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;
171 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
172 return -EADDRNOTAVAIL;
173 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
174 return -EADDRNOTAVAIL;
177 /* Verify our destination addresses do not overlap.
178 * If we alloed overlapping destination addresses
179 * through very weird things can happen with no
180 * easy explanation as one segment stops on another.
182 for (i = 0; i < nr_segments; i++) {
183 unsigned long mstart, mend;
186 mstart = image->segment[i].mem;
187 mend = mstart + image->segment[i].memsz;
188 for (j = 0; j < i; j++) {
189 unsigned long pstart, pend;
191 pstart = image->segment[j].mem;
192 pend = pstart + image->segment[j].memsz;
193 /* Do the segments overlap ? */
194 if ((mend > pstart) && (mstart < pend))
199 /* Ensure our buffer sizes are strictly less than
200 * our memory sizes. This should always be the case,
201 * and it is easier to check up front than to be surprised
204 for (i = 0; i < nr_segments; i++) {
205 if (image->segment[i].bufsz > image->segment[i].memsz)
210 * Verify we have good destination addresses. Normally
211 * the caller is responsible for making certain we don't
212 * attempt to load the new image into invalid or reserved
213 * areas of RAM. But crash kernels are preloaded into a
214 * reserved area of ram. We must ensure the addresses
215 * are in the reserved area otherwise preloading the
216 * kernel could corrupt things.
219 if (image->type == KEXEC_TYPE_CRASH) {
220 for (i = 0; i < nr_segments; i++) {
221 unsigned long mstart, mend;
223 mstart = image->segment[i].mem;
224 mend = mstart + image->segment[i].memsz - 1;
225 /* Ensure we are within the crash kernel limits */
226 if ((mstart < crashk_res.start) ||
227 (mend > crashk_res.end))
228 return -EADDRNOTAVAIL;
235 struct kimage *do_kimage_alloc_init(void)
237 struct kimage *image;
239 /* Allocate a controlling structure */
240 image = kzalloc(sizeof(*image), GFP_KERNEL);
245 image->entry = &image->head;
246 image->last_entry = &image->head;
247 image->control_page = ~0; /* By default this does not apply */
248 image->type = KEXEC_TYPE_DEFAULT;
250 /* Initialize the list of control pages */
251 INIT_LIST_HEAD(&image->control_pages);
253 /* Initialize the list of destination pages */
254 INIT_LIST_HEAD(&image->dest_pages);
256 /* Initialize the list of unusable pages */
257 INIT_LIST_HEAD(&image->unusable_pages);
262 int kimage_is_destination_range(struct kimage *image,
268 for (i = 0; i < image->nr_segments; i++) {
269 unsigned long mstart, mend;
271 mstart = image->segment[i].mem;
272 mend = mstart + image->segment[i].memsz;
273 if ((end > mstart) && (start < mend))
280 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
284 pages = alloc_pages(gfp_mask, order);
286 unsigned int count, i;
288 pages->mapping = NULL;
289 set_page_private(pages, order);
291 for (i = 0; i < count; i++)
292 SetPageReserved(pages + i);
298 static void kimage_free_pages(struct page *page)
300 unsigned int order, count, i;
302 order = page_private(page);
304 for (i = 0; i < count; i++)
305 ClearPageReserved(page + i);
306 __free_pages(page, order);
309 void kimage_free_page_list(struct list_head *list)
311 struct page *page, *next;
313 list_for_each_entry_safe(page, next, list, lru) {
314 list_del(&page->lru);
315 kimage_free_pages(page);
319 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
322 /* Control pages are special, they are the intermediaries
323 * that are needed while we copy the rest of the pages
324 * to their final resting place. As such they must
325 * not conflict with either the destination addresses
326 * or memory the kernel is already using.
328 * The only case where we really need more than one of
329 * these are for architectures where we cannot disable
330 * the MMU and must instead generate an identity mapped
331 * page table for all of the memory.
333 * At worst this runs in O(N) of the image size.
