drivers: power: report battery voltage in AOSP compatible format
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / kernel / kexec.c
CommitLineData
dc009d92
EB
1/*
2 * kexec.c - kexec system call
3 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
4 *
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
7 */
8
c59ede7b 9#include <linux/capability.h>
dc009d92
EB
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>
8c5a1cf0 15#include <linux/mutex.h>
dc009d92
EB
16#include <linux/list.h>
17#include <linux/highmem.h>
18#include <linux/syscalls.h>
19#include <linux/reboot.h>
dc009d92 20#include <linux/ioport.h>
6e274d14 21#include <linux/hardirq.h>
85916f81
MD
22#include <linux/elf.h>
23#include <linux/elfcore.h>
fd59d231
KO
24#include <linux/utsname.h>
25#include <linux/numa.h>
3ab83521
HY
26#include <linux/suspend.h>
27#include <linux/device.h>
89081d17
HY
28#include <linux/freezer.h>
29#include <linux/pm.h>
30#include <linux/cpu.h>
31#include <linux/console.h>
5f41b8cd 32#include <linux/vmalloc.h>
06a7f711 33#include <linux/swap.h>
19234c08 34#include <linux/syscore_ops.h>
6e274d14 35
dc009d92
EB
36#include <asm/page.h>
37#include <asm/uaccess.h>
38#include <asm/io.h>
fd59d231 39#include <asm/sections.h>
dc009d92 40
cc571658 41/* Per cpu memory for storing cpu states in case of system crash. */
43cf38eb 42note_buf_t __percpu *crash_notes;
cc571658 43
fd59d231 44/* vmcoreinfo stuff */
edb79a21 45static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
fd59d231 46u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
d768281e
KO
47size_t vmcoreinfo_size;
48size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
fd59d231 49
2a038881
KA
50/* Flag to indicate we are going to kexec a new kernel */
51bool kexec_in_progress = false;
52
dc009d92
EB
53/* Location of the reserved area for the crash kernel */
54struct resource crashk_res = {
55 .name = "Crash kernel",
56 .start = 0,
57 .end = 0,
58 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
59};
0212f915 60struct resource crashk_low_res = {
157752d8 61 .name = "Crash kernel",
0212f915
YL
62 .start = 0,
63 .end = 0,
64 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
65};
dc009d92 66
6e274d14
AN
67int kexec_should_crash(struct task_struct *p)
68{
b460cbc5 69 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
6e274d14
AN
70 return 1;
71 return 0;
72}
73
dc009d92
EB
74/*
75 * When kexec transitions to the new kernel there is a one-to-one
76 * mapping between physical and virtual addresses. On processors
77 * where you can disable the MMU this is trivial, and easy. For
78 * others it is still a simple predictable page table to setup.
79 *
80 * In that environment kexec copies the new kernel to its final
81 * resting place. This means I can only support memory whose
82 * physical address can fit in an unsigned long. In particular
83 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
84 * If the assembly stub has more restrictive requirements
85 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
86 * defined more restrictively in <asm/kexec.h>.
87 *
88 * The code for the transition from the current kernel to the
89 * the new kernel is placed in the control_code_buffer, whose size
163f6876 90 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
dc009d92
EB
91 * page of memory is necessary, but some architectures require more.
92 * Because this memory must be identity mapped in the transition from
93 * virtual to physical addresses it must live in the range
94 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
95 * modifiable.
96 *
97 * The assembly stub in the control code buffer is passed a linked list
98 * of descriptor pages detailing the source pages of the new kernel,
99 * and the destination addresses of those source pages. As this data
100 * structure is not used in the context of the current OS, it must
101 * be self-contained.
102 *
103 * The code has been made to work with highmem pages and will use a
104 * destination page in its final resting place (if it happens
105 * to allocate it). The end product of this is that most of the
106 * physical address space, and most of RAM can be used.
107 *
108 * Future directions include:
109 * - allocating a page table with the control code buffer identity
110 * mapped, to simplify machine_kexec and make kexec_on_panic more
111 * reliable.
112 */
113
114/*
115 * KIMAGE_NO_DEST is an impossible destination address..., for
116 * allocating pages whose destination address we do not care about.
117 */
118#define KIMAGE_NO_DEST (-1UL)
119
72414d3f
MS
120static int kimage_is_destination_range(struct kimage *image,
121 unsigned long start, unsigned long end);
122static struct page *kimage_alloc_page(struct kimage *image,
9796fdd8 123 gfp_t gfp_mask,
72414d3f 124 unsigned long dest);
dc009d92
EB
125
126static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
72414d3f
MS
127 unsigned long nr_segments,
128 struct kexec_segment __user *segments)
dc009d92
EB
129{
130 size_t segment_bytes;
131 struct kimage *image;
132 unsigned long i;
133 int result;
134
135 /* Allocate a controlling structure */
136 result = -ENOMEM;
4668edc3 137 image = kzalloc(sizeof(*image), GFP_KERNEL);
72414d3f 138 if (!image)
dc009d92 139 goto out;
72414d3f 140
dc009d92
EB
141 image->head = 0;
142 image->entry = &image->head;
143 image->last_entry = &image->head;
144 image->control_page = ~0; /* By default this does not apply */
145 image->start = entry;
146 image->type = KEXEC_TYPE_DEFAULT;
147
148 /* Initialize the list of control pages */
149 INIT_LIST_HEAD(&image->control_pages);
150
151 /* Initialize the list of destination pages */
152 INIT_LIST_HEAD(&image->dest_pages);
153
25985edc 154 /* Initialize the list of unusable pages */
dc009d92
EB
155 INIT_LIST_HEAD(&image->unuseable_pages);
156
157 /* Read in the segments */
158 image->nr_segments = nr_segments;
159 segment_bytes = nr_segments * sizeof(*segments);
160 result = copy_from_user(image->segment, segments, segment_bytes);
f65a03f6
DC
161 if (result) {
162 result = -EFAULT;
dc009d92 163 goto out;
f65a03f6 164 }
dc009d92
EB
165
166 /*
167 * Verify we have good destination addresses. The caller is
168 * responsible for making certain we don't attempt to load
169 * the new image into invalid or reserved areas of RAM. This
170 * just verifies it is an address we can use.
171 *
172 * Since the kernel does everything in page size chunks ensure
b595076a 173 * the destination addresses are page aligned. Too many
dc009d92
EB
174 * special cases crop of when we don't do this. The most
175 * insidious is getting overlapping destination addresses
176 * simply because addresses are changed to page size
177 * granularity.
178 */
179 result = -EADDRNOTAVAIL;
180 for (i = 0; i < nr_segments; i++) {
181 unsigned long mstart, mend;
72414d3f 182
dc009d92
EB
183 mstart = image->segment[i].mem;
184 mend = mstart + image->segment[i].memsz;
185 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
186 goto out;
187 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
188 goto out;
189 }
190
191 /* Verify our destination addresses do not overlap.
192 * If we alloed overlapping destination addresses
193 * through very weird things can happen with no
194 * easy explanation as one segment stops on another.
195 */
196 result = -EINVAL;
72414d3f 197 for (i = 0; i < nr_segments; i++) {
dc009d92
EB
198 unsigned long mstart, mend;
199 unsigned long j;
72414d3f 200
dc009d92
EB
201 mstart = image->segment[i].mem;
202 mend = mstart + image->segment[i].memsz;
72414d3f 203 for (j = 0; j < i; j++) {
dc009d92
EB
204 unsigned long pstart, pend;
205 pstart = image->segment[j].mem;
206 pend = pstart + image->segment[j].memsz;
207 /* Do the segments overlap ? */
208 if ((mend > pstart) && (mstart < pend))
209 goto out;
210 }
211 }
212
213 /* Ensure our buffer sizes are strictly less than
214 * our memory sizes. This should always be the case,
215 * and it is easier to check up front than to be surprised
216 * later on.
