Commit | Line | Data |
---|---|---|
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 | ||
cb105258 VG |
9 | #define pr_fmt(fmt) "kexec: " fmt |
10 | ||
c59ede7b | 11 | #include <linux/capability.h> |
dc009d92 EB |
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> | |
8c5a1cf0 | 17 | #include <linux/mutex.h> |
dc009d92 EB |
18 | #include <linux/list.h> |
19 | #include <linux/highmem.h> | |
20 | #include <linux/syscalls.h> | |
21 | #include <linux/reboot.h> | |
dc009d92 | 22 | #include <linux/ioport.h> |
6e274d14 | 23 | #include <linux/hardirq.h> |
85916f81 MD |
24 | #include <linux/elf.h> |
25 | #include <linux/elfcore.h> | |
fd59d231 KO |
26 | #include <linux/utsname.h> |
27 | #include <linux/numa.h> | |
3ab83521 HY |
28 | #include <linux/suspend.h> |
29 | #include <linux/device.h> | |
89081d17 HY |
30 | #include <linux/freezer.h> |
31 | #include <linux/pm.h> | |
32 | #include <linux/cpu.h> | |
33 | #include <linux/console.h> | |
5f41b8cd | 34 | #include <linux/vmalloc.h> |
06a7f711 | 35 | #include <linux/swap.h> |
19234c08 | 36 | #include <linux/syscore_ops.h> |
52f5684c | 37 | #include <linux/compiler.h> |
8f1d26d0 | 38 | #include <linux/hugetlb.h> |
6e274d14 | 39 | |
dc009d92 EB |
40 | #include <asm/page.h> |
41 | #include <asm/uaccess.h> | |
42 | #include <asm/io.h> | |
fd59d231 | 43 | #include <asm/sections.h> |
dc009d92 | 44 | |
12db5562 VG |
45 | #include <crypto/hash.h> |
46 | #include <crypto/sha.h> | |
47 | ||
cc571658 | 48 | /* Per cpu memory for storing cpu states in case of system crash. */ |
43cf38eb | 49 | note_buf_t __percpu *crash_notes; |
cc571658 | 50 | |
fd59d231 | 51 | /* vmcoreinfo stuff */ |
edb79a21 | 52 | static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES]; |
fd59d231 | 53 | u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4]; |
d768281e KO |
54 | size_t vmcoreinfo_size; |
55 | size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data); | |
fd59d231 | 56 | |
4fc9bbf9 KA |
57 | /* Flag to indicate we are going to kexec a new kernel */ |
58 | bool kexec_in_progress = false; | |
59 | ||
12db5562 VG |
60 | /* |
61 | * Declare these symbols weak so that if architecture provides a purgatory, | |
62 | * these will be overridden. | |
63 | */ | |
64 | char __weak kexec_purgatory[0]; | |
65 | size_t __weak kexec_purgatory_size = 0; | |
66 | ||
74ca317c | 67 | #ifdef CONFIG_KEXEC_FILE |
12db5562 | 68 | static int kexec_calculate_store_digests(struct kimage *image); |
74ca317c | 69 | #endif |
12db5562 | 70 | |
dc009d92 EB |
71 | /* Location of the reserved area for the crash kernel */ |
72 | struct resource crashk_res = { | |
73 | .name = "Crash kernel", | |
74 | .start = 0, | |
75 | .end = 0, | |
76 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM | |
77 | }; | |
0212f915 | 78 | struct resource crashk_low_res = { |
157752d8 | 79 | .name = "Crash kernel", |
0212f915 YL |
80 | .start = 0, |
81 | .end = 0, | |
82 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM | |
83 | }; | |
dc009d92 | 84 | |
6e274d14 AN |
85 | int kexec_should_crash(struct task_struct *p) |
86 | { | |
5375b708 HD |
87 | /* |
88 | * If crash_kexec_post_notifiers is enabled, don't run | |
89 | * crash_kexec() here yet, which must be run after panic | |
90 | * notifiers in panic(). | |
91 | */ | |
92 | if (crash_kexec_post_notifiers) | |
93 | return 0; | |
94 | /* | |
95 | * There are 4 panic() calls in do_exit() path, each of which | |
96 | * corresponds to each of these 4 conditions. | |
97 | */ | |
b460cbc5 | 98 | if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops) |
6e274d14 AN |
99 | return 1; |
100 | return 0; | |
101 | } | |
102 | ||
dc009d92 EB |
103 | /* |
104 | * When kexec transitions to the new kernel there is a one-to-one | |
105 | * mapping between physical and virtual addresses. On processors | |
106 | * where you can disable the MMU this is trivial, and easy. For | |
107 | * others it is still a simple predictable page table to setup. | |
108 | * | |
109 | * In that environment kexec copies the new kernel to its final | |
110 | * resting place. This means I can only support memory whose | |
111 | * physical address can fit in an unsigned long. In particular | |
112 | * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled. | |
113 | * If the assembly stub has more restrictive requirements | |
114 | * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be | |
115 | * defined more restrictively in <asm/kexec.h>. | |
116 | * | |
117 | * The code for the transition from the current kernel to the | |
118 | * the new kernel is placed in the control_code_buffer, whose size | |
163f6876 | 119 | * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single |
dc009d92 EB |
120 | * page of memory is necessary, but some architectures require more. |
121 | * Because this memory must be identity mapped in the transition from | |
122 | * virtual to physical addresses it must live in the range | |
123 | * 0 - TASK_SIZE, as only the user space mappings are arbitrarily | |
124 | * modifiable. | |
125 | * | |
126 | * The assembly stub in the control code buffer is passed a linked list | |
127 | * of descriptor pages detailing the source pages of the new kernel, | |
128 | * and the destination addresses of those source pages. As this data | |
129 | * structure is not used in the context of the current OS, it must | |
130 | * be self-contained. | |
131 | * | |
132 | * The code has been made to work with highmem pages and will use a | |
133 | * destination page in its final resting place (if it happens | |
134 | * to allocate it). The end product of this is that most of the | |
135 | * physical address space, and most of RAM can be used. | |
136 | * | |
137 | * Future directions include: | |
138 | * - allocating a page table with the control code buffer identity | |
139 | * mapped, to simplify machine_kexec and make kexec_on_panic more | |
140 | * reliable. | |
141 | */ | |
142 | ||
143 | /* | |
144 | * KIMAGE_NO_DEST is an impossible destination address..., for | |
145 | * allocating pages whose destination address we do not care about. | |
146 | */ | |
147 | #define KIMAGE_NO_DEST (-1UL) | |
148 | ||
72414d3f MS |
149 | static int kimage_is_destination_range(struct kimage *image, |
150 | unsigned long start, unsigned long end); | |
151 | static struct page *kimage_alloc_page(struct kimage *image, | |
9796fdd8 | 152 | gfp_t gfp_mask, |
72414d3f | 153 | unsigned long dest); |
dc009d92 | 154 | |
dabe7862 VG |
155 | static int copy_user_segment_list(struct kimage *image, |
156 | unsigned long nr_segments, | |
157 | struct kexec_segment __user *segments) | |
dc009d92 | 158 | { |
dabe7862 | 159 | int ret; |
dc009d92 | 160 | size_t segment_bytes; |
dc009d92 EB |
161 | |
162 | /* Read in the segments */ | |
163 | image->nr_segments = nr_segments; | |
164 | segment_bytes = nr_segments * sizeof(*segments); | |
dabe7862 VG |
165 | ret = copy_from_user(image->segment, segments, segment_bytes); |
166 | if (ret) | |
167 | ret = -EFAULT; | |
168 | ||
169 | return ret; | |
170 | } | |
171 | ||
172 | static int sanity_check_segment_list(struct kimage *image) | |
173 | { | |
174 | int result, i; | |
175 | unsigned long nr_segments = image->nr_segments; | |
dc009d92 EB |
176 | |
177 | /* | |
178 | * Verify we have good destination addresses. The caller is | |
179 | * responsible for making certain we don't attempt to load | |
180 | * the new image into invalid or reserved areas of RAM. This | |
181 | * just verifies it is an address we can use. | |
182 | * | |
183 | * Since the kernel does everything in page size chunks ensure | |
b595076a | 184 | * the destination addresses are page aligned. Too many |
dc009d92 EB |
185 | * special cases crop of when we don't do this. The most |
186 | * insidious is getting overlapping destination addresses | |
187 | * simply because addresses are changed to page size | |
188 | * granularity. | |
189 | */ | |
190 | result = -EADDRNOTAVAIL; | |
191 | for (i = 0; i < nr_segments; i++) { | |
192 | unsigned long mstart, mend; | |
72414d3f | 193 | |
dc009d92 EB |
194 | mstart = image->segment[i].mem; |
195 | mend = mstart + image->segment[i].memsz; | |
196 | if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK)) | |
dabe7862 | 197 | return result; |
dc009d92 | 198 | if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT) |
dabe7862 | 199 | return result; |
dc009d92 EB |
200 | } |
201 | ||
202 | /* Verify our destination addresses do not overlap. | |
203 | * If we alloed overlapping destination addresses | |
204 | * through very weird things can happen with no | |
205 | * easy explanation as one segment stops on another. | |
206 | */ | |
207 | result = -EINVAL; | |
72414d3f | 208 | for (i = 0; i < nr_segments; i++) { |
dc009d92 EB |
209 | unsigned long mstart, mend; |
210 | unsigned long j; | |
72414d3f | 211 | |
dc009d92 EB |
212 | mstart = image->segment[i].mem; |
213 | mend = mstart + image->segment[i].memsz; | |
72414d3f | 214 | for (j = 0; j < i; j++) { |
dc009d92 EB |
215 | unsigned long pstart, pend; |
216 | pstart = image->segment[j].mem; | |
217 | pend = pstart + image->segment[j].memsz; | |
218 | /* Do the segments overlap ? */ | |
219 | if ((mend > pstart) && (mstart < pend)) | |
dabe7862 | 220 | return result; |
dc009d92 EB |
221 | } |
222 | } | |
223 | ||
224 | /* Ensure our buffer sizes are strictly less than | |
225 | * our memory sizes. This should always be the case, | |
226 | * and it is easier to check up front than to be surprised | |
227 | * later on. | |
228 | */ | |
229 | result = -EINVAL; | |
72414d3f | 230 | for (i = 0; i < nr_segments; i++) { |
dc009d92 | 231 | if (image->segment[i].bufsz > image->segment[i].memsz) |
dabe7862 | 232 | return result; |
dc009d92 EB |
233 | } |
234 | ||
dabe7862 VG |
235 | /* |
236 | * Verify we have good destination addresses. Normally | |
237 | * the caller is responsible for making certain we don't | |
238 | * attempt to load the new image into invalid or reserved | |
239 | * areas of RAM. But crash kernels are preloaded into a | |
240 | * reserved area of ram. We must ensure the addresses | |
241 | * are in the reserved area otherwise preloading the | |
242 | * kernel could corrupt things. | |
243 | */ | |
72414d3f | 244 | |
dabe7862 VG |
245 | if (image->type == KEXEC_TYPE_CRASH) { |
246 | result = -EADDRNOTAVAIL; | |
247 | for (i = 0; i < nr_segments; i++) { | |
248 | unsigned long mstart, mend; | |
249 | ||
250 | mstart = image->segment[i].mem; | |
251 | mend = mstart + image->segment[i].memsz - 1; | |
252 | /* Ensure we are within the crash kernel limits */ | |
253 | if ((mstart < crashk_res.start) || | |
254 | (mend > crashk_res.end)) | |
255 | return result; | |
256 | } | |
257 | } | |
dc009d92 | 258 | |
dabe7862 VG |
259 | return 0; |
260 | } | |
261 | ||
262 | static struct kimage *do_kimage_alloc_init(void) | |
263 | { | |
264 | struct kimage *image; | |
265 | ||
266 | /* Allocate a controlling structure */ | |
267 | image = kzalloc(sizeof(*image), GFP_KERNEL); | |
268 | if (!image) | |
269 | return NULL; | |
270 | ||
271 | image->head = 0; | |
272 | image->entry = &image->head; | |
273 | image->last_entry = &image->head; | |
274 | image->control_page = ~0; /* By default this does not apply */ | |
275 | image->type = KEXEC_TYPE_DEFAULT; | |
276 | ||
277 | /* Initialize the list of control pages */ | |
278 | INIT_LIST_HEAD(&image->control_pages); | |
279 | ||
280 | /* Initialize the list of destination pages */ | |
281 | INIT_LIST_HEAD(&image->dest_pages); | |
282 | ||
283 | /* Initialize the list of unusable pages */ | |
284 | INIT_LIST_HEAD(&image->unusable_pages); | |
285 | ||
286 | return image; | |
dc009d92 EB |
287 | } |
288 | ||
b92e7e0d ZY |
289 | static void kimage_free_page_list(struct list_head *list); |
290 | ||
255aedd9 VG |
291 | static int kimage_alloc_init(struct kimage **rimage, unsigned long entry, |
292 | unsigned long nr_segments, | |
293 | struct kexec_segment __user *segments, | |
294 | unsigned long flags) | |
dc009d92 | 295 | { |
255aedd9 | 296 | int ret; |
dc009d92 | 297 | struct kimage *image; |
255aedd9 VG |
298 | bool kexec_on_panic = flags & KEXEC_ON_CRASH; |
299 | ||
300 | if (kexec_on_panic) { | |
301 | /* Verify we have a valid entry point */ | |
302 | if ((entry < crashk_res.start) || (entry > crashk_res.end)) | |
303 | return -EADDRNOTAVAIL; | |
304 | } | |
dc009d92 EB |
305 | |
306 | /* Allocate and initialize a controlling structure */ | |
dabe7862 VG |
307 | image = do_kimage_alloc_init(); |
308 | if (!image) | |
309 | return -ENOMEM; | |
310 | ||
311 | image->start = entry; | |
312 | ||
255aedd9 VG |
