xref: /linux/kernel/kexec.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
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 #include <linux/capability.h>
10 #include <linux/mm.h>
11 #include <linux/file.h>
12 #include <linux/slab.h>
13 #include <linux/fs.h>
14 #include <linux/kexec.h>
15 #include <linux/spinlock.h>
16 #include <linux/list.h>
17 #include <linux/highmem.h>
18 #include <linux/syscalls.h>
19 #include <linux/reboot.h>
20 #include <linux/syscalls.h>
21 #include <linux/ioport.h>
22 #include <linux/hardirq.h>
23 
24 #include <asm/page.h>
25 #include <asm/uaccess.h>
26 #include <asm/io.h>
27 #include <asm/system.h>
28 #include <asm/semaphore.h>
29 
30 /* Per cpu memory for storing cpu states in case of system crash. */
31 note_buf_t* crash_notes;
32 
33 /* Location of the reserved area for the crash kernel */
34 struct resource crashk_res = {
35 	.name  = "Crash kernel",
36 	.start = 0,
37 	.end   = 0,
38 	.flags = IORESOURCE_BUSY | IORESOURCE_MEM
39 };
40 
41 int kexec_should_crash(struct task_struct *p)
42 {
43 	if (in_interrupt() || !p->pid || p->pid == 1 || panic_on_oops)
44 		return 1;
45 	return 0;
46 }
47 
48 /*
49  * When kexec transitions to the new kernel there is a one-to-one
50  * mapping between physical and virtual addresses.  On processors
51  * where you can disable the MMU this is trivial, and easy.  For
52  * others it is still a simple predictable page table to setup.
53  *
54  * In that environment kexec copies the new kernel to its final
55  * resting place.  This means I can only support memory whose
56  * physical address can fit in an unsigned long.  In particular
57  * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
58  * If the assembly stub has more restrictive requirements
59  * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
60  * defined more restrictively in <asm/kexec.h>.
61  *
62  * The code for the transition from the current kernel to the
63  * the new kernel is placed in the control_code_buffer, whose size
64  * is given by KEXEC_CONTROL_CODE_SIZE.  In the best case only a single
65  * page of memory is necessary, but some architectures require more.
66  * Because this memory must be identity mapped in the transition from
67  * virtual to physical addresses it must live in the range
68  * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
69  * modifiable.
70  *
71  * The assembly stub in the control code buffer is passed a linked list
72  * of descriptor pages detailing the source pages of the new kernel,
73  * and the destination addresses of those source pages.  As this data
74  * structure is not used in the context of the current OS, it must
75  * be self-contained.
76  *
77  * The code has been made to work with highmem pages and will use a
78  * destination page in its final resting place (if it happens
79  * to allocate it).  The end product of this is that most of the
80  * physical address space, and most of RAM can be used.
81  *
82  * Future directions include:
83  *  - allocating a page table with the control code buffer identity
84  *    mapped, to simplify machine_kexec and make kexec_on_panic more
85  *    reliable.
86  */
87 
88 /*
89  * KIMAGE_NO_DEST is an impossible destination address..., for
90  * allocating pages whose destination address we do not care about.
91  */
92 #define KIMAGE_NO_DEST (-1UL)
93 
94 static int kimage_is_destination_range(struct kimage *image,
95 				       unsigned long start, unsigned long end);
96 static struct page *kimage_alloc_page(struct kimage *image,
97 				       gfp_t gfp_mask,
98 				       unsigned long dest);
99 
100 static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
101 	                    unsigned long nr_segments,
102                             struct kexec_segment __user *segments)
103 {
104 	size_t segment_bytes;
105 	struct kimage *image;
106 	unsigned long i;
107 	int result;
108 
109 	/* Allocate a controlling structure */
110 	result = -ENOMEM;
111 	image = kmalloc(sizeof(*image), GFP_KERNEL);
112 	if (!image)
113 		goto out;
114 
115 	memset(image, 0, sizeof(*image));
116 	image->head = 0;
117 	image->entry = &image->head;
118 	image->last_entry = &image->head;
119 	image->control_page = ~0; /* By default this does not apply */
120 	image->start = entry;
121 	image->type = KEXEC_TYPE_DEFAULT;
122 
123 	/* Initialize the list of control pages */
124 	INIT_LIST_HEAD(&image->control_pages);
125 
126 	/* Initialize the list of destination pages */
127 	INIT_LIST_HEAD(&image->dest_pages);
128 
129 	/* Initialize the list of unuseable pages */
130 	INIT_LIST_HEAD(&image->unuseable_pages);
131 
132 	/* Read in the segments */
133 	image->nr_segments = nr_segments;
134 	segment_bytes = nr_segments * sizeof(*segments);
135 	result = copy_from_user(image->segment, segments, segment_bytes);
136 	if (result)
137 		goto out;
138 
139 	/*
140 	 * Verify we have good destination addresses.  The caller is
141 	 * responsible for making certain we don't attempt to load
142 	 * the new image into invalid or reserved areas of RAM.  This
143 	 * just verifies it is an address we can use.