335 struct list_head extra_pages;
340 INIT_LIST_HEAD(&extra_pages);
342 /* Loop while I can allocate a page and the page allocated
343 * is a destination page.
346 unsigned long pfn, epfn, addr, eaddr;
348 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
351 pfn = page_to_pfn(pages);
353 addr = pfn << PAGE_SHIFT;
354 eaddr = epfn << PAGE_SHIFT;
355 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
356 kimage_is_destination_range(image, addr, eaddr)) {
357 list_add(&pages->lru, &extra_pages);
363 /* Remember the allocated page... */
364 list_add(&pages->lru, &image->control_pages);
366 /* Because the page is already in it's destination
367 * location we will never allocate another page at
368 * that address. Therefore kimage_alloc_pages
369 * will not return it (again) and we don't need
370 * to give it an entry in image->segment[].
373 /* Deal with the destination pages I have inadvertently allocated.
375 * Ideally I would convert multi-page allocations into single
376 * page allocations, and add everything to image->dest_pages.
378 * For now it is simpler to just free the pages.
380 kimage_free_page_list(&extra_pages);
385 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
388 /* Control pages are special, they are the intermediaries
389 * that are needed while we copy the rest of the pages
390 * to their final resting place. As such they must
391 * not conflict with either the destination addresses
392 * or memory the kernel is already using.
394 * Control pages are also the only pags we must allocate
395 * when loading a crash kernel. All of the other pages
396 * are specified by the segments and we just memcpy
397 * into them directly.
399 * The only case where we really need more than one of
400 * these are for architectures where we cannot disable
401 * the MMU and must instead generate an identity mapped
402 * page table for all of the memory.
404 * Given the low demand this implements a very simple
405 * allocator that finds the first hole of the appropriate
406 * size in the reserved memory region, and allocates all
407 * of the memory up to and including the hole.
409 unsigned long hole_start, hole_end, size;
413 size = (1 << order) << PAGE_SHIFT;
414 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
415 hole_end = hole_start + size - 1;
416 while (hole_end <= crashk_res.end) {
419 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
421 /* See if I overlap any of the segments */
422 for (i = 0; i < image->nr_segments; i++) {
423 unsigned long mstart, mend;
425 mstart = image->segment[i].mem;
426 mend = mstart + image->segment[i].memsz - 1;
427 if ((hole_end >= mstart) && (hole_start <= mend)) {
428 /* Advance the hole to the end of the segment */
429 hole_start = (mend + (size - 1)) & ~(size - 1);
430 hole_end = hole_start + size - 1;
434 /* If I don't overlap any segments I have found my hole! */
435 if (i == image->nr_segments) {
436 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
437 image->control_page = hole_end;
446 struct page *kimage_alloc_control_pages(struct kimage *image,
449 struct page *pages = NULL;
451 switch (image->type) {
452 case KEXEC_TYPE_DEFAULT:
453 pages = kimage_alloc_normal_control_pages(image, order);
455 case KEXEC_TYPE_CRASH:
456 pages = kimage_alloc_crash_control_pages(image, order);
463 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
465 if (*image->entry != 0)
468 if (image->entry == image->last_entry) {
469 kimage_entry_t *ind_page;
472 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
476 ind_page = page_address(page);
477 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
478 image->entry = ind_page;
479 image->last_entry = ind_page +
480 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
482 *image->entry = entry;
489 static int kimage_set_destination(struct kimage *image,
490 unsigned long destination)
494 destination &= PAGE_MASK;
495 result = kimage_add_entry(image, destination | IND_DESTINATION);
501 static int kimage_add_page(struct kimage *image, unsigned long page)
506 result = kimage_add_entry(image, page | IND_SOURCE);
