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 #ifdef CONFIG_KEXEC_FILE
68 static int kexec_calculate_store_digests(struct kimage *image);
71 /* Location of the reserved area for the crash kernel */
72 struct resource crashk_res = {
73 .name = "Crash kernel",
76 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
78 struct resource crashk_low_res = {
79 .name = "Crash kernel",
82 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
85 int kexec_should_crash(struct task_struct *p)
87 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
93 * When kexec transitions to the new kernel there is a one-to-one
94 * mapping between physical and virtual addresses. On processors
95 * where you can disable the MMU this is trivial, and easy. For
96 * others it is still a simple predictable page table to setup.
98 * In that environment kexec copies the new kernel to its final
99 * resting place. This means I can only support memory whose
100 * physical address can fit in an unsigned long. In particular
101 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
102 * If the assembly stub has more restrictive requirements
103 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
104 * defined more restrictively in <asm/kexec.h>.
106 * The code for the transition from the current kernel to the
107 * the new kernel is placed in the control_code_buffer, whose size
108 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
109 * page of memory is necessary, but some architectures require more.
110 * Because this memory must be identity mapped in the transition from
111 * virtual to physical addresses it must live in the range
112 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
115 * The assembly stub in the control code buffer is passed a linked list
116 * of descriptor pages detailing the source pages of the new kernel,
117 * and the destination addresses of those source pages. As this data
118 * structure is not used in the context of the current OS, it must
121 * The code has been made to work with highmem pages and will use a
122 * destination page in its final resting place (if it happens
123 * to allocate it). The end product of this is that most of the
124 * physical address space, and most of RAM can be used.
126 * Future directions include:
127 * - allocating a page table with the control code buffer identity
128 * mapped, to simplify machine_kexec and make kexec_on_panic more
133 * KIMAGE_NO_DEST is an impossible destination address..., for
134 * allocating pages whose destination address we do not care about.
136 #define KIMAGE_NO_DEST (-1UL)
138 static int kimage_is_destination_range(struct kimage *image,
139 unsigned long start, unsigned long end);
140 static struct page *kimage_alloc_page(struct kimage *image,
144 static int copy_user_segment_list(struct kimage *image,
145 unsigned long nr_segments,
146 struct kexec_segment __user *segments)
149 size_t segment_bytes;
151 /* Read in the segments */
152 image->nr_segments = nr_segments;
153 segment_bytes = nr_segments * sizeof(*segments);
154 ret = copy_from_user(image->segment, segments, segment_bytes);
161 static int sanity_check_segment_list(struct kimage *image)
164 unsigned long nr_segments = image->nr_segments;
167 * Verify we have good destination addresses. The caller is
168 * responsible for making certain we don't attempt to load
169 * the new image into invalid or reserved areas of RAM. This
170 * just verifies it is an address we can use.
172 * Since the kernel does everything in page size chunks ensure
173 * the destination addresses are page aligned. Too many
174 * special cases crop of when we don't do this. The most
175 * insidious is getting overlapping destination addresses
176 * simply because addresses are changed to page size
179 result = -EADDRNOTAVAIL;
180 for (i = 0; i < nr_segments; i++) {
181 unsigned long mstart, mend;
183 mstart = image->segment[i].mem;
184 mend = mstart + image->segment[i].memsz;
185 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
187 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
191 /* Verify our destination addresses do not overlap.
192 * If we alloed overlapping destination addresses
193 * through very weird things can happen with no
194 * easy explanation as one segment stops on another.
197 for (i = 0; i < nr_segments; i++) {
198 unsigned long mstart, mend;
201 mstart = image->segment[i].mem;
202 mend = mstart + image->segment[i].memsz;
203 for (j = 0; j < i; j++) {
204 unsigned long pstart, pend;
205 pstart = image->segment[j].mem;
206 pend = pstart + image->segment[j].memsz;
207 /* Do the segments overlap ? */
208 if ((mend > pstart) && (mstart < pend))
213 /* Ensure our buffer sizes are strictly less than
214 * our memory sizes. This should always be the case,
215 * and it is easier to check up front than to be surprised
219 for (i = 0; i < nr_segments; i++) {
220 if (image->segment[i].bufsz > image->segment[i].memsz)
225 * Verify we have good destination addresses. Normally
226 * the caller is responsible for making certain we don't
227 * attempt to load the new image into invalid or reserved
228 * areas of RAM. But crash kernels are preloaded into a
229 * reserved area of ram. We must ensure the addresses
230 * are in the reserved area otherwise preloading the
231 * kernel could corrupt things.
234 if (image->type == KEXEC_TYPE_CRASH) {
235 result = -EADDRNOTAVAIL;
236 for (i = 0; i < nr_segments; i++) {
237 unsigned long mstart, mend;
239 mstart = image->segment[i].mem;
240 mend = mstart + image->segment[i].memsz - 1;
241 /* Ensure we are within the crash kernel limits */
242 if ((mstart < crashk_res.start) ||
243 (mend > crashk_res.end))
251 static struct kimage *do_kimage_alloc_init(void)
253 struct kimage *image;
255 /* Allocate a controlling structure */
256 image = kzalloc(sizeof(*image), GFP_KERNEL);
261 image->entry = &image->head;
262 image->last_entry = &image->head;
263 image->control_page = ~0; /* By default this does not apply */
264 image->type = KEXEC_TYPE_DEFAULT;
266 /* Initialize the list of control pages */
267 INIT_LIST_HEAD(&image->control_pages);
269 /* Initialize the list of destination pages */
270 INIT_LIST_HEAD(&image->dest_pages);
272 /* Initialize the list of unusable pages */
273 INIT_LIST_HEAD(&image->unusable_pages);
278 static void kimage_free_page_list(struct list_head *list);
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 #ifdef CONFIG_KEXEC_FILE
347 static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len)
349 struct fd f = fdget(fd);
358 ret = vfs_getattr(&f.file->f_path, &stat);
362 if (stat.size > INT_MAX) {
367 /* Don't hand 0 to vmalloc, it whines. */
368 if (stat.size == 0) {
373 *buf = vmalloc(stat.size);
380 while (pos < stat.size) {
381 bytes = kernel_read(f.file, pos, (char *)(*buf) + pos,
394 if (pos != stat.size) {
406 /* Architectures can provide this probe function */
407 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
408 unsigned long buf_len)
413 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
415 return ERR_PTR(-ENOEXEC);
418 void __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
422 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
423 unsigned long buf_len)
425 return -EKEYREJECTED;
428 /* Apply relocations of type RELA */
430 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
433 pr_err("RELA relocation unsupported.\n");
437 /* Apply relocations of type REL */
439 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
442 pr_err("REL relocation unsupported.\n");
447 * Free up memory used by kernel, initrd, and command line. This is temporary
448 * memory allocation which is not needed any more after these buffers have
449 * been loaded into separate segments and have been copied elsewhere.
451 static void kimage_file_post_load_cleanup(struct kimage *image)
453 struct purgatory_info *pi = &image->purgatory_info;
455 vfree(image->kernel_buf);
456 image->kernel_buf = NULL;
458 vfree(image->initrd_buf);
459 image->initrd_buf = NULL;
461 kfree(image->cmdline_buf);
462 image->cmdline_buf = NULL;
464 vfree(pi->purgatory_buf);
465 pi->purgatory_buf = NULL;
470 /* See if architecture has anything to cleanup post load */
471 arch_kimage_file_post_load_cleanup(image);
474 * Above call should have called into bootloader to free up
475 * any data stored in kimage->image_loader_data. It should
476 * be ok now to free it up.
