Loading...
Note: File does not exist in v3.15.
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}
592
593int __weak machine_kexec_post_load(struct kimage *image)
594{
595 return 0;
596}
597
598void kimage_terminate(struct kimage *image)
599{
600 if (*image->entry != 0)
601 image->entry++;
602
603 *image->entry = IND_DONE;
604}
605
606#define for_each_kimage_entry(image, ptr, entry) \
607 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
608 ptr = (entry & IND_INDIRECTION) ? \
609 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
610
611static void kimage_free_entry(kimage_entry_t entry)
612{
613 struct page *page;
614
615 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
616 kimage_free_pages(page);
617}
618
619void kimage_free(struct kimage *image)
620{
621 kimage_entry_t *ptr, entry;
622 kimage_entry_t ind = 0;
623
624 if (!image)
625 return;
626
627 if (image->vmcoreinfo_data_copy) {
628 crash_update_vmcoreinfo_safecopy(NULL);
629 vunmap(image->vmcoreinfo_data_copy);
630 }
631
632 kimage_free_extra_pages(image);
633 for_each_kimage_entry(image, ptr, entry) {
634 if (entry & IND_INDIRECTION) {
635 /* Free the previous indirection page */
636 if (ind & IND_INDIRECTION)
637 kimage_free_entry(ind);
638 /* Save this indirection page until we are
639 * done with it.
640 */
641 ind = entry;
642 } else if (entry & IND_SOURCE)
643 kimage_free_entry(entry);
644 }
645 /* Free the final indirection page */
646 if (ind & IND_INDIRECTION)
647 kimage_free_entry(ind);
648
649 /* Handle any machine specific cleanup */
650 machine_kexec_cleanup(image);
651
652 /* Free the kexec control pages... */
653 kimage_free_page_list(&image->control_pages);
654
655 /*
656 * Free up any temporary buffers allocated. This might hit if
657 * error occurred much later after buffer allocation.
658 */
659 if (image->file_mode)
660 kimage_file_post_load_cleanup(image);
661
662 kfree(image);
663}
664
665static kimage_entry_t *kimage_dst_used(struct kimage *image,
666 unsigned long page)
667{
668 kimage_entry_t *ptr, entry;
669 unsigned long destination = 0;
670
671 for_each_kimage_entry(image, ptr, entry) {
672 if (entry & IND_DESTINATION)
673 destination = entry & PAGE_MASK;
674 else if (entry & IND_SOURCE) {
675 if (page == destination)
676 return ptr;
677 destination += PAGE_SIZE;
678 }
679 }
680
681 return NULL;
682}
683
684static struct page *kimage_alloc_page(struct kimage *image,
685 gfp_t gfp_mask,
686 unsigned long destination)
687{
688 /*
689 * Here we implement safeguards to ensure that a source page
690 * is not copied to its destination page before the data on
691 * the destination page is no longer useful.
692 *
693 * To do this we maintain the invariant that a source page is
694 * either its own destination page, or it is not a
695 * destination page at all.
696 *
697 * That is slightly stronger than required, but the proof
698 * that no problems will not occur is trivial, and the
699 * implementation is simply to verify.
700 *
701 * When allocating all pages normally this algorithm will run
702 * in O(N) time, but in the worst case it will run in O(N^2)
703 * time. If the runtime is a problem the data structures can
704 * be fixed.
705 */
706 struct page *page;
707 unsigned long addr;
708
709 /*
710 * Walk through the list of destination pages, and see if I
711 * have a match.
712 */
713 list_for_each_entry(page, &image->dest_pages, lru) {
714 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
715 if (addr == destination) {
716 list_del(&page->lru);
717 return page;
718 }
719 }
720 page = NULL;
721 while (1) {
722 kimage_entry_t *old;
723
724 /* Allocate a page, if we run out of memory give up */
725 page = kimage_alloc_pages(gfp_mask, 0);
726 if (!page)
727 return NULL;
728 /* If the page cannot be used file it away */
729 if (page_to_boot_pfn(page) >
730 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
731 list_add(&page->lru, &image->unusable_pages);
732 continue;
733 }
734 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
735
736 /* If it is the destination page we want use it */
737 if (addr == destination)
738 break;
739
740 /* If the page is not a destination page use it */
741 if (!kimage_is_destination_range(image, addr,
742 addr + PAGE_SIZE))
743 break;
744
745 /*
746 * I know that the page is someones destination page.
747 * See if there is already a source page for this
748 * destination page. And if so swap the source pages.
749 */
750 old = kimage_dst_used(image, addr);
751 if (old) {
752 /* If so move it */
753 unsigned long old_addr;
754 struct page *old_page;
755
756 old_addr = *old & PAGE_MASK;
757 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
758 copy_highpage(page, old_page);
759 *old = addr | (*old & ~PAGE_MASK);
760
761 /* The old page I have found cannot be a
762 * destination page, so return it if it's
763 * gfp_flags honor the ones passed in.
