projects/linux·blob v3.7 · refs

1601 lines · 39.6 KiB

1/*
2 * kexec.c - kexec system call
3 * Copyright (C) 2002-2004 Eric Biederman <[email protected]>
4 *
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
7 */
8
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/utsname.h>
25#include <linux/numa.h>
26#include <linux/suspend.h>
27#include <linux/device.h>
28#include <linux/freezer.h>
29#include <linux/pm.h>
30#include <linux/cpu.h>
31#include <linux/console.h>
32#include <linux/vmalloc.h>
33#include <linux/swap.h>
34#include <linux/syscore_ops.h>
35
36#include <asm/page.h>
37#include <asm/uaccess.h>
38#include <asm/io.h>
39#include <asm/sections.h>
40
41/* Per cpu memory for storing cpu states in case of system crash. */
42note_buf_t __percpu *crash_notes;
43
44/* vmcoreinfo stuff */
45static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
46u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
47size_t vmcoreinfo_size;
48size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
49
50/* Location of the reserved area for the crash kernel */
51struct resource crashk_res = {
52 .name = "Crash kernel",
53 .start = 0,
54 .end = 0,
55 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
56};
57
58int kexec_should_crash(struct task_struct *p)
59{
60 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
61 return 1;
62 return 0;
63}
64
65/*
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.
70 *
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>.
78 *
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.
87 *
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.
93 *
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.
98 *
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.
103 */
104
105/*
106 * KIMAGE_NO_DEST is an impossible destination address..., for
107 * allocating pages whose destination address we do not care about.
108 */
109#define KIMAGE_NO_DEST (-1UL)
110
111static int kimage_is_destination_range(struct kimage *image,
112 unsigned long start, unsigned long end);
113static struct page *kimage_alloc_page(struct kimage *image,
114 gfp_t gfp_mask,
115 unsigned long dest);
116
117static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
118 unsigned long nr_segments,
119 struct kexec_segment __user *segments)
120{
121 size_t segment_bytes;
122 struct kimage *image;
123 unsigned long i;
124 int result;
125
126 /* Allocate a controlling structure */
127 result = -ENOMEM;
128 image = kzalloc(sizeof(*image), GFP_KERNEL);
129 if (!image)
130 goto out;
131
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;
138
139 /* Initialize the list of control pages */
140 INIT_LIST_HEAD(&image->control_pages);
141
142 /* Initialize the list of destination pages */
143 INIT_LIST_HEAD(&image->dest_pages);
144
145 /* Initialize the list of unusable pages */
146 INIT_LIST_HEAD(&image->unuseable_pages);
147
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 result = -EFAULT;
154 goto out;
155 }
156
157 /*
158 * Verify we have good destination addresses. The caller is
159 * responsible for making certain we don't attempt to load
160 * the new image into invalid or reserved areas of RAM. This
161 * just verifies it is an address we can use.
162 *
163 * Since the kernel does everything in page size chunks ensure
164 * the destination addresses are page aligned. Too many
165 * special cases crop of when we don't do this. The most
166 * insidious is getting overlapping destination addresses
167 * simply because addresses are changed to page size
168 * granularity.
169 */
170 result = -EADDRNOTAVAIL;
171 for (i = 0; i < nr_segments; i++) {
172 unsigned long mstart, mend;
173
174 mstart = image->segment[i].mem;
175 mend = mstart + image->segment[i].memsz;
176 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
177 goto out;
178 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
179 goto out;
180 }
181
182 /* Verify our destination addresses do not overlap.
183 * If we alloed overlapping destination addresses
184 * through very weird things can happen with no
185 * easy explanation as one segment stops on another.
186 */
187 result = -EINVAL;
188 for (i = 0; i < nr_segments; i++) {
189 unsigned long mstart, mend;
190 unsigned long j;
191
192 mstart = image->segment[i].mem;
193 mend = mstart + image->segment[i].memsz;
194 for (j = 0; j < i; j++) {
195 unsigned long pstart, pend;
196 pstart = image->segment[j].mem;
197 pend = pstart + image->segment[j].memsz;
198 /* Do the segments overlap ? */
199 if ((mend > pstart) && (mstart < pend))
200 goto out;
201 }
202 }
203
204 /* Ensure our buffer sizes are strictly less than
205 * our memory sizes. This should always be the case,
206 * and it is easier to check up front than to be surprised
207 * later on.
208 */
209 result = -EINVAL;
210 for (i = 0; i < nr_segments; i++) {
211 if (image->segment[i].bufsz > image->segment[i].memsz)
212 goto out;
213 }
214
215 result = 0;
216out:
217 if (result == 0)
218 *rimage = image;
219 else
220 kfree(image);
221
222 return result;
223
224}
225
226static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
227 unsigned long nr_segments,
228 struct kexec_segment __user *segments)
229{
230 int result;
231 struct kimage *image;
232
233 /* Allocate and initialize a controlling structure */
234 image = NULL;
235 result = do_kimage_alloc(&image, entry, nr_segments, segments);
236 if (result)
237 goto out;
238
239 *rimage = image;
240
241 /*
242 * Find a location for the control code buffer, and add it
243 * the vector of segments so that it's pages will also be
244 * counted as destination pages.