478 kfree(image->image_loader_data);
479 image->image_loader_data = NULL;
483 * In file mode list of segments is prepared by kernel. Copy relevant
484 * data from user space, do error checking, prepare segment list
487 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
488 const char __user *cmdline_ptr,
489 unsigned long cmdline_len, unsigned flags)
494 ret = copy_file_from_fd(kernel_fd, &image->kernel_buf,
495 &image->kernel_buf_len);
499 /* Call arch image probe handlers */
500 ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
501 image->kernel_buf_len);
506 #ifdef CONFIG_KEXEC_VERIFY_SIG
507 ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
508 image->kernel_buf_len);
510 pr_debug("kernel signature verification failed.\n");
513 pr_debug("kernel signature verification successful.\n");
515 /* It is possible that there no initramfs is being loaded */
516 if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
517 ret = copy_file_from_fd(initrd_fd, &image->initrd_buf,
518 &image->initrd_buf_len);
524 image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
525 if (!image->cmdline_buf) {
530 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
537 image->cmdline_buf_len = cmdline_len;
539 /* command line should be a string with last byte null */
540 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
546 /* Call arch image load handlers */
547 ldata = arch_kexec_kernel_image_load(image);
550 ret = PTR_ERR(ldata);
554 image->image_loader_data = ldata;
556 /* In case of error, free up all allocated memory in this function */
558 kimage_file_post_load_cleanup(image);
563 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
564 int initrd_fd, const char __user *cmdline_ptr,
565 unsigned long cmdline_len, unsigned long flags)
568 struct kimage *image;
569 bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
571 image = do_kimage_alloc_init();
575 image->file_mode = 1;
577 if (kexec_on_panic) {
578 /* Enable special crash kernel control page alloc policy. */
579 image->control_page = crashk_res.start;
580 image->type = KEXEC_TYPE_CRASH;
583 ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
584 cmdline_ptr, cmdline_len, flags);
588 ret = sanity_check_segment_list(image);
590 goto out_free_post_load_bufs;
593 image->control_code_page = kimage_alloc_control_pages(image,
594 get_order(KEXEC_CONTROL_PAGE_SIZE));
595 if (!image->control_code_page) {
596 pr_err("Could not allocate control_code_buffer\n");
597 goto out_free_post_load_bufs;
600 if (!kexec_on_panic) {
601 image->swap_page = kimage_alloc_control_pages(image, 0);
602 if (!image->swap_page) {
603 pr_err("Could not allocate swap buffer\n");
604 goto out_free_control_pages;
610 out_free_control_pages:
611 kimage_free_page_list(&image->control_pages);
612 out_free_post_load_bufs:
613 kimage_file_post_load_cleanup(image);
618 #else /* CONFIG_KEXEC_FILE */
619 static inline void kimage_file_post_load_cleanup(struct kimage *image) { }
620 #endif /* CONFIG_KEXEC_FILE */
622 static int kimage_is_destination_range(struct kimage *image,
628 for (i = 0; i < image->nr_segments; i++) {
629 unsigned long mstart, mend;
631 mstart = image->segment[i].mem;
632 mend = mstart + image->segment[i].memsz;
633 if ((end > mstart) && (start < mend))
640 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
644 pages = alloc_pages(gfp_mask, order);
646 unsigned int count, i;
647 pages->mapping = NULL;
648 set_page_private(pages, order);
650 for (i = 0; i < count; i++)
651 SetPageReserved(pages + i);
657 static void kimage_free_pages(struct page *page)
659 unsigned int order, count, i;
661 order = page_private(page);
663 for (i = 0; i < count; i++)
664 ClearPageReserved(page + i);
665 __free_pages(page, order);
668 static void kimage_free_page_list(struct list_head *list)
670 struct list_head *pos, *next;
672 list_for_each_safe(pos, next, list) {
675 page = list_entry(pos, struct page, lru);
676 list_del(&page->lru);
677 kimage_free_pages(page);
681 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
684 /* Control pages are special, they are the intermediaries
685 * that are needed while we copy the rest of the pages
686 * to their final resting place. As such they must
687 * not conflict with either the destination addresses
688 * or memory the kernel is already using.
690 * The only case where we really need more than one of
691 * these are for architectures where we cannot disable
692 * the MMU and must instead generate an identity mapped
693 * page table for all of the memory.
695 * At worst this runs in O(N) of the image size.
697 struct list_head extra_pages;
702 INIT_LIST_HEAD(&extra_pages);
704 /* Loop while I can allocate a page and the page allocated
705 * is a destination page.
708 unsigned long pfn, epfn, addr, eaddr;
710 pages = kimage_alloc_pages(GFP_KERNEL, order);
713 pfn = page_to_pfn(pages);
715 addr = pfn << PAGE_SHIFT;
716 eaddr = epfn << PAGE_SHIFT;
717 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
718 kimage_is_destination_range(image, addr, eaddr)) {
719 list_add(&pages->lru, &extra_pages);
725 /* Remember the allocated page... */
726 list_add(&pages->lru, &image->control_pages);
728 /* Because the page is already in it's destination
729 * location we will never allocate another page at
730 * that address. Therefore kimage_alloc_pages
731 * will not return it (again) and we don't need
732 * to give it an entry in image->segment[].
735 /* Deal with the destination pages I have inadvertently allocated.
737 * Ideally I would convert multi-page allocations into single
738 * page allocations, and add everything to image->dest_pages.
740 * For now it is simpler to just free the pages.
742 kimage_free_page_list(&extra_pages);
747 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
750 /* Control pages are special, they are the intermediaries
751 * that are needed while we copy the rest of the pages
752 * to their final resting place. As such they must
753 * not conflict with either the destination addresses
754 * or memory the kernel is already using.
756 * Control pages are also the only pags we must allocate
757 * when loading a crash kernel. All of the other pages
758 * are specified by the segments and we just memcpy
759 * into them directly.
761 * The only case where we really need more than one of
762 * these are for architectures where we cannot disable
763 * the MMU and must instead generate an identity mapped
764 * page table for all of the memory.
766 * Given the low demand this implements a very simple
767 * allocator that finds the first hole of the appropriate
768 * size in the reserved memory region, and allocates all
769 * of the memory up to and including the hole.
