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>
32 #include <linux/cpu.h>
33 #include <linux/console.h>
34 #include <linux/vmalloc.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>
48 /* Per cpu memory for storing cpu states in case of system crash. */
49 note_buf_t __percpu *crash_notes;
51 /* vmcoreinfo stuff */
52 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
53 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
54 size_t vmcoreinfo_size;
55 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
57 /* Flag to indicate we are going to kexec a new kernel */
58 bool kexec_in_progress = false;
61 * Declare these symbols weak so that if architecture provides a purgatory,
62 * these will be overridden.
64 char __weak kexec_purgatory[0];
65 size_t __weak kexec_purgatory_size = 0;
67 static int kexec_calculate_store_digests(struct kimage *image);
69 /* Location of the reserved area for the crash kernel */
70 struct resource crashk_res = {
71 .name = "Crash kernel",
74 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
76 struct resource crashk_low_res = {
77 .name = "Crash kernel",
80 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
83 int kexec_should_crash(struct task_struct *p)
85 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
91 * When kexec transitions to the new kernel there is a one-to-one
92 * mapping between physical and virtual addresses. On processors
93 * where you can disable the MMU this is trivial, and easy. For
94 * others it is still a simple predictable page table to setup.
96 * In that environment kexec copies the new kernel to its final
97 * resting place. This means I can only support memory whose
98 * physical address can fit in an unsigned long. In particular
99 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
100 * If the assembly stub has more restrictive requirements
101 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
102 * defined more restrictively in <asm/kexec.h>.
104 * The code for the transition from the current kernel to the
105 * the new kernel is placed in the control_code_buffer, whose size
106 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
107 * page of memory is necessary, but some architectures require more.
108 * Because this memory must be identity mapped in the transition from
109 * virtual to physical addresses it must live in the range
110 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
113 * The assembly stub in the control code buffer is passed a linked list
114 * of descriptor pages detailing the source pages of the new kernel,
115 * and the destination addresses of those source pages. As this data
116 * structure is not used in the context of the current OS, it must
119 * The code has been made to work with highmem pages and will use a
120 * destination page in its final resting place (if it happens
121 * to allocate it). The end product of this is that most of the
122 * physical address space, and most of RAM can be used.
124 * Future directions include:
125 * - allocating a page table with the control code buffer identity
126 * mapped, to simplify machine_kexec and make kexec_on_panic more
131 * KIMAGE_NO_DEST is an impossible destination address..., for
132 * allocating pages whose destination address we do not care about.
134 #define KIMAGE_NO_DEST (-1UL)
136 static int kimage_is_destination_range(struct kimage *image,
137 unsigned long start, unsigned long end);
138 static struct page *kimage_alloc_page(struct kimage *image,
142 static int copy_user_segment_list(struct kimage *image,
143 unsigned long nr_segments,
144 struct kexec_segment __user *segments)
147 size_t segment_bytes;
149 /* Read in the segments */
150 image->nr_segments = nr_segments;
151 segment_bytes = nr_segments * sizeof(*segments);
152 ret = copy_from_user(image->segment, segments, segment_bytes);
159 static int sanity_check_segment_list(struct kimage *image)
162 unsigned long nr_segments = image->nr_segments;
165 * Verify we have good destination addresses. The caller is
166 * responsible for making certain we don't attempt to load
167 * the new image into invalid or reserved areas of RAM. This
168 * just verifies it is an address we can use.
170 * Since the kernel does everything in page size chunks ensure
171 * the destination addresses are page aligned. Too many
172 * special cases crop of when we don't do this. The most
173 * insidious is getting overlapping destination addresses
174 * simply because addresses are changed to page size
177 result = -EADDRNOTAVAIL;
178 for (i = 0; i < nr_segments; i++) {
179 unsigned long mstart, mend;
181 mstart = image->segment[i].mem;
182 mend = mstart + image->segment[i].memsz;
183 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
185 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
189 /* Verify our destination addresses do not overlap.
190 * If we alloed overlapping destination addresses
191 * through very weird things can happen with no
192 * easy explanation as one segment stops on another.
195 for (i = 0; i < nr_segments; i++) {
196 unsigned long mstart, mend;
199 mstart = image->segment[i].mem;
200 mend = mstart + image->segment[i].memsz;
201 for (j = 0; j < i; j++) {
202 unsigned long pstart, pend;
203 pstart = image->segment[j].mem;
204 pend = pstart + image->segment[j].memsz;
205 /* Do the segments overlap ? */
206 if ((mend > pstart) && (mstart < pend))
211 /* Ensure our buffer sizes are strictly less than
212 * our memory sizes. This should always be the case,
213 * and it is easier to check up front than to be surprised
217 for (i = 0; i < nr_segments; i++) {
218 if (image->segment[i].bufsz > image->segment[i].memsz)
223 * Verify we have good destination addresses. Normally
224 * the caller is responsible for making certain we don't
225 * attempt to load the new image into invalid or reserved
226 * areas of RAM. But crash kernels are preloaded into a
227 * reserved area of ram. We must ensure the addresses
228 * are in the reserved area otherwise preloading the
229 * kernel could corrupt things.
232 if (image->type == KEXEC_TYPE_CRASH) {
233 result = -EADDRNOTAVAIL;
234 for (i = 0; i < nr_segments; i++) {
235 unsigned long mstart, mend;
237 mstart = image->segment[i].mem;
238 mend = mstart + image->segment[i].memsz - 1;
239 /* Ensure we are within the crash kernel limits */
240 if ((mstart < crashk_res.start) ||
241 (mend > crashk_res.end))
249 static struct kimage *do_kimage_alloc_init(void)
251 struct kimage *image;
253 /* Allocate a controlling structure */
254 image = kzalloc(sizeof(*image), GFP_KERNEL);
259 image->entry = &image->head;
260 image->last_entry = &image->head;
261 image->control_page = ~0; /* By default this does not apply */
262 image->type = KEXEC_TYPE_DEFAULT;
264 /* Initialize the list of control pages */
265 INIT_LIST_HEAD(&image->control_pages);
267 /* Initialize the list of destination pages */
268 INIT_LIST_HEAD(&image->dest_pages);
270 /* Initialize the list of unusable pages */
271 INIT_LIST_HEAD(&image->unusable_pages);
276 static void kimage_free_page_list(struct list_head *list);
278 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
279 unsigned long nr_segments,
280 struct kexec_segment __user *segments,
284 struct kimage *image;
285 bool kexec_on_panic = flags & KEXEC_ON_CRASH;
287 if (kexec_on_panic) {
288 /* Verify we have a valid entry point */
289 if ((entry < crashk_res.start) || (entry > crashk_res.end))
290 return -EADDRNOTAVAIL;
293 /* Allocate and initialize a controlling structure */
294 image = do_kimage_alloc_init();
298 image->start = entry;
300 ret = copy_user_segment_list(image, nr_segments, segments);
304 ret = sanity_check_segment_list(image);
308 /* Enable the special crash kernel control page allocation policy. */
309 if (kexec_on_panic) {
310 image->control_page = crashk_res.start;
311 image->type = KEXEC_TYPE_CRASH;
315 * Find a location for the control code buffer, and add it
316 * the vector of segments so that it's pages will also be
317 * counted as destination pages.
