2 * kexec.c - kexec system call
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
9 #define pr_fmt(fmt) "kexec: " fmt
11 #include <linux/capability.h>
13 #include <linux/file.h>
14 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/mutex.h>
18 #include <linux/list.h>
19 #include <linux/highmem.h>
20 #include <linux/syscalls.h>
21 #include <linux/reboot.h>
22 #include <linux/ioport.h>
23 #include <linux/hardirq.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/utsname.h>
27 #include <linux/numa.h>
28 #include <linux/suspend.h>
29 #include <linux/device.h>
30 #include <linux/freezer.h>
31 #include <linux/vmalloc.h>
33 #include <linux/cpu.h>
34 #include <linux/console.h>
35 #include <linux/swap.h>
36 #include <linux/syscore_ops.h>
37 #include <linux/compiler.h>
38 #include <linux/hugetlb.h>
41 #include <asm/uaccess.h>
43 #include <asm/sections.h>
45 #include <crypto/hash.h>
46 #include <crypto/sha.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 /* vmcoreinfo stuff */
55 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
56 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
57 size_t vmcoreinfo_size;
58 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
60 /* Flag to indicate we are going to kexec a new kernel */
61 bool kexec_in_progress = false;
64 /* Location of the reserved area for the crash kernel */
65 struct resource crashk_res = {
66 .name = "Crash kernel",
69 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
71 struct resource crashk_low_res = {
72 .name = "Crash kernel",
75 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
78 int kexec_should_crash(struct task_struct *p)
81 * If crash_kexec_post_notifiers is enabled, don't run
82 * crash_kexec() here yet, which must be run after panic
83 * notifiers in panic().
85 if (crash_kexec_post_notifiers)
88 * There are 4 panic() calls in do_exit() path, each of which
89 * corresponds to each of these 4 conditions.
91 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
97 * When kexec transitions to the new kernel there is a one-to-one
98 * mapping between physical and virtual addresses. On processors
99 * where you can disable the MMU this is trivial, and easy. For
100 * others it is still a simple predictable page table to setup.
102 * In that environment kexec copies the new kernel to its final
103 * resting place. This means I can only support memory whose
104 * physical address can fit in an unsigned long. In particular
105 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
106 * If the assembly stub has more restrictive requirements
107 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
108 * defined more restrictively in <asm/kexec.h>.
110 * The code for the transition from the current kernel to the
111 * the new kernel is placed in the control_code_buffer, whose size
112 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
113 * page of memory is necessary, but some architectures require more.
114 * Because this memory must be identity mapped in the transition from
115 * virtual to physical addresses it must live in the range
116 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
119 * The assembly stub in the control code buffer is passed a linked list
120 * of descriptor pages detailing the source pages of the new kernel,
121 * and the destination addresses of those source pages. As this data
122 * structure is not used in the context of the current OS, it must
125 * The code has been made to work with highmem pages and will use a
126 * destination page in its final resting place (if it happens
127 * to allocate it). The end product of this is that most of the
128 * physical address space, and most of RAM can be used.
130 * Future directions include:
131 * - allocating a page table with the control code buffer identity
132 * mapped, to simplify machine_kexec and make kexec_on_panic more
137 * KIMAGE_NO_DEST is an impossible destination address..., for
138 * allocating pages whose destination address we do not care about.
140 #define KIMAGE_NO_DEST (-1UL)
142 static struct page *kimage_alloc_page(struct kimage *image,
146 static int copy_user_segment_list(struct kimage *image,
147 unsigned long nr_segments,
148 struct kexec_segment __user *segments)
151 size_t segment_bytes;
153 /* Read in the segments */
154 image->nr_segments = nr_segments;
155 segment_bytes = nr_segments * sizeof(*segments);
156 ret = copy_from_user(image->segment, segments, segment_bytes);
163 int sanity_check_segment_list(struct kimage *image)
166 unsigned long nr_segments = image->nr_segments;
169 * Verify we have good destination addresses. The caller is
170 * responsible for making certain we don't attempt to load
171 * the new image into invalid or reserved areas of RAM. This
172 * just verifies it is an address we can use.
174 * Since the kernel does everything in page size chunks ensure
175 * the destination addresses are page aligned. Too many
176 * special cases crop of when we don't do this. The most
177 * insidious is getting overlapping destination addresses
178 * simply because addresses are changed to page size
181 result = -EADDRNOTAVAIL;
182 for (i = 0; i < nr_segments; i++) {
183 unsigned long mstart, mend;
185 mstart = image->segment[i].mem;
186 mend = mstart + image->segment[i].memsz;
187 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
189 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
193 /* Verify our destination addresses do not overlap.
194 * If we alloed overlapping destination addresses
195 * through very weird things can happen with no
196 * easy explanation as one segment stops on another.
