Commit | Line | Data |
---|---|---|
40b0b3f8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
2965faa5 DY |
2 | /* |
3 | * kexec.c - kexec system call core code. | |
4 | * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com> | |
2965faa5 DY |
5 | */ |
6 | ||
de90a6bc | 7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
2965faa5 | 8 | |
400031e0 | 9 | #include <linux/btf.h> |
2965faa5 DY |
10 | #include <linux/capability.h> |
11 | #include <linux/mm.h> | |
12 | #include <linux/file.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/fs.h> | |
15 | #include <linux/kexec.h> | |
16 | #include <linux/mutex.h> | |
17 | #include <linux/list.h> | |
18 | #include <linux/highmem.h> | |
19 | #include <linux/syscalls.h> | |
20 | #include <linux/reboot.h> | |
21 | #include <linux/ioport.h> | |
22 | #include <linux/hardirq.h> | |
23 | #include <linux/elf.h> | |
24 | #include <linux/elfcore.h> | |
25 | #include <linux/utsname.h> | |
26 | #include <linux/numa.h> | |
27 | #include <linux/suspend.h> | |
28 | #include <linux/device.h> | |
29 | #include <linux/freezer.h> | |
f39650de | 30 | #include <linux/panic_notifier.h> |
2965faa5 DY |
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> | |
00089c04 | 41 | #include <linux/objtool.h> |
b2075dbb | 42 | #include <linux/kmsg_dump.h> |
2965faa5 DY |
43 | |
44 | #include <asm/page.h> | |
45 | #include <asm/sections.h> | |
46 | ||
47 | #include <crypto/hash.h> | |
2965faa5 DY |
48 | #include "kexec_internal.h" |
49 | ||
05c62574 | 50 | atomic_t __kexec_lock = ATOMIC_INIT(0); |
2965faa5 DY |
51 | |
52 | /* Per cpu memory for storing cpu states in case of system crash. */ | |
53 | note_buf_t __percpu *crash_notes; | |
54 | ||
2965faa5 DY |
55 | /* Flag to indicate we are going to kexec a new kernel */ |
56 | bool kexec_in_progress = false; | |
57 | ||
58 | ||
59 | /* Location of the reserved area for the crash kernel */ | |
60 | struct resource crashk_res = { | |
61 | .name = "Crash kernel", | |
62 | .start = 0, | |
63 | .end = 0, | |
1a085d07 TK |
64 | .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM, |
65 | .desc = IORES_DESC_CRASH_KERNEL | |
2965faa5 DY |
66 | }; |
67 | struct resource crashk_low_res = { | |
68 | .name = "Crash kernel", | |
69 | .start = 0, | |
70 | .end = 0, | |
1a085d07 TK |
71 | .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM, |
72 | .desc = IORES_DESC_CRASH_KERNEL | |
2965faa5 DY |
73 | }; |
74 | ||
75 | int kexec_should_crash(struct task_struct *p) | |
76 | { | |
77 | /* | |
78 | * If crash_kexec_post_notifiers is enabled, don't run | |
79 | * crash_kexec() here yet, which must be run after panic | |
80 | * notifiers in panic(). | |
81 | */ | |
82 | if (crash_kexec_post_notifiers) | |
83 | return 0; | |
84 | /* | |
05ea0424 | 85 | * There are 4 panic() calls in make_task_dead() path, each of which |
2965faa5 DY |
86 | * corresponds to each of these 4 conditions. |
87 | */ | |
88 | if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops) | |
89 | return 1; | |
90 | return 0; | |
91 | } | |
92 | ||
21db79e8 PT |
93 | int kexec_crash_loaded(void) |
94 | { | |
95 | return !!kexec_crash_image; | |
96 | } | |
97 | EXPORT_SYMBOL_GPL(kexec_crash_loaded); | |
98 | ||
2965faa5 DY |
99 | /* |
100 | * When kexec transitions to the new kernel there is a one-to-one | |
101 | * mapping between physical and virtual addresses. On processors | |
102 | * where you can disable the MMU this is trivial, and easy. For | |
103 | * others it is still a simple predictable page table to setup. | |
104 | * | |
105 | * In that environment kexec copies the new kernel to its final | |
106 | * resting place. This means I can only support memory whose | |
107 | * physical address can fit in an unsigned long. In particular | |
108 | * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled. | |
109 | * If the assembly stub has more restrictive requirements | |
110 | * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be | |
111 | * defined more restrictively in <asm/kexec.h>. | |
112 | * | |
113 | * The code for the transition from the current kernel to the | |
7b7b8a2c | 114 | * new kernel is placed in the control_code_buffer, whose size |
2965faa5 DY |
115 | * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single |
116 | * page of memory is necessary, but some architectures require more. | |
117 | * Because this memory must be identity mapped in the transition from | |
118 | * virtual to physical addresses it must live in the range | |
119 | * 0 - TASK_SIZE, as only the user space mappings are arbitrarily | |
120 | * modifiable. | |
121 | * | |
122 | * The assembly stub in the control code buffer is passed a linked list | |
123 | * of descriptor pages detailing the source pages of the new kernel, | |
124 | * and the destination addresses of those source pages. As this data | |
125 | * structure is not used in the context of the current OS, it must | |
126 | * be self-contained. | |
127 | * | |
128 | * The code has been made to work with highmem pages and will use a | |
129 | * destination page in its final resting place (if it happens | |
130 | * to allocate it). The end product of this is that most of the | |
131 | * physical address space, and most of RAM can be used. | |
132 | * | |
133 | * Future directions include: | |
134 | * - allocating a page table with the control code buffer identity | |
135 | * mapped, to simplify machine_kexec and make kexec_on_panic more | |
136 | * reliable. | |
137 | */ | |
138 | ||
139 | /* | |
140 | * KIMAGE_NO_DEST is an impossible destination address..., for | |
141 | * allocating pages whose destination address we do not care about. | |
142 | */ | |
143 | #define KIMAGE_NO_DEST (-1UL) | |
1730f146 | 144 | #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT) |
2965faa5 DY |
145 | |
146 | static struct page *kimage_alloc_page(struct kimage *image, | |
