2 * kexec.c - kexec system call
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
9 #include <linux/capability.h>
11 #include <linux/file.h>
12 #include <linux/slab.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>
30 #include <linux/cpu.h>
31 #include <linux/console.h>
32 #include <linux/vmalloc.h>
33 #include <linux/swap.h>
34 #include <linux/syscore_ops.h>
37 #include <asm/uaccess.h>
39 #include <asm/sections.h>
41 /* Per cpu memory for storing cpu states in case of system crash. */
42 note_buf_t __percpu
*crash_notes
;
44 /* vmcoreinfo stuff */
45 static unsigned char vmcoreinfo_data
[VMCOREINFO_BYTES
];
46 u32 vmcoreinfo_note
[VMCOREINFO_NOTE_SIZE
/4];
47 size_t vmcoreinfo_size
;
48 size_t vmcoreinfo_max_size
= sizeof(vmcoreinfo_data
);
50 /* Location of the reserved area for the crash kernel */
51 struct resource crashk_res
= {
52 .name
= "Crash kernel",
55 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
57 struct resource crashk_low_res
= {
58 .name
= "Crash kernel low",
61 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
64 int kexec_should_crash(struct task_struct
*p
)
66 if (in_interrupt() || !p
->pid
|| is_global_init(p
) || panic_on_oops
)
72 * When kexec transitions to the new kernel there is a one-to-one
73 * mapping between physical and virtual addresses. On processors
74 * where you can disable the MMU this is trivial, and easy. For
75 * others it is still a simple predictable page table to setup.
77 * In that environment kexec copies the new kernel to its final
78 * resting place. This means I can only support memory whose
79 * physical address can fit in an unsigned long. In particular
80 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
81 * If the assembly stub has more restrictive requirements
82 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
83 * defined more restrictively in <asm/kexec.h>.
85 * The code for the transition from the current kernel to the
86 * the new kernel is placed in the control_code_buffer, whose size
87 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
88 * page of memory is necessary, but some architectures require more.
89 * Because this memory must be identity mapped in the transition from
90 * virtual to physical addresses it must live in the range
91 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
94 * The assembly stub in the control code buffer is passed a linked list
95 * of descriptor pages detailing the source pages of the new kernel,
96 * and the destination addresses of those source pages. As this data
97 * structure is not used in the context of the current OS, it must
100 * The code has been made to work with highmem pages and will use a
101 * destination page in its final resting place (if it happens
102 * to allocate it). The end product of this is that most of the
103 * physical address space, and most of RAM can be used.
105 * Future directions include:
106 * - allocating a page table with the control code buffer identity
107 * mapped, to simplify machine_kexec and make kexec_on_panic more
112 * KIMAGE_NO_DEST is an impossible destination address..., for
113 * allocating pages whose destination address we do not care about.
115 #define KIMAGE_NO_DEST (-1UL)
117 static int kimage_is_destination_range(struct kimage
*image
,
118 unsigned long start
, unsigned long end
);
119 static struct page
*kimage_alloc_page(struct kimage
*image
,
123 static int do_kimage_alloc(struct kimage
**rimage
, unsigned long entry
,
124 unsigned long nr_segments
,
125 struct kexec_segment __user
*segments
)
127 size_t segment_bytes
;
128 struct kimage
*image
;
132 /* Allocate a controlling structure */
134 image
= kzalloc(sizeof(*image
), GFP_KERNEL
);
139 image
->entry
= &image
->head
;
140 image
->last_entry
= &image
->head
;
141 image
->control_page
= ~0; /* By default this does not apply */
142 image
->start
= entry
;
143 image
->type
= KEXEC_TYPE_DEFAULT
;
145 /* Initialize the list of control pages */
146 INIT_LIST_HEAD(&image
->control_pages
);
148 /* Initialize the list of destination pages */
149 INIT_LIST_HEAD(&image
->dest_pages
);
151 /* Initialize the list of unusable pages */
152 INIT_LIST_HEAD(&image
->unuseable_pages
);
154 /* Read in the segments */
155 image
->nr_segments
= nr_segments
;
156 segment_bytes
= nr_segments
* sizeof(*segments
);
157 result
= copy_from_user(image
->segment
, segments
, segment_bytes
);
164 * Verify we have good destination addresses. The caller is
165 * responsible for making certain we don't attempt to load
166 * the new image into invalid or reserved areas of RAM. This
167 * just verifies it is an address we can use.
169 * Since the kernel does everything in page size chunks ensure
170 * the destination addresses are page aligned. Too many
171 * special cases crop of when we don't do this. The most
172 * insidious is getting overlapping destination addresses
173 * simply because addresses are changed to page size
176 result
= -EADDRNOTAVAIL
;
177 for (i
= 0; i
< nr_segments
; i
++) {
178 unsigned long mstart
, mend
;
180 mstart
= image
->segment
[i
].mem
;
181 mend
= mstart
+ image
->segment
[i
].memsz
;
182 if ((mstart
& ~PAGE_MASK
) || (mend
& ~PAGE_MASK
))
184 if (mend
>= KEXEC_DESTINATION_MEMORY_LIMIT
)
188 /* Verify our destination addresses do not overlap.
189 * If we alloed overlapping destination addresses
190 * through very weird things can happen with no
191 * easy explanation as one segment stops on another.
