Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * kexec.c - kexec_load system call
4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/capability.h>
10#include <linux/mm.h>
11#include <linux/file.h>
12#include <linux/security.h>
13#include <linux/kexec.h>
14#include <linux/mutex.h>
15#include <linux/list.h>
16#include <linux/syscalls.h>
17#include <linux/vmalloc.h>
18#include <linux/slab.h>
19
20#include "kexec_internal.h"
21
22static int copy_user_segment_list(struct kimage *image,
23 unsigned long nr_segments,
24 struct kexec_segment __user *segments)
25{
26 int ret;
27 size_t segment_bytes;
28
29 /* Read in the segments */
30 image->nr_segments = nr_segments;
31 segment_bytes = nr_segments * sizeof(*segments);
32 ret = copy_from_user(image->segment, segments, segment_bytes);
33 if (ret)
34 ret = -EFAULT;
35
36 return ret;
37}
38
39static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
40 unsigned long nr_segments,
41 struct kexec_segment __user *segments,
42 unsigned long flags)
43{
44 int ret;
45 struct kimage *image;
46 bool kexec_on_panic = flags & KEXEC_ON_CRASH;
47
48 if (kexec_on_panic) {
49 /* Verify we have a valid entry point */
50 if ((entry < phys_to_boot_phys(crashk_res.start)) ||
51 (entry > phys_to_boot_phys(crashk_res.end)))
52 return -EADDRNOTAVAIL;
53 }
54
55 /* Allocate and initialize a controlling structure */
56 image = do_kimage_alloc_init();
57 if (!image)
58 return -ENOMEM;
59
60 image->start = entry;
61
62 ret = copy_user_segment_list(image, nr_segments, segments);
63 if (ret)
64 goto out_free_image;
65
66 if (kexec_on_panic) {
67 /* Enable special crash kernel control page alloc policy. */
68 image->control_page = crashk_res.start;
69 image->type = KEXEC_TYPE_CRASH;
70 }
71
72 ret = sanity_check_segment_list(image);
73 if (ret)
74 goto out_free_image;
75
76 /*
77 * Find a location for the control code buffer, and add it
78 * the vector of segments so that it's pages will also be
79 * counted as destination pages.
80 */
81 ret = -ENOMEM;
82 image->control_code_page = kimage_alloc_control_pages(image,
83 get_order(KEXEC_CONTROL_PAGE_SIZE));
84 if (!image->control_code_page) {
85 pr_err("Could not allocate control_code_buffer\n");
86 goto out_free_image;
87 }
88
89 if (!kexec_on_panic) {
90 image->swap_page = kimage_alloc_control_pages(image, 0);
91 if (!image->swap_page) {
92 pr_err("Could not allocate swap buffer\n");
93 goto out_free_control_pages;
94 }
95 }
96
97 *rimage = image;
98 return 0;
99out_free_control_pages:
100 kimage_free_page_list(&image->control_pages);
101out_free_image:
102 kfree(image);
103 return ret;
104}
105
106static int do_kexec_load(unsigned long entry, unsigned long nr_segments,
107 struct kexec_segment __user *segments, unsigned long flags)
108{
109 struct kimage **dest_image, *image;
110 unsigned long i;
111 int ret;
112
113 if (flags & KEXEC_ON_CRASH) {
114 dest_image = &kexec_crash_image;
115 if (kexec_crash_image)
116 arch_kexec_unprotect_crashkres();
117 } else {
118 dest_image = &kexec_image;
119 }
120
121 if (nr_segments == 0) {
122 /* Uninstall image */
123 kimage_free(xchg(dest_image, NULL));
124 return 0;
125 }
126 if (flags & KEXEC_ON_CRASH) {
127 /*
128 * Loading another kernel to switch to if this one
129 * crashes. Free any current crash dump kernel before
130 * we corrupt it.
131 */
132 kimage_free(xchg(&kexec_crash_image, NULL));
133 }
134
135 ret = kimage_alloc_init(&image, entry, nr_segments, segments, flags);
136 if (ret)
137 return ret;
138
139 if (flags & KEXEC_PRESERVE_CONTEXT)
140 image->preserve_context = 1;
141
142 ret = machine_kexec_prepare(image);
143 if (ret)
144 goto out;
145
146 /*
147 * Some architecture(like S390) may touch the crash memory before
148 * machine_kexec_prepare(), we must copy vmcoreinfo data after it.
149 */
150 ret = kimage_crash_copy_vmcoreinfo(image);
151 if (ret)
152 goto out;
153
154 for (i = 0; i < nr_segments; i++) {
155 ret = kimage_load_segment(image, &image->segment[i]);
156 if (ret)
157 goto out;
158 }
159
160 kimage_terminate(image);
161
162 ret = machine_kexec_post_load(image);
163 if (ret)
164 goto out;
165
166 /* Install the new kernel and uninstall the old */
167 image = xchg(dest_image, image);
168
169out:
170 if ((flags & KEXEC_ON_CRASH) && kexec_crash_image)
171 arch_kexec_protect_crashkres();
172
173 kimage_free(image);
174 return ret;
175}
176
177/*
178 * Exec Kernel system call: for obvious reasons only root may call it.
179 *
180 * This call breaks up into three pieces.
181 * - A generic part which loads the new kernel from the current
182 * address space, and very carefully places the data in the
183 * allocated pages.
184 *
185 * - A generic part that interacts with the kernel and tells all of
186 * the devices to shut down. Preventing on-going dmas, and placing
187 * the devices in a consistent state so a later kernel can
188 * reinitialize them.
189 *
190 * - A machine specific part that includes the syscall number
191 * and then copies the image to it's final destination. And
192 * jumps into the image at entry.
193 *
194 * kexec does not sync, or unmount filesystems so if you need
195 * that to happen you need to do that yourself.
196 */
197
198static inline int kexec_load_check(unsigned long nr_segments,
199 unsigned long flags)
200{
201 int result;
202
203 /* We only trust the superuser with rebooting the system. */
204 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
205 return -EPERM;
206
207 /* Permit LSMs and IMA to fail the kexec */
208 result = security_kernel_load_data(LOADING_KEXEC_IMAGE);
209 if (result < 0)
210 return result;
211
212 /*
213 * kexec can be used to circumvent module loading restrictions, so
214 * prevent loading in that case
215 */
216 result = security_locked_down(LOCKDOWN_KEXEC);
217 if (result)
218 return result;
219
220 /*
221 * Verify we have a legal set of flags
222 * This leaves us room for future extensions.
223 */
224 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
225 return -EINVAL;
226
227 /* Put an artificial cap on the number
228 * of segments passed to kexec_load.
229 */
230 if (nr_segments > KEXEC_SEGMENT_MAX)
231 return -EINVAL;
232
233 return 0;
234}
235
236SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
237 struct kexec_segment __user *, segments, unsigned long, flags)
238{
239 int result;
240
241 result = kexec_load_check(nr_segments, flags);
242 if (result)
243 return result;
244
245 /* Verify we are on the appropriate architecture */
246 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
247 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
248 return -EINVAL;
249
250 /* Because we write directly to the reserved memory
251 * region when loading crash kernels we need a mutex here to
252 * prevent multiple crash kernels from attempting to load
253 * simultaneously, and to prevent a crash kernel from loading
254 * over the top of a in use crash kernel.
