1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * kexec.c - kexec system call core code. 3 * kexec.c - kexec system call core code.
4 * Copyright (C) 2002-2004 Eric Biederman <eb 4 * Copyright (C) 2002-2004 Eric Biederman <ebiederm@xmission.com>
5 */ 5 */
6 6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 8
9 #include <linux/btf.h> 9 #include <linux/btf.h>
10 #include <linux/capability.h> 10 #include <linux/capability.h>
11 #include <linux/mm.h> 11 #include <linux/mm.h>
12 #include <linux/file.h> 12 #include <linux/file.h>
13 #include <linux/slab.h> 13 #include <linux/slab.h>
14 #include <linux/fs.h> 14 #include <linux/fs.h>
15 #include <linux/kexec.h> 15 #include <linux/kexec.h>
16 #include <linux/mutex.h> 16 #include <linux/mutex.h>
17 #include <linux/list.h> 17 #include <linux/list.h>
18 #include <linux/highmem.h> 18 #include <linux/highmem.h>
19 #include <linux/syscalls.h> 19 #include <linux/syscalls.h>
20 #include <linux/reboot.h> 20 #include <linux/reboot.h>
21 #include <linux/ioport.h> 21 #include <linux/ioport.h>
22 #include <linux/hardirq.h> 22 #include <linux/hardirq.h>
23 #include <linux/elf.h> 23 #include <linux/elf.h>
24 #include <linux/elfcore.h> 24 #include <linux/elfcore.h>
25 #include <linux/utsname.h> 25 #include <linux/utsname.h>
26 #include <linux/numa.h> 26 #include <linux/numa.h>
27 #include <linux/suspend.h> 27 #include <linux/suspend.h>
28 #include <linux/device.h> 28 #include <linux/device.h>
29 #include <linux/freezer.h> 29 #include <linux/freezer.h>
30 #include <linux/panic_notifier.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> 41 #include <linux/objtool.h>
42 #include <linux/kmsg_dump.h> 42 #include <linux/kmsg_dump.h>
43 43
44 #include <asm/page.h> 44 #include <asm/page.h>
45 #include <asm/sections.h> 45 #include <asm/sections.h>
46 46
47 #include <crypto/hash.h> 47 #include <crypto/hash.h>
48 #include "kexec_internal.h" 48 #include "kexec_internal.h"
49 49
50 atomic_t __kexec_lock = ATOMIC_INIT(0); 50 atomic_t __kexec_lock = ATOMIC_INIT(0);
51 51
52 /* Flag to indicate we are going to kexec a ne 52 /* Flag to indicate we are going to kexec a new kernel */
53 bool kexec_in_progress = false; 53 bool kexec_in_progress = false;
54 54
55 bool kexec_file_dbg_print; 55 bool kexec_file_dbg_print;
56 56
>> 57 int kexec_should_crash(struct task_struct *p)
>> 58 {
>> 59 /*
>> 60 * If crash_kexec_post_notifiers is enabled, don't run
>> 61 * crash_kexec() here yet, which must be run after panic
>> 62 * notifiers in panic().
>> 63 */
>> 64 if (crash_kexec_post_notifiers)
>> 65 return 0;
>> 66 /*
>> 67 * There are 4 panic() calls in make_task_dead() path, each of which
>> 68 * corresponds to each of these 4 conditions.
>> 69 */
>> 70 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
>> 71 return 1;
>> 72 return 0;
>> 73 }
>> 74
>> 75 int kexec_crash_loaded(void)
>> 76 {
>> 77 return !!kexec_crash_image;
>> 78 }
>> 79 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
>> 80
57 /* 81 /*
58 * When kexec transitions to the new kernel th 82 * When kexec transitions to the new kernel there is a one-to-one
59 * mapping between physical and virtual addres 83 * mapping between physical and virtual addresses. On processors
60 * where you can disable the MMU this is trivi 84 * where you can disable the MMU this is trivial, and easy. For
61 * others it is still a simple predictable pag 85 * others it is still a simple predictable page table to setup.
62 * 86 *
63 * In that environment kexec copies the new ke 87 * In that environment kexec copies the new kernel to its final
64 * resting place. This means I can only suppo 88 * resting place. This means I can only support memory whose
65 * physical address can fit in an unsigned lon 89 * physical address can fit in an unsigned long. In particular
66 * addresses where (pfn << PAGE_SHIFT) > ULONG 90 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
67 * If the assembly stub has more restrictive r 91 * If the assembly stub has more restrictive requirements
68 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_ME 92 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
69 * defined more restrictively in <asm/kexec.h> 93 * defined more restrictively in <asm/kexec.h>.
70 * 94 *
71 * The code for the transition from the curren 95 * The code for the transition from the current kernel to the
72 * new kernel is placed in the control_code_bu 96 * new kernel is placed in the control_code_buffer, whose size
73 * is given by KEXEC_CONTROL_PAGE_SIZE. In th 97 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
74 * page of memory is necessary, but some archi 98 * page of memory is necessary, but some architectures require more.
75 * Because this memory must be identity mapped 99 * Because this memory must be identity mapped in the transition from
76 * virtual to physical addresses it must live 100 * virtual to physical addresses it must live in the range
77 * 0 - TASK_SIZE, as only the user space mappi 101 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
78 * modifiable. 102 * modifiable.
79 * 103 *
80 * The assembly stub in the control code buffe 104 * The assembly stub in the control code buffer is passed a linked list
81 * of descriptor pages detailing the source pa 105 * of descriptor pages detailing the source pages of the new kernel,
82 * and the destination addresses of those sour 106 * and the destination addresses of those source pages. As this data
83 * structure is not used in the context of the 107 * structure is not used in the context of the current OS, it must
84 * be self-contained. 108 * be self-contained.
85 * 109 *
86 * The code has been made to work with highmem 110 * The code has been made to work with highmem pages and will use a
87 * destination page in its final resting place 111 * destination page in its final resting place (if it happens
88 * to allocate it). The end product of this i 112 * to allocate it). The end product of this is that most of the
89 * physical address space, and most of RAM can 113 * physical address space, and most of RAM can be used.
90 * 114 *
91 * Future directions include: 115 * Future directions include:
92 * - allocating a page table with the control 116 * - allocating a page table with the control code buffer identity
93 * mapped, to simplify machine_kexec and ma 117 * mapped, to simplify machine_kexec and make kexec_on_panic more
94 * reliable. 118 * reliable.
95 */ 119 */
96 120
97 /* 121 /*
98 * KIMAGE_NO_DEST is an impossible destination 122 * KIMAGE_NO_DEST is an impossible destination address..., for
99 * allocating pages whose destination address 123 * allocating pages whose destination address we do not care about.
100 */ 124 */
101 #define KIMAGE_NO_DEST (-1UL) 125 #define KIMAGE_NO_DEST (-1UL)
102 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) > 126 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
103 127
104 static struct page *kimage_alloc_page(struct k 128 static struct page *kimage_alloc_page(struct kimage *image,
105 gfp_t g 129 gfp_t gfp_mask,
106 unsigne 130 unsigned long dest);
107 131
108 int sanity_check_segment_list(struct kimage *i 132 int sanity_check_segment_list(struct kimage *image)
109 { 133 {
110 int i; 134 int i;
111 unsigned long nr_segments = image->nr_ 135 unsigned long nr_segments = image->nr_segments;
112 unsigned long total_pages = 0; 136 unsigned long total_pages = 0;
113 unsigned long nr_pages = totalram_page 137 unsigned long nr_pages = totalram_pages();
114 138
115 /* 139 /*
116 * Verify we have good destination add 140 * Verify we have good destination addresses. The caller is
117 * responsible for making certain we d 141 * responsible for making certain we don't attempt to load
118 * the new image into invalid or reser 142 * the new image into invalid or reserved areas of RAM. This
119 * just verifies it is an address we c 143 * just verifies it is an address we can use.
120 * 144 *
121 * Since the kernel does everything in 145 * Since the kernel does everything in page size chunks ensure
122 * the destination addresses are page 146 * the destination addresses are page aligned. Too many
123 * special cases crop of when we don't 147 * special cases crop of when we don't do this. The most
124 * insidious is getting overlapping de 148 * insidious is getting overlapping destination addresses
125 * simply because addresses are change 149 * simply because addresses are changed to page size
126 * granularity. 150 * granularity.