245 */
246 result = -ENOMEM;
247 image->control_code_page = kimage_alloc_control_pages(image,
248 get_order(KEXEC_CONTROL_PAGE_SIZE));
249 if (!image->control_code_page) {
250 printk(KERN_ERR "Could not allocate control_code_buffer\n");
251 goto out;
252 }
253
254 image->swap_page = kimage_alloc_control_pages(image, 0);
255 if (!image->swap_page) {
256 printk(KERN_ERR "Could not allocate swap buffer\n");
257 goto out;
258 }
259
260 result = 0;
261 out:
262 if (result == 0)
263 *rimage = image;
264 else
265 kfree(image);
266
267 return result;
268}
269
270static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
271 unsigned long nr_segments,
272 struct kexec_segment __user *segments)
273{
274 int result;
275 struct kimage *image;
276 unsigned long i;
277
278 image = NULL;
279 /* Verify we have a valid entry point */
280 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
281 result = -EADDRNOTAVAIL;
282 goto out;
283 }
284
285 /* Allocate and initialize a controlling structure */
286 result = do_kimage_alloc(&image, entry, nr_segments, segments);
287 if (result)
288 goto out;
289
290 /* Enable the special crash kernel control page
291 * allocation policy.
292 */
293 image->control_page = crashk_res.start;
294 image->type = KEXEC_TYPE_CRASH;
295
296 /*
297 * Verify we have good destination addresses. Normally
298 * the caller is responsible for making certain we don't
299 * attempt to load the new image into invalid or reserved
300 * areas of RAM. But crash kernels are preloaded into a
301 * reserved area of ram. We must ensure the addresses
302 * are in the reserved area otherwise preloading the
303 * kernel could corrupt things.
304 */
305 result = -EADDRNOTAVAIL;
306 for (i = 0; i < nr_segments; i++) {
307 unsigned long mstart, mend;
308
309 mstart = image->segment[i].mem;
310 mend = mstart + image->segment[i].memsz - 1;
311 /* Ensure we are within the crash kernel limits */
312 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
313 goto out;
314 }
315
316 /*
317 * Find a location for the control code buffer, and add
318 * the vector of segments so that it's pages will also be
319 * counted as destination pages.
320 */
321 result = -ENOMEM;
322 image->control_code_page = kimage_alloc_control_pages(image,
323 get_order(KEXEC_CONTROL_PAGE_SIZE));
324 if (!image->control_code_page) {
325 printk(KERN_ERR "Could not allocate control_code_buffer\n");
326 goto out;
327 }
328
329 result = 0;
330out:
331 if (result == 0)
332 *rimage = image;
333 else
334 kfree(image);
335
336 return result;
337}
338
339static int kimage_is_destination_range(struct kimage *image,
340 unsigned long start,
341 unsigned long end)
342{
343 unsigned long i;
344
345 for (i = 0; i < image->nr_segments; i++) {
346 unsigned long mstart, mend;
347
348 mstart = image->segment[i].mem;
349 mend = mstart + image->segment[i].memsz;
350 if ((end > mstart) && (start < mend))
351 return 1;
352 }
353
354 return 0;
355}
356
357static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
358{
359 struct page *pages;
360
361 pages = alloc_pages(gfp_mask, order);
362 if (pages) {
363 unsigned int count, i;
364 pages->mapping = NULL;
365 set_page_private(pages, order);
366 count = 1 << order;
367 for (i = 0; i < count; i++)
368 SetPageReserved(pages + i);
369 }
370
371 return pages;
372}
373
374static void kimage_free_pages(struct page *page)
375{
376 unsigned int order, count, i;
377
378 order = page_private(page);
379 count = 1 << order;
380 for (i = 0; i < count; i++)
381 ClearPageReserved(page + i);
382 __free_pages(page, order);
383}
384
385static void kimage_free_page_list(struct list_head *list)
386{
387 struct list_head *pos, *next;
388
389 list_for_each_safe(pos, next, list) {
390 struct page *page;
391
392 page = list_entry(pos, struct page, lru);
393 list_del(&page->lru);
394 kimage_free_pages(page);
395 }
396}
397
398static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
399 unsigned int order)
400{
401 /* Control pages are special, they are the intermediaries
402 * that are needed while we copy the rest of the pages
403 * to their final resting place. As such they must
404 * not conflict with either the destination addresses
405 * or memory the kernel is already using.
406 *
407 * The only case where we really need more than one of
408 * these are for architectures where we cannot disable
409 * the MMU and must instead generate an identity mapped
410 * page table for all of the memory.
411 *
412 * At worst this runs in O(N) of the image size.
413 */
414 struct list_head extra_pages;
415 struct page *pages;
416 unsigned int count;
417
418 count = 1 << order;
419 INIT_LIST_HEAD(&extra_pages);
420
421 /* Loop while I can allocate a page and the page allocated
422 * is a destination page.
423 */
424 do {
425 unsigned long pfn, epfn, addr, eaddr;
426
427 pages = kimage_alloc_pages(GFP_KERNEL, order);
428 if (!pages)
429 break;
430 pfn = page_to_pfn(pages);
431 epfn = pfn + count;
432 addr = pfn << PAGE_SHIFT;
433 eaddr = epfn << PAGE_SHIFT;
434 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
435 kimage_is_destination_range(image, addr, eaddr)) {
436 list_add(&pages->lru, &extra_pages);
437 pages = NULL;
438 }
439 } while (!pages);
440
441 if (pages) {
442 /* Remember the allocated page... */
443 list_add(&pages->lru, &image->control_pages);
444
445 /* Because the page is already in it's destination
446 * location we will never allocate another page at
447 * that address. Therefore kimage_alloc_pages
448 * will not return it (again) and we don't need
449 * to give it an entry in image->segment[].
450 */
451 }
452 /* Deal with the destination pages I have inadvertently allocated.
453 *
454 * Ideally I would convert multi-page allocations into single
455 * page allocations, and add everything to image->dest_pages.
456 *
457 * For now it is simpler to just free the pages.