217 */
218 result = -EINVAL;
72414d3f 219 for (i = 0; i < nr_segments; i++) {
dc009d92
EB
220 if (image->segment[i].bufsz > image->segment[i].memsz)
221 goto out;
222 }
223
dc009d92 224 result = 0;
72414d3f
MS
225out:
226 if (result == 0)
dc009d92 227 *rimage = image;
72414d3f 228 else
dc009d92 229 kfree(image);
72414d3f 230
dc009d92
EB
231 return result;
232
233}
234
b92e7e0d
ZY
235static void kimage_free_page_list(struct list_head *list);
236
dc009d92 237static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
72414d3f
MS
238 unsigned long nr_segments,
239 struct kexec_segment __user *segments)
dc009d92
EB
240{
241 int result;
242 struct kimage *image;
243
244 /* Allocate and initialize a controlling structure */
245 image = NULL;
246 result = do_kimage_alloc(&image, entry, nr_segments, segments);
72414d3f 247 if (result)
dc009d92 248 goto out;
72414d3f 249
dc009d92
EB
250 /*
251 * Find a location for the control code buffer, and add it
252 * the vector of segments so that it's pages will also be
253 * counted as destination pages.
254 */
255 result = -ENOMEM;
256 image->control_code_page = kimage_alloc_control_pages(image,
163f6876 257 get_order(KEXEC_CONTROL_PAGE_SIZE));
dc009d92
EB
258 if (!image->control_code_page) {
259 printk(KERN_ERR "Could not allocate control_code_buffer\n");
b92e7e0d 260 goto out_free;
dc009d92
EB
261 }
262
3ab83521
HY
263 image->swap_page = kimage_alloc_control_pages(image, 0);
264 if (!image->swap_page) {
265 printk(KERN_ERR "Could not allocate swap buffer\n");
b92e7e0d 266 goto out_free;
3ab83521
HY
267 }
268
b92e7e0d
ZY
269 *rimage = image;
270 return 0;
72414d3f 271
b92e7e0d
ZY
272out_free:
273 kimage_free_page_list(&image->control_pages);
274 kfree(image);
275out:
dc009d92
EB
276 return result;
277}
278
279static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
72414d3f 280 unsigned long nr_segments,
314b6a4d 281 struct kexec_segment __user *segments)
dc009d92
EB
282{
283 int result;
284 struct kimage *image;
285 unsigned long i;
286
287 image = NULL;
288 /* Verify we have a valid entry point */
289 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
290 result = -EADDRNOTAVAIL;
291 goto out;
292 }
293
294 /* Allocate and initialize a controlling structure */
295 result = do_kimage_alloc(&image, entry, nr_segments, segments);
72414d3f 296 if (result)
dc009d92 297 goto out;
dc009d92
EB
298
299 /* Enable the special crash kernel control page
300 * allocation policy.
301 */
302 image->control_page = crashk_res.start;
303 image->type = KEXEC_TYPE_CRASH;
304
305 /*
306 * Verify we have good destination addresses. Normally
307 * the caller is responsible for making certain we don't
308 * attempt to load the new image into invalid or reserved
309 * areas of RAM. But crash kernels are preloaded into a
310 * reserved area of ram. We must ensure the addresses
311 * are in the reserved area otherwise preloading the
312 * kernel could corrupt things.
313 */
314 result = -EADDRNOTAVAIL;
315 for (i = 0; i < nr_segments; i++) {
316 unsigned long mstart, mend;
72414d3f 317
dc009d92 318 mstart = image->segment[i].mem;
50cccc69 319 mend = mstart + image->segment[i].memsz - 1;
dc009d92
EB
320 /* Ensure we are within the crash kernel limits */
321 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
8c333ac2 322 goto out_free;
dc009d92
EB
323 }
324
dc009d92
EB
325 /*
326 * Find a location for the control code buffer, and add
327 * the vector of segments so that it's pages will also be
328 * counted as destination pages.
329 */
330 result = -ENOMEM;
331 image->control_code_page = kimage_alloc_control_pages(image,
163f6876 332 get_order(KEXEC_CONTROL_PAGE_SIZE));
dc009d92
EB
333 if (!image->control_code_page) {
334 printk(KERN_ERR "Could not allocate control_code_buffer\n");
8c333ac2 335 goto out_free;
dc009d92
EB
336 }
337
8c333ac2
ZY
338 *rimage = image;
339 return 0;
72414d3f 340
8c333ac2
ZY
341out_free:
342 kfree(image);
343out:
dc009d92
EB
344 return result;
345}
346
72414d3f
MS
347static int kimage_is_destination_range(struct kimage *image,
348 unsigned long start,
349 unsigned long end)
dc009d92
EB
350{
351 unsigned long i;
352
353 for (i = 0; i < image->nr_segments; i++) {
354 unsigned long mstart, mend;
72414d3f 355
dc009d92 356 mstart = image->segment[i].mem;
72414d3f
MS
357 mend = mstart + image->segment[i].memsz;
358 if ((end > mstart) && (start < mend))
dc009d92 359 return 1;
dc009d92 360 }
72414d3f 361
dc009d92
EB
362 return 0;
363}
364
9796fdd8 365static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
dc009d92
EB
366{
367 struct page *pages;
72414d3f 368
dc009d92
EB
369 pages = alloc_pages(gfp_mask, order);
370 if (pages) {
371 unsigned int count, i;
372 pages->mapping = NULL;
4c21e2f2 373 set_page_private(pages, order);
dc009d92 374 count = 1 << order;
72414d3f 375 for (i = 0; i < count; i++)
dc009d92 376 SetPageReserved(pages + i);
dc009d92 377 }
72414d3f 378
dc009d92
EB
379 return pages;
380}
381
382static void kimage_free_pages(struct page *page)
383{
384 unsigned int order, count, i;
72414d3f 385
4c21e2f2 386 order = page_private(page);
dc009d92 387 count = 1 << order;
72414d3f 388 for (i = 0; i < count; i++)
dc009d92 389 ClearPageReserved(page + i);
dc009d92
EB
390 __free_pages(page, order);
391}
392
393static void kimage_free_page_list(struct list_head *list)
394{
395 struct list_head *pos, *next;
72414d3f 396
dc009d92
EB
397 list_for_each_safe(pos, next, list) {
398 struct page *page;
399
400 page = list_entry(pos, struct page, lru);
401 list_del(&page->lru);
dc009d92
EB
402 kimage_free_pages(page);
403 }
404}
405
72414d3f
MS
406static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
407 unsigned int order)
dc009d92
EB
408{
409 /* Control pages are special, they are the intermediaries
410 * that are needed while we copy the rest of the pages
411 * to their final resting place. As such they must
412 * not conflict with either the destination addresses
413 * or memory the kernel is already using.
414 *
415 * The only case where we really need more than one of
416 * these are for architectures where we cannot disable
417 * the MMU and must instead generate an identity mapped
418 * page table for all of the memory.
419 *
420 * At worst this runs in O(N) of the image size.
421 */
422 struct list_head extra_pages;
423 struct page *pages;
424 unsigned int count;
425
426 count = 1 << order;
427 INIT_LIST_HEAD(&extra_pages);
428
429 /* Loop while I can allocate a page and the page allocated
430 * is a destination page.