127 */ 151 */
128 for (i = 0; i < nr_segments; i++) { 152 for (i = 0; i < nr_segments; i++) {
129 unsigned long mstart, mend; 153 unsigned long mstart, mend;
130 154
131 mstart = image->segment[i].mem 155 mstart = image->segment[i].mem;
132 mend = mstart + image->segme 156 mend = mstart + image->segment[i].memsz;
133 if (mstart > mend) 157 if (mstart > mend)
134 return -EADDRNOTAVAIL; 158 return -EADDRNOTAVAIL;
135 if ((mstart & ~PAGE_MASK) || ( 159 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
136 return -EADDRNOTAVAIL; 160 return -EADDRNOTAVAIL;
137 if (mend >= KEXEC_DESTINATION_ 161 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
138 return -EADDRNOTAVAIL; 162 return -EADDRNOTAVAIL;
139 } 163 }
140 164
141 /* Verify our destination addresses do 165 /* Verify our destination addresses do not overlap.
142 * If we alloed overlapping destinatio 166 * If we alloed overlapping destination addresses
143 * through very weird things can happe 167 * through very weird things can happen with no
144 * easy explanation as one segment sto 168 * easy explanation as one segment stops on another.
145 */ 169 */
146 for (i = 0; i < nr_segments; i++) { 170 for (i = 0; i < nr_segments; i++) {
147 unsigned long mstart, mend; 171 unsigned long mstart, mend;
148 unsigned long j; 172 unsigned long j;
149 173
150 mstart = image->segment[i].mem 174 mstart = image->segment[i].mem;
151 mend = mstart + image->segme 175 mend = mstart + image->segment[i].memsz;
152 for (j = 0; j < i; j++) { 176 for (j = 0; j < i; j++) {
153 unsigned long pstart, 177 unsigned long pstart, pend;
154 178
155 pstart = image->segmen 179 pstart = image->segment[j].mem;
156 pend = pstart + imag 180 pend = pstart + image->segment[j].memsz;
157 /* Do the segments ove 181 /* Do the segments overlap ? */
158 if ((mend > pstart) && 182 if ((mend > pstart) && (mstart < pend))
159 return -EINVAL 183 return -EINVAL;
160 } 184 }
161 } 185 }
162 186
163 /* Ensure our buffer sizes are strictl 187 /* Ensure our buffer sizes are strictly less than
164 * our memory sizes. This should alwa 188 * our memory sizes. This should always be the case,
165 * and it is easier to check up front 189 * and it is easier to check up front than to be surprised
166 * later on. 190 * later on.
167 */ 191 */
168 for (i = 0; i < nr_segments; i++) { 192 for (i = 0; i < nr_segments; i++) {
169 if (image->segment[i].bufsz > 193 if (image->segment[i].bufsz > image->segment[i].memsz)
170 return -EINVAL; 194 return -EINVAL;
171 } 195 }
172 196
173 /* 197 /*
174 * Verify that no more than half of me 198 * Verify that no more than half of memory will be consumed. If the
175 * request from userspace is too large 199 * request from userspace is too large, a large amount of time will be
176 * wasted allocating pages, which can 200 * wasted allocating pages, which can cause a soft lockup.
177 */ 201 */
178 for (i = 0; i < nr_segments; i++) { 202 for (i = 0; i < nr_segments; i++) {
179 if (PAGE_COUNT(image->segment[ 203 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
180 return -EINVAL; 204 return -EINVAL;
181 205
182 total_pages += PAGE_COUNT(imag 206 total_pages += PAGE_COUNT(image->segment[i].memsz);
183 } 207 }
184 208
185 if (total_pages > nr_pages / 2) 209 if (total_pages > nr_pages / 2)
186 return -EINVAL; 210 return -EINVAL;
187 211
188 #ifdef CONFIG_CRASH_DUMP <<
189 /* 212 /*
190 * Verify we have good destination add 213 * Verify we have good destination addresses. Normally
191 * the caller is responsible for makin 214 * the caller is responsible for making certain we don't
192 * attempt to load the new image into 215 * attempt to load the new image into invalid or reserved
193 * areas of RAM. But crash kernels ar 216 * areas of RAM. But crash kernels are preloaded into a
194 * reserved area of ram. We must ensu 217 * reserved area of ram. We must ensure the addresses
195 * are in the reserved area otherwise 218 * are in the reserved area otherwise preloading the
196 * kernel could corrupt things. 219 * kernel could corrupt things.
197 */ 220 */
198 221
199 if (image->type == KEXEC_TYPE_CRASH) { 222 if (image->type == KEXEC_TYPE_CRASH) {
200 for (i = 0; i < nr_segments; i 223 for (i = 0; i < nr_segments; i++) {
201 unsigned long mstart, 224 unsigned long mstart, mend;
202 225
203 mstart = image->segmen 226 mstart = image->segment[i].mem;
204 mend = mstart + image- 227 mend = mstart + image->segment[i].memsz - 1;
205 /* Ensure we are withi 228 /* Ensure we are within the crash kernel limits */
206 if ((mstart < phys_to_ 229 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
207 (mend > phys_to_bo 230 (mend > phys_to_boot_phys(crashk_res.end)))
208 return -EADDRN 231 return -EADDRNOTAVAIL;
209 } 232 }
210 } 233 }
211 #endif <<
212 234
213 return 0; 235 return 0;
214 } 236 }
215 237
216 struct kimage *do_kimage_alloc_init(void) 238 struct kimage *do_kimage_alloc_init(void)
217 { 239 {
218 struct kimage *image; 240 struct kimage *image;
219 241
220 /* Allocate a controlling structure */ 242 /* Allocate a controlling structure */
221 image = kzalloc(sizeof(*image), GFP_KE 243 image = kzalloc(sizeof(*image), GFP_KERNEL);
222 if (!image) 244 if (!image)
223 return NULL; 245 return NULL;
224 246
225 image->head = 0; 247 image->head = 0;
226 image->entry = &image->head; 248 image->entry = &image->head;
227 image->last_entry = &image->head; 249 image->last_entry = &image->head;
228 image->control_page = ~0; /* By defaul 250 image->control_page = ~0; /* By default this does not apply */
229 image->type = KEXEC_TYPE_DEFAULT; 251 image->type = KEXEC_TYPE_DEFAULT;
230 252
231 /* Initialize the list of control page 253 /* Initialize the list of control pages */
232 INIT_LIST_HEAD(&image->control_pages); 254 INIT_LIST_HEAD(&image->control_pages);
233 255
234 /* Initialize the list of destination 256 /* Initialize the list of destination pages */
235 INIT_LIST_HEAD(&image->dest_pages); 257 INIT_LIST_HEAD(&image->dest_pages);
236 258
237 /* Initialize the list of unusable pag 259 /* Initialize the list of unusable pages */
238 INIT_LIST_HEAD(&image->unusable_pages) 260 INIT_LIST_HEAD(&image->unusable_pages);
239 261
240 #ifdef CONFIG_CRASH_HOTPLUG 262 #ifdef CONFIG_CRASH_HOTPLUG
241 image->hp_action = KEXEC_CRASH_HP_NONE 263 image->hp_action = KEXEC_CRASH_HP_NONE;
242 image->elfcorehdr_index = -1; 264 image->elfcorehdr_index = -1;
243 image->elfcorehdr_updated = false; 265 image->elfcorehdr_updated = false;
244 #endif 266 #endif
245 267
246 return image; 268 return image;
247 } 269 }
248 270
249 int kimage_is_destination_range(struct kimage 271 int kimage_is_destination_range(struct kimage *image,
250 unsign 272 unsigned long start,
251 unsign 273 unsigned long end)
252 { 274 {
253 unsigned long i; 275 unsigned long i;
254 276
255 for (i = 0; i < image->nr_segments; i+ 277 for (i = 0; i < image->nr_segments; i++) {
256 unsigned long mstart, mend; 278 unsigned long mstart, mend;
257 279
258 mstart = image->segment[i].mem 280 mstart = image->segment[i].mem;
259 mend = mstart + image->segment 281 mend = mstart + image->segment[i].memsz - 1;
260 if ((end >= mstart) && (start 282 if ((end >= mstart) && (start <= mend))
261 return 1; 283 return 1;
262 } 284 }
263 285
264 return 0; 286 return 0;
265 } 287 }
266 288
267 static struct page *kimage_alloc_pages(gfp_t g 289 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
268 { 290 {
269 struct page *pages; 291 struct page *pages;
270 292
271 if (fatal_signal_pending(current)) 293 if (fatal_signal_pending(current))
272 return NULL; 294 return NULL;
273 pages = alloc_pages(gfp_mask & ~__GFP_ 295 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
274 if (pages) { 296 if (pages) {
275 unsigned int count, i; 297 unsigned int count, i;
276 298
277 pages->mapping = NULL; 299 pages->mapping = NULL;
278 set_page_private(pages, order) 300 set_page_private(pages, order);
279 count = 1 << order; 301 count = 1 << order;
280 for (i = 0; i < count; i++) 302 for (i = 0; i < count; i++)
281 SetPageReserved(pages 303 SetPageReserved(pages + i);
282 304
283 arch_kexec_post_alloc_pages(pa 305 arch_kexec_post_alloc_pages(page_address(pages), count,
284 gf 306 gfp_mask);
285 307
286 if (gfp_mask & __GFP_ZERO) 308 if (gfp_mask & __GFP_ZERO)
287 for (i = 0; i < count; 309 for (i = 0; i < count; i++)
288 clear_highpage 310 clear_highpage(pages + i);
289 } 311 }
290 312
291 return pages; 313 return pages;
292 } 314 }
293 315
294 static void kimage_free_pages(struct page *pag 316 static void kimage_free_pages(struct page *page)
295 { 317 {
296 unsigned int order, count, i; 318 unsigned int order, count, i;
297 319
298 order = page_private(page); 320 order = page_private(page);
299 count = 1 << order; 321 count = 1 << order;
300 322
301 arch_kexec_pre_free_pages(page_address 323 arch_kexec_pre_free_pages(page_address(page), count);
302 324
303 for (i = 0; i < count; i++) 325 for (i = 0; i < count; i++)
304 ClearPageReserved(page + i); 326 ClearPageReserved(page + i);
305 __free_pages(page, order); 327 __free_pages(page, order);
306 } 328 }
307 329
308 void kimage_free_page_list(struct list_head *l 330 void kimage_free_page_list(struct list_head *list)
309 { 331 {
310 struct page *page, *next; 332 struct page *page, *next;
311 333
312 list_for_each_entry_safe(page, next, l 334 list_for_each_entry_safe(page, next, list, lru) {
313 list_del(&page->lru); 335 list_del(&page->lru);
314 kimage_free_pages(page); 336 kimage_free_pages(page);
315 } 337 }
316 } 338 }
317 339
318 static struct page *kimage_alloc_normal_contro 340 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
319 341 unsigned int order)
320 { 342 {
321 /* Control pages are special, they are 343 /* Control pages are special, they are the intermediaries
322 * that are needed while we copy the r 344 * that are needed while we copy the rest of the pages
323 * to their final resting place. As s 345 * to their final resting place. As such they must
324 * not conflict with either the destin 346 * not conflict with either the destination addresses
325 * or memory the kernel is already usi 347 * or memory the kernel is already using.
