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 #include <linux/delay.h>
25 #include <linux/panic.h>
26
27 #include <asm/page.h>
28 #include <asm/sections.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. */
34 note_buf_t __percpu *crash_notes;
35
36 /* time to wait for possible DMA to finish before starting the kdump kernel
37 * when a CMA reservation is used
38 */
39 #define CMA_DMA_TIMEOUT_SEC 10
40
41 #ifdef CONFIG_CRASH_DUMP
42
kimage_crash_copy_vmcoreinfo(struct kimage * image)43 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
44 {
45 struct page *vmcoreinfo_base;
46 struct page *vmcoreinfo_pages[DIV_ROUND_UP(VMCOREINFO_BYTES, PAGE_SIZE)];
47 unsigned int order, nr_pages;
48 int i;
49 void *safecopy;
50
51 nr_pages = DIV_ROUND_UP(VMCOREINFO_BYTES, PAGE_SIZE);
52 order = get_order(VMCOREINFO_BYTES);
53
54 if (!IS_ENABLED(CONFIG_CRASH_DUMP))
55 return 0;
56 if (image->type != KEXEC_TYPE_CRASH)
57 return 0;
58
59 /*
60 * For kdump, allocate one vmcoreinfo safe copy from the
61 * crash memory. as we have arch_kexec_protect_crashkres()
62 * after kexec syscall, we naturally protect it from write
63 * (even read) access under kernel direct mapping. But on
64 * the other hand, we still need to operate it when crash
65 * happens to generate vmcoreinfo note, hereby we rely on
66 * vmap for this purpose.
67 */
68 vmcoreinfo_base = kimage_alloc_control_pages(image, order);
69 if (!vmcoreinfo_base) {
70 pr_warn("Could not allocate vmcoreinfo buffer\n");
71 return -ENOMEM;
72 }
73 for (i = 0; i < nr_pages; i++)
74 vmcoreinfo_pages[i] = vmcoreinfo_base + i;
75
76 safecopy = vmap(vmcoreinfo_pages, nr_pages, VM_MAP, PAGE_KERNEL);
77 if (!safecopy) {
78 pr_warn("Could not vmap vmcoreinfo buffer\n");
79 return -ENOMEM;
80 }
81
82 image->vmcoreinfo_data_copy = safecopy;
83 crash_update_vmcoreinfo_safecopy(safecopy);
84
85 return 0;
86 }
87
88
89
kexec_should_crash(struct task_struct * p)90 int kexec_should_crash(struct task_struct *p)
91 {
92 /*
93 * If crash_kexec_post_notifiers is enabled, don't run
94 * crash_kexec() here yet, which must be run after panic
95 * notifiers in panic().
96 */
97 if (crash_kexec_post_notifiers)
98 return 0;
99 /*
100 * There are 4 panic() calls in make_task_dead() path, each of which
101 * corresponds to each of these 4 conditions.
102 */
103 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
104 return 1;
105 return 0;
106 }
107
kexec_crash_loaded(void)108 int kexec_crash_loaded(void)
109 {
110 return !!kexec_crash_image;
111 }
112 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
113
crash_cma_clear_pending_dma(void)114 static void crash_cma_clear_pending_dma(void)
115 {
116 if (!crashk_cma_cnt)
117 return;
118
119 mdelay(CMA_DMA_TIMEOUT_SEC * 1000);
120 }
121
122 /*
123 * No panic_cpu check version of crash_kexec(). This function is called
124 * only when panic_cpu holds the current CPU number; this is the only CPU
125 * which processes crash_kexec routines.
126 */
__crash_kexec(struct pt_regs * regs)127 void __noclone __crash_kexec(struct pt_regs *regs)
128 {
129 /* Take the kexec_lock here to prevent sys_kexec_load
130 * running on one cpu from replacing the crash kernel
131 * we are using after a panic on a different cpu.
132 *
133 * If the crash kernel was not located in a fixed area
134 * of memory the xchg(&kexec_crash_image) would be
135 * sufficient. But since I reuse the memory...
