Commit | Line | Data |
---|---|---|
2965faa5 DY |
1 | /* |
2 | * kexec.c - kexec system call core code. | |
3 | * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com> | |
4 | * | |
5 | * This source code is licensed under the GNU General Public License, | |
6 | * Version 2. See the file COPYING for more details. | |
7 | */ | |
8 | ||
de90a6bc | 9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
2965faa5 DY |
10 | |
11 | #include <linux/capability.h> | |
12 | #include <linux/mm.h> | |
13 | #include <linux/file.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/fs.h> | |
16 | #include <linux/kexec.h> | |
17 | #include <linux/mutex.h> | |
18 | #include <linux/list.h> | |
19 | #include <linux/highmem.h> | |
20 | #include <linux/syscalls.h> | |
21 | #include <linux/reboot.h> | |
22 | #include <linux/ioport.h> | |
23 | #include <linux/hardirq.h> | |
24 | #include <linux/elf.h> | |
25 | #include <linux/elfcore.h> | |
26 | #include <linux/utsname.h> | |
27 | #include <linux/numa.h> | |
28 | #include <linux/suspend.h> | |
29 | #include <linux/device.h> | |
30 | #include <linux/freezer.h> | |
31 | #include <linux/pm.h> | |
32 | #include <linux/cpu.h> | |
33 | #include <linux/uaccess.h> | |
34 | #include <linux/io.h> | |
35 | #include <linux/console.h> | |
36 | #include <linux/vmalloc.h> | |
37 | #include <linux/swap.h> | |
38 | #include <linux/syscore_ops.h> | |
39 | #include <linux/compiler.h> | |
40 | #include <linux/hugetlb.h> | |
41 | ||
42 | #include <asm/page.h> | |
43 | #include <asm/sections.h> | |
44 | ||
45 | #include <crypto/hash.h> | |
46 | #include <crypto/sha.h> | |
47 | #include "kexec_internal.h" | |
48 | ||
49 | DEFINE_MUTEX(kexec_mutex); | |
50 | ||
51 | /* Per cpu memory for storing cpu states in case of system crash. */ | |
52 | note_buf_t __percpu *crash_notes; | |
53 | ||
2965faa5 DY |
54 | /* Flag to indicate we are going to kexec a new kernel */ |
55 | bool kexec_in_progress = false; | |
56 | ||
57 | ||
58 | /* Location of the reserved area for the crash kernel */ | |
59 | struct resource crashk_res = { | |
60 | .name = "Crash kernel", | |
61 | .start = 0, | |
62 | .end = 0, | |
1a085d07 TK |
63 | .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM, |
64 | .desc = IORES_DESC_CRASH_KERNEL | |
2965faa5 DY |
65 | }; |
66 | struct resource crashk_low_res = { | |
67 | .name = "Crash kernel", | |
68 | .start = 0, | |
69 | .end = 0, | |
1a085d07 TK |
70 | .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM, |
71 | .desc = IORES_DESC_CRASH_KERNEL | |
2965faa5 DY |
72 | }; |
73 | ||
74 | int kexec_should_crash(struct task_struct *p) | |
75 | { | |
76 | /* | |
77 | * If crash_kexec_post_notifiers is enabled, don't run | |
78 | * crash_kexec() here yet, which must be run after panic | |
79 | * notifiers in panic(). | |
80 | */ | |
81 | if (crash_kexec_post_notifiers) | |
82 | return 0; | |
83 | /* | |
84 | * There are 4 panic() calls in do_exit() path, each of which | |
85 | * corresponds to each of these 4 conditions. | |
86 | */ | |
87 | if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops) | |
88 | return 1; | |
89 | return 0; | |
90 | } | |
91 | ||
21db79e8 PT |
92 | int kexec_crash_loaded(void) |
93 | { | |
94 | return !!kexec_crash_image; | |
95 | } | |
96 | EXPORT_SYMBOL_GPL(kexec_crash_loaded); | |
97 | ||
2965faa5 DY |
98 | /* |
99 | * When kexec transitions to the new kernel there is a one-to-one | |
100 | * mapping between physical and virtual addresses. On processors | |
101 | * where you can disable the MMU this is trivial, and easy. For | |
102 | * others it is still a simple predictable page table to setup. | |
103 | * | |
104 | * In that environment kexec copies the new kernel to its final | |
105 | * resting place. This means I can only support memory whose | |
106 | * physical address can fit in an unsigned long. In particular | |
107 | * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled. | |
108 | * If the assembly stub has more restrictive requirements | |
109 | * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be | |
110 | * defined more restrictively in <asm/kexec.h>. | |
111 | * | |
112 | * The code for the transition from the current kernel to the | |
113 | * the new kernel is placed in the control_code_buffer, whose size | |
114 | * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single | |
115 | * page of memory is necessary, but some architectures require more. | |
116 | * Because this memory must be identity mapped in the transition from | |
117 | * virtual to physical addresses it must live in the range | |
118 | * 0 - TASK_SIZE, as only the user space mappings are arbitrarily | |
