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> <<
10 #include <linux/capability.h> 9 #include <linux/capability.h>
11 #include <linux/mm.h> 10 #include <linux/mm.h>
12 #include <linux/file.h> 11 #include <linux/file.h>
13 #include <linux/slab.h> 12 #include <linux/slab.h>
14 #include <linux/fs.h> 13 #include <linux/fs.h>
15 #include <linux/kexec.h> 14 #include <linux/kexec.h>
16 #include <linux/mutex.h> 15 #include <linux/mutex.h>
17 #include <linux/list.h> 16 #include <linux/list.h>
18 #include <linux/highmem.h> 17 #include <linux/highmem.h>
19 #include <linux/syscalls.h> 18 #include <linux/syscalls.h>
20 #include <linux/reboot.h> 19 #include <linux/reboot.h>
21 #include <linux/ioport.h> 20 #include <linux/ioport.h>
22 #include <linux/hardirq.h> 21 #include <linux/hardirq.h>
23 #include <linux/elf.h> 22 #include <linux/elf.h>
24 #include <linux/elfcore.h> 23 #include <linux/elfcore.h>
25 #include <linux/utsname.h> 24 #include <linux/utsname.h>
26 #include <linux/numa.h> 25 #include <linux/numa.h>
27 #include <linux/suspend.h> 26 #include <linux/suspend.h>
28 #include <linux/device.h> 27 #include <linux/device.h>
29 #include <linux/freezer.h> 28 #include <linux/freezer.h>
30 #include <linux/panic_notifier.h> 29 #include <linux/panic_notifier.h>
31 #include <linux/pm.h> 30 #include <linux/pm.h>
32 #include <linux/cpu.h> 31 #include <linux/cpu.h>
33 #include <linux/uaccess.h> 32 #include <linux/uaccess.h>
34 #include <linux/io.h> 33 #include <linux/io.h>
35 #include <linux/console.h> 34 #include <linux/console.h>
36 #include <linux/vmalloc.h> 35 #include <linux/vmalloc.h>
37 #include <linux/swap.h> 36 #include <linux/swap.h>
38 #include <linux/syscore_ops.h> 37 #include <linux/syscore_ops.h>
39 #include <linux/compiler.h> 38 #include <linux/compiler.h>
40 #include <linux/hugetlb.h> 39 #include <linux/hugetlb.h>
41 #include <linux/objtool.h> 40 #include <linux/objtool.h>
42 #include <linux/kmsg_dump.h> 41 #include <linux/kmsg_dump.h>
43 42
44 #include <asm/page.h> 43 #include <asm/page.h>
45 #include <asm/sections.h> 44 #include <asm/sections.h>
46 45
47 #include <crypto/hash.h> 46 #include <crypto/hash.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;
51 53
52 /* Flag to indicate we are going to kexec a ne 54 /* Flag to indicate we are going to kexec a new kernel */
53 bool kexec_in_progress = false; 55 bool kexec_in_progress = false;
54 56
55 bool kexec_file_dbg_print; !! 57
>> 58 /* Location of the reserved area for the crash kernel */
>> 59 struct resource crashk_res = {
>> 60 .name = "Crash kernel",
>> 61 .start = 0,
>> 62 .end = 0,
>> 63 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
>> 64 .desc = IORES_DESC_CRASH_KERNEL
>> 65 };
>> 66 struct resource crashk_low_res = {
>> 67 .name = "Crash kernel",
>> 68 .start = 0,
>> 69 .end = 0,
>> 70 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM,
>> 71 .desc = IORES_DESC_CRASH_KERNEL
>> 72 };
>> 73
>> 74 int kexec_should_crash(struct task_struct *p)
>> 75 {
>> 76 /*
>> 77 * If crash_kexec_post_notifiers is enabled, don't run
>> 78 * crash_kexec() here yet, which must be run after panic
>> 79 * notifiers in panic().
>> 80 */
>> 81 if (crash_kexec_post_notifiers)
>> 82 return 0;
>> 83 /*
>> 84 * There are 4 panic() calls in make_task_dead() path, each of which
>> 85 * corresponds to each of these 4 conditions.
>> 86 */
>> 87 if (in_interrupt() || !p->pid || is_global_init(p) || panic_on_oops)
>> 88 return 1;
>> 89 return 0;
>> 90 }
>> 91
>> 92 int kexec_crash_loaded(void)
>> 93 {
>> 94 return !!kexec_crash_image;
>> 95 }
>> 96 EXPORT_SYMBOL_GPL(kexec_crash_loaded);
56 97
57 /* 98 /*
58 * When kexec transitions to the new kernel th 99 * When kexec transitions to the new kernel there is a one-to-one
59 * mapping between physical and virtual addres 100 * mapping between physical and virtual addresses. On processors
60 * where you can disable the MMU this is trivi 101 * where you can disable the MMU this is trivial, and easy. For
61 * others it is still a simple predictable pag 102 * others it is still a simple predictable page table to setup.
62 * 103 *
63 * In that environment kexec copies the new ke 104 * In that environment kexec copies the new kernel to its final
64 * resting place. This means I can only suppo 105 * resting place. This means I can only support memory whose
65 * physical address can fit in an unsigned lon 106 * physical address can fit in an unsigned long. In particular
66 * addresses where (pfn << PAGE_SHIFT) > ULONG 107 * addresses where (pfn << PAGE_SHIFT) > ULONG_MAX cannot be handled.
67 * If the assembly stub has more restrictive r 108 * If the assembly stub has more restrictive requirements
68 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_ME 109 * KEXEC_SOURCE_MEMORY_LIMIT and KEXEC_DEST_MEMORY_LIMIT can be
69 * defined more restrictively in <asm/kexec.h> 110 * defined more restrictively in <asm/kexec.h>.
70 * 111 *
71 * The code for the transition from the curren 112 * The code for the transition from the current kernel to the
72 * new kernel is placed in the control_code_bu 113 * new kernel is placed in the control_code_buffer, whose size
73 * is given by KEXEC_CONTROL_PAGE_SIZE. In th 114 * is given by KEXEC_CONTROL_PAGE_SIZE. In the best case only a single
74 * page of memory is necessary, but some archi 115 * page of memory is necessary, but some architectures require more.
75 * Because this memory must be identity mapped 116 * Because this memory must be identity mapped in the transition from
76 * virtual to physical addresses it must live 117 * virtual to physical addresses it must live in the range
77 * 0 - TASK_SIZE, as only the user space mappi 118 * 0 - TASK_SIZE, as only the user space mappings are arbitrarily
78 * modifiable. 119 * modifiable.
79 * 120 *
80 * The assembly stub in the control code buffe 121 * The assembly stub in the control code buffer is passed a linked list
81 * of descriptor pages detailing the source pa 122 * of descriptor pages detailing the source pages of the new kernel,
82 * and the destination addresses of those sour 123 * and the destination addresses of those source pages. As this data
83 * structure is not used in the context of the 124 * structure is not used in the context of the current OS, it must
84 * be self-contained. 125 * be self-contained.
85 * 126 *
86 * The code has been made to work with highmem 127 * The code has been made to work with highmem pages and will use a
87 * destination page in its final resting place 128 * destination page in its final resting place (if it happens
88 * to allocate it). The end product of this i 129 * to allocate it). The end product of this is that most of the
89 * physical address space, and most of RAM can 130 * physical address space, and most of RAM can be used.
90 * 131 *
91 * Future directions include: 132 * Future directions include:
92 * - allocating a page table with the control 133 * - allocating a page table with the control code buffer identity
93 * mapped, to simplify machine_kexec and ma 134 * mapped, to simplify machine_kexec and make kexec_on_panic more
94 * reliable. 135 * reliable.
95 */ 136 */
96 137
97 /* 138 /*
98 * KIMAGE_NO_DEST is an impossible destination 139 * KIMAGE_NO_DEST is an impossible destination address..., for
99 * allocating pages whose destination address 140 * allocating pages whose destination address we do not care about.
100 */ 141 */
101 #define KIMAGE_NO_DEST (-1UL) 142 #define KIMAGE_NO_DEST (-1UL)
102 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) > 143 #define PAGE_COUNT(x) (((x) + PAGE_SIZE - 1) >> PAGE_SHIFT)
103 144
104 static struct page *kimage_alloc_page(struct k 145 static struct page *kimage_alloc_page(struct kimage *image,
105 gfp_t g 146 gfp_t gfp_mask,
106 unsigne 147 unsigned long dest);
107 148
108 int sanity_check_segment_list(struct kimage *i 149 int sanity_check_segment_list(struct kimage *image)
109 { 150 {
110 int i; 151 int i;
111 unsigned long nr_segments = image->nr_ 152 unsigned long nr_segments = image->nr_segments;
112 unsigned long total_pages = 0; 153 unsigned long total_pages = 0;
113 unsigned long nr_pages = totalram_page 154 unsigned long nr_pages = totalram_pages();
114 155
115 /* 156 /*
116 * Verify we have good destination add 157 * Verify we have good destination addresses. The caller is
117 * responsible for making certain we d 158 * responsible for making certain we don't attempt to load
118 * the new image into invalid or reser 159 * the new image into invalid or reserved areas of RAM. This
119 * just verifies it is an address we c 160 * just verifies it is an address we can use.
