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 <linux/utsname.h>
25#include <linux/numa.h>
26#include <linux/suspend.h>
27#include <linux/device.h>
28#include <linux/freezer.h>
29#include <linux/pm.h>
30#include <linux/cpu.h>
31#include <linux/console.h>
32#include <linux/vmalloc.h>
33#include <linux/swap.h>
34#include <linux/syscore_ops.h>
35#include <linux/compiler.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/* Flag to indicate we are going to kexec a new kernel */
52bool kexec_in_progress = false;
53
54/* Location of the reserved area for the crash kernel */
55struct resource crashk_res = {
56 .name = "Crash kernel",
57 .start = 0,
58 .end = 0,
59 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
60};
61struct resource crashk_low_res = {
62 .name = "Crash kernel",
63 .start = 0,
64 .end = 0,
65 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
66};
67
68int kexec_should_crash(struct task_struct *p)
69{
70 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
71 return 1;
72 return 0;
73}
74
75/*
76 * When kexec transitions to the new kernel there is a one-to-one
77 * mapping between physical and virtual addresses. On processors
78 * where you can disable the MMU this is trivial, and easy. For
79 * others it is still a simple predictable page table to setup.
80 *
81 * In that environment kexec copies the new kernel to its final
82 * resting place. This means I can only support memory whose
83 * physical address can fit in an unsigned long. In particular
84 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
85 * If the assembly stub has more restrictive requirements
86 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
87 * defined more restrictively in <asm/kexec.h>.
88 *
89 * The code for the transition from the current kernel to the
90 * the new kernel is placed in the control_code_buffer, whose size
91 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
92 * page of memory is necessary, but some architectures require more.
93 * Because this memory must be identity mapped in the transition from
94 * virtual to physical addresses it must live in the range
95 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
96 * modifiable.
97 *
98 * The assembly stub in the control code buffer is passed a linked list
99 * of descriptor pages detailing the source pages of the new kernel,
100 * and the destination addresses of those source pages. As this data
101 * structure is not used in the context of the current OS, it must
102 * be self-contained.
103 *
104 * The code has been made to work with highmem pages and will use a
105 * destination page in its final resting place (if it happens
106 * to allocate it). The end product of this is that most of the
107 * physical address space, and most of RAM can be used.
108 *
109 * Future directions include:
110 * - allocating a page table with the control code buffer identity
111 * mapped, to simplify machine_kexec and make kexec_on_panic more
112 * reliable.
113 */
114
115/*
116 * KIMAGE_NO_DEST is an impossible destination address..., for
117 * allocating pages whose destination address we do not care about.
118 */
119#define KIMAGE_NO_DEST (-1UL)
120
121static int kimage_is_destination_range(struct kimage *image,
122 unsigned long start, unsigned long end);
123static struct page *kimage_alloc_page(struct kimage *image,
124 gfp_t gfp_mask,
125 unsigned long dest);
126
127static int do_kimage_alloc(struct kimage **rimage, unsigned long entry,
128 unsigned long nr_segments,
129 struct kexec_segment __user *segments)
130{
131 size_t segment_bytes;
132 struct kimage *image;
133 unsigned long i;
134 int result;
135
136 /* Allocate a controlling structure */
137 result = -ENOMEM;
138 image = kzalloc(sizeof(*image), GFP_KERNEL);
139 if (!image)
140 goto out;
141
142 image->head = 0;
143 image->entry = &image->head;
144 image->last_entry = &image->head;
145 image->control_page = ~0; /* By default this does not apply */
146 image->start = entry;
147 image->type = KEXEC_TYPE_DEFAULT;
148
149 /* Initialize the list of control pages */
150 INIT_LIST_HEAD(&image->control_pages);
151
152 /* Initialize the list of destination pages */
153 INIT_LIST_HEAD(&image->dest_pages);
154
155 /* Initialize the list of unusable pages */
156 INIT_LIST_HEAD(&image->unuseable_pages);
157
158 /* Read in the segments */
159 image->nr_segments = nr_segments;
160 segment_bytes = nr_segments * sizeof(*segments);
161 result = copy_from_user(image->segment, segments, segment_bytes);
162 if (result) {
163 result = -EFAULT;
164 goto out;
165 }
166
167 /*
168 * Verify we have good destination addresses. The caller is
169 * responsible for making certain we don't attempt to load
170 * the new image into invalid or reserved areas of RAM. This
171 * just verifies it is an address we can use.
172 *
173 * Since the kernel does everything in page size chunks ensure
174 * the destination addresses are page aligned. Too many
175 * special cases crop of when we don't do this. The most
176 * insidious is getting overlapping destination addresses
177 * simply because addresses are changed to page size
178 * granularity.
179 */
180 result = -EADDRNOTAVAIL;
181 for (i = 0; i < nr_segments; i++) {
182 unsigned long mstart, mend;
183
184 mstart = image->segment[i].mem;
185 mend = mstart + image->segment[i].memsz;
186 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
187 goto out;
188 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
189 goto out;
190 }
191
192 /* Verify our destination addresses do not overlap.
193 * If we alloed overlapping destination addresses
194 * through very weird things can happen with no
195 * easy explanation as one segment stops on another.
196 */
197 result = -EINVAL;
198 for (i = 0; i < nr_segments; i++) {
199 unsigned long mstart, mend;
200 unsigned long j;
201
202 mstart = image->segment[i].mem;
203 mend = mstart + image->segment[i].memsz;
204 for (j = 0; j < i; j++) {
205 unsigned long pstart, pend;
206 pstart = image->segment[j].mem;
207 pend = pstart + image->segment[j].memsz;
208 /* Do the segments overlap ? */
209 if ((mend > pstart) && (mstart < pend))
210 goto out;
211 }
212 }
213
214 /* Ensure our buffer sizes are strictly less than
215 * our memory sizes. This should always be the case,
216 * and it is easier to check up front than to be surprised
217 * later on.
