pmcstat_log.c revision 266590
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 266590 2014-05-23 17:46:00Z emaste $");
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_offset & (-ph.p_align)) == 0)
720					image->pi_vaddr = ph.p_vaddr & (-ph.p_align);
721				break;
722			}
723		}
724	}
725
726	/*
727	 * Get the min and max VA associated with this ELF object.
728	 */
729	if (elf_getshnum(e, &nsh) == 0) {
730		warnx(
731"WARNING: Could not determine the number of sections for \"%s\": %s.",
732		    buffer, elf_errmsg(-1));
733		goto done;
734	}
735
736	for (i = 0; i < nsh; i++) {
737		if ((scn = elf_getscn(e, i)) == NULL ||
738		    gelf_getshdr(scn, &sh) != &sh) {
739			warnx(
740"WARNING: Could not retrieve section header #%ju in \"%s\": %s.",
741			    (uintmax_t) i, buffer, elf_errmsg(-1));
742			goto done;
743		}
744		if (sh.sh_flags & SHF_EXECINSTR) {
745			minva = min(minva, sh.sh_addr);
746			maxva = max(maxva, sh.sh_addr + sh.sh_size);
747		}
748		if (sh.sh_type == SHT_SYMTAB || sh.sh_type == SHT_DYNSYM)
749			pmcstat_image_add_symbols(image, e, scn, &sh);
750	}
751
752	image->pi_start = minva;
753	image->pi_end   = maxva;
754	image->pi_type  = image_type;
755	image->pi_fullpath = pmcstat_string_intern(buffer);
756
757	/* Build display name
758	 */
759	endp = buffer;
760	for (p = buffer; *p; p++)
761		if (*p == '/')
762			endp = p+1;
763	image->pi_name = pmcstat_string_intern(endp);
764
765 done:
766	(void) elf_end(e);
767	if (fd >= 0)
768		(void) close(fd);
769	return;
770}
771
772/*
773 * Given an image descriptor, determine whether it is an ELF, or AOUT.
774 * If no handler claims the image, set its type to 'INDETERMINABLE'.
775 */
776
777void
778pmcstat_image_determine_type(struct pmcstat_image *image)
779{
780	assert(image->pi_type == PMCSTAT_IMAGE_UNKNOWN);
781
782	/* Try each kind of handler in turn */
783	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
784		pmcstat_image_get_elf_params(image);
785	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
786		pmcstat_image_get_aout_params(image);
787
788	/*
789	 * Otherwise, remember that we tried to determine
790	 * the object's type and had failed.
791	 */
792	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
793		image->pi_type = PMCSTAT_IMAGE_INDETERMINABLE;
794}
795
796/*
797 * Locate an image descriptor given an interned path, adding a fresh
798 * descriptor to the cache if necessary.  This function also finds a
799 * suitable name for this image's sample file.
800 *
801 * We defer filling in the file format specific parts of the image
802 * structure till the time we actually see a sample that would fall
803 * into this image.
804 */
805
806static struct pmcstat_image *
807pmcstat_image_from_path(pmcstat_interned_string internedpath,
808    int iskernelmodule)
809{
810	int hash;
811	struct pmcstat_image *pi;
812
813	hash = pmcstat_string_lookup_hash(internedpath);
814
815	/* First, look for an existing entry. */
816	LIST_FOREACH(pi, &pmcstat_image_hash[hash], pi_next)
817	    if (pi->pi_execpath == internedpath &&
818		  pi->pi_iskernelmodule == iskernelmodule)
819		    return (pi);
820
821	/*
822	 * Allocate a new entry and place it at the head of the hash
823	 * and LRU lists.
824	 */
825	pi = malloc(sizeof(*pi));
826	if (pi == NULL)
827		return (NULL);
828
829	pi->pi_type = PMCSTAT_IMAGE_UNKNOWN;
830	pi->pi_execpath = internedpath;
831	pi->pi_start = ~0;
832	pi->pi_end = 0;
833	pi->pi_entry = 0;
834	pi->pi_vaddr = 0;
835	pi->pi_isdynamic = 0;
836	pi->pi_iskernelmodule = iskernelmodule;
837	pi->pi_dynlinkerpath = NULL;
838	pi->pi_symbols = NULL;
839	pi->pi_symcount = 0;
840	pi->pi_addr2line = NULL;
841
842	if (plugins[args.pa_pplugin].pl_initimage != NULL)
843		plugins[args.pa_pplugin].pl_initimage(pi);
844	if (plugins[args.pa_plugin].pl_initimage != NULL)
845		plugins[args.pa_plugin].pl_initimage(pi);
846
847	LIST_INSERT_HEAD(&pmcstat_image_hash[hash], pi, pi_next);
848
849	return (pi);
850}
851
852/*
853 * Record the fact that PC values from 'start' to 'end' come from
854 * image 'image'.
855 */
856
857static void
858pmcstat_image_link(struct pmcstat_process *pp, struct pmcstat_image *image,
859    uintfptr_t start)
860{
861	struct pmcstat_pcmap *pcm, *pcmnew;
862	uintfptr_t offset;
863
864	assert(image->pi_type != PMCSTAT_IMAGE_UNKNOWN &&
865	    image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE);
866
867	if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
868		err(EX_OSERR, "ERROR: Cannot create a map entry");
869
870	/*
871	 * Adjust the map entry to only cover the text portion
872	 * of the object.
873	 */
874
875	offset = start - image->pi_vaddr;
876	pcmnew->ppm_lowpc  = image->pi_start + offset;
877	pcmnew->ppm_highpc = image->pi_end + offset;
878	pcmnew->ppm_image  = image;
879
880	assert(pcmnew->ppm_lowpc < pcmnew->ppm_highpc);
881
882	/* Overlapped mmap()'s are assumed to never occur. */
883	TAILQ_FOREACH(pcm, &pp->pp_map, ppm_next)
884	    if (pcm->ppm_lowpc >= pcmnew->ppm_highpc)
885		    break;
886
887	if (pcm == NULL)
888		TAILQ_INSERT_TAIL(&pp->pp_map, pcmnew, ppm_next);
889	else
890		TAILQ_INSERT_BEFORE(pcm, pcmnew, ppm_next);
891}
892
893/*
894 * Unmap images in the range [start..end) associated with process
895 * 'pp'.
