blob: ec19b92c7ebdac942fa3e8dc269d737c7032593a [file] [log] [blame]
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001/*
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
Randy.Dunlapc59ede72006-01-11 12:17:46 -08009#include <linux/capability.h>
Eric W. Biedermandc009d92005-06-25 14:57:52 -070010#include <linux/mm.h>
11#include <linux/file.h>
12#include <linux/slab.h>
13#include <linux/fs.h>
14#include <linux/kexec.h>
Andrew Morton8c5a1cf2008-08-15 00:40:27 -070015#include <linux/mutex.h>
Eric W. Biedermandc009d92005-06-25 14:57:52 -070016#include <linux/list.h>
17#include <linux/highmem.h>
18#include <linux/syscalls.h>
19#include <linux/reboot.h>
Eric W. Biedermandc009d92005-06-25 14:57:52 -070020#include <linux/ioport.h>
Alexander Nyberg6e274d12005-06-25 14:58:26 -070021#include <linux/hardirq.h>
Magnus Damm85916f82006-12-06 20:40:41 -080022#include <linux/elf.h>
23#include <linux/elfcore.h>
Sam Ravnborg273b2812009-10-18 00:52:28 +020024#include <generated/utsrelease.h>
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070025#include <linux/utsname.h>
26#include <linux/numa.h>
Huang Ying3ab83522008-07-25 19:45:07 -070027#include <linux/suspend.h>
28#include <linux/device.h>
Huang Ying89081d12008-07-25 19:45:10 -070029#include <linux/freezer.h>
30#include <linux/pm.h>
31#include <linux/cpu.h>
32#include <linux/console.h>
Luck, Tony5f41b8c2008-10-20 15:23:40 -070033#include <linux/vmalloc.h>
Amerigo Wang06a7f712009-12-15 16:47:46 -080034#include <linux/swap.h>
KOSAKI Motohiro0f4bd462009-12-22 03:15:43 +000035#include <linux/kmsg_dump.h>
Alexander Nyberg6e274d12005-06-25 14:58:26 -070036
Eric W. Biedermandc009d92005-06-25 14:57:52 -070037#include <asm/page.h>
38#include <asm/uaccess.h>
39#include <asm/io.h>
40#include <asm/system.h>
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070041#include <asm/sections.h>
Eric W. Biedermandc009d92005-06-25 14:57:52 -070042
Vivek Goyalcc571652006-01-09 20:51:41 -080043/* Per cpu memory for storing cpu states in case of system crash. */
Tejun Heo43cf38e2010-02-02 14:38:57 +090044note_buf_t __percpu *crash_notes;
Vivek Goyalcc571652006-01-09 20:51:41 -080045
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070046/* vmcoreinfo stuff */
Dmitri Vorobievedb79a22009-04-02 16:58:58 -070047static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070048u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
Ken'ichi Ohmichid7682812007-10-16 23:27:28 -070049size_t vmcoreinfo_size;
50size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -070051
Eric W. Biedermandc009d92005-06-25 14:57:52 -070052/* Location of the reserved area for the crash kernel */
53struct resource crashk_res = {
54 .name = "Crash kernel",
55 .start = 0,
56 .end = 0,
57 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
58};
59
Alexander Nyberg6e274d12005-06-25 14:58:26 -070060int kexec_should_crash(struct task_struct *p)
61{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -070062 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
Alexander Nyberg6e274d12005-06-25 14:58:26 -070063 return 1;
64 return 0;
65}
66
Eric W. Biedermandc009d92005-06-25 14:57:52 -070067/*
68 * When kexec transitions to the new kernel there is a one-to-one
69 * mapping between physical and virtual addresses. On processors
70 * where you can disable the MMU this is trivial, and easy. For
71 * others it is still a simple predictable page table to setup.
72 *
73 * In that environment kexec copies the new kernel to its final
74 * resting place. This means I can only support memory whose
75 * physical address can fit in an unsigned long. In particular
76 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
77 * If the assembly stub has more restrictive requirements
78 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
79 * defined more restrictively in <asm/kexec.h>.
80 *
81 * The code for the transition from the current kernel to the
82 * the new kernel is placed in the control_code_buffer, whose size
Huang Ying163f6872008-08-15 00:40:22 -070083 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
Eric W. Biedermandc009d92005-06-25 14:57:52 -070084 * page of memory is necessary, but some architectures require more.
85 * Because this memory must be identity mapped in the transition from
86 * virtual to physical addresses it must live in the range
87 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
88 * modifiable.
89 *
90 * The assembly stub in the control code buffer is passed a linked list
91 * of descriptor pages detailing the source pages of the new kernel,
92 * and the destination addresses of those source pages. As this data
93 * structure is not used in the context of the current OS, it must
94 * be self-contained.
95 *
96 * The code has been made to work with highmem pages and will use a
97 * destination page in its final resting place (if it happens
98 * to allocate it). The end product of this is that most of the
99 * physical address space, and most of RAM can be used.
100 *
101 * Future directions include:
102 * - allocating a page table with the control code buffer identity
103 * mapped, to simplify machine_kexec and make kexec_on_panic more
104 * reliable.
105 */
106
107/*
108 * KIMAGE_NO_DEST is an impossible destination address..., for
109 * allocating pages whose destination address we do not care about.
110 */
111#define KIMAGE_NO_DEST (-1UL)
112
Maneesh Soni72414d32005-06-25 14:58:28 -0700113static int kimage_is_destination_range(struct kimage *image,
114 unsigned long start, unsigned long end);
115static struct page *kimage_alloc_page(struct kimage *image,
Al Viro9796fdd2005-10-21 03:22:03 -0400116 gfp_t gfp_mask,
Maneesh Soni72414d32005-06-25 14:58:28 -0700117 unsigned long dest);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700118
119static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
Maneesh Soni72414d32005-06-25 14:58:28 -0700120 unsigned long nr_segments,
121 struct kexec_segment __user *segments)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700122{
123 size_t segment_bytes;
124 struct kimage *image;
125 unsigned long i;
126 int result;
127
128 /* Allocate a controlling structure */
129 result = -ENOMEM;
Burman Yan4668edc2006-12-06 20:38:51 -0800130 image = kzalloc(sizeof(*image), GFP_KERNEL);
Maneesh Soni72414d32005-06-25 14:58:28 -0700131 if (!image)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700132 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -0700133
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700134 image->head = 0;
135 image->entry = &image->head;
136 image->last_entry = &image->head;
137 image->control_page = ~0; /* By default this does not apply */
138 image->start = entry;
139 image->type = KEXEC_TYPE_DEFAULT;
140
141 /* Initialize the list of control pages */
142 INIT_LIST_HEAD(&image->control_pages);
143
144 /* Initialize the list of destination pages */
145 INIT_LIST_HEAD(&image->dest_pages);
146
147 /* Initialize the list of unuseable pages */
148 INIT_LIST_HEAD(&image->unuseable_pages);
149
150 /* Read in the segments */
151 image->nr_segments = nr_segments;
152 segment_bytes = nr_segments * sizeof(*segments);
153 result = copy_from_user(image->segment, segments, segment_bytes);
Dan Carpenterf65a03f2010-08-10 18:03:31 -0700154 if (result) {
155 result = -EFAULT;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700156 goto out;
Dan Carpenterf65a03f2010-08-10 18:03:31 -0700157 }
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700158
159 /*
160 * Verify we have good destination addresses. The caller is
161 * responsible for making certain we don't attempt to load
162 * the new image into invalid or reserved areas of RAM. This
163 * just verifies it is an address we can use.
164 *
165 * Since the kernel does everything in page size chunks ensure
Uwe Kleine-Königb5950762010-11-01 15:38:34 -0400166 * the destination addresses are page aligned. Too many
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700167 * special cases crop of when we don't do this. The most
168 * insidious is getting overlapping destination addresses
169 * simply because addresses are changed to page size
170 * granularity.
