Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / kernel / kexec.c
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
9 #define pr_fmt(fmt)     "kexec: " fmt
10
11 #include <linux/capability.h>
12 #include <linux/mm.h>
13 #include <linux/file.h>
14 #include <linux/slab.h>
15 #include <linux/fs.h>
16 #include <linux/kexec.h>
17 #include <linux/mutex.h>
18 #include <linux/list.h>
19 #include <linux/highmem.h>
20 #include <linux/syscalls.h>
21 #include <linux/reboot.h>
22 #include <linux/ioport.h>
23 #include <linux/hardirq.h>
24 #include <linux/elf.h>
25 #include <linux/elfcore.h>
26 #include <linux/utsname.h>
27 #include <linux/numa.h>
28 #include <linux/suspend.h>
29 #include <linux/device.h>
30 #include <linux/freezer.h>
31 #include <linux/pm.h>
32 #include <linux/cpu.h>
33 #include <linux/console.h>
34 #include <linux/vmalloc.h>
35 #include <linux/swap.h>
36 #include <linux/syscore_ops.h>
37 #include <linux/compiler.h>
38 #include <linux/hugetlb.h>
39
40 #include <asm/page.h>
41 #include <asm/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/sections.h>
44
45 #include <crypto/hash.h>
46 #include <crypto/sha.h>
47
48 /* Per cpu memory for storing cpu states in case of system crash. */
49 note_buf_t __percpu *crash_notes;
50
51 /* vmcoreinfo stuff */
52 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
53 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
54 size_t vmcoreinfo_size;
55 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
56
57 /* Flag to indicate we are going to kexec a new kernel */
58 bool kexec_in_progress = false;
59
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
67 #ifdef CONFIG_KEXEC_FILE
68 static int kexec_calculate_store_digests(struct kimage *image);
69 #endif
70
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 };
78 struct resource crashk_low_res = {
79         .name  = "Crash kernel",
80         .start = 0,
81         .end   = 0,
82         .flags = IORESOURCE_BUSY | IORESOURCE_MEM
83 };
84
85 int kexec_should_crash(struct task_struct *p)
86 {
87         if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
88                 return 1;
89         return 0;
90 }
91
92 /*
93  * When kexec transitions to the new kernel there is a one-to-one
94  * mapping between physical and virtual addresses.  On processors
95  * where you can disable the MMU this is trivial, and easy.  For
96  * others it is still a simple predictable page table to setup.
97  *
98  * In that environment kexec copies the new kernel to its final
99  * resting place.  This means I can only support memory whose
100  * physical address can fit in an unsigned long.  In particular
101  * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
102  * If the assembly stub has more restrictive requirements
103  * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
104  * defined more restrictively in <asm/kexec.h>.
105  *
106  * The code for the transition from the current kernel to the
107  * the new kernel is placed in the control_code_buffer, whose size
108  * is given by KEXEC_CONTROL_PAGE_SIZE.  In the best case only a single
109  * page of memory is necessary, but some architectures require more.
110  * Because this memory must be identity mapped in the transition from
111  * virtual to physical addresses it must live in the range
112  * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
113  * modifiable.
114  *
115  * The assembly stub in the control code buffer is passed a linked list
116  * of descriptor pages detailing the source pages of the new kernel,
117  * and the destination addresses of those source pages.  As this data
118  * structure is not used in the context of the current OS, it must
119  * be self-contained.
120  *
121  * The code has been made to work with highmem pages and will use a
122  * destination page in its final resting place (if it happens
123  * to allocate it).  The end product of this is that most of the
124  * physical address space, and most of RAM can be used.
125  *
126  * Future directions include:
127  *  - allocating a page table with the control code buffer identity
128  *    mapped, to simplify machine_kexec and make kexec_on_panic more
129  *    reliable.
130  */
131
132 /*
133  * KIMAGE_NO_DEST is an impossible destination address..., for
134  * allocating pages whose destination address we do not care about.
135  */
136 #define KIMAGE_NO_DEST (-1UL)
137
138 static int kimage_is_destination_range(struct kimage *image,
139                                        unsigned long start, unsigned long end);
140 static struct page *kimage_alloc_page(struct kimage *image,
141                                        gfp_t gfp_mask,
142                                        unsigned long dest);
143
144 static int copy_user_segment_list(struct kimage *image,
145                                   unsigned long nr_segments,
146                                   struct kexec_segment __user *segments)
147 {
148         int ret;
149         size_t segment_bytes;
150
151         /* Read in the segments */
152         image->nr_segments = nr_segments;
153         segment_bytes = nr_segments * sizeof(*segments);
154         ret = copy_from_user(image->segment, segments, segment_bytes);
155         if (ret)
156                 ret = -EFAULT;
157
158         return ret;
159 }
160
161 static int sanity_check_segment_list(struct kimage *image)
162 {
163         int result, i;
164         unsigned long nr_segments = image->nr_segments;
165
166         /*
167          * Verify we have good destination addresses.  The caller is
168          * responsible for making certain we don't attempt to load
169          * the new image into invalid or reserved areas of RAM.  This
170          * just verifies it is an address we can use.
171          *
172          * Since the kernel does everything in page size chunks ensure
173          * the destination addresses are page aligned.  Too many
174          * special cases crop of when we don't do this.  The most
175          * insidious is getting overlapping destination addresses
176          * simply because addresses are changed to page size
177          * granularity.
178          */
179         result = -EADDRNOTAVAIL;
180         for (i = 0; i < nr_segments; i++) {
181                 unsigned long mstart, mend;
182
183                 mstart = image->segment[i].mem;
184                 mend   = mstart + image->segment[i].memsz;
185                 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
186                         return result;
187                 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
188                         return result;
189         }
190
191         /* Verify our destination addresses do not overlap.
192          * If we alloed overlapping destination addresses
193          * through very weird things can happen with no
194          * easy explanation as one segment stops on another.
195          */
196         result = -EINVAL;
197         for (i = 0; i < nr_segments; i++) {
198                 unsigned long mstart, mend;
199                 unsigned long j;
200
201                 mstart = image->segment[i].mem;
202                 mend   = mstart + image->segment[i].memsz;
203                 for (j = 0; j < i; j++) {
204                         unsigned long pstart, pend;
205                         pstart = image->segment[j].mem;
206                         pend   = pstart + image->segment[j].memsz;
207                         /* Do the segments overlap ? */
208                         if ((mend > pstart) && (mstart < pend))
209                                 return result;
210                 }
211         }
212
213         /* Ensure our buffer sizes are strictly less than
214          * our memory sizes.  This should always be the case,
215          * and it is easier to check up front than to be surprised
216          * later on.
217          */
218         result = -EINVAL;
219         for (i = 0; i < nr_segments; i++) {
220                 if (image->segment[i].bufsz > image->segment[i].memsz)
221                         return result;
222         }
223
224         /*
225          * Verify we have good destination addresses.  Normally
226          * the caller is responsible for making certain we don't
227          * attempt to load the new image into invalid or reserved
228          * areas of RAM.  But crash kernels are preloaded into a
229          * reserved area of ram.  We must ensure the addresses
230          * are in the reserved area otherwise preloading the
231          * kernel could corrupt things.
232          */
233
234         if (image->type == KEXEC_TYPE_CRASH) {
235                 result = -EADDRNOTAVAIL;
236                 for (i = 0; i < nr_segments; i++) {
237                         unsigned long mstart, mend;
238
239                         mstart = image->segment[i].mem;
240                         mend = mstart + image->segment[i].memsz - 1;
241                         /* Ensure we are within the crash kernel limits */
242                         if ((mstart < crashk_res.start) ||
243                             (mend > crashk_res.end))
244                                 return result;
245                 }
246         }
247
248         return 0;
249 }
250
251 static struct kimage *do_kimage_alloc_init(void)
252 {
253         struct kimage *image;
254
255         /* Allocate a controlling structure */
256         image = kzalloc(sizeof(*image), GFP_KERNEL);
257         if (!image)
258                 return NULL;
259
260         image->head = 0;
261         image->entry = &image->head;
262         image->last_entry = &image->head;
263         image->control_page = ~0; /* By default this does not apply */
264         image->type = KEXEC_TYPE_DEFAULT;
265
266         /* Initialize the list of control pages */
267         INIT_LIST_HEAD(&image->control_pages);
268
269         /* Initialize the list of destination pages */
270         INIT_LIST_HEAD(&image->dest_pages);
271
272         /* Initialize the list of unusable pages */
273         INIT_LIST_HEAD(&image->unusable_pages);
274
275         return image;
276 }
277
278 static void kimage_free_page_list(struct list_head *list);
279
280 static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
281                              unsigned long nr_segments,
282                              struct kexec_segment __user *segments,
283                              unsigned long flags)
284 {
285         int ret;
286         struct kimage *image;
287         bool kexec_on_panic = flags & KEXEC_ON_CRASH;
288
289         if (kexec_on_panic) {
290                 /* Verify we have a valid entry point */
291                 if ((entry < crashk_res.start) || (entry > crashk_res.end))
292                         return -EADDRNOTAVAIL;
293         }
294
295         /* Allocate and initialize a controlling structure */
296         image = do_kimage_alloc_init();
297         if (!image)
298                 return -ENOMEM;
299
300         image->start = entry;
301
302         ret = copy_user_segment_list(image, nr_segments, segments);
303         if (ret)
304                 goto out_free_image;
305
306         ret = sanity_check_segment_list(image);
307         if (ret)
308                 goto out_free_image;
309
310          /* Enable the special crash kernel control page allocation policy. */
311         if (kexec_on_panic) {
312                 image->control_page = crashk_res.start;
313                 image->type = KEXEC_TYPE_CRASH;
314         }
315
316         /*
317          * Find a location for the control code buffer, and add it
318          * the vector of segments so that it's pages will also be
319          * counted as destination pages.