218 */
219 result = -EINVAL;
220 for (i = 0; i < nr_segments; i++) {
221 if (image->segment[i].bufsz > image->segment[i].memsz)
222 goto out;
223 }
224
225 result = 0;
226out:
227 if (result == 0)
228 *rimage = image;
229 else
230 kfree(image);
231
232 return result;
233
234}
235
236static void kimage_free_page_list(struct list_head *list);
237
238static int kimage_normal_alloc(struct kimage **rimage, unsigned long entry,
239 unsigned long nr_segments,
240 struct kexec_segment __user *segments)
241{
242 int result;
243 struct kimage *image;
244
245 /* Allocate and initialize a controlling structure */
246 image = NULL;
247 result = do_kimage_alloc(&image, entry, nr_segments, segments);
248 if (result)
249 goto out;
250
251 /*
252 * Find a location for the control code buffer, and add it
253 * the vector of segments so that it's pages will also be
254 * counted as destination pages.
255 */
256 result = -ENOMEM;
257 image->control_code_page = kimage_alloc_control_pages(image,
258 get_order(KEXEC_CONTROL_PAGE_SIZE));
259 if (!image->control_code_page) {
260 printk(KERN_ERR "Could not allocate control_code_buffer\n");
261 goto out_free;
262 }
263
264 image->swap_page = kimage_alloc_control_pages(image, 0);
265 if (!image->swap_page) {
266 printk(KERN_ERR "Could not allocate swap buffer\n");
267 goto out_free;
268 }
269
270 *rimage = image;
271 return 0;
272
273out_free:
274 kimage_free_page_list(&image->control_pages);
275 kfree(image);
276out:
277 return result;
278}
279
280static int kimage_crash_alloc(struct kimage **rimage, unsigned long entry,
281 unsigned long nr_segments,
282 struct kexec_segment __user *segments)
283{
284 int result;
285 struct kimage *image;
286 unsigned long i;
287
288 image = NULL;
289 /* Verify we have a valid entry point */
290 if ((entry < crashk_res.start) || (entry > crashk_res.end)) {
291 result = -EADDRNOTAVAIL;
292 goto out;
293 }
294
295 /* Allocate and initialize a controlling structure */
296 result = do_kimage_alloc(&image, entry, nr_segments, segments);
297 if (result)
298 goto out;
299
300 /* Enable the special crash kernel control page
301 * allocation policy.
302 */
303 image->control_page = crashk_res.start;
304 image->type = KEXEC_TYPE_CRASH;
305
306 /*
307 * Verify we have good destination addresses. Normally
308 * the caller is responsible for making certain we don't
309 * attempt to load the new image into invalid or reserved
310 * areas of RAM. But crash kernels are preloaded into a
311 * reserved area of ram. We must ensure the addresses
312 * are in the reserved area otherwise preloading the
313 * kernel could corrupt things.
314 */
315 result = -EADDRNOTAVAIL;
316 for (i = 0; i < nr_segments; i++) {
317 unsigned long mstart, mend;
318
319 mstart = image->segment[i].mem;
320 mend = mstart + image->segment[i].memsz - 1;
321 /* Ensure we are within the crash kernel limits */
322 if ((mstart < crashk_res.start) || (mend > crashk_res.end))
323 goto out_free;
324 }
325
326 /*
327 * Find a location for the control code buffer, and add
328 * the vector of segments so that it's pages will also be
329 * counted as destination pages.
330 */
331 result = -ENOMEM;
332 image->control_code_page = kimage_alloc_control_pages(image,
333 get_order(KEXEC_CONTROL_PAGE_SIZE));
334 if (!image->control_code_page) {
335 printk(KERN_ERR "Could not allocate control_code_buffer\n");
336 goto out_free;
337 }
338
339 *rimage = image;
340 return 0;
341
342out_free:
343 kfree(image);
344out:
345 return result;
346}
347
348static int kimage_is_destination_range(struct kimage *image,
349 unsigned long start,
350 unsigned long end)
351{
352 unsigned long i;
353
354 for (i = 0; i < image->nr_segments; i++) {
355 unsigned long mstart, mend;
356
357 mstart = image->segment[i].mem;
358 mend = mstart + image->segment[i].memsz;
359 if ((end > mstart) && (start < mend))
360 return 1;
361 }
362
363 return 0;
364}
365
366static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
367{
368 struct page *pages;
369
370 pages = alloc_pages(gfp_mask, order);
371 if (pages) {
372 unsigned int count, i;
373 pages->mapping = NULL;
374 set_page_private(pages, order);
375 count = 1 << order;
376 for (i = 0; i < count; i++)
377 SetPageReserved(pages + i);
378 }
379
380 return pages;
381}
382
383static void kimage_free_pages(struct page *page)
384{
385 unsigned int order, count, i;
386
387 order = page_private(page);
388 count = 1 << order;
389 for (i = 0; i < count; i++)
390 ClearPageReserved(page + i);
391 __free_pages(page, order);
392}
393
394static void kimage_free_page_list(struct list_head *list)
395{
396 struct list_head *pos, *next;
397
398 list_for_each_safe(pos, next, list) {
399 struct page *page;
400
401 page = list_entry(pos, struct page, lru);
402 list_del(&page->lru);
403 kimage_free_pages(page);
404 }
405}
406
407static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
408 unsigned int order)
409{
410 /* Control pages are special, they are the intermediaries
411 * that are needed while we copy the rest of the pages
412 * to their final resting place. As such they must
413 * not conflict with either the destination addresses
414 * or memory the kernel is already using.
415 *
416 * The only case where we really need more than one of
417 * these are for architectures where we cannot disable
418 * the MMU and must instead generate an identity mapped
419 * page table for all of the memory.