136 */
137 if (kexec_trylock()) {
138 if (kexec_crash_image) {
139 struct pt_regs fixed_regs;
140
141 crash_setup_regs(&fixed_regs, regs);
142 crash_save_vmcoreinfo();
143 machine_crash_shutdown(&fixed_regs);
144 crash_cma_clear_pending_dma();
145 machine_kexec(kexec_crash_image);
146 }
147 kexec_unlock();
148 }
149 }
150 STACK_FRAME_NON_STANDARD(__crash_kexec);
151
crash_kexec(struct pt_regs * regs)152 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
153 {
154 if (panic_try_start()) {
155 /* This is the 1st CPU which comes here, so go ahead. */
156 __crash_kexec(regs);
157
158 /*
159 * Reset panic_cpu to allow another panic()/crash_kexec()
160 * call.
161 */
162 panic_reset();
163 }
164 }
165
crash_resource_size(const struct resource * res)166 static inline resource_size_t crash_resource_size(const struct resource *res)
167 {
168 return !res->end ? 0 : resource_size(res);
169 }
170
crash_prepare_elf64_headers(struct crash_mem * mem,int need_kernel_map,void ** addr,unsigned long * sz)171 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
172 void **addr, unsigned long *sz)
173 {
174 Elf64_Ehdr *ehdr;
175 Elf64_Phdr *phdr;
176 unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
177 unsigned char *buf;
178 unsigned int cpu, i;
179 unsigned long long notes_addr;
180 unsigned long mstart, mend;
181
182 /* extra phdr for vmcoreinfo ELF note */
183 nr_phdr = nr_cpus + 1;
184 nr_phdr += mem->nr_ranges;
185
186 /*
187 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
188 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
189 * I think this is required by tools like gdb. So same physical
190 * memory will be mapped in two ELF headers. One will contain kernel
191 * text virtual addresses and other will have __va(physical) addresses.
192 */
193
194 nr_phdr++;
195 elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
196 elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
197
198 buf = vzalloc(elf_sz);
199 if (!buf)
200 return -ENOMEM;
201
202 ehdr = (Elf64_Ehdr *)buf;
203 phdr = (Elf64_Phdr *)(ehdr + 1);
204 memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
205 ehdr->e_ident[EI_CLASS] = ELFCLASS64;
206 ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
207 ehdr->e_ident[EI_VERSION] = EV_CURRENT;
208 ehdr->e_ident[EI_OSABI] = ELF_OSABI;
209 memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
210 ehdr->e_type = ET_CORE;
211 ehdr->e_machine = ELF_ARCH;
212 ehdr->e_version = EV_CURRENT;
213 ehdr->e_phoff = sizeof(Elf64_Ehdr);
214 ehdr->e_ehsize = sizeof(Elf64_Ehdr);
215 ehdr->e_phentsize = sizeof(Elf64_Phdr);
216
217 /* Prepare one phdr of type PT_NOTE for each possible CPU */
218 for_each_possible_cpu(cpu) {
219 phdr->p_type = PT_NOTE;
220 notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
221 phdr->p_offset = phdr->p_paddr = notes_addr;
222 phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
223 (ehdr->e_phnum)++;
224 phdr++;
225 }
226
227 /* Prepare one PT_NOTE header for vmcoreinfo */
228 phdr->p_type = PT_NOTE;
229 phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
230 phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
231 (ehdr->e_phnum)++;
232 phdr++;
233
234 /* Prepare PT_LOAD type program header for kernel text region */
235 if (need_kernel_map) {
236 phdr->p_type = PT_LOAD;
237 phdr->p_flags = PF_R|PF_W|PF_X;
238 phdr->p_vaddr = (unsigned long) _text;
239 phdr->p_filesz = phdr->p_memsz = _end - _text;
240 phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
241 ehdr->e_phnum++;
242 phdr++;
243 }
244
245 /* Go through all the ranges in mem->ranges[] and prepare phdr */
246 for (i = 0; i < mem->nr_ranges; i++) {
