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
9 #include <linux/capability.h>
11 #include <linux/file.h>
12 #include <linux/slab.h>
14 #include <linux/kexec.h>
15 #include <linux/mutex.h>
16 #include <linux/list.h>
17 #include <linux/highmem.h>
18 #include <linux/syscalls.h>
19 #include <linux/reboot.h>
20 #include <linux/ioport.h>
21 #include <linux/hardirq.h>
22 #include <linux/elf.h>
23 #include <linux/elfcore.h>
24 #include <linux/utsname.h>
25 #include <linux/numa.h>
26 #include <linux/suspend.h>
27 #include <linux/device.h>
28 #include <linux/freezer.h>
29 #include <linux/panic_notifier.h>
31 #include <linux/cpu.h>
32 #include <linux/uaccess.h>
34 #include <linux/console.h>
35 #include <linux/vmalloc.h>
36 #include <linux/swap.h>
37 #include <linux/syscore_ops.h>
38 #include <linux/compiler.h>
39 #include <linux/hugetlb.h>
40 #include <linux/objtool.h>
41 #include <linux/kmsg_dump.h>
44 #include <asm/sections.h>
46 #include <crypto/hash.h>
47 #include "kexec_internal.h"
49 DEFINE_MUTEX(kexec_mutex);
51 /* Per cpu memory for storing cpu states in case of system crash. */
52 note_buf_t __percpu *crash_notes;
54 /* Flag to indicate we are going to kexec a new kernel */
55 bool kexec_in_progress = false;
58 /* Location of the reserved area for the crash kernel */
59 struct resource crashk_res = {
60 .name = "Crash kernel",
63 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
64 .desc = IORES_DESC_CRASH_KERNEL
66 struct resource crashk_low_res = {
67 .name = "Crash kernel",
70 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
71 .desc = IORES_DESC_CRASH_KERNEL
74 int kexec_should_crash(struct task_struct *p)
77 * If crash_kexec_post_notifiers is enabled, don't run
78 * crash_kexec() here yet, which must be run after panic
79 * notifiers in panic().
81 if (crash_kexec_post_notifiers)
84 * There are 4 panic() calls in make_task_dead() path, each of which
85 * corresponds to each of these 4 conditions.
87 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
92 int kexec_crash_loaded(void)
94 return !!kexec_crash_image;
96 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
99 * When kexec transitions to the new kernel there is a one-to-one
100 * mapping between physical and virtual addresses. On processors
101 * where you can disable the MMU this is trivial, and easy. For
102 * others it is still a simple predictable page table to setup.
104 * In that environment kexec copies the new kernel to its final
105 * resting place. This means I can only support memory whose
106 * physical address can fit in an unsigned long. In particular
107 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
108 * If the assembly stub has more restrictive requirements
109 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
110 * defined more restrictively in <asm/kexec.h>.
112 * The code for the transition from the current kernel to the
113 * new kernel is placed in the control_code_buffer, whose size
114 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
115 * page of memory is necessary, but some architectures require more.
116 * Because this memory must be identity mapped in the transition from
117 * virtual to physical addresses it must live in the range
118 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
121 * The assembly stub in the control code buffer is passed a linked list
122 * of descriptor pages detailing the source pages of the new kernel,
123 * and the destination addresses of those source pages. As this data
124 * structure is not used in the context of the current OS, it must
127 * The code has been made to work with highmem pages and will use a
128 * destination page in its final resting place (if it happens
129 * to allocate it). The end product of this is that most of the
130 * physical address space, and most of RAM can be used.
132 * Future directions include:
133 * - allocating a page table with the control code buffer identity
134 * mapped, to simplify machine_kexec and make kexec_on_panic more
139 * KIMAGE_NO_DEST is an impossible destination address..., for
140 * allocating pages whose destination address we do not care about.