120 * 161 *
121 * Since the kernel does everything in 162 * Since the kernel does everything in page size chunks ensure
122 * the destination addresses are page 163 * the destination addresses are page aligned. Too many
123 * special cases crop of when we don't 164 * special cases crop of when we don't do this. The most
124 * insidious is getting overlapping de 165 * insidious is getting overlapping destination addresses
125 * simply because addresses are change 166 * simply because addresses are changed to page size
126 * granularity. 167 * granularity.
127 */ 168 */
128 for (i = 0; i < nr_segments; i++) { 169 for (i = 0; i < nr_segments; i++) {
129 unsigned long mstart, mend; 170 unsigned long mstart, mend;
130 171
131 mstart = image->segment[i].mem 172 mstart = image->segment[i].mem;
132 mend = mstart + image->segme 173 mend = mstart + image->segment[i].memsz;
133 if (mstart > mend) 174 if (mstart > mend)
134 return -EADDRNOTAVAIL; 175 return -EADDRNOTAVAIL;
135 if ((mstart & ~PAGE_MASK) || ( 176 if ((mstart & ~PAGE_MASK) || (mend & ~PAGE_MASK))
136 return -EADDRNOTAVAIL; 177 return -EADDRNOTAVAIL;
137 if (mend >= KEXEC_DESTINATION_ 178 if (mend >= KEXEC_DESTINATION_MEMORY_LIMIT)
138 return -EADDRNOTAVAIL; 179 return -EADDRNOTAVAIL;
139 } 180 }
140 181
141 /* Verify our destination addresses do 182 /* Verify our destination addresses do not overlap.
142 * If we alloed overlapping destinatio 183 * If we alloed overlapping destination addresses
143 * through very weird things can happe 184 * through very weird things can happen with no
144 * easy explanation as one segment sto 185 * easy explanation as one segment stops on another.
145 */ 186 */
146 for (i = 0; i < nr_segments; i++) { 187 for (i = 0; i < nr_segments; i++) {
147 unsigned long mstart, mend; 188 unsigned long mstart, mend;
148 unsigned long j; 189 unsigned long j;
149 190
150 mstart = image->segment[i].mem 191 mstart = image->segment[i].mem;
151 mend = mstart + image->segme 192 mend = mstart + image->segment[i].memsz;
152 for (j = 0; j < i; j++) { 193 for (j = 0; j < i; j++) {
153 unsigned long pstart, 194 unsigned long pstart, pend;
154 195
155 pstart = image->segmen 196 pstart = image->segment[j].mem;
156 pend = pstart + imag 197 pend = pstart + image->segment[j].memsz;
157 /* Do the segments ove 198 /* Do the segments overlap ? */
158 if ((mend > pstart) && 199 if ((mend > pstart) && (mstart < pend))
159 return -EINVAL 200 return -EINVAL;
160 } 201 }
161 } 202 }
162 203
163 /* Ensure our buffer sizes are strictl 204 /* Ensure our buffer sizes are strictly less than
164 * our memory sizes. This should alwa 205 * our memory sizes. This should always be the case,
165 * and it is easier to check up front 206 * and it is easier to check up front than to be surprised
166 * later on. 207 * later on.
167 */ 208 */
168 for (i = 0; i < nr_segments; i++) { 209 for (i = 0; i < nr_segments; i++) {
169 if (image->segment[i].bufsz > 210 if (image->segment[i].bufsz > image->segment[i].memsz)
170 return -EINVAL; 211 return -EINVAL;
171 } 212 }
172 213
173 /* 214 /*
174 * Verify that no more than half of me 215 * Verify that no more than half of memory will be consumed. If the
175 * request from userspace is too large 216 * request from userspace is too large, a large amount of time will be
176 * wasted allocating pages, which can 217 * wasted allocating pages, which can cause a soft lockup.
177 */ 218 */
178 for (i = 0; i < nr_segments; i++) { 219 for (i = 0; i < nr_segments; i++) {
179 if (PAGE_COUNT(image->segment[ 220 if (PAGE_COUNT(image->segment[i].memsz) > nr_pages / 2)
180 return -EINVAL; 221 return -EINVAL;
181 222
182 total_pages += PAGE_COUNT(imag 223 total_pages += PAGE_COUNT(image->segment[i].memsz);
183 } 224 }
184 225
185 if (total_pages > nr_pages / 2) 226 if (total_pages > nr_pages / 2)
186 return -EINVAL; 227 return -EINVAL;
187 228
188 #ifdef CONFIG_CRASH_DUMP <<
189 /* 229 /*
190 * Verify we have good destination add 230 * Verify we have good destination addresses. Normally
191 * the caller is responsible for makin 231 * the caller is responsible for making certain we don't
192 * attempt to load the new image into 232 * attempt to load the new image into invalid or reserved
193 * areas of RAM. But crash kernels ar 233 * areas of RAM. But crash kernels are preloaded into a
194 * reserved area of ram. We must ensu 234 * reserved area of ram. We must ensure the addresses
195 * are in the reserved area otherwise 235 * are in the reserved area otherwise preloading the
196 * kernel could corrupt things. 236 * kernel could corrupt things.
197 */ 237 */
198 238
199 if (image->type == KEXEC_TYPE_CRASH) { 239 if (image->type == KEXEC_TYPE_CRASH) {
200 for (i = 0; i < nr_segments; i 240 for (i = 0; i < nr_segments; i++) {
201 unsigned long mstart, 241 unsigned long mstart, mend;
202 242
203 mstart = image->segmen 243 mstart = image->segment[i].mem;
204 mend = mstart + image- 244 mend = mstart + image->segment[i].memsz - 1;
205 /* Ensure we are withi 245 /* Ensure we are within the crash kernel limits */
206 if ((mstart < phys_to_ 246 if ((mstart < phys_to_boot_phys(crashk_res.start)) ||
207 (mend > phys_to_bo 247 (mend > phys_to_boot_phys(crashk_res.end)))
208 return -EADDRN 248 return -EADDRNOTAVAIL;
209 } 249 }
210 } 250 }
211 #endif <<
212 251
213 return 0; 252 return 0;
214 } 253 }
215 254
216 struct kimage *do_kimage_alloc_init(void) 255 struct kimage *do_kimage_alloc_init(void)
217 { 256 {
218 struct kimage *image; 257 struct kimage *image;
219 258
220 /* Allocate a controlling structure */ 259 /* Allocate a controlling structure */
221 image = kzalloc(sizeof(*image), GFP_KE 260 image = kzalloc(sizeof(*image), GFP_KERNEL);
222 if (!image) 261 if (!image)
223 return NULL; 262 return NULL;
224 263
225 image->head = 0; 264 image->head = 0;
226 image->entry = &image->head; 265 image->entry = &image->head;
227 image->last_entry = &image->head; 266 image->last_entry = &image->head;
228 image->control_page = ~0; /* By defaul 267 image->control_page = ~0; /* By default this does not apply */
229 image->type = KEXEC_TYPE_DEFAULT; 268 image->type = KEXEC_TYPE_DEFAULT;
230 269
231 /* Initialize the list of control page 270 /* Initialize the list of control pages */
232 INIT_LIST_HEAD(&image->control_pages); 271 INIT_LIST_HEAD(&image->control_pages);
233 272
234 /* Initialize the list of destination 273 /* Initialize the list of destination pages */
235 INIT_LIST_HEAD(&image->dest_pages); 274 INIT_LIST_HEAD(&image->dest_pages);
236 275
237 /* Initialize the list of unusable pag 276 /* Initialize the list of unusable pages */
238 INIT_LIST_HEAD(&image->unusable_pages) 277 INIT_LIST_HEAD(&image->unusable_pages);
239 278
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; 279 return image;
247 } 280 }
248 281
249 int kimage_is_destination_range(struct kimage 282 int kimage_is_destination_range(struct kimage *image,
250 unsign 283 unsigned long start,
251 unsign 284 unsigned long end)
252 { 285 {
253 unsigned long i; 286 unsigned long i;
254 287
255 for (i = 0; i < image->nr_segments; i+ 288 for (i = 0; i < image->nr_segments; i++) {
256 unsigned long mstart, mend; 289 unsigned long mstart, mend;
257 290
258 mstart = image->segment[i].mem 291 mstart = image->segment[i].mem;
259 mend = mstart + image->segment !! 292 mend = mstart + image->segment[i].memsz;
260 if ((end >= mstart) && (start !! 293 if ((end > mstart) && (start < mend))
261 return 1; 294 return 1;
262 } 295 }
263 296
264 return 0; 297 return 0;
265 } 298 }
266 299
267 static struct page *kimage_alloc_pages(gfp_t g 300 static struct page *kimage_alloc_pages(gfp_t gfp_mask, unsigned int order)
268 { 301 {