194 for (i
= 0; i
< nr_segments
; i
++) {
195 unsigned long mstart
, mend
;
198 mstart
= image
->segment
[i
].mem
;
199 mend
= mstart
+ image
->segment
[i
].memsz
;
200 for (j
= 0; j
< i
; j
++) {
201 unsigned long pstart
, pend
;
202 pstart
= image
->segment
[j
].mem
;
203 pend
= pstart
+ image
->segment
[j
].memsz
;
204 /* Do the segments overlap ? */
205 if ((mend
> pstart
) && (mstart
< pend
))
210 /* Ensure our buffer sizes are strictly less than
211 * our memory sizes. This should always be the case,
212 * and it is easier to check up front than to be surprised
216 for (i
= 0; i
< nr_segments
; i
++) {
217 if (image
->segment
[i
].bufsz
> image
->segment
[i
].memsz
)
232 static int kimage_normal_alloc(struct kimage
**rimage
, unsigned long entry
,
233 unsigned long nr_segments
,
234 struct kexec_segment __user
*segments
)
237 struct kimage
*image
;
239 /* Allocate and initialize a controlling structure */
241 result
= do_kimage_alloc(&image
, entry
, nr_segments
, segments
);
248 * Find a location for the control code buffer, and add it
249 * the vector of segments so that it's pages will also be
250 * counted as destination pages.
253 image
->control_code_page
= kimage_alloc_control_pages(image
,
254 get_order(KEXEC_CONTROL_PAGE_SIZE
));
255 if (!image
->control_code_page
) {
256 printk(KERN_ERR
"Could not allocate control_code_buffer\n");
260 image
->swap_page
= kimage_alloc_control_pages(image
, 0);
261 if (!image
->swap_page
) {
262 printk(KERN_ERR
"Could not allocate swap buffer\n");
276 static int kimage_crash_alloc(struct kimage
**rimage
, unsigned long entry
,
277 unsigned long nr_segments
,
278 struct kexec_segment __user
*segments
)
281 struct kimage
*image
;
285 /* Verify we have a valid entry point */
286 if ((entry
< crashk_res
.start
) || (entry
> crashk_res
.end
)) {
287 result
= -EADDRNOTAVAIL
;
291 /* Allocate and initialize a controlling structure */
292 result
= do_kimage_alloc(&image
, entry
, nr_segments
, segments
);
296 /* Enable the special crash kernel control page
299 image
->control_page
= crashk_res
.start
;
300 image
->type
= KEXEC_TYPE_CRASH
;
303 * Verify we have good destination addresses. Normally
304 * the caller is responsible for making certain we don't
305 * attempt to load the new image into invalid or reserved
306 * areas of RAM. But crash kernels are preloaded into a
307 * reserved area of ram. We must ensure the addresses
308 * are in the reserved area otherwise preloading the
309 * kernel could corrupt things.
311 result
= -EADDRNOTAVAIL
;
312 for (i
= 0; i
< nr_segments
; i
++) {
313 unsigned long mstart
, mend
;
315 mstart
= image
->segment
[i
].mem
;
316 mend
= mstart
+ image
->segment
[i
].memsz
- 1;
317 /* Ensure we are within the crash kernel limits */
318 if ((mstart
< crashk_res
.start
) || (mend
> crashk_res
.end
))
323 * Find a location for the control code buffer, and add
324 * the vector of segments so that it's pages will also be
325 * counted as destination pages.
328 image
->control_code_page
= kimage_alloc_control_pages(image
,
329 get_order(KEXEC_CONTROL_PAGE_SIZE
));
330 if (!image
->control_code_page
) {
331 printk(KERN_ERR
"Could not allocate control_code_buffer\n");
345 static int kimage_is_destination_range(struct kimage
*image
,
351 for (i
= 0; i
< image
->nr_segments
; i
++) {
352 unsigned long mstart
, mend
;
354 mstart
= image
->segment
[i
].mem
;
355 mend
= mstart
+ image
->segment
[i
].memsz
;
356 if ((end
> mstart
) && (start
< mend
))
363 static struct page
*kimage_alloc_pages(gfp_t gfp_mask
, unsigned int order
)
367 pages
= alloc_pages(gfp_mask
, order
);
369 unsigned int count
, i
;
370 pages
->mapping
= NULL
;
371 set_page_private(pages
, order
);
373 for (i
= 0; i
< count
; i
++)
374 SetPageReserved(pages
+ i
);
380 static void kimage_free_pages(struct page
*page
)
382 unsigned int order
, count
, i
;
384 order
= page_private(page
);
386 for (i
= 0; i
< count
; i
++)
387 ClearPageReserved(page
+ i
);
388 __free_pages(page
, order
);
391 static void kimage_free_page_list(struct list_head
*list
)
393 struct list_head
*pos
, *next
;
395 list_for_each_safe(pos
, next
, list
) {
398 page
= list_entry(pos
, struct page
, lru
);
399 list_del(&page
->lru
);
400 kimage_free_pages(page
);
404 static struct page
*kimage_alloc_normal_control_pages(struct kimage
*image
,
407 /* Control pages are special, they are the intermediaries
408 * that are needed while we copy the rest of the pages
409 * to their final resting place. As such they must
410 * not conflict with either the destination addresses
411 * or memory the kernel is already using.
413 * The only case where we really need more than one of
414 * these are for architectures where we cannot disable
415 * the MMU and must instead generate an identity mapped
416 * page table for all of the memory.
418 * At worst this runs in O(N) of the image size.
420 struct list_head extra_pages
;
425 INIT_LIST_HEAD(&extra_pages
);
427 /* Loop while I can allocate a page and the page allocated
428 * is a destination page.
