minidump_machdep.c

Version: - FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
    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 
   27 #include <sys/cdefs.h>
   28 __FBSDID("$FreeBSD$");
   29 
   30 #include "opt_pmap.h"
   31 #include "opt_watchdog.h"
   32 
   33 #include <sys/param.h>
   34 #include <sys/systm.h>
   35 #include <sys/conf.h>
   36 #include <sys/cons.h>
   37 #include <sys/kernel.h>
   38 #include <sys/kerneldump.h>
   39 #include <sys/msgbuf.h>
   40 #include <sys/watchdog.h>
   41 #include <vm/vm.h>
   42 #include <vm/vm_page.h>
   43 #include <vm/pmap.h>
   44 #include <machine/atomic.h>
   45 #include <machine/elf.h>
   46 #include <machine/md_var.h>
   47 #include <machine/vmparam.h>
   48 #include <machine/minidump.h>
   49 
   50 CTASSERT(sizeof(struct kerneldumpheader) == 512);
   51 
   52 /*
   53  * Don't touch the first SIZEOF_METADATA bytes on the dump device. This
   54  * is to protect us from metadata and to protect metadata from us.
   55  */
   56 #define SIZEOF_METADATA         (64*1024)
   57 
   58 #define MD_ALIGN(x)     (((off_t)(x) + PAGE_MASK) & ~PAGE_MASK)
   59 #define DEV_ALIGN(x)    (((off_t)(x) + (DEV_BSIZE-1)) & ~(DEV_BSIZE-1))
   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, dumpsize;
   71 
   72 CTASSERT(sizeof(*vm_page_dump) == 8);
   73 
   74 static int
   75 is_dumpable(vm_paddr_t pa)
   76 {
   77         vm_page_t m;
   78         int i;
   79 
   80         if ((m = vm_phys_paddr_to_vm_page(pa)) != NULL)
   81                 return ((m->flags & PG_NODUMP) == 0);
   82         for (i = 0; dump_avail[i] != 0 || dump_avail[i + 1] != 0; i += 2) {
   83                 if (pa >= dump_avail[i] && pa < dump_avail[i + 1])
   84                         return (1);
   85         }
   86         return (0);
   87 }
   88 
   89 #define PG2MB(pgs) (((pgs) + (1 << 8) - 1) >> 8)
   90 
   91 static int
   92 blk_flush(struct dumperinfo *di)
   93 {
   94         int error;
   95 
   96         if (fragsz == 0)
   97                 return (0);
   98 
   99         error = dump_write(di, dump_va, 0, dumplo, fragsz);
  100         dumplo += fragsz;
  101         fragsz = 0;
  102         return (error);
  103 }
  104 
  105 static struct {
  106         int min_per;
  107         int max_per;
  108         int visited;
  109 } progress_track[10] = {
  110         {  0,  10, 0},
  111         { 10,  20, 0},
  112         { 20,  30, 0},
  113         { 30,  40, 0},
  114         { 40,  50, 0},
  115         { 50,  60, 0},
  116         { 60,  70, 0},
  117         { 70,  80, 0},
  118         { 80,  90, 0},
  119         { 90, 100, 0}
  120 };
  121 
  122 static void
  123 report_progress(size_t progress, size_t dumpsize)
  124 {
  125         int sofar, i;
  126 
  127         sofar = 100 - ((progress * 100) / dumpsize);
  128         for (i = 0; i < nitems(progress_track); i++) {
  129                 if (sofar < progress_track[i].min_per ||
  130                     sofar > progress_track[i].max_per)
  131                         continue;
  132                 if (progress_track[i].visited)
  133                         return;
  134                 progress_track[i].visited = 1;
  135                 printf("..%d%%", sofar);
  136                 return;
  137         }
  138 }
  139 
  140 static int
  141 blk_write(struct dumperinfo *di, char *ptr, vm_paddr_t pa, size_t sz)
  142 {
  143         size_t len;
  144         int error, i, c;
  145         u_int maxdumpsz;
  146 
  147         maxdumpsz = min(di->maxiosize, MAXDUMPPGS * PAGE_SIZE);
  148         if (maxdumpsz == 0)     /* seatbelt */
  149                 maxdumpsz = PAGE_SIZE;
  150         error = 0;
  151         if ((sz % PAGE_SIZE) != 0) {
  152                 printf("size not page aligned\n");
  153                 return (EINVAL);
  154         }
  155         if (ptr != NULL && pa != 0) {