764 */
765 if (!(gfp_mask & __GFP_HIGHMEM) &&
766 PageHighMem(old_page)) {
767 kimage_free_pages(old_page);
768 continue;
769 }
770 addr = old_addr;
771 page = old_page;
772 break;
773 }
774 /* Place the page on the destination list, to be used later */
775 list_add(&page->lru, &image->dest_pages);
776 }
777
778 return page;
779}
780
781static int kimage_load_normal_segment(struct kimage *image,
782 struct kexec_segment *segment)
783{
784 unsigned long maddr;
785 size_t ubytes, mbytes;
786 int result;
787 unsigned char __user *buf = NULL;
788 unsigned char *kbuf = NULL;
789
790 result = 0;
791 if (image->file_mode)
792 kbuf = segment->kbuf;
793 else
794 buf = segment->buf;
795 ubytes = segment->bufsz;
796 mbytes = segment->memsz;
797 maddr = segment->mem;
798
799 result = kimage_set_destination(image, maddr);
800 if (result < 0)
801 goto out;
802
803 while (mbytes) {
804 struct page *page;
805 char *ptr;
806 size_t uchunk, mchunk;
807
808 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
809 if (!page) {
810 result = -ENOMEM;
811 goto out;
812 }
813 result = kimage_add_page(image, page_to_boot_pfn(page)
814 << PAGE_SHIFT);
815 if (result < 0)
816 goto out;
817
818 ptr = kmap(page);
819 /* Start with a clear page */
820 clear_page(ptr);
821 ptr += maddr & ~PAGE_MASK;
822 mchunk = min_t(size_t, mbytes,
823 PAGE_SIZE - (maddr & ~PAGE_MASK));
824 uchunk = min(ubytes, mchunk);
825
826 /* For file based kexec, source pages are in kernel memory */
827 if (image->file_mode)
828 memcpy(ptr, kbuf, uchunk);
829 else
830 result = copy_from_user(ptr, buf, uchunk);
831 kunmap(page);
832 if (result) {
833 result = -EFAULT;
834 goto out;
835 }
836 ubytes -= uchunk;
837 maddr += mchunk;
838 if (image->file_mode)
839 kbuf += mchunk;
840 else
841 buf += mchunk;
842 mbytes -= mchunk;
843
844 cond_resched();
845 }
846out:
847 return result;
848}
849
850static int kimage_load_crash_segment(struct kimage *image,
851 struct kexec_segment *segment)
852{
853 /* For crash dumps kernels we simply copy the data from
854 * user space to it's destination.
855 * We do things a page at a time for the sake of kmap.
856 */
857 unsigned long maddr;
858 size_t ubytes, mbytes;
859 int result;
860 unsigned char __user *buf = NULL;
861 unsigned char *kbuf = NULL;
862
863 result = 0;
864 if (image->file_mode)
865 kbuf = segment->kbuf;
866 else
867 buf = segment->buf;
868 ubytes = segment->bufsz;
869 mbytes = segment->memsz;
870 maddr = segment->mem;
871 while (mbytes) {
872 struct page *page;
873 char *ptr;
874 size_t uchunk, mchunk;
875
876 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
877 if (!page) {
878 result = -ENOMEM;
879 goto out;
880 }
881 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
882 ptr = kmap(page);
883 ptr += maddr & ~PAGE_MASK;
884 mchunk = min_t(size_t, mbytes,
885 PAGE_SIZE - (maddr & ~PAGE_MASK));
886 uchunk = min(ubytes, mchunk);
887 if (mchunk > uchunk) {
888 /* Zero the trailing part of the page */
889 memset(ptr + uchunk, 0, mchunk - uchunk);
890 }
891
892 /* For file based kexec, source pages are in kernel memory */
893 if (image->file_mode)
894 memcpy(ptr, kbuf, uchunk);
895 else
896 result = copy_from_user(ptr, buf, uchunk);
897 kexec_flush_icache_page(page);
898 kunmap(page);
899 arch_kexec_pre_free_pages(page_address(page), 1);
900 if (result) {
901 result = -EFAULT;
902 goto out;
903 }
904 ubytes -= uchunk;
905 maddr += mchunk;
906 if (image->file_mode)
907 kbuf += mchunk;
908 else
909 buf += mchunk;
910 mbytes -= mchunk;
911
912 cond_resched();
913 }
914out:
915 return result;
916}
917
918int kimage_load_segment(struct kimage *image,
919 struct kexec_segment *segment)
920{
921 int result = -ENOMEM;
922
923 switch (image->type) {
924 case KEXEC_TYPE_DEFAULT:
925 result = kimage_load_normal_segment(image, segment);
926 break;
927 case KEXEC_TYPE_CRASH:
928 result = kimage_load_crash_segment(image, segment);
929 break;
930 }
931
932 return result;
933}
934
935struct kimage *kexec_image;
936struct kimage *kexec_crash_image;
937int kexec_load_disabled;
938
939/*
940 * No panic_cpu check version of crash_kexec(). This function is called
941 * only when panic_cpu holds the current CPU number; this is the only CPU
942 * which processes crash_kexec routines.