896 */
897
898static void
899pmcstat_image_unmap(struct pmcstat_process *pp, uintfptr_t start,
900    uintfptr_t end)
901{
902	struct pmcstat_pcmap *pcm, *pcmtmp, *pcmnew;
903
904	assert(pp != NULL);
905	assert(start < end);
906
907	/*
908	 * Cases:
909	 * - we could have the range completely in the middle of an
910	 *   existing pcmap; in this case we have to split the pcmap
911	 *   structure into two (i.e., generate a 'hole').
912	 * - we could have the range covering multiple pcmaps; these
913	 *   will have to be removed.
914	 * - we could have either 'start' or 'end' falling in the
915	 *   middle of a pcmap; in this case shorten the entry.
916	 */
917	TAILQ_FOREACH_SAFE(pcm, &pp->pp_map, ppm_next, pcmtmp) {
918		assert(pcm->ppm_lowpc < pcm->ppm_highpc);
919		if (pcm->ppm_highpc <= start)
920			continue;
921		if (pcm->ppm_lowpc >= end)
922			return;
923		if (pcm->ppm_lowpc >= start && pcm->ppm_highpc <= end) {
924			/*
925			 * The current pcmap is completely inside the
926			 * unmapped range: remove it entirely.
927			 */
928			TAILQ_REMOVE(&pp->pp_map, pcm, ppm_next);
929			free(pcm);
930		} else if (pcm->ppm_lowpc < start && pcm->ppm_highpc > end) {
931			/*
932			 * Split this pcmap into two; curtail the
933			 * current map to end at [start-1], and start
934			 * the new one at [end].
935			 */
936			if ((pcmnew = malloc(sizeof(*pcmnew))) == NULL)
937				err(EX_OSERR,
938				    "ERROR: Cannot split a map entry");
939
940			pcmnew->ppm_image = pcm->ppm_image;
941
942			pcmnew->ppm_lowpc = end;
943			pcmnew->ppm_highpc = pcm->ppm_highpc;
944
945			pcm->ppm_highpc = start;
946
947			TAILQ_INSERT_AFTER(&pp->pp_map, pcm, pcmnew, ppm_next);
948
949			return;
950		} else if (pcm->ppm_lowpc < start && pcm->ppm_highpc <= end)
951			pcm->ppm_highpc = start;
952		else if (pcm->ppm_lowpc >= start && pcm->ppm_highpc > end)
953			pcm->ppm_lowpc = end;
954		else
955			assert(0);
956	}
957}
958
959/*
960 * Resolve file name and line number for the given address.
961 */
962int
963pmcstat_image_addr2line(struct pmcstat_image *image, uintfptr_t addr,
964    char *sourcefile, size_t sourcefile_len, unsigned *sourceline,
965    char *funcname, size_t funcname_len)
966{
967	static int addr2line_warn = 0;
968	unsigned l;
969
970	char *sep, cmdline[PATH_MAX], imagepath[PATH_MAX];
971	int fd;
972
973	if (image->pi_addr2line == NULL) {
974		snprintf(imagepath, sizeof(imagepath), "%s%s.symbols",
975		    args.pa_fsroot,
976		    pmcstat_string_unintern(image->pi_fullpath));
977		fd = open(imagepath, O_RDONLY);
978		if (fd < 0) {
979			snprintf(imagepath, sizeof(imagepath), "%s%s",
980			    args.pa_fsroot,
981			    pmcstat_string_unintern(image->pi_fullpath));
982		} else
983			close(fd);
984		/*
985		 * New addr2line support recursive inline function with -i
986		 * but the format does not add a marker when no more entries
987		 * are available.
988		 */
989		snprintf(cmdline, sizeof(cmdline), "addr2line -Cfe \"%s\"",
990		    imagepath);
991		image->pi_addr2line = popen(cmdline, "r+");
992		if (image->pi_addr2line == NULL) {
993			if (!addr2line_warn) {
994				addr2line_warn = 1;
995				warnx(
996"WARNING: addr2line is needed for source code information."
997				    );
998			}
999			return (0);
1000		}
1001	}
1002
1003	if (feof(image->pi_addr2line) || ferror(image->pi_addr2line)) {
1004		warnx("WARNING: addr2line pipe error");
1005		pclose(image->pi_addr2line);
1006		image->pi_addr2line = NULL;
1007		return (0);
1008	}
1009
1010	fprintf(image->pi_addr2line, "%p\n", (void *)addr);
1011
1012	if (fgets(funcname, funcname_len, image->pi_addr2line) == NULL) {
1013		warnx("WARNING: addr2line function name read error");
1014		return (0);
1015	}
1016	sep = strchr(funcname, '\n');
1017	if (sep != NULL)
1018		*sep = '\0';
1019
1020	if (fgets(sourcefile, sourcefile_len, image->pi_addr2line) == NULL) {
1021		warnx("WARNING: addr2line source file read error");
1022		return (0);
1023	}
1024	sep = strchr(sourcefile, ':');
1025	if (sep == NULL) {
1026		warnx("WARNING: addr2line source line separator missing");
1027		return (0);
1028	}
1029	*sep = '\0';
1030	l = atoi(sep+1);
1031	if (l == 0)
1032		return (0);
1033	*sourceline = l;
1034	return (1);
1035}
1036
1037/*
1038 * Add a {pmcid,name} mapping.
1039 */
1040
1041static void
1042pmcstat_pmcid_add(pmc_id_t pmcid, pmcstat_interned_string ps)
1043{
1044	struct pmcstat_pmcrecord *pr, *prm;
1045
1046	/* Replace an existing name for the PMC. */
1047	prm = NULL;
1048	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1049		if (pr->pr_pmcid == pmcid) {
1050			pr->pr_pmcname = ps;
1051			return;
1052		} else if (pr->pr_pmcname == ps)
1053			prm = pr;
1054
1055	/*
1056	 * Otherwise, allocate a new descriptor and call the
1057	 * plugins hook.
