Loading...
1/*
2 * kexec.c - kexec system call core code.
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 */
8
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/capability.h>
12#include <linux/mm.h>
13#include <linux/file.h>
14#include <linux/slab.h>
15#include <linux/fs.h>
16#include <linux/kexec.h>
17#include <linux/mutex.h>
18#include <linux/list.h>
19#include <linux/highmem.h>
20#include <linux/syscalls.h>
21#include <linux/reboot.h>
22#include <linux/ioport.h>
23#include <linux/hardirq.h>
24#include <linux/elf.h>
25#include <linux/elfcore.h>
26#include <linux/utsname.h>
27#include <linux/numa.h>
28#include <linux/suspend.h>
29#include <linux/device.h>
30#include <linux/freezer.h>
31#include <linux/pm.h>
32#include <linux/cpu.h>
33#include <linux/uaccess.h>
34#include <linux/io.h>
35#include <linux/console.h>
36#include <linux/vmalloc.h>
37#include <linux/swap.h>
38#include <linux/syscore_ops.h>
39#include <linux/compiler.h>
40#include <linux/hugetlb.h>
41
42#include <asm/page.h>
43#include <asm/sections.h>
44
45#include <crypto/hash.h>
46#include <crypto/sha.h>
47#include "kexec_internal.h"
48
49DEFINE_MUTEX(kexec_mutex);
50
51/* Per cpu memory for storing cpu states in case of system crash. */
52note_buf_t __percpu *crash_notes;
53
54/* vmcoreinfo stuff */
55static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
56u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
57size_t vmcoreinfo_size;
58size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
59
60/* Flag to indicate we are going to kexec a new kernel */
61bool kexec_in_progress = false;
62
63
64/* Location of the reserved area for the crash kernel */
65struct resource crashk_res = {
66 .name = "Crash kernel",
67 .start = 0,
68 .end = 0,
69 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
70 .desc = IORES_DESC_CRASH_KERNEL
71};
72struct resource crashk_low_res = {
73 .name = "Crash kernel",
74 .start = 0,
75 .end = 0,
76 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
77 .desc = IORES_DESC_CRASH_KERNEL
78};
79
80int kexec_should_crash(struct task_struct *p)
81{
82 /*
83 * If crash_kexec_post_notifiers is enabled, don't run
84 * crash_kexec() here yet, which must be run after panic
85 * notifiers in panic().
86 */
87 if (crash_kexec_post_notifiers)
88 return 0;
89 /*
90 * There are 4 panic() calls in do_exit() path, each of which
91 * corresponds to each of these 4 conditions.
92 */
93 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
94 return 1;
95 return 0;
96}
97
98/*
99 * When kexec transitions to the new kernel there is a one-to-one
100 * mapping between physical and virtual addresses. On processors
101 * where you can disable the MMU this is trivial, and easy. For
102 * others it is still a simple predictable page table to setup.
103 *
104 * In that environment kexec copies the new kernel to its final
105 * resting place. This means I can only support memory whose
106 * physical address can fit in an unsigned long. In particular
107 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
108 * If the assembly stub has more restrictive requirements
109 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
110 * defined more restrictively in <asm/kexec.h>.
111 *
112 * The code for the transition from the current kernel to the
113 * the new kernel is placed in the control_code_buffer, whose size
114 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
115 * page of memory is necessary, but some architectures require more.
116 * Because this memory must be identity mapped in the transition from
117 * virtual to physical addresses it must live in the range
118 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
119 * modifiable.
120 *
121 * The assembly stub in the control code buffer is passed a linked list
122 * of descriptor pages detailing the source pages of the new kernel,
123 * and the destination addresses of those source pages. As this data
124 * structure is not used in the context of the current OS, it must
125 * be self-contained.
126 *
127 * The code has been made to work with highmem pages and will use a
128 * destination page in its final resting place (if it happens
129 * to allocate it). The end product of this is that most of the
130 * physical address space, and most of RAM can be used.
131 *
132 * Future directions include:
133 * - allocating a page table with the control code buffer identity
134 * mapped, to simplify machine_kexec and make kexec_on_panic more
135 * reliable.
136 */
137
138/*
139 * KIMAGE_NO_DEST is an impossible destination address..., for
140 * allocating pages whose destination address we do not care about.
141 */
142#define KIMAGE_NO_DEST (-1UL)
143
144static struct page *kimage_alloc_page(struct kimage *image,
145 gfp_t gfp_mask,
146 unsigned long dest);
147
148int sanity_check_segment_list(struct kimage *image)
149{
150 int result, i;
151 unsigned long nr_segments = image->nr_segments;
152
153 /*
154 * Verify we have good destination addresses. The caller is
155 * responsible for making certain we don't attempt to load
156 * the new image into invalid or reserved areas of RAM. This
157 * just verifies it is an address we can use.
158 *
159 * Since the kernel does everything in page size chunks ensure
160 * the destination addresses are page aligned. Too many
161 * special cases crop of when we don't do this. The most
162 * insidious is getting overlapping destination addresses
163 * simply because addresses are changed to page size
164 * granularity.
165 */
166 result = -EADDRNOTAVAIL;
167 for (i = 0; i < nr_segments; i++) {
168 unsigned long mstart, mend;
169
170 mstart = image->segment[i].mem;
171 mend = mstart + image->segment[i].memsz;
172 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
173 return result;
174 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
175 return result;
176 }
177
178 /* Verify our destination addresses do not overlap.
179 * If we alloed overlapping destination addresses
180 * through very weird things can happen with no
181 * easy explanation as one segment stops on another.
182 */
183 result = -EINVAL;
184 for (i = 0; i < nr_segments; i++) {
185 unsigned long mstart, mend;
186 unsigned long j;
187
188 mstart = image->segment[i].mem;
189 mend = mstart + image->segment[i].memsz;
190 for (j = 0; j < i; j++) {
191 unsigned long pstart, pend;
192
193 pstart = image->segment[j].mem;
194 pend = pstart + image->segment[j].memsz;
195 /* Do the segments overlap ? */
196 if ((mend > pstart) && (mstart < pend))
197 return result;
198 }
199 }
200
201 /* Ensure our buffer sizes are strictly less than
202 * our memory sizes. This should always be the case,
203 * and it is easier to check up front than to be surprised
204 * later on.
205 */
206 result = -EINVAL;
207 for (i = 0; i < nr_segments; i++) {
208 if (image->segment[i].bufsz > image->segment[i].memsz)
209 return result;
210 }
211
212 /*
213 * Verify we have good destination addresses. Normally
214 * the caller is responsible for making certain we don't
215 * attempt to load the new image into invalid or reserved
216 * areas of RAM. But crash kernels are preloaded into a
217 * reserved area of ram. We must ensure the addresses
218 * are in the reserved area otherwise preloading the
219 * kernel could corrupt things.
220 */
221
222 if (image->type == KEXEC_TYPE_CRASH) {
223 result = -EADDRNOTAVAIL;
224 for (i = 0; i < nr_segments; i++) {
225 unsigned long mstart, mend;
226
227 mstart = image->segment[i].mem;
228 mend = mstart + image->segment[i].memsz - 1;
229 /* Ensure we are within the crash kernel limits */
230 if ((mstart < crashk_res.start) ||
231 (mend > crashk_res.end))
232 return result;
233 }
234 }
235
236 return 0;
237}
238
239struct kimage *do_kimage_alloc_init(void)
240{
241 struct kimage *image;
242
243 /* Allocate a controlling structure */
244 image = kzalloc(sizeof(*image), GFP_KERNEL);
245 if (!image)
246 return NULL;
247
248 image->head = 0;
249 image->entry = &image->head;
250 image->last_entry = &image->head;
251 image->control_page = ~0; /* By default this does not apply */
252 image->type = KEXEC_TYPE_DEFAULT;
253
254 /* Initialize the list of control pages */
255 INIT_LIST_HEAD(&image->control_pages);
256
257 /* Initialize the list of destination pages */
258 INIT_LIST_HEAD(&image->dest_pages);
259
260 /* Initialize the list of unusable pages */
261 INIT_LIST_HEAD(&image->unusable_pages);
262
263 return image;
264}
265
266int kimage_is_destination_range(struct kimage *image,
267 unsigned long start,
268 unsigned long end)
269{
270 unsigned long i;
271
272 for (i = 0; i < image->nr_segments; i++) {
273 unsigned long mstart, mend;
274
275 mstart = image->segment[i].mem;
276 mend = mstart + image->segment[i].memsz;
277 if ((end > mstart) && (start < mend))
278 return 1;
279 }
280
281 return 0;
282}
283
284static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
285{
286 struct page *pages;
287
288 pages = alloc_pages(gfp_mask, order);
289 if (pages) {
290 unsigned int count, i;
291
292 pages->mapping = NULL;
293 set_page_private(pages, order);
294 count = 1 << order;
295 for (i = 0; i < count; i++)
296 SetPageReserved(pages + i);
297 }
298
299 return pages;
300}
301
302static void kimage_free_pages(struct page *page)
303{
304 unsigned int order, count, i;
305
306 order = page_private(page);
307 count = 1 << order;
308 for (i = 0; i < count; i++)
309 ClearPageReserved(page + i);
310 __free_pages(page, order);
311}
312
313void kimage_free_page_list(struct list_head *list)
314{
315 struct page *page, *next;
316
317 list_for_each_entry_safe(page, next, list, lru) {
318 list_del(&page->lru);
319 kimage_free_pages(page);
320 }
321}
322
323static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
324 unsigned int order)
325{
326 /* Control pages are special, they are the intermediaries
327 * that are needed while we copy the rest of the pages
328 * to their final resting place. As such they must
329 * not conflict with either the destination addresses
330 * or memory the kernel is already using.
331 *
332 * The only case where we really need more than one of
333 * these are for architectures where we cannot disable
334 * the MMU and must instead generate an identity mapped
335 * page table for all of the memory.
336 *
337 * At worst this runs in O(N) of the image size.
