2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2005-2007 Joseph Koshy
5 * Copyright (c) 2007 The FreeBSD Foundation
8 * Portions of this software were developed by A. Joseph Koshy under
9 * sponsorship from the FreeBSD Foundation and Google, Inc.
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/param.h>
38 #include <sys/pmclog.h>
51 #include <machine/pmc_mdep.h>
53 #include "libpmcinternal.h"
55 #define PMCLOG_BUFFER_SIZE 512*1024
60 * The pmclog(3) API is oriented towards parsing an event stream in
61 * "realtime", i.e., from an data source that may or may not preserve
62 * record boundaries -- for example when the data source is elsewhere
63 * on a network. The API allows data to be fed into the parser zero
64 * or more bytes at a time.
66 * The state for a log file parser is maintained in a 'struct
67 * pmclog_parse_state'. Parser invocations are done by calling
68 * 'pmclog_read()'; this function will inform the caller when a
69 * complete event is parsed.
71 * The parser first assembles a complete log file event in an internal
72 * work area (see "ps_saved" below). Once a complete log file event
73 * is read, the parser then parses it and converts it to an event
74 * descriptor usable by the client. We could possibly avoid this two
75 * step process by directly parsing the input log to set fields in the
76 * event record. However the parser's state machine would get
77 * insanely complicated, and this code is unlikely to be used in
78 * performance critical paths.
81 #define PMCLOG_HEADER_FROM_SAVED_STATE(PS) \
82 (* ((uint32_t *) &(PS)->ps_saved))
84 #define PMCLOG_INITIALIZE_READER(LE,A) LE = (uint32_t *) &(A)
85 #define PMCLOG_READ32(LE,V) do { \
88 #define PMCLOG_READ64(LE,V) do { \
90 _v = (uint64_t) *(LE)++; \
91 _v |= ((uint64_t) *(LE)++) << 32; \
95 #define PMCLOG_READSTRING(LE,DST,LEN) strlcpy((DST), (char *) (LE), (LEN))
98 * Assemble a log record from '*len' octets starting from address '*data'.
99 * Update 'data' and 'len' to reflect the number of bytes consumed.
101 * '*data' is potentially an unaligned address and '*len' octets may
102 * not be enough to complete a event record.
105 static enum pmclog_parser_state
106 pmclog_get_record(struct pmclog_parse_state *ps, char **data, ssize_t *len)
108 int avail, copylen, recordsize, used;
110 const int HEADERSIZE = sizeof(uint32_t);
113 if ((avail = *len) <= 0)
114 return (ps->ps_state = PL_STATE_ERROR);
119 if (ps->ps_state == PL_STATE_NEW_RECORD)
122 dst = (char *) &ps->ps_saved + ps->ps_svcount;
124 switch (ps->ps_state) {
125 case PL_STATE_NEW_RECORD:
130 * Case A: avail < headersize
131 * -> 'expecting header'
133 * Case B: avail >= headersize
134 * B.1: avail < recordsize
135 * -> 'partial record'
136 * B.2: avail >= recordsize
140 copylen = avail < HEADERSIZE ? avail : HEADERSIZE;
141 bcopy(src, dst, copylen);
142 ps->ps_svcount = used = copylen;
144 if (copylen < HEADERSIZE) {
145 ps->ps_state = PL_STATE_EXPECTING_HEADER;
152 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
153 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
158 if (recordsize <= avail) { /* full record available */
159 bcopy(src, dst, recordsize - copylen);
160 ps->ps_svcount = used = recordsize;
164 /* header + a partial record is available */
165 bcopy(src, dst, avail - copylen);
166 ps->ps_svcount = used = avail;
167 ps->ps_state = PL_STATE_PARTIAL_RECORD;
171 case PL_STATE_EXPECTING_HEADER:
176 * Case C: avail+saved < headersize
177 * -> 'expecting header'
179 * Case D: avail+saved >= headersize
180 * D.1: avail+saved < recordsize
181 * -> 'partial record'
182 * D.2: avail+saved >= recordsize
184 * (see PARTIAL_RECORD handling below)
187 if (avail + ps->ps_svcount < HEADERSIZE) {
188 bcopy(src, dst, avail);
189 ps->ps_svcount += avail;
194 used = copylen = HEADERSIZE - ps->ps_svcount;
195 bcopy(src, dst, copylen);
199 ps->ps_svcount += copylen;
203 case PL_STATE_PARTIAL_RECORD:
208 * Case E: avail+saved < recordsize
209 * -> 'partial record'
211 * Case F: avail+saved >= recordsize
215 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
216 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
221 if (avail + ps->ps_svcount < recordsize) {
223 ps->ps_state = PL_STATE_PARTIAL_RECORD;
225 copylen = recordsize - ps->ps_svcount;
226 ps->ps_state = PL_STATE_NEW_RECORD;
229 bcopy(src, dst, copylen);
230 ps->ps_svcount += copylen;
244 ps->ps_state = PL_STATE_ERROR;
249 * Get an event from the stream pointed to by '*data'. '*len'
