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>
35 #include <linux/compiler.h>
38 #include <asm/uaccess.h>
40 #include <asm/sections.h>
42 /* Per cpu memory for storing cpu states in case of system crash. */
43 note_buf_t __percpu *crash_notes;
45 /* vmcoreinfo stuff */
46 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
47 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
48 size_t vmcoreinfo_size;
49 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
51 /* Flag to indicate we are going to kexec a new kernel */
52 bool kexec_in_progress = false;
54 /* Location of the reserved area for the crash kernel */
55 struct resource crashk_res = {
56 .name = "Crash kernel",
59 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
61 struct resource crashk_low_res = {
62 .name = "Crash kernel",
65 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
68 int kexec_should_crash(struct task_struct *p)
70 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
76 * When kexec transitions to the new kernel there is a one-to-one
77 * mapping between physical and virtual addresses. On processors
78 * where you can disable the MMU this is trivial, and easy. For
79 * others it is still a simple predictable page table to setup.
81 * In that environment kexec copies the new kernel to its final
82 * resting place. This means I can only support memory whose
83 * physical address can fit in an unsigned long. In particular
84 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
85 * If the assembly stub has more restrictive requirements
86 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
87 * defined more restrictively in <asm/kexec.h>.
89 * The code for the transition from the current kernel to the
90 * the new kernel is placed in the control_code_buffer, whose size
91 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
92 * page of memory is necessary, but some architectures require more.
93 * Because this memory must be identity mapped in the transition from
94 * virtual to physical addresses it must live in the range
95 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
98 * The assembly stub in the control code buffer is passed a linked list
99 * of descriptor pages detailing the source pages of the new kernel,
100 * and the destination addresses of those source pages. As this data
101 * structure is not used in the context of the current OS, it must
104 * The code has been made to work with highmem pages and will use a
105 * destination page in its final resting place (if it happens
106 * to allocate it). The end product of this is that most of the
107 * physical address space, and most of RAM can be used.
109 * Future directions include:
110 * - allocating a page table with the control code buffer identity
111 * mapped, to simplify machine_kexec and make kexec_on_panic more
116 * KIMAGE_NO_DEST is an impossible destination address..., for
117 * allocating pages whose destination address we do not care about.
119 #define KIMAGE_NO_DEST (-1UL)
121 static int kimage_is_destination_range(struct kimage *image,
122 unsigned long start, unsigned long end);
123 static struct page *kimage_alloc_page(struct kimage *image,
127 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
128 unsigned long nr_segments,
129 struct kexec_segment __user *segments)
131 size_t segment_bytes;
132 struct kimage *image;
136 /* Allocate a controlling structure */
138 image = kzalloc(sizeof(*image), GFP_KERNEL);
143 image->entry = &image->head;
144 image->last_entry = &image->head;
145 image->control_page = ~0; /* By default this does not apply */
146 image->start = entry;
147 image->type = KEXEC_TYPE_DEFAULT;
149 /* Initialize the list of control pages */
150 INIT_LIST_HEAD(&image->control_pages);
152 /* Initialize the list of destination pages */
153 INIT_LIST_HEAD(&image->dest_pages);
155 /* Initialize the list of unusable pages */
156 INIT_LIST_HEAD(&image->unuseable_pages);
158 /* Read in the segments */
159 image->nr_segments = nr_segments;
160 segment_bytes = nr_segments * sizeof(*segments);
161 result = copy_from_user(image->segment, segments, segment_bytes);
168 * Verify we have good destination addresses. The caller is
169 * responsible for making certain we don't attempt to load
170 * the new image into invalid or reserved areas of RAM. This
171 * just verifies it is an address we can use.
173 * Since the kernel does everything in page size chunks ensure
174 * the destination addresses are page aligned. Too many
175 * special cases crop of when we don't do this. The most
176 * insidious is getting overlapping destination addresses
177 * simply because addresses are changed to page size
180 result = -EADDRNOTAVAIL;
181 for (i = 0; i < nr_segments; i++) {
182 unsigned long mstart, mend;
184 mstart = image->segment[i].mem;
185 mend = mstart + image->segment[i].memsz;
186 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
188 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
192 /* Verify our destination addresses do not overlap.
193 * If we alloed overlapping destination addresses
194 * through very weird things can happen with no
195 * easy explanation as one segment stops on another.
