ia64/xen-unstable

view linux-2.6-xen-sparse/kernel/kexec.c @ 14321:f421ccd1141f

Added section on references vs UUIDs.

Signed-off-by: Ewan Mellor <ewan@xensource.com>
author Ewan Mellor <ewan@xensource.com>
date Fri Mar 09 00:44:10 2007 +0000 (2007-03-09)
parents 8475a4e0425e
children a1daade92952
line source
1 /*
2 * kexec.c - kexec system call
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
4 *
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/spinlock.h>
16 #include <linux/list.h>
17 #include <linux/highmem.h>
18 #include <linux/syscalls.h>
19 #include <linux/reboot.h>
20 #include <linux/syscalls.h>
21 #include <linux/ioport.h>
22 #include <linux/hardirq.h>
24 #include <asm/page.h>
25 #include <asm/uaccess.h>
26 #include <asm/io.h>
27 #include <asm/system.h>
28 #include <asm/semaphore.h>
30 /* Per cpu memory for storing cpu states in case of system crash. */
31 note_buf_t* crash_notes;
33 /* Location of the reserved area for the crash kernel */
34 struct resource crashk_res = {
35 .name = "Crash kernel",
36 .start = 0,
37 .end = 0,
38 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
39 };
41 int kexec_should_crash(struct task_struct *p)
42 {
43 if (in_interrupt() || !p->pid || p->pid == 1 || panic_on_oops)
44 return 1;
45 return 0;
46 }
48 /*
49 * When kexec transitions to the new kernel there is a one-to-one
50 * mapping between physical and virtual addresses. On processors
51 * where you can disable the MMU this is trivial, and easy. For
52 * others it is still a simple predictable page table to setup.
53 *
54 * In that environment kexec copies the new kernel to its final
55 * resting place. This means I can only support memory whose
56 * physical address can fit in an unsigned long. In particular
57 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
58 * If the assembly stub has more restrictive requirements
59 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
60 * defined more restrictively in <asm/kexec.h>.
61 *
62 * The code for the transition from the current kernel to the
63 * the new kernel is placed in the control_code_buffer, whose size
64 * is given by KEXEC_CONTROL_CODE_SIZE. In the best case only a single
65 * page of memory is necessary, but some architectures require more.
66 * Because this memory must be identity mapped in the transition from
67 * virtual to physical addresses it must live in the range
68 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
69 * modifiable.
70 *
71 * The assembly stub in the control code buffer is passed a linked list
72 * of descriptor pages detailing the source pages of the new kernel,
73 * and the destination addresses of those source pages. As this data
74 * structure is not used in the context of the current OS, it must
75 * be self-contained.
76 *
77 * The code has been made to work with highmem pages and will use a
78 * destination page in its final resting place (if it happens
79 * to allocate it). The end product of this is that most of the
80 * physical address space, and most of RAM can be used.
81 *
82 * Future directions include:
83 * - allocating a page table with the control code buffer identity
84 * mapped, to simplify machine_kexec and make kexec_on_panic more
85 * reliable.
86 */
88 /*
89 * KIMAGE_NO_DEST is an impossible destination address..., for
90 * allocating pages whose destination address we do not care about.