171 */
172 result = -EADDRNOTAVAIL;
173 for (i = 0; i < nr_segments; i++) {
174 unsigned long mstart, mend;
Maneesh Soni72414d32005-06-25 14:58:28 -0700175
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700176 mstart = image->segment[i].mem;
177 mend = mstart + image->segment[i].memsz;
178 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
179 goto out;
180 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
181 goto out;
182 }
183
184 /* Verify our destination addresses do not overlap.
185 * If we alloed overlapping destination addresses
186 * through very weird things can happen with no
187 * easy explanation as one segment stops on another.
188 */
189 result = -EINVAL;
Maneesh Soni72414d32005-06-25 14:58:28 -0700190 for (i = 0; i < nr_segments; i++) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700191 unsigned long mstart, mend;
192 unsigned long j;
Maneesh Soni72414d32005-06-25 14:58:28 -0700193
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700194 mstart = image->segment[i].mem;
195 mend = mstart + image->segment[i].memsz;
Maneesh Soni72414d32005-06-25 14:58:28 -0700196 for (j = 0; j < i; j++) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700197 unsigned long pstart, pend;
198 pstart = image->segment[j].mem;
199 pend = pstart + image->segment[j].memsz;
200 /* Do the segments overlap ? */
201 if ((mend > pstart) && (mstart < pend))
202 goto out;
203 }
204 }
205
206 /* Ensure our buffer sizes are strictly less than
207 * our memory sizes. This should always be the case,
208 * and it is easier to check up front than to be surprised
209 * later on.
210 */
211 result = -EINVAL;
Maneesh Soni72414d32005-06-25 14:58:28 -0700212 for (i = 0; i < nr_segments; i++) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700213 if (image->segment[i].bufsz > image->segment[i].memsz)
214 goto out;
215 }
216
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700217 result = 0;
Maneesh Soni72414d32005-06-25 14:58:28 -0700218out:
219 if (result == 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700220 *rimage = image;
Maneesh Soni72414d32005-06-25 14:58:28 -0700221 else
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700222 kfree(image);
Maneesh Soni72414d32005-06-25 14:58:28 -0700223
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700224 return result;
225
226}
227
228static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
Maneesh Soni72414d32005-06-25 14:58:28 -0700229 unsigned long nr_segments,
230 struct kexec_segment __user *segments)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700231{
232 int result;
233 struct kimage *image;
234
235 /* Allocate and initialize a controlling structure */
236 image = NULL;
237 result = do_kimage_alloc(&image, entry, nr_segments, segments);
Maneesh Soni72414d32005-06-25 14:58:28 -0700238 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700239 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -0700240
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700241 *rimage = image;
242
243 /*
244 * Find a location for the control code buffer, and add it
245 * the vector of segments so that it's pages will also be
246 * counted as destination pages.
247 */
248 result = -ENOMEM;
249 image->control_code_page = kimage_alloc_control_pages(image,
Huang Ying163f6872008-08-15 00:40:22 -0700250 get_order(KEXEC_CONTROL_PAGE_SIZE));
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700251 if (!image->control_code_page) {
252 printk(KERN_ERR "Could not allocate control_code_buffer\n");
253 goto out;
254 }
255
Huang Ying3ab83522008-07-25 19:45:07 -0700256 image->swap_page = kimage_alloc_control_pages(image, 0);
257 if (!image->swap_page) {
258 printk(KERN_ERR "Could not allocate swap buffer\n");
259 goto out;
260 }
261
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700262 result = 0;
263 out:
Maneesh Soni72414d32005-06-25 14:58:28 -0700264 if (result == 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700265 *rimage = image;
Maneesh Soni72414d32005-06-25 14:58:28 -0700266 else
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700267 kfree(image);
Maneesh Soni72414d32005-06-25 14:58:28 -0700268
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700269 return result;
270}
271
272static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
Maneesh Soni72414d32005-06-25 14:58:28 -0700273 unsigned long nr_segments,
Alexey Dobriyan314b6a42005-06-27 22:29:33 -0700274 struct kexec_segment __user *segments)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700275{
276 int result;
277 struct kimage *image;
278 unsigned long i;
279
280 image = NULL;
281 /* Verify we have a valid entry point */
282 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
283 result = -EADDRNOTAVAIL;
284 goto out;
285 }
286
287 /* Allocate and initialize a controlling structure */
288 result = do_kimage_alloc(&image, entry, nr_segments, segments);
Maneesh Soni72414d32005-06-25 14:58:28 -0700289 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700290 goto out;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700291
292 /* Enable the special crash kernel control page
293 * allocation policy.
294 */
295 image->control_page = crashk_res.start;
296 image->type = KEXEC_TYPE_CRASH;
297
298 /*
299 * Verify we have good destination addresses. Normally
300 * the caller is responsible for making certain we don't
301 * attempt to load the new image into invalid or reserved
302 * areas of RAM. But crash kernels are preloaded into a
303 * reserved area of ram. We must ensure the addresses
304 * are in the reserved area otherwise preloading the
305 * kernel could corrupt things.
306 */
307 result = -EADDRNOTAVAIL;
308 for (i = 0; i < nr_segments; i++) {
309 unsigned long mstart, mend;
Maneesh Soni72414d32005-06-25 14:58:28 -0700310
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700311 mstart = image->segment[i].mem;
Vivek Goyal50cccc62005-06-25 14:57:55 -0700312 mend = mstart + image->segment[i].memsz - 1;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700313 /* Ensure we are within the crash kernel limits */
314 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
315 goto out;
316 }
317
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700318 /*
319 * Find a location for the control code buffer, and add
320 * the vector of segments so that it's pages will also be
321 * counted as destination pages.
322 */
323 result = -ENOMEM;
324 image->control_code_page = kimage_alloc_control_pages(image,
Huang Ying163f6872008-08-15 00:40:22 -0700325 get_order(KEXEC_CONTROL_PAGE_SIZE));
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700326 if (!image->control_code_page) {
327 printk(KERN_ERR "Could not allocate control_code_buffer\n");
328 goto out;
329 }
330
331 result = 0;
Maneesh Soni72414d32005-06-25 14:58:28 -0700332out:
333 if (result == 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700334 *rimage = image;
Maneesh Soni72414d32005-06-25 14:58:28 -0700335 else
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700336 kfree(image);
Maneesh Soni72414d32005-06-25 14:58:28 -0700337
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700338 return result;
339}
340
Maneesh Soni72414d32005-06-25 14:58:28 -0700341static int kimage_is_destination_range(struct kimage *image,
342 unsigned long start,
343 unsigned long end)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700344{
345 unsigned long i;
346
347 for (i = 0; i < image->nr_segments; i++) {
348 unsigned long mstart, mend;
Maneesh Soni72414d32005-06-25 14:58:28 -0700349
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700350 mstart = image->segment[i].mem;
Maneesh Soni72414d32005-06-25 14:58:28 -0700351 mend = mstart + image->segment[i].memsz;
352 if ((end > mstart) && (start < mend))
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700353 return 1;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700354 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700355
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700356 return 0;
357}
358
Al Viro9796fdd2005-10-21 03:22:03 -0400359static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700360{
361 struct page *pages;
Maneesh Soni72414d32005-06-25 14:58:28 -0700362
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700363 pages = alloc_pages(gfp_mask, order);
364 if (pages) {
365 unsigned int count, i;
366 pages->mapping = NULL;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700367 set_page_private(pages, order);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700368 count = 1 << order;
Maneesh Soni72414d32005-06-25 14:58:28 -0700369 for (i = 0; i < count; i++)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700370 SetPageReserved(pages + i);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700371 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700372
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700373 return pages;
374}
375
376static void kimage_free_pages(struct page *page)
377{
378 unsigned int order, count, i;
Maneesh Soni72414d32005-06-25 14:58:28 -0700379
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700380 order = page_private(page);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700381 count = 1 << order;
Maneesh Soni72414d32005-06-25 14:58:28 -0700382 for (i = 0; i < count; i++)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700383 ClearPageReserved(page + i);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700384 __free_pages(page, order);
385}
386
387static void kimage_free_page_list(struct list_head *list)
388{
389 struct list_head *pos, *next;
Maneesh Soni72414d32005-06-25 14:58:28 -0700390
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700391 list_for_each_safe(pos, next, list) {
392 struct page *page;
393
394 page = list_entry(pos, struct page, lru);
395 list_del(&page->lru);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700396 kimage_free_pages(page);
397 }
398}
399
Maneesh Soni72414d32005-06-25 14:58:28 -0700400static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
401 unsigned int order)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700402{
403 /* Control pages are special, they are the intermediaries
404 * that are needed while we copy the rest of the pages
405 * to their final resting place. As such they must
406 * not conflict with either the destination addresses
407 * or memory the kernel is already using.