458 */
459 kimage_free_page_list(&extra_pages);
460
461 return pages;
462}
463
464static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
465 unsigned int order)
466{
467 /* Control pages are special, they are the intermediaries
468 * that are needed while we copy the rest of the pages
469 * to their final resting place. As such they must
470 * not conflict with either the destination addresses
471 * or memory the kernel is already using.
472 *
473 * Control pages are also the only pags we must allocate
474 * when loading a crash kernel. All of the other pages
475 * are specified by the segments and we just memcpy
476 * into them directly.
477 *
478 * The only case where we really need more than one of
479 * these are for architectures where we cannot disable
480 * the MMU and must instead generate an identity mapped
481 * page table for all of the memory.
482 *
483 * Given the low demand this implements a very simple
484 * allocator that finds the first hole of the appropriate
485 * size in the reserved memory region, and allocates all
486 * of the memory up to and including the hole.
487 */
488 unsigned long hole_start, hole_end, size;
489 struct page *pages;
490
491 pages = NULL;
492 size = (1 << order) << PAGE_SHIFT;
493 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
494 hole_end = hole_start + size - 1;
495 while (hole_end <= crashk_res.end) {
496 unsigned long i;
497
498 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
499 break;
500 if (hole_end > crashk_res.end)
501 break;
502 /* See if I overlap any of the segments */
503 for (i = 0; i < image->nr_segments; i++) {
504 unsigned long mstart, mend;
505
506 mstart = image->segment[i].mem;
507 mend = mstart + image->segment[i].memsz - 1;
508 if ((hole_end >= mstart) && (hole_start <= mend)) {
509 /* Advance the hole to the end of the segment */
510 hole_start = (mend + (size - 1)) & ~(size - 1);
511 hole_end = hole_start + size - 1;
512 break;
513 }
514 }
515 /* If I don't overlap any segments I have found my hole! */
516 if (i == image->nr_segments) {
517 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
518 break;
519 }
520 }
521 if (pages)
522 image->control_page = hole_end;
523
524 return pages;
525}
526
527
528struct page *kimage_alloc_control_pages(struct kimage *image,
529 unsigned int order)
530{
531 struct page *pages = NULL;
532
533 switch (image->type) {
534 case KEXEC_TYPE_DEFAULT:
535 pages = kimage_alloc_normal_control_pages(image, order);
536 break;
537 case KEXEC_TYPE_CRASH:
538 pages = kimage_alloc_crash_control_pages(image, order);
539 break;
540 }
541
542 return pages;
543}
544
545static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
546{
547 if (*image->entry != 0)
548 image->entry++;
549
550 if (image->entry == image->last_entry) {
551 kimage_entry_t *ind_page;
552 struct page *page;
553
554 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
555 if (!page)
556 return -ENOMEM;
557
558 ind_page = page_address(page);
559 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
560 image->entry = ind_page;
561 image->last_entry = ind_page +
562 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
563 }
564 *image->entry = entry;
565 image->entry++;
566 *image->entry = 0;
567
568 return 0;
569}
570
571static int kimage_set_destination(struct kimage *image,
572 unsigned long destination)
573{
574 int result;
575
576 destination &= PAGE_MASK;
577 result = kimage_add_entry(image, destination | IND_DESTINATION);
578 if (result == 0)
579 image->destination = destination;
580
581 return result;
582}
583
584
585static int kimage_add_page(struct kimage *image, unsigned long page)
586{
587 int result;
588
589 page &= PAGE_MASK;
590 result = kimage_add_entry(image, page | IND_SOURCE);
591 if (result == 0)
592 image->destination += PAGE_SIZE;
593
594 return result;
595}
596
597
598static void kimage_free_extra_pages(struct kimage *image)
599{
600 /* Walk through and free any extra destination pages I may have */
601 kimage_free_page_list(&image->dest_pages);
602
603 /* Walk through and free any unusable pages I have cached */
604 kimage_free_page_list(&image->unuseable_pages);
605
606}
607static void kimage_terminate(struct kimage *image)
608{
609 if (*image->entry != 0)
610 image->entry++;
611
612 *image->entry = IND_DONE;
613}
614
615#define for_each_kimage_entry(image, ptr, entry) \
616 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
617 ptr = (entry & IND_INDIRECTION)? \
618 phys_to_virt((entry & PAGE_MASK)): ptr +1)
619
620static void kimage_free_entry(kimage_entry_t entry)
621{
622 struct page *page;
623
624 page = pfn_to_page(entry >> PAGE_SHIFT);
625 kimage_free_pages(page);
626}
627
628static void kimage_free(struct kimage *image)
629{
630 kimage_entry_t *ptr, entry;
631 kimage_entry_t ind = 0;
632
633 if (!image)
634 return;
635
636 kimage_free_extra_pages(image);
637 for_each_kimage_entry(image, ptr, entry) {
638 if (entry & IND_INDIRECTION) {
639 /* Free the previous indirection page */
640 if (ind & IND_INDIRECTION)
641 kimage_free_entry(ind);
642 /* Save this indirection page until we are
643 * done with it.
644 */
645 ind = entry;
646 }
647 else if (entry & IND_SOURCE)
648 kimage_free_entry(entry);
649 }
650 /* Free the final indirection page */
651 if (ind & IND_INDIRECTION)
652 kimage_free_entry(ind);
653
654 /* Handle any machine specific cleanup */
655 machine_kexec_cleanup(image);
656
657 /* Free the kexec control pages... */
658 kimage_free_page_list(&image->control_pages);
659 kfree(image);
660}
661
662static kimage_entry_t *kimage_dst_used(struct kimage *image,
663 unsigned long page)
664{
665 kimage_entry_t *ptr, entry;
666 unsigned long destination = 0;
667
668 for_each_kimage_entry(image, ptr, entry) {
669 if (entry & IND_DESTINATION)
670 destination = entry & PAGE_MASK;
671 else if (entry & IND_SOURCE) {
672 if (page == destination)
673 return ptr;
674 destination += PAGE_SIZE;
675 }
676 }
677
678 return NULL;
679}
680
681static struct page *kimage_alloc_page(struct kimage *image,
682 gfp_t gfp_mask,
683 unsigned long destination)
684{
685 /*
686 * Here we implement safeguards to ensure that a source page
687 * is not copied to its destination page before the data on
688 * the destination page is no longer useful.
