1 // SPDX-License-Identifier: GPL-2.0-only <<
2 /* 1 /*
3 * kexec.c - kexec system call core code. 2 * kexec.c - kexec system call core code.
4 * Copyright (C) 2002-2004 Eric Biederman <eb 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.
5 */ 7 */
6 8
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 10
9 #include <linux/btf.h> <<
10 #include <linux/capability.h> 11 #include <linux/capability.h>
11 #include <linux/mm.h> 12 #include <linux/mm.h>
12 #include <linux/file.h> 13 #include <linux/file.h>
13 #include <linux/slab.h> 14 #include <linux/slab.h>
14 #include <linux/fs.h> 15 #include <linux/fs.h>
15 #include <linux/kexec.h> 16 #include <linux/kexec.h>
16 #include <linux/mutex.h> 17 #include <linux/mutex.h>
17 #include <linux/list.h> 18 #include <linux/list.h>
18 #include <linux/highmem.h> 19 #include <linux/highmem.h>
19 #include <linux/syscalls.h> 20 #include <linux/syscalls.h>
20 #include <linux/reboot.h> 21 #include <linux/reboot.h>
21 #include <linux/ioport.h> 22 #include <linux/ioport.h>
22 #include <linux/hardirq.h> 23 #include <linux/hardirq.h>
23 #include <linux/elf.h> 24 #include <linux/elf.h>
24 #include <linux/elfcore.h> 25 #include <linux/elfcore.h>
25 #include <linux/utsname.h> 26 #include <linux/utsname.h>
26 #include <linux/numa.h> 27 #include <linux/numa.h>
27 #include <linux/suspend.h> 28 #include <linux/suspend.h>
28 #include <linux/device.h> 29 #include <linux/device.h>
29 #include <linux/freezer.h> 30 #include <linux/freezer.h>
30 #include <linux/panic_notifier.h> <<
31 #include <linux/pm.h> 31 #include <linux/pm.h>
32 #include <linux/cpu.h> 32 #include <linux/cpu.h>
33 #include <linux/uaccess.h> 33 #include <linux/uaccess.h>
34 #include <linux/io.h> 34 #include <linux/io.h>
35 #include <linux/console.h> 35 #include <linux/console.h>
36 #include <linux/vmalloc.h> 36 #include <linux/vmalloc.h>
37 #include <linux/swap.h> 37 #include <linux/swap.h>
38 #include <linux/syscore_ops.h> 38 #include <linux/syscore_ops.h>
39 #include <linux/compiler.h> 39 #include <linux/compiler.h>
40 #include <linux/hugetlb.h> 40 #include <linux/hugetlb.h>
41 #include <linux/objtool.h> <<
42 #include <linux/kmsg_dump.h> <<
43 41
44 #include <asm/page.h> 42 #include <asm/page.h>
45 #include <asm/sections.h> 43 #include <asm/sections.h>
46 44
47 #include <crypto/hash.h> 45 #include <crypto/hash.h>
>> 46 #include <crypto/sha.h>
48 #include "kexec_internal.h" 47 #include "kexec_internal.h"
49 48
50 atomic_t __kexec_lock = ATOMIC_INIT(0); !! 49 DEFINE_MUTEX(kexec_mutex);
>> 50
>> 51 /* Per cpu memory for storing cpu states in case of system crash. */
>> 52 note_buf_t __percpu *crash_notes;
>> 53
>> 54 /* vmcoreinfo stuff */
>> 55 static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
>> 56 u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
>> 57 size_t vmcoreinfo_size;
>> 58 size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
51 59
52 /* Flag to indicate we are going to kexec a ne 60 /* Flag to indicate we are going to kexec a new kernel */
53 bool kexec_in_progress = false; 61 bool kexec_in_progress = false;
54 62
55 bool kexec_file_dbg_print; !! 63
>> 64 /* Location of the reserved area for the crash kernel */
>> 65 struct resource crashk_res = {
>> 66 .name = "Crash kernel",
>> 67 .start = 0,
>> 68 .end = 0,
>> 69 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
>> 70 .desc = IORES_DESC_CRASH_KERNEL
>> 71 };
>> 72 struct resource crashk_low_res = {
>> 73 .name = "Crash kernel",
>> 74 .start = 0,
>> 75 .end = 0,
>> 76 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
>> 77 .desc = IORES_DESC_CRASH_KERNEL
>> 78 };
>> 79
>> 80 int kexec_should_crash(struct task_struct *p)
>> 81 {
>> 82 /*
>> 83 * If crash_kexec_post_notifiers is enabled, don't run
>> 84 * crash_kexec() here yet, which must be run after panic
>> 85 * notifiers in panic().
>> 86 */
>> 87 if (crash_kexec_post_notifiers)
>> 88 return 0;
>> 89 /*
>> 90 * There are 4 panic() calls in do_exit() path, each of which
>> 91 * corresponds to each of these 4 conditions.
>> 92 */
>> 93 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
>> 94 return 1;
>> 95 return 0;
>> 96 }
>> 97
>> 98 int kexec_crash_loaded(void)
>> 99 {
>> 100 return !!kexec_crash_image;
>> 101 }
>> 102 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
56 103
57 /* 104 /*
58 * When kexec transitions to the new kernel th 105 * When kexec transitions to the new kernel there is a one-to-one
59 * mapping between physical and virtual addres 106 * mapping between physical and virtual addresses. On processors
60 * where you can disable the MMU this is trivi 107 * where you can disable the MMU this is trivial, and easy. For
61 * others it is still a simple predictable pag 108 * others it is still a simple predictable page table to setup.
62 * 109 *
63 * In that environment kexec copies the new ke 110 * In that environment kexec copies the new kernel to its final
64 * resting place. This means I can only suppo 111 * resting place. This means I can only support memory whose
65 * physical address can fit in an unsigned lon 112 * physical address can fit in an unsigned long. In particular
66 * addresses where (pfn << PAGE_SHIFT) > ULONG 113 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
67 * If the assembly stub has more restrictive r 114 * If the assembly stub has more restrictive requirements
68 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_ME 115 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
69 * defined more restrictively in <asm/kexec.h> 116 * defined more restrictively in <asm/kexec.h>.
70 * 117 *
71 * The code for the transition from the curren 118 * The code for the transition from the current kernel to the
72 * new kernel is placed in the control_code_bu !! 119 * the new kernel is placed in the control_code_buffer, whose size
73 * is given by KEXEC_CONTROL_PAGE_SIZE. In th 120 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
74 * page of memory is necessary, but some archi 121 * page of memory is necessary, but some architectures require more.
75 * Because this memory must be identity mapped 122 * Because this memory must be identity mapped in the transition from
76 * virtual to physical addresses it must live 123 * virtual to physical addresses it must live in the range
77 * 0 - TASK_SIZE, as only the user space mappi 124 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
78 * modifiable. 125 * modifiable.
79 * 126 *
80 * The assembly stub in the control code buffe 127 * The assembly stub in the control code buffer is passed a linked list
81 * of descriptor pages detailing the source pa 128 * of descriptor pages detailing the source pages of the new kernel,
82 * and the destination addresses of those sour 129 * and the destination addresses of those source pages. As this data
83 * structure is not used in the context of the 130 * structure is not used in the context of the current OS, it must
84 * be self-contained. 131 * be self-contained.
85 * 132 *
86 * The code has been made to work with highmem 133 * The code has been made to work with highmem pages and will use a
87 * destination page in its final resting place 134 * destination page in its final resting place (if it happens
88 * to allocate it). The end product of this i 135 * to allocate it). The end product of this is that most of the
89 * physical address space, and most of RAM can 136 * physical address space, and most of RAM can be used.
90 * 137 *
91 * Future directions include: 138 * Future directions include:
92 * - allocating a page table with the control 139 * - allocating a page table with the control code buffer identity
93 * mapped, to simplify machine_kexec and ma 140 * mapped, to simplify machine_kexec and make kexec_on_panic more
94 * reliable. 141 * reliable.
95 */ 142 */
96 143
97 /* 144 /*
98 * KIMAGE_NO_DEST is an impossible destination 145 * KIMAGE_NO_DEST is an impossible destination address..., for
99 * allocating pages whose destination address 146 * allocating pages whose destination address we do not care about.
100 */ 147 */
101 #define KIMAGE_NO_DEST (-1UL) 148 #define KIMAGE_NO_DEST (-1UL)
102 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) > 149 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
103 150
104 static struct page *kimage_alloc_page(struct k 151 static struct page *kimage_alloc_page(struct kimage *image,
105 gfp_t g 152 gfp_t gfp_mask,
106 unsigne 153 unsigned long dest);
107 154
108 int sanity_check_segment_list(struct kimage *i 155 int sanity_check_segment_list(struct kimage *image)
109 { 156 {
110 int i; 157 int i;
111 unsigned long nr_segments = image->nr_ 158 unsigned long nr_segments = image->nr_segments;
112 unsigned long total_pages = 0; 159 unsigned long total_pages = 0;
113 unsigned long nr_pages = totalram_page <<
114 160
115 /* 161 /*
116 * Verify we have good destination add 162 * Verify we have good destination addresses. The caller is
117 * responsible for making certain we d 163 * responsible for making certain we don't attempt to load
118 * the new image into invalid or reser 164 * the new image into invalid or reserved areas of RAM. This
119 * just verifies it is an address we c 165 * just verifies it is an address we can use.
120 * 166 *
121 * Since the kernel does everything in 167 * Since the kernel does everything in page size chunks ensure
122 * the destination addresses are page 168 * the destination addresses are page aligned. Too many
123 * special cases crop of when we don't 169 * special cases crop of when we don't do this. The most
124 * insidious is getting overlapping de 170 * insidious is getting overlapping destination addresses
125 * simply because addresses are change 171 * simply because addresses are changed to page size
126 * granularity. 172 * granularity.
127 */ 173 */
128 for (i = 0; i < nr_segments; i++) { 174 for (i = 0; i < nr_segments; i++) {
129 unsigned long mstart, mend; 175 unsigned long mstart, mend;
130 176
131 mstart = image->segment[i].mem 177 mstart = image->segment[i].mem;
132 mend = mstart + image->segme 178 mend = mstart + image->segment[i].memsz;
133 if (mstart > mend) 179 if (mstart > mend)
134 return -EADDRNOTAVAIL; 180 return -EADDRNOTAVAIL;
135 if ((mstart & ~PAGE_MASK) || ( 181 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
136 return -EADDRNOTAVAIL; 182 return -EADDRNOTAVAIL;
137 if (mend >= KEXEC_DESTINATION_ 183 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
138 return -EADDRNOTAVAIL; 184 return -EADDRNOTAVAIL;
139 } 185 }
140 186
141 /* Verify our destination addresses do 187 /* Verify our destination addresses do not overlap.