255 *
256 * KISS: always take the mutex.
257 */
258 if (!mutex_trylock(&kexec_mutex))
259 return -EBUSY;
260
261 result = do_kexec_load(entry, nr_segments, segments, flags);
262
263 mutex_unlock(&kexec_mutex);
264
265 return result;
266}
267
268#ifdef CONFIG_COMPAT
269COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
270 compat_ulong_t, nr_segments,
271 struct compat_kexec_segment __user *, segments,
272 compat_ulong_t, flags)
273{
274 struct compat_kexec_segment in;
275 struct kexec_segment out, __user *ksegments;
276 unsigned long i, result;
277
278 result = kexec_load_check(nr_segments, flags);
279 if (result)
280 return result;
281
282 /* Don't allow clients that don't understand the native
283 * architecture to do anything.
284 */
285 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
286 return -EINVAL;
287
288 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
289 for (i = 0; i < nr_segments; i++) {
290 result = copy_from_user(&in, &segments[i], sizeof(in));
291 if (result)
292 return -EFAULT;
293
294 out.buf = compat_ptr(in.buf);
295 out.bufsz = in.bufsz;
296 out.mem = in.mem;
297 out.memsz = in.memsz;
298
299 result = copy_to_user(&ksegments[i], &out, sizeof(out));
300 if (result)
301 return -EFAULT;
302 }
303
304 /* Because we write directly to the reserved memory
305 * region when loading crash kernels we need a mutex here to
306 * prevent multiple crash kernels from attempting to load
307 * simultaneously, and to prevent a crash kernel from loading
308 * over the top of a in use crash kernel.
309 *
310 * KISS: always take the mutex.
311 */
312 if (!mutex_trylock(&kexec_mutex))
313 return -EBUSY;
314
315 result = do_kexec_load(entry, nr_segments, ksegments, flags);
316
317 mutex_unlock(&kexec_mutex);
318
319 return result;
320}
321#endif
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/mutex.h>
16#include <linux/list.h>
17#include <linux/highmem.h>
18#include <linux/syscalls.h>
19#include <linux/reboot.h>
20#include <linux/ioport.h>
21#include <linux/hardirq.h>
22#include <linux/elf.h>
23#include <linux/elfcore.h>
24#include <generated/utsrelease.h>
25#include <linux/utsname.h>
26#include <linux/numa.h>
27#include <linux/suspend.h>
28#include <linux/device.h>
29#include <linux/freezer.h>
30#include <linux/pm.h>
31#include <linux/cpu.h>
32#include <linux/console.h>
33#include <linux/vmalloc.h>
34#include <linux/swap.h>
35#include <linux/syscore_ops.h>
36
37#include <asm/page.h>
38#include <asm/uaccess.h>
39#include <asm/io.h>
40#include <asm/sections.h>
41
42/* Per cpu memory for storing cpu states in case of system crash. */
43note_buf_t __percpu *crash_notes;
44
45/* vmcoreinfo stuff */
46static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
47u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
48size_t vmcoreinfo_size;
49size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
50
51/* Location of the reserved area for the crash kernel */
52struct resource crashk_res = {
53 .name = "Crash kernel",
54 .start = 0,
55 .end = 0,
56 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
57};
58
59int kexec_should_crash(struct task_struct *p)
60{
61 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
62 return 1;
63 return 0;
64}
65
66/*
67 * When kexec transitions to the new kernel there is a one-to-one
68 * mapping between physical and virtual addresses. On processors
69 * where you can disable the MMU this is trivial, and easy. For
70 * others it is still a simple predictable page table to setup.
71 *
72 * In that environment kexec copies the new kernel to its final
73 * resting place. This means I can only support memory whose
74 * physical address can fit in an unsigned long. In particular
75 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
76 * If the assembly stub has more restrictive requirements
77 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
78 * defined more restrictively in <asm/kexec.h>.
79 *
80 * The code for the transition from the current kernel to the
81 * the new kernel is placed in the control_code_buffer, whose size
82 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
83 * page of memory is necessary, but some architectures require more.
84 * Because this memory must be identity mapped in the transition from
85 * virtual to physical addresses it must live in the range
86 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
87 * modifiable.
88 *
89 * The assembly stub in the control code buffer is passed a linked list
90 * of descriptor pages detailing the source pages of the new kernel,
91 * and the destination addresses of those source pages. As this data
92 * structure is not used in the context of the current OS, it must
93 * be self-contained.
94 *
95 * The code has been made to work with highmem pages and will use a
96 * destination page in its final resting place (if it happens
97 * to allocate it). The end product of this is that most of the
98 * physical address space, and most of RAM can be used.
99 *
100 * Future directions include:
101 * - allocating a page table with the control code buffer identity
102 * mapped, to simplify machine_kexec and make kexec_on_panic more
103 * reliable.
104 */
105
106/*
107 * KIMAGE_NO_DEST is an impossible destination address..., for
108 * allocating pages whose destination address we do not care about.
109 */
110#define KIMAGE_NO_DEST (-1UL)
111
112static int kimage_is_destination_range(struct kimage *image,
113 unsigned long start, unsigned long end);
114static struct page *kimage_alloc_page(struct kimage *image,
115 gfp_t gfp_mask,
116 unsigned long dest);
117
118static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
119 unsigned long nr_segments,
120 struct kexec_segment __user *segments)
121{
122 size_t segment_bytes;
123 struct kimage *image;
124 unsigned long i;
125 int result;
126
127 /* Allocate a controlling structure */
128 result = -ENOMEM;
129 image = kzalloc(sizeof(*image), GFP_KERNEL);
130 if (!image)
131 goto out;
132
133 image->head = 0;
134 image->entry = &image->head;
135 image->last_entry = &image->head;
136 image->control_page = ~0; /* By default this does not apply */
137 image->start = entry;
138 image->type = KEXEC_TYPE_DEFAULT;
139
140 /* Initialize the list of control pages */
141 INIT_LIST_HEAD(&image->control_pages);
142
143 /* Initialize the list of destination pages */
144 INIT_LIST_HEAD(&image->dest_pages);
145
146 /* Initialize the list of unusable pages */
147 INIT_LIST_HEAD(&image->unuseable_pages);
148
149 /* Read in the segments */
150 image->nr_segments = nr_segments;
151 segment_bytes = nr_segments * sizeof(*segments);
152 result = copy_from_user(image->segment, segments, segment_bytes);
153 if (result) {
154 result = -EFAULT;
155 goto out;
156 }
157
158 /*
159 * Verify we have good destination addresses. The caller is
160 * responsible for making certain we don't attempt to load
161 * the new image into invalid or reserved areas of RAM. This
162 * just verifies it is an address we can use.
163 *
164 * Since the kernel does everything in page size chunks ensure
165 * the destination addresses are page aligned. Too many
166 * special cases crop of when we don't do this. The most
167 * insidious is getting overlapping destination addresses
168 * simply because addresses are changed to page size
169 * granularity.