198 for (i = 0; i < nr_segments; i++) {
199 unsigned long mstart, mend;
202 mstart = image->segment[i].mem;
203 mend = mstart + image->segment[i].memsz;
204 for (j = 0; j < i; j++) {
205 unsigned long pstart, pend;
206 pstart = image->segment[j].mem;
207 pend = pstart + image->segment[j].memsz;
208 /* Do the segments overlap ? */
209 if ((mend > pstart) && (mstart < pend))
214 /* Ensure our buffer sizes are strictly less than
215 * our memory sizes. This should always be the case,
216 * and it is easier to check up front than to be surprised
220 for (i = 0; i < nr_segments; i++) {
221 if (image->segment[i].bufsz > image->segment[i].memsz)
236 static void kimage_free_page_list(struct list_head *list);
238 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
239 unsigned long nr_segments,
240 struct kexec_segment __user *segments)
243 struct kimage *image;
245 /* Allocate and initialize a controlling structure */
247 result = do_kimage_alloc(&image, entry, nr_segments, segments);
252 * Find a location for the control code buffer, and add it
253 * the vector of segments so that it's pages will also be
254 * counted as destination pages.
257 image->control_code_page = kimage_alloc_control_pages(image,
258 get_order(KEXEC_CONTROL_PAGE_SIZE));
259 if (!image->control_code_page) {
260 pr_err("Could not allocate control_code_buffer\n");
264 image->swap_page = kimage_alloc_control_pages(image, 0);
265 if (!image->swap_page) {
266 pr_err("Could not allocate swap buffer\n");
274 kimage_free_page_list(&image->control_pages);
280 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
281 unsigned long nr_segments,
282 struct kexec_segment __user *segments)
285 struct kimage *image;
289 /* Verify we have a valid entry point */
290 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
291 result = -EADDRNOTAVAIL;
295 /* Allocate and initialize a controlling structure */
296 result = do_kimage_alloc(&image, entry, nr_segments, segments);
300 /* Enable the special crash kernel control page
303 image->control_page = crashk_res.start;
304 image->type = KEXEC_TYPE_CRASH;
307 * Verify we have good destination addresses. Normally
308 * the caller is responsible for making certain we don't
309 * attempt to load the new image into invalid or reserved
310 * areas of RAM. But crash kernels are preloaded into a
311 * reserved area of ram. We must ensure the addresses
312 * are in the reserved area otherwise preloading the
313 * kernel could corrupt things.
315 result = -EADDRNOTAVAIL;
316 for (i = 0; i < nr_segments; i++) {
317 unsigned long mstart, mend;
319 mstart = image->segment[i].mem;
320 mend = mstart + image->segment[i].memsz - 1;
321 /* Ensure we are within the crash kernel limits */
322 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
327 * Find a location for the control code buffer, and add
328 * the vector of segments so that it's pages will also be
329 * counted as destination pages.
332 image->control_code_page = kimage_alloc_control_pages(image,
333 get_order(KEXEC_CONTROL_PAGE_SIZE));
334 if (!image->control_code_page) {
335 pr_err("Could not allocate control_code_buffer\n");
348 static int kimage_is_destination_range(struct kimage *image,
354 for (i = 0; i < image->nr_segments; i++) {
355 unsigned long mstart, mend;
357 mstart = image->segment[i].mem;
358 mend = mstart + image->segment[i].memsz;
359 if ((end > mstart) && (start < mend))
366 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
370 pages = alloc_pages(gfp_mask, order);
372 unsigned int count, i;
373 pages->mapping = NULL;
374 set_page_private(pages, order);
376 for (i = 0; i < count; i++)
377 SetPageReserved(pages + i);
383 static void kimage_free_pages(struct page *page)
385 unsigned int order, count, i;
387 order = page_private(page);
389 for (i = 0; i < count; i++)
390 ClearPageReserved(page + i);
391 __free_pages(page, order);
394 static void kimage_free_page_list(struct list_head *list)
396 struct list_head *pos, *next;
398 list_for_each_safe(pos, next, list) {
401 page = list_entry(pos, struct page, lru);
402 list_del(&page->lru);
403 kimage_free_pages(page);
407 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
410 /* Control pages are special, they are the intermediaries
411 * that are needed while we copy the rest of the pages
412 * to their final resting place. As such they must
413 * not conflict with either the destination addresses
414 * or memory the kernel is already using.
416 * The only case where we really need more than one of
417 * these are for architectures where we cannot disable
418 * the MMU and must instead generate an identity mapped
419 * page table for all of the memory.