313 | ret = copy_user_segment_list(image, nr_segments, segments); |
314 | if (ret) | |
dabe7862 VG |
315 | goto out_free_image; |
316 | ||
255aedd9 VG |
317 | ret = sanity_check_segment_list(image); |
318 | if (ret) | |
dabe7862 | 319 | goto out_free_image; |
72414d3f | 320 | |
255aedd9 VG |
321 | /* Enable the special crash kernel control page allocation policy. */ |
322 | if (kexec_on_panic) { | |
323 | image->control_page = crashk_res.start; | |
324 | image->type = KEXEC_TYPE_CRASH; | |
325 | } | |
326 | ||
dc009d92 EB |
327 | /* |
328 | * Find a location for the control code buffer, and add it | |
329 | * the vector of segments so that it's pages will also be | |
330 | * counted as destination pages. | |
331 | */ | |
255aedd9 | 332 | ret = -ENOMEM; |
dc009d92 | 333 | image->control_code_page = kimage_alloc_control_pages(image, |
163f6876 | 334 | get_order(KEXEC_CONTROL_PAGE_SIZE)); |
dc009d92 | 335 | if (!image->control_code_page) { |
e1bebcf4 | 336 | pr_err("Could not allocate control_code_buffer\n"); |
dabe7862 | 337 | goto out_free_image; |
dc009d92 EB |
338 | } |
339 | ||
255aedd9 VG |
340 | if (!kexec_on_panic) { |
341 | image->swap_page = kimage_alloc_control_pages(image, 0); | |
342 | if (!image->swap_page) { | |
343 | pr_err("Could not allocate swap buffer\n"); | |
344 | goto out_free_control_pages; | |
345 | } | |
3ab83521 HY |
346 | } |
347 | ||
b92e7e0d ZY |
348 | *rimage = image; |
349 | return 0; | |
dabe7862 | 350 | out_free_control_pages: |
b92e7e0d | 351 | kimage_free_page_list(&image->control_pages); |
dabe7862 | 352 | out_free_image: |
b92e7e0d | 353 | kfree(image); |
255aedd9 | 354 | return ret; |
dc009d92 EB |
355 | } |
356 | ||
74ca317c | 357 | #ifdef CONFIG_KEXEC_FILE |
cb105258 VG |
358 | static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len) |
359 | { | |
360 | struct fd f = fdget(fd); | |
361 | int ret; | |
362 | struct kstat stat; | |
363 | loff_t pos; | |
364 | ssize_t bytes = 0; | |
365 | ||
366 | if (!f.file) | |
367 | return -EBADF; | |
368 | ||
369 | ret = vfs_getattr(&f.file->f_path, &stat); | |
370 | if (ret) | |
371 | goto out; | |
372 | ||
373 | if (stat.size > INT_MAX) { | |
374 | ret = -EFBIG; | |
375 | goto out; | |
376 | } | |
377 | ||
378 | /* Don't hand 0 to vmalloc, it whines. */ | |
379 | if (stat.size == 0) { | |
380 | ret = -EINVAL; | |
381 | goto out; | |
382 | } | |
383 | ||
384 | *buf = vmalloc(stat.size); | |
385 | if (!*buf) { | |
386 | ret = -ENOMEM; | |
387 | goto out; | |
388 | } | |
389 | ||
390 | pos = 0; | |
391 | while (pos < stat.size) { | |
392 | bytes = kernel_read(f.file, pos, (char *)(*buf) + pos, | |
393 | stat.size - pos); | |
394 | if (bytes < 0) { | |
395 | vfree(*buf); | |
396 | ret = bytes; | |
397 | goto out; | |
398 | } | |
399 | ||
400 | if (bytes == 0) | |
401 | break; | |
402 | pos += bytes; | |
403 | } | |
404 | ||
405 | if (pos != stat.size) { | |
406 | ret = -EBADF; | |
407 | vfree(*buf); | |
408 | goto out; | |
409 | } | |
410 | ||
411 | *buf_len = pos; | |
412 | out: | |
413 | fdput(f); | |
414 | return ret; | |
415 | } | |
416 | ||
417 | /* Architectures can provide this probe function */ | |
418 | int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf, | |
419 | unsigned long buf_len) | |
420 | { | |
421 | return -ENOEXEC; | |
422 | } | |
423 | ||
424 | void * __weak arch_kexec_kernel_image_load(struct kimage *image) | |
425 | { | |
426 | return ERR_PTR(-ENOEXEC); | |
427 | } | |
428 | ||
429 | void __weak arch_kimage_file_post_load_cleanup(struct kimage *image) | |
430 | { | |
431 | } | |
432 | ||
8e7d8381 VG |
433 | int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf, |
434 | unsigned long buf_len) | |
435 | { | |
436 | return -EKEYREJECTED; | |
437 | } | |
438 | ||
12db5562 VG |
439 | /* Apply relocations of type RELA */ |
440 | int __weak | |
441 | arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
442 | unsigned int relsec) | |
443 | { | |
444 | pr_err("RELA relocation unsupported.\n"); | |
445 | return -ENOEXEC; | |
446 | } | |
447 | ||
448 | /* Apply relocations of type REL */ | |
449 | int __weak | |
450 | arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs, | |
451 | unsigned int relsec) | |
452 | { | |
453 | pr_err("REL relocation unsupported.\n"); | |
454 | return -ENOEXEC; | |
455 | } | |
456 | ||
cb105258 | 457 | /* |
ad699349 | 458 | * Free up memory used by kernel, initrd, and command line. This is temporary |
cb105258 VG |
459 | * memory allocation which is not needed any more after these buffers have |
460 | * been loaded into separate segments and have been copied elsewhere. | |
461 | */ | |
462 | static void kimage_file_post_load_cleanup(struct kimage *image) | |
463 | { | |
12db5562 VG |
464 | struct purgatory_info *pi = &image->purgatory_info; |
465 | ||
cb105258 VG |
466 | vfree(image->kernel_buf); |
467 | image->kernel_buf = NULL; | |
468 | ||
469 | vfree(image->initrd_buf); | |
470 | image->initrd_buf = NULL; | |
471 | ||
472 | kfree(image->cmdline_buf); | |
473 | image->cmdline_buf = NULL; | |
474 | ||
12db5562 VG |
475 | vfree(pi->purgatory_buf); |
476 | pi->purgatory_buf = NULL; | |
477 | ||
478 | vfree(pi->sechdrs); | |
479 | pi->sechdrs = NULL; | |
480 | ||
cb105258 VG |
481 | /* See if architecture has anything to cleanup post load */ |
482 | arch_kimage_file_post_load_cleanup(image); | |
27f48d3e VG |
483 | |
484 | /* | |
485 | * Above call should have called into bootloader to free up | |
486 | * any data stored in kimage->image_loader_data. It should | |
487 | * be ok now to free it up. | |
488 | */ | |
489 | kfree(image->image_loader_data); | |
490 | image->image_loader_data = NULL; | |
cb105258 VG |
491 | } |
492 | ||
493 | /* | |
494 | * In file mode list of segments is prepared by kernel. Copy relevant | |
495 | * data from user space, do error checking, prepare segment list | |
496 | */ | |
497 | static int | |
498 | kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd, | |
499 | const char __user *cmdline_ptr, | |
500 | unsigned long cmdline_len, unsigned flags) | |
501 | { | |
502 | int ret = 0; | |
503 | void *ldata; | |
504 | ||
505 | ret = copy_file_from_fd(kernel_fd, &image->kernel_buf, | |
506 | &image->kernel_buf_len); | |
507 | if (ret) | |
508 | return ret; | |
509 | ||
510 | /* Call arch image probe handlers */ | |
511 | ret = arch_kexec_kernel_image_probe(image, image->kernel_buf, | |
512 | image->kernel_buf_len); | |
513 | ||
514 | if (ret) | |
515 | goto out; | |
516 | ||
8e7d8381 VG |
517 | #ifdef CONFIG_KEXEC_VERIFY_SIG |
518 | ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf, | |
519 | image->kernel_buf_len); | |
520 | if (ret) { | |
521 | pr_debug("kernel signature verification failed.\n"); | |
522 | goto out; | |
523 | } | |
524 | pr_debug("kernel signature verification successful.\n"); | |
525 | #endif | |
cb105258 VG |
526 | /* It is possible that there no initramfs is being loaded */ |
527 | if (!(flags & KEXEC_FILE_NO_INITRAMFS)) { | |
528 | ret = copy_file_from_fd(initrd_fd, &image->initrd_buf, | |
529 | &image->initrd_buf_len); | |
530 | if (ret) | |
531 | goto out; | |
532 | } | |
533 | ||
534 | if (cmdline_len) { | |
535 | image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL); | |
536 | if (!image->cmdline_buf) { | |
537 | ret = -ENOMEM; | |
538 | goto out; | |
539 | } | |
540 | ||
541 | ret = copy_from_user(image->cmdline_buf, cmdline_ptr, | |
542 | cmdline_len); | |
543 | if (ret) { | |
544 | ret = -EFAULT; | |
545 | goto out; | |
546 | } | |
547 | ||
548 | image->cmdline_buf_len = cmdline_len; | |
549 | ||
550 | /* command line should be a string with last byte null */ | |
551 | if (image->cmdline_buf[cmdline_len - 1] != '\0') { | |
552 | ret = -EINVAL; | |
553 | goto out; | |
554 | } | |
555 | } | |
556 | ||
557 | /* Call arch image load handlers */ | |
558 | ldata = arch_kexec_kernel_image_load(image); | |
559 | ||
560 | if (IS_ERR(ldata)) { | |
561 | ret = PTR_ERR(ldata); | |
562 | goto out; | |
563 | } | |
564 | ||
565 | image->image_loader_data = ldata; | |
566 | out: | |
567 | /* In case of error, free up all allocated memory in this function */ | |
568 | if (ret) | |
569 | kimage_file_post_load_cleanup(image); | |
570 | return ret; | |
571 | } | |
572 | ||
573 | static int | |
574 | kimage_file_alloc_init(struct kimage **rimage, int kernel_fd, | |
575 | int initrd_fd, const char __user *cmdline_ptr, | |
576 | unsigned long cmdline_len, unsigned long flags) | |
577 | { | |
578 | int ret; | |
579 | struct kimage *image; | |
dd5f7260 | 580 | bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH; |
cb105258 VG |
581 | |
582 | image = do_kimage_alloc_init(); | |
583 | if (!image) | |
584 | return -ENOMEM; | |
585 | ||
586 | image->file_mode = 1; | |
587 | ||
dd5f7260 VG |
588 | if (kexec_on_panic) { |
589 | /* Enable special crash kernel control page alloc policy. */ | |
590 | image->control_page = crashk_res.start; | |
591 | image->type = KEXEC_TYPE_CRASH; | |
592 | } | |
593 | ||
cb105258 VG |
594 | ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd, |
595 | cmdline_ptr, cmdline_len, flags); | |
596 | if (ret) | |
597 | goto out_free_image; | |
598 | ||
599 | ret = sanity_check_segment_list(image); | |
600 | if (ret) | |
601 | goto out_free_post_load_bufs; | |
602 | ||
603 | ret = -ENOMEM; | |
604 | image->control_code_page = kimage_alloc_control_pages(image, | |
605 | get_order(KEXEC_CONTROL_PAGE_SIZE)); | |
606 | if (!image->control_code_page) { | |
607 | pr_err("Could not allocate control_code_buffer\n"); | |
608 | goto out_free_post_load_bufs; | |
609 | } | |
610 | ||
dd5f7260 VG |
611 | if (!kexec_on_panic) { |
612 | image->swap_page = kimage_alloc_control_pages(image, 0); | |
613 | if (!image->swap_page) { | |
d5393955 | 614 | pr_err("Could not allocate swap buffer\n"); |
dd5f7260 VG |
615 | goto out_free_control_pages; |
616 | } | |
cb105258 VG |
617 | } |
618 | ||
619 | *rimage = image; | |
620 | return 0; | |
621 | out_free_control_pages: | |
622 | kimage_free_page_list(&image->control_pages); | |
623 | out_free_post_load_bufs: | |
624 | kimage_file_post_load_cleanup(image); | |
cb105258 VG |
625 | out_free_image: |
626 | kfree(image); | |
627 | return ret; | |
628 | } | |
74ca317c VG |
629 | #else /* CONFIG_KEXEC_FILE */ |
630 | static inline void kimage_file_post_load_cleanup(struct kimage *image) { } | |
631 | #endif /* CONFIG_KEXEC_FILE */ | |
cb105258 | 632 | |
72414d3f MS |
633 | static int kimage_is_destination_range(struct kimage *image, |
634 | unsigned long start, | |
635 | unsigned long end) | |
dc009d92 EB |
636 | { |
637 | unsigned long i; | |
638 | ||
639 | for (i = 0; i < image->nr_segments; i++) { | |
640 | unsigned long mstart, mend; | |
72414d3f | 641 | |
dc009d92 | 642 | mstart = image->segment[i].mem; |
72414d3f MS |
643 | mend = mstart + image->segment[i].memsz; |
644 | if ((end > mstart) && (start < mend)) | |
dc009d92 | 645 | return 1; |
dc009d92 | 646 | } |
72414d3f | 647 | |
dc009d92 EB |
648 | return 0; |
649 | } | |
650 | ||
9796fdd8 | 651 | static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order) |
dc009d92 EB |
652 | { |
653 | struct page *pages; | |
72414d3f | 654 | |
dc009d92 EB |
655 | pages = alloc_pages(gfp_mask, order); |
656 | if (pages) { | |
657 | unsigned int count, i; | |
658 | pages->mapping = NULL; | |
4c21e2f2 | 659 | set_page_private(pages, order); |
dc009d92 | 660 | count = 1 << order; |
72414d3f | 661 | for (i = 0; i < count; i++) |
dc009d92 | 662 | SetPageReserved(pages + i); |
dc009d92 | 663 | } |
72414d3f | 664 | |
dc009d92 EB |
665 | return pages; |
666 | } | |
667 | ||
668 | static void kimage_free_pages(struct page *page) | |
669 | { | |
670 | unsigned int order, count, i; | |
72414d3f | 671 | |
4c21e2f2 | 672 | order = page_private(page); |
dc009d92 | 673 | count = 1 << order; |
72414d3f | 674 | for (i = 0; i < count; i++) |
dc009d92 | 675 | ClearPageReserved(page + i); |
dc009d92 EB |
676 | __free_pages(page, order); |
677 | } | |
678 | ||
679 | static void kimage_free_page_list(struct list_head *list) | |
680 | { | |
681 | struct list_head *pos, *next; | |
72414d3f | 682 | |
dc009d92 EB |
683 | list_for_each_safe(pos, next, list) { |
684 | struct page *page; | |
685 | ||
686 | page = list_entry(pos, struct page, lru); | |
687 | list_del(&page->lru); | |
dc009d92 EB |
688 | kimage_free_pages(page); |
689 | } | |
690 | } | |
691 | ||
72414d3f MS |
692 | static struct page *kimage_alloc_normal_control_pages(struct kimage *image, |
693 | unsigned int order) | |
dc009d92 EB |
694 | { |
695 | /* Control pages are special, they are the intermediaries | |
696 | * that are needed while we copy the rest of the pages | |