771 unsigned long hole_start, hole_end, size;
775 size = (1 << order) << PAGE_SHIFT;
776 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
777 hole_end = hole_start + size - 1;
778 while (hole_end <= crashk_res.end) {
781 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
783 /* See if I overlap any of the segments */
784 for (i = 0; i < image->nr_segments; i++) {
785 unsigned long mstart, mend;
787 mstart = image->segment[i].mem;
788 mend = mstart + image->segment[i].memsz - 1;
789 if ((hole_end >= mstart) && (hole_start <= mend)) {
790 /* Advance the hole to the end of the segment */
791 hole_start = (mend + (size - 1)) & ~(size - 1);
792 hole_end = hole_start + size - 1;
796 /* If I don't overlap any segments I have found my hole! */
797 if (i == image->nr_segments) {
798 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
803 image->control_page = hole_end;
809 struct page *kimage_alloc_control_pages(struct kimage *image,
812 struct page *pages = NULL;
814 switch (image->type) {
815 case KEXEC_TYPE_DEFAULT:
816 pages = kimage_alloc_normal_control_pages(image, order);
818 case KEXEC_TYPE_CRASH:
819 pages = kimage_alloc_crash_control_pages(image, order);
826 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
828 if (*image->entry != 0)
831 if (image->entry == image->last_entry) {
832 kimage_entry_t *ind_page;
835 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
839 ind_page = page_address(page);
840 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
841 image->entry = ind_page;
842 image->last_entry = ind_page +
843 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
845 *image->entry = entry;
852 static int kimage_set_destination(struct kimage *image,
853 unsigned long destination)
857 destination &= PAGE_MASK;
858 result = kimage_add_entry(image, destination | IND_DESTINATION);
864 static int kimage_add_page(struct kimage *image, unsigned long page)
869 result = kimage_add_entry(image, page | IND_SOURCE);
875 static void kimage_free_extra_pages(struct kimage *image)
877 /* Walk through and free any extra destination pages I may have */
878 kimage_free_page_list(&image->dest_pages);
880 /* Walk through and free any unusable pages I have cached */
881 kimage_free_page_list(&image->unusable_pages);
884 static void kimage_terminate(struct kimage *image)
886 if (*image->entry != 0)
889 *image->entry = IND_DONE;
892 #define for_each_kimage_entry(image, ptr, entry) \
893 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
894 ptr = (entry & IND_INDIRECTION) ? \
895 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
897 static void kimage_free_entry(kimage_entry_t entry)
901 page = pfn_to_page(entry >> PAGE_SHIFT);
902 kimage_free_pages(page);
905 static void kimage_free(struct kimage *image)
907 kimage_entry_t *ptr, entry;
908 kimage_entry_t ind = 0;
913 kimage_free_extra_pages(image);
914 for_each_kimage_entry(image, ptr, entry) {
915 if (entry & IND_INDIRECTION) {
916 /* Free the previous indirection page */
917 if (ind & IND_INDIRECTION)
918 kimage_free_entry(ind);
919 /* Save this indirection page until we are
923 } else if (entry & IND_SOURCE)
924 kimage_free_entry(entry);
926 /* Free the final indirection page */
927 if (ind & IND_INDIRECTION)
928 kimage_free_entry(ind);
930 /* Handle any machine specific cleanup */
931 machine_kexec_cleanup(image);
933 /* Free the kexec control pages... */
934 kimage_free_page_list(&image->control_pages);
937 * Free up any temporary buffers allocated. This might hit if
938 * error occurred much later after buffer allocation.
940 if (image->file_mode)
941 kimage_file_post_load_cleanup(image);
946 static kimage_entry_t *kimage_dst_used(struct kimage *image,
949 kimage_entry_t *ptr, entry;
950 unsigned long destination = 0;
952 for_each_kimage_entry(image, ptr, entry) {
953 if (entry & IND_DESTINATION)
954 destination = entry & PAGE_MASK;
955 else if (entry & IND_SOURCE) {
956 if (page == destination)
958 destination += PAGE_SIZE;
965 static struct page *kimage_alloc_page(struct kimage *image,
967 unsigned long destination)
970 * Here we implement safeguards to ensure that a source page
971 * is not copied to its destination page before the data on
972 * the destination page is no longer useful.
974 * To do this we maintain the invariant that a source page is
975 * either its own destination page, or it is not a
976 * destination page at all.
978 * That is slightly stronger than required, but the proof
979 * that no problems will not occur is trivial, and the
980 * implementation is simply to verify.
982 * When allocating all pages normally this algorithm will run
983 * in O(N) time, but in the worst case it will run in O(N^2)
984 * time. If the runtime is a problem the data structures can
991 * Walk through the list of destination pages, and see if I
994 list_for_each_entry(page, &image->dest_pages, lru) {
995 addr = page_to_pfn(page) << PAGE_SHIFT;
996 if (addr == destination) {
997 list_del(&page->lru);
1003 kimage_entry_t *old;
1005 /* Allocate a page, if we run out of memory give up */
1006 page = kimage_alloc_pages(gfp_mask, 0);
1009 /* If the page cannot be used file it away */
1010 if (page_to_pfn(page) >
1011 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
1012 list_add(&page->lru, &image->unusable_pages);
1015 addr = page_to_pfn(page) << PAGE_SHIFT;
1017 /* If it is the destination page we want use it */
1018 if (addr == destination)
1021 /* If the page is not a destination page use it */
1022 if (!kimage_is_destination_range(image, addr,
1027 * I know that the page is someones destination page.
1028 * See if there is already a source page for this
1029 * destination page. And if so swap the source pages.
1031 old = kimage_dst_used(image, addr);
1034 unsigned long old_addr;
1035 struct page *old_page;
1037 old_addr = *old & PAGE_MASK;
1038 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
1039 copy_highpage(page, old_page);
1040 *old = addr | (*old & ~PAGE_MASK);
1042 /* The old page I have found cannot be a
1043 * destination page, so return it if it's
1044 * gfp_flags honor the ones passed in.
1046 if (!(gfp_mask & __GFP_HIGHMEM) &&
1047 PageHighMem(old_page)) {
1048 kimage_free_pages(old_page);
1055 /* Place the page on the destination list I
1056 * will use it later.
1058 list_add(&page->lru, &image->dest_pages);
1065 static int kimage_load_normal_segment(struct kimage *image,
1066 struct kexec_segment *segment)
1068 unsigned long maddr;
1069 size_t ubytes, mbytes;
1071 unsigned char __user *buf = NULL;
1072 unsigned char *kbuf = NULL;
1075 if (image->file_mode)
1076 kbuf = segment->kbuf;
1079 ubytes = segment->bufsz;
1080 mbytes = segment->memsz;
1081 maddr = segment->mem;
1083 result = kimage_set_destination(image, maddr);
1090 size_t uchunk, mchunk;
1092 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
1097 result = kimage_add_page(image, page_to_pfn(page)
1103 /* Start with a clear page */
1105 ptr += maddr & ~PAGE_MASK;
1106 mchunk = min_t(size_t, mbytes,
1107 PAGE_SIZE - (maddr & ~PAGE_MASK));
1108 uchunk = min(ubytes, mchunk);
1110 /* For file based kexec, source pages are in kernel memory */
1111 if (image->file_mode)
1112 memcpy(ptr, kbuf, uchunk);
1114 result = copy_from_user(ptr, buf, uchunk);
1122 if (image->file_mode)
1132 static int kimage_load_crash_segment(struct kimage *image,
1133 struct kexec_segment *segment)
1135 /* For crash dumps kernels we simply copy the data from
1136 * user space to it's destination.
1137 * We do things a page at a time for the sake of kmap.
1139 unsigned long maddr;
1140 size_t ubytes, mbytes;
1142 unsigned char __user *buf = NULL;
1143 unsigned char *kbuf = NULL;
1146 if (image->file_mode)
1147 kbuf = segment->kbuf;
1150 ubytes = segment->bufsz;
1151 mbytes = segment->memsz;
1152 maddr = segment->mem;
1156 size_t uchunk, mchunk;
1158 page = pfn_to_page(maddr >> PAGE_SHIFT);
1164 ptr += maddr & ~PAGE_MASK;
1165 mchunk = min_t(size_t, mbytes,
1166 PAGE_SIZE - (maddr & ~PAGE_MASK));
1167 uchunk = min(ubytes, mchunk);
1168 if (mchunk > uchunk) {
1169 /* Zero the trailing part of the page */
1170 memset(ptr + uchunk, 0, mchunk - uchunk);
1173 /* For file based kexec, source pages are in kernel memory */
1174 if (image->file_mode)
1175 memcpy(ptr, kbuf, uchunk);
1177 result = copy_from_user(ptr, buf, uchunk);
1178 kexec_flush_icache_page(page);
1186 if (image->file_mode)
1196 static int kimage_load_segment(struct kimage *image,
1197 struct kexec_segment *segment)
1199 int result = -ENOMEM;
1201 switch (image->type) {
1202 case KEXEC_TYPE_DEFAULT:
1203 result = kimage_load_normal_segment(image, segment);
1205 case KEXEC_TYPE_CRASH:
1206 result = kimage_load_crash_segment(image, segment);
1214 * Exec Kernel system call: for obvious reasons only root may call it.