421 * At worst this runs in O(N) of the image size.
423 struct list_head extra_pages;
428 INIT_LIST_HEAD(&extra_pages);
430 /* Loop while I can allocate a page and the page allocated
431 * is a destination page.
434 unsigned long pfn, epfn, addr, eaddr;
436 pages = kimage_alloc_pages(GFP_KERNEL, order);
439 pfn = page_to_pfn(pages);
441 addr = pfn << PAGE_SHIFT;
442 eaddr = epfn << PAGE_SHIFT;
443 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
444 kimage_is_destination_range(image, addr, eaddr)) {
445 list_add(&pages->lru, &extra_pages);
451 /* Remember the allocated page... */
452 list_add(&pages->lru, &image->control_pages);
454 /* Because the page is already in it's destination
455 * location we will never allocate another page at
456 * that address. Therefore kimage_alloc_pages
457 * will not return it (again) and we don't need
458 * to give it an entry in image->segment[].
461 /* Deal with the destination pages I have inadvertently allocated.
463 * Ideally I would convert multi-page allocations into single
464 * page allocations, and add everything to image->dest_pages.
466 * For now it is simpler to just free the pages.
468 kimage_free_page_list(&extra_pages);
473 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
476 /* Control pages are special, they are the intermediaries
477 * that are needed while we copy the rest of the pages
478 * to their final resting place. As such they must
479 * not conflict with either the destination addresses
480 * or memory the kernel is already using.
482 * Control pages are also the only pags we must allocate
483 * when loading a crash kernel. All of the other pages
484 * are specified by the segments and we just memcpy
485 * into them directly.
487 * The only case where we really need more than one of
488 * these are for architectures where we cannot disable
489 * the MMU and must instead generate an identity mapped
490 * page table for all of the memory.
492 * Given the low demand this implements a very simple
493 * allocator that finds the first hole of the appropriate
494 * size in the reserved memory region, and allocates all
495 * of the memory up to and including the hole.
497 unsigned long hole_start, hole_end, size;
501 size = (1 << order) << PAGE_SHIFT;
502 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
503 hole_end = hole_start + size - 1;
504 while (hole_end <= crashk_res.end) {
507 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
509 /* See if I overlap any of the segments */
510 for (i = 0; i < image->nr_segments; i++) {
511 unsigned long mstart, mend;
513 mstart = image->segment[i].mem;
514 mend = mstart + image->segment[i].memsz - 1;
515 if ((hole_end >= mstart) && (hole_start <= mend)) {
516 /* Advance the hole to the end of the segment */
517 hole_start = (mend + (size - 1)) & ~(size - 1);
518 hole_end = hole_start + size - 1;
522 /* If I don't overlap any segments I have found my hole! */
523 if (i == image->nr_segments) {
524 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
529 image->control_page = hole_end;
535 struct page *kimage_alloc_control_pages(struct kimage *image,
538 struct page *pages = NULL;
540 switch (image->type) {
541 case KEXEC_TYPE_DEFAULT:
542 pages = kimage_alloc_normal_control_pages(image, order);
544 case KEXEC_TYPE_CRASH:
545 pages = kimage_alloc_crash_control_pages(image, order);
552 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
554 if (*image->entry != 0)
557 if (image->entry == image->last_entry) {
558 kimage_entry_t *ind_page;
561 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
565 ind_page = page_address(page);
566 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
567 image->entry = ind_page;
568 image->last_entry = ind_page +
569 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
571 *image->entry = entry;
578 static int kimage_set_destination(struct kimage *image,
579 unsigned long destination)
583 destination &= PAGE_MASK;
584 result = kimage_add_entry(image, destination | IND_DESTINATION);
586 image->destination = destination;
592 static int kimage_add_page(struct kimage *image, unsigned long page)
597 result = kimage_add_entry(image, page | IND_SOURCE);
599 image->destination += PAGE_SIZE;
605 static void kimage_free_extra_pages(struct kimage *image)
607 /* Walk through and free any extra destination pages I may have */
608 kimage_free_page_list(&image->dest_pages);
610 /* Walk through and free any unusable pages I have cached */
611 kimage_free_page_list(&image->unuseable_pages);
614 static void kimage_terminate(struct kimage *image)
616 if (*image->entry != 0)
