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 <generated/utsrelease.h>
25 #include <linux/utsname.h>
26 #include <linux/numa.h>
27 #include <linux/suspend.h>
28 #include <linux/device.h>
29 #include <linux/freezer.h>
31 #include <linux/cpu.h>
32 #include <linux/console.h>
33 #include <linux/vmalloc.h>
34 #include <linux/swap.h>
35 #include <linux/syscore_ops.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 /* Location of the reserved area for the crash kernel */
52 struct resource crashk_res = {
53 .name = "Crash kernel",
56 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
59 int kexec_should_crash(struct task_struct *p)
61 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
67 * When kexec transitions to the new kernel there is a one-to-one
68 * mapping between physical and virtual addresses. On processors
69 * where you can disable the MMU this is trivial, and easy. For
70 * others it is still a simple predictable page table to setup.
72 * In that environment kexec copies the new kernel to its final
73 * resting place. This means I can only support memory whose
74 * physical address can fit in an unsigned long. In particular
75 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
76 * If the assembly stub has more restrictive requirements
77 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
78 * defined more restrictively in <asm/kexec.h>.
80 * The code for the transition from the current kernel to the
81 * the new kernel is placed in the control_code_buffer, whose size
82 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
83 * page of memory is necessary, but some architectures require more.
84 * Because this memory must be identity mapped in the transition from
85 * virtual to physical addresses it must live in the range
86 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
89 * The assembly stub in the control code buffer is passed a linked list
90 * of descriptor pages detailing the source pages of the new kernel,
91 * and the destination addresses of those source pages. As this data
92 * structure is not used in the context of the current OS, it must
95 * The code has been made to work with highmem pages and will use a
96 * destination page in its final resting place (if it happens
97 * to allocate it). The end product of this is that most of the
98 * physical address space, and most of RAM can be used.
100 * Future directions include:
101 * - allocating a page table with the control code buffer identity
102 * mapped, to simplify machine_kexec and make kexec_on_panic more
107 * KIMAGE_NO_DEST is an impossible destination address..., for
108 * allocating pages whose destination address we do not care about.
110 #define KIMAGE_NO_DEST (-1UL)
112 static int kimage_is_destination_range(struct kimage *image,
113 unsigned long start, unsigned long end);
114 static struct page *kimage_alloc_page(struct kimage *image,
118 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
119 unsigned long nr_segments,
120 struct kexec_segment __user *segments)
122 size_t segment_bytes;
123 struct kimage *image;
127 /* Allocate a controlling structure */
129 image = kzalloc(sizeof(*image), GFP_KERNEL);
134 image->entry = &image->head;
135 image->last_entry = &image->head;
136 image->control_page = ~0; /* By default this does not apply */
137 image->start = entry;
138 image->type = KEXEC_TYPE_DEFAULT;
140 /* Initialize the list of control pages */
141 INIT_LIST_HEAD(&image->control_pages);
143 /* Initialize the list of destination pages */
144 INIT_LIST_HEAD(&image->dest_pages);
146 /* Initialize the list of unusable pages */
147 INIT_LIST_HEAD(&image->unuseable_pages);
149 /* Read in the segments */
150 image->nr_segments = nr_segments;
151 segment_bytes = nr_segments * sizeof(*segments);
152 result = copy_from_user(image->segment, segments, segment_bytes);
159 * Verify we have good destination addresses. The caller is
160 * responsible for making certain we don't attempt to load
161 * the new image into invalid or reserved areas of RAM. This
162 * just verifies it is an address we can use.
164 * Since the kernel does everything in page size chunks ensure
165 * the destination addresses are page aligned. Too many
166 * special cases crop of when we don't do this. The most
167 * insidious is getting overlapping destination addresses
168 * simply because addresses are changed to page size
171 result = -EADDRNOTAVAIL;
172 for (i = 0; i < nr_segments; i++) {
173 unsigned long mstart, mend;
175 mstart = image->segment[i].mem;
176 mend = mstart + image->segment[i].memsz;
177 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
179 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
183 /* Verify our destination addresses do not overlap.
184 * If we alloed overlapping destination addresses
185 * through very weird things can happen with no
186 * easy explanation as one segment stops on another.
