qos queue
[cor_2_6_31.git] / kernel / kexec.c
blobf336e2107f980e2546ecc118d2dd3239f655ebae
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 <linux/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>
35 #include <asm/page.h>
36 #include <asm/uaccess.h>
37 #include <asm/io.h>
38 #include <asm/system.h>
39 #include <asm/sections.h>
41 /* Per cpu memory for storing cpu states in case of system crash. */
42 note_buf_t* crash_notes;
44 /* vmcoreinfo stuff */
45 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
46 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
47 size_t vmcoreinfo_size;
48 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
50 /* Location of the reserved area for the crash kernel */
51 struct resource crashk_res = {
52 .name = "Crash kernel",
53 .start = 0,
54 .end = 0,
55 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
58 int kexec_should_crash(struct task_struct *p)
60 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
61 return 1;
62 return 0;
66 * When kexec transitions to the new kernel there is a one-to-one
67 * mapping between physical and virtual addresses. On processors
68 * where you can disable the MMU this is trivial, and easy. For
69 * others it is still a simple predictable page table to setup.
71 * In that environment kexec copies the new kernel to its final
72 * resting place. This means I can only support memory whose
73 * physical address can fit in an unsigned long. In particular
74 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
75 * If the assembly stub has more restrictive requirements
76 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
77 * defined more restrictively in <asm/kexec.h>.
79 * The code for the transition from the current kernel to the
80 * the new kernel is placed in the control_code_buffer, whose size
81 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
82 * page of memory is necessary, but some architectures require more.
83 * Because this memory must be identity mapped in the transition from
84 * virtual to physical addresses it must live in the range
85 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
86 * modifiable.
88 * The assembly stub in the control code buffer is passed a linked list
89 * of descriptor pages detailing the source pages of the new kernel,
90 * and the destination addresses of those source pages. As this data
91 * structure is not used in the context of the current OS, it must
92 * be self-contained.
94 * The code has been made to work with highmem pages and will use a
95 * destination page in its final resting place (if it happens
96 * to allocate it). The end product of this is that most of the
97 * physical address space, and most of RAM can be used.
99 * Future directions include:
100 * - allocating a page table with the control code buffer identity
101 * mapped, to simplify machine_kexec and make kexec_on_panic more
102 * reliable.
106 * KIMAGE_NO_DEST is an impossible destination address..., for
107 * allocating pages whose destination address we do not care about.
109 #define KIMAGE_NO_DEST (-1UL)
111 static int kimage_is_destination_range(struct kimage *image,
112 unsigned long start, unsigned long end);
113 static struct page *kimage_alloc_page(struct kimage *image,
114 gfp_t gfp_mask,
115 unsigned long dest);
117 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
118 unsigned long nr_segments,
119 struct kexec_segment __user *segments)
121 size_t segment_bytes;
122 struct kimage *image;
123 unsigned long i;
124 int result;
126 /* Allocate a controlling structure */
127 result = -ENOMEM;
128 image = kzalloc(sizeof(*image), GFP_KERNEL);
129 if (!image)
130 goto out;
132 image->head = 0;
133 image->entry = &image->head;
134 image->last_entry = &image->head;
135 image->control_page = ~0; /* By default this does not apply */
136 image->start = entry;
137 image->type = KEXEC_TYPE_DEFAULT;
139 /* Initialize the list of control pages */
140 INIT_LIST_HEAD(&image->control_pages);
142 /* Initialize the list of destination pages */
143 INIT_LIST_HEAD(&image->dest_pages);
145 /* Initialize the list of unuseable pages */
146 INIT_LIST_HEAD(&image->unuseable_pages);
148 /* Read in the segments */
149 image->nr_segments = nr_segments;
150 segment_bytes = nr_segments * sizeof(*segments);
151 result = copy_from_user(image->segment, segments, segment_bytes);
152 if (result)
153 goto out;
156 * Verify we have good destination addresses. The caller is
157 * responsible for making certain we don't attempt to load
158 * the new image into invalid or reserved areas of RAM. This
159 * just verifies it is an address we can use.
161 * Since the kernel does everything in page size chunks ensure
162 * the destination addreses are page aligned. Too many
163 * special cases crop of when we don't do this. The most
164 * insidious is getting overlapping destination addresses
165 * simply because addresses are changed to page size
166 * granularity.
168 result = -EADDRNOTAVAIL;
169 for (i = 0; i < nr_segments; i++) {
170 unsigned long mstart, mend;
172 mstart = image->segment[i].mem;
173 mend = mstart + image->segment[i].memsz;
174 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
175 goto out;
176 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
177 goto out;
180 /* Verify our destination addresses do not overlap.
181 * If we alloed overlapping destination addresses
182 * through very weird things can happen with no
183 * easy explanation as one segment stops on another.
185 result = -EINVAL;
186 for (i = 0; i < nr_segments; i++) {
187 unsigned long mstart, mend;
188 unsigned long j;
190 mstart = image->segment[i].mem;
191 mend = mstart + image->segment[i].memsz;
192 for (j = 0; j < i; j++) {
193 unsigned long pstart, pend;
194 pstart = image->segment[j].mem;
195 pend = pstart + image->segment[j].memsz;
196 /* Do the segments overlap ? */
197 if ((mend > pstart) && (mstart < pend))
198 goto out;
202 /* Ensure our buffer sizes are strictly less than
203 * our memory sizes. This should always be the case,
204 * and it is easier to check up front than to be surprised
205 * later on.