689 *
690 * To do this we maintain the invariant that a source page is
691 * either its own destination page, or it is not a
692 * destination page at all.
693 *
694 * That is slightly stronger than required, but the proof
695 * that no problems will not occur is trivial, and the
696 * implementation is simply to verify.
697 *
698 * When allocating all pages normally this algorithm will run
699 * in O(N) time, but in the worst case it will run in O(N^2)
700 * time. If the runtime is a problem the data structures can
701 * be fixed.
702 */
703 struct page *page;
704 unsigned long addr;
705
706 /*
707 * Walk through the list of destination pages, and see if I
708 * have a match.
709 */
710 list_for_each_entry(page, &image->dest_pages, lru) {
711 addr = page_to_pfn(page) << PAGE_SHIFT;
712 if (addr == destination) {
713 list_del(&page->lru);
714 return page;
715 }
716 }
717 page = NULL;
718 while (1) {
719 kimage_entry_t *old;
720
721 /* Allocate a page, if we run out of memory give up */
722 page = kimage_alloc_pages(gfp_mask, 0);
723 if (!page)
724 return NULL;
725 /* If the page cannot be used file it away */
726 if (page_to_pfn(page) >
727 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
728 list_add(&page->lru, &image->unuseable_pages);
729 continue;
730 }
731 addr = page_to_pfn(page) << PAGE_SHIFT;
732
733 /* If it is the destination page we want use it */
734 if (addr == destination)
735 break;
736
737 /* If the page is not a destination page use it */
738 if (!kimage_is_destination_range(image, addr,
739 addr + PAGE_SIZE))
740 break;
741
742 /*
743 * I know that the page is someones destination page.
744 * See if there is already a source page for this
745 * destination page. And if so swap the source pages.
746 */
747 old = kimage_dst_used(image, addr);
748 if (old) {
749 /* If so move it */
750 unsigned long old_addr;
751 struct page *old_page;
752
753 old_addr = *old & PAGE_MASK;
754 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
755 copy_highpage(page, old_page);
756 *old = addr | (*old & ~PAGE_MASK);
757
758 /* The old page I have found cannot be a
759 * destination page, so return it if it's
760 * gfp_flags honor the ones passed in.
761 */
762 if (!(gfp_mask & __GFP_HIGHMEM) &&
763 PageHighMem(old_page)) {
764 kimage_free_pages(old_page);
765 continue;
766 }
767 addr = old_addr;
768 page = old_page;
769 break;
770 }
771 else {
772 /* Place the page on the destination list I
773 * will use it later.
774 */
775 list_add(&page->lru, &image->dest_pages);
776 }
777 }
778
779 return page;
780}
781
782static int kimage_load_normal_segment(struct kimage *image,
783 struct kexec_segment *segment)
784{
785 unsigned long maddr;
786 unsigned long ubytes, mbytes;
787 int result;
788 unsigned char __user *buf;
789
790 result = 0;
791 buf = segment->buf;
792 ubytes = segment->bufsz;
793 mbytes = segment->memsz;
794 maddr = segment->mem;
795
796 result = kimage_set_destination(image, maddr);
797 if (result < 0)
798 goto out;
799
800 while (mbytes) {
801 struct page *page;
802 char *ptr;
803 size_t uchunk, mchunk;
804
805 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
806 if (!page) {
807 result = -ENOMEM;
808 goto out;
809 }
810 result = kimage_add_page(image, page_to_pfn(page)
811 << PAGE_SHIFT);
812 if (result < 0)
813 goto out;
814
815 ptr = kmap(page);
816 /* Start with a clear page */
817 clear_page(ptr);
818 ptr += maddr & ~PAGE_MASK;
819 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
820 if (mchunk > mbytes)
821 mchunk = mbytes;
822
823 uchunk = mchunk;
824 if (uchunk > ubytes)
825 uchunk = ubytes;
826
827 result = copy_from_user(ptr, buf, uchunk);
828 kunmap(page);
829 if (result) {
830 result = -EFAULT;
831 goto out;
832 }
833 ubytes -= uchunk;
834 maddr += mchunk;
835 buf += mchunk;
836 mbytes -= mchunk;
837 }
838out:
839 return result;
840}
841
842static int kimage_load_crash_segment(struct kimage *image,
843 struct kexec_segment *segment)
844{
845 /* For crash dumps kernels we simply copy the data from
846 * user space to it's destination.
847 * We do things a page at a time for the sake of kmap.
