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; 107 108 error = 0; 109 if ((sz % PAGE_SIZE) != 0) { 110 kprintf("size not page aligned\n"); 111 return (EINVAL); 112 } 113 if (ptr != NULL && pa != 0) { 114 kprintf("can't have both va and pa!\n"); 115 return (EINVAL); 116 } 117 if (pa != 0 && (((uintptr_t)ptr) % PAGE_SIZE) != 0) { 118 kprintf("address not page aligned\n"); 119 return (EINVAL); 120 } 121 if (ptr != NULL) { 122 /* If we're doing a virtual dump, flush any pre-existing pa pages */ 123 error = blk_flush(di); 124 if (error) 125 return (error); 126 } 127 while (sz) { 128 len = (MAXDUMPPGS * PAGE_SIZE) - fragsz; 129 if (len > sz) 130 len = sz; 131 counter += len; 132 progress -= len; 133 if (counter >> 24) { 134 kprintf(" %ld", PG2MB(progress >> PAGE_SHIFT)); 135 counter &= (1<<24) - 1; 136 } 137 if (ptr) { 138 error = dev_ddump(di->priv, ptr, 0, dumplo, len); 139 if (error) 140 return (error); 141 dumplo += len; 142 ptr += len; 143 sz -= len; 144 } else { 145 for (i = 0; i < len; i += PAGE_SIZE) 146 dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT); 147 fragsz += len; 148 pa += len; 149 sz -= len; 150 if (fragsz == (MAXDUMPPGS * PAGE_SIZE)) { 151 error = blk_flush(di); 152 if (error) 153 return (error); 154 } 155 } 156 157 /* Check for user abort. */ 158 c = cncheckc(); 159 if (c == 0x03) 160 return (ECANCELED); 161 if (c != -1) 162 kprintf(" (CTRL-C to abort) "); 163 } 164 165 return (0); 166 } 167 168 /* A fake page table page, to avoid having to handle both 4K and 2M pages */ 169 static pt_entry_t fakept[NPTEPG]; 170 171 void 172 minidumpsys(struct dumperinfo *di) 173 { 174 uint64_t dumpsize; 175 uint32_t ptesize; 176 vm_offset_t va; 177 vm_offset_t kern_end; 178 int error; 179 uint64_t bits; 180 uint64_t *pdp, *pd, *pt, pa; 181 int i, j, k, bit; 182 struct minidumphdr mdhdr; 183 struct mdglobaldata *md; 184 185 counter = 0; 186 /* 187 * Walk page table pages, set bits in vm_page_dump. 188 * 189 * NOTE: kernel_vm_end can actually be below KERNBASE. 190 * Just use KvaEnd. Also note that loops which go 191 * all the way to the end of the address space might 192 * overflow the loop variable. 193 */ 194 ptesize = 0; 195 196 md = (struct mdglobaldata *)globaldata_find(0); 197 198 kern_end = KvaEnd; 199 if (kern_end < (vm_offset_t)&(md[ncpus])) 200 kern_end = (vm_offset_t)&(md[ncpus]); 201 202 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 203 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) { 204 /* 205 * The loop probably overflows a 64-bit int due to NBPDR. 206 */ 207 if (va < VM_MIN_KERNEL_ADDRESS) 208 break; 209 210 /* 211 * We always write a page, even if it is zero. Each 212 * page written corresponds to 2MB of space 213 */ 214 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1); 215 ptesize += PAGE_SIZE; 216 if ((pdp[i] & PG_V) == 0) 217 continue; 218 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 219 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); 220 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) { 221 /* This is an entire 2M page. */ 222 pa = pd[j] & PG_PS_FRAME; 223 for (k = 0; k < NPTEPG; k++) { 224 if (is_dumpable(pa)) 225 dump_add_page(pa); 226 pa += PAGE_SIZE; 227 } 228 continue; 229 } 230 if ((pd[j] & PG_V) == PG_V) { 231 /* set bit for each valid page in this 2MB block */ 232 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 233 for (k = 0; k < NPTEPG; k++) { 234 if ((pt[k] & PG_V) == PG_V) { 235 pa = pt[k] & PG_FRAME; 236 if (is_dumpable(pa)) 237 dump_add_page(pa); 238 } 239 } 240 } else { 241 /* nothing, we're going to dump a null page */ 242 } 243 } 244 245 /* Calculate dump size. */ 246 dumpsize = ptesize; 247 dumpsize += round_page(msgbufp->msg_size); 248 dumpsize += round_page(vm_page_dump_size); 249 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 250 bits = vm_page_dump[i]; 251 while (bits) { 252 bit = bsfq(bits); 253 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 254 /* Clear out undumpable pages now if needed */ 255 if (is_dumpable(pa)) { 256 dumpsize += PAGE_SIZE; 257 } else { 258 dump_drop_page(pa); 259 } 260 bits &= ~(1ul << bit); 261 } 262 } 263 dumpsize += PAGE_SIZE; 264 265 /* Determine dump offset on device. */ 266 if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) { 267 error = ENOSPC; 268 goto fail; 269 } 270 dumplo = di->mediaoffset + di->mediasize - dumpsize; 271 dumplo -= sizeof(kdh) * 2; 272 progress = dumpsize; 273 274 /* Initialize mdhdr */ 275 bzero(&mdhdr, sizeof(mdhdr)); 276 strcpy(mdhdr.magic, MINIDUMP_MAGIC); 277 mdhdr.version = MINIDUMP_VERSION; 278 mdhdr.msgbufsize = msgbufp->msg_size; 279 mdhdr.bitmapsize = vm_page_dump_size; 280 mdhdr.ptesize = ptesize; 281 mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; 282 mdhdr.dmapbase = DMAP_MIN_ADDRESS; 283 mdhdr.dmapend = DMAP_MAX_ADDRESS; 284 285 mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, 286 dumpsize, di->blocksize); 287 288 kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576); 289 kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20); 290 291 /* Dump leader */ 292 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 293 if (error) 294 goto fail; 295 dumplo += sizeof(kdh); 296 297 /* Dump my header */ 298 bzero(&fakept, sizeof(fakept)); 299 bcopy(&mdhdr, &fakept, sizeof(mdhdr)); 300 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 301 if (error) 302 goto fail; 303 304 /* Dump msgbuf up front */ 305 error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size)); 306 if (error) 307 goto fail; 308 309 /* Dump bitmap */ 310 error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size)); 311 if (error) 312 goto fail; 313 314 /* Dump kernel page table pages */ 315 pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys); 316 for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) { 317 /* 318 * The loop probably overflows a 64-bit int due to NBPDR. 319 */ 320 if (va < VM_MIN_KERNEL_ADDRESS) 321 break; 322 323 /* 324 * We always write a page, even if it is zero 325 */ 326 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1); 327 if ((pdp[i] & PG_V) == 0) { 328 bzero(fakept, sizeof(fakept)); 329 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 330 if (error) 331 goto fail; 332 /* flush, in case we reuse fakept in the same block */ 333 error = blk_flush(di); 334 if (error) 335 goto fail; 336 continue; 337 } 338 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME); 339 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1)); 340 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V)) { 341 /* This is a single 2M block. Generate a fake PTP */ 342 pa = pd[j] & PG_PS_FRAME; 343 for (k = 0; k < NPTEPG; k++) { 344 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M; 345 } 346 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 347 if (error) 348 goto fail; 349 /* flush, in case we reuse fakept in the same block */ 350 error = blk_flush(di); 351 if (error) 352 goto fail; 353 continue; 354 } 355 if ((pd[j] & PG_V) == PG_V) { 356 pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME); 357 error = blk_write(di, (char *)pt, 0, PAGE_SIZE); 358 if (error) 359 goto fail; 360 } else { 361 bzero(fakept, sizeof(fakept)); 362 error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE); 363 if (error) 364 goto fail; 365 /* flush, in case we reuse fakept in the same block */ 366 error = blk_flush(di); 367 if (error) 368 goto fail; 369 } 370 } 371 372 /* Dump memory chunks */ 373 /* XXX cluster it up and use blk_dump() */ 374 for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) { 375 bits = vm_page_dump[i]; 376 while (bits) { 377 bit = bsfq(bits); 378 pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE; 379 error = blk_write(di, 0, pa, PAGE_SIZE); 380 if (error) 381 goto fail; 382 bits &= ~(1ul << bit); 383 } 384 } 385 386 error = blk_flush(di); 387 if (error) 388 goto fail; 389 390 /* Dump trailer */ 391 error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh)); 392 if (error) 393 goto fail; 394 dumplo += sizeof(kdh); 395 396 /* Signal completion, signoff and exit stage left. */ 397 dev_ddump(di->priv, NULL, 0, 0, 0); 398 kprintf("\nDump complete\n"); 399 return; 400 401 fail: 402 if (error < 0) 403 error = -error; 404 405 if (error == ECANCELED) 406 kprintf("\nDump aborted\n"); 407 else if (error == ENOSPC) 408 kprintf("\nDump failed. Partition too small.\n"); 409 else 410 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error); 411 } 412 413 void 414 dump_add_page(vm_paddr_t pa) 415 { 416 int idx, bit; 417 418 pa >>= PAGE_SHIFT; 419 idx = pa >> 6; /* 2^6 = 64 */ 420 bit = pa & 63; 421 atomic_set_long(&vm_page_dump[idx], 1ul << bit); 422 } 423 424 void 425 dump_drop_page(vm_paddr_t pa) 426 { 427 int idx, bit; 428 429 pa >>= PAGE_SHIFT; 430 idx = pa >> 6; /* 2^6 = 64 */ 431 bit = pa & 63; 432 atomic_clear_long(&vm_page_dump[idx], 1ul << bit); 433 } 434