2 * kexec.c - kexec system call core code.
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/capability.h>
13 #include <linux/file.h>
14 #include <linux/slab.h>
16 #include <linux/kexec.h>
17 #include <linux/mutex.h>
18 #include <linux/list.h>
19 #include <linux/highmem.h>
20 #include <linux/syscalls.h>
21 #include <linux/reboot.h>
22 #include <linux/ioport.h>
23 #include <linux/hardirq.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/utsname.h>
27 #include <linux/numa.h>
28 #include <linux/suspend.h>
29 #include <linux/device.h>
30 #include <linux/freezer.h>
32 #include <linux/cpu.h>
33 #include <linux/uaccess.h>
35 #include <linux/console.h>
36 #include <linux/vmalloc.h>
37 #include <linux/swap.h>
38 #include <linux/syscore_ops.h>
39 #include <linux/compiler.h>
40 #include <linux/hugetlb.h>
43 #include <asm/sections.h>
45 #include <crypto/hash.h>
46 #include <crypto/sha.h>
47 #include "kexec_internal.h"
49 DEFINE_MUTEX(kexec_mutex);
51 /* Per cpu memory for storing cpu states in case of system crash. */
52 note_buf_t __percpu *crash_notes;
54 /* vmcoreinfo stuff */
55 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
56 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
57 size_t vmcoreinfo_size;
58 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
60 /* Flag to indicate we are going to kexec a new kernel */
61 bool kexec_in_progress = false;
64 /* Location of the reserved area for the crash kernel */
65 struct resource crashk_res = {
66 .name = "Crash kernel",
69 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
70 .desc = IORES_DESC_CRASH_KERNEL
72 struct resource crashk_low_res = {
73 .name = "Crash kernel",
76 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
77 .desc = IORES_DESC_CRASH_KERNEL
80 int kexec_should_crash(struct task_struct *p)
83 * If crash_kexec_post_notifiers is enabled, don't run
84 * crash_kexec() here yet, which must be run after panic
85 * notifiers in panic().
87 if (crash_kexec_post_notifiers)
90 * There are 4 panic() calls in do_exit() path, each of which
91 * corresponds to each of these 4 conditions.
93 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
99 * When kexec transitions to the new kernel there is a one-to-one
100 * mapping between physical and virtual addresses. On processors
101 * where you can disable the MMU this is trivial, and easy. For
102 * others it is still a simple predictable page table to setup.
104 * In that environment kexec copies the new kernel to its final
105 * resting place. This means I can only support memory whose
106 * physical address can fit in an unsigned long. In particular
107 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
108 * If the assembly stub has more restrictive requirements
109 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
110 * defined more restrictively in <asm/kexec.h>.
112 * The code for the transition from the current kernel to the
113 * the new kernel is placed in the control_code_buffer, whose size
114 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
115 * page of memory is necessary, but some architectures require more.
116 * Because this memory must be identity mapped in the transition from
117 * virtual to physical addresses it must live in the range
118 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
121 * The assembly stub in the control code buffer is passed a linked list
122 * of descriptor pages detailing the source pages of the new kernel,
123 * and the destination addresses of those source pages. As this data
124 * structure is not used in the context of the current OS, it must
127 * The code has been made to work with highmem pages and will use a
128 * destination page in its final resting place (if it happens
129 * to allocate it). The end product of this is that most of the
130 * physical address space, and most of RAM can be used.
132 * Future directions include:
133 * - allocating a page table with the control code buffer identity
134 * mapped, to simplify machine_kexec and make kexec_on_panic more
139 * KIMAGE_NO_DEST is an impossible destination address..., for
140 * allocating pages whose destination address we do not care about.
142 #define KIMAGE_NO_DEST (-1UL)
144 static struct page *kimage_alloc_page(struct kimage *image,
148 int sanity_check_segment_list(struct kimage *image)
151 unsigned long nr_segments = image->nr_segments;
154 * Verify we have good destination addresses. The caller is
155 * responsible for making certain we don't attempt to load
156 * the new image into invalid or reserved areas of RAM. This
157 * just verifies it is an address we can use.
159 * Since the kernel does everything in page size chunks ensure
160 * the destination addresses are page aligned. Too many
161 * special cases crop of when we don't do this. The most
162 * insidious is getting overlapping destination addresses
163 * simply because addresses are changed to page size
166 for (i = 0; i < nr_segments; i++) {
167 unsigned long mstart, mend;
169 mstart = image->segment[i].mem;
170 mend = mstart + image->segment[i].memsz;
172 return -EADDRNOTAVAIL;
173 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
174 return -EADDRNOTAVAIL;
175 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
176 return -EADDRNOTAVAIL;
179 /* Verify our destination addresses do not overlap.
