Add linux-next specific files for 20110831
[linux-2.6/next.git] / kernel / kexec.c
blob7204fb982ed599e9fa20d1e7572fe3808e091d8e
1 /*
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.
7 */
9 #include <linux/capability.h>
10 #include <linux/mm.h>
11 #include <linux/file.h>
12 #include <linux/slab.h>
13 #include <linux/fs.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>
30 #include <linux/pm.h>
31 #include <linux/cpu.h>
32 #include <linux/console.h>
33 #include <linux/vmalloc.h>
34 #include <linux/swap.h>
35 #include <linux/kmsg_dump.h>
36 #include <linux/syscore_ops.h>
38 #include <asm/page.h>
39 #include <asm/uaccess.h>
40 #include <asm/io.h>
41 #include <asm/system.h>
42 #include <asm/sections.h>
44 /* Per cpu memory for storing cpu states in case of system crash. */
45 note_buf_t __percpu *crash_notes;
47 /* vmcoreinfo stuff */
48 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
49 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
50 size_t vmcoreinfo_size;
51 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
53 /* Location of the reserved area for the crash kernel */
54 struct resource crashk_res = {
55 .name = "Crash kernel",
56 .start = 0,
57 .end = 0,
58 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
61 int kexec_should_crash(struct task_struct *p)
63 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
64 return 1;
65 return 0;
69 * When kexec transitions to the new kernel there is a one-to-one
70 * mapping between physical and virtual addresses. On processors
71 * where you can disable the MMU this is trivial, and easy. For
72 * others it is still a simple predictable page table to setup.
74 * In that environment kexec copies the new kernel to its final
75 * resting place. This means I can only support memory whose
76 * physical address can fit in an unsigned long. In particular
77 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
78 * If the assembly stub has more restrictive requirements
79 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
80 * defined more restrictively in <asm/kexec.h>.
82 * The code for the transition from the current kernel to the
83 * the new kernel is placed in the control_code_buffer, whose size
84 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
85 * page of memory is necessary, but some architectures require more.
86 * Because this memory must be identity mapped in the transition from
87 * virtual to physical addresses it must live in the range
88 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
89 * modifiable.
91 * The assembly stub in the control code buffer is passed a linked list
92 * of descriptor pages detailing the source pages of the new kernel,
93 * and the destination addresses of those source pages. As this data
94 * structure is not used in the context of the current OS, it must
95 * be self-contained.
97 * The code has been made to work with highmem pages and will use a
98 * destination page in its final resting place (if it happens
99 * to allocate it). The end product of this is that most of the
100 * physical address space, and most of RAM can be used.
102 * Future directions include:
103 * - allocating a page table with the control code buffer identity
104 * mapped, to simplify machine_kexec and make kexec_on_panic more
105 * reliable.
109 * KIMAGE_NO_DEST is an impossible destination address..., for
110 * allocating pages whose destination address we do not care about.
112 #define KIMAGE_NO_DEST (-1UL)
114 static int kimage_is_destination_range(struct kimage *image,
115 unsigned long start, unsigned long end);
116 static struct page *kimage_alloc_page(struct kimage *image,
117 gfp_t gfp_mask,
118 unsigned long dest);
120 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
121 unsigned long nr_segments,
122 struct kexec_segment __user *segments)
124 size_t segment_bytes;
125 struct kimage *image;
126 unsigned long i;
127 int result;
129 /* Allocate a controlling structure */
130 result = -ENOMEM;
131 image = kzalloc(sizeof(*image), GFP_KERNEL);
132 if (!image)
133 goto out;
135 image->head = 0;
136 image->entry = &image->head;
137 image->last_entry = &image->head;
138 image->control_page = ~0; /* By default this does not apply */
139 image->start = entry;
140 image->type = KEXEC_TYPE_DEFAULT;
142 /* Initialize the list of control pages */
143 INIT_LIST_HEAD(&image->control_pages);
145 /* Initialize the list of destination pages */
146 INIT_LIST_HEAD(&image->dest_pages);
148 /* Initialize the list of unusable pages */
149 INIT_LIST_HEAD(&image->unuseable_pages);
151 /* Read in the segments */
152 image->nr_segments = nr_segments;
153 segment_bytes = nr_segments * sizeof(*segments);
154 result = copy_from_user(image->segment, segments, segment_bytes);
155 if (result) {
156 result = -EFAULT;
157 goto out;
161 * Verify we have good destination addresses. The caller is
162 * responsible for making certain we don't attempt to load
163 * the new image into invalid or reserved areas of RAM. This
164 * just verifies it is an address we can use.
166 * Since the kernel does everything in page size chunks ensure
167 * the destination addresses are page aligned. Too many
168 * special cases crop of when we don't do this. The most
169 * insidious is getting overlapping destination addresses
170 * simply because addresses are changed to page size
171 * granularity.
173 result = -EADDRNOTAVAIL;
174 for (i = 0; i < nr_segments; i++) {
175 unsigned long mstart, mend;
177 mstart = image->segment[i].mem;
178 mend = mstart + image->segment[i].memsz;
179 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
180 goto out;
181 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
182 goto out;
185 /* Verify our destination addresses do not overlap.
186 * If we alloed overlapping destination addresses
187 * through very weird things can happen with no
188 * easy explanation as one segment stops on another.
