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 #include <linux/capability.h>
11 #include <linux/file.h>
12 #include <linux/slab.h>
14 #include <linux/kexec.h>
15 #include <linux/mutex.h>
16 #include <linux/list.h>
17 #include <linux/highmem.h>
18 #include <linux/syscalls.h>
19 #include <linux/reboot.h>
20 #include <linux/ioport.h>
21 #include <linux/hardirq.h>
22 #include <linux/elf.h>
23 #include <linux/elfcore.h>
24 #include <linux/utsname.h>
25 #include <linux/numa.h>
26 #include <linux/suspend.h>
27 #include <linux/device.h>
28 #include <linux/freezer.h>
30 #include <linux/cpu.h>
31 #include <linux/console.h>
32 #include <linux/vmalloc.h>
33 #include <linux/swap.h>
34 #include <linux/syscore_ops.h>
37 #include <asm/uaccess.h>
39 #include <asm/sections.h>
41 /* Per cpu memory for storing cpu states in case of system crash. */
42 note_buf_t __percpu *crash_notes;
44 /* vmcoreinfo stuff */
45 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
46 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
47 size_t vmcoreinfo_size;
48 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
50 /* Flag to indicate we are going to kexec a new kernel */
51 bool kexec_in_progress = false;
53 /* Location of the reserved area for the crash kernel */
54 struct resource crashk_res = {
55 .name = "Crash kernel",
58 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
60 struct resource crashk_low_res = {
61 .name = "Crash kernel",
64 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
67 int kexec_should_crash(struct task_struct *p)
69 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
75 * When kexec transitions to the new kernel there is a one-to-one
76 * mapping between physical and virtual addresses. On processors
77 * where you can disable the MMU this is trivial, and easy. For
78 * others it is still a simple predictable page table to setup.
80 * In that environment kexec copies the new kernel to its final
81 * resting place. This means I can only support memory whose
82 * physical address can fit in an unsigned long. In particular
83 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
84 * If the assembly stub has more restrictive requirements
85 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
86 * defined more restrictively in <asm/kexec.h>.
88 * The code for the transition from the current kernel to the
89 * the new kernel is placed in the control_code_buffer, whose size
90 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
91 * page of memory is necessary, but some architectures require more.
92 * Because this memory must be identity mapped in the transition from
93 * virtual to physical addresses it must live in the range
94 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
97 * The assembly stub in the control code buffer is passed a linked list
98 * of descriptor pages detailing the source pages of the new kernel,
99 * and the destination addresses of those source pages. As this data
100 * structure is not used in the context of the current OS, it must
103 * The code has been made to work with highmem pages and will use a
104 * destination page in its final resting place (if it happens
105 * to allocate it). The end product of this is that most of the
106 * physical address space, and most of RAM can be used.
108 * Future directions include:
109 * - allocating a page table with the control code buffer identity
110 * mapped, to simplify machine_kexec and make kexec_on_panic more
115 * KIMAGE_NO_DEST is an impossible destination address..., for
116 * allocating pages whose destination address we do not care about.
118 #define KIMAGE_NO_DEST (-1UL)
120 static int kimage_is_destination_range(struct kimage *image,
121 unsigned long start, unsigned long end);
122 static struct page *kimage_alloc_page(struct kimage *image,
126 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
127 unsigned long nr_segments,
128 struct kexec_segment __user *segments)
130 size_t segment_bytes;
131 struct kimage *image;
135 /* Allocate a controlling structure */
137 image = kzalloc(sizeof(*image), GFP_KERNEL);
142 image->entry = &image->head;
143 image->last_entry = &image->head;
144 image->control_page = ~0; /* By default this does not apply */
145 image->start = entry;
146 image->type = KEXEC_TYPE_DEFAULT;
148 /* Initialize the list of control pages */
149 INIT_LIST_HEAD(&image->control_pages);
151 /* Initialize the list of destination pages */
152 INIT_LIST_HEAD(&image->dest_pages);
154 /* Initialize the list of unusable pages */
155 INIT_LIST_HEAD(&image->unuseable_pages);
157 /* Read in the segments */
158 image->nr_segments = nr_segments;
159 segment_bytes = nr_segments * sizeof(*segments);
160 result = copy_from_user(image->segment, segments, segment_bytes);
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)
235 static void kimage_free_page_list(struct list_head *list);
237 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
238 unsigned long nr_segments,
239 struct kexec_segment __user *segments)
242 struct kimage *image;
244 /* Allocate and initialize a controlling structure */
246 result = do_kimage_alloc(&image, entry, nr_segments, segments);
251 * Find a location for the control code buffer, and add it
252 * the vector of segments so that it's pages will also be
253 * counted as destination pages.