180 * If we alloed overlapping destination addresses
181 * through very weird things can happen with no
182 * easy explanation as one segment stops on another.
184 for (i = 0; i < nr_segments; i++) {
185 unsigned long mstart, mend;
188 mstart = image->segment[i].mem;
189 mend = mstart + image->segment[i].memsz;
190 for (j = 0; j < i; j++) {
191 unsigned long pstart, pend;
193 pstart = image->segment[j].mem;
194 pend = pstart + image->segment[j].memsz;
195 /* Do the segments overlap ? */
196 if ((mend > pstart) && (mstart < pend))
201 /* Ensure our buffer sizes are strictly less than
202 * our memory sizes. This should always be the case,
203 * and it is easier to check up front than to be surprised
206 for (i = 0; i < nr_segments; i++) {
207 if (image->segment[i].bufsz > image->segment[i].memsz)
212 * Verify we have good destination addresses. Normally
213 * the caller is responsible for making certain we don't
214 * attempt to load the new image into invalid or reserved
215 * areas of RAM. But crash kernels are preloaded into a
216 * reserved area of ram. We must ensure the addresses
217 * are in the reserved area otherwise preloading the
218 * kernel could corrupt things.
221 if (image->type == KEXEC_TYPE_CRASH) {
222 for (i = 0; i < nr_segments; i++) {
223 unsigned long mstart, mend;
225 mstart = image->segment[i].mem;
226 mend = mstart + image->segment[i].memsz - 1;
227 /* Ensure we are within the crash kernel limits */
228 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
229 (mend > phys_to_boot_phys(crashk_res.end)))
230 return -EADDRNOTAVAIL;
237 struct kimage *do_kimage_alloc_init(void)
239 struct kimage *image;
241 /* Allocate a controlling structure */
242 image = kzalloc(sizeof(*image), GFP_KERNEL);
247 image->entry = &image->head;
248 image->last_entry = &image->head;
249 image->control_page = ~0; /* By default this does not apply */
250 image->type = KEXEC_TYPE_DEFAULT;
252 /* Initialize the list of control pages */
253 INIT_LIST_HEAD(&image->control_pages);
255 /* Initialize the list of destination pages */
256 INIT_LIST_HEAD(&image->dest_pages);
258 /* Initialize the list of unusable pages */
259 INIT_LIST_HEAD(&image->unusable_pages);
264 int kimage_is_destination_range(struct kimage *image,
270 for (i = 0; i < image->nr_segments; i++) {
271 unsigned long mstart, mend;
273 mstart = image->segment[i].mem;
274 mend = mstart + image->segment[i].memsz;
275 if ((end > mstart) && (start < mend))
282 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
286 pages = alloc_pages(gfp_mask, order);
288 unsigned int count, i;
290 pages->mapping = NULL;
291 set_page_private(pages, order);
293 for (i = 0; i < count; i++)
294 SetPageReserved(pages + i);
300 static void kimage_free_pages(struct page *page)
302 unsigned int order, count, i;
304 order = page_private(page);
306 for (i = 0; i < count; i++)
307 ClearPageReserved(page + i);
308 __free_pages(page, order);
311 void kimage_free_page_list(struct list_head *list)
313 struct page *page, *next;
315 list_for_each_entry_safe(page, next, list, lru) {
316 list_del(&page->lru);
317 kimage_free_pages(page);
321 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
324 /* Control pages are special, they are the intermediaries
325 * that are needed while we copy the rest of the pages
326 * to their final resting place. As such they must
327 * not conflict with either the destination addresses
328 * or memory the kernel is already using.
330 * The only case where we really need more than one of
331 * these are for architectures where we cannot disable
332 * the MMU and must instead generate an identity mapped
333 * page table for all of the memory.
335 * At worst this runs in O(N) of the image size.
337 struct list_head extra_pages;
342 INIT_LIST_HEAD(&extra_pages);
344 /* Loop while I can allocate a page and the page allocated
345 * is a destination page.
348 unsigned long pfn, epfn, addr, eaddr;
350 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
353 pfn = page_to_boot_pfn(pages);
355 addr = pfn << PAGE_SHIFT;
356 eaddr = epfn << PAGE_SHIFT;
357 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
358 kimage_is_destination_range(image, addr, eaddr)) {
359 list_add(&pages->lru, &extra_pages);
365 /* Remember the allocated page... */
366 list_add(&pages->lru, &image->control_pages);
368 /* Because the page is already in it's destination
369 * location we will never allocate another page at
370 * that address. Therefore kimage_alloc_pages
371 * will not return it (again) and we don't need
372 * to give it an entry in image->segment[].