408 *
409 * The only case where we really need more than one of
410 * these are for architectures where we cannot disable
411 * the MMU and must instead generate an identity mapped
412 * page table for all of the memory.
413 *
414 * At worst this runs in O(N) of the image size.
415 */
416 struct list_head extra_pages;
417 struct page *pages;
418 unsigned int count;
419
420 count = 1 << order;
421 INIT_LIST_HEAD(&extra_pages);
422
423 /* Loop while I can allocate a page and the page allocated
424 * is a destination page.
425 */
426 do {
427 unsigned long pfn, epfn, addr, eaddr;
Maneesh Soni72414d32005-06-25 14:58:28 -0700428
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700429 pages = kimage_alloc_pages(GFP_KERNEL, order);
430 if (!pages)
431 break;
432 pfn = page_to_pfn(pages);
433 epfn = pfn + count;
434 addr = pfn << PAGE_SHIFT;
435 eaddr = epfn << PAGE_SHIFT;
436 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
Maneesh Soni72414d32005-06-25 14:58:28 -0700437 kimage_is_destination_range(image, addr, eaddr)) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700438 list_add(&pages->lru, &extra_pages);
439 pages = NULL;
440 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700441 } while (!pages);
442
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700443 if (pages) {
444 /* Remember the allocated page... */
445 list_add(&pages->lru, &image->control_pages);
446
447 /* Because the page is already in it's destination
448 * location we will never allocate another page at
449 * that address. Therefore kimage_alloc_pages
450 * will not return it (again) and we don't need
451 * to give it an entry in image->segment[].
452 */
453 }
454 /* Deal with the destination pages I have inadvertently allocated.
455 *
456 * Ideally I would convert multi-page allocations into single
457 * page allocations, and add everyting to image->dest_pages.
458 *
459 * For now it is simpler to just free the pages.
460 */
461 kimage_free_page_list(&extra_pages);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700462
Maneesh Soni72414d32005-06-25 14:58:28 -0700463 return pages;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700464}
465
Maneesh Soni72414d32005-06-25 14:58:28 -0700466static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
467 unsigned int order)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700468{
469 /* Control pages are special, they are the intermediaries
470 * that are needed while we copy the rest of the pages
471 * to their final resting place. As such they must
472 * not conflict with either the destination addresses
473 * or memory the kernel is already using.
474 *
475 * Control pages are also the only pags we must allocate
476 * when loading a crash kernel. All of the other pages
477 * are specified by the segments and we just memcpy
478 * into them directly.
479 *
480 * The only case where we really need more than one of
481 * these are for architectures where we cannot disable
482 * the MMU and must instead generate an identity mapped
483 * page table for all of the memory.
484 *
485 * Given the low demand this implements a very simple
486 * allocator that finds the first hole of the appropriate
487 * size in the reserved memory region, and allocates all
488 * of the memory up to and including the hole.
489 */
490 unsigned long hole_start, hole_end, size;
491 struct page *pages;
Maneesh Soni72414d32005-06-25 14:58:28 -0700492
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700493 pages = NULL;
494 size = (1 << order) << PAGE_SHIFT;
495 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
496 hole_end = hole_start + size - 1;
Maneesh Soni72414d32005-06-25 14:58:28 -0700497 while (hole_end <= crashk_res.end) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700498 unsigned long i;
Maneesh Soni72414d32005-06-25 14:58:28 -0700499
500 if (hole_end > KEXEC_CONTROL_MEMORY_LIMIT)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700501 break;
Maneesh Soni72414d32005-06-25 14:58:28 -0700502 if (hole_end > crashk_res.end)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700503 break;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700504 /* See if I overlap any of the segments */
Maneesh Soni72414d32005-06-25 14:58:28 -0700505 for (i = 0; i < image->nr_segments; i++) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700506 unsigned long mstart, mend;
Maneesh Soni72414d32005-06-25 14:58:28 -0700507
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700508 mstart = image->segment[i].mem;
509 mend = mstart + image->segment[i].memsz - 1;
510 if ((hole_end >= mstart) && (hole_start <= mend)) {
511 /* Advance the hole to the end of the segment */
512 hole_start = (mend + (size - 1)) & ~(size - 1);
513 hole_end = hole_start + size - 1;
514 break;
515 }
516 }
517 /* If I don't overlap any segments I have found my hole! */
518 if (i == image->nr_segments) {
519 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
520 break;
521 }
522 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700523 if (pages)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700524 image->control_page = hole_end;
Maneesh Soni72414d32005-06-25 14:58:28 -0700525
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700526 return pages;
527}
528
529
Maneesh Soni72414d32005-06-25 14:58:28 -0700530struct page *kimage_alloc_control_pages(struct kimage *image,
531 unsigned int order)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700532{
533 struct page *pages = NULL;
Maneesh Soni72414d32005-06-25 14:58:28 -0700534
535 switch (image->type) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700536 case KEXEC_TYPE_DEFAULT:
537 pages = kimage_alloc_normal_control_pages(image, order);
538 break;
539 case KEXEC_TYPE_CRASH:
540 pages = kimage_alloc_crash_control_pages(image, order);
541 break;
542 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700543
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700544 return pages;
545}
546
547static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
548{
Maneesh Soni72414d32005-06-25 14:58:28 -0700549 if (*image->entry != 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700550 image->entry++;
Maneesh Soni72414d32005-06-25 14:58:28 -0700551
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700552 if (image->entry == image->last_entry) {
553 kimage_entry_t *ind_page;
554 struct page *page;
Maneesh Soni72414d32005-06-25 14:58:28 -0700555
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700556 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
Maneesh Soni72414d32005-06-25 14:58:28 -0700557 if (!page)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700558 return -ENOMEM;
Maneesh Soni72414d32005-06-25 14:58:28 -0700559
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700560 ind_page = page_address(page);
561 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
562 image->entry = ind_page;
Maneesh Soni72414d32005-06-25 14:58:28 -0700563 image->last_entry = ind_page +
564 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700565 }
566 *image->entry = entry;
567 image->entry++;
568 *image->entry = 0;
Maneesh Soni72414d32005-06-25 14:58:28 -0700569
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700570 return 0;
571}
572
Maneesh Soni72414d32005-06-25 14:58:28 -0700573static int kimage_set_destination(struct kimage *image,
574 unsigned long destination)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700575{
576 int result;
577
578 destination &= PAGE_MASK;
579 result = kimage_add_entry(image, destination | IND_DESTINATION);
Maneesh Soni72414d32005-06-25 14:58:28 -0700580 if (result == 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700581 image->destination = destination;
Maneesh Soni72414d32005-06-25 14:58:28 -0700582
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700583 return result;
584}
585
586
587static int kimage_add_page(struct kimage *image, unsigned long page)
588{
589 int result;
590
591 page &= PAGE_MASK;
592 result = kimage_add_entry(image, page | IND_SOURCE);
Maneesh Soni72414d32005-06-25 14:58:28 -0700593 if (result == 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700594 image->destination += PAGE_SIZE;
Maneesh Soni72414d32005-06-25 14:58:28 -0700595
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700596 return result;
597}
598
599
600static void kimage_free_extra_pages(struct kimage *image)
601{
602 /* Walk through and free any extra destination pages I may have */
603 kimage_free_page_list(&image->dest_pages);
604
605 /* Walk through and free any unuseable pages I have cached */
606 kimage_free_page_list(&image->unuseable_pages);
607
608}
WANG Cong7fccf032008-07-25 19:45:02 -0700609static void kimage_terminate(struct kimage *image)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700610{
Maneesh Soni72414d32005-06-25 14:58:28 -0700611 if (*image->entry != 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700612 image->entry++;
Maneesh Soni72414d32005-06-25 14:58:28 -0700613
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700614 *image->entry = IND_DONE;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700615}
616
617#define for_each_kimage_entry(image, ptr, entry) \
618 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
619 ptr = (entry & IND_INDIRECTION)? \
620 phys_to_virt((entry & PAGE_MASK)): ptr +1)
621
622static void kimage_free_entry(kimage_entry_t entry)
623{
624 struct page *page;
625
626 page = pfn_to_page(entry >> PAGE_SHIFT);
627 kimage_free_pages(page);
628}
629
630static void kimage_free(struct kimage *image)
631{
632 kimage_entry_t *ptr, entry;
633 kimage_entry_t ind = 0;
634
635 if (!image)
636 return;
Maneesh Soni72414d32005-06-25 14:58:28 -0700637
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700638 kimage_free_extra_pages(image);
639 for_each_kimage_entry(image, ptr, entry) {
640 if (entry & IND_INDIRECTION) {
641 /* Free the previous indirection page */
Maneesh Soni72414d32005-06-25 14:58:28 -0700642 if (ind & IND_INDIRECTION)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700643 kimage_free_entry(ind);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700644 /* Save this indirection page until we are
645 * done with it.