91 */
92 #define KIMAGE_NO_DEST (-1UL)
94 static int kimage_is_destination_range(struct kimage *image,
95 unsigned long start, unsigned long end);
96 static struct page *kimage_alloc_page(struct kimage *image,
97 gfp_t gfp_mask,
98 unsigned long dest);
100 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
101 unsigned long nr_segments,
102 struct kexec_segment __user *segments)
103 {
104 size_t segment_bytes;
105 struct kimage *image;
106 unsigned long i;
107 int result;
109 /* Allocate a controlling structure */
110 result = -ENOMEM;
111 image = kmalloc(sizeof(*image), GFP_KERNEL);
112 if (!image)
113 goto out;
115 memset(image, 0, sizeof(*image));
116 image->head = 0;
117 image->entry = &image->head;
118 image->last_entry = &image->head;
119 image->control_page = ~0; /* By default this does not apply */
120 image->start = entry;
121 image->type = KEXEC_TYPE_DEFAULT;
123 /* Initialize the list of control pages */
124 INIT_LIST_HEAD(&image->control_pages);
126 /* Initialize the list of destination pages */
127 INIT_LIST_HEAD(&image->dest_pages);
129 /* Initialize the list of unuseable pages */
130 INIT_LIST_HEAD(&image->unuseable_pages);
132 /* Read in the segments */
133 image->nr_segments = nr_segments;
134 segment_bytes = nr_segments * sizeof(*segments);
135 result = copy_from_user(image->segment, segments, segment_bytes);
136 if (result)
137 goto out;
139 /*
140 * Verify we have good destination addresses. The caller is
141 * responsible for making certain we don't attempt to load
142 * the new image into invalid or reserved areas of RAM. This
143 * just verifies it is an address we can use.
144 *
145 * Since the kernel does everything in page size chunks ensure
146 * the destination addreses are page aligned. Too many
147 * special cases crop of when we don't do this. The most
148 * insidious is getting overlapping destination addresses
149 * simply because addresses are changed to page size
150 * granularity.
151 */
152 result = -EADDRNOTAVAIL;
153 for (i = 0; i < nr_segments; i++) {
154 unsigned long mstart, mend;
156 mstart = image->segment[i].mem;
157 mend = mstart + image->segment[i].memsz;
158 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
159 goto out;
160 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
161 goto out;
162 }
164 /* Verify our destination addresses do not overlap.
165 * If we alloed overlapping destination addresses
166 * through very weird things can happen with no
167 * easy explanation as one segment stops on another.
168 */
169 result = -EINVAL;
170 for (i = 0; i < nr_segments; i++) {
171 unsigned long mstart, mend;
172 unsigned long j;
174 mstart = image->segment[i].mem;
175 mend = mstart + image->segment[i].memsz;
176 for (j = 0; j < i; j++) {
177 unsigned long pstart, pend;
178 pstart = image->segment[j].mem;
179 pend = pstart + image->segment[j].memsz;
180 /* Do the segments overlap ? */
181 if ((mend > pstart) && (mstart < pend))
182 goto out;
183 }
184 }
186 /* Ensure our buffer sizes are strictly less than
187 * our memory sizes. This should always be the case,
188 * and it is easier to check up front than to be surprised
189 * later on.
190 */
191 result = -EINVAL;
192 for (i = 0; i < nr_segments; i++) {
193 if (image->segment[i].bufsz > image->segment[i].memsz)
194 goto out;
195 }
197 result = 0;
198 out:
199 if (result == 0)
200 *rimage = image;
201 else
202 kfree(image);
204 return result;
206 }
208 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
209 unsigned long nr_segments,
210 struct kexec_segment __user *segments)
211 {
212 int result;
213 struct kimage *image;
215 /* Allocate and initialize a controlling structure */
216 image = NULL;
217 result = do_kimage_alloc(&image, entry, nr_segments, segments);
218 if (result)
219 goto out;
221 *rimage = image;
223 /*
224 * Find a location for the control code buffer, and add it
225 * the vector of segments so that it's pages will also be
226 * counted as destination pages.
227 */
228 result = -ENOMEM;
229 image->control_code_page = kimage_alloc_control_pages(image,
230 get_order(KEXEC_CONTROL_CODE_SIZE));
231 if (!image->control_code_page) {
232 printk(KERN_ERR "Could not allocate control_code_buffer\n");
233 goto out;
234 }
236 result = 0;
237 out:
238 if (result == 0)
239 *rimage = image;
240 else
241 kfree(image);
243 return result;
244 }
246 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
247 unsigned long nr_segments,
248 struct kexec_segment __user *segments)
249 {
250 int result;
251 struct kimage *image;
252 unsigned long i;
254 image = NULL;
255 /* Verify we have a valid entry point */
256 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
257 result = -EADDRNOTAVAIL;
258 goto out;
259 }
261 /* Allocate and initialize a controlling structure */
262 result = do_kimage_alloc(&image, entry, nr_segments, segments);
263 if (result)
264 goto out;
266 /* Enable the special crash kernel control page
267 * allocation policy.