512 static void kimage_free_extra_pages(struct kimage *image)
514 /* Walk through and free any extra destination pages I may have */
515 kimage_free_page_list(&image->dest_pages);
517 /* Walk through and free any unusable pages I have cached */
518 kimage_free_page_list(&image->unusable_pages);
521 void kimage_terminate(struct kimage *image)
523 if (*image->entry != 0)
526 *image->entry = IND_DONE;
529 #define for_each_kimage_entry(image, ptr, entry) \
530 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
531 ptr = (entry & IND_INDIRECTION) ? \
532 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
534 static void kimage_free_entry(kimage_entry_t entry)
538 page = pfn_to_page(entry >> PAGE_SHIFT);
539 kimage_free_pages(page);
542 void kimage_free(struct kimage *image)
544 kimage_entry_t *ptr, entry;
545 kimage_entry_t ind = 0;
550 kimage_free_extra_pages(image);
551 for_each_kimage_entry(image, ptr, entry) {
552 if (entry & IND_INDIRECTION) {
553 /* Free the previous indirection page */
554 if (ind & IND_INDIRECTION)
555 kimage_free_entry(ind);
556 /* Save this indirection page until we are
560 } else if (entry & IND_SOURCE)
561 kimage_free_entry(entry);
563 /* Free the final indirection page */
564 if (ind & IND_INDIRECTION)
565 kimage_free_entry(ind);
567 /* Handle any machine specific cleanup */
568 machine_kexec_cleanup(image);
570 /* Free the kexec control pages... */
571 kimage_free_page_list(&image->control_pages);
574 * Free up any temporary buffers allocated. This might hit if
575 * error occurred much later after buffer allocation.
577 if (image->file_mode)
578 kimage_file_post_load_cleanup(image);
583 static kimage_entry_t *kimage_dst_used(struct kimage *image,
586 kimage_entry_t *ptr, entry;
587 unsigned long destination = 0;
589 for_each_kimage_entry(image, ptr, entry) {
590 if (entry & IND_DESTINATION)
591 destination = entry & PAGE_MASK;
592 else if (entry & IND_SOURCE) {
593 if (page == destination)
595 destination += PAGE_SIZE;
602 static struct page *kimage_alloc_page(struct kimage *image,
604 unsigned long destination)
607 * Here we implement safeguards to ensure that a source page
608 * is not copied to its destination page before the data on
609 * the destination page is no longer useful.
611 * To do this we maintain the invariant that a source page is
612 * either its own destination page, or it is not a
613 * destination page at all.
615 * That is slightly stronger than required, but the proof
616 * that no problems will not occur is trivial, and the
617 * implementation is simply to verify.
619 * When allocating all pages normally this algorithm will run
620 * in O(N) time, but in the worst case it will run in O(N^2)
621 * time. If the runtime is a problem the data structures can
628 * Walk through the list of destination pages, and see if I
631 list_for_each_entry(page, &image->dest_pages, lru) {
632 addr = page_to_pfn(page) << PAGE_SHIFT;
633 if (addr == destination) {
634 list_del(&page->lru);
642 /* Allocate a page, if we run out of memory give up */
643 page = kimage_alloc_pages(gfp_mask, 0);
646 /* If the page cannot be used file it away */
647 if (page_to_pfn(page) >
648 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
649 list_add(&page->lru, &image->unusable_pages);
652 addr = page_to_pfn(page) << PAGE_SHIFT;
654 /* If it is the destination page we want use it */
655 if (addr == destination)
658 /* If the page is not a destination page use it */
659 if (!kimage_is_destination_range(image, addr,
664 * I know that the page is someones destination page.
665 * See if there is already a source page for this
666 * destination page. And if so swap the source pages.
668 old = kimage_dst_used(image, addr);
671 unsigned long old_addr;
672 struct page *old_page;
674 old_addr = *old & PAGE_MASK;
675 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
676 copy_highpage(page, old_page);
677 *old = addr | (*old & ~PAGE_MASK);
679 /* The old page I have found cannot be a
680 * destination page, so return it if it's
681 * gfp_flags honor the ones passed in.