420 *
421 * At worst this runs in O(N) of the image size.
422 */
423 struct list_head extra_pages;
424 struct page *pages;
425 unsigned int count;
426
427 count = 1 << order;
428 INIT_LIST_HEAD(&extra_pages);
429
430 /* Loop while I can allocate a page and the page allocated
431 * is a destination page.
432 */
433 do {
434 unsigned long pfn, epfn, addr, eaddr;
435
436 pages = kimage_alloc_pages(GFP_KERNEL, order);
437 if (!pages)
438 break;
439 pfn = page_to_pfn(pages);
440 epfn = pfn + count;
441 addr = pfn << PAGE_SHIFT;
442 eaddr = epfn << PAGE_SHIFT;
443 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
444 kimage_is_destination_range(image, addr, eaddr)) {
445 list_add(&pages->lru, &extra_pages);
446 pages = NULL;
447 }
448 } while (!pages);
449
450 if (pages) {
451 /* Remember the allocated page... */
452 list_add(&pages->lru, &image->control_pages);
453
454 /* Because the page is already in it's destination
455 * location we will never allocate another page at
456 * that address. Therefore kimage_alloc_pages
457 * will not return it (again) and we don't need
458 * to give it an entry in image->segment[].
459 */
460 }
461 /* Deal with the destination pages I have inadvertently allocated.
462 *
463 * Ideally I would convert multi-page allocations into single
464 * page allocations, and add everything to image->dest_pages.
465 *
466 * For now it is simpler to just free the pages.
467 */
468 kimage_free_page_list(&extra_pages);
469
470 return pages;
471}
472
473static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
474 unsigned int order)
475{
476 /* Control pages are special, they are the intermediaries
477 * that are needed while we copy the rest of the pages
478 * to their final resting place. As such they must
479 * not conflict with either the destination addresses
480 * or memory the kernel is already using.
481 *
482 * Control pages are also the only pags we must allocate
483 * when loading a crash kernel. All of the other pages
484 * are specified by the segments and we just memcpy
485 * into them directly.
486 *
487 * The only case where we really need more than one of
488 * these are for architectures where we cannot disable
489 * the MMU and must instead generate an identity mapped
490 * page table for all of the memory.
491 *
492 * Given the low demand this implements a very simple
493 * allocator that finds the first hole of the appropriate
494 * size in the reserved memory region, and allocates all
495 * of the memory up to and including the hole.
496 */
497 unsigned long hole_start, hole_end, size;
498 struct page *pages;
499
500 pages = NULL;
501 size = (1 << order) << PAGE_SHIFT;
502 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
503 hole_end = hole_start + size - 1;
504 while (hole_end <= crashk_res.end) {
505 unsigned long i;
506
507 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
508 break;
509 /* See if I overlap any of the segments */
510 for (i = 0; i < image->nr_segments; i++) {
511 unsigned long mstart, mend;
512
513 mstart = image->segment[i].mem;
514 mend = mstart + image->segment[i].memsz - 1;
515 if ((hole_end >= mstart) && (hole_start <= mend)) {
516 /* Advance the hole to the end of the segment */
517 hole_start = (mend + (size - 1)) & ~(size - 1);
518 hole_end = hole_start + size - 1;
519 break;
520 }
521 }
522 /* If I don't overlap any segments I have found my hole! */
523 if (i == image->nr_segments) {
524 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
525 break;
526 }
527 }
528 if (pages)
529 image->control_page = hole_end;
530
531 return pages;
532}
533
534
535struct page *kimage_alloc_control_pages(struct kimage *image,
536 unsigned int order)
537{
538 struct page *pages = NULL;
539
540 switch (image->type) {
541 case KEXEC_TYPE_DEFAULT:
542 pages = kimage_alloc_normal_control_pages(image, order);
543 break;
544 case KEXEC_TYPE_CRASH:
545 pages = kimage_alloc_crash_control_pages(image, order);
546 break;
547 }
548
549 return pages;
550}
551
552static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
553{
554 if (*image->entry != 0)
555 image->entry++;
556
557 if (image->entry == image->last_entry) {
558 kimage_entry_t *ind_page;
559 struct page *page;
560
561 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
562 if (!page)
563 return -ENOMEM;
564
565 ind_page = page_address(page);
566 *image->entry = virt_to_phys(ind_page) | IND_INDIRECTION;
567 image->entry = ind_page;
568 image->last_entry = ind_page +
569 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
570 }
571 *image->entry = entry;
572 image->entry++;
573 *image->entry = 0;
574
575 return 0;
576}
577
578static int kimage_set_destination(struct kimage *image,
579 unsigned long destination)
580{
581 int result;
582
583 destination &= PAGE_MASK;
584 result = kimage_add_entry(image, destination | IND_DESTINATION);
585 if (result == 0)
586 image->destination = destination;
587
588 return result;
589}
590
591
592static int kimage_add_page(struct kimage *image, unsigned long page)
593{
594 int result;
595
596 page &= PAGE_MASK;
597 result = kimage_add_entry(image, page | IND_SOURCE);
598 if (result == 0)
599 image->destination += PAGE_SIZE;
600
601 return result;
602}
603
604
605static void kimage_free_extra_pages(struct kimage *image)
606{
607 /* Walk through and free any extra destination pages I may have */
608 kimage_free_page_list(&image->dest_pages);
609
610 /* Walk through and free any unusable pages I have cached */
611 kimage_free_page_list(&image->unuseable_pages);
612
613}
614static void kimage_terminate(struct kimage *image)
615{
616 if (*image->entry != 0)
617 image->entry++;
618
619 *image->entry = IND_DONE;
620}
621
622#define for_each_kimage_entry(image, ptr, entry) \
623 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
624 ptr = (entry & IND_INDIRECTION)? \
625 phys_to_virt((entry & PAGE_MASK)): ptr +1)
626
627static void kimage_free_entry(kimage_entry_t entry)
628{
629 struct page *page;
630
631 page = pfn_to_page(entry >> PAGE_SHIFT);
632 kimage_free_pages(page);
633}
634
635static void kimage_free(struct kimage *image)
636{
637 kimage_entry_t *ptr, entry;
638 kimage_entry_t ind = 0;
639
640 if (!image)
641 return;
642
643 kimage_free_extra_pages(image);
644 for_each_kimage_entry(image, ptr, entry) {
645 if (entry & IND_INDIRECTION) {
646 /* Free the previous indirection page */
647 if (ind & IND_INDIRECTION)
648 kimage_free_entry(ind);
649 /* Save this indirection page until we are
650 * done with it.