320          */
321         ret = -ENOMEM;
322         image->control_code_page = kimage_alloc_control_pages(image,
323                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
324         if (!image->control_code_page) {
325                 pr_err("Could not allocate control_code_buffer\n");
326                 goto out_free_image;
327         }
328
329         if (!kexec_on_panic) {
330                 image->swap_page = kimage_alloc_control_pages(image, 0);
331                 if (!image->swap_page) {
332                         pr_err("Could not allocate swap buffer\n");
333                         goto out_free_control_pages;
334                 }
335         }
336
337         *rimage = image;
338         return 0;
339 out_free_control_pages:
340         kimage_free_page_list(&image->control_pages);
341 out_free_image:
342         kfree(image);
343         return ret;
344 }
345
346 #ifdef CONFIG_KEXEC_FILE
347 static int copy_file_from_fd(int fd, void **buf, unsigned long *buf_len)
348 {
349         struct fd f = fdget(fd);
350         int ret;
351         struct kstat stat;
352         loff_t pos;
353         ssize_t bytes = 0;
354
355         if (!f.file)
356                 return -EBADF;
357
358         ret = vfs_getattr(&f.file->f_path, &stat);
359         if (ret)
360                 goto out;
361
362         if (stat.size > INT_MAX) {
363                 ret = -EFBIG;
364                 goto out;
365         }
366
367         /* Don't hand 0 to vmalloc, it whines. */
368         if (stat.size == 0) {
369                 ret = -EINVAL;
370                 goto out;
371         }
372
373         *buf = vmalloc(stat.size);
374         if (!*buf) {
375                 ret = -ENOMEM;
376                 goto out;
377         }
378
379         pos = 0;
380         while (pos < stat.size) {
381                 bytes = kernel_read(f.file, pos, (char *)(*buf) + pos,
382                                     stat.size - pos);
383                 if (bytes < 0) {
384                         vfree(*buf);
385                         ret = bytes;
386                         goto out;
387                 }
388
389                 if (bytes == 0)
390                         break;
391                 pos += bytes;
392         }
393
394         if (pos != stat.size) {
395                 ret = -EBADF;
396                 vfree(*buf);
397                 goto out;
398         }
399
400         *buf_len = pos;
401 out:
402         fdput(f);
403         return ret;
404 }
405
406 /* Architectures can provide this probe function */
407 int __weak arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
408                                          unsigned long buf_len)
409 {
410         return -ENOEXEC;
411 }
412
413 void * __weak arch_kexec_kernel_image_load(struct kimage *image)
414 {
415         return ERR_PTR(-ENOEXEC);
416 }
417
418 void __weak arch_kimage_file_post_load_cleanup(struct kimage *image)
419 {
420 }
421
422 int __weak arch_kexec_kernel_verify_sig(struct kimage *image, void *buf,
423                                         unsigned long buf_len)
424 {
425         return -EKEYREJECTED;
426 }
427
428 /* Apply relocations of type RELA */
429 int __weak
430 arch_kexec_apply_relocations_add(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
431                                  unsigned int relsec)
432 {
433         pr_err("RELA relocation unsupported.\n");
434         return -ENOEXEC;
435 }
436
437 /* Apply relocations of type REL */
438 int __weak
439 arch_kexec_apply_relocations(const Elf_Ehdr *ehdr, Elf_Shdr *sechdrs,
440                              unsigned int relsec)
441 {
442         pr_err("REL relocation unsupported.\n");
443         return -ENOEXEC;
444 }
445
446 /*
447  * Free up memory used by kernel, initrd, and command line. This is temporary
448  * memory allocation which is not needed any more after these buffers have
449  * been loaded into separate segments and have been copied elsewhere.
450  */
451 static void kimage_file_post_load_cleanup(struct kimage *image)
452 {
453         struct purgatory_info *pi = &image->purgatory_info;
454
455         vfree(image->kernel_buf);
456         image->kernel_buf = NULL;
457
458         vfree(image->initrd_buf);
459         image->initrd_buf = NULL;
460
461         kfree(image->cmdline_buf);
462         image->cmdline_buf = NULL;
463
464         vfree(pi->purgatory_buf);
465         pi->purgatory_buf = NULL;
466
467         vfree(pi->sechdrs);
468         pi->sechdrs = NULL;
469
470         /* See if architecture has anything to cleanup post load */
471         arch_kimage_file_post_load_cleanup(image);
472
473         /*
474          * Above call should have called into bootloader to free up
475          * any data stored in kimage->image_loader_data. It should
476          * be ok now to free it up.
477          */
478         kfree(image->image_loader_data);
479         image->image_loader_data = NULL;
480 }
481
482 /*
483  * In file mode list of segments is prepared by kernel. Copy relevant
484  * data from user space, do error checking, prepare segment list
485  */
486 static int
487 kimage_file_prepare_segments(struct kimage *image, int kernel_fd, int initrd_fd,
488                              const char __user *cmdline_ptr,
489                              unsigned long cmdline_len, unsigned flags)
490 {
491         int ret = 0;
492         void *ldata;
493
494         ret = copy_file_from_fd(kernel_fd, &image->kernel_buf,
495                                 &image->kernel_buf_len);
496         if (ret)
497                 return ret;
498
499         /* Call arch image probe handlers */
500         ret = arch_kexec_kernel_image_probe(image, image->kernel_buf,
501                                             image->kernel_buf_len);
502
503         if (ret)
504                 goto out;
505
506 #ifdef CONFIG_KEXEC_VERIFY_SIG
507         ret = arch_kexec_kernel_verify_sig(image, image->kernel_buf,
508                                            image->kernel_buf_len);
509         if (ret) {
510                 pr_debug("kernel signature verification failed.\n");
511                 goto out;
512         }
513         pr_debug("kernel signature verification successful.\n");
514 #endif
515         /* It is possible that there no initramfs is being loaded */
516         if (!(flags & KEXEC_FILE_NO_INITRAMFS)) {
517                 ret = copy_file_from_fd(initrd_fd, &image->initrd_buf,
518                                         &image->initrd_buf_len);
519                 if (ret)
520                         goto out;
521         }
522
523         if (cmdline_len) {
524                 image->cmdline_buf = kzalloc(cmdline_len, GFP_KERNEL);
525                 if (!image->cmdline_buf) {
526                         ret = -ENOMEM;
527                         goto out;
528                 }
529
530                 ret = copy_from_user(image->cmdline_buf, cmdline_ptr,
531                                      cmdline_len);
532                 if (ret) {
533                         ret = -EFAULT;
534                         goto out;
535                 }
536
537                 image->cmdline_buf_len = cmdline_len;
538
539                 /* command line should be a string with last byte null */
540                 if (image->cmdline_buf[cmdline_len - 1] != '\0') {
541                         ret = -EINVAL;
542                         goto out;
543                 }
544         }
545
546         /* Call arch image load handlers */
547         ldata = arch_kexec_kernel_image_load(image);
548
549         if (IS_ERR(ldata)) {
550                 ret = PTR_ERR(ldata);
551                 goto out;
552         }
553
554         image->image_loader_data = ldata;
555 out:
556         /* In case of error, free up all allocated memory in this function */
557         if (ret)
558                 kimage_file_post_load_cleanup(image);
559         return ret;
560 }
561
562 static int
563 kimage_file_alloc_init(struct kimage **rimage, int kernel_fd,
564                        int initrd_fd, const char __user *cmdline_ptr,
565                        unsigned long cmdline_len, unsigned long flags)
566 {
567         int ret;
568         struct kimage *image;
569         bool kexec_on_panic = flags & KEXEC_FILE_ON_CRASH;
570
571         image = do_kimage_alloc_init();
572         if (!image)
573                 return -ENOMEM;
574
575         image->file_mode = 1;
576
577         if (kexec_on_panic) {
578                 /* Enable special crash kernel control page alloc policy. */
579                 image->control_page = crashk_res.start;
580                 image->type = KEXEC_TYPE_CRASH;
581         }
582
583         ret = kimage_file_prepare_segments(image, kernel_fd, initrd_fd,
584                                            cmdline_ptr, cmdline_len, flags);
585         if (ret)
586                 goto out_free_image;
587
588         ret = sanity_check_segment_list(image);
589         if (ret)
590                 goto out_free_post_load_bufs;
591
592         ret = -ENOMEM;
593         image->control_code_page = kimage_alloc_control_pages(image,
594                                            get_order(KEXEC_CONTROL_PAGE_SIZE));
595         if (!image->control_code_page) {
596                 pr_err("Could not allocate control_code_buffer\n");
597                 goto out_free_post_load_bufs;
598         }
599
600         if (!kexec_on_panic) {
601                 image->swap_page = kimage_alloc_control_pages(image, 0);
602                 if (!image->swap_page) {
603                         pr_err("Could not allocate swap buffer\n");
604                         goto out_free_control_pages;
605                 }
606         }
607
608         *rimage = image;
609         return 0;
610 out_free_control_pages:
611         kimage_free_page_list(&image->control_pages);
612 out_free_post_load_bufs:
613         kimage_file_post_load_cleanup(image);
614 out_free_image:
615         kfree(image);
616         return ret;
617 }
618 #else /* CONFIG_KEXEC_FILE */
619 static inline void kimage_file_post_load_cleanup(struct kimage *image) { }
620 #endif /* CONFIG_KEXEC_FILE */
621
622 static int kimage_is_destination_range(struct kimage *image,
623                                         unsigned long start,
624                                         unsigned long end)
625 {
626         unsigned long i;
627
628         for (i = 0; i < image->nr_segments; i++) {
629                 unsigned long mstart, mend;
630
631                 mstart = image->segment[i].mem;
632                 mend = mstart + image->segment[i].memsz;
633                 if ((end > mstart) && (start < mend))
634                         return 1;
635         }
636
637         return 0;
638 }
639
640 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
641 {
642         struct page *pages;
643
644         pages = alloc_pages(gfp_mask, order);
645         if (pages) {
646                 unsigned int count, i;
647                 pages->mapping = NULL;
648                 set_page_private(pages, order);
649                 count = 1 << order;
650                 for (i = 0; i < count; i++)
651                         SetPageReserved(pages + i);
652         }
653
654         return pages;
655 }
656
657 static void kimage_free_pages(struct page *page)
658 {
659         unsigned int order, count, i;
660
661         order = page_private(page);
662         count = 1 << order;
663         for (i = 0; i < count; i++)
664                 ClearPageReserved(page + i);
665         __free_pages(page, order);
666 }
667
668 static void kimage_free_page_list(struct list_head *list)
669 {
670         struct list_head *pos, *next;
671
672         list_for_each_safe(pos, next, list) {
673                 struct page *page;
674
675                 page = list_entry(pos, struct page, lru);
676                 list_del(&page->lru);
677                 kimage_free_pages(page);
678         }
679 }
680
681 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
682                                                         unsigned int order)
683 {
684         /* Control pages are special, they are the intermediaries
685          * that are needed while we copy the rest of the pages
686          * to their final resting place.  As such they must
687          * not conflict with either the destination addresses
688          * or memory the kernel is already using.
689          *
690          * The only case where we really need more than one of
691          * these are for architectures where we cannot disable
692          * the MMU and must instead generate an identity mapped
693          * page table for all of the memory.
694          *
695          * At worst this runs in O(N) of the image size.
696          */
697         struct list_head extra_pages;
698         struct page *pages;
699         unsigned int count;
700
701         count = 1 << order;
702         INIT_LIST_HEAD(&extra_pages);
703
704         /* Loop while I can allocate a page and the page allocated
705          * is a destination page.
706          */
707         do {
708                 unsigned long pfn, epfn, addr, eaddr;
709
710                 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
711                 if (!pages)
712                         break;
713                 pfn   = page_to_pfn(pages);
714                 epfn  = pfn + count;
715                 addr  = pfn << PAGE_SHIFT;
716                 eaddr = epfn << PAGE_SHIFT;
717                 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
718                               kimage_is_destination_range(image, addr, eaddr)) {
719                         list_add(&pages->lru, &extra_pages);
720                         pages = NULL;
721                 }
722         } while (!pages);
723
724         if (pages) {
725                 /* Remember the allocated page... */
726                 list_add(&pages->lru, &image->control_pages);
727
728                 /* Because the page is already in it's destination
729                  * location we will never allocate another page at
730                  * that address.  Therefore kimage_alloc_pages
731                  * will not return it (again) and we don't need
732                  * to give it an entry in image->segment[].
733                  */
734         }
735         /* Deal with the destination pages I have inadvertently allocated.
736          *
737          * Ideally I would convert multi-page allocations into single
738          * page allocations, and add everything to image->dest_pages.
739          *
740          * For now it is simpler to just free the pages.
741          */
742         kimage_free_page_list(&extra_pages);
743
744         return pages;
745 }
746
747 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
748                                                       unsigned int order)
749 {
750         /* Control pages are special, they are the intermediaries
751          * that are needed while we copy the rest of the pages
752          * to their final resting place.  As such they must
753          * not conflict with either the destination addresses
754          * or memory the kernel is already using.
755          *
756          * Control pages are also the only pags we must allocate
757          * when loading a crash kernel.  All of the other pages
758          * are specified by the segments and we just memcpy
759          * into them directly.