326 * 348 *
327 * The only case where we really need 349 * The only case where we really need more than one of
328 * these are for architectures where w 350 * these are for architectures where we cannot disable
329 * the MMU and must instead generate a 351 * the MMU and must instead generate an identity mapped
330 * page table for all of the memory. 352 * page table for all of the memory.
331 * 353 *
332 * At worst this runs in O(N) of the i 354 * At worst this runs in O(N) of the image size.
333 */ 355 */
334 struct list_head extra_pages; 356 struct list_head extra_pages;
335 struct page *pages; 357 struct page *pages;
336 unsigned int count; 358 unsigned int count;
337 359
338 count = 1 << order; 360 count = 1 << order;
339 INIT_LIST_HEAD(&extra_pages); 361 INIT_LIST_HEAD(&extra_pages);
340 362
341 /* Loop while I can allocate a page an 363 /* Loop while I can allocate a page and the page allocated
342 * is a destination page. 364 * is a destination page.
343 */ 365 */
344 do { 366 do {
345 unsigned long pfn, epfn, addr, 367 unsigned long pfn, epfn, addr, eaddr;
346 368
347 pages = kimage_alloc_pages(KEX 369 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
348 if (!pages) 370 if (!pages)
349 break; 371 break;
350 pfn = page_to_boot_pfn(pages 372 pfn = page_to_boot_pfn(pages);
351 epfn = pfn + count; 373 epfn = pfn + count;
352 addr = pfn << PAGE_SHIFT; 374 addr = pfn << PAGE_SHIFT;
353 eaddr = (epfn << PAGE_SHIFT) - 375 eaddr = (epfn << PAGE_SHIFT) - 1;
354 if ((epfn >= (KEXEC_CONTROL_ME 376 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
355 kimage_is_destin 377 kimage_is_destination_range(image, addr, eaddr)) {
356 list_add(&pages->lru, 378 list_add(&pages->lru, &extra_pages);
357 pages = NULL; 379 pages = NULL;
358 } 380 }
359 } while (!pages); 381 } while (!pages);
360 382
361 if (pages) { 383 if (pages) {
362 /* Remember the allocated page 384 /* Remember the allocated page... */
363 list_add(&pages->lru, &image-> 385 list_add(&pages->lru, &image->control_pages);
364 386
365 /* Because the page is already 387 /* Because the page is already in it's destination
366 * location we will never allo 388 * location we will never allocate another page at
367 * that address. Therefore ki 389 * that address. Therefore kimage_alloc_pages
368 * will not return it (again) 390 * will not return it (again) and we don't need
369 * to give it an entry in imag 391 * to give it an entry in image->segment[].
370 */ 392 */
371 } 393 }
372 /* Deal with the destination pages I h 394 /* Deal with the destination pages I have inadvertently allocated.
373 * 395 *
374 * Ideally I would convert multi-page 396 * Ideally I would convert multi-page allocations into single
375 * page allocations, and add everythin 397 * page allocations, and add everything to image->dest_pages.
376 * 398 *
377 * For now it is simpler to just free 399 * For now it is simpler to just free the pages.
378 */ 400 */
379 kimage_free_page_list(&extra_pages); 401 kimage_free_page_list(&extra_pages);
380 402
381 return pages; 403 return pages;
382 } 404 }
383 405
384 #ifdef CONFIG_CRASH_DUMP <<
385 static struct page *kimage_alloc_crash_control 406 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
386 407 unsigned int order)
387 { 408 {
388 /* Control pages are special, they are 409 /* Control pages are special, they are the intermediaries
389 * that are needed while we copy the r 410 * that are needed while we copy the rest of the pages
390 * to their final resting place. As s 411 * to their final resting place. As such they must
391 * not conflict with either the destin 412 * not conflict with either the destination addresses
392 * or memory the kernel is already usi 413 * or memory the kernel is already using.
393 * 414 *
394 * Control pages are also the only pag 415 * Control pages are also the only pags we must allocate
395 * when loading a crash kernel. All o 416 * when loading a crash kernel. All of the other pages
396 * are specified by the segments and w 417 * are specified by the segments and we just memcpy
397 * into them directly. 418 * into them directly.
398 * 419 *
399 * The only case where we really need 420 * The only case where we really need more than one of
400 * these are for architectures where w 421 * these are for architectures where we cannot disable
401 * the MMU and must instead generate a 422 * the MMU and must instead generate an identity mapped
402 * page table for all of the memory. 423 * page table for all of the memory.
403 * 424 *
404 * Given the low demand this implement 425 * Given the low demand this implements a very simple
405 * allocator that finds the first hole 426 * allocator that finds the first hole of the appropriate
406 * size in the reserved memory region, 427 * size in the reserved memory region, and allocates all
407 * of the memory up to and including t 428 * of the memory up to and including the hole.