142 #define KIMAGE_NO_DEST (-1UL)
143 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
145 static struct page *kimage_alloc_page(struct kimage *image,
149 int sanity_check_segment_list(struct kimage *image)
152 unsigned long nr_segments = image->nr_segments;
153 unsigned long total_pages = 0;
154 unsigned long nr_pages = totalram_pages();
157 * Verify we have good destination addresses. The caller is
158 * responsible for making certain we don't attempt to load
159 * the new image into invalid or reserved areas of RAM. This
160 * just verifies it is an address we can use.
162 * Since the kernel does everything in page size chunks ensure
163 * the destination addresses are page aligned. Too many
164 * special cases crop of when we don't do this. The most
165 * insidious is getting overlapping destination addresses
166 * simply because addresses are changed to page size
169 for (i = 0; i < nr_segments; i++) {
170 unsigned long mstart, mend;
172 mstart = image->segment[i].mem;
173 mend = mstart + image->segment[i].memsz;
175 return -EADDRNOTAVAIL;
176 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
177 return -EADDRNOTAVAIL;
178 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
179 return -EADDRNOTAVAIL;
182 /* Verify our destination addresses do not overlap.
183 * If we alloed overlapping destination addresses
184 * through very weird things can happen with no
185 * easy explanation as one segment stops on another.
187 for (i = 0; i < nr_segments; i++) {
188 unsigned long mstart, mend;
191 mstart = image->segment[i].mem;
192 mend = mstart + image->segment[i].memsz;
193 for (j = 0; j < i; j++) {
194 unsigned long pstart, pend;
196 pstart = image->segment[j].mem;
197 pend = pstart + image->segment[j].memsz;
198 /* Do the segments overlap ? */
199 if ((mend > pstart) && (mstart < pend))
204 /* Ensure our buffer sizes are strictly less than
205 * our memory sizes. This should always be the case,
206 * and it is easier to check up front than to be surprised
209 for (i = 0; i < nr_segments; i++) {
210 if (image->segment[i].bufsz > image->segment[i].memsz)
215 * Verify that no more than half of memory will be consumed. If the
216 * request from userspace is too large, a large amount of time will be
217 * wasted allocating pages, which can cause a soft lockup.
219 for (i = 0; i < nr_segments; i++) {
220 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
223 total_pages += PAGE_COUNT(image->segment[i].memsz);
226 if (total_pages > nr_pages / 2)
230 * Verify we have good destination addresses. Normally
231 * the caller is responsible for making certain we don't
232 * attempt to load the new image into invalid or reserved
233 * areas of RAM. But crash kernels are preloaded into a
234 * reserved area of ram. We must ensure the addresses
235 * are in the reserved area otherwise preloading the
236 * kernel could corrupt things.
239 if (image->type == KEXEC_TYPE_CRASH) {
240 for (i = 0; i < nr_segments; i++) {
241 unsigned long mstart, mend;
243 mstart = image->segment[i].mem;
244 mend = mstart + image->segment[i].memsz - 1;
245 /* Ensure we are within the crash kernel limits */
246 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
247 (mend > phys_to_boot_phys(crashk_res.end)))
248 return -EADDRNOTAVAIL;
255 struct kimage *do_kimage_alloc_init(void)
257 struct kimage *image;
259 /* Allocate a controlling structure */
260 image = kzalloc(sizeof(*image), GFP_KERNEL);
265 image->entry = &image->head;
266 image->last_entry = &image->head;
267 image->control_page = ~0; /* By default this does not apply */
268 image->type = KEXEC_TYPE_DEFAULT;
270 /* Initialize the list of control pages */
271 INIT_LIST_HEAD(&image->control_pages);
273 /* Initialize the list of destination pages */
274 INIT_LIST_HEAD(&image->dest_pages);
276 /* Initialize the list of unusable pages */
277 INIT_LIST_HEAD(&image->unusable_pages);
282 int kimage_is_destination_range(struct kimage *image,
288 for (i = 0; i < image->nr_segments; i++) {
289 unsigned long mstart, mend;
291 mstart = image->segment[i].mem;
292 mend = mstart + image->segment[i].memsz;
293 if ((end > mstart) && (start < mend))
300 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
304 if (fatal_signal_pending(current))
306 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
308 unsigned int count, i;
310 pages->mapping = NULL;
311 set_page_private(pages, order);
313 for (i = 0; i < count; i++)
314 SetPageReserved(pages + i);
316 arch_kexec_post_alloc_pages(page_address(pages), count,
319 if (gfp_mask & __GFP_ZERO)
320 for (i = 0; i < count; i++)
321 clear_highpage(pages + i);
327 static void kimage_free_pages(struct page *page)
329 unsigned int order, count, i;
331 order = page_private(page);
334 arch_kexec_pre_free_pages(page_address(page), count);
336 for (i = 0; i < count; i++)
337 ClearPageReserved(page + i);
338 __free_pages(page, order);
341 void kimage_free_page_list(struct list_head *list)
343 struct page *page, *next;
345 list_for_each_entry_safe(page, next, list, lru) {
346 list_del(&page->lru);
347 kimage_free_pages(page);
351 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
354 /* Control pages are special, they are the intermediaries
355 * that are needed while we copy the rest of the pages
356 * to their final resting place. As such they must
357 * not conflict with either the destination addresses
358 * or memory the kernel is already using.