431 unsigned long pfn
, epfn
, addr
, eaddr
;
433 pages
= kimage_alloc_pages(GFP_KERNEL
, order
);
436 pfn
= page_to_pfn(pages
);
438 addr
= pfn
<< PAGE_SHIFT
;
439 eaddr
= epfn
<< PAGE_SHIFT
;
440 if ((epfn
>= (KEXEC_CONTROL_MEMORY_LIMIT
>> PAGE_SHIFT
)) ||
441 kimage_is_destination_range(image
, addr
, eaddr
)) {
442 list_add(&pages
->lru
, &extra_pages
);
448 /* Remember the allocated page... */
449 list_add(&pages
->lru
, &image
->control_pages
);
451 /* Because the page is already in it's destination
452 * location we will never allocate another page at
453 * that address. Therefore kimage_alloc_pages
454 * will not return it (again) and we don't need
455 * to give it an entry in image->segment[].
458 /* Deal with the destination pages I have inadvertently allocated.
460 * Ideally I would convert multi-page allocations into single
461 * page allocations, and add everything to image->dest_pages.
463 * For now it is simpler to just free the pages.
465 kimage_free_page_list(&extra_pages
);
470 static struct page
*kimage_alloc_crash_control_pages(struct kimage
*image
,
473 /* Control pages are special, they are the intermediaries
474 * that are needed while we copy the rest of the pages
475 * to their final resting place. As such they must
476 * not conflict with either the destination addresses
477 * or memory the kernel is already using.
479 * Control pages are also the only pags we must allocate
480 * when loading a crash kernel. All of the other pages
481 * are specified by the segments and we just memcpy
482 * into them directly.
484 * The only case where we really need more than one of
485 * these are for architectures where we cannot disable
486 * the MMU and must instead generate an identity mapped
487 * page table for all of the memory.
489 * Given the low demand this implements a very simple
490 * allocator that finds the first hole of the appropriate
491 * size in the reserved memory region, and allocates all
492 * of the memory up to and including the hole.
494 unsigned long hole_start
, hole_end
, size
;
498 size
= (1 << order
) << PAGE_SHIFT
;
499 hole_start
= (image
->control_page
+ (size
- 1)) & ~(size
- 1);
500 hole_end
= hole_start
+ size
- 1;
501 while (hole_end
<= crashk_res
.end
) {
504 if (hole_end
> KEXEC_CRASH_CONTROL_MEMORY_LIMIT
)
506 /* See if I overlap any of the segments */
507 for (i
= 0; i
< image
->nr_segments
; i
++) {
508 unsigned long mstart
, mend
;
510 mstart
= image
->segment
[i
].mem
;
511 mend
= mstart
+ image
->segment
[i
].memsz
- 1;
512 if ((hole_end
>= mstart
) && (hole_start
<= mend
)) {
513 /* Advance the hole to the end of the segment */
514 hole_start
= (mend
+ (size
- 1)) & ~(size
- 1);
515 hole_end
= hole_start
+ size
- 1;
519 /* If I don't overlap any segments I have found my hole! */
520 if (i
== image
->nr_segments
) {
521 pages
= pfn_to_page(hole_start
>> PAGE_SHIFT
);
526 image
->control_page
= hole_end
;
532 struct page
*kimage_alloc_control_pages(struct kimage
*image
,
535 struct page
*pages
= NULL
;
537 switch (image
->type
) {
538 case KEXEC_TYPE_DEFAULT
:
539 pages
= kimage_alloc_normal_control_pages(image
, order
);
541 case KEXEC_TYPE_CRASH
:
542 pages
= kimage_alloc_crash_control_pages(image
, order
);
549 static int kimage_add_entry(struct kimage
*image
, kimage_entry_t entry
)
551 if (*image
->entry
!= 0)
554 if (image
->entry
== image
->last_entry
) {
555 kimage_entry_t
*ind_page
;
558 page
= kimage_alloc_page(image
, GFP_KERNEL
, KIMAGE_NO_DEST
);
562 ind_page
= page_address(page
);
563 *image
->entry
= virt_to_phys(ind_page
) | IND_INDIRECTION
;
564 image
->entry
= ind_page
;
565 image
->last_entry
= ind_page
+
566 ((PAGE_SIZE
/sizeof(kimage_entry_t
)) - 1);
568 *image
->entry
= entry
;
575 static int kimage_set_destination(struct kimage
*image
,
576 unsigned long destination
)
580 destination
&= PAGE_MASK
;
581 result
= kimage_add_entry(image
, destination
| IND_DESTINATION
);
583 image
->destination
= destination
;
589 static int kimage_add_page(struct kimage
*image
, unsigned long page
)
594 result
= kimage_add_entry(image
, page
| IND_SOURCE
);
596 image
->destination
+= PAGE_SIZE
;
602 static void kimage_free_extra_pages(struct kimage
*image
)
604 /* Walk through and free any extra destination pages I may have */
605 kimage_free_page_list(&image
->dest_pages
);
607 /* Walk through and free any unusable pages I have cached */
608 kimage_free_page_list(&image
->unuseable_pages
);
611 static void kimage_terminate(struct kimage
*image
)
613 if (*image
->entry
!= 0)
616 *image
->entry
= IND_DONE
;