848 */
849 unsigned long maddr;
850 unsigned long ubytes, mbytes;
851 int result;
852 unsigned char __user *buf;
853
854 result = 0;
855 buf = segment->buf;
856 ubytes = segment->bufsz;
857 mbytes = segment->memsz;
858 maddr = segment->mem;
859 while (mbytes) {
860 struct page *page;
861 char *ptr;
862 size_t uchunk, mchunk;
863
864 page = pfn_to_page(maddr >> PAGE_SHIFT);
865 if (!page) {
866 result = -ENOMEM;
867 goto out;
868 }
869 ptr = kmap(page);
870 ptr += maddr & ~PAGE_MASK;
871 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
872 if (mchunk > mbytes)
873 mchunk = mbytes;
874
875 uchunk = mchunk;
876 if (uchunk > ubytes) {
877 uchunk = ubytes;
878 /* Zero the trailing part of the page */
879 memset(ptr + uchunk, 0, mchunk - uchunk);
880 }
881 result = copy_from_user(ptr, buf, uchunk);
882 kexec_flush_icache_page(page);
883 kunmap(page);
884 if (result) {
885 result = -EFAULT;
886 goto out;
887 }
888 ubytes -= uchunk;
889 maddr += mchunk;
890 buf += mchunk;
891 mbytes -= mchunk;
892 }
893out:
894 return result;
895}
896
897static int kimage_load_segment(struct kimage *image,
898 struct kexec_segment *segment)
899{
900 int result = -ENOMEM;
901
902 switch (image->type) {
903 case KEXEC_TYPE_DEFAULT:
904 result = kimage_load_normal_segment(image, segment);
905 break;
906 case KEXEC_TYPE_CRASH:
907 result = kimage_load_crash_segment(image, segment);
908 break;
909 }
910
911 return result;
912}
913
914/*
915 * Exec Kernel system call: for obvious reasons only root may call it.
916 *
917 * This call breaks up into three pieces.
918 * - A generic part which loads the new kernel from the current
919 * address space, and very carefully places the data in the
920 * allocated pages.
921 *
922 * - A generic part that interacts with the kernel and tells all of
923 * the devices to shut down. Preventing on-going dmas, and placing
924 * the devices in a consistent state so a later kernel can
925 * reinitialize them.
926 *
927 * - A machine specific part that includes the syscall number
928 * and the copies the image to it's final destination. And
929 * jumps into the image at entry.
930 *
931 * kexec does not sync, or unmount filesystems so if you need
932 * that to happen you need to do that yourself.
933 */
934struct kimage *kexec_image;
935struct kimage *kexec_crash_image;
936
937static DEFINE_MUTEX(kexec_mutex);
938
939SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
940 struct kexec_segment __user *, segments, unsigned long, flags)
941{
942 struct kimage **dest_image, *image;
943 int result;
944
945 /* We only trust the superuser with rebooting the system. */
946 if (!capable(CAP_SYS_BOOT))
947 return -EPERM;
948
949 /*
950 * Verify we have a legal set of flags
951 * This leaves us room for future extensions.
952 */
953 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
954 return -EINVAL;
955
956 /* Verify we are on the appropriate architecture */
957 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
958 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
959 return -EINVAL;
960
961 /* Put an artificial cap on the number
962 * of segments passed to kexec_load.
963 */
964 if (nr_segments > KEXEC_SEGMENT_MAX)
965 return -EINVAL;
966
967 image = NULL;
968 result = 0;
969
970 /* Because we write directly to the reserved memory
971 * region when loading crash kernels we need a mutex here to
972 * prevent multiple crash kernels from attempting to load
973 * simultaneously, and to prevent a crash kernel from loading
974 * over the top of a in use crash kernel.
975 *
976 * KISS: always take the mutex.
977 */
978 if (!mutex_trylock(&kexec_mutex))
979 return -EBUSY;
980
981 dest_image = &kexec_image;
982 if (flags & KEXEC_ON_CRASH)
983 dest_image = &kexec_crash_image;
984 if (nr_segments > 0) {
985 unsigned long i;
986
987 /* Loading another kernel to reboot into */
988 if ((flags & KEXEC_ON_CRASH) == 0)
989 result = kimage_normal_alloc(&image, entry,
990 nr_segments, segments);
991 /* Loading another kernel to switch to if this one crashes */
992 else if (flags & KEXEC_ON_CRASH) {
993 /* Free any current crash dump kernel before
994 * we corrupt it.
995 */
996 kimage_free(xchg(&kexec_crash_image, NULL));
997 result = kimage_crash_alloc(&image, entry,
998 nr_segments, segments);
999 crash_map_reserved_pages();
1000 }
1001 if (result)
1002 goto out;
1003
1004 if (flags & KEXEC_PRESERVE_CONTEXT)
1005 image->preserve_context = 1;
1006 result = machine_kexec_prepare(image);
1007 if (result)
1008 goto out;
1009
1010 for (i = 0; i < nr_segments; i++) {
1011 result = kimage_load_segment(image, &image->segment[i]);
1012 if (result)
1013 goto out;
1014 }
1015 kimage_terminate(image);
1016 if (flags & KEXEC_ON_CRASH)
1017 crash_unmap_reserved_pages();
1018 }
1019 /* Install the new kernel, and Uninstall the old */
1020 image = xchg(dest_image, image);
1021
1022out:
1023 mutex_unlock(&kexec_mutex);
1024 kimage_free(image);
1025
1026 return result;
1027}
1028
1029/*
1030 * Add and remove page tables for crashkernel memory
1031 *
1032 * Provide an empty default implementation here -- architecture
1033 * code may override this
1034 */
1035void __weak crash_map_reserved_pages(void)
1036{}
1037
1038void __weak crash_unmap_reserved_pages(void)
1039{}
1040
1041#ifdef CONFIG_COMPAT
1042asmlinkage long compat_sys_kexec_load(unsigned long entry,
1043 unsigned long nr_segments,
1044 struct compat_kexec_segment __user *segments,
1045 unsigned long flags)
1046{
1047 struct compat_kexec_segment in;
1048 struct kexec_segment out, __user *ksegments;
1049 unsigned long i, result;
1050
1051 /* Don't allow clients that don't understand the native
1052 * architecture to do anything.