247 mstart = mem->ranges[i].start;
248 mend = mem->ranges[i].end;
249
250 phdr->p_type = PT_LOAD;
251 phdr->p_flags = PF_R|PF_W|PF_X;
252 phdr->p_offset = mstart;
253
254 phdr->p_paddr = mstart;
255 phdr->p_vaddr = (unsigned long) __va(mstart);
256 phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
257 phdr->p_align = 0;
258 ehdr->e_phnum++;
259 #ifdef CONFIG_KEXEC_FILE
260 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",
261 phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
262 ehdr->e_phnum, phdr->p_offset);
263 #endif
264 phdr++;
265 }
266
267 *addr = buf;
268 *sz = elf_sz;
269 return 0;
270 }
271
alloc_cmem(unsigned int nr_ranges)272 static struct crash_mem *alloc_cmem(unsigned int nr_ranges)
273 {
274 struct crash_mem *cmem;
275
276 cmem = kvzalloc_flex(*cmem, ranges, nr_ranges);
277 if (!cmem)
278 return NULL;
279
280 cmem->max_nr_ranges = nr_ranges;
281 return cmem;
282 }
283
arch_get_system_nr_ranges(void)284 unsigned int __weak arch_get_system_nr_ranges(void) { return 0; }
arch_crash_populate_cmem(struct crash_mem * cmem)285 int __weak arch_crash_populate_cmem(struct crash_mem *cmem) { return -1; }
arch_crash_exclude_ranges(struct crash_mem * cmem)286 int __weak arch_crash_exclude_ranges(struct crash_mem *cmem) { return 0; }
287
arch_crash_exclude_mem_range(struct crash_mem ** mem,unsigned long long mstart,unsigned long long mend)288 int __weak arch_crash_exclude_mem_range(struct crash_mem **mem,
289 unsigned long long mstart,
290 unsigned long long mend)
291 {
292 return crash_exclude_mem_range(*mem, mstart, mend);
293 }
294
crash_exclude_core_ranges(struct crash_mem ** cmem)295 int crash_exclude_core_ranges(struct crash_mem **cmem)
296 {
297 int ret, i;
298
299 /* Exclude crashkernel region */
300 ret = arch_crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
301 if (ret)
302 return ret;
303
304 if (crashk_low_res.end) {
305 ret = arch_crash_exclude_mem_range(cmem, crashk_low_res.start, crashk_low_res.end);
306 if (ret)
307 return ret;
308 }
309
310 for (i = 0; i < crashk_cma_cnt; ++i) {
311 ret = arch_crash_exclude_mem_range(cmem, crashk_cma_ranges[i].start,
312 crashk_cma_ranges[i].end);
313 if (ret)
314 return ret;
315 }
316
317 return 0;
318 }
319
crash_prepare_headers(int need_kernel_map,void ** addr,unsigned long * sz,unsigned long * nr_mem_ranges)320 int crash_prepare_headers(int need_kernel_map, void **addr, unsigned long *sz,
321 unsigned long *nr_mem_ranges)
322 {
323 unsigned int max_nr_ranges;
324 struct crash_mem *cmem;
325 int ret;
326
327 max_nr_ranges = arch_get_system_nr_ranges();
328 if (!max_nr_ranges)
329 return -ENOMEM;
330
331 cmem = alloc_cmem(max_nr_ranges);
332 if (!cmem)
333 return -ENOMEM;
334
335 ret = arch_crash_populate_cmem(cmem);
336 if (ret)
337 goto out;
338
339 ret = crash_exclude_core_ranges(&cmem);
340 if (ret)
341 goto out;
342
343 ret = arch_crash_exclude_ranges(cmem);
344 if (ret)
345 goto out;
346
347 /* Return the computed number of memory ranges, for hotplug usage */
348 if (nr_mem_ranges)
349 *nr_mem_ranges = cmem->nr_ranges;
350
351 ret = crash_prepare_elf64_headers(cmem, need_kernel_map, addr, sz);
352
353 out:
354 kvfree(cmem);
355 return ret;
356 }
357
358 /**
359 * crash_exclude_mem_range - exclude a mem range for existing ranges
360 * @mem: mem->range contains an array of ranges sorted in ascending order
361 * @mstart: the start of to-be-excluded range
362 * @mend: the start of to-be-excluded range
363 *
364 * If you are unsure if a range split will happen, to avoid function call
365 * failure because of -ENOMEM, always make sure
366 * mem->max_nr_ranges == mem->nr_ranges + 1
367 * before calling the function each time.