651 */
652 ind = entry;
653 }
654 else if (entry & IND_SOURCE)
655 kimage_free_entry(entry);
656 }
657 /* Free the final indirection page */
658 if (ind & IND_INDIRECTION)
659 kimage_free_entry(ind);
660
661 /* Handle any machine specific cleanup */
662 machine_kexec_cleanup(image);
663
664 /* Free the kexec control pages... */
665 kimage_free_page_list(&image->control_pages);
666 kfree(image);
667}
668
669static kimage_entry_t *kimage_dst_used(struct kimage *image,
670 unsigned long page)
671{
672 kimage_entry_t *ptr, entry;
673 unsigned long destination = 0;
674
675 for_each_kimage_entry(image, ptr, entry) {
676 if (entry & IND_DESTINATION)
677 destination = entry & PAGE_MASK;
678 else if (entry & IND_SOURCE) {
679 if (page == destination)
680 return ptr;
681 destination += PAGE_SIZE;
682 }
683 }
684
685 return NULL;
686}
687
688static struct page *kimage_alloc_page(struct kimage *image,
689 gfp_t gfp_mask,
690 unsigned long destination)
691{
692 /*
693 * Here we implement safeguards to ensure that a source page
694 * is not copied to its destination page before the data on
695 * the destination page is no longer useful.
696 *
697 * To do this we maintain the invariant that a source page is
698 * either its own destination page, or it is not a
699 * destination page at all.
700 *
701 * That is slightly stronger than required, but the proof
702 * that no problems will not occur is trivial, and the
703 * implementation is simply to verify.
704 *
705 * When allocating all pages normally this algorithm will run
706 * in O(N) time, but in the worst case it will run in O(N^2)
707 * time. If the runtime is a problem the data structures can
708 * be fixed.
709 */
710 struct page *page;
711 unsigned long addr;
712
713 /*
714 * Walk through the list of destination pages, and see if I
715 * have a match.
716 */
717 list_for_each_entry(page, &image->dest_pages, lru) {
718 addr = page_to_pfn(page) << PAGE_SHIFT;
719 if (addr == destination) {
720 list_del(&page->lru);
721 return page;
722 }
723 }
724 page = NULL;
725 while (1) {
726 kimage_entry_t *old;
727
728 /* Allocate a page, if we run out of memory give up */
729 page = kimage_alloc_pages(gfp_mask, 0);
730 if (!page)
731 return NULL;
732 /* If the page cannot be used file it away */
733 if (page_to_pfn(page) >
734 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
735 list_add(&page->lru, &image->unuseable_pages);
736 continue;
737 }
738 addr = page_to_pfn(page) << PAGE_SHIFT;
739
740 /* If it is the destination page we want use it */
741 if (addr == destination)
742 break;
743
744 /* If the page is not a destination page use it */
745 if (!kimage_is_destination_range(image, addr,
746 addr + PAGE_SIZE))
747 break;
748
749 /*
750 * I know that the page is someones destination page.
751 * See if there is already a source page for this
752 * destination page. And if so swap the source pages.
753 */
754 old = kimage_dst_used(image, addr);
755 if (old) {
756 /* If so move it */
757 unsigned long old_addr;
758 struct page *old_page;
759
760 old_addr = *old & PAGE_MASK;
761 old_page = pfn_to_page(old_addr >> PAGE_SHIFT);
762 copy_highpage(page, old_page);
763 *old = addr | (*old & ~PAGE_MASK);
764
765 /* The old page I have found cannot be a
766 * destination page, so return it if it's
767 * gfp_flags honor the ones passed in.
768 */
769 if (!(gfp_mask & __GFP_HIGHMEM) &&
770 PageHighMem(old_page)) {
771 kimage_free_pages(old_page);
772 continue;
773 }
774 addr = old_addr;
775 page = old_page;
776 break;
777 }
778 else {
779 /* Place the page on the destination list I
780 * will use it later.
781 */
782 list_add(&page->lru, &image->dest_pages);
783 }
784 }
785
786 return page;
787}
788
789static int kimage_load_normal_segment(struct kimage *image,
790 struct kexec_segment *segment)
791{
792 unsigned long maddr;
793 size_t ubytes, mbytes;
794 int result;
795 unsigned char __user *buf;
796
797 result = 0;
798 buf = segment->buf;
799 ubytes = segment->bufsz;
800 mbytes = segment->memsz;
801 maddr = segment->mem;
802
803 result = kimage_set_destination(image, maddr);
804 if (result < 0)
805 goto out;
806
807 while (mbytes) {
808 struct page *page;
809 char *ptr;
810 size_t uchunk, mchunk;
811
812 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
813 if (!page) {
814 result = -ENOMEM;
815 goto out;
816 }
817 result = kimage_add_page(image, page_to_pfn(page)
818 << PAGE_SHIFT);
819 if (result < 0)
820 goto out;
821
822 ptr = kmap(page);
823 /* Start with a clear page */
824 clear_page(ptr);
825 ptr += maddr & ~PAGE_MASK;
826 mchunk = min_t(size_t, mbytes,
827 PAGE_SIZE - (maddr & ~PAGE_MASK));
828 uchunk = min(ubytes, mchunk);
829
830 result = copy_from_user(ptr, buf, uchunk);
831 kunmap(page);
832 if (result) {
833 result = -EFAULT;
834 goto out;
835 }
836 ubytes -= uchunk;
837 maddr += mchunk;
838 buf += mchunk;
839 mbytes -= mchunk;
840 }
841out:
842 return result;
843}
844
845static int kimage_load_crash_segment(struct kimage *image,
846 struct kexec_segment *segment)
847{
848 /* For crash dumps kernels we simply copy the data from
849 * user space to it's destination.