683 if (!(gfp_mask & __GFP_HIGHMEM) &&
684 PageHighMem(old_page)) {
685 kimage_free_pages(old_page);
692 /* Place the page on the destination list, to be used later */
693 list_add(&page->lru, &image->dest_pages);
699 static int kimage_load_normal_segment(struct kimage *image,
700 struct kexec_segment *segment)
703 size_t ubytes, mbytes;
705 unsigned char __user *buf = NULL;
706 unsigned char *kbuf = NULL;
709 if (image->file_mode)
710 kbuf = segment->kbuf;
713 ubytes = segment->bufsz;
714 mbytes = segment->memsz;
715 maddr = segment->mem;
717 result = kimage_set_destination(image, maddr);
724 size_t uchunk, mchunk;
726 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
731 result = kimage_add_page(image, page_to_pfn(page)
737 /* Start with a clear page */
739 ptr += maddr & ~PAGE_MASK;
740 mchunk = min_t(size_t, mbytes,
741 PAGE_SIZE - (maddr & ~PAGE_MASK));
742 uchunk = min(ubytes, mchunk);
744 /* For file based kexec, source pages are in kernel memory */
745 if (image->file_mode)
746 memcpy(ptr, kbuf, uchunk);
748 result = copy_from_user(ptr, buf, uchunk);
756 if (image->file_mode)
766 static int kimage_load_crash_segment(struct kimage *image,
767 struct kexec_segment *segment)
769 /* For crash dumps kernels we simply copy the data from
770 * user space to it's destination.
771 * We do things a page at a time for the sake of kmap.
774 size_t ubytes, mbytes;
776 unsigned char __user *buf = NULL;
777 unsigned char *kbuf = NULL;
780 if (image->file_mode)
781 kbuf = segment->kbuf;
784 ubytes = segment->bufsz;
785 mbytes = segment->memsz;
786 maddr = segment->mem;
790 size_t uchunk, mchunk;
792 page = pfn_to_page(maddr >> PAGE_SHIFT);
798 ptr += maddr & ~PAGE_MASK;
799 mchunk = min_t(size_t, mbytes,
800 PAGE_SIZE - (maddr & ~PAGE_MASK));
801 uchunk = min(ubytes, mchunk);
802 if (mchunk > uchunk) {
803 /* Zero the trailing part of the page */
804 memset(ptr + uchunk, 0, mchunk - uchunk);
807 /* For file based kexec, source pages are in kernel memory */
808 if (image->file_mode)
809 memcpy(ptr, kbuf, uchunk);
811 result = copy_from_user(ptr, buf, uchunk);
812 kexec_flush_icache_page(page);
820 if (image->file_mode)
830 int kimage_load_segment(struct kimage *image,
831 struct kexec_segment *segment)
833 int result = -ENOMEM;
835 switch (image->type) {
836 case KEXEC_TYPE_DEFAULT:
837 result = kimage_load_normal_segment(image, segment);
839 case KEXEC_TYPE_CRASH:
840 result = kimage_load_crash_segment(image, segment);
847 struct kimage *kexec_image;
848 struct kimage *kexec_crash_image;
849 int kexec_load_disabled;
852 * No panic_cpu check version of crash_kexec(). This function is called
853 * only when panic_cpu holds the current CPU number; this is the only CPU
854 * which processes crash_kexec routines.
856 void __crash_kexec(struct pt_regs *regs)
858 /* Take the kexec_mutex here to prevent sys_kexec_load
859 * running on one cpu from replacing the crash kernel
860 * we are using after a panic on a different cpu.
862 * If the crash kernel was not located in a fixed area
863 * of memory the xchg(&kexec_crash_image) would be
864 * sufficient. But since I reuse the memory...
866 if (mutex_trylock(&kexec_mutex)) {
867 if (kexec_crash_image) {
868 struct pt_regs fixed_regs;
870 crash_setup_regs(&fixed_regs, regs);
871 crash_save_vmcoreinfo();
872 machine_crash_shutdown(&fixed_regs);
873 machine_kexec(kexec_crash_image);
875 mutex_unlock(&kexec_mutex);
879 void crash_kexec(struct pt_regs *regs)
881 int old_cpu, this_cpu;
884 * Only one CPU is allowed to execute the crash_kexec() code as with
885 * panic(). Otherwise parallel calls of panic() and crash_kexec()
886 * may stop each other. To exclude them, we use panic_cpu here too.
