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