375 /* Deal with the destination pages I have inadvertently allocated.
377 * Ideally I would convert multi-page allocations into single
378 * page allocations, and add everything to image->dest_pages.
380 * For now it is simpler to just free the pages.
382 kimage_free_page_list(&extra_pages);
387 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
390 /* Control pages are special, they are the intermediaries
391 * that are needed while we copy the rest of the pages
392 * to their final resting place. As such they must
393 * not conflict with either the destination addresses
394 * or memory the kernel is already using.
396 * Control pages are also the only pags we must allocate
397 * when loading a crash kernel. All of the other pages
398 * are specified by the segments and we just memcpy
399 * into them directly.
401 * The only case where we really need more than one of
402 * these are for architectures where we cannot disable
403 * the MMU and must instead generate an identity mapped
404 * page table for all of the memory.
406 * Given the low demand this implements a very simple
407 * allocator that finds the first hole of the appropriate
408 * size in the reserved memory region, and allocates all
409 * of the memory up to and including the hole.
411 unsigned long hole_start, hole_end, size;
415 size = (1 << order) << PAGE_SHIFT;
416 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
417 hole_end = hole_start + size - 1;
418 while (hole_end <= crashk_res.end) {
421 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
423 /* See if I overlap any of the segments */
424 for (i = 0; i < image->nr_segments; i++) {
425 unsigned long mstart, mend;
427 mstart = image->segment[i].mem;
428 mend = mstart + image->segment[i].memsz - 1;
429 if ((hole_end >= mstart) && (hole_start <= mend)) {
430 /* Advance the hole to the end of the segment */
431 hole_start = (mend + (size - 1)) & ~(size - 1);
432 hole_end = hole_start + size - 1;
436 /* If I don't overlap any segments I have found my hole! */
437 if (i == image->nr_segments) {
438 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
439 image->control_page = hole_end;
448 struct page *kimage_alloc_control_pages(struct kimage *image,
451 struct page *pages = NULL;
453 switch (image->type) {
454 case KEXEC_TYPE_DEFAULT:
455 pages = kimage_alloc_normal_control_pages(image, order);
457 case KEXEC_TYPE_CRASH:
458 pages = kimage_alloc_crash_control_pages(image, order);
465 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
467 if (*image->entry != 0)
470 if (image->entry == image->last_entry) {
471 kimage_entry_t *ind_page;
474 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
478 ind_page = page_address(page);
479 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
480 image->entry = ind_page;
481 image->last_entry = ind_page +
482 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
484 *image->entry = entry;
491 static int kimage_set_destination(struct kimage *image,
492 unsigned long destination)
496 destination &= PAGE_MASK;
497 result = kimage_add_entry(image, destination | IND_DESTINATION);
503 static int kimage_add_page(struct kimage *image, unsigned long page)
508 result = kimage_add_entry(image, page | IND_SOURCE);
514 static void kimage_free_extra_pages(struct kimage *image)
516 /* Walk through and free any extra destination pages I may have */
517 kimage_free_page_list(&image->dest_pages);
519 /* Walk through and free any unusable pages I have cached */
520 kimage_free_page_list(&image->unusable_pages);
523 void kimage_terminate(struct kimage *image)
525 if (*image->entry != 0)
528 *image->entry = IND_DONE;
531 #define for_each_kimage_entry(image, ptr, entry) \
532 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
533 ptr = (entry & IND_INDIRECTION) ? \
534 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
536 static void kimage_free_entry(kimage_entry_t entry)
540 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
541 kimage_free_pages(page);
544 void kimage_free(struct kimage *image)
546 kimage_entry_t *ptr, entry;
547 kimage_entry_t ind = 0;
552 kimage_free_extra_pages(image);
553 for_each_kimage_entry(image, ptr, entry) {
554 if (entry & IND_INDIRECTION) {
555 /* Free the previous indirection page */
556 if (ind & IND_INDIRECTION)
557 kimage_free_entry(ind);
558 /* Save this indirection page until we are
562 } else if (entry & IND_SOURCE)
563 kimage_free_entry(entry);
565 /* Free the final indirection page */
566 if (ind & IND_INDIRECTION)
567 kimage_free_entry(ind);
569 /* Handle any machine specific cleanup */
570 machine_kexec_cleanup(image);
572 /* Free the kexec control pages... */
573 kimage_free_page_list(&image->control_pages);
576 * Free up any temporary buffers allocated. This might hit if
577 * error occurred much later after buffer allocation.
