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/param.h>
35 #include <sys/fnv_hash.h>
40 #include <sys/linker.h>
62 #include "kvm_private.h"
65 * Routines private to libkvm.
68 /* from src/lib/libc/gen/nlist.c */
69 int __fdnlist(int, struct nlist *);
72 * Report an error using printf style arguments. "program" is kd->program
73 * on hard errors, and 0 on soft errors, so that under sun error emulation,
74 * only hard errors are printed out (otherwise, programs like gdb will
75 * generate tons of error messages when trying to access bogus pointers).
78 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
83 if (program != NULL) {
84 (void)fprintf(stderr, "%s: ", program);
85 (void)vfprintf(stderr, fmt, ap);
86 (void)fputc('\n', stderr);
88 (void)vsnprintf(kd->errbuf,
89 sizeof(kd->errbuf), fmt, ap);
95 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
101 if (program != NULL) {
102 (void)fprintf(stderr, "%s: ", program);
103 (void)vfprintf(stderr, fmt, ap);
104 (void)fprintf(stderr, ": %s\n", strerror(errno));
106 char *cp = kd->errbuf;
108 (void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
110 (void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
117 _kvm_malloc(kvm_t *kd, size_t n)
121 if ((p = calloc(n, sizeof(char))) == NULL)
122 _kvm_err(kd, kd->program, "can't allocate %zu bytes: %s",
128 _kvm_probe_elf_kernel(kvm_t *kd, int class, int machine)
131 return (kd->nlehdr.e_ident[EI_CLASS] == class &&
132 ((machine == EM_PPC || machine == EM_PPC64) ?
133 kd->nlehdr.e_type == ET_DYN : kd->nlehdr.e_type == ET_EXEC) &&
134 kd->nlehdr.e_machine == machine);
138 _kvm_is_minidump(kvm_t *kd)
144 if (pread(kd->pmfd, &minihdr, 8, 0) == 8 &&
145 memcmp(&minihdr, "minidump", 8) == 0)
151 * The powerpc backend has a hack to strip a leading kerneldump
152 * header from the core before treating it as an ELF header.
154 * We can add that here if we can get a change to libelf to support
155 * an initial offset into the file. Alternatively we could patch
156 * savecore to extract cores from a regular file instead.
159 _kvm_read_core_phdrs(kvm_t *kd, size_t *phnump, GElf_Phdr **phdrp)
166 elf = elf_begin(kd->pmfd, ELF_C_READ, NULL);
168 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
171 if (elf_kind(elf) != ELF_K_ELF) {
172 _kvm_err(kd, kd->program, "invalid core");
175 if (gelf_getclass(elf) != kd->nlehdr.e_ident[EI_CLASS]) {
176 _kvm_err(kd, kd->program, "invalid core");
179 if (gelf_getehdr(elf, &ehdr) == NULL) {
180 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
183 if (ehdr.e_type != ET_CORE) {
184 _kvm_err(kd, kd->program, "invalid core");
187 if (ehdr.e_machine != kd->nlehdr.e_machine) {
188 _kvm_err(kd, kd->program, "invalid core");
192 if (elf_getphdrnum(elf, &phnum) == -1) {
193 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
197 phdr = calloc(phnum, sizeof(*phdr));
199 _kvm_err(kd, kd->program, "failed to allocate phdrs");
203 for (i = 0; i < phnum; i++) {
204 if (gelf_getphdr(elf, i, &phdr[i]) == NULL) {
206 _kvm_err(kd, kd->program, "%s", elf_errmsg(0));
221 * Transform v such that only bits [bit0, bitN) may be set. Generates a
222 * bitmask covering the number of bits, then shifts so +bit0+ is the first.
225 bitmask_range(uint64_t v, uint64_t bit0, uint64_t bitN)
227 if (bit0 == 0 && bitN == BITS_IN(v))
230 return (v & (((1ULL << (bitN - bit0)) - 1ULL) << bit0));
234 * Returns the number of bits in a given byte array range starting at a
235 * given base, from bit0 to bitN. bit0 may be non-zero in the case of
236 * counting backwards from bitN.