199 for (i = 0; i < nr_segments; i++) {
200 unsigned long mstart, mend;
203 mstart = image->segment[i].mem;
204 mend = mstart + image->segment[i].memsz;
205 for (j = 0; j < i; j++) {
206 unsigned long pstart, pend;
207 pstart = image->segment[j].mem;
208 pend = pstart + image->segment[j].memsz;
209 /* Do the segments overlap ? */
210 if ((mend > pstart) && (mstart < pend))
215 /* Ensure our buffer sizes are strictly less than
216 * our memory sizes. This should always be the case,
217 * and it is easier to check up front than to be surprised
221 for (i = 0; i < nr_segments; i++) {
222 if (image->segment[i].bufsz > image->segment[i].memsz)
227 * Verify we have good destination addresses. Normally
228 * the caller is responsible for making certain we don't
229 * attempt to load the new image into invalid or reserved
230 * areas of RAM. But crash kernels are preloaded into a
231 * reserved area of ram. We must ensure the addresses
232 * are in the reserved area otherwise preloading the
233 * kernel could corrupt things.
236 if (image->type == KEXEC_TYPE_CRASH) {
237 result = -EADDRNOTAVAIL;
238 for (i = 0; i < nr_segments; i++) {
239 unsigned long mstart, mend;
241 mstart = image->segment[i].mem;
242 mend = mstart + image->segment[i].memsz - 1;
243 /* Ensure we are within the crash kernel limits */
244 if ((mstart < crashk_res.start) ||
245 (mend > crashk_res.end))
253 struct kimage *do_kimage_alloc_init(void)
255 struct kimage *image;
257 /* Allocate a controlling structure */
258 image = kzalloc(sizeof(*image), GFP_KERNEL);
263 image->entry = &image->head;
264 image->last_entry = &image->head;
265 image->control_page = ~0; /* By default this does not apply */
266 image->type = KEXEC_TYPE_DEFAULT;
268 /* Initialize the list of control pages */
269 INIT_LIST_HEAD(&image->control_pages);
271 /* Initialize the list of destination pages */
272 INIT_LIST_HEAD(&image->dest_pages);
274 /* Initialize the list of unusable pages */
275 INIT_LIST_HEAD(&image->unusable_pages);
280 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
281 unsigned long nr_segments,
282 struct kexec_segment __user *segments,
286 struct kimage *image;
287 bool kexec_on_panic = flags & KEXEC_ON_CRASH;
289 if (kexec_on_panic) {
290 /* Verify we have a valid entry point */
291 if ((entry < crashk_res.start) || (entry > crashk_res.end))
292 return -EADDRNOTAVAIL;
295 /* Allocate and initialize a controlling structure */
296 image = do_kimage_alloc_init();
300 image->start = entry;
302 ret = copy_user_segment_list(image, nr_segments, segments);
306 ret = sanity_check_segment_list(image);
310 /* Enable the special crash kernel control page allocation policy. */
311 if (kexec_on_panic) {
312 image->control_page = crashk_res.start;
313 image->type = KEXEC_TYPE_CRASH;
317 * Find a location for the control code buffer, and add it
318 * the vector of segments so that it's pages will also be
319 * counted as destination pages.
322 image->control_code_page = kimage_alloc_control_pages(image,
323 get_order(KEXEC_CONTROL_PAGE_SIZE));
324 if (!image->control_code_page) {
325 pr_err("Could not allocate control_code_buffer\n");
329 if (!kexec_on_panic) {
330 image->swap_page = kimage_alloc_control_pages(image, 0);
331 if (!image->swap_page) {
332 pr_err("Could not allocate swap buffer\n");
333 goto out_free_control_pages;
339 out_free_control_pages:
340 kimage_free_page_list(&image->control_pages);
346 int kimage_is_destination_range(struct kimage *image,
352 for (i = 0; i < image->nr_segments; i++) {
353 unsigned long mstart, mend;
355 mstart = image->segment[i].mem;
356 mend = mstart + image->segment[i].memsz;
357 if ((end > mstart) && (start < mend))
364 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
368 pages = alloc_pages(gfp_mask, order);
370 unsigned int count, i;
371 pages->mapping = NULL;
372 set_page_private(pages, order);
374 for (i = 0; i < count; i++)
375 SetPageReserved(pages + i);
381 static void kimage_free_pages(struct page *page)
383 unsigned int order, count, i;
385 order = page_private(page);
387 for (i = 0; i < count; i++)
388 ClearPageReserved(page + i);
389 __free_pages(page, order);
392 void kimage_free_page_list(struct list_head *list)
394 struct list_head *pos, *next;
396 list_for_each_safe(pos, next, list) {
399 page = list_entry(pos, struct page, lru);
400 list_del(&page->lru);
401 kimage_free_pages(page);
405 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
408 /* Control pages are special, they are the intermediaries
409 * that are needed while we copy the rest of the pages
410 * to their final resting place. As such they must
411 * not conflict with either the destination addresses
412 * or memory the kernel is already using.
414 * The only case where we really need more than one of
415 * these are for architectures where we cannot disable
416 * the MMU and must instead generate an identity mapped
417 * page table for all of the memory.
419 * At worst this runs in O(N) of the image size.