256 image->control_code_page = kimage_alloc_control_pages(image,
257 get_order(KEXEC_CONTROL_PAGE_SIZE));
258 if (!image->control_code_page) {
259 printk(KERN_ERR "Could not allocate control_code_buffer\n");
263 image->swap_page = kimage_alloc_control_pages(image, 0);
264 if (!image->swap_page) {
265 printk(KERN_ERR "Could not allocate swap buffer\n");
273 kimage_free_page_list(&image->control_pages);
279 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
280 unsigned long nr_segments,
281 struct kexec_segment __user *segments)
284 struct kimage *image;
288 /* Verify we have a valid entry point */
289 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
290 result = -EADDRNOTAVAIL;
294 /* Allocate and initialize a controlling structure */
295 result = do_kimage_alloc(&image, entry, nr_segments, segments);
299 /* Enable the special crash kernel control page
302 image->control_page = crashk_res.start;
303 image->type = KEXEC_TYPE_CRASH;
306 * Verify we have good destination addresses. Normally
307 * the caller is responsible for making certain we don't
308 * attempt to load the new image into invalid or reserved
309 * areas of RAM. But crash kernels are preloaded into a
310 * reserved area of ram. We must ensure the addresses
311 * are in the reserved area otherwise preloading the
312 * kernel could corrupt things.
314 result = -EADDRNOTAVAIL;
315 for (i = 0; i < nr_segments; i++) {
316 unsigned long mstart, mend;
318 mstart = image->segment[i].mem;
319 mend = mstart + image->segment[i].memsz - 1;
320 /* Ensure we are within the crash kernel limits */
321 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
326 * Find a location for the control code buffer, and add
327 * the vector of segments so that it's pages will also be
328 * counted as destination pages.
331 image->control_code_page = kimage_alloc_control_pages(image,
332 get_order(KEXEC_CONTROL_PAGE_SIZE));
333 if (!image->control_code_page) {
334 printk(KERN_ERR "Could not allocate control_code_buffer\n");
347 static int kimage_is_destination_range(struct kimage *image,
353 for (i = 0; i < image->nr_segments; i++) {
354 unsigned long mstart, mend;
356 mstart = image->segment[i].mem;
357 mend = mstart + image->segment[i].memsz;
358 if ((end > mstart) && (start < mend))
365 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
369 pages = alloc_pages(gfp_mask, order);
371 unsigned int count, i;
372 pages->mapping = NULL;
373 set_page_private(pages, order);
375 for (i = 0; i < count; i++)
376 SetPageReserved(pages + i);
382 static void kimage_free_pages(struct page *page)
384 unsigned int order, count, i;
386 order = page_private(page);
388 for (i = 0; i < count; i++)
389 ClearPageReserved(page + i);
390 __free_pages(page, order);
393 static void kimage_free_page_list(struct list_head *list)
395 struct list_head *pos, *next;
397 list_for_each_safe(pos, next, list) {
400 page = list_entry(pos, struct page, lru);
401 list_del(&page->lru);
402 kimage_free_pages(page);
406 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
409 /* Control pages are special, they are the intermediaries
410 * that are needed while we copy the rest of the pages
411 * to their final resting place. As such they must
412 * not conflict with either the destination addresses
413 * or memory the kernel is already using.
415 * The only case where we really need more than one of
416 * these are for architectures where we cannot disable
417 * the MMU and must instead generate an identity mapped
418 * page table for all of the memory.
420 * At worst this runs in O(N) of the image size.
422 struct list_head extra_pages;
427 INIT_LIST_HEAD(&extra_pages);
429 /* Loop while I can allocate a page and the page allocated
430 * is a destination page.
433 unsigned long pfn, epfn, addr, eaddr;
435 pages = kimage_alloc_pages(GFP_KERNEL, order);
438 pfn = page_to_pfn(pages);
440 addr = pfn << PAGE_SHIFT;
441 eaddr = epfn << PAGE_SHIFT;
442 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
443 kimage_is_destination_range(image, addr, eaddr)) {
444 list_add(&pages->lru, &extra_pages);
450 /* Remember the allocated page... */
451 list_add(&pages->lru, &image->control_pages);
453 /* Because the page is already in it's destination
454 * location we will never allocate another page at
455 * that address. Therefore kimage_alloc_pages
456 * will not return it (again) and we don't need
457 * to give it an entry in image->segment[].
460 /* Deal with the destination pages I have inadvertently allocated.
462 * Ideally I would convert multi-page allocations into single
463 * page allocations, and add everything to image->dest_pages.
465 * For now it is simpler to just free the pages.
