2 * Copyright (c) 1989, 1992, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software developed by the Computer Systems
6 * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
7 * BG 91-66 and contributed to Berkeley.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include <sys/param.h>
38 #include <sys/fnv_hash.h>
43 #include <sys/linker.h>
65 #include "kvm_private.h"
68 * Routines private to libkvm.
71 /* from src/lib/libc/gen/nlist.c */
72 int __fdnlist(int, struct nlist *);
75 * Report an error using printf style arguments. "program" is kd->program
76 * on hard errors, and 0 on soft errors, so that under sun error emulation,
77 * only hard errors are printed out (otherwise, programs like gdb will
78 * generate tons of error messages when trying to access bogus pointers).
81 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
86 if (program != NULL) {
87 (void)fprintf(stderr, "%s: ", program);
88 (void)vfprintf(stderr, fmt, ap);
89 (void)fputc('\n', stderr);
91 (void)vsnprintf(kd->errbuf,
92 sizeof(kd->errbuf), fmt, ap);
98 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
104 if (program != NULL) {
105 (void)fprintf(stderr, "%s: ", program);
106 (void)vfprintf(stderr, fmt, ap);
107 (void)fprintf(stderr, ": %s\n", strerror(errno));
109 char *cp = kd->errbuf;
111 (void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
113 (void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
120 _kvm_malloc(kvm_t *kd, size_t n)
124 if ((p = calloc(n, sizeof(char))) == NULL)
125 _kvm_err(kd, kd->program, "can't allocate %zu bytes: %s",
131 _kvm_probe_elf_kernel(kvm_t *kd, int class, int machine)
134 return (kd->nlehdr.e_ident[EI_CLASS] == class &&
135 ((machine == EM_PPC || machine == EM_PPC64) ?
136 kd->nlehdr.e_type == ET_DYN : kd->nlehdr.e_type == ET_EXEC) &&
137 kd->nlehdr.e_machine == machine);
141 _kvm_is_minidump(kvm_t *kd)
147 if (pread(kd->pmfd, &minihdr, 8, 0) == 8 &&
148 memcmp(&minihdr, "minidump", 8) == 0)
154 * The powerpc backend has a hack to strip a leading kerneldump
155 * header from the core before treating it as an ELF header.
157 * We can add that here if we can get a change to libelf to support
158 * an initial offset into the file. Alternatively we could patch
159 * savecore to extract cores from a regular file instead.
162 _kvm_read_core_phdrs(kvm_t *kd, size_t *phnump, GElf_Phdr **phdrp)
169 elf = elf_begin(kd->pmfd, ELF_C_READ, NULL);
171 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
174 if (elf_kind(elf) != ELF_K_ELF) {
175 _kvm_err(kd, kd->program, "invalid core");
178 if (gelf_getclass(elf) != kd->nlehdr.e_ident[EI_CLASS]) {
179 _kvm_err(kd, kd->program, "invalid core");
182 if (gelf_getehdr(elf, &ehdr) == NULL) {
183 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
186 if (ehdr.e_type != ET_CORE) {
187 _kvm_err(kd, kd->program, "invalid core");
190 if (ehdr.e_machine != kd->nlehdr.e_machine) {
191 _kvm_err(kd, kd->program, "invalid core");
195 if (elf_getphdrnum(elf, &phnum) == -1) {
196 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
200 phdr = calloc(phnum, sizeof(*phdr));
202 _kvm_err(kd, kd->program, "failed to allocate phdrs");
206 for (i = 0; i < phnum; i++) {
207 if (gelf_getphdr(elf, i, &phdr[i]) == NULL) {
209 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
224 * Transform v such that only bits [bit0, bitN) may be set. Generates a
225 * bitmask covering the number of bits, then shifts so +bit0+ is the first.
228 bitmask_range(uint64_t v, uint64_t bit0, uint64_t bitN)
230 if (bit0 == 0 && bitN == BITS_IN(v))
233 return (v & (((1ULL << (bitN - bit0)) - 1ULL) << bit0));
237 * Returns the number of bits in a given byte array range starting at a
238 * given base, from bit0 to bitN. bit0 may be non-zero in the case of
239 * counting backwards from bitN.