579 if (image->file_mode)
580 kimage_file_post_load_cleanup(image);
585 static kimage_entry_t *kimage_dst_used(struct kimage *image,
588 kimage_entry_t *ptr, entry;
589 unsigned long destination = 0;
591 for_each_kimage_entry(image, ptr, entry) {
592 if (entry & IND_DESTINATION)
593 destination = entry & PAGE_MASK;
594 else if (entry & IND_SOURCE) {
595 if (page == destination)
597 destination += PAGE_SIZE;
604 static struct page *kimage_alloc_page(struct kimage *image,
606 unsigned long destination)
609 * Here we implement safeguards to ensure that a source page
610 * is not copied to its destination page before the data on
611 * the destination page is no longer useful.
613 * To do this we maintain the invariant that a source page is
614 * either its own destination page, or it is not a
615 * destination page at all.
617 * That is slightly stronger than required, but the proof
618 * that no problems will not occur is trivial, and the
619 * implementation is simply to verify.
621 * When allocating all pages normally this algorithm will run
622 * in O(N) time, but in the worst case it will run in O(N^2)
623 * time. If the runtime is a problem the data structures can
630 * Walk through the list of destination pages, and see if I
633 list_for_each_entry(page, &image->dest_pages, lru) {
634 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
635 if (addr == destination) {
636 list_del(&page->lru);
644 /* Allocate a page, if we run out of memory give up */
645 page = kimage_alloc_pages(gfp_mask, 0);
648 /* If the page cannot be used file it away */
649 if (page_to_boot_pfn(page) >
650 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
651 list_add(&page->lru, &image->unusable_pages);
654 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
656 /* If it is the destination page we want use it */
657 if (addr == destination)
660 /* If the page is not a destination page use it */
661 if (!kimage_is_destination_range(image, addr,
666 * I know that the page is someones destination page.
667 * See if there is already a source page for this
668 * destination page. And if so swap the source pages.
670 old = kimage_dst_used(image, addr);
673 unsigned long old_addr;
674 struct page *old_page;
676 old_addr = *old & PAGE_MASK;
677 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
678 copy_highpage(page, old_page);
679 *old = addr | (*old & ~PAGE_MASK);
681 /* The old page I have found cannot be a
682 * destination page, so return it if it's
683 * gfp_flags honor the ones passed in.
685 if (!(gfp_mask & __GFP_HIGHMEM) &&
686 PageHighMem(old_page)) {
687 kimage_free_pages(old_page);
694 /* Place the page on the destination list, to be used later */
695 list_add(&page->lru, &image->dest_pages);
701 static int kimage_load_normal_segment(struct kimage *image,
702 struct kexec_segment *segment)
705 size_t ubytes, mbytes;
707 unsigned char __user *buf = NULL;
708 unsigned char *kbuf = NULL;
711 if (image->file_mode)
712 kbuf = segment->kbuf;
715 ubytes = segment->bufsz;
716 mbytes = segment->memsz;
717 maddr = segment->mem;
719 result = kimage_set_destination(image, maddr);
726 size_t uchunk, mchunk;
728 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
733 result = kimage_add_page(image, page_to_boot_pfn(page)
739 /* Start with a clear page */
741 ptr += maddr & ~PAGE_MASK;
742 mchunk = min_t(size_t, mbytes,
743 PAGE_SIZE - (maddr & ~PAGE_MASK));
744 uchunk = min(ubytes, mchunk);
746 /* For file based kexec, source pages are in kernel memory */
747 if (image->file_mode)
748 memcpy(ptr, kbuf, uchunk);
750 result = copy_from_user(ptr, buf, uchunk);
758 if (image->file_mode)
768 static int kimage_load_crash_segment(struct kimage *image,
769 struct kexec_segment *segment)
771 /* For crash dumps kernels we simply copy the data from
772 * user space to it's destination.
773 * We do things a page at a time for the sake of kmap.