760          *
761          * The only case where we really need more than one of
762          * these are for architectures where we cannot disable
763          * the MMU and must instead generate an identity mapped
764          * page table for all of the memory.
765          *
766          * Given the low demand this implements a very simple
767          * allocator that finds the first hole of the appropriate
768          * size in the reserved memory region, and allocates all
769          * of the memory up to and including the hole.
770          */
771         unsigned long hole_start, hole_end, size;
772         struct page *pages;
773
774         pages = NULL;
775         size = (1 << order) << PAGE_SHIFT;
776         hole_start = (image->control_page + (size - 1)) & ~(size - 1);
777         hole_end   = hole_start + size - 1;
778         while (hole_end <= crashk_res.end) {
779                 unsigned long i;
780
781                 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
782                         break;
783                 /* See if I overlap any of the segments */
784                 for (i = 0; i < image->nr_segments; i++) {
785                         unsigned long mstart, mend;
786
787                         mstart = image->segment[i].mem;
788                         mend   = mstart + image->segment[i].memsz - 1;
789                         if ((hole_end >= mstart) && (hole_start <= mend)) {
790                                 /* Advance the hole to the end of the segment */
791                                 hole_start = (mend + (size - 1)) & ~(size - 1);
792                                 hole_end   = hole_start + size - 1;
793                                 break;
794                         }
795                 }
796                 /* If I don't overlap any segments I have found my hole! */
797                 if (i == image->nr_segments) {
798                         pages = pfn_to_page(hole_start >> PAGE_SHIFT);
799                         break;
800                 }
801         }
802         if (pages)
803                 image->control_page = hole_end;
804
805         return pages;
806 }
807
808
809 struct page *kimage_alloc_control_pages(struct kimage *image,
810                                          unsigned int order)
811 {
812         struct page *pages = NULL;
813
814         switch (image->type) {
815         case KEXEC_TYPE_DEFAULT:
816                 pages = kimage_alloc_normal_control_pages(image, order);
817                 break;
818         case KEXEC_TYPE_CRASH:
819                 pages = kimage_alloc_crash_control_pages(image, order);
820                 break;
821         }
822
823         return pages;
824 }
825
826 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
827 {
828         if (*image->entry != 0)
829                 image->entry++;
830
831         if (image->entry == image->last_entry) {
832                 kimage_entry_t *ind_page;
833                 struct page *page;
834
835                 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
836                 if (!page)
837                         return -ENOMEM;
838
839                 ind_page = page_address(page);
840                 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
841                 image->entry = ind_page;
842                 image->last_entry = ind_page +
843                                       ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
844         }
845         *image->entry = entry;
846         image->entry++;
847         *image->entry = 0;
848
849         return 0;
850 }
851
852 static int kimage_set_destination(struct kimage *image,
853                                    unsigned long destination)
854 {
855         int result;
856
857         destination &= PAGE_MASK;
858         result = kimage_add_entry(image, destination | IND_DESTINATION);
859
860         return result;
861 }
862
863
864 static int kimage_add_page(struct kimage *image, unsigned long page)
865 {
866         int result;
867
868         page &= PAGE_MASK;
869         result = kimage_add_entry(image, page | IND_SOURCE);
870
871         return result;
872 }
873
874
875 static void kimage_free_extra_pages(struct kimage *image)
876 {
877         /* Walk through and free any extra destination pages I may have */
878         kimage_free_page_list(&image->dest_pages);
879
880         /* Walk through and free any unusable pages I have cached */
881         kimage_free_page_list(&image->unusable_pages);
882
883 }
884 static void kimage_terminate(struct kimage *image)
885 {
886         if (*image->entry != 0)
887                 image->entry++;
888
889         *image->entry = IND_DONE;
890 }
891
892 #define for_each_kimage_entry(image, ptr, entry) \
893         for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
894                 ptr = (entry & IND_INDIRECTION) ? \
895                         phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
896
897 static void kimage_free_entry(kimage_entry_t entry)
898 {
899         struct page *page;
900
901         page = pfn_to_page(entry >> PAGE_SHIFT);
902         kimage_free_pages(page);
903 }
904
905 static void kimage_free(struct kimage *image)
906 {
907         kimage_entry_t *ptr, entry;
908         kimage_entry_t ind = 0;
909
910         if (!image)
911                 return;
912
913         kimage_free_extra_pages(image);
914         for_each_kimage_entry(image, ptr, entry) {
915                 if (entry & IND_INDIRECTION) {
916                         /* Free the previous indirection page */
917                         if (ind & IND_INDIRECTION)
918                                 kimage_free_entry(ind);
919                         /* Save this indirection page until we are
920                          * done with it.
921                          */
922                         ind = entry;
923                 } else if (entry & IND_SOURCE)
924                         kimage_free_entry(entry);
925         }
926         /* Free the final indirection page */
927         if (ind & IND_INDIRECTION)
928                 kimage_free_entry(ind);
929
930         /* Handle any machine specific cleanup */
931         machine_kexec_cleanup(image);
932
933         /* Free the kexec control pages... */
934         kimage_free_page_list(&image->control_pages);
935
936         /*
937          * Free up any temporary buffers allocated. This might hit if
938          * error occurred much later after buffer allocation.
939          */
940         if (image->file_mode)
941                 kimage_file_post_load_cleanup(image);
942
943         kfree(image);
944 }
945
946 static kimage_entry_t *kimage_dst_used(struct kimage *image,
947                                         unsigned long page)
948 {
949         kimage_entry_t *ptr, entry;
950         unsigned long destination = 0;
951
952         for_each_kimage_entry(image, ptr, entry) {
953                 if (entry & IND_DESTINATION)
954                         destination = entry & PAGE_MASK;
955                 else if (entry & IND_SOURCE) {
956                         if (page == destination)
957                                 return ptr;
958                         destination += PAGE_SIZE;
959                 }
960         }
961
962         return NULL;
963 }
964
965 static struct page *kimage_alloc_page(struct kimage *image,
966                                         gfp_t gfp_mask,
967                                         unsigned long destination)
968 {
969         /*
970          * Here we implement safeguards to ensure that a source page
971          * is not copied to its destination page before the data on
972          * the destination page is no longer useful.
973          *
974          * To do this we maintain the invariant that a source page is
975          * either its own destination page, or it is not a
976          * destination page at all.
977          *
978          * That is slightly stronger than required, but the proof
979          * that no problems will not occur is trivial, and the
980          * implementation is simply to verify.
981          *
982          * When allocating all pages normally this algorithm will run
983          * in O(N) time, but in the worst case it will run in O(N^2)
984          * time.   If the runtime is a problem the data structures can
985          * be fixed.
986          */
987         struct page *page;
988         unsigned long addr;
989
990         /*
991          * Walk through the list of destination pages, and see if I
992          * have a match.
993          */
994         list_for_each_entry(page, &image->dest_pages, lru) {
995                 addr = page_to_pfn(page) << PAGE_SHIFT;
996                 if (addr == destination) {
997                         list_del(&page->lru);
998                         return page;
999                 }
1000         }
1001         page = NULL;
1002         while (1) {
1003                 kimage_entry_t *old;
1004
1005                 /* Allocate a page, if we run out of memory give up */
1006                 page = kimage_alloc_pages(gfp_mask, 0);
1007                 if (!page)
1008                         return NULL;
1009                 /* If the page cannot be used file it away */
1010                 if (page_to_pfn(page) >
1011                                 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
1012                         list_add(&page->lru, &image->unusable_pages);
1013                         continue;
1014                 }
1015                 addr = page_to_pfn(page) << PAGE_SHIFT;
1016
1017                 /* If it is the destination page we want use it */
1018                 if (addr == destination)
1019                         break;
1020
1021                 /* If the page is not a destination page use it */
1022                 if (!kimage_is_destination_range(image, addr,
1023                                                   addr + PAGE_SIZE))
1024                         break;
1025
1026                 /*
1027                  * I know that the page is someones destination page.
1028                  * See if there is already a source page for this
1029                  * destination page.  And if so swap the source pages.
1030                  */
1031                 old = kimage_dst_used(image, addr);
1032                 if (old) {
1033                         /* If so move it */
1034                         unsigned long old_addr;
1035                         struct page *old_page;
1036
1037                         old_addr = *old & PAGE_MASK;
1038                         old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
1039                         copy_highpage(page, old_page);
1040                         *old = addr | (*old & ~PAGE_MASK);
1041
1042                         /* The old page I have found cannot be a
1043                          * destination page, so return it if it's
1044                          * gfp_flags honor the ones passed in.
1045                          */
1046                         if (!(gfp_mask & __GFP_HIGHMEM) &&
1047                             PageHighMem(old_page)) {
1048                                 kimage_free_pages(old_page);
1049                                 continue;
1050                         }
1051                         addr = old_addr;
1052                         page = old_page;
1053                         break;
1054                 } else {
1055                         /* Place the page on the destination list I
1056                          * will use it later.
1057                          */
1058                         list_add(&page->lru, &image->dest_pages);
1059                 }
1060         }
1061
1062         return page;
1063 }
1064
1065 static int kimage_load_normal_segment(struct kimage *image,
1066                                          struct kexec_segment *segment)
1067 {
1068         unsigned long maddr;
1069         size_t ubytes, mbytes;
1070         int result;
1071         unsigned char __user *buf = NULL;
1072         unsigned char *kbuf = NULL;
1073
1074         result = 0;
1075         if (image->file_mode)
1076                 kbuf = segment->kbuf;
1077         else
1078                 buf = segment->buf;
1079         ubytes = segment->bufsz;
1080         mbytes = segment->memsz;
1081         maddr = segment->mem;
1082
1083         result = kimage_set_destination(image, maddr);
1084         if (result < 0)
1085                 goto out;
1086
1087         while (mbytes) {
1088                 struct page *page;
1089                 char *ptr;
1090                 size_t uchunk, mchunk;
1091
1092                 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
1093                 if (!page) {
1094                         result  = -ENOMEM;
1095                         goto out;
1096                 }
1097                 result = kimage_add_page(image, page_to_pfn(page)
1098                                                                 << PAGE_SHIFT);
1099                 if (result < 0)
1100                         goto out;
1101
1102                 ptr = kmap(page);
1103                 /* Start with a clear page */
1104                 clear_page(ptr);
1105                 ptr += maddr & ~PAGE_MASK;
1106                 mchunk = min_t(size_t, mbytes,
1107                                 PAGE_SIZE - (maddr & ~PAGE_MASK));
1108                 uchunk = min(ubytes, mchunk);
1109
1110                 /* For file based kexec, source pages are in kernel memory */
1111                 if (image->file_mode)
1112                         memcpy(ptr, kbuf, uchunk);
1113                 else
1114                         result = copy_from_user(ptr, buf, uchunk);
1115                 kunmap(page);
1116                 if (result) {
1117                         result = -EFAULT;
1118                         goto out;
1119                 }
1120                 ubytes -= uchunk;
1121                 maddr  += mchunk;
1122                 if (image->file_mode)
1123                         kbuf += mchunk;
1124                 else
1125                         buf += mchunk;
1126                 mbytes -= mchunk;
1127         }
1128 out:
1129         return result;
1130 }
1131
1132 static int kimage_load_crash_segment(struct kimage *image,
1133                                         struct kexec_segment *segment)
1134 {
1135         /* For crash dumps kernels we simply copy the data from
1136          * user space to it's destination.
1137          * We do things a page at a time for the sake of kmap.
