pmcstat_log.c revision 299826
1/*-
2 * Copyright (c) 2005-2007, Joseph Koshy
3 * Copyright (c) 2007 The FreeBSD Foundation
4 * All rights reserved.
5 *
6 * Portions of this software were developed by A. Joseph Koshy under
7 * sponsorship from the FreeBSD Foundation and Google, Inc.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
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 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31/*
32 * Transform a hwpmc(4) log into human readable form, and into
33 * gprof(1) compatible profiles.
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: stable/10/usr.sbin/pmcstat/pmcstat_log.c 299826 2016-05-15 03:15:36Z pfg $");
38
39#include <sys/param.h>
40#include <sys/endian.h>
41#include <sys/cpuset.h>
42#include <sys/gmon.h>
43#include <sys/imgact_aout.h>
44#include <sys/imgact_elf.h>
45#include <sys/mman.h>
46#include <sys/pmc.h>
47#include <sys/queue.h>
48#include <sys/socket.h>
49#include <sys/stat.h>
50#include <sys/wait.h>
51
52#include <netinet/in.h>
53
54#include <assert.h>
55#include <curses.h>
56#include <err.h>
57#include <errno.h>
58#include <fcntl.h>
59#include <gelf.h>
60#include <libgen.h>
61#include <limits.h>
62#include <netdb.h>
63#include <pmc.h>
64#include <pmclog.h>
65#include <sysexits.h>
66#include <stdint.h>
67#include <stdio.h>
68#include <stdlib.h>
69#include <string.h>
70#include <unistd.h>
71
72#include "pmcstat.h"
73#include "pmcstat_log.h"
74#include "pmcstat_top.h"
75
76#define	PMCSTAT_ALLOCATE		1
77
78/*
79 * PUBLIC INTERFACES
80 *
81 * pmcstat_initialize_logging()	initialize this module, called first
82 * pmcstat_shutdown_logging()		orderly shutdown, called last
83 * pmcstat_open_log()			open an eventlog for processing
84 * pmcstat_process_log()		print/convert an event log
85 * pmcstat_display_log()		top mode display for the log
86 * pmcstat_close_log()			finish processing an event log
87 *
88 * IMPLEMENTATION NOTES
89 *
90 * We correlate each 'callchain' or 'sample' entry seen in the event
91 * log back to an executable object in the system. Executable objects
92 * include:
93 * 	- program executables,
94 *	- shared libraries loaded by the runtime loader,
95 *	- dlopen()'ed objects loaded by the program,
96 *	- the runtime loader itself,
97 *	- the kernel and kernel modules.
98 *
99 * Each process that we know about is treated as a set of regions that
100 * map to executable objects.  Processes are described by
101 * 'pmcstat_process' structures.  Executable objects are tracked by
102 * 'pmcstat_image' structures.  The kernel and kernel modules are
103 * common to all processes (they reside at the same virtual addresses
104 * for all processes).  Individual processes can have their text
105 * segments and shared libraries loaded at process-specific locations.
106 *
107 * A given executable object can be in use by multiple processes
108 * (e.g., libc.so) and loaded at a different address in each.
109 * pmcstat_pcmap structures track per-image mappings.
110 *
111 * The sample log could have samples from multiple PMCs; we
112 * generate one 'gmon.out' profile per PMC.
113 *
114 * IMPLEMENTATION OF GMON OUTPUT
115 *
116 * Each executable object gets one 'gmon.out' profile, per PMC in
117 * use.  Creation of 'gmon.out' profiles is done lazily.  The
118 * 'gmon.out' profiles generated for a given sampling PMC are
119 * aggregates of all the samples for that particular executable
120 * object.
121 *
122 * IMPLEMENTATION OF SYSTEM-WIDE CALLGRAPH OUTPUT
123 *
124 * Each active pmcid has its own callgraph structure, described by a
125 * 'struct pmcstat_callgraph'.  Given a process id and a list of pc
126 * values, we map each pc value to a tuple (image, symbol), where
127 * 'image' denotes an executable object and 'symbol' is the closest
128 * symbol that precedes the pc value.  Each pc value in the list is
129 * also given a 'rank' that reflects its depth in the call stack.
130 */
131
132struct pmcstat_pmcs pmcstat_pmcs = LIST_HEAD_INITIALIZER(pmcstat_pmcs);
133
134/*
135 * All image descriptors are kept in a hash table.
136 */
137struct pmcstat_image_hash_list pmcstat_image_hash[PMCSTAT_NHASH];
138
139/*
140 * All process descriptors are kept in a hash table.
141 */
142struct pmcstat_process_hash_list pmcstat_process_hash[PMCSTAT_NHASH];
143
144struct pmcstat_stats pmcstat_stats; /* statistics */
145static int ps_samples_period; /* samples count between top refresh. */
146
147struct pmcstat_process *pmcstat_kernproc; /* kernel 'process' */
148
149#include "pmcpl_gprof.h"
150#include "pmcpl_callgraph.h"
151#include "pmcpl_annotate.h"
152#include "pmcpl_annotate_cg.h"
153#include "pmcpl_calltree.h"
154
155static struct pmc_plugins  {
156	const char 	*pl_name;	/* name */
157
158	/* configure */
159	int (*pl_configure)(char *opt);
160
161	/* init and shutdown */
162	int (*pl_init)(void);
163	void (*pl_shutdown)(FILE *mf);
164
165	/* sample processing */
166	void (*pl_process)(struct pmcstat_process *pp,
167	    struct pmcstat_pmcrecord *pmcr, uint32_t nsamples,
168	    uintfptr_t *cc, int usermode, uint32_t cpu);
169
170	/* image */
171	void (*pl_initimage)(struct pmcstat_image *pi);
172	void (*pl_shutdownimage)(struct pmcstat_image *pi);
173
174	/* pmc */
175	void (*pl_newpmc)(pmcstat_interned_string ps,
176		struct pmcstat_pmcrecord *pr);
177
178	/* top display */
179	void (*pl_topdisplay)(void);
180
181	/* top keypress */
182	int (*pl_topkeypress)(int c, WINDOW *w);
183
184} plugins[] = {
185	{
186		.pl_name		= "none",
187	},
188	{
189		.pl_name		= "callgraph",
190		.pl_init		= pmcpl_cg_init,
191		.pl_shutdown		= pmcpl_cg_shutdown,
192		.pl_process		= pmcpl_cg_process,
193		.pl_topkeypress		= pmcpl_cg_topkeypress,
194		.pl_topdisplay		= pmcpl_cg_topdisplay
195	},
196	{
197		.pl_name		= "gprof",
198		.pl_shutdown		= pmcpl_gmon_shutdown,
199		.pl_process		= pmcpl_gmon_process,
200		.pl_initimage		= pmcpl_gmon_initimage,
201		.pl_shutdownimage	= pmcpl_gmon_shutdownimage,
202		.pl_newpmc		= pmcpl_gmon_newpmc
203	},
204	{
205		.pl_name		= "annotate",
206		.pl_process		= pmcpl_annotate_process
207	},
208	{
209		.pl_name		= "calltree",
210		.pl_configure		= pmcpl_ct_configure,
211		.pl_init		= pmcpl_ct_init,
212		.pl_shutdown		= pmcpl_ct_shutdown,
213		.pl_process		= pmcpl_ct_process,
214		.pl_topkeypress		= pmcpl_ct_topkeypress,
215		.pl_topdisplay		= pmcpl_ct_topdisplay
216	},
217	{
218		.pl_name		= "annotate_cg",
219		.pl_process		= pmcpl_annotate_cg_process
220	},
221
222	{
223		.pl_name		= NULL
224	}
225};
226
227static int pmcstat_mergepmc;
228
229int pmcstat_pmcinfilter = 0; /* PMC filter for top mode. */
230float pmcstat_threshold = 0.5; /* Cost filter for top mode. */
231
232/*
233 * Prototypes
234 */
235
236static struct pmcstat_image *pmcstat_image_from_path(pmcstat_interned_string
237    _path, int _iskernelmodule);
238static void pmcstat_image_get_aout_params(struct pmcstat_image *_image);
239static void pmcstat_image_get_elf_params(struct pmcstat_image *_image);
240static void	pmcstat_image_link(struct pmcstat_process *_pp,
241    struct pmcstat_image *_i, uintfptr_t _lpc);
242
243static void	pmcstat_pmcid_add(pmc_id_t _pmcid,
244    pmcstat_interned_string _name);
245
246static void	pmcstat_process_aout_exec(struct pmcstat_process *_pp,
247    struct pmcstat_image *_image, uintfptr_t _entryaddr);
248static void	pmcstat_process_elf_exec(struct pmcstat_process *_pp,
249    struct pmcstat_image *_image, uintfptr_t _entryaddr);
250static void	pmcstat_process_exec(struct pmcstat_process *_pp,
251    pmcstat_interned_string _path, uintfptr_t _entryaddr);
252static struct pmcstat_process *pmcstat_process_lookup(pid_t _pid,
253    int _allocate);
254static int	pmcstat_string_compute_hash(const char *_string);
255static void pmcstat_string_initialize(void);
256static int	pmcstat_string_lookup_hash(pmcstat_interned_string _is);
257static void pmcstat_string_shutdown(void);
258static void pmcstat_stats_reset(int _reset_global);
259
260/*
261 * A simple implementation of interned strings.  Each interned string
262 * is assigned a unique address, so that subsequent string compares
263 * can be done by a simple pointer comparison instead of using
264 * strcmp().  This speeds up hash table lookups and saves memory if
265 * duplicate strings are the norm.
266 */
267struct pmcstat_string {
268	LIST_ENTRY(pmcstat_string)	ps_next;	/* hash link */
269	int		ps_len;
270	int		ps_hash;
271	char		*ps_string;
272};
273
274static LIST_HEAD(,pmcstat_string)	pmcstat_string_hash[PMCSTAT_NHASH];
275
276/*
277 * PMC count.
278 */
279int pmcstat_npmcs;
280
281/*
282 * PMC Top mode pause state.
