sys/platform/pc64/x86_64/minidump_machdep.c

   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 int 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         uint32_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         struct minidumphdr mdhdr;
 194         struct mdglobaldata *md;
 195
 196         cnpoll(TRUE);
 197         counter = 0;
 198         /*
 199          * Walk page table pages, set bits in vm_page_dump.
 200          *
 201          * NOTE: kernel_vm_end can actually be below KERNBASE.
 202          *       Just use KvaEnd.  Also note that loops which go
 203          *       all the way to the end of the address space might
 204          *       overflow the loop variable.
 205          */
 206         ptesize = 0;
 207
 208         md = (struct mdglobaldata *)globaldata_find(0);
 209
 210         kern_end = KvaEnd;
 211         if (kern_end < (vm_offset_t)&(md[ncpus]))
 212                 kern_end = (vm_offset_t)&(md[ncpus]);
 213
 214         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
 215         for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
 216                 /*
 217                  * The loop probably overflows a 64-bit int due to NBPDR.
 218                  */
 219                 if (va < VM_MIN_KERNEL_ADDRESS)
 220                         break;
 221
 222                 /*
 223                  * We always write a page, even if it is zero. Each
 224                  * page written corresponds to 2MB of space
 225                  */
 226                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
 227                 ptesize += PAGE_SIZE;
 228                 if ((pdp[i] & kernel_pmap.pmap_bits[PG_V_IDX]) == 0)
 229                         continue;
 230                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
 231                 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
 232                 if ((pd[j] & (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX])) ==
 233                     (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX]))  {
 234                         /* This is an entire 2M page. */
 235                         pa = pd[j] & PG_PS_FRAME;
 236                         for (k = 0; k < NPTEPG; k++) {
 237                                 if (is_dumpable(pa))
 238                                         dump_add_page(pa);
 239                                 pa += PAGE_SIZE;
 240                         }
 241                         continue;
 242                 }
 243                 if ((pd[j] & kernel_pmap.pmap_bits[PG_V_IDX]) == kernel_pmap.pmap_bits[PG_V_IDX]) {
 244                         /* set bit for each valid page in this 2MB block */
 245                         pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
 246                         for (k = 0; k < NPTEPG; k++) {
 247                                 if ((pt[k] & kernel_pmap.pmap_bits[PG_V_IDX]) == kernel_pmap.pmap_bits[PG_V_IDX]) {
 248                                         pa = pt[k] & PG_FRAME;
 249                                         if (is_dumpable(pa))
 250                                                 dump_add_page(pa);
 251                                 }
 252                         }
 253                 } else {
 254                         /* nothing, we're going to dump a null page */
 255                 }
 256         }
 257
 258         /* Calculate dump size. */
 259         dumpsize = ptesize;
 260         dumpsize += round_page(msgbufp->msg_size);
 261         dumpsize += round_page(vm_page_dump_size);
 262         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
 263                 bits = vm_page_dump[i];
 264                 while (bits) {
 265                         bit = bsfq(bits);
 266                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
 267                         /* Clear out undumpable pages now if needed */
 268                         if (is_dumpable(pa)) {
 269                                 dumpsize += PAGE_SIZE;
 270                         } else {
 271                                 dump_drop_page(pa);
 272                         }
 273                         bits &= ~(1ul << bit);
 274                 }
 275         }
 276         dumpsize += PAGE_SIZE;
 277
 278         /* Determine dump offset on device. */
 279         if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
 280                 error = ENOSPC;
 281                 goto fail;
 282         }
 283         dumplo = di->mediaoffset + di->mediasize - dumpsize;
 284         dumplo -= sizeof(kdh) * 2;
 285         progress = dumpsize;
 286
 287         /* Initialize mdhdr */
 288         bzero(&mdhdr, sizeof(mdhdr));
 289         strcpy(mdhdr.magic, MINIDUMP_MAGIC);
 290         mdhdr.version = MINIDUMP_VERSION;
 291         mdhdr.msgbufsize = msgbufp->msg_size;
 292         mdhdr.bitmapsize = vm_page_dump_size;
 293         mdhdr.ptesize = ptesize;
 294         mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
 295         mdhdr.dmapbase = DMAP_MIN_ADDRESS;
 296         mdhdr.dmapend = DMAP_MAX_ADDRESS;
 297
 298         mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION,
 299             dumpsize, di->blocksize);
 300
 301         kprintf("Physical memory: %jd MB\n", (intmax_t)ptoa(physmem) / 1048576);
 302         kprintf("Dumping %jd MB:", (intmax_t)dumpsize >> 20);
 303
 304         /* Dump leader */
 305         error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
 306         if (error)
 307                 goto fail;
 308         dumplo += sizeof(kdh);
 309
 310         /* Dump my header */
 311         bzero(&fakept, sizeof(fakept));
 312         bcopy(&mdhdr, &fakept, sizeof(mdhdr));
 313         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 314         if (error)
 315                 goto fail;
 316
 317         /* Dump msgbuf up front */
 318         error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
 319         if (error)
 320                 goto fail;
 321
 322         /* Dump bitmap */
 323         error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
 324         if (error)
 325                 goto fail;
 326
 327         /* Dump kernel page table pages */
 328         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
 329         for (va = VM_MIN_KERNEL_ADDRESS; va < kern_end; va += NBPDR) {
 330                 /*
 331                  * The loop probably overflows a 64-bit int due to NBPDR.