421 struct list_head extra_pages;
426 INIT_LIST_HEAD(&extra_pages);
428 /* Loop while I can allocate a page and the page allocated
429 * is a destination page.
432 unsigned long pfn, epfn, addr, eaddr;
434 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
437 pfn = page_to_pfn(pages);
439 addr = pfn << PAGE_SHIFT;
440 eaddr = epfn << PAGE_SHIFT;
441 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
442 kimage_is_destination_range(image, addr, eaddr)) {
443 list_add(&pages->lru, &extra_pages);
449 /* Remember the allocated page... */
450 list_add(&pages->lru, &image->control_pages);
452 /* Because the page is already in it's destination
453 * location we will never allocate another page at
454 * that address. Therefore kimage_alloc_pages
455 * will not return it (again) and we don't need
456 * to give it an entry in image->segment[].
459 /* Deal with the destination pages I have inadvertently allocated.
461 * Ideally I would convert multi-page allocations into single
462 * page allocations, and add everything to image->dest_pages.
464 * For now it is simpler to just free the pages.
466 kimage_free_page_list(&extra_pages);
471 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
474 /* Control pages are special, they are the intermediaries
475 * that are needed while we copy the rest of the pages
476 * to their final resting place. As such they must
477 * not conflict with either the destination addresses
478 * or memory the kernel is already using.
480 * Control pages are also the only pags we must allocate
481 * when loading a crash kernel. All of the other pages
482 * are specified by the segments and we just memcpy
483 * into them directly.
485 * The only case where we really need more than one of
486 * these are for architectures where we cannot disable
487 * the MMU and must instead generate an identity mapped
488 * page table for all of the memory.
490 * Given the low demand this implements a very simple
491 * allocator that finds the first hole of the appropriate
492 * size in the reserved memory region, and allocates all
493 * of the memory up to and including the hole.
495 unsigned long hole_start, hole_end, size;
499 size = (1 << order) << PAGE_SHIFT;
500 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
501 hole_end = hole_start + size - 1;
502 while (hole_end <= crashk_res.end) {
505 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
507 /* See if I overlap any of the segments */
508 for (i = 0; i < image->nr_segments; i++) {
509 unsigned long mstart, mend;
511 mstart = image->segment[i].mem;
512 mend = mstart + image->segment[i].memsz - 1;
513 if ((hole_end >= mstart) && (hole_start <= mend)) {
514 /* Advance the hole to the end of the segment */
515 hole_start = (mend + (size - 1)) & ~(size - 1);
516 hole_end = hole_start + size - 1;
520 /* If I don't overlap any segments I have found my hole! */
521 if (i == image->nr_segments) {
522 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
527 image->control_page = hole_end;
533 struct page *kimage_alloc_control_pages(struct kimage *image,
536 struct page *pages = NULL;
538 switch (image->type) {
539 case KEXEC_TYPE_DEFAULT:
540 pages = kimage_alloc_normal_control_pages(image, order);
542 case KEXEC_TYPE_CRASH:
543 pages = kimage_alloc_crash_control_pages(image, order);
550 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
552 if (*image->entry != 0)
555 if (image->entry == image->last_entry) {
556 kimage_entry_t *ind_page;
559 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
563 ind_page = page_address(page);
564 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
565 image->entry = ind_page;
566 image->last_entry = ind_page +
567 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
569 *image->entry = entry;
576 static int kimage_set_destination(struct kimage *image,
577 unsigned long destination)
581 destination &= PAGE_MASK;
582 result = kimage_add_entry(image, destination | IND_DESTINATION);
588 static int kimage_add_page(struct kimage *image, unsigned long page)
593 result = kimage_add_entry(image, page | IND_SOURCE);
599 static void kimage_free_extra_pages(struct kimage *image)
601 /* Walk through and free any extra destination pages I may have */
602 kimage_free_page_list(&image->dest_pages);
604 /* Walk through and free any unusable pages I have cached */
605 kimage_free_page_list(&image->unusable_pages);
608 void kimage_terminate(struct kimage *image)
610 if (*image->entry != 0)
613 *image->entry = IND_DONE;
616 #define for_each_kimage_entry(image, ptr, entry) \
617 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
618 ptr = (entry & IND_INDIRECTION) ? \
619 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
621 static void kimage_free_entry(kimage_entry_t entry)
625 page = pfn_to_page(entry >> PAGE_SHIFT);
626 kimage_free_pages(page);
629 void kimage_free(struct kimage *image)
631 kimage_entry_t *ptr, entry;
632 kimage_entry_t ind = 0;
637 kimage_free_extra_pages(image);
638 for_each_kimage_entry(image, ptr, entry) {
639 if (entry & IND_INDIRECTION) {
640 /* Free the previous indirection page */
641 if (ind & IND_INDIRECTION)
642 kimage_free_entry(ind);
643 /* Save this indirection page until we are
647 } else if (entry & IND_SOURCE)
648 kimage_free_entry(entry);
650 /* Free the final indirection page */
651 if (ind & IND_INDIRECTION)
652 kimage_free_entry(ind);
654 /* Handle any machine specific cleanup */
655 machine_kexec_cleanup(image);
657 /* Free the kexec control pages... */
658 kimage_free_page_list(&image->control_pages);
661 * Free up any temporary buffers allocated. This might hit if
662 * error occurred much later after buffer allocation.