619 *image->entry = IND_DONE;
622 #define for_each_kimage_entry(image, ptr, entry) \
623 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
624 ptr = (entry & IND_INDIRECTION) ? \
625 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
627 static void kimage_free_entry(kimage_entry_t entry)
631 page = pfn_to_page(entry >> PAGE_SHIFT);
632 kimage_free_pages(page);
635 static void kimage_free(struct kimage *image)
637 kimage_entry_t *ptr, entry;
638 kimage_entry_t ind = 0;
643 kimage_free_extra_pages(image);
644 for_each_kimage_entry(image, ptr, entry) {
645 if (entry & IND_INDIRECTION) {
646 /* Free the previous indirection page */
647 if (ind & IND_INDIRECTION)
648 kimage_free_entry(ind);
649 /* 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);
777 /* Place the page on the destination list I
780 list_add(&page->lru, &image->dest_pages);
787 static int kimage_load_normal_segment(struct kimage *image,
788 struct kexec_segment *segment)
791 size_t ubytes, mbytes;
793 unsigned char __user *buf;
797 ubytes = segment->bufsz;
798 mbytes = segment->memsz;
799 maddr = segment->mem;
801 result = kimage_set_destination(image, maddr);
808 size_t uchunk, mchunk;
810 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
815 result = kimage_add_page(image, page_to_pfn(page)
821 /* Start with a clear page */
823 ptr += maddr & ~PAGE_MASK;
824 mchunk = min_t(size_t, mbytes,
825 PAGE_SIZE - (maddr & ~PAGE_MASK));
826 uchunk = min(ubytes, mchunk);
828 result = copy_from_user(ptr, buf, uchunk);
843 static int kimage_load_crash_segment(struct kimage *image,
844 struct kexec_segment *segment)
846 /* For crash dumps kernels we simply copy the data from
847 * user space to it's destination.
848 * We do things a page at a time for the sake of kmap.
851 size_t ubytes, mbytes;
853 unsigned char __user *buf;
857 ubytes = segment->bufsz;
858 mbytes = segment->memsz;
859 maddr = segment->mem;
863 size_t uchunk, mchunk;
865 page = pfn_to_page(maddr >> PAGE_SHIFT);
871 ptr += maddr & ~PAGE_MASK;
872 mchunk = min_t(size_t, mbytes,
873 PAGE_SIZE - (maddr & ~PAGE_MASK));
874 uchunk = min(ubytes, mchunk);
875 if (mchunk > uchunk) {
876 /* Zero the trailing part of the page */
877 memset(ptr + uchunk, 0, mchunk - uchunk);
879 result = copy_from_user(ptr, buf, uchunk);
880 kexec_flush_icache_page(page);
895 static int kimage_load_segment(struct kimage *image,
896 struct kexec_segment *segment)
898 int result = -ENOMEM;
900 switch (image->type) {
901 case KEXEC_TYPE_DEFAULT:
902 result = kimage_load_normal_segment(image, segment);
904 case KEXEC_TYPE_CRASH:
905 result = kimage_load_crash_segment(image, segment);
913 * Exec Kernel system call: for obvious reasons only root may call it.
915 * This call breaks up into three pieces.
916 * - A generic part which loads the new kernel from the current
917 * address space, and very carefully places the data in the
920 * - A generic part that interacts with the kernel and tells all of
921 * the devices to shut down. Preventing on-going dmas, and placing
922 * the devices in a consistent state so a later kernel can
925 * - A machine specific part that includes the syscall number
926 * and then copies the image to it's final destination. And
927 * jumps into the image at entry.
929 * kexec does not sync, or unmount filesystems so if you need
930 * that to happen you need to do that yourself.
932 struct kimage *kexec_image;
933 struct kimage *kexec_crash_image;
934 int kexec_load_disabled;
936 static DEFINE_MUTEX(kexec_mutex);
938 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
939 struct kexec_segment __user *, segments, unsigned long, flags)
941 struct kimage **dest_image, *image;
944 /* We only trust the superuser with rebooting the system. */
945 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
949 * Verify we have a legal set of flags
950 * This leaves us room for future extensions.
952 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
955 /* Verify we are on the appropriate architecture */
956 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
957 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
960 /* Put an artificial cap on the number
961 * of segments passed to kexec_load.
963 if (nr_segments > KEXEC_SEGMENT_MAX)
969 /* Because we write directly to the reserved memory
970 * region when loading crash kernels we need a mutex here to
971 * prevent multiple crash kernels from attempting to load
972 * simultaneously, and to prevent a crash kernel from loading
973 * over the top of a in use crash kernel.
975 * KISS: always take the mutex.