170 */
171 result = -EADDRNOTAVAIL;
172 for (i = 0; i < nr_segments; i++) {
173 unsigned long mstart, mend;
174
175 mstart = image->segment[i].mem;
176 mend = mstart + image->segment[i].memsz;
177 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
178 goto out;
179 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
180 goto out;
181 }
182
183 /* Verify our destination addresses do not overlap.
184 * If we alloed overlapping destination addresses
185 * through very weird things can happen with no
186 * easy explanation as one segment stops on another.
187 */
188 result = -EINVAL;
189 for (i = 0; i < nr_segments; i++) {
190 unsigned long mstart, mend;
191 unsigned long j;
192
193 mstart = image->segment[i].mem;
194 mend = mstart + image->segment[i].memsz;
195 for (j = 0; j < i; j++) {
196 unsigned long pstart, pend;
197 pstart = image->segment[j].mem;
198 pend = pstart + image->segment[j].memsz;
199 /* Do the segments overlap ? */
200 if ((mend > pstart) && (mstart < pend))
201 goto out;
202 }
203 }
204
205 /* Ensure our buffer sizes are strictly less than
206 * our memory sizes. This should always be the case,
207 * and it is easier to check up front than to be surprised
208 * later on.
209 */
210 result = -EINVAL;
211 for (i = 0; i < nr_segments; i++) {
212 if (image->segment[i].bufsz > image->segment[i].memsz)
213 goto out;
214 }
215
216 result = 0;
217out:
218 if (result == 0)
219 *rimage = image;
220 else
221 kfree(image);
222
223 return result;
224
225}
226
227static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
228 unsigned long nr_segments,
229 struct kexec_segment __user *segments)
230{
231 int result;
232 struct kimage *image;
233
234 /* Allocate and initialize a controlling structure */
235 image = NULL;
236 result = do_kimage_alloc(&image, entry, nr_segments, segments);
237 if (result)
238 goto out;
239
240 *rimage = image;
241
242 /*
243 * Find a location for the control code buffer, and add it
244 * the vector of segments so that it's pages will also be
245 * counted as destination pages.
246 */
247 result = -ENOMEM;
248 image->control_code_page = kimage_alloc_control_pages(image,
249 get_order(KEXEC_CONTROL_PAGE_SIZE));
250 if (!image->control_code_page) {
251 printk(KERN_ERR "Could not allocate control_code_buffer\n");
252 goto out;
253 }
254
255 image->swap_page = kimage_alloc_control_pages(image, 0);
256 if (!image->swap_page) {
257 printk(KERN_ERR "Could not allocate swap buffer\n");
258 goto out;
259 }
260
261 result = 0;
262 out:
263 if (result == 0)
264 *rimage = image;
265 else
266 kfree(image);
267
268 return result;
269}
270
271static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
272 unsigned long nr_segments,
273 struct kexec_segment __user *segments)
274{
275 int result;
276 struct kimage *image;
277 unsigned long i;
278
279 image = NULL;
280 /* Verify we have a valid entry point */
281 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
282 result = -EADDRNOTAVAIL;
283 goto out;
284 }
285
286 /* Allocate and initialize a controlling structure */
287 result = do_kimage_alloc(&image, entry, nr_segments, segments);
288 if (result)
289 goto out;
290
291 /* Enable the special crash kernel control page
292 * allocation policy.
293 */
294 image->control_page = crashk_res.start;
295 image->type = KEXEC_TYPE_CRASH;
296
297 /*
298 * Verify we have good destination addresses. Normally
299 * the caller is responsible for making certain we don't
300 * attempt to load the new image into invalid or reserved
301 * areas of RAM. But crash kernels are preloaded into a
302 * reserved area of ram. We must ensure the addresses
303 * are in the reserved area otherwise preloading the
304 * kernel could corrupt things.
305 */
306 result = -EADDRNOTAVAIL;
307 for (i = 0; i < nr_segments; i++) {
308 unsigned long mstart, mend;
309
310 mstart = image->segment[i].mem;
311 mend = mstart + image->segment[i].memsz - 1;
312 /* Ensure we are within the crash kernel limits */
313 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
314 goto out;
315 }
316
317 /*
318 * Find a location for the control code buffer, and add
319 * the vector of segments so that it's pages will also be
320 * counted as destination pages.
321 */
322 result = -ENOMEM;
323 image->control_code_page = kimage_alloc_control_pages(image,
324 get_order(KEXEC_CONTROL_PAGE_SIZE));
325 if (!image->control_code_page) {
326 printk(KERN_ERR "Could not allocate control_code_buffer\n");
327 goto out;
328 }
329
330 result = 0;
331out:
332 if (result == 0)
333 *rimage = image;
334 else
335 kfree(image);
336
337 return result;
338}
339
340static int kimage_is_destination_range(struct kimage *image,
341 unsigned long start,
342 unsigned long end)
343{
344 unsigned long i;
345
346 for (i = 0; i < image->nr_segments; i++) {
347 unsigned long mstart, mend;
348
349 mstart = image->segment[i].mem;
350 mend = mstart + image->segment[i].memsz;
351 if ((end > mstart) && (start < mend))
352 return 1;
353 }
354
355 return 0;
356}
357
358static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
359{
360 struct page *pages;
361
362 pages = alloc_pages(gfp_mask, order);
363 if (pages) {
364 unsigned int count, i;
365 pages->mapping = NULL;
366 set_page_private(pages, order);
367 count = 1 << order;
368 for (i = 0; i < count; i++)
369 SetPageReserved(pages + i);
370 }
371
372 return pages;
373}
374
375static void kimage_free_pages(struct page *page)
376{
377 unsigned int order, count, i;
378
379 order = page_private(page);
380 count = 1 << order;
381 for (i = 0; i < count; i++)
382 ClearPageReserved(page + i);
383 __free_pages(page, order);
384}
385
386static void kimage_free_page_list(struct list_head *list)
387{
388 struct list_head *pos, *next;
389
390 list_for_each_safe(pos, next, list) {
391 struct page *page;
392
393 page = list_entry(pos, struct page, lru);
394 list_del(&page->lru);
395 kimage_free_pages(page);
396 }
397}
398
399static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
400 unsigned int order)
401{
402 /* Control pages are special, they are the intermediaries
403 * that are needed while we copy the rest of the pages
404 * to their final resting place. As such they must
405 * not conflict with either the destination addresses
406 * or memory the kernel is already using.
407 *
408 * The only case where we really need more than one of
409 * these are for architectures where we cannot disable
410 * the MMU and must instead generate an identity mapped
411 * page table for all of the memory.
412 *
413 * At worst this runs in O(N) of the image size.
414 */
415 struct list_head extra_pages;
416 struct page *pages;
417 unsigned int count;
418
419 count = 1 << order;
420 INIT_LIST_HEAD(&extra_pages);
421
422 /* Loop while I can allocate a page and the page allocated
423 * is a destination page.
424 */
425 do {
426 unsigned long pfn, epfn, addr, eaddr;
427
428 pages = kimage_alloc_pages(GFP_KERNEL, order);
429 if (!pages)
430 break;
431 pfn = page_to_pfn(pages);
432 epfn = pfn + count;
433 addr = pfn << PAGE_SHIFT;
434 eaddr = epfn << PAGE_SHIFT;
435 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
436 kimage_is_destination_range(image, addr, eaddr)) {
437 list_add(&pages->lru, &extra_pages);
438 pages = NULL;
439 }
440 } while (!pages);
441
442 if (pages) {
443 /* Remember the allocated page... */
444 list_add(&pages->lru, &image->control_pages);
445
446 /* Because the page is already in it's destination
447 * location we will never allocate another page at
448 * that address. Therefore kimage_alloc_pages
449 * will not return it (again) and we don't need
450 * to give it an entry in image->segment[].