360 * The only case where we really need more than one of
361 * these are for architectures where we cannot disable
362 * the MMU and must instead generate an identity mapped
363 * page table for all of the memory.
365 * At worst this runs in O(N) of the image size.
367 struct list_head extra_pages;
372 INIT_LIST_HEAD(&extra_pages);
374 /* Loop while I can allocate a page and the page allocated
375 * is a destination page.
378 unsigned long pfn, epfn, addr, eaddr;
380 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
383 pfn = page_to_boot_pfn(pages);
385 addr = pfn << PAGE_SHIFT;
386 eaddr = epfn << PAGE_SHIFT;
387 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
388 kimage_is_destination_range(image, addr, eaddr)) {
389 list_add(&pages->lru, &extra_pages);
395 /* Remember the allocated page... */
396 list_add(&pages->lru, &image->control_pages);
398 /* Because the page is already in it's destination
399 * location we will never allocate another page at
400 * that address. Therefore kimage_alloc_pages
401 * will not return it (again) and we don't need
402 * to give it an entry in image->segment[].
405 /* Deal with the destination pages I have inadvertently allocated.
407 * Ideally I would convert multi-page allocations into single
408 * page allocations, and add everything to image->dest_pages.
410 * For now it is simpler to just free the pages.
412 kimage_free_page_list(&extra_pages);
417 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
420 /* Control pages are special, they are the intermediaries
421 * that are needed while we copy the rest of the pages
422 * to their final resting place. As such they must
423 * not conflict with either the destination addresses
424 * or memory the kernel is already using.
426 * Control pages are also the only pags we must allocate
427 * when loading a crash kernel. All of the other pages
428 * are specified by the segments and we just memcpy
429 * into them directly.
431 * The only case where we really need more than one of
432 * these are for architectures where we cannot disable
433 * the MMU and must instead generate an identity mapped
434 * page table for all of the memory.
436 * Given the low demand this implements a very simple
437 * allocator that finds the first hole of the appropriate
438 * size in the reserved memory region, and allocates all
439 * of the memory up to and including the hole.
441 unsigned long hole_start, hole_end, size;
445 size = (1 << order) << PAGE_SHIFT;
446 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
447 hole_end = hole_start + size - 1;
448 while (hole_end <= crashk_res.end) {
453 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
455 /* See if I overlap any of the segments */
456 for (i = 0; i < image->nr_segments; i++) {
457 unsigned long mstart, mend;
459 mstart = image->segment[i].mem;
460 mend = mstart + image->segment[i].memsz - 1;
461 if ((hole_end >= mstart) && (hole_start <= mend)) {
462 /* Advance the hole to the end of the segment */
463 hole_start = (mend + (size - 1)) & ~(size - 1);
464 hole_end = hole_start + size - 1;
468 /* If I don't overlap any segments I have found my hole! */
469 if (i == image->nr_segments) {
470 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
471 image->control_page = hole_end;
476 /* Ensure that these pages are decrypted if SME is enabled. */
478 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
484 struct page *kimage_alloc_control_pages(struct kimage *image,
487 struct page *pages = NULL;
489 switch (image->type) {
490 case KEXEC_TYPE_DEFAULT:
491 pages = kimage_alloc_normal_control_pages(image, order);
493 case KEXEC_TYPE_CRASH:
494 pages = kimage_alloc_crash_control_pages(image, order);
501 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
503 struct page *vmcoreinfo_page;
506 if (image->type != KEXEC_TYPE_CRASH)
510 * For kdump, allocate one vmcoreinfo safe copy from the
511 * crash memory. as we have arch_kexec_protect_crashkres()
512 * after kexec syscall, we naturally protect it from write
513 * (even read) access under kernel direct mapping. But on
514 * the other hand, we still need to operate it when crash
515 * happens to generate vmcoreinfo note, hereby we rely on
516 * vmap for this purpose.