119 | * modifiable. | |
120 | * | |
121 | * The assembly stub in the control code buffer is passed a linked list | |
122 | * of descriptor pages detailing the source pages of the new kernel, | |
123 | * and the destination addresses of those source pages. As this data | |
124 | * structure is not used in the context of the current OS, it must | |
125 | * be self-contained. | |
126 | * | |
127 | * The code has been made to work with highmem pages and will use a | |
128 | * destination page in its final resting place (if it happens | |
129 | * to allocate it). The end product of this is that most of the | |
130 | * physical address space, and most of RAM can be used. | |
131 | * | |
132 | * Future directions include: | |
133 | * - allocating a page table with the control code buffer identity | |
134 | * mapped, to simplify machine_kexec and make kexec_on_panic more | |
135 | * reliable. | |
136 | */ | |
137 | ||
138 | /* | |
139 | * KIMAGE_NO_DEST is an impossible destination address..., for | |
140 | * allocating pages whose destination address we do not care about. | |
141 | */ | |
142 | #define KIMAGE_NO_DEST (-1UL) | |
1730f146 | 143 | #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT) |
2965faa5 DY |
144 | |
145 | static struct page *kimage_alloc_page(struct kimage *image, | |
146 | gfp_t gfp_mask, | |
147 | unsigned long dest); | |
148 | ||
149 | int sanity_check_segment_list(struct kimage *image) | |
150 | { | |
4caf9615 | 151 | int i; |
2965faa5 | 152 | unsigned long nr_segments = image->nr_segments; |
1730f146 | 153 | unsigned long total_pages = 0; |
2965faa5 DY |
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 | */ | |
2965faa5 DY |
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; | |
465d3777 RK |
173 | if (mstart > mend) |
174 | return -EADDRNOTAVAIL; | |
2965faa5 | 175 | if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK)) |
4caf9615 | 176 | return -EADDRNOTAVAIL; |
2965faa5 | 177 | if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT) |
4caf9615 | 178 | return -EADDRNOTAVAIL; |
2965faa5 DY |
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 | */ | |
2965faa5 DY |
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)) | |
4caf9615 | 199 | return -EINVAL; |
2965faa5 DY |
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 | */ | |
2965faa5 DY |
208 | for (i = 0; i < nr_segments; i++) { |
209 | if (image->segment[i].bufsz > image->segment[i].memsz) | |
4caf9615 | 210 | return -EINVAL; |
2965faa5 DY |
211 | } |
212 | ||
1730f146 | 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) > totalram_pages / 2) | |
220 | return -EINVAL; | |
221 | ||
222 | total_pages += PAGE_COUNT(image->segment[i].memsz); | |
223 | } | |
224 | ||
225 | if (total_pages > totalram_pages / 2) | |
226 | return -EINVAL; | |
227 | ||
2965faa5 DY |
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) { | |
2965faa5 DY |
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 */ | |
43546d86 RK |
245 | if ((mstart < phys_to_boot_phys(crashk_res.start)) || |
246 | (mend > phys_to_boot_phys(crashk_res.end))) | |
4caf9615 | 247 | return -EADDRNOTAVAIL; |
2965faa5 DY |
248 | } |
249 | } | |
250 | ||
251 | return 0; | |
252 | } | |
253 | ||
254 | struct 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 | ||
281 | int 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 | ||
299 | static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order) | |
300 | { | |
301 | struct page *pages; | |
302 | ||
303 | pages = alloc_pages(gfp_mask, order); | |
304 | if (pages) { | |
305 | unsigned int count, i; | |
306 | ||
307 | pages->mapping = NULL; | |
308 | set_page_private(pages, order); | |
309 | count = 1 << order; | |
310 | for (i = 0; i < count; i++) | |
311 | SetPageReserved(pages + i); | |
312 | } | |
313 | ||
314 | return pages; | |
315 | } | |
316 | ||
317 | static void kimage_free_pages(struct page *page) | |
318 | { | |
319 | unsigned int order, count, i; | |
320 | ||
321 | order = page_private(page); | |
322 | count = 1 << order; | |
323 | for (i = 0; i < count; i++) | |
324 | ClearPageReserved(page + i); | |
325 | __free_pages(page, order); | |
326 | } | |
327 | ||
328 | void kimage_free_page_list(struct list_head *list) | |
329 | { | |
2b24692b | 330 | struct page *page, *next; |
2965faa5 | 331 | |
2b24692b | 332 | list_for_each_entry_safe(page, next, list, lru) { |
2965faa5 DY |
333 | list_del(&page->lru); |
334 | kimage_free_pages(page); | |
335 | } | |
336 | } | |
337 | ||
338 | static struct page *kimage_alloc_normal_control_pages(struct kimage *image, | |