142 * If we alloed overlapping destinatio 188 * If we alloed overlapping destination addresses
143 * through very weird things can happe 189 * through very weird things can happen with no
144 * easy explanation as one segment sto 190 * easy explanation as one segment stops on another.
145 */ 191 */
146 for (i = 0; i < nr_segments; i++) { 192 for (i = 0; i < nr_segments; i++) {
147 unsigned long mstart, mend; 193 unsigned long mstart, mend;
148 unsigned long j; 194 unsigned long j;
149 195
150 mstart = image->segment[i].mem 196 mstart = image->segment[i].mem;
151 mend = mstart + image->segme 197 mend = mstart + image->segment[i].memsz;
152 for (j = 0; j < i; j++) { 198 for (j = 0; j < i; j++) {
153 unsigned long pstart, 199 unsigned long pstart, pend;
154 200
155 pstart = image->segmen 201 pstart = image->segment[j].mem;
156 pend = pstart + imag 202 pend = pstart + image->segment[j].memsz;
157 /* Do the segments ove 203 /* Do the segments overlap ? */
158 if ((mend > pstart) && 204 if ((mend > pstart) && (mstart < pend))
159 return -EINVAL 205 return -EINVAL;
160 } 206 }
161 } 207 }
162 208
163 /* Ensure our buffer sizes are strictl 209 /* Ensure our buffer sizes are strictly less than
164 * our memory sizes. This should alwa 210 * our memory sizes. This should always be the case,
165 * and it is easier to check up front 211 * and it is easier to check up front than to be surprised
166 * later on. 212 * later on.
167 */ 213 */
168 for (i = 0; i < nr_segments; i++) { 214 for (i = 0; i < nr_segments; i++) {
169 if (image->segment[i].bufsz > 215 if (image->segment[i].bufsz > image->segment[i].memsz)
170 return -EINVAL; 216 return -EINVAL;
171 } 217 }
172 218
173 /* 219 /*
174 * Verify that no more than half of me 220 * Verify that no more than half of memory will be consumed. If the
175 * request from userspace is too large 221 * request from userspace is too large, a large amount of time will be
176 * wasted allocating pages, which can 222 * wasted allocating pages, which can cause a soft lockup.
177 */ 223 */
178 for (i = 0; i < nr_segments; i++) { 224 for (i = 0; i < nr_segments; i++) {
179 if (PAGE_COUNT(image->segment[ !! 225 if (PAGE_COUNT(image->segment[i].memsz) > totalram_pages / 2)
180 return -EINVAL; 226 return -EINVAL;
181 227
182 total_pages += PAGE_COUNT(imag 228 total_pages += PAGE_COUNT(image->segment[i].memsz);
183 } 229 }
184 230
185 if (total_pages > nr_pages / 2) !! 231 if (total_pages > totalram_pages / 2)
186 return -EINVAL; 232 return -EINVAL;
187 233
188 #ifdef CONFIG_CRASH_DUMP <<
189 /* 234 /*
190 * Verify we have good destination add 235 * Verify we have good destination addresses. Normally
191 * the caller is responsible for makin 236 * the caller is responsible for making certain we don't
192 * attempt to load the new image into 237 * attempt to load the new image into invalid or reserved
193 * areas of RAM. But crash kernels ar 238 * areas of RAM. But crash kernels are preloaded into a
194 * reserved area of ram. We must ensu 239 * reserved area of ram. We must ensure the addresses
195 * are in the reserved area otherwise 240 * are in the reserved area otherwise preloading the
196 * kernel could corrupt things. 241 * kernel could corrupt things.
197 */ 242 */
198 243
199 if (image->type == KEXEC_TYPE_CRASH) { 244 if (image->type == KEXEC_TYPE_CRASH) {
200 for (i = 0; i < nr_segments; i 245 for (i = 0; i < nr_segments; i++) {
201 unsigned long mstart, 246 unsigned long mstart, mend;
202 247
203 mstart = image->segmen 248 mstart = image->segment[i].mem;
204 mend = mstart + image- 249 mend = mstart + image->segment[i].memsz - 1;
205 /* Ensure we are withi 250 /* Ensure we are within the crash kernel limits */
206 if ((mstart < phys_to_ 251 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
207 (mend > phys_to_bo 252 (mend > phys_to_boot_phys(crashk_res.end)))
208 return -EADDRN 253 return -EADDRNOTAVAIL;
209 } 254 }
210 } 255 }
211 #endif <<
212 256
213 return 0; 257 return 0;
214 } 258 }
215 259
216 struct kimage *do_kimage_alloc_init(void) 260 struct kimage *do_kimage_alloc_init(void)
217 { 261 {
218 struct kimage *image; 262 struct kimage *image;
219 263
220 /* Allocate a controlling structure */ 264 /* Allocate a controlling structure */
221 image = kzalloc(sizeof(*image), GFP_KE 265 image = kzalloc(sizeof(*image), GFP_KERNEL);
222 if (!image) 266 if (!image)
223 return NULL; 267 return NULL;
224 268
225 image->head = 0; 269 image->head = 0;
226 image->entry = &image->head; 270 image->entry = &image->head;
227 image->last_entry = &image->head; 271 image->last_entry = &image->head;
228 image->control_page = ~0; /* By defaul 272 image->control_page = ~0; /* By default this does not apply */
229 image->type = KEXEC_TYPE_DEFAULT; 273 image->type = KEXEC_TYPE_DEFAULT;
230 274
231 /* Initialize the list of control page 275 /* Initialize the list of control pages */
232 INIT_LIST_HEAD(&image->control_pages); 276 INIT_LIST_HEAD(&image->control_pages);
233 277
234 /* Initialize the list of destination 278 /* Initialize the list of destination pages */
235 INIT_LIST_HEAD(&image->dest_pages); 279 INIT_LIST_HEAD(&image->dest_pages);
236 280
237 /* Initialize the list of unusable pag 281 /* Initialize the list of unusable pages */
238 INIT_LIST_HEAD(&image->unusable_pages) 282 INIT_LIST_HEAD(&image->unusable_pages);
239 283
240 #ifdef CONFIG_CRASH_HOTPLUG <<
241 image->hp_action = KEXEC_CRASH_HP_NONE <<
242 image->elfcorehdr_index = -1; <<
243 image->elfcorehdr_updated = false; <<
244 #endif <<
245 <<
246 return image; 284 return image;
247 } 285 }
248 286
249 int kimage_is_destination_range(struct kimage 287 int kimage_is_destination_range(struct kimage *image,
250 unsign 288 unsigned long start,
251 unsign 289 unsigned long end)
252 { 290 {
253 unsigned long i; 291 unsigned long i;
254 292
255 for (i = 0; i < image->nr_segments; i+ 293 for (i = 0; i < image->nr_segments; i++) {
256 unsigned long mstart, mend; 294 unsigned long mstart, mend;
257 295
258 mstart = image->segment[i].mem 296 mstart = image->segment[i].mem;
259 mend = mstart + image->segment !! 297 mend = mstart + image->segment[i].memsz;
260 if ((end >= mstart) && (start !! 298 if ((end > mstart) && (start < mend))
261 return 1; 299 return 1;
262 } 300 }
263 301
264 return 0; 302 return 0;
265 } 303 }
266 304
267 static struct page *kimage_alloc_pages(gfp_t g 305 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
268 { 306 {
269 struct page *pages; 307 struct page *pages;
270 308
271 if (fatal_signal_pending(current)) !! 309 pages = alloc_pages(gfp_mask, order);
272 return NULL; <<
273 pages = alloc_pages(gfp_mask & ~__GFP_ <<
274 if (pages) { 310 if (pages) {
275 unsigned int count, i; 311 unsigned int count, i;
276 312
277 pages->mapping = NULL; 313 pages->mapping = NULL;
278 set_page_private(pages, order) 314 set_page_private(pages, order);
279 count = 1 << order; 315 count = 1 << order;
280 for (i = 0; i < count; i++) 316 for (i = 0; i < count; i++)
281 SetPageReserved(pages 317 SetPageReserved(pages + i);
282 <<
283 arch_kexec_post_alloc_pages(pa <<
284 gf <<
285 <<
286 if (gfp_mask & __GFP_ZERO) <<
287 for (i = 0; i < count; <<
288 clear_highpage <<
289 } 318 }
290 319
291 return pages; 320 return pages;
292 } 321 }
293 322
294 static void kimage_free_pages(struct page *pag 323 static void kimage_free_pages(struct page *page)
295 { 324 {
296 unsigned int order, count, i; 325 unsigned int order, count, i;
297 326
298 order = page_private(page); 327 order = page_private(page);
299 count = 1 << order; 328 count = 1 << order;
300 <<
301 arch_kexec_pre_free_pages(page_address <<
302 <<
303 for (i = 0; i < count; i++) 329 for (i = 0; i < count; i++)
304 ClearPageReserved(page + i); 330 ClearPageReserved(page + i);
305 __free_pages(page, order); 331 __free_pages(page, order);
306 } 332 }
307 333
308 void kimage_free_page_list(struct list_head *l 334 void kimage_free_page_list(struct list_head *list)
309 { 335 {
310 struct page *page, *next; 336 struct page *page, *next;
311 337
312 list_for_each_entry_safe(page, next, l 338 list_for_each_entry_safe(page, next, list, lru) {
313 list_del(&page->lru); 339 list_del(&page->lru);
314 kimage_free_pages(page); 340 kimage_free_pages(page);
315 } 341 }
316 } 342 }
317 343
318 static struct page *kimage_alloc_normal_contro 344 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
319 345 unsigned int order)
320 { 346 {
321 /* Control pages are special, they are 347 /* Control pages are special, they are the intermediaries
322 * that are needed while we copy the r 348 * that are needed while we copy the rest of the pages
323 * to their final resting place. As s 349 * to their final resting place. As such they must
324 * not conflict with either the destin 350 * not conflict with either the destination addresses
325 * or memory the kernel is already usi 351 * or memory the kernel is already using.
326 * 352 *
327 * The only case where we really need 353 * The only case where we really need more than one of
328 * these are for architectures where w 354 * these are for architectures where we cannot disable
329 * the MMU and must instead generate a 355 * the MMU and must instead generate an identity mapped
330 * page table for all of the memory. 356 * page table for all of the memory.