888 this_cpu = raw_smp_processor_id();
889 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
890 if (old_cpu == PANIC_CPU_INVALID) {
891 /* This is the 1st CPU which comes here, so go ahead. */
892 printk_nmi_flush_on_panic();
896 * Reset panic_cpu to allow another panic()/crash_kexec()
899 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
903 size_t crash_get_memory_size(void)
907 mutex_lock(&kexec_mutex);
908 if (crashk_res.end != crashk_res.start)
909 size = resource_size(&crashk_res);
910 mutex_unlock(&kexec_mutex);
914 void __weak crash_free_reserved_phys_range(unsigned long begin,
919 for (addr = begin; addr < end; addr += PAGE_SIZE)
920 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
923 int crash_shrink_memory(unsigned long new_size)
926 unsigned long start, end;
927 unsigned long old_size;
928 struct resource *ram_res;
930 mutex_lock(&kexec_mutex);
932 if (kexec_crash_image) {
936 start = crashk_res.start;
937 end = crashk_res.end;
938 old_size = (end == 0) ? 0 : end - start + 1;
939 if (new_size >= old_size) {
940 ret = (new_size == old_size) ? 0 : -EINVAL;
944 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
950 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
951 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
953 crash_free_reserved_phys_range(end, crashk_res.end);
955 if ((start == end) && (crashk_res.parent != NULL))
956 release_resource(&crashk_res);
958 ram_res->start = end;
959 ram_res->end = crashk_res.end;
960 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
961 ram_res->name = "System RAM";
963 crashk_res.end = end - 1;
965 insert_resource(&iomem_resource, ram_res);
968 mutex_unlock(&kexec_mutex);
972 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
975 struct elf_note note;
977 note.n_namesz = strlen(name) + 1;
978 note.n_descsz = data_len;
980 memcpy(buf, ¬e, sizeof(note));
981 buf += (sizeof(note) + 3)/4;
982 memcpy(buf, name, note.n_namesz);
983 buf += (note.n_namesz + 3)/4;
984 memcpy(buf, data, note.n_descsz);
985 buf += (note.n_descsz + 3)/4;
990 static void final_note(u32 *buf)
992 struct elf_note note;
997 memcpy(buf, ¬e, sizeof(note));
1000 void crash_save_cpu(struct pt_regs *regs, int cpu)
1002 struct elf_prstatus prstatus;
1005 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1008 /* Using ELF notes here is opportunistic.
1009 * I need a well defined structure format
1010 * for the data I pass, and I need tags
1011 * on the data to indicate what information I have
1012 * squirrelled away. ELF notes happen to provide
1013 * all of that, so there is no need to invent something new.
1015 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1018 memset(&prstatus, 0, sizeof(prstatus));
1019 prstatus.pr_pid = current->pid;
1020 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1021 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1022 &prstatus, sizeof(prstatus));
1026 static int __init crash_notes_memory_init(void)
1028 /* Allocate memory for saving cpu registers. */
1032 * crash_notes could be allocated across 2 vmalloc pages when percpu
1033 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1034 * pages are also on 2 continuous physical pages. In this case the
1035 * 2nd part of crash_notes in 2nd page could be lost since only the
1036 * starting address and size of crash_notes are exported through sysfs.
1037 * Here round up the size of crash_notes to the nearest power of two
1038 * and pass it to __alloc_percpu as align value. This can make sure
1039 * crash_notes is allocated inside one physical page.
1041 size = sizeof(note_buf_t);
1042 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1045 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1046 * definitely will be in 2 pages with that.