 332                  */
 333                 if (va < VM_MIN_KERNEL_ADDRESS)
 334                         break;
 335
 336                 /*
 337                  * We always write a page, even if it is zero
 338                  */
 339                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
 340                 if ((pdp[i] & kernel_pmap.pmap_bits[PG_V_IDX]) == 0) {
 341                         bzero(fakept, sizeof(fakept));
 342                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 343                         if (error)
 344                                 goto fail;
 345                         /* flush, in case we reuse fakept in the same block */
 346                         error = blk_flush(di);
 347                         if (error)
 348                                 goto fail;
 349                         continue;
 350                 }
 351                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
 352                 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
 353                 if ((pd[j] & (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX])) ==
 354                     (kernel_pmap.pmap_bits[PG_PS_IDX] | kernel_pmap.pmap_bits[PG_V_IDX]))  {
 355                         /* This is a single 2M block. Generate a fake PTP */
 356                         pa = pd[j] & PG_PS_FRAME;
 357                         for (k = 0; k < NPTEPG; k++) {
 358                                 fakept[k] = (pa + (k * PAGE_SIZE)) |
 359                                     kernel_pmap.pmap_bits[PG_V_IDX] |
 360                                     kernel_pmap.pmap_bits[PG_RW_IDX] |
 361                                     kernel_pmap.pmap_bits[PG_A_IDX] |
 362                                     kernel_pmap.pmap_bits[PG_M_IDX];
 363                         }
 364                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 365                         if (error)
 366                                 goto fail;
 367                         /* flush, in case we reuse fakept in the same block */
 368                         error = blk_flush(di);
 369                         if (error)
 370                                 goto fail;
 371                         continue;
 372                 }
 373                 if ((pd[j] & kernel_pmap.pmap_bits[PG_V_IDX]) == kernel_pmap.pmap_bits[PG_V_IDX]) {
 374                         pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
 375                         error = blk_write(di, (char *)pt, 0, PAGE_SIZE);
 376                         if (error)
 377                                 goto fail;
 378                 } else {
 379                         bzero(fakept, sizeof(fakept));
 380                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 381                         if (error)
 382                                 goto fail;
 383                         /* flush, in case we reuse fakept in the same block */
 384                         error = blk_flush(di);
 385                         if (error)
 386                                 goto fail;
 387                 }
 388         }
 389
 390         /* Dump memory chunks */
 391         /* XXX cluster it up and use blk_dump() */
 392         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
 393                 bits = vm_page_dump[i];
 394                 while (bits) {
 395                         bit = bsfq(bits);
 396                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
 397                         error = blk_write(di, 0, pa, PAGE_SIZE);
 398                         if (error)
 399                                 goto fail;
 400                         bits &= ~(1ul << bit);
 401                 }
 402         }
 403
 404         error = blk_flush(di);
 405         if (error)
 406                 goto fail;
 407
 408         /* Dump trailer */
 409         error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
 410         if (error)
 411                 goto fail;
 412         dumplo += sizeof(kdh);
 413
 414         /* Signal completion, signoff and exit stage left. */
 415         dev_ddump(di->priv, NULL, 0, 0, 0);
 416         kprintf("\nDump complete\n");
 417         cnpoll(FALSE);
 418         return;
 419
 420  fail:
 421         cnpoll(FALSE);
 422         if (error < 0)
 423                 error = -error;
 424
 425         if (error == ECANCELED)
 426                 kprintf("\nDump aborted\n");
 427         else if (error == ENOSPC)
 428                 kprintf("\nDump failed. Partition too small.\n");
 429         else
 430                 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error);
 431 }
 432
 433 void
 434 dump_add_page(vm_paddr_t pa)
 435 {
 436         int idx, bit;
 437
 438         pa >>= PAGE_SHIFT;
 439         idx = pa >> 6;          /* 2^6 = 64 */
 440         bit = pa & 63;
 441         atomic_set_long(&vm_page_dump[idx], 1ul << bit);
 442 }
 443
 444 void
 445 dump_drop_page(vm_paddr_t pa)
 446 {
 447         int idx, bit;
 448
 449         pa >>= PAGE_SHIFT;
 450         idx = pa >> 6;          /* 2^6 = 64 */
 451         bit = pa & 63;
 452         atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
 453 }