1216 * This call breaks up into three pieces.
1217 * - A generic part which loads the new kernel from the current
1218 * address space, and very carefully places the data in the
1221 * - A generic part that interacts with the kernel and tells all of
1222 * the devices to shut down. Preventing on-going dmas, and placing
1223 * the devices in a consistent state so a later kernel can
1224 * reinitialize them.
1226 * - A machine specific part that includes the syscall number
1227 * and then copies the image to it's final destination. And
1228 * jumps into the image at entry.
1230 * kexec does not sync, or unmount filesystems so if you need
1231 * that to happen you need to do that yourself.
1233 struct kimage *kexec_image;
1234 struct kimage *kexec_crash_image;
1235 int kexec_load_disabled;
1237 static DEFINE_MUTEX(kexec_mutex);
1239 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
1240 struct kexec_segment __user *, segments, unsigned long, flags)
1242 struct kimage **dest_image, *image;
1245 /* We only trust the superuser with rebooting the system. */
1246 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1250 * Verify we have a legal set of flags
1251 * This leaves us room for future extensions.
1253 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
1256 /* Verify we are on the appropriate architecture */
1257 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
1258 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
1261 /* Put an artificial cap on the number
1262 * of segments passed to kexec_load.
1264 if (nr_segments > KEXEC_SEGMENT_MAX)
1270 /* Because we write directly to the reserved memory
1271 * region when loading crash kernels we need a mutex here to
1272 * prevent multiple crash kernels from attempting to load
1273 * simultaneously, and to prevent a crash kernel from loading
1274 * over the top of a in use crash kernel.
1276 * KISS: always take the mutex.
1278 if (!mutex_trylock(&kexec_mutex))
1281 dest_image = &kexec_image;
1282 if (flags & KEXEC_ON_CRASH)
1283 dest_image = &kexec_crash_image;
1284 if (nr_segments > 0) {
1287 /* Loading another kernel to reboot into */
1288 if ((flags & KEXEC_ON_CRASH) == 0)
1289 result = kimage_alloc_init(&image, entry, nr_segments,
1291 /* Loading another kernel to switch to if this one crashes */
1292 else if (flags & KEXEC_ON_CRASH) {
1293 /* Free any current crash dump kernel before
1296 kimage_free(xchg(&kexec_crash_image, NULL));
1297 result = kimage_alloc_init(&image, entry, nr_segments,
1299 crash_map_reserved_pages();
1304 if (flags & KEXEC_PRESERVE_CONTEXT)
1305 image->preserve_context = 1;
1306 result = machine_kexec_prepare(image);
1310 for (i = 0; i < nr_segments; i++) {
1311 result = kimage_load_segment(image, &image->segment[i]);
1315 kimage_terminate(image);
1316 if (flags & KEXEC_ON_CRASH)
1317 crash_unmap_reserved_pages();
1319 /* Install the new kernel, and Uninstall the old */
1320 image = xchg(dest_image, image);
1323 mutex_unlock(&kexec_mutex);
1330 * Add and remove page tables for crashkernel memory
1332 * Provide an empty default implementation here -- architecture
1333 * code may override this
1335 void __weak crash_map_reserved_pages(void)
1338 void __weak crash_unmap_reserved_pages(void)
1341 #ifdef CONFIG_COMPAT
1342 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1343 compat_ulong_t, nr_segments,
1344 struct compat_kexec_segment __user *, segments,
1345 compat_ulong_t, flags)
1347 struct compat_kexec_segment in;
1348 struct kexec_segment out, __user *ksegments;
1349 unsigned long i, result;
1351 /* Don't allow clients that don't understand the native
1352 * architecture to do anything.
1354 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1357 if (nr_segments > KEXEC_SEGMENT_MAX)
1360 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1361 for (i = 0; i < nr_segments; i++) {
1362 result = copy_from_user(&in, &segments[i], sizeof(in));
1366 out.buf = compat_ptr(in.buf);
1367 out.bufsz = in.bufsz;
1369 out.memsz = in.memsz;
1371 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1376 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1380 #ifdef CONFIG_KEXEC_FILE
1381 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
1382 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
1383 unsigned long, flags)
1386 struct kimage **dest_image, *image;
1388 /* We only trust the superuser with rebooting the system. */
1389 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1392 /* Make sure we have a legal set of flags */
1393 if (flags != (flags & KEXEC_FILE_FLAGS))
1398 if (!mutex_trylock(&kexec_mutex))
1401 dest_image = &kexec_image;
1402 if (flags & KEXEC_FILE_ON_CRASH)
1403 dest_image = &kexec_crash_image;
1405 if (flags & KEXEC_FILE_UNLOAD)
1409 * In case of crash, new kernel gets loaded in reserved region. It is
1410 * same memory where old crash kernel might be loaded. Free any
1411 * current crash dump kernel before we corrupt it.
1413 if (flags & KEXEC_FILE_ON_CRASH)
1414 kimage_free(xchg(&kexec_crash_image, NULL));
1416 ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
1417 cmdline_len, flags);
1421 ret = machine_kexec_prepare(image);
1425 ret = kexec_calculate_store_digests(image);
1429 for (i = 0; i < image->nr_segments; i++) {
1430 struct kexec_segment *ksegment;
1432 ksegment = &image->segment[i];
1433 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
1434 i, ksegment->buf, ksegment->bufsz, ksegment->mem,
1437 ret = kimage_load_segment(image, &image->segment[i]);
1442 kimage_terminate(image);
1445 * Free up any temporary buffers allocated which are not needed
1446 * after image has been loaded
1448 kimage_file_post_load_cleanup(image);
1450 image = xchg(dest_image, image);
1452 mutex_unlock(&kexec_mutex);
1457 #endif /* CONFIG_KEXEC_FILE */
1459 void crash_kexec(struct pt_regs *regs)
1461 /* Take the kexec_mutex here to prevent sys_kexec_load
1462 * running on one cpu from replacing the crash kernel
1463 * we are using after a panic on a different cpu.
1465 * If the crash kernel was not located in a fixed area
1466 * of memory the xchg(&kexec_crash_image) would be
1467 * sufficient. But since I reuse the memory...