189 for (i = 0; i < nr_segments; i++) {
190 unsigned long mstart, mend;
193 mstart = image->segment[i].mem;
194 mend = mstart + image->segment[i].memsz;
195 for (j = 0; j < i; j++) {
196 unsigned long pstart, pend;
197 pstart = image->segment[j].mem;
198 pend = pstart + image->segment[j].memsz;
199 /* Do the segments overlap ? */
200 if ((mend > pstart) && (mstart < pend))
205 /* Ensure our buffer sizes are strictly less than
206 * our memory sizes. This should always be the case,
207 * and it is easier to check up front than to be surprised
211 for (i = 0; i < nr_segments; i++) {
212 if (image->segment[i].bufsz > image->segment[i].memsz)
227 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
228 unsigned long nr_segments,
229 struct kexec_segment __user *segments)
232 struct kimage *image;
234 /* Allocate and initialize a controlling structure */
236 result = do_kimage_alloc(&image, entry, nr_segments, segments);
243 * Find a location for the control code buffer, and add it
244 * the vector of segments so that it's pages will also be
245 * counted as destination pages.
248 image->control_code_page = kimage_alloc_control_pages(image,
249 get_order(KEXEC_CONTROL_PAGE_SIZE));
250 if (!image->control_code_page) {
251 printk(KERN_ERR "Could not allocate control_code_buffer\n");
255 image->swap_page = kimage_alloc_control_pages(image, 0);
256 if (!image->swap_page) {
257 printk(KERN_ERR "Could not allocate swap buffer\n");
271 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
272 unsigned long nr_segments,
273 struct kexec_segment __user *segments)
276 struct kimage *image;
280 /* Verify we have a valid entry point */
281 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
282 result = -EADDRNOTAVAIL;
286 /* Allocate and initialize a controlling structure */
287 result = do_kimage_alloc(&image, entry, nr_segments, segments);
291 /* Enable the special crash kernel control page
294 image->control_page = crashk_res.start;
295 image->type = KEXEC_TYPE_CRASH;
298 * Verify we have good destination addresses. Normally
299 * the caller is responsible for making certain we don't
300 * attempt to load the new image into invalid or reserved
301 * areas of RAM. But crash kernels are preloaded into a
302 * reserved area of ram. We must ensure the addresses
303 * are in the reserved area otherwise preloading the
304 * kernel could corrupt things.
306 result = -EADDRNOTAVAIL;
307 for (i = 0; i < nr_segments; i++) {
308 unsigned long mstart, mend;
310 mstart = image->segment[i].mem;
311 mend = mstart + image->segment[i].memsz - 1;
312 /* Ensure we are within the crash kernel limits */
313 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
318 * Find a location for the control code buffer, and add
319 * the vector of segments so that it's pages will also be
320 * counted as destination pages.
323 image->control_code_page = kimage_alloc_control_pages(image,
324 get_order(KEXEC_CONTROL_PAGE_SIZE));
325 if (!image->control_code_page) {
326 printk(KERN_ERR "Could not allocate control_code_buffer\n");
340 static int kimage_is_destination_range(struct kimage *image,
346 for (i = 0; i < image->nr_segments; i++) {
347 unsigned long mstart, mend;
349 mstart = image->segment[i].mem;
350 mend = mstart + image->segment[i].memsz;
351 if ((end > mstart) && (start < mend))
358 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
362 pages = alloc_pages(gfp_mask, order);
364 unsigned int count, i;
365 pages->mapping = NULL;
366 set_page_private(pages, order);
368 for (i = 0; i < count; i++)
369 SetPageReserved(pages + i);
375 static void kimage_free_pages(struct page *page)
377 unsigned int order, count, i;
379 order = page_private(page);
381 for (i = 0; i < count; i++)
382 ClearPageReserved(page + i);
383 __free_pages(page, order);
386 static void kimage_free_page_list(struct list_head *list)
388 struct list_head *pos, *next;
390 list_for_each_safe(pos, next, list) {
393 page = list_entry(pos, struct page, lru);
394 list_del(&page->lru);
395 kimage_free_pages(page);
399 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
402 /* Control pages are special, they are the intermediaries
403 * that are needed while we copy the rest of the pages
404 * to their final resting place. As such they must
405 * not conflict with either the destination addresses
406 * or memory the kernel is already using.
408 * The only case where we really need more than one of
409 * these are for architectures where we cannot disable
410 * the MMU and must instead generate an identity mapped
411 * page table for all of the memory.