338 */
339 struct list_head extra_pages;
340 struct page *pages;
341 unsigned int count;
342
343 count = 1 << order;
344 INIT_LIST_HEAD(&extra_pages);
345
346 /* Loop while I can allocate a page and the page allocated
347 * is a destination page.
348 */
349 do {
350 unsigned long pfn, epfn, addr, eaddr;
351
352 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
353 if (!pages)
354 break;
355 pfn = page_to_pfn(pages);
356 epfn = pfn + count;
357 addr = pfn << PAGE_SHIFT;
358 eaddr = epfn << PAGE_SHIFT;
359 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
360 kimage_is_destination_range(image, addr, eaddr)) {
361 list_add(&pages->lru, &extra_pages);
362 pages = NULL;
363 }
364 } while (!pages);
365
366 if (pages) {
367 /* Remember the allocated page... */
368 list_add(&pages->lru, &image->control_pages);
369
370 /* Because the page is already in it's destination
371 * location we will never allocate another page at
372 * that address. Therefore kimage_alloc_pages
373 * will not return it (again) and we don't need
374 * to give it an entry in image->segment[].
375 */
376 }
377 /* Deal with the destination pages I have inadvertently allocated.
378 *
379 * Ideally I would convert multi-page allocations into single
380 * page allocations, and add everything to image->dest_pages.
381 *
382 * For now it is simpler to just free the pages.
383 */
384 kimage_free_page_list(&extra_pages);
385
386 return pages;
387}
388
389static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
390 unsigned int order)
391{
392 /* Control pages are special, they are the intermediaries
393 * that are needed while we copy the rest of the pages
394 * to their final resting place. As such they must
395 * not conflict with either the destination addresses
396 * or memory the kernel is already using.
397 *
398 * Control pages are also the only pags we must allocate
399 * when loading a crash kernel. All of the other pages
400 * are specified by the segments and we just memcpy
401 * into them directly.
402 *
403 * The only case where we really need more than one of
404 * these are for architectures where we cannot disable
405 * the MMU and must instead generate an identity mapped
406 * page table for all of the memory.
407 *
408 * Given the low demand this implements a very simple
409 * allocator that finds the first hole of the appropriate
410 * size in the reserved memory region, and allocates all
411 * of the memory up to and including the hole.
412 */
413 unsigned long hole_start, hole_end, size;
414 struct page *pages;
415
416 pages = NULL;
417 size = (1 << order) << PAGE_SHIFT;
418 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
419 hole_end = hole_start + size - 1;
420 while (hole_end <= crashk_res.end) {
421 unsigned long i;
422
423 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
424 break;
425 /* See if I overlap any of the segments */
426 for (i = 0; i < image->nr_segments; i++) {
427 unsigned long mstart, mend;
428
429 mstart = image->segment[i].mem;
430 mend = mstart + image->segment[i].memsz - 1;
431 if ((hole_end >= mstart) && (hole_start <= mend)) {
432 /* Advance the hole to the end of the segment */
433 hole_start = (mend + (size - 1)) & ~(size - 1);
434 hole_end = hole_start + size - 1;
435 break;
436 }
437 }
438 /* If I don't overlap any segments I have found my hole! */
439 if (i == image->nr_segments) {
440 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
441 image->control_page = hole_end;
442 break;
443 }
444 }
445
446 return pages;
447}
448
449
450struct page *kimage_alloc_control_pages(struct kimage *image,
451 unsigned int order)
452{
453 struct page *pages = NULL;
454
455 switch (image->type) {
456 case KEXEC_TYPE_DEFAULT:
457 pages = kimage_alloc_normal_control_pages(image, order);
458 break;
459 case KEXEC_TYPE_CRASH:
460 pages = kimage_alloc_crash_control_pages(image, order);
461 break;
462 }
463
464 return pages;
465}
466
467static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
468{
469 if (*image->entry != 0)
470 image->entry++;
471
472 if (image->entry == image->last_entry) {
473 kimage_entry_t *ind_page;
474 struct page *page;
475
476 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
477 if (!page)
478 return -ENOMEM;
479
480 ind_page = page_address(page);
481 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
482 image->entry = ind_page;
483 image->last_entry = ind_page +
484 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
485 }
486 *image->entry = entry;
487 image->entry++;
488 *image->entry = 0;
489
490 return 0;
491}
492
493static int kimage_set_destination(struct kimage *image,
494 unsigned long destination)
495{
496 int result;
497
498 destination &= PAGE_MASK;
499 result = kimage_add_entry(image, destination | IND_DESTINATION);
500
501 return result;
502}
503
504
505static int kimage_add_page(struct kimage *image, unsigned long page)
506{
507 int result;
508
509 page &= PAGE_MASK;
510 result = kimage_add_entry(image, page | IND_SOURCE);
511
512 return result;
513}
514
515
516static void kimage_free_extra_pages(struct kimage *image)
517{
518 /* Walk through and free any extra destination pages I may have */
519 kimage_free_page_list(&image->dest_pages);
520
521 /* Walk through and free any unusable pages I have cached */
522 kimage_free_page_list(&image->unusable_pages);
523
524}
525void kimage_terminate(struct kimage *image)
526{
527 if (*image->entry != 0)
528 image->entry++;
529
530 *image->entry = IND_DONE;
531}
532
533#define for_each_kimage_entry(image, ptr, entry) \
534 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
535 ptr = (entry & IND_INDIRECTION) ? \
536 phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
537
538static void kimage_free_entry(kimage_entry_t entry)
539{
540 struct page *page;
541
542 page = pfn_to_page(entry >> PAGE_SHIFT);
543 kimage_free_pages(page);
544}
545
546void kimage_free(struct kimage *image)
547{
548 kimage_entry_t *ptr, entry;
549 kimage_entry_t ind = 0;
550
551 if (!image)
552 return;
553
554 kimage_free_extra_pages(image);
555 for_each_kimage_entry(image, ptr, entry) {
556 if (entry & IND_INDIRECTION) {
557 /* Free the previous indirection page */
558 if (ind & IND_INDIRECTION)
559 kimage_free_entry(ind);
560 /* Save this indirection page until we are
561 * done with it.
562 */
563 ind = entry;
564 } else if (entry & IND_SOURCE)
565 kimage_free_entry(entry);
566 }
567 /* Free the final indirection page */
568 if (ind & IND_INDIRECTION)
569 kimage_free_entry(ind);
570
571 /* Handle any machine specific cleanup */
572 machine_kexec_cleanup(image);
573
574 /* Free the kexec control pages... */
575 kimage_free_page_list(&image->control_pages);
576
577 /*
578 * Free up any temporary buffers allocated. This might hit if
579 * error occurred much later after buffer allocation.
580 */
581 if (image->file_mode)
582 kimage_file_post_load_cleanup(image);
583
584 kfree(image);
585}
586
587static kimage_entry_t *kimage_dst_used(struct kimage *image,
588 unsigned long page)
589{
590 kimage_entry_t *ptr, entry;
591 unsigned long destination = 0;
592
593 for_each_kimage_entry(image, ptr, entry) {
594 if (entry & IND_DESTINATION)
595 destination = entry & PAGE_MASK;
596 else if (entry & IND_SOURCE) {
597 if (page == destination)
598 return ptr;
599 destination += PAGE_SIZE;
600 }
601 }
602
603 return NULL;
604}
605
606static struct page *kimage_alloc_page(struct kimage *image,
607 gfp_t gfp_mask,
608 unsigned long destination)
609{
610 /*
611 * Here we implement safeguards to ensure that a source page
612 * is not copied to its destination page before the data on
613 * the destination page is no longer useful.
614 *
615 * To do this we maintain the invariant that a source page is
616 * either its own destination page, or it is not a
617 * destination page at all.
618 *
619 * That is slightly stronger than required, but the proof
620 * that no problems will not occur is trivial, and the
621 * implementation is simply to verify.
622 *
623 * When allocating all pages normally this algorithm will run
624 * in O(N) time, but in the worst case it will run in O(N^2)
625 * time. If the runtime is a problem the data structures can
626 * be fixed.
627 */
628 struct page *page;
629 unsigned long addr;
630
631 /*
632 * Walk through the list of destination pages, and see if I
633 * have a match.
634 */
635 list_for_each_entry(page, &image->dest_pages, lru) {
636 addr = page_to_pfn(page) << PAGE_SHIFT;
637 if (addr == destination) {
638 list_del(&page->lru);
639 return page;
640 }
641 }
642 page = NULL;
643 while (1) {
644 kimage_entry_t *old;
645
646 /* Allocate a page, if we run out of memory give up */
647 page = kimage_alloc_pages(gfp_mask, 0);
648 if (!page)
649 return NULL;
650 /* If the page cannot be used file it away */
651 if (page_to_pfn(page) >
652 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
653 list_add(&page->lru, &image->unusable_pages);
654 continue;
655 }
656 addr = page_to_pfn(page) << PAGE_SHIFT;
657
658 /* If it is the destination page we want use it */
659 if (addr == destination)
660 break;
661
662 /* If the page is not a destination page use it */
663 if (!kimage_is_destination_range(image, addr,
664 addr + PAGE_SIZE))
665 break;
666
667 /*
668 * I know that the page is someones destination page.
669 * See if there is already a source page for this
670 * destination page. And if so swap the source pages.
671 */
672 old = kimage_dst_used(image, addr);
673 if (old) {
674 /* If so move it */
675 unsigned long old_addr;
676 struct page *old_page;
677
678 old_addr = *old & PAGE_MASK;
679 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
680 copy_highpage(page, old_page);
681 *old = addr | (*old & ~PAGE_MASK);
682
683 /* The old page I have found cannot be a
684 * destination page, so return it if it's
685 * gfp_flags honor the ones passed in.