619 #define for_each_kimage_entry(image, ptr, entry) \
620 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
621 ptr = (entry & IND_INDIRECTION)? \
622 phys_to_virt((entry & PAGE_MASK)): ptr +1)
624 static void kimage_free_entry(kimage_entry_t entry
)
628 page
= pfn_to_page(entry
>> PAGE_SHIFT
);
629 kimage_free_pages(page
);
632 static void kimage_free(struct kimage
*image
)
634 kimage_entry_t
*ptr
, entry
;
635 kimage_entry_t ind
= 0;
640 kimage_free_extra_pages(image
);
641 for_each_kimage_entry(image
, ptr
, entry
) {
642 if (entry
& IND_INDIRECTION
) {
643 /* Free the previous indirection page */
644 if (ind
& IND_INDIRECTION
)
645 kimage_free_entry(ind
);
646 /* Save this indirection page until we are
651 else if (entry
& IND_SOURCE
)
652 kimage_free_entry(entry
);
654 /* Free the final indirection page */
655 if (ind
& IND_INDIRECTION
)
656 kimage_free_entry(ind
);
658 /* Handle any machine specific cleanup */
659 machine_kexec_cleanup(image
);
661 /* Free the kexec control pages... */
662 kimage_free_page_list(&image
->control_pages
);
666 static kimage_entry_t
*kimage_dst_used(struct kimage
*image
,
669 kimage_entry_t
*ptr
, entry
;
670 unsigned long destination
= 0;
672 for_each_kimage_entry(image
, ptr
, entry
) {
673 if (entry
& IND_DESTINATION
)
674 destination
= entry
& PAGE_MASK
;
675 else if (entry
& IND_SOURCE
) {
676 if (page
== destination
)
678 destination
+= PAGE_SIZE
;
685 static struct page
*kimage_alloc_page(struct kimage
*image
,
687 unsigned long destination
)
690 * Here we implement safeguards to ensure that a source page
691 * is not copied to its destination page before the data on
692 * the destination page is no longer useful.
694 * To do this we maintain the invariant that a source page is
695 * either its own destination page, or it is not a
696 * destination page at all.
698 * That is slightly stronger than required, but the proof
699 * that no problems will not occur is trivial, and the
700 * implementation is simply to verify.
702 * When allocating all pages normally this algorithm will run
703 * in O(N) time, but in the worst case it will run in O(N^2)
704 * time. If the runtime is a problem the data structures can
711 * Walk through the list of destination pages, and see if I
714 list_for_each_entry(page
, &image
->dest_pages
, lru
) {
715 addr
= page_to_pfn(page
) << PAGE_SHIFT
;
716 if (addr
== destination
) {
717 list_del(&page
->lru
);
725 /* Allocate a page, if we run out of memory give up */
726 page
= kimage_alloc_pages(gfp_mask
, 0);
729 /* If the page cannot be used file it away */
730 if (page_to_pfn(page
) >
731 (KEXEC_SOURCE_MEMORY_LIMIT
>> PAGE_SHIFT
)) {
732 list_add(&page
->lru
, &image
->unuseable_pages
);
735 addr
= page_to_pfn(page
) << PAGE_SHIFT
;
737 /* If it is the destination page we want use it */
738 if (addr
== destination
)
741 /* If the page is not a destination page use it */
742 if (!kimage_is_destination_range(image
, addr
,
747 * I know that the page is someones destination page.
748 * See if there is already a source page for this
749 * destination page. And if so swap the source pages.
751 old
= kimage_dst_used(image
, addr
);
754 unsigned long old_addr
;
755 struct page
*old_page
;
757 old_addr
= *old
& PAGE_MASK
;
758 old_page
= pfn_to_page(old_addr
>> PAGE_SHIFT
);
759 copy_highpage(page
, old_page
);
760 *old
= addr
| (*old
& ~PAGE_MASK
);
762 /* The old page I have found cannot be a
763 * destination page, so return it if it's
764 * gfp_flags honor the ones passed in.
766 if (!(gfp_mask
& __GFP_HIGHMEM
) &&
767 PageHighMem(old_page
)) {
768 kimage_free_pages(old_page
);
776 /* Place the page on the destination list I
779 list_add(&page
->lru
, &image
->dest_pages
);
786 static int kimage_load_normal_segment(struct kimage
*image
,
787 struct kexec_segment
*segment
)
790 unsigned long ubytes
, mbytes
;
792 unsigned char __user
*buf
;
796 ubytes
= segment
->bufsz
;
797 mbytes
= segment
->memsz
;
798 maddr
= segment
->mem
;
800 result
= kimage_set_destination(image
, maddr
);
807 size_t uchunk
, mchunk
;
809 page
= kimage_alloc_page(image
, GFP_HIGHUSER
, maddr
);
814 result
= kimage_add_page(image
, page_to_pfn(page
)
820 /* Start with a clear page */
822 ptr
+= maddr
& ~PAGE_MASK
;
823 mchunk
= PAGE_SIZE
- (maddr
& ~PAGE_MASK
);
831 result
= copy_from_user(ptr
, buf
, uchunk
);
846 static int kimage_load_crash_segment(struct kimage
*image
,
847 struct kexec_segment
*segment
)
849 /* For crash dumps kernels we simply copy the data from
850 * user space to it's destination.
851 * We do things a page at a time for the sake of kmap.