943 */
944void __noclone __crash_kexec(struct pt_regs *regs)
945{
946 /* Take the kexec_mutex here to prevent sys_kexec_load
947 * running on one cpu from replacing the crash kernel
948 * we are using after a panic on a different cpu.
949 *
950 * If the crash kernel was not located in a fixed area
951 * of memory the xchg(&kexec_crash_image) would be
952 * sufficient. But since I reuse the memory...
953 */
954 if (mutex_trylock(&kexec_mutex)) {
955 if (kexec_crash_image) {
956 struct pt_regs fixed_regs;
957
958 crash_setup_regs(&fixed_regs, regs);
959 crash_save_vmcoreinfo();
960 machine_crash_shutdown(&fixed_regs);
961 machine_kexec(kexec_crash_image);
962 }
963 mutex_unlock(&kexec_mutex);
964 }
965}
966STACK_FRAME_NON_STANDARD(__crash_kexec);
967
968void crash_kexec(struct pt_regs *regs)
969{
970 int old_cpu, this_cpu;
971
972 /*
973 * Only one CPU is allowed to execute the crash_kexec() code as with
974 * panic(). Otherwise parallel calls of panic() and crash_kexec()
975 * may stop each other. To exclude them, we use panic_cpu here too.
976 */
977 this_cpu = raw_smp_processor_id();
978 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
979 if (old_cpu == PANIC_CPU_INVALID) {
980 /* This is the 1st CPU which comes here, so go ahead. */
981 printk_safe_flush_on_panic();
982 __crash_kexec(regs);
983
984 /*
985 * Reset panic_cpu to allow another panic()/crash_kexec()
986 * call.
987 */
988 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
989 }
990}
991
992size_t crash_get_memory_size(void)
993{
994 size_t size = 0;
995
996 mutex_lock(&kexec_mutex);
997 if (crashk_res.end != crashk_res.start)
998 size = resource_size(&crashk_res);
999 mutex_unlock(&kexec_mutex);
1000 return size;
1001}
1002
1003void __weak crash_free_reserved_phys_range(unsigned long begin,
1004 unsigned long end)
1005{
1006 unsigned long addr;
1007
1008 for (addr = begin; addr < end; addr += PAGE_SIZE)
1009 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
1010}
1011
1012int crash_shrink_memory(unsigned long new_size)
1013{
1014 int ret = 0;
1015 unsigned long start, end;
1016 unsigned long old_size;
1017 struct resource *ram_res;
1018
1019 mutex_lock(&kexec_mutex);
1020
1021 if (kexec_crash_image) {
1022 ret = -ENOENT;
1023 goto unlock;
1024 }
1025 start = crashk_res.start;
1026 end = crashk_res.end;
1027 old_size = (end == 0) ? 0 : end - start + 1;
1028 if (new_size >= old_size) {
1029 ret = (new_size == old_size) ? 0 : -EINVAL;
1030 goto unlock;
1031 }
1032
1033 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1034 if (!ram_res) {
1035 ret = -ENOMEM;
1036 goto unlock;
1037 }
1038
1039 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1040 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1041
1042 crash_free_reserved_phys_range(end, crashk_res.end);
1043
1044 if ((start == end) && (crashk_res.parent != NULL))
1045 release_resource(&crashk_res);
1046
1047 ram_res->start = end;
1048 ram_res->end = crashk_res.end;
1049 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
1050 ram_res->name = "System RAM";
1051
1052 crashk_res.end = end - 1;
1053
1054 insert_resource(&iomem_resource, ram_res);
1055
1056unlock:
1057 mutex_unlock(&kexec_mutex);
1058 return ret;
1059}
1060
1061void crash_save_cpu(struct pt_regs *regs, int cpu)
1062{
1063 struct elf_prstatus prstatus;
1064 u32 *buf;
1065
1066 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1067 return;
1068
1069 /* Using ELF notes here is opportunistic.
1070 * I need a well defined structure format
1071 * for the data I pass, and I need tags
1072 * on the data to indicate what information I have
1073 * squirrelled away. ELF notes happen to provide
1074 * all of that, so there is no need to invent something new.