1138          */
1139         unsigned long maddr;
1140         size_t ubytes, mbytes;
1141         int result;
1142         unsigned char __user *buf = NULL;
1143         unsigned char *kbuf = NULL;
1144
1145         result = 0;
1146         if (image->file_mode)
1147                 kbuf = segment->kbuf;
1148         else
1149                 buf = segment->buf;
1150         ubytes = segment->bufsz;
1151         mbytes = segment->memsz;
1152         maddr = segment->mem;
1153         while (mbytes) {
1154                 struct page *page;
1155                 char *ptr;
1156                 size_t uchunk, mchunk;
1157
1158                 page = pfn_to_page(maddr >> PAGE_SHIFT);
1159                 if (!page) {
1160                         result  = -ENOMEM;
1161                         goto out;
1162                 }
1163                 ptr = kmap(page);
1164                 ptr += maddr & ~PAGE_MASK;
1165                 mchunk = min_t(size_t, mbytes,
1166                                 PAGE_SIZE - (maddr & ~PAGE_MASK));
1167                 uchunk = min(ubytes, mchunk);
1168                 if (mchunk > uchunk) {
1169                         /* Zero the trailing part of the page */
1170                         memset(ptr + uchunk, 0, mchunk - uchunk);
1171                 }
1172
1173                 /* For file based kexec, source pages are in kernel memory */
1174                 if (image->file_mode)
1175                         memcpy(ptr, kbuf, uchunk);
1176                 else
1177                         result = copy_from_user(ptr, buf, uchunk);
1178                 kexec_flush_icache_page(page);
1179                 kunmap(page);
1180                 if (result) {
1181                         result = -EFAULT;
1182                         goto out;
1183                 }
1184                 ubytes -= uchunk;
1185                 maddr  += mchunk;
1186                 if (image->file_mode)
1187                         kbuf += mchunk;
1188                 else
1189                         buf += mchunk;
1190                 mbytes -= mchunk;
1191         }
1192 out:
1193         return result;
1194 }
1195
1196 static int kimage_load_segment(struct kimage *image,
1197                                 struct kexec_segment *segment)
1198 {
1199         int result = -ENOMEM;
1200
1201         switch (image->type) {
1202         case KEXEC_TYPE_DEFAULT:
1203                 result = kimage_load_normal_segment(image, segment);
1204                 break;
1205         case KEXEC_TYPE_CRASH:
1206                 result = kimage_load_crash_segment(image, segment);
1207                 break;
1208         }
1209
1210         return result;
1211 }
1212
1213 /*
1214  * Exec Kernel system call: for obvious reasons only root may call it.
1215  *
1216  * This call breaks up into three pieces.
1217  * - A generic part which loads the new kernel from the current
1218  *   address space, and very carefully places the data in the
1219  *   allocated pages.
1220  *
1221  * - A generic part that interacts with the kernel and tells all of
1222  *   the devices to shut down.  Preventing on-going dmas, and placing
1223  *   the devices in a consistent state so a later kernel can
1224  *   reinitialize them.
1225  *
1226  * - A machine specific part that includes the syscall number
1227  *   and then copies the image to it's final destination.  And
1228  *   jumps into the image at entry.
1229  *
1230  * kexec does not sync, or unmount filesystems so if you need
1231  * that to happen you need to do that yourself.
1232  */
1233 struct kimage *kexec_image;
1234 struct kimage *kexec_crash_image;
1235 int kexec_load_disabled;
1236
1237 static DEFINE_MUTEX(kexec_mutex);
1238
1239 SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
1240                 struct kexec_segment __user *, segments, unsigned long, flags)
1241 {
1242         struct kimage **dest_image, *image;
1243         int result;
1244
1245         /* We only trust the superuser with rebooting the system. */
1246         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1247                 return -EPERM;
1248
1249         /*
1250          * Verify we have a legal set of flags
1251          * This leaves us room for future extensions.
1252          */
1253         if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
1254                 return -EINVAL;
1255
1256         /* Verify we are on the appropriate architecture */
1257         if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
1258                 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
1259                 return -EINVAL;
1260
1261         /* Put an artificial cap on the number
1262          * of segments passed to kexec_load.
1263          */
1264         if (nr_segments > KEXEC_SEGMENT_MAX)
1265                 return -EINVAL;
1266
1267         image = NULL;
1268         result = 0;
1269
1270         /* Because we write directly to the reserved memory
1271          * region when loading crash kernels we need a mutex here to
1272          * prevent multiple crash  kernels from attempting to load
1273          * simultaneously, and to prevent a crash kernel from loading
1274          * over the top of a in use crash kernel.
1275          *
1276          * KISS: always take the mutex.
1277          */
1278         if (!mutex_trylock(&kexec_mutex))
1279                 return -EBUSY;
1280
1281         dest_image = &kexec_image;
1282         if (flags & KEXEC_ON_CRASH)
1283                 dest_image = &kexec_crash_image;
1284         if (nr_segments > 0) {
1285                 unsigned long i;
1286
1287                 if (flags & KEXEC_ON_CRASH) {
1288                         /*
1289                          * Loading another kernel to switch to if this one
1290                          * crashes.  Free any current crash dump kernel before
1291                          * we corrupt it.
1292                          */
1293
1294                         kimage_free(xchg(&kexec_crash_image, NULL));
1295                         result = kimage_alloc_init(&image, entry, nr_segments,
1296                                                    segments, flags);
1297                         crash_map_reserved_pages();
1298                 } else {
1299                         /* Loading another kernel to reboot into. */
1300
1301                         result = kimage_alloc_init(&image, entry, nr_segments,
1302                                                    segments, flags);
1303                 }
1304                 if (result)
1305                         goto out;
1306
1307                 if (flags & KEXEC_PRESERVE_CONTEXT)
1308                         image->preserve_context = 1;
1309                 result = machine_kexec_prepare(image);
1310                 if (result)
1311                         goto out;
1312
1313                 for (i = 0; i < nr_segments; i++) {
1314                         result = kimage_load_segment(image, &image->segment[i]);
1315                         if (result)
1316                                 goto out;
1317                 }
1318                 kimage_terminate(image);
1319                 if (flags & KEXEC_ON_CRASH)
1320                         crash_unmap_reserved_pages();
1321         }
1322         /* Install the new kernel, and  Uninstall the old */
1323         image = xchg(dest_image, image);
1324
1325 out:
1326         mutex_unlock(&kexec_mutex);
1327         kimage_free(image);
1328
1329         return result;
1330 }
1331
1332 /*
1333  * Add and remove page tables for crashkernel memory
1334  *
1335  * Provide an empty default implementation here -- architecture
1336  * code may override this
1337  */
1338 void __weak crash_map_reserved_pages(void)
1339 {}
1340
1341 void __weak crash_unmap_reserved_pages(void)
1342 {}
1343
1344 #ifdef CONFIG_COMPAT
1345 COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1346                        compat_ulong_t, nr_segments,
1347                        struct compat_kexec_segment __user *, segments,
1348                        compat_ulong_t, flags)
1349 {
1350         struct compat_kexec_segment in;
1351         struct kexec_segment out, __user *ksegments;
1352         unsigned long i, result;
1353
1354         /* Don't allow clients that don't understand the native
1355          * architecture to do anything.
1356          */
1357         if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1358                 return -EINVAL;
1359
1360         if (nr_segments > KEXEC_SEGMENT_MAX)
1361                 return -EINVAL;
1362
1363         ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1364         for (i = 0; i < nr_segments; i++) {
1365                 result = copy_from_user(&in, &segments[i], sizeof(in));
1366                 if (result)
1367                         return -EFAULT;
1368
1369                 out.buf   = compat_ptr(in.buf);
1370                 out.bufsz = in.bufsz;
1371                 out.mem   = in.mem;
1372                 out.memsz = in.memsz;
1373
1374                 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1375                 if (result)
1376                         return -EFAULT;
1377         }
1378
1379         return sys_kexec_load(entry, nr_segments, ksegments, flags);
1380 }
1381 #endif
1382
1383 #ifdef CONFIG_KEXEC_FILE
1384 SYSCALL_DEFINE5(kexec_file_load, int, kernel_fd, int, initrd_fd,
1385                 unsigned long, cmdline_len, const char __user *, cmdline_ptr,
1386                 unsigned long, flags)
1387 {
1388         int ret = 0, i;
1389         struct kimage **dest_image, *image;
1390
1391         /* We only trust the superuser with rebooting the system. */
1392         if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
1393                 return -EPERM;
1394
1395         /* Make sure we have a legal set of flags */
1396         if (flags != (flags & KEXEC_FILE_FLAGS))
1397                 return -EINVAL;
1398
1399         image = NULL;
1400
1401         if (!mutex_trylock(&kexec_mutex))
1402                 return -EBUSY;
1403
1404         dest_image = &kexec_image;
1405         if (flags & KEXEC_FILE_ON_CRASH)
1406                 dest_image = &kexec_crash_image;
1407
1408         if (flags & KEXEC_FILE_UNLOAD)
1409                 goto exchange;
1410
1411         /*
1412          * In case of crash, new kernel gets loaded in reserved region. It is
1413          * same memory where old crash kernel might be loaded. Free any
1414          * current crash dump kernel before we corrupt it.
1415          */
1416         if (flags & KEXEC_FILE_ON_CRASH)
1417                 kimage_free(xchg(&kexec_crash_image, NULL));
1418
1419         ret = kimage_file_alloc_init(&image, kernel_fd, initrd_fd, cmdline_ptr,
1420                                      cmdline_len, flags);
1421         if (ret)
1422                 goto out;
1423
1424         ret = machine_kexec_prepare(image);
1425         if (ret)
1426                 goto out;
1427
1428         ret = kexec_calculate_store_digests(image);
1429         if (ret)
1430                 goto out;
1431
1432         for (i = 0; i < image->nr_segments; i++) {
1433                 struct kexec_segment *ksegment;
1434
1435                 ksegment = &image->segment[i];
1436                 pr_debug("Loading segment %d: buf=0x%p bufsz=0x%zx mem=0x%lx memsz=0x%zx\n",
1437                          i, ksegment->buf, ksegment->bufsz, ksegment->mem,
1438                          ksegment->memsz);
1439
1440                 ret = kimage_load_segment(image, &image->segment[i]);
1441                 if (ret)
1442                         goto out;
1443         }
1444
1445         kimage_terminate(image);
1446
1447         /*
1448          * Free up any temporary buffers allocated which are not needed
1449          * after image has been loaded
1450          */
1451         kimage_file_post_load_cleanup(image);
1452 exchange:
1453         image = xchg(dest_image, image);
1454 out:
1455         mutex_unlock(&kexec_mutex);
1456         kimage_free(image);
1457         return ret;
1458 }
1459
1460 #endif /* CONFIG_KEXEC_FILE */
1461
1462 void crash_kexec(struct pt_regs *regs)
1463 {
1464         /* Take the kexec_mutex here to prevent sys_kexec_load
1465          * running on one cpu from replacing the crash kernel
1466          * we are using after a panic on a different cpu.
1467          *
1468          * If the crash kernel was not located in a fixed area
1469          * of memory the xchg(&kexec_crash_image) would be
1470          * sufficient.  But since I reuse the memory...
