1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2006 Peter Wemm 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_pmap.h" 33 #include "opt_watchdog.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/conf.h> 38 #include <sys/cons.h> 39 #include <sys/kernel.h> 40 #include <sys/kerneldump.h> 41 #include <sys/msgbuf.h> 42 #include <sys/sysctl.h> 43 #include <sys/watchdog.h> 44 #include <sys/vmmeter.h> 45 #include <vm/vm.h> 46 #include <vm/vm_param.h> 47 #include <vm/vm_page.h> 48 #include <vm/vm_phys.h> 49 #include <vm/pmap.h> 50 #include <machine/atomic.h> 51 #include <machine/elf.h> 52 #include <machine/md_var.h> 53 #include <machine/minidump.h> 54 55 CTASSERT(sizeof(struct kerneldumpheader) == 512); 56 57 uint64_t *vm_page_dump; 58 int vm_page_dump_size; 59 60 static struct kerneldumpheader kdh; 61 62 /* Handle chunked writes. */ 63 static size_t fragsz; 64 static void *dump_va; 65 static size_t counter, progress, dumpsize, wdog_next; 66 67 CTASSERT(sizeof(*vm_page_dump) == 8); 68 static int dump_retry_count = 5; 69 SYSCTL_INT(_machdep, OID_AUTO, dump_retry_count, CTLFLAG_RWTUN, 70 &dump_retry_count, 0, "Number of times dump has to retry before bailing out"); 71 72 static int 73 is_dumpable(vm_paddr_t pa) 74 { 75 vm_page_t m; 76 int i; 77 78 if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL) 79 return ((m->flags & PG_NODUMP) == 0); 80 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) { 81 if (pa >= dump_avail[i] && pa < dump_avail[i + 1]) 82 return (1); 83 } 84 return (0); 85 } 86 87 #define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8) 88 89 static int 90 blk_flush(struct dumperinfo *di) 91 { 92 int error; 93 94 if (fragsz == 0) 95 return (0); 96 97 error = dump_append(di, dump_va, 0, fragsz); 98 fragsz = 0; 99 return (error); 100 } 101 102 static struct { 103 int min_per; 104 int max_per; 105 int visited; 106 } progress_track[10] = { 107 { 0, 10, 0}, 108 { 10, 20, 0}, 109 { 20, 30, 0}, 110 { 30, 40, 0}, 111 { 40, 50, 0}, 112 { 50, 60, 0}, 113 { 60, 70, 0}, 114 { 70, 80, 0}, 115 { 80, 90, 0}, 116 { 90, 100, 0} 117 }; 118 119 static void 120 report_progress(size_t progress, size_t dumpsize) 121 { 122 int sofar, i; 123 124 sofar = 100 - ((progress * 100) / dumpsize); 125 for (i = 0; i < nitems(progress_track); i++) { 126 if (sofar < progress_track[i].min_per || 127 sofar > progress_track[i].max_per) 128 continue; 129 if (progress_track[i].visited) 130 return; 131 progress_track[i].visited = 1; 132 printf("..%d%%", sofar); 133 return; 134 } 135 } 136 137 /* Pat the watchdog approximately every 128MB of the dump. */ 138 #define WDOG_DUMP_INTERVAL (128 * 1024 * 1024) 139 140 static int 141 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz) 142 { 143 size_t len; 144 int error, i, c; 145 u_int maxdumpsz; 146 147 maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE); 148 if (maxdumpsz == 0) /* seatbelt */ 149 maxdumpsz = PAGE_SIZE; 150 error = 0; 151 if ((sz % PAGE_SIZE) != 0) { 152 printf("size not page aligned\n"); 153 return (EINVAL); 154 } 155 if (ptr != NULL && pa != 0) { 156 printf("cant have both va and pa!\n"); 157 return (EINVAL); 158 } 159 if ((((uintptr_t)pa) % PAGE_SIZE) != 0) { 160 printf("address not page aligned %p\n", ptr); 161 return (EINVAL); 162 } 163 if (ptr != NULL) { 164 /* If we're doing a virtual dump, flush any pre-existing pa pages */ 165 error = blk_flush(di); 166 if (error) 167 return (error); 168 } 169 while (sz) { 170 len = maxdumpsz - fragsz; 171 if (len > sz) 172 len = sz; 173 counter += len; 174 progress -= len; 175 if (counter >> 24) { 176 report_progress(progress, dumpsize); 177 counter &= (1<<24) - 1; 178 } 179 if (progress <= wdog_next) { 180 wdog_kern_pat(WD_LASTVAL); 181 if (wdog_next > WDOG_DUMP_INTERVAL) 182 wdog_next -= WDOG_DUMP_INTERVAL; 183 else 184 wdog_next = 0; 185 } 186 187 if (ptr) { 188 error = dump_append(di, ptr, 0, len); 189 if (error) 190 return (error); 191 ptr += len; 192 sz -= len; 193 } else { 194 for (i = 0; i < len; i += PAGE_SIZE) 195 dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT); 196 fragsz += len; 197 pa += len; 198 sz -= len; 199 if (fragsz == maxdumpsz) { 200 error = blk_flush(di); 201 if (error) 202 return (error); 203 } 204 } 205 206 /* Check for user abort. */ 207 c = cncheckc(); 208 if (c == 0x03) 209 return (ECANCELED); 210 if (c != -1) 211 printf(" (CTRL-C to abort) "); 212 } 213 214 return (0); 215 } 216 217 /* A fake page table page, to avoid having to handle both 4K and 2M pages */ 218 static pd_entry_t fakepd[NPDEPG]; 219 220 int 221 minidumpsys(struct dumperinfo *di) 222 { 223 uint32_t pmapsize; 224 vm_offset_t va; 225 int error; 226 uint64_t bits; 227 uint64_t *pml4, *pdp, *pd, *pt, pa; 228 int i, ii, j, k, n, bit; 229 int retry_count; 230 struct minidumphdr mdhdr; 231 232 retry_count = 0; 233 retry: 234 retry_count++; 235 counter = 0; 236 for (i = 0; i < nitems(progress_track); i++) 237 progress_track[i].visited = 0; 238 /* Walk page table pages, set bits in vm_page_dump */ 239 pmapsize = 0; 240 for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + nkpt * NBPDR, 241 kernel_vm_end); ) { 242 /* 243 * We always write a page, even if it is zero. Each 244 * page written corresponds to 1GB of space 245 */ 246 pmapsize += PAGE_SIZE; 247 ii = pmap_pml4e_index(va); 248 pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii; 249 pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); 250 i = pmap_pdpe_index(va); 251 if ((pdp[i] & PG_V) == 0) { 252 va += NBPDP; 253 continue; 254 } 255 256 /* 257 * 1GB page is represented as 512 2MB pages in a dump. 258 */ 259 if ((pdp[i] & PG_PS) != 0) { 260 va += NBPDP; 261 pa = pdp[i] & PG_PS_FRAME; 262 for (n = 0; n < NPDEPG * NPTEPG; n++) { 263 if (is_dumpable(pa)) 264 dump_add_page(pa); 265 pa += PAGE_SIZE; 266 } 267 continue; 268 } 269 270 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 271 for (n = 0; n < NPDEPG; n++, va += NBPDR) { 272 j = pmap_pde_index(va); 273 274 if ((pd[j] & PG_V) == 0) 275 continue; 276 277 if ((pd[j] & PG_PS) != 0) { 278 /* This is an entire 2M page. */ 279 pa = pd[j] & PG_PS_FRAME; 280 for (k = 0; k < NPTEPG; k++) { 281 if (is_dumpable(pa)) 282 dump_add_page(pa); 283 pa += PAGE_SIZE; 284 } 285 continue; 286 } 287 288 pa = pd[j] & PG_FRAME; 289 /* set bit for this PTE page */ 290 if (is_dumpable(pa)) 291 dump_add_page(pa); 292 /* and for each valid page in this 2MB block */ 293 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 294 for (k = 0; k < NPTEPG; k++) { 295 if ((pt[k] & PG_V) == 0) 296 continue; 297 pa = pt[k] & PG_FRAME; 298 if (is_dumpable(pa)) 299 dump_add_page(pa); 300 } 301 } 302 } 303 304 /* Calculate dump size. */ 305 dumpsize = pmapsize; 306 dumpsize += round_page(msgbufp->msg_size); 307 dumpsize += round_page(vm_page_dump_size); 308 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 309 bits = vm_page_dump[i]; 310 while (bits) { 311 bit = bsfq(bits); 312 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 313 /* Clear out undumpable pages now if needed */ 314 if (is_dumpable(pa)) { 315 dumpsize += PAGE_SIZE; 316 } else { 317 dump_drop_page(pa); 318 } 319 bits &= ~(1ul << bit); 320 } 321 } 322 dumpsize += PAGE_SIZE; 323 324 wdog_next = progress = dumpsize; 325 326 /* Initialize mdhdr */ 327 bzero(&mdhdr, sizeof(mdhdr)); 328 strcpy(mdhdr.magic, MINIDUMP_MAGIC); 329 mdhdr.version = MINIDUMP_VERSION; 330 mdhdr.msgbufsize = msgbufp->msg_size; 331 mdhdr.bitmapsize = vm_page_dump_size; 332 mdhdr.pmapsize = pmapsize; 333 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; 334 mdhdr.dmapbase = DMAP_MIN_ADDRESS; 335 mdhdr.dmapend = DMAP_MAX_ADDRESS; 336 337 dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, 338 dumpsize); 339 340 error = dump_start(di, &kdh); 341 if (error != 0) 342 goto fail; 343 344 printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20, 345 ptoa((uintmax_t)physmem) / 1048576); 346 347 /* Dump my header */ 348 bzero(&fakepd, sizeof(fakepd)); 349 bcopy(&mdhdr, &fakepd, sizeof(mdhdr)); 350 error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE); 351 if (error) 352 goto fail; 353 354 /* Dump msgbuf up front */ 355 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size)); 356 if (error) 357 goto fail; 358 359 /* Dump bitmap */ 360 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size)); 361 if (error) 362 goto fail; 363 364 /* Dump kernel page directory pages */ 365 bzero(fakepd, sizeof(fakepd)); 366 for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + nkpt * NBPDR, 367 kernel_vm_end); va += NBPDP) { 368 ii = pmap_pml4e_index(va); 369 pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii; 370 pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); 371 i = pmap_pdpe_index(va); 372 373 /* We always write a page, even if it is zero */ 374 if ((pdp[i] & PG_V) == 0) { 375 error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE); 376 if (error) 377 goto fail; 378 /* flush, in case we reuse fakepd in the same block */ 379 error = blk_flush(di); 380 if (error) 381 goto fail; 382 continue; 383 } 384 385 /* 1GB page is represented as 512 2MB pages in a dump */ 386 if ((pdp[i] & PG_PS) != 0) { 387 /* PDPE and PDP have identical layout in this case */ 388 fakepd[0] = pdp[i]; 389 for (j = 1; j < NPDEPG; j++) 390 fakepd[j] = fakepd[j - 1] + NBPDR; 391 error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE); 392 if (error) 393 goto fail; 394 /* flush, in case we reuse fakepd in the same block */ 395 error = blk_flush(di); 396 if (error) 397 goto fail; 398 bzero(fakepd, sizeof(fakepd)); 399 continue; 400 } 401 402 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 403 error = blk_write(di, (char *)pd, 0, PAGE_SIZE); 404 if (error) 405 goto fail; 406 error = blk_flush(di); 407 if (error) 408 goto fail; 409 } 410 411 /* Dump memory chunks */ 412 /* XXX cluster it up and use blk_dump() */ 413 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 414 bits = vm_page_dump[i]; 415 while (bits) { 416 bit = bsfq(bits); 417 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 418 error = blk_write(di, 0, pa, PAGE_SIZE); 419 if (error) 420 goto fail; 421 bits &= ~(1ul << bit); 422 } 423 } 424 425 error = blk_flush(di); 426 if (error) 427 goto fail; 428 429 error = dump_finish(di, &kdh); 430 if (error != 0) 431 goto fail; 432 433 printf("\nDump complete\n"); 434 return (0); 435 436 fail: 437 if (error < 0) 438 error = -error; 439 440 printf("\n"); 441 if (error == ENOSPC) { 442 printf("Dump map grown while dumping. "); 443 if (retry_count < dump_retry_count) { 444 printf("Retrying...\n"); 445 goto retry; 446 } 447 printf("Dump failed.\n"); 448 } 449 else if (error == ECANCELED) 450 printf("Dump aborted\n"); 451 else if (error == E2BIG) { 452 printf("Dump failed. Partition too small (about %lluMB were " 453 "needed this time).\n", (long long)dumpsize >> 20); 454 } else 455 printf("** DUMP FAILED (ERROR %d) **\n", error); 456 return (error); 457 } 458 459 void 460 dump_add_page(vm_paddr_t pa) 461 { 462 int idx, bit; 463 464 pa >>= PAGE_SHIFT; 465 idx = pa >> 6; /* 2^6 = 64 */ 466 bit = pa & 63; 467 atomic_set_long(&vm_page_dump[idx], 1ul << bit); 468 } 469 470 void 471 dump_drop_page(vm_paddr_t pa) 472 { 473 int idx, bit; 474 475 pa >>= PAGE_SHIFT; 476 idx = pa >> 6; /* 2^6 = 64 */ 477 bit = pa & 63; 478 atomic_clear_long(&vm_page_dump[idx], 1ul << bit); 479 } 480