331 * 357 *
332 * At worst this runs in O(N) of the i 358 * At worst this runs in O(N) of the image size.
333 */ 359 */
334 struct list_head extra_pages; 360 struct list_head extra_pages;
335 struct page *pages; 361 struct page *pages;
336 unsigned int count; 362 unsigned int count;
337 363
338 count = 1 << order; 364 count = 1 << order;
339 INIT_LIST_HEAD(&extra_pages); 365 INIT_LIST_HEAD(&extra_pages);
340 366
341 /* Loop while I can allocate a page an 367 /* Loop while I can allocate a page and the page allocated
342 * is a destination page. 368 * is a destination page.
343 */ 369 */
344 do { 370 do {
345 unsigned long pfn, epfn, addr, 371 unsigned long pfn, epfn, addr, eaddr;
346 372
347 pages = kimage_alloc_pages(KEX 373 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
348 if (!pages) 374 if (!pages)
349 break; 375 break;
350 pfn = page_to_boot_pfn(pages 376 pfn = page_to_boot_pfn(pages);
351 epfn = pfn + count; 377 epfn = pfn + count;
352 addr = pfn << PAGE_SHIFT; 378 addr = pfn << PAGE_SHIFT;
353 eaddr = (epfn << PAGE_SHIFT) - !! 379 eaddr = epfn << PAGE_SHIFT;
354 if ((epfn >= (KEXEC_CONTROL_ME 380 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
355 kimage_is_destin 381 kimage_is_destination_range(image, addr, eaddr)) {
356 list_add(&pages->lru, 382 list_add(&pages->lru, &extra_pages);
357 pages = NULL; 383 pages = NULL;
358 } 384 }
359 } while (!pages); 385 } while (!pages);
360 386
361 if (pages) { 387 if (pages) {
362 /* Remember the allocated page 388 /* Remember the allocated page... */
363 list_add(&pages->lru, &image-> 389 list_add(&pages->lru, &image->control_pages);
364 390
365 /* Because the page is already 391 /* Because the page is already in it's destination
366 * location we will never allo 392 * location we will never allocate another page at
367 * that address. Therefore ki 393 * that address. Therefore kimage_alloc_pages
368 * will not return it (again) 394 * will not return it (again) and we don't need
369 * to give it an entry in imag 395 * to give it an entry in image->segment[].
370 */ 396 */
371 } 397 }
372 /* Deal with the destination pages I h 398 /* Deal with the destination pages I have inadvertently allocated.
373 * 399 *
374 * Ideally I would convert multi-page 400 * Ideally I would convert multi-page allocations into single
375 * page allocations, and add everythin 401 * page allocations, and add everything to image->dest_pages.
376 * 402 *
377 * For now it is simpler to just free 403 * For now it is simpler to just free the pages.
378 */ 404 */
379 kimage_free_page_list(&extra_pages); 405 kimage_free_page_list(&extra_pages);
380 406
381 return pages; 407 return pages;
382 } 408 }
383 409
384 #ifdef CONFIG_CRASH_DUMP <<
385 static struct page *kimage_alloc_crash_control 410 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
386 411 unsigned int order)
387 { 412 {
388 /* Control pages are special, they are 413 /* Control pages are special, they are the intermediaries
389 * that are needed while we copy the r 414 * that are needed while we copy the rest of the pages
390 * to their final resting place. As s 415 * to their final resting place. As such they must
391 * not conflict with either the destin 416 * not conflict with either the destination addresses
392 * or memory the kernel is already usi 417 * or memory the kernel is already using.
393 * 418 *
394 * Control pages are also the only pag 419 * Control pages are also the only pags we must allocate
395 * when loading a crash kernel. All o 420 * when loading a crash kernel. All of the other pages
396 * are specified by the segments and w 421 * are specified by the segments and we just memcpy
397 * into them directly. 422 * into them directly.
398 * 423 *
399 * The only case where we really need 424 * The only case where we really need more than one of
400 * these are for architectures where w 425 * these are for architectures where we cannot disable
401 * the MMU and must instead generate a 426 * the MMU and must instead generate an identity mapped
402 * page table for all of the memory. 427 * page table for all of the memory.
403 * 428 *
404 * Given the low demand this implement 429 * Given the low demand this implements a very simple
405 * allocator that finds the first hole 430 * allocator that finds the first hole of the appropriate
406 * size in the reserved memory region, 431 * size in the reserved memory region, and allocates all
407 * of the memory up to and including t 432 * of the memory up to and including the hole.
408 */ 433 */
409 unsigned long hole_start, hole_end, si 434 unsigned long hole_start, hole_end, size;
410 struct page *pages; 435 struct page *pages;
411 436
412 pages = NULL; 437 pages = NULL;
413 size = (1 << order) << PAGE_SHIFT; 438 size = (1 << order) << PAGE_SHIFT;
414 hole_start = ALIGN(image->control_page !! 439 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
415 hole_end = hole_start + size - 1; 440 hole_end = hole_start + size - 1;
416 while (hole_end <= crashk_res.end) { 441 while (hole_end <= crashk_res.end) {
417 unsigned long i; 442 unsigned long i;
418 443
419 cond_resched(); 444 cond_resched();
420 445
421 if (hole_end > KEXEC_CRASH_CON 446 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
422 break; 447 break;
423 /* See if I overlap any of the 448 /* See if I overlap any of the segments */
424 for (i = 0; i < image->nr_segm 449 for (i = 0; i < image->nr_segments; i++) {
425 unsigned long mstart, 450 unsigned long mstart, mend;
426 451
427 mstart = image->segmen 452 mstart = image->segment[i].mem;
428 mend = mstart + imag 453 mend = mstart + image->segment[i].memsz - 1;
429 if ((hole_end >= mstar 454 if ((hole_end >= mstart) && (hole_start <= mend)) {
430 /* Advance the 455 /* Advance the hole to the end of the segment */
431 hole_start = A !! 456 hole_start = (mend + (size - 1)) & ~(size - 1);
432 hole_end = h 457 hole_end = hole_start + size - 1;
433 break; 458 break;
434 } 459 }
435 } 460 }
436 /* If I don't overlap any segm 461 /* If I don't overlap any segments I have found my hole! */
437 if (i == image->nr_segments) { 462 if (i == image->nr_segments) {
438 pages = pfn_to_page(ho 463 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
439 image->control_page = !! 464 image->control_page = hole_end;
440 break; 465 break;
441 } 466 }
442 } 467 }
443 468
444 /* Ensure that these pages are decrypt <<
445 if (pages) <<
446 arch_kexec_post_alloc_pages(pa <<
447 <<
448 return pages; 469 return pages;
449 } 470 }
450 #endif <<
451 471
452 472
453 struct page *kimage_alloc_control_pages(struct 473 struct page *kimage_alloc_control_pages(struct kimage *image,
454 unsig 474 unsigned int order)
455 { 475 {
456 struct page *pages = NULL; 476 struct page *pages = NULL;
457 477
458 switch (image->type) { 478 switch (image->type) {
459 case KEXEC_TYPE_DEFAULT: 479 case KEXEC_TYPE_DEFAULT:
460 pages = kimage_alloc_normal_co 480 pages = kimage_alloc_normal_control_pages(image, order);
461 break; 481 break;
462 #ifdef CONFIG_CRASH_DUMP <<
463 case KEXEC_TYPE_CRASH: 482 case KEXEC_TYPE_CRASH:
464 pages = kimage_alloc_crash_con 483 pages = kimage_alloc_crash_control_pages(image, order);
465 break; 484 break;
466 #endif <<
467 } 485 }
468 486
469 return pages; 487 return pages;
470 } 488 }
471 489
472 static int kimage_add_entry(struct kimage *ima 490 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
473 { 491 {
474 if (*image->entry != 0) 492 if (*image->entry != 0)
475 image->entry++; 493 image->entry++;
476 494
477 if (image->entry == image->last_entry) 495 if (image->entry == image->last_entry) {
478 kimage_entry_t *ind_page; 496 kimage_entry_t *ind_page;
479 struct page *page; 497 struct page *page;
480 498
481 page = kimage_alloc_page(image 499 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
482 if (!page) 500 if (!page)
483 return -ENOMEM; 501 return -ENOMEM;
484 502
485 ind_page = page_address(page); 503 ind_page = page_address(page);
486 *image->entry = virt_to_boot_p 504 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
487 image->entry = ind_page; 505 image->entry = ind_page;
488 image->last_entry = ind_page + 506 image->last_entry = ind_page +
489 ((PAGE_S 507 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
490 } 508 }
491 *image->entry = entry; 509 *image->entry = entry;
492 image->entry++; 510 image->entry++;
493 *image->entry = 0; 511 *image->entry = 0;
494 512
495 return 0; 513 return 0;
496 } 514 }
497 515
498 static int kimage_set_destination(struct kimag 516 static int kimage_set_destination(struct kimage *image,
499 unsigned lo 517 unsigned long destination)
500 { 518 {
>> 519 int result;
>> 520
501 destination &= PAGE_MASK; 521 destination &= PAGE_MASK;
>> 522 result = kimage_add_entry(image, destination | IND_DESTINATION);
502 523
503 return kimage_add_entry(image, destina !! 524 return result;
504 } 525 }
505 526
506 527
507 static int kimage_add_page(struct kimage *imag 528 static int kimage_add_page(struct kimage *image, unsigned long page)
508 { 529 {
>> 530 int result;
>> 531
509 page &= PAGE_MASK; 532 page &= PAGE_MASK;
>> 533 result = kimage_add_entry(image, page | IND_SOURCE);
510 534
511 return kimage_add_entry(image, page | !! 535 return result;
512 } 536 }
513 537
514 538
515 static void kimage_free_extra_pages(struct kim 539 static void kimage_free_extra_pages(struct kimage *image)
516 { 540 {
517 /* Walk through and free any extra des 541 /* Walk through and free any extra destination pages I may have */
518 kimage_free_page_list(&image->dest_pag 542 kimage_free_page_list(&image->dest_pages);
519 543
520 /* Walk through and free any unusable 544 /* Walk through and free any unusable pages I have cached */
521 kimage_free_page_list(&image->unusable 545 kimage_free_page_list(&image->unusable_pages);
522 546
523 } 547 }
524 <<
525 void kimage_terminate(struct kimage *image) 548 void kimage_terminate(struct kimage *image)
526 { 549 {
527 if (*image->entry != 0) 550 if (*image->entry != 0)
528 image->entry++; 551 image->entry++;
529 552
530 *image->entry = IND_DONE; 553 *image->entry = IND_DONE;
531 } 554 }
532 555
533 #define for_each_kimage_entry(image, ptr, entr 556 #define for_each_kimage_entry(image, ptr, entry) \
534 for (ptr = &image->head; (entry = *ptr 557 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
535 ptr = (entry & IND_INDIRECTION 558 ptr = (entry & IND_INDIRECTION) ? \
536 boot_phys_to_virt((ent 559 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
537 560
538 static void kimage_free_entry(kimage_entry_t e 561 static void kimage_free_entry(kimage_entry_t entry)
539 { 562 {
540 struct page *page; 563 struct page *page;
541 564
542 page = boot_pfn_to_page(entry >> PAGE_ 565 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
543 kimage_free_pages(page); 566 kimage_free_pages(page);
544 } 567 }
545 568
546 void kimage_free(struct kimage *image) 569 void kimage_free(struct kimage *image)
547 { 570 {
548 kimage_entry_t *ptr, entry; 571 kimage_entry_t *ptr, entry;
549 kimage_entry_t ind = 0; 572 kimage_entry_t ind = 0;
550 573
551 if (!image) 574 if (!image)
552 return; 575 return;
553 576
554 #ifdef CONFIG_CRASH_DUMP <<
555 if (image->vmcoreinfo_data_copy) { <<
556 crash_update_vmcoreinfo_safeco <<
557 vunmap(image->vmcoreinfo_data_ <<
558 } <<
559 #endif <<
560 <<
561 kimage_free_extra_pages(image); 577 kimage_free_extra_pages(image);
562 for_each_kimage_entry(image, ptr, entr 578 for_each_kimage_entry(image, ptr, entry) {
563 if (entry & IND_INDIRECTION) { 579 if (entry & IND_INDIRECTION) {
564 /* Free the previous i 580 /* Free the previous indirection page */
565 if (ind & IND_INDIRECT 581 if (ind & IND_INDIRECTION)
566 kimage_free_en 582 kimage_free_entry(ind);
567 /* Save this indirecti 583 /* Save this indirection page until we are
568 * done with it. 584 * done with it.