408 */ 429 */
409 unsigned long hole_start, hole_end, si 430 unsigned long hole_start, hole_end, size;
410 struct page *pages; 431 struct page *pages;
411 432
412 pages = NULL; 433 pages = NULL;
413 size = (1 << order) << PAGE_SHIFT; 434 size = (1 << order) << PAGE_SHIFT;
414 hole_start = ALIGN(image->control_page 435 hole_start = ALIGN(image->control_page, size);
415 hole_end = hole_start + size - 1; 436 hole_end = hole_start + size - 1;
416 while (hole_end <= crashk_res.end) { 437 while (hole_end <= crashk_res.end) {
417 unsigned long i; 438 unsigned long i;
418 439
419 cond_resched(); 440 cond_resched();
420 441
421 if (hole_end > KEXEC_CRASH_CON 442 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
422 break; 443 break;
423 /* See if I overlap any of the 444 /* See if I overlap any of the segments */
424 for (i = 0; i < image->nr_segm 445 for (i = 0; i < image->nr_segments; i++) {
425 unsigned long mstart, 446 unsigned long mstart, mend;
426 447
427 mstart = image->segmen 448 mstart = image->segment[i].mem;
428 mend = mstart + imag 449 mend = mstart + image->segment[i].memsz - 1;
429 if ((hole_end >= mstar 450 if ((hole_end >= mstart) && (hole_start <= mend)) {
430 /* Advance the 451 /* Advance the hole to the end of the segment */
431 hole_start = A 452 hole_start = ALIGN(mend, size);
432 hole_end = h 453 hole_end = hole_start + size - 1;
433 break; 454 break;
434 } 455 }
435 } 456 }
436 /* If I don't overlap any segm 457 /* If I don't overlap any segments I have found my hole! */
437 if (i == image->nr_segments) { 458 if (i == image->nr_segments) {
438 pages = pfn_to_page(ho 459 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
439 image->control_page = 460 image->control_page = hole_end + 1;
440 break; 461 break;
441 } 462 }
442 } 463 }
443 464
444 /* Ensure that these pages are decrypt 465 /* Ensure that these pages are decrypted if SME is enabled. */
445 if (pages) 466 if (pages)
446 arch_kexec_post_alloc_pages(pa 467 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
447 468
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
>> 490 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
>> 491 {
>> 492 struct page *vmcoreinfo_page;
>> 493 void *safecopy;
>> 494
>> 495 if (image->type != KEXEC_TYPE_CRASH)
>> 496 return 0;
>> 497
>> 498 /*
>> 499 * For kdump, allocate one vmcoreinfo safe copy from the
>> 500 * crash memory. as we have arch_kexec_protect_crashkres()
>> 501 * after kexec syscall, we naturally protect it from write
>> 502 * (even read) access under kernel direct mapping. But on
>> 503 * the other hand, we still need to operate it when crash
>> 504 * happens to generate vmcoreinfo note, hereby we rely on
>> 505 * vmap for this purpose.
>> 506 */
>> 507 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
>> 508 if (!vmcoreinfo_page) {
>> 509 pr_warn("Could not allocate vmcoreinfo buffer\n");
>> 510 return -ENOMEM;
>> 511 }
>> 512 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
>> 513 if (!safecopy) {
>> 514 pr_warn("Could not vmap vmcoreinfo buffer\n");
>> 515 return -ENOMEM;
>> 516 }
>> 517
>> 518 image->vmcoreinfo_data_copy = safecopy;
>> 519 crash_update_vmcoreinfo_safecopy(safecopy);
>> 520
>> 521 return 0;
>> 522 }
>> 523
472 static int kimage_add_entry(struct kimage *ima 524 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
473 { 525 {
474 if (*image->entry != 0) 526 if (*image->entry != 0)
475 image->entry++; 527 image->entry++;
476 528
477 if (image->entry == image->last_entry) 529 if (image->entry == image->last_entry) {
478 kimage_entry_t *ind_page; 530 kimage_entry_t *ind_page;
479 struct page *page; 531 struct page *page;
480 532
481 page = kimage_alloc_page(image 533 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
482 if (!page) 534 if (!page)
483 return -ENOMEM; 535 return -ENOMEM;
484 536
485 ind_page = page_address(page); 537 ind_page = page_address(page);
486 *image->entry = virt_to_boot_p 538 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
487 image->entry = ind_page; 539 image->entry = ind_page;
488 image->last_entry = ind_page + 540 image->last_entry = ind_page +
489 ((PAGE_S 541 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
490 } 542 }
491 *image->entry = entry; 543 *image->entry = entry;
492 image->entry++; 544 image->entry++;
493 *image->entry = 0; 545 *image->entry = 0;
494 546
495 return 0; 547 return 0;
496 } 548 }
497 549
498 static int kimage_set_destination(struct kimag 550 static int kimage_set_destination(struct kimage *image,
499 unsigned lo 551 unsigned long destination)
500 { 552 {
501 destination &= PAGE_MASK; 553 destination &= PAGE_MASK;
502 554
503 return kimage_add_entry(image, destina 555 return kimage_add_entry(image, destination | IND_DESTINATION);
504 } 556 }
505 557
506 558
507 static int kimage_add_page(struct kimage *imag 559 static int kimage_add_page(struct kimage *image, unsigned long page)
508 { 560 {
509 page &= PAGE_MASK; 561 page &= PAGE_MASK;
510 562
511 return kimage_add_entry(image, page | 563 return kimage_add_entry(image, page | IND_SOURCE);
512 } 564 }
513 565
514 566
515 static void kimage_free_extra_pages(struct kim 567 static void kimage_free_extra_pages(struct kimage *image)
516 { 568 {
517 /* Walk through and free any extra des 569 /* Walk through and free any extra destination pages I may have */
518 kimage_free_page_list(&image->dest_pag 570 kimage_free_page_list(&image->dest_pages);
519 571
520 /* Walk through and free any unusable 572 /* Walk through and free any unusable pages I have cached */
521 kimage_free_page_list(&image->unusable 573 kimage_free_page_list(&image->unusable_pages);
522 574
523 } 575 }
524 576
525 void kimage_terminate(struct kimage *image) 577 void kimage_terminate(struct kimage *image)
526 { 578 {
527 if (*image->entry != 0) 579 if (*image->entry != 0)
528 image->entry++; 580 image->entry++;
529 581
530 *image->entry = IND_DONE; 582 *image->entry = IND_DONE;
531 } 583 }
532 584
533 #define for_each_kimage_entry(image, ptr, entr 585 #define for_each_kimage_entry(image, ptr, entry) \
534 for (ptr = &image->head; (entry = *ptr 586 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
535 ptr = (entry & IND_INDIRECTION 587 ptr = (entry & IND_INDIRECTION) ? \
536 boot_phys_to_virt((ent 588 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
537 589
538 static void kimage_free_entry(kimage_entry_t e 590 static void kimage_free_entry(kimage_entry_t entry)
539 { 591 {
540 struct page *page; 592 struct page *page;
541 593
542 page = boot_pfn_to_page(entry >> PAGE_ 594 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
543 kimage_free_pages(page); 595 kimage_free_pages(page);
544 } 596 }
545 597
546 void kimage_free(struct kimage *image) 598 void kimage_free(struct kimage *image)
547 { 599 {
548 kimage_entry_t *ptr, entry; 600 kimage_entry_t *ptr, entry;
549 kimage_entry_t ind = 0; 601 kimage_entry_t ind = 0;
550 602
551 if (!image) 603 if (!image)
552 return; 604 return;
553 605
554 #ifdef CONFIG_CRASH_DUMP <<
555 if (image->vmcoreinfo_data_copy) { 606 if (image->vmcoreinfo_data_copy) {
556 crash_update_vmcoreinfo_safeco 607 crash_update_vmcoreinfo_safecopy(NULL);
557 vunmap(image->vmcoreinfo_data_ 608 vunmap(image->vmcoreinfo_data_copy);
558 } 609 }
559 #endif <<
560 610
561 kimage_free_extra_pages(image); 611 kimage_free_extra_pages(image);
562 for_each_kimage_entry(image, ptr, entr 612 for_each_kimage_entry(image, ptr, entry) {
563 if (entry & IND_INDIRECTION) { 613 if (entry & IND_INDIRECTION) {
564 /* Free the previous i 614 /* Free the previous indirection page */
565 if (ind & IND_INDIRECT 615 if (ind & IND_INDIRECTION)
566 kimage_free_en 616 kimage_free_entry(ind);
567 /* Save this indirecti 617 /* Save this indirection page until we are
568 * done with it. 618 * done with it.
569 */ 619 */
570 ind = entry; 620 ind = entry;
571 } else if (entry & IND_SOURCE) 621 } else if (entry & IND_SOURCE)
572 kimage_free_entry(entr 622 kimage_free_entry(entry);
573 } 623 }
574 /* Free the final indirection page */ 624 /* Free the final indirection page */
575 if (ind & IND_INDIRECTION) 625 if (ind & IND_INDIRECTION)
576 kimage_free_entry(ind); 626 kimage_free_entry(ind);
577 627
578 /* Handle any machine specific cleanup 628 /* Handle any machine specific cleanup */
579 machine_kexec_cleanup(image); 629 machine_kexec_cleanup(image);
580 630
581 /* Free the kexec control pages... */ 631 /* Free the kexec control pages... */
582 kimage_free_page_list(&image->control_ 632 kimage_free_page_list(&image->control_pages);
583 633
584 /* 634 /*
585 * Free up any temporary buffers alloc 635 * Free up any temporary buffers allocated. This might hit if
586 * error occurred much later after buf 636 * error occurred much later after buffer allocation.