268 */
269 image->control_page = crashk_res.start;
270 image->type = KEXEC_TYPE_CRASH;
272 /*
273 * Verify we have good destination addresses. Normally
274 * the caller is responsible for making certain we don't
275 * attempt to load the new image into invalid or reserved
276 * areas of RAM. But crash kernels are preloaded into a
277 * reserved area of ram. We must ensure the addresses
278 * are in the reserved area otherwise preloading the
279 * kernel could corrupt things.
280 */
281 result = -EADDRNOTAVAIL;
282 for (i = 0; i < nr_segments; i++) {
283 unsigned long mstart, mend;
285 mstart = image->segment[i].mem;
286 mend = mstart + image->segment[i].memsz - 1;
287 /* Ensure we are within the crash kernel limits */
288 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
289 goto out;
290 }
292 /*
293 * Find a location for the control code buffer, and add
294 * the vector of segments so that it's pages will also be
295 * counted as destination pages.
296 */
297 result = -ENOMEM;
298 image->control_code_page = kimage_alloc_control_pages(image,
299 get_order(KEXEC_CONTROL_CODE_SIZE));
300 if (!image->control_code_page) {
301 printk(KERN_ERR "Could not allocate control_code_buffer\n");
302 goto out;
303 }
305 result = 0;
306 out:
307 if (result == 0)
308 *rimage = image;
309 else
310 kfree(image);
312 return result;
313 }
315 static int kimage_is_destination_range(struct kimage *image,
316 unsigned long start,
317 unsigned long end)
318 {
319 unsigned long i;
321 for (i = 0; i < image->nr_segments; i++) {
322 unsigned long mstart, mend;
324 mstart = image->segment[i].mem;
325 mend = mstart + image->segment[i].memsz;
326 if ((end > mstart) && (start < mend))
327 return 1;
328 }
330 return 0;
331 }
333 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
334 {
335 struct page *pages;
337 pages = alloc_pages(gfp_mask, order);
338 if (pages) {
339 unsigned int count, i;
340 pages->mapping = NULL;
341 set_page_private(pages, order);
342 count = 1 << order;
343 for (i = 0; i < count; i++)
344 SetPageReserved(pages + i);
345 }
347 return pages;
348 }
350 static void kimage_free_pages(struct page *page)
351 {
352 unsigned int order, count, i;
354 order = page_private(page);
355 count = 1 << order;
356 for (i = 0; i < count; i++)
357 ClearPageReserved(page + i);
358 __free_pages(page, order);
359 }
361 static void kimage_free_page_list(struct list_head *list)
362 {
363 struct list_head *pos, *next;
365 list_for_each_safe(pos, next, list) {
366 struct page *page;
368 page = list_entry(pos, struct page, lru);
369 list_del(&page->lru);
370 kimage_free_pages(page);
371 }
372 }
374 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
375 unsigned int order)
376 {
377 /* Control pages are special, they are the intermediaries
378 * that are needed while we copy the rest of the pages
379 * to their final resting place. As such they must
380 * not conflict with either the destination addresses
381 * or memory the kernel is already using.
382 *
383 * The only case where we really need more than one of
384 * these are for architectures where we cannot disable
385 * the MMU and must instead generate an identity mapped
386 * page table for all of the memory.
387 *
388 * At worst this runs in O(N) of the image size.
389 */
390 struct list_head extra_pages;
391 struct page *pages;
392 unsigned int count;
394 count = 1 << order;
395 INIT_LIST_HEAD(&extra_pages);
397 /* Loop while I can allocate a page and the page allocated
398 * is a destination page.