239 popcount_bytes(uint64_t *addr, uint32_t bit0, uint32_t bitN)
241 uint32_t res = bitN - bit0;
245 /* Align to 64-bit boundary on the left side if needed. */
246 if ((bit0 % BITS_IN(*addr)) != 0) {
247 bound = MIN(bitN, roundup2(bit0, BITS_IN(*addr)));
248 count += __bitcount64(bitmask_range(*addr, bit0, bound));
249 res -= (bound - bit0);
254 bound = MIN(res, BITS_IN(*addr));
255 count += __bitcount64(bitmask_range(*addr, 0, bound));
264 _kvm_pmap_get(kvm_t *kd, u_long idx, size_t len)
266 uintptr_t off = idx * len;
268 if ((off_t)off >= kd->pt_sparse_off)
270 return (void *)((uintptr_t)kd->page_map + off);
274 _kvm_map_get(kvm_t *kd, u_long pa, unsigned int page_size)
279 off = _kvm_pt_find(kd, pa, page_size);
283 addr = (uintptr_t)kd->page_map + off;
284 if (off >= kd->pt_sparse_off)
285 addr = (uintptr_t)kd->sparse_map + (off - kd->pt_sparse_off);
290 _kvm_pt_init(kvm_t *kd, size_t dump_avail_size, off_t dump_avail_off,
291 size_t map_len, off_t map_off, off_t sparse_off, int page_size)
294 uint32_t *popcount_bin;
295 int bin_popcounts = 0;
296 uint64_t pc_bins, res;
299 kd->dump_avail_size = dump_avail_size;
300 if (dump_avail_size > 0) {
301 kd->dump_avail = mmap(NULL, kd->dump_avail_size, PROT_READ,
302 MAP_PRIVATE, kd->pmfd, dump_avail_off);
305 * Older version minidumps don't provide dump_avail[],
306 * so the bitmap is fully populated from 0 to
307 * last_pa. Create an implied dump_avail that
310 kd->dump_avail = calloc(4, sizeof(uint64_t));
311 kd->dump_avail[1] = _kvm64toh(kd, map_len * 8 * page_size);
315 * Map the bitmap specified by the arguments.
317 kd->pt_map = _kvm_malloc(kd, map_len);
318 if (kd->pt_map == NULL) {
319 _kvm_err(kd, kd->program, "cannot allocate %zu bytes for bitmap",
323 rd = pread(kd->pmfd, kd->pt_map, map_len, map_off);
324 if (rd < 0 || rd != (ssize_t)map_len) {
325 _kvm_err(kd, kd->program, "cannot read %zu bytes for bitmap",
329 kd->pt_map_size = map_len;
332 * Generate a popcount cache for every POPCOUNT_BITS in the bitmap,
333 * so lookups only have to calculate the number of bits set between
334 * a cache point and their bit. This reduces lookups to O(1),
335 * without significantly increasing memory requirements.
337 * Round up the number of bins so that 'upper half' lookups work for
338 * the final bin, if needed. The first popcount is 0, since no bits
339 * precede bit 0, so add 1 for that also. Without this, extra work
340 * would be needed to handle the first PTEs in _kvm_pt_find().
344 pc_bins = 1 + (res * NBBY + POPCOUNT_BITS / 2) / POPCOUNT_BITS;
345 kd->pt_popcounts = calloc(pc_bins, sizeof(uint32_t));
346 if (kd->pt_popcounts == NULL) {
347 _kvm_err(kd, kd->program, "cannot allocate popcount bins");
351 for (popcount_bin = &kd->pt_popcounts[1]; res > 0;
352 addr++, res -= sizeof(*addr)) {
353 *popcount_bin += popcount_bytes(addr, 0,
354 MIN(res * NBBY, BITS_IN(*addr)));
355 if (++bin_popcounts == POPCOUNTS_IN(*addr)) {
357 *popcount_bin = *(popcount_bin - 1);
362 assert(pc_bins * sizeof(*popcount_bin) ==
363 ((uintptr_t)popcount_bin - (uintptr_t)kd->pt_popcounts));
365 kd->pt_sparse_off = sparse_off;
366 kd->pt_sparse_size = (uint64_t)*popcount_bin * page_size;
367 kd->pt_page_size = page_size;
370 * Map the sparse page array. This is useful for performing point
371 * lookups of specific pages, e.g. for kvm_walk_pages. Generally,
372 * this is much larger than is reasonable to read in up front, so
373 * mmap it in instead.