283 */
284static int pmcstat_pause;
285
286static void
287pmcstat_stats_reset(int reset_global)
288{
289	struct pmcstat_pmcrecord *pr;
290
291	/* Flush PMCs stats. */
292	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next) {
293		pr->pr_samples = 0;
294		pr->pr_dubious_frames = 0;
295	}
296	ps_samples_period = 0;
297
298	/* Flush global stats. */
299	if (reset_global)
300		bzero(&pmcstat_stats, sizeof(struct pmcstat_stats));
301}
302
303/*
304 * Compute a 'hash' value for a string.
305 */
306
307static int
308pmcstat_string_compute_hash(const char *s)
309{
310	unsigned hash;
311
312	for (hash = 2166136261; *s; s++)
313		hash = (hash ^ *s) * 16777619;
314
315	return (hash & PMCSTAT_HASH_MASK);
316}
317
318/*
319 * Intern a copy of string 's', and return a pointer to the
320 * interned structure.
321 */
322
323pmcstat_interned_string
324pmcstat_string_intern(const char *s)
325{
326	struct pmcstat_string *ps;
327	const struct pmcstat_string *cps;
328	int hash, len;
329
330	if ((cps = pmcstat_string_lookup(s)) != NULL)
331		return (cps);
332
333	hash = pmcstat_string_compute_hash(s);
334	len  = strlen(s);
335
336	if ((ps = malloc(sizeof(*ps))) == NULL)
337		err(EX_OSERR, "ERROR: Could not intern string");
338	ps->ps_len = len;
339	ps->ps_hash = hash;
340	ps->ps_string = strdup(s);
341	LIST_INSERT_HEAD(&pmcstat_string_hash[hash], ps, ps_next);
342	return ((pmcstat_interned_string) ps);
343}
344
345const char *
346pmcstat_string_unintern(pmcstat_interned_string str)
347{
348	const char *s;
349
350	s = ((const struct pmcstat_string *) str)->ps_string;
351	return (s);
352}
353
354pmcstat_interned_string
355pmcstat_string_lookup(const char *s)
356{
357	struct pmcstat_string *ps;
358	int hash, len;
359
360	hash = pmcstat_string_compute_hash(s);
361	len = strlen(s);
362
363	LIST_FOREACH(ps, &pmcstat_string_hash[hash], ps_next)
364	    if (ps->ps_len == len && ps->ps_hash == hash &&
365		strcmp(ps->ps_string, s) == 0)
366		    return (ps);
367	return (NULL);
368}
369
370static int
371pmcstat_string_lookup_hash(pmcstat_interned_string s)
372{
373	const struct pmcstat_string *ps;
374
375	ps = (const struct pmcstat_string *) s;
376	return (ps->ps_hash);
377}
378
379/*
380 * Initialize the string interning facility.
381 */
382
383static void
384pmcstat_string_initialize(void)
385{
386	int i;
387
388	for (i = 0; i < PMCSTAT_NHASH; i++)
389		LIST_INIT(&pmcstat_string_hash[i]);
390}
391
392/*
393 * Destroy the string table, free'ing up space.
394 */
395
396static void
397pmcstat_string_shutdown(void)
398{
399	int i;
400	struct pmcstat_string *ps, *pstmp;
401
402	for (i = 0; i < PMCSTAT_NHASH; i++)
403		LIST_FOREACH_SAFE(ps, &pmcstat_string_hash[i], ps_next,
404		    pstmp) {
405			LIST_REMOVE(ps, ps_next);
406			free(ps->ps_string);
407			free(ps);
408		}
409}
410
411/*
412 * Determine whether a given executable image is an A.OUT object, and
413 * if so, fill in its parameters from the text file.
414 * Sets image->pi_type.
415 */
416
417static void
418pmcstat_image_get_aout_params(struct pmcstat_image *image)
419{
420	int fd;
421	ssize_t nbytes;
422	struct exec ex;
423	const char *path;
424	char buffer[PATH_MAX];
425
426	path = pmcstat_string_unintern(image->pi_execpath);
427	assert(path != NULL);
428
429	if (image->pi_iskernelmodule)
430		errx(EX_SOFTWARE,
431		    "ERROR: a.out kernel modules are unsupported \"%s\"", path);
432
433	(void) snprintf(buffer, sizeof(buffer), "%s%s",
434	    args.pa_fsroot, path);
435
436	if ((fd = open(buffer, O_RDONLY, 0)) < 0 ||
437	    (nbytes = read(fd, &ex, sizeof(ex))) < 0) {
438		if (args.pa_verbosity >= 2)
439			warn("WARNING: Cannot determine type of \"%s\"",
440			    path);
441		image->pi_type = PMCSTAT_IMAGE_INDETERMINABLE;
442		if (fd != -1)
443			(void) close(fd);
444		return;
445	}
446
447	(void) close(fd);
448
449	if ((unsigned) nbytes != sizeof(ex) ||
450	    N_BADMAG(ex))
451		return;
452
453	image->pi_type = PMCSTAT_IMAGE_AOUT;
454
455	/* TODO: the rest of a.out processing */
456
457	return;
458}
459
460/*
461 * Helper function.
462 */
463
464static int
465pmcstat_symbol_compare(const void *a, const void *b)
466{
467	const struct pmcstat_symbol *sym1, *sym2;
468
469	sym1 = (const struct pmcstat_symbol *) a;
470	sym2 = (const struct pmcstat_symbol *) b;
471
472	if (sym1->ps_end <= sym2->ps_start)
473		return (-1);
474	if (sym1->ps_start >= sym2->ps_end)
475		return (1);
476	return (0);
477}
478
479/*
480 * Map an address to a symbol in an image.
481 */
482
483struct pmcstat_symbol *
484pmcstat_symbol_search(struct pmcstat_image *image, uintfptr_t addr)
485{
486	struct pmcstat_symbol sym;
487
488	if (image->pi_symbols == NULL)
489		return (NULL);
490
491	sym.ps_name  = NULL;
492	sym.ps_start = addr;
493	sym.ps_end   = addr + 1;
494
495	return (bsearch((void *) &sym, image->pi_symbols,
496		    image->pi_symcount, sizeof(struct pmcstat_symbol),
497		    pmcstat_symbol_compare));
498}
499
500/*
501 * Add the list of symbols in the given section to the list associated
502 * with the object.
503 */
504static void
505pmcstat_image_add_symbols(struct pmcstat_image *image, Elf *e,
506    Elf_Scn *scn, GElf_Shdr *sh)
507{
508	int firsttime;
509	size_t n, newsyms, nshsyms, nfuncsyms;
510	struct pmcstat_symbol *symptr;
511	char *fnname;
512	GElf_Sym sym;
513	Elf_Data *data;
514
515	if ((data = elf_getdata(scn, NULL)) == NULL)
516		return;
517
518	/*
519	 * Determine the number of functions named in this
520	 * section.
521	 */
522
523	nshsyms = sh->sh_size / sh->sh_entsize;
524	for (n = nfuncsyms = 0; n < nshsyms; n++) {
525		if (gelf_getsym(data, (int) n, &sym) != &sym)
526			return;
527		if (GELF_ST_TYPE(sym.st_info) == STT_FUNC)
528			nfuncsyms++;
529	}
530
531	if (nfuncsyms == 0)
532		return;
533
534	/*
535	 * Allocate space for the new entries.
536	 */
537	firsttime = image->pi_symbols == NULL;
538	symptr = realloc(image->pi_symbols,
539	    sizeof(*symptr) * (image->pi_symcount + nfuncsyms));
540	if (symptr == image->pi_symbols) /* realloc() failed. */
541		return;
542	image->pi_symbols = symptr;
543
544	/*
545	 * Append new symbols to the end of the current table.
546	 */
547	symptr += image->pi_symcount;
548
549	for (n = newsyms = 0; n < nshsyms; n++) {
550		if (gelf_getsym(data, (int) n, &sym) != &sym)
551			return;
552		if (GELF_ST_TYPE(sym.st_info) != STT_FUNC)
553			continue;
554		if (sym.st_shndx == STN_UNDEF)
555			continue;
556
557		if (!firsttime && pmcstat_symbol_search(image, sym.st_value))
558			continue; /* We've seen this symbol already. */
559
560		if ((fnname = elf_strptr(e, sh->sh_link, sym.st_name))
561		    == NULL)
562			continue;
563#ifdef __arm__
564		/* Remove spurious ARM function name. */
565		if (fnname[0] == '$' &&
566		    (fnname[1] == 'a' || fnname[1] == 't' ||
567		    fnname[1] == 'd') &&
568		    fnname[2] == '\0')
569			continue;
570#endif
571
572		symptr->ps_name  = pmcstat_string_intern(fnname);
573		symptr->ps_start = sym.st_value - image->pi_vaddr;
574		symptr->ps_end   = symptr->ps_start + sym.st_size;
575		symptr++;
576
577		newsyms++;
578	}
579
580	image->pi_symcount += newsyms;
581	if (image->pi_symcount == 0)
582		return;
583
584	assert(newsyms <= nfuncsyms);
585
586	/*
587	 * Return space to the system if there were duplicates.
588	 */
589	if (newsyms < nfuncsyms)
590		image->pi_symbols = realloc(image->pi_symbols,
591		    sizeof(*symptr) * image->pi_symcount);
592
593	/*
594	 * Keep the list of symbols sorted.
595	 */
596	qsort(image->pi_symbols, image->pi_symcount, sizeof(*symptr),
597	    pmcstat_symbol_compare);
598
599	/*
600	 * Deal with function symbols that have a size of 'zero' by
601	 * making them extend to the next higher address.  These
602	 * symbols are usually defined in assembly code.
603	 */
604	for (symptr = image->pi_symbols;
605	     symptr < image->pi_symbols + (image->pi_symcount - 1);
606	     symptr++)
607		if (symptr->ps_start == symptr->ps_end)
608			symptr->ps_end = (symptr+1)->ps_start;
609}
610
611/*
612 * Examine an ELF file to determine the size of its text segment.
613 * Sets image->pi_type if anything conclusive can be determined about
614 * this image.
