xref: /linux/kernel/crash_core.c (revision e406d57be7bd2a4e73ea512c1ae36a40a44e499e)
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 <crypto/sha1.h>
31 
32 #include "kallsyms_internal.h"
33 #include "kexec_internal.h"
34 
35 /* Per cpu memory for storing cpu states in case of system crash. */
36 note_buf_t __percpu *crash_notes;
37 
38 /* time to wait for possible DMA to finish before starting the kdump kernel
39  * when a CMA reservation is used
40  */
41 #define CMA_DMA_TIMEOUT_SEC 10
42 
43 #ifdef CONFIG_CRASH_DUMP
44 
kimage_crash_copy_vmcoreinfo(struct kimage * image)45 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
46 {
47 	struct page *vmcoreinfo_page;
48 	void *safecopy;
49 
50 	if (!IS_ENABLED(CONFIG_CRASH_DUMP))
51 		return 0;
52 	if (image->type != KEXEC_TYPE_CRASH)
53 		return 0;
54 
55 	/*
56 	 * For kdump, allocate one vmcoreinfo safe copy from the
57 	 * crash memory. as we have arch_kexec_protect_crashkres()
58 	 * after kexec syscall, we naturally protect it from write
59 	 * (even read) access under kernel direct mapping. But on
60 	 * the other hand, we still need to operate it when crash
61 	 * happens to generate vmcoreinfo note, hereby we rely on
62 	 * vmap for this purpose.
63 	 */
64 	vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
65 	if (!vmcoreinfo_page) {
66 		pr_warn("Could not allocate vmcoreinfo buffer\n");
67 		return -ENOMEM;
68 	}
69 	safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
70 	if (!safecopy) {
71 		pr_warn("Could not vmap vmcoreinfo buffer\n");
72 		return -ENOMEM;
73 	}
74 
75 	image->vmcoreinfo_data_copy = safecopy;
76 	crash_update_vmcoreinfo_safecopy(safecopy);
77 
78 	return 0;
79 }
80 
81 
82 
kexec_should_crash(struct task_struct * p)83 int kexec_should_crash(struct task_struct *p)
84 {
85 	/*
86 	 * If crash_kexec_post_notifiers is enabled, don't run
87 	 * crash_kexec() here yet, which must be run after panic
88 	 * notifiers in panic().
89 	 */
90 	if (crash_kexec_post_notifiers)
91 		return 0;
92 	/*
93 	 * There are 4 panic() calls in make_task_dead() path, each of which
94 	 * corresponds to each of these 4 conditions.
95 	 */
96 	if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
97 		return 1;
98 	return 0;
99 }
100 
kexec_crash_loaded(void)101 int kexec_crash_loaded(void)
102 {
103 	return !!kexec_crash_image;
104 }
105 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
106 
crash_cma_clear_pending_dma(void)107 static void crash_cma_clear_pending_dma(void)
108 {
109 	if (!crashk_cma_cnt)
110 		return;
111 
112 	mdelay(CMA_DMA_TIMEOUT_SEC * 1000);
113 }
114 
115 /*
116  * No panic_cpu check version of crash_kexec().  This function is called
117  * only when panic_cpu holds the current CPU number; this is the only CPU
118  * which processes crash_kexec routines.
119  */
__crash_kexec(struct pt_regs * regs)120 void __noclone __crash_kexec(struct pt_regs *regs)
121 {
122 	/* Take the kexec_lock here to prevent sys_kexec_load
123 	 * running on one cpu from replacing the crash kernel
124 	 * we are using after a panic on a different cpu.
125 	 *
126 	 * If the crash kernel was not located in a fixed area
127 	 * of memory the xchg(&kexec_crash_image) would be
128 	 * sufficient.  But since I reuse the memory...
129 	 */
130 	if (kexec_trylock()) {
131 		if (kexec_crash_image) {
132 			struct pt_regs fixed_regs;
133 
134 			crash_setup_regs(&fixed_regs, regs);
135 			crash_save_vmcoreinfo();
136 			machine_crash_shutdown(&fixed_regs);
137 			crash_cma_clear_pending_dma();
138 			machine_kexec(kexec_crash_image);
139 		}
140 		kexec_unlock();
141 	}
142 }
143 STACK_FRAME_NON_STANDARD(__crash_kexec);
144 
crash_kexec(struct pt_regs * regs)145 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
146 {
147 	if (panic_try_start()) {
148 		/* This is the 1st CPU which comes here, so go ahead. */
149 		__crash_kexec(regs);
150 
151 		/*
152 		 * Reset panic_cpu to allow another panic()/crash_kexec()
153 		 * call.