242 popcount_bytes(uint64_t *addr, uint32_t bit0, uint32_t bitN)
244 uint32_t res = bitN - bit0;
248 /* Align to 64-bit boundary on the left side if needed. */
249 if ((bit0 % BITS_IN(*addr)) != 0) {
250 bound = MIN(bitN, roundup2(bit0, BITS_IN(*addr)));
251 count += __bitcount64(bitmask_range(*addr, bit0, bound));
252 res -= (bound - bit0);
257 bound = MIN(res, BITS_IN(*addr));
258 count += __bitcount64(bitmask_range(*addr, 0, bound));
267 _kvm_pmap_get(kvm_t *kd, u_long idx, size_t len)
269 uintptr_t off = idx * len;
271 if ((off_t)off >= kd->pt_sparse_off)
273 return (void *)((uintptr_t)kd->page_map + off);
277 _kvm_map_get(kvm_t *kd, u_long pa, unsigned int page_size)
282 off = _kvm_pt_find(kd, pa, page_size);
286 addr = (uintptr_t)kd->page_map + off;
287 if (off >= kd->pt_sparse_off)
288 addr = (uintptr_t)kd->sparse_map + (off - kd->pt_sparse_off);
293 _kvm_pt_init(kvm_t *kd, size_t dump_avail_size, off_t dump_avail_off,
294 size_t map_len, off_t map_off, off_t sparse_off, int page_size)
297 uint32_t *popcount_bin;
298 int bin_popcounts = 0;
299 uint64_t pc_bins, res;
302 kd->dump_avail_size = dump_avail_size;
303 if (dump_avail_size > 0) {
304 kd->dump_avail = mmap(NULL, kd->dump_avail_size, PROT_READ,
305 MAP_PRIVATE, kd->pmfd, dump_avail_off);
308 * Older version minidumps don't provide dump_avail[],
309 * so the bitmap is fully populated from 0 to
310 * last_pa. Create an implied dump_avail that
313 kd->dump_avail = calloc(4, sizeof(uint64_t));
314 kd->dump_avail[1] = _kvm64toh(kd, map_len * 8 * page_size);
318 * Map the bitmap specified by the arguments.
320 kd->pt_map = _kvm_malloc(kd, map_len);
321 if (kd->pt_map == NULL) {
322 _kvm_err(kd, kd->program, "cannot allocate %zu bytes for bitmap",
326 rd = pread(kd->pmfd, kd->pt_map, map_len, map_off);
327 if (rd < 0 || rd != (ssize_t)map_len) {
328 _kvm_err(kd, kd->program, "cannot read %zu bytes for bitmap",
332 kd->pt_map_size = map_len;
335 * Generate a popcount cache for every POPCOUNT_BITS in the bitmap,
336 * so lookups only have to calculate the number of bits set between
337 * a cache point and their bit. This reduces lookups to O(1),
338 * without significantly increasing memory requirements.
340 * Round up the number of bins so that 'upper half' lookups work for
341 * the final bin, if needed. The first popcount is 0, since no bits
342 * precede bit 0, so add 1 for that also. Without this, extra work
343 * would be needed to handle the first PTEs in _kvm_pt_find().
347 pc_bins = 1 + (res * NBBY + POPCOUNT_BITS / 2) / POPCOUNT_BITS;
348 kd->pt_popcounts = calloc(pc_bins, sizeof(uint32_t));
349 if (kd->pt_popcounts == NULL) {
350 _kvm_err(kd, kd->program, "cannot allocate popcount bins");
354 for (popcount_bin = &kd->pt_popcounts[1]; res > 0;
355 addr++, res -= sizeof(*addr)) {
356 *popcount_bin += popcount_bytes(addr, 0,
357 MIN(res * NBBY, BITS_IN(*addr)));
358 if (++bin_popcounts == POPCOUNTS_IN(*addr)) {
360 *popcount_bin = *(popcount_bin - 1);
365 assert(pc_bins * sizeof(*popcount_bin) ==
366 ((uintptr_t)popcount_bin - (uintptr_t)kd->pt_popcounts));
368 kd->pt_sparse_off = sparse_off;
369 kd->pt_sparse_size = (uint64_t)*popcount_bin * page_size;
370 kd->pt_page_size = page_size;
373 * Map the sparse page array. This is useful for performing point
374 * lookups of specific pages, e.g. for kvm_walk_pages. Generally,
375 * this is much larger than is reasonable to read in up front, so
376 * mmap it in instead.