646 */
647 ind = entry;
648 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700649 else if (entry & IND_SOURCE)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700650 kimage_free_entry(entry);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700651 }
652 /* Free the final indirection page */
Maneesh Soni72414d32005-06-25 14:58:28 -0700653 if (ind & IND_INDIRECTION)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700654 kimage_free_entry(ind);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700655
656 /* Handle any machine specific cleanup */
657 machine_kexec_cleanup(image);
658
659 /* Free the kexec control pages... */
660 kimage_free_page_list(&image->control_pages);
661 kfree(image);
662}
663
Maneesh Soni72414d32005-06-25 14:58:28 -0700664static kimage_entry_t *kimage_dst_used(struct kimage *image,
665 unsigned long page)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700666{
667 kimage_entry_t *ptr, entry;
668 unsigned long destination = 0;
669
670 for_each_kimage_entry(image, ptr, entry) {
Maneesh Soni72414d32005-06-25 14:58:28 -0700671 if (entry & IND_DESTINATION)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700672 destination = entry & PAGE_MASK;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700673 else if (entry & IND_SOURCE) {
Maneesh Soni72414d32005-06-25 14:58:28 -0700674 if (page == destination)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700675 return ptr;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700676 destination += PAGE_SIZE;
677 }
678 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700679
Alexey Dobriyan314b6a42005-06-27 22:29:33 -0700680 return NULL;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700681}
682
Maneesh Soni72414d32005-06-25 14:58:28 -0700683static struct page *kimage_alloc_page(struct kimage *image,
Al Viro9796fdd2005-10-21 03:22:03 -0400684 gfp_t gfp_mask,
Maneesh Soni72414d32005-06-25 14:58:28 -0700685 unsigned long destination)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700686{
687 /*
688 * Here we implement safeguards to ensure that a source page
689 * is not copied to its destination page before the data on
690 * the destination page is no longer useful.
691 *
692 * To do this we maintain the invariant that a source page is
693 * either its own destination page, or it is not a
694 * destination page at all.
695 *
696 * That is slightly stronger than required, but the proof
697 * that no problems will not occur is trivial, and the
698 * implementation is simply to verify.
699 *
700 * When allocating all pages normally this algorithm will run
701 * in O(N) time, but in the worst case it will run in O(N^2)
702 * time. If the runtime is a problem the data structures can
703 * be fixed.
704 */
705 struct page *page;
706 unsigned long addr;
707
708 /*
709 * Walk through the list of destination pages, and see if I
710 * have a match.
711 */
712 list_for_each_entry(page, &image->dest_pages, lru) {
713 addr = page_to_pfn(page) << PAGE_SHIFT;
714 if (addr == destination) {
715 list_del(&page->lru);
716 return page;
717 }
718 }
719 page = NULL;
720 while (1) {
721 kimage_entry_t *old;
722
723 /* Allocate a page, if we run out of memory give up */
724 page = kimage_alloc_pages(gfp_mask, 0);
Maneesh Soni72414d32005-06-25 14:58:28 -0700725 if (!page)
Alexey Dobriyan314b6a42005-06-27 22:29:33 -0700726 return NULL;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700727 /* If the page cannot be used file it away */
Maneesh Soni72414d32005-06-25 14:58:28 -0700728 if (page_to_pfn(page) >
729 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700730 list_add(&page->lru, &image->unuseable_pages);
731 continue;
732 }
733 addr = page_to_pfn(page) << PAGE_SHIFT;
734
735 /* If it is the destination page we want use it */
736 if (addr == destination)
737 break;
738
739 /* If the page is not a destination page use it */
Maneesh Soni72414d32005-06-25 14:58:28 -0700740 if (!kimage_is_destination_range(image, addr,
741 addr + PAGE_SIZE))
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700742 break;
743
744 /*
745 * I know that the page is someones destination page.
746 * See if there is already a source page for this
747 * destination page. And if so swap the source pages.
748 */
749 old = kimage_dst_used(image, addr);
750 if (old) {
751 /* If so move it */
752 unsigned long old_addr;
753 struct page *old_page;
754
755 old_addr = *old & PAGE_MASK;
756 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
757 copy_highpage(page, old_page);
758 *old = addr | (*old & ~PAGE_MASK);
759
760 /* The old page I have found cannot be a
Jonathan Steelf9092f32008-09-22 13:57:45 -0700761 * destination page, so return it if it's
762 * gfp_flags honor the ones passed in.
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700763 */
Jonathan Steelf9092f32008-09-22 13:57:45 -0700764 if (!(gfp_mask & __GFP_HIGHMEM) &&
765 PageHighMem(old_page)) {
766 kimage_free_pages(old_page);
767 continue;
768 }
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700769 addr = old_addr;
770 page = old_page;
771 break;
772 }
773 else {
774 /* Place the page on the destination list I
775 * will use it later.
776 */
777 list_add(&page->lru, &image->dest_pages);
778 }
779 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700780
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700781 return page;
782}
783
784static int kimage_load_normal_segment(struct kimage *image,
Maneesh Soni72414d32005-06-25 14:58:28 -0700785 struct kexec_segment *segment)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700786{
787 unsigned long maddr;
788 unsigned long ubytes, mbytes;
789 int result;
Alexey Dobriyan314b6a42005-06-27 22:29:33 -0700790 unsigned char __user *buf;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700791
792 result = 0;
793 buf = segment->buf;
794 ubytes = segment->bufsz;
795 mbytes = segment->memsz;
796 maddr = segment->mem;
797
798 result = kimage_set_destination(image, maddr);
Maneesh Soni72414d32005-06-25 14:58:28 -0700799 if (result < 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700800 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -0700801
802 while (mbytes) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700803 struct page *page;
804 char *ptr;
805 size_t uchunk, mchunk;
Maneesh Soni72414d32005-06-25 14:58:28 -0700806
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700807 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700808 if (!page) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700809 result = -ENOMEM;
810 goto out;
811 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700812 result = kimage_add_page(image, page_to_pfn(page)
813 << PAGE_SHIFT);
814 if (result < 0)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700815 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -0700816
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700817 ptr = kmap(page);
818 /* Start with a clear page */
Jan Beulich3ecb01d2010-10-26 14:22:27 -0700819 clear_page(ptr);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700820 ptr += maddr & ~PAGE_MASK;
821 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
Maneesh Soni72414d32005-06-25 14:58:28 -0700822 if (mchunk > mbytes)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700823 mchunk = mbytes;
Maneesh Soni72414d32005-06-25 14:58:28 -0700824
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700825 uchunk = mchunk;
Maneesh Soni72414d32005-06-25 14:58:28 -0700826 if (uchunk > ubytes)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700827 uchunk = ubytes;
Maneesh Soni72414d32005-06-25 14:58:28 -0700828
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700829 result = copy_from_user(ptr, buf, uchunk);
830 kunmap(page);
831 if (result) {
Dan Carpenterf65a03f2010-08-10 18:03:31 -0700832 result = -EFAULT;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700833 goto out;
834 }
835 ubytes -= uchunk;
836 maddr += mchunk;
837 buf += mchunk;
838 mbytes -= mchunk;
839 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700840out:
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700841 return result;
842}
843
844static int kimage_load_crash_segment(struct kimage *image,
Maneesh Soni72414d32005-06-25 14:58:28 -0700845 struct kexec_segment *segment)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700846{
847 /* For crash dumps kernels we simply copy the data from
848 * user space to it's destination.