854 unsigned long ubytes
, mbytes
;
856 unsigned char __user
*buf
;
860 ubytes
= segment
->bufsz
;
861 mbytes
= segment
->memsz
;
862 maddr
= segment
->mem
;
866 size_t uchunk
, mchunk
;
868 page
= pfn_to_page(maddr
>> PAGE_SHIFT
);
874 ptr
+= maddr
& ~PAGE_MASK
;
875 mchunk
= PAGE_SIZE
- (maddr
& ~PAGE_MASK
);
880 if (uchunk
> ubytes
) {
882 /* Zero the trailing part of the page */
883 memset(ptr
+ uchunk
, 0, mchunk
- uchunk
);
885 result
= copy_from_user(ptr
, buf
, uchunk
);
886 kexec_flush_icache_page(page
);
901 static int kimage_load_segment(struct kimage
*image
,
902 struct kexec_segment
*segment
)
904 int result
= -ENOMEM
;
906 switch (image
->type
) {
907 case KEXEC_TYPE_DEFAULT
:
908 result
= kimage_load_normal_segment(image
, segment
);
910 case KEXEC_TYPE_CRASH
:
911 result
= kimage_load_crash_segment(image
, segment
);
919 * Exec Kernel system call: for obvious reasons only root may call it.
921 * This call breaks up into three pieces.
922 * - A generic part which loads the new kernel from the current
923 * address space, and very carefully places the data in the
926 * - A generic part that interacts with the kernel and tells all of
927 * the devices to shut down. Preventing on-going dmas, and placing
928 * the devices in a consistent state so a later kernel can
931 * - A machine specific part that includes the syscall number
932 * and the copies the image to it's final destination. And
933 * jumps into the image at entry.
935 * kexec does not sync, or unmount filesystems so if you need
936 * that to happen you need to do that yourself.
938 struct kimage
*kexec_image
;
939 struct kimage
*kexec_crash_image
;
941 static DEFINE_MUTEX(kexec_mutex
);
943 SYSCALL_DEFINE4(kexec_load
, unsigned long, entry
, unsigned long, nr_segments
,
944 struct kexec_segment __user
*, segments
, unsigned long, flags
)
946 struct kimage
**dest_image
, *image
;
949 /* We only trust the superuser with rebooting the system. */
950 if (!capable(CAP_SYS_BOOT
))
954 * Verify we have a legal set of flags
955 * This leaves us room for future extensions.
957 if ((flags
& KEXEC_FLAGS
) != (flags
& ~KEXEC_ARCH_MASK
))
960 /* Verify we are on the appropriate architecture */
961 if (((flags
& KEXEC_ARCH_MASK
) != KEXEC_ARCH
) &&
962 ((flags
& KEXEC_ARCH_MASK
) != KEXEC_ARCH_DEFAULT
))
965 /* Put an artificial cap on the number
966 * of segments passed to kexec_load.
968 if (nr_segments
> KEXEC_SEGMENT_MAX
)
974 /* Because we write directly to the reserved memory
975 * region when loading crash kernels we need a mutex here to
976 * prevent multiple crash kernels from attempting to load
977 * simultaneously, and to prevent a crash kernel from loading
978 * over the top of a in use crash kernel.
980 * KISS: always take the mutex.
982 if (!mutex_trylock(&kexec_mutex
))
985 dest_image
= &kexec_image
;
986 if (flags
& KEXEC_ON_CRASH
)
987 dest_image
= &kexec_crash_image
;
988 if (nr_segments
> 0) {
991 /* Loading another kernel to reboot into */
992 if ((flags
& KEXEC_ON_CRASH
) == 0)
993 result
= kimage_normal_alloc(&image
, entry
,
994 nr_segments
, segments
);
995 /* Loading another kernel to switch to if this one crashes */
996 else if (flags
& KEXEC_ON_CRASH
) {
997 /* Free any current crash dump kernel before
1000 kimage_free(xchg(&kexec_crash_image
, NULL
));
1001 result
= kimage_crash_alloc(&image
, entry
,
1002 nr_segments
, segments
);
1003 crash_map_reserved_pages();
1008 if (flags
& KEXEC_PRESERVE_CONTEXT
)
1009 image
->preserve_context
= 1;
1010 result
= machine_kexec_prepare(image
);
1014 for (i
= 0; i
< nr_segments
; i
++) {
1015 result
= kimage_load_segment(image
, &image
->segment
[i
]);
1019 kimage_terminate(image
);
1020 if (flags
& KEXEC_ON_CRASH
)
1021 crash_unmap_reserved_pages();
1023 /* Install the new kernel, and Uninstall the old */
1024 image
= xchg(dest_image
, image
);
1027 mutex_unlock(&kexec_mutex
);
1034 * Add and remove page tables for crashkernel memory
1036 * Provide an empty default implementation here -- architecture
1037 * code may override this
1039 void __weak
crash_map_reserved_pages(void)
1042 void __weak
crash_unmap_reserved_pages(void)
1045 #ifdef CONFIG_COMPAT
1046 asmlinkage
long compat_sys_kexec_load(unsigned long entry
,
1047 unsigned long nr_segments
,
1048 struct compat_kexec_segment __user
*segments
,
1049 unsigned long flags
)
1051 struct compat_kexec_segment in
;
1052 struct kexec_segment out
, __user
*ksegments
;
1053 unsigned long i
, result
;
1055 /* Don't allow clients that don't understand the native
1056 * architecture to do anything.