664 if (image->file_mode)
665 kimage_file_post_load_cleanup(image);
670 static kimage_entry_t *kimage_dst_used(struct kimage *image,
673 kimage_entry_t *ptr, entry;
674 unsigned long destination = 0;
676 for_each_kimage_entry(image, ptr, entry) {
677 if (entry & IND_DESTINATION)
678 destination = entry & PAGE_MASK;
679 else if (entry & IND_SOURCE) {
680 if (page == destination)
682 destination += PAGE_SIZE;
689 static struct page *kimage_alloc_page(struct kimage *image,
691 unsigned long destination)
694 * Here we implement safeguards to ensure that a source page
695 * is not copied to its destination page before the data on
696 * the destination page is no longer useful.
698 * To do this we maintain the invariant that a source page is
699 * either its own destination page, or it is not a
700 * destination page at all.
702 * That is slightly stronger than required, but the proof
703 * that no problems will not occur is trivial, and the
704 * implementation is simply to verify.
706 * When allocating all pages normally this algorithm will run
707 * in O(N) time, but in the worst case it will run in O(N^2)
708 * time. If the runtime is a problem the data structures can
715 * Walk through the list of destination pages, and see if I
718 list_for_each_entry(page, &image->dest_pages, lru) {
719 addr = page_to_pfn(page) << PAGE_SHIFT;
720 if (addr == destination) {
721 list_del(&page->lru);
729 /* Allocate a page, if we run out of memory give up */
730 page = kimage_alloc_pages(gfp_mask, 0);
733 /* If the page cannot be used file it away */
734 if (page_to_pfn(page) >
735 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
736 list_add(&page->lru, &image->unusable_pages);
739 addr = page_to_pfn(page) << PAGE_SHIFT;
741 /* If it is the destination page we want use it */
742 if (addr == destination)
745 /* If the page is not a destination page use it */
746 if (!kimage_is_destination_range(image, addr,
751 * I know that the page is someones destination page.
752 * See if there is already a source page for this
753 * destination page. And if so swap the source pages.
755 old = kimage_dst_used(image, addr);
758 unsigned long old_addr;
759 struct page *old_page;
761 old_addr = *old & PAGE_MASK;
762 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
763 copy_highpage(page, old_page);
764 *old = addr | (*old & ~PAGE_MASK);
766 /* The old page I have found cannot be a
767 * destination page, so return it if it's
768 * gfp_flags honor the ones passed in.
770 if (!(gfp_mask & __GFP_HIGHMEM) &&
771 PageHighMem(old_page)) {
772 kimage_free_pages(old_page);
779 /* Place the page on the destination list I
782 list_add(&page->lru, &image->dest_pages);
789 static int kimage_load_normal_segment(struct kimage *image,
790 struct kexec_segment *segment)
793 size_t ubytes, mbytes;
795 unsigned char __user *buf = NULL;
796 unsigned char *kbuf = NULL;
799 if (image->file_mode)
800 kbuf = segment->kbuf;
803 ubytes = segment->bufsz;
804 mbytes = segment->memsz;
805 maddr = segment->mem;
807 result = kimage_set_destination(image, maddr);
814 size_t uchunk, mchunk;
816 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
821 result = kimage_add_page(image, page_to_pfn(page)
827 /* Start with a clear page */
829 ptr += maddr & ~PAGE_MASK;
830 mchunk = min_t(size_t, mbytes,
831 PAGE_SIZE - (maddr & ~PAGE_MASK));
832 uchunk = min(ubytes, mchunk);
834 /* For file based kexec, source pages are in kernel memory */
835 if (image->file_mode)
836 memcpy(ptr, kbuf, uchunk);
838 result = copy_from_user(ptr, buf, uchunk);
846 if (image->file_mode)
856 static int kimage_load_crash_segment(struct kimage *image,
857 struct kexec_segment *segment)
859 /* For crash dumps kernels we simply copy the data from
860 * user space to it's destination.
861 * We do things a page at a time for the sake of kmap.