686 */
687 if (!(gfp_mask & __GFP_HIGHMEM) &&
688 PageHighMem(old_page)) {
689 kimage_free_pages(old_page);
690 continue;
691 }
692 addr = old_addr;
693 page = old_page;
694 break;
695 }
696 /* Place the page on the destination list, to be used later */
697 list_add(&page->lru, &image->dest_pages);
698 }
699
700 return page;
701}
702
703static int kimage_load_normal_segment(struct kimage *image,
704 struct kexec_segment *segment)
705{
706 unsigned long maddr;
707 size_t ubytes, mbytes;
708 int result;
709 unsigned char __user *buf = NULL;
710 unsigned char *kbuf = NULL;
711
712 result = 0;
713 if (image->file_mode)
714 kbuf = segment->kbuf;
715 else
716 buf = segment->buf;
717 ubytes = segment->bufsz;
718 mbytes = segment->memsz;
719 maddr = segment->mem;
720
721 result = kimage_set_destination(image, maddr);
722 if (result < 0)
723 goto out;
724
725 while (mbytes) {
726 struct page *page;
727 char *ptr;
728 size_t uchunk, mchunk;
729
730 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
731 if (!page) {
732 result = -ENOMEM;
733 goto out;
734 }
735 result = kimage_add_page(image, page_to_pfn(page)
736 << PAGE_SHIFT);
737 if (result < 0)
738 goto out;
739
740 ptr = kmap(page);
741 /* Start with a clear page */
742 clear_page(ptr);
743 ptr += maddr & ~PAGE_MASK;
744 mchunk = min_t(size_t, mbytes,
745 PAGE_SIZE - (maddr & ~PAGE_MASK));
746 uchunk = min(ubytes, mchunk);
747
748 /* For file based kexec, source pages are in kernel memory */
749 if (image->file_mode)
750 memcpy(ptr, kbuf, uchunk);
751 else
752 result = copy_from_user(ptr, buf, uchunk);
753 kunmap(page);
754 if (result) {
755 result = -EFAULT;
756 goto out;
757 }
758 ubytes -= uchunk;
759 maddr += mchunk;
760 if (image->file_mode)
761 kbuf += mchunk;
762 else
763 buf += mchunk;
764 mbytes -= mchunk;
765 }
766out:
767 return result;
768}
769
770static int kimage_load_crash_segment(struct kimage *image,
771 struct kexec_segment *segment)
772{
773 /* For crash dumps kernels we simply copy the data from
774 * user space to it's destination.
775 * We do things a page at a time for the sake of kmap.
776 */
777 unsigned long maddr;
778 size_t ubytes, mbytes;
779 int result;
780 unsigned char __user *buf = NULL;
781 unsigned char *kbuf = NULL;
782
783 result = 0;
784 if (image->file_mode)
785 kbuf = segment->kbuf;
786 else
787 buf = segment->buf;
788 ubytes = segment->bufsz;
789 mbytes = segment->memsz;
790 maddr = segment->mem;
791 while (mbytes) {
792 struct page *page;
793 char *ptr;
794 size_t uchunk, mchunk;
795
796 page = pfn_to_page(maddr >> PAGE_SHIFT);
797 if (!page) {
798 result = -ENOMEM;
799 goto out;
800 }
801 ptr = kmap(page);
802 ptr += maddr & ~PAGE_MASK;
803 mchunk = min_t(size_t, mbytes,
804 PAGE_SIZE - (maddr & ~PAGE_MASK));
805 uchunk = min(ubytes, mchunk);
806 if (mchunk > uchunk) {
807 /* Zero the trailing part of the page */
808 memset(ptr + uchunk, 0, mchunk - uchunk);
809 }
810
811 /* For file based kexec, source pages are in kernel memory */
812 if (image->file_mode)
813 memcpy(ptr, kbuf, uchunk);
814 else
815 result = copy_from_user(ptr, buf, uchunk);
816 kexec_flush_icache_page(page);
817 kunmap(page);
818 if (result) {
819 result = -EFAULT;
820 goto out;
821 }
822 ubytes -= uchunk;
823 maddr += mchunk;
824 if (image->file_mode)
825 kbuf += mchunk;
826 else
827 buf += mchunk;
828 mbytes -= mchunk;
829 }
830out:
831 return result;
832}
833
834int kimage_load_segment(struct kimage *image,
835 struct kexec_segment *segment)
836{
837 int result = -ENOMEM;
838
839 switch (image->type) {
840 case KEXEC_TYPE_DEFAULT:
841 result = kimage_load_normal_segment(image, segment);
842 break;
843 case KEXEC_TYPE_CRASH:
844 result = kimage_load_crash_segment(image, segment);
845 break;
846 }
847
848 return result;
849}
850
851struct kimage *kexec_image;
852struct kimage *kexec_crash_image;
853int kexec_load_disabled;
854
855/*
856 * No panic_cpu check version of crash_kexec(). This function is called
857 * only when panic_cpu holds the current CPU number; this is the only CPU
858 * which processes crash_kexec routines.
859 */
860void __crash_kexec(struct pt_regs *regs)
861{
862 /* Take the kexec_mutex here to prevent sys_kexec_load
863 * running on one cpu from replacing the crash kernel
864 * we are using after a panic on a different cpu.
865 *
866 * If the crash kernel was not located in a fixed area
867 * of memory the xchg(&kexec_crash_image) would be
868 * sufficient. But since I reuse the memory...
869 */
870 if (mutex_trylock(&kexec_mutex)) {
871 if (kexec_crash_image) {
872 struct pt_regs fixed_regs;
873
874 crash_setup_regs(&fixed_regs, regs);
875 crash_save_vmcoreinfo();
876 machine_crash_shutdown(&fixed_regs);
877 machine_kexec(kexec_crash_image);
878 }
879 mutex_unlock(&kexec_mutex);
880 }
881}
882
883void crash_kexec(struct pt_regs *regs)
884{
885 int old_cpu, this_cpu;
886
887 /*
888 * Only one CPU is allowed to execute the crash_kexec() code as with
889 * panic(). Otherwise parallel calls of panic() and crash_kexec()
890 * may stop each other. To exclude them, we use panic_cpu here too.
891 */
892 this_cpu = raw_smp_processor_id();
893 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
894 if (old_cpu == PANIC_CPU_INVALID) {
895 /* This is the 1st CPU which comes here, so go ahead. */
896 __crash_kexec(regs);
897
898 /*
899 * Reset panic_cpu to allow another panic()/crash_kexec()
900 * call.
901 */
902 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
903 }
904}
905
906size_t crash_get_memory_size(void)
907{
908 size_t size = 0;
909
910 mutex_lock(&kexec_mutex);
911 if (crashk_res.end != crashk_res.start)
912 size = resource_size(&crashk_res);
913 mutex_unlock(&kexec_mutex);
914 return size;
915}
916
917void __weak crash_free_reserved_phys_range(unsigned long begin,
918 unsigned long end)
919{
920 unsigned long addr;
921
922 for (addr = begin; addr < end; addr += PAGE_SIZE)
923 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
924}
925
926int crash_shrink_memory(unsigned long new_size)
927{
928 int ret = 0;
929 unsigned long start, end;
930 unsigned long old_size;
931 struct resource *ram_res;
932
933 mutex_lock(&kexec_mutex);
934
935 if (kexec_crash_image) {
936 ret = -ENOENT;
937 goto unlock;
938 }
939 start = crashk_res.start;
940 end = crashk_res.end;
941 old_size = (end == 0) ? 0 : end - start + 1;
942 if (new_size >= old_size) {
943 ret = (new_size == old_size) ? 0 : -EINVAL;
944 goto unlock;
945 }
946
947 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
948 if (!ram_res) {
949 ret = -ENOMEM;
950 goto unlock;
951 }
952
953 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
954 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
955
956 crash_map_reserved_pages();
957 crash_free_reserved_phys_range(end, crashk_res.end);
958
959 if ((start == end) && (crashk_res.parent != NULL))
960 release_resource(&crashk_res);
961
962 ram_res->start = end;
963 ram_res->end = crashk_res.end;
964 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
965 ram_res->name = "System RAM";
966
967 crashk_res.end = end - 1;
968
969 insert_resource(&iomem_resource, ram_res);
970 crash_unmap_reserved_pages();
971
972unlock:
973 mutex_unlock(&kexec_mutex);
974 return ret;
975}
976
977static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
978 size_t data_len)
979{
980 struct elf_note note;
981
982 note.n_namesz = strlen(name) + 1;
983 note.n_descsz = data_len;
984 note.n_type = type;
985 memcpy(buf, ¬e, sizeof(note));
986 buf += (sizeof(note) + 3)/4;
987 memcpy(buf, name, note.n_namesz);
988 buf += (note.n_namesz + 3)/4;
989 memcpy(buf, data, note.n_descsz);
990 buf += (note.n_descsz + 3)/4;
991
992 return buf;
993}
994
995static void final_note(u32 *buf)
996{
997 struct elf_note note;
998
999 note.n_namesz = 0;
1000 note.n_descsz = 0;
1001 note.n_type = 0;
1002 memcpy(buf, ¬e, sizeof(note));
1003}
1004
1005void crash_save_cpu(struct pt_regs *regs, int cpu)
1006{
1007 struct elf_prstatus prstatus;
1008 u32 *buf;
1009
1010 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1011 return;
1012
1013 /* Using ELF notes here is opportunistic.
1014 * I need a well defined structure format
1015 * for the data I pass, and I need tags
1016 * on the data to indicate what information I have
1017 * squirrelled away. ELF notes happen to provide
1018 * all of that, so there is no need to invent something new.
1019 */
1020 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1021 if (!buf)
1022 return;
1023 memset(&prstatus, 0, sizeof(prstatus));
1024 prstatus.pr_pid = current->pid;
1025 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1026 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1027 &prstatus, sizeof(prstatus));
1028 final_note(buf);
1029}
1030
1031static int __init crash_notes_memory_init(void)
1032{
1033 /* Allocate memory for saving cpu registers. */
1034 size_t size, align;
1035
1036 /*
1037 * crash_notes could be allocated across 2 vmalloc pages when percpu
1038 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1039 * pages are also on 2 continuous physical pages. In this case the
1040 * 2nd part of crash_notes in 2nd page could be lost since only the
1041 * starting address and size of crash_notes are exported through sysfs.
1042 * Here round up the size of crash_notes to the nearest power of two
1043 * and pass it to __alloc_percpu as align value. This can make sure
1044 * crash_notes is allocated inside one physical page.
