2 * SPDX-License-Identifier: BSD-2-Clause
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 #include <sys/param.h>
36 #include <sys/pmclog.h>
49 #include <machine/pmc_mdep.h>
51 #include "libpmcinternal.h"
53 #define PMCLOG_BUFFER_SIZE 512*1024
58 * The pmclog(3) API is oriented towards parsing an event stream in
59 * "realtime", i.e., from an data source that may or may not preserve
60 * record boundaries -- for example when the data source is elsewhere
61 * on a network. The API allows data to be fed into the parser zero
62 * or more bytes at a time.
64 * The state for a log file parser is maintained in a 'struct
65 * pmclog_parse_state'. Parser invocations are done by calling
66 * 'pmclog_read()'; this function will inform the caller when a
67 * complete event is parsed.
69 * The parser first assembles a complete log file event in an internal
70 * work area (see "ps_saved" below). Once a complete log file event
71 * is read, the parser then parses it and converts it to an event
72 * descriptor usable by the client. We could possibly avoid this two
73 * step process by directly parsing the input log to set fields in the
74 * event record. However the parser's state machine would get
75 * insanely complicated, and this code is unlikely to be used in
76 * performance critical paths.
79 #define PMCLOG_HEADER_FROM_SAVED_STATE(PS) \
80 (* ((uint32_t *) &(PS)->ps_saved))
82 #define PMCLOG_INITIALIZE_READER(LE,A) LE = (uint32_t *) &(A)
83 #define PMCLOG_SKIP32(LE) (LE)++
84 #define PMCLOG_READ32(LE,V) do { \
87 #define PMCLOG_READ64(LE,V) do { \
89 _v = (uint64_t) *(LE)++; \
90 _v |= ((uint64_t) *(LE)++) << 32; \
94 #define PMCLOG_READSTRING(LE,DST,LEN) strlcpy((DST), (char *) (LE), (LEN))
97 * Assemble a log record from '*len' octets starting from address '*data'.
98 * Update 'data' and 'len' to reflect the number of bytes consumed.
100 * '*data' is potentially an unaligned address and '*len' octets may
101 * not be enough to complete a event record.
104 static enum pmclog_parser_state
105 pmclog_get_record(struct pmclog_parse_state *ps, char **data, ssize_t *len)
107 int avail, copylen, recordsize, used;
109 const int HEADERSIZE = sizeof(uint32_t);
112 if ((avail = *len) <= 0)
113 return (ps->ps_state = PL_STATE_ERROR);
118 if (ps->ps_state == PL_STATE_NEW_RECORD)
121 dst = (char *) &ps->ps_saved + ps->ps_svcount;
123 switch (ps->ps_state) {
124 case PL_STATE_NEW_RECORD:
129 * Case A: avail < headersize
130 * -> 'expecting header'
132 * Case B: avail >= headersize
133 * B.1: avail < recordsize
134 * -> 'partial record'
135 * B.2: avail >= recordsize
139 copylen = avail < HEADERSIZE ? avail : HEADERSIZE;
140 bcopy(src, dst, copylen);
141 ps->ps_svcount = used = copylen;
143 if (copylen < HEADERSIZE) {
144 ps->ps_state = PL_STATE_EXPECTING_HEADER;
151 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
152 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
157 if (recordsize <= avail) { /* full record available */
158 bcopy(src, dst, recordsize - copylen);
159 ps->ps_svcount = used = recordsize;
163 /* header + a partial record is available */
164 bcopy(src, dst, avail - copylen);
165 ps->ps_svcount = used = avail;
166 ps->ps_state = PL_STATE_PARTIAL_RECORD;
170 case PL_STATE_EXPECTING_HEADER:
175 * Case C: avail+saved < headersize
176 * -> 'expecting header'
178 * Case D: avail+saved >= headersize
179 * D.1: avail+saved < recordsize
180 * -> 'partial record'
181 * D.2: avail+saved >= recordsize
183 * (see PARTIAL_RECORD handling below)
186 if (avail + ps->ps_svcount < HEADERSIZE) {