207 result = -EINVAL;
208 for (i = 0; i < nr_segments; i++) {
209 if (image->segment[i].bufsz > image->segment[i].memsz)
210 goto out;
213 result = 0;
214 out:
215 if (result == 0)
216 *rimage = image;
217 else
218 kfree(image);
220 return result;
224 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
225 unsigned long nr_segments,
226 struct kexec_segment __user *segments)
228 int result;
229 struct kimage *image;
231 /* Allocate and initialize a controlling structure */
232 image = NULL;
233 result = do_kimage_alloc(&image, entry, nr_segments, segments);
234 if (result)
235 goto out;
237 *rimage = image;
240 * Find a location for the control code buffer, and add it
241 * the vector of segments so that it's pages will also be
242 * counted as destination pages.
244 result = -ENOMEM;
245 image->control_code_page = kimage_alloc_control_pages(image,
246 get_order(KEXEC_CONTROL_PAGE_SIZE));
247 if (!image->control_code_page) {
248 printk(KERN_ERR "Could not allocate control_code_buffer\n");
249 goto out;
252 image->swap_page = kimage_alloc_control_pages(image, 0);
253 if (!image->swap_page) {
254 printk(KERN_ERR "Could not allocate swap buffer\n");
255 goto out;
258 result = 0;
259 out:
260 if (result == 0)
261 *rimage = image;
262 else
263 kfree(image);
265 return result;
268 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
269 unsigned long nr_segments,
270 struct kexec_segment __user *segments)
272 int result;
273 struct kimage *image;
274 unsigned long i;
276 image = NULL;
277 /* Verify we have a valid entry point */
278 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
279 result = -EADDRNOTAVAIL;
280 goto out;
283 /* Allocate and initialize a controlling structure */
284 result = do_kimage_alloc(&image, entry, nr_segments, segments);
285 if (result)
286 goto out;
288 /* Enable the special crash kernel control page
289 * allocation policy.
291 image->control_page = crashk_res.start;
292 image->type = KEXEC_TYPE_CRASH;
295 * Verify we have good destination addresses. Normally
296 * the caller is responsible for making certain we don't
297 * attempt to load the new image into invalid or reserved
298 * areas of RAM. But crash kernels are preloaded into a
299 * reserved area of ram. We must ensure the addresses
300 * are in the reserved area otherwise preloading the
301 * kernel could corrupt things.
303 result = -EADDRNOTAVAIL;
304 for (i = 0; i < nr_segments; i++) {
305 unsigned long mstart, mend;
307 mstart = image->segment[i].mem;
308 mend = mstart + image->segment[i].memsz - 1;
309 /* Ensure we are within the crash kernel limits */
310 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
311 goto out;
315 * Find a location for the control code buffer, and add
316 * the vector of segments so that it's pages will also be
317 * counted as destination pages.
319 result = -ENOMEM;
320 image->control_code_page = kimage_alloc_control_pages(image,
321 get_order(KEXEC_CONTROL_PAGE_SIZE));
322 if (!image->control_code_page) {
323 printk(KERN_ERR "Could not allocate control_code_buffer\n");
324 goto out;
327 result = 0;
328 out:
329 if (result == 0)
330 *rimage = image;
331 else
332 kfree(image);
334 return result;
337 static int kimage_is_destination_range(struct kimage *image,
338 unsigned long start,
339 unsigned long end)
341 unsigned long i;
343 for (i = 0; i < image->nr_segments; i++) {
344 unsigned long mstart, mend;
346 mstart = image->segment[i].mem;
347 mend = mstart + image->segment[i].memsz;
348 if ((end > mstart) && (start < mend))
349 return 1;
352 return 0;
355 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
357 struct page *pages;
359 pages = alloc_pages(gfp_mask, order);
360 if (pages) {
361 unsigned int count, i;
362 pages->mapping = NULL;
363 set_page_private(pages, order);
364 count = 1 << order;
365 for (i = 0; i < count; i++)
366 SetPageReserved(pages + i);
369 return pages;
372 static void kimage_free_pages(struct page *page)
374 unsigned int order, count, i;
376 order = page_private(page);
377 count = 1 << order;
378 for (i = 0; i < count; i++)
379 ClearPageReserved(page + i);
380 __free_pages(page, order);
383 static void kimage_free_page_list(struct list_head *list)
385 struct list_head *pos, *next;
387 list_for_each_safe(pos, next, list) {
388 struct page *page;
390 page = list_entry(pos, struct page, lru);
391 list_del(&page->lru);
392 kimage_free_pages(page);
396 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
397 unsigned int order)
399 /* Control pages are special, they are the intermediaries
400 * that are needed while we copy the rest of the pages
401 * to their final resting place. As such they must
402 * not conflict with either the destination addresses
403 * or memory the kernel is already using.
