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 <sys/kbio.h> 39 #include <vm/vm.h> 40 #include <vm/vm_kern.h> 41 #include <vm/pmap.h> 42 #include <machine/atomic.h> 43 #include <machine/elf.h> 44 #include <machine/globaldata.h> 45 #include <machine/md_var.h> 46 #include <machine/vmparam.h> 47 #include <machine/minidump.h> 48 49 CTASSERT(sizeof(struct kerneldumpheader) == 512); 50 51 /* 52 * Don't touch the first SIZEOF_METADATA bytes on the dump device. This 53 * is to protect us from metadata and to protect metadata from us. 54 */ 55 #define SIZEOF_METADATA (64*1024) 56 57 #define MD_ALIGN(x) (((off_t)(x) + PAGE_MASK) & ~PAGE_MASK) 58 #define DEV_ALIGN(x) roundup2((off_t)(x), DEV_BSIZE) 59 60 extern uint64_t KPDPphys; 61 62 uint64_t *vm_page_dump; 63 vm_offset_t vm_page_dump_size; 64 65 static struct kerneldumpheader kdh; 66 static off_t dumplo; 67 68 /* Handle chunked writes. */ 69 static size_t fragsz; 70 static void *dump_va; 71 static size_t counter, progress; 72 73 CTASSERT(sizeof(*vm_page_dump) == 8); 74 75 static int 76 is_dumpable(vm_paddr_t pa) 77 { 78 int i; 79 80 for (i = 0; dump_avail[i].phys_beg || dump_avail[i].phys_end; ++i) { 81 if (pa >= dump_avail[i].phys_beg && pa < dump_avail[i].phys_end) 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 = dev_ddump(di->priv, dump_va, 0, dumplo, fragsz); 98 dumplo += fragsz; 99 fragsz = 0; 100 return (error); 101 } 102 103 static int 104 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz) 105 { 106 size_t len; 107 int error, i, c; 108 int max_iosize; 109 110 error = 0; 111 if ((sz & PAGE_MASK)) { 112 kprintf("size not page aligned\n"); 113 return (EINVAL); 114 } 115 if (ptr != NULL && pa != 0) { 116 kprintf("can't have both va and pa!\n"); 117 return (EINVAL); 118 } 119 if (pa != 0 && (((uintptr_t)pa) & PAGE_MASK) != 0) { 120 kprintf("address not page aligned\n"); 121 return (EINVAL); 122 } 123 if (ptr != NULL) { 124 /* 125 * If we're doing a virtual dump, flush any 126 * pre-existing pa pages 127 */ 128 error = blk_flush(di); 129 if (error) 130 return (error); 131 } 132 max_iosize = min(MAXPHYS, di->maxiosize); 133 while (sz) { 134 len = max_iosize - fragsz; 135 if (len > sz) 136 len = sz; 137 counter += len; 138 progress -= len; 139 if (counter >> 24) { 140 kprintf(" %ld", PG2MB(progress >> PAGE_SHIFT)); 141 counter &= (1<<24) - 1; 142 } 143 if (ptr) { 144 /*kprintf("s");*/ 145 error = dev_ddump(di->priv, ptr, 0, dumplo, len); 146 /* kprintf("t");*/ 147 if (error) 148 return (error); 149 dumplo += len; 150 ptr += len; 151 sz -= len; 152 } else { 153 for (i = 0; i < len; i += PAGE_SIZE) { 154 dump_va = pmap_kenter_temporary(pa + i, 155 (i + fragsz) >> PAGE_SHIFT); 156 } 157 smp_invltlb(); 158 fragsz += len; 159 pa += len; 160 sz -= len; 161 if (fragsz == max_iosize) { 162 error = blk_flush(di); 163 if (error) 164 return (error); 165 } 166 } 167 } 168 169 /* Check for user abort. */ 170 c = cncheckc(); 171 if (c == 0x03) 172 return (ECANCELED); 173 if (c != -1 && c != NOKEY) 174 kprintf(" (CTRL-C to abort) "); 175 176 return (0); 177 } 178 179 /* A fake page table page, to avoid having to handle both 4K and 2M pages */ 180 static pt_entry_t fakept[NPTEPG]; 181 182 void 183 minidumpsys(struct dumperinfo *di) 184 { 185 uint64_t dumpsize; 186 uint64_t ptesize; 187 vm_offset_t va; 188 vm_offset_t kern_end; 189 int error; 190 uint64_t bits; 191 uint64_t *pdp, *pd, *pt, pa; 192 int i, j, k, bit; 193 int kpdp, klo, khi; 194 int lpdp = -1; 195 long lpdpttl = 0; 196 struct minidumphdr2 mdhdr; 197 struct mdglobaldata *md; 198 199 cnpoll(TRUE); 200 counter = 0; 201 202 /* 203 * minidump page table format is an array of PD entries (1GB pte's), 204 * representing the entire user and kernel virtual address space 205 * (256TB). 