320 image->control_code_page = kimage_alloc_control_pages(image,
321 get_order(KEXEC_CONTROL_PAGE_SIZE));
322 if (!image->control_code_page) {
323 pr_err("Could not allocate control_code_buffer\n");
327 if (!kexec_on_panic) {
328 image->swap_page = kimage_alloc_control_pages(image, 0);
329 if (!image->swap_page) {
330 pr_err("Could not allocate swap buffer\n");
331 goto out_free_control_pages;
337 out_free_control_pages:
338 kimage_free_page_list(&image->control_pages);
344 static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len)
346 struct fd f = fdget(fd);
355 ret = vfs_getattr(&f.file->f_path, &stat);
359 if (stat.size > INT_MAX) {
364 /* Don't hand 0 to vmalloc, it whines. */
365 if (stat.size == 0) {
370 *buf = vmalloc(stat.size);
377 while (pos < stat.size) {
378 bytes = kernel_read(f.file, pos, (char *)(*buf) + pos,
391 if (pos != stat.size) {
403 /* Architectures can provide this probe function */
404 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
405 unsigned long buf_len)
410 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
412 return ERR_PTR(-ENOEXEC);
415 void __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
419 /* Apply relocations of type RELA */
421 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
424 pr_err("RELA relocation unsupported.\n");
428 /* Apply relocations of type REL */
430 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
433 pr_err("REL relocation unsupported.\n");
438 * Free up memory used by kernel, initrd, and comand line. This is temporary
439 * memory allocation which is not needed any more after these buffers have
440 * been loaded into separate segments and have been copied elsewhere.
442 static void kimage_file_post_load_cleanup(struct kimage *image)
444 struct purgatory_info *pi = &image->purgatory_info;
446 vfree(image->kernel_buf);
447 image->kernel_buf = NULL;
449 vfree(image->initrd_buf);
450 image->initrd_buf = NULL;
452 kfree(image->cmdline_buf);
453 image->cmdline_buf = NULL;
455 vfree(pi->purgatory_buf);
456 pi->purgatory_buf = NULL;
461 /* See if architecture has anything to cleanup post load */
462 arch_kimage_file_post_load_cleanup(image);
465 * Above call should have called into bootloader to free up
466 * any data stored in kimage->image_loader_data. It should
467 * be ok now to free it up.
469 kfree(image->image_loader_data);
470 image->image_loader_data = NULL;
474 * In file mode list of segments is prepared by kernel. Copy relevant
475 * data from user space, do error checking, prepare segment list
478 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
479 const char __user *cmdline_ptr,
480 unsigned long cmdline_len, unsigned flags)
485 ret = copy_file_from_fd(kernel_fd, &image->kernel_buf,
486 &image->kernel_buf_len);
490 /* Call arch image probe handlers */
491 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
492 image->kernel_buf_len);
497 /* It is possible that there no initramfs is being loaded */
498 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
499 ret = copy_file_from_fd(initrd_fd, &image->initrd_buf,
500 &image->initrd_buf_len);
506 image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
507 if (!image->cmdline_buf) {
512 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
519 image->cmdline_buf_len = cmdline_len;
521 /* command line should be a string with last byte null */
522 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
528 /* Call arch image load handlers */
529 ldata = arch_kexec_kernel_image_load(image);
532 ret = PTR_ERR(ldata);
536 image->image_loader_data = ldata;
538 /* In case of error, free up all allocated memory in this function */
540 kimage_file_post_load_cleanup(image);
545 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
546 int initrd_fd, const char __user *cmdline_ptr,
547 unsigned long cmdline_len, unsigned long flags)
550 struct kimage *image;
551 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
553 image = do_kimage_alloc_init();
557 image->file_mode = 1;
559 if (kexec_on_panic) {
560 /* Enable special crash kernel control page alloc policy. */
561 image->control_page = crashk_res.start;
562 image->type = KEXEC_TYPE_CRASH;
565 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
566 cmdline_ptr, cmdline_len, flags);
570 ret = sanity_check_segment_list(image);
572 goto out_free_post_load_bufs;
575 image->control_code_page = kimage_alloc_control_pages(image,
576 get_order(KEXEC_CONTROL_PAGE_SIZE));
577 if (!image->control_code_page) {
578 pr_err("Could not allocate control_code_buffer\n");
579 goto out_free_post_load_bufs;
582 if (!kexec_on_panic) {
583 image->swap_page = kimage_alloc_control_pages(image, 0);
584 if (!image->swap_page) {
585 pr_err(KERN_ERR "Could not allocate swap buffer\n");
586 goto out_free_control_pages;
592 out_free_control_pages:
593 kimage_free_page_list(&image->control_pages);
594 out_free_post_load_bufs:
595 kimage_file_post_load_cleanup(image);
601 static int kimage_is_destination_range(struct kimage *image,
607 for (i = 0; i < image->nr_segments; i++) {
608 unsigned long mstart, mend;
610 mstart = image->segment[i].mem;
611 mend = mstart + image->segment[i].memsz;
612 if ((end > mstart) && (start < mend))
619 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
623 pages = alloc_pages(gfp_mask, order);
625 unsigned int count, i;
626 pages->mapping = NULL;
627 set_page_private(pages, order);
629 for (i = 0; i < count; i++)
630 SetPageReserved(pages + i);
636 static void kimage_free_pages(struct page *page)
638 unsigned int order, count, i;
640 order = page_private(page);
642 for (i = 0; i < count; i++)
643 ClearPageReserved(page + i);
644 __free_pages(page, order);
647 static void kimage_free_page_list(struct list_head *list)
649 struct list_head *pos, *next;
651 list_for_each_safe(pos, next, list) {
654 page = list_entry(pos, struct page, lru);
655 list_del(&page->lru);
656 kimage_free_pages(page);
660 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
663 /* Control pages are special, they are the intermediaries
664 * that are needed while we copy the rest of the pages
665 * to their final resting place. As such they must
666 * not conflict with either the destination addresses
667 * or memory the kernel is already using.
669 * The only case where we really need more than one of
670 * these are for architectures where we cannot disable
671 * the MMU and must instead generate an identity mapped
672 * page table for all of the memory.
674 * At worst this runs in O(N) of the image size.
676 struct list_head extra_pages;
681 INIT_LIST_HEAD(&extra_pages);
683 /* Loop while I can allocate a page and the page allocated
684 * is a destination page.
687 unsigned long pfn, epfn, addr, eaddr;
689 pages = kimage_alloc_pages(GFP_KERNEL, order);
692 pfn = page_to_pfn(pages);
694 addr = pfn << PAGE_SHIFT;
695 eaddr = epfn << PAGE_SHIFT;
696 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
697 kimage_is_destination_range(image, addr, eaddr)) {
698 list_add(&pages->lru, &extra_pages);
704 /* Remember the allocated page... */
705 list_add(&pages->lru, &image->control_pages);
707 /* Because the page is already in it's destination
708 * location we will never allocate another page at
709 * that address. Therefore kimage_alloc_pages
710 * will not return it (again) and we don't need
711 * to give it an entry in image->segment[].
714 /* Deal with the destination pages I have inadvertently allocated.
716 * Ideally I would convert multi-page allocations into single
717 * page allocations, and add everything to image->dest_pages.
719 * For now it is simpler to just free the pages.
721 kimage_free_page_list(&extra_pages);
726 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
729 /* Control pages are special, they are the intermediaries
730 * that are needed while we copy the rest of the pages
731 * to their final resting place. As such they must
732 * not conflict with either the destination addresses
733 * or memory the kernel is already using.
735 * Control pages are also the only pags we must allocate
736 * when loading a crash kernel. All of the other pages
737 * are specified by the segments and we just memcpy
738 * into them directly.
740 * The only case where we really need more than one of
741 * these are for architectures where we cannot disable
742 * the MMU and must instead generate an identity mapped
743 * page table for all of the memory.
745 * Given the low demand this implements a very simple
746 * allocator that finds the first hole of the appropriate
747 * size in the reserved memory region, and allocates all
748 * of the memory up to and including the hole.