144 	 *
145 	 * Since the kernel does everything in page size chunks ensure
146 	 * the destination addreses are page aligned.  Too many
147 	 * special cases crop of when we don't do this.  The most
148 	 * insidious is getting overlapping destination addresses
149 	 * simply because addresses are changed to page size
150 	 * granularity.
151 	 */
152 	result = -EADDRNOTAVAIL;
153 	for (i = 0; i < nr_segments; i++) {
154 		unsigned long mstart, mend;
155 
156 		mstart = image->segment[i].mem;
157 		mend   = mstart + image->segment[i].memsz;
158 		if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
159 			goto out;
160 		if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
161 			goto out;
162 	}
163 
164 	/* Verify our destination addresses do not overlap.
165 	 * If we alloed overlapping destination addresses
166 	 * through very weird things can happen with no
167 	 * easy explanation as one segment stops on another.
168 	 */
169 	result = -EINVAL;
170 	for (i = 0; i < nr_segments; i++) {
171 		unsigned long mstart, mend;
172 		unsigned long j;
173 
174 		mstart = image->segment[i].mem;
175 		mend   = mstart + image->segment[i].memsz;
176 		for (j = 0; j < i; j++) {
177 			unsigned long pstart, pend;
178 			pstart = image->segment[j].mem;
179 			pend   = pstart + image->segment[j].memsz;
180 			/* Do the segments overlap ? */
181 			if ((mend > pstart) && (mstart < pend))
182 				goto out;
183 		}
184 	}
185 
186 	/* Ensure our buffer sizes are strictly less than
187 	 * our memory sizes.  This should always be the case,
188 	 * and it is easier to check up front than to be surprised
189 	 * later on.
190 	 */
191 	result = -EINVAL;
192 	for (i = 0; i < nr_segments; i++) {
193 		if (image->segment[i].bufsz > image->segment[i].memsz)
194 			goto out;
195 	}
196 
197 	result = 0;
198 out:
199 	if (result == 0)
200 		*rimage = image;
201 	else
202 		kfree(image);
203 
204 	return result;
205 
206 }
207 
208 static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
209 				unsigned long nr_segments,
210 				struct kexec_segment __user *segments)
211 {
212 	int result;
213 	struct kimage *image;
214 
215 	/* Allocate and initialize a controlling structure */
216 	image = NULL;
217 	result = do_kimage_alloc(&image, entry, nr_segments, segments);
218 	if (result)
219 		goto out;
220 
221 	*rimage = image;
222 
223 	/*
224 	 * Find a location for the control code buffer, and add it
225 	 * the vector of segments so that it's pages will also be
226 	 * counted as destination pages.
227 	 */
228 	result = -ENOMEM;
229 	image->control_code_page = kimage_alloc_control_pages(image,
230 					   get_order(KEXEC_CONTROL_CODE_SIZE));
231 	if (!image->control_code_page) {
232 		printk(KERN_ERR "Could not allocate control_code_buffer\n");
233 		goto out;
234 	}
235 
236 	result = 0;
237  out:
238 	if (result == 0)
239 		*rimage = image;
240 	else
241 		kfree(image);
242 
243 	return result;
244 }
245 
246 static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
247 				unsigned long nr_segments,
248 				struct kexec_segment __user *segments)
249 {
250 	int result;
251 	struct kimage *image;
252 	unsigned long i;
253 
254 	image = NULL;
255 	/* Verify we have a valid entry point */
256 	if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
257 		result = -EADDRNOTAVAIL;
258 		goto out;
259 	}
260 
261 	/* Allocate and initialize a controlling structure */
262 	result = do_kimage_alloc(&image, entry, nr_segments, segments);
263 	if (result)
264 		goto out;
265 
266 	/* Enable the special crash kernel control page
267 	 * allocation policy.