451 */
452 }
453 /* Deal with the destination pages I have inadvertently allocated.
454 *
455 * Ideally I would convert multi-page allocations into single
456 * page allocations, and add everything to image->dest_pages.
457 *
458 * For now it is simpler to just free the pages.
459 */
460 kimage_free_page_list(&extra_pages);
461
462 return pages;
463}
464
465static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
466 unsigned int order)
467{
468 /* Control pages are special, they are the intermediaries
469 * that are needed while we copy the rest of the pages
470 * to their final resting place. As such they must
471 * not conflict with either the destination addresses
472 * or memory the kernel is already using.
473 *
474 * Control pages are also the only pags we must allocate
475 * when loading a crash kernel. All of the other pages
476 * are specified by the segments and we just memcpy
477 * into them directly.
478 *
479 * The only case where we really need more than one of
480 * these are for architectures where we cannot disable
481 * the MMU and must instead generate an identity mapped
482 * page table for all of the memory.
483 *
484 * Given the low demand this implements a very simple
485 * allocator that finds the first hole of the appropriate
486 * size in the reserved memory region, and allocates all
487 * of the memory up to and including the hole.
488 */
489 unsigned long hole_start, hole_end, size;
490 struct page *pages;
491
492 pages = NULL;
493 size = (1 << order) << PAGE_SHIFT;
494 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
495 hole_end = hole_start + size - 1;
496 while (hole_end <= crashk_res.end) {
497 unsigned long i;
498
499 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
500 break;
501 if (hole_end > crashk_res.end)
502 break;
503 /* See if I overlap any of the segments */
504 for (i = 0; i < image->nr_segments; i++) {
505 unsigned long mstart, mend;
506
507 mstart = image->segment[i].mem;
508 mend = mstart + image->segment[i].memsz - 1;
509 if ((hole_end >= mstart) && (hole_start <= mend)) {
510 /* Advance the hole to the end of the segment */
511 hole_start = (mend + (size - 1)) & ~(size - 1);
512 hole_end = hole_start + size - 1;
513 break;
514 }
515 }
516 /* If I don't overlap any segments I have found my hole! */
517 if (i == image->nr_segments) {
518 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
519 break;
520 }
521 }
522 if (pages)
523 image->control_page = hole_end;
524
525 return pages;
526}
527
528
529struct page *kimage_alloc_control_pages(struct kimage *image,
530 unsigned int order)
531{
532 struct page *pages = NULL;
533
534 switch (image->type) {
535 case KEXEC_TYPE_DEFAULT:
536 pages = kimage_alloc_normal_control_pages(image, order);
537 break;
538 case KEXEC_TYPE_CRASH:
539 pages = kimage_alloc_crash_control_pages(image, order);
540 break;
541 }
542
543 return pages;
544}
545
546static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
547{
548 if (*image->entry != 0)
549 image->entry++;
550
551 if (image->entry == image->last_entry) {
552 kimage_entry_t *ind_page;
553 struct page *page;
554
555 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
556 if (!page)
557 return -ENOMEM;
558
559 ind_page = page_address(page);
560 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
561 image->entry = ind_page;
562 image->last_entry = ind_page +
563 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
564 }
565 *image->entry = entry;
566 image->entry++;
567 *image->entry = 0;
568
569 return 0;
570}
571
572static int kimage_set_destination(struct kimage *image,
573 unsigned long destination)
574{
575 int result;
576
577 destination &= PAGE_MASK;
578 result = kimage_add_entry(image, destination | IND_DESTINATION);
579 if (result == 0)
580 image->destination = destination;
581
582 return result;
583}
584
585
586static int kimage_add_page(struct kimage *image, unsigned long page)
587{
588 int result;
589
590 page &= PAGE_MASK;
591 result = kimage_add_entry(image, page | IND_SOURCE);
592 if (result == 0)
593 image->destination += PAGE_SIZE;
594
595 return result;
596}
597
598
599static void kimage_free_extra_pages(struct kimage *image)
600{
601 /* Walk through and free any extra destination pages I may have */
602 kimage_free_page_list(&image->dest_pages);
603
604 /* Walk through and free any unusable pages I have cached */
605 kimage_free_page_list(&image->unuseable_pages);
606
607}
608static void kimage_terminate(struct kimage *image)
609{
610 if (*image->entry != 0)
611 image->entry++;
612
613 *image->entry = IND_DONE;
614}
615
616#define for_each_kimage_entry(image, ptr, entry) \
617 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
618 ptr = (entry & IND_INDIRECTION)? \
619 phys_to_virt((entry & PAGE_MASK)): ptr +1)
620
621static void kimage_free_entry(kimage_entry_t entry)
622{
623 struct page *page;
624
625 page = pfn_to_page(entry >> PAGE_SHIFT);
626 kimage_free_pages(page);
627}
628
629static void kimage_free(struct kimage *image)
630{
631 kimage_entry_t *ptr, entry;
632 kimage_entry_t ind = 0;
633
634 if (!image)
635 return;
636
637 kimage_free_extra_pages(image);
638 for_each_kimage_entry(image, ptr, entry) {
639 if (entry & IND_INDIRECTION) {
640 /* Free the previous indirection page */
641 if (ind & IND_INDIRECTION)
642 kimage_free_entry(ind);
643 /* Save this indirection page until we are
644 * done with it.
645 */
646 ind = entry;
647 }
648 else if (entry & IND_SOURCE)
649 kimage_free_entry(entry);
650 }
651 /* Free the final indirection page */
652 if (ind & IND_INDIRECTION)
653 kimage_free_entry(ind);
654
655 /* Handle any machine specific cleanup */
656 machine_kexec_cleanup(image);
657
658 /* Free the kexec control pages... */
659 kimage_free_page_list(&image->control_pages);
660 kfree(image);
661}
662
663static kimage_entry_t *kimage_dst_used(struct kimage *image,
664 unsigned long page)
665{
666 kimage_entry_t *ptr, entry;
667 unsigned long destination = 0;
668
669 for_each_kimage_entry(image, ptr, entry) {
670 if (entry & IND_DESTINATION)
671 destination = entry & PAGE_MASK;
672 else if (entry & IND_SOURCE) {
673 if (page == destination)
674 return ptr;
675 destination += PAGE_SIZE;
676 }
677 }
678
679 return NULL;
680}
681
682static struct page *kimage_alloc_page(struct kimage *image,
683 gfp_t gfp_mask,
684 unsigned long destination)
685{
686 /*
687 * Here we implement safeguards to ensure that a source page
688 * is not copied to its destination page before the data on
689 * the destination page is no longer useful.
690 *
691 * To do this we maintain the invariant that a source page is
692 * either its own destination page, or it is not a
693 * destination page at all.
694 *
695 * That is slightly stronger than required, but the proof
696 * that no problems will not occur is trivial, and the
697 * implementation is simply to verify.