1045 */
1046 size = sizeof(note_buf_t);
1047 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1048
1049 /*
1050 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1051 * definitely will be in 2 pages with that.
1052 */
1053 BUILD_BUG_ON(size > PAGE_SIZE);
1054
1055 crash_notes = __alloc_percpu(size, align);
1056 if (!crash_notes) {
1057 pr_warn("Memory allocation for saving cpu register states failed\n");
1058 return -ENOMEM;
1059 }
1060 return 0;
1061}
1062subsys_initcall(crash_notes_memory_init);
1063
1064
1065/*
1066 * parsing the "crashkernel" commandline
1067 *
1068 * this code is intended to be called from architecture specific code
1069 */
1070
1071
1072/*
1073 * This function parses command lines in the format
1074 *
1075 * crashkernel=ramsize-range:size[,...][@offset]
1076 *
1077 * The function returns 0 on success and -EINVAL on failure.
1078 */
1079static int __init parse_crashkernel_mem(char *cmdline,
1080 unsigned long long system_ram,
1081 unsigned long long *crash_size,
1082 unsigned long long *crash_base)
1083{
1084 char *cur = cmdline, *tmp;
1085
1086 /* for each entry of the comma-separated list */
1087 do {
1088 unsigned long long start, end = ULLONG_MAX, size;
1089
1090 /* get the start of the range */
1091 start = memparse(cur, &tmp);
1092 if (cur == tmp) {
1093 pr_warn("crashkernel: Memory value expected\n");
1094 return -EINVAL;
1095 }
1096 cur = tmp;
1097 if (*cur != '-') {
1098 pr_warn("crashkernel: '-' expected\n");
1099 return -EINVAL;
1100 }
1101 cur++;
1102
1103 /* if no ':' is here, than we read the end */
1104 if (*cur != ':') {
1105 end = memparse(cur, &tmp);
1106 if (cur == tmp) {
1107 pr_warn("crashkernel: Memory value expected\n");
1108 return -EINVAL;
1109 }
1110 cur = tmp;
1111 if (end <= start) {
1112 pr_warn("crashkernel: end <= start\n");
1113 return -EINVAL;
1114 }
1115 }
1116
1117 if (*cur != ':') {
1118 pr_warn("crashkernel: ':' expected\n");
1119 return -EINVAL;
1120 }
1121 cur++;
1122
1123 size = memparse(cur, &tmp);
1124 if (cur == tmp) {
1125 pr_warn("Memory value expected\n");
1126 return -EINVAL;
1127 }
1128 cur = tmp;
1129 if (size >= system_ram) {
1130 pr_warn("crashkernel: invalid size\n");
1131 return -EINVAL;
1132 }
1133
1134 /* match ? */
1135 if (system_ram >= start && system_ram < end) {
1136 *crash_size = size;
1137 break;
1138 }
1139 } while (*cur++ == ',');
1140
1141 if (*crash_size > 0) {
1142 while (*cur && *cur != ' ' && *cur != '@')
1143 cur++;
1144 if (*cur == '@') {
1145 cur++;
1146 *crash_base = memparse(cur, &tmp);
1147 if (cur == tmp) {
1148 pr_warn("Memory value expected after '@'\n");
1149 return -EINVAL;
1150 }
1151 }
1152 }
1153
1154 return 0;
1155}
1156
1157/*
1158 * That function parses "simple" (old) crashkernel command lines like
1159 *
1160 * crashkernel=size[@offset]
1161 *
1162 * It returns 0 on success and -EINVAL on failure.
1163 */
1164static int __init parse_crashkernel_simple(char *cmdline,
1165 unsigned long long *crash_size,
1166 unsigned long long *crash_base)
1167{
1168 char *cur = cmdline;
1169
1170 *crash_size = memparse(cmdline, &cur);
1171 if (cmdline == cur) {
1172 pr_warn("crashkernel: memory value expected\n");
1173 return -EINVAL;
1174 }
1175
1176 if (*cur == '@')
1177 *crash_base = memparse(cur+1, &cur);
1178 else if (*cur != ' ' && *cur != '\0') {
1179 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1180 return -EINVAL;
1181 }
1182
1183 return 0;
1184}
1185
1186#define SUFFIX_HIGH 0
1187#define SUFFIX_LOW 1
1188#define SUFFIX_NULL 2
1189static __initdata char *suffix_tbl[] = {
1190 [SUFFIX_HIGH] = ",high",
1191 [SUFFIX_LOW] = ",low",
1192 [SUFFIX_NULL] = NULL,
1193};
1194
1195/*
1196 * That function parses "suffix" crashkernel command lines like
1197 *
1198 * crashkernel=size,[high|low]
1199 *
1200 * It returns 0 on success and -EINVAL on failure.
1201 */
1202static int __init parse_crashkernel_suffix(char *cmdline,
1203 unsigned long long *crash_size,
1204 const char *suffix)
1205{
1206 char *cur = cmdline;
1207
1208 *crash_size = memparse(cmdline, &cur);
1209 if (cmdline == cur) {
1210 pr_warn("crashkernel: memory value expected\n");
1211 return -EINVAL;
1212 }
1213
1214 /* check with suffix */
1215 if (strncmp(cur, suffix, strlen(suffix))) {
1216 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1217 return -EINVAL;
1218 }
1219 cur += strlen(suffix);
1220 if (*cur != ' ' && *cur != '\0') {
1221 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
1222 return -EINVAL;
1223 }
1224
1225 return 0;
1226}
1227
1228static __init char *get_last_crashkernel(char *cmdline,
1229 const char *name,
1230 const char *suffix)
1231{
1232 char *p = cmdline, *ck_cmdline = NULL;
1233
1234 /* find crashkernel and use the last one if there are more */
1235 p = strstr(p, name);
1236 while (p) {
1237 char *end_p = strchr(p, ' ');
1238 char *q;
1239
1240 if (!end_p)
1241 end_p = p + strlen(p);
1242
1243 if (!suffix) {
1244 int i;
1245
1246 /* skip the one with any known suffix */
1247 for (i = 0; suffix_tbl[i]; i++) {
1248 q = end_p - strlen(suffix_tbl[i]);
1249 if (!strncmp(q, suffix_tbl[i],
1250 strlen(suffix_tbl[i])))
1251 goto next;
1252 }
1253 ck_cmdline = p;
1254 } else {
1255 q = end_p - strlen(suffix);
1256 if (!strncmp(q, suffix, strlen(suffix)))
1257 ck_cmdline = p;
1258 }
1259next:
1260 p = strstr(p+1, name);
1261 }
1262
1263 if (!ck_cmdline)
1264 return NULL;
1265
1266 return ck_cmdline;
1267}
1268
1269static int __init __parse_crashkernel(char *cmdline,
1270 unsigned long long system_ram,
1271 unsigned long long *crash_size,
1272 unsigned long long *crash_base,
1273 const char *name,
1274 const char *suffix)
1275{
1276 char *first_colon, *first_space;
1277 char *ck_cmdline;
1278
1279 BUG_ON(!crash_size || !crash_base);
1280 *crash_size = 0;
1281 *crash_base = 0;
1282
1283 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1284
1285 if (!ck_cmdline)
1286 return -EINVAL;
1287
1288 ck_cmdline += strlen(name);
1289
1290 if (suffix)
1291 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1292 suffix);
1293 /*
1294 * if the commandline contains a ':', then that's the extended
1295 * syntax -- if not, it must be the classic syntax
1296 */
1297 first_colon = strchr(ck_cmdline, ':');
1298 first_space = strchr(ck_cmdline, ' ');
1299 if (first_colon && (!first_space || first_colon < first_space))
1300 return parse_crashkernel_mem(ck_cmdline, system_ram,
1301 crash_size, crash_base);
1302
1303 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1304}
1305
1306/*
1307 * That function is the entry point for command line parsing and should be
1308 * called from the arch-specific code.
1309 */
1310int __init parse_crashkernel(char *cmdline,
1311 unsigned long long system_ram,
1312 unsigned long long *crash_size,
1313 unsigned long long *crash_base)
1314{
1315 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1316 "crashkernel=", NULL);
1317}
1318
1319int __init parse_crashkernel_high(char *cmdline,
1320 unsigned long long system_ram,
1321 unsigned long long *crash_size,
1322 unsigned long long *crash_base)
1323{
1324 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1325 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1326}
1327
1328int __init parse_crashkernel_low(char *cmdline,
1329 unsigned long long system_ram,
1330 unsigned long long *crash_size,
1331 unsigned long long *crash_base)
1332{
1333 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1334 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1335}
1336
1337static void update_vmcoreinfo_note(void)
1338{
1339 u32 *buf = vmcoreinfo_note;
1340
1341 if (!vmcoreinfo_size)
1342 return;
1343 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1344 vmcoreinfo_size);
1345 final_note(buf);
1346}
1347
1348void crash_save_vmcoreinfo(void)
1349{
1350 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1351 update_vmcoreinfo_note();
1352}
1353
1354void vmcoreinfo_append_str(const char *fmt, ...)