1048 BUILD_BUG_ON(size > PAGE_SIZE);
1050 crash_notes = __alloc_percpu(size, align);
1052 pr_warn("Memory allocation for saving cpu register states failed\n");
1057 subsys_initcall(crash_notes_memory_init);
1061 * parsing the "crashkernel" commandline
1063 * this code is intended to be called from architecture specific code
1068 * This function parses command lines in the format
1070 * crashkernel=ramsize-range:size[,...][@offset]
1072 * The function returns 0 on success and -EINVAL on failure.
1074 static int __init parse_crashkernel_mem(char *cmdline,
1075 unsigned long long system_ram,
1076 unsigned long long *crash_size,
1077 unsigned long long *crash_base)
1079 char *cur = cmdline, *tmp;
1081 /* for each entry of the comma-separated list */
1083 unsigned long long start, end = ULLONG_MAX, size;
1085 /* get the start of the range */
1086 start = memparse(cur, &tmp);
1088 pr_warn("crashkernel: Memory value expected\n");
1093 pr_warn("crashkernel: '-' expected\n");
1098 /* if no ':' is here, than we read the end */
1100 end = memparse(cur, &tmp);
1102 pr_warn("crashkernel: Memory value expected\n");
1107 pr_warn("crashkernel: end <= start\n");
1113 pr_warn("crashkernel: ':' expected\n");
1118 size = memparse(cur, &tmp);
1120 pr_warn("Memory value expected\n");
1124 if (size >= system_ram) {
1125 pr_warn("crashkernel: invalid size\n");
1130 if (system_ram >= start && system_ram < end) {
1134 } while (*cur++ == ',');
1136 if (*crash_size > 0) {
1137 while (*cur && *cur != ' ' && *cur != '@')
1141 *crash_base = memparse(cur, &tmp);
1143 pr_warn("Memory value expected after '@'\n");
1153 * That function parses "simple" (old) crashkernel command lines like
1155 * crashkernel=size[@offset]
1157 * It returns 0 on success and -EINVAL on failure.
1159 static int __init parse_crashkernel_simple(char *cmdline,
1160 unsigned long long *crash_size,
1161 unsigned long long *crash_base)
1163 char *cur = cmdline;
1165 *crash_size = memparse(cmdline, &cur);
1166 if (cmdline == cur) {
1167 pr_warn("crashkernel: memory value expected\n");
1172 *crash_base = memparse(cur+1, &cur);
1173 else if (*cur != ' ' && *cur != '\0') {
1174 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1181 #define SUFFIX_HIGH 0
1182 #define SUFFIX_LOW 1
1183 #define SUFFIX_NULL 2
1184 static __initdata char *suffix_tbl[] = {
1185 [SUFFIX_HIGH] = ",high",
1186 [SUFFIX_LOW] = ",low",
1187 [SUFFIX_NULL] = NULL,
1191 * That function parses "suffix" crashkernel command lines like
1193 * crashkernel=size,[high|low]
1195 * It returns 0 on success and -EINVAL on failure.
1197 static int __init parse_crashkernel_suffix(char *cmdline,
1198 unsigned long long *crash_size,
1201 char *cur = cmdline;
1203 *crash_size = memparse(cmdline, &cur);
1204 if (cmdline == cur) {
1205 pr_warn("crashkernel: memory value expected\n");
1209 /* check with suffix */
1210 if (strncmp(cur, suffix, strlen(suffix))) {
1211 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1214 cur += strlen(suffix);
1215 if (*cur != ' ' && *cur != '\0') {
1216 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1223 static __init char *get_last_crashkernel(char *cmdline,
1227 char *p = cmdline, *ck_cmdline = NULL;
1229 /* find crashkernel and use the last one if there are more */
1230 p = strstr(p, name);
1232 char *end_p = strchr(p, ' ');
1236 end_p = p + strlen(p);
1241 /* skip the one with any known suffix */
1242 for (i = 0; suffix_tbl[i]; i++) {
1243 q = end_p - strlen(suffix_tbl[i]);
1244 if (!strncmp(q, suffix_tbl[i],
1245 strlen(suffix_tbl[i])))
1250 q = end_p - strlen(suffix);
1251 if (!strncmp(q, suffix, strlen(suffix)))
1255 p = strstr(p+1, name);
1264 static int __init __parse_crashkernel(char *cmdline,
1265 unsigned long long system_ram,
1266 unsigned long long *crash_size,
1267 unsigned long long *crash_base,
1271 char *first_colon, *first_space;
1274 BUG_ON(!crash_size || !crash_base);
1278 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1283 ck_cmdline += strlen(name);
1286 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1289 * if the commandline contains a ':', then that's the extended
1290 * syntax -- if not, it must be the classic syntax
1292 first_colon = strchr(ck_cmdline, ':');
1293 first_space = strchr(ck_cmdline, ' ');
1294 if (first_colon && (!first_space || first_colon < first_space))
1295 return parse_crashkernel_mem(ck_cmdline, system_ram,
1296 crash_size, crash_base);
1298 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1302 * That function is the entry point for command line parsing and should be
1303 * called from the arch-specific code.