750 unsigned long hole_start, hole_end, size;
754 size = (1 << order) << PAGE_SHIFT;
755 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
756 hole_end = hole_start + size - 1;
757 while (hole_end <= crashk_res.end) {
760 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
762 /* See if I overlap any of the segments */
763 for (i = 0; i < image->nr_segments; i++) {
764 unsigned long mstart, mend;
766 mstart = image->segment[i].mem;
767 mend = mstart + image->segment[i].memsz - 1;
768 if ((hole_end >= mstart) && (hole_start <= mend)) {
769 /* Advance the hole to the end of the segment */
770 hole_start = (mend + (size - 1)) & ~(size - 1);
771 hole_end = hole_start + size - 1;
775 /* If I don't overlap any segments I have found my hole! */
776 if (i == image->nr_segments) {
777 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
782 image->control_page = hole_end;
788 struct page *kimage_alloc_control_pages(struct kimage *image,
791 struct page *pages = NULL;
793 switch (image->type) {
794 case KEXEC_TYPE_DEFAULT:
795 pages = kimage_alloc_normal_control_pages(image, order);
797 case KEXEC_TYPE_CRASH:
798 pages = kimage_alloc_crash_control_pages(image, order);
805 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
807 if (*image->entry != 0)
810 if (image->entry == image->last_entry) {
811 kimage_entry_t *ind_page;
814 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
818 ind_page = page_address(page);
819 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
820 image->entry = ind_page;
821 image->last_entry = ind_page +
822 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
824 *image->entry = entry;
831 static int kimage_set_destination(struct kimage *image,
832 unsigned long destination)
836 destination &= PAGE_MASK;
837 result = kimage_add_entry(image, destination | IND_DESTINATION);
839 image->destination = destination;
845 static int kimage_add_page(struct kimage *image, unsigned long page)
850 result = kimage_add_entry(image, page | IND_SOURCE);
852 image->destination += PAGE_SIZE;
858 static void kimage_free_extra_pages(struct kimage *image)
860 /* Walk through and free any extra destination pages I may have */
861 kimage_free_page_list(&image->dest_pages);
863 /* Walk through and free any unusable pages I have cached */
864 kimage_free_page_list(&image->unusable_pages);
867 static void kimage_terminate(struct kimage *image)
869 if (*image->entry != 0)
872 *image->entry = IND_DONE;
875 #define for_each_kimage_entry(image, ptr, entry) \
876 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
877 ptr = (entry & IND_INDIRECTION) ? \
878 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
880 static void kimage_free_entry(kimage_entry_t entry)
884 page = pfn_to_page(entry >> PAGE_SHIFT);
885 kimage_free_pages(page);
888 static void kimage_free(struct kimage *image)
890 kimage_entry_t *ptr, entry;
891 kimage_entry_t ind = 0;
896 kimage_free_extra_pages(image);
897 for_each_kimage_entry(image, ptr, entry) {
898 if (entry & IND_INDIRECTION) {
899 /* Free the previous indirection page */
900 if (ind & IND_INDIRECTION)
901 kimage_free_entry(ind);
902 /* Save this indirection page until we are
906 } else if (entry & IND_SOURCE)
907 kimage_free_entry(entry);
909 /* Free the final indirection page */
910 if (ind & IND_INDIRECTION)
911 kimage_free_entry(ind);
913 /* Handle any machine specific cleanup */
914 machine_kexec_cleanup(image);
916 /* Free the kexec control pages... */
917 kimage_free_page_list(&image->control_pages);
920 * Free up any temporary buffers allocated. This might hit if
921 * error occurred much later after buffer allocation.
923 if (image->file_mode)
924 kimage_file_post_load_cleanup(image);
929 static kimage_entry_t *kimage_dst_used(struct kimage *image,
932 kimage_entry_t *ptr, entry;
933 unsigned long destination = 0;
935 for_each_kimage_entry(image, ptr, entry) {
936 if (entry & IND_DESTINATION)
937 destination = entry & PAGE_MASK;
938 else if (entry & IND_SOURCE) {
939 if (page == destination)
941 destination += PAGE_SIZE;
948 static struct page *kimage_alloc_page(struct kimage *image,
950 unsigned long destination)
953 * Here we implement safeguards to ensure that a source page
954 * is not copied to its destination page before the data on
955 * the destination page is no longer useful.
957 * To do this we maintain the invariant that a source page is
958 * either its own destination page, or it is not a
959 * destination page at all.
961 * That is slightly stronger than required, but the proof
962 * that no problems will not occur is trivial, and the
963 * implementation is simply to verify.
965 * When allocating all pages normally this algorithm will run
966 * in O(N) time, but in the worst case it will run in O(N^2)
967 * time. If the runtime is a problem the data structures can
974 * Walk through the list of destination pages, and see if I
977 list_for_each_entry(page, &image->dest_pages, lru) {
978 addr = page_to_pfn(page) << PAGE_SHIFT;
979 if (addr == destination) {
980 list_del(&page->lru);
988 /* Allocate a page, if we run out of memory give up */
989 page = kimage_alloc_pages(gfp_mask, 0);
992 /* If the page cannot be used file it away */
993 if (page_to_pfn(page) >
994 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
995 list_add(&page->lru, &image->unusable_pages);
998 addr = page_to_pfn(page) << PAGE_SHIFT;
1000 /* If it is the destination page we want use it */
1001 if (addr == destination)
1004 /* If the page is not a destination page use it */
1005 if (!kimage_is_destination_range(image, addr,
1010 * I know that the page is someones destination page.
1011 * See if there is already a source page for this
1012 * destination page. And if so swap the source pages.
1014 old = kimage_dst_used(image, addr);
1017 unsigned long old_addr;
1018 struct page *old_page;
1020 old_addr = *old & PAGE_MASK;
1021 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
1022 copy_highpage(page, old_page);
1023 *old = addr | (*old & ~PAGE_MASK);
1025 /* The old page I have found cannot be a
1026 * destination page, so return it if it's
1027 * gfp_flags honor the ones passed in.
1029 if (!(gfp_mask & __GFP_HIGHMEM) &&
1030 PageHighMem(old_page)) {
1031 kimage_free_pages(old_page);
1038 /* Place the page on the destination list I
1039 * will use it later.
1041 list_add(&page->lru, &image->dest_pages);
1048 static int kimage_load_normal_segment(struct kimage *image,
1049 struct kexec_segment *segment)
1051 unsigned long maddr;
1052 size_t ubytes, mbytes;
1054 unsigned char __user *buf = NULL;
1055 unsigned char *kbuf = NULL;
1058 if (image->file_mode)
1059 kbuf = segment->kbuf;
1062 ubytes = segment->bufsz;
1063 mbytes = segment->memsz;
1064 maddr = segment->mem;
1066 result = kimage_set_destination(image, maddr);
1073 size_t uchunk, mchunk;
1075 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
1080 result = kimage_add_page(image, page_to_pfn(page)
1086 /* Start with a clear page */
1088 ptr += maddr & ~PAGE_MASK;
1089 mchunk = min_t(size_t, mbytes,
1090 PAGE_SIZE - (maddr & ~PAGE_MASK));
1091 uchunk = min(ubytes, mchunk);
1093 /* For file based kexec, source pages are in kernel memory */
1094 if (image->file_mode)
1095 memcpy(ptr, kbuf, uchunk);
1097 result = copy_from_user(ptr, buf, uchunk);
1105 if (image->file_mode)
1115 static int kimage_load_crash_segment(struct kimage *image,
1116 struct kexec_segment *segment)
1118 /* For crash dumps kernels we simply copy the data from
1119 * user space to it's destination.