431 */
432 do {
433 unsigned long pfn, epfn, addr, eaddr;
72414d3f 434
dc009d92
EB
435 pages = kimage_alloc_pages(GFP_KERNEL, order);
436 if (!pages)
437 break;
438 pfn = page_to_pfn(pages);
439 epfn = pfn + count;
440 addr = pfn << PAGE_SHIFT;
441 eaddr = epfn << PAGE_SHIFT;
442 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
72414d3f 443 kimage_is_destination_range(image, addr, eaddr)) {
dc009d92
EB
444 list_add(&pages->lru, &extra_pages);
445 pages = NULL;
446 }
72414d3f
MS
447 } while (!pages);
448
dc009d92
EB
449 if (pages) {
450 /* Remember the allocated page... */
451 list_add(&pages->lru, &image->control_pages);
452
453 /* Because the page is already in it's destination
454 * location we will never allocate another page at
455 * that address. Therefore kimage_alloc_pages
456 * will not return it (again) and we don't need
457 * to give it an entry in image->segment[].
458 */
459 }
460 /* Deal with the destination pages I have inadvertently allocated.
461 *
462 * Ideally I would convert multi-page allocations into single
25985edc 463 * page allocations, and add everything to image->dest_pages.
dc009d92
EB
464 *
465 * For now it is simpler to just free the pages.
466 */
467 kimage_free_page_list(&extra_pages);
dc009d92 468
72414d3f 469 return pages;
dc009d92
EB
470}
471
72414d3f
MS
472static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
473 unsigned int order)
dc009d92
EB
474{
475 /* Control pages are special, they are the intermediaries
476 * that are needed while we copy the rest of the pages
477 * to their final resting place. As such they must
478 * not conflict with either the destination addresses
479 * or memory the kernel is already using.
480 *
481 * Control pages are also the only pags we must allocate
482 * when loading a crash kernel. All of the other pages
483 * are specified by the segments and we just memcpy
484 * into them directly.
485 *
486 * The only case where we really need more than one of
487 * these are for architectures where we cannot disable
488 * the MMU and must instead generate an identity mapped
489 * page table for all of the memory.
490 *
491 * Given the low demand this implements a very simple
492 * allocator that finds the first hole of the appropriate
493 * size in the reserved memory region, and allocates all
494 * of the memory up to and including the hole.
495 */
496 unsigned long hole_start, hole_end, size;
497 struct page *pages;
72414d3f 498
dc009d92
EB
499 pages = NULL;
500 size = (1 << order) << PAGE_SHIFT;
501 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
502 hole_end = hole_start + size - 1;
72414d3f 503 while (hole_end <= crashk_res.end) {
dc009d92 504 unsigned long i;
72414d3f 505
3d214fae 506 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
dc009d92 507 break;
dc009d92 508 /* See if I overlap any of the segments */
72414d3f 509 for (i = 0; i < image->nr_segments; i++) {
dc009d92 510 unsigned long mstart, mend;
72414d3f 511
dc009d92
EB
512 mstart = image->segment[i].mem;
513 mend = mstart + image->segment[i].memsz - 1;
514 if ((hole_end >= mstart) && (hole_start <= mend)) {
515 /* Advance the hole to the end of the segment */
516 hole_start = (mend + (size - 1)) & ~(size - 1);
517 hole_end = hole_start + size - 1;
518 break;
519 }
520 }
521 /* If I don't overlap any segments I have found my hole! */
522 if (i == image->nr_segments) {
523 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
524 break;
525 }
526 }
72414d3f 527 if (pages)
dc009d92 528 image->control_page = hole_end;
72414d3f 529
dc009d92
EB
530 return pages;
531}
532
533
72414d3f
MS
534struct page *kimage_alloc_control_pages(struct kimage *image,
535 unsigned int order)
dc009d92
EB
536{
537 struct page *pages = NULL;
72414d3f
MS
538
539 switch (image->type) {
dc009d92
EB
540 case KEXEC_TYPE_DEFAULT:
541 pages = kimage_alloc_normal_control_pages(image, order);
542 break;
543 case KEXEC_TYPE_CRASH:
544 pages = kimage_alloc_crash_control_pages(image, order);
545 break;
546 }
72414d3f 547
dc009d92
EB
548 return pages;
549}
550
551static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
552{
72414d3f 553 if (*image->entry != 0)
dc009d92 554 image->entry++;
72414d3f 555
dc009d92
EB
556 if (image->entry == image->last_entry) {
557 kimage_entry_t *ind_page;
558 struct page *page;
72414d3f 559
dc009d92 560 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
72414d3f 561 if (!page)
dc009d92 562 return -ENOMEM;
72414d3f 563
dc009d92
EB
564 ind_page = page_address(page);
565 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
566 image->entry = ind_page;
72414d3f
MS
567 image->last_entry = ind_page +
568 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
dc009d92
EB
569 }
570 *image->entry = entry;
571 image->entry++;
572 *image->entry = 0;
72414d3f 573
dc009d92
EB
574 return 0;
575}
576
72414d3f
MS
577static int kimage_set_destination(struct kimage *image,
578 unsigned long destination)
dc009d92
EB
579{
580 int result;
581
582 destination &= PAGE_MASK;
583 result = kimage_add_entry(image, destination | IND_DESTINATION);
72414d3f 584 if (result == 0)
dc009d92 585 image->destination = destination;
72414d3f 586
dc009d92
EB
587 return result;
588}
589
590
591static int kimage_add_page(struct kimage *image, unsigned long page)
592{
593 int result;
594
595 page &= PAGE_MASK;
596 result = kimage_add_entry(image, page | IND_SOURCE);
72414d3f 597 if (result == 0)
dc009d92 598 image->destination += PAGE_SIZE;
72414d3f 599
dc009d92
EB
600 return result;
601}
602
603
604static void kimage_free_extra_pages(struct kimage *image)
605{
606 /* Walk through and free any extra destination pages I may have */
607 kimage_free_page_list(&image->dest_pages);
608
25985edc 609 /* Walk through and free any unusable pages I have cached */
dc009d92
EB
610 kimage_free_page_list(&image->unuseable_pages);
611
612}
7fccf032 613static void kimage_terminate(struct kimage *image)
dc009d92 614{
72414d3f 615 if (*image->entry != 0)
dc009d92 616 image->entry++;
72414d3f 617
dc009d92 618 *image->entry = IND_DONE;
dc009d92
EB
619}
620
621#define for_each_kimage_entry(image, ptr, entry) \
622 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
623 ptr = (entry & IND_INDIRECTION)? \
624 phys_to_virt((entry & PAGE_MASK)): ptr +1)
625
626static void kimage_free_entry(kimage_entry_t entry)
627{
628 struct page *page;
629
630 page = pfn_to_page(entry >> PAGE_SHIFT);
631 kimage_free_pages(page);
632}
633
634static void kimage_free(struct kimage *image)
635{
636 kimage_entry_t *ptr, entry;
637 kimage_entry_t ind = 0;
638
639 if (!image)
640 return;
72414d3f 641
dc009d92
EB
642 kimage_free_extra_pages(image);
643 for_each_kimage_entry(image, ptr, entry) {
644 if (entry & IND_INDIRECTION) {
645 /* Free the previous indirection page */
72414d3f 646 if (ind & IND_INDIRECTION)
dc009d92 647 kimage_free_entry(ind);
dc009d92
EB
648 /* Save this indirection page until we are
649 * done with it.