1058 if ((flags
& KEXEC_ARCH_MASK
) == KEXEC_ARCH_DEFAULT
)
1061 if (nr_segments
> KEXEC_SEGMENT_MAX
)
1064 ksegments
= compat_alloc_user_space(nr_segments
* sizeof(out
));
1065 for (i
=0; i
< nr_segments
; i
++) {
1066 result
= copy_from_user(&in
, &segments
[i
], sizeof(in
));
1070 out
.buf
= compat_ptr(in
.buf
);
1071 out
.bufsz
= in
.bufsz
;
1073 out
.memsz
= in
.memsz
;
1075 result
= copy_to_user(&ksegments
[i
], &out
, sizeof(out
));
1080 return sys_kexec_load(entry
, nr_segments
, ksegments
, flags
);
1084 void crash_kexec(struct pt_regs
*regs
)
1086 /* Take the kexec_mutex here to prevent sys_kexec_load
1087 * running on one cpu from replacing the crash kernel
1088 * we are using after a panic on a different cpu.
1090 * If the crash kernel was not located in a fixed area
1091 * of memory the xchg(&kexec_crash_image) would be
1092 * sufficient. But since I reuse the memory...
1094 if (mutex_trylock(&kexec_mutex
)) {
1095 if (kexec_crash_image
) {
1096 struct pt_regs fixed_regs
;
1098 crash_setup_regs(&fixed_regs
, regs
);
1099 crash_save_vmcoreinfo();
1100 machine_crash_shutdown(&fixed_regs
);
1101 machine_kexec(kexec_crash_image
);
1103 mutex_unlock(&kexec_mutex
);
1107 size_t crash_get_memory_size(void)
1110 mutex_lock(&kexec_mutex
);
1111 if (crashk_res
.end
!= crashk_res
.start
)
1112 size
= resource_size(&crashk_res
);
1113 mutex_unlock(&kexec_mutex
);
1117 void __weak
crash_free_reserved_phys_range(unsigned long begin
,
1122 for (addr
= begin
; addr
< end
; addr
+= PAGE_SIZE
) {
1123 ClearPageReserved(pfn_to_page(addr
>> PAGE_SHIFT
));
1124 init_page_count(pfn_to_page(addr
>> PAGE_SHIFT
));
1125 free_page((unsigned long)__va(addr
));
1130 int crash_shrink_memory(unsigned long new_size
)
1133 unsigned long start
, end
;
1134 unsigned long old_size
;
1135 struct resource
*ram_res
;
1137 mutex_lock(&kexec_mutex
);
1139 if (kexec_crash_image
) {
1143 start
= crashk_res
.start
;
1144 end
= crashk_res
.end
;
1145 old_size
= (end
== 0) ? 0 : end
- start
+ 1;
1146 if (new_size
>= old_size
) {
1147 ret
= (new_size
== old_size
) ? 0 : -EINVAL
;
1151 ram_res
= kzalloc(sizeof(*ram_res
), GFP_KERNEL
);
1157 start
= roundup(start
, KEXEC_CRASH_MEM_ALIGN
);
1158 end
= roundup(start
+ new_size
, KEXEC_CRASH_MEM_ALIGN
);
1160 crash_map_reserved_pages();
1161 crash_free_reserved_phys_range(end
, crashk_res
.end
);
1163 if ((start
== end
) && (crashk_res
.parent
!= NULL
))
1164 release_resource(&crashk_res
);
1166 ram_res
->start
= end
;
1167 ram_res
->end
= crashk_res
.end
;
1168 ram_res
->flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
;
1169 ram_res
->name
= "System RAM";
1171 crashk_res
.end
= end
- 1;
1173 insert_resource(&iomem_resource
, ram_res
);
1174 crash_unmap_reserved_pages();
1177 mutex_unlock(&kexec_mutex
);
1181 static u32
*append_elf_note(u32
*buf
, char *name
, unsigned type
, void *data
,
1184 struct elf_note note
;
1186 note
.n_namesz
= strlen(name
) + 1;
1187 note
.n_descsz
= data_len
;
1189 memcpy(buf
, ¬e
, sizeof(note
));
1190 buf
+= (sizeof(note
) + 3)/4;
1191 memcpy(buf
, name
, note
.n_namesz
);
1192 buf
+= (note
.n_namesz
+ 3)/4;
1193 memcpy(buf
, data
, note
.n_descsz
);
1194 buf
+= (note
.n_descsz
+ 3)/4;
1199 static void final_note(u32
*buf
)
1201 struct elf_note note
;
1206 memcpy(buf
, ¬e
, sizeof(note
));
1209 void crash_save_cpu(struct pt_regs
*regs
, int cpu
)
1211 struct elf_prstatus prstatus
;
1214 if ((cpu
< 0) || (cpu
>= nr_cpu_ids
))
1217 /* Using ELF notes here is opportunistic.
1218 * I need a well defined structure format
1219 * for the data I pass, and I need tags
1220 * on the data to indicate what information I have
1221 * squirrelled away. ELF notes happen to provide
1222 * all of that, so there is no need to invent something new.
1224 buf
= (u32
*)per_cpu_ptr(crash_notes
, cpu
);
1227 memset(&prstatus
, 0, sizeof(prstatus
));
1228 prstatus
.pr_pid
= current
->pid
;
1229 elf_core_copy_kernel_regs(&prstatus
.pr_reg
, regs
);
1230 buf
= append_elf_note(buf
, KEXEC_CORE_NOTE_NAME
, NT_PRSTATUS
,
1231 &prstatus
, sizeof(prstatus
));
1235 static int __init
crash_notes_memory_init(void)
1237 /* Allocate memory for saving cpu registers. */
1238 crash_notes
= alloc_percpu(note_buf_t
);
1240 printk("Kexec: Memory allocation for saving cpu register"
1241 " states failed\n");
1246 module_init(crash_notes_memory_init
)
1250 * parsing the "crashkernel" commandline
1252 * this code is intended to be called from architecture specific code
1257 * This function parses command lines in the format
1259 * crashkernel=ramsize-range:size[,...][@offset]
1261 * The function returns 0 on success and -EINVAL on failure.