399 */
400 do {
401 unsigned long pfn, epfn, addr, eaddr;
403 pages = kimage_alloc_pages(GFP_KERNEL, order);
404 if (!pages)
405 break;
406 pfn = kexec_page_to_pfn(pages);
407 epfn = pfn + count;
408 addr = pfn << PAGE_SHIFT;
409 eaddr = epfn << PAGE_SHIFT;
410 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
411 kimage_is_destination_range(image, addr, eaddr)) {
412 list_add(&pages->lru, &extra_pages);
413 pages = NULL;
414 }
415 } while (!pages);
417 if (pages) {
418 /* Remember the allocated page... */
419 list_add(&pages->lru, &image->control_pages);
421 /* Because the page is already in it's destination
422 * location we will never allocate another page at
423 * that address. Therefore kimage_alloc_pages
424 * will not return it (again) and we don't need
425 * to give it an entry in image->segment[].
426 */
427 }
428 /* Deal with the destination pages I have inadvertently allocated.
429 *
430 * Ideally I would convert multi-page allocations into single
431 * page allocations, and add everyting to image->dest_pages.
432 *
433 * For now it is simpler to just free the pages.
434 */
435 kimage_free_page_list(&extra_pages);
437 return pages;
438 }
440 #ifndef CONFIG_XEN
441 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
442 unsigned int order)
443 {
444 /* Control pages are special, they are the intermediaries
445 * that are needed while we copy the rest of the pages
446 * to their final resting place. As such they must
447 * not conflict with either the destination addresses
448 * or memory the kernel is already using.
449 *
450 * Control pages are also the only pags we must allocate
451 * when loading a crash kernel. All of the other pages
452 * are specified by the segments and we just memcpy
453 * into them directly.
454 *
455 * The only case where we really need more than one of
456 * these are for architectures where we cannot disable
457 * the MMU and must instead generate an identity mapped
458 * page table for all of the memory.
459 *
460 * Given the low demand this implements a very simple
461 * allocator that finds the first hole of the appropriate
462 * size in the reserved memory region, and allocates all
463 * of the memory up to and including the hole.
464 */
465 unsigned long hole_start, hole_end, size;
466 struct page *pages;
468 pages = NULL;
469 size = (1 << order) << PAGE_SHIFT;
470 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
471 hole_end = hole_start + size - 1;
472 while (hole_end <= crashk_res.end) {
473 unsigned long i;
475 if (hole_end > KEXEC_CONTROL_MEMORY_LIMIT)
476 break;
477 if (hole_end > crashk_res.end)
478 break;
479 /* See if I overlap any of the segments */
480 for (i = 0; i < image->nr_segments; i++) {
481 unsigned long mstart, mend;
483 mstart = image->segment[i].mem;
484 mend = mstart + image->segment[i].memsz - 1;
485 if ((hole_end >= mstart) && (hole_start <= mend)) {
486 /* Advance the hole to the end of the segment */
487 hole_start = (mend + (size - 1)) & ~(size - 1);
488 hole_end = hole_start + size - 1;
489 break;
490 }
491 }
492 /* If I don't overlap any segments I have found my hole! */
493 if (i == image->nr_segments) {
494 pages = kexec_pfn_to_page(hole_start >> PAGE_SHIFT);
495 break;
496 }
497 }
498 if (pages)
499 image->control_page = hole_end;
501 return pages;
502 }
505 struct page *kimage_alloc_control_pages(struct kimage *image,
506 unsigned int order)
507 {
508 struct page *pages = NULL;
510 switch (image->type) {
511 case KEXEC_TYPE_DEFAULT:
512 pages = kimage_alloc_normal_control_pages(image, order);
513 break;
514 case KEXEC_TYPE_CRASH:
515 pages = kimage_alloc_crash_control_pages(image, order);
516 break;
517 }
519 return pages;
520 }
521 #else /* !CONFIG_XEN */
522 struct page *kimage_alloc_control_pages(struct kimage *image,