154 		 */
155 		panic_reset();
156 	}
157 }
158 
crash_resource_size(const struct resource * res)159 static inline resource_size_t crash_resource_size(const struct resource *res)
160 {
161 	return !res->end ? 0 : resource_size(res);
162 }
163 
164 
165 
166 
crash_prepare_elf64_headers(struct crash_mem * mem,int need_kernel_map,void ** addr,unsigned long * sz)167 int crash_prepare_elf64_headers(struct crash_mem *mem, int need_kernel_map,
168 			  void **addr, unsigned long *sz)
169 {
170 	Elf64_Ehdr *ehdr;
171 	Elf64_Phdr *phdr;
172 	unsigned long nr_cpus = num_possible_cpus(), nr_phdr, elf_sz;
173 	unsigned char *buf;
174 	unsigned int cpu, i;
175 	unsigned long long notes_addr;
176 	unsigned long mstart, mend;
177 
178 	/* extra phdr for vmcoreinfo ELF note */
179 	nr_phdr = nr_cpus + 1;
180 	nr_phdr += mem->nr_ranges;
181 
182 	/*
183 	 * kexec-tools creates an extra PT_LOAD phdr for kernel text mapping
184 	 * area (for example, ffffffff80000000 - ffffffffa0000000 on x86_64).
185 	 * I think this is required by tools like gdb. So same physical
186 	 * memory will be mapped in two ELF headers. One will contain kernel
187 	 * text virtual addresses and other will have __va(physical) addresses.
188 	 */
189 
190 	nr_phdr++;
191 	elf_sz = sizeof(Elf64_Ehdr) + nr_phdr * sizeof(Elf64_Phdr);
192 	elf_sz = ALIGN(elf_sz, ELF_CORE_HEADER_ALIGN);
193 
194 	buf = vzalloc(elf_sz);
195 	if (!buf)
196 		return -ENOMEM;
197 
198 	ehdr = (Elf64_Ehdr *)buf;
199 	phdr = (Elf64_Phdr *)(ehdr + 1);
200 	memcpy(ehdr->e_ident, ELFMAG, SELFMAG);
201 	ehdr->e_ident[EI_CLASS] = ELFCLASS64;
202 	ehdr->e_ident[EI_DATA] = ELFDATA2LSB;
203 	ehdr->e_ident[EI_VERSION] = EV_CURRENT;
204 	ehdr->e_ident[EI_OSABI] = ELF_OSABI;
205 	memset(ehdr->e_ident + EI_PAD, 0, EI_NIDENT - EI_PAD);
206 	ehdr->e_type = ET_CORE;
207 	ehdr->e_machine = ELF_ARCH;
208 	ehdr->e_version = EV_CURRENT;
209 	ehdr->e_phoff = sizeof(Elf64_Ehdr);
210 	ehdr->e_ehsize = sizeof(Elf64_Ehdr);
211 	ehdr->e_phentsize = sizeof(Elf64_Phdr);
212 
213 	/* Prepare one phdr of type PT_NOTE for each possible CPU */
214 	for_each_possible_cpu(cpu) {
215 		phdr->p_type = PT_NOTE;
216 		notes_addr = per_cpu_ptr_to_phys(per_cpu_ptr(crash_notes, cpu));
217 		phdr->p_offset = phdr->p_paddr = notes_addr;
218 		phdr->p_filesz = phdr->p_memsz = sizeof(note_buf_t);
219 		(ehdr->e_phnum)++;
220 		phdr++;
221 	}
222 
223 	/* Prepare one PT_NOTE header for vmcoreinfo */
224 	phdr->p_type = PT_NOTE;
225 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
226 	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
227 	(ehdr->e_phnum)++;
228 	phdr++;
229 
230 	/* Prepare PT_LOAD type program header for kernel text region */
231 	if (need_kernel_map) {
232 		phdr->p_type = PT_LOAD;
233 		phdr->p_flags = PF_R|PF_W|PF_X;
234 		phdr->p_vaddr = (unsigned long) _text;
235 		phdr->p_filesz = phdr->p_memsz = _end - _text;
236 		phdr->p_offset = phdr->p_paddr = __pa_symbol(_text);
237 		ehdr->e_phnum++;
238 		phdr++;
239 	}
240 
241 	/* Go through all the ranges in mem->ranges[] and prepare phdr */
242 	for (i = 0; i < mem->nr_ranges; i++) {
243 		mstart = mem->ranges[i].start;
244 		mend = mem->ranges[i].end;
245 
246 		phdr->p_type = PT_LOAD;
247 		phdr->p_flags = PF_R|PF_W|PF_X;
248 		phdr->p_offset  = mstart;
249 
250 		phdr->p_paddr = mstart;
251 		phdr->p_vaddr = (unsigned long) __va(mstart);
252 		phdr->p_filesz = phdr->p_memsz = mend - mstart + 1;
253 		phdr->p_align = 0;
254 		ehdr->e_phnum++;
255 #ifdef CONFIG_KEXEC_FILE