615 */
616
617static void
618pmcstat_image_get_elf_params(struct pmcstat_image *image)
619{
620	int fd;
621	size_t i, nph, nsh;
622	const char *path, *elfbase;
623	char *p, *endp;
624	uintfptr_t minva, maxva;
625	Elf *e;
626	Elf_Scn *scn;
627	GElf_Ehdr eh;
628	GElf_Phdr ph;
629	GElf_Shdr sh;
630	enum pmcstat_image_type image_type;
631	char buffer[PATH_MAX];
632
633	assert(image->pi_type == PMCSTAT_IMAGE_UNKNOWN);
634
635	image->pi_start = minva = ~(uintfptr_t) 0;
636	image->pi_end = maxva = (uintfptr_t) 0;
637	image->pi_type = image_type = PMCSTAT_IMAGE_INDETERMINABLE;
638	image->pi_isdynamic = 0;
639	image->pi_dynlinkerpath = NULL;
640	image->pi_vaddr = 0;
641
642	path = pmcstat_string_unintern(image->pi_execpath);
643	assert(path != NULL);
644
645	/*
646	 * Look for kernel modules under FSROOT/KERNELPATH/NAME,
647	 * and user mode executable objects under FSROOT/PATHNAME.
648	 */
649	if (image->pi_iskernelmodule)
650		(void) snprintf(buffer, sizeof(buffer), "%s%s/%s",
651		    args.pa_fsroot, args.pa_kernel, path);
652	else
653		(void) snprintf(buffer, sizeof(buffer), "%s%s",
654		    args.pa_fsroot, path);
655
656	e = NULL;
657	if ((fd = open(buffer, O_RDONLY, 0)) < 0 ||
658	    (e = elf_begin(fd, ELF_C_READ, NULL)) == NULL ||
659	    (elf_kind(e) != ELF_K_ELF)) {
660		if (args.pa_verbosity >= 2)
661			warnx("WARNING: Cannot determine the type of \"%s\".",
662			    buffer);
663		goto done;
664	}
665
666	if (gelf_getehdr(e, &eh) != &eh) {
667		warnx(
668		    "WARNING: Cannot retrieve the ELF Header for \"%s\": %s.",
669		    buffer, elf_errmsg(-1));
670		goto done;
671	}
672
673	if (eh.e_type != ET_EXEC && eh.e_type != ET_DYN &&
674	    !(image->pi_iskernelmodule && eh.e_type == ET_REL)) {
675		warnx("WARNING: \"%s\" is of an unsupported ELF type.",
676		    buffer);
677		goto done;
678	}
679
680	image_type = eh.e_ident[EI_CLASS] == ELFCLASS32 ?
681	    PMCSTAT_IMAGE_ELF32 : PMCSTAT_IMAGE_ELF64;
682
683	/*
684	 * Determine the virtual address where an executable would be
685	 * loaded.  Additionally, for dynamically linked executables,
686	 * save the pathname to the runtime linker.
687	 */
688	if (eh.e_type == ET_EXEC) {
689		if (elf_getphnum(e, &nph) == 0) {
690			warnx(
691"WARNING: Could not determine the number of program headers in \"%s\": %s.",
692			    buffer,
693			    elf_errmsg(-1));
694			goto done;
695		}
696		for (i = 0; i < eh.e_phnum; i++) {
697			if (gelf_getphdr(e, i, &ph) != &ph) {
698				warnx(
699"WARNING: Retrieval of PHDR entry #%ju in \"%s\" failed: %s.",
700				    (uintmax_t) i, buffer, elf_errmsg(-1));
701				goto done;
702			}
703			switch (ph.p_type) {
704			case PT_DYNAMIC:
705				image->pi_isdynamic = 1;
706				break;
707			case PT_INTERP:
708				if ((elfbase = elf_rawfile(e, NULL)) == NULL) {
709					warnx(
710"WARNING: Cannot retrieve the interpreter for \"%s\": %s.",
711					    buffer, elf_errmsg(-1));
712					goto done;
713				}
714				image->pi_dynlinkerpath =
715				    pmcstat_string_intern(elfbase +
716				        ph.p_offset);
717				break;
718			case PT_LOAD:
719				if ((ph.p_flags & PF_X) != 0 &&
720				    (ph.p_offset & (-ph.p_align)) == 0)
721					image->pi_vaddr = ph.p_vaddr & (-ph.p_align);
722				break;
723			}
724		}
725	}
726
727	/*
728	 * Get the min and max VA associated with this ELF object.
729	 */
730	if (elf_getshnum(e, &nsh) == 0) {
731		warnx(
732"WARNING: Could not determine the number of sections for \"%s\": %s.",
733		    buffer, elf_errmsg(-1));
734		goto done;
735	}
736
737	for (i = 0; i < nsh; i++) {
738		if ((scn = elf_getscn(e, i)) == NULL ||
739		    gelf_getshdr(scn, &sh) != &sh) {
740			warnx(
741"WARNING: Could not retrieve section header #%ju in \"%s\": %s.",
742			    (uintmax_t) i, buffer, elf_errmsg(-1));
743			goto done;
744		}
745		if (sh.sh_flags & SHF_EXECINSTR) {
746			minva = min(minva, sh.sh_addr);
747			maxva = max(maxva, sh.sh_addr + sh.sh_size);
748		}
749		if (sh.sh_type == SHT_SYMTAB || sh.sh_type == SHT_DYNSYM)
750			pmcstat_image_add_symbols(image, e, scn, &sh);
751	}
752
753	image->pi_start = minva;
754	image->pi_end   = maxva;
755	image->pi_type  = image_type;
756	image->pi_fullpath = pmcstat_string_intern(buffer);
757
758	/* Build display name
759	 */
760	endp = buffer;
761	for (p = buffer; *p; p++)
762		if (*p == '/')
763			endp = p+1;
764	image->pi_name = pmcstat_string_intern(endp);
765
766 done:
767	(void) elf_end(e);
768	if (fd >= 0)
769		(void) close(fd);
770	return;
771}
772
773/*
774 * Given an image descriptor, determine whether it is an ELF, or AOUT.
775 * If no handler claims the image, set its type to 'INDETERMINABLE'.
776 */
777
778void
779pmcstat_image_determine_type(struct pmcstat_image *image)
780{
781	assert(image->pi_type == PMCSTAT_IMAGE_UNKNOWN);
782
783	/* Try each kind of handler in turn */
784	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
785		pmcstat_image_get_elf_params(image);
786	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
787		pmcstat_image_get_aout_params(image);
788
789	/*
790	 * Otherwise, remember that we tried to determine
791	 * the object's type and had failed.
792	 */
793	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
794		image->pi_type = PMCSTAT_IMAGE_INDETERMINABLE;
795}
796
797/*
798 * Locate an image descriptor given an interned path, adding a fresh
799 * descriptor to the cache if necessary.  This function also finds a
800 * suitable name for this image's sample file.
801 *
802 * We defer filling in the file format specific parts of the image
803 * structure till the time we actually see a sample that would fall
804 * into this image.
805 */
806
807static struct pmcstat_image *
808pmcstat_image_from_path(pmcstat_interned_string internedpath,
809    int iskernelmodule)
810{
811	int hash;
812	struct pmcstat_image *pi;
813
814	hash = pmcstat_string_lookup_hash(internedpath);
815
816	/* First, look for an existing entry. */
817	LIST_FOREACH(pi, &pmcstat_image_hash[hash], pi_next)
818	    if (pi->pi_execpath == internedpath &&
819		  pi->pi_iskernelmodule == iskernelmodule)
820		    return (pi);
821
822	/*
823	 * Allocate a new entry and place it at the head of the hash
824	 * and LRU lists.
825	 */
826	pi = malloc(sizeof(*pi));
827	if (pi == NULL)
828		return (NULL);
829
830	pi->pi_type = PMCSTAT_IMAGE_UNKNOWN;
831	pi->pi_execpath = internedpath;
832	pi->pi_start = ~0;
833	pi->pi_end = 0;
834	pi->pi_entry = 0;
835	pi->pi_vaddr = 0;
836	pi->pi_isdynamic = 0;
837	pi->pi_iskernelmodule = iskernelmodule;
838	pi->pi_dynlinkerpath = NULL;
839	pi->pi_symbols = NULL;
840	pi->pi_symcount = 0;
841	pi->pi_addr2line = NULL;
842
843	if (plugins[args.pa_pplugin].pl_initimage != NULL)
844		plugins[args.pa_pplugin].pl_initimage(pi);
845	if (plugins[args.pa_plugin].pl_initimage != NULL)
846		plugins[args.pa_plugin].pl_initimage(pi);
847
848	LIST_INSERT_HEAD(&pmcstat_image_hash[hash], pi, pi_next);
849
850	return (pi);
851}
852
853/*
854 * Record the fact that PC values from 'start' to 'end' come from
855 * image 'image'.
856 */
857
858static void
859pmcstat_image_link(struct pmcstat_process *pp, struct pmcstat_image *image,
860    uintfptr_t start)
861{
862	struct pmcstat_pcmap *pcm, *pcmnew;
863	uintfptr_t offset;
864
865	assert(image->pi_type != PMCSTAT_IMAGE_UNKNOWN &&
866	    image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE);
867
868	if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
869		err(EX_OSERR, "ERROR: Cannot create a map entry");
870
871	/*
872	 * Adjust the map entry to only cover the text portion
873	 * of the object.
874	 */
875
876	offset = start - image->pi_vaddr;
877	pcmnew->ppm_lowpc  = image->pi_start + offset;
878	pcmnew->ppm_highpc = image->pi_end + offset;
879	pcmnew->ppm_image  = image;
880
881	assert(pcmnew->ppm_lowpc < pcmnew->ppm_highpc);
882
883	/* Overlapped mmap()'s are assumed to never occur. */
884	TAILQ_FOREACH(pcm, &pp->pp_map, ppm_next)
885	    if (pcm->ppm_lowpc >= pcmnew->ppm_highpc)
886		    break;
887
888	if (pcm == NULL)
889		TAILQ_INSERT_TAIL(&pp->pp_map, pcmnew, ppm_next);
890	else
891		TAILQ_INSERT_BEFORE(pcm, pcmnew, ppm_next);
892}
893
894/*
895 * Unmap images in the range [start..end) associated with process
896 * 'pp'.