413 * At worst this runs in O(N) of the image size.
415 struct list_head extra_pages;
420 INIT_LIST_HEAD(&extra_pages);
422 /* Loop while I can allocate a page and the page allocated
423 * is a destination page.
426 unsigned long pfn, epfn, addr, eaddr;
428 pages = kimage_alloc_pages(GFP_KERNEL, order);
431 pfn = page_to_pfn(pages);
433 addr = pfn << PAGE_SHIFT;
434 eaddr = epfn << PAGE_SHIFT;
435 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
436 kimage_is_destination_range(image, addr, eaddr)) {
437 list_add(&pages->lru, &extra_pages);
443 /* Remember the allocated page... */
444 list_add(&pages->lru, &image->control_pages);
446 /* Because the page is already in it's destination
447 * location we will never allocate another page at
448 * that address. Therefore kimage_alloc_pages
449 * will not return it (again) and we don't need
450 * to give it an entry in image->segment[].
453 /* Deal with the destination pages I have inadvertently allocated.
455 * Ideally I would convert multi-page allocations into single
456 * page allocations, and add everything to image->dest_pages.
458 * For now it is simpler to just free the pages.
460 kimage_free_page_list(&extra_pages);
465 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
468 /* Control pages are special, they are the intermediaries
469 * that are needed while we copy the rest of the pages
470 * to their final resting place. As such they must
471 * not conflict with either the destination addresses
472 * or memory the kernel is already using.
474 * Control pages are also the only pags we must allocate
475 * when loading a crash kernel. All of the other pages
476 * are specified by the segments and we just memcpy
477 * into them directly.
479 * The only case where we really need more than one of
480 * these are for architectures where we cannot disable
481 * the MMU and must instead generate an identity mapped
482 * page table for all of the memory.
484 * Given the low demand this implements a very simple
485 * allocator that finds the first hole of the appropriate
486 * size in the reserved memory region, and allocates all
487 * of the memory up to and including the hole.
489 unsigned long hole_start, hole_end, size;
493 size = (1 << order) << PAGE_SHIFT;
494 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
495 hole_end = hole_start + size - 1;
496 while (hole_end <= crashk_res.end) {
499 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
501 if (hole_end > crashk_res.end)
503 /* See if I overlap any of the segments */
504 for (i = 0; i < image->nr_segments; i++) {
505 unsigned long mstart, mend;
507 mstart = image->segment[i].mem;
508 mend = mstart + image->segment[i].memsz - 1;
509 if ((hole_end >= mstart) && (hole_start <= mend)) {
510 /* Advance the hole to the end of the segment */
511 hole_start = (mend + (size - 1)) & ~(size - 1);
512 hole_end = hole_start + size - 1;
516 /* If I don't overlap any segments I have found my hole! */
517 if (i == image->nr_segments) {
518 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
523 image->control_page = hole_end;
529 struct page *kimage_alloc_control_pages(struct kimage *image,
532 struct page *pages = NULL;
534 switch (image->type) {
535 case KEXEC_TYPE_DEFAULT:
536 pages = kimage_alloc_normal_control_pages(image, order);
538 case KEXEC_TYPE_CRASH:
539 pages = kimage_alloc_crash_control_pages(image, order);
546 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
548 if (*image->entry != 0)
551 if (image->entry == image->last_entry) {
552 kimage_entry_t *ind_page;
555 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
559 ind_page = page_address(page);
560 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
561 image->entry = ind_page;
562 image->last_entry = ind_page +
563 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
565 *image->entry = entry;
572 static int kimage_set_destination(struct kimage *image,
573 unsigned long destination)
577 destination &= PAGE_MASK;
578 result = kimage_add_entry(image, destination | IND_DESTINATION);
580 image->destination = destination;