1305 int __init parse_crashkernel(char *cmdline,
1306 unsigned long long system_ram,
1307 unsigned long long *crash_size,
1308 unsigned long long *crash_base)
1310 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1311 "crashkernel=", NULL);
1314 int __init parse_crashkernel_high(char *cmdline,
1315 unsigned long long system_ram,
1316 unsigned long long *crash_size,
1317 unsigned long long *crash_base)
1319 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1320 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1323 int __init parse_crashkernel_low(char *cmdline,
1324 unsigned long long system_ram,
1325 unsigned long long *crash_size,
1326 unsigned long long *crash_base)
1328 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1329 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1332 static void update_vmcoreinfo_note(void)
1334 u32 *buf = vmcoreinfo_note;
1336 if (!vmcoreinfo_size)
1338 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1343 void crash_save_vmcoreinfo(void)
1345 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1346 update_vmcoreinfo_note();
1349 void vmcoreinfo_append_str(const char *fmt, ...)
1355 va_start(args, fmt);
1356 r = vscnprintf(buf, sizeof(buf), fmt, args);
1359 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1361 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1363 vmcoreinfo_size += r;
1367 * provide an empty default implementation here -- architecture
1368 * code may override this
1370 void __weak arch_crash_save_vmcoreinfo(void)
1373 unsigned long __weak paddr_vmcoreinfo_note(void)
1375 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1378 static int __init crash_save_vmcoreinfo_init(void)
1380 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1381 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1383 VMCOREINFO_SYMBOL(init_uts_ns);
1384 VMCOREINFO_SYMBOL(node_online_map);
1386 VMCOREINFO_SYMBOL(swapper_pg_dir);
1388 VMCOREINFO_SYMBOL(_stext);
1389 VMCOREINFO_SYMBOL(vmap_area_list);
1391 #ifndef CONFIG_NEED_MULTIPLE_NODES
1392 VMCOREINFO_SYMBOL(mem_map);
1393 VMCOREINFO_SYMBOL(contig_page_data);
1395 #ifdef CONFIG_SPARSEMEM
1396 VMCOREINFO_SYMBOL(mem_section);
1397 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1398 VMCOREINFO_STRUCT_SIZE(mem_section);
1399 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1401 VMCOREINFO_STRUCT_SIZE(page);
1402 VMCOREINFO_STRUCT_SIZE(pglist_data);
1403 VMCOREINFO_STRUCT_SIZE(zone);
1404 VMCOREINFO_STRUCT_SIZE(free_area);
1405 VMCOREINFO_STRUCT_SIZE(list_head);
1406 VMCOREINFO_SIZE(nodemask_t);
1407 VMCOREINFO_OFFSET(page, flags);
1408 VMCOREINFO_OFFSET(page, _refcount);
1409 VMCOREINFO_OFFSET(page, mapping);
1410 VMCOREINFO_OFFSET(page, lru);
1411 VMCOREINFO_OFFSET(page, _mapcount);
1412 VMCOREINFO_OFFSET(page, private);
1413 VMCOREINFO_OFFSET(page, compound_dtor);
1414 VMCOREINFO_OFFSET(page, compound_order);
1415 VMCOREINFO_OFFSET(page, compound_head);
1416 VMCOREINFO_OFFSET(pglist_data, node_zones);
1417 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1418 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1419 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1421 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1422 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1423 VMCOREINFO_OFFSET(pglist_data, node_id);