  156                 printf("cant have both va and pa!\n");
  157                 return (EINVAL);
  158         }
  159         if ((((uintptr_t)pa) % PAGE_SIZE) != 0) {
  160                 printf("address not page aligned %p\n", ptr);
  161                 return (EINVAL);
  162         }
  163         if (ptr != NULL) {
  164                 /* If we're doing a virtual dump, flush any pre-existing pa pages */
  165                 error = blk_flush(di);
  166                 if (error)
  167                         return (error);
  168         }
  169         while (sz) {
  170                 len = maxdumpsz - fragsz;
  171                 if (len > sz)
  172                         len = sz;
  173                 counter += len;
  174                 progress -= len;
  175                 if (counter >> 24) {
  176                         report_progress(progress, dumpsize);
  177                         counter &= (1<<24) - 1;
  178                 }
  179 
  180                 wdog_kern_pat(WD_LASTVAL);
  181 
  182                 if (ptr) {
  183                         error = dump_write(di, ptr, 0, dumplo, len);
  184                         if (error)
  185                                 return (error);
  186                         dumplo += len;
  187                         ptr += len;
  188                         sz -= len;
  189                 } else {
  190                         for (i = 0; i < len; i += PAGE_SIZE)
  191                                 dump_va = pmap_kenter_temporary(pa + i, (i + fragsz) >> PAGE_SHIFT);
  192                         fragsz += len;
  193                         pa += len;
  194                         sz -= len;
  195                         if (fragsz == maxdumpsz) {
  196                                 error = blk_flush(di);
  197                                 if (error)
  198                                         return (error);
  199                         }
  200                 }
  201 
  202                 /* Check for user abort. */
  203                 c = cncheckc();
  204                 if (c == 0x03)
  205                         return (ECANCELED);
  206                 if (c != -1)
  207                         printf(" (CTRL-C to abort) ");
  208         }
  209 
  210         return (0);
  211 }
  212 
  213 /* A fake page table page, to avoid having to handle both 4K and 2M pages */
  214 static pd_entry_t fakepd[NPDEPG];
  215 
  216 void
  217 minidumpsys(struct dumperinfo *di)
  218 {
  219         uint32_t pmapsize;
  220         vm_offset_t va;
  221         int error;
  222         uint64_t bits;
  223         uint64_t *pdp, *pd, *pt, pa;
  224         int i, j, k, n, bit;
  225         int retry_count;
  226         struct minidumphdr mdhdr;
  227 
  228         retry_count = 0;
  229  retry:
  230         retry_count++;
  231         counter = 0;
  232         for (i = 0; i < nitems(progress_track); i++)
  233                 progress_track[i].visited = 0;
  234         /* Walk page table pages, set bits in vm_page_dump */
  235         pmapsize = 0;
  236         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
  237         for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + NKPT * NBPDR,
  238             kernel_vm_end); ) {
  239                 /*
  240                  * We always write a page, even if it is zero. Each
  241                  * page written corresponds to 1GB of space
  242                  */
  243                 pmapsize += PAGE_SIZE;
  244                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
  245                 if ((pdp[i] & PG_V) == 0) {
  246                         va += NBPDP;
  247                         continue;
  248                 }
  249 
  250                 /*
  251                  * 1GB page is represented as 512 2MB pages in a dump.