378 kd->sparse_map = mmap(NULL, kd->pt_sparse_size, PROT_READ,
379 MAP_PRIVATE, kd->pmfd, kd->pt_sparse_off);
380 if (kd->sparse_map == MAP_FAILED) {
381 _kvm_err(kd, kd->program, "cannot map %" PRIu64
382 " bytes from fd %d offset %jd for sparse map: %s",
383 kd->pt_sparse_size, kd->pmfd,
384 (intmax_t)kd->pt_sparse_off, strerror(errno));
391 _kvm_pmap_init(kvm_t *kd, uint32_t pmap_size, off_t pmap_off)
393 ssize_t exp_len = pmap_size;
395 kd->page_map_size = pmap_size;
396 kd->page_map_off = pmap_off;
397 kd->page_map = _kvm_malloc(kd, pmap_size);
398 if (kd->page_map == NULL) {
399 _kvm_err(kd, kd->program, "cannot allocate %u bytes "
400 "for page map", pmap_size);
403 if (pread(kd->pmfd, kd->page_map, pmap_size, pmap_off) != exp_len) {
404 _kvm_err(kd, kd->program, "cannot read %d bytes from "
405 "offset %jd for page map", pmap_size, (intmax_t)pmap_off);
411 static inline uint64_t
412 dump_avail_n(kvm_t *kd, long i)
414 return (_kvm64toh(kd, kd->dump_avail[i]));
418 _kvm_pa_bit_id(kvm_t *kd, uint64_t pa, unsigned int page_size)
424 for (i = 0; dump_avail_n(kd, i + 1) != 0; i += 2) {
425 if (pa >= dump_avail_n(kd, i + 1)) {
426 adj += howmany(dump_avail_n(kd, i + 1), page_size) -
427 dump_avail_n(kd, i) / page_size;
429 return (pa / page_size -
430 dump_avail_n(kd, i) / page_size + adj);
433 return (_KVM_BIT_ID_INVALID);
437 _kvm_bit_id_pa(kvm_t *kd, uint64_t bit_id, unsigned int page_size)
442 for (i = 0; dump_avail_n(kd, i + 1) != 0; i += 2) {
443 sz = howmany(dump_avail_n(kd, i + 1), page_size) -
444 dump_avail_n(kd, i) / page_size;
446 return (rounddown2(dump_avail_n(kd, i), page_size) +
451 return (_KVM_PA_INVALID);
455 * Find the offset for the given physical page address; returns -1 otherwise.
457 * A page's offset is represented by the sparse page base offset plus the
458 * number of bits set before its bit multiplied by page size. This means
459 * that if a page exists in the dump, it's necessary to know how many pages
460 * in the dump precede it. Reduce this O(n) counting to O(1) by caching the
461 * number of bits set at POPCOUNT_BITS intervals.
463 * Then to find the number of pages before the requested address, simply
464 * index into the cache and count the number of bits set between that cache
465 * bin and the page's bit. Halve the number of bytes that have to be
466 * checked by also counting down from the next higher bin if it's closer.
469 _kvm_pt_find(kvm_t *kd, uint64_t pa, unsigned int page_size)
471 uint64_t *bitmap = kd->pt_map;
472 uint64_t pte_bit_id = _kvm_pa_bit_id(kd, pa, page_size);
473 uint64_t pte_u64 = pte_bit_id / BITS_IN(*bitmap);
474 uint64_t popcount_id = pte_bit_id / POPCOUNT_BITS;
475 uint64_t pte_mask = 1ULL << (pte_bit_id % BITS_IN(*bitmap));
479 /* Check whether the page address requested is in the dump. */
480 if (pte_bit_id == _KVM_BIT_ID_INVALID ||
481 pte_bit_id >= (kd->pt_map_size * NBBY) ||
482 (bitmap[pte_u64] & pte_mask) == 0)
486 * Add/sub popcounts from the bitmap until the PTE's bit is reached.
487 * For bits that are in the upper half between the calculated
488 * popcount id and the next one, use the next one and subtract to
489 * minimize the number of popcounts required.
491 if ((pte_bit_id % POPCOUNT_BITS) < (POPCOUNT_BITS / 2)) {
492 count = kd->pt_popcounts[popcount_id] + popcount_bytes(
493 bitmap + popcount_id * POPCOUNTS_IN(*bitmap),
494 0, pte_bit_id - popcount_id * POPCOUNT_BITS);
497 * Counting in reverse is trickier, since we must avoid
498 * reading from bytes that are not in range, and invert.