977 if (!mutex_trylock(&kexec_mutex))
980 dest_image = &kexec_image;
981 if (flags & KEXEC_ON_CRASH)
982 dest_image = &kexec_crash_image;
983 if (nr_segments > 0) {
986 /* Loading another kernel to reboot into */
987 if ((flags & KEXEC_ON_CRASH) == 0)
988 result = kimage_normal_alloc(&image, entry,
989 nr_segments, segments);
990 /* Loading another kernel to switch to if this one crashes */
991 else if (flags & KEXEC_ON_CRASH) {
992 /* Free any current crash dump kernel before
995 kimage_free(xchg(&kexec_crash_image, NULL));
996 result = kimage_crash_alloc(&image, entry,
997 nr_segments, segments);
998 crash_map_reserved_pages();
1003 if (flags & KEXEC_PRESERVE_CONTEXT)
1004 image->preserve_context = 1;
1005 result = machine_kexec_prepare(image);
1009 for (i = 0; i < nr_segments; i++) {
1010 result = kimage_load_segment(image, &image->segment[i]);
1014 kimage_terminate(image);
1015 if (flags & KEXEC_ON_CRASH)
1016 crash_unmap_reserved_pages();
1018 /* Install the new kernel, and Uninstall the old */
1019 image = xchg(dest_image, image);
1022 mutex_unlock(&kexec_mutex);
1029 * Add and remove page tables for crashkernel memory
1031 * Provide an empty default implementation here -- architecture
1032 * code may override this
1034 void __weak crash_map_reserved_pages(void)
1037 void __weak crash_unmap_reserved_pages(void)
1040 #ifdef CONFIG_COMPAT
1041 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1042 compat_ulong_t, nr_segments,
1043 struct compat_kexec_segment __user *, segments,
1044 compat_ulong_t, flags)
1046 struct compat_kexec_segment in;
1047 struct kexec_segment out, __user *ksegments;
1048 unsigned long i, result;
1050 /* Don't allow clients that don't understand the native
1051 * architecture to do anything.
1053 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1056 if (nr_segments > KEXEC_SEGMENT_MAX)
1059 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1060 for (i = 0; i < nr_segments; i++) {
1061 result = copy_from_user(&in, &segments[i], sizeof(in));
1065 out.buf = compat_ptr(in.buf);
1066 out.bufsz = in.bufsz;
1068 out.memsz = in.memsz;
1070 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1075 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1079 void crash_kexec(struct pt_regs *regs)
1081 /* Take the kexec_mutex here to prevent sys_kexec_load
1082 * running on one cpu from replacing the crash kernel
1083 * we are using after a panic on a different cpu.
1085 * If the crash kernel was not located in a fixed area
1086 * of memory the xchg(&kexec_crash_image) would be
1087 * sufficient. But since I reuse the memory...
1089 if (mutex_trylock(&kexec_mutex)) {
1090 if (kexec_crash_image) {
1091 struct pt_regs fixed_regs;
1093 crash_setup_regs(&fixed_regs, regs);
1094 crash_save_vmcoreinfo();
1095 machine_crash_shutdown(&fixed_regs);
1096 machine_kexec(kexec_crash_image);
1098 mutex_unlock(&kexec_mutex);
1102 size_t crash_get_memory_size(void)
1105 mutex_lock(&kexec_mutex);
1106 if (crashk_res.end != crashk_res.start)
1107 size = resource_size(&crashk_res);
1108 mutex_unlock(&kexec_mutex);
1112 void __weak crash_free_reserved_phys_range(unsigned long begin,
1117 for (addr = begin; addr < end; addr += PAGE_SIZE)
1118 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1121 int crash_shrink_memory(unsigned long new_size)
1124 unsigned long start, end;
1125 unsigned long old_size;
1126 struct resource *ram_res;
1128 mutex_lock(&kexec_mutex);
1130 if (kexec_crash_image) {
1134 start = crashk_res.start;
1135 end = crashk_res.end;
1136 old_size = (end == 0) ? 0 : end - start + 1;
1137 if (new_size >= old_size) {
1138 ret = (new_size == old_size) ? 0 : -EINVAL;
1142 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1148 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1149 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1151 crash_map_reserved_pages();
1152 crash_free_reserved_phys_range(end, crashk_res.end);
1154 if ((start == end) && (crashk_res.parent != NULL))
1155 release_resource(&crashk_res);
1157 ram_res->start = end;
1158 ram_res->end = crashk_res.end;
1159 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1160 ram_res->name = "System RAM";
1162 crashk_res.end = end - 1;
1164 insert_resource(&iomem_resource, ram_res);
1165 crash_unmap_reserved_pages();
1168 mutex_unlock(&kexec_mutex);
1172 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1175 struct elf_note note;
1177 note.n_namesz = strlen(name) + 1;
1178 note.n_descsz = data_len;
1180 memcpy(buf, ¬e, sizeof(note));
1181 buf += (sizeof(note) + 3)/4;
1182 memcpy(buf, name, note.n_namesz);
1183 buf += (note.n_namesz + 3)/4;
1184 memcpy(buf, data, note.n_descsz);
1185 buf += (note.n_descsz + 3)/4;
1190 static void final_note(u32 *buf)
1192 struct elf_note note;
1197 memcpy(buf, ¬e, sizeof(note));
1200 void crash_save_cpu(struct pt_regs *regs, int cpu)
1202 struct elf_prstatus prstatus;
1205 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1208 /* Using ELF notes here is opportunistic.