586 static int kimage_add_page(struct kimage *image, unsigned long page)
591 result = kimage_add_entry(image, page | IND_SOURCE);
593 image->destination += PAGE_SIZE;
599 static void kimage_free_extra_pages(struct kimage *image)
601 /* Walk through and free any extra destination pages I may have */
602 kimage_free_page_list(&image->dest_pages);
604 /* Walk through and free any unusable pages I have cached */
605 kimage_free_page_list(&image->unuseable_pages);
608 static void kimage_terminate(struct kimage *image)
610 if (*image->entry != 0)
613 *image->entry = IND_DONE;
616 #define for_each_kimage_entry(image, ptr, entry) \
617 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
618 ptr = (entry & IND_INDIRECTION)? \
619 phys_to_virt((entry & PAGE_MASK)): ptr +1)
621 static void kimage_free_entry(kimage_entry_t entry)
625 page = pfn_to_page(entry >> PAGE_SHIFT);
626 kimage_free_pages(page);
629 static void kimage_free(struct kimage *image)
631 kimage_entry_t *ptr, entry;
632 kimage_entry_t ind = 0;
637 kimage_free_extra_pages(image);
638 for_each_kimage_entry(image, ptr, entry) {
639 if (entry & IND_INDIRECTION) {
640 /* Free the previous indirection page */
641 if (ind & IND_INDIRECTION)
642 kimage_free_entry(ind);
643 /* Save this indirection page until we are
648 else if (entry & IND_SOURCE)
649 kimage_free_entry(entry);
651 /* Free the final indirection page */
652 if (ind & IND_INDIRECTION)
653 kimage_free_entry(ind);
655 /* Handle any machine specific cleanup */
656 machine_kexec_cleanup(image);
658 /* Free the kexec control pages... */
659 kimage_free_page_list(&image->control_pages);
663 static kimage_entry_t *kimage_dst_used(struct kimage *image,
666 kimage_entry_t *ptr, entry;
667 unsigned long destination = 0;
669 for_each_kimage_entry(image, ptr, entry) {
670 if (entry & IND_DESTINATION)
671 destination = entry & PAGE_MASK;
672 else if (entry & IND_SOURCE) {
673 if (page == destination)
675 destination += PAGE_SIZE;
682 static struct page *kimage_alloc_page(struct kimage *image,
684 unsigned long destination)
687 * Here we implement safeguards to ensure that a source page
688 * is not copied to its destination page before the data on
689 * the destination page is no longer useful.
691 * To do this we maintain the invariant that a source page is
692 * either its own destination page, or it is not a
693 * destination page at all.
695 * That is slightly stronger than required, but the proof
696 * that no problems will not occur is trivial, and the
697 * implementation is simply to verify.
699 * When allocating all pages normally this algorithm will run
700 * in O(N) time, but in the worst case it will run in O(N^2)
701 * time. If the runtime is a problem the data structures can
708 * Walk through the list of destination pages, and see if I
711 list_for_each_entry(page, &image->dest_pages, lru) {
712 addr = page_to_pfn(page) << PAGE_SHIFT;
713 if (addr == destination) {
714 list_del(&page->lru);
722 /* Allocate a page, if we run out of memory give up */
723 page = kimage_alloc_pages(gfp_mask, 0);
726 /* If the page cannot be used file it away */
727 if (page_to_pfn(page) >
728 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
729 list_add(&page->lru, &image->unuseable_pages);
732 addr = page_to_pfn(page) << PAGE_SHIFT;
734 /* If it is the destination page we want use it */
735 if (addr == destination)
738 /* If the page is not a destination page use it */
739 if (!kimage_is_destination_range(image, addr,
744 * I know that the page is someones destination page.
745 * See if there is already a source page for this
746 * destination page. And if so swap the source pages.
748 old = kimage_dst_used(image, addr);
751 unsigned long old_addr;
752 struct page *old_page;
754 old_addr = *old & PAGE_MASK;
755 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
756 copy_highpage(page, old_page);
757 *old = addr | (*old & ~PAGE_MASK);
759 /* The old page I have found cannot be a
760 * destination page, so return it if it's
761 * gfp_flags honor the ones passed in.