467 kimage_free_page_list(&extra_pages);
472 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
475 /* Control pages are special, they are the intermediaries
476 * that are needed while we copy the rest of the pages
477 * to their final resting place. As such they must
478 * not conflict with either the destination addresses
479 * or memory the kernel is already using.
481 * Control pages are also the only pags we must allocate
482 * when loading a crash kernel. All of the other pages
483 * are specified by the segments and we just memcpy
484 * into them directly.
486 * The only case where we really need more than one of
487 * these are for architectures where we cannot disable
488 * the MMU and must instead generate an identity mapped
489 * page table for all of the memory.
491 * Given the low demand this implements a very simple
492 * allocator that finds the first hole of the appropriate
493 * size in the reserved memory region, and allocates all
494 * of the memory up to and including the hole.
496 unsigned long hole_start, hole_end, size;
500 size = (1 << order) << PAGE_SHIFT;
501 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
502 hole_end = hole_start + size - 1;
503 while (hole_end <= crashk_res.end) {
506 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
508 /* See if I overlap any of the segments */
509 for (i = 0; i < image->nr_segments; i++) {
510 unsigned long mstart, mend;
512 mstart = image->segment[i].mem;
513 mend = mstart + image->segment[i].memsz - 1;
514 if ((hole_end >= mstart) && (hole_start <= mend)) {
515 /* Advance the hole to the end of the segment */
516 hole_start = (mend + (size - 1)) & ~(size - 1);
517 hole_end = hole_start + size - 1;
521 /* If I don't overlap any segments I have found my hole! */
522 if (i == image->nr_segments) {
523 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
528 image->control_page = hole_end;
534 struct page *kimage_alloc_control_pages(struct kimage *image,
537 struct page *pages = NULL;
539 switch (image->type) {
540 case KEXEC_TYPE_DEFAULT:
541 pages = kimage_alloc_normal_control_pages(image, order);
543 case KEXEC_TYPE_CRASH:
544 pages = kimage_alloc_crash_control_pages(image, order);
551 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
553 if (*image->entry != 0)
556 if (image->entry == image->last_entry) {
557 kimage_entry_t *ind_page;
560 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
564 ind_page = page_address(page);
565 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
566 image->entry = ind_page;
567 image->last_entry = ind_page +
568 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
570 *image->entry = entry;
577 static int kimage_set_destination(struct kimage *image,
578 unsigned long destination)
582 destination &= PAGE_MASK;
583 result = kimage_add_entry(image, destination | IND_DESTINATION);
585 image->destination = destination;
591 static int kimage_add_page(struct kimage *image, unsigned long page)
596 result = kimage_add_entry(image, page | IND_SOURCE);
598 image->destination += PAGE_SIZE;
604 static void kimage_free_extra_pages(struct kimage *image)
606 /* Walk through and free any extra destination pages I may have */
607 kimage_free_page_list(&image->dest_pages);
609 /* Walk through and free any unusable pages I have cached */
610 kimage_free_page_list(&image->unuseable_pages);
613 static void kimage_terminate(struct kimage *image)
615 if (*image->entry != 0)
618 *image->entry = IND_DONE;
621 #define for_each_kimage_entry(image, ptr, entry) \
622 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
623 ptr = (entry & IND_INDIRECTION)? \
624 phys_to_virt((entry & PAGE_MASK)): ptr +1)
626 static void kimage_free_entry(kimage_entry_t entry)
630 page = pfn_to_page(entry >> PAGE_SHIFT);
631 kimage_free_pages(page);
634 static void kimage_free(struct kimage *image)
636 kimage_entry_t *ptr, entry;
637 kimage_entry_t ind = 0;
642 kimage_free_extra_pages(image);
643 for_each_kimage_entry(image, ptr, entry) {
644 if (entry & IND_INDIRECTION) {
645 /* Free the previous indirection page */
646 if (ind & IND_INDIRECTION)
647 kimage_free_entry(ind);
648 /* Save this indirection page until we are
653 else if (entry & IND_SOURCE)
654 kimage_free_entry(entry);
656 /* Free the final indirection page */
657 if (ind & IND_INDIRECTION)
658 kimage_free_entry(ind);
660 /* Handle any machine specific cleanup */
661 machine_kexec_cleanup(image);
663 /* Free the kexec control pages... */
664 kimage_free_page_list(&image->control_pages);
668 static kimage_entry_t *kimage_dst_used(struct kimage *image,
671 kimage_entry_t *ptr, entry;
672 unsigned long destination = 0;
674 for_each_kimage_entry(image, ptr, entry) {
675 if (entry & IND_DESTINATION)
676 destination = entry & PAGE_MASK;
677 else if (entry & IND_SOURCE) {
678 if (page == destination)
680 destination += PAGE_SIZE;
687 static struct page *kimage_alloc_page(struct kimage *image,
689 unsigned long destination)
692 * Here we implement safeguards to ensure that a source page
693 * is not copied to its destination page before the data on
694 * the destination page is no longer useful.
