1 // SPDX-License-Identifier: GPL-2.0-only
3 * kexec.c - kexec system call core code.
4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/capability.h>
12 #include <linux/file.h>
13 #include <linux/slab.h>
15 #include <linux/kexec.h>
16 #include <linux/mutex.h>
17 #include <linux/list.h>
18 #include <linux/highmem.h>
19 #include <linux/syscalls.h>
20 #include <linux/reboot.h>
21 #include <linux/ioport.h>
22 #include <linux/hardirq.h>
23 #include <linux/elf.h>
24 #include <linux/elfcore.h>
25 #include <linux/utsname.h>
26 #include <linux/numa.h>
27 #include <linux/suspend.h>
28 #include <linux/device.h>
29 #include <linux/freezer.h>
30 #include <linux/panic_notifier.h>
32 #include <linux/cpu.h>
33 #include <linux/uaccess.h>
35 #include <linux/console.h>
36 #include <linux/vmalloc.h>
37 #include <linux/swap.h>
38 #include <linux/syscore_ops.h>
39 #include <linux/compiler.h>
40 #include <linux/hugetlb.h>
41 #include <linux/objtool.h>
42 #include <linux/kmsg_dump.h>
45 #include <asm/sections.h>
47 #include <crypto/hash.h>
48 #include "kexec_internal.h"
50 atomic_t __kexec_lock = ATOMIC_INIT(0);
52 /* Flag to indicate we are going to kexec a new kernel */
53 bool kexec_in_progress = false;
55 bool kexec_file_dbg_print;
57 int kexec_should_crash(struct task_struct *p)
60 * If crash_kexec_post_notifiers is enabled, don't run
61 * crash_kexec() here yet, which must be run after panic
62 * notifiers in panic().
64 if (crash_kexec_post_notifiers)
67 * There are 4 panic() calls in make_task_dead() path, each of which
68 * corresponds to each of these 4 conditions.
70 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
75 int kexec_crash_loaded(void)
77 return !!kexec_crash_image;
79 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
82 * When kexec transitions to the new kernel there is a one-to-one
83 * mapping between physical and virtual addresses. On processors
84 * where you can disable the MMU this is trivial, and easy. For
85 * others it is still a simple predictable page table to setup.
87 * In that environment kexec copies the new kernel to its final
88 * resting place. This means I can only support memory whose
89 * physical address can fit in an unsigned long. In particular
90 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
91 * If the assembly stub has more restrictive requirements
92 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
93 * defined more restrictively in <asm/kexec.h>.
95 * The code for the transition from the current kernel to the
96 * new kernel is placed in the control_code_buffer, whose size
97 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
98 * page of memory is necessary, but some architectures require more.
99 * Because this memory must be identity mapped in the transition from
100 * virtual to physical addresses it must live in the range
101 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
104 * The assembly stub in the control code buffer is passed a linked list
105 * of descriptor pages detailing the source pages of the new kernel,
106 * and the destination addresses of those source pages. As this data
107 * structure is not used in the context of the current OS, it must
110 * The code has been made to work with highmem pages and will use a
111 * destination page in its final resting place (if it happens
112 * to allocate it). The end product of this is that most of the
113 * physical address space, and most of RAM can be used.
115 * Future directions include:
116 * - allocating a page table with the control code buffer identity
117 * mapped, to simplify machine_kexec and make kexec_on_panic more
122 * KIMAGE_NO_DEST is an impossible destination address..., for
123 * allocating pages whose destination address we do not care about.
125 #define KIMAGE_NO_DEST (-1UL)
126 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
128 static struct page *kimage_alloc_page(struct kimage *image,
132 int sanity_check_segment_list(struct kimage *image)
135 unsigned long nr_segments = image->nr_segments;
136 unsigned long total_pages = 0;
137 unsigned long nr_pages = totalram_pages();
140 * Verify we have good destination addresses. The caller is
141 * responsible for making certain we don't attempt to load
142 * the new image into invalid or reserved areas of RAM. This
143 * just verifies it is an address we can use.