864 size_t ubytes, mbytes;
866 unsigned char __user *buf = NULL;
867 unsigned char *kbuf = NULL;
870 if (image->file_mode)
871 kbuf = segment->kbuf;
874 ubytes = segment->bufsz;
875 mbytes = segment->memsz;
876 maddr = segment->mem;
880 size_t uchunk, mchunk;
882 page = pfn_to_page(maddr >> PAGE_SHIFT);
888 ptr += maddr & ~PAGE_MASK;
889 mchunk = min_t(size_t, mbytes,
890 PAGE_SIZE - (maddr & ~PAGE_MASK));
891 uchunk = min(ubytes, mchunk);
892 if (mchunk > uchunk) {
893 /* Zero the trailing part of the page */
894 memset(ptr + uchunk, 0, mchunk - uchunk);
897 /* For file based kexec, source pages are in kernel memory */
898 if (image->file_mode)
899 memcpy(ptr, kbuf, uchunk);
901 result = copy_from_user(ptr, buf, uchunk);
902 kexec_flush_icache_page(page);
910 if (image->file_mode)
920 int kimage_load_segment(struct kimage *image,
921 struct kexec_segment *segment)
923 int result = -ENOMEM;
925 switch (image->type) {
926 case KEXEC_TYPE_DEFAULT:
927 result = kimage_load_normal_segment(image, segment);
929 case KEXEC_TYPE_CRASH:
930 result = kimage_load_crash_segment(image, segment);
938 * Exec Kernel system call: for obvious reasons only root may call it.
940 * This call breaks up into three pieces.
941 * - A generic part which loads the new kernel from the current
942 * address space, and very carefully places the data in the
945 * - A generic part that interacts with the kernel and tells all of
946 * the devices to shut down. Preventing on-going dmas, and placing
947 * the devices in a consistent state so a later kernel can
950 * - A machine specific part that includes the syscall number
951 * and then copies the image to it's final destination. And
952 * jumps into the image at entry.
954 * kexec does not sync, or unmount filesystems so if you need
955 * that to happen you need to do that yourself.
957 struct kimage *kexec_image;
958 struct kimage *kexec_crash_image;
959 int kexec_load_disabled;
961 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
962 struct kexec_segment __user *, segments, unsigned long, flags)
964 struct kimage **dest_image, *image;
967 /* We only trust the superuser with rebooting the system. */
968 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
972 * Verify we have a legal set of flags
973 * This leaves us room for future extensions.
975 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
978 /* Verify we are on the appropriate architecture */
979 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
980 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
983 /* Put an artificial cap on the number
984 * of segments passed to kexec_load.
986 if (nr_segments > KEXEC_SEGMENT_MAX)
992 /* Because we write directly to the reserved memory
993 * region when loading crash kernels we need a mutex here to
994 * prevent multiple crash kernels from attempting to load
995 * simultaneously, and to prevent a crash kernel from loading
996 * over the top of a in use crash kernel.
998 * KISS: always take the mutex.
1000 if (!mutex_trylock(&kexec_mutex))
1003 dest_image = &kexec_image;
1004 if (flags & KEXEC_ON_CRASH)
1005 dest_image = &kexec_crash_image;
1006 if (nr_segments > 0) {
1009 if (flags & KEXEC_ON_CRASH) {
1011 * Loading another kernel to switch to if this one
1012 * crashes. Free any current crash dump kernel before
1016 kimage_free(xchg(&kexec_crash_image, NULL));
1017 result = kimage_alloc_init(&image, entry, nr_segments,
1019 crash_map_reserved_pages();
1021 /* Loading another kernel to reboot into. */
1023 result = kimage_alloc_init(&image, entry, nr_segments,
1029 if (flags & KEXEC_PRESERVE_CONTEXT)
1030 image->preserve_context = 1;
1031 result = machine_kexec_prepare(image);
1035 for (i = 0; i < nr_segments; i++) {
1036 result = kimage_load_segment(image, &image->segment[i]);
1040 kimage_terminate(image);
1041 if (flags & KEXEC_ON_CRASH)
1042 crash_unmap_reserved_pages();
1044 /* Install the new kernel, and Uninstall the old */
1045 image = xchg(dest_image, image);
1048 mutex_unlock(&kexec_mutex);
1055 * Add and remove page tables for crashkernel memory
1057 * Provide an empty default implementation here -- architecture
1058 * code may override this
1060 void __weak crash_map_reserved_pages(void)
1063 void __weak crash_unmap_reserved_pages(void)
1066 #ifdef CONFIG_COMPAT
1067 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1068 compat_ulong_t, nr_segments,
1069 struct compat_kexec_segment __user *, segments,
1070 compat_ulong_t, flags)
1072 struct compat_kexec_segment in;
1073 struct kexec_segment out, __user *ksegments;
1074 unsigned long i, result;
1076 /* Don't allow clients that don't understand the native
1077 * architecture to do anything.
1079 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1082 if (nr_segments > KEXEC_SEGMENT_MAX)
1085 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1086 for (i = 0; i < nr_segments; i++) {
1087 result = copy_from_user(&in, &segments[i], sizeof(in));
1091 out.buf = compat_ptr(in.buf);
1092 out.bufsz = in.bufsz;
1094 out.memsz = in.memsz;
1096 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1101 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1105 void crash_kexec(struct pt_regs *regs)
1107 /* Take the kexec_mutex here to prevent sys_kexec_load
1108 * running on one cpu from replacing the crash kernel
1109 * we are using after a panic on a different cpu.
1111 * If the crash kernel was not located in a fixed area
1112 * of memory the xchg(&kexec_crash_image) would be
1113 * sufficient. But since I reuse the memory...