776 size_t ubytes, mbytes;
778 unsigned char __user *buf = NULL;
779 unsigned char *kbuf = NULL;
782 if (image->file_mode)
783 kbuf = segment->kbuf;
786 ubytes = segment->bufsz;
787 mbytes = segment->memsz;
788 maddr = segment->mem;
792 size_t uchunk, mchunk;
794 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
800 ptr += maddr & ~PAGE_MASK;
801 mchunk = min_t(size_t, mbytes,
802 PAGE_SIZE - (maddr & ~PAGE_MASK));
803 uchunk = min(ubytes, mchunk);
804 if (mchunk > uchunk) {
805 /* Zero the trailing part of the page */
806 memset(ptr + uchunk, 0, mchunk - uchunk);
809 /* For file based kexec, source pages are in kernel memory */
810 if (image->file_mode)
811 memcpy(ptr, kbuf, uchunk);
813 result = copy_from_user(ptr, buf, uchunk);
814 kexec_flush_icache_page(page);
822 if (image->file_mode)
832 int kimage_load_segment(struct kimage *image,
833 struct kexec_segment *segment)
835 int result = -ENOMEM;
837 switch (image->type) {
838 case KEXEC_TYPE_DEFAULT:
839 result = kimage_load_normal_segment(image, segment);
841 case KEXEC_TYPE_CRASH:
842 result = kimage_load_crash_segment(image, segment);
849 struct kimage *kexec_image;
850 struct kimage *kexec_crash_image;
851 int kexec_load_disabled;
854 * No panic_cpu check version of crash_kexec(). This function is called
855 * only when panic_cpu holds the current CPU number; this is the only CPU
856 * which processes crash_kexec routines.
858 void __crash_kexec(struct pt_regs *regs)
860 /* Take the kexec_mutex here to prevent sys_kexec_load
861 * running on one cpu from replacing the crash kernel
862 * we are using after a panic on a different cpu.
864 * If the crash kernel was not located in a fixed area
865 * of memory the xchg(&kexec_crash_image) would be
866 * sufficient. But since I reuse the memory...
868 if (mutex_trylock(&kexec_mutex)) {
869 if (kexec_crash_image) {
870 struct pt_regs fixed_regs;
872 crash_setup_regs(&fixed_regs, regs);
873 crash_save_vmcoreinfo();
874 machine_crash_shutdown(&fixed_regs);
875 machine_kexec(kexec_crash_image);
877 mutex_unlock(&kexec_mutex);
881 void crash_kexec(struct pt_regs *regs)
883 int old_cpu, this_cpu;
886 * Only one CPU is allowed to execute the crash_kexec() code as with
887 * panic(). Otherwise parallel calls of panic() and crash_kexec()
888 * may stop each other. To exclude them, we use panic_cpu here too.
890 this_cpu = raw_smp_processor_id();
891 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
892 if (old_cpu == PANIC_CPU_INVALID) {
893 /* This is the 1st CPU which comes here, so go ahead. */
894 printk_nmi_flush_on_panic();
898 * Reset panic_cpu to allow another panic()/crash_kexec()
901 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
905 size_t crash_get_memory_size(void)
909 mutex_lock(&kexec_mutex);
910 if (crashk_res.end != crashk_res.start)
911 size = resource_size(&crashk_res);
912 mutex_unlock(&kexec_mutex);
916 void __weak crash_free_reserved_phys_range(unsigned long begin,
921 for (addr = begin; addr < end; addr += PAGE_SIZE)
922 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
925 int crash_shrink_memory(unsigned long new_size)
928 unsigned long start, end;
929 unsigned long old_size;
930 struct resource *ram_res;
932 mutex_lock(&kexec_mutex);
934 if (kexec_crash_image) {
938 start = crashk_res.start;
939 end = crashk_res.end;
940 old_size = (end == 0) ? 0 : end - start + 1;
941 if (new_size >= old_size) {
942 ret = (new_size == old_size) ? 0 : -EINVAL;
946 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
952 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
953 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
955 crash_free_reserved_phys_range(end, crashk_res.end);
957 if ((start == end) && (crashk_res.parent != NULL))
958 release_resource(&crashk_res);
960 ram_res->start = end;
961 ram_res->end = crashk_res.end;
962 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
963 ram_res->name = "System RAM";
965 crashk_res.end = end - 1;
967 insert_resource(&iomem_resource, ram_res);
970 mutex_unlock(&kexec_mutex);
974 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
977 struct elf_note note;
979 note.n_namesz = strlen(name) + 1;
980 note.n_descsz = data_len;
982 memcpy(buf, ¬e, sizeof(note));
983 buf += (sizeof(note) + 3)/4;
984 memcpy(buf, name, note.n_namesz);
985 buf += (note.n_namesz + 3)/4;
986 memcpy(buf, data, note.n_descsz);
987 buf += (note.n_descsz + 3)/4;
992 static void final_note(u32 *buf)
994 struct elf_note note;
999 memcpy(buf, ¬e, sizeof(note));
1002 void crash_save_cpu(struct pt_regs *regs, int cpu)
1004 struct elf_prstatus prstatus;
1007 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1010 /* Using ELF notes here is opportunistic.