1120 * We do things a page at a time for the sake of kmap.
1122 unsigned long maddr;
1123 size_t ubytes, mbytes;
1125 unsigned char __user *buf = NULL;
1126 unsigned char *kbuf = NULL;
1129 if (image->file_mode)
1130 kbuf = segment->kbuf;
1133 ubytes = segment->bufsz;
1134 mbytes = segment->memsz;
1135 maddr = segment->mem;
1139 size_t uchunk, mchunk;
1141 page = pfn_to_page(maddr >> PAGE_SHIFT);
1147 ptr += maddr & ~PAGE_MASK;
1148 mchunk = min_t(size_t, mbytes,
1149 PAGE_SIZE - (maddr & ~PAGE_MASK));
1150 uchunk = min(ubytes, mchunk);
1151 if (mchunk > uchunk) {
1152 /* Zero the trailing part of the page */
1153 memset(ptr + uchunk, 0, mchunk - uchunk);
1156 /* For file based kexec, source pages are in kernel memory */
1157 if (image->file_mode)
1158 memcpy(ptr, kbuf, uchunk);
1160 result = copy_from_user(ptr, buf, uchunk);
1161 kexec_flush_icache_page(page);
1169 if (image->file_mode)
1179 static int kimage_load_segment(struct kimage *image,
1180 struct kexec_segment *segment)
1182 int result = -ENOMEM;
1184 switch (image->type) {
1185 case KEXEC_TYPE_DEFAULT:
1186 result = kimage_load_normal_segment(image, segment);
1188 case KEXEC_TYPE_CRASH:
1189 result = kimage_load_crash_segment(image, segment);
1197 * Exec Kernel system call: for obvious reasons only root may call it.
1199 * This call breaks up into three pieces.
1200 * - A generic part which loads the new kernel from the current
1201 * address space, and very carefully places the data in the
1204 * - A generic part that interacts with the kernel and tells all of
1205 * the devices to shut down. Preventing on-going dmas, and placing
1206 * the devices in a consistent state so a later kernel can
1207 * reinitialize them.
1209 * - A machine specific part that includes the syscall number
1210 * and then copies the image to it's final destination. And
1211 * jumps into the image at entry.
1213 * kexec does not sync, or unmount filesystems so if you need
1214 * that to happen you need to do that yourself.
1216 struct kimage *kexec_image;
1217 struct kimage *kexec_crash_image;
1218 int kexec_load_disabled;
1220 static DEFINE_MUTEX(kexec_mutex);
1222 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
1223 struct kexec_segment __user *, segments, unsigned long, flags)
1225 struct kimage **dest_image, *image;
1228 /* We only trust the superuser with rebooting the system. */
1229 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1233 * Verify we have a legal set of flags
1234 * This leaves us room for future extensions.
1236 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
1239 /* Verify we are on the appropriate architecture */
1240 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
1241 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
1244 /* Put an artificial cap on the number
1245 * of segments passed to kexec_load.
1247 if (nr_segments > KEXEC_SEGMENT_MAX)
1253 /* Because we write directly to the reserved memory
1254 * region when loading crash kernels we need a mutex here to
1255 * prevent multiple crash kernels from attempting to load
1256 * simultaneously, and to prevent a crash kernel from loading
1257 * over the top of a in use crash kernel.
1259 * KISS: always take the mutex.
1261 if (!mutex_trylock(&kexec_mutex))
1264 dest_image = &kexec_image;
1265 if (flags & KEXEC_ON_CRASH)
1266 dest_image = &kexec_crash_image;
1267 if (nr_segments > 0) {
1270 /* Loading another kernel to reboot into */
1271 if ((flags & KEXEC_ON_CRASH) == 0)
1272 result = kimage_alloc_init(&image, entry, nr_segments,
1274 /* Loading another kernel to switch to if this one crashes */
1275 else if (flags & KEXEC_ON_CRASH) {
1276 /* Free any current crash dump kernel before
1279 kimage_free(xchg(&kexec_crash_image, NULL));
1280 result = kimage_alloc_init(&image, entry, nr_segments,
1282 crash_map_reserved_pages();
1287 if (flags & KEXEC_PRESERVE_CONTEXT)
1288 image->preserve_context = 1;
1289 result = machine_kexec_prepare(image);
1293 for (i = 0; i < nr_segments; i++) {
1294 result = kimage_load_segment(image, &image->segment[i]);
1298 kimage_terminate(image);
1299 if (flags & KEXEC_ON_CRASH)
1300 crash_unmap_reserved_pages();
1302 /* Install the new kernel, and Uninstall the old */
1303 image = xchg(dest_image, image);
1306 mutex_unlock(&kexec_mutex);
1313 * Add and remove page tables for crashkernel memory
1315 * Provide an empty default implementation here -- architecture
1316 * code may override this
1318 void __weak crash_map_reserved_pages(void)
1321 void __weak crash_unmap_reserved_pages(void)
1324 #ifdef CONFIG_COMPAT
1325 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1326 compat_ulong_t, nr_segments,
1327 struct compat_kexec_segment __user *, segments,
1328 compat_ulong_t, flags)
1330 struct compat_kexec_segment in;
1331 struct kexec_segment out, __user *ksegments;
1332 unsigned long i, result;
1334 /* Don't allow clients that don't understand the native
1335 * architecture to do anything.
1337 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1340 if (nr_segments > KEXEC_SEGMENT_MAX)
1343 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1344 for (i = 0; i < nr_segments; i++) {
1345 result = copy_from_user(&in, &segments[i], sizeof(in));
1349 out.buf = compat_ptr(in.buf);
1350 out.bufsz = in.bufsz;
1352 out.memsz = in.memsz;
1354 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1359 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1363 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
1364 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
1365 unsigned long, flags)
1368 struct kimage **dest_image, *image;
1370 /* We only trust the superuser with rebooting the system. */
1371 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1374 /* Make sure we have a legal set of flags */
1375 if (flags != (flags & KEXEC_FILE_FLAGS))
1380 if (!mutex_trylock(&kexec_mutex))
1383 dest_image = &kexec_image;
1384 if (flags & KEXEC_FILE_ON_CRASH)
1385 dest_image = &kexec_crash_image;
1387 if (flags & KEXEC_FILE_UNLOAD)
1391 * In case of crash, new kernel gets loaded in reserved region. It is
1392 * same memory where old crash kernel might be loaded. Free any
1393 * current crash dump kernel before we corrupt it.
1395 if (flags & KEXEC_FILE_ON_CRASH)
1396 kimage_free(xchg(&kexec_crash_image, NULL));
1398 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
1399 cmdline_len, flags);
1403 ret = machine_kexec_prepare(image);
1407 ret = kexec_calculate_store_digests(image);
1411 for (i = 0; i < image->nr_segments; i++) {
1412 struct kexec_segment *ksegment;
1414 ksegment = &image->segment[i];
1415 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
1416 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
1419 ret = kimage_load_segment(image, &image->segment[i]);
1424 kimage_terminate(image);
1427 * Free up any temporary buffers allocated which are not needed
1428 * after image has been loaded
1430 kimage_file_post_load_cleanup(image);
1432 image = xchg(dest_image, image);
1434 mutex_unlock(&kexec_mutex);
1439 void crash_kexec(struct pt_regs *regs)
1441 /* Take the kexec_mutex here to prevent sys_kexec_load
1442 * running on one cpu from replacing the crash kernel
1443 * we are using after a panic on a different cpu.
1445 * If the crash kernel was not located in a fixed area
1446 * of memory the xchg(&kexec_crash_image) would be
1447 * sufficient. But since I reuse the memory...