1355{
1356 va_list args;
1357 char buf[0x50];
1358 size_t r;
1359
1360 va_start(args, fmt);
1361 r = vscnprintf(buf, sizeof(buf), fmt, args);
1362 va_end(args);
1363
1364 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1365
1366 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1367
1368 vmcoreinfo_size += r;
1369}
1370
1371/*
1372 * provide an empty default implementation here -- architecture
1373 * code may override this
1374 */
1375void __weak arch_crash_save_vmcoreinfo(void)
1376{}
1377
1378unsigned long __weak paddr_vmcoreinfo_note(void)
1379{
1380 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1381}
1382
1383static int __init crash_save_vmcoreinfo_init(void)
1384{
1385 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1386 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1387
1388 VMCOREINFO_SYMBOL(init_uts_ns);
1389 VMCOREINFO_SYMBOL(node_online_map);
1390#ifdef CONFIG_MMU
1391 VMCOREINFO_SYMBOL(swapper_pg_dir);
1392#endif
1393 VMCOREINFO_SYMBOL(_stext);
1394 VMCOREINFO_SYMBOL(vmap_area_list);
1395
1396#ifndef CONFIG_NEED_MULTIPLE_NODES
1397 VMCOREINFO_SYMBOL(mem_map);
1398 VMCOREINFO_SYMBOL(contig_page_data);
1399#endif
1400#ifdef CONFIG_SPARSEMEM
1401 VMCOREINFO_SYMBOL(mem_section);
1402 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1403 VMCOREINFO_STRUCT_SIZE(mem_section);
1404 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1405#endif
1406 VMCOREINFO_STRUCT_SIZE(page);
1407 VMCOREINFO_STRUCT_SIZE(pglist_data);
1408 VMCOREINFO_STRUCT_SIZE(zone);
1409 VMCOREINFO_STRUCT_SIZE(free_area);
1410 VMCOREINFO_STRUCT_SIZE(list_head);
1411 VMCOREINFO_SIZE(nodemask_t);
1412 VMCOREINFO_OFFSET(page, flags);
1413 VMCOREINFO_OFFSET(page, _count);
1414 VMCOREINFO_OFFSET(page, mapping);
1415 VMCOREINFO_OFFSET(page, lru);
1416 VMCOREINFO_OFFSET(page, _mapcount);
1417 VMCOREINFO_OFFSET(page, private);
1418 VMCOREINFO_OFFSET(page, compound_dtor);
1419 VMCOREINFO_OFFSET(page, compound_order);
1420 VMCOREINFO_OFFSET(page, compound_head);
1421 VMCOREINFO_OFFSET(pglist_data, node_zones);
1422 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1423#ifdef CONFIG_FLAT_NODE_MEM_MAP
1424 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1425#endif
1426 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1427 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1428 VMCOREINFO_OFFSET(pglist_data, node_id);
1429 VMCOREINFO_OFFSET(zone, free_area);
1430 VMCOREINFO_OFFSET(zone, vm_stat);
1431 VMCOREINFO_OFFSET(zone, spanned_pages);
1432 VMCOREINFO_OFFSET(free_area, free_list);
1433 VMCOREINFO_OFFSET(list_head, next);
1434 VMCOREINFO_OFFSET(list_head, prev);
1435 VMCOREINFO_OFFSET(vmap_area, va_start);
1436 VMCOREINFO_OFFSET(vmap_area, list);
1437 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1438 log_buf_kexec_setup();
1439 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1440 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1441 VMCOREINFO_NUMBER(PG_lru);
1442 VMCOREINFO_NUMBER(PG_private);
1443 VMCOREINFO_NUMBER(PG_swapcache);
1444 VMCOREINFO_NUMBER(PG_slab);
1445#ifdef CONFIG_MEMORY_FAILURE
1446 VMCOREINFO_NUMBER(PG_hwpoison);
1447#endif
1448 VMCOREINFO_NUMBER(PG_head_mask);
1449 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1450#ifdef CONFIG_X86
1451 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
1452#endif
1453#ifdef CONFIG_HUGETLB_PAGE
1454 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
1455#endif
1456
1457 arch_crash_save_vmcoreinfo();
1458 update_vmcoreinfo_note();
1459
1460 return 0;
1461}
1462
1463subsys_initcall(crash_save_vmcoreinfo_init);
1464
1465/*
1466 * Move into place and start executing a preloaded standalone
1467 * executable. If nothing was preloaded return an error.
1468 */
1469int kernel_kexec(void)
1470{
1471 int error = 0;
1472
1473 if (!mutex_trylock(&kexec_mutex))
1474 return -EBUSY;
1475 if (!kexec_image) {
1476 error = -EINVAL;
1477 goto Unlock;
1478 }
1479
1480#ifdef CONFIG_KEXEC_JUMP
1481 if (kexec_image->preserve_context) {
1482 lock_system_sleep();
1483 pm_prepare_console();
1484 error = freeze_processes();
1485 if (error) {
1486 error = -EBUSY;
1487 goto Restore_console;
1488 }
1489 suspend_console();
1490 error = dpm_suspend_start(PMSG_FREEZE);
1491 if (error)
1492 goto Resume_console;
1493 /* At this point, dpm_suspend_start() has been called,
1494 * but *not* dpm_suspend_end(). We *must* call
1495 * dpm_suspend_end() now. Otherwise, drivers for
1496 * some devices (e.g. interrupt controllers) become
1497 * desynchronized with the actual state of the
1498 * hardware at resume time, and evil weirdness ensues.
1499 */
1500 error = dpm_suspend_end(PMSG_FREEZE);
1501 if (error)
1502 goto Resume_devices;
1503 error = disable_nonboot_cpus();
1504 if (error)
1505 goto Enable_cpus;
1506 local_irq_disable();
1507 error = syscore_suspend();
1508 if (error)
1509 goto Enable_irqs;
1510 } else
1511#endif
1512 {
1513 kexec_in_progress = true;
1514 kernel_restart_prepare(NULL);
1515 migrate_to_reboot_cpu();
1516
1517 /*
1518 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1519 * no further code needs to use CPU hotplug (which is true in
1520 * the reboot case). However, the kexec path depends on using
1521 * CPU hotplug again; so re-enable it here.
1522 */
1523 cpu_hotplug_enable();
1524 pr_emerg("Starting new kernel\n");
1525 machine_shutdown();
1526 }
1527
1528 machine_kexec(kexec_image);
1529
1530#ifdef CONFIG_KEXEC_JUMP
1531 if (kexec_image->preserve_context) {
1532 syscore_resume();
1533 Enable_irqs:
1534 local_irq_enable();
1535 Enable_cpus:
1536 enable_nonboot_cpus();
1537 dpm_resume_start(PMSG_RESTORE);
1538 Resume_devices:
1539 dpm_resume_end(PMSG_RESTORE);
1540 Resume_console:
1541 resume_console();
1542 thaw_processes();
1543 Restore_console:
1544 pm_restore_console();
1545 unlock_system_sleep();
1546 }
1547#endif
1548
1549 Unlock:
1550 mutex_unlock(&kexec_mutex);
1551 return error;
1552}
1553
1554/*
1555 * Add and remove page tables for crashkernel memory
1556 *
1557 * Provide an empty default implementation here -- architecture
1558 * code may override this
1559 */
1560void __weak crash_map_reserved_pages(void)
1561{}
1562
1563void __weak crash_unmap_reserved_pages(void)
1564{}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kexec.c - kexec system call core code.
4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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/uaccess.h>
32#include <linux/io.h>
33#include <linux/console.h>
34#include <linux/vmalloc.h>
35#include <linux/swap.h>
36#include <linux/syscore_ops.h>
37#include <linux/compiler.h>
38#include <linux/hugetlb.h>
39#include <linux/frame.h>
40
41#include <asm/page.h>
42#include <asm/sections.h>
43
44#include <crypto/hash.h>
45#include <crypto/sha.h>
46#include "kexec_internal.h"
47
48DEFINE_MUTEX(kexec_mutex);
49
50/* Per cpu memory for storing cpu states in case of system crash. */
51note_buf_t __percpu *crash_notes;
52
53/* Flag to indicate we are going to kexec a new kernel */
54bool kexec_in_progress = false;
55
56
57/* Location of the reserved area for the crash kernel */
58struct resource crashk_res = {
59 .name = "Crash kernel",
60 .start = 0,
61 .end = 0,
62 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
63 .desc = IORES_DESC_CRASH_KERNEL
64};
65struct resource crashk_low_res = {
66 .name = "Crash kernel",
67 .start = 0,
68 .end = 0,
69 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
70 .desc = IORES_DESC_CRASH_KERNEL
71};
72
73int kexec_should_crash(struct task_struct *p)
74{
75 /*
76 * If crash_kexec_post_notifiers is enabled, don't run
77 * crash_kexec() here yet, which must be run after panic
78 * notifiers in panic().
79 */
80 if (crash_kexec_post_notifiers)
81 return 0;
82 /*
83 * There are 4 panic() calls in do_exit() path, each of which
84 * corresponds to each of these 4 conditions.
85 */
86 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
87 return 1;
88 return 0;
89}
90
91int kexec_crash_loaded(void)
92{
93 return !!kexec_crash_image;
94}
95EXPORT_SYMBOL_GPL(kexec_crash_loaded);
96
97/*
98 * When kexec transitions to the new kernel there is a one-to-one
99 * mapping between physical and virtual addresses. On processors
100 * where you can disable the MMU this is trivial, and easy. For
101 * others it is still a simple predictable page table to setup.
102 *
103 * In that environment kexec copies the new kernel to its final
104 * resting place. This means I can only support memory whose
105 * physical address can fit in an unsigned long. In particular
106 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
107 * If the assembly stub has more restrictive requirements
108 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
109 * defined more restrictively in <asm/kexec.h>.
110 *
111 * The code for the transition from the current kernel to the
112 * the new kernel is placed in the control_code_buffer, whose size
113 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
114 * page of memory is necessary, but some architectures require more.
115 * Because this memory must be identity mapped in the transition from
116 * virtual to physical addresses it must live in the range
117 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
118 * modifiable.
119 *
120 * The assembly stub in the control code buffer is passed a linked list
121 * of descriptor pages detailing the source pages of the new kernel,
122 * and the destination addresses of those source pages. As this data
123 * structure is not used in the context of the current OS, it must
124 * be self-contained.
125 *
126 * The code has been made to work with highmem pages and will use a
127 * destination page in its final resting place (if it happens
128 * to allocate it). The end product of this is that most of the
129 * physical address space, and most of RAM can be used.
130 *
131 * Future directions include:
132 * - allocating a page table with the control code buffer identity
133 * mapped, to simplify machine_kexec and make kexec_on_panic more
134 * reliable.
135 */
136
137/*
138 * KIMAGE_NO_DEST is an impossible destination address..., for
139 * allocating pages whose destination address we do not care about.
140 */
141#define KIMAGE_NO_DEST (-1UL)
142#define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
143
144static struct page *kimage_alloc_page(struct kimage *image,
145 gfp_t gfp_mask,
146 unsigned long dest);
147
148int sanity_check_segment_list(struct kimage *image)
149{
150 int i;
151 unsigned long nr_segments = image->nr_segments;
152 unsigned long total_pages = 0;
153 unsigned long nr_pages = totalram_pages();
154
155 /*
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.