405 * The only case where we really need more than one of
406 * these are for architectures where we cannot disable
407 * the MMU and must instead generate an identity mapped
408 * page table for all of the memory.
410 * At worst this runs in O(N) of the image size.
412 struct list_head extra_pages;
413 struct page *pages;
414 unsigned int count;
416 count = 1 << order;
417 INIT_LIST_HEAD(&extra_pages);
419 /* Loop while I can allocate a page and the page allocated
420 * is a destination page.
422 do {
423 unsigned long pfn, epfn, addr, eaddr;
425 pages = kimage_alloc_pages(GFP_KERNEL, order);
426 if (!pages)
427 break;
428 pfn = page_to_pfn(pages);
429 epfn = pfn + count;
430 addr = pfn << PAGE_SHIFT;
431 eaddr = epfn << PAGE_SHIFT;
432 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
433 kimage_is_destination_range(image, addr, eaddr)) {
434 list_add(&pages->lru, &extra_pages);
435 pages = NULL;
437 } while (!pages);
439 if (pages) {
440 /* Remember the allocated page... */
441 list_add(&pages->lru, &image->control_pages);
443 /* Because the page is already in it's destination
444 * location we will never allocate another page at
445 * that address. Therefore kimage_alloc_pages
446 * will not return it (again) and we don't need
447 * to give it an entry in image->segment[].
450 /* Deal with the destination pages I have inadvertently allocated.
452 * Ideally I would convert multi-page allocations into single
453 * page allocations, and add everyting to image->dest_pages.
455 * For now it is simpler to just free the pages.
457 kimage_free_page_list(&extra_pages);
459 return pages;
462 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
463 unsigned int order)
465 /* Control pages are special, they are the intermediaries
466 * that are needed while we copy the rest of the pages
467 * to their final resting place. As such they must
468 * not conflict with either the destination addresses
469 * or memory the kernel is already using.
471 * Control pages are also the only pags we must allocate
472 * when loading a crash kernel. All of the other pages
473 * are specified by the segments and we just memcpy
474 * into them directly.
476 * The only case where we really need more than one of
477 * these are for architectures where we cannot disable
478 * the MMU and must instead generate an identity mapped
479 * page table for all of the memory.
481 * Given the low demand this implements a very simple
482 * allocator that finds the first hole of the appropriate
483 * size in the reserved memory region, and allocates all
484 * of the memory up to and including the hole.
486 unsigned long hole_start, hole_end, size;
487 struct page *pages;
489 pages = NULL;
490 size = (1 << order) << PAGE_SHIFT;
491 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
492 hole_end = hole_start + size - 1;
493 while (hole_end <= crashk_res.end) {
494 unsigned long i;
496 if (hole_end > KEXEC_CONTROL_MEMORY_LIMIT)
497 break;
498 if (hole_end > crashk_res.end)
499 break;
500 /* See if I overlap any of the segments */
501 for (i = 0; i < image->nr_segments; i++) {
502 unsigned long mstart, mend;
504 mstart = image->segment[i].mem;
505 mend = mstart + image->segment[i].memsz - 1;
506 if ((hole_end >= mstart) && (hole_start <= mend)) {
507 /* Advance the hole to the end of the segment */
508 hole_start = (mend + (size - 1)) & ~(size - 1);
509 hole_end = hole_start + size - 1;
510 break;
513 /* If I don't overlap any segments I have found my hole! */
514 if (i == image->nr_segments) {
515 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
516 break;
519 if (pages)
520 image->control_page = hole_end;
522 return pages;
526 struct page *kimage_alloc_control_pages(struct kimage *image,
527 unsigned int order)
529 struct page *pages = NULL;
531 switch (image->type) {
532 case KEXEC_TYPE_DEFAULT:
533 pages = kimage_alloc_normal_control_pages(image, order);
534 break;
535 case KEXEC_TYPE_CRASH:
536 pages = kimage_alloc_crash_control_pages(image, order);
537 break;
540 return pages;
543 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
545 if (*image->entry != 0)
546 image->entry++;
548 if (image->entry == image->last_entry) {
549 kimage_entry_t *ind_page;
550 struct page *page;
552 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
553 if (!page)
554 return -ENOMEM;
556 ind_page = page_address(page);
557 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
558 image->entry = ind_page;
559 image->last_entry = ind_page +
560 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
562 *image->entry = entry;
563 image->entry++;
564 *image->entry = 0;
566 return 0;
569 static int kimage_set_destination(struct kimage *image,
570 unsigned long destination)
572 int result;
574 destination &= PAGE_MASK;
575 result = kimage_add_entry(image, destination | IND_DESTINATION);
576 if (result == 0)
577 image->destination = destination;
579 return result;
583 static int kimage_add_page(struct kimage *image, unsigned long page)
585 int result;
587 page &= PAGE_MASK;
588 result = kimage_add_entry(image, page | IND_SOURCE);
589 if (result == 0)
590 image->destination += PAGE_SIZE;
592 return result;
596 static void kimage_free_extra_pages(struct kimage *image)
598 /* Walk through and free any extra destination pages I may have */
599 kimage_free_page_list(&image->dest_pages);
601 /* Walk through and free any unuseable pages I have cached */
602 kimage_free_page_list(&image->unuseable_pages);
605 static void kimage_terminate(struct kimage *image)
607 if (*image->entry != 0)
608 image->entry++;
610 *image->entry = IND_DONE;
613 #define for_each_kimage_entry(image, ptr, entry) \
614 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
615 ptr = (entry & IND_INDIRECTION)? \
616 phys_to_virt((entry & PAGE_MASK)): ptr +1)
618 static void kimage_free_entry(kimage_entry_t entry)
620 struct page *page;
622 page = pfn_to_page(entry >> PAGE_SHIFT);
623 kimage_free_pages(page);
626 static void kimage_free(struct kimage *image)
628 kimage_entry_t *ptr, entry;
629 kimage_entry_t ind = 0;
631 if (!image)
632 return;
634 kimage_free_extra_pages(image);
635 for_each_kimage_entry(image, ptr, entry) {
636 if (entry & IND_INDIRECTION) {
637 /* Free the previous indirection page */
638 if (ind & IND_INDIRECTION)
639 kimage_free_entry(ind);
640 /* Save this indirection page until we are
641 * done with it.