190 result = -EINVAL;
191 for (i = 0; i < nr_segments; i++) {
192 unsigned long mstart, mend;
193 unsigned long j;
195 mstart = image->segment[i].mem;
196 mend = mstart + image->segment[i].memsz;
197 for (j = 0; j < i; j++) {
198 unsigned long pstart, pend;
199 pstart = image->segment[j].mem;
200 pend = pstart + image->segment[j].memsz;
201 /* Do the segments overlap ? */
202 if ((mend > pstart) && (mstart < pend))
203 goto out;
207 /* Ensure our buffer sizes are strictly less than
208 * our memory sizes. This should always be the case,
209 * and it is easier to check up front than to be surprised
210 * later on.
212 result = -EINVAL;
213 for (i = 0; i < nr_segments; i++) {
214 if (image->segment[i].bufsz > image->segment[i].memsz)
215 goto out;
218 result = 0;
219 out:
220 if (result == 0)
221 *rimage = image;
222 else
223 kfree(image);
225 return result;
229 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
230 unsigned long nr_segments,
231 struct kexec_segment __user *segments)
233 int result;
234 struct kimage *image;
236 /* Allocate and initialize a controlling structure */
237 image = NULL;
238 result = do_kimage_alloc(&image, entry, nr_segments, segments);
239 if (result)
240 goto out;
242 *rimage = image;
245 * Find a location for the control code buffer, and add it
246 * the vector of segments so that it's pages will also be
247 * counted as destination pages.
249 result = -ENOMEM;
250 image->control_code_page = kimage_alloc_control_pages(image,
251 get_order(KEXEC_CONTROL_PAGE_SIZE));
252 if (!image->control_code_page) {
253 printk(KERN_ERR "Could not allocate control_code_buffer\n");
254 goto out;
257 image->swap_page = kimage_alloc_control_pages(image, 0);
258 if (!image->swap_page) {
259 printk(KERN_ERR "Could not allocate swap buffer\n");
260 goto out;
263 result = 0;
264 out:
265 if (result == 0)
266 *rimage = image;
267 else
268 kfree(image);
270 return result;
273 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
274 unsigned long nr_segments,
275 struct kexec_segment __user *segments)
277 int result;
278 struct kimage *image;
279 unsigned long i;
281 image = NULL;
282 /* Verify we have a valid entry point */
283 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
284 result = -EADDRNOTAVAIL;
285 goto out;
288 /* Allocate and initialize a controlling structure */
289 result = do_kimage_alloc(&image, entry, nr_segments, segments);
290 if (result)
291 goto out;
293 /* Enable the special crash kernel control page
294 * allocation policy.
296 image->control_page = crashk_res.start;
297 image->type = KEXEC_TYPE_CRASH;
300 * Verify we have good destination addresses. Normally
301 * the caller is responsible for making certain we don't
302 * attempt to load the new image into invalid or reserved
303 * areas of RAM. But crash kernels are preloaded into a
304 * reserved area of ram. We must ensure the addresses
305 * are in the reserved area otherwise preloading the
306 * kernel could corrupt things.
308 result = -EADDRNOTAVAIL;
309 for (i = 0; i < nr_segments; i++) {
310 unsigned long mstart, mend;
312 mstart = image->segment[i].mem;
313 mend = mstart + image->segment[i].memsz - 1;
314 /* Ensure we are within the crash kernel limits */
315 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
316 goto out;
320 * Find a location for the control code buffer, and add
321 * the vector of segments so that it's pages will also be
322 * counted as destination pages.
324 result = -ENOMEM;
325 image->control_code_page = kimage_alloc_control_pages(image,
326 get_order(KEXEC_CONTROL_PAGE_SIZE));
327 if (!image->control_code_page) {
328 printk(KERN_ERR "Could not allocate control_code_buffer\n");
329 goto out;
332 result = 0;
333 out:
334 if (result == 0)
335 *rimage = image;
336 else
337 kfree(image);
339 return result;
342 static int kimage_is_destination_range(struct kimage *image,
343 unsigned long start,
344 unsigned long end)
346 unsigned long i;
348 for (i = 0; i < image->nr_segments; i++) {
349 unsigned long mstart, mend;
351 mstart = image->segment[i].mem;
352 mend = mstart + image->segment[i].memsz;
353 if ((end > mstart) && (start < mend))
354 return 1;
357 return 0;
360 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
362 struct page *pages;
364 pages = alloc_pages(gfp_mask, order);
365 if (pages) {
366 unsigned int count, i;
367 pages->mapping = NULL;
368 set_page_private(pages, order);
369 count = 1 << order;
370 for (i = 0; i < count; i++)
371 SetPageReserved(pages + i);
374 return pages;
377 static void kimage_free_pages(struct page *page)
379 unsigned int order, count, i;
381 order = page_private(page);
382 count = 1 << order;
383 for (i = 0; i < count; i++)
384 ClearPageReserved(page + i);
385 __free_pages(page, order);
388 static void kimage_free_page_list(struct list_head *list)
390 struct list_head *pos, *next;
392 list_for_each_safe(pos, next, list) {
393 struct page *page;
395 page = list_entry(pos, struct page, lru);
396 list_del(&page->lru);
397 kimage_free_pages(page);
401 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
402 unsigned int order)
404 /* Control pages are special, they are the intermediaries
405 * that are needed while we copy the rest of the pages
406 * to their final resting place. As such they must
407 * not conflict with either the destination addresses
408 * or memory the kernel is already using.