1058	 */
1059	if ((pr = malloc(sizeof(*pr))) == NULL)
1060		err(EX_OSERR, "ERROR: Cannot allocate pmc record");
1061
1062	pr->pr_pmcid = pmcid;
1063	pr->pr_pmcname = ps;
1064	pr->pr_pmcin = pmcstat_npmcs++;
1065	pr->pr_samples = 0;
1066	pr->pr_dubious_frames = 0;
1067	pr->pr_merge = prm == NULL ? pr : prm;
1068
1069	LIST_INSERT_HEAD(&pmcstat_pmcs, pr, pr_next);
1070
1071	if (plugins[args.pa_pplugin].pl_newpmc != NULL)
1072		plugins[args.pa_pplugin].pl_newpmc(ps, pr);
1073	if (plugins[args.pa_plugin].pl_newpmc != NULL)
1074		plugins[args.pa_plugin].pl_newpmc(ps, pr);
1075}
1076
1077/*
1078 * Given a pmcid in use, find its human-readable name.
1079 */
1080
1081const char *
1082pmcstat_pmcid_to_name(pmc_id_t pmcid)
1083{
1084	struct pmcstat_pmcrecord *pr;
1085
1086	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1087	    if (pr->pr_pmcid == pmcid)
1088		    return (pmcstat_string_unintern(pr->pr_pmcname));
1089
1090	return NULL;
1091}
1092
1093/*
1094 * Convert PMC index to name.
1095 */
1096
1097const char *
1098pmcstat_pmcindex_to_name(int pmcin)
1099{
1100	struct pmcstat_pmcrecord *pr;
1101
1102	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1103		if (pr->pr_pmcin == pmcin)
1104			return pmcstat_string_unintern(pr->pr_pmcname);
1105
1106	return NULL;
1107}
1108
1109/*
1110 * Return PMC record with given index.
1111 */
1112
1113struct pmcstat_pmcrecord *
1114pmcstat_pmcindex_to_pmcr(int pmcin)
1115{
1116	struct pmcstat_pmcrecord *pr;
1117
1118	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next)
1119		if (pr->pr_pmcin == pmcin)
1120			return pr;
1121
1122	return NULL;
1123}
1124
1125/*
1126 * Get PMC record by id, apply merge policy.
1127 */
1128
1129static struct pmcstat_pmcrecord *
1130pmcstat_lookup_pmcid(pmc_id_t pmcid)
1131{
1132	struct pmcstat_pmcrecord *pr;
1133
1134	LIST_FOREACH(pr, &pmcstat_pmcs, pr_next) {
1135		if (pr->pr_pmcid == pmcid) {
1136			if (pmcstat_mergepmc)
1137				return pr->pr_merge;
1138			return pr;
1139		}
1140	}
1141
1142	return NULL;
1143}
1144
1145/*
1146 * Associate an AOUT image with a process.
1147 */
1148
1149static void
1150pmcstat_process_aout_exec(struct pmcstat_process *pp,
1151    struct pmcstat_image *image, uintfptr_t entryaddr)
1152{
1153	(void) pp;
1154	(void) image;
1155	(void) entryaddr;
1156	/* TODO Implement a.out handling */
1157}
1158
1159/*
1160 * Associate an ELF image with a process.
1161 */
1162
1163static void
1164pmcstat_process_elf_exec(struct pmcstat_process *pp,
1165    struct pmcstat_image *image, uintfptr_t entryaddr)
1166{
1167	uintmax_t libstart;
1168	struct pmcstat_image *rtldimage;
1169
1170	assert(image->pi_type == PMCSTAT_IMAGE_ELF32 ||
1171	    image->pi_type == PMCSTAT_IMAGE_ELF64);
1172
1173	/* Create a map entry for the base executable. */
1174	pmcstat_image_link(pp, image, image->pi_vaddr);
1175
1176	/*
1177	 * For dynamically linked executables we need to determine
1178	 * where the dynamic linker was mapped to for this process,
1179	 * Subsequent executable objects that are mapped in by the
1180	 * dynamic linker will be tracked by log events of type
1181	 * PMCLOG_TYPE_MAP_IN.
1182	 */
1183
1184	if (image->pi_isdynamic) {
1185
1186		/*
1187		 * The runtime loader gets loaded just after the maximum
1188		 * possible heap address.  Like so:
1189		 *
1190		 * [  TEXT DATA BSS HEAP -->*RTLD  SHLIBS   <--STACK]
1191		 * ^					            ^
1192		 * 0				   VM_MAXUSER_ADDRESS
1193
1194		 *
1195		 * The exact address where the loader gets mapped in
1196		 * will vary according to the size of the executable
1197		 * and the limits on the size of the process'es data
1198		 * segment at the time of exec().  The entry address
1199		 * recorded at process exec time corresponds to the
1200		 * 'start' address inside the dynamic linker.  From
1201		 * this we can figure out the address where the
1202		 * runtime loader's file object had been mapped to.
1203		 */
1204		rtldimage = pmcstat_image_from_path(image->pi_dynlinkerpath, 0);
1205		if (rtldimage == NULL) {
1206			warnx("WARNING: Cannot find image for \"%s\".",
1207			    pmcstat_string_unintern(image->pi_dynlinkerpath));
1208			pmcstat_stats.ps_exec_errors++;
1209			return;
1210		}
1211
1212		if (rtldimage->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1213			pmcstat_image_get_elf_params(rtldimage);
1214
1215		if (rtldimage->pi_type != PMCSTAT_IMAGE_ELF32 &&
1216		    rtldimage->pi_type != PMCSTAT_IMAGE_ELF64) {
1217			warnx("WARNING: rtld not an ELF object \"%s\".",
1218			    pmcstat_string_unintern(image->pi_dynlinkerpath));
1219			return;
1220		}
1221
1222		libstart = entryaddr - rtldimage->pi_entry;
1223		pmcstat_image_link(pp, rtldimage, libstart);
1224	}
1225}
1226
1227/*
1228 * Find the process descriptor corresponding to a PID.  If 'allocate'
1229 * is zero, we return a NULL if a pid descriptor could not be found or
1230 * a process descriptor process.  If 'allocate' is non-zero, then we
1231 * will attempt to allocate a fresh process descriptor.  Zombie
1232 * process descriptors are only removed if a fresh allocation for the
1233 * same PID is requested.