643 ind = entry;
645 else if (entry & IND_SOURCE)
646 kimage_free_entry(entry);
648 /* Free the final indirection page */
649 if (ind & IND_INDIRECTION)
650 kimage_free_entry(ind);
652 /* Handle any machine specific cleanup */
653 machine_kexec_cleanup(image);
655 /* Free the kexec control pages... */
656 kimage_free_page_list(&image->control_pages);
657 kfree(image);
660 static kimage_entry_t *kimage_dst_used(struct kimage *image,
661 unsigned long page)
663 kimage_entry_t *ptr, entry;
664 unsigned long destination = 0;
666 for_each_kimage_entry(image, ptr, entry) {
667 if (entry & IND_DESTINATION)
668 destination = entry & PAGE_MASK;
669 else if (entry & IND_SOURCE) {
670 if (page == destination)
671 return ptr;
672 destination += PAGE_SIZE;
676 return NULL;
679 static struct page *kimage_alloc_page(struct kimage *image,
680 gfp_t gfp_mask,
681 unsigned long destination)
684 * Here we implement safeguards to ensure that a source page
685 * is not copied to its destination page before the data on
686 * the destination page is no longer useful.
688 * To do this we maintain the invariant that a source page is
689 * either its own destination page, or it is not a
690 * destination page at all.
692 * That is slightly stronger than required, but the proof
693 * that no problems will not occur is trivial, and the
694 * implementation is simply to verify.
696 * When allocating all pages normally this algorithm will run
697 * in O(N) time, but in the worst case it will run in O(N^2)
698 * time. If the runtime is a problem the data structures can
699 * be fixed.
701 struct page *page;
702 unsigned long addr;
705 * Walk through the list of destination pages, and see if I
706 * have a match.
708 list_for_each_entry(page, &image->dest_pages, lru) {
709 addr = page_to_pfn(page) << PAGE_SHIFT;
710 if (addr == destination) {
711 list_del(&page->lru);
712 return page;
715 page = NULL;
716 while (1) {
717 kimage_entry_t *old;
719 /* Allocate a page, if we run out of memory give up */
720 page = kimage_alloc_pages(gfp_mask, 0);
721 if (!page)
722 return NULL;
723 /* If the page cannot be used file it away */
724 if (page_to_pfn(page) >
725 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
726 list_add(&page->lru, &image->unuseable_pages);
727 continue;
729 addr = page_to_pfn(page) << PAGE_SHIFT;
731 /* If it is the destination page we want use it */
732 if (addr == destination)
733 break;
735 /* If the page is not a destination page use it */
736 if (!kimage_is_destination_range(image, addr,
737 addr + PAGE_SIZE))
738 break;
741 * I know that the page is someones destination page.
742 * See if there is already a source page for this
743 * destination page. And if so swap the source pages.
745 old = kimage_dst_used(image, addr);
746 if (old) {
747 /* If so move it */
748 unsigned long old_addr;
749 struct page *old_page;
751 old_addr = *old & PAGE_MASK;
752 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
753 copy_highpage(page, old_page);
754 *old = addr | (*old & ~PAGE_MASK);
756 /* The old page I have found cannot be a
757 * destination page, so return it if it's
758 * gfp_flags honor the ones passed in.
760 if (!(gfp_mask & __GFP_HIGHMEM) &&
761 PageHighMem(old_page)) {
762 kimage_free_pages(old_page);
763 continue;
765 addr = old_addr;
766 page = old_page;
767 break;
769 else {
770 /* Place the page on the destination list I
771 * will use it later.