697 | * to their final resting place. As such they must | |
698 | * not conflict with either the destination addresses | |
699 | * or memory the kernel is already using. | |
700 | * | |
701 | * The only case where we really need more than one of | |
702 | * these are for architectures where we cannot disable | |
703 | * the MMU and must instead generate an identity mapped | |
704 | * page table for all of the memory. | |
705 | * | |
706 | * At worst this runs in O(N) of the image size. | |
707 | */ | |
708 | struct list_head extra_pages; | |
709 | struct page *pages; | |
710 | unsigned int count; | |
711 | ||
712 | count = 1 << order; | |
713 | INIT_LIST_HEAD(&extra_pages); | |
714 | ||
715 | /* Loop while I can allocate a page and the page allocated | |
716 | * is a destination page. | |
717 | */ | |
718 | do { | |
719 | unsigned long pfn, epfn, addr, eaddr; | |
72414d3f | 720 | |
7e01b5ac | 721 | pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order); |
dc009d92 EB |
722 | if (!pages) |
723 | break; | |
724 | pfn = page_to_pfn(pages); | |
725 | epfn = pfn + count; | |
726 | addr = pfn << PAGE_SHIFT; | |
727 | eaddr = epfn << PAGE_SHIFT; | |
728 | if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) || | |
72414d3f | 729 | kimage_is_destination_range(image, addr, eaddr)) { |
dc009d92 EB |
730 | list_add(&pages->lru, &extra_pages); |
731 | pages = NULL; | |
732 | } | |
72414d3f MS |
733 | } while (!pages); |
734 | ||
dc009d92 EB |
735 | if (pages) { |
736 | /* Remember the allocated page... */ | |
737 | list_add(&pages->lru, &image->control_pages); | |
738 | ||
739 | /* Because the page is already in it's destination | |
740 | * location we will never allocate another page at | |
741 | * that address. Therefore kimage_alloc_pages | |
742 | * will not return it (again) and we don't need | |
743 | * to give it an entry in image->segment[]. | |
744 | */ | |
745 | } | |
746 | /* Deal with the destination pages I have inadvertently allocated. | |
747 | * | |
748 | * Ideally I would convert multi-page allocations into single | |
25985edc | 749 | * page allocations, and add everything to image->dest_pages. |
dc009d92 EB |
750 | * |
751 | * For now it is simpler to just free the pages. | |
752 | */ | |
753 | kimage_free_page_list(&extra_pages); | |
dc009d92 | 754 | |
72414d3f | 755 | return pages; |
dc009d92 EB |
756 | } |
757 | ||
72414d3f MS |
758 | static struct page *kimage_alloc_crash_control_pages(struct kimage *image, |
759 | unsigned int order) | |
dc009d92 EB |
760 | { |
761 | /* Control pages are special, they are the intermediaries | |
762 | * that are needed while we copy the rest of the pages | |
763 | * to their final resting place. As such they must | |
764 | * not conflict with either the destination addresses | |
765 | * or memory the kernel is already using. | |
766 | * | |
767 | * Control pages are also the only pags we must allocate | |
768 | * when loading a crash kernel. All of the other pages | |
769 | * are specified by the segments and we just memcpy | |
770 | * into them directly. | |
771 | * | |
772 | * The only case where we really need more than one of | |
773 | * these are for architectures where we cannot disable | |
774 | * the MMU and must instead generate an identity mapped | |
775 | * page table for all of the memory. | |
776 | * | |
777 | * Given the low demand this implements a very simple | |
778 | * allocator that finds the first hole of the appropriate | |
779 | * size in the reserved memory region, and allocates all | |
780 | * of the memory up to and including the hole. | |
781 | */ | |
782 | unsigned long hole_start, hole_end, size; | |
783 | struct page *pages; | |
72414d3f | 784 | |
dc009d92 EB |
785 | pages = NULL; |
786 | size = (1 << order) << PAGE_SHIFT; | |
787 | hole_start = (image->control_page + (size - 1)) & ~(size - 1); | |
788 | hole_end = hole_start + size - 1; | |
72414d3f | 789 | while (hole_end <= crashk_res.end) { |
dc009d92 | 790 | unsigned long i; |
72414d3f | 791 | |
3d214fae | 792 | if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT) |
dc009d92 | 793 | break; |
dc009d92 | 794 | /* See if I overlap any of the segments */ |
72414d3f | 795 | for (i = 0; i < image->nr_segments; i++) { |
dc009d92 | 796 | unsigned long mstart, mend; |
72414d3f | 797 | |
dc009d92 EB |
798 | mstart = image->segment[i].mem; |
799 | mend = mstart + image->segment[i].memsz - 1; | |
800 | if ((hole_end >= mstart) && (hole_start <= mend)) { | |
801 | /* Advance the hole to the end of the segment */ | |
802 | hole_start = (mend + (size - 1)) & ~(size - 1); | |
803 | hole_end = hole_start + size - 1; | |
804 | break; | |
805 | } | |
806 | } | |
807 | /* If I don't overlap any segments I have found my hole! */ | |
808 | if (i == image->nr_segments) { | |
809 | pages = pfn_to_page(hole_start >> PAGE_SHIFT); | |
810 | break; | |
811 | } | |
812 | } | |
72414d3f | 813 | if (pages) |
dc009d92 | 814 | image->control_page = hole_end; |
72414d3f | 815 | |
dc009d92 EB |
816 | return pages; |
817 | } | |
818 | ||
819 | ||
72414d3f MS |
820 | struct page *kimage_alloc_control_pages(struct kimage *image, |
821 | unsigned int order) | |
dc009d92 EB |
822 | { |
823 | struct page *pages = NULL; | |
72414d3f MS |
824 | |
825 | switch (image->type) { | |
dc009d92 EB |
826 | case KEXEC_TYPE_DEFAULT: |
827 | pages = kimage_alloc_normal_control_pages(image, order); | |
828 | break; | |
829 | case KEXEC_TYPE_CRASH: | |
830 | pages = kimage_alloc_crash_control_pages(image, order); | |
831 | break; | |
832 | } | |
72414d3f | 833 | |
dc009d92 EB |
834 | return pages; |
835 | } | |
836 | ||
837 | static int kimage_add_entry(struct kimage *image, kimage_entry_t entry) | |
838 | { | |
72414d3f | 839 | if (*image->entry != 0) |
dc009d92 | 840 | image->entry++; |
72414d3f | 841 | |
dc009d92 EB |
842 | if (image->entry == image->last_entry) { |
843 | kimage_entry_t *ind_page; | |
844 | struct page *page; | |
72414d3f | 845 | |
dc009d92 | 846 | page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST); |
72414d3f | 847 | if (!page) |
dc009d92 | 848 | return -ENOMEM; |
72414d3f | 849 | |
dc009d92 EB |
850 | ind_page = page_address(page); |
851 | *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION; | |
852 | image->entry = ind_page; | |
72414d3f MS |
853 | image->last_entry = ind_page + |
854 | ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1); | |
dc009d92 EB |
855 | } |
856 | *image->entry = entry; | |
857 | image->entry++; | |
858 | *image->entry = 0; | |
72414d3f | 859 | |
dc009d92 EB |
860 | return 0; |
861 | } | |
862 | ||
72414d3f MS |
863 | static int kimage_set_destination(struct kimage *image, |
864 | unsigned long destination) | |
dc009d92 EB |
865 | { |
866 | int result; | |
867 | ||
868 | destination &= PAGE_MASK; | |
869 | result = kimage_add_entry(image, destination | IND_DESTINATION); | |
72414d3f | 870 | |
dc009d92 EB |
871 | return result; |
872 | } | |
873 | ||
874 | ||
875 | static int kimage_add_page(struct kimage *image, unsigned long page) | |
876 | { | |
877 | int result; | |
878 | ||
879 | page &= PAGE_MASK; | |
880 | result = kimage_add_entry(image, page | IND_SOURCE); | |
72414d3f | 881 | |
dc009d92 EB |
882 | return result; |
883 | } | |
884 | ||
885 | ||
886 | static void kimage_free_extra_pages(struct kimage *image) | |
887 | { | |
888 | /* Walk through and free any extra destination pages I may have */ | |
889 | kimage_free_page_list(&image->dest_pages); | |
890 | ||
25985edc | 891 | /* Walk through and free any unusable pages I have cached */ |
7d3e2bca | 892 | kimage_free_page_list(&image->unusable_pages); |
dc009d92 EB |
893 | |
894 | } | |
7fccf032 | 895 | static void kimage_terminate(struct kimage *image) |
dc009d92 | 896 | { |
72414d3f | 897 | if (*image->entry != 0) |
dc009d92 | 898 | image->entry++; |
72414d3f | 899 | |
dc009d92 | 900 | *image->entry = IND_DONE; |
dc009d92 EB |
901 | } |
902 | ||
903 | #define for_each_kimage_entry(image, ptr, entry) \ | |
904 | for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \ | |
e1bebcf4 FF |
905 | ptr = (entry & IND_INDIRECTION) ? \ |
906 | phys_to_virt((entry & PAGE_MASK)) : ptr + 1) | |
dc009d92 EB |
907 | |
908 | static void kimage_free_entry(kimage_entry_t entry) | |
909 | { | |
910 | struct page *page; | |
911 | ||
912 | page = pfn_to_page(entry >> PAGE_SHIFT); | |
913 | kimage_free_pages(page); | |
914 | } | |
915 | ||
916 | static void kimage_free(struct kimage *image) | |
917 | { | |
918 | kimage_entry_t *ptr, entry; | |
919 | kimage_entry_t ind = 0; | |
920 | ||
921 | if (!image) | |
922 | return; | |
72414d3f | 923 | |
dc009d92 EB |
924 | kimage_free_extra_pages(image); |
925 | for_each_kimage_entry(image, ptr, entry) { | |
926 | if (entry & IND_INDIRECTION) { | |
927 | /* Free the previous indirection page */ | |
72414d3f | 928 | if (ind & IND_INDIRECTION) |
dc009d92 | 929 | kimage_free_entry(ind); |
dc009d92 EB |
930 | /* Save this indirection page until we are |
931 | * done with it. | |
932 | */ | |
933 | ind = entry; | |
e1bebcf4 | 934 | } else if (entry & IND_SOURCE) |
dc009d92 | 935 | kimage_free_entry(entry); |
dc009d92 EB |
936 | } |
937 | /* Free the final indirection page */ | |
72414d3f | 938 | if (ind & IND_INDIRECTION) |
dc009d92 | 939 | kimage_free_entry(ind); |
dc009d92 EB |
940 | |
941 | /* Handle any machine specific cleanup */ | |
942 | machine_kexec_cleanup(image); | |
943 | ||
944 | /* Free the kexec control pages... */ | |
945 | kimage_free_page_list(&image->control_pages); | |
cb105258 | 946 | |
cb105258 VG |
947 | /* |
948 | * Free up any temporary buffers allocated. This might hit if | |
949 | * error occurred much later after buffer allocation. | |
950 | */ | |
951 | if (image->file_mode) | |
952 | kimage_file_post_load_cleanup(image); | |
953 | ||
dc009d92 EB |
954 | kfree(image); |
955 | } | |
956 | ||
72414d3f MS |
957 | static kimage_entry_t *kimage_dst_used(struct kimage *image, |
958 | unsigned long page) | |
dc009d92 EB |
959 | { |
960 | kimage_entry_t *ptr, entry; | |
961 | unsigned long destination = 0; | |
962 | ||
963 | for_each_kimage_entry(image, ptr, entry) { | |
72414d3f | 964 | if (entry & IND_DESTINATION) |
dc009d92 | 965 | destination = entry & PAGE_MASK; |
dc009d92 | 966 | else if (entry & IND_SOURCE) { |
72414d3f | 967 | if (page == destination) |
dc009d92 | 968 | return ptr; |
dc009d92 EB |
969 | destination += PAGE_SIZE; |
970 | } | |
971 | } | |
72414d3f | 972 | |
314b6a4d | 973 | return NULL; |
dc009d92 EB |
974 | } |
975 | ||
72414d3f | 976 | static struct page *kimage_alloc_page(struct kimage *image, |
9796fdd8 | 977 | gfp_t gfp_mask, |
72414d3f | 978 | unsigned long destination) |
dc009d92 EB |
979 | { |
980 | /* | |
981 | * Here we implement safeguards to ensure that a source page | |
982 | * is not copied to its destination page before the data on | |
983 | * the destination page is no longer useful. | |
984 | * | |
985 | * To do this we maintain the invariant that a source page is | |
986 | * either its own destination page, or it is not a | |
987 | * destination page at all. | |
988 | * | |
989 | * That is slightly stronger than required, but the proof | |
990 | * that no problems will not occur is trivial, and the | |
991 | * implementation is simply to verify. | |
992 | * | |
993 | * When allocating all pages normally this algorithm will run | |
994 | * in O(N) time, but in the worst case it will run in O(N^2) | |
995 | * time. If the runtime is a problem the data structures can | |
996 | * be fixed. | |
997 | */ | |
998 | struct page *page; | |
999 | unsigned long addr; | |
1000 | ||
1001 | /* | |
1002 | * Walk through the list of destination pages, and see if I | |
1003 | * have a match. | |
1004 | */ | |
1005 | list_for_each_entry(page, &image->dest_pages, lru) { | |
1006 | addr = page_to_pfn(page) << PAGE_SHIFT; | |
1007 | if (addr == destination) { | |
1008 | list_del(&page->lru); | |
1009 | return page; | |
1010 | } | |
1011 | } | |
1012 | page = NULL; | |
1013 | while (1) { | |
1014 | kimage_entry_t *old; | |
1015 | ||
1016 | /* Allocate a page, if we run out of memory give up */ | |
1017 | page = kimage_alloc_pages(gfp_mask, 0); | |
72414d3f | 1018 | if (!page) |
314b6a4d | 1019 | return NULL; |
dc009d92 | 1020 | /* If the page cannot be used file it away */ |
72414d3f MS |
1021 | if (page_to_pfn(page) > |
1022 | (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) { | |
7d3e2bca | 1023 | list_add(&page->lru, &image->unusable_pages); |
dc009d92 EB |
1024 | continue; |
1025 | } | |
1026 | addr = page_to_pfn(page) << PAGE_SHIFT; | |
1027 | ||
1028 | /* If it is the destination page we want use it */ | |