339 | unsigned int order) | |
340 | { | |
341 | /* Control pages are special, they are the intermediaries | |
342 | * that are needed while we copy the rest of the pages | |
343 | * to their final resting place. As such they must | |
344 | * not conflict with either the destination addresses | |
345 | * or memory the kernel is already using. | |
346 | * | |
347 | * The only case where we really need more than one of | |
348 | * these are for architectures where we cannot disable | |
349 | * the MMU and must instead generate an identity mapped | |
350 | * page table for all of the memory. | |
351 | * | |
352 | * At worst this runs in O(N) of the image size. | |
353 | */ | |
354 | struct list_head extra_pages; | |
355 | struct page *pages; | |
356 | unsigned int count; | |
357 | ||
358 | count = 1 << order; | |
359 | INIT_LIST_HEAD(&extra_pages); | |
360 | ||
361 | /* Loop while I can allocate a page and the page allocated | |
362 | * is a destination page. | |
363 | */ | |
364 | do { | |
365 | unsigned long pfn, epfn, addr, eaddr; | |
366 | ||
367 | pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order); | |
368 | if (!pages) | |
369 | break; | |
43546d86 | 370 | pfn = page_to_boot_pfn(pages); |
2965faa5 DY |
371 | epfn = pfn + count; |
372 | addr = pfn << PAGE_SHIFT; | |
373 | eaddr = epfn << PAGE_SHIFT; | |
374 | if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) || | |
375 | kimage_is_destination_range(image, addr, eaddr)) { | |
376 | list_add(&pages->lru, &extra_pages); | |
377 | pages = NULL; | |
378 | } | |
379 | } while (!pages); | |
380 | ||
381 | if (pages) { | |
382 | /* Remember the allocated page... */ | |
383 | list_add(&pages->lru, &image->control_pages); | |
384 | ||
385 | /* Because the page is already in it's destination | |
386 | * location we will never allocate another page at | |
387 | * that address. Therefore kimage_alloc_pages | |
388 | * will not return it (again) and we don't need | |
389 | * to give it an entry in image->segment[]. | |
390 | */ | |
391 | } | |
392 | /* Deal with the destination pages I have inadvertently allocated. | |
393 | * | |
394 | * Ideally I would convert multi-page allocations into single | |
395 | * page allocations, and add everything to image->dest_pages. | |
396 | * | |
397 | * For now it is simpler to just free the pages. | |
398 | */ | |
399 | kimage_free_page_list(&extra_pages); | |
400 | ||
401 | return pages; | |
402 | } | |
403 | ||
404 | static struct page *kimage_alloc_crash_control_pages(struct kimage *image, | |
405 | unsigned int order) | |
406 | { | |
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. | |
412 | * | |
413 | * Control pages are also the only pags we must allocate | |
414 | * when loading a crash kernel. All of the other pages | |
415 | * are specified by the segments and we just memcpy | |
416 | * into them directly. | |
417 | * | |
418 | * The only case where we really need more than one of | |
419 | * these are for architectures where we cannot disable | |
420 | * the MMU and must instead generate an identity mapped | |
421 | * page table for all of the memory. | |
422 | * | |
423 | * Given the low demand this implements a very simple | |
424 | * allocator that finds the first hole of the appropriate | |
425 | * size in the reserved memory region, and allocates all | |
426 | * of the memory up to and including the hole. | |
427 | */ | |
428 | unsigned long hole_start, hole_end, size; | |
429 | struct page *pages; | |
430 | ||
431 | pages = NULL; | |
432 | size = (1 << order) << PAGE_SHIFT; | |
433 | hole_start = (image->control_page + (size - 1)) & ~(size - 1); | |
434 | hole_end = hole_start + size - 1; | |
435 | while (hole_end <= crashk_res.end) { | |
436 | unsigned long i; | |
437 | ||
8e53c073 | 438 | cond_resched(); |
439 | ||
2965faa5 DY |
440 | if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT) |
441 | break; | |
442 | /* See if I overlap any of the segments */ | |
443 | for (i = 0; i < image->nr_segments; i++) { | |
444 | unsigned long mstart, mend; | |
445 | ||
446 | mstart = image->segment[i].mem; | |
447 | mend = mstart + image->segment[i].memsz - 1; | |
448 | if ((hole_end >= mstart) && (hole_start <= mend)) { | |
449 | /* Advance the hole to the end of the segment */ | |
450 | hole_start = (mend + (size - 1)) & ~(size - 1); | |
451 | hole_end = hole_start + size - 1; | |
452 | break; | |
453 | } | |
454 | } | |
455 | /* If I don't overlap any segments I have found my hole! */ | |
456 | if (i == image->nr_segments) { | |
457 | pages = pfn_to_page(hole_start >> PAGE_SHIFT); | |
04e9949b | 458 | image->control_page = hole_end; |