897 */
898
899static void
900pmcstat_image_unmap(struct pmcstat_process *pp, uintfptr_t start,
901    uintfptr_t end)
902{
903	struct pmcstat_pcmap *pcm, *pcmtmp, *pcmnew;
904
905	assert(pp != NULL);
906	assert(start < end);
907
908	/*
909	 * Cases:
910	 * - we could have the range completely in the middle of an
911	 *   existing pcmap; in this case we have to split the pcmap
912	 *   structure into two (i.e., generate a 'hole').
913	 * - we could have the range covering multiple pcmaps; these
914	 *   will have to be removed.
915	 * - we could have either 'start' or 'end' falling in the
916	 *   middle of a pcmap; in this case shorten the entry.
917	 */
918	TAILQ_FOREACH_SAFE(pcm, &pp->pp_map, ppm_next, pcmtmp) {
919		assert(pcm->ppm_lowpc < pcm->ppm_highpc);
920		if (pcm->ppm_highpc <= start)
921			continue;
922		if (pcm->ppm_lowpc >= end)
923			return;
924		if (pcm->ppm_lowpc >= start && pcm->ppm_highpc <= end) {
925			/*
926			 * The current pcmap is completely inside the
927			 * unmapped range: remove it entirely.
928			 */
929			TAILQ_REMOVE(&pp->pp_map, pcm, ppm_next);
930			free(pcm);
931		} else if (pcm->ppm_lowpc < start && pcm->ppm_highpc > end) {
932			/*
933			 * Split this pcmap into two; curtail the
934			 * current map to end at [start-1], and start
935			 * the new one at [end].
936			 */
937			if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
938				err(EX_OSERR,
939				    "ERROR: Cannot split a map entry");
940
941			pcmnew->ppm_image = pcm->ppm_image;
942
943			pcmnew->ppm_lowpc = end;
944			pcmnew->ppm_highpc = pcm->ppm_highpc;
945
946			pcm->ppm_highpc = start;
947
948			TAILQ_INSERT_AFTER(&pp->pp_map, pcm, pcmnew, ppm_next);
949
950			return;
951		} else if (pcm->ppm_lowpc < start && pcm->ppm_highpc <= end)
952			pcm->ppm_highpc = start;
953		else if (pcm->ppm_lowpc >= start && pcm->ppm_highpc > end)
954			pcm->ppm_lowpc = end;
955		else
956			assert(0);
957	}
958}
959
960/*
961 * Resolve file name and line number for the given address.
962 */
963int
964pmcstat_image_addr2line(struct pmcstat_image *image, uintfptr_t addr,
965    char *sourcefile, size_t sourcefile_len, unsigned *sourceline,
966    char *funcname, size_t funcname_len)
967{
968	static int addr2line_warn = 0;
969	unsigned l;
970
971	char *sep, cmdline[PATH_MAX], imagepath[PATH_MAX];
972	int fd;
973
974	if (image->pi_addr2line == NULL) {
975		snprintf(imagepath, sizeof(imagepath), "%s%s.symbols",
976		    args.pa_fsroot,
977		    pmcstat_string_unintern(image->pi_fullpath));
978		fd = open(imagepath, O_RDONLY);
979		if (fd < 0) {
980			snprintf(imagepath, sizeof(imagepath), "%s%s",
981			    args.pa_fsroot,
982			    pmcstat_string_unintern(image->pi_fullpath));
983		} else
984			close(fd);
985		/*
986		 * New addr2line support recursive inline function with -i
987		 * but the format does not add a marker when no more entries
988		 * are available.
989		 */
990		snprintf(cmdline, sizeof(cmdline), "addr2line -Cfe \"%s\"",
991		    imagepath);
992		image->pi_addr2line = popen(cmdline, "r+");
993		if (image->pi_addr2line == NULL) {
994			if (!addr2line_warn) {
995				addr2line_warn = 1;
996				warnx(
997"WARNING: addr2line is needed for source code information."
998				    );
999			}
1000			return (0);
1001		}
1002	}
1003
1004	if (feof(image->pi_addr2line) || ferror(image->pi_addr2line)) {
1005		warnx("WARNING: addr2line pipe error");
1006		pclose(image->pi_addr2line);
1007		image->pi_addr2line = NULL;
1008		return (0);
1009	}
1010
1011	fprintf(image->pi_addr2line, "%p\n", (void *)addr);
1012
1013	if (fgets(funcname, funcname_len, image->pi_addr2line) == NULL) {
1014		warnx("WARNING: addr2line function name read error");
1015		return (0);
1016	}
1017	sep = strchr(funcname, '\n');
1018	if (sep != NULL)
1019		*sep = '\0';
1020
1021	if (fgets(sourcefile, sourcefile_len, image->pi_addr2line) == NULL) {
1022		warnx("WARNING: addr2line source file read error");
1023		return (0);
1024	}
1025	sep = strchr(sourcefile, ':');
1026	if (sep == NULL) {
1027		warnx("WARNING: addr2line source line separator missing");
1028		return (0);
1029	}
1030	*sep = '\0';
1031	l = atoi(sep+1);
1032	if (l == 0)
1033		return (0);
1034	*sourceline = l;
1035	return (1);
1036}
1037
1038/*
1039 * Add a {pmcid,name} mapping.
1040 */
1041
1042static void
1043pmcstat_pmcid_add(pmc_id_t pmcid, pmcstat_interned_string ps)
1044{
1045	struct pmcstat_pmcrecord *pr, *prm;
1046
1047	/* Replace an existing name for the PMC. */
1048	prm = NULL;
1049	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1050		if (pr->pr_pmcid == pmcid) {
1051			pr->pr_pmcname = ps;
1052			return;
1053		} else if (pr->pr_pmcname == ps)
1054			prm = pr;
1055
1056	/*
1057	 * Otherwise, allocate a new descriptor and call the
1058	 * plugins hook.
1059	 */
1060	if ((pr = malloc(sizeof(*pr))) == NULL)
1061		err(EX_OSERR, "ERROR: Cannot allocate pmc record");
1062
1063	pr->pr_pmcid = pmcid;
1064	pr->pr_pmcname = ps;
1065	pr->pr_pmcin = pmcstat_npmcs++;
1066	pr->pr_samples = 0;
1067	pr->pr_dubious_frames = 0;
1068	pr->pr_merge = prm == NULL ? pr : prm;
1069
1070	LIST_INSERT_HEAD(&pmcstat_pmcs, pr, pr_next);
1071
1072	if (plugins[args.pa_pplugin].pl_newpmc != NULL)
1073		plugins[args.pa_pplugin].pl_newpmc(ps, pr);
1074	if (plugins[args.pa_plugin].pl_newpmc != NULL)
1075		plugins[args.pa_plugin].pl_newpmc(ps, pr);
1076}
1077
1078/*
1079 * Given a pmcid in use, find its human-readable name.
1080 */
1081
1082const char *
1083pmcstat_pmcid_to_name(pmc_id_t pmcid)
1084{
1085	struct pmcstat_pmcrecord *pr;
1086
1087	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1088	    if (pr->pr_pmcid == pmcid)
1089		    return (pmcstat_string_unintern(pr->pr_pmcname));
1090
1091	return NULL;
1092}
1093
1094/*
1095 * Convert PMC index to name.
1096 */
1097
1098const char *
1099pmcstat_pmcindex_to_name(int pmcin)
1100{
1101	struct pmcstat_pmcrecord *pr;
1102
1103	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1104		if (pr->pr_pmcin == pmcin)
1105			return pmcstat_string_unintern(pr->pr_pmcname);
1106
1107	return NULL;
1108}
1109
1110/*
1111 * Return PMC record with given index.
1112 */
1113
1114struct pmcstat_pmcrecord *
1115pmcstat_pmcindex_to_pmcr(int pmcin)
1116{
1117	struct pmcstat_pmcrecord *pr;
1118
1119	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1120		if (pr->pr_pmcin == pmcin)
1121			return pr;
1122
1123	return NULL;
1124}
1125
1126/*
1127 * Get PMC record by id, apply merge policy.
1128 */
1129
1130static struct pmcstat_pmcrecord *
1131pmcstat_lookup_pmcid(pmc_id_t pmcid)
1132{
1133	struct pmcstat_pmcrecord *pr;
1134
1135	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next) {
1136		if (pr->pr_pmcid == pmcid) {
1137			if (pmcstat_mergepmc)
1138				return pr->pr_merge;
1139			return pr;
1140		}
1141	}
1142
1143	return NULL;
1144}
1145
1146/*
1147 * Associate an AOUT image with a process.
1148 */
1149
1150static void
1151pmcstat_process_aout_exec(struct pmcstat_process *pp,
1152    struct pmcstat_image *image, uintfptr_t entryaddr)
1153{
1154	(void) pp;
1155	(void) image;
1156	(void) entryaddr;
1157	/* TODO Implement a.out handling */
1158}
1159
1160/*
1161 * Associate an ELF image with a process.
1162 */
1163
1164static void
1165pmcstat_process_elf_exec(struct pmcstat_process *pp,
1166    struct pmcstat_image *image, uintfptr_t entryaddr)
1167{
1168	uintmax_t libstart;
1169	struct pmcstat_image *rtldimage;
1170
1171	assert(image->pi_type == PMCSTAT_IMAGE_ELF32 ||
1172	    image->pi_type == PMCSTAT_IMAGE_ELF64);
1173
1174	/* Create a map entry for the base executable. */
1175	pmcstat_image_link(pp, image, image->pi_vaddr);
1176
1177	/*
1178	 * For dynamically linked executables we need to determine
1179	 * where the dynamic linker was mapped to for this process,
1180	 * Subsequent executable objects that are mapped in by the
1181	 * dynamic linker will be tracked by log events of type
1182	 * PMCLOG_TYPE_MAP_IN.