650 */
651 ind = entry;
652 }
72414d3f 653 else if (entry & IND_SOURCE)
dc009d92 654 kimage_free_entry(entry);
dc009d92
EB
655 }
656 /* Free the final indirection page */
72414d3f 657 if (ind & IND_INDIRECTION)
dc009d92 658 kimage_free_entry(ind);
dc009d92
EB
659
660 /* Handle any machine specific cleanup */
661 machine_kexec_cleanup(image);
662
663 /* Free the kexec control pages... */
664 kimage_free_page_list(&image->control_pages);
665 kfree(image);
666}
667
72414d3f
MS
668static kimage_entry_t *kimage_dst_used(struct kimage *image,
669 unsigned long page)
dc009d92
EB
670{
671 kimage_entry_t *ptr, entry;
672 unsigned long destination = 0;
673
674 for_each_kimage_entry(image, ptr, entry) {
72414d3f 675 if (entry & IND_DESTINATION)
dc009d92 676 destination = entry & PAGE_MASK;
dc009d92 677 else if (entry & IND_SOURCE) {
72414d3f 678 if (page == destination)
dc009d92 679 return ptr;
dc009d92
EB
680 destination += PAGE_SIZE;
681 }
682 }
72414d3f 683
314b6a4d 684 return NULL;
dc009d92
EB
685}
686
72414d3f 687static struct page *kimage_alloc_page(struct kimage *image,
9796fdd8 688 gfp_t gfp_mask,
72414d3f 689 unsigned long destination)
dc009d92
EB
690{
691 /*
692 * Here we implement safeguards to ensure that a source page
693 * is not copied to its destination page before the data on
694 * the destination page is no longer useful.
695 *
696 * To do this we maintain the invariant that a source page is
697 * either its own destination page, or it is not a
698 * destination page at all.
699 *
700 * That is slightly stronger than required, but the proof
701 * that no problems will not occur is trivial, and the
702 * implementation is simply to verify.
703 *
704 * When allocating all pages normally this algorithm will run
705 * in O(N) time, but in the worst case it will run in O(N^2)
706 * time. If the runtime is a problem the data structures can
707 * be fixed.
708 */
709 struct page *page;
710 unsigned long addr;
711
712 /*
713 * Walk through the list of destination pages, and see if I
714 * have a match.
715 */
716 list_for_each_entry(page, &image->dest_pages, lru) {
717 addr = page_to_pfn(page) << PAGE_SHIFT;
718 if (addr == destination) {
719 list_del(&page->lru);
720 return page;
721 }
722 }
723 page = NULL;
724 while (1) {
725 kimage_entry_t *old;
726
727 /* Allocate a page, if we run out of memory give up */
728 page = kimage_alloc_pages(gfp_mask, 0);
72414d3f 729 if (!page)
314b6a4d 730 return NULL;
dc009d92 731 /* If the page cannot be used file it away */
72414d3f
MS
732 if (page_to_pfn(page) >
733 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
dc009d92
EB
734 list_add(&page->lru, &image->unuseable_pages);
735 continue;
736 }
737 addr = page_to_pfn(page) << PAGE_SHIFT;
738
739 /* If it is the destination page we want use it */
740 if (addr == destination)
741 break;
742
743 /* If the page is not a destination page use it */
72414d3f
MS
744 if (!kimage_is_destination_range(image, addr,
745 addr + PAGE_SIZE))
dc009d92
EB
746 break;
747
748 /*
749 * I know that the page is someones destination page.
750 * See if there is already a source page for this
751 * destination page. And if so swap the source pages.
752 */
753 old = kimage_dst_used(image, addr);
754 if (old) {
755 /* If so move it */
756 unsigned long old_addr;
757 struct page *old_page;
758
759 old_addr = *old & PAGE_MASK;
760 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
761 copy_highpage(page, old_page);
762 *old = addr | (*old & ~PAGE_MASK);
763
764 /* The old page I have found cannot be a
f9092f35
JS
765 * destination page, so return it if it's
766 * gfp_flags honor the ones passed in.
dc009d92 767 */
f9092f35
JS
768 if (!(gfp_mask & __GFP_HIGHMEM) &&
769 PageHighMem(old_page)) {
770 kimage_free_pages(old_page);
771 continue;
772 }
dc009d92
EB
773 addr = old_addr;
774 page = old_page;
775 break;
776 }
777 else {
778 /* Place the page on the destination list I
779 * will use it later.
780 */
781 list_add(&page->lru, &image->dest_pages);
782 }
783 }
72414d3f 784
dc009d92
EB
785 return page;
786}
787
788static int kimage_load_normal_segment(struct kimage *image,
72414d3f 789 struct kexec_segment *segment)
dc009d92
EB
790{
791 unsigned long maddr;
310faaa9 792 size_t ubytes, mbytes;
dc009d92 793 int result;
314b6a4d 794 unsigned char __user *buf;
dc009d92
EB
795
796 result = 0;
797 buf = segment->buf;
798 ubytes = segment->bufsz;
799 mbytes = segment->memsz;
800 maddr = segment->mem;
801
802 result = kimage_set_destination(image, maddr);
72414d3f 803 if (result < 0)
dc009d92 804 goto out;
72414d3f
MS
805
806 while (mbytes) {
dc009d92
EB
807 struct page *page;
808 char *ptr;
809 size_t uchunk, mchunk;
72414d3f 810
dc009d92 811 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
c80544dc 812 if (!page) {
dc009d92
EB
813 result = -ENOMEM;
814 goto out;
815 }
72414d3f
MS
816 result = kimage_add_page(image, page_to_pfn(page)
817 << PAGE_SHIFT);
818 if (result < 0)
dc009d92 819 goto out;
72414d3f 820
dc009d92
EB
821 ptr = kmap(page);
822 /* Start with a clear page */
3ecb01df 823 clear_page(ptr);
dc009d92 824 ptr += maddr & ~PAGE_MASK;
31c3a3fe
ZY
825 mchunk = min_t(size_t, mbytes,
826 PAGE_SIZE - (maddr & ~PAGE_MASK));
827 uchunk = min(ubytes, mchunk);
72414d3f 828
dc009d92
EB
829 result = copy_from_user(ptr, buf, uchunk);
830 kunmap(page);
831 if (result) {
f65a03f6 832 result = -EFAULT;
dc009d92
EB
833 goto out;
834 }
835 ubytes -= uchunk;
836 maddr += mchunk;
837 buf += mchunk;
838 mbytes -= mchunk;
839 }
72414d3f 840out:
dc009d92
EB
841 return result;
842}
843
844static int kimage_load_crash_segment(struct kimage *image,
72414d3f 845 struct kexec_segment *segment)
dc009d92
EB
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;
310faaa9 852 size_t ubytes, mbytes;
dc009d92 853 int result;
314b6a4d 854 unsigned char __user *buf;
dc009d92
EB
855
856 result = 0;
857 buf = segment->buf;
858 ubytes = segment->bufsz;
859 mbytes = segment->memsz;
860 maddr = segment->mem;
72414d3f 861 while (mbytes) {
dc009d92
EB
862 struct page *page;
863 char *ptr;
864 size_t uchunk, mchunk;
72414d3f 865
dc009d92 866 page = pfn_to_page(maddr >> PAGE_SHIFT);
c80544dc 867 if (!page) {
dc009d92
EB
868 result = -ENOMEM;
869 goto out;
870 }
871 ptr = kmap(page);
872 ptr += maddr & ~PAGE_MASK;
31c3a3fe
ZY
873 mchunk = min_t(size_t, mbytes,
874 PAGE_SIZE - (maddr & ~PAGE_MASK));
875 uchunk = min(ubytes, mchunk);
876 if (mchunk > uchunk) {
dc009d92
EB
877 /* Zero the trailing part of the page */
878 memset(ptr + uchunk, 0, mchunk - uchunk);
879 }
880 result = copy_from_user(ptr, buf, uchunk);
a7956113 881 kexec_flush_icache_page(page);
dc009d92
EB
882 kunmap(page);
883 if (result) {
f65a03f6 884 result = -EFAULT;
dc009d92
EB
885 goto out;
886 }
887 ubytes -= uchunk;
888 maddr += mchunk;
889 buf += mchunk;
890 mbytes -= mchunk;
891 }
72414d3f 892out:
dc009d92
EB
893 return result;
894}
895
896static int kimage_load_segment(struct kimage *image,
72414d3f 897 struct kexec_segment *segment)
dc009d92
EB
898{
899 int result = -ENOMEM;
72414d3f
MS
900
901 switch (image->type) {
dc009d92
EB
902 case KEXEC_TYPE_DEFAULT:
903 result = kimage_load_normal_segment(image, segment);
904 break;
905 case KEXEC_TYPE_CRASH:
906 result = kimage_load_crash_segment(image, segment);
907 break;
908 }
72414d3f 909
dc009d92
EB
910 return result;
911}
912
913/*
914 * Exec Kernel system call: for obvious reasons only root may call it.