268 	 */
269 	image->control_page = crashk_res.start;
270 	image->type = KEXEC_TYPE_CRASH;
271 
272 	/*
273 	 * Verify we have good destination addresses.  Normally
274 	 * the caller is responsible for making certain we don't
275 	 * attempt to load the new image into invalid or reserved
276 	 * areas of RAM.  But crash kernels are preloaded into a
277 	 * reserved area of ram.  We must ensure the addresses
278 	 * are in the reserved area otherwise preloading the
279 	 * kernel could corrupt things.
280 	 */
281 	result = -EADDRNOTAVAIL;
282 	for (i = 0; i < nr_segments; i++) {
283 		unsigned long mstart, mend;
284 
285 		mstart = image->segment[i].mem;
286 		mend = mstart + image->segment[i].memsz - 1;
287 		/* Ensure we are within the crash kernel limits */
288 		if ((mstart < crashk_res.start) || (mend > crashk_res.end))
289 			goto out;
290 	}
291 
292 	/*
293 	 * Find a location for the control code buffer, and add
294 	 * the vector of segments so that it's pages will also be
295 	 * counted as destination pages.
296 	 */
297 	result = -ENOMEM;
298 	image->control_code_page = kimage_alloc_control_pages(image,
299 					   get_order(KEXEC_CONTROL_CODE_SIZE));
300 	if (!image->control_code_page) {
301 		printk(KERN_ERR "Could not allocate control_code_buffer\n");
302 		goto out;
303 	}
304 
305 	result = 0;
306 out:
307 	if (result == 0)
308 		*rimage = image;
309 	else
310 		kfree(image);
311 
312 	return result;
313 }
314 
315 static int kimage_is_destination_range(struct kimage *image,
316 					unsigned long start,
317 					unsigned long end)
318 {
319 	unsigned long i;
320 
321 	for (i = 0; i < image->nr_segments; i++) {
322 		unsigned long mstart, mend;
323 
324 		mstart = image->segment[i].mem;
325 		mend = mstart + image->segment[i].memsz;
326 		if ((end > mstart) && (start < mend))
327 			return 1;
328 	}
329 
330 	return 0;
331 }
332 
333 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
334 {
335 	struct page *pages;
336 
337 	pages = alloc_pages(gfp_mask, order);
338 	if (pages) {
339 		unsigned int count, i;
340 		pages->mapping = NULL;
341 		set_page_private(pages, order);
342 		count = 1 << order;
343 		for (i = 0; i < count; i++)
344 			SetPageReserved(pages + i);
345 	}
346 
347 	return pages;
348 }
349 
350 static void kimage_free_pages(struct page *page)
351 {
352 	unsigned int order, count, i;
353 
354 	order = page_private(page);
355 	count = 1 << order;
356 	for (i = 0; i < count; i++)
357 		ClearPageReserved(page + i);
358 	__free_pages(page, order);
359 }
360 
361 static void kimage_free_page_list(struct list_head *list)
362 {
363 	struct list_head *pos, *next;
364 
365 	list_for_each_safe(pos, next, list) {
366 		struct page *page;
367 
368 		page = list_entry(pos, struct page, lru);
369 		list_del(&page->lru);
370 		kimage_free_pages(page);
371 	}
372 }
373 
374 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
375 							unsigned int order)
376 {
377 	/* Control pages are special, they are the intermediaries
378 	 * that are needed while we copy the rest of the pages
379 	 * to their final resting place.  As such they must
380 	 * not conflict with either the destination addresses
381 	 * or memory the kernel is already using.
382 	 *
383 	 * The only case where we really need more than one of
384 	 * these are for architectures where we cannot disable
385 	 * the MMU and must instead generate an identity mapped
386 	 * page table for all of the memory.
387 	 *
388 	 * At worst this runs in O(N) of the image size.
389 	 */
390 	struct list_head extra_pages;
391 	struct page *pages;
392 	unsigned int count;
393 
394 	count = 1 << order;
395 	INIT_LIST_HEAD(&extra_pages);
396 
397 	/* Loop while I can allocate a page and the page allocated
398 	 * is a destination page.