410 * The only case where we really need more than one of
411 * these are for architectures where we cannot disable
412 * the MMU and must instead generate an identity mapped
413 * page table for all of the memory.
415 * At worst this runs in O(N) of the image size.
417 struct list_head extra_pages;
418 struct page *pages;
419 unsigned int count;
421 count = 1 << order;
422 INIT_LIST_HEAD(&extra_pages);
424 /* Loop while I can allocate a page and the page allocated
425 * is a destination page.
427 do {
428 unsigned long pfn, epfn, addr, eaddr;
430 pages = kimage_alloc_pages(GFP_KERNEL, order);
431 if (!pages)
432 break;
433 pfn = page_to_pfn(pages);
434 epfn = pfn + count;
435 addr = pfn << PAGE_SHIFT;
436 eaddr = epfn << PAGE_SHIFT;
437 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
438 kimage_is_destination_range(image, addr, eaddr)) {
439 list_add(&pages->lru, &extra_pages);
440 pages = NULL;
442 } while (!pages);
444 if (pages) {
445 /* Remember the allocated page... */
446 list_add(&pages->lru, &image->control_pages);
448 /* Because the page is already in it's destination
449 * location we will never allocate another page at
450 * that address. Therefore kimage_alloc_pages
451 * will not return it (again) and we don't need
452 * to give it an entry in image->segment[].
455 /* Deal with the destination pages I have inadvertently allocated.
457 * Ideally I would convert multi-page allocations into single
458 * page allocations, and add everything to image->dest_pages.
460 * For now it is simpler to just free the pages.
462 kimage_free_page_list(&extra_pages);
464 return pages;
467 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
468 unsigned int order)
470 /* Control pages are special, they are the intermediaries
471 * that are needed while we copy the rest of the pages
472 * to their final resting place. As such they must
473 * not conflict with either the destination addresses
474 * or memory the kernel is already using.
476 * Control pages are also the only pags we must allocate
477 * when loading a crash kernel. All of the other pages
478 * are specified by the segments and we just memcpy
479 * into them directly.
481 * The only case where we really need more than one of
482 * these are for architectures where we cannot disable
483 * the MMU and must instead generate an identity mapped
484 * page table for all of the memory.
486 * Given the low demand this implements a very simple
487 * allocator that finds the first hole of the appropriate
488 * size in the reserved memory region, and allocates all
489 * of the memory up to and including the hole.
491 unsigned long hole_start, hole_end, size;
492 struct page *pages;
494 pages = NULL;
495 size = (1 << order) << PAGE_SHIFT;
496 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
497 hole_end = hole_start + size - 1;
498 while (hole_end <= crashk_res.end) {
499 unsigned long i;
501 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
502 break;
503 if (hole_end > crashk_res.end)
504 break;
505 /* See if I overlap any of the segments */
506 for (i = 0; i < image->nr_segments; i++) {
507 unsigned long mstart, mend;
509 mstart = image->segment[i].mem;
510 mend = mstart + image->segment[i].memsz - 1;
511 if ((hole_end >= mstart) && (hole_start <= mend)) {
512 /* Advance the hole to the end of the segment */
513 hole_start = (mend + (size - 1)) & ~(size - 1);
514 hole_end = hole_start + size - 1;
515 break;
518 /* If I don't overlap any segments I have found my hole! */
519 if (i == image->nr_segments) {
520 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
521 break;
524 if (pages)
525 image->control_page = hole_end;
527 return pages;
531 struct page *kimage_alloc_control_pages(struct kimage *image,
532 unsigned int order)
534 struct page *pages = NULL;
536 switch (image->type) {
537 case KEXEC_TYPE_DEFAULT:
538 pages = kimage_alloc_normal_control_pages(image, order);
539 break;
540 case KEXEC_TYPE_CRASH:
541 pages = kimage_alloc_crash_control_pages(image, order);
542 break;
545 return pages;
548 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
550 if (*image->entry != 0)
551 image->entry++;
553 if (image->entry == image->last_entry) {
554 kimage_entry_t *ind_page;
555 struct page *page;
557 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
558 if (!page)
559 return -ENOMEM;
561 ind_page = page_address(page);
562 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
563 image->entry = ind_page;
564 image->last_entry = ind_page +
565 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
567 *image->entry = entry;
568 image->entry++;
569 *image->entry = 0;
571 return 0;
574 static int kimage_set_destination(struct kimage *image,
575 unsigned long destination)
577 int result;
579 destination &= PAGE_MASK;
580 result = kimage_add_entry(image, destination | IND_DESTINATION);
581 if (result == 0)
582 image->destination = destination;
584 return result;
588 static int kimage_add_page(struct kimage *image, unsigned long page)
590 int result;
592 page &= PAGE_MASK;
593 result = kimage_add_entry(image, page | IND_SOURCE);
594 if (result == 0)
595 image->destination += PAGE_SIZE;
597 return result;
601 static void kimage_free_extra_pages(struct kimage *image)
603 /* Walk through and free any extra destination pages I may have */
604 kimage_free_page_list(&image->dest_pages);
606 /* Walk through and free any unusable pages I have cached */
607 kimage_free_page_list(&image->unuseable_pages);
610 static void kimage_terminate(struct kimage *image)
612 if (*image->entry != 0)
613 image->entry++;
615 *image->entry = IND_DONE;
618 #define for_each_kimage_entry(image, ptr, entry) \
619 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
620 ptr = (entry & IND_INDIRECTION)? \
621 phys_to_virt((entry & PAGE_MASK)): ptr +1)
623 static void kimage_free_entry(kimage_entry_t entry)
625 struct page *page;
627 page = pfn_to_page(entry >> PAGE_SHIFT);
628 kimage_free_pages(page);
631 static void kimage_free(struct kimage *image)
633 kimage_entry_t *ptr, entry;
634 kimage_entry_t ind = 0;
636 if (!image)
637 return;
639 kimage_free_extra_pages(image);
640 for_each_kimage_entry(image, ptr, entry) {
641 if (entry & IND_INDIRECTION) {
642 /* Free the previous indirection page */
643 if (ind & IND_INDIRECTION)
644 kimage_free_entry(ind);
645 /* Save this indirection page until we are
646 * done with it.