569 */ 585 */
570 ind = entry; 586 ind = entry;
571 } else if (entry & IND_SOURCE) 587 } else if (entry & IND_SOURCE)
572 kimage_free_entry(entr 588 kimage_free_entry(entry);
573 } 589 }
574 /* Free the final indirection page */ 590 /* Free the final indirection page */
575 if (ind & IND_INDIRECTION) 591 if (ind & IND_INDIRECTION)
576 kimage_free_entry(ind); 592 kimage_free_entry(ind);
577 593
578 /* Handle any machine specific cleanup 594 /* Handle any machine specific cleanup */
579 machine_kexec_cleanup(image); 595 machine_kexec_cleanup(image);
580 596
581 /* Free the kexec control pages... */ 597 /* Free the kexec control pages... */
582 kimage_free_page_list(&image->control_ 598 kimage_free_page_list(&image->control_pages);
583 599
584 /* 600 /*
585 * Free up any temporary buffers alloc 601 * Free up any temporary buffers allocated. This might hit if
586 * error occurred much later after buf 602 * error occurred much later after buffer allocation.
587 */ 603 */
588 if (image->file_mode) 604 if (image->file_mode)
589 kimage_file_post_load_cleanup( 605 kimage_file_post_load_cleanup(image);
590 606
591 kfree(image); 607 kfree(image);
592 } 608 }
593 609
594 static kimage_entry_t *kimage_dst_used(struct 610 static kimage_entry_t *kimage_dst_used(struct kimage *image,
595 unsign 611 unsigned long page)
596 { 612 {
597 kimage_entry_t *ptr, entry; 613 kimage_entry_t *ptr, entry;
598 unsigned long destination = 0; 614 unsigned long destination = 0;
599 615
600 for_each_kimage_entry(image, ptr, entr 616 for_each_kimage_entry(image, ptr, entry) {
601 if (entry & IND_DESTINATION) 617 if (entry & IND_DESTINATION)
602 destination = entry & 618 destination = entry & PAGE_MASK;
603 else if (entry & IND_SOURCE) { 619 else if (entry & IND_SOURCE) {
604 if (page == destinatio 620 if (page == destination)
605 return ptr; 621 return ptr;
606 destination += PAGE_SI 622 destination += PAGE_SIZE;
607 } 623 }
608 } 624 }
609 625
610 return NULL; 626 return NULL;
611 } 627 }
612 628
613 static struct page *kimage_alloc_page(struct k 629 static struct page *kimage_alloc_page(struct kimage *image,
614 gfp_t 630 gfp_t gfp_mask,
615 unsign 631 unsigned long destination)
616 { 632 {
617 /* 633 /*
618 * Here we implement safeguards to ens 634 * Here we implement safeguards to ensure that a source page
619 * is not copied to its destination pa 635 * is not copied to its destination page before the data on
620 * the destination page is no longer u 636 * the destination page is no longer useful.
621 * 637 *
622 * To do this we maintain the invarian 638 * To do this we maintain the invariant that a source page is
623 * either its own destination page, or 639 * either its own destination page, or it is not a
624 * destination page at all. 640 * destination page at all.
625 * 641 *
626 * That is slightly stronger than requ 642 * That is slightly stronger than required, but the proof
627 * that no problems will not occur is 643 * that no problems will not occur is trivial, and the
628 * implementation is simply to verify. 644 * implementation is simply to verify.
629 * 645 *
630 * When allocating all pages normally 646 * When allocating all pages normally this algorithm will run
631 * in O(N) time, but in the worst case 647 * in O(N) time, but in the worst case it will run in O(N^2)
632 * time. If the runtime is a problem 648 * time. If the runtime is a problem the data structures can
633 * be fixed. 649 * be fixed.
634 */ 650 */
635 struct page *page; 651 struct page *page;
636 unsigned long addr; 652 unsigned long addr;
637 653
638 /* 654 /*
639 * Walk through the list of destinatio 655 * Walk through the list of destination pages, and see if I
640 * have a match. 656 * have a match.
641 */ 657 */
642 list_for_each_entry(page, &image->dest 658 list_for_each_entry(page, &image->dest_pages, lru) {
643 addr = page_to_boot_pfn(page) 659 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
644 if (addr == destination) { 660 if (addr == destination) {
645 list_del(&page->lru); 661 list_del(&page->lru);
646 return page; 662 return page;
647 } 663 }
648 } 664 }
649 page = NULL; 665 page = NULL;
650 while (1) { 666 while (1) {
651 kimage_entry_t *old; 667 kimage_entry_t *old;
652 668
653 /* Allocate a page, if we run 669 /* Allocate a page, if we run out of memory give up */
654 page = kimage_alloc_pages(gfp_ 670 page = kimage_alloc_pages(gfp_mask, 0);
655 if (!page) 671 if (!page)
656 return NULL; 672 return NULL;
657 /* If the page cannot be used 673 /* If the page cannot be used file it away */
658 if (page_to_boot_pfn(page) > 674 if (page_to_boot_pfn(page) >
659 (KEXEC_SOURCE_ 675 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
660 list_add(&page->lru, & 676 list_add(&page->lru, &image->unusable_pages);
661 continue; 677 continue;
662 } 678 }
663 addr = page_to_boot_pfn(page) 679 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
664 680
665 /* If it is the destination pa 681 /* If it is the destination page we want use it */
666 if (addr == destination) 682 if (addr == destination)
667 break; 683 break;
668 684
669 /* If the page is not a destin 685 /* If the page is not a destination page use it */
670 if (!kimage_is_destination_ran 686 if (!kimage_is_destination_range(image, addr,
671 !! 687 addr + PAGE_SIZE))
672 break; 688 break;
673 689
674 /* 690 /*
675 * I know that the page is som 691 * I know that the page is someones destination page.
676 * See if there is already a s 692 * See if there is already a source page for this
677 * destination page. And if s 693 * destination page. And if so swap the source pages.
678 */ 694 */
679 old = kimage_dst_used(image, a 695 old = kimage_dst_used(image, addr);
680 if (old) { 696 if (old) {
681 /* If so move it */ 697 /* If so move it */
682 unsigned long old_addr 698 unsigned long old_addr;
683 struct page *old_page; 699 struct page *old_page;
684 700
685 old_addr = *old & PAGE 701 old_addr = *old & PAGE_MASK;
686 old_page = boot_pfn_to 702 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
687 copy_highpage(page, ol 703 copy_highpage(page, old_page);
688 *old = addr | (*old & 704 *old = addr | (*old & ~PAGE_MASK);
689 705
690 /* The old page I have 706 /* The old page I have found cannot be a
691 * destination page, s 707 * destination page, so return it if it's
692 * gfp_flags honor the 708 * gfp_flags honor the ones passed in.