1183	 */
1184
1185	if (image->pi_isdynamic) {
1186
1187		/*
1188		 * The runtime loader gets loaded just after the maximum
1189		 * possible heap address.  Like so:
1190		 *
1191		 * [  TEXT DATA BSS HEAP -->*RTLD  SHLIBS   <--STACK]
1192		 * ^					            ^
1193		 * 0				   VM_MAXUSER_ADDRESS
1194
1195		 *
1196		 * The exact address where the loader gets mapped in
1197		 * will vary according to the size of the executable
1198		 * and the limits on the size of the process'es data
1199		 * segment at the time of exec().  The entry address
1200		 * recorded at process exec time corresponds to the
1201		 * 'start' address inside the dynamic linker.  From
1202		 * this we can figure out the address where the
1203		 * runtime loader's file object had been mapped to.
1204		 */
1205		rtldimage = pmcstat_image_from_path(image->pi_dynlinkerpath, 0);
1206		if (rtldimage == NULL) {
1207			warnx("WARNING: Cannot find image for \"%s\".",
1208			    pmcstat_string_unintern(image->pi_dynlinkerpath));
1209			pmcstat_stats.ps_exec_errors++;
1210			return;
1211		}
1212
1213		if (rtldimage->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1214			pmcstat_image_get_elf_params(rtldimage);
1215
1216		if (rtldimage->pi_type != PMCSTAT_IMAGE_ELF32 &&
1217		    rtldimage->pi_type != PMCSTAT_IMAGE_ELF64) {
1218			warnx("WARNING: rtld not an ELF object \"%s\".",
1219			    pmcstat_string_unintern(image->pi_dynlinkerpath));
1220			return;
1221		}
1222
1223		libstart = entryaddr - rtldimage->pi_entry;
1224		pmcstat_image_link(pp, rtldimage, libstart);
1225	}
1226}
1227
1228/*
1229 * Find the process descriptor corresponding to a PID.  If 'allocate'
1230 * is zero, we return a NULL if a pid descriptor could not be found or
1231 * a process descriptor process.  If 'allocate' is non-zero, then we
1232 * will attempt to allocate a fresh process descriptor.  Zombie
1233 * process descriptors are only removed if a fresh allocation for the
1234 * same PID is requested.
1235 */
1236
1237static struct pmcstat_process *
1238pmcstat_process_lookup(pid_t pid, int allocate)
1239{
1240	uint32_t hash;
1241	struct pmcstat_pcmap *ppm, *ppmtmp;
1242	struct pmcstat_process *pp, *pptmp;
1243
1244	hash = (uint32_t) pid & PMCSTAT_HASH_MASK;	/* simplicity wins */
1245
1246	LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[hash], pp_next, pptmp)
1247		if (pp->pp_pid == pid) {
1248			/* Found a descriptor, check and process zombies */
1249			if (allocate && pp->pp_isactive == 0) {
1250				/* remove maps */
1251				TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next,
1252				    ppmtmp) {
1253					TAILQ_REMOVE(&pp->pp_map, ppm,
1254					    ppm_next);
1255					free(ppm);
1256				}
1257				/* remove process entry */
1258				LIST_REMOVE(pp, pp_next);
1259				free(pp);
1260				break;
1261			}
1262			return (pp);
1263		}
1264
1265	if (!allocate)
1266		return (NULL);
1267
1268	if ((pp = malloc(sizeof(*pp))) == NULL)
1269		err(EX_OSERR, "ERROR: Cannot allocate pid descriptor");
1270
1271	pp->pp_pid = pid;
1272	pp->pp_isactive = 1;
1273
1274	TAILQ_INIT(&pp->pp_map);
1275
1276	LIST_INSERT_HEAD(&pmcstat_process_hash[hash], pp, pp_next);
1277	return (pp);
1278}
1279
1280/*
1281 * Associate an image and a process.
1282 */
1283
1284static void
1285pmcstat_process_exec(struct pmcstat_process *pp,
1286    pmcstat_interned_string path, uintfptr_t entryaddr)
1287{
1288	struct pmcstat_image *image;
1289
1290	if ((image = pmcstat_image_from_path(path, 0)) == NULL) {
1291		pmcstat_stats.ps_exec_errors++;
1292		return;
1293	}
1294
1295	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1296		pmcstat_image_determine_type(image);
1297
1298	assert(image->pi_type != PMCSTAT_IMAGE_UNKNOWN);
1299
1300	switch (image->pi_type) {
1301	case PMCSTAT_IMAGE_ELF32:
1302	case PMCSTAT_IMAGE_ELF64:
1303		pmcstat_stats.ps_exec_elf++;
1304		pmcstat_process_elf_exec(pp, image, entryaddr);
1305		break;
1306
1307	case PMCSTAT_IMAGE_AOUT:
1308		pmcstat_stats.ps_exec_aout++;
1309		pmcstat_process_aout_exec(pp, image, entryaddr);
1310		break;
1311
1312	case PMCSTAT_IMAGE_INDETERMINABLE:
1313		pmcstat_stats.ps_exec_indeterminable++;
1314		break;
1315
1316	default:
1317		err(EX_SOFTWARE,
1318		    "ERROR: Unsupported executable type for \"%s\"",
1319		    pmcstat_string_unintern(path));
1320	}
1321}
1322
1323
1324/*
1325 * Find the map entry associated with process 'p' at PC value 'pc'.
1326 */
1327
1328struct pmcstat_pcmap *
1329pmcstat_process_find_map(struct pmcstat_process *p, uintfptr_t pc)
1330{
1331	struct pmcstat_pcmap *ppm;
1332
1333	TAILQ_FOREACH(ppm, &p->pp_map, ppm_next) {
1334		if (pc >= ppm->ppm_lowpc && pc < ppm->ppm_highpc)
1335			return (ppm);
1336		if (pc < ppm->ppm_lowpc)
1337			return (NULL);
1338	}
1339
1340	return (NULL);
1341}
1342
1343/*
1344 * Convert a hwpmc(4) log to profile information.  A system-wide
1345 * callgraph is generated if FLAG_DO_CALLGRAPHS is set.  gmon.out
1346 * files usable by gprof(1) are created if FLAG_DO_GPROF is set.
1347 */
1348static int
1349pmcstat_analyze_log(void)
1350{
1351	uint32_t cpu, cpuflags;
1352	uintfptr_t pc;
1353	pid_t pid;
1354	struct pmcstat_image *image;
1355	struct pmcstat_process *pp, *ppnew;
1356	struct pmcstat_pcmap *ppm, *ppmtmp;
1357	struct pmclog_ev ev;
1358	struct pmcstat_pmcrecord *pmcr;
1359	pmcstat_interned_string image_path;
1360
1361	assert(args.pa_flags & FLAG_DO_ANALYSIS);
1362
1363	if (elf_version(EV_CURRENT) == EV_NONE)
1364		err(EX_UNAVAILABLE, "Elf library initialization failed");
1365
1366	while (pmclog_read(args.pa_logparser, &ev) == 0) {
1367		assert(ev.pl_state == PMCLOG_OK);
1368
1369		switch (ev.pl_type) {
1370		case PMCLOG_TYPE_INITIALIZE:
1371			if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
1372			    PMC_VERSION_MAJOR << 24 && args.pa_verbosity > 0)
1373				warnx(
1374"WARNING: Log version 0x%x does not match compiled version 0x%x.",
1375				    ev.pl_u.pl_i.pl_version, PMC_VERSION_MAJOR);
1376			break;
1377
1378		case PMCLOG_TYPE_MAP_IN:
1379			/*
1380			 * Introduce an address range mapping for a
1381			 * userland process or the kernel (pid == -1).
1382			 *
1383			 * We always allocate a process descriptor so
1384			 * that subsequent samples seen for this
1385			 * address range are mapped to the current
1386			 * object being mapped in.
1387			 */
1388			pid = ev.pl_u.pl_mi.pl_pid;
1389			if (pid == -1)
1390				pp = pmcstat_kernproc;
1391			else
1392				pp = pmcstat_process_lookup(pid,
1393				    PMCSTAT_ALLOCATE);
1394
1395			assert(pp != NULL);
1396
1397			image_path = pmcstat_string_intern(ev.pl_u.pl_mi.
1398			    pl_pathname);
1399			image = pmcstat_image_from_path(image_path, pid == -1);
1400			if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1401				pmcstat_image_determine_type(image);
1402			if (image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE)
1403				pmcstat_image_link(pp, image,
1404				    ev.pl_u.pl_mi.pl_start);
1405			break;
1406
1407		case PMCLOG_TYPE_MAP_OUT:
1408			/*
1409			 * Remove an address map.
1410			 */
1411			pid = ev.pl_u.pl_mo.pl_pid;
1412			if (pid == -1)
1413				pp = pmcstat_kernproc;
1414			else
1415				pp = pmcstat_process_lookup(pid, 0);
1416
1417			if (pp == NULL)	/* unknown process */
1418				break;
1419
1420			pmcstat_image_unmap(pp, ev.pl_u.pl_mo.pl_start,
1421			    ev.pl_u.pl_mo.pl_end);
1422			break;
1423
1424		case PMCLOG_TYPE_PCSAMPLE:
1425			/*
1426			 * Note: the `PCSAMPLE' log entry is not
1427			 * generated by hpwmc(4) after version 2.
1428			 */
1429
1430			/*
1431			 * We bring in the gmon file for the image
1432			 * currently associated with the PMC & pid
1433			 * pair and increment the appropriate entry
1434			 * bin inside this.