1011 * I need a well defined structure format
1012 * for the data I pass, and I need tags
1013 * on the data to indicate what information I have
1014 * squirrelled away. ELF notes happen to provide
1015 * all of that, so there is no need to invent something new.
1017 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1020 memset(&prstatus, 0, sizeof(prstatus));
1021 prstatus.pr_pid = current->pid;
1022 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1023 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1024 &prstatus, sizeof(prstatus));
1028 static int __init crash_notes_memory_init(void)
1030 /* Allocate memory for saving cpu registers. */
1034 * crash_notes could be allocated across 2 vmalloc pages when percpu
1035 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1036 * pages are also on 2 continuous physical pages. In this case the
1037 * 2nd part of crash_notes in 2nd page could be lost since only the
1038 * starting address and size of crash_notes are exported through sysfs.
1039 * Here round up the size of crash_notes to the nearest power of two
1040 * and pass it to __alloc_percpu as align value. This can make sure
1041 * crash_notes is allocated inside one physical page.
1043 size = sizeof(note_buf_t);
1044 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1047 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1048 * definitely will be in 2 pages with that.
1050 BUILD_BUG_ON(size > PAGE_SIZE);
1052 crash_notes = __alloc_percpu(size, align);
1054 pr_warn("Memory allocation for saving cpu register states failed\n");
1059 subsys_initcall(crash_notes_memory_init);
1063 * parsing the "crashkernel" commandline
1065 * this code is intended to be called from architecture specific code
1070 * This function parses command lines in the format
1072 * crashkernel=ramsize-range:size[,...][@offset]
1074 * The function returns 0 on success and -EINVAL on failure.
1076 static int __init parse_crashkernel_mem(char *cmdline,
1077 unsigned long long system_ram,
1078 unsigned long long *crash_size,
1079 unsigned long long *crash_base)
1081 char *cur = cmdline, *tmp;
1083 /* for each entry of the comma-separated list */
1085 unsigned long long start, end = ULLONG_MAX, size;
1087 /* get the start of the range */
1088 start = memparse(cur, &tmp);
1090 pr_warn("crashkernel: Memory value expected\n");
1095 pr_warn("crashkernel: '-' expected\n");
1100 /* if no ':' is here, than we read the end */
1102 end = memparse(cur, &tmp);
1104 pr_warn("crashkernel: Memory value expected\n");
1109 pr_warn("crashkernel: end <= start\n");
1115 pr_warn("crashkernel: ':' expected\n");
1120 size = memparse(cur, &tmp);
1122 pr_warn("Memory value expected\n");
1126 if (size >= system_ram) {
1127 pr_warn("crashkernel: invalid size\n");
1132 if (system_ram >= start && system_ram < end) {
1136 } while (*cur++ == ',');
1138 if (*crash_size > 0) {
1139 while (*cur && *cur != ' ' && *cur != '@')
1143 *crash_base = memparse(cur, &tmp);
1145 pr_warn("Memory value expected after '@'\n");
1155 * That function parses "simple" (old) crashkernel command lines like
1157 * crashkernel=size[@offset]
1159 * It returns 0 on success and -EINVAL on failure.
1161 static int __init parse_crashkernel_simple(char *cmdline,
1162 unsigned long long *crash_size,
1163 unsigned long long *crash_base)
1165 char *cur = cmdline;
1167 *crash_size = memparse(cmdline, &cur);
1168 if (cmdline == cur) {
1169 pr_warn("crashkernel: memory value expected\n");
1174 *crash_base = memparse(cur+1, &cur);
1175 else if (*cur != ' ' && *cur != '\0') {
1176 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1183 #define SUFFIX_HIGH 0
1184 #define SUFFIX_LOW 1
1185 #define SUFFIX_NULL 2
1186 static __initdata char *suffix_tbl[] = {
1187 [SUFFIX_HIGH] = ",high",
1188 [SUFFIX_LOW] = ",low",
1189 [SUFFIX_NULL] = NULL,
1193 * That function parses "suffix" crashkernel command lines like
1195 * crashkernel=size,[high|low]
1197 * It returns 0 on success and -EINVAL on failure.