  252                  */
  253                 if ((pdp[i] & PG_PS) != 0) {
  254                         va += NBPDP;
  255                         pa = pdp[i] & PG_PS_FRAME;
  256                         for (n = 0; n < NPDEPG * NPTEPG; n++) {
  257                                 if (is_dumpable(pa))
  258                                         dump_add_page(pa);
  259                                 pa += PAGE_SIZE;
  260                         }
  261                         continue;
  262                 }
  263 
  264                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
  265                 for (n = 0; n < NPDEPG; n++, va += NBPDR) {
  266                         j = (va >> PDRSHIFT) & ((1ul << NPDEPGSHIFT) - 1);
  267 
  268                         if ((pd[j] & PG_V) == 0)
  269                                 continue;
  270 
  271                         if ((pd[j] & PG_PS) != 0) {
  272                                 /* This is an entire 2M page. */
  273                                 pa = pd[j] & PG_PS_FRAME;
  274                                 for (k = 0; k < NPTEPG; k++) {
  275                                         if (is_dumpable(pa))
  276                                                 dump_add_page(pa);
  277                                         pa += PAGE_SIZE;
  278                                 }
  279                                 continue;
  280                         }
  281 
  282                         pa = pd[j] & PG_FRAME;
  283                         /* set bit for this PTE page */
  284                         if (is_dumpable(pa))
  285                                 dump_add_page(pa);
  286                         /* and 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] & PG_V) == 0)
  290                                         continue;
  291                                 pa = pt[k] & PG_FRAME;
  292                                 if (is_dumpable(pa))
  293                                         dump_add_page(pa);
  294                         }
  295                 }
  296         }
  297 
  298         /* Calculate dump size. */
  299         dumpsize = pmapsize;
  300         dumpsize += round_page(msgbufp->msg_size);
  301         dumpsize += round_page(vm_page_dump_size);
  302         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
  303                 bits = vm_page_dump[i];
  304                 while (bits) {
  305                         bit = bsfq(bits);
  306                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
  307                         /* Clear out undumpable pages now if needed */
  308                         if (is_dumpable(pa)) {
  309                                 dumpsize += PAGE_SIZE;
  310                         } else {
  311                                 dump_drop_page(pa);
  312                         }
  313                         bits &= ~(1ul << bit);
  314                 }
  315         }
  316         dumpsize += PAGE_SIZE;
  317 
  318         /* Determine dump offset on device. */
  319         if (di->mediasize < SIZEOF_METADATA + dumpsize + sizeof(kdh) * 2) {
  320                 error = E2BIG;
  321                 goto fail;
  322         }
  323         dumplo = di->mediaoffset + di->mediasize - dumpsize;
  324         dumplo -= sizeof(kdh) * 2;
  325         progress = dumpsize;
  326 
  327         /* Initialize mdhdr */
  328         bzero(&mdhdr, sizeof(mdhdr));
  329         strcpy(mdhdr.magic, MINIDUMP_MAGIC);
  330         mdhdr.version = MINIDUMP_VERSION;
  331         mdhdr.msgbufsize = msgbufp->msg_size;
  332         mdhdr.bitmapsize = vm_page_dump_size;
  333         mdhdr.pmapsize = pmapsize;
  334         mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS;
  335         mdhdr.dmapbase = DMAP_MIN_ADDRESS;
  336         mdhdr.dmapend = DMAP_MAX_ADDRESS;
  337 
  338         mkdumpheader(&kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, dumpsize, di->blocksize);
  339 
  340         printf("Dumping %llu out of %ju MB:", (long long)dumpsize >> 20,
  341             ptoa((uintmax_t)physmem) / 1048576);
  342 
  343         /* Dump leader */
  344         error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
  345         if (error)
  346                 goto fail;
  347         dumplo += sizeof(kdh);
  348 
  349         /* Dump my header */
  350         bzero(&fakepd, sizeof(fakepd));
  351         bcopy(&mdhdr, &fakepd, sizeof(mdhdr));
  352         error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
  353         if (error)
  354                 goto fail;
  355 
  356         /* Dump msgbuf up front */
  357         error = blk_write(di, (char *)msgbufp->msg_ptr, 0, round_page(msgbufp->msg_size));
  358         if (error)
  359                 goto fail;
  360 
  361         /* Dump bitmap */
  362         error = blk_write(di, (char *)vm_page_dump, 0, round_page(vm_page_dump_size));
  363         if (error)