145 * Since the kernel does everything in page size chunks ensure
146 * the destination addresses are page aligned. Too many
147 * special cases crop of when we don't do this. The most
148 * insidious is getting overlapping destination addresses
149 * simply because addresses are changed to page size
152 for (i = 0; i < nr_segments; i++) {
153 unsigned long mstart, mend;
155 mstart = image->segment[i].mem;
156 mend = mstart + image->segment[i].memsz;
158 return -EADDRNOTAVAIL;
159 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
160 return -EADDRNOTAVAIL;
161 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
162 return -EADDRNOTAVAIL;
165 /* Verify our destination addresses do not overlap.
166 * If we alloed overlapping destination addresses
167 * through very weird things can happen with no
168 * easy explanation as one segment stops on another.
170 for (i = 0; i < nr_segments; i++) {
171 unsigned long mstart, mend;
174 mstart = image->segment[i].mem;
175 mend = mstart + image->segment[i].memsz;
176 for (j = 0; j < i; j++) {
177 unsigned long pstart, pend;
179 pstart = image->segment[j].mem;
180 pend = pstart + image->segment[j].memsz;
181 /* Do the segments overlap ? */
182 if ((mend > pstart) && (mstart < pend))
187 /* Ensure our buffer sizes are strictly less than
188 * our memory sizes. This should always be the case,
189 * and it is easier to check up front than to be surprised
192 for (i = 0; i < nr_segments; i++) {
193 if (image->segment[i].bufsz > image->segment[i].memsz)
198 * Verify that no more than half of memory will be consumed. If the
199 * request from userspace is too large, a large amount of time will be
200 * wasted allocating pages, which can cause a soft lockup.
202 for (i = 0; i < nr_segments; i++) {
203 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
206 total_pages += PAGE_COUNT(image->segment[i].memsz);
209 if (total_pages > nr_pages / 2)
213 * Verify we have good destination addresses. Normally
214 * the caller is responsible for making certain we don't
215 * attempt to load the new image into invalid or reserved
216 * areas of RAM. But crash kernels are preloaded into a
217 * reserved area of ram. We must ensure the addresses
218 * are in the reserved area otherwise preloading the
219 * kernel could corrupt things.
222 if (image->type == KEXEC_TYPE_CRASH) {
223 for (i = 0; i < nr_segments; i++) {
224 unsigned long mstart, mend;
226 mstart = image->segment[i].mem;
227 mend = mstart + image->segment[i].memsz - 1;
228 /* Ensure we are within the crash kernel limits */
229 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
230 (mend > phys_to_boot_phys(crashk_res.end)))
231 return -EADDRNOTAVAIL;
238 struct kimage *do_kimage_alloc_init(void)
240 struct kimage *image;
242 /* Allocate a controlling structure */
243 image = kzalloc(sizeof(*image), GFP_KERNEL);
248 image->entry = &image->head;
249 image->last_entry = &image->head;
250 image->control_page = ~0; /* By default this does not apply */
251 image->type = KEXEC_TYPE_DEFAULT;
253 /* Initialize the list of control pages */
254 INIT_LIST_HEAD(&image->control_pages);
256 /* Initialize the list of destination pages */
257 INIT_LIST_HEAD(&image->dest_pages);
259 /* Initialize the list of unusable pages */
260 INIT_LIST_HEAD(&image->unusable_pages);
262 #ifdef CONFIG_CRASH_HOTPLUG
263 image->hp_action = KEXEC_CRASH_HP_NONE;
264 image->elfcorehdr_index = -1;
265 image->elfcorehdr_updated = false;
271 int kimage_is_destination_range(struct kimage *image,
277 for (i = 0; i < image->nr_segments; i++) {
278 unsigned long mstart, mend;
280 mstart = image->segment[i].mem;
281 mend = mstart + image->segment[i].memsz - 1;
282 if ((end >= mstart) && (start <= mend))
289 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
293 if (fatal_signal_pending(current))
295 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
297 unsigned int count, i;
299 pages->mapping = NULL;
300 set_page_private(pages, order);
302 for (i = 0; i < count; i++)
303 SetPageReserved(pages + i);
305 arch_kexec_post_alloc_pages(page_address(pages), count,
308 if (gfp_mask & __GFP_ZERO)
309 for (i = 0; i < count; i++)
310 clear_highpage(pages + i);
316 static void kimage_free_pages(struct page *page)
318 unsigned int order, count, i;
320 order = page_private(page);
323 arch_kexec_pre_free_pages(page_address(page), count);
325 for (i = 0; i < count; i++)
326 ClearPageReserved(page + i);
327 __free_pages(page, order);
330 void kimage_free_page_list(struct list_head *list)
332 struct page *page, *next;
334 list_for_each_entry_safe(page, next, list, lru) {
335 list_del(&page->lru);
336 kimage_free_pages(page);
340 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
343 /* Control pages are special, they are the intermediaries
344 * that are needed while we copy the rest of the pages
345 * to their final resting place. As such they must
346 * not conflict with either the destination addresses
347 * or memory the kernel is already using.