518 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
519 if (!vmcoreinfo_page) {
520 pr_warn("Could not allocate vmcoreinfo buffer\n");
523 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
525 pr_warn("Could not vmap vmcoreinfo buffer\n");
529 image->vmcoreinfo_data_copy = safecopy;
530 crash_update_vmcoreinfo_safecopy(safecopy);
535 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
537 if (*image->entry != 0)
540 if (image->entry == image->last_entry) {
541 kimage_entry_t *ind_page;
544 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
548 ind_page = page_address(page);
549 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
550 image->entry = ind_page;
551 image->last_entry = ind_page +
552 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
554 *image->entry = entry;
561 static int kimage_set_destination(struct kimage *image,
562 unsigned long destination)
566 destination &= PAGE_MASK;
567 result = kimage_add_entry(image, destination | IND_DESTINATION);
573 static int kimage_add_page(struct kimage *image, unsigned long page)
578 result = kimage_add_entry(image, page | IND_SOURCE);
584 static void kimage_free_extra_pages(struct kimage *image)
586 /* Walk through and free any extra destination pages I may have */
587 kimage_free_page_list(&image->dest_pages);
589 /* Walk through and free any unusable pages I have cached */
590 kimage_free_page_list(&image->unusable_pages);
594 void kimage_terminate(struct kimage *image)
596 if (*image->entry != 0)
599 *image->entry = IND_DONE;
602 #define for_each_kimage_entry(image, ptr, entry) \
603 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
604 ptr = (entry & IND_INDIRECTION) ? \
605 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
607 static void kimage_free_entry(kimage_entry_t entry)
611 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
612 kimage_free_pages(page);
615 void kimage_free(struct kimage *image)
617 kimage_entry_t *ptr, entry;
618 kimage_entry_t ind = 0;
623 if (image->vmcoreinfo_data_copy) {
624 crash_update_vmcoreinfo_safecopy(NULL);
625 vunmap(image->vmcoreinfo_data_copy);
628 kimage_free_extra_pages(image);
629 for_each_kimage_entry(image, ptr, entry) {
630 if (entry & IND_INDIRECTION) {
631 /* Free the previous indirection page */
632 if (ind & IND_INDIRECTION)
633 kimage_free_entry(ind);
634 /* Save this indirection page until we are
638 } else if (entry & IND_SOURCE)
639 kimage_free_entry(entry);
641 /* Free the final indirection page */
642 if (ind & IND_INDIRECTION)
643 kimage_free_entry(ind);
645 /* Handle any machine specific cleanup */
646 machine_kexec_cleanup(image);
648 /* Free the kexec control pages... */
649 kimage_free_page_list(&image->control_pages);
652 * Free up any temporary buffers allocated. This might hit if
653 * error occurred much later after buffer allocation.
655 if (image->file_mode)
656 kimage_file_post_load_cleanup(image);
661 static kimage_entry_t *kimage_dst_used(struct kimage *image,
664 kimage_entry_t *ptr, entry;
665 unsigned long destination = 0;
667 for_each_kimage_entry(image, ptr, entry) {
668 if (entry & IND_DESTINATION)
669 destination = entry & PAGE_MASK;
670 else if (entry & IND_SOURCE) {
671 if (page == destination)
673 destination += PAGE_SIZE;
680 static struct page *kimage_alloc_page(struct kimage *image,
682 unsigned long destination)
685 * Here we implement safeguards to ensure that a source page
686 * is not copied to its destination page before the data on
687 * the destination page is no longer useful.