269 struct page *pages; 302 struct page *pages;
270 303
271 if (fatal_signal_pending(current)) 304 if (fatal_signal_pending(current))
272 return NULL; 305 return NULL;
273 pages = alloc_pages(gfp_mask & ~__GFP_ 306 pages = alloc_pages(gfp_mask & ~__GFP_ZERO, order);
274 if (pages) { 307 if (pages) {
275 unsigned int count, i; 308 unsigned int count, i;
276 309
277 pages->mapping = NULL; 310 pages->mapping = NULL;
278 set_page_private(pages, order) 311 set_page_private(pages, order);
279 count = 1 << order; 312 count = 1 << order;
280 for (i = 0; i < count; i++) 313 for (i = 0; i < count; i++)
281 SetPageReserved(pages 314 SetPageReserved(pages + i);
282 315
283 arch_kexec_post_alloc_pages(pa 316 arch_kexec_post_alloc_pages(page_address(pages), count,
284 gf 317 gfp_mask);
285 318
286 if (gfp_mask & __GFP_ZERO) 319 if (gfp_mask & __GFP_ZERO)
287 for (i = 0; i < count; 320 for (i = 0; i < count; i++)
288 clear_highpage 321 clear_highpage(pages + i);
289 } 322 }
290 323
291 return pages; 324 return pages;
292 } 325 }
293 326
294 static void kimage_free_pages(struct page *pag 327 static void kimage_free_pages(struct page *page)
295 { 328 {
296 unsigned int order, count, i; 329 unsigned int order, count, i;
297 330
298 order = page_private(page); 331 order = page_private(page);
299 count = 1 << order; 332 count = 1 << order;
300 333
301 arch_kexec_pre_free_pages(page_address 334 arch_kexec_pre_free_pages(page_address(page), count);
302 335
303 for (i = 0; i < count; i++) 336 for (i = 0; i < count; i++)
304 ClearPageReserved(page + i); 337 ClearPageReserved(page + i);
305 __free_pages(page, order); 338 __free_pages(page, order);
306 } 339 }
307 340
308 void kimage_free_page_list(struct list_head *l 341 void kimage_free_page_list(struct list_head *list)
309 { 342 {
310 struct page *page, *next; 343 struct page *page, *next;
311 344
312 list_for_each_entry_safe(page, next, l 345 list_for_each_entry_safe(page, next, list, lru) {
313 list_del(&page->lru); 346 list_del(&page->lru);
314 kimage_free_pages(page); 347 kimage_free_pages(page);
315 } 348 }
316 } 349 }
317 350
318 static struct page *kimage_alloc_normal_contro 351 static struct page *kimage_alloc_normal_control_pages(struct kimage *image,
319 352 unsigned int order)
320 { 353 {
321 /* Control pages are special, they are 354 /* Control pages are special, they are the intermediaries
322 * that are needed while we copy the r 355 * that are needed while we copy the rest of the pages
323 * to their final resting place. As s 356 * to their final resting place. As such they must
324 * not conflict with either the destin 357 * not conflict with either the destination addresses
325 * or memory the kernel is already usi 358 * or memory the kernel is already using.
326 * 359 *
327 * The only case where we really need 360 * The only case where we really need more than one of
328 * these are for architectures where w 361 * these are for architectures where we cannot disable
329 * the MMU and must instead generate a 362 * the MMU and must instead generate an identity mapped
330 * page table for all of the memory. 363 * page table for all of the memory.
331 * 364 *
332 * At worst this runs in O(N) of the i 365 * At worst this runs in O(N) of the image size.
333 */ 366 */
334 struct list_head extra_pages; 367 struct list_head extra_pages;
335 struct page *pages; 368 struct page *pages;
336 unsigned int count; 369 unsigned int count;
337 370
338 count = 1 << order; 371 count = 1 << order;
339 INIT_LIST_HEAD(&extra_pages); 372 INIT_LIST_HEAD(&extra_pages);
340 373
341 /* Loop while I can allocate a page an 374 /* Loop while I can allocate a page and the page allocated
342 * is a destination page. 375 * is a destination page.
343 */ 376 */
344 do { 377 do {
345 unsigned long pfn, epfn, addr, 378 unsigned long pfn, epfn, addr, eaddr;
346 379
347 pages = kimage_alloc_pages(KEX 380 pages = kimage_alloc_pages(KEXEC_CONTROL_MEMORY_GFP, order);
348 if (!pages) 381 if (!pages)
349 break; 382 break;
350 pfn = page_to_boot_pfn(pages 383 pfn = page_to_boot_pfn(pages);
351 epfn = pfn + count; 384 epfn = pfn + count;
352 addr = pfn << PAGE_SHIFT; 385 addr = pfn << PAGE_SHIFT;
353 eaddr = (epfn << PAGE_SHIFT) - !! 386 eaddr = epfn << PAGE_SHIFT;
354 if ((epfn >= (KEXEC_CONTROL_ME 387 if ((epfn >= (KEXEC_CONTROL_MEMORY_LIMIT >> PAGE_SHIFT)) ||
355 kimage_is_destin 388 kimage_is_destination_range(image, addr, eaddr)) {
356 list_add(&pages->lru, 389 list_add(&pages->lru, &extra_pages);
357 pages = NULL; 390 pages = NULL;
358 } 391 }
359 } while (!pages); 392 } while (!pages);
360 393
361 if (pages) { 394 if (pages) {
362 /* Remember the allocated page 395 /* Remember the allocated page... */
363 list_add(&pages->lru, &image-> 396 list_add(&pages->lru, &image->control_pages);
364 397
365 /* Because the page is already 398 /* Because the page is already in it's destination
366 * location we will never allo 399 * location we will never allocate another page at
367 * that address. Therefore ki 400 * that address. Therefore kimage_alloc_pages
368 * will not return it (again) 401 * will not return it (again) and we don't need
369 * to give it an entry in imag 402 * to give it an entry in image->segment[].
370 */ 403 */
371 } 404 }
372 /* Deal with the destination pages I h 405 /* Deal with the destination pages I have inadvertently allocated.
373 * 406 *
374 * Ideally I would convert multi-page 407 * Ideally I would convert multi-page allocations into single
375 * page allocations, and add everythin 408 * page allocations, and add everything to image->dest_pages.
376 * 409 *
377 * For now it is simpler to just free 410 * For now it is simpler to just free the pages.
378 */ 411 */
379 kimage_free_page_list(&extra_pages); 412 kimage_free_page_list(&extra_pages);
380 413
381 return pages; 414 return pages;
382 } 415 }
383 416
384 #ifdef CONFIG_CRASH_DUMP <<
385 static struct page *kimage_alloc_crash_control 417 static struct page *kimage_alloc_crash_control_pages(struct kimage *image,
386 418 unsigned int order)
387 { 419 {
388 /* Control pages are special, they are 420 /* Control pages are special, they are the intermediaries
389 * that are needed while we copy the r 421 * that are needed while we copy the rest of the pages
390 * to their final resting place. As s 422 * to their final resting place. As such they must
391 * not conflict with either the destin 423 * not conflict with either the destination addresses
392 * or memory the kernel is already usi 424 * or memory the kernel is already using.
393 * 425 *
394 * Control pages are also the only pag 426 * Control pages are also the only pags we must allocate
395 * when loading a crash kernel. All o 427 * when loading a crash kernel. All of the other pages
396 * are specified by the segments and w 428 * are specified by the segments and we just memcpy
397 * into them directly. 429 * into them directly.
398 * 430 *
399 * The only case where we really need 431 * The only case where we really need more than one of
400 * these are for architectures where w 432 * these are for architectures where we cannot disable
401 * the MMU and must instead generate a 433 * the MMU and must instead generate an identity mapped
402 * page table for all of the memory. 434 * page table for all of the memory.
403 * 435 *
404 * Given the low demand this implement 436 * Given the low demand this implements a very simple
405 * allocator that finds the first hole 437 * allocator that finds the first hole of the appropriate
406 * size in the reserved memory region, 438 * size in the reserved memory region, and allocates all
407 * of the memory up to and including t 439 * of the memory up to and including the hole.