698 *
699 * When allocating all pages normally this algorithm will run
700 * in O(N) time, but in the worst case it will run in O(N^2)
701 * time. If the runtime is a problem the data structures can
702 * be fixed.
703 */
704 struct page *page;
705 unsigned long addr;
706
707 /*
708 * Walk through the list of destination pages, and see if I
709 * have a match.
710 */
711 list_for_each_entry(page, &image->dest_pages, lru) {
712 addr = page_to_pfn(page) << PAGE_SHIFT;
713 if (addr == destination) {
714 list_del(&page->lru);
715 return page;
716 }
717 }
718 page = NULL;
719 while (1) {
720 kimage_entry_t *old;
721
722 /* Allocate a page, if we run out of memory give up */
723 page = kimage_alloc_pages(gfp_mask, 0);
724 if (!page)
725 return NULL;
726 /* If the page cannot be used file it away */
727 if (page_to_pfn(page) >
728 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
729 list_add(&page->lru, &image->unuseable_pages);
730 continue;
731 }
732 addr = page_to_pfn(page) << PAGE_SHIFT;
733
734 /* If it is the destination page we want use it */
735 if (addr == destination)
736 break;
737
738 /* If the page is not a destination page use it */
739 if (!kimage_is_destination_range(image, addr,
740 addr + PAGE_SIZE))
741 break;
742
743 /*
744 * I know that the page is someones destination page.
745 * See if there is already a source page for this
746 * destination page. And if so swap the source pages.
747 */
748 old = kimage_dst_used(image, addr);
749 if (old) {
750 /* If so move it */
751 unsigned long old_addr;
752 struct page *old_page;
753
754 old_addr = *old & PAGE_MASK;
755 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
756 copy_highpage(page, old_page);
757 *old = addr | (*old & ~PAGE_MASK);
758
759 /* The old page I have found cannot be a
760 * destination page, so return it if it's
761 * gfp_flags honor the ones passed in.
762 */
763 if (!(gfp_mask & __GFP_HIGHMEM) &&
764 PageHighMem(old_page)) {
765 kimage_free_pages(old_page);
766 continue;
767 }
768 addr = old_addr;
769 page = old_page;
770 break;
771 }
772 else {
773 /* Place the page on the destination list I
774 * will use it later.
775 */
776 list_add(&page->lru, &image->dest_pages);
777 }
778 }
779
780 return page;
781}
782
783static int kimage_load_normal_segment(struct kimage *image,
784 struct kexec_segment *segment)
785{
786 unsigned long maddr;
787 unsigned long ubytes, mbytes;
788 int result;
789 unsigned char __user *buf;
790
791 result = 0;
792 buf = segment->buf;
793 ubytes = segment->bufsz;
794 mbytes = segment->memsz;
795 maddr = segment->mem;
796
797 result = kimage_set_destination(image, maddr);
798 if (result < 0)
799 goto out;
800
801 while (mbytes) {
802 struct page *page;
803 char *ptr;
804 size_t uchunk, mchunk;
805
806 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
807 if (!page) {
808 result = -ENOMEM;
809 goto out;
810 }
811 result = kimage_add_page(image, page_to_pfn(page)
812 << PAGE_SHIFT);
813 if (result < 0)
814 goto out;
815
816 ptr = kmap(page);
817 /* Start with a clear page */
818 clear_page(ptr);
819 ptr += maddr & ~PAGE_MASK;
820 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
821 if (mchunk > mbytes)
822 mchunk = mbytes;
823
824 uchunk = mchunk;
825 if (uchunk > ubytes)
826 uchunk = ubytes;
827
828 result = copy_from_user(ptr, buf, uchunk);
829 kunmap(page);
830 if (result) {
831 result = -EFAULT;
832 goto out;
833 }
834 ubytes -= uchunk;
835 maddr += mchunk;
836 buf += mchunk;
837 mbytes -= mchunk;
838 }
839out:
840 return result;
841}
842
843static int kimage_load_crash_segment(struct kimage *image,
844 struct kexec_segment *segment)
845{
846 /* For crash dumps kernels we simply copy the data from
847 * user space to it's destination.
848 * We do things a page at a time for the sake of kmap.
849 */
850 unsigned long maddr;
851 unsigned long ubytes, mbytes;
852 int result;
853 unsigned char __user *buf;
854
855 result = 0;
856 buf = segment->buf;
857 ubytes = segment->bufsz;
858 mbytes = segment->memsz;
859 maddr = segment->mem;
860 while (mbytes) {
861 struct page *page;
862 char *ptr;
863 size_t uchunk, mchunk;
864
865 page = pfn_to_page(maddr >> PAGE_SHIFT);
866 if (!page) {
867 result = -ENOMEM;
868 goto out;
869 }
870 ptr = kmap(page);
871 ptr += maddr & ~PAGE_MASK;
872 mchunk = PAGE_SIZE - (maddr & ~PAGE_MASK);
873 if (mchunk > mbytes)
874 mchunk = mbytes;
875
876 uchunk = mchunk;
877 if (uchunk > ubytes) {
878 uchunk = ubytes;
879 /* Zero the trailing part of the page */
880 memset(ptr + uchunk, 0, mchunk - uchunk);
881 }
882 result = copy_from_user(ptr, buf, uchunk);
883 kexec_flush_icache_page(page);
884 kunmap(page);
885 if (result) {
886 result = -EFAULT;
887 goto out;
888 }
889 ubytes -= uchunk;
890 maddr += mchunk;
891 buf += mchunk;
892 mbytes -= mchunk;
893 }
894out:
895 return result;
896}
897
898static int kimage_load_segment(struct kimage *image,
899 struct kexec_segment *segment)
900{
901 int result = -ENOMEM;
902
903 switch (image->type) {
904 case KEXEC_TYPE_DEFAULT:
905 result = kimage_load_normal_segment(image, segment);
906 break;
907 case KEXEC_TYPE_CRASH:
908 result = kimage_load_crash_segment(image, segment);
909 break;
910 }
911
912 return result;
913}
914
915/*
916 * Exec Kernel system call: for obvious reasons only root may call it.
917 *
918 * This call breaks up into three pieces.
919 * - A generic part which loads the new kernel from the current
920 * address space, and very carefully places the data in the
921 * allocated pages.
922 *
923 * - A generic part that interacts with the kernel and tells all of
924 * the devices to shut down. Preventing on-going dmas, and placing
925 * the devices in a consistent state so a later kernel can
926 * reinitialize them.
927 *
928 * - A machine specific part that includes the syscall number
929 * and the copies the image to it's final destination. And
930 * jumps into the image at entry.
931 *
932 * kexec does not sync, or unmount filesystems so if you need
933 * that to happen you need to do that yourself.
934 */
935struct kimage *kexec_image;
936struct kimage *kexec_crash_image;
937
938static DEFINE_MUTEX(kexec_mutex);
939
940SYSCALL_DEFINE4(kexec_load, unsigned long, entry, unsigned long, nr_segments,
941 struct kexec_segment __user *, segments, unsigned long, flags)
942{
943 struct kimage **dest_image, *image;
944 int result;
945
946 /* We only trust the superuser with rebooting the system. */
947 if (!capable(CAP_SYS_BOOT))
948 return -EPERM;
949
950 /*
951 * Verify we have a legal set of flags
952 * This leaves us room for future extensions.