368 *
369 * returns 0 if a memory range is excluded successfully
370 * return -ENOMEM if mem->ranges doesn't have space to hold split ranges
371 */
crash_exclude_mem_range(struct crash_mem * mem,unsigned long long mstart,unsigned long long mend)372 int crash_exclude_mem_range(struct crash_mem *mem,
373 unsigned long long mstart, unsigned long long mend)
374 {
375 int i;
376 unsigned long long start, end, p_start, p_end;
377
378 for (i = 0; i < mem->nr_ranges; i++) {
379 start = mem->ranges[i].start;
380 end = mem->ranges[i].end;
381 p_start = mstart;
382 p_end = mend;
383
384 if (p_start > end)
385 continue;
386
387 /*
388 * Because the memory ranges in mem->ranges are stored in
389 * ascending order, when we detect `p_end < start`, we can
390 * immediately exit the for loop, as the subsequent memory
391 * ranges will definitely be outside the range we are looking
392 * for.
393 */
394 if (p_end < start)
395 break;
396
397 /* Truncate any area outside of range */
398 if (p_start < start)
399 p_start = start;
400 if (p_end > end)
401 p_end = end;
402
403 /* Found completely overlapping range */
404 if (p_start == start && p_end == end) {
405 memmove(&mem->ranges[i], &mem->ranges[i + 1],
406 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
407 i--;
408 mem->nr_ranges--;
409 } else if (p_start > start && p_end < end) {
410 /* Split original range */
411 if (mem->nr_ranges >= mem->max_nr_ranges)
412 return -ENOMEM;
413
414 memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
415 (mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
416
417 mem->ranges[i].end = p_start - 1;
418 mem->ranges[i + 1].start = p_end + 1;
419 mem->ranges[i + 1].end = end;
420
421 i++;
422 mem->nr_ranges++;
423 } else if (p_start != start)
424 mem->ranges[i].end = p_start - 1;
425 else
426 mem->ranges[i].start = p_end + 1;
427 }
428
429 return 0;
430 }
431 EXPORT_SYMBOL_GPL(crash_exclude_mem_range);
432
crash_get_memory_size(void)433 ssize_t crash_get_memory_size(void)
434 {
435 ssize_t size = 0;
436
437 if (!kexec_trylock())
438 return -EBUSY;
439
440 size += crash_resource_size(&crashk_res);
441 size += crash_resource_size(&crashk_low_res);
442
443 kexec_unlock();
444 return size;
445 }
446
__crash_shrink_memory(struct resource * old_res,unsigned long new_size)447 static int __crash_shrink_memory(struct resource *old_res,
448 unsigned long new_size)
449 {
450 struct resource *ram_res;
451
452 ram_res = kzalloc_obj(*ram_res);
453 if (!ram_res)
454 return -ENOMEM;
455
456 ram_res->start = old_res->start + new_size;
457 ram_res->end = old_res->end;
458 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
459 ram_res->name = "System RAM";
460
461 if (!new_size) {
462 release_resource(old_res);
463 old_res->start = 0;
464 old_res->end = 0;
465 } else {
466 old_res->end = ram_res->start - 1;
467 }
468
469 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
470 insert_resource(&iomem_resource, ram_res);
471
472 return 0;
473 }
474
crash_shrink_memory(unsigned long new_size)475 int crash_shrink_memory(unsigned long new_size)
476 {
477 int ret = 0;
478 unsigned long old_size, low_size;
479
480 if (!kexec_trylock())
481 return -EBUSY;
482
483 if (kexec_crash_image) {
484 ret = -ENOENT;
485 goto unlock;
486 }
487
488 low_size = crash_resource_size(&crashk_low_res);
489 old_size = crash_resource_size(&crashk_res) + low_size;
490 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
491 if (new_size >= old_size) {
492 ret = (new_size == old_size) ? 0 : -EINVAL;
493 goto unlock;
494 }
495
496 /*
497 * (low_size > new_size) implies that low_size is greater than zero.
498 * This also means that if low_size is zero, the else branch is taken.
499 *
500 * If low_size is greater than 0, (low_size > new_size) indicates that
501 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
502 * needs to be shrunken.