696 * To do this we maintain the invariant that a source page is
697 * either its own destination page, or it is not a
698 * destination page at all.
700 * That is slightly stronger than required, but the proof
701 * that no problems will not occur is trivial, and the
702 * implementation is simply to verify.
704 * When allocating all pages normally this algorithm will run
705 * in O(N) time, but in the worst case it will run in O(N^2)
706 * time. If the runtime is a problem the data structures can
713 * Walk through the list of destination pages, and see if I
716 list_for_each_entry(page, &image->dest_pages, lru) {
717 addr = page_to_pfn(page) << PAGE_SHIFT;
718 if (addr == destination) {
719 list_del(&page->lru);
727 /* Allocate a page, if we run out of memory give up */
728 page = kimage_alloc_pages(gfp_mask, 0);
731 /* If the page cannot be used file it away */
732 if (page_to_pfn(page) >
733 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
734 list_add(&page->lru, &image->unuseable_pages);
737 addr = page_to_pfn(page) << PAGE_SHIFT;
739 /* If it is the destination page we want use it */
740 if (addr == destination)
743 /* If the page is not a destination page use it */
744 if (!kimage_is_destination_range(image, addr,
749 * I know that the page is someones destination page.
750 * See if there is already a source page for this
751 * destination page. And if so swap the source pages.
753 old = kimage_dst_used(image, addr);
756 unsigned long old_addr;
757 struct page *old_page;
759 old_addr = *old & PAGE_MASK;
760 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
761 copy_highpage(page, old_page);
762 *old = addr | (*old & ~PAGE_MASK);
764 /* The old page I have found cannot be a
765 * destination page, so return it if it's
766 * gfp_flags honor the ones passed in.
768 if (!(gfp_mask & __GFP_HIGHMEM) &&
769 PageHighMem(old_page)) {
770 kimage_free_pages(old_page);
778 /* Place the page on the destination list I
781 list_add(&page->lru, &image->dest_pages);
788 static int kimage_load_normal_segment(struct kimage *image,
789 struct kexec_segment *segment)
792 size_t ubytes, mbytes;
794 unsigned char __user *buf;
798 ubytes = segment->bufsz;
799 mbytes = segment->memsz;
800 maddr = segment->mem;
802 result = kimage_set_destination(image, maddr);
809 size_t uchunk, mchunk;
811 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
816 result = kimage_add_page(image, page_to_pfn(page)
822 /* Start with a clear page */
824 ptr += maddr & ~PAGE_MASK;
825 mchunk = min_t(size_t, mbytes,
826 PAGE_SIZE - (maddr & ~PAGE_MASK));
827 uchunk = min(ubytes, mchunk);
829 result = copy_from_user(ptr, buf, uchunk);
844 static int kimage_load_crash_segment(struct kimage *image,
845 struct kexec_segment *segment)
847 /* For crash dumps kernels we simply copy the data from
848 * user space to it's destination.
849 * We do things a page at a time for the sake of kmap.
852 size_t ubytes, mbytes;
854 unsigned char __user *buf;
858 ubytes = segment->bufsz;
859 mbytes = segment->memsz;
860 maddr = segment->mem;
864 size_t uchunk, mchunk;
866 page = pfn_to_page(maddr >> PAGE_SHIFT);
872 ptr += maddr & ~PAGE_MASK;
873 mchunk = min_t(size_t, mbytes,
874 PAGE_SIZE - (maddr & ~PAGE_MASK));
875 uchunk = min(ubytes, mchunk);
876 if (mchunk > uchunk) {
877 /* Zero the trailing part of the page */
878 memset(ptr + uchunk, 0, mchunk - uchunk);
880 result = copy_from_user(ptr, buf, uchunk);
881 kexec_flush_icache_page(page);
896 static int kimage_load_segment(struct kimage *image,
897 struct kexec_segment *segment)
899 int result = -ENOMEM;
901 switch (image->type) {
902 case KEXEC_TYPE_DEFAULT:
903 result = kimage_load_normal_segment(image, segment);
905 case KEXEC_TYPE_CRASH:
906 result = kimage_load_crash_segment(image, segment);
914 * Exec Kernel system call: for obvious reasons only root may call it.
916 * This call breaks up into three pieces.