408 */ 440 */
409 unsigned long hole_start, hole_end, si 441 unsigned long hole_start, hole_end, size;
410 struct page *pages; 442 struct page *pages;
411 443
412 pages = NULL; 444 pages = NULL;
413 size = (1 << order) << PAGE_SHIFT; 445 size = (1 << order) << PAGE_SHIFT;
414 hole_start = ALIGN(image->control_page !! 446 hole_start = (image->control_page + (size - 1)) & ~(size - 1);
415 hole_end = hole_start + size - 1; 447 hole_end = hole_start + size - 1;
416 while (hole_end <= crashk_res.end) { 448 while (hole_end <= crashk_res.end) {
417 unsigned long i; 449 unsigned long i;
418 450
419 cond_resched(); 451 cond_resched();
420 452
421 if (hole_end > KEXEC_CRASH_CON 453 if (hole_end > KEXEC_CRASH_CONTROL_MEMORY_LIMIT)
422 break; 454 break;
423 /* See if I overlap any of the 455 /* See if I overlap any of the segments */
424 for (i = 0; i < image->nr_segm 456 for (i = 0; i < image->nr_segments; i++) {
425 unsigned long mstart, 457 unsigned long mstart, mend;
426 458
427 mstart = image->segmen 459 mstart = image->segment[i].mem;
428 mend = mstart + imag 460 mend = mstart + image->segment[i].memsz - 1;
429 if ((hole_end >= mstar 461 if ((hole_end >= mstart) && (hole_start <= mend)) {
430 /* Advance the 462 /* Advance the hole to the end of the segment */
431 hole_start = A !! 463 hole_start = (mend + (size - 1)) & ~(size - 1);
432 hole_end = h 464 hole_end = hole_start + size - 1;
433 break; 465 break;
434 } 466 }
435 } 467 }
436 /* If I don't overlap any segm 468 /* If I don't overlap any segments I have found my hole! */
437 if (i == image->nr_segments) { 469 if (i == image->nr_segments) {
438 pages = pfn_to_page(ho 470 pages = pfn_to_page(hole_start >> PAGE_SHIFT);
439 image->control_page = !! 471 image->control_page = hole_end;
440 break; 472 break;
441 } 473 }
442 } 474 }
443 475
444 /* Ensure that these pages are decrypt 476 /* Ensure that these pages are decrypted if SME is enabled. */
445 if (pages) 477 if (pages)
446 arch_kexec_post_alloc_pages(pa 478 arch_kexec_post_alloc_pages(page_address(pages), 1 << order, 0);
447 479
448 return pages; 480 return pages;
449 } 481 }
450 #endif <<
451 482
452 483
453 struct page *kimage_alloc_control_pages(struct 484 struct page *kimage_alloc_control_pages(struct kimage *image,
454 unsig 485 unsigned int order)
455 { 486 {
456 struct page *pages = NULL; 487 struct page *pages = NULL;
457 488
458 switch (image->type) { 489 switch (image->type) {
459 case KEXEC_TYPE_DEFAULT: 490 case KEXEC_TYPE_DEFAULT:
460 pages = kimage_alloc_normal_co 491 pages = kimage_alloc_normal_control_pages(image, order);
461 break; 492 break;
462 #ifdef CONFIG_CRASH_DUMP <<
463 case KEXEC_TYPE_CRASH: 493 case KEXEC_TYPE_CRASH:
464 pages = kimage_alloc_crash_con 494 pages = kimage_alloc_crash_control_pages(image, order);
465 break; 495 break;
466 #endif <<
467 } 496 }
468 497
469 return pages; 498 return pages;
470 } 499 }
471 500
>> 501 int kimage_crash_copy_vmcoreinfo(struct kimage *image)
>> 502 {
>> 503 struct page *vmcoreinfo_page;
>> 504 void *safecopy;
>> 505
>> 506 if (image->type != KEXEC_TYPE_CRASH)
>> 507 return 0;
>> 508
>> 509 /*
>> 510 * For kdump, allocate one vmcoreinfo safe copy from the
>> 511 * crash memory. as we have arch_kexec_protect_crashkres()
>> 512 * after kexec syscall, we naturally protect it from write
>> 513 * (even read) access under kernel direct mapping. But on
>> 514 * the other hand, we still need to operate it when crash
>> 515 * happens to generate vmcoreinfo note, hereby we rely on
>> 516 * vmap for this purpose.
>> 517 */
>> 518 vmcoreinfo_page = kimage_alloc_control_pages(image, 0);
>> 519 if (!vmcoreinfo_page) {
>> 520 pr_warn("Could not allocate vmcoreinfo buffer\n");
>> 521 return -ENOMEM;
>> 522 }
>> 523 safecopy = vmap(&vmcoreinfo_page, 1, VM_MAP, PAGE_KERNEL);
>> 524 if (!safecopy) {
>> 525 pr_warn("Could not vmap vmcoreinfo buffer\n");
>> 526 return -ENOMEM;
>> 527 }
>> 528
>> 529 image->vmcoreinfo_data_copy = safecopy;
>> 530 crash_update_vmcoreinfo_safecopy(safecopy);
>> 531
>> 532 return 0;
>> 533 }
>> 534
472 static int kimage_add_entry(struct kimage *ima 535 static int kimage_add_entry(struct kimage *image, kimage_entry_t entry)
473 { 536 {
474 if (*image->entry != 0) 537 if (*image->entry != 0)
475 image->entry++; 538 image->entry++;
476 539
477 if (image->entry == image->last_entry) 540 if (image->entry == image->last_entry) {
478 kimage_entry_t *ind_page; 541 kimage_entry_t *ind_page;
479 struct page *page; 542 struct page *page;
480 543
481 page = kimage_alloc_page(image 544 page = kimage_alloc_page(image, GFP_KERNEL, KIMAGE_NO_DEST);
482 if (!page) 545 if (!page)
483 return -ENOMEM; 546 return -ENOMEM;
484 547
485 ind_page = page_address(page); 548 ind_page = page_address(page);
486 *image->entry = virt_to_boot_p 549 *image->entry = virt_to_boot_phys(ind_page) | IND_INDIRECTION;
487 image->entry = ind_page; 550 image->entry = ind_page;
488 image->last_entry = ind_page + 551 image->last_entry = ind_page +
489 ((PAGE_S 552 ((PAGE_SIZE/sizeof(kimage_entry_t)) - 1);
490 } 553 }
491 *image->entry = entry; 554 *image->entry = entry;
492 image->entry++; 555 image->entry++;
493 *image->entry = 0; 556 *image->entry = 0;
494 557
495 return 0; 558 return 0;
496 } 559 }
497 560
498 static int kimage_set_destination(struct kimag 561 static int kimage_set_destination(struct kimage *image,
499 unsigned lo 562 unsigned long destination)
500 { 563 {
>> 564 int result;
>> 565
501 destination &= PAGE_MASK; 566 destination &= PAGE_MASK;
>> 567 result = kimage_add_entry(image, destination | IND_DESTINATION);
502 568
503 return kimage_add_entry(image, destina !! 569 return result;
504 } 570 }
505 571
506 572
507 static int kimage_add_page(struct kimage *imag 573 static int kimage_add_page(struct kimage *image, unsigned long page)
508 { 574 {
>> 575 int result;
>> 576
509 page &= PAGE_MASK; 577 page &= PAGE_MASK;
>> 578 result = kimage_add_entry(image, page | IND_SOURCE);
510 579
511 return kimage_add_entry(image, page | !! 580 return result;
512 } 581 }
513 582
514 583
515 static void kimage_free_extra_pages(struct kim 584 static void kimage_free_extra_pages(struct kimage *image)
516 { 585 {
517 /* Walk through and free any extra des 586 /* Walk through and free any extra destination pages I may have */
518 kimage_free_page_list(&image->dest_pag 587 kimage_free_page_list(&image->dest_pages);
519 588
520 /* Walk through and free any unusable 589 /* Walk through and free any unusable pages I have cached */
521 kimage_free_page_list(&image->unusable 590 kimage_free_page_list(&image->unusable_pages);
522 591
523 } 592 }
524 593
>> 594 int __weak machine_kexec_post_load(struct kimage *image)
>> 595 {
>> 596 return 0;
>> 597 }
>> 598
525 void kimage_terminate(struct kimage *image) 599 void kimage_terminate(struct kimage *image)
526 { 600 {
527 if (*image->entry != 0) 601 if (*image->entry != 0)
528 image->entry++; 602 image->entry++;
529 603
530 *image->entry = IND_DONE; 604 *image->entry = IND_DONE;
531 } 605 }
532 606
533 #define for_each_kimage_entry(image, ptr, entr 607 #define for_each_kimage_entry(image, ptr, entry) \
534 for (ptr = &image->head; (entry = *ptr 608 for (ptr = &image->head; (entry = *ptr) && !(entry & IND_DONE); \
535 ptr = (entry & IND_INDIRECTION 609 ptr = (entry & IND_INDIRECTION) ? \
536 boot_phys_to_virt((ent 610 boot_phys_to_virt((entry & PAGE_MASK)) : ptr + 1)
537 611
538 static void kimage_free_entry(kimage_entry_t e 612 static void kimage_free_entry(kimage_entry_t entry)
539 { 613 {
540 struct page *page; 614 struct page *page;
541 615
542 page = boot_pfn_to_page(entry >> PAGE_ 616 page = boot_pfn_to_page(entry >> PAGE_SHIFT);
543 kimage_free_pages(page); 617 kimage_free_pages(page);
544 } 618 }
545 619
546 void kimage_free(struct kimage *image) 620 void kimage_free(struct kimage *image)
547 { 621 {
548 kimage_entry_t *ptr, entry; 622 kimage_entry_t *ptr, entry;
549 kimage_entry_t ind = 0; 623 kimage_entry_t ind = 0;
550 624
551 if (!image) 625 if (!image)
552 return; 626 return;
553 627
554 #ifdef CONFIG_CRASH_DUMP <<
555 if (image->vmcoreinfo_data_copy) { 628 if (image->vmcoreinfo_data_copy) {
556 crash_update_vmcoreinfo_safeco 629 crash_update_vmcoreinfo_safecopy(NULL);
557 vunmap(image->vmcoreinfo_data_ 630 vunmap(image->vmcoreinfo_data_copy);
558 } 631 }
559 #endif <<
560 632
561 kimage_free_extra_pages(image); 633 kimage_free_extra_pages(image);
562 for_each_kimage_entry(image, ptr, entr 634 for_each_kimage_entry(image, ptr, entry) {
563 if (entry & IND_INDIRECTION) { 635 if (entry & IND_INDIRECTION) {
564 /* Free the previous i 636 /* Free the previous indirection page */
565 if (ind & IND_INDIRECT 637 if (ind & IND_INDIRECTION)
566 kimage_free_en 638 kimage_free_entry(ind);
567 /* Save this indirecti 639 /* Save this indirection page until we are
568 * done with it. 640 * done with it.