850 * We do things a page at a time for the sake of kmap.
851 */
852 unsigned long maddr;
853 size_t ubytes, mbytes;
854 int result;
855 unsigned char __user *buf;
856
857 result = 0;
858 buf = segment->buf;
859 ubytes = segment->bufsz;
860 mbytes = segment->memsz;
861 maddr = segment->mem;
862 while (mbytes) {
863 struct page *page;
864 char *ptr;
865 size_t uchunk, mchunk;
866
867 page = pfn_to_page(maddr >> PAGE_SHIFT);
868 if (!page) {
869 result = -ENOMEM;
870 goto out;
871 }
872 ptr = kmap(page);
873 ptr += maddr & ~PAGE_MASK;
874 mchunk = min_t(size_t, mbytes,
875 PAGE_SIZE - (maddr & ~PAGE_MASK));
876 uchunk = min(ubytes, mchunk);
877 if (mchunk > uchunk) {
878 /* Zero the trailing part of the page */
879 memset(ptr + uchunk, 0, mchunk - uchunk);
880 }
881 result = copy_from_user(ptr, buf, uchunk);
882 kexec_flush_icache_page(page);
883 kunmap(page);
884 if (result) {
885 result = -EFAULT;
886 goto out;
887 }
888 ubytes -= uchunk;
889 maddr += mchunk;
890 buf += mchunk;
891 mbytes -= mchunk;
892 }
893out:
894 return result;
895}
896
897static int kimage_load_segment(struct kimage *image,
898 struct kexec_segment *segment)
899{
900 int result = -ENOMEM;
901
902 switch (image->type) {
903 case KEXEC_TYPE_DEFAULT:
904 result = kimage_load_normal_segment(image, segment);
905 break;
906 case KEXEC_TYPE_CRASH:
907 result = kimage_load_crash_segment(image, segment);
908 break;
909 }
910
911 return result;
912}
913
914/*
915 * Exec Kernel system call: for obvious reasons only root may call it.
916 *
917 * This call breaks up into three pieces.
918 * - A generic part which loads the new kernel from the current
919 * address space, and very carefully places the data in the
920 * allocated pages.
921 *
922 * - A generic part that interacts with the kernel and tells all of
923 * the devices to shut down. Preventing on-going dmas, and placing
924 * the devices in a consistent state so a later kernel can
925 * reinitialize them.
926 *
927 * - A machine specific part that includes the syscall number
928 * and then copies the image to it's final destination. And
929 * jumps into the image at entry.
930 *
931 * kexec does not sync, or unmount filesystems so if you need
932 * that to happen you need to do that yourself.
933 */
934struct kimage *kexec_image;
935struct kimage *kexec_crash_image;
936int kexec_load_disabled;
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) || kexec_load_disabled)
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
1043COMPAT_SYSCALL_DEFINE4(kexec_load, compat_ulong_t, entry,
1044 compat_ulong_t, nr_segments,
1045 struct compat_kexec_segment __user *, segments,
1046 compat_ulong_t, 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 free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT));
1121}
1122
1123int crash_shrink_memory(unsigned long new_size)
1124{
1125 int ret = 0;
1126 unsigned long start, end;
1127 unsigned long old_size;
1128 struct resource *ram_res;
1129
1130 mutex_lock(&kexec_mutex);
1131
1132 if (kexec_crash_image) {
1133 ret = -ENOENT;
1134 goto unlock;
1135 }
1136 start = crashk_res.start;
1137 end = crashk_res.end;
1138 old_size = (end == 0) ? 0 : end - start + 1;
1139 if (new_size >= old_size) {
1140 ret = (new_size == old_size) ? 0 : -EINVAL;
1141 goto unlock;
1142 }
1143
1144 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
1145 if (!ram_res) {
1146 ret = -ENOMEM;
1147 goto unlock;
1148 }
1149
1150 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
1151 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
1152
1153 crash_map_reserved_pages();
1154 crash_free_reserved_phys_range(end, crashk_res.end);
1155
1156 if ((start == end) && (crashk_res.parent != NULL))
1157 release_resource(&crashk_res);
1158
1159 ram_res->start = end;
1160 ram_res->end = crashk_res.end;
1161 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_MEM;
1162 ram_res->name = "System RAM";
1163
1164 crashk_res.end = end - 1;
1165
1166 insert_resource(&iomem_resource, ram_res);
1167 crash_unmap_reserved_pages();
1168
1169unlock:
1170 mutex_unlock(&kexec_mutex);
1171 return ret;
1172}
1173
1174static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
1175 size_t data_len)
1176{
1177 struct elf_note note;
1178
1179 note.n_namesz = strlen(name) + 1;
1180 note.n_descsz = data_len;
1181 note.n_type = type;
1182 memcpy(buf, ¬e, sizeof(note));
1183 buf += (sizeof(note) + 3)/4;
1184 memcpy(buf, name, note.n_namesz);
1185 buf += (note.n_namesz + 3)/4;
1186 memcpy(buf, data, note.n_descsz);
1187 buf += (note.n_descsz + 3)/4;
1188
1189 return buf;
1190}
1191
1192static void final_note(u32 *buf)
1193{
1194 struct elf_note note;
1195
1196 note.n_namesz = 0;
1197 note.n_descsz = 0;
1198 note.n_type = 0;
1199 memcpy(buf, ¬e, sizeof(note));
1200}
1201
1202void crash_save_cpu(struct pt_regs *regs, int cpu)
1203{
1204 struct elf_prstatus prstatus;
1205 u32 *buf;
1206
1207 if ((cpu < 0) || (cpu >= nr_cpu_ids))
1208 return;
1209
1210 /* Using ELF notes here is opportunistic.