206 * 207 * However, we will only dump the KVM portion of this space. And we 208 * only copy the PDP pages for direct access, the PD and PT pages 209 * will be included in the dump as part of the physical map. 210 */ 211 ptesize = NPML4EPG * NPDPEPG * 8; 212 213 /* 214 * Walk page table pages, set bits in vm_page_dump. 215 * 216 * NOTE: kernel_vm_end can actually be below KERNBASE. 217 * Just use KvaEnd. Also note that loops which go 218 * all the way to the end of the address space might 219 * overflow the loop variable. 220 */ 221 md = (struct mdglobaldata *)globaldata_find(0); 222 223 kern_end = KvaEnd; 224 if (kern_end < (vm_offset_t)&(md[ncpus])) 225 kern_end = (vm_offset_t)&(md[ncpus]); 226 227 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 228 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) { 229 /* 230 * The loop probably overflows a 64-bit int due to NBPDR. 231 */ 232 if (va < VM_MIN_KERNEL_ADDRESS) 233 break; 234 235 /* 236 * KPDPphys[] is relative to VM_MIN_KERNEL_ADDRESS. It 237 * contains NKPML4E PDP pages (so we can get to all kernel 238 * PD entries from this array). 239 */ 240 i = ((va - VM_MIN_KERNEL_ADDRESS) >> PDPSHIFT) & 241 (NPML4EPG * NPDPEPG - 1); 242 if (i != lpdp) { 243 lpdp = i; 244 lpdpttl = 0; 245 } 246 247 /* 248 * Calculate the PD index in the PDP. Each PD represents 1GB. 249 * KVA space can cover multiple PDP pages. The PDP array 250 * has been initialized for the entire kernel address space. 251 * 252 * We include the PD entries in the PDP in the dump 253 */ 254 i = ((va - VM_MIN_KERNEL_ADDRESS) >> PDPSHIFT) & 255 (NPML4EPG * NPDPEPG - 1); 256 if ((pdp[i] & kernel_pmap.pmap_bits[PG_V_IDX]) == 0) 257 continue; 258 259 /* 260 * Add the PD page from the PDP to the dump 261 */ 262 dump_add_page(pdp[i] & PG_FRAME); 263 lpdpttl += PAGE_SIZE; 264 265 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 266 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); 267 if ((pd[j] & (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX])) == 268 (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX])) { 269 /* This is an entire 2M page. */ 270 lpdpttl += PAGE_SIZE * NPTEPG; 271 pa = pd[j] & PG_PS_FRAME; 272 for (k = 0; k < NPTEPG; k++) { 273 if (is_dumpable(pa)) 274 dump_add_page(pa); 275 pa += PAGE_SIZE; 276 } 277 } else if ((pd[j] & kernel_pmap.pmap_bits[PG_V_IDX]) == 278 kernel_pmap.pmap_bits[PG_V_IDX]) { 279 /* 280 * Add the PT page from the PD to the dump (it is no 281 * longer included in the ptemap. 282 */ 283 dump_add_page(pd[j] & PG_FRAME); 284 lpdpttl += PAGE_SIZE; 285 286 /* set bit for each valid page in this 2MB block */ 287 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 288 for (k = 0; k < NPTEPG; k++) { 289 if ((pt[k] & kernel_pmap.pmap_bits[PG_V_IDX]) == kernel_pmap.pmap_bits[PG_V_IDX]) { 290 pa = pt[k] & PG_FRAME; 291 lpdpttl += PAGE_SIZE; 292 if (is_dumpable(pa)) 293 dump_add_page(pa); 294 } 295 } 296 } else { 297 /* nothing, we're going to dump a null page */ 298 } 299 } 300 301 /* Calculate dump size. */ 302 dumpsize = ptesize; 303 dumpsize += round_page(msgbufp->msg_size); 304 dumpsize += round_page(vm_page_dump_size); 305 306 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 307 bits = vm_page_dump[i]; 308 while (bits) { 309 bit = bsfq(bits); 310 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 311 /* Clear out undumpable pages now if needed */ 312 if (is_dumpable(pa)) { 313 dumpsize += PAGE_SIZE; 314 } else { 315 dump_drop_page(pa); 316 } 317 bits &= ~(1ul << bit); 318 } 319 } 320 dumpsize += PAGE_SIZE; 321 322 /* Determine dump offset on device. */ 323 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) { 324 error = ENOSPC; 325 goto fail; 326 } 327 dumplo = di->mediaoffset + di->mediasize - dumpsize; 328 dumplo -= sizeof(kdh) * 2; 329 progress = dumpsize; 330 331 /* Initialize mdhdr */ 332 bzero(&mdhdr, sizeof(mdhdr)); 333 strcpy(mdhdr.magic, MINIDUMP2_MAGIC); 