  364                 goto fail;
  365 
  366         /* Dump kernel page directory pages */
  367         bzero(fakepd, sizeof(fakepd));
  368         pdp = (uint64_t *)PHYS_TO_DMAP(KPDPphys);
  369         for (va = VM_MIN_KERNEL_ADDRESS; va < MAX(KERNBASE + NKPT * NBPDR,
  370             kernel_vm_end); va += NBPDP) {
  371                 i = (va >> PDPSHIFT) & ((1ul << NPDPEPGSHIFT) - 1);
  372 
  373                 /* We always write a page, even if it is zero */
  374                 if ((pdp[i] & PG_V) == 0) {
  375                         error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
  376                         if (error)
  377                                 goto fail;
  378                         /* flush, in case we reuse fakepd in the same block */
  379                         error = blk_flush(di);
  380                         if (error)
  381                                 goto fail;
  382                         continue;
  383                 }
  384 
  385                 /* 1GB page is represented as 512 2MB pages in a dump */
  386                 if ((pdp[i] & PG_PS) != 0) {
  387                         /* PDPE and PDP have identical layout in this case */
  388                         fakepd[0] = pdp[i];
  389                         for (j = 1; j < NPDEPG; j++)
  390                                 fakepd[j] = fakepd[j - 1] + NBPDR;
  391                         error = blk_write(di, (char *)&fakepd, 0, PAGE_SIZE);
  392                         if (error)
  393                                 goto fail;
  394                         /* flush, in case we reuse fakepd in the same block */
  395                         error = blk_flush(di);
  396                         if (error)
  397                                 goto fail;
  398                         bzero(fakepd, sizeof(fakepd));
  399                         continue;
  400                 }
  401 
  402                 pd = (uint64_t *)PHYS_TO_DMAP(pdp[i] & PG_FRAME);
  403                 error = blk_write(di, (char *)pd, 0, PAGE_SIZE);
  404                 if (error)
  405                         goto fail;
  406                 error = blk_flush(di);
  407                 if (error)
  408                         goto fail;
  409         }
  410 
  411         /* Dump memory chunks */
  412         /* XXX cluster it up and use blk_dump() */
  413         for (i = 0; i < vm_page_dump_size / sizeof(*vm_page_dump); i++) {
  414                 bits = vm_page_dump[i];
  415                 while (bits) {
  416                         bit = bsfq(bits);
  417                         pa = (((uint64_t)i * sizeof(*vm_page_dump) * NBBY) + bit) * PAGE_SIZE;
  418                         error = blk_write(di, 0, pa, PAGE_SIZE);
  419                         if (error)
  420                                 goto fail;
  421                         bits &= ~(1ul << bit);
  422                 }
  423         }
  424 
  425         error = blk_flush(di);
  426         if (error)
  427                 goto fail;
  428 
  429         /* Dump trailer */
  430         error = dump_write(di, &kdh, 0, dumplo, sizeof(kdh));
  431         if (error)
  432                 goto fail;
  433         dumplo += sizeof(kdh);
  434 
  435         /* Signal completion, signoff and exit stage left. */
  436         dump_write(di, NULL, 0, 0, 0);
  437         printf("\nDump complete\n");
  438         return;
  439 
  440  fail:
  441         if (error < 0)
  442                 error = -error;
  443 
  444         printf("\n");
  445         if (error == ENOSPC) {
  446                 printf("Dump map grown while dumping. ");
  447                 if (retry_count < 5) {
  448                         printf("Retrying...\n");
  449                         goto retry;
  450                 }
  451                 printf("Dump failed.\n");
  452         }
  453         else if (error == ECANCELED)
  454                 printf("Dump aborted\n");
  455         else if (error == E2BIG)
  456                 printf("Dump failed. Partition too small.\n");
  457         else
  458                 printf("** DUMP FAILED (ERROR %d) **\n", error);
  459 }
  460 
  461 void
  462 dump_add_page(vm_paddr_t pa)
  463 {
  464         int idx, bit;
  465 
  466         pa >>= PAGE_SHIFT;
  467         idx = pa >> 6;          /* 2^6 = 64 */
  468         bit = pa & 63;
  469         atomic_set_long(&vm_page_dump[idx], 1ul << bit);
  470 }
  471 
  472 void
  473 dump_drop_page(vm_paddr_t pa)
  474 {
  475         int idx, bit;
  476 
  477         pa >>= PAGE_SHIFT;
  478         idx = pa >> 6;          /* 2^6 = 64 */
  479         bit = pa & 63;
  480         atomic_clear_long(&vm_page_dump[idx], 1ul << bit);
  481 }
Cache object: 46e9d5f6088e8a66e3467377994b38e9
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.
FreeBSD/Linux Kernel Cross Reference sys/amd64/amd64/minidump_machdep.c

FreeBSD/Linux Kernel Cross Reference
sys/amd64/amd64/minidump_machdep.c