773 list_add(&page->lru, &image->dest_pages);
777 return page;
780 static int kimage_load_normal_segment(struct kimage *image,
781 struct kexec_segment *segment)
783 unsigned long maddr;
784 unsigned long ubytes, mbytes;
785 int result;
786 unsigned char __user *buf;
788 result = 0;
789 buf = segment->buf;
790 ubytes = segment->bufsz;
791 mbytes = segment->memsz;
792 maddr = segment->mem;
794 result = kimage_set_destination(image, maddr);
795 if (result < 0)
796 goto out;
798 while (mbytes) {
799 struct page *page;
800 char *ptr;
801 size_t uchunk, mchunk;
803 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
804 if (!page) {
805 result = -ENOMEM;
806 goto out;
808 result = kimage_add_page(image, page_to_pfn(page)
809 << PAGE_SHIFT);
810 if (result < 0)
811 goto out;
813 ptr = kmap(page);
814 /* Start with a clear page */
815 memset(ptr, 0, PAGE_SIZE);
816 ptr += maddr & ~PAGE_MASK;
817 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
818 if (mchunk > mbytes)
819 mchunk = mbytes;
821 uchunk = mchunk;
822 if (uchunk > ubytes)
823 uchunk = ubytes;
825 result = copy_from_user(ptr, buf, uchunk);
826 kunmap(page);
827 if (result) {
828 result = (result < 0) ? result : -EIO;
829 goto out;
831 ubytes -= uchunk;
832 maddr += mchunk;
833 buf += mchunk;
834 mbytes -= mchunk;
836 out:
837 return result;
840 static int kimage_load_crash_segment(struct kimage *image,
841 struct kexec_segment *segment)
843 /* For crash dumps kernels we simply copy the data from
844 * user space to it's destination.
845 * We do things a page at a time for the sake of kmap.
847 unsigned long maddr;
848 unsigned long ubytes, mbytes;
849 int result;
850 unsigned char __user *buf;
852 result = 0;
853 buf = segment->buf;
854 ubytes = segment->bufsz;
855 mbytes = segment->memsz;
856 maddr = segment->mem;
857 while (mbytes) {
858 struct page *page;
859 char *ptr;
860 size_t uchunk, mchunk;
862 page = pfn_to_page(maddr >> PAGE_SHIFT);
863 if (!page) {
864 result = -ENOMEM;
865 goto out;
867 ptr = kmap(page);
868 ptr += maddr & ~PAGE_MASK;
869 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
870 if (mchunk > mbytes)
871 mchunk = mbytes;
873 uchunk = mchunk;
874 if (uchunk > ubytes) {
875 uchunk = ubytes;
876 /* Zero the trailing part of the page */
877 memset(ptr + uchunk, 0, mchunk - uchunk);
879 result = copy_from_user(ptr, buf, uchunk);
880 kexec_flush_icache_page(page);
881 kunmap(page);
882 if (result) {
883 result = (result < 0) ? result : -EIO;
884 goto out;
886 ubytes -= uchunk;
887 maddr += mchunk;
888 buf += mchunk;
889 mbytes -= mchunk;
891 out:
892 return result;
895 static int kimage_load_segment(struct kimage *image,
896 struct kexec_segment *segment)
898 int result = -ENOMEM;
900 switch (image->type) {
901 case KEXEC_TYPE_DEFAULT:
902 result = kimage_load_normal_segment(image, segment);
903 break;
904 case KEXEC_TYPE_CRASH:
905 result = kimage_load_crash_segment(image, segment);
906 break;
909 return result;
913 * Exec Kernel system call: for obvious reasons only root may call it.
915 * This call breaks up into three pieces.
916 * - A generic part which loads the new kernel from the current
917 * address space, and very carefully places the data in the
918 * allocated pages.
920 * - A generic part that interacts with the kernel and tells all of
921 * the devices to shut down. Preventing on-going dmas, and placing
922 * the devices in a consistent state so a later kernel can
923 * reinitialize them.
925 * - A machine specific part that includes the syscall number
926 * and the copies the image to it's final destination. And
927 * jumps into the image at entry.
929 * kexec does not sync, or unmount filesystems so if you need
930 * that to happen you need to do that yourself.
932 struct kimage *kexec_image;
933 struct kimage *kexec_crash_image;
935 static DEFINE_MUTEX(kexec_mutex);
937 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
938 struct kexec_segment __user *, segments, unsigned long, flags)
940 struct kimage **dest_image, *image;
941 int result;
943 /* We only trust the superuser with rebooting the system. */
944 if (!capable(CAP_SYS_BOOT))
945 return -EPERM;
948 * Verify we have a legal set of flags
949 * This leaves us room for future extensions.
951 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
952 return -EINVAL;
954 /* Verify we are on the appropriate architecture */
955 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
956 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
957 return -EINVAL;
959 /* Put an artificial cap on the number
960 * of segments passed to kexec_load.
962 if (nr_segments > KEXEC_SEGMENT_MAX)
963 return -EINVAL;
965 image = NULL;
966 result = 0;
968 /* Because we write directly to the reserved memory
969 * region when loading crash kernels we need a mutex here to
970 * prevent multiple crash kernels from attempting to load
971 * simultaneously, and to prevent a crash kernel from loading
972 * over the top of a in use crash kernel.
974 * KISS: always take the mutex.