763 if (!(gfp_mask & __GFP_HIGHMEM) &&
764 PageHighMem(old_page)) {
765 kimage_free_pages(old_page);
773 /* Place the page on the destination list I
776 list_add(&page->lru, &image->dest_pages);
783 static int kimage_load_normal_segment(struct kimage *image,
784 struct kexec_segment *segment)
787 unsigned long ubytes, mbytes;
789 unsigned char __user *buf;
793 ubytes = segment->bufsz;
794 mbytes = segment->memsz;
795 maddr = segment->mem;
797 result = kimage_set_destination(image, maddr);
804 size_t uchunk, mchunk;
806 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
811 result = kimage_add_page(image, page_to_pfn(page)
817 /* Start with a clear page */
819 ptr += maddr & ~PAGE_MASK;
820 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
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 unsigned long 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 = PAGE_SIZE - (maddr & ~PAGE_MASK);
877 if (uchunk > ubytes) {
879 /* Zero the trailing part of the page */
880 memset(ptr + uchunk, 0, mchunk - uchunk);
882 result = copy_from_user(ptr, buf, uchunk);
883 kexec_flush_icache_page(page);
898 static int kimage_load_segment(struct kimage *image,
899 struct kexec_segment *segment)
901 int result = -ENOMEM;
903 switch (image->type) {
904 case KEXEC_TYPE_DEFAULT:
905 result = kimage_load_normal_segment(image, segment);
907 case KEXEC_TYPE_CRASH:
908 result = kimage_load_crash_segment(image, segment);
916 * Exec Kernel system call: for obvious reasons only root may call it.
918 * This call breaks up into three pieces.
919 * - A generic part which loads the new kernel from the current
920 * address space, and very carefully places the data in the
923 * - A generic part that interacts with the kernel and tells all of
924 * the devices to shut down. Preventing on-going dmas, and placing
925 * the devices in a consistent state so a later kernel can
928 * - A machine specific part that includes the syscall number
929 * and the copies the image to it's final destination. And
930 * jumps into the image at entry.
932 * kexec does not sync, or unmount filesystems so if you need
933 * that to happen you need to do that yourself.
935 struct kimage *kexec_image;
936 struct kimage *kexec_crash_image;
938 static DEFINE_MUTEX(kexec_mutex);
940 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
941 struct kexec_segment __user *, segments, unsigned long, flags)
943 struct kimage **dest_image, *image;
946 /* We only trust the superuser with rebooting the system. */
947 if (!capable(CAP_SYS_BOOT))
951 * Verify we have a legal set of flags
952 * This leaves us room for future extensions.
954 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
957 /* Verify we are on the appropriate architecture */
958 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
959 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
962 /* Put an artificial cap on the number
963 * of segments passed to kexec_load.
965 if (nr_segments > KEXEC_SEGMENT_MAX)
971 /* Because we write directly to the reserved memory
972 * region when loading crash kernels we need a mutex here to
973 * prevent multiple crash kernels from attempting to load
974 * simultaneously, and to prevent a crash kernel from loading
975 * over the top of a in use crash kernel.
977 * KISS: always take the mutex.
979 if (!mutex_trylock(&kexec_mutex))
982 dest_image = &kexec_image;
983 if (flags & KEXEC_ON_CRASH)
984 dest_image = &kexec_crash_image;
985 if (nr_segments > 0) {
988 /* Loading another kernel to reboot into */
989 if ((flags & KEXEC_ON_CRASH) == 0)
990 result = kimage_normal_alloc(&image, entry,
991 nr_segments, segments);
992 /* Loading another kernel to switch to if this one crashes */
993 else if (flags & KEXEC_ON_CRASH) {
994 /* Free any current crash dump kernel before
997 kimage_free(xchg(&kexec_crash_image, NULL));
998 result = kimage_crash_alloc(&image, entry,
999 nr_segments, segments);
1000 crash_map_reserved_pages();
1005 if (flags & KEXEC_PRESERVE_CONTEXT)
1006 image->preserve_context = 1;
1007 result = machine_kexec_prepare(image);
1011 for (i = 0; i < nr_segments; i++) {
1012 result = kimage_load_segment(image, &image->segment[i]);
1016 kimage_terminate(image);
1017 if (flags & KEXEC_ON_CRASH)
1018 crash_unmap_reserved_pages();
1020 /* Install the new kernel, and Uninstall the old */
1021 image = xchg(dest_image, image);
1024 mutex_unlock(&kexec_mutex);
1031 * Add and remove page tables for crashkernel memory
1033 * Provide an empty default implementation here -- architecture
1034 * code may override this
1036 void __weak crash_map_reserved_pages(void)
1039 void __weak crash_unmap_reserved_pages(void)
1042 #ifdef CONFIG_COMPAT
1043 asmlinkage long compat_sys_kexec_load(unsigned long entry,
1044 unsigned long nr_segments,
1045 struct compat_kexec_segment __user *segments,
1046 unsigned long flags)
1048 struct compat_kexec_segment in;
1049 struct kexec_segment out, __user *ksegments;
1050 unsigned long i, result;
1052 /* Don't allow clients that don't understand the native
1053 * architecture to do anything.