569 */ 641 */
570 ind = entry; 642 ind = entry;
571 } else if (entry & IND_SOURCE) 643 } else if (entry & IND_SOURCE)
572 kimage_free_entry(entr 644 kimage_free_entry(entry);
573 } 645 }
574 /* Free the final indirection page */ 646 /* Free the final indirection page */
575 if (ind & IND_INDIRECTION) 647 if (ind & IND_INDIRECTION)
576 kimage_free_entry(ind); 648 kimage_free_entry(ind);
577 649
578 /* Handle any machine specific cleanup 650 /* Handle any machine specific cleanup */
579 machine_kexec_cleanup(image); 651 machine_kexec_cleanup(image);
580 652
581 /* Free the kexec control pages... */ 653 /* Free the kexec control pages... */
582 kimage_free_page_list(&image->control_ 654 kimage_free_page_list(&image->control_pages);
583 655
584 /* 656 /*
585 * Free up any temporary buffers alloc 657 * Free up any temporary buffers allocated. This might hit if
586 * error occurred much later after buf 658 * error occurred much later after buffer allocation.
587 */ 659 */
588 if (image->file_mode) 660 if (image->file_mode)
589 kimage_file_post_load_cleanup( 661 kimage_file_post_load_cleanup(image);
590 662
591 kfree(image); 663 kfree(image);
592 } 664 }
593 665
594 static kimage_entry_t *kimage_dst_used(struct 666 static kimage_entry_t *kimage_dst_used(struct kimage *image,
595 unsign 667 unsigned long page)
596 { 668 {
597 kimage_entry_t *ptr, entry; 669 kimage_entry_t *ptr, entry;
598 unsigned long destination = 0; 670 unsigned long destination = 0;
599 671
600 for_each_kimage_entry(image, ptr, entr 672 for_each_kimage_entry(image, ptr, entry) {
601 if (entry & IND_DESTINATION) 673 if (entry & IND_DESTINATION)
602 destination = entry & 674 destination = entry & PAGE_MASK;
603 else if (entry & IND_SOURCE) { 675 else if (entry & IND_SOURCE) {
604 if (page == destinatio 676 if (page == destination)
605 return ptr; 677 return ptr;
606 destination += PAGE_SI 678 destination += PAGE_SIZE;
607 } 679 }
608 } 680 }
609 681
610 return NULL; 682 return NULL;
611 } 683 }
612 684
613 static struct page *kimage_alloc_page(struct k 685 static struct page *kimage_alloc_page(struct kimage *image,
614 gfp_t 686 gfp_t gfp_mask,
615 unsign 687 unsigned long destination)
616 { 688 {
617 /* 689 /*
618 * Here we implement safeguards to ens 690 * Here we implement safeguards to ensure that a source page
619 * is not copied to its destination pa 691 * is not copied to its destination page before the data on
620 * the destination page is no longer u 692 * the destination page is no longer useful.
621 * 693 *
622 * To do this we maintain the invarian 694 * To do this we maintain the invariant that a source page is
623 * either its own destination page, or 695 * either its own destination page, or it is not a
624 * destination page at all. 696 * destination page at all.
625 * 697 *
626 * That is slightly stronger than requ 698 * That is slightly stronger than required, but the proof
627 * that no problems will not occur is 699 * that no problems will not occur is trivial, and the
628 * implementation is simply to verify. 700 * implementation is simply to verify.
629 * 701 *
630 * When allocating all pages normally 702 * When allocating all pages normally this algorithm will run
631 * in O(N) time, but in the worst case 703 * in O(N) time, but in the worst case it will run in O(N^2)
632 * time. If the runtime is a problem 704 * time. If the runtime is a problem the data structures can
633 * be fixed. 705 * be fixed.
634 */ 706 */
635 struct page *page; 707 struct page *page;
636 unsigned long addr; 708 unsigned long addr;
637 709
638 /* 710 /*
639 * Walk through the list of destinatio 711 * Walk through the list of destination pages, and see if I
640 * have a match. 712 * have a match.
641 */ 713 */
642 list_for_each_entry(page, &image->dest 714 list_for_each_entry(page, &image->dest_pages, lru) {
643 addr = page_to_boot_pfn(page) 715 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
644 if (addr == destination) { 716 if (addr == destination) {
645 list_del(&page->lru); 717 list_del(&page->lru);
646 return page; 718 return page;
647 } 719 }
648 } 720 }
649 page = NULL; 721 page = NULL;
650 while (1) { 722 while (1) {
651 kimage_entry_t *old; 723 kimage_entry_t *old;
652 724
653 /* Allocate a page, if we run 725 /* Allocate a page, if we run out of memory give up */
654 page = kimage_alloc_pages(gfp_ 726 page = kimage_alloc_pages(gfp_mask, 0);
655 if (!page) 727 if (!page)
656 return NULL; 728 return NULL;
657 /* If the page cannot be used 729 /* If the page cannot be used file it away */
658 if (page_to_boot_pfn(page) > 730 if (page_to_boot_pfn(page) >
659 (KEXEC_SOURCE_ 731 (KEXEC_SOURCE_MEMORY_LIMIT >> PAGE_SHIFT)) {
660 list_add(&page->lru, & 732 list_add(&page->lru, &image->unusable_pages);
661 continue; 733 continue;
662 } 734 }
663 addr = page_to_boot_pfn(page) 735 addr = page_to_boot_pfn(page) << PAGE_SHIFT;
664 736
665 /* If it is the destination pa 737 /* If it is the destination page we want use it */
666 if (addr == destination) 738 if (addr == destination)
667 break; 739 break;
668 740
669 /* If the page is not a destin 741 /* If the page is not a destination page use it */
670 if (!kimage_is_destination_ran 742 if (!kimage_is_destination_range(image, addr,
671 !! 743 addr + PAGE_SIZE))
672 break; 744 break;
673 745
674 /* 746 /*
675 * I know that the page is som 747 * I know that the page is someones destination page.
676 * See if there is already a s 748 * See if there is already a source page for this
677 * destination page. And if s 749 * destination page. And if so swap the source pages.
678 */ 750 */
679 old = kimage_dst_used(image, a 751 old = kimage_dst_used(image, addr);
680 if (old) { 752 if (old) {
681 /* If so move it */ 753 /* If so move it */
682 unsigned long old_addr 754 unsigned long old_addr;
683 struct page *old_page; 755 struct page *old_page;
684 756
685 old_addr = *old & PAGE 757 old_addr = *old & PAGE_MASK;
686 old_page = boot_pfn_to 758 old_page = boot_pfn_to_page(old_addr >> PAGE_SHIFT);
687 copy_highpage(page, ol 759 copy_highpage(page, old_page);
688 *old = addr | (*old & 760 *old = addr | (*old & ~PAGE_MASK);
689 761
690 /* The old page I have 762 /* The old page I have found cannot be a
691 * destination page, s 763 * destination page, so return it if it's
692 * gfp_flags honor the 764 * gfp_flags honor the ones passed in.