917 * - A generic part which loads the new kernel from the current
918 * address space, and very carefully places the data in the
921 * - A generic part that interacts with the kernel and tells all of
922 * the devices to shut down. Preventing on-going dmas, and placing
923 * the devices in a consistent state so a later kernel can
926 * - A machine specific part that includes the syscall number
927 * and then copies the image to it's final destination. And
928 * jumps into the image at entry.
930 * kexec does not sync, or unmount filesystems so if you need
931 * that to happen you need to do that yourself.
933 struct kimage *kexec_image;
934 struct kimage *kexec_crash_image;
935 int kexec_load_disabled;
937 static DEFINE_MUTEX(kexec_mutex);
939 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
940 struct kexec_segment __user *, segments, unsigned long, flags)
942 struct kimage **dest_image, *image;
945 /* We only trust the superuser with rebooting the system. */
946 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
950 * Verify we have a legal set of flags
951 * This leaves us room for future extensions.
953 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
956 /* Verify we are on the appropriate architecture */
957 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
958 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
961 /* Put an artificial cap on the number
962 * of segments passed to kexec_load.
964 if (nr_segments > KEXEC_SEGMENT_MAX)
970 /* Because we write directly to the reserved memory
971 * region when loading crash kernels we need a mutex here to
972 * prevent multiple crash kernels from attempting to load
973 * simultaneously, and to prevent a crash kernel from loading
974 * over the top of a in use crash kernel.
976 * KISS: always take the mutex.
978 if (!mutex_trylock(&kexec_mutex))
981 dest_image = &kexec_image;
982 if (flags & KEXEC_ON_CRASH)
983 dest_image = &kexec_crash_image;
984 if (nr_segments > 0) {
987 /* Loading another kernel to reboot into */
988 if ((flags & KEXEC_ON_CRASH) == 0)
989 result = kimage_normal_alloc(&image, entry,
990 nr_segments, segments);
991 /* Loading another kernel to switch to if this one crashes */
992 else if (flags & KEXEC_ON_CRASH) {
993 /* Free any current crash dump kernel before
996 kimage_free(xchg(&kexec_crash_image, NULL));
997 result = kimage_crash_alloc(&image, entry,
998 nr_segments, segments);
999 crash_map_reserved_pages();
1004 if (flags & KEXEC_PRESERVE_CONTEXT)
1005 image->preserve_context = 1;
1006 result = machine_kexec_prepare(image);
1010 for (i = 0; i < nr_segments; i++) {
1011 result = kimage_load_segment(image, &image->segment[i]);
1015 kimage_terminate(image);
1016 if (flags & KEXEC_ON_CRASH)
1017 crash_unmap_reserved_pages();
1019 /* Install the new kernel, and Uninstall the old */
1020 image = xchg(dest_image, image);
1023 mutex_unlock(&kexec_mutex);
1030 * Add and remove page tables for crashkernel memory
1032 * Provide an empty default implementation here -- architecture
1033 * code may override this
1035 void __weak crash_map_reserved_pages(void)
1038 void __weak crash_unmap_reserved_pages(void)
1041 #ifdef CONFIG_COMPAT
1042 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1043 compat_ulong_t, nr_segments,
1044 struct compat_kexec_segment __user *, segments,
1045 compat_ulong_t, flags)
1047 struct compat_kexec_segment in;
1048 struct kexec_segment out, __user *ksegments;
1049 unsigned long i, result;
1051 /* Don't allow clients that don't understand the native
1052 * architecture to do anything.
1054 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1057 if (nr_segments > KEXEC_SEGMENT_MAX)
1060 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1061 for (i=0; i < nr_segments; i++) {
1062 result = copy_from_user(&in, &segments[i], sizeof(in));
1066 out.buf = compat_ptr(in.buf);
1067 out.bufsz = in.bufsz;
1069 out.memsz = in.memsz;
1071 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1076 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1080 void crash_kexec(struct pt_regs *regs)
1082 /* Take the kexec_mutex here to prevent sys_kexec_load
1083 * running on one cpu from replacing the crash kernel
1084 * we are using after a panic on a different cpu.
1086 * If the crash kernel was not located in a fixed area
1087 * of memory the xchg(&kexec_crash_image) would be
1088 * sufficient. But since I reuse the memory...