1055 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1058 if (nr_segments > KEXEC_SEGMENT_MAX)
1061 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1062 for (i=0; i < nr_segments; i++) {
1063 result = copy_from_user(&in, &segments[i], sizeof(in));
1067 out.buf = compat_ptr(in.buf);
1068 out.bufsz = in.bufsz;
1070 out.memsz = in.memsz;
1072 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1077 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1081 void crash_kexec(struct pt_regs *regs)
1083 /* Take the kexec_mutex here to prevent sys_kexec_load
1084 * running on one cpu from replacing the crash kernel
1085 * we are using after a panic on a different cpu.
1087 * If the crash kernel was not located in a fixed area
1088 * of memory the xchg(&kexec_crash_image) would be
1089 * sufficient. But since I reuse the memory...
1091 if (mutex_trylock(&kexec_mutex)) {
1092 if (kexec_crash_image) {
1093 struct pt_regs fixed_regs;
1095 crash_setup_regs(&fixed_regs, regs);
1096 crash_save_vmcoreinfo();
1097 machine_crash_shutdown(&fixed_regs);
1098 machine_kexec(kexec_crash_image);
1100 mutex_unlock(&kexec_mutex);
1104 size_t crash_get_memory_size(void)
1107 mutex_lock(&kexec_mutex);
1108 if (crashk_res.end != crashk_res.start)
1109 size = resource_size(&crashk_res);
1110 mutex_unlock(&kexec_mutex);
1114 void __weak crash_free_reserved_phys_range(unsigned long begin,
1119 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1120 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1121 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1122 free_page((unsigned long)__va(addr));
1127 int crash_shrink_memory(unsigned long new_size)
1130 unsigned long start, end;
1131 unsigned long old_size;
1132 struct resource *ram_res;
1134 mutex_lock(&kexec_mutex);
1136 if (kexec_crash_image) {
1140 start = crashk_res.start;
1141 end = crashk_res.end;
1142 old_size = (end == 0) ? 0 : end - start + 1;
1143 if (new_size >= old_size) {
1144 ret = (new_size == old_size) ? 0 : -EINVAL;
1148 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1154 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1155 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1157 crash_map_reserved_pages();
1158 crash_free_reserved_phys_range(end, crashk_res.end);
1160 if ((start == end) && (crashk_res.parent != NULL))
1161 release_resource(&crashk_res);
1163 ram_res->start = end;
1164 ram_res->end = crashk_res.end;
1165 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1166 ram_res->name = "System RAM";
1168 crashk_res.end = end - 1;
1170 insert_resource(&iomem_resource, ram_res);
1171 crash_unmap_reserved_pages();
1174 mutex_unlock(&kexec_mutex);
1178 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1181 struct elf_note note;
1183 note.n_namesz = strlen(name) + 1;
1184 note.n_descsz = data_len;
1186 memcpy(buf, ¬e, sizeof(note));
1187 buf += (sizeof(note) + 3)/4;
1188 memcpy(buf, name, note.n_namesz);
1189 buf += (note.n_namesz + 3)/4;
1190 memcpy(buf, data, note.n_descsz);
1191 buf += (note.n_descsz + 3)/4;
1196 static void final_note(u32 *buf)
1198 struct elf_note note;
1203 memcpy(buf, ¬e, sizeof(note));
1206 void crash_save_cpu(struct pt_regs *regs, int cpu)
1208 struct elf_prstatus prstatus;
1211 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1214 /* Using ELF notes here is opportunistic.
1215 * I need a well defined structure format
1216 * for the data I pass, and I need tags
1217 * on the data to indicate what information I have
1218 * squirrelled away. ELF notes happen to provide
1219 * all of that, so there is no need to invent something new.
