1 /*- 2 * Copyright (c) 2006 Peter Wemm 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 * 26 * $FreeBSD: src/sys/amd64/amd64/minidump_machdep.c,v 1.10 2009/05/29 21:27:12 jamie Exp $ 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/conf.h> 32 #include <sys/cons.h> 33 #include <sys/device.h> 34 #include <sys/globaldata.h> 35 #include <sys/kernel.h> 36 #include <sys/kerneldump.h> 37 #include <sys/msgbuf.h> 38 #include <vm/vm.h> 39 #include <vm/vm_kern.h> 40 #include <vm/pmap.h> 41 #include <machine/atomic.h> 42 #include <machine/elf.h> 43 #include <machine/globaldata.h> 44 #include <machine/md_var.h> 45 #include <machine/vmparam.h> 46 #include <machine/minidump.h> 47 48 CTASSERT(sizeof(struct kerneldumpheader) == 512); 49 50 /* 51 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This 52 * is to protect us from metadata and to protect metadata from us. 53 */ 54 #define SIZEOF_METADATA (64*1024) 55 56 #define MD_ALIGN(x) (((off_t)(x) + PAGE_MASK) & ~PAGE_MASK) 57 #define DEV_ALIGN(x) (((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1)) 58 59 extern uint64_t KPDPphys; 60 61 uint64_t *vm_page_dump; 62 int vm_page_dump_size; 63 64 static struct kerneldumpheader kdh; 65 static off_t dumplo; 66 67 /* Handle chunked writes. */ 68 static size_t fragsz; 69 static void *dump_va; 70 static size_t counter, progress; 71 72 CTASSERT(sizeof(*vm_page_dump) == 8); 73 74 static int 75 is_dumpable(vm_paddr_t pa) 76 { 77 int i; 78 79 for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) { 80 if (pa >= dump_avail[i] && pa < dump_avail[i + 1]) 81 return (1); 82 } 83 return (0); 84 } 85 86 #define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8) 87 88 static int 89 blk_flush(struct dumperinfo *di) 90 { 91 int error; 92 93 if (fragsz == 0) 94 return (0); 95 96 error = dev_ddump(di->priv, dump_va, 0, dumplo, fragsz); 97 dumplo += fragsz; 98 fragsz = 0; 99 return (error); 100 } 101 102 static int 103 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz) 104 { 105 size_t len; 106 int error, i, c; 107 int max_iosize; 108 109 error = 0; 110 if ((sz & PAGE_MASK)) { 111 kprintf("size not page aligned\n"); 112 return (EINVAL); 113 } 114 if (ptr != NULL && pa != 0) { 115 kprintf("can't have both va and pa!\n"); 116 return (EINVAL); 117 } 118 if (pa != 0 && (((uintptr_t)pa) & PAGE_MASK) != 0) { 119 kprintf("address not page aligned\n"); 120 return (EINVAL); 121 } 122 if (ptr != NULL) { 123 /* 124 * If we're doing a virtual dump, flush any 125 * pre-existing pa pages 126 */ 127 error = blk_flush(di); 128 if (error) 129 return (error); 130 } 131 max_iosize = min(MAXPHYS, di->maxiosize); 132 while (sz) { 133 len = max_iosize - fragsz; 134 if (len > sz) 135 len = sz; 136 counter += len; 137 progress -= len; 138 if (counter >> 24) { 139 kprintf(" %ld", PG2MB(progress >> PAGE_SHIFT)); 140 counter &= (1<<24) - 1; 141 } 142 if (ptr) { 143 /*kprintf("s");*/ 144 error = dev_ddump(di->priv, ptr, 0, dumplo, len); 145 /* kprintf("t");*/ 146 if (error) 147 return (error); 148 dumplo += len; 149 ptr += len; 150 sz -= len; 151 } else { 152 for (i = 0; i < len; i += PAGE_SIZE) { 153 dump_va = pmap_kenter_temporary(pa + i, 154 (i + fragsz) >> PAGE_SHIFT); 155 } 156 smp_invltlb(); 157 fragsz += len; 158 pa += len; 159 sz -= len; 160 if (fragsz == max_iosize) { 161 error = blk_flush(di); 162 if (error) 163 return (error); 164 } 165 } 166 } 167 168 /* Check for user abort. */ 169 c = cncheckc(); 170 if (c == 0x03) 171 return (ECANCELED); 172 if (c != -1) 173 kprintf(" (CTRL-C to abort) "); 174 175 return (0); 176 } 177 178 /* A fake page table page, to avoid having to handle both 4K and 2M pages */ 179 static pt_entry_t fakept[NPTEPG]; 180 181 void 182 minidumpsys(struct dumperinfo *di) 183 { 184 uint64_t dumpsize; 185 uint32_t ptesize; 186 vm_offset_t va; 187 vm_offset_t kern_end; 188 int error; 189 uint64_t bits; 190 uint64_t *pdp, *pd, *pt, pa; 191 int i, j, k, bit; 192 struct minidumphdr mdhdr; 193 struct mdglobaldata *md; 194 195 counter = 0; 196 /* 197 * Walk page table pages, set bits in vm_page_dump. 198 * 199 * NOTE: kernel_vm_end can actually be below KERNBASE. 