1471          */
1472         if (mutex_trylock(&kexec_mutex)) {
1473                 if (kexec_crash_image) {
1474                         struct pt_regs fixed_regs;
1475
1476                         crash_setup_regs(&fixed_regs, regs);
1477                         crash_save_vmcoreinfo();
1478                         machine_crash_shutdown(&fixed_regs);
1479                         machine_kexec(kexec_crash_image);
1480                 }
1481                 mutex_unlock(&kexec_mutex);
1482         }
1483 }
1484
1485 size_t crash_get_memory_size(void)
1486 {
1487         size_t size = 0;
1488         mutex_lock(&kexec_mutex);
1489         if (crashk_res.end != crashk_res.start)
1490                 size = resource_size(&crashk_res);
1491         mutex_unlock(&kexec_mutex);
1492         return size;
1493 }
1494
1495 void __weak crash_free_reserved_phys_range(unsigned long begin,
1496                                            unsigned long end)
1497 {
1498         unsigned long addr;
1499
1500         for (addr = begin; addr < end; addr += PAGE_SIZE)
1501                 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1502 }
1503
1504 int crash_shrink_memory(unsigned long new_size)
1505 {
1506         int ret = 0;
1507         unsigned long start, end;
1508         unsigned long old_size;
1509         struct resource *ram_res;
1510
1511         mutex_lock(&kexec_mutex);
1512
1513         if (kexec_crash_image) {
1514                 ret = -ENOENT;
1515                 goto unlock;
1516         }
1517         start = crashk_res.start;
1518         end = crashk_res.end;
1519         old_size = (end == 0) ? 0 : end - start + 1;
1520         if (new_size >= old_size) {
1521                 ret = (new_size == old_size) ? 0 : -EINVAL;
1522                 goto unlock;
1523         }
1524
1525         ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1526         if (!ram_res) {
1527                 ret = -ENOMEM;
1528                 goto unlock;
1529         }
1530
1531         start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1532         end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1533
1534         crash_map_reserved_pages();
1535         crash_free_reserved_phys_range(end, crashk_res.end);
1536
1537         if ((start == end) && (crashk_res.parent != NULL))
1538                 release_resource(&crashk_res);
1539
1540         ram_res->start = end;
1541         ram_res->end = crashk_res.end;
1542         ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1543         ram_res->name = "System RAM";
1544
1545         crashk_res.end = end - 1;
1546
1547         insert_resource(&iomem_resource, ram_res);
1548         crash_unmap_reserved_pages();
1549
1550 unlock:
1551         mutex_unlock(&kexec_mutex);
1552         return ret;
1553 }
1554
1555 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1556                             size_t data_len)
1557 {
1558         struct elf_note note;
1559
1560         note.n_namesz = strlen(name) + 1;
1561         note.n_descsz = data_len;
1562         note.n_type   = type;
1563         memcpy(buf, &note, sizeof(note));
1564         buf += (sizeof(note) + 3)/4;
1565         memcpy(buf, name, note.n_namesz);
1566         buf += (note.n_namesz + 3)/4;
1567         memcpy(buf, data, note.n_descsz);
1568         buf += (note.n_descsz + 3)/4;
1569
1570         return buf;
1571 }
1572
1573 static void final_note(u32 *buf)
1574 {
1575         struct elf_note note;
1576
1577         note.n_namesz = 0;
1578         note.n_descsz = 0;
1579         note.n_type   = 0;
1580         memcpy(buf, &note, sizeof(note));
1581 }
1582
1583 void crash_save_cpu(struct pt_regs *regs, int cpu)
1584 {
1585         struct elf_prstatus prstatus;
1586         u32 *buf;
1587
1588         if ((cpu < 0) || (cpu >= nr_cpu_ids))
1589                 return;
1590
1591         /* Using ELF notes here is opportunistic.
1592          * I need a well defined structure format
1593          * for the data I pass, and I need tags
1594          * on the data to indicate what information I have
1595          * squirrelled away.  ELF notes happen to provide
1596          * all of that, so there is no need to invent something new.
1597          */
1598         buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
1599         if (!buf)
1600                 return;
1601         memset(&prstatus, 0, sizeof(prstatus));
1602         prstatus.pr_pid = current->pid;
1603         elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1604         buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1605                               &prstatus, sizeof(prstatus));
1606         final_note(buf);
1607 }
1608
1609 static int __init crash_notes_memory_init(void)
1610 {
1611         /* Allocate memory for saving cpu registers. */
1612         crash_notes = alloc_percpu(note_buf_t);
1613         if (!crash_notes) {
1614                 pr_warn("Kexec: Memory allocation for saving cpu register states failed\n");
1615                 return -ENOMEM;
1616         }
1617         return 0;
1618 }
1619 subsys_initcall(crash_notes_memory_init);
1620
1621
1622 /*
1623  * parsing the "crashkernel" commandline
1624  *
1625  * this code is intended to be called from architecture specific code
1626  */
1627
1628
1629 /*
1630  * This function parses command lines in the format
1631  *
1632  *   crashkernel=ramsize-range:size[,...][@offset]
1633  *
1634  * The function returns 0 on success and -EINVAL on failure.
1635  */
1636 static int __init parse_crashkernel_mem(char *cmdline,
1637                                         unsigned long long system_ram,
1638                                         unsigned long long *crash_size,
1639                                         unsigned long long *crash_base)
1640 {
1641         char *cur = cmdline, *tmp;
1642
1643         /* for each entry of the comma-separated list */
1644         do {
1645                 unsigned long long start, end = ULLONG_MAX, size;
1646
1647                 /* get the start of the range */
1648                 start = memparse(cur, &tmp);
1649                 if (cur == tmp) {
1650                         pr_warn("crashkernel: Memory value expected\n");
1651                         return -EINVAL;
1652                 }
1653                 cur = tmp;
1654                 if (*cur != '-') {
1655                         pr_warn("crashkernel: '-' expected\n");
1656                         return -EINVAL;
1657                 }
1658                 cur++;
1659
1660                 /* if no ':' is here, than we read the end */
1661                 if (*cur != ':') {
1662                         end = memparse(cur, &tmp);
1663                         if (cur == tmp) {
1664                                 pr_warn("crashkernel: Memory value expected\n");
1665                                 return -EINVAL;
1666                         }
1667                         cur = tmp;
1668                         if (end <= start) {
1669                                 pr_warn("crashkernel: end <= start\n");
1670                                 return -EINVAL;
1671                         }
1672                 }
1673
1674                 if (*cur != ':') {
1675                         pr_warn("crashkernel: ':' expected\n");
1676                         return -EINVAL;
1677                 }
1678                 cur++;
1679
1680                 size = memparse(cur, &tmp);
1681                 if (cur == tmp) {
1682                         pr_warn("Memory value expected\n");
1683                         return -EINVAL;
1684                 }
1685                 cur = tmp;
1686                 if (size >= system_ram) {
1687                         pr_warn("crashkernel: invalid size\n");
1688                         return -EINVAL;
1689                 }
1690
1691                 /* match ? */
1692                 if (system_ram >= start && system_ram < end) {
1693                         *crash_size = size;
1694                         break;
1695                 }
1696         } while (*cur++ == ',');
1697
1698         if (*crash_size > 0) {
1699                 while (*cur && *cur != ' ' && *cur != '@')
1700                         cur++;
1701                 if (*cur == '@') {
1702                         cur++;
1703                         *crash_base = memparse(cur, &tmp);
1704                         if (cur == tmp) {
1705                                 pr_warn("Memory value expected after '@'\n");
1706                                 return -EINVAL;
1707                         }
1708                 }
1709         }
1710
1711         return 0;
1712 }
1713
1714 /*
1715  * That function parses "simple" (old) crashkernel command lines like
1716  *
1717  *      crashkernel=size[@offset]
1718  *
1719  * It returns 0 on success and -EINVAL on failure.
1720  */
1721 static int __init parse_crashkernel_simple(char *cmdline,
1722                                            unsigned long long *crash_size,
1723                                            unsigned long long *crash_base)
1724 {
1725         char *cur = cmdline;
1726
1727         *crash_size = memparse(cmdline, &cur);
1728         if (cmdline == cur) {
1729                 pr_warn("crashkernel: memory value expected\n");
1730                 return -EINVAL;
1731         }
1732
1733         if (*cur == '@')
1734                 *crash_base = memparse(cur+1, &cur);
1735         else if (*cur != ' ' && *cur != '\0') {
1736                 pr_warn("crashkernel: unrecognized char\n");
1737                 return -EINVAL;
1738         }
1739
1740         return 0;
1741 }
1742
1743 #define SUFFIX_HIGH 0
1744 #define SUFFIX_LOW  1
1745 #define SUFFIX_NULL 2
1746 static __initdata char *suffix_tbl[] = {
1747         [SUFFIX_HIGH] = ",high",
1748         [SUFFIX_LOW]  = ",low",
1749         [SUFFIX_NULL] = NULL,
1750 };
1751
1752 /*
1753  * That function parses "suffix"  crashkernel command lines like
1754  *
1755  *      crashkernel=size,[high|low]
1756  *
1757  * It returns 0 on success and -EINVAL on failure.
1758  */
1759 static int __init parse_crashkernel_suffix(char *cmdline,
1760                                            unsigned long long   *crash_size,
1761                                            const char *suffix)
1762 {
1763         char *cur = cmdline;
1764
1765         *crash_size = memparse(cmdline, &cur);
1766         if (cmdline == cur) {
1767                 pr_warn("crashkernel: memory value expected\n");
1768                 return -EINVAL;
1769         }
1770
1771         /* check with suffix */
1772         if (strncmp(cur, suffix, strlen(suffix))) {
1773                 pr_warn("crashkernel: unrecognized char\n");
1774                 return -EINVAL;
1775         }
1776         cur += strlen(suffix);
1777         if (*cur != ' ' && *cur != '\0') {
1778                 pr_warn("crashkernel: unrecognized char\n");
1779                 return -EINVAL;
1780         }
1781
1782         return 0;
1783 }
1784
1785 static __init char *get_last_crashkernel(char *cmdline,
1786                              const char *name,
1787                              const char *suffix)
1788 {
1789         char *p = cmdline, *ck_cmdline = NULL;
1790
1791         /* find crashkernel and use the last one if there are more */
1792         p = strstr(p, name);
1793         while (p) {
1794                 char *end_p = strchr(p, ' ');
1795                 char *q;
1796
1797                 if (!end_p)
1798                         end_p = p + strlen(p);
1799
1800                 if (!suffix) {
1801                         int i;
1802
1803                         /* skip the one with any known suffix */
1804                         for (i = 0; suffix_tbl[i]; i++) {
1805                                 q = end_p - strlen(suffix_tbl[i]);
1806                                 if (!strncmp(q, suffix_tbl[i],
1807                                              strlen(suffix_tbl[i])))
1808                                         goto next;
1809                         }
1810                         ck_cmdline = p;
1811                 } else {
1812                         q = end_p - strlen(suffix);
1813                         if (!strncmp(q, suffix, strlen(suffix)))
1814                                 ck_cmdline = p;
1815                 }
1816 next:
1817                 p = strstr(p+1, name);
1818         }
1819
1820         if (!ck_cmdline)
1821                 return NULL;
1822
1823         return ck_cmdline;
1824 }
1825
1826 static int __init __parse_crashkernel(char *cmdline,
1827                              unsigned long long system_ram,
1828                              unsigned long long *crash_size,
1829                              unsigned long long *crash_base,
1830                              const char *name,
1831                              const char *suffix)
1832 {
1833         char    *first_colon, *first_space;
1834         char    *ck_cmdline;
1835
1836         BUG_ON(!crash_size || !crash_base);
1837         *crash_size = 0;
1838         *crash_base = 0;
1839
1840         ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1841
1842         if (!ck_cmdline)
1843                 return -EINVAL;
1844
1845         ck_cmdline += strlen(name);
1846
1847         if (suffix)
1848                 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1849                                 suffix);
1850         /*
1851          * if the commandline contains a ':', then that's the extended
1852          * syntax -- if not, it must be the classic syntax
1853          */
1854         first_colon = strchr(ck_cmdline, ':');
1855         first_space = strchr(ck_cmdline, ' ');
1856         if (first_colon && (!first_space || first_colon < first_space))
1857                 return parse_crashkernel_mem(ck_cmdline, system_ram,
1858                                 crash_size, crash_base);
1859
1860         return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1861 }
1862
1863 /*
1864  * That function is the entry point for command line parsing and should be
1865  * called from the arch-specific code.