587 */ 637 */
588 if (image->file_mode) 638 if (image->file_mode)
589 kimage_file_post_load_cleanup( 639 kimage_file_post_load_cleanup(image);
590 640
591 kfree(image); 641 kfree(image);
592 } 642 }
593 643
594 static kimage_entry_t *kimage_dst_used(struct 644 static kimage_entry_t *kimage_dst_used(struct kimage *image,
595 unsign 645 unsigned long page)
596 { 646 {
597 kimage_entry_t *ptr, entry; 647 kimage_entry_t *ptr, entry;
598 unsigned long destination = 0; 648 unsigned long destination = 0;
599 649
600 for_each_kimage_entry(image, ptr, entr 650 for_each_kimage_entry(image, ptr, entry) {
601 if (entry & IND_DESTINATION) 651 if (entry & IND_DESTINATION)
602 destination = entry & 652 destination = entry & PAGE_MASK;
603 else if (entry & IND_SOURCE) { 653 else if (entry & IND_SOURCE) {
604 if (page == destinatio 654 if (page == destination)
605 return ptr; 655 return ptr;
606 destination += PAGE_SI 656 destination += PAGE_SIZE;
607 } 657 }
608 } 658 }
609 659
610 return NULL; 660 return NULL;
611 } 661 }
612 662
613 static struct page *kimage_alloc_page(struct k 663 static struct page *kimage_alloc_page(struct kimage *image,
614 gfp_t 664 gfp_t gfp_mask,
615 unsign 665 unsigned long destination)
616 { 666 {
617 /* 667 /*
618 * Here we implement safeguards to ens 668 * Here we implement safeguards to ensure that a source page
619 * is not copied to its destination pa 669 * is not copied to its destination page before the data on
620 * the destination page is no longer u 670 * the destination page is no longer useful.
621 * 671 *
622 * To do this we maintain the invarian 672 * To do this we maintain the invariant that a source page is
623 * either its own destination page, or 673 * either its own destination page, or it is not a
624 * destination page at all. 674 * destination page at all.
625 * 675 *
626 * That is slightly stronger than requ 676 * That is slightly stronger than required, but the proof
627 * that no problems will not occur is 677 * that no problems will not occur is trivial, and the
628 * implementation is simply to verify. 678 * implementation is simply to verify.
629 * 679 *
630 * When allocating all pages normally 680 * When allocating all pages normally this algorithm will run
631 * in O(N) time, but in the worst case 681 * in O(N) time, but in the worst case it will run in O(N^2)
632 * time. If the runtime is a problem 682 * time. If the runtime is a problem the data structures can
633 * be fixed. 683 * be fixed.
634 */ 684 */
635 struct page *page; 685 struct page *page;
636 unsigned long addr; 686 unsigned long addr;
637 687
638 /* 688 /*
639 * Walk through the list of destinatio 689 * Walk through the list of destination pages, and see if I
640 * have a match. 690 * have a match.
641 */ 691 */
642 list_for_each_entry(page, &image->dest 692 list_for_each_entry(page, &image->dest_pages, lru) {
643 addr = page_to_boot_pfn(page) 693 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
644 if (addr == destination) { 694 if (addr == destination) {
645 list_del(&page->lru); 695 list_del(&page->lru);
646 return page; 696 return page;
647 } 697 }
648 } 698 }
649 page = NULL; 699 page = NULL;
650 while (1) { 700 while (1) {
651 kimage_entry_t *old; 701 kimage_entry_t *old;
652 702
653 /* Allocate a page, if we run 703 /* Allocate a page, if we run out of memory give up */
654 page = kimage_alloc_pages(gfp_ 704 page = kimage_alloc_pages(gfp_mask, 0);
655 if (!page) 705 if (!page)
656 return NULL; 706 return NULL;
657 /* If the page cannot be used 707 /* If the page cannot be used file it away */
658 if (page_to_boot_pfn(page) > 708 if (page_to_boot_pfn(page) >
659 (KEXEC_SOURCE_ 709 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
660 list_add(&page->lru, & 710 list_add(&page->lru, &image->unusable_pages);
661 continue; 711 continue;
662 } 712 }
663 addr = page_to_boot_pfn(page) 713 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
664 714
665 /* If it is the destination pa 715 /* If it is the destination page we want use it */
666 if (addr == destination) 716 if (addr == destination)
667 break; 717 break;
668 718
669 /* If the page is not a destin 719 /* If the page is not a destination page use it */
670 if (!kimage_is_destination_ran 720 if (!kimage_is_destination_range(image, addr,
671 721 addr + PAGE_SIZE - 1))
672 break; 722 break;
673 723
674 /* 724 /*
675 * I know that the page is som 725 * I know that the page is someones destination page.
676 * See if there is already a s 726 * See if there is already a source page for this
677 * destination page. And if s 727 * destination page. And if so swap the source pages.
678 */ 728 */
679 old = kimage_dst_used(image, a 729 old = kimage_dst_used(image, addr);
680 if (old) { 730 if (old) {
681 /* If so move it */ 731 /* If so move it */
682 unsigned long old_addr 732 unsigned long old_addr;
683 struct page *old_page; 733 struct page *old_page;
684 734
685 old_addr = *old & PAGE 735 old_addr = *old & PAGE_MASK;
686 old_page = boot_pfn_to 736 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
687 copy_highpage(page, ol 737 copy_highpage(page, old_page);
688 *old = addr | (*old & 738 *old = addr | (*old & ~PAGE_MASK);
689 739
690 /* The old page I have 740 /* The old page I have found cannot be a
691 * destination page, s 741 * destination page, so return it if it's
692 * gfp_flags honor the 742 * gfp_flags honor the ones passed in.
693 */ 743 */
694 if (!(gfp_mask & __GFP 744 if (!(gfp_mask & __GFP_HIGHMEM) &&
695 PageHighMem(old_pa 745 PageHighMem(old_page)) {
696 kimage_free_pa 746 kimage_free_pages(old_page);
697 continue; 747 continue;
698 } 748 }
699 page = old_page; 749 page = old_page;
700 break; 750 break;
701 } 751 }
702 /* Place the page on the desti 752 /* Place the page on the destination list, to be used later */
703 list_add(&page->lru, &image->d 753 list_add(&page->lru, &image->dest_pages);
704 } 754 }
705 755
706 return page; 756 return page;
707 } 757 }
708 758
709 static int kimage_load_normal_segment(struct k 759 static int kimage_load_normal_segment(struct kimage *image,
710 struc 760 struct kexec_segment *segment)
711 { 761 {
712 unsigned long maddr; 762 unsigned long maddr;
713 size_t ubytes, mbytes; 763 size_t ubytes, mbytes;
714 int result; 764 int result;
715 unsigned char __user *buf = NULL; 765 unsigned char __user *buf = NULL;
716 unsigned char *kbuf = NULL; 766 unsigned char *kbuf = NULL;
717 767
718 if (image->file_mode) 768 if (image->file_mode)
719 kbuf = segment->kbuf; 769 kbuf = segment->kbuf;
720 else 770 else
721 buf = segment->buf; 771 buf = segment->buf;
722 ubytes = segment->bufsz; 772 ubytes = segment->bufsz;
723 mbytes = segment->memsz; 773 mbytes = segment->memsz;
724 maddr = segment->mem; 774 maddr = segment->mem;
725 775
726 result = kimage_set_destination(image, 776 result = kimage_set_destination(image, maddr);
727 if (result < 0) 777 if (result < 0)
728 goto out; 778 goto out;
729 779
730 while (mbytes) { 780 while (mbytes) {
731 struct page *page; 781 struct page *page;
732 char *ptr; 782 char *ptr;
733 size_t uchunk, mchunk; 783 size_t uchunk, mchunk;
734 784
735 page = kimage_alloc_page(image 785 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
736 if (!page) { 786 if (!page) {
737 result = -ENOMEM; 787 result = -ENOMEM;
738 goto out; 788 goto out;
739 } 789 }
740 result = kimage_add_page(image 790 result = kimage_add_page(image, page_to_boot_pfn(page)
741 791 << PAGE_SHIFT);
742 if (result < 0) 792 if (result < 0)
743 goto out; 793 goto out;
744 794
745 ptr = kmap_local_page(page); 795 ptr = kmap_local_page(page);
746 /* Start with a clear page */ 796 /* Start with a clear page */
747 clear_page(ptr); 797 clear_page(ptr);
748 ptr += maddr & ~PAGE_MASK; 798 ptr += maddr & ~PAGE_MASK;
749 mchunk = min_t(size_t, mbytes, 799 mchunk = min_t(size_t, mbytes,
750 PAGE_SIZE - (m 800 PAGE_SIZE - (maddr & ~PAGE_MASK));
751 uchunk = min(ubytes, mchunk); 801 uchunk = min(ubytes, mchunk);
752 802
753 if (uchunk) { !! 803 /* For file based kexec, source pages are in kernel memory */
754 /* For file based kexe !! 804 if (image->file_mode)
755 if (image->file_mode) !! 805 memcpy(ptr, kbuf, uchunk);
756 memcpy(ptr, kb !! 806 else
757 else !! 807 result = copy_from_user(ptr, buf, uchunk);
758 result = copy_ <<
759 ubytes -= uchunk; <<
760 if (image->file_mode) <<
761 kbuf += uchunk <<
762 else <<
763 buf += uchunk; <<
764 } <<
765 kunmap_local(ptr); 808 kunmap_local(ptr);
766 if (result) { 809 if (result) {
767 result = -EFAULT; 810 result = -EFAULT;
768 goto out; 811 goto out;
769 } 812 }
>> 813 ubytes -= uchunk;
770 maddr += mchunk; 814 maddr += mchunk;
>> 815 if (image->file_mode)
>> 816 kbuf += mchunk;
>> 817 else
>> 818 buf += mchunk;
771 mbytes -= mchunk; 819 mbytes -= mchunk;
772 820
773 cond_resched(); 821 cond_resched();
774 } 822 }
775 out: 823 out:
776 return result; 824 return result;
777 } 825 }
778 826
779 #ifdef CONFIG_CRASH_DUMP <<
780 static int kimage_load_crash_segment(struct ki 827 static int kimage_load_crash_segment(struct kimage *image,
781 struct 828 struct kexec_segment *segment)
782 { 829 {
783 /* For crash dumps kernels we simply c 830 /* For crash dumps kernels we simply copy the data from
784 * user space to it's destination. 831 * user space to it's destination.