1209 * I need a well defined structure format
1210 * for the data I pass, and I need tags
1211 * on the data to indicate what information I have
1212 * squirrelled away. ELF notes happen to provide
1213 * all of that, so there is no need to invent something new.
1215 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1218 memset(&prstatus, 0, sizeof(prstatus));
1219 prstatus.pr_pid = current->pid;
1220 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1221 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1222 &prstatus, sizeof(prstatus));
1226 static int __init crash_notes_memory_init(void)
1228 /* Allocate memory for saving cpu registers. */
1229 crash_notes = alloc_percpu(note_buf_t);
1231 pr_warn("Kexec: Memory allocation for saving cpu register states failed\n");
1236 subsys_initcall(crash_notes_memory_init);
1240 * parsing the "crashkernel" commandline
1242 * this code is intended to be called from architecture specific code
1247 * This function parses command lines in the format
1249 * crashkernel=ramsize-range:size[,...][@offset]
1251 * The function returns 0 on success and -EINVAL on failure.
1253 static int __init parse_crashkernel_mem(char *cmdline,
1254 unsigned long long system_ram,
1255 unsigned long long *crash_size,
1256 unsigned long long *crash_base)
1258 char *cur = cmdline, *tmp;
1260 /* for each entry of the comma-separated list */
1262 unsigned long long start, end = ULLONG_MAX, size;
1264 /* get the start of the range */
1265 start = memparse(cur, &tmp);
1267 pr_warn("crashkernel: Memory value expected\n");
1272 pr_warn("crashkernel: '-' expected\n");
1277 /* if no ':' is here, than we read the end */
1279 end = memparse(cur, &tmp);
1281 pr_warn("crashkernel: Memory value expected\n");
1286 pr_warn("crashkernel: end <= start\n");
1292 pr_warn("crashkernel: ':' expected\n");
1297 size = memparse(cur, &tmp);
1299 pr_warn("Memory value expected\n");
1303 if (size >= system_ram) {
1304 pr_warn("crashkernel: invalid size\n");
1309 if (system_ram >= start && system_ram < end) {
1313 } while (*cur++ == ',');
1315 if (*crash_size > 0) {
1316 while (*cur && *cur != ' ' && *cur != '@')
1320 *crash_base = memparse(cur, &tmp);
1322 pr_warn("Memory value expected after '@'\n");
1332 * That function parses "simple" (old) crashkernel command lines like
1334 * crashkernel=size[@offset]
1336 * It returns 0 on success and -EINVAL on failure.
1338 static int __init parse_crashkernel_simple(char *cmdline,
1339 unsigned long long *crash_size,
1340 unsigned long long *crash_base)
1342 char *cur = cmdline;
1344 *crash_size = memparse(cmdline, &cur);
1345 if (cmdline == cur) {
1346 pr_warn("crashkernel: memory value expected\n");
1351 *crash_base = memparse(cur+1, &cur);
1352 else if (*cur != ' ' && *cur != '\0') {
1353 pr_warn("crashkernel: unrecognized char\n");
1360 #define SUFFIX_HIGH 0
1361 #define SUFFIX_LOW 1
1362 #define SUFFIX_NULL 2
1363 static __initdata char *suffix_tbl[] = {
1364 [SUFFIX_HIGH] = ",high",
1365 [SUFFIX_LOW] = ",low",
1366 [SUFFIX_NULL] = NULL,
1370 * That function parses "suffix" crashkernel command lines like
1372 * crashkernel=size,[high|low]
1374 * It returns 0 on success and -EINVAL on failure.