1090 if (mutex_trylock(&kexec_mutex)) {
1091 if (kexec_crash_image) {
1092 struct pt_regs fixed_regs;
1094 crash_setup_regs(&fixed_regs, regs);
1095 crash_save_vmcoreinfo();
1096 machine_crash_shutdown(&fixed_regs);
1097 machine_kexec(kexec_crash_image);
1099 mutex_unlock(&kexec_mutex);
1103 size_t crash_get_memory_size(void)
1106 mutex_lock(&kexec_mutex);
1107 if (crashk_res.end != crashk_res.start)
1108 size = resource_size(&crashk_res);
1109 mutex_unlock(&kexec_mutex);
1113 void __weak crash_free_reserved_phys_range(unsigned long begin,
1118 for (addr = begin; addr < end; addr += PAGE_SIZE)
1119 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1122 int crash_shrink_memory(unsigned long new_size)
1125 unsigned long start, end;
1126 unsigned long old_size;
1127 struct resource *ram_res;
1129 mutex_lock(&kexec_mutex);
1131 if (kexec_crash_image) {
1135 start = crashk_res.start;
1136 end = crashk_res.end;
1137 old_size = (end == 0) ? 0 : end - start + 1;
1138 if (new_size >= old_size) {
1139 ret = (new_size == old_size) ? 0 : -EINVAL;
1143 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1149 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1150 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1152 crash_map_reserved_pages();
1153 crash_free_reserved_phys_range(end, crashk_res.end);
1155 if ((start == end) && (crashk_res.parent != NULL))
1156 release_resource(&crashk_res);
1158 ram_res->start = end;
1159 ram_res->end = crashk_res.end;
1160 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1161 ram_res->name = "System RAM";
1163 crashk_res.end = end - 1;
1165 insert_resource(&iomem_resource, ram_res);
1166 crash_unmap_reserved_pages();
1169 mutex_unlock(&kexec_mutex);
1173 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1176 struct elf_note note;
1178 note.n_namesz = strlen(name) + 1;
1179 note.n_descsz = data_len;
1181 memcpy(buf, ¬e, sizeof(note));
1182 buf += (sizeof(note) + 3)/4;
1183 memcpy(buf, name, note.n_namesz);
1184 buf += (note.n_namesz + 3)/4;
1185 memcpy(buf, data, note.n_descsz);
1186 buf += (note.n_descsz + 3)/4;
1191 static void final_note(u32 *buf)
1193 struct elf_note note;
1198 memcpy(buf, ¬e, sizeof(note));
1201 void crash_save_cpu(struct pt_regs *regs, int cpu)
1203 struct elf_prstatus prstatus;
1206 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1209 /* Using ELF notes here is opportunistic.
1210 * I need a well defined structure format
1211 * for the data I pass, and I need tags
1212 * on the data to indicate what information I have
1213 * squirrelled away. ELF notes happen to provide
1214 * all of that, so there is no need to invent something new.
1216 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1219 memset(&prstatus, 0, sizeof(prstatus));
1220 prstatus.pr_pid = current->pid;
1221 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1222 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1223 &prstatus, sizeof(prstatus));
1227 static int __init crash_notes_memory_init(void)
1229 /* Allocate memory for saving cpu registers. */
1230 crash_notes = alloc_percpu(note_buf_t);
1232 printk("Kexec: Memory allocation for saving cpu register"
1233 " states failed\n");
1238 subsys_initcall(crash_notes_memory_init);
1242 * parsing the "crashkernel" commandline
1244 * this code is intended to be called from architecture specific code
1249 * This function parses command lines in the format
1251 * crashkernel=ramsize-range:size[,...][@offset]
1253 * The function returns 0 on success and -EINVAL on failure.
1255 static int __init parse_crashkernel_mem(char *cmdline,
1256 unsigned long long system_ram,
1257 unsigned long long *crash_size,
1258 unsigned long long *crash_base)
1260 char *cur = cmdline, *tmp;
1262 /* for each entry of the comma-separated list */
1264 unsigned long long start, end = ULLONG_MAX, size;
1266 /* get the start of the range */
1267 start = memparse(cur, &tmp);
1269 pr_warning("crashkernel: Memory value expected\n");
1274 pr_warning("crashkernel: '-' expected\n");
1279 /* if no ':' is here, than we read the end */
1281 end = memparse(cur, &tmp);
1283 pr_warning("crashkernel: Memory "
1284 "value expected\n");
1289 pr_warning("crashkernel: end <= start\n");
1295 pr_warning("crashkernel: ':' expected\n");
1300 size = memparse(cur, &tmp);
1302 pr_warning("Memory value expected\n");
1306 if (size >= system_ram) {
1307 pr_warning("crashkernel: invalid size\n");
1312 if (system_ram >= start && system_ram < end) {
1316 } while (*cur++ == ',');
1318 if (*crash_size > 0) {
1319 while (*cur && *cur != ' ' && *cur != '@')
1323 *crash_base = memparse(cur, &tmp);
1325 pr_warning("Memory value expected "
1336 * That function parses "simple" (old) crashkernel command lines like
1338 * crashkernel=size[@offset]
1340 * It returns 0 on success and -EINVAL on failure.