2965faa5 DY |
459 | break; |
460 | } | |
461 | } | |
2965faa5 DY |
462 | |
463 | return pages; | |
464 | } | |
465 | ||
466 | ||
467 | struct page *kimage_alloc_control_pages(struct kimage *image, | |
468 | unsigned int order) | |
469 | { | |
470 | struct page *pages = NULL; | |
471 | ||
472 | switch (image->type) { | |
473 | case KEXEC_TYPE_DEFAULT: | |
474 | pages = kimage_alloc_normal_control_pages(image, order); | |
475 | break; | |
476 | case KEXEC_TYPE_CRASH: | |
477 | pages = kimage_alloc_crash_control_pages(image, order); | |
478 | break; | |
479 | } | |
480 | ||
481 | return pages; | |
482 | } | |
483 | ||
484 | static int kimage_add_entry(struct kimage *image, kimage_entry_t entry) | |
485 | { | |
486 | if (*image->entry != 0) | |
487 | image->entry++; | |
488 | ||
489 | if (image->entry == image->last_entry) { | |
490 | kimage_entry_t *ind_page; | |
491 | struct page *page; | |
492 | ||
493 | page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST); | |
494 | if (!page) | |
495 | return -ENOMEM; | |
496 | ||
497 | ind_page = page_address(page); | |
43546d86 | 498 | *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION; |
2965faa5 DY |
499 | image->entry = ind_page; |
500 | image->last_entry = ind_page + | |
501 | ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1); | |
502 | } | |
503 | *image->entry = entry; | |
504 | image->entry++; | |
505 | *image->entry = 0; | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | static int kimage_set_destination(struct kimage *image, | |
511 | unsigned long destination) | |
512 | { | |
513 | int result; | |
514 | ||
515 | destination &= PAGE_MASK; | |
516 | result = kimage_add_entry(image, destination | IND_DESTINATION); | |
517 | ||
518 | return result; | |
519 | } | |
520 | ||
521 | ||
522 | static int kimage_add_page(struct kimage *image, unsigned long page) | |
523 | { | |
524 | int result; | |
525 | ||
526 | page &= PAGE_MASK; | |
527 | result = kimage_add_entry(image, page | IND_SOURCE); | |
528 | ||
529 | return result; | |
530 | } | |
531 | ||
532 | ||
533 | static void kimage_free_extra_pages(struct kimage *image) | |
534 | { | |
535 | /* Walk through and free any extra destination pages I may have */ | |
536 | kimage_free_page_list(&image->dest_pages); | |
537 | ||
538 | /* Walk through and free any unusable pages I have cached */ | |
539 | kimage_free_page_list(&image->unusable_pages); | |
540 | ||
541 | } | |
542 | void kimage_terminate(struct kimage *image) | |
543 | { | |
544 | if (*image->entry != 0) | |
545 | image->entry++; | |
546 | ||
547 | *image->entry = IND_DONE; | |
548 | } | |
549 | ||
550 | #define for_each_kimage_entry(image, ptr, entry) \ | |
551 | for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \ | |
552 | ptr = (entry & IND_INDIRECTION) ? \ | |
43546d86 | 553 | boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1) |
2965faa5 DY |
554 | |
555 | static void kimage_free_entry(kimage_entry_t entry) | |
556 | { | |
557 | struct page *page; | |
558 | ||
43546d86 | 559 | page = boot_pfn_to_page(entry >> PAGE_SHIFT); |
2965faa5 DY |
560 | kimage_free_pages(page); |
561 | } | |
562 | ||
563 | void kimage_free(struct kimage *image) | |
564 | { | |
565 | kimage_entry_t *ptr, entry; | |
566 | kimage_entry_t ind = 0; | |
567 | ||
568 | if (!image) | |
569 | return; | |
570 | ||
571 | kimage_free_extra_pages(image); | |
572 | for_each_kimage_entry(image, ptr, entry) { | |
573 | if (entry & IND_INDIRECTION) { | |
574 | /* Free the previous indirection page */ | |
575 | if (ind & IND_INDIRECTION) | |
576 | kimage_free_entry(ind); | |
577 | /* Save this indirection page until we are | |
578 | * done with it. | |
579 | */ | |
580 | ind = entry; | |
581 | } else if (entry & IND_SOURCE) | |
582 | kimage_free_entry(entry); | |
583 | } | |
584 | /* Free the final indirection page */ | |
585 | if (ind & IND_INDIRECTION) | |
586 | kimage_free_entry(ind); | |
587 | ||
588 | /* Handle any machine specific cleanup */ | |
589 | machine_kexec_cleanup(image); | |
590 | ||
591 | /* Free the kexec control pages... */ | |
592 | kimage_free_page_list(&image->control_pages); | |
593 | ||
594 | /* | |
595 | * Free up any temporary buffers allocated. This might hit if | |
596 | * error occurred much later after buffer allocation. | |
597 | */ | |
598 | if (image->file_mode) | |
599 | kimage_file_post_load_cleanup(image); | |
600 | ||
601 | kfree(image); | |
602 | } | |
603 | ||
604 | static kimage_entry_t *kimage_dst_used(struct kimage *image, | |
605 | unsigned long page) | |
606 | { | |
607 | kimage_entry_t *ptr, entry; | |