1424 VMCOREINFO_OFFSET(zone, free_area);
1425 VMCOREINFO_OFFSET(zone, vm_stat);
1426 VMCOREINFO_OFFSET(zone, spanned_pages);
1427 VMCOREINFO_OFFSET(free_area, free_list);
1428 VMCOREINFO_OFFSET(list_head, next);
1429 VMCOREINFO_OFFSET(list_head, prev);
1430 VMCOREINFO_OFFSET(vmap_area, va_start);
1431 VMCOREINFO_OFFSET(vmap_area, list);
1432 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1433 log_buf_kexec_setup();
1434 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1435 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1436 VMCOREINFO_NUMBER(PG_lru);
1437 VMCOREINFO_NUMBER(PG_private);
1438 VMCOREINFO_NUMBER(PG_swapcache);
1439 VMCOREINFO_NUMBER(PG_slab);
1440 #ifdef CONFIG_MEMORY_FAILURE
1441 VMCOREINFO_NUMBER(PG_hwpoison);
1443 VMCOREINFO_NUMBER(PG_head_mask);
1444 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1446 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
1448 #ifdef CONFIG_HUGETLB_PAGE
1449 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
1452 arch_crash_save_vmcoreinfo();
1453 update_vmcoreinfo_note();
1458 subsys_initcall(crash_save_vmcoreinfo_init);
1461 * Move into place and start executing a preloaded standalone
1462 * executable. If nothing was preloaded return an error.
1464 int kernel_kexec(void)
1468 if (!mutex_trylock(&kexec_mutex))
1475 #ifdef CONFIG_KEXEC_JUMP
1476 if (kexec_image->preserve_context) {
1477 lock_system_sleep();
1478 pm_prepare_console();
1479 error = freeze_processes();
1482 goto Restore_console;
1485 error = dpm_suspend_start(PMSG_FREEZE);
1487 goto Resume_console;
1488 /* At this point, dpm_suspend_start() has been called,
1489 * but *not* dpm_suspend_end(). We *must* call
1490 * dpm_suspend_end() now. Otherwise, drivers for
1491 * some devices (e.g. interrupt controllers) become
1492 * desynchronized with the actual state of the
1493 * hardware at resume time, and evil weirdness ensues.
1495 error = dpm_suspend_end(PMSG_FREEZE);
1497 goto Resume_devices;
1498 error = disable_nonboot_cpus();
1501 local_irq_disable();
1502 error = syscore_suspend();
1508 kexec_in_progress = true;
1509 kernel_restart_prepare(NULL);
1510 migrate_to_reboot_cpu();
1513 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1514 * no further code needs to use CPU hotplug (which is true in
1515 * the reboot case). However, the kexec path depends on using
1516 * CPU hotplug again; so re-enable it here.
1518 cpu_hotplug_enable();
1519 pr_emerg("Starting new kernel\n");
1523 machine_kexec(kexec_image);
1525 #ifdef CONFIG_KEXEC_JUMP
1526 if (kexec_image->preserve_context) {
1531 enable_nonboot_cpus();
1532 dpm_resume_start(PMSG_RESTORE);
1534 dpm_resume_end(PMSG_RESTORE);
1539 pm_restore_console();
1540 unlock_system_sleep();
1545 mutex_unlock(&kexec_mutex);
1550 * Protection mechanism for crashkernel reserved memory after
1551 * the kdump kernel is loaded.
1553 * Provide an empty default implementation here -- architecture
1554 * code may override this
1556 void __weak arch_kexec_protect_crashkres(void)
1559 void __weak arch_kexec_unprotect_crashkres(void)