1449 if (mutex_trylock(&kexec_mutex)) {
1450 if (kexec_crash_image) {
1451 struct pt_regs fixed_regs;
1453 crash_setup_regs(&fixed_regs, regs);
1454 crash_save_vmcoreinfo();
1455 machine_crash_shutdown(&fixed_regs);
1456 machine_kexec(kexec_crash_image);
1458 mutex_unlock(&kexec_mutex);
1462 size_t crash_get_memory_size(void)
1465 mutex_lock(&kexec_mutex);
1466 if (crashk_res.end != crashk_res.start)
1467 size = resource_size(&crashk_res);
1468 mutex_unlock(&kexec_mutex);
1472 void __weak crash_free_reserved_phys_range(unsigned long begin,
1477 for (addr = begin; addr < end; addr += PAGE_SIZE)
1478 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1481 int crash_shrink_memory(unsigned long new_size)
1484 unsigned long start, end;
1485 unsigned long old_size;
1486 struct resource *ram_res;
1488 mutex_lock(&kexec_mutex);
1490 if (kexec_crash_image) {
1494 start = crashk_res.start;
1495 end = crashk_res.end;
1496 old_size = (end == 0) ? 0 : end - start + 1;
1497 if (new_size >= old_size) {
1498 ret = (new_size == old_size) ? 0 : -EINVAL;
1502 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1508 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1509 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1511 crash_map_reserved_pages();
1512 crash_free_reserved_phys_range(end, crashk_res.end);
1514 if ((start == end) && (crashk_res.parent != NULL))
1515 release_resource(&crashk_res);
1517 ram_res->start = end;
1518 ram_res->end = crashk_res.end;
1519 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1520 ram_res->name = "System RAM";
1522 crashk_res.end = end - 1;
1524 insert_resource(&iomem_resource, ram_res);
1525 crash_unmap_reserved_pages();
1528 mutex_unlock(&kexec_mutex);
1532 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1535 struct elf_note note;
1537 note.n_namesz = strlen(name) + 1;
1538 note.n_descsz = data_len;
1540 memcpy(buf, ¬e, sizeof(note));
1541 buf += (sizeof(note) + 3)/4;
1542 memcpy(buf, name, note.n_namesz);
1543 buf += (note.n_namesz + 3)/4;
1544 memcpy(buf, data, note.n_descsz);
1545 buf += (note.n_descsz + 3)/4;
1550 static void final_note(u32 *buf)
1552 struct elf_note note;
1557 memcpy(buf, ¬e, sizeof(note));
1560 void crash_save_cpu(struct pt_regs *regs, int cpu)
1562 struct elf_prstatus prstatus;
1565 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1568 /* Using ELF notes here is opportunistic.
1569 * I need a well defined structure format
1570 * for the data I pass, and I need tags
1571 * on the data to indicate what information I have
1572 * squirrelled away. ELF notes happen to provide
1573 * all of that, so there is no need to invent something new.
1575 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1578 memset(&prstatus, 0, sizeof(prstatus));
1579 prstatus.pr_pid = current->pid;
1580 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1581 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1582 &prstatus, sizeof(prstatus));
1586 static int __init crash_notes_memory_init(void)
1588 /* Allocate memory for saving cpu registers. */
1589 crash_notes = alloc_percpu(note_buf_t);
1591 pr_warn("Kexec: Memory allocation for saving cpu register states failed\n");
1596 subsys_initcall(crash_notes_memory_init);
1600 * parsing the "crashkernel" commandline
1602 * this code is intended to be called from architecture specific code
1607 * This function parses command lines in the format
1609 * crashkernel=ramsize-range:size[,...][@offset]
1611 * The function returns 0 on success and -EINVAL on failure.
1613 static int __init parse_crashkernel_mem(char *cmdline,
1614 unsigned long long system_ram,
1615 unsigned long long *crash_size,
1616 unsigned long long *crash_base)
1618 char *cur = cmdline, *tmp;
1620 /* for each entry of the comma-separated list */
1622 unsigned long long start, end = ULLONG_MAX, size;
1624 /* get the start of the range */
1625 start = memparse(cur, &tmp);
1627 pr_warn("crashkernel: Memory value expected\n");
1632 pr_warn("crashkernel: '-' expected\n");
1637 /* if no ':' is here, than we read the end */
1639 end = memparse(cur, &tmp);
1641 pr_warn("crashkernel: Memory value expected\n");
1646 pr_warn("crashkernel: end <= start\n");
1652 pr_warn("crashkernel: ':' expected\n");
1657 size = memparse(cur, &tmp);
1659 pr_warn("Memory value expected\n");
1663 if (size >= system_ram) {
1664 pr_warn("crashkernel: invalid size\n");
1669 if (system_ram >= start && system_ram < end) {
1673 } while (*cur++ == ',');
1675 if (*crash_size > 0) {
1676 while (*cur && *cur != ' ' && *cur != '@')
1680 *crash_base = memparse(cur, &tmp);
1682 pr_warn("Memory value expected after '@'\n");
1692 * That function parses "simple" (old) crashkernel command lines like
1694 * crashkernel=size[@offset]
1696 * It returns 0 on success and -EINVAL on failure.
1698 static int __init parse_crashkernel_simple(char *cmdline,
1699 unsigned long long *crash_size,
1700 unsigned long long *crash_base)
1702 char *cur = cmdline;
1704 *crash_size = memparse(cmdline, &cur);
1705 if (cmdline == cur) {
1706 pr_warn("crashkernel: memory value expected\n");
1711 *crash_base = memparse(cur+1, &cur);
1712 else if (*cur != ' ' && *cur != '\0') {
1713 pr_warn("crashkernel: unrecognized char\n");
1720 #define SUFFIX_HIGH 0
1721 #define SUFFIX_LOW 1
1722 #define SUFFIX_NULL 2
1723 static __initdata char *suffix_tbl[] = {
1724 [SUFFIX_HIGH] = ",high",
1725 [SUFFIX_LOW] = ",low",
1726 [SUFFIX_NULL] = NULL,
1730 * That function parses "suffix" crashkernel command lines like
1732 * crashkernel=size,[high|low]
1734 * It returns 0 on success and -EINVAL on failure.
1736 static int __init parse_crashkernel_suffix(char *cmdline,
1737 unsigned long long *crash_size,
1738 unsigned long long *crash_base,
1741 char *cur = cmdline;
1743 *crash_size = memparse(cmdline, &cur);
1744 if (cmdline == cur) {
1745 pr_warn("crashkernel: memory value expected\n");
1749 /* check with suffix */
1750 if (strncmp(cur, suffix, strlen(suffix))) {
1751 pr_warn("crashkernel: unrecognized char\n");
1754 cur += strlen(suffix);
1755 if (*cur != ' ' && *cur != '\0') {
1756 pr_warn("crashkernel: unrecognized char\n");
1763 static __init char *get_last_crashkernel(char *cmdline,
1767 char *p = cmdline, *ck_cmdline = NULL;
1769 /* find crashkernel and use the last one if there are more */
1770 p = strstr(p, name);
1772 char *end_p = strchr(p, ' ');
1776 end_p = p + strlen(p);
1781 /* skip the one with any known suffix */
1782 for (i = 0; suffix_tbl[i]; i++) {
1783 q = end_p - strlen(suffix_tbl[i]);
1784 if (!strncmp(q, suffix_tbl[i],
1785 strlen(suffix_tbl[i])))
1790 q = end_p - strlen(suffix);
1791 if (!strncmp(q, suffix, strlen(suffix)))
1795 p = strstr(p+1, name);
1804 static int __init __parse_crashkernel(char *cmdline,
1805 unsigned long long system_ram,
1806 unsigned long long *crash_size,
1807 unsigned long long *crash_base,
1811 char *first_colon, *first_space;
1814 BUG_ON(!crash_size || !crash_base);
1818 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1823 ck_cmdline += strlen(name);
1826 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1827 crash_base, suffix);
1829 * if the commandline contains a ':', then that's the extended
1830 * syntax -- if not, it must be the classic syntax
1832 first_colon = strchr(ck_cmdline, ':');
1833 first_space = strchr(ck_cmdline, ' ');
1834 if (first_colon && (!first_space || first_colon < first_space))
1835 return parse_crashkernel_mem(ck_cmdline, system_ram,
1836 crash_size, crash_base);
1838 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1842 * That function is the entry point for command line parsing and should be
1843 * called from the arch-specific code.