375 kd->sparse_map = mmap(NULL, kd->pt_sparse_size, PROT_READ,
376 MAP_PRIVATE, kd->pmfd, kd->pt_sparse_off);
377 if (kd->sparse_map == MAP_FAILED) {
378 _kvm_err(kd, kd->program, "cannot map %" PRIu64
379 " bytes from fd %d offset %jd for sparse map: %s",
380 kd->pt_sparse_size, kd->pmfd,
381 (intmax_t)kd->pt_sparse_off, strerror(errno));
388 _kvm_pmap_init(kvm_t *kd, uint32_t pmap_size, off_t pmap_off)
390 ssize_t exp_len = pmap_size;
392 kd->page_map_size = pmap_size;
393 kd->page_map_off = pmap_off;
394 kd->page_map = _kvm_malloc(kd, pmap_size);
395 if (kd->page_map == NULL) {
396 _kvm_err(kd, kd->program, "cannot allocate %u bytes "
397 "for page map", pmap_size);
400 if (pread(kd->pmfd, kd->page_map, pmap_size, pmap_off) != exp_len) {
401 _kvm_err(kd, kd->program, "cannot read %d bytes from "
402 "offset %jd for page map", pmap_size, (intmax_t)pmap_off);
408 static inline uint64_t
409 dump_avail_n(kvm_t *kd, long i)
411 return (_kvm64toh(kd, kd->dump_avail[i]));
415 _kvm_pa_bit_id(kvm_t *kd, uint64_t pa, unsigned int page_size)
421 for (i = 0; dump_avail_n(kd, i + 1) != 0; i += 2) {
422 if (pa >= dump_avail_n(kd, i + 1)) {
423 adj += howmany(dump_avail_n(kd, i + 1), page_size) -
424 dump_avail_n(kd, i) / page_size;
426 return (pa / page_size -
427 dump_avail_n(kd, i) / page_size + adj);
430 return (_KVM_BIT_ID_INVALID);
434 _kvm_bit_id_pa(kvm_t *kd, uint64_t bit_id, unsigned int page_size)
439 for (i = 0; dump_avail_n(kd, i + 1) != 0; i += 2) {
440 sz = howmany(dump_avail_n(kd, i + 1), page_size) -
441 dump_avail_n(kd, i) / page_size;
443 return (rounddown2(dump_avail_n(kd, i), page_size) +
448 return (_KVM_PA_INVALID);
452 * Find the offset for the given physical page address; returns -1 otherwise.
454 * A page's offset is represented by the sparse page base offset plus the
455 * number of bits set before its bit multiplied by page size. This means
456 * that if a page exists in the dump, it's necessary to know how many pages
457 * in the dump precede it. Reduce this O(n) counting to O(1) by caching the
458 * number of bits set at POPCOUNT_BITS intervals.
460 * Then to find the number of pages before the requested address, simply
461 * index into the cache and count the number of bits set between that cache
462 * bin and the page's bit. Halve the number of bytes that have to be
463 * checked by also counting down from the next higher bin if it's closer.
466 _kvm_pt_find(kvm_t *kd, uint64_t pa, unsigned int page_size)
468 uint64_t *bitmap = kd->pt_map;
469 uint64_t pte_bit_id = _kvm_pa_bit_id(kd, pa, page_size);
470 uint64_t pte_u64 = pte_bit_id / BITS_IN(*bitmap);
471 uint64_t popcount_id = pte_bit_id / POPCOUNT_BITS;
472 uint64_t pte_mask = 1ULL << (pte_bit_id % BITS_IN(*bitmap));
476 /* Check whether the page address requested is in the dump. */
477 if (pte_bit_id == _KVM_BIT_ID_INVALID ||
478 pte_bit_id >= (kd->pt_map_size * NBBY) ||
479 (bitmap[pte_u64] & pte_mask) == 0)
483 * Add/sub popcounts from the bitmap until the PTE's bit is reached.
484 * For bits that are in the upper half between the calculated
485 * popcount id and the next one, use the next one and subtract to
486 * minimize the number of popcounts required.
488 if ((pte_bit_id % POPCOUNT_BITS) < (POPCOUNT_BITS / 2)) {
489 count = kd->pt_popcounts[popcount_id] + popcount_bytes(
490 bitmap + popcount_id * POPCOUNTS_IN(*bitmap),
491 0, pte_bit_id - popcount_id * POPCOUNT_BITS);
494 * Counting in reverse is trickier, since we must avoid
495 * reading from bytes that are not in range, and invert.