953 */
954 if ((flags & KEXEC_FLAGS) != (flags & ~KEXEC_ARCH_MASK))
955 return -EINVAL;
956
957 /* Verify we are on the appropriate architecture */
958 if (((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH) &&
959 ((flags & KEXEC_ARCH_MASK) != KEXEC_ARCH_DEFAULT))
960 return -EINVAL;
961
962 /* Put an artificial cap on the number
963 * of segments passed to kexec_load.
964 */
965 if (nr_segments > KEXEC_SEGMENT_MAX)
966 return -EINVAL;
967
968 image = NULL;
969 result = 0;
970
971 /* Because we write directly to the reserved memory
972 * region when loading crash kernels we need a mutex here to
973 * prevent multiple crash kernels from attempting to load
974 * simultaneously, and to prevent a crash kernel from loading
975 * over the top of a in use crash kernel.
976 *
977 * KISS: always take the mutex.
978 */
979 if (!mutex_trylock(&kexec_mutex))
980 return -EBUSY;
981
982 dest_image = &kexec_image;
983 if (flags & KEXEC_ON_CRASH)
984 dest_image = &kexec_crash_image;
985 if (nr_segments > 0) {
986 unsigned long i;
987
988 /* Loading another kernel to reboot into */
989 if ((flags & KEXEC_ON_CRASH) == 0)
990 result = kimage_normal_alloc(&image, entry,
991 nr_segments, segments);
992 /* Loading another kernel to switch to if this one crashes */
993 else if (flags & KEXEC_ON_CRASH) {
994 /* Free any current crash dump kernel before
995 * we corrupt it.
996 */
997 kimage_free(xchg(&kexec_crash_image, NULL));
998 result = kimage_crash_alloc(&image, entry,
999 nr_segments, segments);
1000 crash_map_reserved_pages();
1001 }
1002 if (result)
1003 goto out;
1004
1005 if (flags & KEXEC_PRESERVE_CONTEXT)
1006 image->preserve_context = 1;
1007 result = machine_kexec_prepare(image);
1008 if (result)
1009 goto out;
1010
1011 for (i = 0; i < nr_segments; i++) {
1012 result = kimage_load_segment(image, &image->segment[i]);
1013 if (result)
1014 goto out;
1015 }
1016 kimage_terminate(image);
1017 if (flags & KEXEC_ON_CRASH)
1018 crash_unmap_reserved_pages();
1019 }
1020 /* Install the new kernel, and Uninstall the old */
1021 image = xchg(dest_image, image);
1022
1023out:
1024 mutex_unlock(&kexec_mutex);
1025 kimage_free(image);
1026
1027 return result;
1028}
1029
1030/*
1031 * Add and remove page tables for crashkernel memory
1032 *
1033 * Provide an empty default implementation here -- architecture
1034 * code may override this
1035 */
1036void __weak crash_map_reserved_pages(void)
1037{}
1038
1039void __weak crash_unmap_reserved_pages(void)
1040{}
1041
1042#ifdef CONFIG_COMPAT
1043asmlinkage long compat_sys_kexec_load(unsigned long entry,
1044 unsigned long nr_segments,
1045 struct compat_kexec_segment __user *segments,
1046 unsigned long flags)
1047{
1048 struct compat_kexec_segment in;
1049 struct kexec_segment out, __user *ksegments;
1050 unsigned long i, result;
1051
1052 /* Don't allow clients that don't understand the native
1053 * architecture to do anything.
1054 */
1055 if ((flags & KEXEC_ARCH_MASK) == KEXEC_ARCH_DEFAULT)
1056 return -EINVAL;
1057
1058 if (nr_segments > KEXEC_SEGMENT_MAX)
1059 return -EINVAL;
1060
1061 ksegments = compat_alloc_user_space(nr_segments * sizeof(out));
1062 for (i=0; i < nr_segments; i++) {
1063 result = copy_from_user(&in, &segments[i], sizeof(in));
1064 if (result)
1065 return -EFAULT;
1066
1067 out.buf = compat_ptr(in.buf);
1068 out.bufsz = in.bufsz;
1069 out.mem = in.mem;
1070 out.memsz = in.memsz;
1071
1072 result = copy_to_user(&ksegments[i], &out, sizeof(out));
1073 if (result)
1074 return -EFAULT;
1075 }
1076
1077 return sys_kexec_load(entry, nr_segments, ksegments, flags);
1078}
1079#endif
1080
1081void crash_kexec(struct pt_regs *regs)
1082{
1083 /* Take the kexec_mutex here to prevent sys_kexec_load
1084 * running on one cpu from replacing the crash kernel
1085 * we are using after a panic on a different cpu.
1086 *
1087 * If the crash kernel was not located in a fixed area
1088 * of memory the xchg(&kexec_crash_image) would be
1089 * sufficient. But since I reuse the memory...
1090 */
1091 if (mutex_trylock(&kexec_mutex)) {
1092 if (kexec_crash_image) {
1093 struct pt_regs fixed_regs;
1094
1095 crash_setup_regs(&fixed_regs, regs);
1096 crash_save_vmcoreinfo();
1097 machine_crash_shutdown(&fixed_regs);
1098 machine_kexec(kexec_crash_image);
1099 }
1100 mutex_unlock(&kexec_mutex);
1101 }
1102}
1103
1104size_t crash_get_memory_size(void)
1105{
1106 size_t size = 0;
1107 mutex_lock(&kexec_mutex);
1108 if (crashk_res.end != crashk_res.start)
1109 size = resource_size(&crashk_res);
1110 mutex_unlock(&kexec_mutex);
1111 return size;
1112}
1113
1114void __weak crash_free_reserved_phys_range(unsigned long begin,
1115 unsigned long end)
1116{
1117 unsigned long addr;
1118
1119 for (addr = begin; addr < end; addr += PAGE_SIZE) {
1120 ClearPageReserved(pfn_to_page(addr >> PAGE_SHIFT));
1121 init_page_count(pfn_to_page(addr >> PAGE_SHIFT));
1122 free_page((unsigned long)__va(addr));
1123 totalram_pages++;
1124 }
1125}
1126
1127int crash_shrink_memory(unsigned long new_size)
1128{
1129 int ret = 0;
1130 unsigned long start, end;
1131 unsigned long old_size;
1132 struct resource *ram_res;
1133
1134 mutex_lock(&kexec_mutex);
1135
1136 if (kexec_crash_image) {
1137 ret = -ENOENT;
1138 goto unlock;
1139 }
1140 start = crashk_res.start;
1141 end = crashk_res.end;
1142 old_size = (end == 0) ? 0 : end - start + 1;
1143 if (new_size >= old_size) {
1144 ret = (new_size == old_size) ? 0 : -EINVAL;
1145 goto unlock;
1146 }
1147
1148 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1149 if (!ram_res) {
1150 ret = -ENOMEM;
1151 goto unlock;
1152 }
1153
1154 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1155 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1156
1157 crash_map_reserved_pages();
1158 crash_free_reserved_phys_range(end, crashk_res.end);
1159
1160 if ((start == end) && (crashk_res.parent != NULL))
1161 release_resource(&crashk_res);
1162
1163 ram_res->start = end;
1164 ram_res->end = crashk_res.end;
1165 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1166 ram_res->name = "System RAM";
1167
1168 crashk_res.end = end - 1;
1169
1170 insert_resource(&iomem_resource, ram_res);
1171 crash_unmap_reserved_pages();
1172
1173unlock:
1174 mutex_unlock(&kexec_mutex);
1175 return ret;
1176}
1177
1178static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1179 size_t data_len)
1180{
1181 struct elf_note note;
1182
1183 note.n_namesz = strlen(name) + 1;
1184 note.n_descsz = data_len;
1185 note.n_type = type;
1186 memcpy(buf, ¬e, sizeof(note));
1187 buf += (sizeof(note) + 3)/4;
1188 memcpy(buf, name, note.n_namesz);
1189 buf += (note.n_namesz + 3)/4;
1190 memcpy(buf, data, note.n_descsz);
1191 buf += (note.n_descsz + 3)/4;
1192
1193 return buf;
1194}
1195
1196static void final_note(u32 *buf)
1197{
1198 struct elf_note note;
1199
1200 note.n_namesz = 0;
1201 note.n_descsz = 0;
1202 note.n_type = 0;
1203 memcpy(buf, ¬e, sizeof(note));
1204}
1205
1206void crash_save_cpu(struct pt_regs *regs, int cpu)
1207{
1208 struct elf_prstatus prstatus;
1209 u32 *buf;
1210
1211 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1212 return;
1213
1214 /* Using ELF notes here is opportunistic.