1342 static int __init parse_crashkernel_simple(char *cmdline,
1343 unsigned long long *crash_size,
1344 unsigned long long *crash_base)
1346 char *cur = cmdline;
1348 *crash_size = memparse(cmdline, &cur);
1349 if (cmdline == cur) {
1350 pr_warning("crashkernel: memory value expected\n");
1355 *crash_base = memparse(cur+1, &cur);
1356 else if (*cur != ' ' && *cur != '\0') {
1357 pr_warning("crashkernel: unrecognized char\n");
1364 #define SUFFIX_HIGH 0
1365 #define SUFFIX_LOW 1
1366 #define SUFFIX_NULL 2
1367 static __initdata char *suffix_tbl[] = {
1368 [SUFFIX_HIGH] = ",high",
1369 [SUFFIX_LOW] = ",low",
1370 [SUFFIX_NULL] = NULL,
1374 * That function parses "suffix" crashkernel command lines like
1376 * crashkernel=size,[high|low]
1378 * It returns 0 on success and -EINVAL on failure.
1380 static int __init parse_crashkernel_suffix(char *cmdline,
1381 unsigned long long *crash_size,
1382 unsigned long long *crash_base,
1385 char *cur = cmdline;
1387 *crash_size = memparse(cmdline, &cur);
1388 if (cmdline == cur) {
1389 pr_warn("crashkernel: memory value expected\n");
1393 /* check with suffix */
1394 if (strncmp(cur, suffix, strlen(suffix))) {
1395 pr_warn("crashkernel: unrecognized char\n");
1398 cur += strlen(suffix);
1399 if (*cur != ' ' && *cur != '\0') {
1400 pr_warn("crashkernel: unrecognized char\n");
1407 static __init char *get_last_crashkernel(char *cmdline,
1411 char *p = cmdline, *ck_cmdline = NULL;
1413 /* find crashkernel and use the last one if there are more */
1414 p = strstr(p, name);
1416 char *end_p = strchr(p, ' ');
1420 end_p = p + strlen(p);
1425 /* skip the one with any known suffix */
1426 for (i = 0; suffix_tbl[i]; i++) {
1427 q = end_p - strlen(suffix_tbl[i]);
1428 if (!strncmp(q, suffix_tbl[i],
1429 strlen(suffix_tbl[i])))
1434 q = end_p - strlen(suffix);
1435 if (!strncmp(q, suffix, strlen(suffix)))
1439 p = strstr(p+1, name);
1448 static int __init __parse_crashkernel(char *cmdline,
1449 unsigned long long system_ram,
1450 unsigned long long *crash_size,
1451 unsigned long long *crash_base,
1455 char *first_colon, *first_space;
1458 BUG_ON(!crash_size || !crash_base);
1462 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1467 ck_cmdline += strlen(name);
1470 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1471 crash_base, suffix);
1473 * if the commandline contains a ':', then that's the extended
1474 * syntax -- if not, it must be the classic syntax
1476 first_colon = strchr(ck_cmdline, ':');
1477 first_space = strchr(ck_cmdline, ' ');
1478 if (first_colon && (!first_space || first_colon < first_space))
1479 return parse_crashkernel_mem(ck_cmdline, system_ram,
1480 crash_size, crash_base);
1482 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1486 * That function is the entry point for command line parsing and should be
1487 * called from the arch-specific code.
1489 int __init parse_crashkernel(char *cmdline,
1490 unsigned long long system_ram,
1491 unsigned long long *crash_size,
1492 unsigned long long *crash_base)
1494 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1495 "crashkernel=", NULL);
1498 int __init parse_crashkernel_high(char *cmdline,
1499 unsigned long long system_ram,
1500 unsigned long long *crash_size,
1501 unsigned long long *crash_base)
1503 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1504 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1507 int __init parse_crashkernel_low(char *cmdline,
1508 unsigned long long system_ram,
1509 unsigned long long *crash_size,
1510 unsigned long long *crash_base)
1512 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1513 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1516 static void update_vmcoreinfo_note(void)
1518 u32 *buf = vmcoreinfo_note;
1520 if (!vmcoreinfo_size)
1522 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1527 void crash_save_vmcoreinfo(void)
1529 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1530 update_vmcoreinfo_note();
1533 void vmcoreinfo_append_str(const char *fmt, ...)