523 unsigned int order)
524 {
525 return kimage_alloc_normal_control_pages(image, order);
526 }
527 #endif
529 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
530 {
531 if (*image->entry != 0)
532 image->entry++;
534 if (image->entry == image->last_entry) {
535 kimage_entry_t *ind_page;
536 struct page *page;
538 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
539 if (!page)
540 return -ENOMEM;
542 ind_page = page_address(page);
543 *image->entry = kexec_virt_to_phys(ind_page) | IND_INDIRECTION;
544 image->entry = ind_page;
545 image->last_entry = ind_page +
546 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
547 }
548 *image->entry = entry;
549 image->entry++;
550 *image->entry = 0;
552 return 0;
553 }
555 static int kimage_set_destination(struct kimage *image,
556 unsigned long destination)
557 {
558 int result;
560 destination &= PAGE_MASK;
561 result = kimage_add_entry(image, destination | IND_DESTINATION);
562 if (result == 0)
563 image->destination = destination;
565 return result;
566 }
569 static int kimage_add_page(struct kimage *image, unsigned long page)
570 {
571 int result;
573 page &= PAGE_MASK;
574 result = kimage_add_entry(image, page | IND_SOURCE);
575 if (result == 0)
576 image->destination += PAGE_SIZE;
578 return result;
579 }
582 static void kimage_free_extra_pages(struct kimage *image)
583 {
584 /* Walk through and free any extra destination pages I may have */
585 kimage_free_page_list(&image->dest_pages);
587 /* Walk through and free any unuseable pages I have cached */
588 kimage_free_page_list(&image->unuseable_pages);
590 }
591 static int kimage_terminate(struct kimage *image)
592 {
593 if (*image->entry != 0)
594 image->entry++;
596 *image->entry = IND_DONE;
598 return 0;
599 }
601 #define for_each_kimage_entry(image, ptr, entry) \
602 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
603 ptr = (entry & IND_INDIRECTION)? \
604 kexec_phys_to_virt((entry & PAGE_MASK)): ptr +1)
606 static void kimage_free_entry(kimage_entry_t entry)
607 {
608 struct page *page;
610 page = kexec_pfn_to_page(entry >> PAGE_SHIFT);
611 kimage_free_pages(page);
612 }
614 static void kimage_free(struct kimage *image)
615 {
616 kimage_entry_t *ptr, entry;
617 kimage_entry_t ind = 0;
619 if (!image)
620 return;
622 #ifdef CONFIG_XEN
623 xen_machine_kexec_unload(image);
624 #endif
626 kimage_free_extra_pages(image);
627 for_each_kimage_entry(image, ptr, entry) {
628 if (entry & IND_INDIRECTION) {
629 /* Free the previous indirection page */
630 if (ind & IND_INDIRECTION)
631 kimage_free_entry(ind);
632 /* Save this indirection page until we are
633 * done with it.
634 */
635 ind = entry;
636 }
637 else if (entry & IND_SOURCE)
638 kimage_free_entry(entry);
639 }
640 /* Free the final indirection page */
641 if (ind & IND_INDIRECTION)
642 kimage_free_entry(ind);
644 /* Handle any machine specific cleanup */
645 machine_kexec_cleanup(image);
647 /* Free the kexec control pages... */
648 kimage_free_page_list(&image->control_pages);
649 kfree(image);
650 }
652 static kimage_entry_t *kimage_dst_used(struct kimage *image,
653 unsigned long page)
654 {
655 kimage_entry_t *ptr, entry;
656 unsigned long destination = 0;
658 for_each_kimage_entry(image, ptr, entry) {
659 if (entry & IND_DESTINATION)
660 destination = entry & PAGE_MASK;
661 else if (entry & IND_SOURCE) {
662 if (page == destination)
663 return ptr;
664 destination += PAGE_SIZE;
665 }
666 }
668 return NULL;
669 }
671 static struct page *kimage_alloc_page(struct kimage *image,
672 gfp_t gfp_mask,
673 unsigned long destination)
674 {
675 /*
676 * Here we implement safeguards to ensure that a source page
677 * is not copied to its destination page before the data on
678 * the destination page is no longer useful.