147 | gfp_t gfp_mask, | |
148 | unsigned long dest); | |
149 | ||
150 | int sanity_check_segment_list(struct kimage *image) | |
151 | { | |
4caf9615 | 152 | int i; |
2965faa5 | 153 | unsigned long nr_segments = image->nr_segments; |
1730f146 | 154 | unsigned long total_pages = 0; |
ca79b0c2 | 155 | unsigned long nr_pages = totalram_pages(); |
2965faa5 DY |
156 | |
157 | /* | |
158 | * Verify we have good destination addresses. The caller is | |
159 | * responsible for making certain we don't attempt to load | |
160 | * the new image into invalid or reserved areas of RAM. This | |
161 | * just verifies it is an address we can use. | |
162 | * | |
163 | * Since the kernel does everything in page size chunks ensure | |
164 | * the destination addresses are page aligned. Too many | |
165 | * special cases crop of when we don't do this. The most | |
166 | * insidious is getting overlapping destination addresses | |
167 | * simply because addresses are changed to page size | |
168 | * granularity. | |
169 | */ | |
2965faa5 DY |
170 | for (i = 0; i < nr_segments; i++) { |
171 | unsigned long mstart, mend; | |
172 | ||
173 | mstart = image->segment[i].mem; | |
174 | mend = mstart + image->segment[i].memsz; | |
465d3777 RK |
175 | if (mstart > mend) |
176 | return -EADDRNOTAVAIL; | |
2965faa5 | 177 | if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK)) |
4caf9615 | 178 | return -EADDRNOTAVAIL; |
2965faa5 | 179 | if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT) |
4caf9615 | 180 | return -EADDRNOTAVAIL; |
2965faa5 DY |
181 | } |
182 | ||
183 | /* Verify our destination addresses do not overlap. | |
184 | * If we alloed overlapping destination addresses | |
185 | * through very weird things can happen with no | |
186 | * easy explanation as one segment stops on another. | |
187 | */ | |
2965faa5 DY |
188 | for (i = 0; i < nr_segments; i++) { |
189 | unsigned long mstart, mend; | |
190 | unsigned long j; | |
191 | ||
192 | mstart = image->segment[i].mem; | |
193 | mend = mstart + image->segment[i].memsz; | |
194 | for (j = 0; j < i; j++) { | |
195 | unsigned long pstart, pend; | |
196 | ||
197 | pstart = image->segment[j].mem; | |
198 | pend = pstart + image->segment[j].memsz; | |
199 | /* Do the segments overlap ? */ | |
200 | if ((mend > pstart) && (mstart < pend)) | |
4caf9615 | 201 | return -EINVAL; |
2965faa5 DY |
202 | } |
203 | } | |
204 | ||
205 | /* Ensure our buffer sizes are strictly less than | |
206 | * our memory sizes. This should always be the case, | |
207 | * and it is easier to check up front than to be surprised | |
208 | * later on. | |
209 | */ | |
2965faa5 DY |
210 | for (i = 0; i < nr_segments; i++) { |
211 | if (image->segment[i].bufsz > image->segment[i].memsz) | |
4caf9615 | 212 | return -EINVAL; |
2965faa5 DY |
213 | } |
214 | ||
1730f146 | 215 | /* |
216 | * Verify that no more than half of memory will be consumed. If the | |
217 | * request from userspace is too large, a large amount of time will be | |
218 | * wasted allocating pages, which can cause a soft lockup. | |
219 | */ | |
220 | for (i = 0; i < nr_segments; i++) { | |
3d6357de | 221 | if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2) |
1730f146 | 222 | return -EINVAL; |
223 | ||
224 | total_pages += PAGE_COUNT(image->segment[i].memsz); | |
225 | } | |
226 | ||
3d6357de | 227 | if (total_pages > nr_pages / 2) |
1730f146 | 228 | return -EINVAL; |
229 | ||
2965faa5 DY |
230 | /* |
231 | * Verify we have good destination addresses. Normally | |
232 | * the caller is responsible for making certain we don't | |
233 | * attempt to load the new image into invalid or reserved | |
234 | * areas of RAM. But crash kernels are preloaded into a | |
235 | * reserved area of ram. We must ensure the addresses | |
236 | * are in the reserved area otherwise preloading the | |
237 | * kernel could corrupt things. | |
238 | */ | |
239 | ||
240 | if (image->type == KEXEC_TYPE_CRASH) { | |
2965faa5 DY |
241 | for (i = 0; i < nr_segments; i++) { |
242 | unsigned long mstart, mend; | |
243 | ||
244 | mstart = image->segment[i].mem; | |
245 | mend = mstart + image->segment[i].memsz - 1; | |
246 | /* Ensure we are within the crash kernel limits */ | |
43546d86 RK |
247 | if ((mstart < phys_to_boot_phys(crashk_res.start)) || |
248 | (mend > phys_to_boot_phys(crashk_res.end))) | |
4caf9615 | 249 | return -EADDRNOTAVAIL; |
2965faa5 DY |
250 | } |
251 | } | |
252 | ||
253 | return 0; | |
254 | } | |
255 | ||
256 | struct kimage *do_kimage_alloc_init(void) | |
257 | { | |
258 | struct kimage *image; | |
259 | ||
260 | /* Allocate a controlling structure */ | |
261 | image = kzalloc(sizeof(*image), GFP_KERNEL); | |
262 | if (!image) | |
263 | return NULL; | |
264 | ||
265 | image->head = 0; | |
266 | image->entry = &image->head; | |
267 | image->last_entry = &image->head; | |
268 | image->control_page = ~0; /* By default this does not apply */ | |
269 | image->type = KEXEC_TYPE_DEFAULT; | |
270 | ||
271 | /* Initialize the list of control pages */ | |
272 | INIT_LIST_HEAD(&image->control_pages); | |
273 | ||
274 | /* Initialize the list of destination pages */ | |
275 | INIT_LIST_HEAD(&image->dest_pages); | |
276 | ||
277 | /* Initialize the list of unusable pages */ | |
278 | INIT_LIST_HEAD(&image->unusable_pages); | |
279 | ||
280 | return image; | |
281 | } | |
282 | ||
283 | int kimage_is_destination_range(struct kimage *image, | |
284 | unsigned long start, | |
285 | unsigned long end) | |
286 | { | |
287 | unsigned long i; | |
288 | ||
289 | for (i = 0; i < image->nr_segments; i++) { | |
290 | unsigned long mstart, mend; | |
291 | ||
292 | mstart = image->segment[i].mem; | |
293 | mend = mstart + image->segment[i].memsz; | |
294 | if ((end > mstart) && (start < mend)) | |
295 | return 1; | |
296 | } | |
297 | ||
298 | return 0; | |