1234 */
1235
1236static struct pmcstat_process *
1237pmcstat_process_lookup(pid_t pid, int allocate)
1238{
1239	uint32_t hash;
1240	struct pmcstat_pcmap *ppm, *ppmtmp;
1241	struct pmcstat_process *pp, *pptmp;
1242
1243	hash = (uint32_t) pid & PMCSTAT_HASH_MASK;	/* simplicity wins */
1244
1245	LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[hash], pp_next, pptmp)
1246		if (pp->pp_pid == pid) {
1247			/* Found a descriptor, check and process zombies */
1248			if (allocate && pp->pp_isactive == 0) {
1249				/* remove maps */
1250				TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next,
1251				    ppmtmp) {
1252					TAILQ_REMOVE(&pp->pp_map, ppm,
1253					    ppm_next);
1254					free(ppm);
1255				}
1256				/* remove process entry */
1257				LIST_REMOVE(pp, pp_next);
1258				free(pp);
1259				break;
1260			}
1261			return (pp);
1262		}
1263
1264	if (!allocate)
1265		return (NULL);
1266
1267	if ((pp = malloc(sizeof(*pp))) == NULL)
1268		err(EX_OSERR, "ERROR: Cannot allocate pid descriptor");
1269
1270	pp->pp_pid = pid;
1271	pp->pp_isactive = 1;
1272
1273	TAILQ_INIT(&pp->pp_map);
1274
1275	LIST_INSERT_HEAD(&pmcstat_process_hash[hash], pp, pp_next);
1276	return (pp);
1277}
1278
1279/*
1280 * Associate an image and a process.
1281 */
1282
1283static void
1284pmcstat_process_exec(struct pmcstat_process *pp,
1285    pmcstat_interned_string path, uintfptr_t entryaddr)
1286{
1287	struct pmcstat_image *image;
1288
1289	if ((image = pmcstat_image_from_path(path, 0)) == NULL) {
1290		pmcstat_stats.ps_exec_errors++;
1291		return;
1292	}
1293
1294	if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1295		pmcstat_image_determine_type(image);
1296
1297	assert(image->pi_type != PMCSTAT_IMAGE_UNKNOWN);
1298
1299	switch (image->pi_type) {
1300	case PMCSTAT_IMAGE_ELF32:
1301	case PMCSTAT_IMAGE_ELF64:
1302		pmcstat_stats.ps_exec_elf++;
1303		pmcstat_process_elf_exec(pp, image, entryaddr);
1304		break;
1305
1306	case PMCSTAT_IMAGE_AOUT:
1307		pmcstat_stats.ps_exec_aout++;
1308		pmcstat_process_aout_exec(pp, image, entryaddr);
1309		break;
1310
1311	case PMCSTAT_IMAGE_INDETERMINABLE:
1312		pmcstat_stats.ps_exec_indeterminable++;
1313		break;
1314
1315	default:
1316		err(EX_SOFTWARE,
1317		    "ERROR: Unsupported executable type for \"%s\"",
1318		    pmcstat_string_unintern(path));
1319	}
1320}
1321
1322
1323/*
1324 * Find the map entry associated with process 'p' at PC value 'pc'.
1325 */
1326
1327struct pmcstat_pcmap *
1328pmcstat_process_find_map(struct pmcstat_process *p, uintfptr_t pc)
1329{
1330	struct pmcstat_pcmap *ppm;
1331
1332	TAILQ_FOREACH(ppm, &p->pp_map, ppm_next) {
1333		if (pc >= ppm->ppm_lowpc && pc < ppm->ppm_highpc)
1334			return (ppm);
1335		if (pc < ppm->ppm_lowpc)
1336			return (NULL);
1337	}
1338
1339	return (NULL);
1340}
1341
1342/*
1343 * Convert a hwpmc(4) log to profile information.  A system-wide
1344 * callgraph is generated if FLAG_DO_CALLGRAPHS is set.  gmon.out
1345 * files usable by gprof(1) are created if FLAG_DO_GPROF is set.
1346 */
1347static int
1348pmcstat_analyze_log(void)
1349{
1350	uint32_t cpu, cpuflags;
1351	uintfptr_t pc;
1352	pid_t pid;
1353	struct pmcstat_image *image;
1354	struct pmcstat_process *pp, *ppnew;
1355	struct pmcstat_pcmap *ppm, *ppmtmp;
1356	struct pmclog_ev ev;
1357	struct pmcstat_pmcrecord *pmcr;
1358	pmcstat_interned_string image_path;
1359
1360	assert(args.pa_flags & FLAG_DO_ANALYSIS);
1361
1362	if (elf_version(EV_CURRENT) == EV_NONE)
1363		err(EX_UNAVAILABLE, "Elf library intialization failed");
1364
1365	while (pmclog_read(args.pa_logparser, &ev) == 0) {
1366		assert(ev.pl_state == PMCLOG_OK);
1367
1368		switch (ev.pl_type) {
1369		case PMCLOG_TYPE_INITIALIZE:
1370			if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
1371			    PMC_VERSION_MAJOR << 24 && args.pa_verbosity > 0)
1372				warnx(
1373"WARNING: Log version 0x%x does not match compiled version 0x%x.",
1374				    ev.pl_u.pl_i.pl_version, PMC_VERSION_MAJOR);
1375			break;
1376
1377		case PMCLOG_TYPE_MAP_IN:
1378			/*
1379			 * Introduce an address range mapping for a
1380			 * userland process or the kernel (pid == -1).
1381			 *
1382			 * We always allocate a process descriptor so
1383			 * that subsequent samples seen for this
1384			 * address range are mapped to the current
1385			 * object being mapped in.
1386			 */
1387			pid = ev.pl_u.pl_mi.pl_pid;
1388			if (pid == -1)
1389				pp = pmcstat_kernproc;
1390			else
1391				pp = pmcstat_process_lookup(pid,
1392				    PMCSTAT_ALLOCATE);
1393
1394			assert(pp != NULL);
1395
1396			image_path = pmcstat_string_intern(ev.pl_u.pl_mi.
1397			    pl_pathname);
1398			image = pmcstat_image_from_path(image_path, pid == -1);
1399			if (image->pi_type == PMCSTAT_IMAGE_UNKNOWN)
1400				pmcstat_image_determine_type(image);
1401			if (image->pi_type != PMCSTAT_IMAGE_INDETERMINABLE)
1402				pmcstat_image_link(pp, image,
1403				    ev.pl_u.pl_mi.pl_start);
1404			break;
1405
1406		case PMCLOG_TYPE_MAP_OUT:
1407			/*
1408			 * Remove an address map.