399 	 */
400 	do {
401 		unsigned long pfn, epfn, addr, eaddr;
402 
403 		pages = kimage_alloc_pages(GFP_KERNEL, order);
404 		if (!pages)
405 			break;
406 		pfn   = page_to_pfn(pages);
407 		epfn  = pfn + count;
408 		addr  = pfn << PAGE_SHIFT;
409 		eaddr = epfn << PAGE_SHIFT;
410 		if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
411 			      kimage_is_destination_range(image, addr, eaddr)) {
412 			list_add(&pages->lru, &extra_pages);
413 			pages = NULL;
414 		}
415 	} while (!pages);
416 
417 	if (pages) {
418 		/* Remember the allocated page... */
419 		list_add(&pages->lru, &image->control_pages);
420 
421 		/* Because the page is already in it's destination
422 		 * location we will never allocate another page at
423 		 * that address.  Therefore kimage_alloc_pages
424 		 * will not return it (again) and we don't need
425 		 * to give it an entry in image->segment[].
426 		 */
427 	}
428 	/* Deal with the destination pages I have inadvertently allocated.
429 	 *
430 	 * Ideally I would convert multi-page allocations into single
431 	 * page allocations, and add everyting to image->dest_pages.
432 	 *
433 	 * For now it is simpler to just free the pages.
434 	 */
435 	kimage_free_page_list(&extra_pages);
436 
437 	return pages;
438 }
439 
440 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
441 						      unsigned int order)
442 {
443 	/* Control pages are special, they are the intermediaries
444 	 * that are needed while we copy the rest of the pages
445 	 * to their final resting place.  As such they must
446 	 * not conflict with either the destination addresses
447 	 * or memory the kernel is already using.
448 	 *
449 	 * Control pages are also the only pags we must allocate
450 	 * when loading a crash kernel.  All of the other pages
451 	 * are specified by the segments and we just memcpy
452 	 * into them directly.
453 	 *
454 	 * The only case where we really need more than one of
455 	 * these are for architectures where we cannot disable
456 	 * the MMU and must instead generate an identity mapped
457 	 * page table for all of the memory.
458 	 *
459 	 * Given the low demand this implements a very simple
460 	 * allocator that finds the first hole of the appropriate
461 	 * size in the reserved memory region, and allocates all
462 	 * of the memory up to and including the hole.
463 	 */
464 	unsigned long hole_start, hole_end, size;
465 	struct page *pages;
466 
467 	pages = NULL;
468 	size = (1 << order) << PAGE_SHIFT;
469 	hole_start = (image->control_page + (size - 1)) & ~(size - 1);
470 	hole_end   = hole_start + size - 1;
471 	while (hole_end <= crashk_res.end) {
472 		unsigned long i;
473 
474 		if (hole_end > KEXEC_CONTROL_MEMORY_LIMIT)
475 			break;
476 		if (hole_end > crashk_res.end)
477 			break;
478 		/* See if I overlap any of the segments */
479 		for (i = 0; i < image->nr_segments; i++) {
480 			unsigned long mstart, mend;
481 
482 			mstart = image->segment[i].mem;
483 			mend   = mstart + image->segment[i].memsz - 1;
484 			if ((hole_end >= mstart) && (hole_start <= mend)) {
485 				/* Advance the hole to the end of the segment */
486 				hole_start = (mend + (size - 1)) & ~(size - 1);
487 				hole_end   = hole_start + size - 1;
488 				break;
489 			}
490 		}
491 		/* If I don't overlap any segments I have found my hole! */
492 		if (i == image->nr_segments) {
493 			pages = pfn_to_page(hole_start >> PAGE_SHIFT);
494 			break;
495 		}
496 	}
497 	if (pages)
498 		image->control_page = hole_end;
499 
500 	return pages;
501 }
502 
503 
504 struct page *kimage_alloc_control_pages(struct kimage *image,
505 					 unsigned int order)
506 {
507 	struct page *pages = NULL;
508 
509 	switch (image->type) {
510 	case KEXEC_TYPE_DEFAULT:
511 		pages = kimage_alloc_normal_control_pages(image, order);
512 		break;
513 	case KEXEC_TYPE_CRASH:
514 		pages = kimage_alloc_crash_control_pages(image, order);
515 		break;
516 	}
517 
518 	return pages;
519 }
520 
521 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
522 {
523 	if (*image->entry != 0)
524 		image->entry++;
525 
526 	if (image->entry == image->last_entry) {
527 		kimage_entry_t *ind_page;
528 		struct page *page;
529 