648 ind = entry;
650 else if (entry & IND_SOURCE)
651 kimage_free_entry(entry);
653 /* Free the final indirection page */
654 if (ind & IND_INDIRECTION)
655 kimage_free_entry(ind);
657 /* Handle any machine specific cleanup */
658 machine_kexec_cleanup(image);
660 /* Free the kexec control pages... */
661 kimage_free_page_list(&image->control_pages);
662 kfree(image);
665 static kimage_entry_t *kimage_dst_used(struct kimage *image,
666 unsigned long page)
668 kimage_entry_t *ptr, entry;
669 unsigned long destination = 0;
671 for_each_kimage_entry(image, ptr, entry) {
672 if (entry & IND_DESTINATION)
673 destination = entry & PAGE_MASK;
674 else if (entry & IND_SOURCE) {
675 if (page == destination)
676 return ptr;
677 destination += PAGE_SIZE;
681 return NULL;
684 static struct page *kimage_alloc_page(struct kimage *image,
685 gfp_t gfp_mask,
686 unsigned long destination)
689 * Here we implement safeguards to ensure that a source page
690 * is not copied to its destination page before the data on
691 * the destination page is no longer useful.
693 * To do this we maintain the invariant that a source page is
694 * either its own destination page, or it is not a
695 * destination page at all.
697 * That is slightly stronger than required, but the proof
698 * that no problems will not occur is trivial, and the
699 * implementation is simply to verify.
701 * When allocating all pages normally this algorithm will run
702 * in O(N) time, but in the worst case it will run in O(N^2)
703 * time. If the runtime is a problem the data structures can
704 * be fixed.
706 struct page *page;
707 unsigned long addr;
710 * Walk through the list of destination pages, and see if I
711 * have a match.
713 list_for_each_entry(page, &image->dest_pages, lru) {
714 addr = page_to_pfn(page) << PAGE_SHIFT;
715 if (addr == destination) {
716 list_del(&page->lru);
717 return page;
720 page = NULL;
721 while (1) {
722 kimage_entry_t *old;
724 /* Allocate a page, if we run out of memory give up */
725 page = kimage_alloc_pages(gfp_mask, 0);
726 if (!page)
727 return NULL;
728 /* If the page cannot be used file it away */
729 if (page_to_pfn(page) >
730 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
731 list_add(&page->lru, &image->unuseable_pages);
732 continue;
734 addr = page_to_pfn(page) << PAGE_SHIFT;
736 /* If it is the destination page we want use it */
737 if (addr == destination)
738 break;
740 /* If the page is not a destination page use it */
741 if (!kimage_is_destination_range(image, addr,
742 addr + PAGE_SIZE))
743 break;
746 * I know that the page is someones destination page.
747 * See if there is already a source page for this
748 * destination page. And if so swap the source pages.
750 old = kimage_dst_used(image, addr);
751 if (old) {
752 /* If so move it */
753 unsigned long old_addr;
754 struct page *old_page;
756 old_addr = *old & PAGE_MASK;
757 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
758 copy_highpage(page, old_page);
759 *old = addr | (*old & ~PAGE_MASK);
761 /* The old page I have found cannot be a
762 * destination page, so return it if it's
763 * gfp_flags honor the ones passed in.
765 if (!(gfp_mask & __GFP_HIGHMEM) &&
766 PageHighMem(old_page)) {
767 kimage_free_pages(old_page);
768 continue;
770 addr = old_addr;
771 page = old_page;
772 break;
774 else {
775 /* Place the page on the destination list I
776 * will use it later.
778 list_add(&page->lru, &image->dest_pages);
782 return page;
785 static int kimage_load_normal_segment(struct kimage *image,
786 struct kexec_segment *segment)
788 unsigned long maddr;
789 unsigned long ubytes, mbytes;
790 int result;
791 unsigned char __user *buf;
793 result = 0;
794 buf = segment->buf;
795 ubytes = segment->bufsz;
796 mbytes = segment->memsz;
797 maddr = segment->mem;
799 result = kimage_set_destination(image, maddr);
800 if (result < 0)
801 goto out;
803 while (mbytes) {
804 struct page *page;
805 char *ptr;
806 size_t uchunk, mchunk;
808 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
809 if (!page) {
810 result = -ENOMEM;
811 goto out;
813 result = kimage_add_page(image, page_to_pfn(page)
814 << PAGE_SHIFT);
815 if (result < 0)
816 goto out;
818 ptr = kmap(page);
819 /* Start with a clear page */
820 clear_page(ptr);
821 ptr += maddr & ~PAGE_MASK;
822 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
823 if (mchunk > mbytes)
824 mchunk = mbytes;
826 uchunk = mchunk;
827 if (uchunk > ubytes)
828 uchunk = ubytes;
830 result = copy_from_user(ptr, buf, uchunk);
831 kunmap(page);
832 if (result) {
833 result = -EFAULT;
834 goto out;
836 ubytes -= uchunk;
837 maddr += mchunk;
838 buf += mchunk;
839 mbytes -= mchunk;
841 out:
842 return result;
845 static int kimage_load_crash_segment(struct kimage *image,
846 struct kexec_segment *segment)
848 /* For crash dumps kernels we simply copy the data from
849 * user space to it's destination.