1053 */
1054 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1055 return -EINVAL;
1056
1057 if (nr_segments > KEXEC_SEGMENT_MAX)
1058 return -EINVAL;
1059
1060 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1061 for (i=0; i < nr_segments; i++) {
1062 result = copy_from_user(&in, &segments[i], sizeof(in));
1063 if (result)
1064 return -EFAULT;
1065
1066 out.buf = compat_ptr(in.buf);
1067 out.bufsz = in.bufsz;
1068 out.mem = in.mem;
1069 out.memsz = in.memsz;
1070
1071 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1072 if (result)
1073 return -EFAULT;
1074 }
1075
1076 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1077}
1078#endif
1079
1080void crash_kexec(struct pt_regs *regs)
1081{
1082 /* Take the kexec_mutex here to prevent sys_kexec_load
1083 * running on one cpu from replacing the crash kernel
1084 * we are using after a panic on a different cpu.
1085 *
1086 * If the crash kernel was not located in a fixed area
1087 * of memory the xchg(&kexec_crash_image) would be
1088 * sufficient. But since I reuse the memory...
1089 */
1090 if (mutex_trylock(&kexec_mutex)) {
1091 if (kexec_crash_image) {
1092 struct pt_regs fixed_regs;
1093
1094 crash_setup_regs(&fixed_regs, regs);
1095 crash_save_vmcoreinfo();
1096 machine_crash_shutdown(&fixed_regs);
1097 machine_kexec(kexec_crash_image);
1098 }
1099 mutex_unlock(&kexec_mutex);
1100 }
1101}
1102
1103size_t crash_get_memory_size(void)
1104{
1105 size_t size = 0;
1106 mutex_lock(&kexec_mutex);
1107 if (crashk_res.end != crashk_res.start)
1108 size = resource_size(&crashk_res);
1109 mutex_unlock(&kexec_mutex);
1110 return size;
1111}
1112
1113void __weak crash_free_reserved_phys_range(unsigned long begin,
1114 unsigned long end)
1115{
1116 unsigned long addr;
1117
1118 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1119 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1120 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1121 free_page((unsigned long)__va(addr));
1122 totalram_pages++;
1123 }
1124}
1125
1126int crash_shrink_memory(unsigned long new_size)
1127{
1128 int ret = 0;
1129 unsigned long start, end;
1130 unsigned long old_size;
1131 struct resource *ram_res;
1132
1133 mutex_lock(&kexec_mutex);
1134
1135 if (kexec_crash_image) {
1136 ret = -ENOENT;
1137 goto unlock;
1138 }
1139 start = crashk_res.start;
1140 end = crashk_res.end;
1141 old_size = (end == 0) ? 0 : end - start + 1;
1142 if (new_size >= old_size) {
1143 ret = (new_size == old_size) ? 0 : -EINVAL;
1144 goto unlock;
1145 }
1146
1147 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1148 if (!ram_res) {
1149 ret = -ENOMEM;
1150 goto unlock;
1151 }
1152
1153 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1154 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1155
1156 crash_map_reserved_pages();
1157 crash_free_reserved_phys_range(end, crashk_res.end);
1158
1159 if ((start == end) && (crashk_res.parent != NULL))
1160 release_resource(&crashk_res);
1161
1162 ram_res->start = end;
1163 ram_res->end = crashk_res.end;
1164 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1165 ram_res->name = "System RAM";
1166
1167 crashk_res.end = end - 1;
1168
1169 insert_resource(&iomem_resource, ram_res);
1170 crash_unmap_reserved_pages();
1171
1172unlock:
1173 mutex_unlock(&kexec_mutex);
1174 return ret;
1175}
1176
1177static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1178 size_t data_len)
1179{
1180 struct elf_note note;
1181
1182 note.n_namesz = strlen(name) + 1;
1183 note.n_descsz = data_len;
1184 note.n_type = type;
1185 memcpy(buf, &note, sizeof(note));
1186 buf += (sizeof(note) + 3)/4;
1187 memcpy(buf, name, note.n_namesz);
1188 buf += (note.n_namesz + 3)/4;
1189 memcpy(buf, data, note.n_descsz);
1190 buf += (note.n_descsz + 3)/4;
1191
1192 return buf;
1193}
1194
1195static void final_note(u32 *buf)
1196{
1197 struct elf_note note;
1198
1199 note.n_namesz = 0;
1200 note.n_descsz = 0;
1201 note.n_type = 0;
1202 memcpy(buf, &note, sizeof(note));
1203}
1204
1205void crash_save_cpu(struct pt_regs *regs, int cpu)
1206{
1207 struct elf_prstatus prstatus;
1208 u32 *buf;
1209
1210 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1211 return;
1212
1213 /* Using ELF notes here is opportunistic.
1214 * I need a well defined structure format
1215 * for the data I pass, and I need tags
1216 * on the data to indicate what information I have
1217 * squirrelled away. ELF notes happen to provide
1218 * all of that, so there is no need to invent something new.
1219 */
1220 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1221 if (!buf)
1222 return;
1223 memset(&prstatus, 0, sizeof(prstatus));
1224 prstatus.pr_pid = current->pid;
1225 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1226 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1227 &prstatus, sizeof(prstatus));
1228 final_note(buf);
1229}
1230
1231static int __init crash_notes_memory_init(void)
1232{
1233 /* Allocate memory for saving cpu registers. */
1234 crash_notes = alloc_percpu(note_buf_t);
1235 if (!crash_notes) {
1236 printk("Kexec: Memory allocation for saving cpu register"
1237 " states failed\n");
1238 return -ENOMEM;
1239 }
1240 return 0;
1241}
1242module_init(crash_notes_memory_init)
1243
1244
1245/*
1246 * parsing the "crashkernel" commandline
1247 *
1248 * this code is intended to be called from architecture specific code
1249 */
1250
1251
1252/*
1253 * This function parses command lines in the format
1254 *
1255 * crashkernel=ramsize-range:size[,...][@offset]
1256 *
1257 * The function returns 0 on success and -EINVAL on failure.