1845 int __init parse_crashkernel(char *cmdline,
1846 unsigned long long system_ram,
1847 unsigned long long *crash_size,
1848 unsigned long long *crash_base)
1850 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1851 "crashkernel=", NULL);
1854 int __init parse_crashkernel_high(char *cmdline,
1855 unsigned long long system_ram,
1856 unsigned long long *crash_size,
1857 unsigned long long *crash_base)
1859 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1860 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1863 int __init parse_crashkernel_low(char *cmdline,
1864 unsigned long long system_ram,
1865 unsigned long long *crash_size,
1866 unsigned long long *crash_base)
1868 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1869 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1872 static void update_vmcoreinfo_note(void)
1874 u32 *buf = vmcoreinfo_note;
1876 if (!vmcoreinfo_size)
1878 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1883 void crash_save_vmcoreinfo(void)
1885 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1886 update_vmcoreinfo_note();
1889 void vmcoreinfo_append_str(const char *fmt, ...)
1895 va_start(args, fmt);
1896 r = vscnprintf(buf, sizeof(buf), fmt, args);
1899 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1901 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1903 vmcoreinfo_size += r;
1907 * provide an empty default implementation here -- architecture
1908 * code may override this
1910 void __weak arch_crash_save_vmcoreinfo(void)
1913 unsigned long __weak paddr_vmcoreinfo_note(void)
1915 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1918 static int __init crash_save_vmcoreinfo_init(void)
1920 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1921 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1923 VMCOREINFO_SYMBOL(init_uts_ns);
1924 VMCOREINFO_SYMBOL(node_online_map);
1926 VMCOREINFO_SYMBOL(swapper_pg_dir);
1928 VMCOREINFO_SYMBOL(_stext);
1929 VMCOREINFO_SYMBOL(vmap_area_list);
1931 #ifndef CONFIG_NEED_MULTIPLE_NODES
1932 VMCOREINFO_SYMBOL(mem_map);
1933 VMCOREINFO_SYMBOL(contig_page_data);
1935 #ifdef CONFIG_SPARSEMEM
1936 VMCOREINFO_SYMBOL(mem_section);
1937 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1938 VMCOREINFO_STRUCT_SIZE(mem_section);
1939 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1941 VMCOREINFO_STRUCT_SIZE(page);
1942 VMCOREINFO_STRUCT_SIZE(pglist_data);
1943 VMCOREINFO_STRUCT_SIZE(zone);
1944 VMCOREINFO_STRUCT_SIZE(free_area);
1945 VMCOREINFO_STRUCT_SIZE(list_head);
1946 VMCOREINFO_SIZE(nodemask_t);
1947 VMCOREINFO_OFFSET(page, flags);
1948 VMCOREINFO_OFFSET(page, _count);
1949 VMCOREINFO_OFFSET(page, mapping);
1950 VMCOREINFO_OFFSET(page, lru);
1951 VMCOREINFO_OFFSET(page, _mapcount);
1952 VMCOREINFO_OFFSET(page, private);
1953 VMCOREINFO_OFFSET(pglist_data, node_zones);
1954 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1955 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1956 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1958 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1959 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1960 VMCOREINFO_OFFSET(pglist_data, node_id);
1961 VMCOREINFO_OFFSET(zone, free_area);
1962 VMCOREINFO_OFFSET(zone, vm_stat);
1963 VMCOREINFO_OFFSET(zone, spanned_pages);
1964 VMCOREINFO_OFFSET(free_area, free_list);
1965 VMCOREINFO_OFFSET(list_head, next);
1966 VMCOREINFO_OFFSET(list_head, prev);
1967 VMCOREINFO_OFFSET(vmap_area, va_start);
1968 VMCOREINFO_OFFSET(vmap_area, list);
1969 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1970 log_buf_kexec_setup();
1971 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1972 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1973 VMCOREINFO_NUMBER(PG_lru);
1974 VMCOREINFO_NUMBER(PG_private);
1975 VMCOREINFO_NUMBER(PG_swapcache);
1976 VMCOREINFO_NUMBER(PG_slab);
1977 #ifdef CONFIG_MEMORY_FAILURE
1978 VMCOREINFO_NUMBER(PG_hwpoison);
1980 VMCOREINFO_NUMBER(PG_head_mask);
1981 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1982 #ifdef CONFIG_HUGETLBFS
1983 VMCOREINFO_SYMBOL(free_huge_page);
1986 arch_crash_save_vmcoreinfo();
1987 update_vmcoreinfo_note();
1992 subsys_initcall(crash_save_vmcoreinfo_init);
1994 static int __kexec_add_segment(struct kimage *image, char *buf,
1995 unsigned long bufsz, unsigned long mem,
1996 unsigned long memsz)
1998 struct kexec_segment *ksegment;
2000 ksegment = &image->segment[image->nr_segments];
2001 ksegment->kbuf = buf;
2002 ksegment->bufsz = bufsz;
2003 ksegment->mem = mem;
2004 ksegment->memsz = memsz;
2005 image->nr_segments++;
2010 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
2011 struct kexec_buf *kbuf)
2013 struct kimage *image = kbuf->image;
2014 unsigned long temp_start, temp_end;
2016 temp_end = min(end, kbuf->buf_max);
2017 temp_start = temp_end - kbuf->memsz;
2020 /* align down start */
2021 temp_start = temp_start & (~(kbuf->buf_align - 1));
2023 if (temp_start < start || temp_start < kbuf->buf_min)
2026 temp_end = temp_start + kbuf->memsz - 1;
2029 * Make sure this does not conflict with any of existing
2032 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2033 temp_start = temp_start - PAGE_SIZE;
2037 /* We found a suitable memory range */
2041 /* If we are here, we found a suitable memory range */
2042 __kexec_add_segment(image, kbuf->buffer, kbuf->bufsz, temp_start,
2045 /* Success, stop navigating through remaining System RAM ranges */
2049 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
2050 struct kexec_buf *kbuf)
2052 struct kimage *image = kbuf->image;
2053 unsigned long temp_start, temp_end;
2055 temp_start = max(start, kbuf->buf_min);
2058 temp_start = ALIGN(temp_start, kbuf->buf_align);
2059 temp_end = temp_start + kbuf->memsz - 1;
2061 if (temp_end > end || temp_end > kbuf->buf_max)
2064 * Make sure this does not conflict with any of existing
2067 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2068 temp_start = temp_start + PAGE_SIZE;
2072 /* We found a suitable memory range */
2076 /* If we are here, we found a suitable memory range */
2077 __kexec_add_segment(image, kbuf->buffer, kbuf->bufsz, temp_start,
2080 /* Success, stop navigating through remaining System RAM ranges */
2084 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
2086 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
2087 unsigned long sz = end - start + 1;
2089 /* Returning 0 will take to next memory range */
2090 if (sz < kbuf->memsz)
2093 if (end < kbuf->buf_min || start > kbuf->buf_max)
2097 * Allocate memory top down with-in ram range. Otherwise bottom up
2101 return locate_mem_hole_top_down(start, end, kbuf);
2102 return locate_mem_hole_bottom_up(start, end, kbuf);
2106 * Helper function for placing a buffer in a kexec segment. This assumes
2107 * that kexec_mutex is held.