200 * Just use KvaEnd. Also note that loops which go 201 * all the way to the end of the address space might 202 * overflow the loop variable. 203 */ 204 ptesize = 0; 205 206 md = (struct mdglobaldata *)globaldata_find(0); 207 208 kern_end = KvaEnd; 209 if (kern_end < (vm_offset_t)&(md[ncpus])) 210 kern_end = (vm_offset_t)&(md[ncpus]); 211 212 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 213 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) { 214 /* 215 * The loop probably overflows a 64-bit int due to NBPDR. 216 */ 217 if (va < VM_MIN_KERNEL_ADDRESS) 218 break; 219 220 /* 221 * We always write a page, even if it is zero. Each 222 * page written corresponds to 2MB of space 223 */ 224 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1); 225 ptesize += PAGE_SIZE; 226 if ((pdp[i] & PG_V) == 0) 227 continue; 228 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 229 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); 230 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) { 231 /* This is an entire 2M page. */ 232 pa = pd[j] & PG_PS_FRAME; 233 for (k = 0; k < NPTEPG; k++) { 234 if (is_dumpable(pa)) 235 dump_add_page(pa); 236 pa += PAGE_SIZE; 237 } 238 continue; 239 } 240 if ((pd[j] & PG_V) == PG_V) { 241 /* set bit for each valid page in this 2MB block */ 242 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 243 for (k = 0; k < NPTEPG; k++) { 244 if ((pt[k] & PG_V) == PG_V) { 245 pa = pt[k] & PG_FRAME; 246 if (is_dumpable(pa)) 247 dump_add_page(pa); 248 } 249 } 250 } else { 251 /* nothing, we're going to dump a null page */ 252 } 253 } 254 255 /* Calculate dump size. */ 256 dumpsize = ptesize; 257 dumpsize += round_page(msgbufp->msg_size); 258 dumpsize += round_page(vm_page_dump_size); 259 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 260 bits = vm_page_dump[i]; 261 while (bits) { 262 bit = bsfq(bits); 263 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 264 /* Clear out undumpable pages now if needed */ 265 if (is_dumpable(pa)) { 266 dumpsize += PAGE_SIZE; 267 } else { 268 dump_drop_page(pa); 269 } 270 bits &= ~(1ul << bit); 271 } 272 } 273 dumpsize += PAGE_SIZE; 274 275 /* Determine dump offset on device. */ 276 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) { 277 error = ENOSPC; 278 goto fail; 279 } 280 dumplo = di->mediaoffset + di->mediasize - dumpsize; 281 dumplo -= sizeof(kdh) * 2; 282 progress = dumpsize; 283 284 /* Initialize mdhdr */ 285 bzero(&mdhdr, sizeof(mdhdr)); 286 strcpy(mdhdr.magic, MINIDUMP_MAGIC); 287 mdhdr.version = MINIDUMP_VERSION; 288 mdhdr.msgbufsize = msgbufp->msg_size; 289 mdhdr.bitmapsize = vm_page_dump_size; 290 mdhdr.ptesize = ptesize; 291 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; 292 mdhdr.dmapbase = DMAP_MIN_ADDRESS; 293 mdhdr.dmapend = DMAP_MAX_ADDRESS; 294 295 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, 296 dumpsize, di->blocksize); 297 298 kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576); 299 kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20); 300 301 /* Dump leader */ 302 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 303 if (error) 304 goto fail; 305 dumplo += sizeof(kdh); 306 307 /* Dump my header */ 308 bzero(&fakept, sizeof(fakept)); 309 bcopy(&mdhdr, &fakept, sizeof(mdhdr)); 310 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 311 if (error) 312 goto fail; 313 314 /* Dump msgbuf up front */ 315 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size)); 316 if (error) 317 goto fail; 318 319 /* Dump bitmap */ 320 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size)); 321 if (error) 322 goto fail; 323 324 /* Dump kernel page table pages */ 325 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 326 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) { 327 /* 328 * The loop probably overflows a 64-bit int due to NBPDR. 329 */ 330 if (va < VM_MIN_KERNEL_ADDRESS) 331 break; 332 333 /* 334 * We always write a page, even if it is zero 335 */ 336 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1); 337 if ((pdp[i] & PG_V) == 0) { 338 bzero(fakept, sizeof(fakept)); 339 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 340 if (error) 341 goto fail; 342 /* flush, in case we reuse fakept in the same block */ 343 error = blk_flush(di); 344 if (error) 345 goto fail; 346 continue; 347 } 348 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 349 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); 350 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) { 351 /* This is a single 2M block. Generate a fake PTP */ 352 pa = pd[j] & PG_PS_FRAME; 353 for (k = 0; k < NPTEPG; k++) { 354 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M; 355 } 356 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 357 if (error) 358 goto fail; 359 /* flush, in case we reuse fakept in the same block */ 360 error = blk_flush(di); 361 if (error) 362 goto fail; 363 continue; 364 } 365 if ((pd[j] & PG_V) == PG_V) { 366 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 367 error = blk_write(di, (char *)pt, 0, PAGE_SIZE); 368 if (error) 369 goto fail; 370 } else { 371 bzero(fakept, sizeof(fakept)); 372 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 373 if (error) 374 goto fail; 375 /* flush, in case we reuse fakept in the same block */ 376 error = blk_flush(di); 377 if (error) 378 goto fail; 379 } 380 } 381 382 /* Dump memory chunks */ 383 /* XXX cluster it up and use blk_dump() */ 384 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 385 bits = vm_page_dump[i]; 386 while (bits) { 387 bit = bsfq(bits); 388 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 389 error = blk_write(di, 0, pa, PAGE_SIZE); 390 if (error) 391 goto fail; 392 bits &= ~(1ul << bit); 393 } 394 } 395 396 error = blk_flush(di); 397 if (error) 398 goto fail; 399 400 /* Dump trailer */ 401 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 402 if (error) 403 goto fail; 404 dumplo += sizeof(kdh); 405 406 /* Signal completion, signoff and exit stage left. */ 407 dev_ddump(di->priv, NULL, 0, 0, 0); 408 kprintf("\nDump complete\n"); 409 return; 410 411 fail: 412 if (error < 0) 413 error = -error; 414 415 if (error == ECANCELED) 416 kprintf("\nDump aborted\n"); 417 else if (error == ENOSPC) 418 kprintf("\nDump failed. Partition too small.\n"); 419 else 420 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error); 421 } 422 423 void 424 dump_add_page(vm_paddr_t pa) 425 { 426 int idx, bit; 427 428 pa >>= PAGE_SHIFT; 429 idx = pa >> 6; /* 2^6 = 64 */ 430 bit = pa & 63; 431 atomic_set_long(&vm_page_dump[idx], 1ul << bit); 432 } 433 434 void 435 dump_drop_page(vm_paddr_t pa) 436 { 437 int idx, bit; 438 439 pa >>= PAGE_SHIFT; 440 idx = pa >> 6; /* 2^6 = 64 */ 441 bit = pa & 63; 442 atomic_clear_long(&vm_page_dump[idx], 1ul << bit); 443 } 444