1199 static int __init parse_crashkernel_suffix(char *cmdline,
1200 unsigned long long *crash_size,
1203 char *cur = cmdline;
1205 *crash_size = memparse(cmdline, &cur);
1206 if (cmdline == cur) {
1207 pr_warn("crashkernel: memory value expected\n");
1211 /* check with suffix */
1212 if (strncmp(cur, suffix, strlen(suffix))) {
1213 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1216 cur += strlen(suffix);
1217 if (*cur != ' ' && *cur != '\0') {
1218 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1225 static __init char *get_last_crashkernel(char *cmdline,
1229 char *p = cmdline, *ck_cmdline = NULL;
1231 /* find crashkernel and use the last one if there are more */
1232 p = strstr(p, name);
1234 char *end_p = strchr(p, ' ');
1238 end_p = p + strlen(p);
1243 /* skip the one with any known suffix */
1244 for (i = 0; suffix_tbl[i]; i++) {
1245 q = end_p - strlen(suffix_tbl[i]);
1246 if (!strncmp(q, suffix_tbl[i],
1247 strlen(suffix_tbl[i])))
1252 q = end_p - strlen(suffix);
1253 if (!strncmp(q, suffix, strlen(suffix)))
1257 p = strstr(p+1, name);
1266 static int __init __parse_crashkernel(char *cmdline,
1267 unsigned long long system_ram,
1268 unsigned long long *crash_size,
1269 unsigned long long *crash_base,
1273 char *first_colon, *first_space;
1276 BUG_ON(!crash_size || !crash_base);
1280 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1285 ck_cmdline += strlen(name);
1288 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1291 * if the commandline contains a ':', then that's the extended
1292 * syntax -- if not, it must be the classic syntax
1294 first_colon = strchr(ck_cmdline, ':');
1295 first_space = strchr(ck_cmdline, ' ');
1296 if (first_colon && (!first_space || first_colon < first_space))
1297 return parse_crashkernel_mem(ck_cmdline, system_ram,
1298 crash_size, crash_base);
1300 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1304 * That function is the entry point for command line parsing and should be
1305 * called from the arch-specific code.
1307 int __init parse_crashkernel(char *cmdline,
1308 unsigned long long system_ram,
1309 unsigned long long *crash_size,
1310 unsigned long long *crash_base)
1312 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1313 "crashkernel=", NULL);
1316 int __init parse_crashkernel_high(char *cmdline,
1317 unsigned long long system_ram,
1318 unsigned long long *crash_size,
1319 unsigned long long *crash_base)
1321 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1322 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1325 int __init parse_crashkernel_low(char *cmdline,
1326 unsigned long long system_ram,
1327 unsigned long long *crash_size,
1328 unsigned long long *crash_base)
1330 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1331 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1334 static void update_vmcoreinfo_note(void)
1336 u32 *buf = vmcoreinfo_note;
1338 if (!vmcoreinfo_size)
1340 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1345 void crash_save_vmcoreinfo(void)
1347 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1348 update_vmcoreinfo_note();
1351 void vmcoreinfo_append_str(const char *fmt, ...)
1357 va_start(args, fmt);
1358 r = vscnprintf(buf, sizeof(buf), fmt, args);
1361 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1363 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1365 vmcoreinfo_size += r;
1369 * provide an empty default implementation here -- architecture
1370 * code may override this
1372 void __weak arch_crash_save_vmcoreinfo(void)
1375 phys_addr_t __weak paddr_vmcoreinfo_note(void)
1377 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1380 static int __init crash_save_vmcoreinfo_init(void)
1382 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1383 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1385 VMCOREINFO_SYMBOL(init_uts_ns);
1386 VMCOREINFO_SYMBOL(node_online_map);
1388 VMCOREINFO_SYMBOL(swapper_pg_dir);
1390 VMCOREINFO_SYMBOL(_stext);
1391 VMCOREINFO_SYMBOL(vmap_area_list);
1393 #ifndef CONFIG_NEED_MULTIPLE_NODES
1394 VMCOREINFO_SYMBOL(mem_map);
1395 VMCOREINFO_SYMBOL(contig_page_data);
1397 #ifdef CONFIG_SPARSEMEM
1398 VMCOREINFO_SYMBOL(mem_section);