850 * We do things a page at a time for the sake of kmap.
852 unsigned long maddr;
853 unsigned long ubytes, mbytes;
854 int result;
855 unsigned char __user *buf;
857 result = 0;
858 buf = segment->buf;
859 ubytes = segment->bufsz;
860 mbytes = segment->memsz;
861 maddr = segment->mem;
862 while (mbytes) {
863 struct page *page;
864 char *ptr;
865 size_t uchunk, mchunk;
867 page = pfn_to_page(maddr >> PAGE_SHIFT);
868 if (!page) {
869 result = -ENOMEM;
870 goto out;
872 ptr = kmap(page);
873 ptr += maddr & ~PAGE_MASK;
874 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
875 if (mchunk > mbytes)
876 mchunk = mbytes;
878 uchunk = mchunk;
879 if (uchunk > ubytes) {
880 uchunk = ubytes;
881 /* Zero the trailing part of the page */
882 memset(ptr + uchunk, 0, mchunk - uchunk);
884 result = copy_from_user(ptr, buf, uchunk);
885 kexec_flush_icache_page(page);
886 kunmap(page);
887 if (result) {
888 result = -EFAULT;
889 goto out;
891 ubytes -= uchunk;
892 maddr += mchunk;
893 buf += mchunk;
894 mbytes -= mchunk;
896 out:
897 return result;
900 static int kimage_load_segment(struct kimage *image,
901 struct kexec_segment *segment)
903 int result = -ENOMEM;
905 switch (image->type) {
906 case KEXEC_TYPE_DEFAULT:
907 result = kimage_load_normal_segment(image, segment);
908 break;
909 case KEXEC_TYPE_CRASH:
910 result = kimage_load_crash_segment(image, segment);
911 break;
914 return result;
918 * Exec Kernel system call: for obvious reasons only root may call it.
920 * This call breaks up into three pieces.
921 * - A generic part which loads the new kernel from the current
922 * address space, and very carefully places the data in the
923 * allocated pages.
925 * - A generic part that interacts with the kernel and tells all of
926 * the devices to shut down. Preventing on-going dmas, and placing
927 * the devices in a consistent state so a later kernel can
928 * reinitialize them.
930 * - A machine specific part that includes the syscall number
931 * and the copies the image to it's final destination. And
932 * jumps into the image at entry.
934 * kexec does not sync, or unmount filesystems so if you need
935 * that to happen you need to do that yourself.
937 struct kimage *kexec_image;
938 struct kimage *kexec_crash_image;
940 static DEFINE_MUTEX(kexec_mutex);
942 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
943 struct kexec_segment __user *, segments, unsigned long, flags)
945 struct kimage **dest_image, *image;
946 int result;
948 /* We only trust the superuser with rebooting the system. */
949 if (!capable(CAP_SYS_BOOT))
950 return -EPERM;
953 * Verify we have a legal set of flags
954 * This leaves us room for future extensions.
956 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
957 return -EINVAL;
959 /* Verify we are on the appropriate architecture */
960 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
961 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
962 return -EINVAL;
964 /* Put an artificial cap on the number
965 * of segments passed to kexec_load.
967 if (nr_segments > KEXEC_SEGMENT_MAX)
968 return -EINVAL;
970 image = NULL;
971 result = 0;
973 /* Because we write directly to the reserved memory
974 * region when loading crash kernels we need a mutex here to
975 * prevent multiple crash kernels from attempting to load
976 * simultaneously, and to prevent a crash kernel from loading
977 * over the top of a in use crash kernel.
979 * KISS: always take the mutex.
981 if (!mutex_trylock(&kexec_mutex))
982 return -EBUSY;
984 dest_image = &kexec_image;
985 if (flags & KEXEC_ON_CRASH)
986 dest_image = &kexec_crash_image;
987 if (nr_segments > 0) {
988 unsigned long i;
990 /* Loading another kernel to reboot into */
991 if ((flags & KEXEC_ON_CRASH) == 0)
992 result = kimage_normal_alloc(&image, entry,
993 nr_segments, segments);
994 /* Loading another kernel to switch to if this one crashes */
995 else if (flags & KEXEC_ON_CRASH) {
996 /* Free any current crash dump kernel before
997 * we corrupt it.