187 bcopy(src, dst, avail);
188 ps->ps_svcount += avail;
193 used = copylen = HEADERSIZE - ps->ps_svcount;
194 bcopy(src, dst, copylen);
198 ps->ps_svcount += copylen;
202 case PL_STATE_PARTIAL_RECORD:
207 * Case E: avail+saved < recordsize
208 * -> 'partial record'
210 * Case F: avail+saved >= recordsize
214 h = PMCLOG_HEADER_FROM_SAVED_STATE(ps);
215 recordsize = PMCLOG_HEADER_TO_LENGTH(h);
220 if (avail + ps->ps_svcount < recordsize) {
222 ps->ps_state = PL_STATE_PARTIAL_RECORD;
224 copylen = recordsize - ps->ps_svcount;
225 ps->ps_state = PL_STATE_NEW_RECORD;
228 bcopy(src, dst, copylen);
229 ps->ps_svcount += copylen;
243 ps->ps_state = PL_STATE_ERROR;
248 * Get an event from the stream pointed to by '*data'. '*len'
249 * indicates the number of bytes available to parse. Arguments
250 * '*data' and '*len' are updated to indicate the number of bytes
255 pmclog_get_event(void *cookie, char **data, ssize_t *len,
256 struct pmclog_ev *ev)
259 uint32_t h, *le, npc;
260 enum pmclog_parser_state e;
261 struct pmclog_parse_state *ps;
262 struct pmclog_header *ph;
264 ps = (struct pmclog_parse_state *) cookie;
266 assert(ps->ps_state != PL_STATE_ERROR);
268 if ((e = pmclog_get_record(ps,data,len)) == PL_STATE_ERROR) {
269 ev->pl_state = PMCLOG_ERROR;
270 printf("state error\n");
274 if (e != PL_STATE_NEW_RECORD) {
275 ev->pl_state = PMCLOG_REQUIRE_DATA;
279 PMCLOG_INITIALIZE_READER(le, ps->ps_saved);
281 ph = (struct pmclog_header *)(uintptr_t)le;
284 if (!PMCLOG_HEADER_CHECK_MAGIC(h)) {
285 printf("bad magic\n");
286 ps->ps_state = PL_STATE_ERROR;
287 ev->pl_state = PMCLOG_ERROR;
291 /* copy out the time stamp */
292 ev->pl_ts.tv_sec = ph->pl_tsc;
295 evlen = PMCLOG_HEADER_TO_LENGTH(h);
297 #define PMCLOG_GET_PATHLEN(P,E,TYPE) do { \
298 (P) = (E) - offsetof(struct TYPE, pl_pathname); \
299 if ((P) > PATH_MAX || (P) < 0) \
303 #define PMCLOG_GET_CALLCHAIN_SIZE(SZ,E) do { \
304 (SZ) = ((E) - offsetof(struct pmclog_callchain, pl_pc)) \
305 / sizeof(uintfptr_t); \
308 switch (ev->pl_type = PMCLOG_HEADER_TO_TYPE(h)) {
309 case PMCLOG_TYPE_CALLCHAIN:
310 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pid);
311 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_tid);
312 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_pmcid);
313 PMCLOG_READ32(le,ev->pl_u.pl_cc.pl_cpuflags);
314 PMCLOG_GET_CALLCHAIN_SIZE(ev->pl_u.pl_cc.pl_npc,evlen);
315 for (npc = 0; npc < ev->pl_u.pl_cc.pl_npc; npc++)
316 PMCLOG_READADDR(le,ev->pl_u.pl_cc.pl_pc[npc]);
317 for (;npc < PMC_CALLCHAIN_DEPTH_MAX; npc++)
318 ev->pl_u.pl_cc.pl_pc[npc] = (uintfptr_t) 0;
320 case PMCLOG_TYPE_CLOSELOG:
321 ev->pl_state = PMCLOG_EOF;
323 case PMCLOG_TYPE_DROPNOTIFY:
326 case PMCLOG_TYPE_INITIALIZE:
327 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_version);
328 PMCLOG_READ32(le,ev->pl_u.pl_i.pl_arch);
329 PMCLOG_READ64(le,ev->pl_u.pl_i.pl_tsc_freq);
330 memcpy(&ev->pl_u.pl_i.pl_ts, le, sizeof(struct timespec));
331 le += sizeof(struct timespec)/4;
332 PMCLOG_READSTRING(le, ev->pl_u.pl_i.pl_cpuid, PMC_CPUID_LEN);
333 memcpy(ev->pl_u.pl_i.pl_cpuid, le, PMC_CPUID_LEN);
334 ps->ps_cpuid = strdup(ev->pl_u.pl_i.pl_cpuid);
335 ps->ps_version = ev->pl_u.pl_i.pl_version;
336 ps->ps_arch = ev->pl_u.pl_i.pl_arch;
337 ps->ps_initialized = 1;
339 case PMCLOG_TYPE_MAP_IN:
340 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_map_in);
341 PMCLOG_READ32(le,ev->pl_u.pl_mi.pl_pid);
343 PMCLOG_READADDR(le,ev->pl_u.pl_mi.pl_start);
344 PMCLOG_READSTRING(le, ev->pl_u.pl_mi.pl_pathname, pathlen);
346 case PMCLOG_TYPE_MAP_OUT:
347 PMCLOG_READ32(le,ev->pl_u.pl_mo.pl_pid);
349 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_start);
350 PMCLOG_READADDR(le,ev->pl_u.pl_mo.pl_end);
352 case PMCLOG_TYPE_PMCALLOCATE:
353 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_pmcid);
354 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_event);
355 PMCLOG_READ32(le,ev->pl_u.pl_a.pl_flags);
357 PMCLOG_READ64(le,ev->pl_u.pl_a.pl_rate);
360 * pl_event could contain either a PMC event code or a PMU
363 if ((ev->pl_u.pl_a.pl_flags & PMC_F_EV_PMU) != 0)
364 ev->pl_u.pl_a.pl_evname =
365 pmc_pmu_event_get_by_idx(ps->ps_cpuid,
366 ev->pl_u.pl_a.pl_event);
367 else if (ev->pl_u.pl_a.pl_event <= PMC_EVENT_LAST)
368 ev->pl_u.pl_a.pl_evname =
369 _pmc_name_of_event(ev->pl_u.pl_a.pl_event,
372 ev->pl_u.pl_a.pl_evname = NULL;
373 if (ev->pl_u.pl_a.pl_evname == NULL) {
374 printf("unknown event\n");
378 case PMCLOG_TYPE_PMCALLOCATEDYN:
379 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_pmcid);
380 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_event);
381 PMCLOG_READ32(le,ev->pl_u.pl_ad.pl_flags);
383 PMCLOG_READSTRING(le,ev->pl_u.pl_ad.pl_evname,PMC_NAME_MAX);
385 case PMCLOG_TYPE_PMCATTACH:
386 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_pmcattach);
387 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pmcid);
388 PMCLOG_READ32(le,ev->pl_u.pl_t.pl_pid);
389 PMCLOG_READSTRING(le,ev->pl_u.pl_t.pl_pathname,pathlen);
391 case PMCLOG_TYPE_PMCDETACH:
392 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pmcid);
393 PMCLOG_READ32(le,ev->pl_u.pl_d.pl_pid);
395 case PMCLOG_TYPE_PROCCSW:
396 PMCLOG_READ64(le,ev->pl_u.pl_c.pl_value);
397 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pmcid);
398 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_pid);
399 PMCLOG_READ32(le,ev->pl_u.pl_c.pl_tid);
401 case PMCLOG_TYPE_PROCEXEC:
402 PMCLOG_GET_PATHLEN(pathlen,evlen,pmclog_procexec);
403 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pid);
404 PMCLOG_READ32(le,ev->pl_u.pl_x.pl_pmcid);
405 PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_baseaddr);
406 PMCLOG_READADDR(le,ev->pl_u.pl_x.pl_dynaddr);
407 PMCLOG_READSTRING(le,ev->pl_u.pl_x.pl_pathname,pathlen);
409 case PMCLOG_TYPE_PROCEXIT:
410 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pmcid);
411 PMCLOG_READ32(le,ev->pl_u.pl_e.pl_pid);
412 PMCLOG_READ64(le,ev->pl_u.pl_e.pl_value);
414 case PMCLOG_TYPE_PROCFORK:
415 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_oldpid);
416 PMCLOG_READ32(le,ev->pl_u.pl_f.pl_newpid);
418 case PMCLOG_TYPE_SYSEXIT:
419 PMCLOG_READ32(le,ev->pl_u.pl_se.pl_pid);
421 case PMCLOG_TYPE_USERDATA:
422 PMCLOG_READ32(le,ev->pl_u.pl_u.pl_userdata);
424 case PMCLOG_TYPE_THR_CREATE:
425 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_tid);
426 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_pid);
427 PMCLOG_READ32(le,ev->pl_u.pl_tc.pl_flags);
429 memcpy(ev->pl_u.pl_tc.pl_tdname, le, MAXCOMLEN+1);
431 case PMCLOG_TYPE_THR_EXIT:
432 PMCLOG_READ32(le,ev->pl_u.pl_te.pl_tid);
434 case PMCLOG_TYPE_PROC_CREATE:
435 PMCLOG_READ32(le,ev->pl_u.pl_pc.pl_pid);
436 PMCLOG_READ32(le,ev->pl_u.pl_pc.pl_flags);
437 memcpy(ev->pl_u.pl_pc.pl_pcomm, le, MAXCOMLEN+1);
439 default: /* unknown record type */
440 ps->ps_state = PL_STATE_ERROR;
441 ev->pl_state = PMCLOG_ERROR;
445 ev->pl_offset = (ps->ps_offset += evlen);
446 ev->pl_count = (ps->ps_count += 1);
448 ev->pl_state = PMCLOG_OK;
452 ev->pl_state = PMCLOG_ERROR;
453 ps->ps_state = PL_STATE_ERROR;
458 * Extract and return the next event from the byte stream.