679 *
680 * To do this we maintain the invariant that a source page is
681 * either its own destination page, or it is not a
682 * destination page at all.
683 *
684 * That is slightly stronger than required, but the proof
685 * that no problems will not occur is trivial, and the
686 * implementation is simply to verify.
687 *
688 * When allocating all pages normally this algorithm will run
689 * in O(N) time, but in the worst case it will run in O(N^2)
690 * time. If the runtime is a problem the data structures can
691 * be fixed.
692 */
693 struct page *page;
694 unsigned long addr;
696 /*
697 * Walk through the list of destination pages, and see if I
698 * have a match.
699 */
700 list_for_each_entry(page, &image->dest_pages, lru) {
701 addr = kexec_page_to_pfn(page) << PAGE_SHIFT;
702 if (addr == destination) {
703 list_del(&page->lru);
704 return page;
705 }
706 }
707 page = NULL;
708 while (1) {
709 kimage_entry_t *old;
711 /* Allocate a page, if we run out of memory give up */
712 page = kimage_alloc_pages(gfp_mask, 0);
713 if (!page)
714 return NULL;
715 /* If the page cannot be used file it away */
716 if (kexec_page_to_pfn(page) >
717 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
718 list_add(&page->lru, &image->unuseable_pages);
719 continue;
720 }
721 addr = kexec_page_to_pfn(page) << PAGE_SHIFT;
723 /* If it is the destination page we want use it */
724 if (addr == destination)
725 break;
727 /* If the page is not a destination page use it */
728 if (!kimage_is_destination_range(image, addr,
729 addr + PAGE_SIZE))
730 break;
732 /*
733 * I know that the page is someones destination page.
734 * See if there is already a source page for this
735 * destination page. And if so swap the source pages.
736 */
737 old = kimage_dst_used(image, addr);
738 if (old) {
739 /* If so move it */
740 unsigned long old_addr;
741 struct page *old_page;
743 old_addr = *old & PAGE_MASK;
744 old_page = kexec_pfn_to_page(old_addr >> PAGE_SHIFT);
745 copy_highpage(page, old_page);
746 *old = addr | (*old & ~PAGE_MASK);
748 /* The old page I have found cannot be a
749 * destination page, so return it.
750 */
751 addr = old_addr;
752 page = old_page;
753 break;
754 }
755 else {
756 /* Place the page on the destination list I
757 * will use it later.
758 */
759 list_add(&page->lru, &image->dest_pages);
760 }
761 }
763 return page;
764 }
766 static int kimage_load_normal_segment(struct kimage *image,
767 struct kexec_segment *segment)
768 {
769 unsigned long maddr;
770 unsigned long ubytes, mbytes;
771 int result;
772 unsigned char __user *buf;
774 result = 0;
775 buf = segment->buf;
776 ubytes = segment->bufsz;
777 mbytes = segment->memsz;
778 maddr = segment->mem;
780 result = kimage_set_destination(image, maddr);
781 if (result < 0)
782 goto out;
784 while (mbytes) {
785 struct page *page;
786 char *ptr;
787 size_t uchunk, mchunk;
789 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
790 if (page == 0) {
791 result = -ENOMEM;
792 goto out;
793 }
794 result = kimage_add_page(image, kexec_page_to_pfn(page)
795 << PAGE_SHIFT);
796 if (result < 0)
797 goto out;
799 ptr = kmap(page);
800 /* Start with a clear page */
801 memset(ptr, 0, PAGE_SIZE);
802 ptr += maddr & ~PAGE_MASK;
803 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
804 if (mchunk > mbytes)
805 mchunk = mbytes;
807 uchunk = mchunk;
808 if (uchunk > ubytes)
809 uchunk = ubytes;
811 result = copy_from_user(ptr, buf, uchunk);
812 kunmap(page);
813 if (result) {
814 result = (result < 0) ? result : -EIO;
815 goto out;
816 }
817 ubytes -= uchunk;
818 maddr += mchunk;
819 buf += mchunk;
820 mbytes -= mchunk;
821 }
822 out:
823 return result;
824 }
826 #ifndef CONFIG_XEN
827 static int kimage_load_crash_segment(struct kimage *image,
828 struct kexec_segment *segment)
829 {
830 /* For crash dumps kernels we simply copy the data from
831 * user space to it's destination.