250 * indicates the number of bytes available to parse. Arguments
251 * '*data' and '*len' are updated to indicate the number of bytes
256 pmclog_get_event(void *cookie, char **data, ssize_t *len,
257 struct pmclog_ev *ev)
260 uint32_t h, *le, npc, noop;
261 enum pmclog_parser_state e;
262 struct pmclog_parse_state *ps;
263 struct pmclog_header *ph;
265 ps = (struct pmclog_parse_state *) cookie;
267 assert(ps->ps_state != PL_STATE_ERROR);
269 if ((e = pmclog_get_record(ps,data,len)) == PL_STATE_ERROR) {
270 ev->pl_state = PMCLOG_ERROR;
271 printf("state error\n");
275 if (e != PL_STATE_NEW_RECORD) {
276 ev->pl_state = PMCLOG_REQUIRE_DATA;
280 PMCLOG_INITIALIZE_READER(le, ps->ps_saved);
282 ph = (struct pmclog_header *)(uintptr_t)le;
285 if (!PMCLOG_HEADER_CHECK_MAGIC(h)) {
286 printf("bad magic\n");
287 ps->ps_state = PL_STATE_ERROR;
288 ev->pl_state = PMCLOG_ERROR;
292 /* copy out the time stamp */
293 ev->pl_ts.tv_sec = ph->pl_tsc;
296 evlen = PMCLOG_HEADER_TO_LENGTH(h);
298 #define PMCLOG_GET_PATHLEN(P,E,TYPE) do { \
299 (P) = (E) - offsetof(struct TYPE, pl_pathname); \
300 if ((P) > PATH_MAX || (P) < 0) \
304 #define PMCLOG_GET_CALLCHAIN_SIZE(SZ,E) do { \
305 (SZ) = ((E) - offsetof(struct pmclog_callchain, pl_pc)) \
306 / sizeof(uintfptr_t); \
309 switch (ev->pl_type = PMCLOG_HEADER_TO_TYPE(h)) {
310 case PMCLOG_TYPE_CALLCHAIN:
311 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pid);
312 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_tid);
313 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pmcid);
314 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_cpuflags);
315 PMCLOG_GET_CALLCHAIN_SIZE(ev->pl_u.pl_cc.pl_npc,evlen);
316 for (npc = 0; npc < ev->pl_u.pl_cc.pl_npc; npc++)
317 PMCLOG_READADDR(le,ev->pl_u.pl_cc.pl_pc[npc]);
318 for (;npc < PMC_CALLCHAIN_DEPTH_MAX; npc++)
319 ev->pl_u.pl_cc.pl_pc[npc] = (uintfptr_t) 0;
321 case PMCLOG_TYPE_CLOSELOG:
322 ev->pl_state = PMCLOG_EOF;
324 case PMCLOG_TYPE_DROPNOTIFY:
327 case PMCLOG_TYPE_INITIALIZE:
328 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_version);
329 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_arch);
330 PMCLOG_READ64(le,ev->pl_u.pl_i.pl_tsc_freq);
331 memcpy(&ev->pl_u.pl_i.pl_ts, le, sizeof(struct timespec));
332 le += sizeof(struct timespec)/4;
333 PMCLOG_READSTRING(le, ev->pl_u.pl_i.pl_cpuid, PMC_CPUID_LEN);
334 memcpy(ev->pl_u.pl_i.pl_cpuid, le, PMC_CPUID_LEN);
335 ps->ps_cpuid = strdup(ev->pl_u.pl_i.pl_cpuid);
336 ps->ps_version = ev->pl_u.pl_i.pl_version;
337 ps->ps_arch = ev->pl_u.pl_i.pl_arch;
338 ps->ps_initialized = 1;
340 case PMCLOG_TYPE_MAP_IN:
341 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
342 PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
343 PMCLOG_READ32(le,noop);
344 PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
345 PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
347 case PMCLOG_TYPE_MAP_OUT:
348 PMCLOG_READ32(le,ev->pl_u.pl_mo.pl_pid);
349 PMCLOG_READ32(le,noop);
350 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_start);
351 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_end);
353 case PMCLOG_TYPE_PMCALLOCATE:
354 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid);
355 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event);
356 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags);
357 PMCLOG_READ32(le,noop);
358 PMCLOG_READ64(le,ev->pl_u.pl_a.pl_rate);
359 ev->pl_u.pl_a.pl_evname = pmc_pmu_event_get_by_idx(ps->ps_cpuid, ev->pl_u.pl_a.pl_event);
360 if (ev->pl_u.pl_a.pl_evname != NULL)
362 else if ((ev->pl_u.pl_a.pl_evname =
363 _pmc_name_of_event(ev->pl_u.pl_a.pl_event, ps->ps_arch))
365 printf("unknown event\n");
369 case PMCLOG_TYPE_PMCALLOCATEDYN:
370 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_pmcid);
371 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_event);
372 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_flags);
373 PMCLOG_READ32(le,noop);
374 PMCLOG_READSTRING(le,ev->pl_u.pl_ad.pl_evname,PMC_NAME_MAX);
376 case PMCLOG_TYPE_PMCATTACH:
377 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach);
378 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid);
379 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid);
380 PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen);
382 case PMCLOG_TYPE_PMCDETACH:
383 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid);
384 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid);
386 case PMCLOG_TYPE_PROCCSW:
387 PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value);
388 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid);
389 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid);
390 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_tid);
392 case PMCLOG_TYPE_PROCEXEC:
393 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec);
394 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid);
395 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pmcid);
396 PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_entryaddr);
397 PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen);
399 case PMCLOG_TYPE_PROCEXIT:
400 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid);
401 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid);
402 PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value);
404 case PMCLOG_TYPE_PROCFORK:
405 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid);
406 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid);
408 case PMCLOG_TYPE_SYSEXIT:
409 PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid);
411 case PMCLOG_TYPE_USERDATA:
412 PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata);
414 case PMCLOG_TYPE_THR_CREATE:
415 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_tid);
416 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_pid);
417 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_flags);
418 PMCLOG_READ32(le,noop);
419 memcpy(ev->pl_u.pl_tc.pl_tdname, le, MAXCOMLEN+1);
421 case PMCLOG_TYPE_THR_EXIT:
422 PMCLOG_READ32(le,ev->pl_u.pl_te.pl_tid);
424 case PMCLOG_TYPE_PROC_CREATE:
425 PMCLOG_READ32(le,ev->pl_u.pl_pc.pl_pid);
426 PMCLOG_READ32(le,ev->pl_u.pl_pc.pl_flags);
427 memcpy(ev->pl_u.pl_pc.pl_pcomm, le, MAXCOMLEN+1);
429 default: /* unknown record type */
430 ps->ps_state = PL_STATE_ERROR;
431 ev->pl_state = PMCLOG_ERROR;
435 ev->pl_offset = (ps->ps_offset += evlen);
436 ev->pl_count = (ps->ps_count += 1);
438 ev->pl_state = PMCLOG_OK;
442 ev->pl_state = PMCLOG_ERROR;
443 ps->ps_state = PL_STATE_ERROR;
448 * Extract and return the next event from the byte stream.
450 * Returns 0 and sets the event's state to PMCLOG_OK in case an event
451 * was successfully parsed. Otherwise this function returns -1 and
452 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data
453 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if
454 * a parse error was encountered.
458 pmclog_read(void *cookie, struct pmclog_ev *ev)
462 struct pmclog_parse_state *ps;
464 ps = (struct pmclog_parse_state *) cookie;
466 if (ps->ps_state == PL_STATE_ERROR) {
467 ev->pl_state = PMCLOG_ERROR;
472 * If there isn't enough data left for a new event try and get
475 if (ps->ps_len == 0) {
476 ev->pl_state = PMCLOG_REQUIRE_DATA;
479 * If we have a valid file descriptor to read from, attempt
480 * to read from that. This read may return with an error,
481 * (which may be EAGAIN or other recoverable error), or
484 if (ps->ps_fd != PMCLOG_FD_NONE) {
486 nread = read(ps->ps_fd, ps->ps_buffer,
491 ev->pl_state = PMCLOG_EOF;
492 else if (errno != EAGAIN) /* not restartable */
493 ev->pl_state = PMCLOG_ERROR;
498 ps->ps_data = ps->ps_buffer;
504 assert(ps->ps_len > 0);
507 /* Retrieve one event from the byte stream. */
508 retval = pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev);
510 * If we need more data and we have a configured fd, try read
513 if (retval < 0 && ev->pl_state == PMCLOG_REQUIRE_DATA &&
515 assert(ps->ps_len == 0);
523 * Feed data to a memory based parser.
525 * The memory area pointed to by 'data' needs to be valid till the
526 * next error return from pmclog_next_event().
530 pmclog_feed(void *cookie, char *data, int len)
532 struct pmclog_parse_state *ps;
534 ps = (struct pmclog_parse_state *) cookie;
536 if (len < 0 || /* invalid length */
537 ps->ps_buffer || /* called for a file parser */
538 ps->ps_len != 0) /* unnecessary call */
548 * Allocate and initialize parser state.
554 struct pmclog_parse_state *ps;
556 if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL)
559 ps->ps_state = PL_STATE_NEW_RECORD;
561 ps->ps_initialized = 0;
563 ps->ps_offset = (off_t) 0;
564 bzero(&ps->ps_saved, sizeof(ps->ps_saved));
569 ps->ps_buffer = NULL;
572 /* allocate space for a work area */
573 if (ps->ps_fd != PMCLOG_FD_NONE) {
574 if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL) {
585 * Free up parser state.
589 pmclog_close(void *cookie)
591 struct pmclog_parse_state *ps;
593 ps = (struct pmclog_parse_state *) cookie;