1866  */
1867 int __init parse_crashkernel(char *cmdline,
1868                              unsigned long long system_ram,
1869                              unsigned long long *crash_size,
1870                              unsigned long long *crash_base)
1871 {
1872         return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1873                                         "crashkernel=", NULL);
1874 }
1875
1876 int __init parse_crashkernel_high(char *cmdline,
1877                              unsigned long long system_ram,
1878                              unsigned long long *crash_size,
1879                              unsigned long long *crash_base)
1880 {
1881         return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1882                                 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1883 }
1884
1885 int __init parse_crashkernel_low(char *cmdline,
1886                              unsigned long long system_ram,
1887                              unsigned long long *crash_size,
1888                              unsigned long long *crash_base)
1889 {
1890         return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1891                                 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1892 }
1893
1894 static void update_vmcoreinfo_note(void)
1895 {
1896         u32 *buf = vmcoreinfo_note;
1897
1898         if (!vmcoreinfo_size)
1899                 return;
1900         buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1901                               vmcoreinfo_size);
1902         final_note(buf);
1903 }
1904
1905 void crash_save_vmcoreinfo(void)
1906 {
1907         vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1908         update_vmcoreinfo_note();
1909 }
1910
1911 void vmcoreinfo_append_str(const char *fmt, ...)
1912 {
1913         va_list args;
1914         char buf[0x50];
1915         size_t r;
1916
1917         va_start(args, fmt);
1918         r = vscnprintf(buf, sizeof(buf), fmt, args);
1919         va_end(args);
1920
1921         r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1922
1923         memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1924
1925         vmcoreinfo_size += r;
1926 }
1927
1928 /*
1929  * provide an empty default implementation here -- architecture
1930  * code may override this
1931  */
1932 void __weak arch_crash_save_vmcoreinfo(void)
1933 {}
1934
1935 unsigned long __weak paddr_vmcoreinfo_note(void)
1936 {
1937         return __pa((unsigned long)(char *)&vmcoreinfo_note);
1938 }
1939
1940 static int __init crash_save_vmcoreinfo_init(void)
1941 {
1942         VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1943         VMCOREINFO_PAGESIZE(PAGE_SIZE);
1944
1945         VMCOREINFO_SYMBOL(init_uts_ns);
1946         VMCOREINFO_SYMBOL(node_online_map);
1947 #ifdef CONFIG_MMU
1948         VMCOREINFO_SYMBOL(swapper_pg_dir);
1949 #endif
1950         VMCOREINFO_SYMBOL(_stext);
1951         VMCOREINFO_SYMBOL(vmap_area_list);
1952
1953 #ifndef CONFIG_NEED_MULTIPLE_NODES
1954         VMCOREINFO_SYMBOL(mem_map);
1955         VMCOREINFO_SYMBOL(contig_page_data);
1956 #endif
1957 #ifdef CONFIG_SPARSEMEM
1958         VMCOREINFO_SYMBOL(mem_section);
1959         VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1960         VMCOREINFO_STRUCT_SIZE(mem_section);
1961         VMCOREINFO_OFFSET(mem_section, section_mem_map);
1962 #endif
1963         VMCOREINFO_STRUCT_SIZE(page);
1964         VMCOREINFO_STRUCT_SIZE(pglist_data);
1965         VMCOREINFO_STRUCT_SIZE(zone);
1966         VMCOREINFO_STRUCT_SIZE(free_area);
1967         VMCOREINFO_STRUCT_SIZE(list_head);
1968         VMCOREINFO_SIZE(nodemask_t);
1969         VMCOREINFO_OFFSET(page, flags);
1970         VMCOREINFO_OFFSET(page, _count);
1971         VMCOREINFO_OFFSET(page, mapping);
1972         VMCOREINFO_OFFSET(page, lru);
1973         VMCOREINFO_OFFSET(page, _mapcount);
1974         VMCOREINFO_OFFSET(page, private);
1975         VMCOREINFO_OFFSET(pglist_data, node_zones);
1976         VMCOREINFO_OFFSET(pglist_data, nr_zones);
1977 #ifdef CONFIG_FLAT_NODE_MEM_MAP
1978         VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1979 #endif
1980         VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1981         VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1982         VMCOREINFO_OFFSET(pglist_data, node_id);
1983         VMCOREINFO_OFFSET(zone, free_area);
1984         VMCOREINFO_OFFSET(zone, vm_stat);
1985         VMCOREINFO_OFFSET(zone, spanned_pages);
1986         VMCOREINFO_OFFSET(free_area, free_list);
1987         VMCOREINFO_OFFSET(list_head, next);
1988         VMCOREINFO_OFFSET(list_head, prev);
1989         VMCOREINFO_OFFSET(vmap_area, va_start);
1990         VMCOREINFO_OFFSET(vmap_area, list);
1991         VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1992         log_buf_kexec_setup();
1993         VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1994         VMCOREINFO_NUMBER(NR_FREE_PAGES);
1995         VMCOREINFO_NUMBER(PG_lru);
1996         VMCOREINFO_NUMBER(PG_private);
1997         VMCOREINFO_NUMBER(PG_swapcache);
1998         VMCOREINFO_NUMBER(PG_slab);
1999 #ifdef CONFIG_MEMORY_FAILURE
2000         VMCOREINFO_NUMBER(PG_hwpoison);
2001 #endif
2002         VMCOREINFO_NUMBER(PG_head_mask);
2003         VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
2004 #ifdef CONFIG_HUGETLBFS
2005         VMCOREINFO_SYMBOL(free_huge_page);
2006 #endif
2007
2008         arch_crash_save_vmcoreinfo();
2009         update_vmcoreinfo_note();
2010
2011         return 0;
2012 }
2013
2014 subsys_initcall(crash_save_vmcoreinfo_init);
2015
2016 #ifdef CONFIG_KEXEC_FILE
2017 static int locate_mem_hole_top_down(unsigned long start, unsigned long end,
2018                                     struct kexec_buf *kbuf)
2019 {
2020         struct kimage *image = kbuf->image;
2021         unsigned long temp_start, temp_end;
2022
2023         temp_end = min(end, kbuf->buf_max);
2024         temp_start = temp_end - kbuf->memsz;
2025
2026         do {
2027                 /* align down start */
2028                 temp_start = temp_start & (~(kbuf->buf_align - 1));
2029
2030                 if (temp_start < start || temp_start < kbuf->buf_min)
2031                         return 0;
2032
2033                 temp_end = temp_start + kbuf->memsz - 1;
2034
2035                 /*
2036                  * Make sure this does not conflict with any of existing
2037                  * segments
2038                  */
2039                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2040                         temp_start = temp_start - PAGE_SIZE;
2041                         continue;
2042                 }
2043
2044                 /* We found a suitable memory range */
2045                 break;
2046         } while (1);
2047
2048         /* If we are here, we found a suitable memory range */
2049         kbuf->mem = temp_start;
2050
2051         /* Success, stop navigating through remaining System RAM ranges */
2052         return 1;
2053 }
2054
2055 static int locate_mem_hole_bottom_up(unsigned long start, unsigned long end,
2056                                      struct kexec_buf *kbuf)
2057 {
2058         struct kimage *image = kbuf->image;
2059         unsigned long temp_start, temp_end;
2060
2061         temp_start = max(start, kbuf->buf_min);
2062
2063         do {
2064                 temp_start = ALIGN(temp_start, kbuf->buf_align);
2065                 temp_end = temp_start + kbuf->memsz - 1;
2066
2067                 if (temp_end > end || temp_end > kbuf->buf_max)
2068                         return 0;
2069                 /*
2070                  * Make sure this does not conflict with any of existing
2071                  * segments
2072                  */
2073                 if (kimage_is_destination_range(image, temp_start, temp_end)) {
2074                         temp_start = temp_start + PAGE_SIZE;
2075                         continue;
2076                 }
2077
2078                 /* We found a suitable memory range */
2079                 break;
2080         } while (1);
2081
2082         /* If we are here, we found a suitable memory range */
2083         kbuf->mem = temp_start;
2084
2085         /* Success, stop navigating through remaining System RAM ranges */
2086         return 1;
2087 }
2088
2089 static int locate_mem_hole_callback(u64 start, u64 end, void *arg)
2090 {
2091         struct kexec_buf *kbuf = (struct kexec_buf *)arg;
2092         unsigned long sz = end - start + 1;
2093
2094         /* Returning 0 will take to next memory range */
2095         if (sz < kbuf->memsz)
2096                 return 0;
2097
2098         if (end < kbuf->buf_min || start > kbuf->buf_max)
2099                 return 0;
2100
2101         /*
2102          * Allocate memory top down with-in ram range. Otherwise bottom up
2103          * allocation.
2104          */
2105         if (kbuf->top_down)
2106                 return locate_mem_hole_top_down(start, end, kbuf);
2107         return locate_mem_hole_bottom_up(start, end, kbuf);
2108 }
2109
2110 /*
2111  * Helper function for placing a buffer in a kexec segment. This assumes
2112  * that kexec_mutex is held.
2113  */
2114 int kexec_add_buffer(struct kimage *image, char *buffer, unsigned long bufsz,
2115                      unsigned long memsz, unsigned long buf_align,
2116                      unsigned long buf_min, unsigned long buf_max,
2117                      bool top_down, unsigned long *load_addr)
2118 {
2119
2120         struct kexec_segment *ksegment;
2121         struct kexec_buf buf, *kbuf;
2122         int ret;
2123
2124         /* Currently adding segment this way is allowed only in file mode */
2125         if (!image->file_mode)
2126                 return -EINVAL;
2127
2128         if (image->nr_segments >= KEXEC_SEGMENT_MAX)
2129                 return -EINVAL;
2130
2131         /*
2132          * Make sure we are not trying to add buffer after allocating
2133          * control pages. All segments need to be placed first before
2134          * any control pages are allocated. As control page allocation
2135          * logic goes through list of segments to make sure there are
2136          * no destination overlaps.