999 kimage_free(xchg(&kexec_crash_image, NULL));
1000 result = kimage_crash_alloc(&image, entry,
1001 nr_segments, segments);
1003 if (result)
1004 goto out;
1006 if (flags & KEXEC_PRESERVE_CONTEXT)
1007 image->preserve_context = 1;
1008 result = machine_kexec_prepare(image);
1009 if (result)
1010 goto out;
1012 for (i = 0; i < nr_segments; i++) {
1013 result = kimage_load_segment(image, &image->segment[i]);
1014 if (result)
1015 goto out;
1017 kimage_terminate(image);
1019 /* Install the new kernel, and Uninstall the old */
1020 image = xchg(dest_image, image);
1022 out:
1023 mutex_unlock(&kexec_mutex);
1024 kimage_free(image);
1026 return result;
1029 #ifdef CONFIG_COMPAT
1030 asmlinkage long compat_sys_kexec_load(unsigned long entry,
1031 unsigned long nr_segments,
1032 struct compat_kexec_segment __user *segments,
1033 unsigned long flags)
1035 struct compat_kexec_segment in;
1036 struct kexec_segment out, __user *ksegments;
1037 unsigned long i, result;
1039 /* Don't allow clients that don't understand the native
1040 * architecture to do anything.
1042 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1043 return -EINVAL;
1045 if (nr_segments > KEXEC_SEGMENT_MAX)
1046 return -EINVAL;
1048 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1049 for (i=0; i < nr_segments; i++) {
1050 result = copy_from_user(&in, &segments[i], sizeof(in));
1051 if (result)
1052 return -EFAULT;
1054 out.buf = compat_ptr(in.buf);
1055 out.bufsz = in.bufsz;
1056 out.mem = in.mem;
1057 out.memsz = in.memsz;
1059 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1060 if (result)
1061 return -EFAULT;
1064 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1066 #endif
1068 void crash_kexec(struct pt_regs *regs)
1070 /* Take the kexec_mutex here to prevent sys_kexec_load
1071 * running on one cpu from replacing the crash kernel
1072 * we are using after a panic on a different cpu.
1074 * If the crash kernel was not located in a fixed area
1075 * of memory the xchg(&kexec_crash_image) would be
1076 * sufficient. But since I reuse the memory...
1078 if (mutex_trylock(&kexec_mutex)) {
1079 if (kexec_crash_image) {
1080 struct pt_regs fixed_regs;
1082 kmsg_dump(KMSG_DUMP_KEXEC);
1084 crash_setup_regs(&fixed_regs, regs);
1085 crash_save_vmcoreinfo();
1086 machine_crash_shutdown(&fixed_regs);
1087 machine_kexec(kexec_crash_image);
1089 mutex_unlock(&kexec_mutex);
1093 size_t crash_get_memory_size(void)
1095 size_t size = 0;
1096 mutex_lock(&kexec_mutex);
1097 if (crashk_res.end != crashk_res.start)
1098 size = resource_size(&crashk_res);
1099 mutex_unlock(&kexec_mutex);
1100 return size;
1103 void __weak crash_free_reserved_phys_range(unsigned long begin,
1104 unsigned long end)
1106 unsigned long addr;
1108 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1109 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1110 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1111 free_page((unsigned long)__va(addr));
1112 totalram_pages++;
1116 int crash_shrink_memory(unsigned long new_size)
1118 int ret = 0;
1119 unsigned long start, end;
1121 mutex_lock(&kexec_mutex);
1123 if (kexec_crash_image) {
1124 ret = -ENOENT;
1125 goto unlock;
1127 start = crashk_res.start;
1128 end = crashk_res.end;
1130 if (new_size >= end - start + 1) {
1131 ret = -EINVAL;
1132 if (new_size == end - start + 1)
1133 ret = 0;
1134 goto unlock;
1137 start = roundup(start, PAGE_SIZE);
1138 end = roundup(start + new_size, PAGE_SIZE);
1140 crash_free_reserved_phys_range(end, crashk_res.end);
1142 if ((start == end) && (crashk_res.parent != NULL))
1143 release_resource(&crashk_res);
1144 crashk_res.end = end - 1;
1146 unlock:
1147 mutex_unlock(&kexec_mutex);
1148 return ret;
1151 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1152 size_t data_len)
1154 struct elf_note note;
1156 note.n_namesz = strlen(name) + 1;
1157 note.n_descsz = data_len;
1158 note.n_type = type;
1159 memcpy(buf, &note, sizeof(note));
1160 buf += (sizeof(note) + 3)/4;
1161 memcpy(buf, name, note.n_namesz);
1162 buf += (note.n_namesz + 3)/4;
1163 memcpy(buf, data, note.n_descsz);
1164 buf += (note.n_descsz + 3)/4;
1166 return buf;
1169 static void final_note(u32 *buf)
1171 struct elf_note note;
1173 note.n_namesz = 0;
1174 note.n_descsz = 0;
1175 note.n_type = 0;
1176 memcpy(buf, &note, sizeof(note));
1179 void crash_save_cpu(struct pt_regs *regs, int cpu)
1181 struct elf_prstatus prstatus;
1182 u32 *buf;
1184 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1185 return;
1187 /* Using ELF notes here is opportunistic.
1188 * I need a well defined structure format
1189 * for the data I pass, and I need tags
1190 * on the data to indicate what information I have
1191 * squirrelled away. ELF notes happen to provide
1192 * all of that, so there is no need to invent something new.