915 *
916 * This call breaks up into three pieces.
917 * - A generic part which loads the new kernel from the current
918 * address space, and very carefully places the data in the
919 * allocated pages.
920 *
921 * - A generic part that interacts with the kernel and tells all of
922 * the devices to shut down. Preventing on-going dmas, and placing
923 * the devices in a consistent state so a later kernel can
924 * reinitialize them.
925 *
926 * - A machine specific part that includes the syscall number
927 * and the copies the image to it's final destination. And
928 * jumps into the image at entry.
929 *
930 * kexec does not sync, or unmount filesystems so if you need
931 * that to happen you need to do that yourself.
932 */
c330dda9
JM
933struct kimage *kexec_image;
934struct kimage *kexec_crash_image;
8c5a1cf0
AM
935
936static DEFINE_MUTEX(kexec_mutex);
dc009d92 937
754fe8d2
HC
938SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
939 struct kexec_segment __user *, segments, unsigned long, flags)
dc009d92
EB
940{
941 struct kimage **dest_image, *image;
dc009d92
EB
942 int result;
943
944 /* We only trust the superuser with rebooting the system. */
945 if (!capable(CAP_SYS_BOOT))
946 return -EPERM;
947
948 /*
949 * Verify we have a legal set of flags
950 * This leaves us room for future extensions.
951 */
952 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
953 return -EINVAL;
954
955 /* Verify we are on the appropriate architecture */
956 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
957 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
dc009d92 958 return -EINVAL;
dc009d92
EB
959
960 /* Put an artificial cap on the number
961 * of segments passed to kexec_load.
962 */
963 if (nr_segments > KEXEC_SEGMENT_MAX)
964 return -EINVAL;
965
966 image = NULL;
967 result = 0;
968
969 /* Because we write directly to the reserved memory
970 * region when loading crash kernels we need a mutex here to
971 * prevent multiple crash kernels from attempting to load
972 * simultaneously, and to prevent a crash kernel from loading
973 * over the top of a in use crash kernel.
974 *
975 * KISS: always take the mutex.
976 */
8c5a1cf0 977 if (!mutex_trylock(&kexec_mutex))
dc009d92 978 return -EBUSY;
72414d3f 979
dc009d92 980 dest_image = &kexec_image;
72414d3f 981 if (flags & KEXEC_ON_CRASH)
dc009d92 982 dest_image = &kexec_crash_image;
dc009d92
EB
983 if (nr_segments > 0) {
984 unsigned long i;
72414d3f 985
dc009d92 986 /* Loading another kernel to reboot into */
72414d3f
MS
987 if ((flags & KEXEC_ON_CRASH) == 0)
988 result = kimage_normal_alloc(&image, entry,
989 nr_segments, segments);
dc009d92
EB
990 /* Loading another kernel to switch to if this one crashes */
991 else if (flags & KEXEC_ON_CRASH) {
992 /* Free any current crash dump kernel before
993 * we corrupt it.
994 */
995 kimage_free(xchg(&kexec_crash_image, NULL));
72414d3f
MS
996 result = kimage_crash_alloc(&image, entry,
997 nr_segments, segments);
558df720 998 crash_map_reserved_pages();
dc009d92 999 }
72414d3f 1000 if (result)
dc009d92 1001 goto out;
72414d3f 1002
3ab83521
HY
1003 if (flags & KEXEC_PRESERVE_CONTEXT)
1004 image->preserve_context = 1;
dc009d92 1005 result = machine_kexec_prepare(image);
72414d3f 1006 if (result)
dc009d92 1007 goto out;
72414d3f
MS
1008
1009 for (i = 0; i < nr_segments; i++) {
dc009d92 1010 result = kimage_load_segment(image, &image->segment[i]);
72414d3f 1011 if (result)
dc009d92 1012 goto out;
dc009d92 1013 }
7fccf032 1014 kimage_terminate(image);
558df720
MH
1015 if (flags & KEXEC_ON_CRASH)
1016 crash_unmap_reserved_pages();
dc009d92
EB
1017 }
1018 /* Install the new kernel, and Uninstall the old */
1019 image = xchg(dest_image, image);
1020
72414d3f 1021out:
8c5a1cf0 1022 mutex_unlock(&kexec_mutex);
dc009d92 1023 kimage_free(image);
72414d3f 1024
dc009d92
EB
1025 return result;
1026}
1027
558df720
MH
1028/*
1029 * Add and remove page tables for crashkernel memory
1030 *
1031 * Provide an empty default implementation here -- architecture
1032 * code may override this
1033 */
1034void __weak crash_map_reserved_pages(void)
1035{}
1036
1037void __weak crash_unmap_reserved_pages(void)
1038{}
1039
dc009d92
EB
1040#ifdef CONFIG_COMPAT
1041asmlinkage long compat_sys_kexec_load(unsigned long entry,
72414d3f
MS
1042 unsigned long nr_segments,
1043 struct compat_kexec_segment __user *segments,
1044 unsigned long flags)
dc009d92
EB
1045{
1046 struct compat_kexec_segment in;
1047 struct kexec_segment out, __user *ksegments;
1048 unsigned long i, result;
1049
1050 /* Don't allow clients that don't understand the native
1051 * architecture to do anything.
1052 */
72414d3f 1053 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
dc009d92 1054 return -EINVAL;
dc009d92 1055
72414d3f 1056 if (nr_segments > KEXEC_SEGMENT_MAX)
dc009d92 1057 return -EINVAL;
dc009d92
EB
1058
1059 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1060 for (i=0; i < nr_segments; i++) {
1061 result = copy_from_user(&in, &segments[i], sizeof(in));
72414d3f 1062 if (result)
dc009d92 1063 return -EFAULT;
dc009d92
EB
1064
1065 out.buf = compat_ptr(in.buf);
1066 out.bufsz = in.bufsz;
1067 out.mem = in.mem;
1068 out.memsz = in.memsz;
1069
1070 result = copy_to_user(&ksegments[i], &out, sizeof(out));
72414d3f 1071 if (result)
dc009d92 1072 return -EFAULT;
dc009d92
EB
1073 }
1074
1075 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1076}
1077#endif
1078
6e274d14 1079void crash_kexec(struct pt_regs *regs)
dc009d92 1080{
8c5a1cf0 1081 /* Take the kexec_mutex here to prevent sys_kexec_load
dc009d92
EB
1082 * running on one cpu from replacing the crash kernel
1083 * we are using after a panic on a different cpu.
1084 *
1085 * If the crash kernel was not located in a fixed area
1086 * of memory the xchg(&kexec_crash_image) would be
1087 * sufficient. But since I reuse the memory...