530 		page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
531 		if (!page)
532 			return -ENOMEM;
533 
534 		ind_page = page_address(page);
535 		*image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
536 		image->entry = ind_page;
537 		image->last_entry = ind_page +
538 				      ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
539 	}
540 	*image->entry = entry;
541 	image->entry++;
542 	*image->entry = 0;
543 
544 	return 0;
545 }
546 
547 static int kimage_set_destination(struct kimage *image,
548 				   unsigned long destination)
549 {
550 	int result;
551 
552 	destination &= PAGE_MASK;
553 	result = kimage_add_entry(image, destination | IND_DESTINATION);
554 	if (result == 0)
555 		image->destination = destination;
556 
557 	return result;
558 }
559 
560 
561 static int kimage_add_page(struct kimage *image, unsigned long page)
562 {
563 	int result;
564 
565 	page &= PAGE_MASK;
566 	result = kimage_add_entry(image, page | IND_SOURCE);
567 	if (result == 0)
568 		image->destination += PAGE_SIZE;
569 
570 	return result;
571 }
572 
573 
574 static void kimage_free_extra_pages(struct kimage *image)
575 {
576 	/* Walk through and free any extra destination pages I may have */
577 	kimage_free_page_list(&image->dest_pages);
578 
579 	/* Walk through and free any unuseable pages I have cached */
580 	kimage_free_page_list(&image->unuseable_pages);
581 
582 }
583 static int kimage_terminate(struct kimage *image)
584 {
585 	if (*image->entry != 0)
586 		image->entry++;
587 
588 	*image->entry = IND_DONE;
589 
590 	return 0;
591 }
592 
593 #define for_each_kimage_entry(image, ptr, entry) \
594 	for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
595 		ptr = (entry & IND_INDIRECTION)? \
596 			phys_to_virt((entry & PAGE_MASK)): ptr +1)
597 
598 static void kimage_free_entry(kimage_entry_t entry)
599 {
600 	struct page *page;
601 
602 	page = pfn_to_page(entry >> PAGE_SHIFT);
603 	kimage_free_pages(page);
604 }
605 
606 static void kimage_free(struct kimage *image)
607 {
608 	kimage_entry_t *ptr, entry;
609 	kimage_entry_t ind = 0;
610 
611 	if (!image)
612 		return;
613 
614 	kimage_free_extra_pages(image);
615 	for_each_kimage_entry(image, ptr, entry) {
616 		if (entry & IND_INDIRECTION) {
617 			/* Free the previous indirection page */
618 			if (ind & IND_INDIRECTION)
619 				kimage_free_entry(ind);
620 			/* Save this indirection page until we are
621 			 * done with it.
622 			 */
623 			ind = entry;
624 		}
625 		else if (entry & IND_SOURCE)
626 			kimage_free_entry(entry);
627 	}
628 	/* Free the final indirection page */
629 	if (ind & IND_INDIRECTION)
630 		kimage_free_entry(ind);
631 
632 	/* Handle any machine specific cleanup */
633 	machine_kexec_cleanup(image);
634 
635 	/* Free the kexec control pages... */
636 	kimage_free_page_list(&image->control_pages);
637 	kfree(image);
638 }
639 
640 static kimage_entry_t *kimage_dst_used(struct kimage *image,
641 					unsigned long page)
642 {
643 	kimage_entry_t *ptr, entry;
644 	unsigned long destination = 0;
645 
646 	for_each_kimage_entry(image, ptr, entry) {
647 		if (entry & IND_DESTINATION)
648 			destination = entry & PAGE_MASK;
649 		else if (entry & IND_SOURCE) {
650 			if (page == destination)
651 				return ptr;
652 			destination += PAGE_SIZE;
653 		}
654 	}
655 
656 	return NULL;
657 }
658 
659 static struct page *kimage_alloc_page(struct kimage *image,
660 					gfp_t gfp_mask,
661 					unsigned long destination)
662 {
663 	/*
664 	 * Here we implement safeguards to ensure that a source page
665 	 * is not copied to its destination page before the data on
666 	 * the destination page is no longer useful.
667 	 *
668 	 * To do this we maintain the invariant that a source page is
669 	 * either its own destination page, or it is not a
670 	 * destination page at all.
671 	 *
672 	 * That is slightly stronger than required, but the proof
673 	 * that no problems will not occur is trivial, and the
674 	 * implementation is simply to verify.
675 	 *
676 	 * When allocating all pages normally this algorithm will run
677 	 * in O(N) time, but in the worst case it will run in O(N^2)
678 	 * time.   If the runtime is a problem the data structures can
679 	 * be fixed.