160 *
161 * Since the kernel does everything in page size chunks ensure
162 * the destination addresses 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.
167 */
168 for (i = 0; i < nr_segments; i++) {
169 unsigned long mstart, mend;
170
171 mstart = image->segment[i].mem;
172 mend = mstart + image->segment[i].memsz;
173 if (mstart > mend)
174 return -EADDRNOTAVAIL;
175 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
176 return -EADDRNOTAVAIL;
177 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
178 return -EADDRNOTAVAIL;
179 }
180
181 /* Verify our destination addresses do not overlap.
182 * If we alloed overlapping destination addresses
183 * through very weird things can happen with no
184 * easy explanation as one segment stops on another.
185 */
186 for (i = 0; i < nr_segments; i++) {
187 unsigned long mstart, mend;
188 unsigned long j;
189
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
195 pstart = image->segment[j].mem;
196 pend = pstart + image->segment[j].memsz;
197 /* Do the segments overlap ? */
198 if ((mend > pstart) && (mstart < pend))
199 return -EINVAL;
200 }
201 }
202
203 /* Ensure our buffer sizes are strictly less than
204 * our memory sizes. This should always be the case,
205 * and it is easier to check up front than to be surprised
206 * later on.
207 */
208 for (i = 0; i < nr_segments; i++) {
209 if (image->segment[i].bufsz > image->segment[i].memsz)
210 return -EINVAL;
211 }
212
213 /*
214 * Verify that no more than half of memory will be consumed. If the
215 * request from userspace is too large, a large amount of time will be
216 * wasted allocating pages, which can cause a soft lockup.
217 */
218 for (i = 0; i < nr_segments; i++) {
219 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
220 return -EINVAL;
221
222 total_pages += PAGE_COUNT(image->segment[i].memsz);
223 }
224
225 if (total_pages > nr_pages / 2)
226 return -EINVAL;
227
228 /*
229 * Verify we have good destination addresses. Normally
230 * the caller is responsible for making certain we don't
231 * attempt to load the new image into invalid or reserved
232 * areas of RAM. But crash kernels are preloaded into a
233 * reserved area of ram. We must ensure the addresses
234 * are in the reserved area otherwise preloading the
235 * kernel could corrupt things.
236 */
237
238 if (image->type == KEXEC_TYPE_CRASH) {
239 for (i = 0; i < nr_segments; i++) {
240 unsigned long mstart, mend;
241
242 mstart = image->segment[i].mem;
243 mend = mstart + image->segment[i].memsz - 1;
244 /* Ensure we are within the crash kernel limits */
245 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
246 (mend > phys_to_boot_phys(crashk_res.end)))
247 return -EADDRNOTAVAIL;
248 }
249 }
250
251 return 0;
252}
253
254struct kimage *do_kimage_alloc_init(void)
255{
256 struct kimage *image;
257
258 /* Allocate a controlling structure */
259 image = kzalloc(sizeof(*image), GFP_KERNEL);
260 if (!image)
261 return NULL;
262
263 image->head = 0;
264 image->entry = &image->head;
265 image->last_entry = &image->head;
266 image->control_page = ~0; /* By default this does not apply */
267 image->type = KEXEC_TYPE_DEFAULT;
268
269 /* Initialize the list of control pages */
270 INIT_LIST_HEAD(&image->control_pages);
271
272 /* Initialize the list of destination pages */
273 INIT_LIST_HEAD(&image->dest_pages);
274
275 /* Initialize the list of unusable pages */
276 INIT_LIST_HEAD(&image->unusable_pages);
277
278 return image;
279}
280
281int kimage_is_destination_range(struct kimage *image,
282 unsigned long start,
283 unsigned long end)
284{
285 unsigned long i;
286
287 for (i = 0; i < image->nr_segments; i++) {
288 unsigned long mstart, mend;
289
290 mstart = image->segment[i].mem;
291 mend = mstart + image->segment[i].memsz;
292 if ((end > mstart) && (start < mend))
293 return 1;
294 }
295
296 return 0;
297}
298
299static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
300{
301 struct page *pages;
302
303 if (fatal_signal_pending(current))
304 return NULL;
305 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
306 if (pages) {
307 unsigned int count, i;
308
309 pages->mapping = NULL;
310 set_page_private(pages, order);
311 count = 1 << order;
312 for (i = 0; i < count; i++)
313 SetPageReserved(pages + i);
314
315 arch_kexec_post_alloc_pages(page_address(pages), count,
316 gfp_mask);
317
318 if (gfp_mask & __GFP_ZERO)
319 for (i = 0; i < count; i++)
320 clear_highpage(pages + i);
321 }
322
323 return pages;
324}
325
326static void kimage_free_pages(struct page *page)
327{
328 unsigned int order, count, i;
329
330 order = page_private(page);
331 count = 1 << order;
332
333 arch_kexec_pre_free_pages(page_address(page), count);
334
335 for (i = 0; i < count; i++)
336 ClearPageReserved(page + i);
337 __free_pages(page, order);
338}
339
340void kimage_free_page_list(struct list_head *list)
341{
342 struct page *page, *next;
343
344 list_for_each_entry_safe(page, next, list, lru) {
345 list_del(&page->lru);
346 kimage_free_pages(page);
347 }
348}
349
350static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
351 unsigned int order)
352{
353 /* Control pages are special, they are the intermediaries
354 * that are needed while we copy the rest of the pages
355 * to their final resting place. As such they must
356 * not conflict with either the destination addresses
357 * or memory the kernel is already using.
358 *
359 * The only case where we really need more than one of
360 * these are for architectures where we cannot disable
361 * the MMU and must instead generate an identity mapped
362 * page table for all of the memory.
363 *
364 * At worst this runs in O(N) of the image size.
365 */
366 struct list_head extra_pages;
367 struct page *pages;
368 unsigned int count;
369
370 count = 1 << order;
371 INIT_LIST_HEAD(&extra_pages);
372
373 /* Loop while I can allocate a page and the page allocated
374 * is a destination page.
375 */
376 do {
377 unsigned long pfn, epfn, addr, eaddr;
378
379 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
380 if (!pages)
381 break;
382 pfn = page_to_boot_pfn(pages);
383 epfn = pfn + count;
384 addr = pfn << PAGE_SHIFT;
385 eaddr = epfn << PAGE_SHIFT;
386 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
387 kimage_is_destination_range(image, addr, eaddr)) {
388 list_add(&pages->lru, &extra_pages);
389 pages = NULL;
390 }
391 } while (!pages);
392
393 if (pages) {
394 /* Remember the allocated page... */
395 list_add(&pages->lru, &image->control_pages);
396
397 /* Because the page is already in it's destination
398 * location we will never allocate another page at
399 * that address. Therefore kimage_alloc_pages
400 * will not return it (again) and we don't need
401 * to give it an entry in image->segment[].
402 */
403 }
404 /* Deal with the destination pages I have inadvertently allocated.
405 *
406 * Ideally I would convert multi-page allocations into single
407 * page allocations, and add everything to image->dest_pages.
408 *
409 * For now it is simpler to just free the pages.
410 */
411 kimage_free_page_list(&extra_pages);
412
413 return pages;
414}
415
416static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
417 unsigned int order)
418{
419 /* Control pages are special, they are the intermediaries
420 * that are needed while we copy the rest of the pages
421 * to their final resting place. As such they must
422 * not conflict with either the destination addresses
423 * or memory the kernel is already using.
424 *
425 * Control pages are also the only pags we must allocate
426 * when loading a crash kernel. All of the other pages
427 * are specified by the segments and we just memcpy
428 * into them directly.
429 *
430 * The only case where we really need more than one of
431 * these are for architectures where we cannot disable
432 * the MMU and must instead generate an identity mapped
433 * page table for all of the memory.
434 *
435 * Given the low demand this implements a very simple
436 * allocator that finds the first hole of the appropriate
437 * size in the reserved memory region, and allocates all
438 * of the memory up to and including the hole.
439 */
440 unsigned long hole_start, hole_end, size;
441 struct page *pages;
442
443 pages = NULL;
444 size = (1 << order) << PAGE_SHIFT;
445 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
446 hole_end = hole_start + size - 1;
447 while (hole_end <= crashk_res.end) {
448 unsigned long i;
449
450 cond_resched();
451
452 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
453 break;
454 /* See if I overlap any of the segments */
455 for (i = 0; i < image->nr_segments; i++) {
456 unsigned long mstart, mend;
457
458 mstart = image->segment[i].mem;
459 mend = mstart + image->segment[i].memsz - 1;
460 if ((hole_end >= mstart) && (hole_start <= mend)) {
461 /* Advance the hole to the end of the segment */
462 hole_start = (mend + (size - 1)) & ~(size - 1);
463 hole_end = hole_start + size - 1;
464 break;
465 }
466 }
467 /* If I don't overlap any segments I have found my hole! */
468 if (i == image->nr_segments) {
469 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
470 image->control_page = hole_end;
471 break;
472 }
473 }
474
475 /* Ensure that these pages are decrypted if SME is enabled. */
476 if (pages)
477 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
478
479 return pages;
480}
481
482
483struct page *kimage_alloc_control_pages(struct kimage *image,
484 unsigned int order)
485{
486 struct page *pages = NULL;
487
488 switch (image->type) {
489 case KEXEC_TYPE_DEFAULT:
490 pages = kimage_alloc_normal_control_pages(image, order);
491 break;
492 case KEXEC_TYPE_CRASH:
493 pages = kimage_alloc_crash_control_pages(image, order);
494 break;
495 }
496
497 return pages;
498}
499
500int kimage_crash_copy_vmcoreinfo(struct kimage *image)
501{
502 struct page *vmcoreinfo_page;
503 void *safecopy;
504
505 if (image->type != KEXEC_TYPE_CRASH)
506 return 0;
507
508 /*
509 * For kdump, allocate one vmcoreinfo safe copy from the
510 * crash memory. as we have arch_kexec_protect_crashkres()
511 * after kexec syscall, we naturally protect it from write
512 * (even read) access under kernel direct mapping. But on
513 * the other hand, we still need to operate it when crash
514 * happens to generate vmcoreinfo note, hereby we rely on
515 * vmap for this purpose.