2109 int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
2110 unsigned long memsz, unsigned long buf_align,
2111 unsigned long buf_min, unsigned long buf_max,
2112 bool top_down, unsigned long *load_addr)
2115 struct kexec_segment *ksegment;
2116 struct kexec_buf buf, *kbuf;
2119 /* Currently adding segment this way is allowed only in file mode */
2120 if (!image->file_mode)
2123 if (image->nr_segments >= KEXEC_SEGMENT_MAX)
2127 * Make sure we are not trying to add buffer after allocating
2128 * control pages. All segments need to be placed first before
2129 * any control pages are allocated. As control page allocation
2130 * logic goes through list of segments to make sure there are
2131 * no destination overlaps.
2133 if (!list_empty(&image->control_pages)) {
2138 memset(&buf, 0, sizeof(struct kexec_buf));
2140 kbuf->image = image;
2141 kbuf->buffer = buffer;
2142 kbuf->bufsz = bufsz;
2144 kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
2145 kbuf->buf_align = max(buf_align, PAGE_SIZE);
2146 kbuf->buf_min = buf_min;
2147 kbuf->buf_max = buf_max;
2148 kbuf->top_down = top_down;
2150 /* Walk the RAM ranges and allocate a suitable range for the buffer */
2151 if (image->type == KEXEC_TYPE_CRASH)
2152 ret = walk_iomem_res("Crash kernel",
2153 IORESOURCE_MEM | IORESOURCE_BUSY,
2154 crashk_res.start, crashk_res.end, kbuf,
2155 locate_mem_hole_callback);
2157 ret = walk_system_ram_res(0, -1, kbuf,
2158 locate_mem_hole_callback);
2160 /* A suitable memory range could not be found for buffer */
2161 return -EADDRNOTAVAIL;
2164 /* Found a suitable memory range */
2165 ksegment = &image->segment[image->nr_segments - 1];
2166 *load_addr = ksegment->mem;
2170 /* Calculate and store the digest of segments */
2171 static int kexec_calculate_store_digests(struct kimage *image)
2173 struct crypto_shash *tfm;
2174 struct shash_desc *desc;
2175 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
2176 size_t desc_size, nullsz;
2179 struct kexec_sha_region *sha_regions;
2180 struct purgatory_info *pi = &image->purgatory_info;
2182 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
2183 zero_buf_sz = PAGE_SIZE;
2185 tfm = crypto_alloc_shash("sha256", 0, 0);
2191 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
2192 desc = kzalloc(desc_size, GFP_KERNEL);
2198 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
2199 sha_regions = vzalloc(sha_region_sz);
2206 ret = crypto_shash_init(desc);
2208 goto out_free_sha_regions;
2210 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
2213 goto out_free_sha_regions;
2216 for (j = i = 0; i < image->nr_segments; i++) {
2217 struct kexec_segment *ksegment;
2219 ksegment = &image->segment[i];
2221 * Skip purgatory as it will be modified once we put digest
2222 * info in purgatory.
2224 if (ksegment->kbuf == pi->purgatory_buf)
2227 ret = crypto_shash_update(desc, ksegment->kbuf,
2233 * Assume rest of the buffer is filled with zero and
2234 * update digest accordingly.
2236 nullsz = ksegment->memsz - ksegment->bufsz;
2238 unsigned long bytes = nullsz;
2240 if (bytes > zero_buf_sz)
2241 bytes = zero_buf_sz;
2242 ret = crypto_shash_update(desc, zero_buf, bytes);
2251 sha_regions[j].start = ksegment->mem;
2252 sha_regions[j].len = ksegment->memsz;
2257 ret = crypto_shash_final(desc, digest);
2259 goto out_free_digest;
2260 ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
2261 sha_regions, sha_region_sz, 0);
2263 goto out_free_digest;
2265 ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
2266 digest, SHA256_DIGEST_SIZE, 0);
2268 goto out_free_digest;
2273 out_free_sha_regions:
2283 /* Actually load purgatory. Lot of code taken from kexec-tools */
2284 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
2285 unsigned long max, int top_down)
2287 struct purgatory_info *pi = &image->purgatory_info;
2288 unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
2289 unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
2290 unsigned char *buf_addr, *src;
2291 int i, ret = 0, entry_sidx = -1;
2292 const Elf_Shdr *sechdrs_c;
2293 Elf_Shdr *sechdrs = NULL;
2294 void *purgatory_buf = NULL;
2297 * sechdrs_c points to section headers in purgatory and are read
2298 * only. No modifications allowed.
2300 sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
2303 * We can not modify sechdrs_c[] and its fields. It is read only.
2304 * Copy it over to a local copy where one can store some temporary
2305 * data and free it at the end. We need to modify ->sh_addr and
2306 * ->sh_offset fields to keep track of permanent and temporary
2307 * locations of sections.
2309 sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2313 memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2316 * We seem to have multiple copies of sections. First copy is which
2317 * is embedded in kernel in read only section. Some of these sections
2318 * will be copied to a temporary buffer and relocated. And these
2319 * sections will finally be copied to their final destination at
2320 * segment load time.
2322 * Use ->sh_offset to reflect section address in memory. It will
2323 * point to original read only copy if section is not allocatable.
2324 * Otherwise it will point to temporary copy which will be relocated.
2326 * Use ->sh_addr to contain final address of the section where it
2327 * will go during execution time.
2329 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2330 if (sechdrs[i].sh_type == SHT_NOBITS)
2333 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
2334 sechdrs[i].sh_offset;
2338 * Identify entry point section and make entry relative to section
2341 entry = pi->ehdr->e_entry;
2342 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2343 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2346 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
2349 /* Make entry section relative */
2350 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
2351 ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
2352 pi->ehdr->e_entry)) {
2354 entry -= sechdrs[i].sh_addr;
2359 /* Determine how much memory is needed to load relocatable object. */
2365 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2366 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2369 align = sechdrs[i].sh_addralign;
2370 if (sechdrs[i].sh_type != SHT_NOBITS) {
2371 if (buf_align < align)
2373 buf_sz = ALIGN(buf_sz, align);
2374 buf_sz += sechdrs[i].sh_size;
2377 if (bss_align < align)
2379 bss_sz = ALIGN(bss_sz, align);
2380 bss_sz += sechdrs[i].sh_size;
2384 /* Determine the bss padding required to align bss properly */
2386 if (buf_sz & (bss_align - 1))
2387 bss_pad = bss_align - (buf_sz & (bss_align - 1));
2389 memsz = buf_sz + bss_pad + bss_sz;
2391 /* Allocate buffer for purgatory */
2392 purgatory_buf = vzalloc(buf_sz);
2393 if (!purgatory_buf) {
2398 if (buf_align < bss_align)
2399 buf_align = bss_align;
2401 /* Add buffer to segment list */
2402 ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
2403 buf_align, min, max, top_down,
2404 &pi->purgatory_load_addr);
2408 /* Load SHF_ALLOC sections */
2409 buf_addr = purgatory_buf;
2410 load_addr = curr_load_addr = pi->purgatory_load_addr;
2411 bss_addr = load_addr + buf_sz + bss_pad;
2413 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2414 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2417 align = sechdrs[i].sh_addralign;
2418 if (sechdrs[i].sh_type != SHT_NOBITS) {
2419 curr_load_addr = ALIGN(curr_load_addr, align);
2420 offset = curr_load_addr - load_addr;
2421 /* We already modifed ->sh_offset to keep src addr */
2422 src = (char *) sechdrs[i].sh_offset;
2423 memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
2425 /* Store load address and source address of section */
2426 sechdrs[i].sh_addr = curr_load_addr;
2429 * This section got copied to temporary buffer. Update
2430 * ->sh_offset accordingly.