689 * To do this we maintain the invariant that a source page is
690 * either its own destination page, or it is not a
691 * destination page at all.
693 * That is slightly stronger than required, but the proof
694 * that no problems will not occur is trivial, and the
695 * implementation is simply to verify.
697 * When allocating all pages normally this algorithm will run
698 * in O(N) time, but in the worst case it will run in O(N^2)
699 * time. If the runtime is a problem the data structures can
706 * Walk through the list of destination pages, and see if I
709 list_for_each_entry(page, &image->dest_pages, lru) {
710 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
711 if (addr == destination) {
712 list_del(&page->lru);
720 /* Allocate a page, if we run out of memory give up */
721 page = kimage_alloc_pages(gfp_mask, 0);
724 /* If the page cannot be used file it away */
725 if (page_to_boot_pfn(page) >
726 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
727 list_add(&page->lru, &image->unusable_pages);
730 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
732 /* If it is the destination page we want use it */
733 if (addr == destination)
736 /* If the page is not a destination page use it */
737 if (!kimage_is_destination_range(image, addr,
742 * I know that the page is someones destination page.
743 * See if there is already a source page for this
744 * destination page. And if so swap the source pages.
746 old = kimage_dst_used(image, addr);
749 unsigned long old_addr;
750 struct page *old_page;
752 old_addr = *old & PAGE_MASK;
753 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
754 copy_highpage(page, old_page);
755 *old = addr | (*old & ~PAGE_MASK);
757 /* The old page I have found cannot be a
758 * destination page, so return it if it's
759 * gfp_flags honor the ones passed in.
761 if (!(gfp_mask & __GFP_HIGHMEM) &&
762 PageHighMem(old_page)) {
763 kimage_free_pages(old_page);
769 /* Place the page on the destination list, to be used later */
770 list_add(&page->lru, &image->dest_pages);
776 static int kimage_load_normal_segment(struct kimage *image,
777 struct kexec_segment *segment)
780 size_t ubytes, mbytes;
782 unsigned char __user *buf = NULL;
783 unsigned char *kbuf = NULL;
785 if (image->file_mode)
786 kbuf = segment->kbuf;
789 ubytes = segment->bufsz;
790 mbytes = segment->memsz;
791 maddr = segment->mem;
793 result = kimage_set_destination(image, maddr);
800 size_t uchunk, mchunk;
802 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
807 result = kimage_add_page(image, page_to_boot_pfn(page)
813 /* Start with a clear page */
815 ptr += maddr & ~PAGE_MASK;
816 mchunk = min_t(size_t, mbytes,
817 PAGE_SIZE - (maddr & ~PAGE_MASK));
818 uchunk = min(ubytes, mchunk);
820 /* For file based kexec, source pages are in kernel memory */
821 if (image->file_mode)
822 memcpy(ptr, kbuf, uchunk);
824 result = copy_from_user(ptr, buf, uchunk);
832 if (image->file_mode)
844 static int kimage_load_crash_segment(struct kimage *image,
845 struct kexec_segment *segment)
847 /* For crash dumps kernels we simply copy the data from
848 * user space to it's destination.
849 * We do things a page at a time for the sake of kmap.