503 */
504 if (low_size > new_size) {
505 ret = __crash_shrink_memory(&crashk_res, 0);
506 if (ret)
507 goto unlock;
508
509 ret = __crash_shrink_memory(&crashk_low_res, new_size);
510 } else {
511 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
512 }
513
514 /* Swap crashk_res and crashk_low_res if needed */
515 if (!crashk_res.end && crashk_low_res.end) {
516 crashk_res.start = crashk_low_res.start;
517 crashk_res.end = crashk_low_res.end;
518 release_resource(&crashk_low_res);
519 crashk_low_res.start = 0;
520 crashk_low_res.end = 0;
521 insert_resource(&iomem_resource, &crashk_res);
522 }
523
524 unlock:
525 kexec_unlock();
526 return ret;
527 }
528
crash_save_cpu(struct pt_regs * regs,int cpu)529 void crash_save_cpu(struct pt_regs *regs, int cpu)
530 {
531 struct elf_prstatus prstatus;
532 u32 *buf;
533
534 if ((cpu < 0) || (cpu >= nr_cpu_ids))
535 return;
536
537 /* Using ELF notes here is opportunistic.
538 * I need a well defined structure format
539 * for the data I pass, and I need tags
540 * on the data to indicate what information I have
541 * squirrelled away. ELF notes happen to provide
542 * all of that, so there is no need to invent something new.
543 */
544 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
545 if (!buf)
546 return;
547 memset(&prstatus, 0, sizeof(prstatus));
548 prstatus.common.pr_pid = current->pid;
549 elf_core_copy_regs(&prstatus.pr_reg, regs);
550 buf = append_elf_note(buf, NN_PRSTATUS, NT_PRSTATUS,
551 &prstatus, sizeof(prstatus));
552 final_note(buf);
553 }
554
555
556
crash_notes_memory_init(void)557 static int __init crash_notes_memory_init(void)
558 {
559 /* Allocate memory for saving cpu registers. */
560 size_t size, align;
561
562 /*
563 * crash_notes could be allocated across 2 vmalloc pages when percpu
564 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
565 * pages are also on 2 continuous physical pages. In this case the
566 * 2nd part of crash_notes in 2nd page could be lost since only the
567 * starting address and size of crash_notes are exported through sysfs.
568 * Here round up the size of crash_notes to the nearest power of two
569 * and pass it to __alloc_percpu as align value. This can make sure
570 * crash_notes is allocated inside one physical page.
571 */
572 size = sizeof(note_buf_t);
573 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
574
575 /*
576 * Break compile if size is bigger than PAGE_SIZE since crash_notes
577 * definitely will be in 2 pages with that.
578 */
579 BUILD_BUG_ON(size > PAGE_SIZE);
580
581 crash_notes = __alloc_percpu(size, align);
582 if (!crash_notes) {
583 pr_warn("Memory allocation for saving cpu register states failed\n");
584 return -ENOMEM;
585 }
586 return 0;
587 }
588 subsys_initcall(crash_notes_memory_init);
589
590 #endif /*CONFIG_CRASH_DUMP*/
591
592 #ifdef CONFIG_CRASH_HOTPLUG
593 #undef pr_fmt
594 #define pr_fmt(fmt) "crash hp: " fmt
595
596 /*
597 * Different than kexec/kdump loading/unloading/jumping/shrinking which
598 * usually rarely happen, there will be many crash hotplug events notified
599 * during one short period, e.g one memory board is hot added and memory
600 * regions are online. So mutex lock __crash_hotplug_lock is used to
601 * serialize the crash hotplug handling specifically.
602 */
603 static DEFINE_MUTEX(__crash_hotplug_lock);
604 #define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
605 #define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
606
607 /*
608 * This routine utilized when the crash_hotplug sysfs node is read.
609 * It reflects the kernel's ability/permission to update the kdump
610 * image directly.
611 */
crash_check_hotplug_support(void)612 int crash_check_hotplug_support(void)
613 {
614 int rc = 0;
615
616 crash_hotplug_lock();
617 /* Obtain lock while reading crash information */
618 if (!kexec_trylock()) {
619 if (!kexec_in_progress)
620 pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
621 crash_hotplug_unlock();
622 return 0;
623 }
624 if (kexec_crash_image) {
625 rc = kexec_crash_image->hotplug_support;
626 }
627 /* Release lock now that update complete */
628 kexec_unlock();
629 crash_hotplug_unlock();
630
631 return rc;
632 }
633
634 /*
635 * To accurately reflect hot un/plug changes of CPU and Memory resources
636 * (including onling and offlining of those resources), the relevant
637 * kexec segments must be updated with latest CPU and Memory resources.
638 *
639 * Architectures must ensure two things for all segments that need
640 * updating during hotplug events:
641 *
642 * 1. Segments must be large enough to accommodate a growing number of
643 * resources.