2432 sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
2434 /* Advance to the next address */
2435 curr_load_addr += sechdrs[i].sh_size;
2437 bss_addr = ALIGN(bss_addr, align);
2438 sechdrs[i].sh_addr = bss_addr;
2439 bss_addr += sechdrs[i].sh_size;
2443 /* Update entry point based on load address of text section */
2444 if (entry_sidx >= 0)
2445 entry += sechdrs[entry_sidx].sh_addr;
2447 /* Make kernel jump to purgatory after shutdown */
2448 image->start = entry;
2450 /* Used later to get/set symbol values */
2451 pi->sechdrs = sechdrs;
2454 * Used later to identify which section is purgatory and skip it
2455 * from checksumming.
2457 pi->purgatory_buf = purgatory_buf;
2461 vfree(purgatory_buf);
2465 static int kexec_apply_relocations(struct kimage *image)
2468 struct purgatory_info *pi = &image->purgatory_info;
2469 Elf_Shdr *sechdrs = pi->sechdrs;
2471 /* Apply relocations */
2472 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2473 Elf_Shdr *section, *symtab;
2475 if (sechdrs[i].sh_type != SHT_RELA &&
2476 sechdrs[i].sh_type != SHT_REL)
2480 * For section of type SHT_RELA/SHT_REL,
2481 * ->sh_link contains section header index of associated
2482 * symbol table. And ->sh_info contains section header
2483 * index of section to which relocations apply.
2485 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
2486 sechdrs[i].sh_link >= pi->ehdr->e_shnum)
2489 section = &sechdrs[sechdrs[i].sh_info];
2490 symtab = &sechdrs[sechdrs[i].sh_link];
2492 if (!(section->sh_flags & SHF_ALLOC))
2496 * symtab->sh_link contain section header index of associated
2499 if (symtab->sh_link >= pi->ehdr->e_shnum)
2500 /* Invalid section number? */
2504 * Respective archicture needs to provide support for applying
2505 * relocations of type SHT_RELA/SHT_REL.
2507 if (sechdrs[i].sh_type == SHT_RELA)
2508 ret = arch_kexec_apply_relocations_add(pi->ehdr,
2510 else if (sechdrs[i].sh_type == SHT_REL)
2511 ret = arch_kexec_apply_relocations(pi->ehdr,
2520 /* Load relocatable purgatory object and relocate it appropriately */
2521 int kexec_load_purgatory(struct kimage *image, unsigned long min,
2522 unsigned long max, int top_down,
2523 unsigned long *load_addr)
2525 struct purgatory_info *pi = &image->purgatory_info;
2528 if (kexec_purgatory_size <= 0)
2531 if (kexec_purgatory_size < sizeof(Elf_Ehdr))
2534 pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
2536 if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
2537 || pi->ehdr->e_type != ET_REL
2538 || !elf_check_arch(pi->ehdr)
2539 || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
2542 if (pi->ehdr->e_shoff >= kexec_purgatory_size
2543 || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
2544 kexec_purgatory_size - pi->ehdr->e_shoff))
2547 ret = __kexec_load_purgatory(image, min, max, top_down);
2551 ret = kexec_apply_relocations(image);
2555 *load_addr = pi->purgatory_load_addr;
2559 vfree(pi->purgatory_buf);
2563 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
2572 if (!pi->sechdrs || !pi->ehdr)
2575 sechdrs = pi->sechdrs;
2578 for (i = 0; i < ehdr->e_shnum; i++) {
2579 if (sechdrs[i].sh_type != SHT_SYMTAB)
2582 if (sechdrs[i].sh_link >= ehdr->e_shnum)
2583 /* Invalid strtab section number */
2585 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
2586 syms = (Elf_Sym *)sechdrs[i].sh_offset;
2588 /* Go through symbols for a match */
2589 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
2590 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
2593 if (strcmp(strtab + syms[k].st_name, name) != 0)
2596 if (syms[k].st_shndx == SHN_UNDEF ||
2597 syms[k].st_shndx >= ehdr->e_shnum) {
2598 pr_debug("Symbol: %s has bad section index %d.\n",
2599 name, syms[k].st_shndx);
2603 /* Found the symbol we are looking for */
2611 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
2613 struct purgatory_info *pi = &image->purgatory_info;
2617 sym = kexec_purgatory_find_symbol(pi, name);
2619 return ERR_PTR(-EINVAL);
2621 sechdr = &pi->sechdrs[sym->st_shndx];
2624 * Returns the address where symbol will finally be loaded after
2625 * kexec_load_segment()
2627 return (void *)(sechdr->sh_addr + sym->st_value);
2631 * Get or set value of a symbol. If "get_value" is true, symbol value is
2632 * returned in buf otherwise symbol value is set based on value in buf.
2634 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
2635 void *buf, unsigned int size, bool get_value)
2639 struct purgatory_info *pi = &image->purgatory_info;
2642 sym = kexec_purgatory_find_symbol(pi, name);
2646 if (sym->st_size != size) {
2647 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
2648 name, (unsigned long)sym->st_size, size);
2652 sechdrs = pi->sechdrs;
2654 if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
2655 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
2656 get_value ? "get" : "set");
2660 sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
2664 memcpy((void *)buf, sym_buf, size);
2666 memcpy((void *)sym_buf, buf, size);
2672 * Move into place and start executing a preloaded standalone
2673 * executable. If nothing was preloaded return an error.
2675 int kernel_kexec(void)
2679 if (!mutex_trylock(&kexec_mutex))
2686 #ifdef CONFIG_KEXEC_JUMP
2687 if (kexec_image->preserve_context) {
2688 lock_system_sleep();
2689 pm_prepare_console();
2690 error = freeze_processes();
2693 goto Restore_console;
2696 error = dpm_suspend_start(PMSG_FREEZE);
2698 goto Resume_console;
2699 /* At this point, dpm_suspend_start() has been called,
2700 * but *not* dpm_suspend_end(). We *must* call
2701 * dpm_suspend_end() now. Otherwise, drivers for
2702 * some devices (e.g. interrupt controllers) become
2703 * desynchronized with the actual state of the
2704 * hardware at resume time, and evil weirdness ensues.
2706 error = dpm_suspend_end(PMSG_FREEZE);
2708 goto Resume_devices;
2709 error = disable_nonboot_cpus();
2712 local_irq_disable();
2713 error = syscore_suspend();
2719 kexec_in_progress = true;
2720 kernel_restart_prepare(NULL);
2721 migrate_to_reboot_cpu();
2724 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
2725 * no further code needs to use CPU hotplug (which is true in
2726 * the reboot case). However, the kexec path depends on using
2727 * CPU hotplug again; so re-enable it here.
2729 cpu_hotplug_enable();
2730 pr_emerg("Starting new kernel\n");
2734 machine_kexec(kexec_image);
2736 #ifdef CONFIG_KEXEC_JUMP
2737 if (kexec_image->preserve_context) {
2742 enable_nonboot_cpus();
2743 dpm_resume_start(PMSG_RESTORE);
2745 dpm_resume_end(PMSG_RESTORE);
2750 pm_restore_console();
2751 unlock_system_sleep();
2756 mutex_unlock(&kexec_mutex);