256 		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",
257 			      phdr, phdr->p_vaddr, phdr->p_paddr, phdr->p_filesz,
258 			      ehdr->e_phnum, phdr->p_offset);
259 #endif
260 		phdr++;
261 	}
262 
263 	*addr = buf;
264 	*sz = elf_sz;
265 	return 0;
266 }
267 
268 /**
269  * crash_exclude_mem_range - exclude a mem range for existing ranges
270  * @mem: mem->range contains an array of ranges sorted in ascending order
271  * @mstart: the start of to-be-excluded range
272  * @mend: the start of to-be-excluded range
273  *
274  * If you are unsure if a range split will happen, to avoid function call
275  * failure because of -ENOMEM, always make sure
276  *    mem->max_nr_ranges == mem->nr_ranges + 1
277  * before calling the function each time.
278  *
279  * returns 0 if a memory range is excluded successfully
280  * return -ENOMEM if mem->ranges doesn't have space to hold split ranges
281  */
crash_exclude_mem_range(struct crash_mem * mem,unsigned long long mstart,unsigned long long mend)282 int crash_exclude_mem_range(struct crash_mem *mem,
283 			    unsigned long long mstart, unsigned long long mend)
284 {
285 	int i;
286 	unsigned long long start, end, p_start, p_end;
287 
288 	for (i = 0; i < mem->nr_ranges; i++) {
289 		start = mem->ranges[i].start;
290 		end = mem->ranges[i].end;
291 		p_start = mstart;
292 		p_end = mend;
293 
294 		if (p_start > end)
295 			continue;
296 
297 		/*
298 		 * Because the memory ranges in mem->ranges are stored in
299 		 * ascending order, when we detect `p_end < start`, we can
300 		 * immediately exit the for loop, as the subsequent memory
301 		 * ranges will definitely be outside the range we are looking
302 		 * for.
303 		 */
304 		if (p_end < start)
305 			break;
306 
307 		/* Truncate any area outside of range */
308 		if (p_start < start)
309 			p_start = start;
310 		if (p_end > end)
311 			p_end = end;
312 
313 		/* Found completely overlapping range */
314 		if (p_start == start && p_end == end) {
315 			memmove(&mem->ranges[i], &mem->ranges[i + 1],
316 				(mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
317 			i--;
318 			mem->nr_ranges--;
319 		} else if (p_start > start && p_end < end) {
320 			/* Split original range */
321 			if (mem->nr_ranges >= mem->max_nr_ranges)
322 				return -ENOMEM;
323 
324 			memmove(&mem->ranges[i + 2], &mem->ranges[i + 1],
325 				(mem->nr_ranges - (i + 1)) * sizeof(mem->ranges[i]));
326 
327 			mem->ranges[i].end = p_start - 1;
328 			mem->ranges[i + 1].start = p_end + 1;
329 			mem->ranges[i + 1].end = end;
330 
331 			i++;
332 			mem->nr_ranges++;
333 		} else if (p_start != start)
334 			mem->ranges[i].end = p_start - 1;
335 		else
336 			mem->ranges[i].start = p_end + 1;
337 	}
338 
339 	return 0;
340 }
341 EXPORT_SYMBOL_GPL(crash_exclude_mem_range);
342 
crash_get_memory_size(void)343 ssize_t crash_get_memory_size(void)
344 {
345 	ssize_t size = 0;
346 
347 	if (!kexec_trylock())
348 		return -EBUSY;
349 
350 	size += crash_resource_size(&crashk_res);
351 	size += crash_resource_size(&crashk_low_res);
352 
353 	kexec_unlock();
354 	return size;
355 }
356 
__crash_shrink_memory(struct resource * old_res,unsigned long new_size)357 static int __crash_shrink_memory(struct resource *old_res,
358 				 unsigned long new_size)
359 {
360 	struct resource *ram_res;
361 
362 	ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
363 	if (!ram_res)
364 		return -ENOMEM;
365 
366 	ram_res->start = old_res->start + new_size;
367 	ram_res->end   = old_res->end;