1215 * I need a well defined structure format
1216 * for the data I pass, and I need tags
1217 * on the data to indicate what information I have
1218 * squirrelled away. ELF notes happen to provide
1219 * all of that, so there is no need to invent something new.
1220 */
1221 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1222 if (!buf)
1223 return;
1224 memset(&prstatus, 0, sizeof(prstatus));
1225 prstatus.pr_pid = current->pid;
1226 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1227 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1228 &prstatus, sizeof(prstatus));
1229 final_note(buf);
1230}
1231
1232static int __init crash_notes_memory_init(void)
1233{
1234 /* Allocate memory for saving cpu registers. */
1235 crash_notes = alloc_percpu(note_buf_t);
1236 if (!crash_notes) {
1237 printk("Kexec: Memory allocation for saving cpu register"
1238 " states failed\n");
1239 return -ENOMEM;
1240 }
1241 return 0;
1242}
1243module_init(crash_notes_memory_init)
1244
1245
1246/*
1247 * parsing the "crashkernel" commandline
1248 *
1249 * this code is intended to be called from architecture specific code
1250 */
1251
1252
1253/*
1254 * This function parses command lines in the format
1255 *
1256 * crashkernel=ramsize-range:size[,...][@offset]
1257 *
1258 * The function returns 0 on success and -EINVAL on failure.
1259 */
1260static int __init parse_crashkernel_mem(char *cmdline,
1261 unsigned long long system_ram,
1262 unsigned long long *crash_size,
1263 unsigned long long *crash_base)
1264{
1265 char *cur = cmdline, *tmp;
1266
1267 /* for each entry of the comma-separated list */
1268 do {
1269 unsigned long long start, end = ULLONG_MAX, size;
1270
1271 /* get the start of the range */
1272 start = memparse(cur, &tmp);
1273 if (cur == tmp) {
1274 pr_warning("crashkernel: Memory value expected\n");
1275 return -EINVAL;
1276 }
1277 cur = tmp;
1278 if (*cur != '-') {
1279 pr_warning("crashkernel: '-' expected\n");
1280 return -EINVAL;
1281 }
1282 cur++;
1283
1284 /* if no ':' is here, than we read the end */
1285 if (*cur != ':') {
1286 end = memparse(cur, &tmp);
1287 if (cur == tmp) {
1288 pr_warning("crashkernel: Memory "
1289 "value expected\n");
1290 return -EINVAL;
1291 }
1292 cur = tmp;
1293 if (end <= start) {
1294 pr_warning("crashkernel: end <= start\n");
1295 return -EINVAL;
1296 }
1297 }
1298
1299 if (*cur != ':') {
1300 pr_warning("crashkernel: ':' expected\n");
1301 return -EINVAL;
1302 }
1303 cur++;
1304
1305 size = memparse(cur, &tmp);
1306 if (cur == tmp) {
1307 pr_warning("Memory value expected\n");
1308 return -EINVAL;
1309 }
1310 cur = tmp;
1311 if (size >= system_ram) {
1312 pr_warning("crashkernel: invalid size\n");
1313 return -EINVAL;
1314 }
1315
1316 /* match ? */
1317 if (system_ram >= start && system_ram < end) {
1318 *crash_size = size;
1319 break;
1320 }
1321 } while (*cur++ == ',');
1322
1323 if (*crash_size > 0) {
1324 while (*cur && *cur != ' ' && *cur != '@')
1325 cur++;
1326 if (*cur == '@') {
1327 cur++;
1328 *crash_base = memparse(cur, &tmp);
1329 if (cur == tmp) {
1330 pr_warning("Memory value expected "
1331 "after '@'\n");
1332 return -EINVAL;
1333 }
1334 }
1335 }
1336
1337 return 0;
1338}
1339
1340/*
1341 * That function parses "simple" (old) crashkernel command lines like
1342 *
1343 * crashkernel=size[@offset]
1344 *
1345 * It returns 0 on success and -EINVAL on failure.
1346 */
1347static int __init parse_crashkernel_simple(char *cmdline,
1348 unsigned long long *crash_size,
1349 unsigned long long *crash_base)
1350{
1351 char *cur = cmdline;
1352
1353 *crash_size = memparse(cmdline, &cur);
1354 if (cmdline == cur) {
1355 pr_warning("crashkernel: memory value expected\n");
1356 return -EINVAL;
1357 }
1358
1359 if (*cur == '@')
1360 *crash_base = memparse(cur+1, &cur);
1361 else if (*cur != ' ' && *cur != '\0') {
1362 pr_warning("crashkernel: unrecognized char\n");
1363 return -EINVAL;
1364 }
1365
1366 return 0;
1367}
1368
1369/*
1370 * That function is the entry point for command line parsing and should be
1371 * called from the arch-specific code.