334 mdhdr.version = MINIDUMP2_VERSION; 335 mdhdr.msgbufsize = msgbufp->msg_size; 336 mdhdr.bitmapsize = vm_page_dump_size; 337 mdhdr.ptesize = ptesize; 338 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; 339 mdhdr.dmapbase = DMAP_MIN_ADDRESS; 340 mdhdr.dmapend = DMAP_MAX_ADDRESS; 341 342 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, 343 dumpsize, di->blocksize); 344 345 kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576); 346 kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20); 347 348 /* Dump leader */ 349 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 350 if (error) 351 goto fail; 352 dumplo += sizeof(kdh); 353 354 /* Dump my header */ 355 bzero(fakept, sizeof(fakept)); 356 bcopy(&mdhdr, fakept, sizeof(mdhdr)); 357 error = blk_write(di, (char *)fakept, 0, PAGE_SIZE); 358 if (error) 359 goto fail; 360 361 /* Dump msgbuf up front */ 362 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size)); 363 if (error) 364 goto fail; 365 366 /* Dump bitmap */ 367 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size)); 368 if (error) 369 goto fail; 370 371 /* 372 * Dump a full PDP array for the entire KVM space, user and kernel. 373 * This is 512*512 1G PD entries (512*512*8 = 2MB). 374 * 375 * The minidump only dumps PD entries related to KVA space. Also 376 * note that pdp[] (aka KPDPphys[]) only covers VM_MIN_KERNEL_ADDRESS 377 * to VM_MAX_KERNEL_ADDRESS. 378 * 379 * The actual KPDPphys[] array covers a KVA space starting at KVA 380 * KPDPPHYS_KVA. 381 * 382 * By dumping a PDP[] array of PDs representing the entire virtual 383 * address space we can expand what we dump in the future. 384 */ 385 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 386 kpdp = (KPDPPHYS_KVA >> PDPSHIFT) & 387 (NPML4EPG * NPDPEPG - 1); 388 klo = (int)(VM_MIN_KERNEL_ADDRESS >> PDPSHIFT) & 389 (NPML4EPG * NPDPEPG - 1); 390 khi = (int)(VM_MAX_KERNEL_ADDRESS >> PDPSHIFT) & 391 (NPML4EPG * NPDPEPG - 1); 392 393 for (i = 0; i < NPML4EPG * NPDPEPG; ++i) { 394 if (i < klo || i > khi) { 395 fakept[i & (NPDPEPG - 1)] = 0; 396 } else { 397 fakept[i & (NPDPEPG - 1)] = pdp[i - kpdp]; 398 } 399 if ((i & (NPDPEPG - 1)) == (NPDPEPG - 1)) { 400 error = blk_write(di, (char *)fakept, 0, PAGE_SIZE); 401 if (error) 402 goto fail; 403 error = blk_flush(di); 404 if (error) 405 goto fail; 406 } 407 } 408 409 /* Dump memory chunks */ 410 /* XXX cluster it up and use blk_dump() */ 411 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 412 bits = vm_page_dump[i]; 413 while (bits) { 414 bit = bsfq(bits); 415 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 416 error = blk_write(di, 0, pa, PAGE_SIZE); 417 if (error) 418 goto fail; 419 bits &= ~(1ul << bit); 420 } 421 } 422 423 error = blk_flush(di); 424 if (error) 425 goto fail; 426 427 /* Dump trailer */ 428 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 429 if (error) 430 goto fail; 431 dumplo += sizeof(kdh); 432 433 /* Signal completion, signoff and exit stage left. */ 434 dev_ddump(di->priv, NULL, 0, 0, 0); 435 kprintf("\nDump complete\n"); 436 cnpoll(FALSE); 437 return; 438 439 fail: 440 cnpoll(FALSE); 441 if (error < 0) 442 error = -error; 443 444 if (error == ECANCELED) 445 kprintf("\nDump aborted\n"); 446 else if (error == ENOSPC) 447 kprintf("\nDump failed. Partition too small.\n"); 448 else 449 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error); 450 } 451 452 void 453 dump_add_page(vm_paddr_t pa) 454 { 455 int idx, bit; 456 457 pa >>= PAGE_SHIFT; 458 idx = pa >> 6; /* 2^6 = 64 */ 459 bit = pa & 63; 460 atomic_set_long(&vm_page_dump[idx], 1ul << bit); 461 } 462 463 void 464 dump_drop_page(vm_paddr_t pa) 465 { 466 int idx, bit; 467 468 pa >>= PAGE_SHIFT; 469 idx = pa >> 6; /* 2^6 = 64 */ 470 bit = pa & 63; 471 atomic_clear_long(&vm_page_dump[idx], 1ul << bit); 472 } 473