2137          */
2138         if (!list_empty(&image->control_pages)) {
2139                 WARN_ON(1);
2140                 return -EINVAL;
2141         }
2142
2143         memset(&buf, 0, sizeof(struct kexec_buf));
2144         kbuf = &buf;
2145         kbuf->image = image;
2146         kbuf->buffer = buffer;
2147         kbuf->bufsz = bufsz;
2148
2149         kbuf->memsz = ALIGN(memsz, PAGE_SIZE);
2150         kbuf->buf_align = max(buf_align, PAGE_SIZE);
2151         kbuf->buf_min = buf_min;
2152         kbuf->buf_max = buf_max;
2153         kbuf->top_down = top_down;
2154
2155         /* Walk the RAM ranges and allocate a suitable range for the buffer */
2156         if (image->type == KEXEC_TYPE_CRASH)
2157                 ret = walk_iomem_res("Crash kernel",
2158                                      IORESOURCE_MEM | IORESOURCE_BUSY,
2159                                      crashk_res.start, crashk_res.end, kbuf,
2160                                      locate_mem_hole_callback);
2161         else
2162                 ret = walk_system_ram_res(0, -1, kbuf,
2163                                           locate_mem_hole_callback);
2164         if (ret != 1) {
2165                 /* A suitable memory range could not be found for buffer */
2166                 return -EADDRNOTAVAIL;
2167         }
2168
2169         /* Found a suitable memory range */
2170         ksegment = &image->segment[image->nr_segments];
2171         ksegment->kbuf = kbuf->buffer;
2172         ksegment->bufsz = kbuf->bufsz;
2173         ksegment->mem = kbuf->mem;
2174         ksegment->memsz = kbuf->memsz;
2175         image->nr_segments++;
2176         *load_addr = ksegment->mem;
2177         return 0;
2178 }
2179
2180 /* Calculate and store the digest of segments */
2181 static int kexec_calculate_store_digests(struct kimage *image)
2182 {
2183         struct crypto_shash *tfm;
2184         struct shash_desc *desc;
2185         int ret = 0, i, j, zero_buf_sz, sha_region_sz;
2186         size_t desc_size, nullsz;
2187         char *digest;
2188         void *zero_buf;
2189         struct kexec_sha_region *sha_regions;
2190         struct purgatory_info *pi = &image->purgatory_info;
2191
2192         zero_buf = __va(page_to_pfn(ZERO_PAGE(0)) << PAGE_SHIFT);
2193         zero_buf_sz = PAGE_SIZE;
2194
2195         tfm = crypto_alloc_shash("sha256", 0, 0);
2196         if (IS_ERR(tfm)) {
2197                 ret = PTR_ERR(tfm);
2198                 goto out;
2199         }
2200
2201         desc_size = crypto_shash_descsize(tfm) + sizeof(*desc);
2202         desc = kzalloc(desc_size, GFP_KERNEL);
2203         if (!desc) {
2204                 ret = -ENOMEM;
2205                 goto out_free_tfm;
2206         }
2207
2208         sha_region_sz = KEXEC_SEGMENT_MAX * sizeof(struct kexec_sha_region);
2209         sha_regions = vzalloc(sha_region_sz);
2210         if (!sha_regions)
2211                 goto out_free_desc;
2212
2213         desc->tfm   = tfm;
2214         desc->flags = 0;
2215
2216         ret = crypto_shash_init(desc);
2217         if (ret < 0)
2218                 goto out_free_sha_regions;
2219
2220         digest = kzalloc(SHA256_DIGEST_SIZE, GFP_KERNEL);
2221         if (!digest) {
2222                 ret = -ENOMEM;
2223                 goto out_free_sha_regions;
2224         }
2225
2226         for (j = i = 0; i < image->nr_segments; i++) {
2227                 struct kexec_segment *ksegment;
2228
2229                 ksegment = &image->segment[i];
2230                 /*
2231                  * Skip purgatory as it will be modified once we put digest
2232                  * info in purgatory.
2233                  */
2234                 if (ksegment->kbuf == pi->purgatory_buf)
2235                         continue;
2236
2237                 ret = crypto_shash_update(desc, ksegment->kbuf,
2238                                           ksegment->bufsz);
2239                 if (ret)
2240                         break;
2241
2242                 /*
2243                  * Assume rest of the buffer is filled with zero and
2244                  * update digest accordingly.
2245                  */
2246                 nullsz = ksegment->memsz - ksegment->bufsz;
2247                 while (nullsz) {
2248                         unsigned long bytes = nullsz;
2249
2250                         if (bytes > zero_buf_sz)
2251                                 bytes = zero_buf_sz;
2252                         ret = crypto_shash_update(desc, zero_buf, bytes);
2253                         if (ret)
2254                                 break;
2255                         nullsz -= bytes;
2256                 }
2257
2258                 if (ret)
2259                         break;
2260
2261                 sha_regions[j].start = ksegment->mem;
2262                 sha_regions[j].len = ksegment->memsz;
2263                 j++;
2264         }
2265
2266         if (!ret) {
2267                 ret = crypto_shash_final(desc, digest);
2268                 if (ret)
2269                         goto out_free_digest;
2270                 ret = kexec_purgatory_get_set_symbol(image, "sha_regions",
2271                                                 sha_regions, sha_region_sz, 0);
2272                 if (ret)
2273                         goto out_free_digest;
2274
2275                 ret = kexec_purgatory_get_set_symbol(image, "sha256_digest",
2276                                                 digest, SHA256_DIGEST_SIZE, 0);
2277                 if (ret)
2278                         goto out_free_digest;
2279         }
2280
2281 out_free_digest:
2282         kfree(digest);
2283 out_free_sha_regions:
2284         vfree(sha_regions);
2285 out_free_desc:
2286         kfree(desc);
2287 out_free_tfm:
2288         kfree(tfm);
2289 out:
2290         return ret;
2291 }
2292
2293 /* Actually load purgatory. Lot of code taken from kexec-tools */
2294 static int __kexec_load_purgatory(struct kimage *image, unsigned long min,
2295                                   unsigned long max, int top_down)
2296 {
2297         struct purgatory_info *pi = &image->purgatory_info;
2298         unsigned long align, buf_align, bss_align, buf_sz, bss_sz, bss_pad;
2299         unsigned long memsz, entry, load_addr, curr_load_addr, bss_addr, offset;
2300         unsigned char *buf_addr, *src;
2301         int i, ret = 0, entry_sidx = -1;
2302         const Elf_Shdr *sechdrs_c;
2303         Elf_Shdr *sechdrs = NULL;
2304         void *purgatory_buf = NULL;
2305
2306         /*
2307          * sechdrs_c points to section headers in purgatory and are read
2308          * only. No modifications allowed.
2309          */
2310         sechdrs_c = (void *)pi->ehdr + pi->ehdr->e_shoff;
2311
2312         /*
2313          * We can not modify sechdrs_c[] and its fields. It is read only.
2314          * Copy it over to a local copy where one can store some temporary
2315          * data and free it at the end. We need to modify ->sh_addr and
2316          * ->sh_offset fields to keep track of permanent and temporary
2317          * locations of sections.
2318          */
2319         sechdrs = vzalloc(pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2320         if (!sechdrs)
2321                 return -ENOMEM;
2322
2323         memcpy(sechdrs, sechdrs_c, pi->ehdr->e_shnum * sizeof(Elf_Shdr));
2324
2325         /*
2326          * We seem to have multiple copies of sections. First copy is which
2327          * is embedded in kernel in read only section. Some of these sections
2328          * will be copied to a temporary buffer and relocated. And these
2329          * sections will finally be copied to their final destination at
2330          * segment load time.
2331          *
2332          * Use ->sh_offset to reflect section address in memory. It will
2333          * point to original read only copy if section is not allocatable.
2334          * Otherwise it will point to temporary copy which will be relocated.
2335          *
2336          * Use ->sh_addr to contain final address of the section where it
2337          * will go during execution time.
2338          */
2339         for (i = 0; i < pi->ehdr->e_shnum; i++) {
2340                 if (sechdrs[i].sh_type == SHT_NOBITS)
2341                         continue;
2342
2343                 sechdrs[i].sh_offset = (unsigned long)pi->ehdr +
2344                                                 sechdrs[i].sh_offset;
2345         }
2346
2347         /*
2348          * Identify entry point section and make entry relative to section
2349          * start.
2350          */
2351         entry = pi->ehdr->e_entry;
2352         for (i = 0; i < pi->ehdr->e_shnum; i++) {
2353                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2354                         continue;
2355
2356                 if (!(sechdrs[i].sh_flags & SHF_EXECINSTR))
2357                         continue;
2358
2359                 /* Make entry section relative */
2360                 if (sechdrs[i].sh_addr <= pi->ehdr->e_entry &&
2361                     ((sechdrs[i].sh_addr + sechdrs[i].sh_size) >
2362                      pi->ehdr->e_entry)) {
2363                         entry_sidx = i;
2364                         entry -= sechdrs[i].sh_addr;
2365                         break;
2366                 }
2367         }
2368
2369         /* Determine how much memory is needed to load relocatable object. */
2370         buf_align = 1;
2371         bss_align = 1;
2372         buf_sz = 0;
2373         bss_sz = 0;
2374
2375         for (i = 0; i < pi->ehdr->e_shnum; i++) {
2376                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2377                         continue;
2378
2379                 align = sechdrs[i].sh_addralign;
2380                 if (sechdrs[i].sh_type != SHT_NOBITS) {
2381                         if (buf_align < align)
2382                                 buf_align = align;
2383                         buf_sz = ALIGN(buf_sz, align);
2384                         buf_sz += sechdrs[i].sh_size;
2385                 } else {
2386                         /* bss section */
2387                         if (bss_align < align)
2388                                 bss_align = align;
2389                         bss_sz = ALIGN(bss_sz, align);
2390                         bss_sz += sechdrs[i].sh_size;
2391                 }
2392         }
2393
2394         /* Determine the bss padding required to align bss properly */
2395         bss_pad = 0;
2396         if (buf_sz & (bss_align - 1))
2397                 bss_pad = bss_align - (buf_sz & (bss_align - 1));
2398
2399         memsz = buf_sz + bss_pad + bss_sz;
2400
2401         /* Allocate buffer for purgatory */
2402         purgatory_buf = vzalloc(buf_sz);
2403         if (!purgatory_buf) {
2404                 ret = -ENOMEM;
2405                 goto out;
2406         }
2407
2408         if (buf_align < bss_align)
2409                 buf_align = bss_align;
2410
2411         /* Add buffer to segment list */
2412         ret = kexec_add_buffer(image, purgatory_buf, buf_sz, memsz,
2413                                 buf_align, min, max, top_down,
2414                                 &pi->purgatory_load_addr);
2415         if (ret)
2416                 goto out;
2417
2418         /* Load SHF_ALLOC sections */
2419         buf_addr = purgatory_buf;
2420         load_addr = curr_load_addr = pi->purgatory_load_addr;
2421         bss_addr = load_addr + buf_sz + bss_pad;
2422
2423         for (i = 0; i < pi->ehdr->e_shnum; i++) {
2424                 if (!(sechdrs[i].sh_flags & SHF_ALLOC))
2425                         continue;
2426
2427                 align = sechdrs[i].sh_addralign;
2428                 if (sechdrs[i].sh_type != SHT_NOBITS) {
2429                         curr_load_addr = ALIGN(curr_load_addr, align);
2430                         offset = curr_load_addr - load_addr;
2431                         /* We already modifed ->sh_offset to keep src addr */
2432                         src = (char *) sechdrs[i].sh_offset;
2433                         memcpy(buf_addr + offset, src, sechdrs[i].sh_size);
2434
2435                         /* Store load address and source address of section */
2436                         sechdrs[i].sh_addr = curr_load_addr;
2437
2438                         /*
2439                          * This section got copied to temporary buffer. Update
2440                          * ->sh_offset accordingly.
2441                          */
2442                         sechdrs[i].sh_offset = (unsigned long)(buf_addr + offset);
2443
2444                         /* Advance to the next address */
2445                         curr_load_addr += sechdrs[i].sh_size;
2446                 } else {
2447                         bss_addr = ALIGN(bss_addr, align);
2448                         sechdrs[i].sh_addr = bss_addr;
2449                         bss_addr += sechdrs[i].sh_size;
2450                 }
2451         }
2452
2453         /* Update entry point based on load address of text section */
2454         if (entry_sidx >= 0)
2455                 entry += sechdrs[entry_sidx].sh_addr;
2456
2457         /* Make kernel jump to purgatory after shutdown */
2458         image->start = entry;
2459
2460         /* Used later to get/set symbol values */
2461         pi->sechdrs = sechdrs;
2462
2463         /*
2464          * Used later to identify which section is purgatory and skip it
2465          * from checksumming.