785 * We do things a page at a time for t 832 * We do things a page at a time for the sake of kmap.
786 */ 833 */
787 unsigned long maddr; 834 unsigned long maddr;
788 size_t ubytes, mbytes; 835 size_t ubytes, mbytes;
789 int result; 836 int result;
790 unsigned char __user *buf = NULL; 837 unsigned char __user *buf = NULL;
791 unsigned char *kbuf = NULL; 838 unsigned char *kbuf = NULL;
792 839
793 result = 0; 840 result = 0;
794 if (image->file_mode) 841 if (image->file_mode)
795 kbuf = segment->kbuf; 842 kbuf = segment->kbuf;
796 else 843 else
797 buf = segment->buf; 844 buf = segment->buf;
798 ubytes = segment->bufsz; 845 ubytes = segment->bufsz;
799 mbytes = segment->memsz; 846 mbytes = segment->memsz;
800 maddr = segment->mem; 847 maddr = segment->mem;
801 while (mbytes) { 848 while (mbytes) {
802 struct page *page; 849 struct page *page;
803 char *ptr; 850 char *ptr;
804 size_t uchunk, mchunk; 851 size_t uchunk, mchunk;
805 852
806 page = boot_pfn_to_page(maddr 853 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
807 if (!page) { 854 if (!page) {
808 result = -ENOMEM; 855 result = -ENOMEM;
809 goto out; 856 goto out;
810 } 857 }
811 arch_kexec_post_alloc_pages(pa 858 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
812 ptr = kmap_local_page(page); 859 ptr = kmap_local_page(page);
813 ptr += maddr & ~PAGE_MASK; 860 ptr += maddr & ~PAGE_MASK;
814 mchunk = min_t(size_t, mbytes, 861 mchunk = min_t(size_t, mbytes,
815 PAGE_SIZE - (m 862 PAGE_SIZE - (maddr & ~PAGE_MASK));
816 uchunk = min(ubytes, mchunk); 863 uchunk = min(ubytes, mchunk);
817 if (mchunk > uchunk) { 864 if (mchunk > uchunk) {
818 /* Zero the trailing p 865 /* Zero the trailing part of the page */
819 memset(ptr + uchunk, 0 866 memset(ptr + uchunk, 0, mchunk - uchunk);
820 } 867 }
821 868
822 if (uchunk) { !! 869 /* For file based kexec, source pages are in kernel memory */
823 /* For file based kexe !! 870 if (image->file_mode)
824 if (image->file_mode) !! 871 memcpy(ptr, kbuf, uchunk);
825 memcpy(ptr, kb !! 872 else
826 else !! 873 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); 874 kexec_flush_icache_page(page);
835 kunmap_local(ptr); 875 kunmap_local(ptr);
836 arch_kexec_pre_free_pages(page 876 arch_kexec_pre_free_pages(page_address(page), 1);
837 if (result) { 877 if (result) {
838 result = -EFAULT; 878 result = -EFAULT;
839 goto out; 879 goto out;
840 } 880 }
>> 881 ubytes -= uchunk;
841 maddr += mchunk; 882 maddr += mchunk;
>> 883 if (image->file_mode)
>> 884 kbuf += mchunk;
>> 885 else
>> 886 buf += mchunk;
842 mbytes -= mchunk; 887 mbytes -= mchunk;
843 888
844 cond_resched(); 889 cond_resched();
845 } 890 }
846 out: 891 out:
847 return result; 892 return result;
848 } 893 }
849 #endif <<
850 894
851 int kimage_load_segment(struct kimage *image, 895 int kimage_load_segment(struct kimage *image,
852 struct kexec_s 896 struct kexec_segment *segment)
853 { 897 {
854 int result = -ENOMEM; 898 int result = -ENOMEM;
855 899
856 switch (image->type) { 900 switch (image->type) {
857 case KEXEC_TYPE_DEFAULT: 901 case KEXEC_TYPE_DEFAULT:
858 result = kimage_load_normal_se 902 result = kimage_load_normal_segment(image, segment);
859 break; 903 break;
860 #ifdef CONFIG_CRASH_DUMP <<
861 case KEXEC_TYPE_CRASH: 904 case KEXEC_TYPE_CRASH:
862 result = kimage_load_crash_seg 905 result = kimage_load_crash_segment(image, segment);
863 break; 906 break;
864 #endif <<
865 } 907 }
866 908
867 return result; 909 return result;
868 } 910 }
869 911
870 struct kexec_load_limit { 912 struct kexec_load_limit {
871 /* Mutex protects the limit count. */ 913 /* Mutex protects the limit count. */
872 struct mutex mutex; 914 struct mutex mutex;
873 int limit; 915 int limit;
874 }; 916 };
875 917
876 static struct kexec_load_limit load_limit_rebo 918 static struct kexec_load_limit load_limit_reboot = {
877 .mutex = __MUTEX_INITIALIZER(load_limi 919 .mutex = __MUTEX_INITIALIZER(load_limit_reboot.mutex),
878 .limit = -1, 920 .limit = -1,
879 }; 921 };
880 922
881 static struct kexec_load_limit load_limit_pani 923 static struct kexec_load_limit load_limit_panic = {
882 .mutex = __MUTEX_INITIALIZER(load_limi 924 .mutex = __MUTEX_INITIALIZER(load_limit_panic.mutex),
883 .limit = -1, 925 .limit = -1,
884 }; 926 };
885 927
886 struct kimage *kexec_image; 928 struct kimage *kexec_image;
887 struct kimage *kexec_crash_image; 929 struct kimage *kexec_crash_image;
888 static int kexec_load_disabled; 930 static int kexec_load_disabled;
889 931
890 #ifdef CONFIG_SYSCTL 932 #ifdef CONFIG_SYSCTL
891 static int kexec_limit_handler(const struct ct !! 933 static int kexec_limit_handler(struct ctl_table *table, int write,
892 void *buffer, s 934 void *buffer, size_t *lenp, loff_t *ppos)
893 { 935 {
894 struct kexec_load_limit *limit = table 936 struct kexec_load_limit *limit = table->data;
895 int val; 937 int val;
896 struct ctl_table tmp = { 938 struct ctl_table tmp = {
897 .data = &val, 939 .data = &val,
898 .maxlen = sizeof(val), 940 .maxlen = sizeof(val),
899 .mode = table->mode, 941 .mode = table->mode,
900 }; 942 };
901 int ret; 943 int ret;
902 944
903 if (write) { 945 if (write) {
904 ret = proc_dointvec(&tmp, writ 946 ret = proc_dointvec(&tmp, write, buffer, lenp, ppos);
905 if (ret) 947 if (ret)
906 return ret; 948 return ret;
907 949
908 if (val < 0) 950 if (val < 0)
909 return -EINVAL; 951 return -EINVAL;
910 952
911 mutex_lock(&limit->mutex); 953 mutex_lock(&limit->mutex);
912 if (limit->limit != -1 && val 954 if (limit->limit != -1 && val >= limit->limit)
913 ret = -EINVAL; 955 ret = -EINVAL;
914 else 956 else
915 limit->limit = val; 957 limit->limit = val;
916 mutex_unlock(&limit->mutex); 958 mutex_unlock(&limit->mutex);
917 959
918 return ret; 960 return ret;
919 } 961 }
920 962
921 mutex_lock(&limit->mutex); 963 mutex_lock(&limit->mutex);
922 val = limit->limit; 964 val = limit->limit;
923 mutex_unlock(&limit->mutex); 965 mutex_unlock(&limit->mutex);
924 966
925 return proc_dointvec(&tmp, write, buff 967 return proc_dointvec(&tmp, write, buffer, lenp, ppos);
926 } 968 }
927 969
928 static struct ctl_table kexec_core_sysctls[] = 970 static struct ctl_table kexec_core_sysctls[] = {
929 { 971 {
930 .procname = "kexec_load_ 972 .procname = "kexec_load_disabled",
931 .data = &kexec_load_ 973 .data = &kexec_load_disabled,
932 .maxlen = sizeof(int), 974 .maxlen = sizeof(int),