852 size_t ubytes, mbytes;
854 unsigned char __user *buf = NULL;
855 unsigned char *kbuf = NULL;
858 if (image->file_mode)
859 kbuf = segment->kbuf;
862 ubytes = segment->bufsz;
863 mbytes = segment->memsz;
864 maddr = segment->mem;
868 size_t uchunk, mchunk;
870 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
875 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
877 ptr += maddr & ~PAGE_MASK;
878 mchunk = min_t(size_t, mbytes,
879 PAGE_SIZE - (maddr & ~PAGE_MASK));
880 uchunk = min(ubytes, mchunk);
881 if (mchunk > uchunk) {
882 /* Zero the trailing part of the page */
883 memset(ptr + uchunk, 0, mchunk - uchunk);
886 /* For file based kexec, source pages are in kernel memory */
887 if (image->file_mode)
888 memcpy(ptr, kbuf, uchunk);
890 result = copy_from_user(ptr, buf, uchunk);
891 kexec_flush_icache_page(page);
893 arch_kexec_pre_free_pages(page_address(page), 1);
900 if (image->file_mode)
912 int kimage_load_segment(struct kimage *image,
913 struct kexec_segment *segment)
915 int result = -ENOMEM;
917 switch (image->type) {
918 case KEXEC_TYPE_DEFAULT:
919 result = kimage_load_normal_segment(image, segment);
921 case KEXEC_TYPE_CRASH:
922 result = kimage_load_crash_segment(image, segment);
929 struct kimage *kexec_image;
930 struct kimage *kexec_crash_image;
931 int kexec_load_disabled;
933 static struct ctl_table kexec_core_sysctls[] = {
935 .procname = "kexec_load_disabled",
936 .data = &kexec_load_disabled,
937 .maxlen = sizeof(int),
939 /* only handle a transition from default "0" to "1" */
940 .proc_handler = proc_dointvec_minmax,
941 .extra1 = SYSCTL_ONE,
942 .extra2 = SYSCTL_ONE,
947 static int __init kexec_core_sysctl_init(void)
949 register_sysctl_init("kernel", kexec_core_sysctls);
952 late_initcall(kexec_core_sysctl_init);
956 * No panic_cpu check version of crash_kexec(). This function is called
957 * only when panic_cpu holds the current CPU number; this is the only CPU
958 * which processes crash_kexec routines.
960 void __noclone __crash_kexec(struct pt_regs *regs)
962 /* Take the kexec_mutex here to prevent sys_kexec_load
963 * running on one cpu from replacing the crash kernel
964 * we are using after a panic on a different cpu.
966 * If the crash kernel was not located in a fixed area
967 * of memory the xchg(&kexec_crash_image) would be
968 * sufficient. But since I reuse the memory...
970 if (mutex_trylock(&kexec_mutex)) {
971 if (kexec_crash_image) {
972 struct pt_regs fixed_regs;
974 crash_setup_regs(&fixed_regs, regs);
975 crash_save_vmcoreinfo();
976 machine_crash_shutdown(&fixed_regs);
977 machine_kexec(kexec_crash_image);
979 mutex_unlock(&kexec_mutex);
982 STACK_FRAME_NON_STANDARD(__crash_kexec);
984 void crash_kexec(struct pt_regs *regs)
986 int old_cpu, this_cpu;
989 * Only one CPU is allowed to execute the crash_kexec() code as with
990 * panic(). Otherwise parallel calls of panic() and crash_kexec()
991 * may stop each other. To exclude them, we use panic_cpu here too.
993 this_cpu = raw_smp_processor_id();
994 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
995 if (old_cpu == PANIC_CPU_INVALID) {
996 /* This is the 1st CPU which comes here, so go ahead. */
1000 * Reset panic_cpu to allow another panic()/crash_kexec()
1003 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
1007 size_t crash_get_memory_size(void)
1011 mutex_lock(&kexec_mutex);
1012 if (crashk_res.end != crashk_res.start)
1013 size = resource_size(&crashk_res);
1014 mutex_unlock(&kexec_mutex);
1018 int crash_shrink_memory(unsigned long new_size)
1021 unsigned long start, end;
1022 unsigned long old_size;
1023 struct resource *ram_res;
1025 mutex_lock(&kexec_mutex);
1027 if (kexec_crash_image) {
1031 start = crashk_res.start;
1032 end = crashk_res.end;
1033 old_size = (end == 0) ? 0 : end - start + 1;
1034 if (new_size >= old_size) {
1035 ret = (new_size == old_size) ? 0 : -EINVAL;
1039 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1045 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1046 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1048 crash_free_reserved_phys_range(end, crashk_res.end);
1050 if ((start == end) && (crashk_res.parent != NULL))
1051 release_resource(&crashk_res);
1053 ram_res->start = end;
1054 ram_res->end = crashk_res.end;
1055 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1056 ram_res->name = "System RAM";
1058 crashk_res.end = end - 1;
1060 insert_resource(&iomem_resource, ram_res);
1063 mutex_unlock(&kexec_mutex);
1067 void crash_save_cpu(struct pt_regs *regs, int cpu)
1069 struct elf_prstatus prstatus;
1072 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1075 /* Using ELF notes here is opportunistic.