500 uint64_t pte_u64_bit_off = pte_u64 * BITS_IN(*bitmap);
503 bitN = MIN(popcount_id * POPCOUNT_BITS,
504 kd->pt_map_size * BITS_IN(uint8_t));
505 count = kd->pt_popcounts[popcount_id] - popcount_bytes(
507 pte_bit_id - pte_u64_bit_off, bitN - pte_u64_bit_off);
511 * This can only happen if the core is truncated. Treat these
512 * entries as if they don't exist, since their backing doesn't.
514 if (count >= (kd->pt_sparse_size / page_size))
517 return (kd->pt_sparse_off + (uint64_t)count * page_size);
521 kvm_fdnlist(kvm_t *kd, struct kvm_nlist *list)
526 if (kd->resolve_symbol == NULL) {
530 for (count = 0; list[count].n_name != NULL &&
531 list[count].n_name[0] != '\0'; count++)
533 nl = calloc(count + 1, sizeof(*nl));
534 for (i = 0; i < count; i++)
535 nl[i].n_name = list[i].n_name;
536 nfail = __fdnlist(kd->nlfd, nl);
537 for (i = 0; i < count; i++) {
538 list[i].n_type = nl[i].n_type;
539 list[i].n_value = nl[i].n_value;
546 while (list->n_name != NULL && list->n_name[0] != '\0') {
547 error = kd->resolve_symbol(list->n_name, &addr);
553 list->n_value = addr;
554 list->n_type = N_DATA | N_EXT;
562 * Walk the list of unresolved symbols, generate a new list and prefix the
563 * symbol names, try again, and merge back what we could resolve.
566 kvm_fdnlist_prefix(kvm_t *kd, struct kvm_nlist *nl, int missing,
567 const char *prefix, kvaddr_t (*validate_fn)(kvm_t *, kvaddr_t))
569 struct kvm_nlist *n, *np, *p;
573 int slen, unresolved;
576 * Calculate the space we need to malloc for nlist and names.
577 * We are going to store the name twice for later lookups: once
578 * with the prefix and once the unmodified name delmited by \0.
582 for (p = nl; p->n_name && p->n_name[0]; ++p) {
583 if (p->n_type != N_UNDF)
585 len += sizeof(struct kvm_nlist) + strlen(prefix) +
586 2 * (strlen(p->n_name) + 1);
591 /* Add space for the terminating nlist entry. */
592 len += sizeof(struct kvm_nlist);
595 /* Alloc one chunk for (nlist, [names]) and setup pointers. */
596 n = np = malloc(len);
600 cp = ce = (char *)np;
601 cp += unresolved * sizeof(struct kvm_nlist);
604 /* Generate shortened nlist with special prefix. */
606 for (p = nl; p->n_name && p->n_name[0]; ++p) {
607 if (p->n_type != N_UNDF)
610 /* Save the new\0orig. name so we can later match it again. */
611 slen = snprintf(cp, ce - cp, "%s%s%c%s", prefix,
612 (prefix[0] != '\0' && p->n_name[0] == '_') ?
613 (p->n_name + 1) : p->n_name, '\0', p->n_name);
614 if (slen < 0 || slen >= ce - cp)
622 /* Do lookup on the reduced list. */
624 unresolved = kvm_fdnlist(kd, np);
626 /* Check if we could resolve further symbols and update the list. */
627 if (unresolved >= 0 && unresolved < missing) {
628 /* Find the first freshly resolved entry. */
629 for (; np->n_name && np->n_name[0]; np++)
630 if (np->n_type != N_UNDF)
633 * The lists are both in the same order,
634 * so we can walk them in parallel.