933 .mode = 0644, 975 .mode = 0644,
934 /* only handle a transition fr 976 /* only handle a transition from default "" to "1" */
935 .proc_handler = proc_dointve 977 .proc_handler = proc_dointvec_minmax,
936 .extra1 = SYSCTL_ONE, 978 .extra1 = SYSCTL_ONE,
937 .extra2 = SYSCTL_ONE, 979 .extra2 = SYSCTL_ONE,
938 }, 980 },
939 { 981 {
940 .procname = "kexec_load_ 982 .procname = "kexec_load_limit_panic",
941 .data = &load_limit_ 983 .data = &load_limit_panic,
942 .mode = 0644, 984 .mode = 0644,
943 .proc_handler = kexec_limit_ 985 .proc_handler = kexec_limit_handler,
944 }, 986 },
945 { 987 {
946 .procname = "kexec_load_ 988 .procname = "kexec_load_limit_reboot",
947 .data = &load_limit_ 989 .data = &load_limit_reboot,
948 .mode = 0644, 990 .mode = 0644,
949 .proc_handler = kexec_limit_ 991 .proc_handler = kexec_limit_handler,
950 }, 992 },
>> 993 { }
951 }; 994 };
952 995
953 static int __init kexec_core_sysctl_init(void) 996 static int __init kexec_core_sysctl_init(void)
954 { 997 {
955 register_sysctl_init("kernel", kexec_c 998 register_sysctl_init("kernel", kexec_core_sysctls);
956 return 0; 999 return 0;
957 } 1000 }
958 late_initcall(kexec_core_sysctl_init); 1001 late_initcall(kexec_core_sysctl_init);
959 #endif 1002 #endif
960 1003
961 bool kexec_load_permitted(int kexec_image_type 1004 bool kexec_load_permitted(int kexec_image_type)
962 { 1005 {
963 struct kexec_load_limit *limit; 1006 struct kexec_load_limit *limit;
964 1007
965 /* 1008 /*
966 * Only the superuser can use the kexe 1009 * Only the superuser can use the kexec syscall and if it has not
967 * been disabled. 1010 * been disabled.
968 */ 1011 */
969 if (!capable(CAP_SYS_BOOT) || kexec_lo 1012 if (!capable(CAP_SYS_BOOT) || kexec_load_disabled)
970 return false; 1013 return false;
971 1014
972 /* Check limit counter and decrease it 1015 /* Check limit counter and decrease it.*/
973 limit = (kexec_image_type == KEXEC_TYP 1016 limit = (kexec_image_type == KEXEC_TYPE_CRASH) ?
974 &load_limit_panic : &load_limi 1017 &load_limit_panic : &load_limit_reboot;
975 mutex_lock(&limit->mutex); 1018 mutex_lock(&limit->mutex);
976 if (!limit->limit) { 1019 if (!limit->limit) {
977 mutex_unlock(&limit->mutex); 1020 mutex_unlock(&limit->mutex);
978 return false; 1021 return false;
979 } 1022 }
980 if (limit->limit != -1) 1023 if (limit->limit != -1)
981 limit->limit--; 1024 limit->limit--;
982 mutex_unlock(&limit->mutex); 1025 mutex_unlock(&limit->mutex);
983 1026
984 return true; 1027 return true;
>> 1028 }
>> 1029
>> 1030 /*
>> 1031 * No panic_cpu check version of crash_kexec(). This function is called
>> 1032 * only when panic_cpu holds the current CPU number; this is the only CPU
>> 1033 * which processes crash_kexec routines.
>> 1034 */
>> 1035 void __noclone __crash_kexec(struct pt_regs *regs)
>> 1036 {
>> 1037 /* Take the kexec_lock here to prevent sys_kexec_load
>> 1038 * running on one cpu from replacing the crash kernel
>> 1039 * we are using after a panic on a different cpu.
>> 1040 *
>> 1041 * If the crash kernel was not located in a fixed area
>> 1042 * of memory the xchg(&kexec_crash_image) would be
>> 1043 * sufficient. But since I reuse the memory...
>> 1044 */
>> 1045 if (kexec_trylock()) {
>> 1046 if (kexec_crash_image) {
>> 1047 struct pt_regs fixed_regs;
>> 1048
>> 1049 crash_setup_regs(&fixed_regs, regs);
>> 1050 crash_save_vmcoreinfo();
>> 1051 machine_crash_shutdown(&fixed_regs);
>> 1052 machine_kexec(kexec_crash_image);
>> 1053 }
>> 1054 kexec_unlock();
>> 1055 }
>> 1056 }
>> 1057 STACK_FRAME_NON_STANDARD(__crash_kexec);
>> 1058
>> 1059 __bpf_kfunc void crash_kexec(struct pt_regs *regs)
>> 1060 {
>> 1061 int old_cpu, this_cpu;
>> 1062
>> 1063 /*
>> 1064 * Only one CPU is allowed to execute the crash_kexec() code as with
>> 1065 * panic(). Otherwise parallel calls of panic() and crash_kexec()
>> 1066 * may stop each other. To exclude them, we use panic_cpu here too.
>> 1067 */
>> 1068 old_cpu = PANIC_CPU_INVALID;
>> 1069 this_cpu = raw_smp_processor_id();
>> 1070
>> 1071 if (atomic_try_cmpxchg(&panic_cpu, &old_cpu, this_cpu)) {
>> 1072 /* This is the 1st CPU which comes here, so go ahead. */
>> 1073 __crash_kexec(regs);
>> 1074
>> 1075 /*
>> 1076 * Reset panic_cpu to allow another panic()/crash_kexec()
>> 1077 * call.
>> 1078 */
>> 1079 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
>> 1080 }
>> 1081 }
>> 1082
>> 1083 static inline resource_size_t crash_resource_size(const struct resource *res)
>> 1084 {
>> 1085 return !res->end ? 0 : resource_size(res);
>> 1086 }
>> 1087
>> 1088 ssize_t crash_get_memory_size(void)
>> 1089 {
>> 1090 ssize_t size = 0;
>> 1091
>> 1092 if (!kexec_trylock())
>> 1093 return -EBUSY;
>> 1094
>> 1095 size += crash_resource_size(&crashk_res);
>> 1096 size += crash_resource_size(&crashk_low_res);
>> 1097
>> 1098 kexec_unlock();
>> 1099 return size;
>> 1100 }
>> 1101
>> 1102 static int __crash_shrink_memory(struct resource *old_res,
>> 1103 unsigned long new_size)
>> 1104 {
>> 1105 struct resource *ram_res;
>> 1106
>> 1107 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
>> 1108 if (!ram_res)
>> 1109 return -ENOMEM;
>> 1110
>> 1111 ram_res->start = old_res->start + new_size;
>> 1112 ram_res->end = old_res->end;
>> 1113 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
>> 1114 ram_res->name = "System RAM";
>> 1115
>> 1116 if (!new_size) {
>> 1117 release_resource(old_res);
>> 1118 old_res->start = 0;
>> 1119 old_res->end = 0;
>> 1120 } else {
>> 1121 crashk_res.end = ram_res->start - 1;
>> 1122 }
>> 1123
>> 1124 crash_free_reserved_phys_range(ram_res->start, ram_res->end);
>> 1125 insert_resource(&iomem_resource, ram_res);
>> 1126
>> 1127 return 0;
>> 1128 }
>> 1129
>> 1130 int crash_shrink_memory(unsigned long new_size)
>> 1131 {
>> 1132 int ret = 0;
>> 1133 unsigned long old_size, low_size;
>> 1134
>> 1135 if (!kexec_trylock())
>> 1136 return -EBUSY;
>> 1137
>> 1138 if (kexec_crash_image) {
>> 1139 ret = -ENOENT;
>> 1140 goto unlock;
>> 1141 }
>> 1142
>> 1143 low_size = crash_resource_size(&crashk_low_res);
>> 1144 old_size = crash_resource_size(&crashk_res) + low_size;
>> 1145 new_size = roundup(new_size, KEXEC_CRASH_MEM_ALIGN);
>> 1146 if (new_size >= old_size) {
>> 1147 ret = (new_size == old_size) ? 0 : -EINVAL;
>> 1148 goto unlock;
>> 1149 }
>> 1150
>> 1151 /*
>> 1152 * (low_size > new_size) implies that low_size is greater than zero.