1409			 */
1410			pid = ev.pl_u.pl_mo.pl_pid;
1411			if (pid == -1)
1412				pp = pmcstat_kernproc;
1413			else
1414				pp = pmcstat_process_lookup(pid, 0);
1415
1416			if (pp == NULL)	/* unknown process */
1417				break;
1418
1419			pmcstat_image_unmap(pp, ev.pl_u.pl_mo.pl_start,
1420			    ev.pl_u.pl_mo.pl_end);
1421			break;
1422
1423		case PMCLOG_TYPE_PCSAMPLE:
1424			/*
1425			 * Note: the `PCSAMPLE' log entry is not
1426			 * generated by hpwmc(4) after version 2.
1427			 */
1428
1429			/*
1430			 * We bring in the gmon file for the image
1431			 * currently associated with the PMC & pid
1432			 * pair and increment the appropriate entry
1433			 * bin inside this.
1434			 */
1435			pmcstat_stats.ps_samples_total++;
1436			ps_samples_period++;
1437
1438			pc = ev.pl_u.pl_s.pl_pc;
1439			pp = pmcstat_process_lookup(ev.pl_u.pl_s.pl_pid,
1440			    PMCSTAT_ALLOCATE);
1441
1442			/* Get PMC record. */
1443			pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_s.pl_pmcid);
1444			assert(pmcr != NULL);
1445			pmcr->pr_samples++;
1446
1447			/*
1448			 * Call the plugins processing
1449			 * TODO: move pmcstat_process_find_map inside plugins
1450			 */
1451
1452			if (plugins[args.pa_pplugin].pl_process != NULL)
1453				plugins[args.pa_pplugin].pl_process(
1454				    pp, pmcr, 1, &pc,
1455				    pmcstat_process_find_map(pp, pc) != NULL, 0);
1456			plugins[args.pa_plugin].pl_process(
1457			    pp, pmcr, 1, &pc,
1458			    pmcstat_process_find_map(pp, pc) != NULL, 0);
1459			break;
1460
1461		case PMCLOG_TYPE_CALLCHAIN:
1462			pmcstat_stats.ps_samples_total++;
1463			ps_samples_period++;
1464
1465			cpuflags = ev.pl_u.pl_cc.pl_cpuflags;
1466			cpu = PMC_CALLCHAIN_CPUFLAGS_TO_CPU(cpuflags);
1467
1468			/* Filter on the CPU id. */
1469			if (!CPU_ISSET(cpu, &(args.pa_cpumask))) {
1470				pmcstat_stats.ps_samples_skipped++;
1471				break;
1472			}
1473
1474			pp = pmcstat_process_lookup(ev.pl_u.pl_cc.pl_pid,
1475			    PMCSTAT_ALLOCATE);
1476
1477			/* Get PMC record. */
1478			pmcr = pmcstat_lookup_pmcid(ev.pl_u.pl_cc.pl_pmcid);
1479			assert(pmcr != NULL);
1480			pmcr->pr_samples++;
1481
1482			/*
1483			 * Call the plugins processing
1484			 */
1485
1486			if (plugins[args.pa_pplugin].pl_process != NULL)
1487				plugins[args.pa_pplugin].pl_process(
1488				    pp, pmcr,
1489				    ev.pl_u.pl_cc.pl_npc,
1490				    ev.pl_u.pl_cc.pl_pc,
1491				    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
1492				    cpu);
1493			plugins[args.pa_plugin].pl_process(
1494			    pp, pmcr,
1495			    ev.pl_u.pl_cc.pl_npc,
1496			    ev.pl_u.pl_cc.pl_pc,
1497			    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(cpuflags),
1498			    cpu);
1499			break;
1500
1501		case PMCLOG_TYPE_PMCALLOCATE:
1502			/*
1503			 * Record the association pmc id between this
1504			 * PMC and its name.
1505			 */
1506			pmcstat_pmcid_add(ev.pl_u.pl_a.pl_pmcid,
1507			    pmcstat_string_intern(ev.pl_u.pl_a.pl_evname));
1508			break;
1509
1510		case PMCLOG_TYPE_PMCALLOCATEDYN:
1511			/*
1512			 * Record the association pmc id between this
1513			 * PMC and its name.
1514			 */
1515			pmcstat_pmcid_add(ev.pl_u.pl_ad.pl_pmcid,
1516			    pmcstat_string_intern(ev.pl_u.pl_ad.pl_evname));
1517			break;
1518
1519		case PMCLOG_TYPE_PROCEXEC:
1520
1521			/*
1522			 * Change the executable image associated with
1523			 * a process.
1524			 */
1525			pp = pmcstat_process_lookup(ev.pl_u.pl_x.pl_pid,
1526			    PMCSTAT_ALLOCATE);
1527
1528			/* delete the current process map */
1529			TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
1530				TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
1531				free(ppm);
1532			}
1533
1534			/* associate this process  image */
1535			image_path = pmcstat_string_intern(
1536				ev.pl_u.pl_x.pl_pathname);
1537			assert(image_path != NULL);
1538			pmcstat_process_exec(pp, image_path,
1539			    ev.pl_u.pl_x.pl_entryaddr);
1540			break;
1541
1542		case PMCLOG_TYPE_PROCEXIT:
1543
1544			/*
1545			 * Due to the way the log is generated, the
1546			 * last few samples corresponding to a process
1547			 * may appear in the log after the process
1548			 * exit event is recorded.  Thus we keep the
1549			 * process' descriptor and associated data
1550			 * structures around, but mark the process as
1551			 * having exited.
1552			 */
1553			pp = pmcstat_process_lookup(ev.pl_u.pl_e.pl_pid, 0);
1554			if (pp == NULL)
1555				break;
1556			pp->pp_isactive = 0;	/* mark as a zombie */
1557			break;
1558
1559		case PMCLOG_TYPE_SYSEXIT:
1560			pp = pmcstat_process_lookup(ev.pl_u.pl_se.pl_pid, 0);
1561			if (pp == NULL)
1562				break;
1563			pp->pp_isactive = 0;	/* make a zombie */
1564			break;
1565
1566		case PMCLOG_TYPE_PROCFORK:
1567
1568			/*
1569			 * Allocate a process descriptor for the new
1570			 * (child) process.