849 * We do things a page at a time for the sake of kmap.
850 */
851 unsigned long maddr;
852 unsigned long ubytes, mbytes;
853 int result;
Alexey Dobriyan314b6a42005-06-27 22:29:33 -0700854 unsigned char __user *buf;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700855
856 result = 0;
857 buf = segment->buf;
858 ubytes = segment->bufsz;
859 mbytes = segment->memsz;
860 maddr = segment->mem;
Maneesh Soni72414d32005-06-25 14:58:28 -0700861 while (mbytes) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700862 struct page *page;
863 char *ptr;
864 size_t uchunk, mchunk;
Maneesh Soni72414d32005-06-25 14:58:28 -0700865
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700866 page = pfn_to_page(maddr >> PAGE_SHIFT);
Stephen Hemmingerc80544d2007-10-18 03:07:05 -0700867 if (!page) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700868 result = -ENOMEM;
869 goto out;
870 }
871 ptr = kmap(page);
872 ptr += maddr & ~PAGE_MASK;
873 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
Maneesh Soni72414d32005-06-25 14:58:28 -0700874 if (mchunk > mbytes)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700875 mchunk = mbytes;
Maneesh Soni72414d32005-06-25 14:58:28 -0700876
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700877 uchunk = mchunk;
878 if (uchunk > ubytes) {
879 uchunk = ubytes;
880 /* Zero the trailing part of the page */
881 memset(ptr + uchunk, 0, mchunk - uchunk);
882 }
883 result = copy_from_user(ptr, buf, uchunk);
Zou Nan haia79561132006-12-07 09:51:35 -0800884 kexec_flush_icache_page(page);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700885 kunmap(page);
886 if (result) {
Dan Carpenterf65a03f2010-08-10 18:03:31 -0700887 result = -EFAULT;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700888 goto out;
889 }
890 ubytes -= uchunk;
891 maddr += mchunk;
892 buf += mchunk;
893 mbytes -= mchunk;
894 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700895out:
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700896 return result;
897}
898
899static int kimage_load_segment(struct kimage *image,
Maneesh Soni72414d32005-06-25 14:58:28 -0700900 struct kexec_segment *segment)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700901{
902 int result = -ENOMEM;
Maneesh Soni72414d32005-06-25 14:58:28 -0700903
904 switch (image->type) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700905 case KEXEC_TYPE_DEFAULT:
906 result = kimage_load_normal_segment(image, segment);
907 break;
908 case KEXEC_TYPE_CRASH:
909 result = kimage_load_crash_segment(image, segment);
910 break;
911 }
Maneesh Soni72414d32005-06-25 14:58:28 -0700912
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700913 return result;
914}
915
916/*
917 * Exec Kernel system call: for obvious reasons only root may call it.
918 *
919 * This call breaks up into three pieces.
920 * - A generic part which loads the new kernel from the current
921 * address space, and very carefully places the data in the
922 * allocated pages.
923 *
924 * - A generic part that interacts with the kernel and tells all of
925 * the devices to shut down. Preventing on-going dmas, and placing
926 * the devices in a consistent state so a later kernel can
927 * reinitialize them.
928 *
929 * - A machine specific part that includes the syscall number
930 * and the copies the image to it's final destination. And
931 * jumps into the image at entry.
932 *
933 * kexec does not sync, or unmount filesystems so if you need
934 * that to happen you need to do that yourself.
935 */
Jeff Moyerc330dda2006-06-23 02:05:07 -0700936struct kimage *kexec_image;
937struct kimage *kexec_crash_image;
Andrew Morton8c5a1cf2008-08-15 00:40:27 -0700938
939static DEFINE_MUTEX(kexec_mutex);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700940
Heiko Carstens754fe8d2009-01-14 14:14:09 +0100941SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
942 struct kexec_segment __user *, segments, unsigned long, flags)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700943{
944 struct kimage **dest_image, *image;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700945 int result;
946
947 /* We only trust the superuser with rebooting the system. */
948 if (!capable(CAP_SYS_BOOT))
949 return -EPERM;
950
951 /*
952 * Verify we have a legal set of flags
953 * This leaves us room for future extensions.
954 */
955 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
956 return -EINVAL;
957
958 /* Verify we are on the appropriate architecture */
959 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
960 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700961 return -EINVAL;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700962
963 /* Put an artificial cap on the number
964 * of segments passed to kexec_load.
965 */
966 if (nr_segments > KEXEC_SEGMENT_MAX)
967 return -EINVAL;
968
969 image = NULL;
970 result = 0;
971
972 /* Because we write directly to the reserved memory
973 * region when loading crash kernels we need a mutex here to
974 * prevent multiple crash kernels from attempting to load
975 * simultaneously, and to prevent a crash kernel from loading
976 * over the top of a in use crash kernel.
977 *
978 * KISS: always take the mutex.
979 */
Andrew Morton8c5a1cf2008-08-15 00:40:27 -0700980 if (!mutex_trylock(&kexec_mutex))
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700981 return -EBUSY;
Maneesh Soni72414d32005-06-25 14:58:28 -0700982
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700983 dest_image = &kexec_image;
Maneesh Soni72414d32005-06-25 14:58:28 -0700984 if (flags & KEXEC_ON_CRASH)
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700985 dest_image = &kexec_crash_image;
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700986 if (nr_segments > 0) {
987 unsigned long i;
Maneesh Soni72414d32005-06-25 14:58:28 -0700988
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700989 /* Loading another kernel to reboot into */
Maneesh Soni72414d32005-06-25 14:58:28 -0700990 if ((flags & KEXEC_ON_CRASH) == 0)
991 result = kimage_normal_alloc(&image, entry,
992 nr_segments, segments);
Eric W. Biedermandc009d92005-06-25 14:57:52 -0700993 /* Loading another kernel to switch to if this one crashes */
994 else if (flags & KEXEC_ON_CRASH) {
995 /* Free any current crash dump kernel before
996 * we corrupt it.