460 * Returns 0 and sets the event's state to PMCLOG_OK in case an event
461 * was successfully parsed. Otherwise this function returns -1 and
462 * sets the event's state to one of PMCLOG_REQUIRE_DATA (if more data
463 * is needed) or PMCLOG_EOF (if an EOF was seen) or PMCLOG_ERROR if
464 * a parse error was encountered.
468 pmclog_read(void *cookie, struct pmclog_ev *ev)
472 struct pmclog_parse_state *ps;
474 ps = (struct pmclog_parse_state *) cookie;
476 if (ps->ps_state == PL_STATE_ERROR) {
477 ev->pl_state = PMCLOG_ERROR;
482 * If there isn't enough data left for a new event try and get
485 if (ps->ps_len == 0) {
486 ev->pl_state = PMCLOG_REQUIRE_DATA;
489 * If we have a valid file descriptor to read from, attempt
490 * to read from that. This read may return with an error,
491 * (which may be EAGAIN or other recoverable error), or
494 if (ps->ps_fd != PMCLOG_FD_NONE) {
496 nread = read(ps->ps_fd, ps->ps_buffer,
501 ev->pl_state = PMCLOG_EOF;
502 else if (errno != EAGAIN) /* not restartable */
503 ev->pl_state = PMCLOG_ERROR;
508 ps->ps_data = ps->ps_buffer;
514 assert(ps->ps_len > 0);
517 /* Retrieve one event from the byte stream. */
518 retval = pmclog_get_event(ps, &ps->ps_data, &ps->ps_len, ev);
520 * If we need more data and we have a configured fd, try read
523 if (retval < 0 && ev->pl_state == PMCLOG_REQUIRE_DATA &&
525 assert(ps->ps_len == 0);
533 * Feed data to a memory based parser.
535 * The memory area pointed to by 'data' needs to be valid till the
536 * next error return from pmclog_next_event().
540 pmclog_feed(void *cookie, char *data, int len)
542 struct pmclog_parse_state *ps;
544 ps = (struct pmclog_parse_state *) cookie;
546 if (len < 0 || /* invalid length */
547 ps->ps_buffer || /* called for a file parser */
548 ps->ps_len != 0) /* unnecessary call */
558 * Allocate and initialize parser state.
564 struct pmclog_parse_state *ps;
566 if ((ps = (struct pmclog_parse_state *) malloc(sizeof(*ps))) == NULL)
569 ps->ps_state = PL_STATE_NEW_RECORD;
571 ps->ps_initialized = 0;
573 ps->ps_offset = (off_t) 0;
574 bzero(&ps->ps_saved, sizeof(ps->ps_saved));
579 ps->ps_buffer = NULL;
582 /* allocate space for a work area */
583 if (ps->ps_fd != PMCLOG_FD_NONE) {
584 if ((ps->ps_buffer = malloc(PMCLOG_BUFFER_SIZE)) == NULL) {
595 * Free up parser state.
599 pmclog_close(void *cookie)
601 struct pmclog_parse_state *ps;
603 ps = (struct pmclog_parse_state *) cookie;