636 for (p = nl; np->n_name && np->n_name[0] &&
637 p->n_name && p->n_name[0]; ++p) {
638 if (p->n_type != N_UNDF)
640 /* Skip expanded name and compare to orig. one. */
641 ccp = np->n_name + strlen(np->n_name) + 1;
642 if (strcmp(ccp, p->n_name) != 0)
644 /* Update nlist with new, translated results. */
645 p->n_type = np->n_type;
647 p->n_value = (*validate_fn)(kd, np->n_value);
649 p->n_value = np->n_value;
651 /* Find next freshly resolved entry. */
652 for (np++; np->n_name && np->n_name[0]; np++)
653 if (np->n_type != N_UNDF)
657 /* We could assert missing = unresolved here. */
664 _kvm_nlist(kvm_t *kd, struct kvm_nlist *nl, int initialize)
668 struct kld_sym_lookup lookup;
670 const char *prefix = "";
671 char symname[1024]; /* XXX-BZ symbol name length limit? */
672 int tried_vnet, tried_dpcpu;
675 * If we can't use the kld symbol lookup, revert to the
679 error = kvm_fdnlist(kd, nl);
680 if (error <= 0) /* Hard error or success. */
683 if (_kvm_vnet_initialized(kd, initialize))
684 error = kvm_fdnlist_prefix(kd, nl, error,
685 VNET_SYMPREFIX, _kvm_vnet_validaddr);
687 if (error > 0 && _kvm_dpcpu_initialized(kd, initialize))
688 error = kvm_fdnlist_prefix(kd, nl, error,
689 DPCPU_SYMPREFIX, _kvm_dpcpu_validaddr);
695 * We can use the kld lookup syscall. Go through each nlist entry
696 * and look it up with a kldsym(2) syscall.
702 for (p = nl; p->n_name && p->n_name[0]; ++p) {
703 if (p->n_type != N_UNDF)
706 lookup.version = sizeof(lookup);
710 error = snprintf(symname, sizeof(symname), "%s%s", prefix,
711 (prefix[0] != '\0' && p->n_name[0] == '_') ?
712 (p->n_name + 1) : p->n_name);
713 if (error < 0 || error >= (int)sizeof(symname))
715 lookup.symname = symname;
716 if (lookup.symname[0] == '_')
719 if (kldsym(0, KLDSYM_LOOKUP, &lookup) != -1) {
721 if (_kvm_vnet_initialized(kd, initialize) &&
722 strcmp(prefix, VNET_SYMPREFIX) == 0)
724 _kvm_vnet_validaddr(kd, lookup.symvalue);
725 else if (_kvm_dpcpu_initialized(kd, initialize) &&
726 strcmp(prefix, DPCPU_SYMPREFIX) == 0)
728 _kvm_dpcpu_validaddr(kd, lookup.symvalue);
730 p->n_value = lookup.symvalue;
737 * Check the number of entries that weren't found. If they exist,
738 * try again with a prefix for virtualized or DPCPU symbol names.
740 error = ((p - nl) - nvalid);
741 if (error && _kvm_vnet_initialized(kd, initialize) && !tried_vnet) {
743 prefix = VNET_SYMPREFIX;
746 if (error && _kvm_dpcpu_initialized(kd, initialize) && !tried_dpcpu) {
748 prefix = DPCPU_SYMPREFIX;
753 * Return the number of entries that weren't found. If they exist,
754 * also fill internal error buffer.
756 error = ((p - nl) - nvalid);
758 _kvm_syserr(kd, kd->program, "kvm_nlist");
763 _kvm_bitmap_init(struct kvm_bitmap *bm, u_long bitmapsize, u_long *idx)
767 bm->map = calloc(bitmapsize, sizeof *bm->map);
770 bm->size = bitmapsize;
775 _kvm_bitmap_set(struct kvm_bitmap *bm, u_long bm_index)
777 uint8_t *byte = &bm->map[bm_index / 8];
779 if (bm_index / 8 < bm->size)
780 *byte |= (1UL << (bm_index % 8));
784 _kvm_bitmap_next(struct kvm_bitmap *bm, u_long *idx)
786 u_long first_invalid = bm->size * CHAR_BIT;
788 if (*idx == ULONG_MAX)
793 /* Find the next valid idx. */
794 for (; *idx < first_invalid; (*idx)++) {
795 unsigned int mask = *idx % CHAR_BIT;
796 if ((bm->map[*idx * CHAR_BIT] & mask) == 0)
800 return (*idx < first_invalid);
804 _kvm_bitmap_deinit(struct kvm_bitmap *bm)
811 _kvm_visit_cb(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg, u_long pa,
812 u_long kmap_vaddr, u_long dmap_vaddr, vm_prot_t prot, size_t len,
813 unsigned int page_size)
815 unsigned int pgsz = page_size ? page_size : len;
816 struct kvm_page p = {
817 .kp_version = LIBKVM_WALK_PAGES_VERSION,
819 .kp_kmap_vaddr = kmap_vaddr,
820 .kp_dmap_vaddr = dmap_vaddr,
822 .kp_offset = _kvm_pt_find(kd, pa, pgsz),