1263 static int __init
parse_crashkernel_mem(char *cmdline
,
1264 unsigned long long system_ram
,
1265 unsigned long long *crash_size
,
1266 unsigned long long *crash_base
)
1268 char *cur
= cmdline
, *tmp
;
1270 /* for each entry of the comma-separated list */
1272 unsigned long long start
, end
= ULLONG_MAX
, size
;
1274 /* get the start of the range */
1275 start
= memparse(cur
, &tmp
);
1277 pr_warning("crashkernel: Memory value expected\n");
1282 pr_warning("crashkernel: '-' expected\n");
1287 /* if no ':' is here, than we read the end */
1289 end
= memparse(cur
, &tmp
);
1291 pr_warning("crashkernel: Memory "
1292 "value expected\n");
1297 pr_warning("crashkernel: end <= start\n");
1303 pr_warning("crashkernel: ':' expected\n");
1308 size
= memparse(cur
, &tmp
);
1310 pr_warning("Memory value expected\n");
1314 if (size
>= system_ram
) {
1315 pr_warning("crashkernel: invalid size\n");
1320 if (system_ram
>= start
&& system_ram
< end
) {
1324 } while (*cur
++ == ',');
1326 if (*crash_size
> 0) {
1327 while (*cur
&& *cur
!= ' ' && *cur
!= '@')
1331 *crash_base
= memparse(cur
, &tmp
);
1333 pr_warning("Memory value expected "
1344 * That function parses "simple" (old) crashkernel command lines like
1346 * crashkernel=size[@offset]
1348 * It returns 0 on success and -EINVAL on failure.
1350 static int __init
parse_crashkernel_simple(char *cmdline
,
1351 unsigned long long *crash_size
,
1352 unsigned long long *crash_base
)
1354 char *cur
= cmdline
;
1356 *crash_size
= memparse(cmdline
, &cur
);
1357 if (cmdline
== cur
) {
1358 pr_warning("crashkernel: memory value expected\n");
1363 *crash_base
= memparse(cur
+1, &cur
);
1364 else if (*cur
!= ' ' && *cur
!= '\0') {
1365 pr_warning("crashkernel: unrecognized char\n");
1373 * That function is the entry point for command line parsing and should be
1374 * called from the arch-specific code.
1376 static int __init
__parse_crashkernel(char *cmdline
,
1377 unsigned long long system_ram
,
1378 unsigned long long *crash_size
,
1379 unsigned long long *crash_base
,
1382 char *p
= cmdline
, *ck_cmdline
= NULL
;
1383 char *first_colon
, *first_space
;
1385 BUG_ON(!crash_size
|| !crash_base
);
1389 /* find crashkernel and use the last one if there are more */
1390 p
= strstr(p
, name
);
1393 p
= strstr(p
+1, name
);
1399 ck_cmdline
+= strlen(name
);
1402 * if the commandline contains a ':', then that's the extended
1403 * syntax -- if not, it must be the classic syntax
1405 first_colon
= strchr(ck_cmdline
, ':');
1406 first_space
= strchr(ck_cmdline
, ' ');
1407 if (first_colon
&& (!first_space
|| first_colon
< first_space
))
1408 return parse_crashkernel_mem(ck_cmdline
, system_ram
,
1409 crash_size
, crash_base
);
1411 return parse_crashkernel_simple(ck_cmdline
, crash_size
,
1417 int __init
parse_crashkernel(char *cmdline
,
1418 unsigned long long system_ram
,
1419 unsigned long long *crash_size
,
1420 unsigned long long *crash_base
)
1422 return __parse_crashkernel(cmdline
, system_ram
, crash_size
, crash_base
,
1426 int __init
parse_crashkernel_low(char *cmdline
,
1427 unsigned long long system_ram
,
1428 unsigned long long *crash_size
,
1429 unsigned long long *crash_base
)
1431 return __parse_crashkernel(cmdline
, system_ram
, crash_size
, crash_base
,
1432 "crashkernel_low=");
1435 static void update_vmcoreinfo_note(void)
1437 u32
*buf
= vmcoreinfo_note
;
1439 if (!vmcoreinfo_size
)
1441 buf
= append_elf_note(buf
, VMCOREINFO_NOTE_NAME
, 0, vmcoreinfo_data
,
1446 void crash_save_vmcoreinfo(void)
1448 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1449 update_vmcoreinfo_note();
1452 void vmcoreinfo_append_str(const char *fmt
, ...)