368 	ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
369 	ram_res->name  = "System RAM";
370 
371 	if (!new_size) {
372 		release_resource(old_res);
373 		old_res->start = 0;
374 		old_res->end   = 0;
375 	} else {
376 		crashk_res.end = ram_res->start - 1;
377 	}
378 
379 	crash_free_reserved_phys_range(ram_res->start, ram_res->end);
380 	insert_resource(&iomem_resource, ram_res);
381 
382 	return 0;
383 }
384 
crash_shrink_memory(unsigned long new_size)385 int crash_shrink_memory(unsigned long new_size)
386 {
387 	int ret = 0;
388 	unsigned long old_size, low_size;
389 
390 	if (!kexec_trylock())
391 		return -EBUSY;
392 
393 	if (kexec_crash_image) {
394 		ret = -ENOENT;
395 		goto unlock;
396 	}
397 
398 	low_size = crash_resource_size(&crashk_low_res);
399 	old_size = crash_resource_size(&crashk_res) + low_size;
400 	new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
401 	if (new_size >= old_size) {
402 		ret = (new_size == old_size) ? 0 : -EINVAL;
403 		goto unlock;
404 	}
405 
406 	/*
407 	 * (low_size > new_size) implies that low_size is greater than zero.
408 	 * This also means that if low_size is zero, the else branch is taken.
409 	 *
410 	 * If low_size is greater than 0, (low_size > new_size) indicates that
411 	 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
412 	 * needs to be shrunken.
413 	 */
414 	if (low_size > new_size) {
415 		ret = __crash_shrink_memory(&crashk_res, 0);
416 		if (ret)
417 			goto unlock;
418 
419 		ret = __crash_shrink_memory(&crashk_low_res, new_size);
420 	} else {
421 		ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
422 	}
423 
424 	/* Swap crashk_res and crashk_low_res if needed */
425 	if (!crashk_res.end && crashk_low_res.end) {
426 		crashk_res.start = crashk_low_res.start;
427 		crashk_res.end   = crashk_low_res.end;
428 		release_resource(&crashk_low_res);
429 		crashk_low_res.start = 0;
430 		crashk_low_res.end   = 0;
431 		insert_resource(&iomem_resource, &crashk_res);
432 	}
433 
434 unlock:
435 	kexec_unlock();
436 	return ret;
437 }
438 
crash_save_cpu(struct pt_regs * regs,int cpu)439 void crash_save_cpu(struct pt_regs *regs, int cpu)
440 {
441 	struct elf_prstatus prstatus;
442 	u32 *buf;
443 
444 	if ((cpu < 0) || (cpu >= nr_cpu_ids))
445 		return;
446 
447 	/* Using ELF notes here is opportunistic.
448 	 * I need a well defined structure format
449 	 * for the data I pass, and I need tags
450 	 * on the data to indicate what information I have
451 	 * squirrelled away.  ELF notes happen to provide
452 	 * all of that, so there is no need to invent something new.
453 	 */
454 	buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
455 	if (!buf)
456 		return;
457 	memset(&prstatus, 0, sizeof(prstatus));
458 	prstatus.common.pr_pid = current->pid;
459 	elf_core_copy_regs(&prstatus.pr_reg, regs);
460 	buf = append_elf_note(buf, NN_PRSTATUS, NT_PRSTATUS,
461 			      &prstatus, sizeof(prstatus));
462 	final_note(buf);
463 }
464 
465 
466 
crash_notes_memory_init(void)467 static int __init crash_notes_memory_init(void)
468 {
469 	/* Allocate memory for saving cpu registers. */
470 	size_t size, align;
471 
472 	/*
473 	 * crash_notes could be allocated across 2 vmalloc pages when percpu
474 	 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
475 	 * pages are also on 2 continuous physical pages. In this case the
476 	 * 2nd part of crash_notes in 2nd page could be lost since only the
477 	 * starting address and size of crash_notes are exported through sysfs.