299 | } | |
300 | ||
301 | static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order) | |
302 | { | |
303 | struct page *pages; | |
304 | ||
7c3a6aed TH |
305 | if (fatal_signal_pending(current)) |
306 | return NULL; | |
bba4ed01 | 307 | pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order); |
2965faa5 DY |
308 | if (pages) { |
309 | unsigned int count, i; | |
310 | ||
311 | pages->mapping = NULL; | |
312 | set_page_private(pages, order); | |
313 | count = 1 << order; | |
314 | for (i = 0; i < count; i++) | |
315 | SetPageReserved(pages + i); | |
bba4ed01 TL |
316 | |
317 | arch_kexec_post_alloc_pages(page_address(pages), count, | |
318 | gfp_mask); | |
319 | ||
320 | if (gfp_mask & __GFP_ZERO) | |
321 | for (i = 0; i < count; i++) | |
322 | clear_highpage(pages + i); | |
2965faa5 DY |
323 | } |
324 | ||
325 | return pages; | |
326 | } | |
327 | ||
328 | static void kimage_free_pages(struct page *page) | |
329 | { | |
330 | unsigned int order, count, i; | |
331 | ||
332 | order = page_private(page); | |
333 | count = 1 << order; | |
bba4ed01 TL |
334 | |
335 | arch_kexec_pre_free_pages(page_address(page), count); | |
336 | ||
2965faa5 DY |
337 | for (i = 0; i < count; i++) |
338 | ClearPageReserved(page + i); | |
339 | __free_pages(page, order); | |
340 | } | |
341 | ||
342 | void kimage_free_page_list(struct list_head *list) | |
343 | { | |
2b24692b | 344 | struct page *page, *next; |
2965faa5 | 345 | |
2b24692b | 346 | list_for_each_entry_safe(page, next, list, lru) { |
2965faa5 DY |
347 | list_del(&page->lru); |
348 | kimage_free_pages(page); | |
349 | } | |
350 | } | |
351 | ||
352 | static struct page *kimage_alloc_normal_control_pages(struct kimage *image, | |
353 | unsigned int order) | |
354 | { | |
355 | /* Control pages are special, they are the intermediaries | |
356 | * that are needed while we copy the rest of the pages | |
357 | * to their final resting place. As such they must | |
358 | * not conflict with either the destination addresses | |
359 | * or memory the kernel is already using. | |
360 | * | |
361 | * The only case where we really need more than one of | |
362 | * these are for architectures where we cannot disable | |
363 | * the MMU and must instead generate an identity mapped | |
364 | * page table for all of the memory. | |
365 | * | |
366 | * At worst this runs in O(N) of the image size. | |
367 | */ | |
368 | struct list_head extra_pages; | |
369 | struct page *pages; | |
370 | unsigned int count; | |
371 | ||
372 | count = 1 << order; | |
373 | INIT_LIST_HEAD(&extra_pages); | |
374 | ||
375 | /* Loop while I can allocate a page and the page allocated | |
376 | * is a destination page. | |
377 | */ | |
378 | do { | |
379 | unsigned long pfn, epfn, addr, eaddr; | |
380 | ||
381 | pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order); | |
382 | if (!pages) | |
383 | break; | |
43546d86 | 384 | pfn = page_to_boot_pfn(pages); |
2965faa5 DY |
385 | epfn = pfn + count; |
386 | addr = pfn << PAGE_SHIFT; | |
387 | eaddr = epfn << PAGE_SHIFT; | |
388 | if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) || | |
389 | kimage_is_destination_range(image, addr, eaddr)) { | |
390 | list_add(&pages->lru, &extra_pages); | |
391 | pages = NULL; | |
392 | } | |
393 | } while (!pages); | |
394 | ||
395 | if (pages) { | |
396 | /* Remember the allocated page... */ | |
397 | list_add(&pages->lru, &image->control_pages); | |
398 | ||
399 | /* Because the page is already in it's destination | |
400 | * location we will never allocate another page at | |
401 | * that address. Therefore kimage_alloc_pages | |
402 | * will not return it (again) and we don't need | |
403 | * to give it an entry in image->segment[]. | |
404 | */ | |
405 | } | |
406 | /* Deal with the destination pages I have inadvertently allocated. | |
407 | * | |
408 | * Ideally I would convert multi-page allocations into single | |
409 | * page allocations, and add everything to image->dest_pages. | |
410 | * | |
411 | * For now it is simpler to just free the pages. | |
412 | */ | |
413 | kimage_free_page_list(&extra_pages); | |
414 | ||
415 | return pages; | |
416 | } | |
417 | ||
418 | static struct page *kimage_alloc_crash_control_pages(struct kimage *image, | |
419 | unsigned int order) | |
420 | { | |
421 | /* Control pages are special, they are the intermediaries | |
422 | * that are needed while we copy the rest of the pages | |
423 | * to their final resting place. As such they must | |
424 | * not conflict with either the destination addresses | |
425 | * or memory the kernel is already using. | |
426 | * | |
427 | * Control pages are also the only pags we must allocate | |
428 | * when loading a crash kernel. All of the other pages | |
429 | * are specified by the segments and we just memcpy | |
430 | * into them directly. | |
431 | * | |
432 | * The only case where we really need more than one of | |
433 | * these are for architectures where we cannot disable | |
434 | * the MMU and must instead generate an identity mapped | |
435 | * page table for all of the memory. | |
436 | * | |
437 | * Given the low demand this implements a very simple | |
438 | * allocator that finds the first hole of the appropriate | |
439 | * size in the reserved memory region, and allocates all | |
440 | * of the memory up to and including the hole. | |
441 | */ | |
442 | unsigned long hole_start, hole_end, size; | |
443 | struct page *pages; | |
444 | ||
445 | pages = NULL; | |
446 | size = (1 << order) << PAGE_SHIFT; | |
447 | hole_start = (image->control_page + (size - 1)) & ~(size - 1); | |
448 | hole_end = hole_start + size - 1; | |
449 | while (hole_end <= crashk_res.end) { | |
450 | unsigned long i; | |
451 | ||
8e53c073 | 452 | cond_resched(); |
453 | ||
2965faa5 DY |
454 | if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT) |