1076 * I need a well defined structure format
1077 * for the data I pass, and I need tags
1078 * on the data to indicate what information I have
1079 * squirrelled away. ELF notes happen to provide
1080 * all of that, so there is no need to invent something new.
1082 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1085 memset(&prstatus, 0, sizeof(prstatus));
1086 prstatus.common.pr_pid = current->pid;
1087 elf_core_copy_regs(&prstatus.pr_reg, regs);
1088 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1089 &prstatus, sizeof(prstatus));
1093 static int __init crash_notes_memory_init(void)
1095 /* Allocate memory for saving cpu registers. */
1099 * crash_notes could be allocated across 2 vmalloc pages when percpu
1100 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1101 * pages are also on 2 continuous physical pages. In this case the
1102 * 2nd part of crash_notes in 2nd page could be lost since only the
1103 * starting address and size of crash_notes are exported through sysfs.
1104 * Here round up the size of crash_notes to the nearest power of two
1105 * and pass it to __alloc_percpu as align value. This can make sure
1106 * crash_notes is allocated inside one physical page.
1108 size = sizeof(note_buf_t);
1109 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1112 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1113 * definitely will be in 2 pages with that.
1115 BUILD_BUG_ON(size > PAGE_SIZE);
1117 crash_notes = __alloc_percpu(size, align);
1119 pr_warn("Memory allocation for saving cpu register states failed\n");
1124 subsys_initcall(crash_notes_memory_init);
1128 * Move into place and start executing a preloaded standalone
1129 * executable. If nothing was preloaded return an error.
1131 int kernel_kexec(void)
1135 if (!mutex_trylock(&kexec_mutex))
1142 #ifdef CONFIG_KEXEC_JUMP
1143 if (kexec_image->preserve_context) {
1144 pm_prepare_console();
1145 error = freeze_processes();
1148 goto Restore_console;
1151 error = dpm_suspend_start(PMSG_FREEZE);
1153 goto Resume_console;
1154 /* At this point, dpm_suspend_start() has been called,
1155 * but *not* dpm_suspend_end(). We *must* call
1156 * dpm_suspend_end() now. Otherwise, drivers for
1157 * some devices (e.g. interrupt controllers) become
1158 * desynchronized with the actual state of the
1159 * hardware at resume time, and evil weirdness ensues.
1161 error = dpm_suspend_end(PMSG_FREEZE);
1163 goto Resume_devices;
1164 error = suspend_disable_secondary_cpus();
1167 local_irq_disable();
1168 error = syscore_suspend();
1174 kexec_in_progress = true;
1175 kernel_restart_prepare("kexec reboot");
1176 migrate_to_reboot_cpu();
1179 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1180 * no further code needs to use CPU hotplug (which is true in
1181 * the reboot case). However, the kexec path depends on using
1182 * CPU hotplug again; so re-enable it here.
1184 cpu_hotplug_enable();
1185 pr_notice("Starting new kernel\n");
1189 kmsg_dump(KMSG_DUMP_SHUTDOWN);
1190 machine_kexec(kexec_image);
1192 #ifdef CONFIG_KEXEC_JUMP
1193 if (kexec_image->preserve_context) {
1198 suspend_enable_secondary_cpus();
1199 dpm_resume_start(PMSG_RESTORE);
1201 dpm_resume_end(PMSG_RESTORE);
1206 pm_restore_console();
1211 mutex_unlock(&kexec_mutex);