349 * The only case where we really need more than one of
350 * these are for architectures where we cannot disable
351 * the MMU and must instead generate an identity mapped
352 * page table for all of the memory.
354 * At worst this runs in O(N) of the image size.
356 struct list_head extra_pages;
361 INIT_LIST_HEAD(&extra_pages);
363 /* Loop while I can allocate a page and the page allocated
364 * is a destination page.
367 unsigned long pfn, epfn, addr, eaddr;
369 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
372 pfn = page_to_boot_pfn(pages);
374 addr = pfn << PAGE_SHIFT;
375 eaddr = (epfn << PAGE_SHIFT) - 1;
376 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
377 kimage_is_destination_range(image, addr, eaddr)) {
378 list_add(&pages->lru, &extra_pages);
384 /* Remember the allocated page... */
385 list_add(&pages->lru, &image->control_pages);
387 /* Because the page is already in it's destination
388 * location we will never allocate another page at
389 * that address. Therefore kimage_alloc_pages
390 * will not return it (again) and we don't need
391 * to give it an entry in image->segment[].
394 /* Deal with the destination pages I have inadvertently allocated.
396 * Ideally I would convert multi-page allocations into single
397 * page allocations, and add everything to image->dest_pages.
399 * For now it is simpler to just free the pages.
401 kimage_free_page_list(&extra_pages);
406 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
409 /* Control pages are special, they are the intermediaries
410 * that are needed while we copy the rest of the pages
411 * to their final resting place. As such they must
412 * not conflict with either the destination addresses
413 * or memory the kernel is already using.
415 * Control pages are also the only pags we must allocate
416 * when loading a crash kernel. All of the other pages
417 * are specified by the segments and we just memcpy
418 * into them directly.
420 * The only case where we really need more than one of
421 * these are for architectures where we cannot disable
422 * the MMU and must instead generate an identity mapped
423 * page table for all of the memory.
425 * Given the low demand this implements a very simple
426 * allocator that finds the first hole of the appropriate
427 * size in the reserved memory region, and allocates all
428 * of the memory up to and including the hole.
430 unsigned long hole_start, hole_end, size;
434 size = (1 << order) << PAGE_SHIFT;
435 hole_start = ALIGN(image->control_page, size);
436 hole_end = hole_start + size - 1;
437 while (hole_end <= crashk_res.end) {
442 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
444 /* See if I overlap any of the segments */
445 for (i = 0; i < image->nr_segments; i++) {
446 unsigned long mstart, mend;
448 mstart = image->segment[i].mem;
449 mend = mstart + image->segment[i].memsz - 1;
450 if ((hole_end >= mstart) && (hole_start <= mend)) {
451 /* Advance the hole to the end of the segment */
452 hole_start = ALIGN(mend, size);
453 hole_end = hole_start + size - 1;
457 /* If I don't overlap any segments I have found my hole! */
458 if (i == image->nr_segments) {
459 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
460 image->control_page = hole_end + 1;
465 /* Ensure that these pages are decrypted if SME is enabled. */
467 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
473 struct page *kimage_alloc_control_pages(struct kimage *image,
476 struct page *pages = NULL;
478 switch (image->type) {
479 case KEXEC_TYPE_DEFAULT:
480 pages = kimage_alloc_normal_control_pages(image, order);
482 case KEXEC_TYPE_CRASH:
483 pages = kimage_alloc_crash_control_pages(image, order);
490 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
492 struct page *vmcoreinfo_page;
495 if (image->type != KEXEC_TYPE_CRASH)
499 * For kdump, allocate one vmcoreinfo safe copy from the
500 * crash memory. as we have arch_kexec_protect_crashkres()
501 * after kexec syscall, we naturally protect it from write
502 * (even read) access under kernel direct mapping. But on
503 * the other hand, we still need to operate it when crash
504 * happens to generate vmcoreinfo note, hereby we rely on
505 * vmap for this purpose.