455 | break; | |
456 | /* See if I overlap any of the segments */ | |
457 | for (i = 0; i < image->nr_segments; i++) { | |
458 | unsigned long mstart, mend; | |
459 | ||
460 | mstart = image->segment[i].mem; | |
461 | mend = mstart + image->segment[i].memsz - 1; | |
462 | if ((hole_end >= mstart) && (hole_start <= mend)) { | |
463 | /* Advance the hole to the end of the segment */ | |
464 | hole_start = (mend + (size - 1)) & ~(size - 1); | |
465 | hole_end = hole_start + size - 1; | |
466 | break; | |
467 | } | |
468 | } | |
469 | /* If I don't overlap any segments I have found my hole! */ | |
470 | if (i == image->nr_segments) { | |
471 | pages = pfn_to_page(hole_start >> PAGE_SHIFT); | |
04e9949b | 472 | image->control_page = hole_end; |
2965faa5 DY |
473 | break; |
474 | } | |
475 | } | |
2965faa5 | 476 | |
9cf38d55 LJ |
477 | /* Ensure that these pages are decrypted if SME is enabled. */ |
478 | if (pages) | |
479 | arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0); | |
480 | ||
2965faa5 DY |
481 | return pages; |
482 | } | |
483 | ||
484 | ||
485 | struct page *kimage_alloc_control_pages(struct kimage *image, | |
486 | unsigned int order) | |
487 | { | |
488 | struct page *pages = NULL; | |
489 | ||
490 | switch (image->type) { | |
491 | case KEXEC_TYPE_DEFAULT: | |
492 | pages = kimage_alloc_normal_control_pages(image, order); | |
493 | break; | |
494 | case KEXEC_TYPE_CRASH: | |
495 | pages = kimage_alloc_crash_control_pages(image, order); | |
496 | break; | |
497 | } | |
498 | ||
499 | return pages; | |
500 | } | |
501 | ||
1229384f XP |
502 | int kimage_crash_copy_vmcoreinfo(struct kimage *image) |
503 | { | |
504 | struct page *vmcoreinfo_page; | |
505 | void *safecopy; | |
506 | ||
507 | if (image->type != KEXEC_TYPE_CRASH) | |
508 | return 0; | |
509 | ||
510 | /* | |
511 | * For kdump, allocate one vmcoreinfo safe copy from the | |
512 | * crash memory. as we have arch_kexec_protect_crashkres() | |
513 | * after kexec syscall, we naturally protect it from write | |
514 | * (even read) access under kernel direct mapping. But on | |
515 | * the other hand, we still need to operate it when crash | |
516 | * happens to generate vmcoreinfo note, hereby we rely on | |
517 | * vmap for this purpose. | |
518 | */ | |
519 | vmcoreinfo_page = kimage_alloc_control_pages(image, 0); | |
520 | if (!vmcoreinfo_page) { | |
521 | pr_warn("Could not allocate vmcoreinfo buffer\n"); | |
522 | return -ENOMEM; | |
523 | } | |
524 | safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL); | |
525 | if (!safecopy) { | |
526 | pr_warn("Could not vmap vmcoreinfo buffer\n"); | |
527 | return -ENOMEM; | |
528 | } | |
529 | ||
530 | image->vmcoreinfo_data_copy = safecopy; | |
531 | crash_update_vmcoreinfo_safecopy(safecopy); | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
2965faa5 DY |
536 | static int kimage_add_entry(struct kimage *image, kimage_entry_t entry) |
537 | { | |
538 | if (*image->entry != 0) | |
539 | image->entry++; | |
540 | ||
541 | if (image->entry == image->last_entry) { | |
542 | kimage_entry_t *ind_page; | |
543 | struct page *page; | |
544 | ||
545 | page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST); | |
546 | if (!page) | |
547 | return -ENOMEM; | |
548 | ||
549 | ind_page = page_address(page); | |
43546d86 | 550 | *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION; |
2965faa5 DY |
551 | image->entry = ind_page; |
552 | image->last_entry = ind_page + | |
553 | ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1); | |
554 | } | |
555 | *image->entry = entry; | |
556 | image->entry++; | |
557 | *image->entry = 0; | |
558 | ||
559 | return 0; | |
560 | } | |
561 | ||
562 | static int kimage_set_destination(struct kimage *image, | |
563 | unsigned long destination) | |
564 | { | |
2965faa5 | 565 | destination &= PAGE_MASK; |
2965faa5 | 566 | |
32d0c98e | 567 | return kimage_add_entry(image, destination | IND_DESTINATION); |
2965faa5 DY |
568 | } |
569 | ||
570 | ||
571 | static int kimage_add_page(struct kimage *image, unsigned long page) | |
572 | { | |
2965faa5 | 573 | page &= PAGE_MASK; |
2965faa5 | 574 | |
32d0c98e | 575 | return kimage_add_entry(image, page | IND_SOURCE); |
2965faa5 DY |
576 | } |
577 | ||
578 | ||
579 | static void kimage_free_extra_pages(struct kimage *image) | |
580 | { | |
581 | /* Walk through and free any extra destination pages I may have */ | |
582 | kimage_free_page_list(&image->dest_pages); | |
583 | ||
584 | /* Walk through and free any unusable pages I have cached */ | |
585 | kimage_free_page_list(&image->unusable_pages); | |
586 | ||
587 | } | |
de68e4da | 588 | |
2965faa5 DY |
589 | void kimage_terminate(struct kimage *image) |
590 | { | |
591 | if (*image->entry != 0) | |
592 | image->entry++; | |
593 | ||
594 | *image->entry = IND_DONE; | |
595 | } | |
596 | ||
597 | #define for_each_kimage_entry(image, ptr, entry) \ | |
598 | for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \ | |
599 | ptr = (entry & IND_INDIRECTION) ? \ | |
43546d86 | 600 | boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1) |
2965faa5 DY |
601 | |
602 | static void kimage_free_entry(kimage_entry_t entry) | |
603 | { | |
604 | struct page *page; | |
605 | ||
43546d86 | 606 | page = boot_pfn_to_page(entry >> PAGE_SHIFT); |
2965faa5 DY |
607 | kimage_free_pages(page); |
608 | } | |
609 | ||
610 | void kimage_free(struct kimage *image) | |
611 | { | |
612 | kimage_entry_t *ptr, entry; | |
613 | kimage_entry_t ind = 0; | |
614 | ||
615 | if (!image) | |
616 | return; | |
617 | ||
1229384f XP |
618 | if (image->vmcoreinfo_data_copy) { |
619 | crash_update_vmcoreinfo_safecopy(NULL); | |
620 | vunmap(image->vmcoreinfo_data_copy); | |
621 | } | |
622 | ||
2965faa5 DY |
623 | kimage_free_extra_pages(image); |
624 | for_each_kimage_entry(image, ptr, entry) { | |