680 	 */
681 	struct page *page;
682 	unsigned long addr;
683 
684 	/*
685 	 * Walk through the list of destination pages, and see if I
686 	 * have a match.
687 	 */
688 	list_for_each_entry(page, &image->dest_pages, lru) {
689 		addr = page_to_pfn(page) << PAGE_SHIFT;
690 		if (addr == destination) {
691 			list_del(&page->lru);
692 			return page;
693 		}
694 	}
695 	page = NULL;
696 	while (1) {
697 		kimage_entry_t *old;
698 
699 		/* Allocate a page, if we run out of memory give up */
700 		page = kimage_alloc_pages(gfp_mask, 0);
701 		if (!page)
702 			return NULL;
703 		/* If the page cannot be used file it away */
704 		if (page_to_pfn(page) >
705 				(KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
706 			list_add(&page->lru, &image->unuseable_pages);
707 			continue;
708 		}
709 		addr = page_to_pfn(page) << PAGE_SHIFT;
710 
711 		/* If it is the destination page we want use it */
712 		if (addr == destination)
713 			break;
714 
715 		/* If the page is not a destination page use it */
716 		if (!kimage_is_destination_range(image, addr,
717 						  addr + PAGE_SIZE))
718 			break;
719 
720 		/*
721 		 * I know that the page is someones destination page.
722 		 * See if there is already a source page for this
723 		 * destination page.  And if so swap the source pages.
724 		 */
725 		old = kimage_dst_used(image, addr);
726 		if (old) {
727 			/* If so move it */
728 			unsigned long old_addr;
729 			struct page *old_page;
730 
731 			old_addr = *old & PAGE_MASK;
732 			old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
733 			copy_highpage(page, old_page);
734 			*old = addr | (*old & ~PAGE_MASK);
735 
736 			/* The old page I have found cannot be a
737 			 * destination page, so return it.
738 			 */
739 			addr = old_addr;
740 			page = old_page;
741 			break;
742 		}
743 		else {
744 			/* Place the page on the destination list I
745 			 * will use it later.
746 			 */
747 			list_add(&page->lru, &image->dest_pages);
748 		}
749 	}
750 
751 	return page;
752 }
753 
754 static int kimage_load_normal_segment(struct kimage *image,
755 					 struct kexec_segment *segment)
756 {
757 	unsigned long maddr;
758 	unsigned long ubytes, mbytes;
759 	int result;
760 	unsigned char __user *buf;
761 
762 	result = 0;
763 	buf = segment->buf;
764 	ubytes = segment->bufsz;
765 	mbytes = segment->memsz;
766 	maddr = segment->mem;
767 
768 	result = kimage_set_destination(image, maddr);
769 	if (result < 0)
770 		goto out;
771 
772 	while (mbytes) {
773 		struct page *page;
774 		char *ptr;
775 		size_t uchunk, mchunk;
776 
777 		page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
778 		if (page == 0) {
779 			result  = -ENOMEM;
780 			goto out;
781 		}
782 		result = kimage_add_page(image, page_to_pfn(page)
783 								<< PAGE_SHIFT);
784 		if (result < 0)
785 			goto out;
786 
787 		ptr = kmap(page);
788 		/* Start with a clear page */
789 		memset(ptr, 0, PAGE_SIZE);
790 		ptr += maddr & ~PAGE_MASK;
791 		mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
792 		if (mchunk > mbytes)
793 			mchunk = mbytes;
794 
795 		uchunk = mchunk;
796 		if (uchunk > ubytes)
797 			uchunk = ubytes;
798 
799 		result = copy_from_user(ptr, buf, uchunk);
800 		kunmap(page);
801 		if (result) {
802 			result = (result < 0) ? result : -EIO;
803 			goto out;
804 		}
805 		ubytes -= uchunk;
806 		maddr  += mchunk;
807 		buf    += mchunk;
808 		mbytes -= mchunk;
809 	}
810 out:
811 	return result;
812 }
813 
814 static int kimage_load_crash_segment(struct kimage *image,
815 					struct kexec_segment *segment)
816 {
817 	/* For crash dumps kernels we simply copy the data from
818 	 * user space to it's destination.
819 	 * We do things a page at a time for the sake of kmap.