1029 | if (addr == destination) | |
1030 | break; | |
1031 | ||
1032 | /* If the page is not a destination page use it */ | |
72414d3f MS |
1033 | if (!kimage_is_destination_range(image, addr, |
1034 | addr + PAGE_SIZE)) | |
dc009d92 EB |
1035 | break; |
1036 | ||
1037 | /* | |
1038 | * I know that the page is someones destination page. | |
1039 | * See if there is already a source page for this | |
1040 | * destination page. And if so swap the source pages. | |
1041 | */ | |
1042 | old = kimage_dst_used(image, addr); | |
1043 | if (old) { | |
1044 | /* If so move it */ | |
1045 | unsigned long old_addr; | |
1046 | struct page *old_page; | |
1047 | ||
1048 | old_addr = *old & PAGE_MASK; | |
1049 | old_page = pfn_to_page(old_addr >> PAGE_SHIFT); | |
1050 | copy_highpage(page, old_page); | |
1051 | *old = addr | (*old & ~PAGE_MASK); | |
1052 | ||
1053 | /* The old page I have found cannot be a | |
f9092f35 JS |
1054 | * destination page, so return it if it's |
1055 | * gfp_flags honor the ones passed in. | |
dc009d92 | 1056 | */ |
f9092f35 JS |
1057 | if (!(gfp_mask & __GFP_HIGHMEM) && |
1058 | PageHighMem(old_page)) { | |
1059 | kimage_free_pages(old_page); | |
1060 | continue; | |
1061 | } | |
dc009d92 EB |
1062 | addr = old_addr; |
1063 | page = old_page; | |
1064 | break; | |
e1bebcf4 | 1065 | } else { |
dc009d92 EB |
1066 | /* Place the page on the destination list I |
1067 | * will use it later. | |
1068 | */ | |
1069 | list_add(&page->lru, &image->dest_pages); | |
1070 | } | |
1071 | } | |
72414d3f | 1072 | |
dc009d92 EB |
1073 | return page; |
1074 | } | |
1075 | ||
1076 | static int kimage_load_normal_segment(struct kimage *image, | |
72414d3f | 1077 | struct kexec_segment *segment) |
dc009d92 EB |
1078 | { |
1079 | unsigned long maddr; | |
310faaa9 | 1080 | size_t ubytes, mbytes; |
dc009d92 | 1081 | int result; |
cb105258 VG |
1082 | unsigned char __user *buf = NULL; |
1083 | unsigned char *kbuf = NULL; | |
dc009d92 EB |
1084 | |
1085 | result = 0; | |
cb105258 VG |
1086 | if (image->file_mode) |
1087 | kbuf = segment->kbuf; | |
1088 | else | |
1089 | buf = segment->buf; | |
dc009d92 EB |
1090 | ubytes = segment->bufsz; |
1091 | mbytes = segment->memsz; | |
1092 | maddr = segment->mem; | |
1093 | ||
1094 | result = kimage_set_destination(image, maddr); | |
72414d3f | 1095 | if (result < 0) |
dc009d92 | 1096 | goto out; |
72414d3f MS |
1097 | |
1098 | while (mbytes) { | |
dc009d92 EB |
1099 | struct page *page; |
1100 | char *ptr; | |
1101 | size_t uchunk, mchunk; | |
72414d3f | 1102 | |
dc009d92 | 1103 | page = kimage_alloc_page(image, GFP_HIGHUSER, maddr); |
c80544dc | 1104 | if (!page) { |
dc009d92 EB |
1105 | result = -ENOMEM; |
1106 | goto out; | |
1107 | } | |
72414d3f MS |
1108 | result = kimage_add_page(image, page_to_pfn(page) |
1109 | << PAGE_SHIFT); | |
1110 | if (result < 0) | |
dc009d92 | 1111 | goto out; |
72414d3f | 1112 | |
dc009d92 EB |
1113 | ptr = kmap(page); |
1114 | /* Start with a clear page */ | |
3ecb01df | 1115 | clear_page(ptr); |
dc009d92 | 1116 | ptr += maddr & ~PAGE_MASK; |
31c3a3fe ZY |
1117 | mchunk = min_t(size_t, mbytes, |
1118 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
1119 | uchunk = min(ubytes, mchunk); | |
72414d3f | 1120 | |
cb105258 VG |
1121 | /* For file based kexec, source pages are in kernel memory */ |
1122 | if (image->file_mode) | |
1123 | memcpy(ptr, kbuf, uchunk); | |
1124 | else | |
1125 | result = copy_from_user(ptr, buf, uchunk); | |
dc009d92 EB |
1126 | kunmap(page); |
1127 | if (result) { | |
f65a03f6 | 1128 | result = -EFAULT; |
dc009d92 EB |
1129 | goto out; |
1130 | } | |
1131 | ubytes -= uchunk; | |
1132 | maddr += mchunk; | |
cb105258 VG |
1133 | if (image->file_mode) |
1134 | kbuf += mchunk; | |
1135 | else | |
1136 | buf += mchunk; | |
dc009d92 EB |
1137 | mbytes -= mchunk; |
1138 | } | |
72414d3f | 1139 | out: |
dc009d92 EB |
1140 | return result; |
1141 | } | |
1142 | ||
1143 | static int kimage_load_crash_segment(struct kimage *image, | |
72414d3f | 1144 | struct kexec_segment *segment) |
dc009d92 EB |
1145 | { |
1146 | /* For crash dumps kernels we simply copy the data from | |
1147 | * user space to it's destination. | |
1148 | * We do things a page at a time for the sake of kmap. | |
1149 | */ | |
1150 | unsigned long maddr; | |
310faaa9 | 1151 | size_t ubytes, mbytes; |
dc009d92 | 1152 | int result; |
dd5f7260 VG |
1153 | unsigned char __user *buf = NULL; |
1154 | unsigned char *kbuf = NULL; | |
dc009d92 EB |
1155 | |
1156 | result = 0; | |
dd5f7260 VG |
1157 | if (image->file_mode) |
1158 | kbuf = segment->kbuf; | |
1159 | else | |
1160 | buf = segment->buf; | |
dc009d92 EB |
1161 | ubytes = segment->bufsz; |
1162 | mbytes = segment->memsz; | |
1163 | maddr = segment->mem; | |
72414d3f | 1164 | while (mbytes) { |
dc009d92 EB |
1165 | struct page *page; |
1166 | char *ptr; | |
1167 | size_t uchunk, mchunk; | |
72414d3f | 1168 | |
dc009d92 | 1169 | page = pfn_to_page(maddr >> PAGE_SHIFT); |
c80544dc | 1170 | if (!page) { |
dc009d92 EB |
1171 | result = -ENOMEM; |
1172 | goto out; | |
1173 | } | |
1174 | ptr = kmap(page); | |
1175 | ptr += maddr & ~PAGE_MASK; | |
31c3a3fe ZY |
1176 | mchunk = min_t(size_t, mbytes, |
1177 | PAGE_SIZE - (maddr & ~PAGE_MASK)); | |
1178 | uchunk = min(ubytes, mchunk); | |
1179 | if (mchunk > uchunk) { | |
dc009d92 EB |
1180 | /* Zero the trailing part of the page */ |
1181 | memset(ptr + uchunk, 0, mchunk - uchunk); | |
1182 | } | |
dd5f7260 VG |
1183 | |
1184 | /* For file based kexec, source pages are in kernel memory */ | |
1185 | if (image->file_mode) | |
1186 | memcpy(ptr, kbuf, uchunk); | |
1187 | else | |
1188 | result = copy_from_user(ptr, buf, uchunk); | |
a7956113 | 1189 | kexec_flush_icache_page(page); |
dc009d92 EB |
1190 | kunmap(page); |
1191 | if (result) { | |
f65a03f6 | 1192 | result = -EFAULT; |
dc009d92 EB |
1193 | goto out; |
1194 | } | |
1195 | ubytes -= uchunk; | |
1196 | maddr += mchunk; | |
dd5f7260 VG |
1197 | if (image->file_mode) |
1198 | kbuf += mchunk; | |
1199 | else | |
1200 | buf += mchunk; | |
dc009d92 EB |
1201 | mbytes -= mchunk; |
1202 | } | |
72414d3f | 1203 | out: |
dc009d92 EB |
1204 | return result; |
1205 | } | |
1206 | ||
1207 | static int kimage_load_segment(struct kimage *image, | |
72414d3f | 1208 | struct kexec_segment *segment) |
dc009d92 EB |
1209 | { |
1210 | int result = -ENOMEM; | |
72414d3f MS |
1211 | |
1212 | switch (image->type) { | |
dc009d92 EB |
1213 | case KEXEC_TYPE_DEFAULT: |
1214 | result = kimage_load_normal_segment(image, segment); | |
1215 | break; | |
1216 | case KEXEC_TYPE_CRASH: | |
1217 | result = kimage_load_crash_segment(image, segment); | |
1218 | break; | |
1219 | } | |
72414d3f | 1220 | |
dc009d92 EB |
1221 | return result; |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * Exec Kernel system call: for obvious reasons only root may call it. | |
1226 | * | |
1227 | * This call breaks up into three pieces. | |
1228 | * - A generic part which loads the new kernel from the current | |
1229 | * address space, and very carefully places the data in the | |
1230 | * allocated pages. | |
1231 | * | |
1232 | * - A generic part that interacts with the kernel and tells all of | |
1233 | * the devices to shut down. Preventing on-going dmas, and placing | |
1234 | * the devices in a consistent state so a later kernel can | |
1235 | * reinitialize them. | |
1236 | * | |
1237 | * - A machine specific part that includes the syscall number | |
002ace78 | 1238 | * and then copies the image to it's final destination. And |
dc009d92 EB |
1239 | * jumps into the image at entry. |
1240 | * | |
1241 | * kexec does not sync, or unmount filesystems so if you need | |
1242 | * that to happen you need to do that yourself. | |
1243 | */ | |
c330dda9 JM |
1244 | struct kimage *kexec_image; |
1245 | struct kimage *kexec_crash_image; | |
7984754b | 1246 | int kexec_load_disabled; |
8c5a1cf0 AM |
1247 | |
1248 | static DEFINE_MUTEX(kexec_mutex); | |
dc009d92 | 1249 | |
754fe8d2 HC |
1250 | SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments, |
1251 | struct kexec_segment __user *, segments, unsigned long, flags) | |
dc009d92 EB |
1252 | { |
1253 | struct kimage **dest_image, *image; | |
dc009d92 EB |
1254 | int result; |
1255 | ||
1256 | /* We only trust the superuser with rebooting the system. */ | |
7984754b | 1257 | if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) |
dc009d92 EB |
1258 | return -EPERM; |
1259 | ||
1260 | /* | |
1261 | * Verify we have a legal set of flags | |
1262 | * This leaves us room for future extensions. | |
1263 | */ | |
1264 | if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK)) | |
1265 | return -EINVAL; | |
1266 | ||
1267 | /* Verify we are on the appropriate architecture */ | |
1268 | if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) && | |
1269 | ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT)) | |
dc009d92 | 1270 | return -EINVAL; |
dc009d92 EB |
1271 | |
1272 | /* Put an artificial cap on the number | |
1273 | * of segments passed to kexec_load. | |
1274 | */ | |
1275 | if (nr_segments > KEXEC_SEGMENT_MAX) | |
1276 | return -EINVAL; | |
1277 | ||
1278 | image = NULL; | |
1279 | result = 0; | |
1280 | ||
1281 | /* Because we write directly to the reserved memory | |
1282 | * region when loading crash kernels we need a mutex here to | |
1283 | * prevent multiple crash kernels from attempting to load | |
1284 | * simultaneously, and to prevent a crash kernel from loading | |
1285 | * over the top of a in use crash kernel. | |
1286 | * | |
1287 | * KISS: always take the mutex. | |
1288 | */ | |
8c5a1cf0 | 1289 | if (!mutex_trylock(&kexec_mutex)) |
dc009d92 | 1290 | return -EBUSY; |
72414d3f | 1291 | |
dc009d92 | 1292 | dest_image = &kexec_image; |
72414d3f | 1293 | if (flags & KEXEC_ON_CRASH) |
dc009d92 | 1294 | dest_image = &kexec_crash_image; |
dc009d92 EB |
1295 | if (nr_segments > 0) { |
1296 | unsigned long i; | |
72414d3f | 1297 | |
518a0c71 GL |
1298 | if (flags & KEXEC_ON_CRASH) { |
1299 | /* | |
1300 | * Loading another kernel to switch to if this one | |
1301 | * crashes. Free any current crash dump kernel before | |
dc009d92 EB |
1302 | * we corrupt it. |
1303 | */ | |
518a0c71 | 1304 | |
dc009d92 | 1305 | kimage_free(xchg(&kexec_crash_image, NULL)); |
255aedd9 VG |
1306 | result = kimage_alloc_init(&image, entry, nr_segments, |
1307 | segments, flags); | |
558df720 | 1308 | crash_map_reserved_pages(); |
518a0c71 GL |
1309 | } else { |
1310 | /* Loading another kernel to reboot into. */ | |
1311 | ||
1312 | result = kimage_alloc_init(&image, entry, nr_segments, | |
1313 | segments, flags); | |
dc009d92 | 1314 | } |
72414d3f | 1315 | if (result) |
dc009d92 | 1316 | goto out; |
72414d3f | 1317 | |
3ab83521 HY |
1318 | if (flags & KEXEC_PRESERVE_CONTEXT) |
1319 | image->preserve_context = 1; | |
dc009d92 | 1320 | result = machine_kexec_prepare(image); |
72414d3f | 1321 | if (result) |
dc009d92 | 1322 | goto out; |
72414d3f MS |
1323 | |
1324 | for (i = 0; i < nr_segments; i++) { | |
dc009d92 | 1325 | result = kimage_load_segment(image, &image->segment[i]); |
72414d3f | 1326 | if (result) |
dc009d92 | 1327 | goto out; |
dc009d92 | 1328 | } |
7fccf032 | 1329 | kimage_terminate(image); |
558df720 MH |
1330 | if (flags & KEXEC_ON_CRASH) |
1331 | crash_unmap_reserved_pages(); | |
dc009d92 EB |
1332 | } |
1333 | /* Install the new kernel, and Uninstall the old */ | |
1334 | image = xchg(dest_image, image); | |
1335 | ||
72414d3f | 1336 | out: |
8c5a1cf0 | 1337 | mutex_unlock(&kexec_mutex); |
dc009d92 | 1338 | kimage_free(image); |
72414d3f | 1339 | |
dc009d92 EB |
1340 | return result; |
1341 | } | |
1342 | ||
558df720 MH |
1343 | /* |
1344 | * Add and remove page tables for crashkernel memory | |
1345 | * | |
1346 | * Provide an empty default implementation here -- architecture | |
1347 | * code may override this | |
1348 | */ | |
1349 | void __weak crash_map_reserved_pages(void) | |
1350 | {} | |
1351 | ||
1352 | void __weak crash_unmap_reserved_pages(void) | |
1353 | {} | |
1354 | ||
dc009d92 | 1355 | #ifdef CONFIG_COMPAT |
ca2c405a HC |
1356 | COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry, |
1357 | compat_ulong_t, nr_segments, | |
1358 | struct compat_kexec_segment __user *, segments, | |
1359 | compat_ulong_t, flags) | |
dc009d92 EB |
1360 | { |
1361 | struct compat_kexec_segment in; | |
1362 | struct kexec_segment out, __user *ksegments; | |
1363 | unsigned long i, result; | |
1364 | ||
1365 | /* Don't allow clients that don't understand the native | |
1366 | * architecture to do anything. | |
1367 | */ | |