1211 * I need a well defined structure format
1212 * for the data I pass, and I need tags
1213 * on the data to indicate what information I have
1214 * squirrelled away. ELF notes happen to provide
1215 * all of that, so there is no need to invent something new.
1216 */
1217 buf = (u32*)per_cpu_ptr(crash_notes, cpu);
1218 if (!buf)
1219 return;
1220 memset(&prstatus, 0, sizeof(prstatus));
1221 prstatus.pr_pid = current->pid;
1222 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
1223 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
1224 &prstatus, sizeof(prstatus));
1225 final_note(buf);
1226}
1227
1228static int __init crash_notes_memory_init(void)
1229{
1230 /* Allocate memory for saving cpu registers. */
1231 crash_notes = alloc_percpu(note_buf_t);
1232 if (!crash_notes) {
1233 printk("Kexec: Memory allocation for saving cpu register"
1234 " states failed\n");
1235 return -ENOMEM;
1236 }
1237 return 0;
1238}
1239subsys_initcall(crash_notes_memory_init);
1240
1241
1242/*
1243 * parsing the "crashkernel" commandline
1244 *
1245 * this code is intended to be called from architecture specific code
1246 */
1247
1248
1249/*
1250 * This function parses command lines in the format
1251 *
1252 * crashkernel=ramsize-range:size[,...][@offset]
1253 *
1254 * The function returns 0 on success and -EINVAL on failure.
1255 */
1256static int __init parse_crashkernel_mem(char *cmdline,
1257 unsigned long long system_ram,
1258 unsigned long long *crash_size,
1259 unsigned long long *crash_base)
1260{
1261 char *cur = cmdline, *tmp;
1262
1263 /* for each entry of the comma-separated list */
1264 do {
1265 unsigned long long start, end = ULLONG_MAX, size;
1266
1267 /* get the start of the range */
1268 start = memparse(cur, &tmp);
1269 if (cur == tmp) {
1270 pr_warning("crashkernel: Memory value expected\n");
1271 return -EINVAL;
1272 }
1273 cur = tmp;
1274 if (*cur != '-') {
1275 pr_warning("crashkernel: '-' expected\n");
1276 return -EINVAL;
1277 }
1278 cur++;
1279
1280 /* if no ':' is here, than we read the end */
1281 if (*cur != ':') {
1282 end = memparse(cur, &tmp);
1283 if (cur == tmp) {
1284 pr_warning("crashkernel: Memory "
1285 "value expected\n");
1286 return -EINVAL;
1287 }
1288 cur = tmp;
1289 if (end <= start) {
1290 pr_warning("crashkernel: end <= start\n");
1291 return -EINVAL;
1292 }
1293 }
1294
1295 if (*cur != ':') {
1296 pr_warning("crashkernel: ':' expected\n");
1297 return -EINVAL;
1298 }
1299 cur++;
1300
1301 size = memparse(cur, &tmp);
1302 if (cur == tmp) {
1303 pr_warning("Memory value expected\n");
1304 return -EINVAL;
1305 }
1306 cur = tmp;
1307 if (size >= system_ram) {
1308 pr_warning("crashkernel: invalid size\n");
1309 return -EINVAL;
1310 }
1311
1312 /* match ? */
1313 if (system_ram >= start && system_ram < end) {
1314 *crash_size = size;
1315 break;
1316 }
1317 } while (*cur++ == ',');
1318
1319 if (*crash_size > 0) {
1320 while (*cur && *cur != ' ' && *cur != '@')
1321 cur++;
1322 if (*cur == '@') {
1323 cur++;
1324 *crash_base = memparse(cur, &tmp);
1325 if (cur == tmp) {
1326 pr_warning("Memory value expected "
1327 "after '@'\n");
1328 return -EINVAL;
1329 }
1330 }
1331 }
1332
1333 return 0;
1334}
1335
1336/*
1337 * That function parses "simple" (old) crashkernel command lines like
1338 *
1339 * crashkernel=size[@offset]
1340 *
1341 * It returns 0 on success and -EINVAL on failure.
1342 */
1343static int __init parse_crashkernel_simple(char *cmdline,
1344 unsigned long long *crash_size,
1345 unsigned long long *crash_base)
1346{
1347 char *cur = cmdline;
1348
1349 *crash_size = memparse(cmdline, &cur);
1350 if (cmdline == cur) {
1351 pr_warning("crashkernel: memory value expected\n");
1352 return -EINVAL;
1353 }
1354
1355 if (*cur == '@')
1356 *crash_base = memparse(cur+1, &cur);
1357 else if (*cur != ' ' && *cur != '\0') {
1358 pr_warning("crashkernel: unrecognized char\n");
1359 return -EINVAL;
1360 }
1361
1362 return 0;
1363}
1364
1365#define SUFFIX_HIGH 0
1366#define SUFFIX_LOW 1
1367#define SUFFIX_NULL 2
1368static __initdata char *suffix_tbl[] = {
1369 [SUFFIX_HIGH] = ",high",
1370 [SUFFIX_LOW] = ",low",
1371 [SUFFIX_NULL] = NULL,
1372};
1373
1374/*
1375 * That function parses "suffix" crashkernel command lines like
1376 *
1377 * crashkernel=size,[high|low]
1378 *
1379 * It returns 0 on success and -EINVAL on failure.