1115 if (mutex_trylock(&kexec_mutex)) {
1116 if (kexec_crash_image) {
1117 struct pt_regs fixed_regs;
1119 crash_setup_regs(&fixed_regs, regs);
1120 crash_save_vmcoreinfo();
1121 machine_crash_shutdown(&fixed_regs);
1122 machine_kexec(kexec_crash_image);
1124 mutex_unlock(&kexec_mutex);
1128 size_t crash_get_memory_size(void)
1131 mutex_lock(&kexec_mutex);
1132 if (crashk_res.end != crashk_res.start)
1133 size = resource_size(&crashk_res);
1134 mutex_unlock(&kexec_mutex);
1138 void __weak crash_free_reserved_phys_range(unsigned long begin,
1143 for (addr = begin; addr < end; addr += PAGE_SIZE)
1144 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1147 int crash_shrink_memory(unsigned long new_size)
1150 unsigned long start, end;
1151 unsigned long old_size;
1152 struct resource *ram_res;
1154 mutex_lock(&kexec_mutex);
1156 if (kexec_crash_image) {
1160 start = crashk_res.start;
1161 end = crashk_res.end;
1162 old_size = (end == 0) ? 0 : end - start + 1;
1163 if (new_size >= old_size) {
1164 ret = (new_size == old_size) ? 0 : -EINVAL;
1168 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1174 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1175 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1177 crash_map_reserved_pages();
1178 crash_free_reserved_phys_range(end, crashk_res.end);
1180 if ((start == end) && (crashk_res.parent != NULL))
1181 release_resource(&crashk_res);
1183 ram_res->start = end;
1184 ram_res->end = crashk_res.end;
1185 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1186 ram_res->name = "System RAM";
1188 crashk_res.end = end - 1;
1190 insert_resource(&iomem_resource, ram_res);
1191 crash_unmap_reserved_pages();
1194 mutex_unlock(&kexec_mutex);
1198 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1201 struct elf_note note;
1203 note.n_namesz = strlen(name) + 1;
1204 note.n_descsz = data_len;
1206 memcpy(buf, ¬e, sizeof(note));
1207 buf += (sizeof(note) + 3)/4;
1208 memcpy(buf, name, note.n_namesz);
1209 buf += (note.n_namesz + 3)/4;
1210 memcpy(buf, data, note.n_descsz);
1211 buf += (note.n_descsz + 3)/4;
1216 static void final_note(u32 *buf)
1218 struct elf_note note;
1223 memcpy(buf, ¬e, sizeof(note));
1226 void crash_save_cpu(struct pt_regs *regs, int cpu)
1228 struct elf_prstatus prstatus;
1231 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1234 /* Using ELF notes here is opportunistic.
1235 * I need a well defined structure format
1236 * for the data I pass, and I need tags
1237 * on the data to indicate what information I have
1238 * squirrelled away. ELF notes happen to provide
1239 * all of that, so there is no need to invent something new.
1241 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1244 memset(&prstatus, 0, sizeof(prstatus));
1245 prstatus.pr_pid = current->pid;
1246 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1247 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1248 &prstatus, sizeof(prstatus));
1252 static int __init crash_notes_memory_init(void)
1254 /* Allocate memory for saving cpu registers. */
1255 crash_notes = alloc_percpu(note_buf_t);
1257 pr_warn("Kexec: Memory allocation for saving cpu register states failed\n");
1262 subsys_initcall(crash_notes_memory_init);
1266 * parsing the "crashkernel" commandline
1268 * this code is intended to be called from architecture specific code
1273 * This function parses command lines in the format
1275 * crashkernel=ramsize-range:size[,...][@offset]
1277 * The function returns 0 on success and -EINVAL on failure.
1279 static int __init parse_crashkernel_mem(char *cmdline,
1280 unsigned long long system_ram,
1281 unsigned long long *crash_size,
1282 unsigned long long *crash_base)
1284 char *cur = cmdline, *tmp;
1286 /* for each entry of the comma-separated list */
1288 unsigned long long start, end = ULLONG_MAX, size;
1290 /* get the start of the range */
1291 start = memparse(cur, &tmp);
1293 pr_warn("crashkernel: Memory value expected\n");
1298 pr_warn("crashkernel: '-' expected\n");
1303 /* if no ':' is here, than we read the end */
1305 end = memparse(cur, &tmp);
1307 pr_warn("crashkernel: Memory value expected\n");
1312 pr_warn("crashkernel: end <= start\n");
1318 pr_warn("crashkernel: ':' expected\n");
1323 size = memparse(cur, &tmp);
1325 pr_warn("Memory value expected\n");
1329 if (size >= system_ram) {
1330 pr_warn("crashkernel: invalid size\n");
1335 if (system_ram >= start && system_ram < end) {
1339 } while (*cur++ == ',');
1341 if (*crash_size > 0) {
1342 while (*cur && *cur != ' ' && *cur != '@')
1346 *crash_base = memparse(cur, &tmp);
1348 pr_warn("Memory value expected after '@'\n");
1358 * That function parses "simple" (old) crashkernel command lines like
1360 * crashkernel=size[@offset]
1362 * It returns 0 on success and -EINVAL on failure.