1258 */
1259static int __init parse_crashkernel_mem(char *cmdline,
1260 unsigned long long system_ram,
1261 unsigned long long *crash_size,
1262 unsigned long long *crash_base)
1263{
1264 char *cur = cmdline, *tmp;
1265
1266 /* for each entry of the comma-separated list */
1267 do {
1268 unsigned long long start, end = ULLONG_MAX, size;
1269
1270 /* get the start of the range */
1271 start = memparse(cur, &tmp);
1272 if (cur == tmp) {
1273 pr_warning("crashkernel: Memory value expected\n");
1274 return -EINVAL;
1275 }
1276 cur = tmp;
1277 if (*cur != '-') {
1278 pr_warning("crashkernel: '-' expected\n");
1279 return -EINVAL;
1280 }
1281 cur++;
1282
1283 /* if no ':' is here, than we read the end */
1284 if (*cur != ':') {
1285 end = memparse(cur, &tmp);
1286 if (cur == tmp) {
1287 pr_warning("crashkernel: Memory "
1288 "value expected\n");
1289 return -EINVAL;
1290 }
1291 cur = tmp;
1292 if (end <= start) {
1293 pr_warning("crashkernel: end <= start\n");
1294 return -EINVAL;
1295 }
1296 }
1297
1298 if (*cur != ':') {
1299 pr_warning("crashkernel: ':' expected\n");
1300 return -EINVAL;
1301 }
1302 cur++;
1303
1304 size = memparse(cur, &tmp);
1305 if (cur == tmp) {
1306 pr_warning("Memory value expected\n");
1307 return -EINVAL;
1308 }
1309 cur = tmp;
1310 if (size >= system_ram) {
1311 pr_warning("crashkernel: invalid size\n");
1312 return -EINVAL;
1313 }
1314
1315 /* match ? */
1316 if (system_ram >= start && system_ram < end) {
1317 *crash_size = size;
1318 break;
1319 }
1320 } while (*cur++ == ',');
1321
1322 if (*crash_size > 0) {
1323 while (*cur && *cur != ' ' && *cur != '@')
1324 cur++;
1325 if (*cur == '@') {
1326 cur++;
1327 *crash_base = memparse(cur, &tmp);
1328 if (cur == tmp) {
1329 pr_warning("Memory value expected "
1330 "after '@'\n");
1331 return -EINVAL;
1332 }
1333 }
1334 }
1335
1336 return 0;
1337}
1338
1339/*
1340 * That function parses "simple" (old) crashkernel command lines like
1341 *
1342 * crashkernel=size[@offset]
1343 *
1344 * It returns 0 on success and -EINVAL on failure.
1345 */
1346static int __init parse_crashkernel_simple(char *cmdline,
1347 unsigned long long *crash_size,
1348 unsigned long long *crash_base)
1349{
1350 char *cur = cmdline;
1351
1352 *crash_size = memparse(cmdline, &cur);
1353 if (cmdline == cur) {
1354 pr_warning("crashkernel: memory value expected\n");
1355 return -EINVAL;
1356 }
1357
1358 if (*cur == '@')
1359 *crash_base = memparse(cur+1, &cur);
1360 else if (*cur != ' ' && *cur != '\0') {
1361 pr_warning("crashkernel: unrecognized char\n");
1362 return -EINVAL;
1363 }
1364
1365 return 0;
1366}
1367
1368/*
1369 * That function is the entry point for command line parsing and should be
1370 * called from the arch-specific code.
1371 */
1372int __init parse_crashkernel(char *cmdline,
1373 unsigned long long system_ram,
1374 unsigned long long *crash_size,
1375 unsigned long long *crash_base)
1376{
1377 char *p = cmdline, *ck_cmdline = NULL;
1378 char *first_colon, *first_space;
1379
1380 BUG_ON(!crash_size || !crash_base);
1381 *crash_size = 0;
1382 *crash_base = 0;
1383
1384 /* find crashkernel and use the last one if there are more */
1385 p = strstr(p, "crashkernel=");
1386 while (p) {
1387 ck_cmdline = p;
1388 p = strstr(p+1, "crashkernel=");
1389 }
1390
1391 if (!ck_cmdline)
1392 return -EINVAL;
1393
1394 ck_cmdline += 12; /* strlen("crashkernel=") */
1395
1396 /*
1397 * if the commandline contains a ':', then that's the extended
1398 * syntax -- if not, it must be the classic syntax
1399 */
1400 first_colon = strchr(ck_cmdline, ':');
1401 first_space = strchr(ck_cmdline, ' ');
1402 if (first_colon && (!first_space || first_colon < first_space))
1403 return parse_crashkernel_mem(ck_cmdline, system_ram,
1404 crash_size, crash_base);
1405 else
1406 return parse_crashkernel_simple(ck_cmdline, crash_size,
1407 crash_base);
1408
1409 return 0;
1410}
1411
1412
1413static void update_vmcoreinfo_note(void)
1414{
1415 u32 *buf = vmcoreinfo_note;
1416
1417 if (!vmcoreinfo_size)
1418 return;
1419 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1420 vmcoreinfo_size);
1421 final_note(buf);
1422}
1423
1424void crash_save_vmcoreinfo(void)
1425{
1426 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1427 update_vmcoreinfo_note();
1428}
1429
1430void vmcoreinfo_append_str(const char *fmt, ...)