1088 */
8c5a1cf0 1089 if (mutex_trylock(&kexec_mutex)) {
c0ce7d08 1090 if (kexec_crash_image) {
e996e581 1091 struct pt_regs fixed_regs;
0f4bd46e 1092
e996e581 1093 crash_setup_regs(&fixed_regs, regs);
fd59d231 1094 crash_save_vmcoreinfo();
e996e581 1095 machine_crash_shutdown(&fixed_regs);
c0ce7d08 1096 machine_kexec(kexec_crash_image);
dc009d92 1097 }
8c5a1cf0 1098 mutex_unlock(&kexec_mutex);
dc009d92
EB
1099 }
1100}
cc571658 1101
06a7f711
AW
1102size_t crash_get_memory_size(void)
1103{
e05bd336 1104 size_t size = 0;
06a7f711 1105 mutex_lock(&kexec_mutex);
e05bd336 1106 if (crashk_res.end != crashk_res.start)
28f65c11 1107 size = resource_size(&crashk_res);
06a7f711
AW
1108 mutex_unlock(&kexec_mutex);
1109 return size;
1110}
1111
c0bb9e45
AB
1112void __weak crash_free_reserved_phys_range(unsigned long begin,
1113 unsigned long end)
06a7f711
AW
1114{
1115 unsigned long addr;
1116
e07cee23
JL
1117 for (addr = begin; addr < end; addr += PAGE_SIZE)
1118 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
06a7f711
AW
1119}
1120
1121int crash_shrink_memory(unsigned long new_size)
1122{
1123 int ret = 0;
1124 unsigned long start, end;
bec013c4 1125 unsigned long old_size;
6480e5a0 1126 struct resource *ram_res;
06a7f711
AW
1127
1128 mutex_lock(&kexec_mutex);
1129
1130 if (kexec_crash_image) {
1131 ret = -ENOENT;
1132 goto unlock;
1133 }
1134 start = crashk_res.start;
1135 end = crashk_res.end;
bec013c4
MH
1136 old_size = (end == 0) ? 0 : end - start + 1;
1137 if (new_size >= old_size) {
1138 ret = (new_size == old_size) ? 0 : -EINVAL;
06a7f711
AW
1139 goto unlock;
1140 }
1141
6480e5a0
MH
1142 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1143 if (!ram_res) {
1144 ret = -ENOMEM;
1145 goto unlock;
1146 }
1147
558df720
MH
1148 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1149 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
06a7f711 1150
558df720 1151 crash_map_reserved_pages();
c0bb9e45 1152 crash_free_reserved_phys_range(end, crashk_res.end);
06a7f711 1153
e05bd336 1154 if ((start == end) && (crashk_res.parent != NULL))
06a7f711 1155 release_resource(&crashk_res);
6480e5a0
MH
1156
1157 ram_res->start = end;
1158 ram_res->end = crashk_res.end;
1159 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1160 ram_res->name = "System RAM";
1161
475f9aa6 1162 crashk_res.end = end - 1;
6480e5a0
MH
1163
1164 insert_resource(&iomem_resource, ram_res);
558df720 1165 crash_unmap_reserved_pages();
06a7f711
AW
1166
1167unlock:
1168 mutex_unlock(&kexec_mutex);
1169 return ret;
1170}
1171
85916f81
MD
1172static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1173 size_t data_len)
1174{
1175 struct elf_note note;
1176
1177 note.n_namesz = strlen(name) + 1;
1178 note.n_descsz = data_len;
1179 note.n_type = type;
1180 memcpy(buf, &note, sizeof(note));
1181 buf += (sizeof(note) + 3)/4;
1182 memcpy(buf, name, note.n_namesz);
1183 buf += (note.n_namesz + 3)/4;
1184 memcpy(buf, data, note.n_descsz);
1185 buf += (note.n_descsz + 3)/4;
1186
1187 return buf;
1188}
1189
1190static void final_note(u32 *buf)
1191{
1192 struct elf_note note;
1193
1194 note.n_namesz = 0;
1195 note.n_descsz = 0;
1196 note.n_type = 0;
1197 memcpy(buf, &note, sizeof(note));
1198}
1199
1200void crash_save_cpu(struct pt_regs *regs, int cpu)
1201{
1202 struct elf_prstatus prstatus;
1203 u32 *buf;
1204
4f4b6c1a 1205 if ((cpu < 0) || (cpu >= nr_cpu_ids))
85916f81
MD
1206 return;
1207
1208 /* Using ELF notes here is opportunistic.
1209 * I need a well defined structure format
1210 * for the data I pass, and I need tags
1211 * on the data to indicate what information I have
1212 * squirrelled away. ELF notes happen to provide
1213 * all of that, so there is no need to invent something new.
1214 */
1215 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1216 if (!buf)
1217 return;
1218 memset(&prstatus, 0, sizeof(prstatus));
1219 prstatus.pr_pid = current->pid;
6cd61c0b 1220 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
6672f76a
SH
1221 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1222 &prstatus, sizeof(prstatus));
85916f81
MD
1223 final_note(buf);
1224}
1225
cc571658
VG
1226static int __init crash_notes_memory_init(void)
1227{
1228 /* Allocate memory for saving cpu registers. */
1229 crash_notes = alloc_percpu(note_buf_t);
1230 if (!crash_notes) {
1231 printk("Kexec: Memory allocation for saving cpu register"
1232 " states failed\n");
1233 return -ENOMEM;
1234 }
1235 return 0;
1236}
1237module_init(crash_notes_memory_init)
fd59d231 1238
cba63c30
BW
1239
1240/*
1241 * parsing the "crashkernel" commandline
1242 *
1243 * this code is intended to be called from architecture specific code
1244 */
1245
1246
1247/*
1248 * This function parses command lines in the format
1249 *
1250 * crashkernel=ramsize-range:size[,...][@offset]
1251 *
1252 * The function returns 0 on success and -EINVAL on failure.
1253 */
1254static int __init parse_crashkernel_mem(char *cmdline,
1255 unsigned long long system_ram,
1256 unsigned long long *crash_size,
1257 unsigned long long *crash_base)
1258{
1259 char *cur = cmdline, *tmp;
1260
1261 /* for each entry of the comma-separated list */
1262 do {
1263 unsigned long long start, end = ULLONG_MAX, size;
1264
1265 /* get the start of the range */
1266 start = memparse(cur, &tmp);
1267 if (cur == tmp) {
1268 pr_warning("crashkernel: Memory value expected\n");
1269 return -EINVAL;
1270 }
1271 cur = tmp;
1272 if (*cur != '-') {
1273 pr_warning("crashkernel: '-' expected\n");
1274 return -EINVAL;
1275 }
1276 cur++;
1277
1278 /* if no ':' is here, than we read the end */
1279 if (*cur != ':') {
1280 end = memparse(cur, &tmp);
1281 if (cur == tmp) {
1282 pr_warning("crashkernel: Memory "
1283 "value expected\n");
1284 return -EINVAL;
1285 }
1286 cur = tmp;
1287 if (end <= start) {
1288 pr_warning("crashkernel: end <= start\n");
1289 return -EINVAL;
1290 }
1291 }
1292
1293 if (*cur != ':') {
1294 pr_warning("crashkernel: ':' expected\n");
1295 return -EINVAL;
1296 }
1297 cur++;
1298
1299 size = memparse(cur, &tmp);
1300 if (cur == tmp) {
1301 pr_warning("Memory value expected\n");
1302 return -EINVAL;
1303 }
1304 cur = tmp;
1305 if (size >= system_ram) {
1306 pr_warning("crashkernel: invalid size\n");
1307 return -EINVAL;
1308 }
1309
1310 /* match ? */
be089d79 1311 if (system_ram >= start && system_ram < end) {
cba63c30
BW
1312 *crash_size = size;
1313 break;
1314 }
1315 } while (*cur++ == ',');
1316
1317 if (*crash_size > 0) {
11c7da4b 1318 while (*cur && *cur != ' ' && *cur != '@')
cba63c30
BW
1319 cur++;
1320 if (*cur == '@') {
1321 cur++;
1322 *crash_base = memparse(cur, &tmp);
1323 if (cur == tmp) {
1324 pr_warning("Memory value expected "
1325 "after '@'\n");
1326 return -EINVAL;
1327 }
1328 }
1329 }
1330
1331 return 0;
1332}
1333
1334/*
1335 * That function parses "simple" (old) crashkernel command lines like
1336 *
1337 * crashkernel=size[@offset]
1338 *
1339 * It returns 0 on success and -EINVAL on failure.