1469 if (mutex_trylock(&kexec_mutex)) {
1470 if (kexec_crash_image) {
1471 struct pt_regs fixed_regs;
1473 crash_setup_regs(&fixed_regs, regs);
1474 crash_save_vmcoreinfo();
1475 machine_crash_shutdown(&fixed_regs);
1476 machine_kexec(kexec_crash_image);
1478 mutex_unlock(&kexec_mutex);
1482 size_t crash_get_memory_size(void)
1485 mutex_lock(&kexec_mutex);
1486 if (crashk_res.end != crashk_res.start)
1487 size = resource_size(&crashk_res);
1488 mutex_unlock(&kexec_mutex);
1492 void __weak crash_free_reserved_phys_range(unsigned long begin,
1497 for (addr = begin; addr < end; addr += PAGE_SIZE)
1498 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1501 int crash_shrink_memory(unsigned long new_size)
1504 unsigned long start, end;
1505 unsigned long old_size;
1506 struct resource *ram_res;
1508 mutex_lock(&kexec_mutex);
1510 if (kexec_crash_image) {
1514 start = crashk_res.start;
1515 end = crashk_res.end;
1516 old_size = (end == 0) ? 0 : end - start + 1;
1517 if (new_size >= old_size) {
1518 ret = (new_size == old_size) ? 0 : -EINVAL;
1522 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1528 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1529 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1531 crash_map_reserved_pages();
1532 crash_free_reserved_phys_range(end, crashk_res.end);
1534 if ((start == end) && (crashk_res.parent != NULL))
1535 release_resource(&crashk_res);
1537 ram_res->start = end;
1538 ram_res->end = crashk_res.end;
1539 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1540 ram_res->name = "System RAM";
1542 crashk_res.end = end - 1;
1544 insert_resource(&iomem_resource, ram_res);
1545 crash_unmap_reserved_pages();
1548 mutex_unlock(&kexec_mutex);
1552 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1555 struct elf_note note;
1557 note.n_namesz = strlen(name) + 1;
1558 note.n_descsz = data_len;
1560 memcpy(buf, ¬e, sizeof(note));
1561 buf += (sizeof(note) + 3)/4;
1562 memcpy(buf, name, note.n_namesz);
1563 buf += (note.n_namesz + 3)/4;
1564 memcpy(buf, data, note.n_descsz);
1565 buf += (note.n_descsz + 3)/4;
1570 static void final_note(u32 *buf)
1572 struct elf_note note;
1577 memcpy(buf, ¬e, sizeof(note));
1580 void crash_save_cpu(struct pt_regs *regs, int cpu)
1582 struct elf_prstatus prstatus;
1585 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1588 /* Using ELF notes here is opportunistic.
1589 * I need a well defined structure format
1590 * for the data I pass, and I need tags
1591 * on the data to indicate what information I have
1592 * squirrelled away. ELF notes happen to provide
1593 * all of that, so there is no need to invent something new.
1595 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1598 memset(&prstatus, 0, sizeof(prstatus));
1599 prstatus.pr_pid = current->pid;
1600 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1601 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1602 &prstatus, sizeof(prstatus));
1606 static int __init crash_notes_memory_init(void)
1608 /* Allocate memory for saving cpu registers. */
1609 crash_notes = alloc_percpu(note_buf_t);
1611 pr_warn("Kexec: Memory allocation for saving cpu register states failed\n");
1616 subsys_initcall(crash_notes_memory_init);
1620 * parsing the "crashkernel" commandline
1622 * this code is intended to be called from architecture specific code
1627 * This function parses command lines in the format
1629 * crashkernel=ramsize-range:size[,...][@offset]
1631 * The function returns 0 on success and -EINVAL on failure.
1633 static int __init parse_crashkernel_mem(char *cmdline,
1634 unsigned long long system_ram,
1635 unsigned long long *crash_size,
1636 unsigned long long *crash_base)
1638 char *cur = cmdline, *tmp;
1640 /* for each entry of the comma-separated list */
1642 unsigned long long start, end = ULLONG_MAX, size;
1644 /* get the start of the range */
1645 start = memparse(cur, &tmp);
1647 pr_warn("crashkernel: Memory value expected\n");
1652 pr_warn("crashkernel: '-' expected\n");
1657 /* if no ':' is here, than we read the end */
1659 end = memparse(cur, &tmp);
1661 pr_warn("crashkernel: Memory value expected\n");
1666 pr_warn("crashkernel: end <= start\n");
1672 pr_warn("crashkernel: ':' expected\n");
1677 size = memparse(cur, &tmp);
1679 pr_warn("Memory value expected\n");
1683 if (size >= system_ram) {
1684 pr_warn("crashkernel: invalid size\n");
1689 if (system_ram >= start && system_ram < end) {
1693 } while (*cur++ == ',');
1695 if (*crash_size > 0) {
1696 while (*cur && *cur != ' ' && *cur != '@')
1700 *crash_base = memparse(cur, &tmp);
1702 pr_warn("Memory value expected after '@'\n");
1712 * That function parses "simple" (old) crashkernel command lines like
1714 * crashkernel=size[@offset]
1716 * It returns 0 on success and -EINVAL on failure.
1718 static int __init parse_crashkernel_simple(char *cmdline,
1719 unsigned long long *crash_size,
1720 unsigned long long *crash_base)
1722 char *cur = cmdline;
1724 *crash_size = memparse(cmdline, &cur);
1725 if (cmdline == cur) {
1726 pr_warn("crashkernel: memory value expected\n");
1731 *crash_base = memparse(cur+1, &cur);
1732 else if (*cur != ' ' && *cur != '\0') {
1733 pr_warn("crashkernel: unrecognized char\n");
1740 #define SUFFIX_HIGH 0
1741 #define SUFFIX_LOW 1
1742 #define SUFFIX_NULL 2
1743 static __initdata char *suffix_tbl[] = {
1744 [SUFFIX_HIGH] = ",high",
1745 [SUFFIX_LOW] = ",low",
1746 [SUFFIX_NULL] = NULL,
1750 * That function parses "suffix" crashkernel command lines like
1752 * crashkernel=size,[high|low]
1754 * It returns 0 on success and -EINVAL on failure.
1756 static int __init parse_crashkernel_suffix(char *cmdline,
1757 unsigned long long *crash_size,
1760 char *cur = cmdline;
1762 *crash_size = memparse(cmdline, &cur);
1763 if (cmdline == cur) {
1764 pr_warn("crashkernel: memory value expected\n");
1768 /* check with suffix */
1769 if (strncmp(cur, suffix, strlen(suffix))) {
1770 pr_warn("crashkernel: unrecognized char\n");
1773 cur += strlen(suffix);
1774 if (*cur != ' ' && *cur != '\0') {
1775 pr_warn("crashkernel: unrecognized char\n");
1782 static __init char *get_last_crashkernel(char *cmdline,
1786 char *p = cmdline, *ck_cmdline = NULL;
1788 /* find crashkernel and use the last one if there are more */
1789 p = strstr(p, name);
1791 char *end_p = strchr(p, ' ');
1795 end_p = p + strlen(p);
1800 /* skip the one with any known suffix */
1801 for (i = 0; suffix_tbl[i]; i++) {
1802 q = end_p - strlen(suffix_tbl[i]);
1803 if (!strncmp(q, suffix_tbl[i],
1804 strlen(suffix_tbl[i])))
1809 q = end_p - strlen(suffix);
1810 if (!strncmp(q, suffix, strlen(suffix)))
1814 p = strstr(p+1, name);
1823 static int __init __parse_crashkernel(char *cmdline,
1824 unsigned long long system_ram,
1825 unsigned long long *crash_size,
1826 unsigned long long *crash_base,
1830 char *first_colon, *first_space;
1833 BUG_ON(!crash_size || !crash_base);
1837 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1842 ck_cmdline += strlen(name);
1845 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1848 * if the commandline contains a ':', then that's the extended
1849 * syntax -- if not, it must be the classic syntax
1851 first_colon = strchr(ck_cmdline, ':');
1852 first_space = strchr(ck_cmdline, ' ');
1853 if (first_colon && (!first_space || first_colon < first_space))
1854 return parse_crashkernel_mem(ck_cmdline, system_ram,
1855 crash_size, crash_base);
1857 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1861 * That function is the entry point for command line parsing and should be
1862 * called from the arch-specific code.