72414d3f | 1368 | if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT) |
dc009d92 | 1369 | return -EINVAL; |
dc009d92 | 1370 | |
72414d3f | 1371 | if (nr_segments > KEXEC_SEGMENT_MAX) |
dc009d92 | 1372 | return -EINVAL; |
dc009d92 EB |
1373 | |
1374 | ksegments = compat_alloc_user_space(nr_segments * sizeof(out)); | |
e1bebcf4 | 1375 | for (i = 0; i < nr_segments; i++) { |
dc009d92 | 1376 | result = copy_from_user(&in, &segments[i], sizeof(in)); |
72414d3f | 1377 | if (result) |
dc009d92 | 1378 | return -EFAULT; |
dc009d92 EB |
1379 | |
1380 | out.buf = compat_ptr(in.buf); | |
1381 | out.bufsz = in.bufsz; | |
1382 | out.mem = in.mem; | |
1383 | out.memsz = in.memsz; | |
1384 | ||
1385 | result = copy_to_user(&ksegments[i], &out, sizeof(out)); | |
72414d3f | 1386 | if (result) |
dc009d92 | 1387 | return -EFAULT; |
dc009d92 EB |
1388 | } |
1389 | ||
1390 | return sys_kexec_load(entry, nr_segments, ksegments, flags); | |
1391 | } | |
1392 | #endif | |
1393 | ||
74ca317c | 1394 | #ifdef CONFIG_KEXEC_FILE |
f0895685 VG |
1395 | SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd, |
1396 | unsigned long, cmdline_len, const char __user *, cmdline_ptr, | |
1397 | unsigned long, flags) | |
1398 | { | |
cb105258 VG |
1399 | int ret = 0, i; |
1400 | struct kimage **dest_image, *image; | |
1401 | ||
1402 | /* We only trust the superuser with rebooting the system. */ | |
1403 | if (!capable(CAP_SYS_BOOT) || kexec_load_disabled) | |
1404 | return -EPERM; | |
1405 | ||
1406 | /* Make sure we have a legal set of flags */ | |
1407 | if (flags != (flags & KEXEC_FILE_FLAGS)) | |
1408 | return -EINVAL; | |
1409 | ||
1410 | image = NULL; | |
1411 | ||
1412 | if (!mutex_trylock(&kexec_mutex)) | |
1413 | return -EBUSY; | |
1414 | ||
1415 | dest_image = &kexec_image; | |
1416 | if (flags & KEXEC_FILE_ON_CRASH) | |
1417 | dest_image = &kexec_crash_image; | |
1418 | ||
1419 | if (flags & KEXEC_FILE_UNLOAD) | |
1420 | goto exchange; | |
1421 | ||
1422 | /* | |
1423 | * In case of crash, new kernel gets loaded in reserved region. It is | |
1424 | * same memory where old crash kernel might be loaded. Free any | |
1425 | * current crash dump kernel before we corrupt it. | |
1426 | */ | |
1427 | if (flags & KEXEC_FILE_ON_CRASH) | |
1428 | kimage_free(xchg(&kexec_crash_image, NULL)); | |
1429 | ||
1430 | ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr, | |
1431 | cmdline_len, flags); | |
1432 | if (ret) | |
1433 | goto out; | |
1434 | ||
1435 | ret = machine_kexec_prepare(image); | |
1436 | if (ret) | |
1437 | goto out; | |
1438 | ||
12db5562 VG |
1439 | ret = kexec_calculate_store_digests(image); |
1440 | if (ret) | |
1441 | goto out; | |
1442 | ||
cb105258 VG |
1443 | for (i = 0; i < image->nr_segments; i++) { |
1444 | struct kexec_segment *ksegment; | |
1445 | ||
1446 | ksegment = &image->segment[i]; | |
1447 | pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n", | |
1448 | i, ksegment->buf, ksegment->bufsz, ksegment->mem, | |
1449 | ksegment->memsz); | |
1450 | ||
1451 | ret = kimage_load_segment(image, &image->segment[i]); | |
1452 | if (ret) | |
1453 | goto out; | |
1454 | } | |
1455 | ||
1456 | kimage_terminate(image); | |
1457 | ||
1458 | /* | |
1459 | * Free up any temporary buffers allocated which are not needed | |
1460 | * after image has been loaded | |
1461 | */ | |
1462 | kimage_file_post_load_cleanup(image); | |
1463 | exchange: | |
1464 | image = xchg(dest_image, image); | |
1465 | out: | |
1466 | mutex_unlock(&kexec_mutex); | |
1467 | kimage_free(image); | |
1468 | return ret; | |
f0895685 VG |
1469 | } |
1470 | ||
74ca317c VG |
1471 | #endif /* CONFIG_KEXEC_FILE */ |
1472 | ||
6e274d14 | 1473 | void crash_kexec(struct pt_regs *regs) |
dc009d92 | 1474 | { |
8c5a1cf0 | 1475 | /* Take the kexec_mutex here to prevent sys_kexec_load |
dc009d92 EB |
1476 | * running on one cpu from replacing the crash kernel |
1477 | * we are using after a panic on a different cpu. | |
1478 | * | |
1479 | * If the crash kernel was not located in a fixed area | |
1480 | * of memory the xchg(&kexec_crash_image) would be | |
1481 | * sufficient. But since I reuse the memory... | |
1482 | */ | |
8c5a1cf0 | 1483 | if (mutex_trylock(&kexec_mutex)) { |
c0ce7d08 | 1484 | if (kexec_crash_image) { |
e996e581 | 1485 | struct pt_regs fixed_regs; |
0f4bd46e | 1486 | |
e996e581 | 1487 | crash_setup_regs(&fixed_regs, regs); |
fd59d231 | 1488 | crash_save_vmcoreinfo(); |
e996e581 | 1489 | machine_crash_shutdown(&fixed_regs); |
c0ce7d08 | 1490 | machine_kexec(kexec_crash_image); |
dc009d92 | 1491 | } |
8c5a1cf0 | 1492 | mutex_unlock(&kexec_mutex); |
dc009d92 EB |
1493 | } |
1494 | } | |
cc571658 | 1495 | |
06a7f711 AW |
1496 | size_t crash_get_memory_size(void) |
1497 | { | |
e05bd336 | 1498 | size_t size = 0; |
06a7f711 | 1499 | mutex_lock(&kexec_mutex); |
e05bd336 | 1500 | if (crashk_res.end != crashk_res.start) |
28f65c11 | 1501 | size = resource_size(&crashk_res); |
06a7f711 AW |
1502 | mutex_unlock(&kexec_mutex); |
1503 | return size; | |
1504 | } | |
1505 | ||
c0bb9e45 AB |
1506 | void __weak crash_free_reserved_phys_range(unsigned long begin, |
1507 | unsigned long end) | |
06a7f711 AW |
1508 | { |
1509 | unsigned long addr; | |
1510 | ||
e07cee23 JL |
1511 | for (addr = begin; addr < end; addr += PAGE_SIZE) |
1512 | free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT)); | |
06a7f711 AW |
1513 | } |
1514 | ||
1515 | int crash_shrink_memory(unsigned long new_size) | |
1516 | { | |
1517 | int ret = 0; | |
1518 | unsigned long start, end; | |
bec013c4 | 1519 | unsigned long old_size; |
6480e5a0 | 1520 | struct resource *ram_res; |
06a7f711 AW |
1521 | |
1522 | mutex_lock(&kexec_mutex); | |
1523 | ||
1524 | if (kexec_crash_image) { | |
1525 | ret = -ENOENT; | |
1526 | goto unlock; | |
1527 | } | |
1528 | start = crashk_res.start; | |
1529 | end = crashk_res.end; | |
bec013c4 MH |
1530 | old_size = (end == 0) ? 0 : end - start + 1; |
1531 | if (new_size >= old_size) { | |
1532 | ret = (new_size == old_size) ? 0 : -EINVAL; | |
06a7f711 AW |
1533 | goto unlock; |
1534 | } | |
1535 | ||
6480e5a0 MH |
1536 | ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL); |
1537 | if (!ram_res) { | |
1538 | ret = -ENOMEM; | |
1539 | goto unlock; | |
1540 | } | |
1541 | ||
558df720 MH |
1542 | start = roundup(start, KEXEC_CRASH_MEM_ALIGN); |
1543 | end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN); | |
06a7f711 | 1544 | |
558df720 | 1545 | crash_map_reserved_pages(); |
c0bb9e45 | 1546 | crash_free_reserved_phys_range(end, crashk_res.end); |
06a7f711 | 1547 | |
e05bd336 | 1548 | if ((start == end) && (crashk_res.parent != NULL)) |
06a7f711 | 1549 | release_resource(&crashk_res); |
6480e5a0 MH |
1550 | |
1551 | ram_res->start = end; | |
1552 | ram_res->end = crashk_res.end; | |
1553 | ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM; | |
1554 | ram_res->name = "System RAM"; | |
1555 | ||
475f9aa6 | 1556 | crashk_res.end = end - 1; |
6480e5a0 MH |
1557 | |
1558 | insert_resource(&iomem_resource, ram_res); | |
558df720 | 1559 | crash_unmap_reserved_pages(); |
06a7f711 AW |
1560 | |
1561 | unlock: | |
1562 | mutex_unlock(&kexec_mutex); | |
1563 | return ret; | |
1564 | } | |
1565 | ||
85916f81 MD |
1566 | static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data, |
1567 | size_t data_len) | |
1568 | { | |
1569 | struct elf_note note; | |
1570 | ||
1571 | note.n_namesz = strlen(name) + 1; | |
1572 | note.n_descsz = data_len; | |
1573 | note.n_type = type; | |
1574 | memcpy(buf, ¬e, sizeof(note)); | |
1575 | buf += (sizeof(note) + 3)/4; | |
1576 | memcpy(buf, name, note.n_namesz); | |
1577 | buf += (note.n_namesz + 3)/4; | |
1578 | memcpy(buf, data, note.n_descsz); | |
1579 | buf += (note.n_descsz + 3)/4; | |
1580 | ||
1581 | return buf; | |
1582 | } | |
1583 | ||
1584 | static void final_note(u32 *buf) | |
1585 | { | |
1586 | struct elf_note note; | |
1587 | ||
1588 | note.n_namesz = 0; | |
1589 | note.n_descsz = 0; | |
1590 | note.n_type = 0; | |
1591 | memcpy(buf, ¬e, sizeof(note)); | |
1592 | } | |
1593 | ||
1594 | void crash_save_cpu(struct pt_regs *regs, int cpu) | |
1595 | { | |
1596 | struct elf_prstatus prstatus; | |
1597 | u32 *buf; | |
1598 | ||
4f4b6c1a | 1599 | if ((cpu < 0) || (cpu >= nr_cpu_ids)) |
85916f81 MD |
1600 | return; |
1601 | ||
1602 | /* Using ELF notes here is opportunistic. | |
1603 | * I need a well defined structure format | |
1604 | * for the data I pass, and I need tags | |
1605 | * on the data to indicate what information I have | |
1606 | * squirrelled away. ELF notes happen to provide | |
1607 | * all of that, so there is no need to invent something new. | |
1608 | */ | |
e1bebcf4 | 1609 | buf = (u32 *)per_cpu_ptr(crash_notes, cpu); |
85916f81 MD |
1610 | if (!buf) |
1611 | return; | |
1612 | memset(&prstatus, 0, sizeof(prstatus)); | |
1613 | prstatus.pr_pid = current->pid; | |
6cd61c0b | 1614 | elf_core_copy_kernel_regs(&prstatus.pr_reg, regs); |
6672f76a | 1615 | buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, |
e1bebcf4 | 1616 | &prstatus, sizeof(prstatus)); |
85916f81 MD |
1617 | final_note(buf); |
1618 | } | |
1619 | ||
cc571658 VG |
1620 | static int __init crash_notes_memory_init(void) |
1621 | { | |
1622 | /* Allocate memory for saving cpu registers. */ | |
1623 | crash_notes = alloc_percpu(note_buf_t); | |
1624 | if (!crash_notes) { | |
e1bebcf4 | 1625 | pr_warn("Kexec: Memory allocation for saving cpu register states failed\n"); |
cc571658 VG |
1626 | return -ENOMEM; |
1627 | } | |
1628 | return 0; | |
1629 | } | |
c96d6660 | 1630 | subsys_initcall(crash_notes_memory_init); |
fd59d231 | 1631 | |
cba63c30 BW |
1632 | |
1633 | /* | |
1634 | * parsing the "crashkernel" commandline | |
1635 | * | |
1636 | * this code is intended to be called from architecture specific code | |
1637 | */ | |
1638 | ||
1639 | ||
1640 | /* | |
1641 | * This function parses command lines in the format | |
1642 | * | |
1643 | * crashkernel=ramsize-range:size[,...][@offset] | |
1644 | * | |
1645 | * The function returns 0 on success and -EINVAL on failure. | |
1646 | */ | |
e1bebcf4 FF |
1647 | static int __init parse_crashkernel_mem(char *cmdline, |
1648 | unsigned long long system_ram, | |
1649 | unsigned long long *crash_size, | |
1650 | unsigned long long *crash_base) | |
cba63c30 BW |
1651 | { |
1652 | char *cur = cmdline, *tmp; | |
1653 | ||
1654 | /* for each entry of the comma-separated list */ | |
1655 | do { | |
1656 | unsigned long long start, end = ULLONG_MAX, size; | |
1657 | ||
1658 | /* get the start of the range */ | |
1659 | start = memparse(cur, &tmp); | |
1660 | if (cur == tmp) { | |
e1bebcf4 | 1661 | pr_warn("crashkernel: Memory value expected\n"); |
cba63c30 BW |
1662 | return -EINVAL; |
1663 | } | |
1664 | cur = tmp; | |
1665 | if (*cur != '-') { | |
e1bebcf4 | 1666 | pr_warn("crashkernel: '-' expected\n"); |
cba63c30 BW |
1667 | return -EINVAL; |
1668 | } | |
1669 | cur++; | |
1670 | ||
1671 | /* if no ':' is here, than we read the end */ | |
1672 | if (*cur != ':') { | |
1673 | end = memparse(cur, &tmp); | |
1674 | if (cur == tmp) { | |
e1bebcf4 | 1675 | pr_warn("crashkernel: Memory value expected\n"); |
cba63c30 BW |
1676 | return -EINVAL; |
1677 | } | |
1678 | cur = tmp; | |
1679 | if (end <= start) { | |
e1bebcf4 | 1680 | pr_warn("crashkernel: end <= start\n"); |
cba63c30 BW |
1681 | return -EINVAL; |
1682 | } | |
1683 | } | |
1684 | ||
1685 | if (*cur != ':') { | |
e1bebcf4 | 1686 | pr_warn("crashkernel: ':' expected\n"); |
cba63c30 BW |
1687 | return -EINVAL; |
1688 | } | |
1689 | cur++; | |
1690 | ||
1691 | size = memparse(cur, &tmp); | |
1692 | if (cur == tmp) { | |
e1bebcf4 | 1693 | pr_warn("Memory value expected\n"); |
cba63c30 BW |
1694 | return -EINVAL; |
1695 | } | |
1696 | cur = tmp; | |
1697 | if (size >= system_ram) { | |
e1bebcf4 | 1698 | pr_warn("crashkernel: invalid size\n"); |
cba63c30 BW |
1699 | return -EINVAL; |
1700 | } | |
1701 | ||
1702 | /* match ? */ | |
be089d79 | 1703 | if (system_ram >= start && system_ram < end) { |
cba63c30 BW |
1704 | *crash_size = size; |
1705 | break; | |
1706 | } | |
1707 | } while (*cur++ == ','); | |
1708 | ||
1709 | if (*crash_size > 0) { | |
11c7da4b | 1710 | while (*cur && *cur != ' ' && *cur != '@') |
cba63c30 BW |
1711 | cur++; |
1712 | if (*cur == '@') { | |
1713 | cur++; | |
1714 | *crash_base = memparse(cur, &tmp); | |
1715 | if (cur == tmp) { | |
e1bebcf4 | 1716 | pr_warn("Memory value expected after '@'\n"); |
cba63c30 BW |
1717 | return -EINVAL; |
1718 | } | |
1719 | } | |
1720 | } | |
1721 | ||
1722 | return 0; | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * That function parses "simple" (old) crashkernel command lines like | |