1194 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1195 if (!buf)
1196 return;
1197 memset(&prstatus, 0, sizeof(prstatus));
1198 prstatus.pr_pid = current->pid;
1199 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1200 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1201 &prstatus, sizeof(prstatus));
1202 final_note(buf);
1205 static int __init crash_notes_memory_init(void)
1207 /* Allocate memory for saving cpu registers. */
1208 crash_notes = alloc_percpu(note_buf_t);
1209 if (!crash_notes) {
1210 printk("Kexec: Memory allocation for saving cpu register"
1211 " states failed\n");
1212 return -ENOMEM;
1214 return 0;
1216 module_init(crash_notes_memory_init)
1220 * parsing the "crashkernel" commandline
1222 * this code is intended to be called from architecture specific code
1227 * This function parses command lines in the format
1229 * crashkernel=ramsize-range:size[,...][@offset]
1231 * The function returns 0 on success and -EINVAL on failure.
1233 static int __init parse_crashkernel_mem(char *cmdline,
1234 unsigned long long system_ram,
1235 unsigned long long *crash_size,
1236 unsigned long long *crash_base)
1238 char *cur = cmdline, *tmp;
1240 /* for each entry of the comma-separated list */
1241 do {
1242 unsigned long long start, end = ULLONG_MAX, size;
1244 /* get the start of the range */
1245 start = memparse(cur, &tmp);
1246 if (cur == tmp) {
1247 pr_warning("crashkernel: Memory value expected\n");
1248 return -EINVAL;
1250 cur = tmp;
1251 if (*cur != '-') {
1252 pr_warning("crashkernel: '-' expected\n");
1253 return -EINVAL;
1255 cur++;
1257 /* if no ':' is here, than we read the end */
1258 if (*cur != ':') {
1259 end = memparse(cur, &tmp);
1260 if (cur == tmp) {
1261 pr_warning("crashkernel: Memory "
1262 "value expected\n");
1263 return -EINVAL;
1265 cur = tmp;
1266 if (end <= start) {
1267 pr_warning("crashkernel: end <= start\n");
1268 return -EINVAL;
1272 if (*cur != ':') {
1273 pr_warning("crashkernel: ':' expected\n");
1274 return -EINVAL;
1276 cur++;
1278 size = memparse(cur, &tmp);
1279 if (cur == tmp) {
1280 pr_warning("Memory value expected\n");
1281 return -EINVAL;
1283 cur = tmp;
1284 if (size >= system_ram) {
1285 pr_warning("crashkernel: invalid size\n");
1286 return -EINVAL;
1289 /* match ? */
1290 if (system_ram >= start && system_ram < end) {
1291 *crash_size = size;
1292 break;
1294 } while (*cur++ == ',');
1296 if (*crash_size > 0) {
1297 while (*cur && *cur != ' ' && *cur != '@')
1298 cur++;
1299 if (*cur == '@') {
1300 cur++;
1301 *crash_base = memparse(cur, &tmp);
1302 if (cur == tmp) {
1303 pr_warning("Memory value expected "
1304 "after '@'\n");
1305 return -EINVAL;
1310 return 0;
1314 * That function parses "simple" (old) crashkernel command lines like
1316 * crashkernel=size[@offset]
1318 * It returns 0 on success and -EINVAL on failure.
1320 static int __init parse_crashkernel_simple(char *cmdline,
1321 unsigned long long *crash_size,
1322 unsigned long long *crash_base)
1324 char *cur = cmdline;
1326 *crash_size = memparse(cmdline, &cur);
1327 if (cmdline == cur) {
1328 pr_warning("crashkernel: memory value expected\n");
1329 return -EINVAL;
1332 if (*cur == '@')
1333 *crash_base = memparse(cur+1, &cur);
1335 return 0;
1339 * That function is the entry point for command line parsing and should be
1340 * called from the arch-specific code.
1342 int __init parse_crashkernel(char *cmdline,
1343 unsigned long long system_ram,
1344 unsigned long long *crash_size,
1345 unsigned long long *crash_base)
1347 char *p = cmdline, *ck_cmdline = NULL;
1348 char *first_colon, *first_space;
1350 BUG_ON(!crash_size || !crash_base);
1351 *crash_size = 0;
1352 *crash_base = 0;
1354 /* find crashkernel and use the last one if there are more */
1355 p = strstr(p, "crashkernel=");
1356 while (p) {
1357 ck_cmdline = p;
1358 p = strstr(p+1, "crashkernel=");
1361 if (!ck_cmdline)
1362 return -EINVAL;
1364 ck_cmdline += 12; /* strlen("crashkernel=") */
1367 * if the commandline contains a ':', then that's the extended
1368 * syntax -- if not, it must be the classic syntax
1370 first_colon = strchr(ck_cmdline, ':');
1371 first_space = strchr(ck_cmdline, ' ');
1372 if (first_colon && (!first_space || first_colon < first_space))
1373 return parse_crashkernel_mem(ck_cmdline, system_ram,
1374 crash_size, crash_base);
1375 else
1376 return parse_crashkernel_simple(ck_cmdline, crash_size,
1377 crash_base);
1379 return 0;
1384 void crash_save_vmcoreinfo(void)
1386 u32 *buf;
1388 if (!vmcoreinfo_size)
1389 return;
1391 vmcoreinfo_append_str("CRASHTIME=%ld", get_seconds());
1393 buf = (u32 *)vmcoreinfo_note;
1395 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1396 vmcoreinfo_size);
1398 final_note(buf);
1401 void vmcoreinfo_append_str(const char *fmt, ...)