1431{
1432 va_list args;
1433 char buf[0x50];
1434 int r;
1435
1436 va_start(args, fmt);
1437 r = vsnprintf(buf, sizeof(buf), fmt, args);
1438 va_end(args);
1439
1440 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1441 r = vmcoreinfo_max_size - vmcoreinfo_size;
1442
1443 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1444
1445 vmcoreinfo_size += r;
1446}
1447
1448/*
1449 * provide an empty default implementation here -- architecture
1450 * code may override this
1451 */
1452void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1453{}
1454
1455unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1456{
1457 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1458}
1459
1460static int __init crash_save_vmcoreinfo_init(void)
1461{
1462 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1463 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1464
1465 VMCOREINFO_SYMBOL(init_uts_ns);
1466 VMCOREINFO_SYMBOL(node_online_map);
1467#ifdef CONFIG_MMU
1468 VMCOREINFO_SYMBOL(swapper_pg_dir);
1469#endif
1470 VMCOREINFO_SYMBOL(_stext);
1471 VMCOREINFO_SYMBOL(vmlist);
1472
1473#ifndef CONFIG_NEED_MULTIPLE_NODES
1474 VMCOREINFO_SYMBOL(mem_map);
1475 VMCOREINFO_SYMBOL(contig_page_data);
1476#endif
1477#ifdef CONFIG_SPARSEMEM
1478 VMCOREINFO_SYMBOL(mem_section);
1479 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1480 VMCOREINFO_STRUCT_SIZE(mem_section);
1481 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1482#endif
1483 VMCOREINFO_STRUCT_SIZE(page);
1484 VMCOREINFO_STRUCT_SIZE(pglist_data);
1485 VMCOREINFO_STRUCT_SIZE(zone);
1486 VMCOREINFO_STRUCT_SIZE(free_area);
1487 VMCOREINFO_STRUCT_SIZE(list_head);
1488 VMCOREINFO_SIZE(nodemask_t);
1489 VMCOREINFO_OFFSET(page, flags);
1490 VMCOREINFO_OFFSET(page, _count);
1491 VMCOREINFO_OFFSET(page, mapping);
1492 VMCOREINFO_OFFSET(page, lru);
1493 VMCOREINFO_OFFSET(pglist_data, node_zones);
1494 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1495#ifdef CONFIG_FLAT_NODE_MEM_MAP
1496 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1497#endif
1498 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1499 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1500 VMCOREINFO_OFFSET(pglist_data, node_id);
1501 VMCOREINFO_OFFSET(zone, free_area);
1502 VMCOREINFO_OFFSET(zone, vm_stat);
1503 VMCOREINFO_OFFSET(zone, spanned_pages);
1504 VMCOREINFO_OFFSET(free_area, free_list);
1505 VMCOREINFO_OFFSET(list_head, next);
1506 VMCOREINFO_OFFSET(list_head, prev);
1507 VMCOREINFO_OFFSET(vm_struct, addr);
1508 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1509 log_buf_kexec_setup();
1510 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1511 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1512 VMCOREINFO_NUMBER(PG_lru);
1513 VMCOREINFO_NUMBER(PG_private);
1514 VMCOREINFO_NUMBER(PG_swapcache);
1515
1516 arch_crash_save_vmcoreinfo();
1517 update_vmcoreinfo_note();
1518
1519 return 0;
1520}
1521
1522module_init(crash_save_vmcoreinfo_init)
1523
1524/*
1525 * Move into place and start executing a preloaded standalone
1526 * executable. If nothing was preloaded return an error.
1527 */
1528int kernel_kexec(void)
1529{
1530 int error = 0;
1531
1532 if (!mutex_trylock(&kexec_mutex))
1533 return -EBUSY;
1534 if (!kexec_image) {
1535 error = -EINVAL;
1536 goto Unlock;
1537 }
1538
1539#ifdef CONFIG_KEXEC_JUMP
1540 if (kexec_image->preserve_context) {
1541 lock_system_sleep();
1542 pm_prepare_console();
1543 error = freeze_processes();
1544 if (error) {
1545 error = -EBUSY;
1546 goto Restore_console;
1547 }
1548 suspend_console();
1549 error = dpm_suspend_start(PMSG_FREEZE);
1550 if (error)
1551 goto Resume_console;
1552 /* At this point, dpm_suspend_start() has been called,
1553 * but *not* dpm_suspend_end(). We *must* call
1554 * dpm_suspend_end() now. Otherwise, drivers for
1555 * some devices (e.g. interrupt controllers) become
1556 * desynchronized with the actual state of the
1557 * hardware at resume time, and evil weirdness ensues.
1558 */
1559 error = dpm_suspend_end(PMSG_FREEZE);
1560 if (error)
1561 goto Resume_devices;
1562 error = disable_nonboot_cpus();
1563 if (error)
1564 goto Enable_cpus;
1565 local_irq_disable();
1566 error = syscore_suspend();
1567 if (error)
1568 goto Enable_irqs;
1569 } else
1570#endif
1571 {
1572 kernel_restart_prepare(NULL);
1573 printk(KERN_EMERG "Starting new kernel\n");
1574 machine_shutdown();
1575 }
1576
1577 machine_kexec(kexec_image);
1578
1579#ifdef CONFIG_KEXEC_JUMP
1580 if (kexec_image->preserve_context) {
1581 syscore_resume();
1582 Enable_irqs:
1583 local_irq_enable();
1584 Enable_cpus:
1585 enable_nonboot_cpus();
1586 dpm_resume_start(PMSG_RESTORE);
1587 Resume_devices:
1588 dpm_resume_end(PMSG_RESTORE);
1589 Resume_console:
1590 resume_console();
1591 thaw_processes();
1592 Restore_console:
1593 pm_restore_console();
1594 unlock_system_sleep();
1595 }
1596#endif
1597
1598 Unlock:
1599 mutex_unlock(&kexec_mutex);
1600 return error;
1601}