1727 | * | |
e1bebcf4 | 1728 | * crashkernel=size[@offset] |
cba63c30 BW |
1729 | * |
1730 | * It returns 0 on success and -EINVAL on failure. | |
1731 | */ | |
e1bebcf4 FF |
1732 | static int __init parse_crashkernel_simple(char *cmdline, |
1733 | unsigned long long *crash_size, | |
1734 | unsigned long long *crash_base) | |
cba63c30 BW |
1735 | { |
1736 | char *cur = cmdline; | |
1737 | ||
1738 | *crash_size = memparse(cmdline, &cur); | |
1739 | if (cmdline == cur) { | |
e1bebcf4 | 1740 | pr_warn("crashkernel: memory value expected\n"); |
cba63c30 BW |
1741 | return -EINVAL; |
1742 | } | |
1743 | ||
1744 | if (*cur == '@') | |
1745 | *crash_base = memparse(cur+1, &cur); | |
eaa3be6a | 1746 | else if (*cur != ' ' && *cur != '\0') { |
e1bebcf4 | 1747 | pr_warn("crashkernel: unrecognized char\n"); |
eaa3be6a ZD |
1748 | return -EINVAL; |
1749 | } | |
cba63c30 BW |
1750 | |
1751 | return 0; | |
1752 | } | |
1753 | ||
adbc742b YL |
1754 | #define SUFFIX_HIGH 0 |
1755 | #define SUFFIX_LOW 1 | |
1756 | #define SUFFIX_NULL 2 | |
1757 | static __initdata char *suffix_tbl[] = { | |
1758 | [SUFFIX_HIGH] = ",high", | |
1759 | [SUFFIX_LOW] = ",low", | |
1760 | [SUFFIX_NULL] = NULL, | |
1761 | }; | |
1762 | ||
cba63c30 | 1763 | /* |
adbc742b YL |
1764 | * That function parses "suffix" crashkernel command lines like |
1765 | * | |
1766 | * crashkernel=size,[high|low] | |
1767 | * | |
1768 | * It returns 0 on success and -EINVAL on failure. | |
cba63c30 | 1769 | */ |
adbc742b YL |
1770 | static int __init parse_crashkernel_suffix(char *cmdline, |
1771 | unsigned long long *crash_size, | |
adbc742b YL |
1772 | const char *suffix) |
1773 | { | |
1774 | char *cur = cmdline; | |
1775 | ||
1776 | *crash_size = memparse(cmdline, &cur); | |
1777 | if (cmdline == cur) { | |
1778 | pr_warn("crashkernel: memory value expected\n"); | |
1779 | return -EINVAL; | |
1780 | } | |
1781 | ||
1782 | /* check with suffix */ | |
1783 | if (strncmp(cur, suffix, strlen(suffix))) { | |
1784 | pr_warn("crashkernel: unrecognized char\n"); | |
1785 | return -EINVAL; | |
1786 | } | |
1787 | cur += strlen(suffix); | |
1788 | if (*cur != ' ' && *cur != '\0') { | |
1789 | pr_warn("crashkernel: unrecognized char\n"); | |
1790 | return -EINVAL; | |
1791 | } | |
1792 | ||
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | static __init char *get_last_crashkernel(char *cmdline, | |
1797 | const char *name, | |
1798 | const char *suffix) | |
1799 | { | |
1800 | char *p = cmdline, *ck_cmdline = NULL; | |
1801 | ||
1802 | /* find crashkernel and use the last one if there are more */ | |
1803 | p = strstr(p, name); | |
1804 | while (p) { | |
1805 | char *end_p = strchr(p, ' '); | |
1806 | char *q; | |
1807 | ||
1808 | if (!end_p) | |
1809 | end_p = p + strlen(p); | |
1810 | ||
1811 | if (!suffix) { | |
1812 | int i; | |
1813 | ||
1814 | /* skip the one with any known suffix */ | |
1815 | for (i = 0; suffix_tbl[i]; i++) { | |
1816 | q = end_p - strlen(suffix_tbl[i]); | |
1817 | if (!strncmp(q, suffix_tbl[i], | |
1818 | strlen(suffix_tbl[i]))) | |
1819 | goto next; | |
1820 | } | |
1821 | ck_cmdline = p; | |
1822 | } else { | |
1823 | q = end_p - strlen(suffix); | |
1824 | if (!strncmp(q, suffix, strlen(suffix))) | |
1825 | ck_cmdline = p; | |
1826 | } | |
1827 | next: | |
1828 | p = strstr(p+1, name); | |
1829 | } | |
1830 | ||
1831 | if (!ck_cmdline) | |
1832 | return NULL; | |
1833 | ||
1834 | return ck_cmdline; | |
1835 | } | |
1836 | ||
0212f915 | 1837 | static int __init __parse_crashkernel(char *cmdline, |
cba63c30 BW |
1838 | unsigned long long system_ram, |
1839 | unsigned long long *crash_size, | |
0212f915 | 1840 | unsigned long long *crash_base, |
adbc742b YL |
1841 | const char *name, |
1842 | const char *suffix) | |
cba63c30 | 1843 | { |
cba63c30 | 1844 | char *first_colon, *first_space; |
adbc742b | 1845 | char *ck_cmdline; |
cba63c30 BW |
1846 | |
1847 | BUG_ON(!crash_size || !crash_base); | |
1848 | *crash_size = 0; | |
1849 | *crash_base = 0; | |
1850 | ||
adbc742b | 1851 | ck_cmdline = get_last_crashkernel(cmdline, name, suffix); |
cba63c30 BW |
1852 | |
1853 | if (!ck_cmdline) | |
1854 | return -EINVAL; | |
1855 | ||
0212f915 | 1856 | ck_cmdline += strlen(name); |
cba63c30 | 1857 | |
adbc742b YL |
1858 | if (suffix) |
1859 | return parse_crashkernel_suffix(ck_cmdline, crash_size, | |
36f3f500 | 1860 | suffix); |
cba63c30 BW |
1861 | /* |
1862 | * if the commandline contains a ':', then that's the extended | |
1863 | * syntax -- if not, it must be the classic syntax | |
1864 | */ | |
1865 | first_colon = strchr(ck_cmdline, ':'); | |
1866 | first_space = strchr(ck_cmdline, ' '); | |
1867 | if (first_colon && (!first_space || first_colon < first_space)) | |
1868 | return parse_crashkernel_mem(ck_cmdline, system_ram, | |
1869 | crash_size, crash_base); | |
cba63c30 | 1870 | |
80c74f6a | 1871 | return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base); |
cba63c30 BW |
1872 | } |
1873 | ||
adbc742b YL |
1874 | /* |
1875 | * That function is the entry point for command line parsing and should be | |
1876 | * called from the arch-specific code. | |
1877 | */ | |
0212f915 YL |
1878 | int __init parse_crashkernel(char *cmdline, |
1879 | unsigned long long system_ram, | |
1880 | unsigned long long *crash_size, | |
1881 | unsigned long long *crash_base) | |
1882 | { | |
1883 | return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base, | |
adbc742b | 1884 | "crashkernel=", NULL); |
0212f915 | 1885 | } |
55a20ee7 YL |
1886 | |
1887 | int __init parse_crashkernel_high(char *cmdline, | |
1888 | unsigned long long system_ram, | |
1889 | unsigned long long *crash_size, | |
1890 | unsigned long long *crash_base) | |
1891 | { | |
1892 | return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base, | |
adbc742b | 1893 | "crashkernel=", suffix_tbl[SUFFIX_HIGH]); |
55a20ee7 | 1894 | } |
0212f915 YL |
1895 | |
1896 | int __init parse_crashkernel_low(char *cmdline, | |
1897 | unsigned long long system_ram, | |
1898 | unsigned long long *crash_size, | |
1899 | unsigned long long *crash_base) | |
1900 | { | |
1901 | return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base, | |
adbc742b | 1902 | "crashkernel=", suffix_tbl[SUFFIX_LOW]); |
0212f915 | 1903 | } |
cba63c30 | 1904 | |
fa8ff292 | 1905 | static void update_vmcoreinfo_note(void) |
fd59d231 | 1906 | { |
fa8ff292 | 1907 | u32 *buf = vmcoreinfo_note; |
fd59d231 KO |
1908 | |
1909 | if (!vmcoreinfo_size) | |
1910 | return; | |
fd59d231 KO |
1911 | buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data, |
1912 | vmcoreinfo_size); | |
fd59d231 KO |
1913 | final_note(buf); |
1914 | } | |
1915 | ||
fa8ff292 MH |
1916 | void crash_save_vmcoreinfo(void) |
1917 | { | |
63dca8d5 | 1918 | vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds()); |
fa8ff292 MH |
1919 | update_vmcoreinfo_note(); |
1920 | } | |
1921 | ||
fd59d231 KO |
1922 | void vmcoreinfo_append_str(const char *fmt, ...) |
1923 | { | |
1924 | va_list args; | |
1925 | char buf[0x50]; | |
310faaa9 | 1926 | size_t r; |
fd59d231 KO |
1927 | |
1928 | va_start(args, fmt); | |
a19428e5 | 1929 | r = vscnprintf(buf, sizeof(buf), fmt, args); |
fd59d231 KO |
1930 | va_end(args); |
1931 | ||
31c3a3fe | 1932 | r = min(r, vmcoreinfo_max_size - vmcoreinfo_size); |
fd59d231 KO |
1933 | |
1934 | memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r); | |
1935 | ||
1936 | vmcoreinfo_size += r; | |
1937 | } | |
1938 | ||
1939 | /* | |
1940 | * provide an empty default implementation here -- architecture | |
1941 | * code may override this | |
1942 | */ | |
52f5684c | 1943 | void __weak arch_crash_save_vmcoreinfo(void) |
fd59d231 KO |
1944 | {} |
1945 | ||
52f5684c | 1946 | unsigned long __weak paddr_vmcoreinfo_note(void) |
fd59d231 KO |
1947 | { |
1948 | return __pa((unsigned long)(char *)&vmcoreinfo_note); | |
1949 | } | |
1950 | ||
1951 | static int __init crash_save_vmcoreinfo_init(void) | |
1952 | { | |
bba1f603 KO |
1953 | VMCOREINFO_OSRELEASE(init_uts_ns.name.release); |
1954 | VMCOREINFO_PAGESIZE(PAGE_SIZE); | |
fd59d231 | 1955 | |
bcbba6c1 KO |
1956 | VMCOREINFO_SYMBOL(init_uts_ns); |
1957 | VMCOREINFO_SYMBOL(node_online_map); | |
d034cfab | 1958 | #ifdef CONFIG_MMU |
bcbba6c1 | 1959 | VMCOREINFO_SYMBOL(swapper_pg_dir); |
d034cfab | 1960 | #endif |
bcbba6c1 | 1961 | VMCOREINFO_SYMBOL(_stext); |
f1c4069e | 1962 | VMCOREINFO_SYMBOL(vmap_area_list); |
fd59d231 KO |
1963 | |
1964 | #ifndef CONFIG_NEED_MULTIPLE_NODES | |
bcbba6c1 KO |
1965 | VMCOREINFO_SYMBOL(mem_map); |
1966 | VMCOREINFO_SYMBOL(contig_page_data); | |
fd59d231 KO |
1967 | #endif |
1968 | #ifdef CONFIG_SPARSEMEM | |
bcbba6c1 KO |
1969 | VMCOREINFO_SYMBOL(mem_section); |
1970 | VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS); | |
c76f860c | 1971 | VMCOREINFO_STRUCT_SIZE(mem_section); |
bcbba6c1 | 1972 | VMCOREINFO_OFFSET(mem_section, section_mem_map); |
fd59d231 | 1973 | #endif |
c76f860c KO |
1974 | VMCOREINFO_STRUCT_SIZE(page); |
1975 | VMCOREINFO_STRUCT_SIZE(pglist_data); | |
1976 | VMCOREINFO_STRUCT_SIZE(zone); | |
1977 | VMCOREINFO_STRUCT_SIZE(free_area); | |
1978 | VMCOREINFO_STRUCT_SIZE(list_head); | |
1979 | VMCOREINFO_SIZE(nodemask_t); | |
bcbba6c1 KO |
1980 | VMCOREINFO_OFFSET(page, flags); |
1981 | VMCOREINFO_OFFSET(page, _count); | |
1982 | VMCOREINFO_OFFSET(page, mapping); | |
1983 | VMCOREINFO_OFFSET(page, lru); | |
8d67091e AK |
1984 | VMCOREINFO_OFFSET(page, _mapcount); |
1985 | VMCOREINFO_OFFSET(page, private); | |
bcbba6c1 KO |
1986 | VMCOREINFO_OFFSET(pglist_data, node_zones); |
1987 | VMCOREINFO_OFFSET(pglist_data, nr_zones); | |
fd59d231 | 1988 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
bcbba6c1 | 1989 | VMCOREINFO_OFFSET(pglist_data, node_mem_map); |
fd59d231 | 1990 | #endif |
bcbba6c1 KO |
1991 | VMCOREINFO_OFFSET(pglist_data, node_start_pfn); |
1992 | VMCOREINFO_OFFSET(pglist_data, node_spanned_pages); | |
1993 | VMCOREINFO_OFFSET(pglist_data, node_id); | |
1994 | VMCOREINFO_OFFSET(zone, free_area); | |
1995 | VMCOREINFO_OFFSET(zone, vm_stat); | |
1996 | VMCOREINFO_OFFSET(zone, spanned_pages); | |
1997 | VMCOREINFO_OFFSET(free_area, free_list); | |
1998 | VMCOREINFO_OFFSET(list_head, next); | |
1999 | VMCOREINFO_OFFSET(list_head, prev); | |
13ba3fcb AK |
2000 | VMCOREINFO_OFFSET(vmap_area, va_start); |
2001 | VMCOREINFO_OFFSET(vmap_area, list); | |
bcbba6c1 | 2002 | VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER); |
04d491ab | 2003 | log_buf_kexec_setup(); |
83a08e7c | 2004 | VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES); |
bcbba6c1 | 2005 | VMCOREINFO_NUMBER(NR_FREE_PAGES); |
122c7a59 KO |
2006 | VMCOREINFO_NUMBER(PG_lru); |
2007 | VMCOREINFO_NUMBER(PG_private); | |
2008 | VMCOREINFO_NUMBER(PG_swapcache); | |
8d67091e | 2009 | VMCOREINFO_NUMBER(PG_slab); |
0d0bf667 MT |
2010 | #ifdef CONFIG_MEMORY_FAILURE |
2011 | VMCOREINFO_NUMBER(PG_hwpoison); | |
2012 | #endif | |
b3acc56b | 2013 | VMCOREINFO_NUMBER(PG_head_mask); |
8d67091e | 2014 | VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE); |
3a1122d2 | 2015 | #ifdef CONFIG_HUGETLBFS |
8f1d26d0 | 2016 | VMCOREINFO_SYMBOL(free_huge_page); |
3a1122d2 | 2017 | #endif |
fd59d231 KO |
2018 | |
2019 | arch_crash_save_vmcoreinfo(); | |
fa8ff292 | 2020 | update_vmcoreinfo_note(); |
fd59d231 KO |
2021 | |
2022 | return 0; | |
2023 | } | |
2024 | ||
c96d6660 | 2025 | subsys_initcall(crash_save_vmcoreinfo_init); |
3ab83521 | 2026 | |
74ca317c | 2027 | #ifdef CONFIG_KEXEC_FILE |
cb105258 VG |
2028 | static int locate_mem_hole_top_down(unsigned long start, unsigned long end, |
2029 | struct kexec_buf *kbuf) | |
2030 | { | |
2031 | struct kimage *image = kbuf->image; | |
2032 | unsigned long temp_start, temp_end; | |
2033 | ||
2034 | temp_end = min(end, kbuf->buf_max); | |
2035 | temp_start = temp_end - kbuf->memsz; | |
2036 | ||
2037 | do { | |
2038 | /* align down start */ | |
2039 | temp_start = temp_start & (~(kbuf->buf_align - 1)); | |
2040 | ||
2041 | if (temp_start < start || temp_start < kbuf->buf_min) | |
2042 | return 0; | |
2043 | ||
2044 | temp_end = temp_start + kbuf->memsz - 1; | |
2045 | ||
2046 | /* | |
2047 | * Make sure this does not conflict with any of existing | |
2048 | * segments | |
2049 | */ | |
2050 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
2051 | temp_start = temp_start - PAGE_SIZE; | |
2052 | continue; | |
2053 | } | |
2054 | ||
2055 | /* We found a suitable memory range */ | |
2056 | break; | |
2057 | } while (1); | |
2058 | ||
2059 | /* If we are here, we found a suitable memory range */ | |
669280a1 | 2060 | kbuf->mem = temp_start; |