1571			 */
1572			ppnew =
1573			    pmcstat_process_lookup(ev.pl_u.pl_f.pl_newpid,
1574				PMCSTAT_ALLOCATE);
1575
1576			/*
1577			 * If we had been tracking the parent, clone
1578			 * its address maps.
1579			 */
1580			pp = pmcstat_process_lookup(ev.pl_u.pl_f.pl_oldpid, 0);
1581			if (pp == NULL)
1582				break;
1583			TAILQ_FOREACH(ppm, &pp->pp_map, ppm_next)
1584			    pmcstat_image_link(ppnew, ppm->ppm_image,
1585				ppm->ppm_lowpc);
1586			break;
1587
1588		default:	/* other types of entries are not relevant */
1589			break;
1590		}
1591	}
1592
1593	if (ev.pl_state == PMCLOG_EOF)
1594		return (PMCSTAT_FINISHED);
1595	else if (ev.pl_state == PMCLOG_REQUIRE_DATA)
1596		return (PMCSTAT_RUNNING);
1597
1598	err(EX_DATAERR,
1599	    "ERROR: event parsing failed (record %jd, offset 0x%jx)",
1600	    (uintmax_t) ev.pl_count + 1, ev.pl_offset);
1601}
1602
1603/*
1604 * Print log entries as text.
1605 */
1606
1607static int
1608pmcstat_print_log(void)
1609{
1610	struct pmclog_ev ev;
1611	uint32_t npc;
1612
1613	while (pmclog_read(args.pa_logparser, &ev) == 0) {
1614		assert(ev.pl_state == PMCLOG_OK);
1615		switch (ev.pl_type) {
1616		case PMCLOG_TYPE_CALLCHAIN:
1617			PMCSTAT_PRINT_ENTRY("callchain",
1618			    "%d 0x%x %d %d %c", ev.pl_u.pl_cc.pl_pid,
1619			    ev.pl_u.pl_cc.pl_pmcid,
1620			    PMC_CALLCHAIN_CPUFLAGS_TO_CPU(ev.pl_u.pl_cc. \
1621				pl_cpuflags), ev.pl_u.pl_cc.pl_npc,
1622			    PMC_CALLCHAIN_CPUFLAGS_TO_USERMODE(ev.pl_u.pl_cc.\
1623			        pl_cpuflags) ? 'u' : 's');
1624			for (npc = 0; npc < ev.pl_u.pl_cc.pl_npc; npc++)
1625				PMCSTAT_PRINT_ENTRY("...", "%p",
1626				    (void *) ev.pl_u.pl_cc.pl_pc[npc]);
1627			break;
1628		case PMCLOG_TYPE_CLOSELOG:
1629			PMCSTAT_PRINT_ENTRY("closelog",);
1630			break;
1631		case PMCLOG_TYPE_DROPNOTIFY:
1632			PMCSTAT_PRINT_ENTRY("drop",);
1633			break;
1634		case PMCLOG_TYPE_INITIALIZE:
1635			PMCSTAT_PRINT_ENTRY("initlog","0x%x \"%s\"",
1636			    ev.pl_u.pl_i.pl_version,
1637			    pmc_name_of_cputype(ev.pl_u.pl_i.pl_arch));
1638			if ((ev.pl_u.pl_i.pl_version & 0xFF000000) !=
1639			    PMC_VERSION_MAJOR << 24 && args.pa_verbosity > 0)
1640				warnx(
1641"WARNING: Log version 0x%x != expected version 0x%x.",
1642				    ev.pl_u.pl_i.pl_version, PMC_VERSION);
1643			break;
1644		case PMCLOG_TYPE_MAP_IN:
1645			PMCSTAT_PRINT_ENTRY("map-in","%d %p \"%s\"",
1646			    ev.pl_u.pl_mi.pl_pid,
1647			    (void *) ev.pl_u.pl_mi.pl_start,
1648			    ev.pl_u.pl_mi.pl_pathname);
1649			break;
1650		case PMCLOG_TYPE_MAP_OUT:
1651			PMCSTAT_PRINT_ENTRY("map-out","%d %p %p",
1652			    ev.pl_u.pl_mo.pl_pid,
1653			    (void *) ev.pl_u.pl_mo.pl_start,
1654			    (void *) ev.pl_u.pl_mo.pl_end);
1655			break;
1656		case PMCLOG_TYPE_PCSAMPLE:
1657			PMCSTAT_PRINT_ENTRY("sample","0x%x %d %p %c",
1658			    ev.pl_u.pl_s.pl_pmcid,
1659			    ev.pl_u.pl_s.pl_pid,
1660			    (void *) ev.pl_u.pl_s.pl_pc,
1661			    ev.pl_u.pl_s.pl_usermode ? 'u' : 's');
1662			break;
1663		case PMCLOG_TYPE_PMCALLOCATE:
1664			PMCSTAT_PRINT_ENTRY("allocate","0x%x \"%s\" 0x%x",
1665			    ev.pl_u.pl_a.pl_pmcid,
1666			    ev.pl_u.pl_a.pl_evname,
1667			    ev.pl_u.pl_a.pl_flags);
1668			break;
1669		case PMCLOG_TYPE_PMCALLOCATEDYN:
1670			PMCSTAT_PRINT_ENTRY("allocatedyn","0x%x \"%s\" 0x%x",
1671			    ev.pl_u.pl_ad.pl_pmcid,
1672			    ev.pl_u.pl_ad.pl_evname,
1673			    ev.pl_u.pl_ad.pl_flags);
1674			break;
1675		case PMCLOG_TYPE_PMCATTACH:
1676			PMCSTAT_PRINT_ENTRY("attach","0x%x %d \"%s\"",
1677			    ev.pl_u.pl_t.pl_pmcid,
1678			    ev.pl_u.pl_t.pl_pid,
1679			    ev.pl_u.pl_t.pl_pathname);
1680			break;
1681		case PMCLOG_TYPE_PMCDETACH:
1682			PMCSTAT_PRINT_ENTRY("detach","0x%x %d",
1683			    ev.pl_u.pl_d.pl_pmcid,
1684			    ev.pl_u.pl_d.pl_pid);
1685			break;
1686		case PMCLOG_TYPE_PROCCSW:
1687			PMCSTAT_PRINT_ENTRY("cswval","0x%x %d %jd",
1688			    ev.pl_u.pl_c.pl_pmcid,
1689			    ev.pl_u.pl_c.pl_pid,
1690			    ev.pl_u.pl_c.pl_value);
1691			break;
1692		case PMCLOG_TYPE_PROCEXEC:
1693			PMCSTAT_PRINT_ENTRY("exec","0x%x %d %p \"%s\"",
1694			    ev.pl_u.pl_x.pl_pmcid,
1695			    ev.pl_u.pl_x.pl_pid,
1696			    (void *) ev.pl_u.pl_x.pl_entryaddr,
1697			    ev.pl_u.pl_x.pl_pathname);