608 | unsigned long destination = 0; | |
609 | ||
610 | for_each_kimage_entry(image, ptr, entry) { | |
611 | if (entry & IND_DESTINATION) | |
612 | destination = entry & PAGE_MASK; | |
613 | else if (entry & IND_SOURCE) { | |
614 | if (page == destination) | |
615 | return ptr; | |
616 | destination += PAGE_SIZE; | |
617 | } | |
618 | } | |
619 | ||
620 | return NULL; | |
621 | } | |
622 | ||
623 | static struct page *kimage_alloc_page(struct kimage *image, | |
624 | gfp_t gfp_mask, | |
625 | unsigned long destination) | |
626 | { | |
627 | /* | |
628 | * Here we implement safeguards to ensure that a source page | |
629 | * is not copied to its destination page before the data on | |
630 | * the destination page is no longer useful. | |
631 | * | |
632 | * To do this we maintain the invariant that a source page is | |
633 | * either its own destination page, or it is not a | |
634 | * destination page at all. | |
635 | * | |
636 | * That is slightly stronger than required, but the proof | |
637 | * that no problems will not occur is trivial, and the | |
638 | * implementation is simply to verify. | |
639 | * | |
640 | * When allocating all pages normally this algorithm will run | |
641 | * in O(N) time, but in the worst case it will run in O(N^2) | |
642 | * time. If the runtime is a problem the data structures can | |
643 | * be fixed. | |
644 | */ | |
645 | struct page *page; | |
646 | unsigned long addr; | |
647 | ||
648 | /* | |
649 | * Walk through the list of destination pages, and see if I | |
650 | * have a match. | |
651 | */ | |
652 | list_for_each_entry(page, &image->dest_pages, lru) { | |
43546d86 | 653 | addr = page_to_boot_pfn(page) << PAGE_SHIFT; |
2965faa5 DY |
654 | if (addr == destination) { |
655 | list_del(&page->lru); | |
656 | return page; | |
657 | } | |
658 | } | |
659 | page = NULL; | |
660 | while (1) { | |
661 | kimage_entry_t *old; | |
662 | ||
663 | /* Allocate a page, if we run out of memory give up */ | |
664 | page = kimage_alloc_pages(gfp_mask, 0); | |
665 | if (!page) | |
666 | return NULL; | |
667 | /* If the page cannot be used file it away */ | |
43546d86 | 668 | if (page_to_boot_pfn(page) > |
2965faa5 DY |
669 | (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) { |
670 | list_add(&page->lru, &image->unusable_pages); | |
671 | continue; | |
672 | } | |
43546d86 | 673 | addr = page_to_boot_pfn(page) << PAGE_SHIFT; |
2965faa5 DY |
674 | |
675 | /* If it is the destination page we want use it */ | |
676 | if (addr == destination) | |
677 | break; | |
678 | ||
679 | /* If the page is not a destination page use it */ | |
680 | if (!kimage_is_destination_range(image, addr, | |
681 | addr + PAGE_SIZE)) | |
682 | break; | |
683 | ||
684 | /* | |
685 | * I know that the page is someones destination page. | |
686 | * See if there is already a source page for this | |
687 | * destination page. And if so swap the source pages. | |
688 | */ | |
689 | old = kimage_dst_used(image, addr); | |
690 | if (old) { | |
691 | /* If so move it */ | |
692 | unsigned long old_addr; | |
693 | struct page *old_page; | |
694 | ||
695 | old_addr = *old & PAGE_MASK; | |
43546d86 | 696 | old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT); |
2965faa5 DY |
697 | copy_highpage(page, old_page); |
698 | *old = addr | (*old & ~PAGE_MASK); | |
699 | ||
700 | /* The old page I have found cannot be a | |
701 | * destination page, so return it if it's | |
702 | * gfp_flags honor the ones passed in. | |
703 | */ | |
704 | if (!(gfp_mask & __GFP_HIGHMEM) && | |
705 | PageHighMem(old_page)) { | |
706 | kimage_free_pages(old_page); | |
707 | continue; | |
708 | } | |
709 | addr = old_addr; | |
710 | page = old_page; | |
711 | break; | |
712 | } | |
713 | /* Place the page on the destination list, to be used later */ | |
714 | list_add(&page->lru, &image->dest_pages); | |
715 | } | |
716 | ||
717 | return page; | |
718 | } | |
719 | ||
720 | static int kimage_load_normal_segment(struct kimage *image, | |
721 | struct kexec_segment *segment) | |
722 | { | |
723 | unsigned long maddr; | |
724 | size_t ubytes, mbytes; | |
725 | int result; | |
726 | unsigned char __user *buf = NULL; | |
727 | unsigned char *kbuf = NULL; | |
728 | ||
729 | result = 0; | |
730 | if (image->file_mode) | |
731 | kbuf = segment->kbuf; | |
732 | else | |
733 | buf = segment->buf; | |
734 | ubytes = segment->bufsz; | |
735 | mbytes = segment->memsz; | |
736 | maddr = segment->mem; | |
737 | ||
738 | result = kimage_set_destination(image, maddr); | |
739 | if (result < 0) | |
740 | goto out; | |
741 | ||
742 | while (mbytes) { | |
743 | struct page *page; | |