cb105258 VG |
2061 | |
2062 | /* Success, stop navigating through remaining System RAM ranges */ | |
2063 | return 1; | |
2064 | } | |
2065 | ||
2066 | static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end, | |
2067 | struct kexec_buf *kbuf) | |
2068 | { | |
2069 | struct kimage *image = kbuf->image; | |
2070 | unsigned long temp_start, temp_end; | |
2071 | ||
2072 | temp_start = max(start, kbuf->buf_min); | |
2073 | ||
2074 | do { | |
2075 | temp_start = ALIGN(temp_start, kbuf->buf_align); | |
2076 | temp_end = temp_start + kbuf->memsz - 1; | |
2077 | ||
2078 | if (temp_end > end || temp_end > kbuf->buf_max) | |
2079 | return 0; | |
2080 | /* | |
2081 | * Make sure this does not conflict with any of existing | |
2082 | * segments | |
2083 | */ | |
2084 | if (kimage_is_destination_range(image, temp_start, temp_end)) { | |
2085 | temp_start = temp_start + PAGE_SIZE; | |
2086 | continue; | |
2087 | } | |
2088 | ||
2089 | /* We found a suitable memory range */ | |
2090 | break; | |
2091 | } while (1); | |
2092 | ||
2093 | /* If we are here, we found a suitable memory range */ | |
669280a1 | 2094 | kbuf->mem = temp_start; |
cb105258 VG |
2095 | |
2096 | /* Success, stop navigating through remaining System RAM ranges */ | |
2097 | return 1; | |
2098 | } | |
2099 | ||
2100 | static int locate_mem_hole_callback(u64 start, u64 end, void *arg) | |
2101 | { | |
2102 | struct kexec_buf *kbuf = (struct kexec_buf *)arg; | |
2103 | unsigned long sz = end - start + 1; | |
2104 | ||
2105 | /* Returning 0 will take to next memory range */ | |
2106 | if (sz < kbuf->memsz) | |
2107 | return 0; | |
2108 | ||
2109 | if (end < kbuf->buf_min || start > kbuf->buf_max) | |
2110 | return 0; | |
2111 | ||
2112 | /* | |
2113 | * Allocate memory top down with-in ram range. Otherwise bottom up | |
2114 | * allocation. | |
2115 | */ | |
2116 | if (kbuf->top_down) | |
2117 | return locate_mem_hole_top_down(start, end, kbuf); | |
2118 | return locate_mem_hole_bottom_up(start, end, kbuf); | |
2119 | } | |
2120 | ||
2121 | /* | |
2122 | * Helper function for placing a buffer in a kexec segment. This assumes | |
2123 | * that kexec_mutex is held. | |
2124 | */ | |
2125 | int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz, | |
2126 | unsigned long memsz, unsigned long buf_align, | |
2127 | unsigned long buf_min, unsigned long buf_max, | |
2128 | bool top_down, unsigned long *load_addr) | |
2129 | { | |
2130 | ||
2131 | struct kexec_segment *ksegment; | |
2132 | struct kexec_buf buf, *kbuf; | |
2133 | int ret; | |
2134 | ||
2135 | /* Currently adding segment this way is allowed only in file mode */ | |
2136 | if (!image->file_mode) | |
2137 | return -EINVAL; | |
2138 | ||
2139 | if (image->nr_segments >= KEXEC_SEGMENT_MAX) | |
2140 | return -EINVAL; | |
2141 | ||
2142 | /* | |
2143 | * Make sure we are not trying to add buffer after allocating | |
2144 | * control pages. All segments need to be placed first before | |
2145 | * any control pages are allocated. As control page allocation | |
2146 | * logic goes through list of segments to make sure there are | |
2147 | * no destination overlaps. | |
2148 | */ | |
2149 | if (!list_empty(&image->control_pages)) { | |
2150 | WARN_ON(1); | |
2151 | return -EINVAL; | |
2152 | } | |
2153 | ||
2154 | memset(&buf, 0, sizeof(struct kexec_buf)); | |
2155 | kbuf = &buf; | |
2156 | kbuf->image = image; | |
2157 | kbuf->buffer = buffer; | |
2158 | kbuf->bufsz = bufsz; | |
2159 | ||
2160 | kbuf->memsz = ALIGN(memsz, PAGE_SIZE); | |
2161 | kbuf->buf_align = max(buf_align, PAGE_SIZE); | |
2162 | kbuf->buf_min = buf_min; | |
2163 | kbuf->buf_max = buf_max; | |
2164 | kbuf->top_down = top_down; | |
2165 | ||
2166 | /* Walk the RAM ranges and allocate a suitable range for the buffer */ | |
dd5f7260 VG |
2167 | if (image->type == KEXEC_TYPE_CRASH) |
2168 | ret = walk_iomem_res("Crash kernel", | |
2169 | IORESOURCE_MEM | IORESOURCE_BUSY, | |
2170 | crashk_res.start, crashk_res.end, kbuf, | |
2171 | locate_mem_hole_callback); | |
2172 | else | |
2173 | ret = walk_system_ram_res(0, -1, kbuf, | |
2174 | locate_mem_hole_callback); | |
cb105258 VG |
2175 | if (ret != 1) { |
2176 | /* A suitable memory range could not be found for buffer */ | |
2177 | return -EADDRNOTAVAIL; | |
2178 | } | |
2179 | ||
2180 | /* Found a suitable memory range */ | |
669280a1 BH |
2181 | ksegment = &image->segment[image->nr_segments]; |
2182 | ksegment->kbuf = kbuf->buffer; | |
2183 | ksegment->bufsz = kbuf->bufsz; | |
2184 | ksegment->mem = kbuf->mem; | |
2185 | ksegment->memsz = kbuf->memsz; | |
2186 | image->nr_segments++; | |
cb105258 VG |
2187 | *load_addr = ksegment->mem; |
2188 | return 0; | |
2189 | } | |
2190 | ||
12db5562 VG |
2191 | /* Calculate and store the digest of segments */ |
2192 | static int kexec_calculate_store_digests(struct kimage *image) | |
2193 | { | |
2194 | struct crypto_shash *tfm; | |
2195 | struct shash_desc *desc; | |
2196 | int ret = 0, i, j, zero_buf_sz, sha_region_sz; | |
2197 | size_t desc_size, nullsz; | |
2198 | char *digest; | |
2199 | void *zero_buf; | |
2200 | struct kexec_sha_region *sha_regions; | |
2201 | struct purgatory_info *pi = &image->purgatory_info; | |
2202 | ||
2203 | zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT); | |
2204 | zero_buf_sz = PAGE_SIZE; | |
2205 | ||
2206 | tfm = crypto_alloc_shash("sha256", 0, 0); | |
2207 | if (IS_ERR(tfm)) { | |
2208 | ret = PTR_ERR(tfm); | |
2209 | goto out; | |
2210 | } | |
2211 | ||
2212 | desc_size = crypto_shash_descsize(tfm) + sizeof(*desc); | |
2213 | desc = kzalloc(desc_size, GFP_KERNEL); | |
2214 | if (!desc) { | |
2215 | ret = -ENOMEM; | |
2216 | goto out_free_tfm; | |
2217 | } | |
2218 | ||
2219 | sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region); | |
2220 | sha_regions = vzalloc(sha_region_sz); | |
2221 | if (!sha_regions) | |
2222 | goto out_free_desc; | |
2223 | ||
2224 | desc->tfm = tfm; | |
2225 | desc->flags = 0; | |
2226 | ||
2227 | ret = crypto_shash_init(desc); | |
2228 | if (ret < 0) | |
2229 | goto out_free_sha_regions; | |
2230 | ||
2231 | digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL); | |
2232 | if (!digest) { | |
2233 | ret = -ENOMEM; | |
2234 | goto out_free_sha_regions; | |
2235 | } | |
2236 | ||
2237 | for (j = i = 0; i < image->nr_segments; i++) { | |
2238 | struct kexec_segment *ksegment; | |
2239 | ||
2240 | ksegment = &image->segment[i]; | |
2241 | /* | |
2242 | * Skip purgatory as it will be modified once we put digest | |
2243 | * info in purgatory. | |
2244 | */ | |
2245 | if (ksegment->kbuf == pi->purgatory_buf) | |
2246 | continue; | |
2247 | ||
2248 | ret = crypto_shash_update(desc, ksegment->kbuf, | |
2249 | ksegment->bufsz); | |
2250 | if (ret) | |
2251 | break; | |
2252 | ||
2253 | /* | |
2254 | * Assume rest of the buffer is filled with zero and | |
2255 | * update digest accordingly. | |
2256 | */ | |
2257 | nullsz = ksegment->memsz - ksegment->bufsz; | |
2258 | while (nullsz) { | |
2259 | unsigned long bytes = nullsz; | |
2260 | ||
2261 | if (bytes > zero_buf_sz) | |
2262 | bytes = zero_buf_sz; | |
2263 | ret = crypto_shash_update(desc, zero_buf, bytes); | |
2264 | if (ret) | |
2265 | break; | |
2266 | nullsz -= bytes; | |
2267 | } | |
2268 | ||
2269 | if (ret) | |
2270 | break; | |
2271 | ||
2272 | sha_regions[j].start = ksegment->mem; | |
2273 | sha_regions[j].len = ksegment->memsz; | |
2274 | j++; | |
2275 | } | |
2276 | ||
2277 | if (!ret) { | |
2278 | ret = crypto_shash_final(desc, digest); | |
2279 | if (ret) | |
2280 | goto out_free_digest; | |
2281 | ret = kexec_purgatory_get_set_symbol(image, "sha_regions", | |
2282 | sha_regions, sha_region_sz, 0); | |
2283 | if (ret) | |
2284 | goto out_free_digest; | |
2285 | ||
2286 | ret = kexec_purgatory_get_set_symbol(image, "sha256_digest", | |
2287 | digest, SHA256_DIGEST_SIZE, 0); | |
2288 | if (ret) | |
2289 | goto out_free_digest; | |
2290 | } | |
2291 | ||
2292 | out_free_digest: | |
2293 | kfree(digest); | |
2294 | out_free_sha_regions: | |
2295 | vfree(sha_regions); | |
2296 | out_free_desc: | |
2297 | kfree(desc); | |
2298 | out_free_tfm: | |
2299 | kfree(tfm); | |
2300 | out: | |
2301 | return ret; | |
2302 | } | |
2303 | ||
2304 | /* Actually load purgatory. Lot of code taken from kexec-tools */ | |
2305 | static int __kexec_load_purgatory(struct kimage *image, unsigned long min, | |
2306 | unsigned long max, int top_down) | |
2307 | { | |
2308 | struct purgatory_info *pi = &image->purgatory_info; | |
2309 | unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad; | |
2310 | unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset; | |
2311 | unsigned char *buf_addr, *src; | |
2312 | int i, ret = 0, entry_sidx = -1; | |
2313 | const Elf_Shdr *sechdrs_c; | |
2314 | Elf_Shdr *sechdrs = NULL; | |
2315 | void *purgatory_buf = NULL; | |
2316 | ||
2317 | /* | |
2318 | * sechdrs_c points to section headers in purgatory and are read | |
2319 | * only. No modifications allowed. | |
2320 | */ | |
2321 | sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff; | |
2322 | ||
2323 | /* | |
2324 | * We can not modify sechdrs_c[] and its fields. It is read only. | |
2325 | * Copy it over to a local copy where one can store some temporary | |
2326 | * data and free it at the end. We need to modify ->sh_addr and | |
2327 | * ->sh_offset fields to keep track of permanent and temporary | |
2328 | * locations of sections. | |
2329 | */ | |
2330 | sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
2331 | if (!sechdrs) | |
2332 | return -ENOMEM; | |
2333 | ||
2334 | memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr)); | |
2335 | ||
2336 | /* | |
2337 | * We seem to have multiple copies of sections. First copy is which | |
2338 | * is embedded in kernel in read only section. Some of these sections | |
2339 | * will be copied to a temporary buffer and relocated. And these | |
2340 | * sections will finally be copied to their final destination at | |
2341 | * segment load time. | |
2342 | * | |
2343 | * Use ->sh_offset to reflect section address in memory. It will | |
2344 | * point to original read only copy if section is not allocatable. | |
2345 | * Otherwise it will point to temporary copy which will be relocated. | |
2346 | * | |
2347 | * Use ->sh_addr to contain final address of the section where it | |
2348 | * will go during execution time. | |
2349 | */ | |
2350 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
2351 | if (sechdrs[i].sh_type == SHT_NOBITS) | |
2352 | continue; | |
2353 | ||
2354 | sechdrs[i].sh_offset = (unsigned long)pi->ehdr + | |
2355 | sechdrs[i].sh_offset; | |
2356 | } | |
2357 | ||
2358 | /* | |
2359 | * Identify entry point section and make entry relative to section | |
2360 | * start. | |
2361 | */ | |
2362 | entry = pi->ehdr->e_entry; | |
2363 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
2364 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
2365 | continue; | |
2366 | ||
2367 | if (!(sechdrs[i].sh_flags & SHF_EXECINSTR)) | |
2368 | continue; | |
2369 | ||
2370 | /* Make entry section relative */ | |
2371 | if (sechdrs[i].sh_addr <= pi->ehdr->e_entry && | |
2372 | ((sechdrs[i].sh_addr + sechdrs[i].sh_size) > | |
2373 | pi->ehdr->e_entry)) { | |
2374 | entry_sidx = i; | |
2375 | entry -= sechdrs[i].sh_addr; | |
2376 | break; | |
2377 | } | |
2378 | } | |
2379 | ||
2380 | /* Determine how much memory is needed to load relocatable object. */ | |
2381 | buf_align = 1; | |
2382 | bss_align = 1; | |
2383 | buf_sz = 0; | |
2384 | bss_sz = 0; | |
2385 | ||
2386 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
2387 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
2388 | continue; | |
2389 | ||
2390 | align = sechdrs[i].sh_addralign; | |
2391 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
2392 | if (buf_align < align) | |
2393 | buf_align = align; | |
2394 | buf_sz = ALIGN(buf_sz, align); | |
2395 | buf_sz += sechdrs[i].sh_size; | |
2396 | } else { | |
2397 | /* bss section */ | |
2398 | if (bss_align < align) | |
2399 | bss_align = align; | |
2400 | bss_sz = ALIGN(bss_sz, align); | |
2401 | bss_sz += sechdrs[i].sh_size; | |
2402 | } | |
2403 | } | |
2404 | ||
2405 | /* Determine the bss padding required to align bss properly */ | |
2406 | bss_pad = 0; | |
2407 | if (buf_sz & (bss_align - 1)) | |
2408 | bss_pad = bss_align - (buf_sz & (bss_align - 1)); | |
2409 | ||
2410 | memsz = buf_sz + bss_pad + bss_sz; | |
2411 | ||
2412 | /* Allocate buffer for purgatory */ | |
2413 | purgatory_buf = vzalloc(buf_sz); | |
2414 | if (!purgatory_buf) { | |
2415 | ret = -ENOMEM; | |
2416 | goto out; | |