820 	 */
821 	unsigned long maddr;
822 	unsigned long ubytes, mbytes;
823 	int result;
824 	unsigned char __user *buf;
825 
826 	result = 0;
827 	buf = segment->buf;
828 	ubytes = segment->bufsz;
829 	mbytes = segment->memsz;
830 	maddr = segment->mem;
831 	while (mbytes) {
832 		struct page *page;
833 		char *ptr;
834 		size_t uchunk, mchunk;
835 
836 		page = pfn_to_page(maddr >> PAGE_SHIFT);
837 		if (page == 0) {
838 			result  = -ENOMEM;
839 			goto out;
840 		}
841 		ptr = kmap(page);
842 		ptr += maddr & ~PAGE_MASK;
843 		mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
844 		if (mchunk > mbytes)
845 			mchunk = mbytes;
846 
847 		uchunk = mchunk;
848 		if (uchunk > ubytes) {
849 			uchunk = ubytes;
850 			/* Zero the trailing part of the page */
851 			memset(ptr + uchunk, 0, mchunk - uchunk);
852 		}
853 		result = copy_from_user(ptr, buf, uchunk);
854 		kunmap(page);
855 		if (result) {
856 			result = (result < 0) ? result : -EIO;
857 			goto out;
858 		}
859 		ubytes -= uchunk;
860 		maddr  += mchunk;
861 		buf    += mchunk;
862 		mbytes -= mchunk;
863 	}
864 out:
865 	return result;
866 }
867 
868 static int kimage_load_segment(struct kimage *image,
869 				struct kexec_segment *segment)
870 {
871 	int result = -ENOMEM;
872 
873 	switch (image->type) {
874 	case KEXEC_TYPE_DEFAULT:
875 		result = kimage_load_normal_segment(image, segment);
876 		break;
877 	case KEXEC_TYPE_CRASH:
878 		result = kimage_load_crash_segment(image, segment);
879 		break;
880 	}
881 
882 	return result;
883 }
884 
885 /*
886  * Exec Kernel system call: for obvious reasons only root may call it.
887  *
888  * This call breaks up into three pieces.
889  * - A generic part which loads the new kernel from the current
890  *   address space, and very carefully places the data in the
891  *   allocated pages.
892  *
893  * - A generic part that interacts with the kernel and tells all of
894  *   the devices to shut down.  Preventing on-going dmas, and placing
895  *   the devices in a consistent state so a later kernel can
896  *   reinitialize them.
897  *
898  * - A machine specific part that includes the syscall number
899  *   and the copies the image to it's final destination.  And
900  *   jumps into the image at entry.
901  *
902  * kexec does not sync, or unmount filesystems so if you need
903  * that to happen you need to do that yourself.
904  */
905 struct kimage *kexec_image = NULL;
906 static struct kimage *kexec_crash_image = NULL;
907 /*
908  * A home grown binary mutex.
909  * Nothing can wait so this mutex is safe to use
910  * in interrupt context :)
911  */
912 static int kexec_lock = 0;
913 
914 asmlinkage long sys_kexec_load(unsigned long entry, unsigned long nr_segments,
915 				struct kexec_segment __user *segments,
916 				unsigned long flags)
917 {
918 	struct kimage **dest_image, *image;
919 	int locked;
920 	int result;
921 
922 	/* We only trust the superuser with rebooting the system. */
923 	if (!capable(CAP_SYS_BOOT))
924 		return -EPERM;
925 
926 	/*
927 	 * Verify we have a legal set of flags
928 	 * This leaves us room for future extensions.
929 	 */
930 	if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
931 		return -EINVAL;
932 
933 	/* Verify we are on the appropriate architecture */
934 	if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
935 		((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
936 		return -EINVAL;
937 
938 	/* Put an artificial cap on the number
939 	 * of segments passed to kexec_load.
940 	 */
941 	if (nr_segments > KEXEC_SEGMENT_MAX)
942 		return -EINVAL;
943 
944 	image = NULL;
945 	result = 0;
946 
947 	/* Because we write directly to the reserved memory
948 	 * region when loading crash kernels we need a mutex here to
949 	 * prevent multiple crash  kernels from attempting to load
950 	 * simultaneously, and to prevent a crash kernel from loading
951 	 * over the top of a in use crash kernel.
952 	 *
953 	 * KISS: always take the mutex.