1403 va_list args;
1404 char buf[0x50];
1405 int r;
1407 va_start(args, fmt);
1408 r = vsnprintf(buf, sizeof(buf), fmt, args);
1409 va_end(args);
1411 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1412 r = vmcoreinfo_max_size - vmcoreinfo_size;
1414 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1416 vmcoreinfo_size += r;
1420 * provide an empty default implementation here -- architecture
1421 * code may override this
1423 void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1426 unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1428 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1431 static int __init crash_save_vmcoreinfo_init(void)
1433 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1434 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1436 VMCOREINFO_SYMBOL(init_uts_ns);
1437 VMCOREINFO_SYMBOL(node_online_map);
1438 VMCOREINFO_SYMBOL(swapper_pg_dir);
1439 VMCOREINFO_SYMBOL(_stext);
1440 VMCOREINFO_SYMBOL(vmlist);
1442 #ifndef CONFIG_NEED_MULTIPLE_NODES
1443 VMCOREINFO_SYMBOL(mem_map);
1444 VMCOREINFO_SYMBOL(contig_page_data);
1445 #endif
1446 #ifdef CONFIG_SPARSEMEM
1447 VMCOREINFO_SYMBOL(mem_section);
1448 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1449 VMCOREINFO_STRUCT_SIZE(mem_section);
1450 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1451 #endif
1452 VMCOREINFO_STRUCT_SIZE(page);
1453 VMCOREINFO_STRUCT_SIZE(pglist_data);
1454 VMCOREINFO_STRUCT_SIZE(zone);
1455 VMCOREINFO_STRUCT_SIZE(free_area);
1456 VMCOREINFO_STRUCT_SIZE(list_head);
1457 VMCOREINFO_SIZE(nodemask_t);
1458 VMCOREINFO_OFFSET(page, flags);
1459 VMCOREINFO_OFFSET(page, _count);
1460 VMCOREINFO_OFFSET(page, mapping);
1461 VMCOREINFO_OFFSET(page, lru);
1462 VMCOREINFO_OFFSET(pglist_data, node_zones);
1463 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1464 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1465 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1466 #endif
1467 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1468 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1469 VMCOREINFO_OFFSET(pglist_data, node_id);
1470 VMCOREINFO_OFFSET(zone, free_area);
1471 VMCOREINFO_OFFSET(zone, vm_stat);
1472 VMCOREINFO_OFFSET(zone, spanned_pages);
1473 VMCOREINFO_OFFSET(free_area, free_list);
1474 VMCOREINFO_OFFSET(list_head, next);
1475 VMCOREINFO_OFFSET(list_head, prev);
1476 VMCOREINFO_OFFSET(vm_struct, addr);
1477 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1478 log_buf_kexec_setup();
1479 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1480 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1481 VMCOREINFO_NUMBER(PG_lru);
1482 VMCOREINFO_NUMBER(PG_private);
1483 VMCOREINFO_NUMBER(PG_swapcache);
1485 arch_crash_save_vmcoreinfo();
1487 return 0;
1490 module_init(crash_save_vmcoreinfo_init)
1493 * Move into place and start executing a preloaded standalone
1494 * executable. If nothing was preloaded return an error.
1496 int kernel_kexec(void)
1498 int error = 0;
1500 if (!mutex_trylock(&kexec_mutex))
1501 return -EBUSY;
1502 if (!kexec_image) {
1503 error = -EINVAL;
1504 goto Unlock;
1507 #ifdef CONFIG_KEXEC_JUMP
1508 if (kexec_image->preserve_context) {
1509 mutex_lock(&pm_mutex);
1510 pm_prepare_console();
1511 error = freeze_processes();
1512 if (error) {
1513 error = -EBUSY;
1514 goto Restore_console;
1516 suspend_console();
1517 error = dpm_suspend_start(PMSG_FREEZE);
1518 if (error)
1519 goto Resume_console;
1520 /* At this point, dpm_suspend_start() has been called,
1521 * but *not* dpm_suspend_noirq(). We *must* call
1522 * dpm_suspend_noirq() now. Otherwise, drivers for
1523 * some devices (e.g. interrupt controllers) become
1524 * desynchronized with the actual state of the
1525 * hardware at resume time, and evil weirdness ensues.
1527 error = dpm_suspend_noirq(PMSG_FREEZE);
1528 if (error)
1529 goto Resume_devices;
1530 error = disable_nonboot_cpus();
1531 if (error)
1532 goto Enable_cpus;
1533 local_irq_disable();
1534 error = syscore_suspend();
1535 if (error)
1536 goto Enable_irqs;
1537 } else
1538 #endif
1540 kernel_restart_prepare(NULL);
1541 printk(KERN_EMERG "Starting new kernel\n");
1542 machine_shutdown();
1545 machine_kexec(kexec_image);
1547 #ifdef CONFIG_KEXEC_JUMP
1548 if (kexec_image->preserve_context) {
1549 syscore_resume();
1550 Enable_irqs:
1551 local_irq_enable();
1552 Enable_cpus:
1553 enable_nonboot_cpus();
1554 dpm_resume_noirq(PMSG_RESTORE);
1555 Resume_devices:
1556 dpm_resume_end(PMSG_RESTORE);
1557 Resume_console:
1558 resume_console();
1559 thaw_processes();
1560 Restore_console:
1561 pm_restore_console();
1562 mutex_unlock(&pm_mutex);
1564 #endif
1566 Unlock:
1567 mutex_unlock(&kexec_mutex);
1568 return error;