997 */
998 kimage_free(xchg(&kexec_crash_image, NULL));
Maneesh Soni72414d32005-06-25 14:58:28 -0700999 result = kimage_crash_alloc(&image, entry,
1000 nr_segments, segments);
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001001 }
Maneesh Soni72414d32005-06-25 14:58:28 -07001002 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001003 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -07001004
Huang Ying3ab83522008-07-25 19:45:07 -07001005 if (flags & KEXEC_PRESERVE_CONTEXT)
1006 image->preserve_context = 1;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001007 result = machine_kexec_prepare(image);
Maneesh Soni72414d32005-06-25 14:58:28 -07001008 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001009 goto out;
Maneesh Soni72414d32005-06-25 14:58:28 -07001010
1011 for (i = 0; i < nr_segments; i++) {
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001012 result = kimage_load_segment(image, &image->segment[i]);
Maneesh Soni72414d32005-06-25 14:58:28 -07001013 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001014 goto out;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001015 }
WANG Cong7fccf032008-07-25 19:45:02 -07001016 kimage_terminate(image);
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001017 }
1018 /* Install the new kernel, and Uninstall the old */
1019 image = xchg(dest_image, image);
1020
Maneesh Soni72414d32005-06-25 14:58:28 -07001021out:
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001022 mutex_unlock(&kexec_mutex);
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001023 kimage_free(image);
Maneesh Soni72414d32005-06-25 14:58:28 -07001024
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001025 return result;
1026}
1027
1028#ifdef CONFIG_COMPAT
1029asmlinkage long compat_sys_kexec_load(unsigned long entry,
Maneesh Soni72414d32005-06-25 14:58:28 -07001030 unsigned long nr_segments,
1031 struct compat_kexec_segment __user *segments,
1032 unsigned long flags)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001033{
1034 struct compat_kexec_segment in;
1035 struct kexec_segment out, __user *ksegments;
1036 unsigned long i, result;
1037
1038 /* Don't allow clients that don't understand the native
1039 * architecture to do anything.
1040 */
Maneesh Soni72414d32005-06-25 14:58:28 -07001041 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001042 return -EINVAL;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001043
Maneesh Soni72414d32005-06-25 14:58:28 -07001044 if (nr_segments > KEXEC_SEGMENT_MAX)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001045 return -EINVAL;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001046
1047 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1048 for (i=0; i < nr_segments; i++) {
1049 result = copy_from_user(&in, &segments[i], sizeof(in));
Maneesh Soni72414d32005-06-25 14:58:28 -07001050 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001051 return -EFAULT;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001052
1053 out.buf = compat_ptr(in.buf);
1054 out.bufsz = in.bufsz;
1055 out.mem = in.mem;
1056 out.memsz = in.memsz;
1057
1058 result = copy_to_user(&ksegments[i], &out, sizeof(out));
Maneesh Soni72414d32005-06-25 14:58:28 -07001059 if (result)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001060 return -EFAULT;
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001061 }
1062
1063 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1064}
1065#endif
1066
Alexander Nyberg6e274d12005-06-25 14:58:26 -07001067void crash_kexec(struct pt_regs *regs)
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001068{
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001069 /* Take the kexec_mutex here to prevent sys_kexec_load
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001070 * running on one cpu from replacing the crash kernel
1071 * we are using after a panic on a different cpu.
1072 *
1073 * If the crash kernel was not located in a fixed area
1074 * of memory the xchg(&kexec_crash_image) would be
1075 * sufficient. But since I reuse the memory...
1076 */
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001077 if (mutex_trylock(&kexec_mutex)) {
David Wilderc0ce7d02006-06-23 15:29:34 -07001078 if (kexec_crash_image) {
Vivek Goyale996e582006-01-09 20:51:44 -08001079 struct pt_regs fixed_regs;
KOSAKI Motohiro0f4bd462009-12-22 03:15:43 +00001080
1081 kmsg_dump(KMSG_DUMP_KEXEC);
1082
Vivek Goyale996e582006-01-09 20:51:44 -08001083 crash_setup_regs(&fixed_regs, regs);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001084 crash_save_vmcoreinfo();
Vivek Goyale996e582006-01-09 20:51:44 -08001085 machine_crash_shutdown(&fixed_regs);
David Wilderc0ce7d02006-06-23 15:29:34 -07001086 machine_kexec(kexec_crash_image);
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001087 }
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001088 mutex_unlock(&kexec_mutex);
Eric W. Biedermandc009d92005-06-25 14:57:52 -07001089 }
1090}
Vivek Goyalcc571652006-01-09 20:51:41 -08001091
Amerigo Wang06a7f712009-12-15 16:47:46 -08001092size_t crash_get_memory_size(void)
1093{
Pavan Naregundie05bd332010-06-29 15:05:28 -07001094 size_t size = 0;
Amerigo Wang06a7f712009-12-15 16:47:46 -08001095 mutex_lock(&kexec_mutex);
Pavan Naregundie05bd332010-06-29 15:05:28 -07001096 if (crashk_res.end != crashk_res.start)
1097 size = crashk_res.end - crashk_res.start + 1;
Amerigo Wang06a7f712009-12-15 16:47:46 -08001098 mutex_unlock(&kexec_mutex);
1099 return size;
1100}
1101
1102static void free_reserved_phys_range(unsigned long begin, unsigned long end)
1103{
1104 unsigned long addr;
1105
1106 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1107 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1108 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1109 free_page((unsigned long)__va(addr));
1110 totalram_pages++;
1111 }
1112}
1113
1114int crash_shrink_memory(unsigned long new_size)
1115{
1116 int ret = 0;
1117 unsigned long start, end;
1118
1119 mutex_lock(&kexec_mutex);
1120
1121 if (kexec_crash_image) {
1122 ret = -ENOENT;
1123 goto unlock;
1124 }
1125 start = crashk_res.start;
1126 end = crashk_res.end;
1127
1128 if (new_size >= end - start + 1) {
1129 ret = -EINVAL;
1130 if (new_size == end - start + 1)
1131 ret = 0;
1132 goto unlock;
1133 }
1134
1135 start = roundup(start, PAGE_SIZE);
1136 end = roundup(start + new_size, PAGE_SIZE);
1137
1138 free_reserved_phys_range(end, crashk_res.end);
1139
Pavan Naregundie05bd332010-06-29 15:05:28 -07001140 if ((start == end) && (crashk_res.parent != NULL))
Amerigo Wang06a7f712009-12-15 16:47:46 -08001141 release_resource(&crashk_res);
Vitaly Mayatskikh475f9aa62010-05-11 14:06:51 -07001142 crashk_res.end = end - 1;
Amerigo Wang06a7f712009-12-15 16:47:46 -08001143
1144unlock:
1145 mutex_unlock(&kexec_mutex);
1146 return ret;
1147}
1148
Magnus Damm85916f82006-12-06 20:40:41 -08001149static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1150 size_t data_len)
1151{
1152 struct elf_note note;
1153
1154 note.n_namesz = strlen(name) + 1;
1155 note.n_descsz = data_len;
1156 note.n_type = type;
1157 memcpy(buf, &note, sizeof(note));
1158 buf += (sizeof(note) + 3)/4;
1159 memcpy(buf, name, note.n_namesz);
1160 buf += (note.n_namesz + 3)/4;
1161 memcpy(buf, data, note.n_descsz);
1162 buf += (note.n_descsz + 3)/4;
1163
1164 return buf;
1165}
1166
1167static void final_note(u32 *buf)
1168{
1169 struct elf_note note;
1170
1171 note.n_namesz = 0;
1172 note.n_descsz = 0;
1173 note.n_type = 0;
1174 memcpy(buf, &note, sizeof(note));
1175}
1176
1177void crash_save_cpu(struct pt_regs *regs, int cpu)
1178{
1179 struct elf_prstatus prstatus;
1180 u32 *buf;
1181
Rusty Russell4f4b6c12009-01-01 10:12:15 +10301182 if ((cpu < 0) || (cpu >= nr_cpu_ids))
Magnus Damm85916f82006-12-06 20:40:41 -08001183 return;
1184
1185 /* Using ELF notes here is opportunistic.
1186 * I need a well defined structure format
1187 * for the data I pass, and I need tags
1188 * on the data to indicate what information I have
1189 * squirrelled away. ELF notes happen to provide
1190 * all of that, so there is no need to invent something new.