1864 int __init parse_crashkernel(char *cmdline,
1865 unsigned long long system_ram,
1866 unsigned long long *crash_size,
1867 unsigned long long *crash_base)
1869 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1870 "crashkernel=", NULL);
1873 int __init parse_crashkernel_high(char *cmdline,
1874 unsigned long long system_ram,
1875 unsigned long long *crash_size,
1876 unsigned long long *crash_base)
1878 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1879 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1882 int __init parse_crashkernel_low(char *cmdline,
1883 unsigned long long system_ram,
1884 unsigned long long *crash_size,
1885 unsigned long long *crash_base)
1887 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1888 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1891 static void update_vmcoreinfo_note(void)
1893 u32 *buf = vmcoreinfo_note;
1895 if (!vmcoreinfo_size)
1897 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1902 void crash_save_vmcoreinfo(void)
1904 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1905 update_vmcoreinfo_note();
1908 void vmcoreinfo_append_str(const char *fmt, ...)
1914 va_start(args, fmt);
1915 r = vscnprintf(buf, sizeof(buf), fmt, args);
1918 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1920 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1922 vmcoreinfo_size += r;
1926 * provide an empty default implementation here -- architecture
1927 * code may override this
1929 void __weak arch_crash_save_vmcoreinfo(void)
1932 unsigned long __weak paddr_vmcoreinfo_note(void)
1934 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1937 static int __init crash_save_vmcoreinfo_init(void)
1939 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1940 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1942 VMCOREINFO_SYMBOL(init_uts_ns);
1943 VMCOREINFO_SYMBOL(node_online_map);
1945 VMCOREINFO_SYMBOL(swapper_pg_dir);
1947 VMCOREINFO_SYMBOL(_stext);
1948 VMCOREINFO_SYMBOL(vmap_area_list);
1950 #ifndef CONFIG_NEED_MULTIPLE_NODES
1951 VMCOREINFO_SYMBOL(mem_map);
1952 VMCOREINFO_SYMBOL(contig_page_data);
1954 #ifdef CONFIG_SPARSEMEM
1955 VMCOREINFO_SYMBOL(mem_section);
1956 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1957 VMCOREINFO_STRUCT_SIZE(mem_section);
1958 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1960 VMCOREINFO_STRUCT_SIZE(page);
1961 VMCOREINFO_STRUCT_SIZE(pglist_data);
1962 VMCOREINFO_STRUCT_SIZE(zone);
1963 VMCOREINFO_STRUCT_SIZE(free_area);
1964 VMCOREINFO_STRUCT_SIZE(list_head);
1965 VMCOREINFO_SIZE(nodemask_t);
1966 VMCOREINFO_OFFSET(page, flags);
1967 VMCOREINFO_OFFSET(page, _count);
1968 VMCOREINFO_OFFSET(page, mapping);
1969 VMCOREINFO_OFFSET(page, lru);
1970 VMCOREINFO_OFFSET(page, _mapcount);
1971 VMCOREINFO_OFFSET(page, private);
1972 VMCOREINFO_OFFSET(pglist_data, node_zones);
1973 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1974 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1975 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1977 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1978 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1979 VMCOREINFO_OFFSET(pglist_data, node_id);
1980 VMCOREINFO_OFFSET(zone, free_area);
1981 VMCOREINFO_OFFSET(zone, vm_stat);
1982 VMCOREINFO_OFFSET(zone, spanned_pages);
1983 VMCOREINFO_OFFSET(free_area, free_list);
1984 VMCOREINFO_OFFSET(list_head, next);
1985 VMCOREINFO_OFFSET(list_head, prev);
1986 VMCOREINFO_OFFSET(vmap_area, va_start);
1987 VMCOREINFO_OFFSET(vmap_area, list);
1988 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1989 log_buf_kexec_setup();
1990 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1991 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1992 VMCOREINFO_NUMBER(PG_lru);
1993 VMCOREINFO_NUMBER(PG_private);
1994 VMCOREINFO_NUMBER(PG_swapcache);
1995 VMCOREINFO_NUMBER(PG_slab);
1996 #ifdef CONFIG_MEMORY_FAILURE
1997 VMCOREINFO_NUMBER(PG_hwpoison);
1999 VMCOREINFO_NUMBER(PG_head_mask);
2000 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
2001 #ifdef CONFIG_HUGETLBFS
2002 VMCOREINFO_SYMBOL(free_huge_page);
2005 arch_crash_save_vmcoreinfo();
2006 update_vmcoreinfo_note();
2011 subsys_initcall(crash_save_vmcoreinfo_init);
2013 #ifdef CONFIG_KEXEC_FILE
2014 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
2015 struct kexec_buf *kbuf)
2017 struct kimage *image = kbuf->image;
2018 unsigned long temp_start, temp_end;
2020 temp_end = min(end, kbuf->buf_max);
2021 temp_start = temp_end - kbuf->memsz;
2024 /* align down start */
2025 temp_start = temp_start & (~(kbuf->buf_align - 1));
2027 if (temp_start < start || temp_start < kbuf->buf_min)
2030 temp_end = temp_start + kbuf->memsz - 1;
2033 * Make sure this does not conflict with any of existing
2036 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2037 temp_start = temp_start - PAGE_SIZE;
2041 /* We found a suitable memory range */
2045 /* If we are here, we found a suitable memory range */
2046 kbuf->mem = temp_start;
2048 /* Success, stop navigating through remaining System RAM ranges */
2052 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
2053 struct kexec_buf *kbuf)
2055 struct kimage *image = kbuf->image;
2056 unsigned long temp_start, temp_end;
2058 temp_start = max(start, kbuf->buf_min);
2061 temp_start = ALIGN(temp_start, kbuf->buf_align);
2062 temp_end = temp_start + kbuf->memsz - 1;
2064 if (temp_end > end || temp_end > kbuf->buf_max)
2067 * Make sure this does not conflict with any of existing
2070 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2071 temp_start = temp_start + PAGE_SIZE;
2075 /* We found a suitable memory range */
2079 /* If we are here, we found a suitable memory range */
2080 kbuf->mem = temp_start;
2082 /* Success, stop navigating through remaining System RAM ranges */
2086 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
2088 struct kexec_buf *kbuf = (struct kexec_buf *)arg;
2089 unsigned long sz = end - start + 1;
2091 /* Returning 0 will take to next memory range */
2092 if (sz < kbuf->memsz)
2095 if (end < kbuf->buf_min || start > kbuf->buf_max)
2099 * Allocate memory top down with-in ram range. Otherwise bottom up
2103 return locate_mem_hole_top_down(start, end, kbuf);
2104 return locate_mem_hole_bottom_up(start, end, kbuf);
2108 * Helper function for placing a buffer in a kexec segment. This assumes
2109 * that kexec_mutex is held.
2111 int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
2112 unsigned long memsz, unsigned long buf_align,
2113 unsigned long buf_min, unsigned long buf_max,
2114 bool top_down, unsigned long *load_addr)
2117 struct kexec_segment *ksegment;
2118 struct kexec_buf buf, *kbuf;
2121 /* Currently adding segment this way is allowed only in file mode */
2122 if (!image->file_mode)
2125 if (image->nr_segments >= KEXEC_SEGMENT_MAX)
2129 * Make sure we are not trying to add buffer after allocating
2130 * control pages. All segments need to be placed first before
2131 * any control pages are allocated. As control page allocation
2132 * logic goes through list of segments to make sure there are
2133 * no destination overlaps.