954 	 */
955 	locked = xchg(&kexec_lock, 1);
956 	if (locked)
957 		return -EBUSY;
958 
959 	dest_image = &kexec_image;
960 	if (flags & KEXEC_ON_CRASH)
961 		dest_image = &kexec_crash_image;
962 	if (nr_segments > 0) {
963 		unsigned long i;
964 
965 		/* Loading another kernel to reboot into */
966 		if ((flags & KEXEC_ON_CRASH) == 0)
967 			result = kimage_normal_alloc(&image, entry,
968 							nr_segments, segments);
969 		/* Loading another kernel to switch to if this one crashes */
970 		else if (flags & KEXEC_ON_CRASH) {
971 			/* Free any current crash dump kernel before
972 			 * we corrupt it.
973 			 */
974 			kimage_free(xchg(&kexec_crash_image, NULL));
975 			result = kimage_crash_alloc(&image, entry,
976 						     nr_segments, segments);
977 		}
978 		if (result)
979 			goto out;
980 
981 		result = machine_kexec_prepare(image);
982 		if (result)
983 			goto out;
984 
985 		for (i = 0; i < nr_segments; i++) {
986 			result = kimage_load_segment(image, &image->segment[i]);
987 			if (result)
988 				goto out;
989 		}
990 		result = kimage_terminate(image);
991 		if (result)
992 			goto out;
993 	}
994 	/* Install the new kernel, and  Uninstall the old */
995 	image = xchg(dest_image, image);
996 
997 out:
998 	xchg(&kexec_lock, 0); /* Release the mutex */
999 	kimage_free(image);
1000 
1001 	return result;
1002 }
1003 
1004 #ifdef CONFIG_COMPAT
1005 asmlinkage long compat_sys_kexec_load(unsigned long entry,
1006 				unsigned long nr_segments,
1007 				struct compat_kexec_segment __user *segments,
1008 				unsigned long flags)
1009 {
1010 	struct compat_kexec_segment in;
1011 	struct kexec_segment out, __user *ksegments;
1012 	unsigned long i, result;
1013 
1014 	/* Don't allow clients that don't understand the native
1015 	 * architecture to do anything.
1016 	 */
1017 	if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1018 		return -EINVAL;
1019 
1020 	if (nr_segments > KEXEC_SEGMENT_MAX)
1021 		return -EINVAL;
1022 
1023 	ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1024 	for (i=0; i < nr_segments; i++) {
1025 		result = copy_from_user(&in, &segments[i], sizeof(in));
1026 		if (result)
1027 			return -EFAULT;
1028 
1029 		out.buf   = compat_ptr(in.buf);
1030 		out.bufsz = in.bufsz;
1031 		out.mem   = in.mem;
1032 		out.memsz = in.memsz;
1033 
1034 		result = copy_to_user(&ksegments[i], &out, sizeof(out));
1035 		if (result)
1036 			return -EFAULT;
1037 	}
1038 
1039 	return sys_kexec_load(entry, nr_segments, ksegments, flags);
1040 }
1041 #endif
1042 
1043 void crash_kexec(struct pt_regs *regs)
1044 {
1045 	struct kimage *image;
1046 	int locked;
1047 
1048 
1049 	/* Take the kexec_lock here to prevent sys_kexec_load
1050 	 * running on one cpu from replacing the crash kernel
1051 	 * we are using after a panic on a different cpu.
1052 	 *
1053 	 * If the crash kernel was not located in a fixed area
1054 	 * of memory the xchg(&kexec_crash_image) would be
1055 	 * sufficient.  But since I reuse the memory...
1056 	 */
1057 	locked = xchg(&kexec_lock, 1);
1058 	if (!locked) {
1059 		image = xchg(&kexec_crash_image, NULL);
1060 		if (image) {
1061 			struct pt_regs fixed_regs;
1062 			crash_setup_regs(&fixed_regs, regs);
1063 			machine_crash_shutdown(&fixed_regs);
1064 			machine_kexec(image);
1065 		}
1066 		xchg(&kexec_lock, 0);
1067 	}
1068 }
1069 
1070 static int __init crash_notes_memory_init(void)
1071 {
1072 	/* Allocate memory for saving cpu registers. */
1073 	crash_notes = alloc_percpu(note_buf_t);
1074 	if (!crash_notes) {
1075 		printk("Kexec: Memory allocation for saving cpu register"
1076 		" states failed\n");
1077 		return -ENOMEM;
1078 	}
1079 	return 0;
1080 }
1081 module_init(crash_notes_memory_init)
1082