1221 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1224 memset(&prstatus, 0, sizeof(prstatus));
1225 prstatus.pr_pid = current->pid;
1226 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1227 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1228 &prstatus, sizeof(prstatus));
1232 static int __init crash_notes_memory_init(void)
1234 /* Allocate memory for saving cpu registers. */
1235 crash_notes = alloc_percpu(note_buf_t);
1237 printk("Kexec: Memory allocation for saving cpu register"
1238 " states failed\n");
1243 module_init(crash_notes_memory_init)
1247 * parsing the "crashkernel" commandline
1249 * this code is intended to be called from architecture specific code
1254 * This function parses command lines in the format
1256 * crashkernel=ramsize-range:size[,...][@offset]
1258 * The function returns 0 on success and -EINVAL on failure.
1260 static int __init parse_crashkernel_mem(char *cmdline,
1261 unsigned long long system_ram,
1262 unsigned long long *crash_size,
1263 unsigned long long *crash_base)
1265 char *cur = cmdline, *tmp;
1267 /* for each entry of the comma-separated list */
1269 unsigned long long start, end = ULLONG_MAX, size;
1271 /* get the start of the range */
1272 start = memparse(cur, &tmp);
1274 pr_warning("crashkernel: Memory value expected\n");
1279 pr_warning("crashkernel: '-' expected\n");
1284 /* if no ':' is here, than we read the end */
1286 end = memparse(cur, &tmp);
1288 pr_warning("crashkernel: Memory "
1289 "value expected\n");
1294 pr_warning("crashkernel: end <= start\n");
1300 pr_warning("crashkernel: ':' expected\n");
1305 size = memparse(cur, &tmp);
1307 pr_warning("Memory value expected\n");
1311 if (size >= system_ram) {
1312 pr_warning("crashkernel: invalid size\n");
1317 if (system_ram >= start && system_ram < end) {
1321 } while (*cur++ == ',');
1323 if (*crash_size > 0) {
1324 while (*cur && *cur != ' ' && *cur != '@')
1328 *crash_base = memparse(cur, &tmp);
1330 pr_warning("Memory value expected "
1341 * That function parses "simple" (old) crashkernel command lines like
1343 * crashkernel=size[@offset]
1345 * It returns 0 on success and -EINVAL on failure.
1347 static int __init parse_crashkernel_simple(char *cmdline,
1348 unsigned long long *crash_size,
1349 unsigned long long *crash_base)
1351 char *cur = cmdline;
1353 *crash_size = memparse(cmdline, &cur);
1354 if (cmdline == cur) {
1355 pr_warning("crashkernel: memory value expected\n");
1360 *crash_base = memparse(cur+1, &cur);
1361 else if (*cur != ' ' && *cur != '\0') {
1362 pr_warning("crashkernel: unrecognized char\n");
1370 * That function is the entry point for command line parsing and should be
1371 * called from the arch-specific code.
1373 int __init parse_crashkernel(char *cmdline,
1374 unsigned long long system_ram,
1375 unsigned long long *crash_size,
1376 unsigned long long *crash_base)
1378 char *p = cmdline, *ck_cmdline = NULL;
1379 char *first_colon, *first_space;
1381 BUG_ON(!crash_size || !crash_base);
1385 /* find crashkernel and use the last one if there are more */
1386 p = strstr(p, "crashkernel=");
1389 p = strstr(p+1, "crashkernel=");
1395 ck_cmdline += 12; /* strlen("crashkernel=") */
1398 * if the commandline contains a ':', then that's the extended
1399 * syntax -- if not, it must be the classic syntax
1401 first_colon = strchr(ck_cmdline, ':');
1402 first_space = strchr(ck_cmdline, ' ');
1403 if (first_colon && (!first_space || first_colon < first_space))
1404 return parse_crashkernel_mem(ck_cmdline, system_ram,
1405 crash_size, crash_base);
1407 return parse_crashkernel_simple(ck_cmdline, crash_size,
1414 static void update_vmcoreinfo_note(void)
1416 u32 *buf = vmcoreinfo_note;
1418 if (!vmcoreinfo_size)
1420 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1425 void crash_save_vmcoreinfo(void)
1427 vmcoreinfo_append_str("CRASHTIME=%ld", get_seconds());
1428 update_vmcoreinfo_note();
1431 void vmcoreinfo_append_str(const char *fmt, ...)