1376 static int __init parse_crashkernel_suffix(char *cmdline,
1377 unsigned long long *crash_size,
1378 unsigned long long *crash_base,
1381 char *cur = cmdline;
1383 *crash_size = memparse(cmdline, &cur);
1384 if (cmdline == cur) {
1385 pr_warn("crashkernel: memory value expected\n");
1389 /* check with suffix */
1390 if (strncmp(cur, suffix, strlen(suffix))) {
1391 pr_warn("crashkernel: unrecognized char\n");
1394 cur += strlen(suffix);
1395 if (*cur != ' ' && *cur != '\0') {
1396 pr_warn("crashkernel: unrecognized char\n");
1403 static __init char *get_last_crashkernel(char *cmdline,
1407 char *p = cmdline, *ck_cmdline = NULL;
1409 /* find crashkernel and use the last one if there are more */
1410 p = strstr(p, name);
1412 char *end_p = strchr(p, ' ');
1416 end_p = p + strlen(p);
1421 /* skip the one with any known suffix */
1422 for (i = 0; suffix_tbl[i]; i++) {
1423 q = end_p - strlen(suffix_tbl[i]);
1424 if (!strncmp(q, suffix_tbl[i],
1425 strlen(suffix_tbl[i])))
1430 q = end_p - strlen(suffix);
1431 if (!strncmp(q, suffix, strlen(suffix)))
1435 p = strstr(p+1, name);
1444 static int __init __parse_crashkernel(char *cmdline,
1445 unsigned long long system_ram,
1446 unsigned long long *crash_size,
1447 unsigned long long *crash_base,
1451 char *first_colon, *first_space;
1454 BUG_ON(!crash_size || !crash_base);
1458 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1463 ck_cmdline += strlen(name);
1466 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1467 crash_base, suffix);
1469 * if the commandline contains a ':', then that's the extended
1470 * syntax -- if not, it must be the classic syntax
1472 first_colon = strchr(ck_cmdline, ':');
1473 first_space = strchr(ck_cmdline, ' ');
1474 if (first_colon && (!first_space || first_colon < first_space))
1475 return parse_crashkernel_mem(ck_cmdline, system_ram,
1476 crash_size, crash_base);
1478 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1482 * That function is the entry point for command line parsing and should be
1483 * called from the arch-specific code.
1485 int __init parse_crashkernel(char *cmdline,
1486 unsigned long long system_ram,
1487 unsigned long long *crash_size,
1488 unsigned long long *crash_base)
1490 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1491 "crashkernel=", NULL);
1494 int __init parse_crashkernel_high(char *cmdline,
1495 unsigned long long system_ram,
1496 unsigned long long *crash_size,
1497 unsigned long long *crash_base)
1499 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1500 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1503 int __init parse_crashkernel_low(char *cmdline,
1504 unsigned long long system_ram,
1505 unsigned long long *crash_size,
1506 unsigned long long *crash_base)
1508 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1509 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1512 static void update_vmcoreinfo_note(void)
1514 u32 *buf = vmcoreinfo_note;
1516 if (!vmcoreinfo_size)
1518 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1523 void crash_save_vmcoreinfo(void)
1525 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1526 update_vmcoreinfo_note();
1529 void vmcoreinfo_append_str(const char *fmt, ...)