744 | char *ptr; | |
745 | size_t uchunk, mchunk; | |
746 | ||
747 | page = kimage_alloc_page(image, GFP_HIGHUSER, maddr); | |
748 | if (!page) { | |
749 | result = -ENOMEM; | |
750 | goto out; | |
751 | } | |
43546d86 | 752 | result = kimage_add_page(image, page_to_boot_pfn(page) |
2965faa5 DY |
753 | << PAGE_SHIFT); |
754 | if (result < 0) | |
755 | goto out; | |
756 | ||
757 | ptr = kmap(page); | |
758 | /* Start with a clear page */ | |
759 | clear_page(ptr); | |
760 | ptr += maddr & ~PAGE_MASK; | |
761 | mchunk = min_t(size_t, mbytes, | |
762 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
763 | uchunk = min(ubytes, mchunk); | |
764 | ||
765 | /* For file based kexec, source pages are in kernel memory */ | |
766 | if (image->file_mode) | |
767 | memcpy(ptr, kbuf, uchunk); | |
768 | else | |
769 | result = copy_from_user(ptr, buf, uchunk); | |
770 | kunmap(page); | |
771 | if (result) { | |
772 | result = -EFAULT; | |
773 | goto out; | |
774 | } | |
775 | ubytes -= uchunk; | |
776 | maddr += mchunk; | |
777 | if (image->file_mode) | |
778 | kbuf += mchunk; | |
779 | else | |
780 | buf += mchunk; | |
781 | mbytes -= mchunk; | |
782 | } | |
783 | out: | |
784 | return result; | |
785 | } | |
786 | ||
787 | static int kimage_load_crash_segment(struct kimage *image, | |
788 | struct kexec_segment *segment) | |
789 | { | |
790 | /* For crash dumps kernels we simply copy the data from | |
791 | * user space to it's destination. | |
792 | * We do things a page at a time for the sake of kmap. | |
793 | */ | |
794 | unsigned long maddr; | |
795 | size_t ubytes, mbytes; | |
796 | int result; | |
797 | unsigned char __user *buf = NULL; | |
798 | unsigned char *kbuf = NULL; | |
799 | ||
800 | result = 0; | |
801 | if (image->file_mode) | |
802 | kbuf = segment->kbuf; | |
803 | else | |
804 | buf = segment->buf; | |
805 | ubytes = segment->bufsz; | |
806 | mbytes = segment->memsz; | |
807 | maddr = segment->mem; | |
808 | while (mbytes) { | |
809 | struct page *page; | |
810 | char *ptr; | |
811 | size_t uchunk, mchunk; | |
812 | ||
43546d86 | 813 | page = boot_pfn_to_page(maddr >> PAGE_SHIFT); |
2965faa5 DY |
814 | if (!page) { |
815 | result = -ENOMEM; | |
816 | goto out; | |
817 | } | |
818 | ptr = kmap(page); | |
819 | ptr += maddr & ~PAGE_MASK; | |
820 | mchunk = min_t(size_t, mbytes, | |
821 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
822 | uchunk = min(ubytes, mchunk); | |
823 | if (mchunk > uchunk) { | |
824 | /* Zero the trailing part of the page */ | |
825 | memset(ptr + uchunk, 0, mchunk - uchunk); | |
826 | } | |
827 | ||
828 | /* For file based kexec, source pages are in kernel memory */ | |
829 | if (image->file_mode) | |
830 | memcpy(ptr, kbuf, uchunk); | |
831 | else | |
832 | result = copy_from_user(ptr, buf, uchunk); | |
833 | kexec_flush_icache_page(page); | |
834 | kunmap(page); | |
835 | if (result) { | |
836 | result = -EFAULT; | |
837 | goto out; | |
838 | } | |
839 | ubytes -= uchunk; | |
840 | maddr += mchunk; | |
841 | if (image->file_mode) | |
842 | kbuf += mchunk; | |
843 | else | |
844 | buf += mchunk; | |
845 | mbytes -= mchunk; | |
846 | } | |
847 | out: | |
848 | return result; | |
849 | } | |
850 | ||
851 | int kimage_load_segment(struct kimage *image, | |
852 | struct kexec_segment *segment) | |
853 | { | |
854 | int result = -ENOMEM; | |
855 | ||
856 | switch (image->type) { | |
857 | case KEXEC_TYPE_DEFAULT: | |
858 | result = kimage_load_normal_segment(image, segment); | |
859 | break; | |
860 | case KEXEC_TYPE_CRASH: | |
861 | result = kimage_load_crash_segment(image, segment); | |
862 | break; | |
863 | } | |
864 | ||
865 | return result; | |
866 | } | |
867 | ||
868 | struct kimage *kexec_image; | |
869 | struct kimage *kexec_crash_image; | |
870 | int kexec_load_disabled; | |
871 | ||
7bbee5ca HK |
872 | /* |
873 | * No panic_cpu check version of crash_kexec(). This function is called | |
874 | * only when panic_cpu holds the current CPU number; this is the only CPU | |
875 | * which processes crash_kexec routines. | |
876 | */ | |
877 | void __crash_kexec(struct pt_regs *regs) | |
2965faa5 DY |
878 | { |
879 | /* Take the kexec_mutex here to prevent sys_kexec_load | |
880 | * running on one cpu from replacing the crash kernel | |
881 | * we are using after a panic on a different cpu. | |
882 | * | |
883 | * If the crash kernel was not located in a fixed area | |
884 | * of memory the xchg(&kexec_crash_image) would be | |
885 | * sufficient. But since I reuse the memory... | |
886 | */ | |
887 | if (mutex_trylock(&kexec_mutex)) { | |
888 | if (kexec_crash_image) { | |