1340 */
1341static int __init parse_crashkernel_simple(char *cmdline,
1342 unsigned long long *crash_size,
1343 unsigned long long *crash_base)
1344{
1345 char *cur = cmdline;
1346
1347 *crash_size = memparse(cmdline, &cur);
1348 if (cmdline == cur) {
1349 pr_warning("crashkernel: memory value expected\n");
1350 return -EINVAL;
1351 }
1352
1353 if (*cur == '@')
1354 *crash_base = memparse(cur+1, &cur);
eaa3be6a
ZD
1355 else if (*cur != ' ' && *cur != '\0') {
1356 pr_warning("crashkernel: unrecognized char\n");
1357 return -EINVAL;
1358 }
cba63c30
BW
1359
1360 return 0;
1361}
1362
adbc742b
YL
1363#define SUFFIX_HIGH 0
1364#define SUFFIX_LOW 1
1365#define SUFFIX_NULL 2
1366static __initdata char *suffix_tbl[] = {
1367 [SUFFIX_HIGH] = ",high",
1368 [SUFFIX_LOW] = ",low",
1369 [SUFFIX_NULL] = NULL,
1370};
1371
cba63c30 1372/*
adbc742b
YL
1373 * That function parses "suffix" crashkernel command lines like
1374 *
1375 * crashkernel=size,[high|low]
1376 *
1377 * It returns 0 on success and -EINVAL on failure.
cba63c30 1378 */
adbc742b
YL
1379static int __init parse_crashkernel_suffix(char *cmdline,
1380 unsigned long long *crash_size,
1381 unsigned long long *crash_base,
1382 const char *suffix)
1383{
1384 char *cur = cmdline;
1385
1386 *crash_size = memparse(cmdline, &cur);
1387 if (cmdline == cur) {
1388 pr_warn("crashkernel: memory value expected\n");
1389 return -EINVAL;
1390 }
1391
1392 /* check with suffix */
1393 if (strncmp(cur, suffix, strlen(suffix))) {
1394 pr_warn("crashkernel: unrecognized char\n");
1395 return -EINVAL;
1396 }
1397 cur += strlen(suffix);
1398 if (*cur != ' ' && *cur != '\0') {
1399 pr_warn("crashkernel: unrecognized char\n");
1400 return -EINVAL;
1401 }
1402
1403 return 0;
1404}
1405
1406static __init char *get_last_crashkernel(char *cmdline,
1407 const char *name,
1408 const char *suffix)
1409{
1410 char *p = cmdline, *ck_cmdline = NULL;
1411
1412 /* find crashkernel and use the last one if there are more */
1413 p = strstr(p, name);
1414 while (p) {
1415 char *end_p = strchr(p, ' ');
1416 char *q;
1417
1418 if (!end_p)
1419 end_p = p + strlen(p);
1420
1421 if (!suffix) {
1422 int i;
1423
1424 /* skip the one with any known suffix */
1425 for (i = 0; suffix_tbl[i]; i++) {
1426 q = end_p - strlen(suffix_tbl[i]);
1427 if (!strncmp(q, suffix_tbl[i],
1428 strlen(suffix_tbl[i])))
1429 goto next;
1430 }
1431 ck_cmdline = p;
1432 } else {
1433 q = end_p - strlen(suffix);
1434 if (!strncmp(q, suffix, strlen(suffix)))
1435 ck_cmdline = p;
1436 }
1437next:
1438 p = strstr(p+1, name);
1439 }
1440
1441 if (!ck_cmdline)
1442 return NULL;
1443
1444 return ck_cmdline;
1445}
1446
0212f915 1447static int __init __parse_crashkernel(char *cmdline,
cba63c30
BW
1448 unsigned long long system_ram,
1449 unsigned long long *crash_size,
0212f915 1450 unsigned long long *crash_base,
adbc742b
YL
1451 const char *name,
1452 const char *suffix)
cba63c30 1453{
cba63c30 1454 char *first_colon, *first_space;
adbc742b 1455 char *ck_cmdline;
cba63c30
BW
1456
1457 BUG_ON(!crash_size || !crash_base);
1458 *crash_size = 0;
1459 *crash_base = 0;
1460
adbc742b 1461 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
cba63c30
BW
1462
1463 if (!ck_cmdline)
1464 return -EINVAL;
1465
0212f915 1466 ck_cmdline += strlen(name);
cba63c30 1467
adbc742b
YL
1468 if (suffix)
1469 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1470 crash_base, suffix);
cba63c30
BW
1471 /*
1472 * if the commandline contains a ':', then that's the extended
1473 * syntax -- if not, it must be the classic syntax
1474 */
1475 first_colon = strchr(ck_cmdline, ':');
1476 first_space = strchr(ck_cmdline, ' ');
1477 if (first_colon && (!first_space || first_colon < first_space))
1478 return parse_crashkernel_mem(ck_cmdline, system_ram,
1479 crash_size, crash_base);
1480 else
1481 return parse_crashkernel_simple(ck_cmdline, crash_size,
1482 crash_base);
1483
1484 return 0;
1485}
1486
adbc742b
YL
1487/*
1488 * That function is the entry point for command line parsing and should be
1489 * called from the arch-specific code.
1490 */
0212f915
YL
1491int __init parse_crashkernel(char *cmdline,
1492 unsigned long long system_ram,
1493 unsigned long long *crash_size,
1494 unsigned long long *crash_base)
1495{
1496 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
adbc742b 1497 "crashkernel=", NULL);
0212f915 1498}
55a20ee7
YL
1499
1500int __init parse_crashkernel_high(char *cmdline,
1501 unsigned long long system_ram,
1502 unsigned long long *crash_size,
1503 unsigned long long *crash_base)
1504{
1505 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
adbc742b 1506 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
55a20ee7 1507}
0212f915
YL
1508
1509int __init parse_crashkernel_low(char *cmdline,
1510 unsigned long long system_ram,
1511 unsigned long long *crash_size,
1512 unsigned long long *crash_base)
1513{
1514 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
adbc742b 1515 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
0212f915 1516}
cba63c30 1517
fa8ff292 1518static void update_vmcoreinfo_note(void)
fd59d231 1519{
fa8ff292 1520 u32 *buf = vmcoreinfo_note;
fd59d231
KO
1521
1522 if (!vmcoreinfo_size)
1523 return;
fd59d231
KO
1524 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1525 vmcoreinfo_size);
fd59d231
KO
1526 final_note(buf);
1527}
1528
fa8ff292
MH
1529void crash_save_vmcoreinfo(void)
1530{
63dca8d5 1531 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
fa8ff292
MH
1532 update_vmcoreinfo_note();
1533}
1534
fd59d231
KO
1535void vmcoreinfo_append_str(const char *fmt, ...)