1458 va_start(args
, fmt
);
1459 r
= vsnprintf(buf
, sizeof(buf
), fmt
, args
);
1462 if (r
+ vmcoreinfo_size
> vmcoreinfo_max_size
)
1463 r
= vmcoreinfo_max_size
- vmcoreinfo_size
;
1465 memcpy(&vmcoreinfo_data
[vmcoreinfo_size
], buf
, r
);
1467 vmcoreinfo_size
+= r
;
1471 * provide an empty default implementation here -- architecture
1472 * code may override this
1474 void __attribute__ ((weak
)) arch_crash_save_vmcoreinfo(void)
1477 unsigned long __attribute__ ((weak
)) paddr_vmcoreinfo_note(void)
1479 return __pa((unsigned long)(char *)&vmcoreinfo_note
);
1482 static int __init
crash_save_vmcoreinfo_init(void)
1484 VMCOREINFO_OSRELEASE(init_uts_ns
.name
.release
);
1485 VMCOREINFO_PAGESIZE(PAGE_SIZE
);
1487 VMCOREINFO_SYMBOL(init_uts_ns
);
1488 VMCOREINFO_SYMBOL(node_online_map
);
1490 VMCOREINFO_SYMBOL(swapper_pg_dir
);
1492 VMCOREINFO_SYMBOL(_stext
);
1493 VMCOREINFO_SYMBOL(vmlist
);
1495 #ifndef CONFIG_NEED_MULTIPLE_NODES
1496 VMCOREINFO_SYMBOL(mem_map
);
1497 VMCOREINFO_SYMBOL(contig_page_data
);
1499 #ifdef CONFIG_SPARSEMEM
1500 VMCOREINFO_SYMBOL(mem_section
);
1501 VMCOREINFO_LENGTH(mem_section
, NR_SECTION_ROOTS
);
1502 VMCOREINFO_STRUCT_SIZE(mem_section
);
1503 VMCOREINFO_OFFSET(mem_section
, section_mem_map
);
1505 VMCOREINFO_STRUCT_SIZE(page
);
1506 VMCOREINFO_STRUCT_SIZE(pglist_data
);
1507 VMCOREINFO_STRUCT_SIZE(zone
);
1508 VMCOREINFO_STRUCT_SIZE(free_area
);
1509 VMCOREINFO_STRUCT_SIZE(list_head
);
1510 VMCOREINFO_SIZE(nodemask_t
);
1511 VMCOREINFO_OFFSET(page
, flags
);
1512 VMCOREINFO_OFFSET(page
, _count
);
1513 VMCOREINFO_OFFSET(page
, mapping
);
1514 VMCOREINFO_OFFSET(page
, lru
);
1515 VMCOREINFO_OFFSET(page
, _mapcount
);
1516 VMCOREINFO_OFFSET(page
, private);
1517 VMCOREINFO_OFFSET(pglist_data
, node_zones
);
1518 VMCOREINFO_OFFSET(pglist_data
, nr_zones
);
1519 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1520 VMCOREINFO_OFFSET(pglist_data
, node_mem_map
);
1522 VMCOREINFO_OFFSET(pglist_data
, node_start_pfn
);
1523 VMCOREINFO_OFFSET(pglist_data
, node_spanned_pages
);
1524 VMCOREINFO_OFFSET(pglist_data
, node_id
);
1525 VMCOREINFO_OFFSET(zone
, free_area
);
1526 VMCOREINFO_OFFSET(zone
, vm_stat
);
1527 VMCOREINFO_OFFSET(zone
, spanned_pages
);
1528 VMCOREINFO_OFFSET(free_area
, free_list
);
1529 VMCOREINFO_OFFSET(list_head
, next
);
1530 VMCOREINFO_OFFSET(list_head
, prev
);
1531 VMCOREINFO_OFFSET(vm_struct
, addr
);
1532 VMCOREINFO_LENGTH(zone
.free_area
, MAX_ORDER
);
1533 log_buf_kexec_setup();
1534 VMCOREINFO_LENGTH(free_area
.free_list
, MIGRATE_TYPES
);
1535 VMCOREINFO_NUMBER(NR_FREE_PAGES
);
1536 VMCOREINFO_NUMBER(PG_lru
);
1537 VMCOREINFO_NUMBER(PG_private
);
1538 VMCOREINFO_NUMBER(PG_swapcache
);
1539 VMCOREINFO_NUMBER(PG_slab
);
1540 #ifdef CONFIG_MEMORY_FAILURE
1541 VMCOREINFO_NUMBER(PG_hwpoison
);
1543 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE
);
1545 arch_crash_save_vmcoreinfo();
1546 update_vmcoreinfo_note();
1551 module_init(crash_save_vmcoreinfo_init
)
1554 * Move into place and start executing a preloaded standalone
1555 * executable. If nothing was preloaded return an error.
1557 int kernel_kexec(void)
1561 if (!mutex_trylock(&kexec_mutex
))
1568 #ifdef CONFIG_KEXEC_JUMP
1569 if (kexec_image
->preserve_context
) {
1570 lock_system_sleep();
1571 pm_prepare_console();
1572 error
= freeze_processes();
1575 goto Restore_console
;
1578 error
= dpm_suspend_start(PMSG_FREEZE
);
1580 goto Resume_console
;
1581 /* At this point, dpm_suspend_start() has been called,
1582 * but *not* dpm_suspend_end(). We *must* call
1583 * dpm_suspend_end() now. Otherwise, drivers for
1584 * some devices (e.g. interrupt controllers) become
1585 * desynchronized with the actual state of the
1586 * hardware at resume time, and evil weirdness ensues.
1588 error
= dpm_suspend_end(PMSG_FREEZE
);
1590 goto Resume_devices
;
1591 error
= disable_nonboot_cpus();
1594 local_irq_disable();
1595 error
= syscore_suspend();
1601 kernel_restart_prepare(NULL
);
1602 printk(KERN_EMERG
"Starting new kernel\n");
1606 machine_kexec(kexec_image
);
1608 #ifdef CONFIG_KEXEC_JUMP
1609 if (kexec_image
->preserve_context
) {
1614 enable_nonboot_cpus();
1615 dpm_resume_start(PMSG_RESTORE
);
1617 dpm_resume_end(PMSG_RESTORE
);
1622 pm_restore_console();
1623 unlock_system_sleep();
1628 mutex_unlock(&kexec_mutex
);