478 	 * Here round up the size of crash_notes to the nearest power of two
479 	 * and pass it to __alloc_percpu as align value. This can make sure
480 	 * crash_notes is allocated inside one physical page.
481 	 */
482 	size = sizeof(note_buf_t);
483 	align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
484 
485 	/*
486 	 * Break compile if size is bigger than PAGE_SIZE since crash_notes
487 	 * definitely will be in 2 pages with that.
488 	 */
489 	BUILD_BUG_ON(size > PAGE_SIZE);
490 
491 	crash_notes = __alloc_percpu(size, align);
492 	if (!crash_notes) {
493 		pr_warn("Memory allocation for saving cpu register states failed\n");
494 		return -ENOMEM;
495 	}
496 	return 0;
497 }
498 subsys_initcall(crash_notes_memory_init);
499 
500 #endif /*CONFIG_CRASH_DUMP*/
501 
502 #ifdef CONFIG_CRASH_HOTPLUG
503 #undef pr_fmt
504 #define pr_fmt(fmt) "crash hp: " fmt
505 
506 /*
507  * Different than kexec/kdump loading/unloading/jumping/shrinking which
508  * usually rarely happen, there will be many crash hotplug events notified
509  * during one short period, e.g one memory board is hot added and memory
510  * regions are online. So mutex lock  __crash_hotplug_lock is used to
511  * serialize the crash hotplug handling specifically.
512  */
513 static DEFINE_MUTEX(__crash_hotplug_lock);
514 #define crash_hotplug_lock() mutex_lock(&__crash_hotplug_lock)
515 #define crash_hotplug_unlock() mutex_unlock(&__crash_hotplug_lock)
516 
517 /*
518  * This routine utilized when the crash_hotplug sysfs node is read.
519  * It reflects the kernel's ability/permission to update the kdump
520  * image directly.
521  */
crash_check_hotplug_support(void)522 int crash_check_hotplug_support(void)
523 {
524 	int rc = 0;
525 
526 	crash_hotplug_lock();
527 	/* Obtain lock while reading crash information */
528 	if (!kexec_trylock()) {
529 		if (!kexec_in_progress)
530 			pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
531 		crash_hotplug_unlock();
532 		return 0;
533 	}
534 	if (kexec_crash_image) {
535 		rc = kexec_crash_image->hotplug_support;
536 	}
537 	/* Release lock now that update complete */
538 	kexec_unlock();
539 	crash_hotplug_unlock();
540 
541 	return rc;
542 }
543 
544 /*
545  * To accurately reflect hot un/plug changes of CPU and Memory resources
546  * (including onling and offlining of those resources), the relevant
547  * kexec segments must be updated with latest CPU and Memory resources.
548  *
549  * Architectures must ensure two things for all segments that need
550  * updating during hotplug events:
551  *
552  * 1. Segments must be large enough to accommodate a growing number of
553  *    resources.
554  * 2. Exclude the segments from SHA verification.
555  *
556  * For example, on most architectures, the elfcorehdr (which is passed
557  * to the crash kernel via the elfcorehdr= parameter) must include the
558  * new list of CPUs and memory. To make changes to the elfcorehdr, it
559  * should be large enough to permit a growing number of CPU and Memory
560  * resources. One can estimate the elfcorehdr memory size based on
561  * NR_CPUS_DEFAULT and CRASH_MAX_MEMORY_RANGES. The elfcorehdr is
562  * excluded from SHA verification by default if the architecture
563  * supports crash hotplug.