976 if (!mutex_trylock(&kexec_mutex))
977 return -EBUSY;
979 dest_image = &kexec_image;
980 if (flags & KEXEC_ON_CRASH)
981 dest_image = &kexec_crash_image;
982 if (nr_segments > 0) {
983 unsigned long i;
985 /* Loading another kernel to reboot into */
986 if ((flags & KEXEC_ON_CRASH) == 0)
987 result = kimage_normal_alloc(&image, entry,
988 nr_segments, segments);
989 /* Loading another kernel to switch to if this one crashes */
990 else if (flags & KEXEC_ON_CRASH) {
991 /* Free any current crash dump kernel before
992 * we corrupt it.
994 kimage_free(xchg(&kexec_crash_image, NULL));
995 result = kimage_crash_alloc(&image, entry,
996 nr_segments, segments);
998 if (result)
999 goto out;
1001 if (flags & KEXEC_PRESERVE_CONTEXT)
1002 image->preserve_context = 1;
1003 result = machine_kexec_prepare(image);
1004 if (result)
1005 goto out;
1007 for (i = 0; i < nr_segments; i++) {
1008 result = kimage_load_segment(image, &image->segment[i]);
1009 if (result)
1010 goto out;
1012 kimage_terminate(image);
1014 /* Install the new kernel, and Uninstall the old */
1015 image = xchg(dest_image, image);
1017 out:
1018 mutex_unlock(&kexec_mutex);
1019 kimage_free(image);
1021 return result;
1024 #ifdef CONFIG_COMPAT
1025 asmlinkage long compat_sys_kexec_load(unsigned long entry,
1026 unsigned long nr_segments,
1027 struct compat_kexec_segment __user *segments,
1028 unsigned long flags)
1030 struct compat_kexec_segment in;
1031 struct kexec_segment out, __user *ksegments;
1032 unsigned long i, result;
1034 /* Don't allow clients that don't understand the native
1035 * architecture to do anything.
1037 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1038 return -EINVAL;
1040 if (nr_segments > KEXEC_SEGMENT_MAX)
1041 return -EINVAL;
1043 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1044 for (i=0; i < nr_segments; i++) {
1045 result = copy_from_user(&in, &segments[i], sizeof(in));
1046 if (result)
1047 return -EFAULT;
1049 out.buf = compat_ptr(in.buf);
1050 out.bufsz = in.bufsz;
1051 out.mem = in.mem;
1052 out.memsz = in.memsz;
1054 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1055 if (result)
1056 return -EFAULT;
1059 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1061 #endif
1063 void crash_kexec(struct pt_regs *regs)
1065 /* Take the kexec_mutex here to prevent sys_kexec_load
1066 * running on one cpu from replacing the crash kernel
1067 * we are using after a panic on a different cpu.
1069 * If the crash kernel was not located in a fixed area
1070 * of memory the xchg(&kexec_crash_image) would be
1071 * sufficient. But since I reuse the memory...
1073 if (mutex_trylock(&kexec_mutex)) {
1074 if (kexec_crash_image) {
1075 struct pt_regs fixed_regs;
1076 crash_setup_regs(&fixed_regs, regs);
1077 crash_save_vmcoreinfo();
1078 machine_crash_shutdown(&fixed_regs);
1079 machine_kexec(kexec_crash_image);
1081 mutex_unlock(&kexec_mutex);
1085 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1086 size_t data_len)
1088 struct elf_note note;
1090 note.n_namesz = strlen(name) + 1;
1091 note.n_descsz = data_len;
1092 note.n_type = type;
1093 memcpy(buf, &note, sizeof(note));
1094 buf += (sizeof(note) + 3)/4;
1095 memcpy(buf, name, note.n_namesz);
1096 buf += (note.n_namesz + 3)/4;
1097 memcpy(buf, data, note.n_descsz);
1098 buf += (note.n_descsz + 3)/4;
1100 return buf;
1103 static void final_note(u32 *buf)
1105 struct elf_note note;
1107 note.n_namesz = 0;
1108 note.n_descsz = 0;
1109 note.n_type = 0;
1110 memcpy(buf, &note, sizeof(note));
1113 void crash_save_cpu(struct pt_regs *regs, int cpu)
1115 struct elf_prstatus prstatus;
1116 u32 *buf;
1118 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1119 return;
1121 /* Using ELF notes here is opportunistic.
1122 * I need a well defined structure format
1123 * for the data I pass, and I need tags
1124 * on the data to indicate what information I have
1125 * squirrelled away. ELF notes happen to provide
1126 * all of that, so there is no need to invent something new.
1128 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1129 if (!buf)
1130 return;
1131 memset(&prstatus, 0, sizeof(prstatus));
1132 prstatus.pr_pid = current->pid;
1133 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1134 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1135 &prstatus, sizeof(prstatus));
1136 final_note(buf);
1139 static int __init crash_notes_memory_init(void)
1141 /* Allocate memory for saving cpu registers. */
1142 crash_notes = alloc_percpu(note_buf_t);
1143 if (!crash_notes) {
1144 printk("Kexec: Memory allocation for saving cpu register"
1145 " states failed\n");
1146 return -ENOMEM;
1148 return 0;
1150 module_init(crash_notes_memory_init)
1154 * parsing the "crashkernel" commandline
1156 * this code is intended to be called from architecture specific code
1161 * This function parses command lines in the format
1163 * crashkernel=ramsize-range:size[,...][@offset]
1165 * The function returns 0 on success and -EINVAL on failure.