507 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
508 if (!vmcoreinfo_page) {
509 pr_warn("Could not allocate vmcoreinfo buffer\n");
512 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
514 pr_warn("Could not vmap vmcoreinfo buffer\n");
518 image->vmcoreinfo_data_copy = safecopy;
519 crash_update_vmcoreinfo_safecopy(safecopy);
524 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
526 if (*image->entry != 0)
529 if (image->entry == image->last_entry) {
530 kimage_entry_t *ind_page;
533 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
537 ind_page = page_address(page);
538 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
539 image->entry = ind_page;
540 image->last_entry = ind_page +
541 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
543 *image->entry = entry;
550 static int kimage_set_destination(struct kimage *image,
551 unsigned long destination)
553 destination &= PAGE_MASK;
555 return kimage_add_entry(image, destination | IND_DESTINATION);
559 static int kimage_add_page(struct kimage *image, unsigned long page)
563 return kimage_add_entry(image, page | IND_SOURCE);
567 static void kimage_free_extra_pages(struct kimage *image)
569 /* Walk through and free any extra destination pages I may have */
570 kimage_free_page_list(&image->dest_pages);
572 /* Walk through and free any unusable pages I have cached */
573 kimage_free_page_list(&image->unusable_pages);
577 void kimage_terminate(struct kimage *image)
579 if (*image->entry != 0)
582 *image->entry = IND_DONE;
585 #define for_each_kimage_entry(image, ptr, entry) \
586 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
587 ptr = (entry & IND_INDIRECTION) ? \
588 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
590 static void kimage_free_entry(kimage_entry_t entry)
594 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
595 kimage_free_pages(page);
598 void kimage_free(struct kimage *image)
600 kimage_entry_t *ptr, entry;
601 kimage_entry_t ind = 0;
606 if (image->vmcoreinfo_data_copy) {
607 crash_update_vmcoreinfo_safecopy(NULL);
608 vunmap(image->vmcoreinfo_data_copy);
611 kimage_free_extra_pages(image);
612 for_each_kimage_entry(image, ptr, entry) {
613 if (entry & IND_INDIRECTION) {
614 /* Free the previous indirection page */
615 if (ind & IND_INDIRECTION)
616 kimage_free_entry(ind);
617 /* Save this indirection page until we are
621 } else if (entry & IND_SOURCE)
622 kimage_free_entry(entry);
624 /* Free the final indirection page */
625 if (ind & IND_INDIRECTION)
626 kimage_free_entry(ind);
628 /* Handle any machine specific cleanup */
629 machine_kexec_cleanup(image);
631 /* Free the kexec control pages... */
632 kimage_free_page_list(&image->control_pages);
635 * Free up any temporary buffers allocated. This might hit if
636 * error occurred much later after buffer allocation.
638 if (image->file_mode)
639 kimage_file_post_load_cleanup(image);
644 static kimage_entry_t *kimage_dst_used(struct kimage *image,
647 kimage_entry_t *ptr, entry;
648 unsigned long destination = 0;
650 for_each_kimage_entry(image, ptr, entry) {
651 if (entry & IND_DESTINATION)
652 destination = entry & PAGE_MASK;
653 else if (entry & IND_SOURCE) {
654 if (page == destination)
656 destination += PAGE_SIZE;
663 static struct page *kimage_alloc_page(struct kimage *image,
665 unsigned long destination)
668 * Here we implement safeguards to ensure that a source page
669 * is not copied to its destination page before the data on
670 * the destination page is no longer useful.
672 * To do this we maintain the invariant that a source page is
673 * either its own destination page, or it is not a
674 * destination page at all.
676 * That is slightly stronger than required, but the proof
677 * that no problems will not occur is trivial, and the
678 * implementation is simply to verify.