564  */
crash_handle_hotplug_event(unsigned int hp_action,unsigned int cpu,void * arg)565 static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu, void *arg)
566 {
567 	struct kimage *image;
568 
569 	crash_hotplug_lock();
570 	/* Obtain lock while changing crash information */
571 	if (!kexec_trylock()) {
572 		if (!kexec_in_progress)
573 			pr_info("kexec_trylock() failed, kdump image may be inaccurate\n");
574 		crash_hotplug_unlock();
575 		return;
576 	}
577 
578 	/* Check kdump is not loaded */
579 	if (!kexec_crash_image)
580 		goto out;
581 
582 	image = kexec_crash_image;
583 
584 	/* Check that kexec segments update is permitted */
585 	if (!image->hotplug_support)
586 		goto out;
587 
588 	if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
589 		hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
590 		pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
591 	else
592 		pr_debug("hp_action %u\n", hp_action);
593 
594 	/*
595 	 * The elfcorehdr_index is set to -1 when the struct kimage
596 	 * is allocated. Find the segment containing the elfcorehdr,
597 	 * if not already found.
598 	 */
599 	if (image->elfcorehdr_index < 0) {
600 		unsigned long mem;
601 		unsigned char *ptr;
602 		unsigned int n;
603 
604 		for (n = 0; n < image->nr_segments; n++) {
605 			mem = image->segment[n].mem;
606 			ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
607 			if (ptr) {
608 				/* The segment containing elfcorehdr */
609 				if (memcmp(ptr, ELFMAG, SELFMAG) == 0)
610 					image->elfcorehdr_index = (int)n;
611 				kunmap_local(ptr);
612 			}
613 		}
614 	}
615 
616 	if (image->elfcorehdr_index < 0) {
617 		pr_err("unable to locate elfcorehdr segment");
618 		goto out;
619 	}
620 
621 	/* Needed in order for the segments to be updated */
622 	arch_kexec_unprotect_crashkres();
623 
624 	/* Differentiate between normal load and hotplug update */
625 	image->hp_action = hp_action;
626 
627 	/* Now invoke arch-specific update handler */
628 	arch_crash_handle_hotplug_event(image, arg);
629 
630 	/* No longer handling a hotplug event */
631 	image->hp_action = KEXEC_CRASH_HP_NONE;
632 	image->elfcorehdr_updated = true;
633 
634 	/* Change back to read-only */
635 	arch_kexec_protect_crashkres();
636 
637 	/* Errors in the callback is not a reason to rollback state */
638 out:
639 	/* Release lock now that update complete */
640 	kexec_unlock();
641 	crash_hotplug_unlock();
642 }
643 
crash_memhp_notifier(struct notifier_block * nb,unsigned long val,void * arg)644 static int crash_memhp_notifier(struct notifier_block *nb, unsigned long val, void *arg)
645 {
646 	switch (val) {
647 	case MEM_ONLINE:
648 		crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_MEMORY,
649 			KEXEC_CRASH_HP_INVALID_CPU, arg);
650 		break;
651 
652 	case MEM_OFFLINE:
653 		crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_MEMORY,
654 			KEXEC_CRASH_HP_INVALID_CPU, arg);
655 		break;
656 	}
657 	return NOTIFY_OK;
658 }
659 
660 static struct notifier_block crash_memhp_nb = {
661 	.notifier_call = crash_memhp_notifier,
662 	.priority = 0
663 };
664 
crash_cpuhp_online(unsigned int cpu)665 static int crash_cpuhp_online(unsigned int cpu)
666 {
667 	crash_handle_hotplug_event(KEXEC_CRASH_HP_ADD_CPU, cpu, NULL);
668 	return 0;
669 }
670 
crash_cpuhp_offline(unsigned int cpu)671 static int crash_cpuhp_offline(unsigned int cpu)
672 {
673 	crash_handle_hotplug_event(KEXEC_CRASH_HP_REMOVE_CPU, cpu, NULL);
674 	return 0;
675 }
676 
crash_hotplug_init(void)677 static int __init crash_hotplug_init(void)
678 {
679 	int result = 0;
680 
681 	if (IS_ENABLED(CONFIG_MEMORY_HOTPLUG))
682 		register_memory_notifier(&crash_memhp_nb);
683 
684 	if (IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
685 		result = cpuhp_setup_state_nocalls(CPUHP_BP_PREPARE_DYN,
686 			"crash/cpuhp", crash_cpuhp_online, crash_cpuhp_offline);
687 	}
688 
689 	return result;
690 }
691 
692 subsys_initcall(crash_hotplug_init);
693 #endif
694