1167 static int __init parse_crashkernel_mem(char *cmdline,
1168 unsigned long long system_ram,
1169 unsigned long long *crash_size,
1170 unsigned long long *crash_base)
1172 char *cur = cmdline, *tmp;
1174 /* for each entry of the comma-separated list */
1175 do {
1176 unsigned long long start, end = ULLONG_MAX, size;
1178 /* get the start of the range */
1179 start = memparse(cur, &tmp);
1180 if (cur == tmp) {
1181 pr_warning("crashkernel: Memory value expected\n");
1182 return -EINVAL;
1184 cur = tmp;
1185 if (*cur != '-') {
1186 pr_warning("crashkernel: '-' expected\n");
1187 return -EINVAL;
1189 cur++;
1191 /* if no ':' is here, than we read the end */
1192 if (*cur != ':') {
1193 end = memparse(cur, &tmp);
1194 if (cur == tmp) {
1195 pr_warning("crashkernel: Memory "
1196 "value expected\n");
1197 return -EINVAL;
1199 cur = tmp;
1200 if (end <= start) {
1201 pr_warning("crashkernel: end <= start\n");
1202 return -EINVAL;
1206 if (*cur != ':') {
1207 pr_warning("crashkernel: ':' expected\n");
1208 return -EINVAL;
1210 cur++;
1212 size = memparse(cur, &tmp);
1213 if (cur == tmp) {
1214 pr_warning("Memory value expected\n");
1215 return -EINVAL;
1217 cur = tmp;
1218 if (size >= system_ram) {
1219 pr_warning("crashkernel: invalid size\n");
1220 return -EINVAL;
1223 /* match ? */
1224 if (system_ram >= start && system_ram < end) {
1225 *crash_size = size;
1226 break;
1228 } while (*cur++ == ',');
1230 if (*crash_size > 0) {
1231 while (*cur && *cur != ' ' && *cur != '@')
1232 cur++;
1233 if (*cur == '@') {
1234 cur++;
1235 *crash_base = memparse(cur, &tmp);
1236 if (cur == tmp) {
1237 pr_warning("Memory value expected "
1238 "after '@'\n");
1239 return -EINVAL;
1244 return 0;
1248 * That function parses "simple" (old) crashkernel command lines like
1250 * crashkernel=size[@offset]
1252 * It returns 0 on success and -EINVAL on failure.
1254 static int __init parse_crashkernel_simple(char *cmdline,
1255 unsigned long long *crash_size,
1256 unsigned long long *crash_base)
1258 char *cur = cmdline;
1260 *crash_size = memparse(cmdline, &cur);
1261 if (cmdline == cur) {
1262 pr_warning("crashkernel: memory value expected\n");
1263 return -EINVAL;
1266 if (*cur == '@')
1267 *crash_base = memparse(cur+1, &cur);
1269 return 0;
1273 * That function is the entry point for command line parsing and should be
1274 * called from the arch-specific code.
1276 int __init parse_crashkernel(char *cmdline,
1277 unsigned long long system_ram,
1278 unsigned long long *crash_size,
1279 unsigned long long *crash_base)
1281 char *p = cmdline, *ck_cmdline = NULL;
1282 char *first_colon, *first_space;
1284 BUG_ON(!crash_size || !crash_base);
1285 *crash_size = 0;
1286 *crash_base = 0;
1288 /* find crashkernel and use the last one if there are more */
1289 p = strstr(p, "crashkernel=");
1290 while (p) {
1291 ck_cmdline = p;
1292 p = strstr(p+1, "crashkernel=");
1295 if (!ck_cmdline)
1296 return -EINVAL;
1298 ck_cmdline += 12; /* strlen("crashkernel=") */
1301 * if the commandline contains a ':', then that's the extended
1302 * syntax -- if not, it must be the classic syntax
1304 first_colon = strchr(ck_cmdline, ':');
1305 first_space = strchr(ck_cmdline, ' ');
1306 if (first_colon && (!first_space || first_colon < first_space))
1307 return parse_crashkernel_mem(ck_cmdline, system_ram,
1308 crash_size, crash_base);
1309 else
1310 return parse_crashkernel_simple(ck_cmdline, crash_size,
1311 crash_base);
1313 return 0;
1318 void crash_save_vmcoreinfo(void)
1320 u32 *buf;
1322 if (!vmcoreinfo_size)
1323 return;
1325 vmcoreinfo_append_str("CRASHTIME=%ld", get_seconds());
1327 buf = (u32 *)vmcoreinfo_note;
1329 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1330 vmcoreinfo_size);
1332 final_note(buf);
1335 void vmcoreinfo_append_str(const char *fmt, ...)