680 * When allocating all pages normally this algorithm will run
681 * in O(N) time, but in the worst case it will run in O(N^2)
682 * time. If the runtime is a problem the data structures can
689 * Walk through the list of destination pages, and see if I
692 list_for_each_entry(page, &image->dest_pages, lru) {
693 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
694 if (addr == destination) {
695 list_del(&page->lru);
703 /* Allocate a page, if we run out of memory give up */
704 page = kimage_alloc_pages(gfp_mask, 0);
707 /* If the page cannot be used file it away */
708 if (page_to_boot_pfn(page) >
709 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
710 list_add(&page->lru, &image->unusable_pages);
713 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
715 /* If it is the destination page we want use it */
716 if (addr == destination)
719 /* If the page is not a destination page use it */
720 if (!kimage_is_destination_range(image, addr,
721 addr + PAGE_SIZE - 1))
725 * I know that the page is someones destination page.
726 * See if there is already a source page for this
727 * destination page. And if so swap the source pages.
729 old = kimage_dst_used(image, addr);
732 unsigned long old_addr;
733 struct page *old_page;
735 old_addr = *old & PAGE_MASK;
736 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
737 copy_highpage(page, old_page);
738 *old = addr | (*old & ~PAGE_MASK);
740 /* The old page I have found cannot be a
741 * destination page, so return it if it's
742 * gfp_flags honor the ones passed in.
744 if (!(gfp_mask & __GFP_HIGHMEM) &&
745 PageHighMem(old_page)) {
746 kimage_free_pages(old_page);
752 /* Place the page on the destination list, to be used later */
753 list_add(&page->lru, &image->dest_pages);
759 static int kimage_load_normal_segment(struct kimage *image,
760 struct kexec_segment *segment)
763 size_t ubytes, mbytes;
765 unsigned char __user *buf = NULL;
766 unsigned char *kbuf = NULL;
768 if (image->file_mode)
769 kbuf = segment->kbuf;
772 ubytes = segment->bufsz;
773 mbytes = segment->memsz;
774 maddr = segment->mem;
776 result = kimage_set_destination(image, maddr);
783 size_t uchunk, mchunk;
785 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
790 result = kimage_add_page(image, page_to_boot_pfn(page)
795 ptr = kmap_local_page(page);
796 /* Start with a clear page */
798 ptr += maddr & ~PAGE_MASK;
799 mchunk = min_t(size_t, mbytes,
800 PAGE_SIZE - (maddr & ~PAGE_MASK));
801 uchunk = min(ubytes, mchunk);
803 /* For file based kexec, source pages are in kernel memory */
804 if (image->file_mode)
805 memcpy(ptr, kbuf, uchunk);
807 result = copy_from_user(ptr, buf, uchunk);
815 if (image->file_mode)
827 static int kimage_load_crash_segment(struct kimage *image,
828 struct kexec_segment *segment)
830 /* For crash dumps kernels we simply copy the data from
831 * user space to it's destination.
832 * We do things a page at a time for the sake of kmap.
835 size_t ubytes, mbytes;
837 unsigned char __user *buf = NULL;
838 unsigned char *kbuf = NULL;
841 if (image->file_mode)
842 kbuf = segment->kbuf;
845 ubytes = segment->bufsz;
846 mbytes = segment->memsz;
847 maddr = segment->mem;
851 size_t uchunk, mchunk;
853 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
858 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
859 ptr = kmap_local_page(page);
860 ptr += maddr & ~PAGE_MASK;
861 mchunk = min_t(size_t, mbytes,
862 PAGE_SIZE - (maddr & ~PAGE_MASK));
863 uchunk = min(ubytes, mchunk);
864 if (mchunk > uchunk) {
865 /* Zero the trailing part of the page */
866 memset(ptr + uchunk, 0, mchunk - uchunk);
869 /* For file based kexec, source pages are in kernel memory */
870 if (image->file_mode)
871 memcpy(ptr, kbuf, uchunk);
873 result = copy_from_user(ptr, buf, uchunk);