1698			break;
1699		case PMCLOG_TYPE_PROCEXIT:
1700			PMCSTAT_PRINT_ENTRY("exitval","0x%x %d %jd",
1701			    ev.pl_u.pl_e.pl_pmcid,
1702			    ev.pl_u.pl_e.pl_pid,
1703			    ev.pl_u.pl_e.pl_value);
1704			break;
1705		case PMCLOG_TYPE_PROCFORK:
1706			PMCSTAT_PRINT_ENTRY("fork","%d %d",
1707			    ev.pl_u.pl_f.pl_oldpid,
1708			    ev.pl_u.pl_f.pl_newpid);
1709			break;
1710		case PMCLOG_TYPE_USERDATA:
1711			PMCSTAT_PRINT_ENTRY("userdata","0x%x",
1712			    ev.pl_u.pl_u.pl_userdata);
1713			break;
1714		case PMCLOG_TYPE_SYSEXIT:
1715			PMCSTAT_PRINT_ENTRY("exit","%d",
1716			    ev.pl_u.pl_se.pl_pid);
1717			break;
1718		default:
1719			fprintf(args.pa_printfile, "unknown event (type %d).\n",
1720			    ev.pl_type);
1721		}
1722	}
1723
1724	if (ev.pl_state == PMCLOG_EOF)
1725		return (PMCSTAT_FINISHED);
1726	else if (ev.pl_state ==  PMCLOG_REQUIRE_DATA)
1727		return (PMCSTAT_RUNNING);
1728
1729	errx(EX_DATAERR,
1730	    "ERROR: event parsing failed (record %jd, offset 0x%jx).",
1731	    (uintmax_t) ev.pl_count + 1, ev.pl_offset);
1732	/*NOTREACHED*/
1733}
1734
1735/*
1736 * Public Interfaces.
1737 */
1738
1739/*
1740 * Close a logfile, after first flushing all in-module queued data.
1741 */
1742
1743int
1744pmcstat_close_log(void)
1745{
1746	/* If a local logfile is configured ask the kernel to stop
1747	 * and flush data. Kernel will close the file when data is flushed
1748	 * so keep the status to EXITING.
1749	 */
1750	if (args.pa_logfd != -1) {
1751		if (pmc_close_logfile() < 0)
1752			err(EX_OSERR, "ERROR: logging failed");
1753	}
1754
1755	return (args.pa_flags & FLAG_HAS_PIPE ? PMCSTAT_EXITING :
1756	    PMCSTAT_FINISHED);
1757}
1758
1759
1760
1761/*
1762 * Open a log file, for reading or writing.
1763 *
1764 * The function returns the fd of a successfully opened log or -1 in
1765 * case of failure.
1766 */
1767
1768int
1769pmcstat_open_log(const char *path, int mode)
1770{
1771	int error, fd, cfd;
1772	size_t hlen;
1773	const char *p, *errstr;
1774	struct addrinfo hints, *res, *res0;
1775	char hostname[MAXHOSTNAMELEN];
1776
1777	errstr = NULL;
1778	fd = -1;
1779
1780	/*
1781	 * If 'path' is "-" then open one of stdin or stdout depending
1782	 * on the value of 'mode'.
1783	 *
1784	 * If 'path' contains a ':' and does not start with a '/' or '.',
1785	 * and is being opened for writing, treat it as a "host:port"
1786	 * specification and open a network socket.
1787	 *
1788	 * Otherwise, treat 'path' as a file name and open that.
1789	 */
1790	if (path[0] == '-' && path[1] == '\0')
1791		fd = (mode == PMCSTAT_OPEN_FOR_READ) ? 0 : 1;
1792	else if (path[0] != '/' &&
1793	    path[0] != '.' && strchr(path, ':') != NULL) {
1794
1795		p = strrchr(path, ':');
1796		hlen = p - path;
1797		if (p == path || hlen >= sizeof(hostname)) {
1798			errstr = strerror(EINVAL);
1799			goto done;
1800		}
1801
1802		assert(hlen < sizeof(hostname));
1803		(void) strncpy(hostname, path, hlen);
1804		hostname[hlen] = '\0';
1805
1806		(void) memset(&hints, 0, sizeof(hints));
1807		hints.ai_family = AF_UNSPEC;
1808		hints.ai_socktype = SOCK_STREAM;
1809		if ((error = getaddrinfo(hostname, p+1, &hints, &res0)) != 0) {
1810			errstr = gai_strerror(error);
1811			goto done;
1812		}
1813
1814		fd = -1;
1815		for (res = res0; res; res = res->ai_next) {
1816			if ((fd = socket(res->ai_family, res->ai_socktype,
1817			    res->ai_protocol)) < 0) {
1818				errstr = strerror(errno);
1819				continue;
1820			}
1821			if (mode == PMCSTAT_OPEN_FOR_READ) {
1822				if (bind(fd, res->ai_addr, res->ai_addrlen) < 0) {
1823					errstr = strerror(errno);
1824					(void) close(fd);
1825					fd = -1;
1826					continue;
1827				}
1828				listen(fd, 1);
1829				cfd = accept(fd, NULL, NULL);
1830				(void) close(fd);
1831				if (cfd < 0) {
1832					errstr = strerror(errno);
1833					fd = -1;
1834					break;
1835				}
1836				fd = cfd;
1837			} else {
1838				if (connect(fd, res->ai_addr, res->ai_addrlen) < 0) {
1839					errstr = strerror(errno);
1840					(void) close(fd);
1841					fd = -1;
1842					continue;
1843				}
1844			}
1845			errstr = NULL;
1846			break;
1847		}
1848		freeaddrinfo(res0);
1849
1850	} else if ((fd = open(path, mode == PMCSTAT_OPEN_FOR_READ ?