693 */ 765 */
694 if (!(gfp_mask & __GFP 766 if (!(gfp_mask & __GFP_HIGHMEM) &&
695 PageHighMem(old_pa 767 PageHighMem(old_page)) {
696 kimage_free_pa 768 kimage_free_pages(old_page);
697 continue; 769 continue;
698 } 770 }
699 page = old_page; 771 page = old_page;
700 break; 772 break;
701 } 773 }
702 /* Place the page on the desti 774 /* Place the page on the destination list, to be used later */
703 list_add(&page->lru, &image->d 775 list_add(&page->lru, &image->dest_pages);
704 } 776 }
705 777
706 return page; 778 return page;
707 } 779 }
708 780
709 static int kimage_load_normal_segment(struct k 781 static int kimage_load_normal_segment(struct kimage *image,
710 struc 782 struct kexec_segment *segment)
711 { 783 {
712 unsigned long maddr; 784 unsigned long maddr;
713 size_t ubytes, mbytes; 785 size_t ubytes, mbytes;
714 int result; 786 int result;
715 unsigned char __user *buf = NULL; 787 unsigned char __user *buf = NULL;
716 unsigned char *kbuf = NULL; 788 unsigned char *kbuf = NULL;
717 789
718 if (image->file_mode) 790 if (image->file_mode)
719 kbuf = segment->kbuf; 791 kbuf = segment->kbuf;
720 else 792 else
721 buf = segment->buf; 793 buf = segment->buf;
722 ubytes = segment->bufsz; 794 ubytes = segment->bufsz;
723 mbytes = segment->memsz; 795 mbytes = segment->memsz;
724 maddr = segment->mem; 796 maddr = segment->mem;
725 797
726 result = kimage_set_destination(image, 798 result = kimage_set_destination(image, maddr);
727 if (result < 0) 799 if (result < 0)
728 goto out; 800 goto out;
729 801
730 while (mbytes) { 802 while (mbytes) {
731 struct page *page; 803 struct page *page;
732 char *ptr; 804 char *ptr;
733 size_t uchunk, mchunk; 805 size_t uchunk, mchunk;
734 806
735 page = kimage_alloc_page(image 807 page = kimage_alloc_page(image, GFP_HIGHUSER, maddr);
736 if (!page) { 808 if (!page) {
737 result = -ENOMEM; 809 result = -ENOMEM;
738 goto out; 810 goto out;
739 } 811 }
740 result = kimage_add_page(image 812 result = kimage_add_page(image, page_to_boot_pfn(page)
741 813 << PAGE_SHIFT);
742 if (result < 0) 814 if (result < 0)
743 goto out; 815 goto out;
744 816
745 ptr = kmap_local_page(page); !! 817 ptr = kmap(page);
746 /* Start with a clear page */ 818 /* Start with a clear page */
747 clear_page(ptr); 819 clear_page(ptr);
748 ptr += maddr & ~PAGE_MASK; 820 ptr += maddr & ~PAGE_MASK;
749 mchunk = min_t(size_t, mbytes, 821 mchunk = min_t(size_t, mbytes,
750 PAGE_SIZE - (m 822 PAGE_SIZE - (maddr & ~PAGE_MASK));
751 uchunk = min(ubytes, mchunk); 823 uchunk = min(ubytes, mchunk);
752 824
753 if (uchunk) { !! 825 /* For file based kexec, source pages are in kernel memory */
754 /* For file based kexe !! 826 if (image->file_mode)
755 if (image->file_mode) !! 827 memcpy(ptr, kbuf, uchunk);
756 memcpy(ptr, kb !! 828 else
757 else !! 829 result = copy_from_user(ptr, buf, uchunk);
758 result = copy_ !! 830 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) { 831 if (result) {
767 result = -EFAULT; 832 result = -EFAULT;
768 goto out; 833 goto out;
769 } 834 }
>> 835 ubytes -= uchunk;
770 maddr += mchunk; 836 maddr += mchunk;
>> 837 if (image->file_mode)
>> 838 kbuf += mchunk;
>> 839 else
>> 840 buf += mchunk;
771 mbytes -= mchunk; 841 mbytes -= mchunk;
772 842
773 cond_resched(); 843 cond_resched();
774 } 844 }
775 out: 845 out:
776 return result; 846 return result;
777 } 847 }
778 848
779 #ifdef CONFIG_CRASH_DUMP <<
780 static int kimage_load_crash_segment(struct ki 849 static int kimage_load_crash_segment(struct kimage *image,
781 struct 850 struct kexec_segment *segment)
782 { 851 {
783 /* For crash dumps kernels we simply c 852 /* For crash dumps kernels we simply copy the data from
784 * user space to it's destination. 853 * user space to it's destination.
785 * We do things a page at a time for t 854 * We do things a page at a time for the sake of kmap.
786 */ 855 */
787 unsigned long maddr; 856 unsigned long maddr;
788 size_t ubytes, mbytes; 857 size_t ubytes, mbytes;
789 int result; 858 int result;
790 unsigned char __user *buf = NULL; 859 unsigned char __user *buf = NULL;
791 unsigned char *kbuf = NULL; 860 unsigned char *kbuf = NULL;
792 861
793 result = 0; 862 result = 0;
794 if (image->file_mode) 863 if (image->file_mode)
795 kbuf = segment->kbuf; 864 kbuf = segment->kbuf;
796 else 865 else
797 buf = segment->buf; 866 buf = segment->buf;
798 ubytes = segment->bufsz; 867 ubytes = segment->bufsz;
799 mbytes = segment->memsz; 868 mbytes = segment->memsz;
800 maddr = segment->mem; 869 maddr = segment->mem;
801 while (mbytes) { 870 while (mbytes) {
802 struct page *page; 871 struct page *page;
803 char *ptr; 872 char *ptr;
804 size_t uchunk, mchunk; 873 size_t uchunk, mchunk;
805 874
806 page = boot_pfn_to_page(maddr 875 page = boot_pfn_to_page(maddr >> PAGE_SHIFT);
807 if (!page) { 876 if (!page) {
808 result = -ENOMEM; 877 result = -ENOMEM;
809 goto out; 878 goto out;
810 } 879 }
811 arch_kexec_post_alloc_pages(pa 880 arch_kexec_post_alloc_pages(page_address(page), 1, 0);
812 ptr = kmap_local_page(page); !! 881 ptr = kmap(page);
813 ptr += maddr & ~PAGE_MASK; 882 ptr += maddr & ~PAGE_MASK;
814 mchunk = min_t(size_t, mbytes, 883 mchunk = min_t(size_t, mbytes,
815 PAGE_SIZE - (m 884 PAGE_SIZE - (maddr & ~PAGE_MASK));
816 uchunk = min(ubytes, mchunk); 885 uchunk = min(ubytes, mchunk);
817 if (mchunk > uchunk) { 886 if (mchunk > uchunk) {
818 /* Zero the trailing p 887 /* Zero the trailing part of the page */
819 memset(ptr + uchunk, 0 888 memset(ptr + uchunk, 0, mchunk - uchunk);
820 } 889 }
821 890
822 if (uchunk) { !! 891 /* For file based kexec, source pages are in kernel memory */
823 /* For file based kexe !! 892 if (image->file_mode)
824 if (image->file_mode) !! 893 memcpy(ptr, kbuf, uchunk);
825 memcpy(ptr, kb !! 894 else
826 else !! 895 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); 896 kexec_flush_icache_page(page);
835 kunmap_local(ptr); !! 897 kunmap(page);
836 arch_kexec_pre_free_pages(page 898 arch_kexec_pre_free_pages(page_address(page), 1);
837 if (result) { 899 if (result) {
838 result = -EFAULT; 900 result = -EFAULT;
839 goto out; 901 goto out;
840 } 902 }
>> 903 ubytes -= uchunk;
841 maddr += mchunk; 904 maddr += mchunk;
>> 905 if (image->file_mode)
>> 906 kbuf += mchunk;
>> 907 else
>> 908 buf += mchunk;
842 mbytes -= mchunk; 909 mbytes -= mchunk;
843 910
844 cond_resched(); 911 cond_resched();
845 } 912 }
846 out: 913 out:
847 return result; 914 return result;
848 } 915 }
849 #endif <<
850 916
851 int kimage_load_segment(struct kimage *image, 917 int kimage_load_segment(struct kimage *image,
852 struct kexec_s 918 struct kexec_segment *segment)
853 { 919 {
854 int result = -ENOMEM; 920 int result = -ENOMEM;
855 921
856 switch (image->type) { 922 switch (image->type) {
857 case KEXEC_TYPE_DEFAULT: 923 case KEXEC_TYPE_DEFAULT:
858 result = kimage_load_normal_se 924 result = kimage_load_normal_segment(image, segment);
859 break; 925 break;
860 #ifdef CONFIG_CRASH_DUMP <<
861 case KEXEC_TYPE_CRASH: 926 case KEXEC_TYPE_CRASH:
862 result = kimage_load_crash_seg 927 result = kimage_load_crash_segment(image, segment);
863 break; 928 break;
864 #endif <<
865 } 929 }
866 930
867 return result; 931 return result;
868 } 932 }
869 933
870 struct kexec_load_limit { <<
871 /* Mutex protects the limit count. */ <<
872 struct mutex mutex; <<
873 int limit; <<
874 }; <<
875 <<
876 static struct kexec_load_limit load_limit_rebo <<
877 .mutex = __MUTEX_INITIALIZER(load_limi <<
878 .limit = -1, <<
879 }; <<
880 <<
881 static struct kexec_load_limit load_limit_pani <<
882 .mutex = __MUTEX_INITIALIZER(load_limi <<
883 .limit = -1, <<
884 }; <<
885 <<
886 struct kimage *kexec_image; 934 struct kimage *kexec_image;
887 struct kimage *kexec_crash_image; 935 struct kimage *kexec_crash_image;
888 static int kexec_load_disabled; !! 936 int kexec_load_disabled;
889 <<
890 #ifdef CONFIG_SYSCTL 937 #ifdef CONFIG_SYSCTL
891 static int kexec_limit_handler(const struct ct <<
892 void *buffer, s <<
893 { <<
894 struct kexec_load_limit *limit = table <<
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 <<
908 if (val < 0) <<
909 return -EINVAL; <<
910 <<
911 mutex_lock(&limit->mutex); <<
912 if (limit->limit != -1 && val <<
913 ret = -EINVAL; <<
914 else <<
915 limit->limit = val; <<
916 mutex_unlock(&limit->mutex); <<
917 <<
918 return ret; <<
919 } <<
920 <<
921 mutex_lock(&limit->mutex); <<
922 val = limit->limit; <<
923 mutex_unlock(&limit->mutex); <<
924 <<
925 return proc_dointvec(&tmp, write, buff <<
926 } <<
927 <<
928 static struct ctl_table kexec_core_sysctls[] = 938 static struct ctl_table kexec_core_sysctls[] = {
929 { 939 {
930 .procname = "kexec_load_ 940 .procname = "kexec_load_disabled",
931 .data = &kexec_load_ 941 .data = &kexec_load_disabled,
932 .maxlen = sizeof(int), 942 .maxlen = sizeof(int),
933 .mode = 0644, 943 .mode = 0644,
934 /* only handle a transition fr 944 /* only handle a transition from default "" to "1" */
935 .proc_handler = proc_dointve 945 .proc_handler = proc_dointvec_minmax,
936 .extra1 = SYSCTL_ONE, 946 .extra1 = SYSCTL_ONE,
937 .extra2 = SYSCTL_ONE, 947 .extra2 = SYSCTL_ONE,
938 }, 948 },
939 { !! 949 { }
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 }; 950 };
952 951
953 static int __init kexec_core_sysctl_init(void) 952 static int __init kexec_core_sysctl_init(void)
954 { 953 {
955 register_sysctl_init("kernel", kexec_c 954 register_sysctl_init("kernel", kexec_core_sysctls);
956 return 0; 955 return 0;
957 } 956 }
958 late_initcall(kexec_core_sysctl_init); 957 late_initcall(kexec_core_sysctl_init);
959 #endif 958 #endif
960 959
961 bool kexec_load_permitted(int kexec_image_type !! 960 /*
>> 961 * No panic_cpu check version of crash_kexec(). This function is called
>> 962 * only when panic_cpu holds the current CPU number; this is the only CPU
>> 963 * which processes crash_kexec routines.
>> 964 */
>> 965 void __noclone __crash_kexec(struct pt_regs *regs)
>> 966 {
>> 967 /* Take the kexec_mutex here to prevent sys_kexec_load
>> 968 * running on one cpu from replacing the crash kernel
>> 969 * we are using after a panic on a different cpu.
>> 970 *
>> 971 * If the crash kernel was not located in a fixed area
>> 972 * of memory the xchg(&kexec_crash_image) would be
>> 973 * sufficient. But since I reuse the memory...
>> 974 */
>> 975 if (mutex_trylock(&kexec_mutex)) {
>> 976 if (kexec_crash_image) {
>> 977 struct pt_regs fixed_regs;
>> 978
>> 979 crash_setup_regs(&fixed_regs, regs);
>> 980 crash_save_vmcoreinfo();
>> 981 machine_crash_shutdown(&fixed_regs);
>> 982 machine_kexec(kexec_crash_image);
>> 983 }
>> 984 mutex_unlock(&kexec_mutex);
>> 985 }
>> 986 }
>> 987 STACK_FRAME_NON_STANDARD(__crash_kexec);
>> 988
>> 989 void crash_kexec(struct pt_regs *regs)
>> 990 {
>> 991 int old_cpu, this_cpu;
>> 992
>> 993 /*
>> 994 * Only one CPU is allowed to execute the crash_kexec() code as with
>> 995 * panic(). Otherwise parallel calls of panic() and crash_kexec()
>> 996 * may stop each other. To exclude them, we use panic_cpu here too.
>> 997 */
>> 998 this_cpu = raw_smp_processor_id();
>> 999 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
>> 1000 if (old_cpu == PANIC_CPU_INVALID) {
>> 1001 /* This is the 1st CPU which comes here, so go ahead. */
>> 1002 __crash_kexec(regs);
>> 1003
>> 1004 /*
>> 1005 * Reset panic_cpu to allow another panic()/crash_kexec()
>> 1006 * call.
>> 1007 */
>> 1008 atomic_set(&panic_cpu, PANIC_CPU_INVALID);
>> 1009 }
>> 1010 }
>> 1011
>> 1012 size_t crash_get_memory_size(void)
>> 1013 {
>> 1014 size_t size = 0;
>> 1015
>> 1016 mutex_lock(&kexec_mutex);
>> 1017 if (crashk_res.end != crashk_res.start)
>> 1018 size = resource_size(&crashk_res);
>> 1019 mutex_unlock(&kexec_mutex);
>> 1020 return size;
>> 1021 }
>> 1022
>> 1023 void __weak crash_free_reserved_phys_range(unsigned long begin,
>> 1024 unsigned long end)
>> 1025 {
>> 1026 unsigned long addr;
>> 1027
>> 1028 for (addr = begin; addr < end; addr += PAGE_SIZE)
>> 1029 free_reserved_page(boot_pfn_to_page(addr >> PAGE_SHIFT));
>> 1030 }
>> 1031
>> 1032 int crash_shrink_memory(unsigned long new_size)
962 { 1033 {
963 struct kexec_load_limit *limit; !! 1034 int ret = 0;
>> 1035 unsigned long start, end;
>> 1036 unsigned long old_size;
>> 1037 struct resource *ram_res;
>> 1038
>> 1039 mutex_lock(&kexec_mutex);
>> 1040
>> 1041 if (kexec_crash_image) {
>> 1042 ret = -ENOENT;
>> 1043 goto unlock;
>> 1044 }
>> 1045 start = crashk_res.start;
>> 1046 end = crashk_res.end;
>> 1047 old_size = (end == 0) ? 0 : end - start + 1;
>> 1048 if (new_size >= old_size) {
>> 1049 ret = (new_size == old_size) ? 0 : -EINVAL;
>> 1050 goto unlock;
>> 1051 }
>> 1052
>> 1053 ram_res = kzalloc(sizeof(*ram_res), GFP_KERNEL);
>> 1054 if (!ram_res) {
>> 1055 ret = -ENOMEM;
>> 1056 goto unlock;
>> 1057 }
>> 1058
>> 1059 start = roundup(start, KEXEC_CRASH_MEM_ALIGN);
>> 1060 end = roundup(start + new_size, KEXEC_CRASH_MEM_ALIGN);
>> 1061
>> 1062 crash_free_reserved_phys_range(end, crashk_res.end);
>> 1063
>> 1064 if ((start == end) && (crashk_res.parent != NULL))
>> 1065 release_resource(&crashk_res);
>> 1066
>> 1067 ram_res->start = end;
>> 1068 ram_res->end = crashk_res.end;
>> 1069 ram_res->flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM;
>> 1070 ram_res->name = "System RAM";
>> 1071
>> 1072 crashk_res.end = end - 1;
>> 1073
>> 1074 insert_resource(&iomem_resource, ram_res);
>> 1075
>> 1076 unlock:
>> 1077 mutex_unlock(&kexec_mutex);
>> 1078 return ret;
>> 1079 }
>> 1080
>> 1081 void crash_save_cpu(struct pt_regs *regs, int cpu)
>> 1082 {
>> 1083 struct elf_prstatus prstatus;
>> 1084 u32 *buf;
>> 1085
>> 1086 if ((cpu < 0) || (cpu >= nr_cpu_ids))
>> 1087 return;
>> 1088
>> 1089 /* Using ELF notes here is opportunistic.
>> 1090 * I need a well defined structure format
>> 1091 * for the data I pass, and I need tags
>> 1092 * on the data to indicate what information I have
>> 1093 * squirrelled away. ELF notes happen to provide
>> 1094 * all of that, so there is no need to invent something new.