1536{
1537 va_list args;
1538 char buf[0x50];
310faaa9 1539 size_t r;
fd59d231
KO
1540
1541 va_start(args, fmt);
1542 r = vsnprintf(buf, sizeof(buf), fmt, args);
1543 va_end(args);
1544
31c3a3fe 1545 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
fd59d231
KO
1546
1547 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1548
1549 vmcoreinfo_size += r;
1550}
1551
1552/*
1553 * provide an empty default implementation here -- architecture
1554 * code may override this
1555 */
1556void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1557{}
1558
1559unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1560{
1561 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1562}
1563
1564static int __init crash_save_vmcoreinfo_init(void)
1565{
bba1f603
KO
1566 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1567 VMCOREINFO_PAGESIZE(PAGE_SIZE);
fd59d231 1568
bcbba6c1
KO
1569 VMCOREINFO_SYMBOL(init_uts_ns);
1570 VMCOREINFO_SYMBOL(node_online_map);
d034cfab 1571#ifdef CONFIG_MMU
bcbba6c1 1572 VMCOREINFO_SYMBOL(swapper_pg_dir);
d034cfab 1573#endif
bcbba6c1 1574 VMCOREINFO_SYMBOL(_stext);
f1c4069e 1575 VMCOREINFO_SYMBOL(vmap_area_list);
fd59d231
KO
1576
1577#ifndef CONFIG_NEED_MULTIPLE_NODES
bcbba6c1
KO
1578 VMCOREINFO_SYMBOL(mem_map);
1579 VMCOREINFO_SYMBOL(contig_page_data);
fd59d231
KO
1580#endif
1581#ifdef CONFIG_SPARSEMEM
bcbba6c1
KO
1582 VMCOREINFO_SYMBOL(mem_section);
1583 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
c76f860c 1584 VMCOREINFO_STRUCT_SIZE(mem_section);
bcbba6c1 1585 VMCOREINFO_OFFSET(mem_section, section_mem_map);
fd59d231 1586#endif
c76f860c
KO
1587 VMCOREINFO_STRUCT_SIZE(page);
1588 VMCOREINFO_STRUCT_SIZE(pglist_data);
1589 VMCOREINFO_STRUCT_SIZE(zone);
1590 VMCOREINFO_STRUCT_SIZE(free_area);
1591 VMCOREINFO_STRUCT_SIZE(list_head);
1592 VMCOREINFO_SIZE(nodemask_t);
bcbba6c1
KO
1593 VMCOREINFO_OFFSET(page, flags);
1594 VMCOREINFO_OFFSET(page, _count);
1595 VMCOREINFO_OFFSET(page, mapping);
1596 VMCOREINFO_OFFSET(page, lru);
8d67091e
AK
1597 VMCOREINFO_OFFSET(page, _mapcount);
1598 VMCOREINFO_OFFSET(page, private);
bcbba6c1
KO
1599 VMCOREINFO_OFFSET(pglist_data, node_zones);
1600 VMCOREINFO_OFFSET(pglist_data, nr_zones);
fd59d231 1601#ifdef CONFIG_FLAT_NODE_MEM_MAP
bcbba6c1 1602 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
fd59d231 1603#endif
bcbba6c1
KO
1604 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1605 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1606 VMCOREINFO_OFFSET(pglist_data, node_id);
1607 VMCOREINFO_OFFSET(zone, free_area);
1608 VMCOREINFO_OFFSET(zone, vm_stat);
1609 VMCOREINFO_OFFSET(zone, spanned_pages);
1610 VMCOREINFO_OFFSET(free_area, free_list);
1611 VMCOREINFO_OFFSET(list_head, next);
1612 VMCOREINFO_OFFSET(list_head, prev);
13ba3fcb
AK
1613 VMCOREINFO_OFFSET(vmap_area, va_start);
1614 VMCOREINFO_OFFSET(vmap_area, list);
bcbba6c1 1615 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
04d491ab 1616 log_buf_kexec_setup();
83a08e7c 1617 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
bcbba6c1 1618 VMCOREINFO_NUMBER(NR_FREE_PAGES);
122c7a59
KO
1619 VMCOREINFO_NUMBER(PG_lru);
1620 VMCOREINFO_NUMBER(PG_private);
1621 VMCOREINFO_NUMBER(PG_swapcache);
8d67091e 1622 VMCOREINFO_NUMBER(PG_slab);
0d0bf667
MT
1623#ifdef CONFIG_MEMORY_FAILURE
1624 VMCOREINFO_NUMBER(PG_hwpoison);
1625#endif
8d67091e 1626 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
fd59d231
KO
1627
1628 arch_crash_save_vmcoreinfo();
fa8ff292 1629 update_vmcoreinfo_note();
fd59d231
KO
1630
1631 return 0;
1632}
1633
1634module_init(crash_save_vmcoreinfo_init)
3ab83521 1635
7ade3fcc
HY
1636/*
1637 * Move into place and start executing a preloaded standalone
1638 * executable. If nothing was preloaded return an error.
3ab83521
HY
1639 */
1640int kernel_kexec(void)
1641{
1642 int error = 0;
1643
8c5a1cf0 1644 if (!mutex_trylock(&kexec_mutex))
3ab83521
HY
1645 return -EBUSY;
1646 if (!kexec_image) {
1647 error = -EINVAL;
1648 goto Unlock;
1649 }
1650
3ab83521 1651#ifdef CONFIG_KEXEC_JUMP
7ade3fcc 1652 if (kexec_image->preserve_context) {
bcda53fa 1653 lock_system_sleep();
89081d17
HY
1654 pm_prepare_console();
1655 error = freeze_processes();
1656 if (error) {
1657 error = -EBUSY;
1658 goto Restore_console;
1659 }
1660 suspend_console();
d1616302 1661 error = dpm_suspend_start(PMSG_FREEZE);
89081d17
HY
1662 if (error)
1663 goto Resume_console;
d1616302 1664 /* At this point, dpm_suspend_start() has been called,
cf579dfb
RW
1665 * but *not* dpm_suspend_end(). We *must* call
1666 * dpm_suspend_end() now. Otherwise, drivers for
89081d17
HY
1667 * some devices (e.g. interrupt controllers) become
1668 * desynchronized with the actual state of the
1669 * hardware at resume time, and evil weirdness ensues.
1670 */
cf579dfb 1671 error = dpm_suspend_end(PMSG_FREEZE);
89081d17 1672 if (error)
749b0afc
RW
1673 goto Resume_devices;
1674 error = disable_nonboot_cpus();
1675 if (error)
1676 goto Enable_cpus;
2ed8d2b3 1677 local_irq_disable();
2e711c04 1678 error = syscore_suspend();
770824bd 1679 if (error)
749b0afc 1680 goto Enable_irqs;
7ade3fcc 1681 } else
3ab83521 1682#endif
7ade3fcc 1683 {
2a038881 1684 kexec_in_progress = true;
ca195b7f 1685 kernel_restart_prepare(NULL);
3ab83521
HY
1686 printk(KERN_EMERG "Starting new kernel\n");
1687 machine_shutdown();
1688 }
1689
1690 machine_kexec(kexec_image);
1691
3ab83521 1692#ifdef CONFIG_KEXEC_JUMP
7ade3fcc 1693 if (kexec_image->preserve_context) {
19234c08 1694 syscore_resume();
749b0afc 1695 Enable_irqs:
3ab83521 1696 local_irq_enable();
749b0afc 1697 Enable_cpus:
89081d17 1698 enable_nonboot_cpus();
cf579dfb 1699 dpm_resume_start(PMSG_RESTORE);
89081d17 1700 Resume_devices:
d1616302 1701 dpm_resume_end(PMSG_RESTORE);
89081d17
HY
1702 Resume_console:
1703 resume_console();
1704 thaw_processes();
1705 Restore_console:
1706 pm_restore_console();
bcda53fa 1707 unlock_system_sleep();
3ab83521 1708 }
7ade3fcc 1709#endif
3ab83521
HY
1710
1711 Unlock:
8c5a1cf0 1712 mutex_unlock(&kexec_mutex);
3ab83521
HY
1713 return error;
1714}