2135 if (!list_empty(&image->control_pages)) {
2140 memset(&buf, 0, sizeof(struct kexec_buf));
2142 kbuf->image = image;
2143 kbuf->buffer = buffer;
2144 kbuf->bufsz = bufsz;
2146 kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
2147 kbuf->buf_align = max(buf_align, PAGE_SIZE);
2148 kbuf->buf_min = buf_min;
2149 kbuf->buf_max = buf_max;
2150 kbuf->top_down = top_down;
2152 /* Walk the RAM ranges and allocate a suitable range for the buffer */
2153 if (image->type == KEXEC_TYPE_CRASH)
2154 ret = walk_iomem_res("Crash kernel",
2155 IORESOURCE_MEM | IORESOURCE_BUSY,
2156 crashk_res.start, crashk_res.end, kbuf,
2157 locate_mem_hole_callback);
2159 ret = walk_system_ram_res(0, -1, kbuf,
2160 locate_mem_hole_callback);
2162 /* A suitable memory range could not be found for buffer */
2163 return -EADDRNOTAVAIL;
2166 /* Found a suitable memory range */
2167 ksegment = &image->segment[image->nr_segments];
2168 ksegment->kbuf = kbuf->buffer;
2169 ksegment->bufsz = kbuf->bufsz;
2170 ksegment->mem = kbuf->mem;
2171 ksegment->memsz = kbuf->memsz;
2172 image->nr_segments++;
2173 *load_addr = ksegment->mem;
2177 /* Calculate and store the digest of segments */
2178 static int kexec_calculate_store_digests(struct kimage *image)
2180 struct crypto_shash *tfm;
2181 struct shash_desc *desc;
2182 int ret = 0, i, j, zero_buf_sz, sha_region_sz;
2183 size_t desc_size, nullsz;
2186 struct kexec_sha_region *sha_regions;
2187 struct purgatory_info *pi = &image->purgatory_info;
2189 zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
2190 zero_buf_sz = PAGE_SIZE;
2192 tfm = crypto_alloc_shash("sha256", 0, 0);
2198 desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
2199 desc = kzalloc(desc_size, GFP_KERNEL);
2205 sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
2206 sha_regions = vzalloc(sha_region_sz);
2213 ret = crypto_shash_init(desc);
2215 goto out_free_sha_regions;
2217 digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
2220 goto out_free_sha_regions;
2223 for (j = i = 0; i < image->nr_segments; i++) {
2224 struct kexec_segment *ksegment;
2226 ksegment = &image->segment[i];
2228 * Skip purgatory as it will be modified once we put digest
2229 * info in purgatory.
2231 if (ksegment->kbuf == pi->purgatory_buf)
2234 ret = crypto_shash_update(desc, ksegment->kbuf,
2240 * Assume rest of the buffer is filled with zero and
2241 * update digest accordingly.
2243 nullsz = ksegment->memsz - ksegment->bufsz;
2245 unsigned long bytes = nullsz;
2247 if (bytes > zero_buf_sz)
2248 bytes = zero_buf_sz;
2249 ret = crypto_shash_update(desc, zero_buf, bytes);
2258 sha_regions[j].start = ksegment->mem;
2259 sha_regions[j].len = ksegment->memsz;
2264 ret = crypto_shash_final(desc, digest);
2266 goto out_free_digest;
2267 ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
2268 sha_regions, sha_region_sz, 0);
2270 goto out_free_digest;
2272 ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
2273 digest, SHA256_DIGEST_SIZE, 0);
2275 goto out_free_digest;
2280 out_free_sha_regions:
2290 /* Actually load purgatory. Lot of code taken from kexec-tools */
2291 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
2292 unsigned long max, int top_down)
2294 struct purgatory_info *pi = &image->purgatory_info;
2295 unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
2296 unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
2297 unsigned char *buf_addr, *src;
2298 int i, ret = 0, entry_sidx = -1;
2299 const Elf_Shdr *sechdrs_c;
2300 Elf_Shdr *sechdrs = NULL;
2301 void *purgatory_buf = NULL;
2304 * sechdrs_c points to section headers in purgatory and are read
2305 * only. No modifications allowed.
2307 sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
2310 * We can not modify sechdrs_c[] and its fields. It is read only.
2311 * Copy it over to a local copy where one can store some temporary
2312 * data and free it at the end. We need to modify ->sh_addr and
2313 * ->sh_offset fields to keep track of permanent and temporary
2314 * locations of sections.
2316 sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2320 memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2323 * We seem to have multiple copies of sections. First copy is which
2324 * is embedded in kernel in read only section. Some of these sections
2325 * will be copied to a temporary buffer and relocated. And these
2326 * sections will finally be copied to their final destination at
2327 * segment load time.
2329 * Use ->sh_offset to reflect section address in memory. It will
2330 * point to original read only copy if section is not allocatable.
2331 * Otherwise it will point to temporary copy which will be relocated.
2333 * Use ->sh_addr to contain final address of the section where it
2334 * will go during execution time.
2336 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2337 if (sechdrs[i].sh_type == SHT_NOBITS)
2340 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
2341 sechdrs[i].sh_offset;
2345 * Identify entry point section and make entry relative to section
2348 entry = pi->ehdr->e_entry;
2349 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2350 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2353 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
2356 /* Make entry section relative */
2357 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
2358 ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
2359 pi->ehdr->e_entry)) {
2361 entry -= sechdrs[i].sh_addr;
2366 /* Determine how much memory is needed to load relocatable object. */
2372 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2373 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2376 align = sechdrs[i].sh_addralign;
2377 if (sechdrs[i].sh_type != SHT_NOBITS) {
2378 if (buf_align < align)
2380 buf_sz = ALIGN(buf_sz, align);
2381 buf_sz += sechdrs[i].sh_size;
2384 if (bss_align < align)
2386 bss_sz = ALIGN(bss_sz, align);
2387 bss_sz += sechdrs[i].sh_size;
2391 /* Determine the bss padding required to align bss properly */
2393 if (buf_sz & (bss_align - 1))
2394 bss_pad = bss_align - (buf_sz & (bss_align - 1));
2396 memsz = buf_sz + bss_pad + bss_sz;
2398 /* Allocate buffer for purgatory */
2399 purgatory_buf = vzalloc(buf_sz);
2400 if (!purgatory_buf) {
2405 if (buf_align < bss_align)
2406 buf_align = bss_align;
2408 /* Add buffer to segment list */
2409 ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
2410 buf_align, min, max, top_down,
2411 &pi->purgatory_load_addr);
2415 /* Load SHF_ALLOC sections */
2416 buf_addr = purgatory_buf;
2417 load_addr = curr_load_addr = pi->purgatory_load_addr;
2418 bss_addr = load_addr + buf_sz + bss_pad;
2420 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2421 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2424 align = sechdrs[i].sh_addralign;
2425 if (sechdrs[i].sh_type != SHT_NOBITS) {
2426 curr_load_addr = ALIGN(curr_load_addr, align);
2427 offset = curr_load_addr - load_addr;
2428 /* We already modifed ->sh_offset to keep src addr */
2429 src = (char *) sechdrs[i].sh_offset;
2430 memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
2432 /* Store load address and source address of section */
2433 sechdrs[i].sh_addr = curr_load_addr;
2436 * This section got copied to temporary buffer. Update
2437 * ->sh_offset accordingly.