1380 */
1381static int __init parse_crashkernel_suffix(char *cmdline,
1382 unsigned long long *crash_size,
1383 unsigned long long *crash_base,
1384 const char *suffix)
1385{
1386 char *cur = cmdline;
1387
1388 *crash_size = memparse(cmdline, &cur);
1389 if (cmdline == cur) {
1390 pr_warn("crashkernel: memory value expected\n");
1391 return -EINVAL;
1392 }
1393
1394 /* check with suffix */
1395 if (strncmp(cur, suffix, strlen(suffix))) {
1396 pr_warn("crashkernel: unrecognized char\n");
1397 return -EINVAL;
1398 }
1399 cur += strlen(suffix);
1400 if (*cur != ' ' && *cur != '\0') {
1401 pr_warn("crashkernel: unrecognized char\n");
1402 return -EINVAL;
1403 }
1404
1405 return 0;
1406}
1407
1408static __init char *get_last_crashkernel(char *cmdline,
1409 const char *name,
1410 const char *suffix)
1411{
1412 char *p = cmdline, *ck_cmdline = NULL;
1413
1414 /* find crashkernel and use the last one if there are more */
1415 p = strstr(p, name);
1416 while (p) {
1417 char *end_p = strchr(p, ' ');
1418 char *q;
1419
1420 if (!end_p)
1421 end_p = p + strlen(p);
1422
1423 if (!suffix) {
1424 int i;
1425
1426 /* skip the one with any known suffix */
1427 for (i = 0; suffix_tbl[i]; i++) {
1428 q = end_p - strlen(suffix_tbl[i]);
1429 if (!strncmp(q, suffix_tbl[i],
1430 strlen(suffix_tbl[i])))
1431 goto next;
1432 }
1433 ck_cmdline = p;
1434 } else {
1435 q = end_p - strlen(suffix);
1436 if (!strncmp(q, suffix, strlen(suffix)))
1437 ck_cmdline = p;
1438 }
1439next:
1440 p = strstr(p+1, name);
1441 }
1442
1443 if (!ck_cmdline)
1444 return NULL;
1445
1446 return ck_cmdline;
1447}
1448
1449static int __init __parse_crashkernel(char *cmdline,
1450 unsigned long long system_ram,
1451 unsigned long long *crash_size,
1452 unsigned long long *crash_base,
1453 const char *name,
1454 const char *suffix)
1455{
1456 char *first_colon, *first_space;
1457 char *ck_cmdline;
1458
1459 BUG_ON(!crash_size || !crash_base);
1460 *crash_size = 0;
1461 *crash_base = 0;
1462
1463 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
1464
1465 if (!ck_cmdline)
1466 return -EINVAL;
1467
1468 ck_cmdline += strlen(name);
1469
1470 if (suffix)
1471 return parse_crashkernel_suffix(ck_cmdline, crash_size,
1472 crash_base, suffix);
1473 /*
1474 * if the commandline contains a ':', then that's the extended
1475 * syntax -- if not, it must be the classic syntax
1476 */
1477 first_colon = strchr(ck_cmdline, ':');
1478 first_space = strchr(ck_cmdline, ' ');
1479 if (first_colon && (!first_space || first_colon < first_space))
1480 return parse_crashkernel_mem(ck_cmdline, system_ram,
1481 crash_size, crash_base);
1482
1483 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
1484}
1485
1486/*
1487 * That function is the entry point for command line parsing and should be
1488 * called from the arch-specific code.
1489 */
1490int __init parse_crashkernel(char *cmdline,
1491 unsigned long long system_ram,
1492 unsigned long long *crash_size,
1493 unsigned long long *crash_base)
1494{
1495 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1496 "crashkernel=", NULL);
1497}
1498
1499int __init parse_crashkernel_high(char *cmdline,
1500 unsigned long long system_ram,
1501 unsigned long long *crash_size,
1502 unsigned long long *crash_base)
1503{
1504 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1505 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
1506}
1507
1508int __init parse_crashkernel_low(char *cmdline,
1509 unsigned long long system_ram,
1510 unsigned long long *crash_size,
1511 unsigned long long *crash_base)
1512{
1513 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
1514 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
1515}
1516
1517static void update_vmcoreinfo_note(void)
1518{
1519 u32 *buf = vmcoreinfo_note;
1520
1521 if (!vmcoreinfo_size)
1522 return;
1523 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
1524 vmcoreinfo_size);
1525 final_note(buf);
1526}
1527
1528void crash_save_vmcoreinfo(void)
1529{
1530 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
1531 update_vmcoreinfo_note();
1532}
1533
1534void vmcoreinfo_append_str(const char *fmt, ...)