889 | struct pt_regs fixed_regs; | |
890 | ||
891 | crash_setup_regs(&fixed_regs, regs); | |
892 | crash_save_vmcoreinfo(); | |
893 | machine_crash_shutdown(&fixed_regs); | |
894 | machine_kexec(kexec_crash_image); | |
895 | } | |
896 | mutex_unlock(&kexec_mutex); | |
897 | } | |
898 | } | |
899 | ||
7bbee5ca HK |
900 | void crash_kexec(struct pt_regs *regs) |
901 | { | |
902 | int old_cpu, this_cpu; | |
903 | ||
904 | /* | |
905 | * Only one CPU is allowed to execute the crash_kexec() code as with | |
906 | * panic(). Otherwise parallel calls of panic() and crash_kexec() | |
907 | * may stop each other. To exclude them, we use panic_cpu here too. | |
908 | */ | |
909 | this_cpu = raw_smp_processor_id(); | |
910 | old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu); | |
911 | if (old_cpu == PANIC_CPU_INVALID) { | |
912 | /* This is the 1st CPU which comes here, so go ahead. */ | |
f92bac3b | 913 | printk_safe_flush_on_panic(); |
7bbee5ca HK |
914 | __crash_kexec(regs); |
915 | ||
916 | /* | |
917 | * Reset panic_cpu to allow another panic()/crash_kexec() | |
918 | * call. | |
919 | */ | |
920 | atomic_set(&panic_cpu, PANIC_CPU_INVALID); | |
921 | } | |
922 | } | |
923 | ||
2965faa5 DY |
924 | size_t crash_get_memory_size(void) |
925 | { | |
926 | size_t size = 0; | |
927 | ||
928 | mutex_lock(&kexec_mutex); | |
929 | if (crashk_res.end != crashk_res.start) | |
930 | size = resource_size(&crashk_res); | |
931 | mutex_unlock(&kexec_mutex); | |
932 | return size; | |
933 | } | |
934 | ||
935 | void __weak crash_free_reserved_phys_range(unsigned long begin, | |
936 | unsigned long end) | |
937 | { | |
938 | unsigned long addr; | |
939 | ||
940 | for (addr = begin; addr < end; addr += PAGE_SIZE) | |
43546d86 | 941 | free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT)); |
2965faa5 DY |
942 | } |
943 | ||
944 | int crash_shrink_memory(unsigned long new_size) | |
945 | { | |
946 | int ret = 0; | |
947 | unsigned long start, end; | |
948 | unsigned long old_size; | |
949 | struct resource *ram_res; | |
950 | ||
951 | mutex_lock(&kexec_mutex); | |
952 | ||
953 | if (kexec_crash_image) { | |
954 | ret = -ENOENT; | |
955 | goto unlock; | |
956 | } | |
957 | start = crashk_res.start; | |
958 | end = crashk_res.end; | |
959 | old_size = (end == 0) ? 0 : end - start + 1; | |
960 | if (new_size >= old_size) { | |
961 | ret = (new_size == old_size) ? 0 : -EINVAL; | |
962 | goto unlock; | |
963 | } | |
964 | ||
965 | ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL); | |
966 | if (!ram_res) { | |
967 | ret = -ENOMEM; | |
968 | goto unlock; | |
969 | } | |
970 | ||
971 | start = roundup(start, KEXEC_CRASH_MEM_ALIGN); | |
972 | end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN); | |
973 | ||
2965faa5 DY |
974 | crash_free_reserved_phys_range(end, crashk_res.end); |
975 | ||
976 | if ((start == end) && (crashk_res.parent != NULL)) | |
977 | release_resource(&crashk_res); | |
978 | ||
979 | ram_res->start = end; | |
980 | ram_res->end = crashk_res.end; | |
1a085d07 | 981 | ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM; |
2965faa5 DY |
982 | ram_res->name = "System RAM"; |
983 | ||
984 | crashk_res.end = end - 1; | |
985 | ||
986 | insert_resource(&iomem_resource, ram_res); | |
2965faa5 DY |
987 | |
988 | unlock: | |
989 | mutex_unlock(&kexec_mutex); | |
990 | return ret; | |
991 | } | |
992 | ||
2965faa5 DY |
993 | void crash_save_cpu(struct pt_regs *regs, int cpu) |
994 | { | |
995 | struct elf_prstatus prstatus; | |
996 | u32 *buf; | |
997 | ||
998 | if ((cpu < 0) || (cpu >= nr_cpu_ids)) | |
999 | return; | |
1000 | ||
1001 | /* Using ELF notes here is opportunistic. | |
1002 | * I need a well defined structure format | |
1003 | * for the data I pass, and I need tags | |
1004 | * on the data to indicate what information I have | |
1005 | * squirrelled away. ELF notes happen to provide | |
1006 | * all of that, so there is no need to invent something new. | |
1007 | */ | |
1008 | buf = (u32 *)per_cpu_ptr(crash_notes, cpu); | |
1009 | if (!buf) | |
1010 | return; | |
1011 | memset(&prstatus, 0, sizeof(prstatus)); | |
1012 | prstatus.pr_pid = current->pid; | |
1013 | elf_core_copy_kernel_regs(&prstatus.pr_reg, regs); | |
1014 | buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, | |
1015 | &prstatus, sizeof(prstatus)); | |
1016 | final_note(buf); | |
1017 | } | |
1018 | ||
1019 | static int __init crash_notes_memory_init(void) | |
1020 | { | |
1021 | /* Allocate memory for saving cpu registers. */ | |
bbb78b8f BH |
1022 | size_t size, align; |
1023 | ||
1024 | /* | |