1075 */
1076 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1077 if (!buf)
1078 return;
1079 memset(&prstatus, 0, sizeof(prstatus));
1080 prstatus.pr_pid = current->pid;
1081 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1082 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1083 &prstatus, sizeof(prstatus));
1084 final_note(buf);
1085}
1086
1087static int __init crash_notes_memory_init(void)
1088{
1089 /* Allocate memory for saving cpu registers. */
1090 size_t size, align;
1091
1092 /*
1093 * crash_notes could be allocated across 2 vmalloc pages when percpu
1094 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
1095 * pages are also on 2 continuous physical pages. In this case the
1096 * 2nd part of crash_notes in 2nd page could be lost since only the
1097 * starting address and size of crash_notes are exported through sysfs.
1098 * Here round up the size of crash_notes to the nearest power of two
1099 * and pass it to __alloc_percpu as align value. This can make sure
1100 * crash_notes is allocated inside one physical page.
1101 */
1102 size = sizeof(note_buf_t);
1103 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
1104
1105 /*
1106 * Break compile if size is bigger than PAGE_SIZE since crash_notes
1107 * definitely will be in 2 pages with that.
1108 */
1109 BUILD_BUG_ON(size > PAGE_SIZE);
1110
1111 crash_notes = __alloc_percpu(size, align);
1112 if (!crash_notes) {
1113 pr_warn("Memory allocation for saving cpu register states failed\n");
1114 return -ENOMEM;
1115 }
1116 return 0;
1117}
1118subsys_initcall(crash_notes_memory_init);
1119
1120
1121/*
1122 * Move into place and start executing a preloaded standalone
1123 * executable. If nothing was preloaded return an error.
1124 */
1125int kernel_kexec(void)
1126{
1127 int error = 0;
1128
1129 if (!mutex_trylock(&kexec_mutex))
1130 return -EBUSY;
1131 if (!kexec_image) {
1132 error = -EINVAL;
1133 goto Unlock;
1134 }
1135
1136#ifdef CONFIG_KEXEC_JUMP
1137 if (kexec_image->preserve_context) {
1138 lock_system_sleep();
1139 pm_prepare_console();
1140 error = freeze_processes();
1141 if (error) {
1142 error = -EBUSY;
1143 goto Restore_console;
1144 }
1145 suspend_console();
1146 error = dpm_suspend_start(PMSG_FREEZE);
1147 if (error)
1148 goto Resume_console;
1149 /* At this point, dpm_suspend_start() has been called,
1150 * but *not* dpm_suspend_end(). We *must* call
1151 * dpm_suspend_end() now. Otherwise, drivers for
1152 * some devices (e.g. interrupt controllers) become
1153 * desynchronized with the actual state of the
1154 * hardware at resume time, and evil weirdness ensues.
1155 */
1156 error = dpm_suspend_end(PMSG_FREEZE);
1157 if (error)
1158 goto Resume_devices;
1159 error = suspend_disable_secondary_cpus();
1160 if (error)
1161 goto Enable_cpus;
1162 local_irq_disable();
1163 error = syscore_suspend();
1164 if (error)
1165 goto Enable_irqs;
1166 } else
1167#endif
1168 {
1169 kexec_in_progress = true;
1170 kernel_restart_prepare(NULL);
1171 migrate_to_reboot_cpu();
1172
1173 /*
1174 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1175 * no further code needs to use CPU hotplug (which is true in
1176 * the reboot case). However, the kexec path depends on using
1177 * CPU hotplug again; so re-enable it here.
1178 */
1179 cpu_hotplug_enable();
1180 pr_notice("Starting new kernel\n");
1181 machine_shutdown();
1182 }
1183
1184 machine_kexec(kexec_image);
1185
1186#ifdef CONFIG_KEXEC_JUMP
1187 if (kexec_image->preserve_context) {
1188 syscore_resume();
1189 Enable_irqs:
1190 local_irq_enable();
1191 Enable_cpus:
1192 suspend_enable_secondary_cpus();
1193 dpm_resume_start(PMSG_RESTORE);
1194 Resume_devices:
1195 dpm_resume_end(PMSG_RESTORE);
1196 Resume_console:
1197 resume_console();
1198 thaw_processes();
1199 Restore_console:
1200 pm_restore_console();
1201 unlock_system_sleep();
1202 }
1203#endif
1204
1205 Unlock:
1206 mutex_unlock(&kexec_mutex);
1207 return error;
1208}
1209
1210/*
1211 * Protection mechanism for crashkernel reserved memory after
1212 * the kdump kernel is loaded.
1213 *
1214 * Provide an empty default implementation here -- architecture
1215 * code may override this
1216 */
1217void __weak arch_kexec_protect_crashkres(void)
1218{}
1219
1220void __weak arch_kexec_unprotect_crashkres(void)
1221{}