497 uint64_t pte_u64_bit_off = pte_u64 * BITS_IN(*bitmap);
500 bitN = MIN(popcount_id * POPCOUNT_BITS,
501 kd->pt_map_size * BITS_IN(uint8_t));
502 count = kd->pt_popcounts[popcount_id] - popcount_bytes(
504 pte_bit_id - pte_u64_bit_off, bitN - pte_u64_bit_off);
508 * This can only happen if the core is truncated. Treat these
509 * entries as if they don't exist, since their backing doesn't.
511 if (count >= (kd->pt_sparse_size / page_size))
514 return (kd->pt_sparse_off + (uint64_t)count * page_size);
518 kvm_fdnlist(kvm_t *kd, struct kvm_nlist *list)
523 if (kd->resolve_symbol == NULL) {
527 for (count = 0; list[count].n_name != NULL &&
528 list[count].n_name[0] != '\0'; count++)
530 nl = calloc(count + 1, sizeof(*nl));
531 for (i = 0; i < count; i++)
532 nl[i].n_name = list[i].n_name;
533 nfail = __fdnlist(kd->nlfd, nl);
534 for (i = 0; i < count; i++) {
535 list[i].n_type = nl[i].n_type;
536 list[i].n_value = nl[i].n_value;
543 while (list->n_name != NULL && list->n_name[0] != '\0') {
544 error = kd->resolve_symbol(list->n_name, &addr);
550 list->n_value = addr;
551 list->n_type = N_DATA | N_EXT;
559 * Walk the list of unresolved symbols, generate a new list and prefix the
560 * symbol names, try again, and merge back what we could resolve.
563 kvm_fdnlist_prefix(kvm_t *kd, struct kvm_nlist *nl, int missing,
564 const char *prefix, kvaddr_t (*validate_fn)(kvm_t *, kvaddr_t))
566 struct kvm_nlist *n, *np, *p;
570 int slen, unresolved;
573 * Calculate the space we need to malloc for nlist and names.
574 * We are going to store the name twice for later lookups: once
575 * with the prefix and once the unmodified name delmited by \0.
579 for (p = nl; p->n_name && p->n_name[0]; ++p) {
580 if (p->n_type != N_UNDF)
582 len += sizeof(struct kvm_nlist) + strlen(prefix) +
583 2 * (strlen(p->n_name) + 1);
588 /* Add space for the terminating nlist entry. */
589 len += sizeof(struct kvm_nlist);
592 /* Alloc one chunk for (nlist, [names]) and setup pointers. */
593 n = np = malloc(len);
597 cp = ce = (char *)np;
598 cp += unresolved * sizeof(struct kvm_nlist);
601 /* Generate shortened nlist with special prefix. */
603 for (p = nl; p->n_name && p->n_name[0]; ++p) {
604 if (p->n_type != N_UNDF)
607 /* Save the new\0orig. name so we can later match it again. */
608 slen = snprintf(cp, ce - cp, "%s%s%c%s", prefix,
609 (prefix[0] != '\0' && p->n_name[0] == '_') ?
610 (p->n_name + 1) : p->n_name, '\0', p->n_name);
611 if (slen < 0 || slen >= ce - cp)
619 /* Do lookup on the reduced list. */
621 unresolved = kvm_fdnlist(kd, np);
623 /* Check if we could resolve further symbols and update the list. */
624 if (unresolved >= 0 && unresolved < missing) {
625 /* Find the first freshly resolved entry. */
626 for (; np->n_name && np->n_name[0]; np++)
627 if (np->n_type != N_UNDF)
630 * The lists are both in the same order,
631 * so we can walk them in parallel.