1437 va_start(args, fmt);
1438 r = vsnprintf(buf, sizeof(buf), fmt, args);
1441 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1442 r = vmcoreinfo_max_size - vmcoreinfo_size;
1444 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1446 vmcoreinfo_size += r;
1450 * provide an empty default implementation here -- architecture
1451 * code may override this
1453 void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1456 unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1458 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1461 static int __init crash_save_vmcoreinfo_init(void)
1463 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1464 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1466 VMCOREINFO_SYMBOL(init_uts_ns);
1467 VMCOREINFO_SYMBOL(node_online_map);
1469 VMCOREINFO_SYMBOL(swapper_pg_dir);
1471 VMCOREINFO_SYMBOL(_stext);
1472 VMCOREINFO_SYMBOL(vmlist);
1474 #ifndef CONFIG_NEED_MULTIPLE_NODES
1475 VMCOREINFO_SYMBOL(mem_map);
1476 VMCOREINFO_SYMBOL(contig_page_data);
1478 #ifdef CONFIG_SPARSEMEM
1479 VMCOREINFO_SYMBOL(mem_section);
1480 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1481 VMCOREINFO_STRUCT_SIZE(mem_section);
1482 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1484 VMCOREINFO_STRUCT_SIZE(page);
1485 VMCOREINFO_STRUCT_SIZE(pglist_data);
1486 VMCOREINFO_STRUCT_SIZE(zone);
1487 VMCOREINFO_STRUCT_SIZE(free_area);
1488 VMCOREINFO_STRUCT_SIZE(list_head);
1489 VMCOREINFO_SIZE(nodemask_t);
1490 VMCOREINFO_OFFSET(page, flags);
1491 VMCOREINFO_OFFSET(page, _count);
1492 VMCOREINFO_OFFSET(page, mapping);
1493 VMCOREINFO_OFFSET(page, lru);
1494 VMCOREINFO_OFFSET(pglist_data, node_zones);
1495 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1496 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1497 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1499 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1500 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1501 VMCOREINFO_OFFSET(pglist_data, node_id);
1502 VMCOREINFO_OFFSET(zone, free_area);
1503 VMCOREINFO_OFFSET(zone, vm_stat);
1504 VMCOREINFO_OFFSET(zone, spanned_pages);
1505 VMCOREINFO_OFFSET(free_area, free_list);
1506 VMCOREINFO_OFFSET(list_head, next);
1507 VMCOREINFO_OFFSET(list_head, prev);
1508 VMCOREINFO_OFFSET(vm_struct, addr);
1509 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1510 log_buf_kexec_setup();
1511 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1512 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1513 VMCOREINFO_NUMBER(PG_lru);
1514 VMCOREINFO_NUMBER(PG_private);
1515 VMCOREINFO_NUMBER(PG_swapcache);
1517 arch_crash_save_vmcoreinfo();
1518 update_vmcoreinfo_note();
1523 module_init(crash_save_vmcoreinfo_init)
1526 * Move into place and start executing a preloaded standalone
1527 * executable. If nothing was preloaded return an error.
1529 int kernel_kexec(void)
1533 if (!mutex_trylock(&kexec_mutex))
1540 #ifdef CONFIG_KEXEC_JUMP
1541 if (kexec_image->preserve_context) {
1542 lock_system_sleep();
1543 pm_prepare_console();
1544 error = freeze_processes();
1547 goto Restore_console;
1550 error = dpm_suspend_start(PMSG_FREEZE);
1552 goto Resume_console;
1553 /* At this point, dpm_suspend_start() has been called,
1554 * but *not* dpm_suspend_end(). We *must* call
1555 * dpm_suspend_end() now. Otherwise, drivers for
1556 * some devices (e.g. interrupt controllers) become
1557 * desynchronized with the actual state of the
1558 * hardware at resume time, and evil weirdness ensues.
1560 error = dpm_suspend_end(PMSG_FREEZE);
1562 goto Resume_devices;
1563 error = disable_nonboot_cpus();
1566 local_irq_disable();
1567 error = syscore_suspend();
1573 kernel_restart_prepare(NULL);
1574 printk(KERN_EMERG "Starting new kernel\n");
1578 machine_kexec(kexec_image);
1580 #ifdef CONFIG_KEXEC_JUMP
1581 if (kexec_image->preserve_context) {
1586 enable_nonboot_cpus();
1587 dpm_resume_start(PMSG_RESTORE);
1589 dpm_resume_end(PMSG_RESTORE);
1594 pm_restore_console();
1595 unlock_system_sleep();
1600 mutex_unlock(&kexec_mutex);