2439 sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
2441 /* Advance to the next address */
2442 curr_load_addr += sechdrs[i].sh_size;
2444 bss_addr = ALIGN(bss_addr, align);
2445 sechdrs[i].sh_addr = bss_addr;
2446 bss_addr += sechdrs[i].sh_size;
2450 /* Update entry point based on load address of text section */
2451 if (entry_sidx >= 0)
2452 entry += sechdrs[entry_sidx].sh_addr;
2454 /* Make kernel jump to purgatory after shutdown */
2455 image->start = entry;
2457 /* Used later to get/set symbol values */
2458 pi->sechdrs = sechdrs;
2461 * Used later to identify which section is purgatory and skip it
2462 * from checksumming.
2464 pi->purgatory_buf = purgatory_buf;
2468 vfree(purgatory_buf);
2472 static int kexec_apply_relocations(struct kimage *image)
2475 struct purgatory_info *pi = &image->purgatory_info;
2476 Elf_Shdr *sechdrs = pi->sechdrs;
2478 /* Apply relocations */
2479 for (i = 0; i < pi->ehdr->e_shnum; i++) {
2480 Elf_Shdr *section, *symtab;
2482 if (sechdrs[i].sh_type != SHT_RELA &&
2483 sechdrs[i].sh_type != SHT_REL)
2487 * For section of type SHT_RELA/SHT_REL,
2488 * ->sh_link contains section header index of associated
2489 * symbol table. And ->sh_info contains section header
2490 * index of section to which relocations apply.
2492 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
2493 sechdrs[i].sh_link >= pi->ehdr->e_shnum)
2496 section = &sechdrs[sechdrs[i].sh_info];
2497 symtab = &sechdrs[sechdrs[i].sh_link];
2499 if (!(section->sh_flags & SHF_ALLOC))
2503 * symtab->sh_link contain section header index of associated
2506 if (symtab->sh_link >= pi->ehdr->e_shnum)
2507 /* Invalid section number? */
2511 * Respective architecture needs to provide support for applying
2512 * relocations of type SHT_RELA/SHT_REL.
2514 if (sechdrs[i].sh_type == SHT_RELA)
2515 ret = arch_kexec_apply_relocations_add(pi->ehdr,
2517 else if (sechdrs[i].sh_type == SHT_REL)
2518 ret = arch_kexec_apply_relocations(pi->ehdr,
2527 /* Load relocatable purgatory object and relocate it appropriately */
2528 int kexec_load_purgatory(struct kimage *image, unsigned long min,
2529 unsigned long max, int top_down,
2530 unsigned long *load_addr)
2532 struct purgatory_info *pi = &image->purgatory_info;
2535 if (kexec_purgatory_size <= 0)
2538 if (kexec_purgatory_size < sizeof(Elf_Ehdr))
2541 pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
2543 if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
2544 || pi->ehdr->e_type != ET_REL
2545 || !elf_check_arch(pi->ehdr)
2546 || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
2549 if (pi->ehdr->e_shoff >= kexec_purgatory_size
2550 || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
2551 kexec_purgatory_size - pi->ehdr->e_shoff))
2554 ret = __kexec_load_purgatory(image, min, max, top_down);
2558 ret = kexec_apply_relocations(image);
2562 *load_addr = pi->purgatory_load_addr;
2566 vfree(pi->purgatory_buf);
2570 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
2579 if (!pi->sechdrs || !pi->ehdr)
2582 sechdrs = pi->sechdrs;
2585 for (i = 0; i < ehdr->e_shnum; i++) {
2586 if (sechdrs[i].sh_type != SHT_SYMTAB)
2589 if (sechdrs[i].sh_link >= ehdr->e_shnum)
2590 /* Invalid strtab section number */
2592 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
2593 syms = (Elf_Sym *)sechdrs[i].sh_offset;
2595 /* Go through symbols for a match */
2596 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
2597 if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
2600 if (strcmp(strtab + syms[k].st_name, name) != 0)
2603 if (syms[k].st_shndx == SHN_UNDEF ||
2604 syms[k].st_shndx >= ehdr->e_shnum) {
2605 pr_debug("Symbol: %s has bad section index %d.\n",
2606 name, syms[k].st_shndx);
2610 /* Found the symbol we are looking for */
2618 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
2620 struct purgatory_info *pi = &image->purgatory_info;
2624 sym = kexec_purgatory_find_symbol(pi, name);
2626 return ERR_PTR(-EINVAL);
2628 sechdr = &pi->sechdrs[sym->st_shndx];
2631 * Returns the address where symbol will finally be loaded after
2632 * kexec_load_segment()
2634 return (void *)(sechdr->sh_addr + sym->st_value);
2638 * Get or set value of a symbol. If "get_value" is true, symbol value is
2639 * returned in buf otherwise symbol value is set based on value in buf.
2641 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
2642 void *buf, unsigned int size, bool get_value)
2646 struct purgatory_info *pi = &image->purgatory_info;
2649 sym = kexec_purgatory_find_symbol(pi, name);
2653 if (sym->st_size != size) {
2654 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
2655 name, (unsigned long)sym->st_size, size);
2659 sechdrs = pi->sechdrs;
2661 if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
2662 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
2663 get_value ? "get" : "set");
2667 sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
2671 memcpy((void *)buf, sym_buf, size);
2673 memcpy((void *)sym_buf, buf, size);
2677 #endif /* CONFIG_KEXEC_FILE */
2680 * Move into place and start executing a preloaded standalone
2681 * executable. If nothing was preloaded return an error.
2683 int kernel_kexec(void)
2687 if (!mutex_trylock(&kexec_mutex))
2694 #ifdef CONFIG_KEXEC_JUMP
2695 if (kexec_image->preserve_context) {
2696 lock_system_sleep();
2697 pm_prepare_console();
2698 error = freeze_processes();
2701 goto Restore_console;
2704 error = dpm_suspend_start(PMSG_FREEZE);
2706 goto Resume_console;
2707 /* At this point, dpm_suspend_start() has been called,
2708 * but *not* dpm_suspend_end(). We *must* call
2709 * dpm_suspend_end() now. Otherwise, drivers for
2710 * some devices (e.g. interrupt controllers) become
2711 * desynchronized with the actual state of the
2712 * hardware at resume time, and evil weirdness ensues.
2714 error = dpm_suspend_end(PMSG_FREEZE);
2716 goto Resume_devices;
2717 error = disable_nonboot_cpus();
2720 local_irq_disable();
2721 error = syscore_suspend();
2727 kexec_in_progress = true;
2728 kernel_restart_prepare(NULL);
2729 migrate_to_reboot_cpu();
2732 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
2733 * no further code needs to use CPU hotplug (which is true in
2734 * the reboot case). However, the kexec path depends on using
2735 * CPU hotplug again; so re-enable it here.
2737 cpu_hotplug_enable();
2738 pr_emerg("Starting new kernel\n");
2742 machine_kexec(kexec_image);
2744 #ifdef CONFIG_KEXEC_JUMP
2745 if (kexec_image->preserve_context) {
2750 enable_nonboot_cpus();
2751 dpm_resume_start(PMSG_RESTORE);
2753 dpm_resume_end(PMSG_RESTORE);
2758 pm_restore_console();
2759 unlock_system_sleep();
2764 mutex_unlock(&kexec_mutex);