625 | if (entry & IND_INDIRECTION) { | |
626 | /* Free the previous indirection page */ | |
627 | if (ind & IND_INDIRECTION) | |
628 | kimage_free_entry(ind); | |
629 | /* Save this indirection page until we are | |
630 | * done with it. | |
631 | */ | |
632 | ind = entry; | |
633 | } else if (entry & IND_SOURCE) | |
634 | kimage_free_entry(entry); | |
635 | } | |
636 | /* Free the final indirection page */ | |
637 | if (ind & IND_INDIRECTION) | |
638 | kimage_free_entry(ind); | |
639 | ||
640 | /* Handle any machine specific cleanup */ | |
641 | machine_kexec_cleanup(image); | |
642 | ||
643 | /* Free the kexec control pages... */ | |
644 | kimage_free_page_list(&image->control_pages); | |
645 | ||
646 | /* | |
647 | * Free up any temporary buffers allocated. This might hit if | |
648 | * error occurred much later after buffer allocation. | |
649 | */ | |
650 | if (image->file_mode) | |
651 | kimage_file_post_load_cleanup(image); | |
652 | ||
653 | kfree(image); | |
654 | } | |
655 | ||
656 | static kimage_entry_t *kimage_dst_used(struct kimage *image, | |
657 | unsigned long page) | |
658 | { | |
659 | kimage_entry_t *ptr, entry; | |
660 | unsigned long destination = 0; | |
661 | ||
662 | for_each_kimage_entry(image, ptr, entry) { | |
663 | if (entry & IND_DESTINATION) | |
664 | destination = entry & PAGE_MASK; | |
665 | else if (entry & IND_SOURCE) { | |
666 | if (page == destination) | |
667 | return ptr; | |
668 | destination += PAGE_SIZE; | |
669 | } | |
670 | } | |
671 | ||
672 | return NULL; | |
673 | } | |
674 | ||
675 | static struct page *kimage_alloc_page(struct kimage *image, | |
676 | gfp_t gfp_mask, | |
677 | unsigned long destination) | |
678 | { | |
679 | /* | |
680 | * Here we implement safeguards to ensure that a source page | |
681 | * is not copied to its destination page before the data on | |
682 | * the destination page is no longer useful. | |
683 | * | |
684 | * To do this we maintain the invariant that a source page is | |
685 | * either its own destination page, or it is not a | |
686 | * destination page at all. | |
687 | * | |
688 | * That is slightly stronger than required, but the proof | |
689 | * that no problems will not occur is trivial, and the | |
690 | * implementation is simply to verify. | |
691 | * | |
692 | * When allocating all pages normally this algorithm will run | |
693 | * in O(N) time, but in the worst case it will run in O(N^2) | |
694 | * time. If the runtime is a problem the data structures can | |
695 | * be fixed. | |
696 | */ | |
697 | struct page *page; | |
698 | unsigned long addr; | |
699 | ||
700 | /* | |
701 | * Walk through the list of destination pages, and see if I | |
702 | * have a match. | |
703 | */ | |
704 | list_for_each_entry(page, &image->dest_pages, lru) { | |
43546d86 | 705 | addr = page_to_boot_pfn(page) << PAGE_SHIFT; |
2965faa5 DY |
706 | if (addr == destination) { |
707 | list_del(&page->lru); | |
708 | return page; | |
709 | } | |
710 | } | |
711 | page = NULL; | |
712 | while (1) { | |
713 | kimage_entry_t *old; | |
714 | ||
715 | /* Allocate a page, if we run out of memory give up */ | |
716 | page = kimage_alloc_pages(gfp_mask, 0); | |
717 | if (!page) | |
718 | return NULL; | |
719 | /* If the page cannot be used file it away */ | |
43546d86 | 720 | if (page_to_boot_pfn(page) > |
2965faa5 DY |
721 | (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) { |
722 | list_add(&page->lru, &image->unusable_pages); | |
723 | continue; | |
724 | } | |
43546d86 | 725 | addr = page_to_boot_pfn(page) << PAGE_SHIFT; |
2965faa5 DY |
726 | |
727 | /* If it is the destination page we want use it */ | |
728 | if (addr == destination) | |
729 | break; | |
730 | ||
731 | /* If the page is not a destination page use it */ | |
732 | if (!kimage_is_destination_range(image, addr, | |
733 | addr + PAGE_SIZE)) | |
734 | break; | |
735 | ||
736 | /* | |
737 | * I know that the page is someones destination page. | |
738 | * See if there is already a source page for this | |
739 | * destination page. And if so swap the source pages. | |
740 | */ | |
741 | old = kimage_dst_used(image, addr); | |
742 | if (old) { | |
743 | /* If so move it */ | |
744 | unsigned long old_addr; | |
745 | struct page *old_page; | |
746 | ||
747 | old_addr = *old & PAGE_MASK; | |
43546d86 | 748 | old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT); |
2965faa5 DY |
749 | copy_highpage(page, old_page); |
750 | *old = addr | (*old & ~PAGE_MASK); | |
751 | ||
752 | /* The old page I have found cannot be a | |
753 | * destination page, so return it if it's | |
754 | * gfp_flags honor the ones passed in. | |
755 | */ | |
756 | if (!(gfp_mask & __GFP_HIGHMEM) && | |
757 | PageHighMem(old_page)) { | |
758 | kimage_free_pages(old_page); | |
759 | continue; | |
760 | } | |
2965faa5 DY |
761 | page = old_page; |
762 | break; | |
763 | } | |
764 | /* Place the page on the destination list, to be used later */ | |
765 | list_add(&page->lru, &image->dest_pages); | |
766 | } | |
767 | ||
768 | return page; | |
769 | } | |
770 | ||
771 | static int kimage_load_normal_segment(struct kimage *image, | |
772 | struct kexec_segment *segment) | |
773 | { | |
774 | unsigned long maddr; | |
775 | size_t ubytes, mbytes; | |
776 | int result; | |
777 | unsigned char __user *buf = NULL; | |
778 | unsigned char *kbuf = NULL; | |
779 | ||
2965faa5 DY |
780 | if (image->file_mode) |
781 | kbuf = segment->kbuf; | |
782 | else | |
783 | buf = segment->buf; | |
784 | ubytes = segment->bufsz; | |
785 | mbytes = segment->memsz; | |
786 | maddr = segment->mem; | |
787 | ||
788 | result = kimage_set_destination(image, maddr); | |
789 | if (result < 0) | |
790 | goto out; | |
791 | ||
792 | while (mbytes) { | |
793 | struct page *page; | |
794 | char *ptr; | |
795 | size_t uchunk, mchunk; | |
796 | ||