1435			 */
1436			pmcstat_stats.ps_samples_total++;
1437			ps_samples_period++;
1438
1439			pc = ev.pl_u.pl_s.pl_pc;
1440			pp = pmcstat_process_lookup(ev.pl_u.pl_s.pl_pid,
1441			    PMCSTAT_ALLOCATE);
1442
1443			/* Get PMC record. */
1444			pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_s.pl_pmcid);
1445			assert(pmcr != NULL);
1446			pmcr->pr_samples++;
1447
1448			/*
1449			 * Call the plugins processing
1450			 * TODO: move pmcstat_process_find_map inside plugins
1451			 */
1452
1453			if (plugins[args.pa_pplugin].pl_process != NULL)
1454				plugins[args.pa_pplugin].pl_process(
1455				    pp, pmcr, 1, &pc,
1456				    pmcstat_process_find_map(pp, pc) != NULL, 0);
1457			plugins[args.pa_plugin].pl_process(
1458			    pp, pmcr, 1, &pc,
1459			    pmcstat_process_find_map(pp, pc) != NULL, 0);
1460			break;
1461
1462		case PMCLOG_TYPE_CALLCHAIN:
1463			pmcstat_stats.ps_samples_total++;
1464			ps_samples_period++;
1465
1466			cpuflags = ev.pl_u.pl_cc.pl_cpuflags;
1467			cpu = PMC_CALLCHAIN_CPUFLAGS_TO_CPU(cpuflags);
1468
1469			/* Filter on the CPU id. */
1470			if (!CPU_ISSET(cpu, &(args.pa_cpumask))) {
1471				pmcstat_stats.ps_samples_skipped++;
1472				break;
1473			}
1474
1475			pp = pmcstat_process_lookup(ev.pl_u.pl_cc.pl_pid,
1476			    PMCSTAT_ALLOCATE);
1477
1478			/* Get PMC record. */
1479			pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_cc.pl_pmcid);
1480			assert(pmcr != NULL);
1481			pmcr->pr_samples++;
1482
1483			/*
1484			 * Call the plugins processing
1485			 */
1486
1487			if (plugins[args.pa_pplugin].pl_process != NULL)
1488				plugins[args.pa_pplugin].pl_process(
1489				    pp, pmcr,
1490				    ev.pl_u.pl_cc.pl_npc,
1491				    ev.pl_u.pl_cc.pl_pc,
1492				    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
1493				    cpu);
1494			plugins[args.pa_plugin].pl_process(
1495			    pp, pmcr,
1496			    ev.pl_u.pl_cc.pl_npc,
1497			    ev.pl_u.pl_cc.pl_pc,
1498			    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
1499			    cpu);
1500			break;
1501
1502		case PMCLOG_TYPE_PMCALLOCATE:
1503			/*
1504			 * Record the association pmc id between this
1505			 * PMC and its name.
1506			 */
1507			pmcstat_pmcid_add(ev.pl_u.pl_a.pl_pmcid,
1508			    pmcstat_string_intern(ev.pl_u.pl_a.pl_evname));
1509			break;
1510
1511		case PMCLOG_TYPE_PMCALLOCATEDYN:
1512			/*
1513			 * Record the association pmc id between this
1514			 * PMC and its name.
1515			 */
1516			pmcstat_pmcid_add(ev.pl_u.pl_ad.pl_pmcid,
1517			    pmcstat_string_intern(ev.pl_u.pl_ad.pl_evname));
1518			break;
1519
1520		case PMCLOG_TYPE_PROCEXEC:
1521
1522			/*
1523			 * Change the executable image associated with
1524			 * a process.
1525			 */
1526			pp = pmcstat_process_lookup(ev.pl_u.pl_x.pl_pid,
1527			    PMCSTAT_ALLOCATE);
1528
1529			/* delete the current process map */
1530			TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
1531				TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
1532				free(ppm);
1533			}
1534
1535			/* associate this process  image */
1536			image_path = pmcstat_string_intern(
1537				ev.pl_u.pl_x.pl_pathname);
1538			assert(image_path != NULL);
1539			pmcstat_process_exec(pp, image_path,
1540			    ev.pl_u.pl_x.pl_entryaddr);
1541			break;
1542
1543		case PMCLOG_TYPE_PROCEXIT:
1544
1545			/*
1546			 * Due to the way the log is generated, the
1547			 * last few samples corresponding to a process
1548			 * may appear in the log after the process
1549			 * exit event is recorded.  Thus we keep the
1550			 * process' descriptor and associated data
1551			 * structures around, but mark the process as
1552			 * having exited.
1553			 */
1554			pp = pmcstat_process_lookup(ev.pl_u.pl_e.pl_pid, 0);
1555			if (pp == NULL)
1556				break;
1557			pp->pp_isactive = 0;	/* mark as a zombie */
1558			break;
1559
1560		case PMCLOG_TYPE_SYSEXIT:
1561			pp = pmcstat_process_lookup(ev.pl_u.pl_se.pl_pid, 0);
1562			if (pp == NULL)
1563				break;
1564			pp->pp_isactive = 0;	/* make a zombie */
1565			break;
1566
1567		case PMCLOG_TYPE_PROCFORK:
1568
1569			/*
1570			 * Allocate a process descriptor for the new
1571			 * (child) process.
1572			 */
1573			ppnew =
1574			    pmcstat_process_lookup(ev.pl_u.pl_f.pl_newpid,
1575				PMCSTAT_ALLOCATE);
1576
1577			/*
1578			 * If we had been tracking the parent, clone
1579			 * its address maps.
1580			 */
1581			pp = pmcstat_process_lookup(ev.pl_u.pl_f.pl_oldpid, 0);
1582			if (pp == NULL)
1583				break;
1584			TAILQ_FOREACH(ppm, &pp->pp_map, ppm_next)
1585			    pmcstat_image_link(ppnew, ppm->ppm_image,
1586				ppm->ppm_lowpc);
1587			break;
1588
1589		default:	/* other types of entries are not relevant */
1590			break;
1591		}
1592	}
1593
1594	if (ev.pl_state == PMCLOG_EOF)
1595		return (PMCSTAT_FINISHED);
1596	else if (ev.pl_state == PMCLOG_REQUIRE_DATA)
1597		return (PMCSTAT_RUNNING);
1598
1599	err(EX_DATAERR,
1600	    "ERROR: event parsing failed (record %jd, offset 0x%jx)",
1601	    (uintmax_t) ev.pl_count + 1, ev.pl_offset);
1602}
1603
1604/*
1605 * Print log entries as text.
1606 */
1607
1608static int
1609pmcstat_print_log(void)
1610{
1611	struct pmclog_ev ev;
1612	uint32_t npc;
1613
1614	while (pmclog_read(args.pa_logparser, &ev) == 0) {
1615		assert(ev.pl_state == PMCLOG_OK);
1616		switch (ev.pl_type) {
1617		case PMCLOG_TYPE_CALLCHAIN:
1618			PMCSTAT_PRINT_ENTRY("callchain",
1619			    "%d 0x%x %d %d %c", ev.pl_u.pl_cc.pl_pid,
1620			    ev.pl_u.pl_cc.pl_pmcid,
1621			    PMC_CALLCHAIN_CPUFLAGS_TO_CPU(ev.pl_u.pl_cc. \
1622				pl_cpuflags), ev.pl_u.pl_cc.pl_npc,
1623			    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(ev.pl_u.pl_cc.\
1624			        pl_cpuflags) ? 'u' : 's');
1625			for (npc = 0; npc < ev.pl_u.pl_cc.pl_npc; npc++)
1626				PMCSTAT_PRINT_ENTRY("...", "%p",
1627				    (void *) ev.pl_u.pl_cc.pl_pc[npc]);
1628			break;
1629		case PMCLOG_TYPE_CLOSELOG:
1630			PMCSTAT_PRINT_ENTRY("closelog",);
1631			break;
1632		case PMCLOG_TYPE_DROPNOTIFY:
1633			PMCSTAT_PRINT_ENTRY("drop",);
1634			break;
1635		case PMCLOG_TYPE_INITIALIZE:
1636			PMCSTAT_PRINT_ENTRY("initlog","0x%x \"%s\"",
1637			    ev.pl_u.pl_i.pl_version,
1638			    pmc_name_of_cputype(ev.pl_u.pl_i.pl_arch));
1639			if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
1640			    PMC_VERSION_MAJOR << 24 && args.pa_verbosity > 0)
1641				warnx(
1642"WARNING: Log version 0x%x != expected version 0x%x.",
1643				    ev.pl_u.pl_i.pl_version, PMC_VERSION);
1644			break;
1645		case PMCLOG_TYPE_MAP_IN:
1646			PMCSTAT_PRINT_ENTRY("map-in","%d %p \"%s\"",
1647			    ev.pl_u.pl_mi.pl_pid,
1648			    (void *) ev.pl_u.pl_mi.pl_start,
1649			    ev.pl_u.pl_mi.pl_pathname);
1650			break;
1651		case PMCLOG_TYPE_MAP_OUT:
1652			PMCSTAT_PRINT_ENTRY("map-out","%d %p %p",
1653			    ev.pl_u.pl_mo.pl_pid,
1654			    (void *) ev.pl_u.pl_mo.pl_start,
1655			    (void *) ev.pl_u.pl_mo.pl_end);
1656			break;
1657		case PMCLOG_TYPE_PCSAMPLE:
1658			PMCSTAT_PRINT_ENTRY("sample","0x%x %d %p %c",
1659			    ev.pl_u.pl_s.pl_pmcid,
1660			    ev.pl_u.pl_s.pl_pid,
1661			    (void *) ev.pl_u.pl_s.pl_pc,
1662			    ev.pl_u.pl_s.pl_usermode ? 'u' : 's');
1663			break;
1664		case PMCLOG_TYPE_PMCALLOCATE:
1665			PMCSTAT_PRINT_ENTRY("allocate","0x%x \"%s\" 0x%x",
1666			    ev.pl_u.pl_a.pl_pmcid,
1667			    ev.pl_u.pl_a.pl_evname,
1668			    ev.pl_u.pl_a.pl_flags);
1669			break;
1670		case PMCLOG_TYPE_PMCALLOCATEDYN:
1671			PMCSTAT_PRINT_ENTRY("allocatedyn","0x%x \"%s\" 0x%x",
1672			    ev.pl_u.pl_ad.pl_pmcid,
1673			    ev.pl_u.pl_ad.pl_evname,
1674			    ev.pl_u.pl_ad.pl_flags);
1675			break;
1676		case PMCLOG_TYPE_PMCATTACH:
1677			PMCSTAT_PRINT_ENTRY("attach","0x%x %d \"%s\"",
1678			    ev.pl_u.pl_t.pl_pmcid,
1679			    ev.pl_u.pl_t.pl_pid,
1680			    ev.pl_u.pl_t.pl_pathname);
1681			break;
1682		case PMCLOG_TYPE_PMCDETACH:
1683			PMCSTAT_PRINT_ENTRY("detach","0x%x %d",
1684			    ev.pl_u.pl_d.pl_pmcid,
1685			    ev.pl_u.pl_d.pl_pid);
1686			break;
1687		case PMCLOG_TYPE_PROCCSW:
1688			PMCSTAT_PRINT_ENTRY("cswval","0x%x %d %jd",
1689			    ev.pl_u.pl_c.pl_pmcid,
1690			    ev.pl_u.pl_c.pl_pid,
1691			    ev.pl_u.pl_c.pl_value);
1692			break;
1693		case PMCLOG_TYPE_PROCEXEC:
1694			PMCSTAT_PRINT_ENTRY("exec","0x%x %d %p \"%s\"",
1695			    ev.pl_u.pl_x.pl_pmcid,
1696			    ev.pl_u.pl_x.pl_pid,
1697			    (void *) ev.pl_u.pl_x.pl_entryaddr,
1698			    ev.pl_u.pl_x.pl_pathname);
1699			break;
1700		case PMCLOG_TYPE_PROCEXIT:
1701			PMCSTAT_PRINT_ENTRY("exitval","0x%x %d %jd",
1702			    ev.pl_u.pl_e.pl_pmcid,
1703			    ev.pl_u.pl_e.pl_pid,
1704			    ev.pl_u.pl_e.pl_value);
1705			break;
1706		case PMCLOG_TYPE_PROCFORK:
1707			PMCSTAT_PRINT_ENTRY("fork","%d %d",
1708			    ev.pl_u.pl_f.pl_oldpid,
1709			    ev.pl_u.pl_f.pl_newpid);
1710			break;
1711		case PMCLOG_TYPE_USERDATA:
1712			PMCSTAT_PRINT_ENTRY("userdata","0x%x",
1713			    ev.pl_u.pl_u.pl_userdata);
1714			break;
1715		case PMCLOG_TYPE_SYSEXIT:
1716			PMCSTAT_PRINT_ENTRY("exit","%d",
1717			    ev.pl_u.pl_se.pl_pid);
1718			break;
1719		default:
1720			fprintf(args.pa_printfile, "unknown event (type %d).\n",
1721			    ev.pl_type);
1722		}
1723	}
1724
1725	if (ev.pl_state == PMCLOG_EOF)
1726		return (PMCSTAT_FINISHED);
1727	else if (ev.pl_state ==  PMCLOG_REQUIRE_DATA)
1728		return (PMCSTAT_RUNNING);
1729
1730	errx(EX_DATAERR,
1731	    "ERROR: event parsing failed (record %jd, offset 0x%jx).",
1732	    (uintmax_t) ev.pl_count + 1, ev.pl_offset);
1733	/*NOTREACHED*/
1734}
1735
1736/*
1737 * Public Interfaces.