693 */ 709 */
694 if (!(gfp_mask & __GFP 710 if (!(gfp_mask & __GFP_HIGHMEM) &&
695 PageHighMem(old_pa 711 PageHighMem(old_page)) {
696 kimage_free_pa 712 kimage_free_pages(old_page);
697 continue; 713 continue;
698 } 714 }
>> 715 addr = old_addr;
699 page = old_page; 716 page = old_page;
700 break; 717 break;
701 } 718 }
702 /* Place the page on the desti 719 /* Place the page on the destination list, to be used later */
703 list_add(&page->lru, &image->d 720 list_add(&page->lru, &image->dest_pages);
704 } 721 }
705 722
706 return page; 723 return page;
707 } 724 }
708 725
709 static int kimage_load_normal_segment(struct k 726 static int kimage_load_normal_segment(struct kimage *image,
710 struc 727 struct kexec_segment *segment)
711 { 728 {
712 unsigned long maddr; 729 unsigned long maddr;
713 size_t ubytes, mbytes; 730 size_t ubytes, mbytes;
714 int result; 731 int result;
715 unsigned char __user *buf = NULL; 732 unsigned char __user *buf = NULL;
716 unsigned char *kbuf = NULL; 733 unsigned char *kbuf = NULL;
717 734
>> 735 result = 0;
718 if (image->file_mode) 736 if (image->file_mode)
719 kbuf = segment->kbuf; 737 kbuf = segment->kbuf;
720 else 738 else
721 buf = segment->buf; 739 buf = segment->buf;
722 ubytes = segment->bufsz; 740 ubytes = segment->bufsz;
723 mbytes = segment->memsz; 741 mbytes = segment->memsz;
724 maddr = segment->mem; 742 maddr = segment->mem;
725 743
726 result = kimage_set_destination(image, 744 result = kimage_set_destination(image, maddr);
727 if (result < 0) 745 if (result < 0)
728 goto out; 746 goto out;
729 747
730 while (mbytes) { 748 while (mbytes) {
731 struct page *page; 749 struct page *page;
732 char *ptr; 750 char *ptr;
733 size_t uchunk, mchunk; 751 size_t uchunk, mchunk;
734 752
735 page = kimage_alloc_page(image 753 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
736 if (!page) { 754 if (!page) {
737 result = -ENOMEM; 755 result = -ENOMEM;
738 goto out; 756 goto out;
739 } 757 }
740 result = kimage_add_page(image 758 result = kimage_add_page(image, page_to_boot_pfn(page)
741 759 << PAGE_SHIFT);
742 if (result < 0) 760 if (result < 0)
743 goto out; 761 goto out;
744 762
745 ptr = kmap_local_page(page); !! 763 ptr = kmap(page);
746 /* Start with a clear page */ 764 /* Start with a clear page */
747 clear_page(ptr); 765 clear_page(ptr);
748 ptr += maddr & ~PAGE_MASK; 766 ptr += maddr & ~PAGE_MASK;
749 mchunk = min_t(size_t, mbytes, 767 mchunk = min_t(size_t, mbytes,
750 PAGE_SIZE - (m 768 PAGE_SIZE - (maddr & ~PAGE_MASK));
751 uchunk = min(ubytes, mchunk); 769 uchunk = min(ubytes, mchunk);
752 770
753 if (uchunk) { !! 771 /* For file based kexec, source pages are in kernel memory */
754 /* For file based kexe !! 772 if (image->file_mode)
755 if (image->file_mode) !! 773 memcpy(ptr, kbuf, uchunk);
756 memcpy(ptr, kb !! 774 else
757 else !! 775 result = copy_from_user(ptr, buf, uchunk);
758 result = copy_ !! 776 kunmap(page);
759 ubytes -= uchunk; <<
760 if (image->file_mode) <<
761 kbuf += uchunk <<
762 else <<
763 buf += uchunk; <<
764 } <<
765 kunmap_local(ptr); <<
766 if (result) { 777 if (result) {
767 result = -EFAULT; 778 result = -EFAULT;
768 goto out; 779 goto out;
769 } 780 }
>> 781 ubytes -= uchunk;
770 maddr += mchunk; 782 maddr += mchunk;
>> 783 if (image->file_mode)
>> 784 kbuf += mchunk;
>> 785 else
>> 786 buf += mchunk;
771 mbytes -= mchunk; 787 mbytes -= mchunk;
772 <<
773 cond_resched(); <<
774 } 788 }
775 out: 789 out:
776 return result; 790 return result;
777 } 791 }
778 792
779 #ifdef CONFIG_CRASH_DUMP <<
780 static int kimage_load_crash_segment(struct ki 793 static int kimage_load_crash_segment(struct kimage *image,
781 struct 794 struct kexec_segment *segment)
782 { 795 {
783 /* For crash dumps kernels we simply c 796 /* For crash dumps kernels we simply copy the data from
784 * user space to it's destination. 797 * user space to it's destination.
785 * We do things a page at a time for t 798 * We do things a page at a time for the sake of kmap.
786 */ 799 */
787 unsigned long maddr; 800 unsigned long maddr;
788 size_t ubytes, mbytes; 801 size_t ubytes, mbytes;
789 int result; 802 int result;
790 unsigned char __user *buf = NULL; 803 unsigned char __user *buf = NULL;
791 unsigned char *kbuf = NULL; 804 unsigned char *kbuf = NULL;
792 805
793 result = 0; 806 result = 0;
794 if (image->file_mode) 807 if (image->file_mode)
795 kbuf = segment->kbuf; 808 kbuf = segment->kbuf;
796 else 809 else
797 buf = segment->buf; 810 buf = segment->buf;
798 ubytes = segment->bufsz; 811 ubytes = segment->bufsz;
799 mbytes = segment->memsz; 812 mbytes = segment->memsz;
800 maddr = segment->mem; 813 maddr = segment->mem;
801 while (mbytes) { 814 while (mbytes) {
802 struct page *page; 815 struct page *page;
803 char *ptr; 816 char *ptr;
804 size_t uchunk, mchunk; 817 size_t uchunk, mchunk;
805 818
806 page = boot_pfn_to_page(maddr 819 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
807 if (!page) { 820 if (!page) {
808 result = -ENOMEM; 821 result = -ENOMEM;
809 goto out; 822 goto out;
810 } 823 }
811 arch_kexec_post_alloc_pages(pa !! 824 ptr = kmap(page);
812 ptr = kmap_local_page(page); <<
813 ptr += maddr & ~PAGE_MASK; 825 ptr += maddr & ~PAGE_MASK;
814 mchunk = min_t(size_t, mbytes, 826 mchunk = min_t(size_t, mbytes,
815 PAGE_SIZE - (m 827 PAGE_SIZE - (maddr & ~PAGE_MASK));
816 uchunk = min(ubytes, mchunk); 828 uchunk = min(ubytes, mchunk);
817 if (mchunk > uchunk) { 829 if (mchunk > uchunk) {
818 /* Zero the trailing p 830 /* Zero the trailing part of the page */
819 memset(ptr + uchunk, 0 831 memset(ptr + uchunk, 0, mchunk - uchunk);
820 } 832 }
821 833
822 if (uchunk) { !! 834 /* For file based kexec, source pages are in kernel memory */
823 /* For file based kexe !! 835 if (image->file_mode)
824 if (image->file_mode) !! 836 memcpy(ptr, kbuf, uchunk);
825 memcpy(ptr, kb !! 837 else
826 else !! 838 result = copy_from_user(ptr, buf, uchunk);
827 result = copy_ <<
828 ubytes -= uchunk; <<
829 if (image->file_mode) <<
830 kbuf += uchunk <<
831 else <<
832 buf += uchunk; <<
833 } <<
834 kexec_flush_icache_page(page); 839 kexec_flush_icache_page(page);
835 kunmap_local(ptr); !! 840 kunmap(page);
836 arch_kexec_pre_free_pages(page <<
837 if (result) { 841 if (result) {
838 result = -EFAULT; 842 result = -EFAULT;
839 goto out; 843 goto out;
840 } 844 }
>> 845 ubytes -= uchunk;
841 maddr += mchunk; 846 maddr += mchunk;
>> 847 if (image->file_mode)
>> 848 kbuf += mchunk;
>> 849 else
>> 850 buf += mchunk;
842 mbytes -= mchunk; 851 mbytes -= mchunk;
843 <<
844 cond_resched(); <<
845 } 852 }
846 out: 853 out:
847 return result; 854 return result;
848 } 855 }
849 #endif <<
850 856
851 int kimage_load_segment(struct kimage *image, 857 int kimage_load_segment(struct kimage *image,
852 struct kexec_s 858 struct kexec_segment *segment)
853 { 859 {
854 int result = -ENOMEM; 860 int result = -ENOMEM;
855 861
856 switch (image->type) { 862 switch (image->type) {
857 case KEXEC_TYPE_DEFAULT: 863 case KEXEC_TYPE_DEFAULT:
858 result = kimage_load_normal_se 864 result = kimage_load_normal_segment(image, segment);
859 break; 865 break;
860 #ifdef CONFIG_CRASH_DUMP <<
861 case KEXEC_TYPE_CRASH: 866 case KEXEC_TYPE_CRASH:
862 result = kimage_load_crash_seg 867 result = kimage_load_crash_segment(image, segment);
863 break; 868 break;
864 #endif <<
865 } 869 }
866 870
867 return result; 871 return result;
868 } 872 }
869 873
870 struct kexec_load_limit { !! 874 struct kimage *kexec_image;
871 /* Mutex protects the limit count. */ !! 875 struct kimage *kexec_crash_image;
872 struct mutex mutex; !! 876 int kexec_load_disabled;
873 int limit; <<
874 }; <<
875 877
876 static struct kexec_load_limit load_limit_rebo !! 878 /*
877 .mutex = __MUTEX_INITIALIZER(load_limi !! 879 * No panic_cpu check version of crash_kexec(). This function is called
878 .limit = -1, !! 880 * only when panic_cpu holds the current CPU number; this is the only CPU
879 }; !! 881 * which processes crash_kexec routines.
>> 882 */
>> 883 void __crash_kexec(struct pt_regs *regs)
>> 884 {
>> 885 /* Take the kexec_mutex here to prevent sys_kexec_load
>> 886 * running on one cpu from replacing the crash kernel
>> 887 * we are using after a panic on a different cpu.
>> 888 *
>> 889 * If the crash kernel was not located in a fixed area
>> 890 * of memory the xchg(&kexec_crash_image) would be
>> 891 * sufficient. But since I reuse the memory...
>> 892 */
>> 893 if (mutex_trylock(&kexec_mutex)) {
>> 894 if (kexec_crash_image) {
>> 895 struct pt_regs fixed_regs;
>> 896
>> 897 crash_setup_regs(&fixed_regs, regs);
>> 898 crash_save_vmcoreinfo();
>> 899 machine_crash_shutdown(&fixed_regs);
>> 900 machine_kexec(kexec_crash_image);
>> 901 }
>> 902 mutex_unlock(&kexec_mutex);
>> 903 }
>> 904 }
880 905
881 static struct kexec_load_limit load_limit_pani !! 906 void crash_kexec(struct pt_regs *regs)
882 .mutex = __MUTEX_INITIALIZER(load_limi !! 907 {
883 .limit = -1, !! 908 int old_cpu, this_cpu;
884 }; !! 909
>> 910 /*
>> 911 * Only one CPU is allowed to execute the crash_kexec() code as with
>> 912 * panic(). Otherwise parallel calls of panic() and crash_kexec()
>> 913 * may stop each other. To exclude them, we use panic_cpu here too.
>> 914 */
>> 915 this_cpu = raw_smp_processor_id();
>> 916 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
>> 917 if (old_cpu == PANIC_CPU_INVALID) {
>> 918 /* This is the 1st CPU which comes here, so go ahead. */
>> 919 printk_safe_flush_on_panic();
>> 920 __crash_kexec(regs);
>> 921
>> 922 /*
>> 923 * Reset panic_cpu to allow another panic()/crash_kexec()
>> 924 * call.