797 | page = kimage_alloc_page(image, GFP_HIGHUSER, maddr); | |
798 | if (!page) { | |
799 | result = -ENOMEM; | |
800 | goto out; | |
801 | } | |
43546d86 | 802 | result = kimage_add_page(image, page_to_boot_pfn(page) |
2965faa5 DY |
803 | << PAGE_SHIFT); |
804 | if (result < 0) | |
805 | goto out; | |
806 | ||
948084f0 | 807 | ptr = kmap_local_page(page); |
2965faa5 DY |
808 | /* Start with a clear page */ |
809 | clear_page(ptr); | |
810 | ptr += maddr & ~PAGE_MASK; | |
811 | mchunk = min_t(size_t, mbytes, | |
812 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
813 | uchunk = min(ubytes, mchunk); | |
814 | ||
815 | /* For file based kexec, source pages are in kernel memory */ | |
816 | if (image->file_mode) | |
817 | memcpy(ptr, kbuf, uchunk); | |
818 | else | |
819 | result = copy_from_user(ptr, buf, uchunk); | |
948084f0 | 820 | kunmap_local(ptr); |
2965faa5 DY |
821 | if (result) { |
822 | result = -EFAULT; | |
823 | goto out; | |
824 | } | |
825 | ubytes -= uchunk; | |
826 | maddr += mchunk; | |
827 | if (image->file_mode) | |
828 | kbuf += mchunk; | |
829 | else | |
830 | buf += mchunk; | |
831 | mbytes -= mchunk; | |
a8311f64 JF |
832 | |
833 | cond_resched(); | |
2965faa5 DY |
834 | } |
835 | out: | |
836 | return result; | |
837 | } | |
838 | ||
839 | static int kimage_load_crash_segment(struct kimage *image, | |
840 | struct kexec_segment *segment) | |
841 | { | |
842 | /* For crash dumps kernels we simply copy the data from | |
843 | * user space to it's destination. | |
844 | * We do things a page at a time for the sake of kmap. | |
845 | */ | |
846 | unsigned long maddr; | |
847 | size_t ubytes, mbytes; | |
848 | int result; | |
849 | unsigned char __user *buf = NULL; | |
850 | unsigned char *kbuf = NULL; | |
851 | ||
852 | result = 0; | |
853 | if (image->file_mode) | |
854 | kbuf = segment->kbuf; | |
855 | else | |
856 | buf = segment->buf; | |
857 | ubytes = segment->bufsz; | |
858 | mbytes = segment->memsz; | |
859 | maddr = segment->mem; | |
860 | while (mbytes) { | |
861 | struct page *page; | |
862 | char *ptr; | |
863 | size_t uchunk, mchunk; | |
864 | ||
43546d86 | 865 | page = boot_pfn_to_page(maddr >> PAGE_SHIFT); |
2965faa5 DY |
866 | if (!page) { |
867 | result = -ENOMEM; | |
868 | goto out; | |
869 | } | |
9cf38d55 | 870 | arch_kexec_post_alloc_pages(page_address(page), 1, 0); |
948084f0 | 871 | ptr = kmap_local_page(page); |
2965faa5 DY |
872 | ptr += maddr & ~PAGE_MASK; |
873 | mchunk = min_t(size_t, mbytes, | |
874 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
875 | uchunk = min(ubytes, mchunk); | |
876 | if (mchunk > uchunk) { | |
877 | /* Zero the trailing part of the page */ | |
878 | memset(ptr + uchunk, 0, mchunk - uchunk); | |
879 | } | |
880 | ||
881 | /* For file based kexec, source pages are in kernel memory */ | |
882 | if (image->file_mode) | |
883 | memcpy(ptr, kbuf, uchunk); | |
884 | else | |
885 | result = copy_from_user(ptr, buf, uchunk); | |
886 | kexec_flush_icache_page(page); | |
948084f0 | 887 | kunmap_local(ptr); |
9cf38d55 | 888 | arch_kexec_pre_free_pages(page_address(page), 1); |
2965faa5 DY |
889 | if (result) { |
890 | result = -EFAULT; | |
891 | goto out; | |
892 | } | |
893 | ubytes -= uchunk; | |
894 | maddr += mchunk; | |
895 | if (image->file_mode) | |
896 | kbuf += mchunk; | |
897 | else | |
898 | buf += mchunk; | |
899 | mbytes -= mchunk; | |
a8311f64 JF |
900 | |
901 | cond_resched(); | |
2965faa5 DY |
902 | } |
903 | out: | |
904 | return result; | |
905 | } | |
906 | ||
907 | int kimage_load_segment(struct kimage *image, | |
908 | struct kexec_segment *segment) | |
909 | { | |
910 | int result = -ENOMEM; | |
911 | ||
912 | switch (image->type) { | |
913 | case KEXEC_TYPE_DEFAULT: | |
914 | result = kimage_load_normal_segment(image, segment); | |
915 | break; | |
916 | case KEXEC_TYPE_CRASH: | |
917 | result = kimage_load_crash_segment(image, segment); | |
918 | break; | |
919 | } | |
920 | ||
921 | return result; | |
922 | } | |
923 | ||
a42aaad2 RR |
924 | struct kexec_load_limit { |
925 | /* Mutex protects the limit count. */ | |
926 | struct mutex mutex; | |
927 | int limit; | |
928 | }; | |
929 | ||
930 | static struct kexec_load_limit load_limit_reboot = { | |
931 | .mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex), | |
932 | .limit = -1, | |
933 | }; | |
934 | ||
935 | static struct kexec_load_limit load_limit_panic = { | |
936 | .mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex), | |
937 | .limit = -1, | |
938 | }; | |
939 | ||
2965faa5 DY |
940 | struct kimage *kexec_image; |
941 | struct kimage *kexec_crash_image; | |
7e99f8b6 | 942 | static int kexec_load_disabled; |
a42aaad2 | 943 | |
a467257f | 944 | #ifdef CONFIG_SYSCTL |
a42aaad2 RR |
945 | static int kexec_limit_handler(struct ctl_table *table, int write, |
946 | void *buffer, size_t *lenp, loff_t *ppos) | |
947 | { | |
948 | struct kexec_load_limit *limit = table->data; | |
949 | int val; | |
950 | struct ctl_table tmp = { | |
951 | .data = &val, | |
952 | .maxlen = sizeof(val), | |
953 | .mode = table->mode, | |
954 | }; | |
955 | int ret; | |
956 | ||
957 | if (write) { | |
958 | ret = proc_dointvec(&tmp, write, buffer, lenp, ppos); | |
959 | if (ret) | |
960 | return ret; | |
961 | ||
962 | if (val < 0) | |
963 | return -EINVAL; | |
964 | ||
965 | mutex_lock(&limit->mutex); | |
966 | if (limit->limit != -1 && val >= limit->limit) | |
967 | ret = -EINVAL; | |
968 | else | |
969 | limit->limit = val; | |
970 | mutex_unlock(&limit->mutex); | |
971 | ||
972 | return ret; | |
973 | } | |
974 | ||
975 | mutex_lock(&limit->mutex); | |
976 | val = limit->limit; | |
977 | mutex_unlock(&limit->mutex); | |
978 | ||
979 | return proc_dointvec(&tmp, write, buffer, lenp, ppos); | |
980 | } | |
981 | ||
a467257f | 982 | static struct ctl_table kexec_core_sysctls[] = { |
983 | { | |
984 | .procname = "kexec_load_disabled", | |