1191 */
1192 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1193 if (!buf)
1194 return;
1195 memset(&prstatus, 0, sizeof(prstatus));
1196 prstatus.pr_pid = current->pid;
Tejun Heo6cd61c0b2009-02-09 22:17:39 +09001197 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
Simon Horman6672f762007-05-08 00:28:22 -07001198 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1199 &prstatus, sizeof(prstatus));
Magnus Damm85916f82006-12-06 20:40:41 -08001200 final_note(buf);
1201}
1202
Vivek Goyalcc571652006-01-09 20:51:41 -08001203static int __init crash_notes_memory_init(void)
1204{
1205 /* Allocate memory for saving cpu registers. */
1206 crash_notes = alloc_percpu(note_buf_t);
1207 if (!crash_notes) {
1208 printk("Kexec: Memory allocation for saving cpu register"
1209 " states failed\n");
1210 return -ENOMEM;
1211 }
1212 return 0;
1213}
1214module_init(crash_notes_memory_init)
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001215
Bernhard Wallecba63c32007-10-18 23:40:58 -07001216
1217/*
1218 * parsing the "crashkernel" commandline
1219 *
1220 * this code is intended to be called from architecture specific code
1221 */
1222
1223
1224/*
1225 * This function parses command lines in the format
1226 *
1227 * crashkernel=ramsize-range:size[,...][@offset]
1228 *
1229 * The function returns 0 on success and -EINVAL on failure.
1230 */
1231static int __init parse_crashkernel_mem(char *cmdline,
1232 unsigned long long system_ram,
1233 unsigned long long *crash_size,
1234 unsigned long long *crash_base)
1235{
1236 char *cur = cmdline, *tmp;
1237
1238 /* for each entry of the comma-separated list */
1239 do {
1240 unsigned long long start, end = ULLONG_MAX, size;
1241
1242 /* get the start of the range */
1243 start = memparse(cur, &tmp);
1244 if (cur == tmp) {
1245 pr_warning("crashkernel: Memory value expected\n");
1246 return -EINVAL;
1247 }
1248 cur = tmp;
1249 if (*cur != '-') {
1250 pr_warning("crashkernel: '-' expected\n");
1251 return -EINVAL;
1252 }
1253 cur++;
1254
1255 /* if no ':' is here, than we read the end */
1256 if (*cur != ':') {
1257 end = memparse(cur, &tmp);
1258 if (cur == tmp) {
1259 pr_warning("crashkernel: Memory "
1260 "value expected\n");
1261 return -EINVAL;
1262 }
1263 cur = tmp;
1264 if (end <= start) {
1265 pr_warning("crashkernel: end <= start\n");
1266 return -EINVAL;
1267 }
1268 }
1269
1270 if (*cur != ':') {
1271 pr_warning("crashkernel: ':' expected\n");
1272 return -EINVAL;
1273 }
1274 cur++;
1275
1276 size = memparse(cur, &tmp);
1277 if (cur == tmp) {
1278 pr_warning("Memory value expected\n");
1279 return -EINVAL;
1280 }
1281 cur = tmp;
1282 if (size >= system_ram) {
1283 pr_warning("crashkernel: invalid size\n");
1284 return -EINVAL;
1285 }
1286
1287 /* match ? */
Michael Ellermanbe089d792008-05-01 04:34:49 -07001288 if (system_ram >= start && system_ram < end) {
Bernhard Wallecba63c32007-10-18 23:40:58 -07001289 *crash_size = size;
1290 break;
1291 }
1292 } while (*cur++ == ',');
1293
1294 if (*crash_size > 0) {
Hidetoshi Seto11c7da42009-07-29 15:02:08 -07001295 while (*cur && *cur != ' ' && *cur != '@')
Bernhard Wallecba63c32007-10-18 23:40:58 -07001296 cur++;
1297 if (*cur == '@') {
1298 cur++;
1299 *crash_base = memparse(cur, &tmp);
1300 if (cur == tmp) {
1301 pr_warning("Memory value expected "
1302 "after '@'\n");
1303 return -EINVAL;
1304 }
1305 }
1306 }
1307
1308 return 0;
1309}
1310
1311/*
1312 * That function parses "simple" (old) crashkernel command lines like
1313 *
1314 * crashkernel=size[@offset]
1315 *
1316 * It returns 0 on success and -EINVAL on failure.
1317 */
1318static int __init parse_crashkernel_simple(char *cmdline,
1319 unsigned long long *crash_size,
1320 unsigned long long *crash_base)
1321{
1322 char *cur = cmdline;
1323
1324 *crash_size = memparse(cmdline, &cur);
1325 if (cmdline == cur) {
1326 pr_warning("crashkernel: memory value expected\n");
1327 return -EINVAL;
1328 }
1329
1330 if (*cur == '@')
1331 *crash_base = memparse(cur+1, &cur);
1332
1333 return 0;
1334}
1335
1336/*
1337 * That function is the entry point for command line parsing and should be
1338 * called from the arch-specific code.
1339 */
1340int __init parse_crashkernel(char *cmdline,
1341 unsigned long long system_ram,
1342 unsigned long long *crash_size,
1343 unsigned long long *crash_base)
1344{
1345 char *p = cmdline, *ck_cmdline = NULL;
1346 char *first_colon, *first_space;
1347
1348 BUG_ON(!crash_size || !crash_base);
1349 *crash_size = 0;
1350 *crash_base = 0;
1351
1352 /* find crashkernel and use the last one if there are more */
1353 p = strstr(p, "crashkernel=");
1354 while (p) {
1355 ck_cmdline = p;
1356 p = strstr(p+1, "crashkernel=");
1357 }
1358
1359 if (!ck_cmdline)
1360 return -EINVAL;
1361
1362 ck_cmdline += 12; /* strlen("crashkernel=") */
1363
1364 /*
1365 * if the commandline contains a ':', then that's the extended
1366 * syntax -- if not, it must be the classic syntax
1367 */
1368 first_colon = strchr(ck_cmdline, ':');
1369 first_space = strchr(ck_cmdline, ' ');
1370 if (first_colon && (!first_space || first_colon < first_space))
1371 return parse_crashkernel_mem(ck_cmdline, system_ram,
1372 crash_size, crash_base);
1373 else
1374 return parse_crashkernel_simple(ck_cmdline, crash_size,
1375 crash_base);
1376
1377 return 0;
1378}
1379
1380
1381
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001382void crash_save_vmcoreinfo(void)
1383{
1384 u32 *buf;
1385
1386 if (!vmcoreinfo_size)
1387 return;
1388
Ken'ichi Ohmichid7682812007-10-16 23:27:28 -07001389 vmcoreinfo_append_str("CRASHTIME=%ld", get_seconds());
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001390
1391 buf = (u32 *)vmcoreinfo_note;
1392
1393 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1394 vmcoreinfo_size);
1395
1396 final_note(buf);
1397}
1398
1399void vmcoreinfo_append_str(const char *fmt, ...)