1364 static int __init parse_crashkernel_simple(char *cmdline,
1365 unsigned long long *crash_size,
1366 unsigned long long *crash_base)
1368 char *cur = cmdline;
1370 *crash_size = memparse(cmdline, &cur);
1371 if (cmdline == cur) {
1372 pr_warn("crashkernel: memory value expected\n");
1377 *crash_base = memparse(cur+1, &cur);
1378 else if (*cur != ' ' && *cur != '\0') {
1379 pr_warn("crashkernel: unrecognized char\n");
1386 #define SUFFIX_HIGH 0
1387 #define SUFFIX_LOW 1
1388 #define SUFFIX_NULL 2
1389 static __initdata char *suffix_tbl[] = {
1390 [SUFFIX_HIGH] = ",high",
1391 [SUFFIX_LOW] = ",low",
1392 [SUFFIX_NULL] = NULL,
1396 * That function parses "suffix" crashkernel command lines like
1398 * crashkernel=size,[high|low]
1400 * It returns 0 on success and -EINVAL on failure.
1402 static int __init parse_crashkernel_suffix(char *cmdline,
1403 unsigned long long *crash_size,
1406 char *cur = cmdline;
1408 *crash_size = memparse(cmdline, &cur);
1409 if (cmdline == cur) {
1410 pr_warn("crashkernel: memory value expected\n");
1414 /* check with suffix */
1415 if (strncmp(cur, suffix, strlen(suffix))) {
1416 pr_warn("crashkernel: unrecognized char\n");
1419 cur += strlen(suffix);
1420 if (*cur != ' ' && *cur != '\0') {
1421 pr_warn("crashkernel: unrecognized char\n");
1428 static __init char *get_last_crashkernel(char *cmdline,
1432 char *p = cmdline, *ck_cmdline = NULL;
1434 /* find crashkernel and use the last one if there are more */
1435 p = strstr(p, name);
1437 char *end_p = strchr(p, ' ');
1441 end_p = p + strlen(p);
1446 /* skip the one with any known suffix */
1447 for (i = 0; suffix_tbl[i]; i++) {
1448 q = end_p - strlen(suffix_tbl[i]);
1449 if (!strncmp(q, suffix_tbl[i],
1450 strlen(suffix_tbl[i])))
1455 q = end_p - strlen(suffix);
1456 if (!strncmp(q, suffix, strlen(suffix)))
1460 p = strstr(p+1, name);
1469 static int __init __parse_crashkernel(char *cmdline,
1470 unsigned long long system_ram,
1471 unsigned long long *crash_size,
1472 unsigned long long *crash_base,
1476 char *first_colon, *first_space;
1479 BUG_ON(!crash_size || !crash_base);
1483 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1488 ck_cmdline += strlen(name);
1491 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1494 * if the commandline contains a ':', then that's the extended
1495 * syntax -- if not, it must be the classic syntax
1497 first_colon = strchr(ck_cmdline, ':');
1498 first_space = strchr(ck_cmdline, ' ');
1499 if (first_colon && (!first_space || first_colon < first_space))
1500 return parse_crashkernel_mem(ck_cmdline, system_ram,
1501 crash_size, crash_base);
1503 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1507 * That function is the entry point for command line parsing and should be
1508 * called from the arch-specific code.
1510 int __init parse_crashkernel(char *cmdline,
1511 unsigned long long system_ram,
1512 unsigned long long *crash_size,
1513 unsigned long long *crash_base)
1515 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1516 "crashkernel=", NULL);
1519 int __init parse_crashkernel_high(char *cmdline,
1520 unsigned long long system_ram,
1521 unsigned long long *crash_size,
1522 unsigned long long *crash_base)
1524 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1525 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1528 int __init parse_crashkernel_low(char *cmdline,
1529 unsigned long long system_ram,
1530 unsigned long long *crash_size,
1531 unsigned long long *crash_base)
1533 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1534 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1537 static void update_vmcoreinfo_note(void)
1539 u32 *buf = vmcoreinfo_note;
1541 if (!vmcoreinfo_size)
1543 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1548 void crash_save_vmcoreinfo(void)
1550 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1551 update_vmcoreinfo_note();
1554 void vmcoreinfo_append_str(const char *fmt, ...)