985 | .data = &kexec_load_disabled, | |
986 | .maxlen = sizeof(int), | |
987 | .mode = 0644, | |
988 | /* only handle a transition from default "0" to "1" */ | |
989 | .proc_handler = proc_dointvec_minmax, | |
990 | .extra1 = SYSCTL_ONE, | |
991 | .extra2 = SYSCTL_ONE, | |
992 | }, | |
a42aaad2 RR |
993 | { |
994 | .procname = "kexec_load_limit_panic", | |
995 | .data = &load_limit_panic, | |
996 | .mode = 0644, | |
997 | .proc_handler = kexec_limit_handler, | |
998 | }, | |
999 | { | |
1000 | .procname = "kexec_load_limit_reboot", | |
1001 | .data = &load_limit_reboot, | |
1002 | .mode = 0644, | |
1003 | .proc_handler = kexec_limit_handler, | |
1004 | }, | |
a467257f | 1005 | { } |
1006 | }; | |
1007 | ||
1008 | static int __init kexec_core_sysctl_init(void) | |
1009 | { | |
1010 | register_sysctl_init("kernel", kexec_core_sysctls); | |
1011 | return 0; | |
1012 | } | |
1013 | late_initcall(kexec_core_sysctl_init); | |
1014 | #endif | |
2965faa5 | 1015 | |
a42aaad2 | 1016 | bool kexec_load_permitted(int kexec_image_type) |
7e99f8b6 | 1017 | { |
a42aaad2 RR |
1018 | struct kexec_load_limit *limit; |
1019 | ||
7e99f8b6 RR |
1020 | /* |
1021 | * Only the superuser can use the kexec syscall and if it has not | |
1022 | * been disabled. | |
1023 | */ | |
a42aaad2 RR |
1024 | if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) |
1025 | return false; | |
1026 | ||
1027 | /* Check limit counter and decrease it.*/ | |
1028 | limit = (kexec_image_type == KEXEC_TYPE_CRASH) ? | |
1029 | &load_limit_panic : &load_limit_reboot; | |
1030 | mutex_lock(&limit->mutex); | |
1031 | if (!limit->limit) { | |
1032 | mutex_unlock(&limit->mutex); | |
1033 | return false; | |
1034 | } | |
1035 | if (limit->limit != -1) | |
1036 | limit->limit--; | |
1037 | mutex_unlock(&limit->mutex); | |
1038 | ||
1039 | return true; | |
7e99f8b6 RR |
1040 | } |
1041 | ||
7bbee5ca HK |
1042 | /* |
1043 | * No panic_cpu check version of crash_kexec(). This function is called | |
1044 | * only when panic_cpu holds the current CPU number; this is the only CPU | |
1045 | * which processes crash_kexec routines. | |
1046 | */ | |
c207aee4 | 1047 | void __noclone __crash_kexec(struct pt_regs *regs) |
2965faa5 | 1048 | { |
05c62574 | 1049 | /* Take the kexec_lock here to prevent sys_kexec_load |
2965faa5 DY |
1050 | * running on one cpu from replacing the crash kernel |
1051 | * we are using after a panic on a different cpu. | |
1052 | * | |
1053 | * If the crash kernel was not located in a fixed area | |
1054 | * of memory the xchg(&kexec_crash_image) would be | |
1055 | * sufficient. But since I reuse the memory... | |
1056 | */ | |
05c62574 | 1057 | if (kexec_trylock()) { |
2965faa5 DY |
1058 | if (kexec_crash_image) { |
1059 | struct pt_regs fixed_regs; | |
1060 | ||
1061 | crash_setup_regs(&fixed_regs, regs); | |
1062 | crash_save_vmcoreinfo(); | |
1063 | machine_crash_shutdown(&fixed_regs); | |
1064 | machine_kexec(kexec_crash_image); | |
1065 | } | |
05c62574 | 1066 | kexec_unlock(); |
2965faa5 DY |
1067 | } |
1068 | } | |
c207aee4 | 1069 | STACK_FRAME_NON_STANDARD(__crash_kexec); |
2965faa5 | 1070 | |
400031e0 | 1071 | __bpf_kfunc void crash_kexec(struct pt_regs *regs) |
7bbee5ca HK |
1072 | { |
1073 | int old_cpu, this_cpu; | |
1074 | ||
1075 | /* | |
1076 | * Only one CPU is allowed to execute the crash_kexec() code as with | |
1077 | * panic(). Otherwise parallel calls of panic() and crash_kexec() | |
1078 | * may stop each other. To exclude them, we use panic_cpu here too. | |
1079 | */ | |
1080 | this_cpu = raw_smp_processor_id(); | |
1081 | old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu); | |
1082 | if (old_cpu == PANIC_CPU_INVALID) { | |
1083 | /* This is the 1st CPU which comes here, so go ahead. */ | |
1084 | __crash_kexec(regs); | |
1085 | ||
1086 | /* | |
1087 | * Reset panic_cpu to allow another panic()/crash_kexec() | |
1088 | * call. | |
1089 | */ | |
1090 | atomic_set(&panic_cpu, PANIC_CPU_INVALID); | |
1091 | } | |
1092 | } | |
1093 | ||
7bb5da0d | 1094 | ssize_t crash_get_memory_size(void) |
2965faa5 | 1095 | { |
7bb5da0d VS |
1096 | ssize_t size = 0; |
1097 | ||
05c62574 | 1098 | if (!kexec_trylock()) |
7bb5da0d | 1099 | return -EBUSY; |
2965faa5 | 1100 | |
2965faa5 DY |
1101 | if (crashk_res.end != crashk_res.start) |
1102 | size = resource_size(&crashk_res); | |
7bb5da0d | 1103 | |
05c62574 | 1104 | kexec_unlock(); |
2965faa5 DY |
1105 | return size; |
1106 | } | |
1107 | ||
2965faa5 DY |
1108 | int crash_shrink_memory(unsigned long new_size) |
1109 | { | |
1110 | int ret = 0; | |
1111 | unsigned long start, end; | |
1112 | unsigned long old_size; | |
1113 | struct resource *ram_res; | |
1114 | ||
05c62574 | 1115 | if (!kexec_trylock()) |
7bb5da0d | 1116 | return -EBUSY; |
2965faa5 DY |
1117 | |
1118 | if (kexec_crash_image) { | |
1119 | ret = -ENOENT; | |
1120 | goto unlock; | |
1121 | } | |
1122 | start = crashk_res.start; | |
1123 | end = crashk_res.end; | |
1124 | old_size = (end == 0) ? 0 : end - start + 1; | |
1125 | if (new_size >= old_size) { | |
1126 | ret = (new_size == old_size) ? 0 : -EINVAL; | |
1127 | goto unlock; | |
1128 | } | |
1129 | ||
1130 | ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL); | |
1131 | if (!ram_res) { | |
1132 | ret = -ENOMEM; | |
1133 | goto unlock; | |
1134 | } | |
1135 | ||
1136 | start = roundup(start, KEXEC_CRASH_MEM_ALIGN); | |
1137 | end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN); | |
1138 | ||
2965faa5 DY |
1139 | crash_free_reserved_phys_range(end, crashk_res.end); |
1140 | ||
1141 | if ((start == end) && (crashk_res.parent != NULL)) | |
1142 | release_resource(&crashk_res); | |
1143 | ||
1144 | ram_res->start = end; | |
1145 | ram_res->end = crashk_res.end; | |
1a085d07 | 1146 | ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM; |
2965faa5 DY |