832 * We do things a page at a time for the sake of kmap.
833 */
834 unsigned long maddr;
835 unsigned long ubytes, mbytes;
836 int result;
837 unsigned char __user *buf;
839 result = 0;
840 buf = segment->buf;
841 ubytes = segment->bufsz;
842 mbytes = segment->memsz;
843 maddr = segment->mem;
844 while (mbytes) {
845 struct page *page;
846 char *ptr;
847 size_t uchunk, mchunk;
849 page = kexec_pfn_to_page(maddr >> PAGE_SHIFT);
850 if (page == 0) {
851 result = -ENOMEM;
852 goto out;
853 }
854 ptr = kmap(page);
855 ptr += maddr & ~PAGE_MASK;
856 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
857 if (mchunk > mbytes)
858 mchunk = mbytes;
860 uchunk = mchunk;
861 if (uchunk > ubytes) {
862 uchunk = ubytes;
863 /* Zero the trailing part of the page */
864 memset(ptr + uchunk, 0, mchunk - uchunk);
865 }
866 result = copy_from_user(ptr, buf, uchunk);
867 kunmap(page);
868 if (result) {
869 result = (result < 0) ? result : -EIO;
870 goto out;
871 }
872 ubytes -= uchunk;
873 maddr += mchunk;
874 buf += mchunk;
875 mbytes -= mchunk;
876 }
877 out:
878 return result;
879 }
881 static int kimage_load_segment(struct kimage *image,
882 struct kexec_segment *segment)
883 {
884 int result = -ENOMEM;
886 switch (image->type) {
887 case KEXEC_TYPE_DEFAULT:
888 result = kimage_load_normal_segment(image, segment);
889 break;
890 case KEXEC_TYPE_CRASH:
891 result = kimage_load_crash_segment(image, segment);
892 break;
893 }
895 return result;
896 }
897 #else /* CONFIG_XEN */
898 static int kimage_load_segment(struct kimage *image,
899 struct kexec_segment *segment)
900 {
901 return kimage_load_normal_segment(image, segment);
902 }
903 #endif
905 /*
906 * Exec Kernel system call: for obvious reasons only root may call it.
907 *
908 * This call breaks up into three pieces.
909 * - A generic part which loads the new kernel from the current
910 * address space, and very carefully places the data in the
911 * allocated pages.
912 *
913 * - A generic part that interacts with the kernel and tells all of
914 * the devices to shut down. Preventing on-going dmas, and placing
915 * the devices in a consistent state so a later kernel can
916 * reinitialize them.
917 *
918 * - A machine specific part that includes the syscall number
919 * and the copies the image to it's final destination. And
920 * jumps into the image at entry.
921 *
922 * kexec does not sync, or unmount filesystems so if you need
923 * that to happen you need to do that yourself.
924 */
925 struct kimage *kexec_image;
926 struct kimage *kexec_crash_image;
927 /*
928 * A home grown binary mutex.
929 * Nothing can wait so this mutex is safe to use
930 * in interrupt context :)
931 */
932 static int kexec_lock;
934 asmlinkage long sys_kexec_load(unsigned long entry, unsigned long nr_segments,
935 struct kexec_segment __user *segments,
936 unsigned long flags)
937 {
938 struct kimage **dest_image, *image;
939 int locked;
940 int result;
942 /* We only trust the superuser with rebooting the system. */
943 if (!capable(CAP_SYS_BOOT))
944 return -EPERM;
946 /*
947 * Verify we have a legal set of flags
948 * This leaves us room for future extensions.