1535{
1536 va_list args;
1537 char buf[0x50];
1538 size_t r;
1539
1540 va_start(args, fmt);
1541 r = vscnprintf(buf, sizeof(buf), fmt, args);
1542 va_end(args);
1543
1544 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
1545
1546 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
1547
1548 vmcoreinfo_size += r;
1549}
1550
1551/*
1552 * provide an empty default implementation here -- architecture
1553 * code may override this
1554 */
1555void __weak arch_crash_save_vmcoreinfo(void)
1556{}
1557
1558unsigned long __weak paddr_vmcoreinfo_note(void)
1559{
1560 return __pa((unsigned long)(char *)&vmcoreinfo_note);
1561}
1562
1563static int __init crash_save_vmcoreinfo_init(void)
1564{
1565 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
1566 VMCOREINFO_PAGESIZE(PAGE_SIZE);
1567
1568 VMCOREINFO_SYMBOL(init_uts_ns);
1569 VMCOREINFO_SYMBOL(node_online_map);
1570#ifdef CONFIG_MMU
1571 VMCOREINFO_SYMBOL(swapper_pg_dir);
1572#endif
1573 VMCOREINFO_SYMBOL(_stext);
1574 VMCOREINFO_SYMBOL(vmap_area_list);
1575
1576#ifndef CONFIG_NEED_MULTIPLE_NODES
1577 VMCOREINFO_SYMBOL(mem_map);
1578 VMCOREINFO_SYMBOL(contig_page_data);
1579#endif
1580#ifdef CONFIG_SPARSEMEM
1581 VMCOREINFO_SYMBOL(mem_section);
1582 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
1583 VMCOREINFO_STRUCT_SIZE(mem_section);
1584 VMCOREINFO_OFFSET(mem_section, section_mem_map);
1585#endif
1586 VMCOREINFO_STRUCT_SIZE(page);
1587 VMCOREINFO_STRUCT_SIZE(pglist_data);
1588 VMCOREINFO_STRUCT_SIZE(zone);
1589 VMCOREINFO_STRUCT_SIZE(free_area);
1590 VMCOREINFO_STRUCT_SIZE(list_head);
1591 VMCOREINFO_SIZE(nodemask_t);
1592 VMCOREINFO_OFFSET(page, flags);
1593 VMCOREINFO_OFFSET(page, _count);
1594 VMCOREINFO_OFFSET(page, mapping);
1595 VMCOREINFO_OFFSET(page, lru);
1596 VMCOREINFO_OFFSET(page, _mapcount);
1597 VMCOREINFO_OFFSET(page, private);
1598 VMCOREINFO_OFFSET(pglist_data, node_zones);
1599 VMCOREINFO_OFFSET(pglist_data, nr_zones);
1600#ifdef CONFIG_FLAT_NODE_MEM_MAP
1601 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
1602#endif
1603 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
1604 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
1605 VMCOREINFO_OFFSET(pglist_data, node_id);
1606 VMCOREINFO_OFFSET(zone, free_area);
1607 VMCOREINFO_OFFSET(zone, vm_stat);
1608 VMCOREINFO_OFFSET(zone, spanned_pages);
1609 VMCOREINFO_OFFSET(free_area, free_list);
1610 VMCOREINFO_OFFSET(list_head, next);
1611 VMCOREINFO_OFFSET(list_head, prev);
1612 VMCOREINFO_OFFSET(vmap_area, va_start);
1613 VMCOREINFO_OFFSET(vmap_area, list);
1614 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1615 log_buf_kexec_setup();
1616 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1617 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1618 VMCOREINFO_NUMBER(PG_lru);
1619 VMCOREINFO_NUMBER(PG_private);
1620 VMCOREINFO_NUMBER(PG_swapcache);
1621 VMCOREINFO_NUMBER(PG_slab);
1622#ifdef CONFIG_MEMORY_FAILURE
1623 VMCOREINFO_NUMBER(PG_hwpoison);
1624#endif
1625 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
1626
1627 arch_crash_save_vmcoreinfo();
1628 update_vmcoreinfo_note();
1629
1630 return 0;
1631}
1632
1633subsys_initcall(crash_save_vmcoreinfo_init);
1634
1635/*
1636 * Move into place and start executing a preloaded standalone
1637 * executable. If nothing was preloaded return an error.
1638 */
1639int kernel_kexec(void)
1640{
1641 int error = 0;
1642
1643 if (!mutex_trylock(&kexec_mutex))
1644 return -EBUSY;
1645 if (!kexec_image) {
1646 error = -EINVAL;
1647 goto Unlock;
1648 }
1649
1650#ifdef CONFIG_KEXEC_JUMP
1651 if (kexec_image->preserve_context) {
1652 lock_system_sleep();
1653 pm_prepare_console();
1654 error = freeze_processes();
1655 if (error) {
1656 error = -EBUSY;
1657 goto Restore_console;
1658 }
1659 suspend_console();
1660 error = dpm_suspend_start(PMSG_FREEZE);
1661 if (error)
1662 goto Resume_console;
1663 /* At this point, dpm_suspend_start() has been called,
1664 * but *not* dpm_suspend_end(). We *must* call
1665 * dpm_suspend_end() now. Otherwise, drivers for
1666 * some devices (e.g. interrupt controllers) become
1667 * desynchronized with the actual state of the
1668 * hardware at resume time, and evil weirdness ensues.
1669 */
1670 error = dpm_suspend_end(PMSG_FREEZE);
1671 if (error)
1672 goto Resume_devices;
1673 error = disable_nonboot_cpus();
1674 if (error)
1675 goto Enable_cpus;
1676 local_irq_disable();
1677 error = syscore_suspend();
1678 if (error)
1679 goto Enable_irqs;
1680 } else
1681#endif
1682 {
1683 kexec_in_progress = true;
1684 kernel_restart_prepare(NULL);
1685 migrate_to_reboot_cpu();
1686
1687 /*
1688 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1689 * no further code needs to use CPU hotplug (which is true in
1690 * the reboot case). However, the kexec path depends on using
1691 * CPU hotplug again; so re-enable it here.
1692 */
1693 cpu_hotplug_enable();
1694 printk(KERN_EMERG "Starting new kernel\n");
1695 machine_shutdown();
1696 }
1697
1698 machine_kexec(kexec_image);
1699
1700#ifdef CONFIG_KEXEC_JUMP
1701 if (kexec_image->preserve_context) {
1702 syscore_resume();
1703 Enable_irqs:
1704 local_irq_enable();
1705 Enable_cpus:
1706 enable_nonboot_cpus();
1707 dpm_resume_start(PMSG_RESTORE);
1708 Resume_devices:
1709 dpm_resume_end(PMSG_RESTORE);
1710 Resume_console:
1711 resume_console();
1712 thaw_processes();
1713 Restore_console:
1714 pm_restore_console();
1715 unlock_system_sleep();
1716 }
1717#endif
1718
1719 Unlock:
1720 mutex_unlock(&kexec_mutex);
1721 return error;
1722}