516 */
517 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
518 if (!vmcoreinfo_page) {
519 pr_warn("Could not allocate vmcoreinfo buffer\n");
520 return -ENOMEM;
521 }
522 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
523 if (!safecopy) {
524 pr_warn("Could not vmap vmcoreinfo buffer\n");
525 return -ENOMEM;
526 }
527
528 image->vmcoreinfo_data_copy = safecopy;
529 crash_update_vmcoreinfo_safecopy(safecopy);
530
531 return 0;
532}
533
534static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
535{
536 if (*image->entry != 0)
537 image->entry++;
538
539 if (image->entry == image->last_entry) {
540 kimage_entry_t *ind_page;
541 struct page *page;
542
543 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
544 if (!page)
545 return -ENOMEM;
546
547 ind_page = page_address(page);
548 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
549 image->entry = ind_page;
550 image->last_entry = ind_page +
551 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
552 }
553 *image->entry = entry;
554 image->entry++;
555 *image->entry = 0;
556
557 return 0;
558}
559
560static int kimage_set_destination(struct kimage *image,
561 unsigned long destination)
562{
563 int result;
564
565 destination &= PAGE_MASK;
566 result = kimage_add_entry(image, destination | IND_DESTINATION);
567
568 return result;
569}
570
571
572static int kimage_add_page(struct kimage *image, unsigned long page)
573{
574 int result;
575
576 page &= PAGE_MASK;
577 result = kimage_add_entry(image, page | IND_SOURCE);
578
579 return result;
580}
581
582
583static void kimage_free_extra_pages(struct kimage *image)
584{
585 /* Walk through and free any extra destination pages I may have */
586 kimage_free_page_list(&image->dest_pages);
587
588 /* Walk through and free any unusable pages I have cached */
589 kimage_free_page_list(&image->unusable_pages);
590
591}
592void kimage_terminate(struct kimage *image)
593{
594 if (*image->entry != 0)
595 image->entry++;
596
597 *image->entry = IND_DONE;
598}
599
600#define for_each_kimage_entry(image, ptr, entry) \
601 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
602 ptr = (entry & IND_INDIRECTION) ? \
603 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
604
605static void kimage_free_entry(kimage_entry_t entry)
606{
607 struct page *page;
608
609 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
610 kimage_free_pages(page);
611}
612
613void kimage_free(struct kimage *image)
614{
615 kimage_entry_t *ptr, entry;
616 kimage_entry_t ind = 0;
617
618 if (!image)
619 return;
620
621 if (image->vmcoreinfo_data_copy) {
622 crash_update_vmcoreinfo_safecopy(NULL);
623 vunmap(image->vmcoreinfo_data_copy);
624 }
625
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 } else if (entry & IND_SOURCE)
637 kimage_free_entry(entry);
638 }
639 /* Free the final indirection page */
640 if (ind & IND_INDIRECTION)
641 kimage_free_entry(ind);
642
643 /* Handle any machine specific cleanup */
644 machine_kexec_cleanup(image);
645
646 /* Free the kexec control pages... */
647 kimage_free_page_list(&image->control_pages);
648
649 /*
650 * Free up any temporary buffers allocated. This might hit if
651 * error occurred much later after buffer allocation.
652 */
653 if (image->file_mode)
654 kimage_file_post_load_cleanup(image);
655
656 kfree(image);
657}
658
659static kimage_entry_t *kimage_dst_used(struct kimage *image,
660 unsigned long page)
661{
662 kimage_entry_t *ptr, entry;
663 unsigned long destination = 0;
664
665 for_each_kimage_entry(image, ptr, entry) {
666 if (entry & IND_DESTINATION)
667 destination = entry & PAGE_MASK;
668 else if (entry & IND_SOURCE) {
669 if (page == destination)
670 return ptr;
671 destination += PAGE_SIZE;
672 }
673 }
674
675 return NULL;
676}
677
678static struct page *kimage_alloc_page(struct kimage *image,
679 gfp_t gfp_mask,
680 unsigned long destination)
681{
682 /*
683 * Here we implement safeguards to ensure that a source page
684 * is not copied to its destination page before the data on
685 * the destination page is no longer useful.
686 *
687 * To do this we maintain the invariant that a source page is
688 * either its own destination page, or it is not a
689 * destination page at all.
690 *
691 * That is slightly stronger than required, but the proof
692 * that no problems will not occur is trivial, and the
693 * implementation is simply to verify.
694 *
695 * When allocating all pages normally this algorithm will run
696 * in O(N) time, but in the worst case it will run in O(N^2)
697 * time. If the runtime is a problem the data structures can
698 * be fixed.
699 */
700 struct page *page;
701 unsigned long addr;
702
703 /*
704 * Walk through the list of destination pages, and see if I
705 * have a match.
706 */
707 list_for_each_entry(page, &image->dest_pages, lru) {
708 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
709 if (addr == destination) {
710 list_del(&page->lru);
711 return page;
712 }
713 }
714 page = NULL;
715 while (1) {
716 kimage_entry_t *old;
717
718 /* Allocate a page, if we run out of memory give up */
719 page = kimage_alloc_pages(gfp_mask, 0);
720 if (!page)
721 return NULL;
722 /* If the page cannot be used file it away */
723 if (page_to_boot_pfn(page) >
724 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
725 list_add(&page->lru, &image->unusable_pages);
726 continue;
727 }
728 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
729
730 /* If it is the destination page we want use it */
731 if (addr == destination)
732 break;
733
734 /* If the page is not a destination page use it */
735 if (!kimage_is_destination_range(image, addr,
736 addr + PAGE_SIZE))
737 break;
738
739 /*
740 * I know that the page is someones destination page.
741 * See if there is already a source page for this
742 * destination page. And if so swap the source pages.
743 */
744 old = kimage_dst_used(image, addr);
745 if (old) {
746 /* If so move it */
747 unsigned long old_addr;
748 struct page *old_page;
749
750 old_addr = *old & PAGE_MASK;
751 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
752 copy_highpage(page, old_page);
753 *old = addr | (*old & ~PAGE_MASK);
754
755 /* The old page I have found cannot be a
756 * destination page, so return it if it's
757 * gfp_flags honor the ones passed in.
758 */
759 if (!(gfp_mask & __GFP_HIGHMEM) &&
760 PageHighMem(old_page)) {
761 kimage_free_pages(old_page);
762 continue;
763 }
764 addr = old_addr;
765 page = old_page;
766 break;
767 }
768 /* Place the page on the destination list, to be used later */
769 list_add(&page->lru, &image->dest_pages);
770 }
771
772 return page;
773}
774
775static int kimage_load_normal_segment(struct kimage *image,
776 struct kexec_segment *segment)
777{
778 unsigned long maddr;
779 size_t ubytes, mbytes;
780 int result;
781 unsigned char __user *buf = NULL;
782 unsigned char *kbuf = NULL;
783
784 result = 0;
785 if (image->file_mode)
786 kbuf = segment->kbuf;
787 else
788 buf = segment->buf;
789 ubytes = segment->bufsz;
790 mbytes = segment->memsz;
791 maddr = segment->mem;
792
793 result = kimage_set_destination(image, maddr);
794 if (result < 0)
795 goto out;
796
797 while (mbytes) {
798 struct page *page;
799 char *ptr;
800 size_t uchunk, mchunk;
801
802 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
803 if (!page) {
804 result = -ENOMEM;
805 goto out;
806 }
807 result = kimage_add_page(image, page_to_boot_pfn(page)
808 << PAGE_SHIFT);
809 if (result < 0)
810 goto out;
811
812 ptr = kmap(page);
813 /* Start with a clear page */
814 clear_page(ptr);
815 ptr += maddr & ~PAGE_MASK;
816 mchunk = min_t(size_t, mbytes,
817 PAGE_SIZE - (maddr & ~PAGE_MASK));
818 uchunk = min(ubytes, mchunk);
819
820 /* For file based kexec, source pages are in kernel memory */
821 if (image->file_mode)
822 memcpy(ptr, kbuf, uchunk);
823 else
824 result = copy_from_user(ptr, buf, uchunk);
825 kunmap(page);
826 if (result) {
827 result = -EFAULT;
828 goto out;
829 }
830 ubytes -= uchunk;
831 maddr += mchunk;
832 if (image->file_mode)
833 kbuf += mchunk;
834 else
835 buf += mchunk;
836 mbytes -= mchunk;
837
838 cond_resched();
839 }
840out:
841 return result;
842}
843
844static int kimage_load_crash_segment(struct kimage *image,
845 struct kexec_segment *segment)
846{
847 /* For crash dumps kernels we simply copy the data from
848 * user space to it's destination.
849 * We do things a page at a time for the sake of kmap.