949 */
950 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
951 return -EINVAL;
953 /* Verify we are on the appropriate architecture */
954 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
955 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
956 return -EINVAL;
958 /* Put an artificial cap on the number
959 * of segments passed to kexec_load.
960 */
961 if (nr_segments > KEXEC_SEGMENT_MAX)
962 return -EINVAL;
964 image = NULL;
965 result = 0;
967 /* Because we write directly to the reserved memory
968 * region when loading crash kernels we need a mutex here to
969 * prevent multiple crash kernels from attempting to load
970 * simultaneously, and to prevent a crash kernel from loading
971 * over the top of a in use crash kernel.
972 *
973 * KISS: always take the mutex.
974 */
975 locked = xchg(&kexec_lock, 1);
976 if (locked)
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.
993 */
994 kimage_free(xchg(&kexec_crash_image, NULL));
995 result = kimage_crash_alloc(&image, entry,
996 nr_segments, segments);
997 }
998 if (result)
999 goto out;
1001 result = machine_kexec_prepare(image);
1002 if (result)
1003 goto out;
1005 for (i = 0; i < nr_segments; i++) {
1006 result = kimage_load_segment(image, &image->segment[i]);
1007 if (result)
1008 goto out;
1010 result = kimage_terminate(image);
1011 if (result)
1012 goto out;
1014 #ifdef CONFIG_XEN
1015 if (image) {
1016 result = xen_machine_kexec_load(image);
1017 if (result)
1018 goto out;
1020 #endif
1021 /* Install the new kernel, and Uninstall the old */
1022 image = xchg(dest_image, image);
1024 out:
1025 xchg(&kexec_lock, 0); /* Release the mutex */
1026 kimage_free(image);
1028 return result;
1031 #ifdef CONFIG_COMPAT
1032 asmlinkage long compat_sys_kexec_load(unsigned long entry,
1033 unsigned long nr_segments,
1034 struct compat_kexec_segment __user *segments,
1035 unsigned long flags)
1037 struct compat_kexec_segment in;
1038 struct kexec_segment out, __user *ksegments;
1039 unsigned long i, result;
1041 /* Don't allow clients that don't understand the native
1042 * architecture to do anything.
1043 */
1044 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1045 return -EINVAL;
1047 if (nr_segments > KEXEC_SEGMENT_MAX)
1048 return -EINVAL;
1050 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1051 for (i=0; i < nr_segments; i++) {
1052 result = copy_from_user(&in, &segments[i], sizeof(in));
1053 if (result)
1054 return -EFAULT;
1056 out.buf = compat_ptr(in.buf);
1057 out.bufsz = in.bufsz;
1058 out.mem = in.mem;
1059 out.memsz = in.memsz;
1061 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1062 if (result)
1063 return -EFAULT;
1066 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1068 #endif
1070 void crash_kexec(struct pt_regs *regs)
1072 int locked;
1074 /* Take the kexec_lock here to prevent sys_kexec_load
1075 * running on one cpu from replacing the crash kernel
1076 * we are using after a panic on a different cpu.
1078 * If the crash kernel was not located in a fixed area
1079 * of memory the xchg(&kexec_crash_image) would be
1080 * sufficient. But since I reuse the memory...
1081 */
1082 locked = xchg(&kexec_lock, 1);
1083 if (!locked) {
1084 if (kexec_crash_image) {
1085 struct pt_regs fixed_regs;
1086 crash_setup_regs(&fixed_regs, regs);
1087 machine_crash_shutdown(&fixed_regs);
1088 machine_kexec(kexec_crash_image);
1090 xchg(&kexec_lock, 0);
1094 static int __init crash_notes_memory_init(void)
1096 /* Allocate memory for saving cpu registers. */
1097 crash_notes = alloc_percpu(note_buf_t);
1098 if (!crash_notes) {
1099 printk("Kexec: Memory allocation for saving cpu register"
1100 " states failed\n");
1101 return -ENOMEM;
1103 return 0;
1105 module_init(crash_notes_memory_init)