874 kexec_flush_icache_page(page);
876 arch_kexec_pre_free_pages(page_address(page), 1);
883 if (image->file_mode)
895 int kimage_load_segment(struct kimage *image,
896 struct kexec_segment *segment)
898 int result = -ENOMEM;
900 switch (image->type) {
901 case KEXEC_TYPE_DEFAULT:
902 result = kimage_load_normal_segment(image, segment);
904 case KEXEC_TYPE_CRASH:
905 result = kimage_load_crash_segment(image, segment);
912 struct kexec_load_limit {
913 /* Mutex protects the limit count. */
918 static struct kexec_load_limit load_limit_reboot = {
919 .mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
923 static struct kexec_load_limit load_limit_panic = {
924 .mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
928 struct kimage *kexec_image;
929 struct kimage *kexec_crash_image;
930 static int kexec_load_disabled;
933 static int kexec_limit_handler(struct ctl_table *table, int write,
934 void *buffer, size_t *lenp, loff_t *ppos)
936 struct kexec_load_limit *limit = table->data;
938 struct ctl_table tmp = {
940 .maxlen = sizeof(val),
946 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
953 mutex_lock(&limit->mutex);
954 if (limit->limit != -1 && val >= limit->limit)
958 mutex_unlock(&limit->mutex);
963 mutex_lock(&limit->mutex);
965 mutex_unlock(&limit->mutex);
967 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
970 static struct ctl_table kexec_core_sysctls[] = {
972 .procname = "kexec_load_disabled",
973 .data = &kexec_load_disabled,
974 .maxlen = sizeof(int),
976 /* only handle a transition from default "0" to "1" */
977 .proc_handler = proc_dointvec_minmax,
978 .extra1 = SYSCTL_ONE,
979 .extra2 = SYSCTL_ONE,
982 .procname = "kexec_load_limit_panic",
983 .data = &load_limit_panic,
985 .proc_handler = kexec_limit_handler,
988 .procname = "kexec_load_limit_reboot",
989 .data = &load_limit_reboot,
991 .proc_handler = kexec_limit_handler,
996 static int __init kexec_core_sysctl_init(void)
998 register_sysctl_init("kernel", kexec_core_sysctls);
1001 late_initcall(kexec_core_sysctl_init);
1004 bool kexec_load_permitted(int kexec_image_type)
1006 struct kexec_load_limit *limit;
1009 * Only the superuser can use the kexec syscall and if it has not
1012 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1015 /* Check limit counter and decrease it.*/
1016 limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
1017 &load_limit_panic : &load_limit_reboot;
1018 mutex_lock(&limit->mutex);
1019 if (!limit->limit) {
1020 mutex_unlock(&limit->mutex);
1023 if (limit->limit != -1)
1025 mutex_unlock(&limit->mutex);
1031 * No panic_cpu check version of crash_kexec(). This function is called
1032 * only when panic_cpu holds the current CPU number; this is the only CPU
1033 * which processes crash_kexec routines.
1035 void __noclone __crash_kexec(struct pt_regs *regs)
1037 /* Take the kexec_lock here to prevent sys_kexec_load
1038 * running on one cpu from replacing the crash kernel
1039 * we are using after a panic on a different cpu.
1041 * If the crash kernel was not located in a fixed area
1042 * of memory the xchg(&kexec_crash_image) would be
1043 * sufficient. But since I reuse the memory...
1045 if (kexec_trylock()) {
1046 if (kexec_crash_image) {
1047 struct pt_regs fixed_regs;
1049 crash_setup_regs(&fixed_regs, regs);
1050 crash_save_vmcoreinfo();
1051 machine_crash_shutdown(&fixed_regs);
1052 machine_kexec(kexec_crash_image);
1057 STACK_FRAME_NON_STANDARD(__crash_kexec);
1059 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
1061 int old_cpu, this_cpu;
1064 * Only one CPU is allowed to execute the crash_kexec() code as with
1065 * panic(). Otherwise parallel calls of panic() and crash_kexec()
1066 * may stop each other. To exclude them, we use panic_cpu here too.