1025 | * crash_notes could be allocated across 2 vmalloc pages when percpu | |
1026 | * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc | |
1027 | * pages are also on 2 continuous physical pages. In this case the | |
1028 | * 2nd part of crash_notes in 2nd page could be lost since only the | |
1029 | * starting address and size of crash_notes are exported through sysfs. | |
1030 | * Here round up the size of crash_notes to the nearest power of two | |
1031 | * and pass it to __alloc_percpu as align value. This can make sure | |
1032 | * crash_notes is allocated inside one physical page. | |
1033 | */ | |
1034 | size = sizeof(note_buf_t); | |
1035 | align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE); | |
1036 | ||
1037 | /* | |
1038 | * Break compile if size is bigger than PAGE_SIZE since crash_notes | |
1039 | * definitely will be in 2 pages with that. | |
1040 | */ | |
1041 | BUILD_BUG_ON(size > PAGE_SIZE); | |
1042 | ||
1043 | crash_notes = __alloc_percpu(size, align); | |
2965faa5 | 1044 | if (!crash_notes) { |
de90a6bc | 1045 | pr_warn("Memory allocation for saving cpu register states failed\n"); |
2965faa5 DY |
1046 | return -ENOMEM; |
1047 | } | |
1048 | return 0; | |
1049 | } | |
1050 | subsys_initcall(crash_notes_memory_init); | |
1051 | ||
1052 | ||
2965faa5 DY |
1053 | /* |
1054 | * Move into place and start executing a preloaded standalone | |
1055 | * executable. If nothing was preloaded return an error. | |
1056 | */ | |
1057 | int kernel_kexec(void) | |
1058 | { | |
1059 | int error = 0; | |
1060 | ||
1061 | if (!mutex_trylock(&kexec_mutex)) | |
1062 | return -EBUSY; | |
1063 | if (!kexec_image) { | |
1064 | error = -EINVAL; | |
1065 | goto Unlock; | |
1066 | } | |
1067 | ||
1068 | #ifdef CONFIG_KEXEC_JUMP | |
1069 | if (kexec_image->preserve_context) { | |
1070 | lock_system_sleep(); | |
1071 | pm_prepare_console(); | |
1072 | error = freeze_processes(); | |
1073 | if (error) { | |
1074 | error = -EBUSY; | |
1075 | goto Restore_console; | |
1076 | } | |
1077 | suspend_console(); | |
1078 | error = dpm_suspend_start(PMSG_FREEZE); | |
1079 | if (error) | |
1080 | goto Resume_console; | |
1081 | /* At this point, dpm_suspend_start() has been called, | |
1082 | * but *not* dpm_suspend_end(). We *must* call | |
1083 | * dpm_suspend_end() now. Otherwise, drivers for | |
1084 | * some devices (e.g. interrupt controllers) become | |
1085 | * desynchronized with the actual state of the | |
1086 | * hardware at resume time, and evil weirdness ensues. | |
1087 | */ | |
1088 | error = dpm_suspend_end(PMSG_FREEZE); | |
1089 | if (error) | |
1090 | goto Resume_devices; | |
1091 | error = disable_nonboot_cpus(); | |
1092 | if (error) | |
1093 | goto Enable_cpus; | |
1094 | local_irq_disable(); | |
1095 | error = syscore_suspend(); | |
1096 | if (error) | |
1097 | goto Enable_irqs; | |
1098 | } else | |
1099 | #endif | |
1100 | { | |
1101 | kexec_in_progress = true; | |
1102 | kernel_restart_prepare(NULL); | |
1103 | migrate_to_reboot_cpu(); | |
1104 | ||
1105 | /* | |
1106 | * migrate_to_reboot_cpu() disables CPU hotplug assuming that | |
1107 | * no further code needs to use CPU hotplug (which is true in | |
1108 | * the reboot case). However, the kexec path depends on using | |
1109 | * CPU hotplug again; so re-enable it here. | |
1110 | */ | |
1111 | cpu_hotplug_enable(); | |
1112 | pr_emerg("Starting new kernel\n"); | |
1113 | machine_shutdown(); | |
1114 | } | |
1115 | ||
1116 | machine_kexec(kexec_image); | |
1117 | ||
1118 | #ifdef CONFIG_KEXEC_JUMP | |
1119 | if (kexec_image->preserve_context) { | |
1120 | syscore_resume(); | |
1121 | Enable_irqs: | |
1122 | local_irq_enable(); | |
1123 | Enable_cpus: | |
1124 | enable_nonboot_cpus(); | |
1125 | dpm_resume_start(PMSG_RESTORE); | |
1126 | Resume_devices: | |
1127 | dpm_resume_end(PMSG_RESTORE); | |
1128 | Resume_console: | |
1129 | resume_console(); | |
1130 | thaw_processes(); | |
1131 | Restore_console: | |
1132 | pm_restore_console(); | |
1133 | unlock_system_sleep(); | |
1134 | } | |
1135 | #endif | |
1136 | ||
1137 | Unlock: | |
1138 | mutex_unlock(&kexec_mutex); | |
1139 | return error; | |
1140 | } | |
1141 | ||
1142 | /* | |
7a0058ec XP |
1143 | * Protection mechanism for crashkernel reserved memory after |
1144 | * the kdump kernel is loaded. | |
2965faa5 DY |
1145 | * |
1146 | * Provide an empty default implementation here -- architecture | |
1147 | * code may override this | |
1148 | */ | |
9b492cf5 XP |
1149 | void __weak arch_kexec_protect_crashkres(void) |
1150 | {} | |
1151 | ||
1152 | void __weak arch_kexec_unprotect_crashkres(void) | |
1153 | {} |