1147 | ram_res->name = "System RAM"; |
1148 | ||
1149 | crashk_res.end = end - 1; | |
1150 | ||
1151 | insert_resource(&iomem_resource, ram_res); | |
2965faa5 DY |
1152 | |
1153 | unlock: | |
05c62574 | 1154 | kexec_unlock(); |
2965faa5 DY |
1155 | return ret; |
1156 | } | |
1157 | ||
2965faa5 DY |
1158 | void crash_save_cpu(struct pt_regs *regs, int cpu) |
1159 | { | |
1160 | struct elf_prstatus prstatus; | |
1161 | u32 *buf; | |
1162 | ||
1163 | if ((cpu < 0) || (cpu >= nr_cpu_ids)) | |
1164 | return; | |
1165 | ||
1166 | /* Using ELF notes here is opportunistic. | |
1167 | * I need a well defined structure format | |
1168 | * for the data I pass, and I need tags | |
1169 | * on the data to indicate what information I have | |
1170 | * squirrelled away. ELF notes happen to provide | |
1171 | * all of that, so there is no need to invent something new. | |
1172 | */ | |
1173 | buf = (u32 *)per_cpu_ptr(crash_notes, cpu); | |
1174 | if (!buf) | |
1175 | return; | |
1176 | memset(&prstatus, 0, sizeof(prstatus)); | |
f2485a2d | 1177 | prstatus.common.pr_pid = current->pid; |
9554e908 | 1178 | elf_core_copy_regs(&prstatus.pr_reg, regs); |
2965faa5 DY |
1179 | buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, |
1180 | &prstatus, sizeof(prstatus)); | |
1181 | final_note(buf); | |
1182 | } | |
1183 | ||
1184 | static int __init crash_notes_memory_init(void) | |
1185 | { | |
1186 | /* Allocate memory for saving cpu registers. */ | |
bbb78b8f BH |
1187 | size_t size, align; |
1188 | ||
1189 | /* | |
1190 | * crash_notes could be allocated across 2 vmalloc pages when percpu | |
1191 | * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc | |
1192 | * pages are also on 2 continuous physical pages. In this case the | |
1193 | * 2nd part of crash_notes in 2nd page could be lost since only the | |
1194 | * starting address and size of crash_notes are exported through sysfs. | |
1195 | * Here round up the size of crash_notes to the nearest power of two | |
1196 | * and pass it to __alloc_percpu as align value. This can make sure | |
1197 | * crash_notes is allocated inside one physical page. | |
1198 | */ | |
1199 | size = sizeof(note_buf_t); | |
1200 | align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE); | |
1201 | ||
1202 | /* | |
1203 | * Break compile if size is bigger than PAGE_SIZE since crash_notes | |
1204 | * definitely will be in 2 pages with that. | |
1205 | */ | |
1206 | BUILD_BUG_ON(size > PAGE_SIZE); | |
1207 | ||
1208 | crash_notes = __alloc_percpu(size, align); | |
2965faa5 | 1209 | if (!crash_notes) { |
de90a6bc | 1210 | pr_warn("Memory allocation for saving cpu register states failed\n"); |
2965faa5 DY |
1211 | return -ENOMEM; |
1212 | } | |
1213 | return 0; | |
1214 | } | |
1215 | subsys_initcall(crash_notes_memory_init); | |
1216 | ||
1217 | ||
2965faa5 DY |
1218 | /* |
1219 | * Move into place and start executing a preloaded standalone | |
1220 | * executable. If nothing was preloaded return an error. | |
1221 | */ | |
1222 | int kernel_kexec(void) | |
1223 | { | |
1224 | int error = 0; | |
1225 | ||
05c62574 | 1226 | if (!kexec_trylock()) |
2965faa5 DY |
1227 | return -EBUSY; |
1228 | if (!kexec_image) { | |
1229 | error = -EINVAL; | |
1230 | goto Unlock; | |
1231 | } | |
1232 | ||
1233 | #ifdef CONFIG_KEXEC_JUMP | |
1234 | if (kexec_image->preserve_context) { | |
2965faa5 DY |
1235 | pm_prepare_console(); |
1236 | error = freeze_processes(); | |
1237 | if (error) { | |
1238 | error = -EBUSY; | |
1239 | goto Restore_console; | |
1240 | } | |
1241 | suspend_console(); | |
1242 | error = dpm_suspend_start(PMSG_FREEZE); | |
1243 | if (error) | |
1244 | goto Resume_console; | |
1245 | /* At this point, dpm_suspend_start() has been called, | |
1246 | * but *not* dpm_suspend_end(). We *must* call | |
1247 | * dpm_suspend_end() now. Otherwise, drivers for | |
1248 | * some devices (e.g. interrupt controllers) become | |
1249 | * desynchronized with the actual state of the | |
1250 | * hardware at resume time, and evil weirdness ensues. | |
1251 | */ | |
1252 | error = dpm_suspend_end(PMSG_FREEZE); | |
1253 | if (error) | |
1254 | goto Resume_devices; | |
2f1a6fbb | 1255 | error = suspend_disable_secondary_cpus(); |
2965faa5 DY |
1256 | if (error) |
1257 | goto Enable_cpus; | |
1258 | local_irq_disable(); | |
1259 | error = syscore_suspend(); | |
1260 | if (error) | |
1261 | goto Enable_irqs; | |
1262 | } else | |
1263 | #endif | |
1264 | { | |
1265 | kexec_in_progress = true; | |
a119b4e5 | 1266 | kernel_restart_prepare("kexec reboot"); |
2965faa5 DY |
1267 | migrate_to_reboot_cpu(); |
1268 | ||
1269 | /* | |
1270 | * migrate_to_reboot_cpu() disables CPU hotplug assuming that | |
1271 | * no further code needs to use CPU hotplug (which is true in | |
1272 | * the reboot case). However, the kexec path depends on using | |
1273 | * CPU hotplug again; so re-enable it here. | |
1274 | */ | |
1275 | cpu_hotplug_enable(); | |
d42cc530 | 1276 | pr_notice("Starting new kernel\n"); |
2965faa5 DY |
1277 | machine_shutdown(); |
1278 | } | |
1279 | ||
b2075dbb | 1280 | kmsg_dump(KMSG_DUMP_SHUTDOWN); |
2965faa5 DY |
1281 | machine_kexec(kexec_image); |
1282 | ||
1283 | #ifdef CONFIG_KEXEC_JUMP | |
1284 | if (kexec_image->preserve_context) { | |
1285 | syscore_resume(); | |
1286 | Enable_irqs: | |
1287 | local_irq_enable(); | |
1288 | Enable_cpus: | |
2f1a6fbb | 1289 | suspend_enable_secondary_cpus(); |
2965faa5 DY |
1290 | dpm_resume_start(PMSG_RESTORE); |
1291 | Resume_devices: | |
1292 | dpm_resume_end(PMSG_RESTORE); | |
1293 | Resume_console: | |
1294 | resume_console(); | |
1295 | thaw_processes(); | |
1296 | Restore_console: | |
1297 | pm_restore_console(); | |
2965faa5 DY |
1298 | } |
1299 | #endif | |
1300 | ||
1301 | Unlock: | |
05c62574 | 1302 | kexec_unlock(); |
2965faa5 DY |
1303 | return error; |
1304 | } |