1399 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1400 VMCOREINFO_STRUCT_SIZE(mem_section);
1401 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1403 VMCOREINFO_STRUCT_SIZE(page);
1404 VMCOREINFO_STRUCT_SIZE(pglist_data);
1405 VMCOREINFO_STRUCT_SIZE(zone);
1406 VMCOREINFO_STRUCT_SIZE(free_area);
1407 VMCOREINFO_STRUCT_SIZE(list_head);
1408 VMCOREINFO_SIZE(nodemask_t);
1409 VMCOREINFO_OFFSET(page, flags);
1410 VMCOREINFO_OFFSET(page, _refcount);
1411 VMCOREINFO_OFFSET(page, mapping);
1412 VMCOREINFO_OFFSET(page, lru);
1413 VMCOREINFO_OFFSET(page, _mapcount);
1414 VMCOREINFO_OFFSET(page, private);
1415 VMCOREINFO_OFFSET(page, compound_dtor);
1416 VMCOREINFO_OFFSET(page, compound_order);
1417 VMCOREINFO_OFFSET(page, compound_head);
1418 VMCOREINFO_OFFSET(pglist_data, node_zones);
1419 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1420 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1421 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1423 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1424 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1425 VMCOREINFO_OFFSET(pglist_data, node_id);
1426 VMCOREINFO_OFFSET(zone, free_area);
1427 VMCOREINFO_OFFSET(zone, vm_stat);
1428 VMCOREINFO_OFFSET(zone, spanned_pages);
1429 VMCOREINFO_OFFSET(free_area, free_list);
1430 VMCOREINFO_OFFSET(list_head, next);
1431 VMCOREINFO_OFFSET(list_head, prev);
1432 VMCOREINFO_OFFSET(vmap_area, va_start);
1433 VMCOREINFO_OFFSET(vmap_area, list);
1434 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1435 log_buf_kexec_setup();
1436 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1437 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1438 VMCOREINFO_NUMBER(PG_lru);
1439 VMCOREINFO_NUMBER(PG_private);
1440 VMCOREINFO_NUMBER(PG_swapcache);
1441 VMCOREINFO_NUMBER(PG_slab);
1442 #ifdef CONFIG_MEMORY_FAILURE
1443 VMCOREINFO_NUMBER(PG_hwpoison);
1445 VMCOREINFO_NUMBER(PG_head_mask);
1446 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1448 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
1450 #ifdef CONFIG_HUGETLB_PAGE
1451 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
1454 arch_crash_save_vmcoreinfo();
1455 update_vmcoreinfo_note();
1460 subsys_initcall(crash_save_vmcoreinfo_init);
1463 * Move into place and start executing a preloaded standalone
1464 * executable. If nothing was preloaded return an error.
1466 int kernel_kexec(void)
1470 if (!mutex_trylock(&kexec_mutex))
1477 #ifdef CONFIG_KEXEC_JUMP
1478 if (kexec_image->preserve_context) {
1479 lock_system_sleep();
1480 pm_prepare_console();
1481 error = freeze_processes();
1484 goto Restore_console;
1487 error = dpm_suspend_start(PMSG_FREEZE);
1489 goto Resume_console;
1490 /* At this point, dpm_suspend_start() has been called,
1491 * but *not* dpm_suspend_end(). We *must* call
1492 * dpm_suspend_end() now. Otherwise, drivers for
1493 * some devices (e.g. interrupt controllers) become
1494 * desynchronized with the actual state of the
1495 * hardware at resume time, and evil weirdness ensues.
1497 error = dpm_suspend_end(PMSG_FREEZE);
1499 goto Resume_devices;
1500 error = disable_nonboot_cpus();
1503 local_irq_disable();
1504 error = syscore_suspend();
1510 kexec_in_progress = true;
1511 kernel_restart_prepare(NULL);
1512 migrate_to_reboot_cpu();
1515 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1516 * no further code needs to use CPU hotplug (which is true in
1517 * the reboot case). However, the kexec path depends on using
1518 * CPU hotplug again; so re-enable it here.
1520 cpu_hotplug_enable();
1521 pr_emerg("Starting new kernel\n");
1525 machine_kexec(kexec_image);
1527 #ifdef CONFIG_KEXEC_JUMP
1528 if (kexec_image->preserve_context) {
1533 enable_nonboot_cpus();
1534 dpm_resume_start(PMSG_RESTORE);
1536 dpm_resume_end(PMSG_RESTORE);
1541 pm_restore_console();
1542 unlock_system_sleep();
1547 mutex_unlock(&kexec_mutex);
1552 * Protection mechanism for crashkernel reserved memory after
1553 * the kdump kernel is loaded.
1555 * Provide an empty default implementation here -- architecture
1556 * code may override this
1558 void __weak arch_kexec_protect_crashkres(void)
1561 void __weak arch_kexec_unprotect_crashkres(void)