1535 va_start(args, fmt);
1536 r = vscnprintf(buf, sizeof(buf), fmt, args);
1539 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1541 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1543 vmcoreinfo_size += r;
1547 * provide an empty default implementation here -- architecture
1548 * code may override this
1550 void __weak arch_crash_save_vmcoreinfo(void)
1553 unsigned long __weak paddr_vmcoreinfo_note(void)
1555 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1558 static int __init crash_save_vmcoreinfo_init(void)
1560 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1561 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1563 VMCOREINFO_SYMBOL(init_uts_ns);
1564 VMCOREINFO_SYMBOL(node_online_map);
1566 VMCOREINFO_SYMBOL(swapper_pg_dir);
1568 VMCOREINFO_SYMBOL(_stext);
1569 VMCOREINFO_SYMBOL(vmap_area_list);
1571 #ifndef CONFIG_NEED_MULTIPLE_NODES
1572 VMCOREINFO_SYMBOL(mem_map);
1573 VMCOREINFO_SYMBOL(contig_page_data);
1575 #ifdef CONFIG_SPARSEMEM
1576 VMCOREINFO_SYMBOL(mem_section);
1577 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1578 VMCOREINFO_STRUCT_SIZE(mem_section);
1579 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1581 VMCOREINFO_STRUCT_SIZE(page);
1582 VMCOREINFO_STRUCT_SIZE(pglist_data);
1583 VMCOREINFO_STRUCT_SIZE(zone);
1584 VMCOREINFO_STRUCT_SIZE(free_area);
1585 VMCOREINFO_STRUCT_SIZE(list_head);
1586 VMCOREINFO_SIZE(nodemask_t);
1587 VMCOREINFO_OFFSET(page, flags);
1588 VMCOREINFO_OFFSET(page, _count);
1589 VMCOREINFO_OFFSET(page, mapping);
1590 VMCOREINFO_OFFSET(page, lru);
1591 VMCOREINFO_OFFSET(page, _mapcount);
1592 VMCOREINFO_OFFSET(page, private);
1593 VMCOREINFO_OFFSET(pglist_data, node_zones);
1594 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1595 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1596 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1598 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1599 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1600 VMCOREINFO_OFFSET(pglist_data, node_id);
1601 VMCOREINFO_OFFSET(zone, free_area);
1602 VMCOREINFO_OFFSET(zone, vm_stat);
1603 VMCOREINFO_OFFSET(zone, spanned_pages);
1604 VMCOREINFO_OFFSET(free_area, free_list);
1605 VMCOREINFO_OFFSET(list_head, next);
1606 VMCOREINFO_OFFSET(list_head, prev);
1607 VMCOREINFO_OFFSET(vmap_area, va_start);
1608 VMCOREINFO_OFFSET(vmap_area, list);
1609 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1610 log_buf_kexec_setup();
1611 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1612 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1613 VMCOREINFO_NUMBER(PG_lru);
1614 VMCOREINFO_NUMBER(PG_private);
1615 VMCOREINFO_NUMBER(PG_swapcache);
1616 VMCOREINFO_NUMBER(PG_slab);
1617 #ifdef CONFIG_MEMORY_FAILURE
1618 VMCOREINFO_NUMBER(PG_hwpoison);
1620 VMCOREINFO_NUMBER(PG_head_mask);
1621 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1623 arch_crash_save_vmcoreinfo();
1624 update_vmcoreinfo_note();
1629 subsys_initcall(crash_save_vmcoreinfo_init);
1632 * Move into place and start executing a preloaded standalone
1633 * executable. If nothing was preloaded return an error.
1635 int kernel_kexec(void)
1639 if (!mutex_trylock(&kexec_mutex))
1646 #ifdef CONFIG_KEXEC_JUMP
1647 if (kexec_image->preserve_context) {
1648 lock_system_sleep();
1649 pm_prepare_console();
1650 error = freeze_processes();
1653 goto Restore_console;
1656 error = dpm_suspend_start(PMSG_FREEZE);
1658 goto Resume_console;
1659 /* At this point, dpm_suspend_start() has been called,
1660 * but *not* dpm_suspend_end(). We *must* call
1661 * dpm_suspend_end() now. Otherwise, drivers for
1662 * some devices (e.g. interrupt controllers) become
1663 * desynchronized with the actual state of the
1664 * hardware at resume time, and evil weirdness ensues.
1666 error = dpm_suspend_end(PMSG_FREEZE);
1668 goto Resume_devices;
1669 error = disable_nonboot_cpus();
1672 local_irq_disable();
1673 error = syscore_suspend();
1679 kexec_in_progress = true;
1680 kernel_restart_prepare(NULL);
1681 migrate_to_reboot_cpu();
1684 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1685 * no further code needs to use CPU hotplug (which is true in
1686 * the reboot case). However, the kexec path depends on using
1687 * CPU hotplug again; so re-enable it here.
1689 cpu_hotplug_enable();
1690 pr_emerg("Starting new kernel\n");
1694 machine_kexec(kexec_image);
1696 #ifdef CONFIG_KEXEC_JUMP
1697 if (kexec_image->preserve_context) {
1702 enable_nonboot_cpus();
1703 dpm_resume_start(PMSG_RESTORE);
1705 dpm_resume_end(PMSG_RESTORE);
1710 pm_restore_console();
1711 unlock_system_sleep();
1716 mutex_unlock(&kexec_mutex);