1738 */
1739
1740/*
1741 * Close a logfile, after first flushing all in-module queued data.
1742 */
1743
1744int
1745pmcstat_close_log(void)
1746{
1747	/* If a local logfile is configured ask the kernel to stop
1748	 * and flush data. Kernel will close the file when data is flushed
1749	 * so keep the status to EXITING.
1750	 */
1751	if (args.pa_logfd != -1) {
1752		if (pmc_close_logfile() < 0)
1753			err(EX_OSERR, "ERROR: logging failed");
1754	}
1755
1756	return (args.pa_flags & FLAG_HAS_PIPE ? PMCSTAT_EXITING :
1757	    PMCSTAT_FINISHED);
1758}
1759
1760
1761
1762/*
1763 * Open a log file, for reading or writing.
1764 *
1765 * The function returns the fd of a successfully opened log or -1 in
1766 * case of failure.
1767 */
1768
1769int
1770pmcstat_open_log(const char *path, int mode)
1771{
1772	int error, fd, cfd;
1773	size_t hlen;
1774	const char *p, *errstr;
1775	struct addrinfo hints, *res, *res0;
1776	char hostname[MAXHOSTNAMELEN];
1777
1778	errstr = NULL;
1779	fd = -1;
1780
1781	/*
1782	 * If 'path' is "-" then open one of stdin or stdout depending
1783	 * on the value of 'mode'.
1784	 *
1785	 * If 'path' contains a ':' and does not start with a '/' or '.',
1786	 * and is being opened for writing, treat it as a "host:port"
1787	 * specification and open a network socket.
1788	 *
1789	 * Otherwise, treat 'path' as a file name and open that.
1790	 */
1791	if (path[0] == '-' && path[1] == '\0')
1792		fd = (mode == PMCSTAT_OPEN_FOR_READ) ? 0 : 1;
1793	else if (path[0] != '/' &&
1794	    path[0] != '.' && strchr(path, ':') != NULL) {
1795
1796		p = strrchr(path, ':');
1797		hlen = p - path;
1798		if (p == path || hlen >= sizeof(hostname)) {
1799			errstr = strerror(EINVAL);
1800			goto done;
1801		}
1802
1803		assert(hlen < sizeof(hostname));
1804		(void) strncpy(hostname, path, hlen);
1805		hostname[hlen] = '\0';
1806
1807		(void) memset(&hints, 0, sizeof(hints));
1808		hints.ai_family = AF_UNSPEC;
1809		hints.ai_socktype = SOCK_STREAM;
1810		if ((error = getaddrinfo(hostname, p+1, &hints, &res0)) != 0) {
1811			errstr = gai_strerror(error);
1812			goto done;
1813		}
1814
1815		fd = -1;
1816		for (res = res0; res; res = res->ai_next) {
1817			if ((fd = socket(res->ai_family, res->ai_socktype,
1818			    res->ai_protocol)) < 0) {
1819				errstr = strerror(errno);
1820				continue;
1821			}
1822			if (mode == PMCSTAT_OPEN_FOR_READ) {
1823				if (bind(fd, res->ai_addr, res->ai_addrlen) < 0) {
1824					errstr = strerror(errno);
1825					(void) close(fd);
1826					fd = -1;
1827					continue;
1828				}
1829				listen(fd, 1);
1830				cfd = accept(fd, NULL, NULL);
1831				(void) close(fd);
1832				if (cfd < 0) {
1833					errstr = strerror(errno);
1834					fd = -1;
1835					break;
1836				}
1837				fd = cfd;
1838			} else {
1839				if (connect(fd, res->ai_addr, res->ai_addrlen) < 0) {
1840					errstr = strerror(errno);
1841					(void) close(fd);
1842					fd = -1;
1843					continue;
1844				}
1845			}
1846			errstr = NULL;
1847			break;
1848		}
1849		freeaddrinfo(res0);
1850
1851	} else if ((fd = open(path, mode == PMCSTAT_OPEN_FOR_READ ?
1852		    O_RDONLY : (O_WRONLY|O_CREAT|O_TRUNC),
1853		    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
1854			errstr = strerror(errno);
1855
1856  done:
1857	if (errstr)
1858		errx(EX_OSERR, "ERROR: Cannot open \"%s\" for %s: %s.", path,
1859		    (mode == PMCSTAT_OPEN_FOR_READ ? "reading" : "writing"),
1860		    errstr);
1861
1862	return (fd);
1863}
1864
1865/*
1866 * Process a log file in offline analysis mode.
1867 */
1868
1869int
1870pmcstat_process_log(void)
1871{
1872
1873	/*
1874	 * If analysis has not been asked for, just print the log to
1875	 * the current output file.
1876	 */
1877	if (args.pa_flags & FLAG_DO_PRINT)
1878		return (pmcstat_print_log());
1879	else
1880		return (pmcstat_analyze_log());
1881}
1882
1883/*
1884 * Refresh top display.
1885 */
1886
1887static void
1888pmcstat_refresh_top(void)
1889{
1890	int v_attrs;
1891	float v;
1892	char pmcname[40];
1893	struct pmcstat_pmcrecord *pmcpr;
1894
1895	/* If in pause mode do not refresh display. */
1896	if (pmcstat_pause)
1897		return;
1898
1899	/* Wait until PMC pop in the log. */
1900	pmcpr = pmcstat_pmcindex_to_pmcr(pmcstat_pmcinfilter);
1901	if (pmcpr == NULL)
1902		return;
1903
1904	/* Format PMC name. */
1905	if (pmcstat_mergepmc)
1906		snprintf(pmcname, sizeof(pmcname), "[%s]",
1907		    pmcstat_string_unintern(pmcpr->pr_pmcname));
1908	else
1909		snprintf(pmcname, sizeof(pmcname), "%s.%d",
1910		    pmcstat_string_unintern(pmcpr->pr_pmcname),
1911		    pmcstat_pmcinfilter);
1912
1913	/* Format samples count. */
1914	if (ps_samples_period > 0)
1915		v = (pmcpr->pr_samples * 100.0) / ps_samples_period;
1916	else
1917		v = 0.;
1918	v_attrs = PMCSTAT_ATTRPERCENT(v);
1919
1920	PMCSTAT_PRINTBEGIN();
1921	PMCSTAT_PRINTW("PMC: %s Samples: %u ",
1922	    pmcname,
1923	    pmcpr->pr_samples);
1924	PMCSTAT_ATTRON(v_attrs);
1925	PMCSTAT_PRINTW("(%.1f%%) ", v);
1926	PMCSTAT_ATTROFF(v_attrs);
1927	PMCSTAT_PRINTW(", %u unresolved\n\n",
1928	    pmcpr->pr_dubious_frames);
1929	if (plugins[args.pa_plugin].pl_topdisplay != NULL)
1930		plugins[args.pa_plugin].pl_topdisplay();
1931	PMCSTAT_PRINTEND();
1932}
1933
1934/*
1935 * Find the next pmc index to display.
1936 */
1937
1938static void
1939pmcstat_changefilter(void)
1940{
1941	int pmcin;
1942	struct pmcstat_pmcrecord *pmcr;
1943
1944	/*
1945	 * Find the next merge target.
1946	 */
1947	if (pmcstat_mergepmc) {
1948		pmcin = pmcstat_pmcinfilter;
1949
1950		do {
1951			pmcr = pmcstat_pmcindex_to_pmcr(pmcstat_pmcinfilter);
1952			if (pmcr == NULL || pmcr == pmcr->pr_merge)
1953				break;
1954
1955			pmcstat_pmcinfilter++;
1956			if (pmcstat_pmcinfilter >= pmcstat_npmcs)
1957				pmcstat_pmcinfilter = 0;
1958
1959		} while (pmcstat_pmcinfilter != pmcin);
1960	}
1961}
1962
1963/*
1964 * Top mode keypress.