1337 va_list args;
1338 char buf[0x50];
1339 int r;
1341 va_start(args, fmt);
1342 r = vsnprintf(buf, sizeof(buf), fmt, args);
1343 va_end(args);
1345 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1346 r = vmcoreinfo_max_size - vmcoreinfo_size;
1348 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1350 vmcoreinfo_size += r;
1354 * provide an empty default implementation here -- architecture
1355 * code may override this
1357 void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1360 unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1362 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1365 static int __init crash_save_vmcoreinfo_init(void)
1367 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1368 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1370 VMCOREINFO_SYMBOL(init_uts_ns);
1371 VMCOREINFO_SYMBOL(node_online_map);
1372 VMCOREINFO_SYMBOL(swapper_pg_dir);
1373 VMCOREINFO_SYMBOL(_stext);
1374 VMCOREINFO_SYMBOL(vmlist);
1376 #ifndef CONFIG_NEED_MULTIPLE_NODES
1377 VMCOREINFO_SYMBOL(mem_map);
1378 VMCOREINFO_SYMBOL(contig_page_data);
1379 #endif
1380 #ifdef CONFIG_SPARSEMEM
1381 VMCOREINFO_SYMBOL(mem_section);
1382 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1383 VMCOREINFO_STRUCT_SIZE(mem_section);
1384 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1385 #endif
1386 VMCOREINFO_STRUCT_SIZE(page);
1387 VMCOREINFO_STRUCT_SIZE(pglist_data);
1388 VMCOREINFO_STRUCT_SIZE(zone);
1389 VMCOREINFO_STRUCT_SIZE(free_area);
1390 VMCOREINFO_STRUCT_SIZE(list_head);
1391 VMCOREINFO_SIZE(nodemask_t);
1392 VMCOREINFO_OFFSET(page, flags);
1393 VMCOREINFO_OFFSET(page, _count);
1394 VMCOREINFO_OFFSET(page, mapping);
1395 VMCOREINFO_OFFSET(page, lru);
1396 VMCOREINFO_OFFSET(pglist_data, node_zones);
1397 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1398 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1399 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1400 #endif
1401 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1402 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1403 VMCOREINFO_OFFSET(pglist_data, node_id);
1404 VMCOREINFO_OFFSET(zone, free_area);
1405 VMCOREINFO_OFFSET(zone, vm_stat);
1406 VMCOREINFO_OFFSET(zone, spanned_pages);
1407 VMCOREINFO_OFFSET(free_area, free_list);
1408 VMCOREINFO_OFFSET(list_head, next);
1409 VMCOREINFO_OFFSET(list_head, prev);
1410 VMCOREINFO_OFFSET(vm_struct, addr);
1411 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1412 log_buf_kexec_setup();
1413 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1414 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1415 VMCOREINFO_NUMBER(PG_lru);
1416 VMCOREINFO_NUMBER(PG_private);
1417 VMCOREINFO_NUMBER(PG_swapcache);
1419 arch_crash_save_vmcoreinfo();
1421 return 0;
1424 module_init(crash_save_vmcoreinfo_init)
1427 * Move into place and start executing a preloaded standalone
1428 * executable. If nothing was preloaded return an error.
1430 int kernel_kexec(void)
1432 int error = 0;
1434 if (!mutex_trylock(&kexec_mutex))
1435 return -EBUSY;
1436 if (!kexec_image) {
1437 error = -EINVAL;
1438 goto Unlock;
1441 #ifdef CONFIG_KEXEC_JUMP
1442 if (kexec_image->preserve_context) {
1443 mutex_lock(&pm_mutex);
1444 pm_prepare_console();
1445 error = freeze_processes();
1446 if (error) {
1447 error = -EBUSY;
1448 goto Restore_console;
1450 suspend_console();
1451 error = dpm_suspend_start(PMSG_FREEZE);
1452 if (error)
1453 goto Resume_console;
1454 /* At this point, dpm_suspend_start() has been called,
1455 * but *not* dpm_suspend_noirq(). We *must* call
1456 * dpm_suspend_noirq() now. Otherwise, drivers for
1457 * some devices (e.g. interrupt controllers) become
1458 * desynchronized with the actual state of the
1459 * hardware at resume time, and evil weirdness ensues.
1461 error = dpm_suspend_noirq(PMSG_FREEZE);
1462 if (error)
1463 goto Resume_devices;
1464 error = disable_nonboot_cpus();
1465 if (error)
1466 goto Enable_cpus;
1467 local_irq_disable();
1468 /* Suspend system devices */
1469 error = sysdev_suspend(PMSG_FREEZE);
1470 if (error)
1471 goto Enable_irqs;
1472 } else
1473 #endif
1475 kernel_restart_prepare(NULL);
1476 printk(KERN_EMERG "Starting new kernel\n");
1477 machine_shutdown();
1480 machine_kexec(kexec_image);
1482 #ifdef CONFIG_KEXEC_JUMP
1483 if (kexec_image->preserve_context) {
1484 sysdev_resume();
1485 Enable_irqs:
1486 local_irq_enable();
1487 Enable_cpus:
1488 enable_nonboot_cpus();
1489 dpm_resume_noirq(PMSG_RESTORE);
1490 Resume_devices:
1491 dpm_resume_end(PMSG_RESTORE);
1492 Resume_console:
1493 resume_console();
1494 thaw_processes();
1495 Restore_console:
1496 pm_restore_console();
1497 mutex_unlock(&pm_mutex);
1499 #endif
1501 Unlock:
1502 mutex_unlock(&kexec_mutex);
1503 return error;