1560 va_start(args, fmt);
1561 r = vscnprintf(buf, sizeof(buf), fmt, args);
1564 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1566 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1568 vmcoreinfo_size += r;
1572 * provide an empty default implementation here -- architecture
1573 * code may override this
1575 void __weak arch_crash_save_vmcoreinfo(void)
1578 unsigned long __weak paddr_vmcoreinfo_note(void)
1580 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1583 static int __init crash_save_vmcoreinfo_init(void)
1585 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1586 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1588 VMCOREINFO_SYMBOL(init_uts_ns);
1589 VMCOREINFO_SYMBOL(node_online_map);
1591 VMCOREINFO_SYMBOL(swapper_pg_dir);
1593 VMCOREINFO_SYMBOL(_stext);
1594 VMCOREINFO_SYMBOL(vmap_area_list);
1596 #ifndef CONFIG_NEED_MULTIPLE_NODES
1597 VMCOREINFO_SYMBOL(mem_map);
1598 VMCOREINFO_SYMBOL(contig_page_data);
1600 #ifdef CONFIG_SPARSEMEM
1601 VMCOREINFO_SYMBOL(mem_section);
1602 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1603 VMCOREINFO_STRUCT_SIZE(mem_section);
1604 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1606 VMCOREINFO_STRUCT_SIZE(page);
1607 VMCOREINFO_STRUCT_SIZE(pglist_data);
1608 VMCOREINFO_STRUCT_SIZE(zone);
1609 VMCOREINFO_STRUCT_SIZE(free_area);
1610 VMCOREINFO_STRUCT_SIZE(list_head);
1611 VMCOREINFO_SIZE(nodemask_t);
1612 VMCOREINFO_OFFSET(page, flags);
1613 VMCOREINFO_OFFSET(page, _count);
1614 VMCOREINFO_OFFSET(page, mapping);
1615 VMCOREINFO_OFFSET(page, lru);
1616 VMCOREINFO_OFFSET(page, _mapcount);
1617 VMCOREINFO_OFFSET(page, private);
1618 VMCOREINFO_OFFSET(pglist_data, node_zones);
1619 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1620 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1621 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1623 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1624 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1625 VMCOREINFO_OFFSET(pglist_data, node_id);
1626 VMCOREINFO_OFFSET(zone, free_area);
1627 VMCOREINFO_OFFSET(zone, vm_stat);
1628 VMCOREINFO_OFFSET(zone, spanned_pages);
1629 VMCOREINFO_OFFSET(free_area, free_list);
1630 VMCOREINFO_OFFSET(list_head, next);
1631 VMCOREINFO_OFFSET(list_head, prev);
1632 VMCOREINFO_OFFSET(vmap_area, va_start);
1633 VMCOREINFO_OFFSET(vmap_area, list);
1634 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1635 log_buf_kexec_setup();
1636 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1637 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1638 VMCOREINFO_NUMBER(PG_lru);
1639 VMCOREINFO_NUMBER(PG_private);
1640 VMCOREINFO_NUMBER(PG_swapcache);
1641 VMCOREINFO_NUMBER(PG_slab);
1642 #ifdef CONFIG_MEMORY_FAILURE
1643 VMCOREINFO_NUMBER(PG_hwpoison);
1645 VMCOREINFO_NUMBER(PG_head_mask);
1646 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1647 #ifdef CONFIG_HUGETLBFS
1648 VMCOREINFO_SYMBOL(free_huge_page);
1651 arch_crash_save_vmcoreinfo();
1652 update_vmcoreinfo_note();
1657 subsys_initcall(crash_save_vmcoreinfo_init);
1660 * Move into place and start executing a preloaded standalone
1661 * executable. If nothing was preloaded return an error.
1663 int kernel_kexec(void)
1667 if (!mutex_trylock(&kexec_mutex))
1674 #ifdef CONFIG_KEXEC_JUMP
1675 if (kexec_image->preserve_context) {
1676 lock_system_sleep();
1677 pm_prepare_console();
1678 error = freeze_processes();
1681 goto Restore_console;
1684 error = dpm_suspend_start(PMSG_FREEZE);
1686 goto Resume_console;
1687 /* At this point, dpm_suspend_start() has been called,
1688 * but *not* dpm_suspend_end(). We *must* call
1689 * dpm_suspend_end() now. Otherwise, drivers for
1690 * some devices (e.g. interrupt controllers) become
1691 * desynchronized with the actual state of the
1692 * hardware at resume time, and evil weirdness ensues.
1694 error = dpm_suspend_end(PMSG_FREEZE);
1696 goto Resume_devices;
1697 error = disable_nonboot_cpus();
1700 local_irq_disable();
1701 error = syscore_suspend();
1707 kexec_in_progress = true;
1708 kernel_restart_prepare(NULL);
1709 migrate_to_reboot_cpu();
1712 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1713 * no further code needs to use CPU hotplug (which is true in
1714 * the reboot case). However, the kexec path depends on using
1715 * CPU hotplug again; so re-enable it here.
1717 cpu_hotplug_enable();
1718 pr_emerg("Starting new kernel\n");
1722 machine_kexec(kexec_image);
1724 #ifdef CONFIG_KEXEC_JUMP
1725 if (kexec_image->preserve_context) {
1730 enable_nonboot_cpus();
1731 dpm_resume_start(PMSG_RESTORE);
1733 dpm_resume_end(PMSG_RESTORE);
1738 pm_restore_console();
1739 unlock_system_sleep();
1744 mutex_unlock(&kexec_mutex);