1068 old_cpu = PANIC_CPU_INVALID;
1069 this_cpu = raw_smp_processor_id();
1071 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
1072 /* This is the 1st CPU which comes here, so go ahead. */
1073 __crash_kexec(regs);
1076 * Reset panic_cpu to allow another panic()/crash_kexec()
1079 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
1083 static inline resource_size_t crash_resource_size(const struct resource *res)
1085 return !res->end ? 0 : resource_size(res);
1088 ssize_t crash_get_memory_size(void)
1092 if (!kexec_trylock())
1095 size += crash_resource_size(&crashk_res);
1096 size += crash_resource_size(&crashk_low_res);
1102 static int __crash_shrink_memory(struct resource *old_res,
1103 unsigned long new_size)
1105 struct resource *ram_res;
1107 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1111 ram_res->start = old_res->start + new_size;
1112 ram_res->end = old_res->end;
1113 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1114 ram_res->name = "System RAM";
1117 release_resource(old_res);
1121 crashk_res.end = ram_res->start - 1;
1124 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
1125 insert_resource(&iomem_resource, ram_res);
1130 int crash_shrink_memory(unsigned long new_size)
1133 unsigned long old_size, low_size;
1135 if (!kexec_trylock())
1138 if (kexec_crash_image) {
1143 low_size = crash_resource_size(&crashk_low_res);
1144 old_size = crash_resource_size(&crashk_res) + low_size;
1145 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
1146 if (new_size >= old_size) {
1147 ret = (new_size == old_size) ? 0 : -EINVAL;
1152 * (low_size > new_size) implies that low_size is greater than zero.
1153 * This also means that if low_size is zero, the else branch is taken.
1155 * If low_size is greater than 0, (low_size > new_size) indicates that
1156 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
1157 * needs to be shrunken.
1159 if (low_size > new_size) {
1160 ret = __crash_shrink_memory(&crashk_res, 0);
1164 ret = __crash_shrink_memory(&crashk_low_res, new_size);
1166 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
1169 /* Swap crashk_res and crashk_low_res if needed */
1170 if (!crashk_res.end && crashk_low_res.end) {
1171 crashk_res.start = crashk_low_res.start;
1172 crashk_res.end = crashk_low_res.end;
1173 release_resource(&crashk_low_res);
1174 crashk_low_res.start = 0;
1175 crashk_low_res.end = 0;
1176 insert_resource(&iomem_resource, &crashk_res);
1184 void crash_save_cpu(struct pt_regs *regs, int cpu)
1186 struct elf_prstatus prstatus;
1189 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1192 /* Using ELF notes here is opportunistic.
1193 * I need a well defined structure format
1194 * for the data I pass, and I need tags
1195 * on the data to indicate what information I have
1196 * squirrelled away. ELF notes happen to provide
1197 * all of that, so there is no need to invent something new.
1199 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1202 memset(&prstatus, 0, sizeof(prstatus));
1203 prstatus.common.pr_pid = current->pid;
1204 elf_core_copy_regs(&prstatus.pr_reg, regs);
1205 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1206 &prstatus, sizeof(prstatus));
1211 * Move into place and start executing a preloaded standalone
1212 * executable. If nothing was preloaded return an error.
1214 int kernel_kexec(void)
1218 if (!kexec_trylock())
1225 #ifdef CONFIG_KEXEC_JUMP
1226 if (kexec_image->preserve_context) {
1227 pm_prepare_console();
1228 error = freeze_processes();
1231 goto Restore_console;
1234 error = dpm_suspend_start(PMSG_FREEZE);
1236 goto Resume_console;
1237 /* At this point, dpm_suspend_start() has been called,
1238 * but *not* dpm_suspend_end(). We *must* call
1239 * dpm_suspend_end() now. Otherwise, drivers for
1240 * some devices (e.g. interrupt controllers) become
1241 * desynchronized with the actual state of the
1242 * hardware at resume time, and evil weirdness ensues.
1244 error = dpm_suspend_end(PMSG_FREEZE);
1246 goto Resume_devices;
1247 error = suspend_disable_secondary_cpus();
1250 local_irq_disable();
1251 error = syscore_suspend();
1257 kexec_in_progress = true;
1258 kernel_restart_prepare("kexec reboot");
1259 migrate_to_reboot_cpu();
1263 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1264 * no further code needs to use CPU hotplug (which is true in
1265 * the reboot case). However, the kexec path depends on using
1266 * CPU hotplug again; so re-enable it here.
1268 cpu_hotplug_enable();
1269 pr_notice("Starting new kernel\n");
1273 kmsg_dump(KMSG_DUMP_SHUTDOWN);
1274 machine_kexec(kexec_image);
1276 #ifdef CONFIG_KEXEC_JUMP
1277 if (kexec_image->preserve_context) {
1282 suspend_enable_secondary_cpus();
1283 dpm_resume_start(PMSG_RESTORE);
1285 dpm_resume_end(PMSG_RESTORE);
1290 pm_restore_console();