633 for (p = nl; np->n_name && np->n_name[0] &&
634 p->n_name && p->n_name[0]; ++p) {
635 if (p->n_type != N_UNDF)
637 /* Skip expanded name and compare to orig. one. */
638 ccp = np->n_name + strlen(np->n_name) + 1;
639 if (strcmp(ccp, p->n_name) != 0)
641 /* Update nlist with new, translated results. */
642 p->n_type = np->n_type;
644 p->n_value = (*validate_fn)(kd, np->n_value);
646 p->n_value = np->n_value;
648 /* Find next freshly resolved entry. */
649 for (np++; np->n_name && np->n_name[0]; np++)
650 if (np->n_type != N_UNDF)
654 /* We could assert missing = unresolved here. */
661 _kvm_nlist(kvm_t *kd, struct kvm_nlist *nl, int initialize)
665 struct kld_sym_lookup lookup;
667 const char *prefix = "";
668 char symname[1024]; /* XXX-BZ symbol name length limit? */
669 int tried_vnet, tried_dpcpu;
672 * If we can't use the kld symbol lookup, revert to the
676 error = kvm_fdnlist(kd, nl);
677 if (error <= 0) /* Hard error or success. */
680 if (_kvm_vnet_initialized(kd, initialize))
681 error = kvm_fdnlist_prefix(kd, nl, error,
682 VNET_SYMPREFIX, _kvm_vnet_validaddr);
684 if (error > 0 && _kvm_dpcpu_initialized(kd, initialize))
685 error = kvm_fdnlist_prefix(kd, nl, error,
686 DPCPU_SYMPREFIX, _kvm_dpcpu_validaddr);
692 * We can use the kld lookup syscall. Go through each nlist entry
693 * and look it up with a kldsym(2) syscall.
699 for (p = nl; p->n_name && p->n_name[0]; ++p) {
700 if (p->n_type != N_UNDF)
703 lookup.version = sizeof(lookup);
707 error = snprintf(symname, sizeof(symname), "%s%s", prefix,
708 (prefix[0] != '\0' && p->n_name[0] == '_') ?
709 (p->n_name + 1) : p->n_name);
710 if (error < 0 || error >= (int)sizeof(symname))
712 lookup.symname = symname;
713 if (lookup.symname[0] == '_')
716 if (kldsym(0, KLDSYM_LOOKUP, &lookup) != -1) {
718 if (_kvm_vnet_initialized(kd, initialize) &&
719 strcmp(prefix, VNET_SYMPREFIX) == 0)
721 _kvm_vnet_validaddr(kd, lookup.symvalue);
722 else if (_kvm_dpcpu_initialized(kd, initialize) &&
723 strcmp(prefix, DPCPU_SYMPREFIX) == 0)
725 _kvm_dpcpu_validaddr(kd, lookup.symvalue);
727 p->n_value = lookup.symvalue;
734 * Check the number of entries that weren't found. If they exist,
735 * try again with a prefix for virtualized or DPCPU symbol names.
737 error = ((p - nl) - nvalid);
738 if (error && _kvm_vnet_initialized(kd, initialize) && !tried_vnet) {
740 prefix = VNET_SYMPREFIX;
743 if (error && _kvm_dpcpu_initialized(kd, initialize) && !tried_dpcpu) {
745 prefix = DPCPU_SYMPREFIX;
750 * Return the number of entries that weren't found. If they exist,
751 * also fill internal error buffer.
753 error = ((p - nl) - nvalid);
755 _kvm_syserr(kd, kd->program, "kvm_nlist");
760 _kvm_bitmap_init(struct kvm_bitmap *bm, u_long bitmapsize, u_long *idx)
764 bm->map = calloc(bitmapsize, sizeof *bm->map);
767 bm->size = bitmapsize;
772 _kvm_bitmap_set(struct kvm_bitmap *bm, u_long bm_index)
774 uint8_t *byte = &bm->map[bm_index / 8];
776 if (bm_index / 8 < bm->size)
777 *byte |= (1UL << (bm_index % 8));
781 _kvm_bitmap_next(struct kvm_bitmap *bm, u_long *idx)
783 u_long first_invalid = bm->size * CHAR_BIT;
785 if (*idx == ULONG_MAX)
790 /* Find the next valid idx. */
791 for (; *idx < first_invalid; (*idx)++) {
792 unsigned int mask = 1U << (*idx % CHAR_BIT);
793 if ((bm->map[*idx / CHAR_BIT] & mask) != 0)
797 return (*idx < first_invalid);
801 _kvm_bitmap_deinit(struct kvm_bitmap *bm)
808 _kvm_visit_cb(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg, u_long pa,
809 u_long kmap_vaddr, u_long dmap_vaddr, vm_prot_t prot, size_t len,
810 unsigned int page_size)
812 unsigned int pgsz = page_size ? page_size : len;
813 struct kvm_page p = {
814 .kp_version = LIBKVM_WALK_PAGES_VERSION,
816 .kp_kmap_vaddr = kmap_vaddr,
817 .kp_dmap_vaddr = dmap_vaddr,
819 .kp_offset = _kvm_pt_find(kd, pa, pgsz),