2417 | } | |
2418 | ||
2419 | if (buf_align < bss_align) | |
2420 | buf_align = bss_align; | |
2421 | ||
2422 | /* Add buffer to segment list */ | |
2423 | ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz, | |
2424 | buf_align, min, max, top_down, | |
2425 | &pi->purgatory_load_addr); | |
2426 | if (ret) | |
2427 | goto out; | |
2428 | ||
2429 | /* Load SHF_ALLOC sections */ | |
2430 | buf_addr = purgatory_buf; | |
2431 | load_addr = curr_load_addr = pi->purgatory_load_addr; | |
2432 | bss_addr = load_addr + buf_sz + bss_pad; | |
2433 | ||
2434 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
2435 | if (!(sechdrs[i].sh_flags & SHF_ALLOC)) | |
2436 | continue; | |
2437 | ||
2438 | align = sechdrs[i].sh_addralign; | |
2439 | if (sechdrs[i].sh_type != SHT_NOBITS) { | |
2440 | curr_load_addr = ALIGN(curr_load_addr, align); | |
2441 | offset = curr_load_addr - load_addr; | |
2442 | /* We already modifed ->sh_offset to keep src addr */ | |
2443 | src = (char *) sechdrs[i].sh_offset; | |
2444 | memcpy(buf_addr + offset, src, sechdrs[i].sh_size); | |
2445 | ||
2446 | /* Store load address and source address of section */ | |
2447 | sechdrs[i].sh_addr = curr_load_addr; | |
2448 | ||
2449 | /* | |
2450 | * This section got copied to temporary buffer. Update | |
2451 | * ->sh_offset accordingly. | |
2452 | */ | |
2453 | sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset); | |
2454 | ||
2455 | /* Advance to the next address */ | |
2456 | curr_load_addr += sechdrs[i].sh_size; | |
2457 | } else { | |
2458 | bss_addr = ALIGN(bss_addr, align); | |
2459 | sechdrs[i].sh_addr = bss_addr; | |
2460 | bss_addr += sechdrs[i].sh_size; | |
2461 | } | |
2462 | } | |
2463 | ||
2464 | /* Update entry point based on load address of text section */ | |
2465 | if (entry_sidx >= 0) | |
2466 | entry += sechdrs[entry_sidx].sh_addr; | |
2467 | ||
2468 | /* Make kernel jump to purgatory after shutdown */ | |
2469 | image->start = entry; | |
2470 | ||
2471 | /* Used later to get/set symbol values */ | |
2472 | pi->sechdrs = sechdrs; | |
2473 | ||
2474 | /* | |
2475 | * Used later to identify which section is purgatory and skip it | |
2476 | * from checksumming. | |
2477 | */ | |
2478 | pi->purgatory_buf = purgatory_buf; | |
2479 | return ret; | |
2480 | out: | |
2481 | vfree(sechdrs); | |
2482 | vfree(purgatory_buf); | |
2483 | return ret; | |
2484 | } | |
2485 | ||
2486 | static int kexec_apply_relocations(struct kimage *image) | |
2487 | { | |
2488 | int i, ret; | |
2489 | struct purgatory_info *pi = &image->purgatory_info; | |
2490 | Elf_Shdr *sechdrs = pi->sechdrs; | |
2491 | ||
2492 | /* Apply relocations */ | |
2493 | for (i = 0; i < pi->ehdr->e_shnum; i++) { | |
2494 | Elf_Shdr *section, *symtab; | |
2495 | ||
2496 | if (sechdrs[i].sh_type != SHT_RELA && | |
2497 | sechdrs[i].sh_type != SHT_REL) | |
2498 | continue; | |
2499 | ||
2500 | /* | |
2501 | * For section of type SHT_RELA/SHT_REL, | |
2502 | * ->sh_link contains section header index of associated | |
2503 | * symbol table. And ->sh_info contains section header | |
2504 | * index of section to which relocations apply. | |
2505 | */ | |
2506 | if (sechdrs[i].sh_info >= pi->ehdr->e_shnum || | |
2507 | sechdrs[i].sh_link >= pi->ehdr->e_shnum) | |
2508 | return -ENOEXEC; | |
2509 | ||
2510 | section = &sechdrs[sechdrs[i].sh_info]; | |
2511 | symtab = &sechdrs[sechdrs[i].sh_link]; | |
2512 | ||
2513 | if (!(section->sh_flags & SHF_ALLOC)) | |
2514 | continue; | |
2515 | ||
2516 | /* | |
2517 | * symtab->sh_link contain section header index of associated | |
2518 | * string table. | |
2519 | */ | |
2520 | if (symtab->sh_link >= pi->ehdr->e_shnum) | |
2521 | /* Invalid section number? */ | |
2522 | continue; | |
2523 | ||
2524 | /* | |
edb0ec07 | 2525 | * Respective architecture needs to provide support for applying |
12db5562 VG |
2526 | * relocations of type SHT_RELA/SHT_REL. |
2527 | */ | |
2528 | if (sechdrs[i].sh_type == SHT_RELA) | |
2529 | ret = arch_kexec_apply_relocations_add(pi->ehdr, | |
2530 | sechdrs, i); | |
2531 | else if (sechdrs[i].sh_type == SHT_REL) | |
2532 | ret = arch_kexec_apply_relocations(pi->ehdr, | |
2533 | sechdrs, i); | |
2534 | if (ret) | |
2535 | return ret; | |
2536 | } | |
2537 | ||
2538 | return 0; | |
2539 | } | |
2540 | ||
2541 | /* Load relocatable purgatory object and relocate it appropriately */ | |
2542 | int kexec_load_purgatory(struct kimage *image, unsigned long min, | |
2543 | unsigned long max, int top_down, | |
2544 | unsigned long *load_addr) | |
2545 | { | |
2546 | struct purgatory_info *pi = &image->purgatory_info; | |
2547 | int ret; | |
2548 | ||
2549 | if (kexec_purgatory_size <= 0) | |
2550 | return -EINVAL; | |
2551 | ||
2552 | if (kexec_purgatory_size < sizeof(Elf_Ehdr)) | |
2553 | return -ENOEXEC; | |
2554 | ||
2555 | pi->ehdr = (Elf_Ehdr *)kexec_purgatory; | |
2556 | ||
2557 | if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0 | |
2558 | || pi->ehdr->e_type != ET_REL | |
2559 | || !elf_check_arch(pi->ehdr) | |
2560 | || pi->ehdr->e_shentsize != sizeof(Elf_Shdr)) | |
2561 | return -ENOEXEC; | |
2562 | ||
2563 | if (pi->ehdr->e_shoff >= kexec_purgatory_size | |
2564 | || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) > | |
2565 | kexec_purgatory_size - pi->ehdr->e_shoff)) | |
2566 | return -ENOEXEC; | |
2567 | ||
2568 | ret = __kexec_load_purgatory(image, min, max, top_down); | |
2569 | if (ret) | |
2570 | return ret; | |
2571 | ||
2572 | ret = kexec_apply_relocations(image); | |
2573 | if (ret) | |
2574 | goto out; | |
2575 | ||
2576 | *load_addr = pi->purgatory_load_addr; | |
2577 | return 0; | |
2578 | out: | |
2579 | vfree(pi->sechdrs); | |
2580 | vfree(pi->purgatory_buf); | |
2581 | return ret; | |
2582 | } | |
2583 | ||
2584 | static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi, | |
2585 | const char *name) | |
2586 | { | |
2587 | Elf_Sym *syms; | |
2588 | Elf_Shdr *sechdrs; | |
2589 | Elf_Ehdr *ehdr; | |
2590 | int i, k; | |
2591 | const char *strtab; | |
2592 | ||
2593 | if (!pi->sechdrs || !pi->ehdr) | |
2594 | return NULL; | |
2595 | ||
2596 | sechdrs = pi->sechdrs; | |
2597 | ehdr = pi->ehdr; | |
2598 | ||
2599 | for (i = 0; i < ehdr->e_shnum; i++) { | |
2600 | if (sechdrs[i].sh_type != SHT_SYMTAB) | |
2601 | continue; | |
2602 | ||
2603 | if (sechdrs[i].sh_link >= ehdr->e_shnum) | |
2604 | /* Invalid strtab section number */ | |
2605 | continue; | |
2606 | strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset; | |
2607 | syms = (Elf_Sym *)sechdrs[i].sh_offset; | |
2608 | ||
2609 | /* Go through symbols for a match */ | |
2610 | for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) { | |
2611 | if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL) | |
2612 | continue; | |
2613 | ||
2614 | if (strcmp(strtab + syms[k].st_name, name) != 0) | |
2615 | continue; | |
2616 | ||
2617 | if (syms[k].st_shndx == SHN_UNDEF || | |
2618 | syms[k].st_shndx >= ehdr->e_shnum) { | |
2619 | pr_debug("Symbol: %s has bad section index %d.\n", | |
2620 | name, syms[k].st_shndx); | |
2621 | return NULL; | |
2622 | } | |
2623 | ||
2624 | /* Found the symbol we are looking for */ | |
2625 | return &syms[k]; | |
2626 | } | |
2627 | } | |
2628 | ||
2629 | return NULL; | |
2630 | } | |
2631 | ||
2632 | void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name) | |
2633 | { | |
2634 | struct purgatory_info *pi = &image->purgatory_info; | |
2635 | Elf_Sym *sym; | |
2636 | Elf_Shdr *sechdr; | |
2637 | ||
2638 | sym = kexec_purgatory_find_symbol(pi, name); | |
2639 | if (!sym) | |
2640 | return ERR_PTR(-EINVAL); | |
2641 | ||
2642 | sechdr = &pi->sechdrs[sym->st_shndx]; | |
2643 | ||
2644 | /* | |
2645 | * Returns the address where symbol will finally be loaded after | |
2646 | * kexec_load_segment() | |
2647 | */ | |
2648 | return (void *)(sechdr->sh_addr + sym->st_value); | |
2649 | } | |
2650 | ||
2651 | /* | |
2652 | * Get or set value of a symbol. If "get_value" is true, symbol value is | |
2653 | * returned in buf otherwise symbol value is set based on value in buf. | |
2654 | */ | |
2655 | int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name, | |
2656 | void *buf, unsigned int size, bool get_value) | |
2657 | { | |
2658 | Elf_Sym *sym; | |
2659 | Elf_Shdr *sechdrs; | |
2660 | struct purgatory_info *pi = &image->purgatory_info; | |
2661 | char *sym_buf; | |
2662 | ||
2663 | sym = kexec_purgatory_find_symbol(pi, name); | |
2664 | if (!sym) | |
2665 | return -EINVAL; | |
2666 | ||
2667 | if (sym->st_size != size) { | |
2668 | pr_err("symbol %s size mismatch: expected %lu actual %u\n", | |
2669 | name, (unsigned long)sym->st_size, size); | |
2670 | return -EINVAL; | |
2671 | } | |
2672 | ||
2673 | sechdrs = pi->sechdrs; | |
2674 | ||
2675 | if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) { | |
2676 | pr_err("symbol %s is in a bss section. Cannot %s\n", name, | |
2677 | get_value ? "get" : "set"); | |
2678 | return -EINVAL; | |
2679 | } | |
2680 | ||
2681 | sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset + | |
2682 | sym->st_value; | |
2683 | ||
2684 | if (get_value) | |
2685 | memcpy((void *)buf, sym_buf, size); | |
2686 | else | |
2687 | memcpy((void *)sym_buf, buf, size); | |
2688 | ||
2689 | return 0; | |
2690 | } | |
74ca317c | 2691 | #endif /* CONFIG_KEXEC_FILE */ |
cb105258 | 2692 | |
7ade3fcc HY |
2693 | /* |
2694 | * Move into place and start executing a preloaded standalone | |
2695 | * executable. If nothing was preloaded return an error. | |
3ab83521 HY |
2696 | */ |
2697 | int kernel_kexec(void) | |
2698 | { | |
2699 | int error = 0; | |
2700 | ||
8c5a1cf0 | 2701 | if (!mutex_trylock(&kexec_mutex)) |
3ab83521 HY |
2702 | return -EBUSY; |
2703 | if (!kexec_image) { | |
2704 | error = -EINVAL; | |
2705 | goto Unlock; | |
2706 | } | |
2707 | ||
3ab83521 | 2708 | #ifdef CONFIG_KEXEC_JUMP |
7ade3fcc | 2709 | if (kexec_image->preserve_context) { |
bcda53fa | 2710 | lock_system_sleep(); |
89081d17 HY |
2711 | pm_prepare_console(); |
2712 | error = freeze_processes(); | |
2713 | if (error) { | |
2714 | error = -EBUSY; | |
2715 | goto Restore_console; | |
2716 | } | |
2717 | suspend_console(); | |
d1616302 | 2718 | error = dpm_suspend_start(PMSG_FREEZE); |
89081d17 HY |
2719 | if (error) |
2720 | goto Resume_console; | |
d1616302 | 2721 | /* At this point, dpm_suspend_start() has been called, |
cf579dfb RW |
2722 | * but *not* dpm_suspend_end(). We *must* call |
2723 | * dpm_suspend_end() now. Otherwise, drivers for | |
89081d17 HY |
2724 | * some devices (e.g. interrupt controllers) become |
2725 | * desynchronized with the actual state of the | |
2726 | * hardware at resume time, and evil weirdness ensues. | |
2727 | */ | |
cf579dfb | 2728 | error = dpm_suspend_end(PMSG_FREEZE); |
89081d17 | 2729 | if (error) |
749b0afc RW |
2730 | goto Resume_devices; |
2731 | error = disable_nonboot_cpus(); | |
2732 | if (error) | |
2733 | goto Enable_cpus; | |
2ed8d2b3 | 2734 | local_irq_disable(); |
2e711c04 | 2735 | error = syscore_suspend(); |
770824bd | 2736 | if (error) |
749b0afc | 2737 | goto Enable_irqs; |
7ade3fcc | 2738 | } else |
3ab83521 | 2739 | #endif |
7ade3fcc | 2740 | { |
4fc9bbf9 | 2741 | kexec_in_progress = true; |
ca195b7f | 2742 | kernel_restart_prepare(NULL); |
c97102ba | 2743 | migrate_to_reboot_cpu(); |
011e4b02 SB |
2744 | |
2745 | /* | |
2746 | * migrate_to_reboot_cpu() disables CPU hotplug assuming that | |
2747 | * no further code needs to use CPU hotplug (which is true in | |
2748 | * the reboot case). However, the kexec path depends on using | |
2749 | * CPU hotplug again; so re-enable it here. | |
2750 | */ | |
2751 | cpu_hotplug_enable(); | |
e1bebcf4 | 2752 | pr_emerg("Starting new kernel\n"); |
3ab83521 HY |
2753 | machine_shutdown(); |
2754 | } | |
2755 | ||
2756 | machine_kexec(kexec_image); | |
2757 | ||
3ab83521 | 2758 | #ifdef CONFIG_KEXEC_JUMP |
7ade3fcc | 2759 | if (kexec_image->preserve_context) { |
19234c08 | 2760 | syscore_resume(); |
749b0afc | 2761 | Enable_irqs: |
3ab83521 | 2762 | local_irq_enable(); |
749b0afc | 2763 | Enable_cpus: |
89081d17 | 2764 | enable_nonboot_cpus(); |
cf579dfb | 2765 | dpm_resume_start(PMSG_RESTORE); |
89081d17 | 2766 | Resume_devices: |
d1616302 | 2767 | dpm_resume_end(PMSG_RESTORE); |
89081d17 HY |
2768 | Resume_console: |
2769 | resume_console(); | |
2770 | thaw_processes(); | |
2771 | Restore_console: | |
2772 | pm_restore_console(); | |
bcda53fa | 2773 | unlock_system_sleep(); |
3ab83521 | 2774 | } |
7ade3fcc | 2775 | #endif |
3ab83521 HY |
2776 | |
2777 | Unlock: | |
8c5a1cf0 | 2778 | mutex_unlock(&kexec_mutex); |
3ab83521 HY |
2779 | return error; |
2780 | } |