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 <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("cant 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         /* Walk page table pages, set bits in vm_page_dump */
 187         ptesize = 0;
 188
 189         md = (struct mdglobaldata *)globaldata_find(0);
 190
 191         kern_end = kernel_vm_end;
 192         if (kern_end < (vm_offset_t)&(md[ncpus]))
 193                 kern_end = (vm_offset_t)&(md[ncpus]);
 194
 195         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
 196         for (va = KERNBASE; va < kern_end; va += NBPDR) {
 197                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
 198                 /*
 199                  * We always write a page, even if it is zero. Each
 200                  * page written corresponds to 2MB of space
 201                  */
 202                 ptesize += PAGE_SIZE;
 203                 if ((pdp[i] & PG_V) == 0)
 204                         continue;
 205                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
 206                 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
 207                 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
 208                         /* This is an entire 2M page. */
 209                         pa = pd[j] & PG_PS_FRAME;
 210                         for (k = 0; k < NPTEPG; k++) {
 211                                 if (is_dumpable(pa))
 212                                         dump_add_page(pa);
 213                                 pa += PAGE_SIZE;
 214                         }
 215                         continue;
 216                 }
 217                 if ((pd[j] & PG_V) == PG_V) {
 218                         /* set bit for each valid page in this 2MB block */
 219                         pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
 220                         for (k = 0; k < NPTEPG; k++) {
 221                                 if ((pt[k] & PG_V) == PG_V) {
 222                                         pa = pt[k] & PG_FRAME;
 223                                         if (is_dumpable(pa))
 224                                                 dump_add_page(pa);
 225                                 }
 226                         }
 227                 } else {
 228                         /* nothing, we're going to dump a null page */
 229                 }
 230         }
 231
 232         /* Calculate dump size. */
 233         dumpsize = ptesize;
 234         dumpsize += round_page(msgbufp->msg_size);
 235         dumpsize += round_page(vm_page_dump_size);
 236         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
 237                 bits = vm_page_dump[i];
 238                 while (bits) {
 239                         bit = bsfq(bits);
 240                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
 241                         /* Clear out undumpable pages now if needed */
 242                         if (is_dumpable(pa)) {
 243                                 dumpsize += PAGE_SIZE;
 244                         } else {
 245                                 dump_drop_page(pa);
 246                         }
 247                         bits &= ~(1ul << bit);
 248                 }
 249         }
 250         dumpsize += PAGE_SIZE;
 251
 252         /* Determine dump offset on device. */
 253         if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
 254                 error = ENOSPC;
 255                 goto fail;
 256         }
 257         dumplo = di->mediaoffset + di->mediasize - dumpsize;
 258         dumplo -= sizeof(kdh) * 2;
 259         progress = dumpsize;
 260
 261         /* Initialize mdhdr */
 262         bzero(&mdhdr, sizeof(mdhdr));
 263         strcpy(mdhdr.magic, MINIDUMP_MAGIC);
 264         mdhdr.version = MINIDUMP_VERSION;
 265         mdhdr.msgbufsize = msgbufp->msg_size;
 266         mdhdr.bitmapsize = vm_page_dump_size;
 267         mdhdr.ptesize = ptesize;
 268         mdhdr.kernbase = KERNBASE;
 269         mdhdr.dmapbase = DMAP_MIN_ADDRESS;
 270         mdhdr.dmapend = DMAP_MAX_ADDRESS;
 271
 272         mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION,
 273             dumpsize, di->blocksize);
 274
 275         kprintf("Physical memory: %ju MB\n", ptoa((uintmax_t)physmem) / 1048576);
 276         kprintf("Dumping %llu MB:", (long long)dumpsize >> 20);
 277
 278         /* Dump leader */
 279         error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
 280         if (error)
 281                 goto fail;
 282         dumplo += sizeof(kdh);
 283
 284         /* Dump my header */
 285         bzero(&fakept, sizeof(fakept));
 286         bcopy(&mdhdr, &fakept, sizeof(mdhdr));
 287         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 288         if (error)
 289                 goto fail;
 290
 291         /* Dump msgbuf up front */
 292         error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