>> 1095 */
>> 1096 buf = (u32 *)per_cpu_ptr(crash_notes, cpu);
>> 1097 if (!buf)
>> 1098 return;
>> 1099 memset(&prstatus, 0, sizeof(prstatus));
>> 1100 prstatus.common.pr_pid = current->pid;
>> 1101 elf_core_copy_regs(&prstatus.pr_reg, regs);
>> 1102 buf = append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS,
>> 1103 &prstatus, sizeof(prstatus));
>> 1104 final_note(buf);
>> 1105 }
>> 1106
>> 1107 static int __init crash_notes_memory_init(void)
>> 1108 {
>> 1109 /* Allocate memory for saving cpu registers. */
>> 1110 size_t size, align;
964 1111
965 /* 1112 /*
966 * Only the superuser can use the kexe !! 1113 * crash_notes could be allocated across 2 vmalloc pages when percpu
967 * been disabled. !! 1114 * is vmalloc based . vmalloc doesn't guarantee 2 continuous vmalloc
>> 1115 * pages are also on 2 continuous physical pages. In this case the
>> 1116 * 2nd part of crash_notes in 2nd page could be lost since only the
>> 1117 * starting address and size of crash_notes are exported through sysfs.
>> 1118 * Here round up the size of crash_notes to the nearest power of two
>> 1119 * and pass it to __alloc_percpu as align value. This can make sure
>> 1120 * crash_notes is allocated inside one physical page.
968 */ 1121 */
969 if (!capable(CAP_SYS_BOOT) || kexec_lo !! 1122 size = sizeof(note_buf_t);
970 return false; !! 1123 align = min(roundup_pow_of_two(sizeof(note_buf_t)), PAGE_SIZE);
971 1124
972 /* Check limit counter and decrease it !! 1125 /*
973 limit = (kexec_image_type == KEXEC_TYP !! 1126 * Break compile if size is bigger than PAGE_SIZE since crash_notes
974 &load_limit_panic : &load_limi !! 1127 * definitely will be in 2 pages with that.
975 mutex_lock(&limit->mutex); !! 1128 */
976 if (!limit->limit) { !! 1129 BUILD_BUG_ON(size > PAGE_SIZE);
977 mutex_unlock(&limit->mutex); <<
978 return false; <<
979 } <<
980 if (limit->limit != -1) <<
981 limit->limit--; <<
982 mutex_unlock(&limit->mutex); <<
983 1130
984 return true; !! 1131 crash_notes = __alloc_percpu(size, align);
>> 1132 if (!crash_notes) {
>> 1133 pr_warn("Memory allocation for saving cpu register states failed\n");
>> 1134 return -ENOMEM;
>> 1135 }
>> 1136 return 0;
985 } 1137 }
>> 1138 subsys_initcall(crash_notes_memory_init);
>> 1139
986 1140
987 /* 1141 /*
988 * Move into place and start executing a prelo 1142 * Move into place and start executing a preloaded standalone
989 * executable. If nothing was preloaded retur 1143 * executable. If nothing was preloaded return an error.
990 */ 1144 */
991 int kernel_kexec(void) 1145 int kernel_kexec(void)
992 { 1146 {
993 int error = 0; 1147 int error = 0;
994 1148
995 if (!kexec_trylock()) !! 1149 if (!mutex_trylock(&kexec_mutex))
996 return -EBUSY; 1150 return -EBUSY;
997 if (!kexec_image) { 1151 if (!kexec_image) {
998 error = -EINVAL; 1152 error = -EINVAL;
999 goto Unlock; 1153 goto Unlock;
1000 } 1154 }
1001 1155
1002 #ifdef CONFIG_KEXEC_JUMP 1156 #ifdef CONFIG_KEXEC_JUMP
1003 if (kexec_image->preserve_context) { 1157 if (kexec_image->preserve_context) {
1004 pm_prepare_console(); 1158 pm_prepare_console();
1005 error = freeze_processes(); 1159 error = freeze_processes();
1006 if (error) { 1160 if (error) {
1007 error = -EBUSY; 1161 error = -EBUSY;
1008 goto Restore_console; 1162 goto Restore_console;
1009 } 1163 }
1010 suspend_console(); 1164 suspend_console();
1011 error = dpm_suspend_start(PMS 1165 error = dpm_suspend_start(PMSG_FREEZE);
1012 if (error) 1166 if (error)
1013 goto Resume_console; 1167 goto Resume_console;
1014 /* At this point, dpm_suspend 1168 /* At this point, dpm_suspend_start() has been called,
1015 * but *not* dpm_suspend_end( 1169 * but *not* dpm_suspend_end(). We *must* call
1016 * dpm_suspend_end() now. Ot 1170 * dpm_suspend_end() now. Otherwise, drivers for
1017 * some devices (e.g. interru 1171 * some devices (e.g. interrupt controllers) become
1018 * desynchronized with the ac 1172 * desynchronized with the actual state of the
1019 * hardware at resume time, a 1173 * hardware at resume time, and evil weirdness ensues.
1020 */ 1174 */
1021 error = dpm_suspend_end(PMSG_ 1175 error = dpm_suspend_end(PMSG_FREEZE);
1022 if (error) 1176 if (error)
1023 goto Resume_devices; 1177 goto Resume_devices;
1024 error = suspend_disable_secon 1178 error = suspend_disable_secondary_cpus();
1025 if (error) 1179 if (error)
1026 goto Enable_cpus; 1180 goto Enable_cpus;
1027 local_irq_disable(); 1181 local_irq_disable();
1028 error = syscore_suspend(); 1182 error = syscore_suspend();
1029 if (error) 1183 if (error)
1030 goto Enable_irqs; 1184 goto Enable_irqs;
1031 } else 1185 } else
1032 #endif 1186 #endif
1033 { 1187 {
1034 kexec_in_progress = true; 1188 kexec_in_progress = true;
1035 kernel_restart_prepare("kexec 1189 kernel_restart_prepare("kexec reboot");
1036 migrate_to_reboot_cpu(); 1190 migrate_to_reboot_cpu();
1037 syscore_shutdown(); <<
1038 1191
1039 /* 1192 /*
1040 * migrate_to_reboot_cpu() di 1193 * migrate_to_reboot_cpu() disables CPU hotplug assuming that
1041 * no further code needs to u 1194 * no further code needs to use CPU hotplug (which is true in
1042 * the reboot case). However, 1195 * the reboot case). However, the kexec path depends on using
1043 * CPU hotplug again; so re-e 1196 * CPU hotplug again; so re-enable it here.
1044 */ 1197 */
1045 cpu_hotplug_enable(); 1198 cpu_hotplug_enable();
1046 pr_notice("Starting new kerne 1199 pr_notice("Starting new kernel\n");
1047 machine_shutdown(); 1200 machine_shutdown();
1048 } 1201 }
1049 1202
1050 kmsg_dump(KMSG_DUMP_SHUTDOWN); 1203 kmsg_dump(KMSG_DUMP_SHUTDOWN);
1051 machine_kexec(kexec_image); 1204 machine_kexec(kexec_image);
1052 1205
1053 #ifdef CONFIG_KEXEC_JUMP 1206 #ifdef CONFIG_KEXEC_JUMP
1054 if (kexec_image->preserve_context) { 1207 if (kexec_image->preserve_context) {
1055 syscore_resume(); 1208 syscore_resume();
1056 Enable_irqs: 1209 Enable_irqs:
1057 local_irq_enable(); 1210 local_irq_enable();
1058 Enable_cpus: 1211 Enable_cpus:
1059 suspend_enable_secondary_cpus 1212 suspend_enable_secondary_cpus();
1060 dpm_resume_start(PMSG_RESTORE 1213 dpm_resume_start(PMSG_RESTORE);
1061 Resume_devices: 1214 Resume_devices:
1062 dpm_resume_end(PMSG_RESTORE); 1215 dpm_resume_end(PMSG_RESTORE);
1063 Resume_console: 1216 Resume_console:
1064 resume_console(); 1217 resume_console();
1065 thaw_processes(); 1218 thaw_processes();
1066 Restore_console: 1219 Restore_console:
1067 pm_restore_console(); 1220 pm_restore_console();
1068 } 1221 }
1069 #endif 1222 #endif
1070 1223
1071 Unlock: 1224 Unlock:
1072 kexec_unlock(); !! 1225 mutex_unlock(&kexec_mutex);
1073 return error; 1226 return error;
1074 } 1227 }
>> 1228
>> 1229 /*
>> 1230 * Protection mechanism for crashkernel reserved memory after
>> 1231 * the kdump kernel is loaded.
>> 1232 *
>> 1233 * Provide an empty default implementation here -- architecture
>> 1234 * code may override this
>> 1235 */
>> 1236 void __weak arch_kexec_protect_crashkres(void)
>> 1237 {}
>> 1238
>> 1239 void __weak arch_kexec_unprotect_crashkres(void)
>> 1240 {}
1075 1241
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.