>> 925 */
>> 926 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
>> 927 }
>> 928 }
>> 929
>> 930 size_t crash_get_memory_size(void)
>> 931 {
>> 932 size_t size = 0;
>> 933
>> 934 mutex_lock(&kexec_mutex);
>> 935 if (crashk_res.end != crashk_res.start)
>> 936 size = resource_size(&crashk_res);
>> 937 mutex_unlock(&kexec_mutex);
>> 938 return size;
>> 939 }
>> 940
>> 941 void __weak crash_free_reserved_phys_range(unsigned long begin,
>> 942 unsigned long end)
>> 943 {
>> 944 unsigned long addr;
>> 945
>> 946 for (addr = begin; addr < end; addr += PAGE_SIZE)
>> 947 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
>> 948 }
>> 949
>> 950 int crash_shrink_memory(unsigned long new_size)
>> 951 {
>> 952 int ret = 0;
>> 953 unsigned long start, end;
>> 954 unsigned long old_size;
>> 955 struct resource *ram_res;
>> 956
>> 957 mutex_lock(&kexec_mutex);
>> 958
>> 959 if (kexec_crash_image) {
>> 960 ret = -ENOENT;
>> 961 goto unlock;
>> 962 }
>> 963 start = crashk_res.start;
>> 964 end = crashk_res.end;
>> 965 old_size = (end == 0) ? 0 : end - start + 1;
>> 966 if (new_size >= old_size) {
>> 967 ret = (new_size == old_size) ? 0 : -EINVAL;
>> 968 goto unlock;
>> 969 }
>> 970
>> 971 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
>> 972 if (!ram_res) {
>> 973 ret = -ENOMEM;
>> 974 goto unlock;
>> 975 }
>> 976
>> 977 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
>> 978 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
>> 979
>> 980 crash_free_reserved_phys_range(end, crashk_res.end);
>> 981
>> 982 if ((start == end) && (crashk_res.parent != NULL))
>> 983 release_resource(&crashk_res);
>> 984
>> 985 ram_res->start = end;
>> 986 ram_res->end = crashk_res.end;
>> 987 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
>> 988 ram_res->name = "System RAM";
>> 989
>> 990 crashk_res.end = end - 1;
>> 991
>> 992 insert_resource(&iomem_resource, ram_res);
>> 993
>> 994 unlock:
>> 995 mutex_unlock(&kexec_mutex);
>> 996 return ret;
>> 997 }
>> 998
>> 999 static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
>> 1000 size_t data_len)
>> 1001 {
>> 1002 struct elf_note note;
>> 1003
>> 1004 note.n_namesz = strlen(name) + 1;
>> 1005 note.n_descsz = data_len;
>> 1006 note.n_type = type;
>> 1007 memcpy(buf, ¬e, sizeof(note));
>> 1008 buf += (sizeof(note) + 3)/4;
>> 1009 memcpy(buf, name, note.n_namesz);
>> 1010 buf += (note.n_namesz + 3)/4;
>> 1011 memcpy(buf, data, note.n_descsz);
>> 1012 buf += (note.n_descsz + 3)/4;
>> 1013
>> 1014 return buf;
>> 1015 }
>> 1016
>> 1017 static void final_note(u32 *buf)
>> 1018 {
>> 1019 struct elf_note note;
>> 1020
>> 1021 note.n_namesz = 0;
>> 1022 note.n_descsz = 0;
>> 1023 note.n_type = 0;
>> 1024 memcpy(buf, ¬e, sizeof(note));
>> 1025 }
>> 1026
>> 1027 void crash_save_cpu(struct pt_regs *regs, int cpu)
>> 1028 {
>> 1029 struct elf_prstatus prstatus;
>> 1030 u32 *buf;
>> 1031
>> 1032 if ((cpu < 0) || (cpu >= nr_cpu_ids))
>> 1033 return;
>> 1034
>> 1035 /* Using ELF notes here is opportunistic.
>> 1036 * I need a well defined structure format
>> 1037 * for the data I pass, and I need tags
>> 1038 * on the data to indicate what information I have
>> 1039 * squirrelled away. ELF notes happen to provide
>> 1040 * all of that, so there is no need to invent something new.
>> 1041 */
>> 1042 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
>> 1043 if (!buf)
>> 1044 return;
>> 1045 memset(&prstatus, 0, sizeof(prstatus));
>> 1046 prstatus.pr_pid = current->pid;
>> 1047 elf_core_copy_kernel_regs(&prstatus.pr_reg, regs);
>> 1048 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
>> 1049 &prstatus, sizeof(prstatus));
>> 1050 final_note(buf);
>> 1051 }
>> 1052
>> 1053 static int __init crash_notes_memory_init(void)
>> 1054 {
>> 1055 /* Allocate memory for saving cpu registers. */
>> 1056 size_t size, align;
>> 1057
>> 1058 /*
>> 1059 * crash_notes could be allocated across 2 vmalloc pages when percpu
>> 1060 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
>> 1061 * pages are also on 2 continuous physical pages. In this case the
>> 1062 * 2nd part of crash_notes in 2nd page could be lost since only the
>> 1063 * starting address and size of crash_notes are exported through sysfs.
>> 1064 * Here round up the size of crash_notes to the nearest power of two
>> 1065 * and pass it to __alloc_percpu as align value. This can make sure
>> 1066 * crash_notes is allocated inside one physical page.
>> 1067 */
>> 1068 size = sizeof(note_buf_t);
>> 1069 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
>> 1070
>> 1071 /*
>> 1072 * Break compile if size is bigger than PAGE_SIZE since crash_notes
>> 1073 * definitely will be in 2 pages with that.
>> 1074 */
>> 1075 BUILD_BUG_ON(size > PAGE_SIZE);
>> 1076
>> 1077 crash_notes = __alloc_percpu(size, align);
>> 1078 if (!crash_notes) {
>> 1079 pr_warn("Memory allocation for saving cpu register states failed\n");
>> 1080 return -ENOMEM;
>> 1081 }
>> 1082 return 0;
>> 1083 }
>> 1084 subsys_initcall(crash_notes_memory_init);
885 1085
886 struct kimage *kexec_image; <<
887 struct kimage *kexec_crash_image; <<
888 static int kexec_load_disabled; <<
889 1086
890 #ifdef CONFIG_SYSCTL !! 1087 /*
891 static int kexec_limit_handler(const struct ct !! 1088 * parsing the "crashkernel" commandline
892 void *buffer, s !! 1089 *
893 { !! 1090 * this code is intended to be called from architecture specific code
894 struct kexec_load_limit *limit = table !! 1091 */
895 int val; <<
896 struct ctl_table tmp = { <<
897 .data = &val, <<
898 .maxlen = sizeof(val), <<
899 .mode = table->mode, <<
900 }; <<
901 int ret; <<
902 <<
903 if (write) { <<
904 ret = proc_dointvec(&tmp, writ <<
905 if (ret) <<
906 return ret; <<
907 1092
908 if (val < 0) !! 1093
>> 1094 /*
>> 1095 * This function parses command lines in the format
>> 1096 *
>> 1097 * crashkernel=ramsize-range:size[,...][@offset]
>> 1098 *
>> 1099 * The function returns 0 on success and -EINVAL on failure.
>> 1100 */
>> 1101 static int __init parse_crashkernel_mem(char *cmdline,
>> 1102 unsigned long long system_ram,
>> 1103 unsigned long long *crash_size,
>> 1104 unsigned long long *crash_base)
>> 1105 {
>> 1106 char *cur = cmdline, *tmp;
>> 1107
>> 1108 /* for each entry of the comma-separated list */
>> 1109 do {
>> 1110 unsigned long long start, end = ULLONG_MAX, size;
>> 1111
>> 1112 /* get the start of the range */
>> 1113 start = memparse(cur, &tmp);
>> 1114 if (cur == tmp) {
>> 1115 pr_warn("crashkernel: Memory value expected\n");
>> 1116 return -EINVAL;
>> 1117 }
>> 1118 cur = tmp;
>> 1119 if (*cur != '-') {
>> 1120 pr_warn("crashkernel: '-' expected\n");
909 return -EINVAL; 1121 return -EINVAL;
>> 1122 }
>> 1123 cur++;
910 1124
911 mutex_lock(&limit->mutex); !! 1125 /* if no ':' is here, than we read the end */
912 if (limit->limit != -1 && val !! 1126 if (*cur != ':') {
913 ret = -EINVAL; !! 1127 end = memparse(cur, &tmp);
914 else !! 1128 if (cur == tmp) {
915 limit->limit = val; !! 1129 pr_warn("crashkernel: Memory value expected\n");
916 mutex_unlock(&limit->mutex); !! 1130 return -EINVAL;
>> 1131 }
>> 1132 cur = tmp;
>> 1133 if (end <= start) {
>> 1134 pr_warn("crashkernel: end <= start\n");
>> 1135 return -EINVAL;
>> 1136 }
>> 1137 }
>> 1138
>> 1139 if (*cur != ':') {
>> 1140 pr_warn("crashkernel: ':' expected\n");
>> 1141 return -EINVAL;
>> 1142 }
>> 1143 cur++;
>> 1144
>> 1145 size = memparse(cur, &tmp);
>> 1146 if (cur == tmp) {
>> 1147 pr_warn("Memory value expected\n");
>> 1148 return -EINVAL;
>> 1149 }
>> 1150 cur = tmp;
>> 1151 if (size >= system_ram) {
>> 1152 pr_warn("crashkernel: invalid size\n");
>> 1153 return -EINVAL;
>> 1154 }
917 1155
918 return ret; !! 1156 /* match ? */
>> 1157 if (system_ram >= start && system_ram < end) {
>> 1158 *crash_size = size;
>> 1159 break;
>> 1160 }
>> 1161 } while (*cur++ == ',');
>> 1162
>> 1163 if (*crash_size > 0) {
>> 1164 while (*cur && *cur != ' ' && *cur != '@')
>> 1165 cur++;
>> 1166 if (*cur == '@') {
>> 1167 cur++;
>> 1168 *crash_base = memparse(cur, &tmp);
>> 1169 if (cur == tmp) {
>> 1170 pr_warn("Memory value expected after '@'\n");
>> 1171 return -EINVAL;
>> 1172 }
>> 1173 }
>> 1174 }
>> 1175
>> 1176 return 0;
>> 1177 }
>> 1178
>> 1179 /*
>> 1180 * That function parses "simple" (old) crashkernel command lines like
>> 1181 *
>> 1182 * crashkernel=size[@offset]
>> 1183 *
>> 1184 * It returns 0 on success and -EINVAL on failure.