 293         if (error)
 294                 goto fail;
 295
 296         /* Dump bitmap */
 297         error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
 298         if (error)
 299                 goto fail;
 300
 301         /* Dump kernel page table pages */
 302         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
 303         for (va = KERNBASE; va < kern_end; va += NBPDR) {
 304                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
 305                 /* We always write a page, even if it is zero */
 306                 if ((pdp[i] & PG_V) == 0) {
 307                         bzero(fakept, sizeof(fakept));
 308                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 309                         if (error)
 310                                 goto fail;
 311                         /* flush, in case we reuse fakept in the same block */
 312                         error = blk_flush(di);
 313                         if (error)
 314                                 goto fail;
 315                         continue;
 316                 }
 317                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
 318                 j = ((va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1));
 319                 if ((pd[j] & (PG_PS | PG_V)) == (PG_PS | PG_V))  {
 320                         /* This is a single 2M block. Generate a fake PTP */
 321                         pa = pd[j] & PG_PS_FRAME;
 322                         for (k = 0; k < NPTEPG; k++) {
 323                                 fakept[k] = (pa + (k * PAGE_SIZE)) | PG_V | PG_RW | PG_A | PG_M;
 324                         }
 325                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 326                         if (error)
 327                                 goto fail;
 328                         /* flush, in case we reuse fakept in the same block */
 329                         error = blk_flush(di);
 330                         if (error)
 331                                 goto fail;
 332                         continue;
 333                 }
 334                 if ((pd[j] & PG_V) == PG_V) {
 335                         pt = (uint64_t *)PHYS_TO_DMAP(pd[j] & PG_FRAME);
 336                         error = blk_write(di, (char *)pt, 0, PAGE_SIZE);
 337                         if (error)
 338                                 goto fail;
 339                 } else {
 340                         bzero(fakept, sizeof(fakept));
 341                         error = blk_write(di, (char *)&fakept, 0, PAGE_SIZE);
 342                         if (error)
 343                                 goto fail;
 344                         /* flush, in case we reuse fakept in the same block */
 345                         error = blk_flush(di);
 346                         if (error)
 347                                 goto fail;
 348                 }
 349         }
 350
 351         /* Dump memory chunks */
 352         /* XXX cluster it up and use blk_dump() */
 353         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
 354                 bits = vm_page_dump[i];
 355                 while (bits) {
 356                         bit = bsfq(bits);
 357                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
 358                         error = blk_write(di, 0, pa, PAGE_SIZE);
 359                         if (error)
 360                                 goto fail;
 361                         bits &= ~(1ul << bit);
 362                 }
 363         }
 364
 365         error = blk_flush(di);
 366         if (error)
 367                 goto fail;
 368
 369         /* Dump trailer */
 370         error = dev_ddump(di->priv, &kdh, 0, dumplo, sizeof(kdh));
 371         if (error)
 372                 goto fail;
 373         dumplo += sizeof(kdh);
 374
 375         /* Signal completion, signoff and exit stage left. */
 376         dev_ddump(di->priv, NULL, 0, 0, 0);
 377         kprintf("\nDump complete\n");
 378         return;
 379
 380  fail:
 381         if (error < 0)
 382                 error = -error;
 383
 384         if (error == ECANCELED)
 385                 kprintf("\nDump aborted\n");
 386         else if (error == ENOSPC)
 387                 kprintf("\nDump failed. Partition too small.\n");
 388         else
 389                 kprintf("\n** DUMP FAILED (ERROR %d) **\n", error);
 390 }
 391
 392 void
 393 dump_add_page(vm_paddr_t pa)
 394 {
 395         int idx, bit;
 396
 397         pa >>= PAGE_SHIFT;
 398         idx = pa >> 6;          /* 2^6 = 64 */
 399         bit = pa & 63;
 400         atomic_set_long(&vm_page_dump[idx], 1ul << bit);
 401 }
 402
 403 void
 404 dump_drop_page(vm_paddr_t pa)
 405 {
 406         int idx, bit;
 407
 408         pa >>= PAGE_SHIFT;
 409         idx = pa >> 6;          /* 2^6 = 64 */
 410         bit = pa & 63;
 411         atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
 412 }