1400{
1401 va_list args;
1402 char buf[0x50];
1403 int r;
1404
1405 va_start(args, fmt);
1406 r = vsnprintf(buf, sizeof(buf), fmt, args);
1407 va_end(args);
1408
1409 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1410 r = vmcoreinfo_max_size - vmcoreinfo_size;
1411
1412 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1413
1414 vmcoreinfo_size += r;
1415}
1416
1417/*
1418 * provide an empty default implementation here -- architecture
1419 * code may override this
1420 */
1421void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1422{}
1423
1424unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1425{
1426 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1427}
1428
1429static int __init crash_save_vmcoreinfo_init(void)
1430{
Ken'ichi Ohmichibba1f602008-02-07 00:15:22 -08001431 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1432 VMCOREINFO_PAGESIZE(PAGE_SIZE);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001433
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001434 VMCOREINFO_SYMBOL(init_uts_ns);
1435 VMCOREINFO_SYMBOL(node_online_map);
1436 VMCOREINFO_SYMBOL(swapper_pg_dir);
1437 VMCOREINFO_SYMBOL(_stext);
Ken'ichi Ohmichiacd99db2008-10-18 20:28:30 -07001438 VMCOREINFO_SYMBOL(vmlist);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001439
1440#ifndef CONFIG_NEED_MULTIPLE_NODES
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001441 VMCOREINFO_SYMBOL(mem_map);
1442 VMCOREINFO_SYMBOL(contig_page_data);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001443#endif
1444#ifdef CONFIG_SPARSEMEM
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001445 VMCOREINFO_SYMBOL(mem_section);
1446 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
Ken'ichi Ohmichic76f8602008-02-07 00:15:20 -08001447 VMCOREINFO_STRUCT_SIZE(mem_section);
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001448 VMCOREINFO_OFFSET(mem_section, section_mem_map);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001449#endif
Ken'ichi Ohmichic76f8602008-02-07 00:15:20 -08001450 VMCOREINFO_STRUCT_SIZE(page);
1451 VMCOREINFO_STRUCT_SIZE(pglist_data);
1452 VMCOREINFO_STRUCT_SIZE(zone);
1453 VMCOREINFO_STRUCT_SIZE(free_area);
1454 VMCOREINFO_STRUCT_SIZE(list_head);
1455 VMCOREINFO_SIZE(nodemask_t);
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001456 VMCOREINFO_OFFSET(page, flags);
1457 VMCOREINFO_OFFSET(page, _count);
1458 VMCOREINFO_OFFSET(page, mapping);
1459 VMCOREINFO_OFFSET(page, lru);
1460 VMCOREINFO_OFFSET(pglist_data, node_zones);
1461 VMCOREINFO_OFFSET(pglist_data, nr_zones);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001462#ifdef CONFIG_FLAT_NODE_MEM_MAP
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001463 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001464#endif
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001465 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1466 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1467 VMCOREINFO_OFFSET(pglist_data, node_id);
1468 VMCOREINFO_OFFSET(zone, free_area);
1469 VMCOREINFO_OFFSET(zone, vm_stat);
1470 VMCOREINFO_OFFSET(zone, spanned_pages);
1471 VMCOREINFO_OFFSET(free_area, free_list);
1472 VMCOREINFO_OFFSET(list_head, next);
1473 VMCOREINFO_OFFSET(list_head, prev);
Ken'ichi Ohmichiacd99db2008-10-18 20:28:30 -07001474 VMCOREINFO_OFFSET(vm_struct, addr);
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001475 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
Neil Horman04d491a2009-04-02 16:58:57 -07001476 log_buf_kexec_setup();
Ken'ichi Ohmichi83a08e72008-01-08 15:33:05 -08001477 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
Ken'ichi Ohmichibcbba6c2007-10-16 23:27:30 -07001478 VMCOREINFO_NUMBER(NR_FREE_PAGES);
Ken'ichi Ohmichi122c7a52008-04-28 02:13:04 -07001479 VMCOREINFO_NUMBER(PG_lru);
1480 VMCOREINFO_NUMBER(PG_private);
1481 VMCOREINFO_NUMBER(PG_swapcache);
Ken'ichi Ohmichifd59d232007-10-16 23:27:27 -07001482
1483 arch_crash_save_vmcoreinfo();
1484
1485 return 0;
1486}
1487
1488module_init(crash_save_vmcoreinfo_init)
Huang Ying3ab83522008-07-25 19:45:07 -07001489
Huang Ying7ade3fc2008-08-15 00:40:21 -07001490/*
1491 * Move into place and start executing a preloaded standalone
1492 * executable. If nothing was preloaded return an error.
Huang Ying3ab83522008-07-25 19:45:07 -07001493 */
1494int kernel_kexec(void)
1495{
1496 int error = 0;
1497
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001498 if (!mutex_trylock(&kexec_mutex))
Huang Ying3ab83522008-07-25 19:45:07 -07001499 return -EBUSY;
1500 if (!kexec_image) {
1501 error = -EINVAL;
1502 goto Unlock;
1503 }
1504
Huang Ying3ab83522008-07-25 19:45:07 -07001505#ifdef CONFIG_KEXEC_JUMP
Huang Ying7ade3fc2008-08-15 00:40:21 -07001506 if (kexec_image->preserve_context) {
Huang Ying89081d12008-07-25 19:45:10 -07001507 mutex_lock(&pm_mutex);
1508 pm_prepare_console();
1509 error = freeze_processes();
1510 if (error) {
1511 error = -EBUSY;
1512 goto Restore_console;
1513 }
1514 suspend_console();
Alan Sternd1616302009-05-24 22:05:42 +02001515 error = dpm_suspend_start(PMSG_FREEZE);
Huang Ying89081d12008-07-25 19:45:10 -07001516 if (error)
1517 goto Resume_console;
Alan Sternd1616302009-05-24 22:05:42 +02001518 /* At this point, dpm_suspend_start() has been called,
1519 * but *not* dpm_suspend_noirq(). We *must* call
1520 * dpm_suspend_noirq() now. Otherwise, drivers for
Huang Ying89081d12008-07-25 19:45:10 -07001521 * some devices (e.g. interrupt controllers) become
1522 * desynchronized with the actual state of the
1523 * hardware at resume time, and evil weirdness ensues.
1524 */
Alan Sternd1616302009-05-24 22:05:42 +02001525 error = dpm_suspend_noirq(PMSG_FREEZE);
Huang Ying89081d12008-07-25 19:45:10 -07001526 if (error)
Rafael J. Wysocki749b0af2009-03-16 22:34:35 +01001527 goto Resume_devices;
1528 error = disable_nonboot_cpus();
1529 if (error)
1530 goto Enable_cpus;
Rafael J. Wysocki2ed8d2b2009-03-16 22:34:06 +01001531 local_irq_disable();
Rafael J. Wysocki770824b2009-02-22 18:38:50 +01001532 /* Suspend system devices */
1533 error = sysdev_suspend(PMSG_FREEZE);
1534 if (error)
Rafael J. Wysocki749b0af2009-03-16 22:34:35 +01001535 goto Enable_irqs;
Huang Ying7ade3fc2008-08-15 00:40:21 -07001536 } else
Huang Ying3ab83522008-07-25 19:45:07 -07001537#endif
Huang Ying7ade3fc2008-08-15 00:40:21 -07001538 {
Huang Yingca195b72008-08-15 00:40:24 -07001539 kernel_restart_prepare(NULL);
Huang Ying3ab83522008-07-25 19:45:07 -07001540 printk(KERN_EMERG "Starting new kernel\n");
1541 machine_shutdown();
1542 }
1543
1544 machine_kexec(kexec_image);
1545
Huang Ying3ab83522008-07-25 19:45:07 -07001546#ifdef CONFIG_KEXEC_JUMP
Huang Ying7ade3fc2008-08-15 00:40:21 -07001547 if (kexec_image->preserve_context) {
Rafael J. Wysocki770824b2009-02-22 18:38:50 +01001548 sysdev_resume();
Rafael J. Wysocki749b0af2009-03-16 22:34:35 +01001549 Enable_irqs:
Huang Ying3ab83522008-07-25 19:45:07 -07001550 local_irq_enable();
Rafael J. Wysocki749b0af2009-03-16 22:34:35 +01001551 Enable_cpus:
Huang Ying89081d12008-07-25 19:45:10 -07001552 enable_nonboot_cpus();
Alan Sternd1616302009-05-24 22:05:42 +02001553 dpm_resume_noirq(PMSG_RESTORE);
Huang Ying89081d12008-07-25 19:45:10 -07001554 Resume_devices:
Alan Sternd1616302009-05-24 22:05:42 +02001555 dpm_resume_end(PMSG_RESTORE);
Huang Ying89081d12008-07-25 19:45:10 -07001556 Resume_console:
1557 resume_console();
1558 thaw_processes();
1559 Restore_console:
1560 pm_restore_console();
1561 mutex_unlock(&pm_mutex);
Huang Ying3ab83522008-07-25 19:45:07 -07001562 }
Huang Ying7ade3fc2008-08-15 00:40:21 -07001563#endif
Huang Ying3ab83522008-07-25 19:45:07 -07001564
1565 Unlock:
Andrew Morton8c5a1cf2008-08-15 00:40:27 -07001566 mutex_unlock(&kexec_mutex);
Huang Ying3ab83522008-07-25 19:45:07 -07001567 return error;
1568}