>> 1153 * This also means that if low_size is zero, the else branch is taken.
>> 1154 *
>> 1155 * If low_size is greater than 0, (low_size > new_size) indicates that
>> 1156 * crashk_low_res also needs to be shrunken. Otherwise, only crashk_res
>> 1157 * needs to be shrunken.
>> 1158 */
>> 1159 if (low_size > new_size) {
>> 1160 ret = __crash_shrink_memory(&crashk_res, 0);
>> 1161 if (ret)
>> 1162 goto unlock;
>> 1163
>> 1164 ret = __crash_shrink_memory(&crashk_low_res, new_size);
>> 1165 } else {
>> 1166 ret = __crash_shrink_memory(&crashk_res, new_size - low_size);
>> 1167 }
>> 1168
>> 1169 /* Swap crashk_res and crashk_low_res if needed */
>> 1170 if (!crashk_res.end && crashk_low_res.end) {
>> 1171 crashk_res.start = crashk_low_res.start;
>> 1172 crashk_res.end = crashk_low_res.end;
>> 1173 release_resource(&crashk_low_res);
>> 1174 crashk_low_res.start = 0;
>> 1175 crashk_low_res.end = 0;
>> 1176 insert_resource(&iomem_resource, &crashk_res);
>> 1177 }
>> 1178
>> 1179 unlock:
>> 1180 kexec_unlock();
>> 1181 return ret;
>> 1182 }
>> 1183
>> 1184 void crash_save_cpu(struct pt_regs *regs, int cpu)
>> 1185 {
>> 1186 struct elf_prstatus prstatus;
>> 1187 u32 *buf;
>> 1188
>> 1189 if ((cpu < 0) || (cpu >= nr_cpu_ids))
>> 1190 return;
>> 1191
>> 1192 /* Using ELF notes here is opportunistic.
>> 1193 * I need a well defined structure format
>> 1194 * for the data I pass, and I need tags
>> 1195 * on the data to indicate what information I have
>> 1196 * squirrelled away. ELF notes happen to provide
>> 1197 * all of that, so there is no need to invent something new.
>> 1198 */
>> 1199 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
>> 1200 if (!buf)
>> 1201 return;
>> 1202 memset(&prstatus, 0, sizeof(prstatus));
>> 1203 prstatus.common.pr_pid = current->pid;
>> 1204 elf_core_copy_regs(&prstatus.pr_reg, regs);
>> 1205 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
>> 1206 &prstatus, sizeof(prstatus));
>> 1207 final_note(buf);
985 } 1208 }
986 1209
987 /* 1210 /*
988 * Move into place and start executing a prelo 1211 * Move into place and start executing a preloaded standalone
989 * executable. If nothing was preloaded retur 1212 * executable. If nothing was preloaded return an error.
990 */ 1213 */
991 int kernel_kexec(void) 1214 int kernel_kexec(void)
992 { 1215 {
993 int error = 0; 1216 int error = 0;
994 1217
995 if (!kexec_trylock()) 1218 if (!kexec_trylock())
996 return -EBUSY; 1219 return -EBUSY;
997 if (!kexec_image) { 1220 if (!kexec_image) {
998 error = -EINVAL; 1221 error = -EINVAL;
999 goto Unlock; 1222 goto Unlock;
1000 } 1223 }
1001 1224
1002 #ifdef CONFIG_KEXEC_JUMP 1225 #ifdef CONFIG_KEXEC_JUMP
1003 if (kexec_image->preserve_context) { 1226 if (kexec_image->preserve_context) {
1004 pm_prepare_console(); 1227 pm_prepare_console();
1005 error = freeze_processes(); 1228 error = freeze_processes();
1006 if (error) { 1229 if (error) {
1007 error = -EBUSY; 1230 error = -EBUSY;
1008 goto Restore_console; 1231 goto Restore_console;
1009 } 1232 }
1010 suspend_console(); 1233 suspend_console();
1011 error = dpm_suspend_start(PMS 1234 error = dpm_suspend_start(PMSG_FREEZE);
1012 if (error) 1235 if (error)
1013 goto Resume_console; 1236 goto Resume_console;
1014 /* At this point, dpm_suspend 1237 /* At this point, dpm_suspend_start() has been called,
1015 * but *not* dpm_suspend_end( 1238 * but *not* dpm_suspend_end(). We *must* call
1016 * dpm_suspend_end() now. Ot 1239 * dpm_suspend_end() now. Otherwise, drivers for
1017 * some devices (e.g. interru 1240 * some devices (e.g. interrupt controllers) become
1018 * desynchronized with the ac 1241 * desynchronized with the actual state of the
1019 * hardware at resume time, a 1242 * hardware at resume time, and evil weirdness ensues.
1020 */ 1243 */
1021 error = dpm_suspend_end(PMSG_ 1244 error = dpm_suspend_end(PMSG_FREEZE);
1022 if (error) 1245 if (error)
1023 goto Resume_devices; 1246 goto Resume_devices;
1024 error = suspend_disable_secon 1247 error = suspend_disable_secondary_cpus();
1025 if (error) 1248 if (error)
1026 goto Enable_cpus; 1249 goto Enable_cpus;
1027 local_irq_disable(); 1250 local_irq_disable();
1028 error = syscore_suspend(); 1251 error = syscore_suspend();
1029 if (error) 1252 if (error)
1030 goto Enable_irqs; 1253 goto Enable_irqs;
1031 } else 1254 } else
1032 #endif 1255 #endif
1033 { 1256 {
1034 kexec_in_progress = true; 1257 kexec_in_progress = true;
1035 kernel_restart_prepare("kexec 1258 kernel_restart_prepare("kexec reboot");
1036 migrate_to_reboot_cpu(); 1259 migrate_to_reboot_cpu();
1037 syscore_shutdown(); 1260 syscore_shutdown();
1038 1261
1039 /* 1262 /*
1040 * migrate_to_reboot_cpu() di 1263 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1041 * no further code needs to u 1264 * no further code needs to use CPU hotplug (which is true in
1042 * the reboot case). However, 1265 * the reboot case). However, the kexec path depends on using
1043 * CPU hotplug again; so re-e 1266 * CPU hotplug again; so re-enable it here.
1044 */ 1267 */
1045 cpu_hotplug_enable(); 1268 cpu_hotplug_enable();
1046 pr_notice("Starting new kerne 1269 pr_notice("Starting new kernel\n");
1047 machine_shutdown(); 1270 machine_shutdown();
1048 } 1271 }
1049 1272
1050 kmsg_dump(KMSG_DUMP_SHUTDOWN); 1273 kmsg_dump(KMSG_DUMP_SHUTDOWN);
1051 machine_kexec(kexec_image); 1274 machine_kexec(kexec_image);
1052 1275
1053 #ifdef CONFIG_KEXEC_JUMP 1276 #ifdef CONFIG_KEXEC_JUMP
1054 if (kexec_image->preserve_context) { 1277 if (kexec_image->preserve_context) {
1055 syscore_resume(); 1278 syscore_resume();
1056 Enable_irqs: 1279 Enable_irqs:
1057 local_irq_enable(); 1280 local_irq_enable();
1058 Enable_cpus: 1281 Enable_cpus:
1059 suspend_enable_secondary_cpus 1282 suspend_enable_secondary_cpus();
1060 dpm_resume_start(PMSG_RESTORE 1283 dpm_resume_start(PMSG_RESTORE);
1061 Resume_devices: 1284 Resume_devices:
1062 dpm_resume_end(PMSG_RESTORE); 1285 dpm_resume_end(PMSG_RESTORE);
1063 Resume_console: 1286 Resume_console:
1064 resume_console(); 1287 resume_console();
1065 thaw_processes(); 1288 thaw_processes();
1066 Restore_console: 1289 Restore_console:
1067 pm_restore_console(); 1290 pm_restore_console();
1068 } 1291 }
1069 #endif 1292 #endif
1070 1293
1071 Unlock: 1294 Unlock:
1072 kexec_unlock(); 1295 kexec_unlock();
1073 return error; 1296 return error;
1074 } 1297 }
1075 1298
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.