644 * 2. Exclude the segments from SHA verification.
645 *
646 * For example, on most architectures, the elfcorehdr (which is passed
647 * to the crash kernel via the elfcorehdr= parameter) must include the
648 * new list of CPUs and memory. To make changes to the elfcorehdr, it
649 * should be large enough to permit a growing number of CPU and Memory
650 * resources. One can estimate the elfcorehdr memory size based on
651 * NR_CPUS and CRASH_MAX_MEMORY_RANGES. The elfcorehdr is
652 * excluded from SHA verification by default if the architecture
653 * supports crash hotplug.
654 */
crash_handle_hotplug_event(unsigned int hp_action,unsigned int cpu,void * arg)655 static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu, void *arg)
656 {
657 struct kimage *image;
658
659 crash_hotplug_lock();
660 /* Obtain lock while changing crash information */
661 if (!kexec_trylock()) {
662 if (!kexec_in_progress)
663 pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
664 crash_hotplug_unlock();
665 return;
666 }
667
668 /* Check kdump is not loaded */
669 if (!kexec_crash_image)
670 goto out;
671
672 image = kexec_crash_image;
673
674 /* Check that kexec segments update is permitted */
675 if (!image->hotplug_support)
676 goto out;
677
678 if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
679 hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
680 pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
681 else
682 pr_debug("hp_action %u\n", hp_action);
683
684 /*
685 * The elfcorehdr_index is set to -1 when the struct kimage
686 * is allocated. Find the segment containing the elfcorehdr,
687 * if not already found.
688 */
689 if (image->elfcorehdr_index < 0) {
690 unsigned long mem;
691 unsigned char *ptr;
692 unsigned int n;
693
694 for (n = 0; n < image->nr_segments; n++) {
695 mem = image->segment[n].mem;
696 ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
697 if (ptr) {
698 /* The segment containing elfcorehdr */
699 if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
700 image->elfcorehdr_index = (int)n;
701 kunmap_local(ptr);
702 }
703 }
704 }
705
706 if (image->elfcorehdr_index < 0) {
707 pr_err("unable to locate elfcorehdr segment");
708 goto out;
709 }
710
711 /* Needed in order for the segments to be updated */
712 arch_kexec_unprotect_crashkres();
713
714 /* Differentiate between normal load and hotplug update */
715 image->hp_action = hp_action;
716
717 /* Now invoke arch-specific update handler */
718 arch_crash_handle_hotplug_event(image, arg);
719
720 /* No longer handling a hotplug event */
721 image->hp_action = KEXEC_CRASH_HP_NONE;
722 image->elfcorehdr_updated = true;
723
724 /* Change back to read-only */
725 arch_kexec_protect_crashkres();
726
727 /* Errors in the callback is not a reason to rollback state */
728 out:
729 /* Release lock now that update complete */
730 kexec_unlock();
731 crash_hotplug_unlock();
732 }
733
crash_memhp_notifier(struct notifier_block * nb,unsigned long val,void * arg)734 static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *arg)
735 {
736 switch (val) {
737 case MEM_ONLINE:
738 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
739 KEXEC_CRASH_HP_INVALID_CPU, arg);
740 break;
741
742 case MEM_OFFLINE:
743 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
744 KEXEC_CRASH_HP_INVALID_CPU, arg);
745 break;
746 }
747 return NOTIFY_OK;
748 }
749
750 static struct notifier_block crash_memhp_nb = {
751 .notifier_call = crash_memhp_notifier,
752 .priority = 0
753 };
754
crash_cpuhp_online(unsigned int cpu)755 static int crash_cpuhp_online(unsigned int cpu)
756 {
757 crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu, NULL);
758 return 0;
759 }
760
crash_cpuhp_offline(unsigned int cpu)761 static int crash_cpuhp_offline(unsigned int cpu)
762 {
763 crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu, NULL);
764 return 0;
765 }
766
crash_hotplug_init(void)767 static int __init crash_hotplug_init(void)
768 {
769 int result = 0;
770
771 if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
772 register_memory_notifier(&crash_memhp_nb);
773
774 if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
775 result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
776 "crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
777 }
778
779 return result;
780 }
781
782 subsys_initcall(crash_hotplug_init);
783 #endif
784