1372 */
1373int __init parse_crashkernel(char *cmdline,
1374 unsigned long long system_ram,
1375 unsigned long long *crash_size,
1376 unsigned long long *crash_base)
1377{
1378 char *p = cmdline, *ck_cmdline = NULL;
1379 char *first_colon, *first_space;
1380
1381 BUG_ON(!crash_size || !crash_base);
1382 *crash_size = 0;
1383 *crash_base = 0;
1384
1385 /* find crashkernel and use the last one if there are more */
1386 p = strstr(p, "crashkernel=");
1387 while (p) {
1388 ck_cmdline = p;
1389 p = strstr(p+1, "crashkernel=");
1390 }
1391
1392 if (!ck_cmdline)
1393 return -EINVAL;
1394
1395 ck_cmdline += 12; /* strlen("crashkernel=") */
1396
1397 /*
1398 * if the commandline contains a ':', then that's the extended
1399 * syntax -- if not, it must be the classic syntax
1400 */
1401 first_colon = strchr(ck_cmdline, ':');
1402 first_space = strchr(ck_cmdline, ' ');
1403 if (first_colon && (!first_space || first_colon < first_space))
1404 return parse_crashkernel_mem(ck_cmdline, system_ram,
1405 crash_size, crash_base);
1406 else
1407 return parse_crashkernel_simple(ck_cmdline, crash_size,
1408 crash_base);
1409
1410 return 0;
1411}
1412
1413
1414static void update_vmcoreinfo_note(void)
1415{
1416 u32 *buf = vmcoreinfo_note;
1417
1418 if (!vmcoreinfo_size)
1419 return;
1420 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1421 vmcoreinfo_size);
1422 final_note(buf);
1423}
1424
1425void crash_save_vmcoreinfo(void)
1426{
1427 vmcoreinfo_append_str("CRASHTIME=%ld", get_seconds());
1428 update_vmcoreinfo_note();
1429}
1430
1431void vmcoreinfo_append_str(const char *fmt, ...)
1432{
1433 va_list args;
1434 char buf[0x50];
1435 int r;
1436
1437 va_start(args, fmt);
1438 r = vsnprintf(buf, sizeof(buf), fmt, args);
1439 va_end(args);
1440
1441 if (r + vmcoreinfo_size > vmcoreinfo_max_size)
1442 r = vmcoreinfo_max_size - vmcoreinfo_size;
1443
1444 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1445
1446 vmcoreinfo_size += r;
1447}
1448
1449/*
1450 * provide an empty default implementation here -- architecture
1451 * code may override this
1452 */
1453void __attribute__ ((weak)) arch_crash_save_vmcoreinfo(void)
1454{}
1455
1456unsigned long __attribute__ ((weak)) paddr_vmcoreinfo_note(void)
1457{
1458 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1459}
1460
1461static int __init crash_save_vmcoreinfo_init(void)
1462{
1463 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1464 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1465
1466 VMCOREINFO_SYMBOL(init_uts_ns);
1467 VMCOREINFO_SYMBOL(node_online_map);
1468#ifdef CONFIG_MMU
1469 VMCOREINFO_SYMBOL(swapper_pg_dir);
1470#endif
1471 VMCOREINFO_SYMBOL(_stext);
1472 VMCOREINFO_SYMBOL(vmlist);
1473
1474#ifndef CONFIG_NEED_MULTIPLE_NODES
1475 VMCOREINFO_SYMBOL(mem_map);
1476 VMCOREINFO_SYMBOL(contig_page_data);
1477#endif
1478#ifdef CONFIG_SPARSEMEM
1479 VMCOREINFO_SYMBOL(mem_section);
1480 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1481 VMCOREINFO_STRUCT_SIZE(mem_section);
1482 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1483#endif
1484 VMCOREINFO_STRUCT_SIZE(page);
1485 VMCOREINFO_STRUCT_SIZE(pglist_data);
1486 VMCOREINFO_STRUCT_SIZE(zone);
1487 VMCOREINFO_STRUCT_SIZE(free_area);
1488 VMCOREINFO_STRUCT_SIZE(list_head);
1489 VMCOREINFO_SIZE(nodemask_t);
1490 VMCOREINFO_OFFSET(page, flags);
1491 VMCOREINFO_OFFSET(page, _count);
1492 VMCOREINFO_OFFSET(page, mapping);
1493 VMCOREINFO_OFFSET(page, lru);
1494 VMCOREINFO_OFFSET(pglist_data, node_zones);
1495 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1496#ifdef CONFIG_FLAT_NODE_MEM_MAP
1497 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1498#endif
1499 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1500 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1501 VMCOREINFO_OFFSET(pglist_data, node_id);
1502 VMCOREINFO_OFFSET(zone, free_area);
1503 VMCOREINFO_OFFSET(zone, vm_stat);
1504 VMCOREINFO_OFFSET(zone, spanned_pages);
1505 VMCOREINFO_OFFSET(free_area, free_list);
1506 VMCOREINFO_OFFSET(list_head, next);
1507 VMCOREINFO_OFFSET(list_head, prev);
1508 VMCOREINFO_OFFSET(vm_struct, addr);
1509 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1510 log_buf_kexec_setup();
1511 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1512 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1513 VMCOREINFO_NUMBER(PG_lru);
1514 VMCOREINFO_NUMBER(PG_private);
1515 VMCOREINFO_NUMBER(PG_swapcache);
1516
1517 arch_crash_save_vmcoreinfo();
1518 update_vmcoreinfo_note();
1519
1520 return 0;
1521}
1522
1523module_init(crash_save_vmcoreinfo_init)
1524
1525/*
1526 * Move into place and start executing a preloaded standalone
1527 * executable. If nothing was preloaded return an error.
1528 */
1529int kernel_kexec(void)
1530{
1531 int error = 0;
1532
1533 if (!mutex_trylock(&kexec_mutex))
1534 return -EBUSY;
1535 if (!kexec_image) {
1536 error = -EINVAL;
1537 goto Unlock;
1538 }
1539
1540#ifdef CONFIG_KEXEC_JUMP
1541 if (kexec_image->preserve_context) {
1542 lock_system_sleep();
1543 pm_prepare_console();
1544 error = freeze_processes();
1545 if (error) {
1546 error = -EBUSY;
1547 goto Restore_console;
1548 }
1549 suspend_console();
1550 error = dpm_suspend_start(PMSG_FREEZE);
1551 if (error)
1552 goto Resume_console;
1553 /* At this point, dpm_suspend_start() has been called,
1554 * but *not* dpm_suspend_end(). We *must* call
1555 * dpm_suspend_end() now. Otherwise, drivers for
1556 * some devices (e.g. interrupt controllers) become
1557 * desynchronized with the actual state of the
1558 * hardware at resume time, and evil weirdness ensues.
1559 */
1560 error = dpm_suspend_end(PMSG_FREEZE);
1561 if (error)
1562 goto Resume_devices;
1563 error = disable_nonboot_cpus();
1564 if (error)
1565 goto Enable_cpus;
1566 local_irq_disable();
1567 error = syscore_suspend();
1568 if (error)
1569 goto Enable_irqs;
1570 } else
1571#endif
1572 {
1573 kernel_restart_prepare(NULL);
1574 printk(KERN_EMERG "Starting new kernel\n");
1575 machine_shutdown();
1576 }
1577
1578 machine_kexec(kexec_image);
1579
1580#ifdef CONFIG_KEXEC_JUMP
1581 if (kexec_image->preserve_context) {
1582 syscore_resume();
1583 Enable_irqs:
1584 local_irq_enable();
1585 Enable_cpus:
1586 enable_nonboot_cpus();
1587 dpm_resume_start(PMSG_RESTORE);
1588 Resume_devices:
1589 dpm_resume_end(PMSG_RESTORE);
1590 Resume_console:
1591 resume_console();
1592 thaw_processes();
1593 Restore_console:
1594 pm_restore_console();
1595 unlock_system_sleep();
1596 }
1597#endif
1598
1599 Unlock:
1600 mutex_unlock(&kexec_mutex);
1601 return error;
1602}