>> 1185 */
>> 1186 static int __init parse_crashkernel_simple(char *cmdline,
>> 1187 unsigned long long *crash_size,
>> 1188 unsigned long long *crash_base)
>> 1189 {
>> 1190 char *cur = cmdline;
>> 1191
>> 1192 *crash_size = memparse(cmdline, &cur);
>> 1193 if (cmdline == cur) {
>> 1194 pr_warn("crashkernel: memory value expected\n");
>> 1195 return -EINVAL;
919 } 1196 }
920 1197
921 mutex_lock(&limit->mutex); !! 1198 if (*cur == '@')
922 val = limit->limit; !! 1199 *crash_base = memparse(cur+1, &cur);
923 mutex_unlock(&limit->mutex); !! 1200 else if (*cur != ' ' && *cur != '\0') {
>> 1201 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
>> 1202 return -EINVAL;
>> 1203 }
924 1204
925 return proc_dointvec(&tmp, write, buff !! 1205 return 0;
926 } 1206 }
927 1207
928 static struct ctl_table kexec_core_sysctls[] = !! 1208 #define SUFFIX_HIGH 0
929 { !! 1209 #define SUFFIX_LOW 1
930 .procname = "kexec_load_ !! 1210 #define SUFFIX_NULL 2
931 .data = &kexec_load_ !! 1211 static __initdata char *suffix_tbl[] = {
932 .maxlen = sizeof(int), !! 1212 [SUFFIX_HIGH] = ",high",
933 .mode = 0644, !! 1213 [SUFFIX_LOW] = ",low",
934 /* only handle a transition fr !! 1214 [SUFFIX_NULL] = NULL,
935 .proc_handler = proc_dointve <<
936 .extra1 = SYSCTL_ONE, <<
937 .extra2 = SYSCTL_ONE, <<
938 }, <<
939 { <<
940 .procname = "kexec_load_ <<
941 .data = &load_limit_ <<
942 .mode = 0644, <<
943 .proc_handler = kexec_limit_ <<
944 }, <<
945 { <<
946 .procname = "kexec_load_ <<
947 .data = &load_limit_ <<
948 .mode = 0644, <<
949 .proc_handler = kexec_limit_ <<
950 }, <<
951 }; 1215 };
952 1216
953 static int __init kexec_core_sysctl_init(void) !! 1217 /*
954 { !! 1218 * That function parses "suffix" crashkernel command lines like
955 register_sysctl_init("kernel", kexec_c !! 1219 *
>> 1220 * crashkernel=size,[high|low]
>> 1221 *
>> 1222 * It returns 0 on success and -EINVAL on failure.
>> 1223 */
>> 1224 static int __init parse_crashkernel_suffix(char *cmdline,
>> 1225 unsigned long long *crash_size,
>> 1226 const char *suffix)
>> 1227 {
>> 1228 char *cur = cmdline;
>> 1229
>> 1230 *crash_size = memparse(cmdline, &cur);
>> 1231 if (cmdline == cur) {
>> 1232 pr_warn("crashkernel: memory value expected\n");
>> 1233 return -EINVAL;
>> 1234 }
>> 1235
>> 1236 /* check with suffix */
>> 1237 if (strncmp(cur, suffix, strlen(suffix))) {
>> 1238 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
>> 1239 return -EINVAL;
>> 1240 }
>> 1241 cur += strlen(suffix);
>> 1242 if (*cur != ' ' && *cur != '\0') {
>> 1243 pr_warn("crashkernel: unrecognized char: %c\n", *cur);
>> 1244 return -EINVAL;
>> 1245 }
>> 1246
956 return 0; 1247 return 0;
957 } 1248 }
958 late_initcall(kexec_core_sysctl_init); <<
959 #endif <<
960 1249
961 bool kexec_load_permitted(int kexec_image_type !! 1250 static __init char *get_last_crashkernel(char *cmdline,
>> 1251 const char *name,
>> 1252 const char *suffix)
>> 1253 {
>> 1254 char *p = cmdline, *ck_cmdline = NULL;
>> 1255
>> 1256 /* find crashkernel and use the last one if there are more */
>> 1257 p = strstr(p, name);
>> 1258 while (p) {
>> 1259 char *end_p = strchr(p, ' ');
>> 1260 char *q;
>> 1261
>> 1262 if (!end_p)
>> 1263 end_p = p + strlen(p);
>> 1264
>> 1265 if (!suffix) {
>> 1266 int i;
>> 1267
>> 1268 /* skip the one with any known suffix */
>> 1269 for (i = 0; suffix_tbl[i]; i++) {
>> 1270 q = end_p - strlen(suffix_tbl[i]);
>> 1271 if (!strncmp(q, suffix_tbl[i],
>> 1272 strlen(suffix_tbl[i])))
>> 1273 goto next;
>> 1274 }
>> 1275 ck_cmdline = p;
>> 1276 } else {
>> 1277 q = end_p - strlen(suffix);
>> 1278 if (!strncmp(q, suffix, strlen(suffix)))
>> 1279 ck_cmdline = p;
>> 1280 }
>> 1281 next:
>> 1282 p = strstr(p+1, name);
>> 1283 }
>> 1284
>> 1285 if (!ck_cmdline)
>> 1286 return NULL;
>> 1287
>> 1288 return ck_cmdline;
>> 1289 }
>> 1290
>> 1291 static int __init __parse_crashkernel(char *cmdline,
>> 1292 unsigned long long system_ram,
>> 1293 unsigned long long *crash_size,
>> 1294 unsigned long long *crash_base,
>> 1295 const char *name,
>> 1296 const char *suffix)
962 { 1297 {
963 struct kexec_load_limit *limit; !! 1298 char *first_colon, *first_space;
>> 1299 char *ck_cmdline;
964 1300
>> 1301 BUG_ON(!crash_size || !crash_base);
>> 1302 *crash_size = 0;
>> 1303 *crash_base = 0;
>> 1304
>> 1305 ck_cmdline = get_last_crashkernel(cmdline, name, suffix);
>> 1306
>> 1307 if (!ck_cmdline)
>> 1308 return -EINVAL;
>> 1309
>> 1310 ck_cmdline += strlen(name);
>> 1311
>> 1312 if (suffix)
>> 1313 return parse_crashkernel_suffix(ck_cmdline, crash_size,
>> 1314 suffix);
965 /* 1315 /*
966 * Only the superuser can use the kexe !! 1316 * if the commandline contains a ':', then that's the extended
967 * been disabled. !! 1317 * syntax -- if not, it must be the classic syntax
968 */ 1318 */
969 if (!capable(CAP_SYS_BOOT) || kexec_lo !! 1319 first_colon = strchr(ck_cmdline, ':');
970 return false; !! 1320 first_space = strchr(ck_cmdline, ' ');
>> 1321 if (first_colon && (!first_space || first_colon < first_space))
>> 1322 return parse_crashkernel_mem(ck_cmdline, system_ram,
>> 1323 crash_size, crash_base);
>> 1324
>> 1325 return parse_crashkernel_simple(ck_cmdline, crash_size, crash_base);
>> 1326 }
>> 1327
>> 1328 /*
>> 1329 * That function is the entry point for command line parsing and should be
>> 1330 * called from the arch-specific code.
>> 1331 */
>> 1332 int __init parse_crashkernel(char *cmdline,
>> 1333 unsigned long long system_ram,
>> 1334 unsigned long long *crash_size,
>> 1335 unsigned long long *crash_base)
>> 1336 {
>> 1337 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
>> 1338 "crashkernel=", NULL);
>> 1339 }
>> 1340
>> 1341 int __init parse_crashkernel_high(char *cmdline,
>> 1342 unsigned long long system_ram,
>> 1343 unsigned long long *crash_size,
>> 1344 unsigned long long *crash_base)
>> 1345 {
>> 1346 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
>> 1347 "crashkernel=", suffix_tbl[SUFFIX_HIGH]);
>> 1348 }
>> 1349
>> 1350 int __init parse_crashkernel_low(char *cmdline,
>> 1351 unsigned long long system_ram,
>> 1352 unsigned long long *crash_size,
>> 1353 unsigned long long *crash_base)
>> 1354 {
>> 1355 return __parse_crashkernel(cmdline, system_ram, crash_size, crash_base,
>> 1356 "crashkernel=", suffix_tbl[SUFFIX_LOW]);
>> 1357 }
>> 1358
>> 1359 static void update_vmcoreinfo_note(void)
>> 1360 {
>> 1361 u32 *buf = vmcoreinfo_note;
>> 1362
>> 1363 if (!vmcoreinfo_size)
>> 1364 return;
>> 1365 buf = append_elf_note(buf, VMCOREINFO_NOTE_NAME, 0, vmcoreinfo_data,
>> 1366 vmcoreinfo_size);
>> 1367 final_note(buf);
>> 1368 }
>> 1369
>> 1370 void crash_save_vmcoreinfo(void)
>> 1371 {
>> 1372 vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
>> 1373 update_vmcoreinfo_note();
>> 1374 }
>> 1375
>> 1376 void vmcoreinfo_append_str(const char *fmt, ...)