2466          */
2467         pi->purgatory_buf = purgatory_buf;
2468         return ret;
2469 out:
2470         vfree(sechdrs);
2471         vfree(purgatory_buf);
2472         return ret;
2473 }
2474
2475 static int kexec_apply_relocations(struct kimage *image)
2476 {
2477         int i, ret;
2478         struct purgatory_info *pi = &image->purgatory_info;
2479         Elf_Shdr *sechdrs = pi->sechdrs;
2480
2481         /* Apply relocations */
2482         for (i = 0; i < pi->ehdr->e_shnum; i++) {
2483                 Elf_Shdr *section, *symtab;
2484
2485                 if (sechdrs[i].sh_type != SHT_RELA &&
2486                     sechdrs[i].sh_type != SHT_REL)
2487                         continue;
2488
2489                 /*
2490                  * For section of type SHT_RELA/SHT_REL,
2491                  * ->sh_link contains section header index of associated
2492                  * symbol table. And ->sh_info contains section header
2493                  * index of section to which relocations apply.
2494                  */
2495                 if (sechdrs[i].sh_info >= pi->ehdr->e_shnum ||
2496                     sechdrs[i].sh_link >= pi->ehdr->e_shnum)
2497                         return -ENOEXEC;
2498
2499                 section = &sechdrs[sechdrs[i].sh_info];
2500                 symtab = &sechdrs[sechdrs[i].sh_link];
2501
2502                 if (!(section->sh_flags & SHF_ALLOC))
2503                         continue;
2504
2505                 /*
2506                  * symtab->sh_link contain section header index of associated
2507                  * string table.
2508                  */
2509                 if (symtab->sh_link >= pi->ehdr->e_shnum)
2510                         /* Invalid section number? */
2511                         continue;
2512
2513                 /*
2514                  * Respective architecture needs to provide support for applying
2515                  * relocations of type SHT_RELA/SHT_REL.
2516                  */
2517                 if (sechdrs[i].sh_type == SHT_RELA)
2518                         ret = arch_kexec_apply_relocations_add(pi->ehdr,
2519                                                                sechdrs, i);
2520                 else if (sechdrs[i].sh_type == SHT_REL)
2521                         ret = arch_kexec_apply_relocations(pi->ehdr,
2522                                                            sechdrs, i);
2523                 if (ret)
2524                         return ret;
2525         }
2526
2527         return 0;
2528 }
2529
2530 /* Load relocatable purgatory object and relocate it appropriately */
2531 int kexec_load_purgatory(struct kimage *image, unsigned long min,
2532                          unsigned long max, int top_down,
2533                          unsigned long *load_addr)
2534 {
2535         struct purgatory_info *pi = &image->purgatory_info;
2536         int ret;
2537
2538         if (kexec_purgatory_size <= 0)
2539                 return -EINVAL;
2540
2541         if (kexec_purgatory_size < sizeof(Elf_Ehdr))
2542                 return -ENOEXEC;
2543
2544         pi->ehdr = (Elf_Ehdr *)kexec_purgatory;
2545
2546         if (memcmp(pi->ehdr->e_ident, ELFMAG, SELFMAG) != 0
2547             || pi->ehdr->e_type != ET_REL
2548             || !elf_check_arch(pi->ehdr)
2549             || pi->ehdr->e_shentsize != sizeof(Elf_Shdr))
2550                 return -ENOEXEC;
2551
2552         if (pi->ehdr->e_shoff >= kexec_purgatory_size
2553             || (pi->ehdr->e_shnum * sizeof(Elf_Shdr) >
2554             kexec_purgatory_size - pi->ehdr->e_shoff))
2555                 return -ENOEXEC;
2556
2557         ret = __kexec_load_purgatory(image, min, max, top_down);
2558         if (ret)
2559                 return ret;
2560
2561         ret = kexec_apply_relocations(image);
2562         if (ret)
2563                 goto out;
2564
2565         *load_addr = pi->purgatory_load_addr;
2566         return 0;
2567 out:
2568         vfree(pi->sechdrs);
2569         vfree(pi->purgatory_buf);
2570         return ret;
2571 }
2572
2573 static Elf_Sym *kexec_purgatory_find_symbol(struct purgatory_info *pi,
2574                                             const char *name)
2575 {
2576         Elf_Sym *syms;
2577         Elf_Shdr *sechdrs;
2578         Elf_Ehdr *ehdr;
2579         int i, k;
2580         const char *strtab;
2581
2582         if (!pi->sechdrs || !pi->ehdr)
2583                 return NULL;
2584
2585         sechdrs = pi->sechdrs;
2586         ehdr = pi->ehdr;
2587
2588         for (i = 0; i < ehdr->e_shnum; i++) {
2589                 if (sechdrs[i].sh_type != SHT_SYMTAB)
2590                         continue;
2591
2592                 if (sechdrs[i].sh_link >= ehdr->e_shnum)
2593                         /* Invalid strtab section number */
2594                         continue;
2595                 strtab = (char *)sechdrs[sechdrs[i].sh_link].sh_offset;
2596                 syms = (Elf_Sym *)sechdrs[i].sh_offset;
2597
2598                 /* Go through symbols for a match */
2599                 for (k = 0; k < sechdrs[i].sh_size/sizeof(Elf_Sym); k++) {
2600                         if (ELF_ST_BIND(syms[k].st_info) != STB_GLOBAL)
2601                                 continue;
2602
2603                         if (strcmp(strtab + syms[k].st_name, name) != 0)
2604                                 continue;
2605
2606                         if (syms[k].st_shndx == SHN_UNDEF ||
2607                             syms[k].st_shndx >= ehdr->e_shnum) {
2608                                 pr_debug("Symbol: %s has bad section index %d.\n",
2609                                                 name, syms[k].st_shndx);
2610                                 return NULL;
2611                         }
2612
2613                         /* Found the symbol we are looking for */
2614                         return &syms[k];
2615                 }
2616         }
2617
2618         return NULL;
2619 }
2620
2621 void *kexec_purgatory_get_symbol_addr(struct kimage *image, const char *name)
2622 {
2623         struct purgatory_info *pi = &image->purgatory_info;
2624         Elf_Sym *sym;
2625         Elf_Shdr *sechdr;
2626
2627         sym = kexec_purgatory_find_symbol(pi, name);
2628         if (!sym)
2629                 return ERR_PTR(-EINVAL);
2630
2631         sechdr = &pi->sechdrs[sym->st_shndx];
2632
2633         /*
2634          * Returns the address where symbol will finally be loaded after
2635          * kexec_load_segment()
2636          */
2637         return (void *)(sechdr->sh_addr + sym->st_value);
2638 }
2639
2640 /*
2641  * Get or set value of a symbol. If "get_value" is true, symbol value is
2642  * returned in buf otherwise symbol value is set based on value in buf.
2643  */
2644 int kexec_purgatory_get_set_symbol(struct kimage *image, const char *name,
2645                                    void *buf, unsigned int size, bool get_value)
2646 {
2647         Elf_Sym *sym;
2648         Elf_Shdr *sechdrs;
2649         struct purgatory_info *pi = &image->purgatory_info;
2650         char *sym_buf;
2651
2652         sym = kexec_purgatory_find_symbol(pi, name);
2653         if (!sym)
2654                 return -EINVAL;
2655
2656         if (sym->st_size != size) {
2657                 pr_err("symbol %s size mismatch: expected %lu actual %u\n",
2658                        name, (unsigned long)sym->st_size, size);
2659                 return -EINVAL;
2660         }
2661
2662         sechdrs = pi->sechdrs;
2663
2664         if (sechdrs[sym->st_shndx].sh_type == SHT_NOBITS) {
2665                 pr_err("symbol %s is in a bss section. Cannot %s\n", name,
2666                        get_value ? "get" : "set");
2667                 return -EINVAL;
2668         }
2669
2670         sym_buf = (unsigned char *)sechdrs[sym->st_shndx].sh_offset +
2671                                         sym->st_value;
2672
2673         if (get_value)
2674                 memcpy((void *)buf, sym_buf, size);
2675         else
2676                 memcpy((void *)sym_buf, buf, size);
2677
2678         return 0;
2679 }
2680 #endif /* CONFIG_KEXEC_FILE */
2681
2682 /*
2683  * Move into place and start executing a preloaded standalone
2684  * executable.  If nothing was preloaded return an error.
2685  */
2686 int kernel_kexec(void)
2687 {
2688         int error = 0;
2689
2690         if (!mutex_trylock(&kexec_mutex))
2691                 return -EBUSY;
2692         if (!kexec_image) {
2693                 error = -EINVAL;
2694                 goto Unlock;
2695         }
2696
2697 #ifdef CONFIG_KEXEC_JUMP
2698         if (kexec_image->preserve_context) {
2699                 lock_system_sleep();
2700                 pm_prepare_console();
2701                 error = freeze_processes();
2702                 if (error) {
2703                         error = -EBUSY;
2704                         goto Restore_console;
2705                 }
2706                 suspend_console();
2707                 error = dpm_suspend_start(PMSG_FREEZE);
2708                 if (error)
2709                         goto Resume_console;
2710                 /* At this point, dpm_suspend_start() has been called,
2711                  * but *not* dpm_suspend_end(). We *must* call
2712                  * dpm_suspend_end() now.  Otherwise, drivers for
2713                  * some devices (e.g. interrupt controllers) become
2714                  * desynchronized with the actual state of the
2715                  * hardware at resume time, and evil weirdness ensues.
2716                  */
2717                 error = dpm_suspend_end(PMSG_FREEZE);
2718                 if (error)
2719                         goto Resume_devices;
2720                 error = disable_nonboot_cpus();
2721                 if (error)
2722                         goto Enable_cpus;
2723                 local_irq_disable();
2724                 error = syscore_suspend();
2725                 if (error)
2726                         goto Enable_irqs;
2727         } else
2728 #endif
2729         {
2730                 kexec_in_progress = true;
2731                 kernel_restart_prepare(NULL);
2732                 migrate_to_reboot_cpu();
2733
2734                 /*
2735                  * migrate_to_reboot_cpu() disables CPU hotplug assuming that
2736                  * no further code needs to use CPU hotplug (which is true in
2737                  * the reboot case). However, the kexec path depends on using
2738                  * CPU hotplug again; so re-enable it here.
2739                  */
2740                 cpu_hotplug_enable();
2741                 pr_emerg("Starting new kernel\n");
2742                 machine_shutdown();
2743         }
2744
2745         machine_kexec(kexec_image);
2746
2747 #ifdef CONFIG_KEXEC_JUMP
2748         if (kexec_image->preserve_context) {
2749                 syscore_resume();
2750  Enable_irqs:
2751                 local_irq_enable();
2752  Enable_cpus:
2753                 enable_nonboot_cpus();
2754                 dpm_resume_start(PMSG_RESTORE);
2755  Resume_devices:
2756                 dpm_resume_end(PMSG_RESTORE);
2757  Resume_console:
2758                 resume_console();
2759                 thaw_processes();
2760  Restore_console:
2761                 pm_restore_console();
2762                 unlock_system_sleep();
2763         }
2764 #endif
2765
2766  Unlock:
2767         mutex_unlock(&kexec_mutex);
2768         return error;
2769 }