 if (!image->hotplug_support)
565 goto out;
566
567 if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
568 hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
569 pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
570 else
571 pr_debug("hp_action %u\n", hp_action);
572
573 /*
574 * The elfcorehdr_index is set to -1 when the struct kimage
575 * is allocated. Find the segment containing the elfcorehdr,
576 * if not already found.
577 */
578 if (image->elfcorehdr_index < 0) {
579 unsigned long mem;
580 unsigned char *ptr;
581 unsigned int n;
582
583 for (n = 0; n < image->nr_segments; n++) {
584 mem = image->segment[n].mem;
585 ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
586 if (ptr) {
587 /* The segment containing elfcorehdr */
588 if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
589 image->elfcorehdr_index = (int)n;
590 kunmap_local(ptr);
591 }
592 }
593 }
594
595 if (image->elfcorehdr_index < 0) {
596 pr_err("unable to locate elfcorehdr segment");
597 goto out;
598 }
599
600 /* Needed in order for the segments to be updated */
601 arch_kexec_unprotect_crashkres();
602
603 /* Differentiate between normal load and hotplug update */
604 image->hp_action = hp_action;
605
606 /* Now invoke arch-specific update handler */
607 arch_crash_handle_hotplug_event(image, arg);
608
609 /* No longer handling a hotplug event */
610 image->hp_action = KEXEC_CRASH_HP_NONE;
611 image->elfcorehdr_updated = true;
612
613 /* Change back to read-only */
614 arch_kexec_protect_crashkres();
615
616 /* Errors in the callback is not a reason to rollback state */
617out:
618 /* Release lock now that update complete */
619 kexec_unlock();
620 crash_hotplug_unlock();
621}
622
623static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *arg)
624{
625 switch (val) {
626 case MEM_ONLINE:
627 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
628 KEXEC_CRASH_HP_INVALID_CPU, arg);
629 break;
630
631 case MEM_OFFLINE:
632 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
633 KEXEC_CRASH_HP_INVALID_CPU, arg);
634 break;
635 }
636 return NOTIFY_OK;
637}
638
639static struct notifier_block crash_memhp_nb = {
640 .notifier_call = crash_memhp_notifier,
641 .priority = 0
642};
643
644static int crash_cpuhp_online(unsigned int cpu)
645{
646 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu, NULL);
647 return 0;
648}
649
650static int crash_cpuhp_offline(unsigned int cpu)
651{
652 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu, NULL);
653 return 0;
654}
655
656static int __init crash_hotplug_init(void)
657{
658 int result = 0;
659
660 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
661 register_memory_notifier(&crash_memhp_nb);
662
663 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
664 result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
665 "crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
666 }
667
668 return result;
669}
670
671subsys_initcall(crash_hotplug_init);
672#endif