1539 va_start(args, fmt);
1540 r = vscnprintf(buf, sizeof(buf), fmt, args);
1543 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1545 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1547 vmcoreinfo_size += r;
1551 * provide an empty default implementation here -- architecture
1552 * code may override this
1554 void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1557 unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1559 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1562 static int __init crash_save_vmcoreinfo_init(void)
1564 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1565 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1567 VMCOREINFO_SYMBOL(init_uts_ns);
1568 VMCOREINFO_SYMBOL(node_online_map);
1570 VMCOREINFO_SYMBOL(swapper_pg_dir);
1572 VMCOREINFO_SYMBOL(_stext);
1573 VMCOREINFO_SYMBOL(vmap_area_list);
1575 #ifndef CONFIG_NEED_MULTIPLE_NODES
1576 VMCOREINFO_SYMBOL(mem_map);
1577 VMCOREINFO_SYMBOL(contig_page_data);
1579 #ifdef CONFIG_SPARSEMEM
1580 VMCOREINFO_SYMBOL(mem_section);
1581 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1582 VMCOREINFO_STRUCT_SIZE(mem_section);
1583 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1585 VMCOREINFO_STRUCT_SIZE(page);
1586 VMCOREINFO_STRUCT_SIZE(pglist_data);
1587 VMCOREINFO_STRUCT_SIZE(zone);
1588 VMCOREINFO_STRUCT_SIZE(free_area);
1589 VMCOREINFO_STRUCT_SIZE(list_head);
1590 VMCOREINFO_SIZE(nodemask_t);
1591 VMCOREINFO_OFFSET(page, flags);
1592 VMCOREINFO_OFFSET(page, _count);
1593 VMCOREINFO_OFFSET(page, mapping);
1594 VMCOREINFO_OFFSET(page, lru);
1595 VMCOREINFO_OFFSET(page, _mapcount);
1596 VMCOREINFO_OFFSET(page, private);
1597 VMCOREINFO_OFFSET(pglist_data, node_zones);
1598 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1599 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1600 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1602 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1603 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1604 VMCOREINFO_OFFSET(pglist_data, node_id);
1605 VMCOREINFO_OFFSET(zone, free_area);
1606 VMCOREINFO_OFFSET(zone, vm_stat);
1607 VMCOREINFO_OFFSET(zone, spanned_pages);
1608 VMCOREINFO_OFFSET(free_area, free_list);
1609 VMCOREINFO_OFFSET(list_head, next);
1610 VMCOREINFO_OFFSET(list_head, prev);
1611 VMCOREINFO_OFFSET(vmap_area, va_start);
1612 VMCOREINFO_OFFSET(vmap_area, list);
1613 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1614 log_buf_kexec_setup();
1615 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1616 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1617 VMCOREINFO_NUMBER(PG_lru);
1618 VMCOREINFO_NUMBER(PG_private);
1619 VMCOREINFO_NUMBER(PG_swapcache);
1620 VMCOREINFO_NUMBER(PG_slab);
1621 #ifdef CONFIG_MEMORY_FAILURE
1622 VMCOREINFO_NUMBER(PG_hwpoison);
1624 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1626 arch_crash_save_vmcoreinfo();
1627 update_vmcoreinfo_note();
1632 subsys_initcall(crash_save_vmcoreinfo_init);
1635 * Move into place and start executing a preloaded standalone
1636 * executable. If nothing was preloaded return an error.
1638 int kernel_kexec(void)
1642 if (!mutex_trylock(&kexec_mutex))
1649 #ifdef CONFIG_KEXEC_JUMP
1650 if (kexec_image->preserve_context) {
1651 lock_system_sleep();
1652 pm_prepare_console();
1653 error = freeze_processes();
1656 goto Restore_console;
1659 error = dpm_suspend_start(PMSG_FREEZE);
1661 goto Resume_console;
1662 /* At this point, dpm_suspend_start() has been called,
1663 * but *not* dpm_suspend_end(). We *must* call
1664 * dpm_suspend_end() now. Otherwise, drivers for
1665 * some devices (e.g. interrupt controllers) become
1666 * desynchronized with the actual state of the
1667 * hardware at resume time, and evil weirdness ensues.
1669 error = dpm_suspend_end(PMSG_FREEZE);
1671 goto Resume_devices;
1672 error = disable_nonboot_cpus();
1675 local_irq_disable();
1676 error = syscore_suspend();
1682 kexec_in_progress = true;
1683 kernel_restart_prepare(NULL);
1684 migrate_to_reboot_cpu();
1685 printk(KERN_EMERG "Starting new kernel\n");
1689 machine_kexec(kexec_image);
1691 #ifdef CONFIG_KEXEC_JUMP
1692 if (kexec_image->preserve_context) {
1697 enable_nonboot_cpus();
1698 dpm_resume_start(PMSG_RESTORE);
1700 dpm_resume_end(PMSG_RESTORE);
1705 pm_restore_console();
1706 unlock_system_sleep();
1711 mutex_unlock(&kexec_mutex);