1965 */
1966
1967int
1968pmcstat_keypress_log(void)
1969{
1970	int c, ret = 0;
1971	WINDOW *w;
1972
1973	w = newwin(1, 0, 1, 0);
1974	c = wgetch(w);
1975	wprintw(w, "Key: %c => ", c);
1976	switch (c) {
1977	case 'c':
1978		wprintw(w, "enter mode 'd' or 'a' => ");
1979		c = wgetch(w);
1980		if (c == 'd') {
1981			args.pa_topmode = PMCSTAT_TOP_DELTA;
1982			wprintw(w, "switching to delta mode");
1983		} else {
1984			args.pa_topmode = PMCSTAT_TOP_ACCUM;
1985			wprintw(w, "switching to accumulation mode");
1986		}
1987		break;
1988	case 'm':
1989		pmcstat_mergepmc = !pmcstat_mergepmc;
1990		/*
1991		 * Changing merge state require data reset.
1992		 */
1993		if (plugins[args.pa_plugin].pl_shutdown != NULL)
1994			plugins[args.pa_plugin].pl_shutdown(NULL);
1995		pmcstat_stats_reset(0);
1996		if (plugins[args.pa_plugin].pl_init != NULL)
1997			plugins[args.pa_plugin].pl_init();
1998
1999		/* Update filter to be on a merge target. */
2000		pmcstat_changefilter();
2001		wprintw(w, "merge PMC %s", pmcstat_mergepmc ? "on" : "off");
2002		break;
2003	case 'n':
2004		/* Close current plugin. */
2005		if (plugins[args.pa_plugin].pl_shutdown != NULL)
2006			plugins[args.pa_plugin].pl_shutdown(NULL);
2007
2008		/* Find next top display available. */
2009		do {
2010			args.pa_plugin++;
2011			if (plugins[args.pa_plugin].pl_name == NULL)
2012				args.pa_plugin = 0;
2013		} while (plugins[args.pa_plugin].pl_topdisplay == NULL);
2014
2015		/* Open new plugin. */
2016		pmcstat_stats_reset(0);
2017		if (plugins[args.pa_plugin].pl_init != NULL)
2018			plugins[args.pa_plugin].pl_init();
2019		wprintw(w, "switching to plugin %s",
2020		    plugins[args.pa_plugin].pl_name);
2021		break;
2022	case 'p':
2023		pmcstat_pmcinfilter++;
2024		if (pmcstat_pmcinfilter >= pmcstat_npmcs)
2025			pmcstat_pmcinfilter = 0;
2026		pmcstat_changefilter();
2027		wprintw(w, "switching to PMC %s.%d",
2028		    pmcstat_pmcindex_to_name(pmcstat_pmcinfilter),
2029		    pmcstat_pmcinfilter);
2030		break;
2031	case ' ':
2032		pmcstat_pause = !pmcstat_pause;
2033		if (pmcstat_pause)
2034			wprintw(w, "pause => press space again to continue");
2035		break;
2036	case 'q':
2037		wprintw(w, "exiting...");
2038		ret = 1;
2039		break;
2040	default:
2041		if (plugins[args.pa_plugin].pl_topkeypress != NULL)
2042			if (plugins[args.pa_plugin].pl_topkeypress(c, w))
2043				ret = 1;
2044	}
2045
2046	wrefresh(w);
2047	delwin(w);
2048	return ret;
2049}
2050
2051
2052/*
2053 * Top mode display.
2054 */
2055
2056void
2057pmcstat_display_log(void)
2058{
2059
2060	pmcstat_refresh_top();
2061
2062	/* Reset everythings if delta mode. */
2063	if (args.pa_topmode == PMCSTAT_TOP_DELTA) {
2064		if (plugins[args.pa_plugin].pl_shutdown != NULL)
2065			plugins[args.pa_plugin].pl_shutdown(NULL);
2066		pmcstat_stats_reset(0);
2067		if (plugins[args.pa_plugin].pl_init != NULL)
2068			plugins[args.pa_plugin].pl_init();
2069	}
2070
2071}
2072
2073/*
2074 * Configure a plugins.
2075 */
2076
2077void
2078pmcstat_pluginconfigure_log(char *opt)
2079{
2080
2081	if (strncmp(opt, "threshold=", 10) == 0) {
2082		pmcstat_threshold = atof(opt+10);
2083	} else {
2084		if (plugins[args.pa_plugin].pl_configure != NULL) {
2085			if (!plugins[args.pa_plugin].pl_configure(opt))
2086				err(EX_USAGE,
2087				    "ERROR: unknown option <%s>.", opt);
2088		}
2089	}
2090}
2091
2092/*
2093 * Initialize module.
2094 */
2095
2096void
2097pmcstat_initialize_logging(void)
2098{
2099	int i;
2100
2101	/* use a convenient format for 'ldd' output */
2102	if (setenv("LD_TRACE_LOADED_OBJECTS_FMT1","%o \"%p\" %x\n",1) != 0)
2103		err(EX_OSERR, "ERROR: Cannot setenv");
2104
2105	/* Initialize hash tables */
2106	pmcstat_string_initialize();
2107	for (i = 0; i < PMCSTAT_NHASH; i++) {
2108		LIST_INIT(&pmcstat_image_hash[i]);
2109		LIST_INIT(&pmcstat_process_hash[i]);
2110	}
2111
2112	/*
2113	 * Create a fake 'process' entry for the kernel with pid -1.
2114	 * hwpmc(4) will subsequently inform us about where the kernel
2115	 * and any loaded kernel modules are mapped.
2116	 */
2117	if ((pmcstat_kernproc = pmcstat_process_lookup((pid_t) -1,
2118		 PMCSTAT_ALLOCATE)) == NULL)
2119		err(EX_OSERR, "ERROR: Cannot initialize logging");
2120
2121	/* PMC count. */
2122	pmcstat_npmcs = 0;
2123
2124	/* Merge PMC with same name. */
2125	pmcstat_mergepmc = args.pa_mergepmc;
2126
2127	/*
2128	 * Initialize plugins
2129	 */
2130
2131	if (plugins[args.pa_pplugin].pl_init != NULL)
2132		plugins[args.pa_pplugin].pl_init();
2133	if (plugins[args.pa_plugin].pl_init != NULL)
2134		plugins[args.pa_plugin].pl_init();
2135}
2136
2137/*
2138 * Shutdown module.
2139 */
2140
2141void
2142pmcstat_shutdown_logging(void)
2143{
2144	int i;
2145	FILE *mf;
2146	struct pmcstat_image *pi, *pitmp;
2147	struct pmcstat_process *pp, *pptmp;
2148	struct pmcstat_pcmap *ppm, *ppmtmp;
2149
2150	/* determine where to send the map file */
2151	mf = NULL;
2152	if (args.pa_mapfilename != NULL)
2153		mf = (strcmp(args.pa_mapfilename, "-") == 0) ?
2154		    args.pa_printfile : fopen(args.pa_mapfilename, "w");
2155
2156	if (mf == NULL && args.pa_flags & FLAG_DO_GPROF &&
2157	    args.pa_verbosity >= 2)
2158		mf = args.pa_printfile;
2159
2160	if (mf)
2161		(void) fprintf(mf, "MAP:\n");
2162
2163	/*
2164	 * Shutdown the plugins
2165	 */
2166
2167	if (plugins[args.pa_plugin].pl_shutdown != NULL)
2168		plugins[args.pa_plugin].pl_shutdown(mf);
2169	if (plugins[args.pa_pplugin].pl_shutdown != NULL)
2170		plugins[args.pa_pplugin].pl_shutdown(mf);
2171
2172	for (i = 0; i < PMCSTAT_NHASH; i++) {
2173		LIST_FOREACH_SAFE(pi, &pmcstat_image_hash[i], pi_next,
2174		    pitmp) {
2175			if (plugins[args.pa_plugin].pl_shutdownimage != NULL)
2176				plugins[args.pa_plugin].pl_shutdownimage(pi);
2177			if (plugins[args.pa_pplugin].pl_shutdownimage != NULL)
2178				plugins[args.pa_pplugin].pl_shutdownimage(pi);
2179
2180			free(pi->pi_symbols);
2181			if (pi->pi_addr2line != NULL)
2182				pclose(pi->pi_addr2line);
2183			LIST_REMOVE(pi, pi_next);
2184			free(pi);
2185		}
2186
2187		LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[i], pp_next,
2188		    pptmp) {
2189			TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
2190				TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
2191				free(ppm);
2192			}
2193			LIST_REMOVE(pp, pp_next);
2194			free(pp);
2195		}
2196	}
2197
2198	pmcstat_string_shutdown();
2199
2200	/*
2201	 * Print errors unless -q was specified.  Print all statistics
2202	 * if verbosity > 1.
2203	 */
2204#define	PRINT(N,V) do {							\
2205		if (pmcstat_stats.ps_##V || args.pa_verbosity >= 2)	\
2206			(void) fprintf(args.pa_printfile, " %-40s %d\n",\
2207			    N, pmcstat_stats.ps_##V);			\
2208	} while (0)
2209
2210	if (args.pa_verbosity >= 1 && (args.pa_flags & FLAG_DO_ANALYSIS)) {
2211		(void) fprintf(args.pa_printfile, "CONVERSION STATISTICS:\n");
2212		PRINT("#exec/a.out", exec_aout);
2213		PRINT("#exec/elf", exec_elf);
2214		PRINT("#exec/unknown", exec_indeterminable);
2215		PRINT("#exec handling errors", exec_errors);
2216		PRINT("#samples/total", samples_total);
2217		PRINT("#samples/unclaimed", samples_unknown_offset);
2218		PRINT("#samples/unknown-object", samples_indeterminable);
2219		PRINT("#samples/unknown-function", samples_unknown_function);
2220		PRINT("#callchain/dubious-frames", callchain_dubious_frames);
2221	}
2222
2223	if (mf)
2224		(void) fclose(mf);
2225}
2226