>> 1377 {
>> 1378 va_list args;
>> 1379 char buf[0x50];
>> 1380 size_t r;
>> 1381
>> 1382 va_start(args, fmt);
>> 1383 r = vscnprintf(buf, sizeof(buf), fmt, args);
>> 1384 va_end(args);
>> 1385
>> 1386 r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
>> 1387
>> 1388 memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
>> 1389
>> 1390 vmcoreinfo_size += r;
>> 1391 }
>> 1392
>> 1393 /*
>> 1394 * provide an empty default implementation here -- architecture
>> 1395 * code may override this
>> 1396 */
>> 1397 void __weak arch_crash_save_vmcoreinfo(void)
>> 1398 {}
>> 1399
>> 1400 phys_addr_t __weak paddr_vmcoreinfo_note(void)
>> 1401 {
>> 1402 return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
>> 1403 }
>> 1404
>> 1405 static int __init crash_save_vmcoreinfo_init(void)
>> 1406 {
>> 1407 VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
>> 1408 VMCOREINFO_PAGESIZE(PAGE_SIZE);
>> 1409
>> 1410 VMCOREINFO_SYMBOL(init_uts_ns);
>> 1411 VMCOREINFO_SYMBOL(node_online_map);
>> 1412 #ifdef CONFIG_MMU
>> 1413 VMCOREINFO_SYMBOL(swapper_pg_dir);
>> 1414 #endif
>> 1415 VMCOREINFO_SYMBOL(_stext);
>> 1416 VMCOREINFO_SYMBOL(vmap_area_list);
>> 1417
>> 1418 #ifndef CONFIG_NEED_MULTIPLE_NODES
>> 1419 VMCOREINFO_SYMBOL(mem_map);
>> 1420 VMCOREINFO_SYMBOL(contig_page_data);
>> 1421 #endif
>> 1422 #ifdef CONFIG_SPARSEMEM
>> 1423 VMCOREINFO_SYMBOL(mem_section);
>> 1424 VMCOREINFO_LENGTH(mem_section, NR_SECTION_ROOTS);
>> 1425 VMCOREINFO_STRUCT_SIZE(mem_section);
>> 1426 VMCOREINFO_OFFSET(mem_section, section_mem_map);
>> 1427 #endif
>> 1428 VMCOREINFO_STRUCT_SIZE(page);
>> 1429 VMCOREINFO_STRUCT_SIZE(pglist_data);
>> 1430 VMCOREINFO_STRUCT_SIZE(zone);
>> 1431 VMCOREINFO_STRUCT_SIZE(free_area);
>> 1432 VMCOREINFO_STRUCT_SIZE(list_head);
>> 1433 VMCOREINFO_SIZE(nodemask_t);
>> 1434 VMCOREINFO_OFFSET(page, flags);
>> 1435 VMCOREINFO_OFFSET(page, _refcount);
>> 1436 VMCOREINFO_OFFSET(page, mapping);
>> 1437 VMCOREINFO_OFFSET(page, lru);
>> 1438 VMCOREINFO_OFFSET(page, _mapcount);
>> 1439 VMCOREINFO_OFFSET(page, private);
>> 1440 VMCOREINFO_OFFSET(page, compound_dtor);
>> 1441 VMCOREINFO_OFFSET(page, compound_order);
>> 1442 VMCOREINFO_OFFSET(page, compound_head);
>> 1443 VMCOREINFO_OFFSET(pglist_data, node_zones);
>> 1444 VMCOREINFO_OFFSET(pglist_data, nr_zones);
>> 1445 #ifdef CONFIG_FLAT_NODE_MEM_MAP
>> 1446 VMCOREINFO_OFFSET(pglist_data, node_mem_map);
>> 1447 #endif
>> 1448 VMCOREINFO_OFFSET(pglist_data, node_start_pfn);
>> 1449 VMCOREINFO_OFFSET(pglist_data, node_spanned_pages);
>> 1450 VMCOREINFO_OFFSET(pglist_data, node_id);
>> 1451 VMCOREINFO_OFFSET(zone, free_area);
>> 1452 VMCOREINFO_OFFSET(zone, vm_stat);
>> 1453 VMCOREINFO_OFFSET(zone, spanned_pages);
>> 1454 VMCOREINFO_OFFSET(free_area, free_list);
>> 1455 VMCOREINFO_OFFSET(list_head, next);
>> 1456 VMCOREINFO_OFFSET(list_head, prev);
>> 1457 VMCOREINFO_OFFSET(vmap_area, va_start);
>> 1458 VMCOREINFO_OFFSET(vmap_area, list);
>> 1459 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
>> 1460 log_buf_kexec_setup();
>> 1461 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
>> 1462 VMCOREINFO_NUMBER(NR_FREE_PAGES);
>> 1463 VMCOREINFO_NUMBER(PG_lru);
>> 1464 VMCOREINFO_NUMBER(PG_private);
>> 1465 VMCOREINFO_NUMBER(PG_swapcache);
>> 1466 VMCOREINFO_NUMBER(PG_slab);
>> 1467 #ifdef CONFIG_MEMORY_FAILURE
>> 1468 VMCOREINFO_NUMBER(PG_hwpoison);
>> 1469 #endif
>> 1470 VMCOREINFO_NUMBER(PG_head_mask);
>> 1471 VMCOREINFO_NUMBER(PAGE_BUDDY_MAPCOUNT_VALUE);
>> 1472 #ifdef CONFIG_HUGETLB_PAGE
>> 1473 VMCOREINFO_NUMBER(HUGETLB_PAGE_DTOR);
>> 1474 #endif
971 1475
972 /* Check limit counter and decrease it !! 1476 arch_crash_save_vmcoreinfo();
973 limit = (kexec_image_type == KEXEC_TYP !! 1477 update_vmcoreinfo_note();
974 &load_limit_panic : &load_limi <<
975 mutex_lock(&limit->mutex); <<
976 if (!limit->limit) { <<
977 mutex_unlock(&limit->mutex); <<
978 return false; <<
979 } <<
980 if (limit->limit != -1) <<
981 limit->limit--; <<
982 mutex_unlock(&limit->mutex); <<
983 1478
984 return true; !! 1479 return 0;
985 } 1480 }
986 1481
>> 1482 subsys_initcall(crash_save_vmcoreinfo_init);
>> 1483
987 /* 1484 /*
988 * Move into place and start executing a prelo 1485 * Move into place and start executing a preloaded standalone
989 * executable. If nothing was preloaded retur 1486 * executable. If nothing was preloaded return an error.
990 */ 1487 */
991 int kernel_kexec(void) 1488 int kernel_kexec(void)
992 { 1489 {
993 int error = 0; 1490 int error = 0;
994 1491
995 if (!kexec_trylock()) !! 1492 if (!mutex_trylock(&kexec_mutex))
996 return -EBUSY; 1493 return -EBUSY;
997 if (!kexec_image) { 1494 if (!kexec_image) {
998 error = -EINVAL; 1495 error = -EINVAL;
999 goto Unlock; 1496 goto Unlock;
1000 } 1497 }
1001 1498
1002 #ifdef CONFIG_KEXEC_JUMP 1499 #ifdef CONFIG_KEXEC_JUMP
1003 if (kexec_image->preserve_context) { 1500 if (kexec_image->preserve_context) {
>> 1501 lock_system_sleep();
1004 pm_prepare_console(); 1502 pm_prepare_console();
1005 error = freeze_processes(); 1503 error = freeze_processes();
1006 if (error) { 1504 if (error) {
1007 error = -EBUSY; 1505 error = -EBUSY;
1008 goto Restore_console; 1506 goto Restore_console;
1009 } 1507 }
1010 suspend_console(); 1508 suspend_console();
1011 error = dpm_suspend_start(PMS 1509 error = dpm_suspend_start(PMSG_FREEZE);
1012 if (error) 1510 if (error)
1013 goto Resume_console; 1511 goto Resume_console;
1014 /* At this point, dpm_suspend 1512 /* At this point, dpm_suspend_start() has been called,
1015 * but *not* dpm_suspend_end( 1513 * but *not* dpm_suspend_end(). We *must* call
1016 * dpm_suspend_end() now. Ot 1514 * dpm_suspend_end() now. Otherwise, drivers for
1017 * some devices (e.g. interru 1515 * some devices (e.g. interrupt controllers) become
1018 * desynchronized with the ac 1516 * desynchronized with the actual state of the
1019 * hardware at resume time, a 1517 * hardware at resume time, and evil weirdness ensues.
1020 */ 1518 */
1021 error = dpm_suspend_end(PMSG_ 1519 error = dpm_suspend_end(PMSG_FREEZE);
1022 if (error) 1520 if (error)
1023 goto Resume_devices; 1521 goto Resume_devices;
1024 error = suspend_disable_secon !! 1522 error = disable_nonboot_cpus();
1025 if (error) 1523 if (error)
1026 goto Enable_cpus; 1524 goto Enable_cpus;
1027 local_irq_disable(); 1525 local_irq_disable();
1028 error = syscore_suspend(); 1526 error = syscore_suspend();
1029 if (error) 1527 if (error)
1030 goto Enable_irqs; 1528 goto Enable_irqs;
1031 } else 1529 } else
1032 #endif 1530 #endif
1033 { 1531 {
1034 kexec_in_progress = true; 1532 kexec_in_progress = true;
1035 kernel_restart_prepare("kexec !! 1533 kernel_restart_prepare(NULL);
1036 migrate_to_reboot_cpu(); 1534 migrate_to_reboot_cpu();
1037 syscore_shutdown(); <<
1038 1535
1039 /* 1536 /*
1040 * migrate_to_reboot_cpu() di 1537 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1041 * no further code needs to u 1538 * no further code needs to use CPU hotplug (which is true in
1042 * the reboot case). However, 1539 * the reboot case). However, the kexec path depends on using
1043 * CPU hotplug again; so re-e 1540 * CPU hotplug again; so re-enable it here.
1044 */ 1541 */
1045 cpu_hotplug_enable(); 1542 cpu_hotplug_enable();
1046 pr_notice("Starting new kerne !! 1543 pr_emerg("Starting new kernel\n");
1047 machine_shutdown(); 1544 machine_shutdown();
1048 } 1545 }
1049 1546
1050 kmsg_dump(KMSG_DUMP_SHUTDOWN); <<
1051 machine_kexec(kexec_image); 1547 machine_kexec(kexec_image);
1052 1548
1053 #ifdef CONFIG_KEXEC_JUMP 1549 #ifdef CONFIG_KEXEC_JUMP
1054 if (kexec_image->preserve_context) { 1550 if (kexec_image->preserve_context) {
1055 syscore_resume(); 1551 syscore_resume();
1056 Enable_irqs: 1552 Enable_irqs:
1057 local_irq_enable(); 1553 local_irq_enable();
1058 Enable_cpus: 1554 Enable_cpus:
1059 suspend_enable_secondary_cpus !! 1555 enable_nonboot_cpus();
1060 dpm_resume_start(PMSG_RESTORE 1556 dpm_resume_start(PMSG_RESTORE);
1061 Resume_devices: 1557 Resume_devices:
1062 dpm_resume_end(PMSG_RESTORE); 1558 dpm_resume_end(PMSG_RESTORE);
1063 Resume_console: 1559 Resume_console:
1064 resume_console(); 1560 resume_console();
1065 thaw_processes(); 1561 thaw_processes();
1066 Restore_console: 1562 Restore_console:
1067 pm_restore_console(); 1563 pm_restore_console();
>> 1564 unlock_system_sleep();
1068 } 1565 }
1069 #endif 1566 #endif
1070 1567
1071 Unlock: 1568 Unlock:
1072 kexec_unlock(); !! 1569 mutex_unlock(&kexec_mutex);
1073 return error; 1570 return error;
1074 } 1571 }
>> 1572
>> 1573 /*
>> 1574 * Protection mechanism for crashkernel reserved memory after
>> 1575 * the kdump kernel is loaded.
>> 1576 *
>> 1577 * Provide an empty default implementation here -- architecture
>> 1578 * code may override this
>> 1579 */
>> 1580 void __weak arch_kexec_protect_crashkres(void)
>> 1581 {}
>> 1582
>> 1583 void __weak arch_kexec_unprotect_crashkres(void)
>> 1584 {}
1075 1585
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.