850 */
851 unsigned long maddr;
852 size_t ubytes, mbytes;
853 int result;
854 unsigned char __user *buf = NULL;
855 unsigned char *kbuf = NULL;
856
857 result = 0;
858 if (image->file_mode)
859 kbuf = segment->kbuf;
860 else
861 buf = segment->buf;
862 ubytes = segment->bufsz;
863 mbytes = segment->memsz;
864 maddr = segment->mem;
865 while (mbytes) {
866 struct page *page;
867 char *ptr;
868 size_t uchunk, mchunk;
869
870 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
871 if (!page) {
872 result = -ENOMEM;
873 goto out;
874 }
875 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
876 ptr = kmap(page);
877 ptr += maddr & ~PAGE_MASK;
878 mchunk = min_t(size_t, mbytes,
879 PAGE_SIZE - (maddr & ~PAGE_MASK));
880 uchunk = min(ubytes, mchunk);
881 if (mchunk > uchunk) {
882 /* Zero the trailing part of the page */
883 memset(ptr + uchunk, 0, mchunk - uchunk);
884 }
885
886 /* For file based kexec, source pages are in kernel memory */
887 if (image->file_mode)
888 memcpy(ptr, kbuf, uchunk);
889 else
890 result = copy_from_user(ptr, buf, uchunk);
891 kexec_flush_icache_page(page);
892 kunmap(page);
893 arch_kexec_pre_free_pages(page_address(page), 1);
894 if (result) {
895 result = -EFAULT;
896 goto out;
897 }
898 ubytes -= uchunk;
899 maddr += mchunk;
900 if (image->file_mode)
901 kbuf += mchunk;
902 else
903 buf += mchunk;
904 mbytes -= mchunk;
905
906 cond_resched();
907 }
908out:
909 return result;
910}
911
912int kimage_load_segment(struct kimage *image,
913 struct kexec_segment *segment)
914{
915 int result = -ENOMEM;
916
917 switch (image->type) {
918 case KEXEC_TYPE_DEFAULT:
919 result = kimage_load_normal_segment(image, segment);
920 break;
921 case KEXEC_TYPE_CRASH:
922 result = kimage_load_crash_segment(image, segment);
923 break;
924 }
925
926 return result;
927}
928
929struct kimage *kexec_image;
930struct kimage *kexec_crash_image;
931int kexec_load_disabled;
932
933/*
934 * No panic_cpu check version of crash_kexec(). This function is called
935 * only when panic_cpu holds the current CPU number; this is the only CPU
936 * which processes crash_kexec routines.
937 */
938void __noclone __crash_kexec(struct pt_regs *regs)
939{
940 /* Take the kexec_mutex here to prevent sys_kexec_load
941 * running on one cpu from replacing the crash kernel
942 * we are using after a panic on a different cpu.
943 *
944 * If the crash kernel was not located in a fixed area
945 * of memory the xchg(&kexec_crash_image) would be
946 * sufficient. But since I reuse the memory...
947 */
948 if (mutex_trylock(&kexec_mutex)) {
949 if (kexec_crash_image) {
950 struct pt_regs fixed_regs;
951
952 crash_setup_regs(&fixed_regs, regs);
953 crash_save_vmcoreinfo();
954 machine_crash_shutdown(&fixed_regs);
955 machine_kexec(kexec_crash_image);
956 }
957 mutex_unlock(&kexec_mutex);
958 }
959}
960STACK_FRAME_NON_STANDARD(__crash_kexec);
961
962void crash_kexec(struct pt_regs *regs)
963{
964 int old_cpu, this_cpu;
965
966 /*
967 * Only one CPU is allowed to execute the crash_kexec() code as with
968 * panic(). Otherwise parallel calls of panic() and crash_kexec()
969 * may stop each other. To exclude them, we use panic_cpu here too.
970 */
971 this_cpu = raw_smp_processor_id();
972 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
973 if (old_cpu == PANIC_CPU_INVALID) {
974 /* This is the 1st CPU which comes here, so go ahead. */
975 printk_safe_flush_on_panic();
976 __crash_kexec(regs);
977
978 /*
979 * Reset panic_cpu to allow another panic()/crash_kexec()
980 * call.
981 */
982 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
983 }
984}
985
986size_t crash_get_memory_size(void)
987{
988 size_t size = 0;
989
990 mutex_lock(&kexec_mutex);
991 if (crashk_res.end != crashk_res.start)
992 size = resource_size(&crashk_res);
993 mutex_unlock(&kexec_mutex);
994 return size;
995}
996
997void __weak crash_free_reserved_phys_range(unsigned long begin,
998 unsigned long end)
999{
1000 unsigned long addr;
1001
1002 for (addr = begin; addr < end; addr += PAGE_SIZE)
1003 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
1004}
1005
1006int crash_shrink_memory(unsigned long new_size)
1007{
1008 int ret = 0;
1009 unsigned long start, end;
1010 unsigned long old_size;
1011 struct resource *ram_res;
1012
1013 mutex_lock(&kexec_mutex);
1014
1015 if (kexec_crash_image) {
1016 ret = -ENOENT;
1017 goto unlock;
1018 }
1019 start = crashk_res.start;
1020 end = crashk_res.end;
1021 old_size = (end == 0) ? 0 : end - start + 1;
1022 if (new_size >= old_size) {
1023 ret = (new_size == old_size) ? 0 : -EINVAL;
1024 goto unlock;
1025 }
1026
1027 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1028 if (!ram_res) {
1029 ret = -ENOMEM;
1030 goto unlock;
1031 }
1032
1033 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1034 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1035
1036 crash_free_reserved_phys_range(end, crashk_res.end);
1037
1038 if ((start == end) && (crashk_res.parent != NULL))
1039 release_resource(&crashk_res);
1040
1041 ram_res->start = end;
1042 ram_res->end = crashk_res.end;
1043 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1044 ram_res->name = "System RAM";
1045
1046 crashk_res.end = end - 1;
1047
1048 insert_resource(&iomem_resource, ram_res);
1049
1050unlock:
1051 mutex_unlock(&kexec_mutex);
1052 return ret;
1053}
1054
1055void crash_save_cpu(struct pt_regs *regs, int cpu)
1056{
1057 struct elf_prstatus prstatus;
1058 u32 *buf;
1059
1060 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1061 return;
1062
1063 /* Using ELF notes here is opportunistic.
1064 * I need a well defined structure format
1065 * for the data I pass, and I need tags
1066 * on the data to indicate what information I have
1067 * squirrelled away. ELF notes happen to provide
1068 * all of that, so there is no need to invent something new.
1069 */
1070 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1071 if (!buf)
1072 return;
1073 memset(&prstatus, 0, sizeof(prstatus));
1074 prstatus.pr_pid = current->pid;
1075 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1076 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1077 &prstatus, sizeof(prstatus));
1078 final_note(buf);
1079}
1080
1081static int __init crash_notes_memory_init(void)
1082{
1083 /* Allocate memory for saving cpu registers. */
1084 size_t size, align;
1085
1086 /*
1087 * crash_notes could be allocated across 2 vmalloc pages when percpu
1088 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1089 * pages are also on 2 continuous physical pages. In this case the
1090 * 2nd part of crash_notes in 2nd page could be lost since only the
1091 * starting address and size of crash_notes are exported through sysfs.
1092 * Here round up the size of crash_notes to the nearest power of two
1093 * and pass it to __alloc_percpu as align value. This can make sure
1094 * crash_notes is allocated inside one physical page.
1095 */
1096 size = sizeof(note_buf_t);
1097 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1098
1099 /*
1100 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1101 * definitely will be in 2 pages with that.
1102 */
1103 BUILD_BUG_ON(size > PAGE_SIZE);
1104
1105 crash_notes = __alloc_percpu(size, align);
1106 if (!crash_notes) {
1107 pr_warn("Memory allocation for saving cpu register states failed\n");
1108 return -ENOMEM;
1109 }
1110 return 0;
1111}
1112subsys_initcall(crash_notes_memory_init);
1113
1114
1115/*
1116 * Move into place and start executing a preloaded standalone
1117 * executable. If nothing was preloaded return an error.
1118 */
1119int kernel_kexec(void)
1120{
1121 int error = 0;
1122
1123 if (!mutex_trylock(&kexec_mutex))
1124 return -EBUSY;
1125 if (!kexec_image) {
1126 error = -EINVAL;
1127 goto Unlock;
1128 }
1129
1130#ifdef CONFIG_KEXEC_JUMP
1131 if (kexec_image->preserve_context) {
1132 lock_system_sleep();
1133 pm_prepare_console();
1134 error = freeze_processes();
1135 if (error) {
1136 error = -EBUSY;
1137 goto Restore_console;
1138 }
1139 suspend_console();
1140 error = dpm_suspend_start(PMSG_FREEZE);
1141 if (error)
1142 goto Resume_console;
1143 /* At this point, dpm_suspend_start() has been called,
1144 * but *not* dpm_suspend_end(). We *must* call
1145 * dpm_suspend_end() now. Otherwise, drivers for
1146 * some devices (e.g. interrupt controllers) become
1147 * desynchronized with the actual state of the
1148 * hardware at resume time, and evil weirdness ensues.
1149 */
1150 error = dpm_suspend_end(PMSG_FREEZE);
1151 if (error)
1152 goto Resume_devices;
1153 error = suspend_disable_secondary_cpus();
1154 if (error)
1155 goto Enable_cpus;
1156 local_irq_disable();
1157 error = syscore_suspend();
1158 if (error)
1159 goto Enable_irqs;
1160 } else
1161#endif
1162 {
1163 kexec_in_progress = true;
1164 kernel_restart_prepare(NULL);
1165 migrate_to_reboot_cpu();
1166
1167 /*
1168 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1169 * no further code needs to use CPU hotplug (which is true in
1170 * the reboot case). However, the kexec path depends on using
1171 * CPU hotplug again; so re-enable it here.
1172 */
1173 cpu_hotplug_enable();
1174 pr_emerg("Starting new kernel\n");
1175 machine_shutdown();
1176 }
1177
1178 machine_kexec(kexec_image);
1179
1180#ifdef CONFIG_KEXEC_JUMP
1181 if (kexec_image->preserve_context) {
1182 syscore_resume();
1183 Enable_irqs:
1184 local_irq_enable();
1185 Enable_cpus:
1186 suspend_enable_secondary_cpus();
1187 dpm_resume_start(PMSG_RESTORE);
1188 Resume_devices:
1189 dpm_resume_end(PMSG_RESTORE);
1190 Resume_console:
1191 resume_console();
1192 thaw_processes();
1193 Restore_console:
1194 pm_restore_console();
1195 unlock_system_sleep();
1196 }
1197#endif
1198
1199 Unlock:
1200 mutex_unlock(&kexec_mutex);
1201 return error;
1202}
1203
1204/*
1205 * Protection mechanism for crashkernel reserved memory after
1206 * the kdump kernel is loaded.
1207 *
1208 * Provide an empty default implementation here -- architecture
1209 * code may override this
1210 */
1211void __weak arch_kexec_protect_crashkres(void)
1212{}
1213
1214void __weak arch_kexec_unprotect_crashkres(void)
1215{}