1851		    O_RDONLY : (O_WRONLY|O_CREAT|O_TRUNC),
1852		    S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)) < 0)
1853			errstr = strerror(errno);
1854
1855  done:
1856	if (errstr)
1857		errx(EX_OSERR, "ERROR: Cannot open \"%s\" for %s: %s.", path,
1858		    (mode == PMCSTAT_OPEN_FOR_READ ? "reading" : "writing"),
1859		    errstr);
1860
1861	return (fd);
1862}
1863
1864/*
1865 * Process a log file in offline analysis mode.
1866 */
1867
1868int
1869pmcstat_process_log(void)
1870{
1871
1872	/*
1873	 * If analysis has not been asked for, just print the log to
1874	 * the current output file.
1875	 */
1876	if (args.pa_flags & FLAG_DO_PRINT)
1877		return (pmcstat_print_log());
1878	else
1879		return (pmcstat_analyze_log());
1880}
1881
1882/*
1883 * Refresh top display.
1884 */
1885
1886static void
1887pmcstat_refresh_top(void)
1888{
1889	int v_attrs;
1890	float v;
1891	char pmcname[40];
1892	struct pmcstat_pmcrecord *pmcpr;
1893
1894	/* If in pause mode do not refresh display. */
1895	if (pmcstat_pause)
1896		return;
1897
1898	/* Wait until PMC pop in the log. */
1899	pmcpr = pmcstat_pmcindex_to_pmcr(pmcstat_pmcinfilter);
1900	if (pmcpr == NULL)
1901		return;
1902
1903	/* Format PMC name. */
1904	if (pmcstat_mergepmc)
1905		snprintf(pmcname, sizeof(pmcname), "[%s]",
1906		    pmcstat_string_unintern(pmcpr->pr_pmcname));
1907	else
1908		snprintf(pmcname, sizeof(pmcname), "%s.%d",
1909		    pmcstat_string_unintern(pmcpr->pr_pmcname),
1910		    pmcstat_pmcinfilter);
1911
1912	/* Format samples count. */
1913	if (ps_samples_period > 0)
1914		v = (pmcpr->pr_samples * 100.0) / ps_samples_period;
1915	else
1916		v = 0.;
1917	v_attrs = PMCSTAT_ATTRPERCENT(v);
1918
1919	PMCSTAT_PRINTBEGIN();
1920	PMCSTAT_PRINTW("PMC: %s Samples: %u ",
1921	    pmcname,
1922	    pmcpr->pr_samples);
1923	PMCSTAT_ATTRON(v_attrs);
1924	PMCSTAT_PRINTW("(%.1f%%) ", v);
1925	PMCSTAT_ATTROFF(v_attrs);
1926	PMCSTAT_PRINTW(", %u unresolved\n\n",
1927	    pmcpr->pr_dubious_frames);
1928	if (plugins[args.pa_plugin].pl_topdisplay != NULL)
1929		plugins[args.pa_plugin].pl_topdisplay();
1930	PMCSTAT_PRINTEND();
1931}
1932
1933/*
1934 * Find the next pmc index to display.
1935 */
1936
1937static void
1938pmcstat_changefilter(void)
1939{
1940	int pmcin;
1941	struct pmcstat_pmcrecord *pmcr;
1942
1943	/*
1944	 * Find the next merge target.
1945	 */
1946	if (pmcstat_mergepmc) {
1947		pmcin = pmcstat_pmcinfilter;
1948
1949		do {
1950			pmcr = pmcstat_pmcindex_to_pmcr(pmcstat_pmcinfilter);
1951			if (pmcr == NULL || pmcr == pmcr->pr_merge)
1952				break;
1953
1954			pmcstat_pmcinfilter++;
1955			if (pmcstat_pmcinfilter >= pmcstat_npmcs)
1956				pmcstat_pmcinfilter = 0;
1957
1958		} while (pmcstat_pmcinfilter != pmcin);
1959	}
1960}
1961
1962/*
1963 * Top mode keypress.
1964 */
1965
1966int
1967pmcstat_keypress_log(void)
1968{
1969	int c, ret = 0;
1970	WINDOW *w;
1971
1972	w = newwin(1, 0, 1, 0);
1973	c = wgetch(w);
1974	wprintw(w, "Key: %c => ", c);
1975	switch (c) {
1976	case 'c':
1977		wprintw(w, "enter mode 'd' or 'a' => ");
1978		c = wgetch(w);
1979		if (c == 'd') {
1980			args.pa_topmode = PMCSTAT_TOP_DELTA;
1981			wprintw(w, "switching to delta mode");
1982		} else {
1983			args.pa_topmode = PMCSTAT_TOP_ACCUM;
1984			wprintw(w, "switching to accumulation mode");
1985		}
1986		break;
1987	case 'm':
1988		pmcstat_mergepmc = !pmcstat_mergepmc;
1989		/*
1990		 * Changing merge state require data reset.
1991		 */
1992		if (plugins[args.pa_plugin].pl_shutdown != NULL)
1993			plugins[args.pa_plugin].pl_shutdown(NULL);
1994		pmcstat_stats_reset(0);
1995		if (plugins[args.pa_plugin].pl_init != NULL)
1996			plugins[args.pa_plugin].pl_init();
1997
1998		/* Update filter to be on a merge target. */
1999		pmcstat_changefilter();
2000		wprintw(w, "merge PMC %s", pmcstat_mergepmc ? "on" : "off");
2001		break;
2002	case 'n':
2003		/* Close current plugin. */
2004		if (plugins[args.pa_plugin].pl_shutdown != NULL)
2005			plugins[args.pa_plugin].pl_shutdown(NULL);
2006
2007		/* Find next top display available. */
2008		do {
2009			args.pa_plugin++;
2010			if (plugins[args.pa_plugin].pl_name == NULL)
2011				args.pa_plugin = 0;
2012		} while (plugins[args.pa_plugin].pl_topdisplay == NULL);
2013
2014		/* Open new plugin. */
2015		pmcstat_stats_reset(0);
2016		if (plugins[args.pa_plugin].pl_init != NULL)
2017			plugins[args.pa_plugin].pl_init();
2018		wprintw(w, "switching to plugin %s",
2019		    plugins[args.pa_plugin].pl_name);
2020		break;
2021	case 'p':
2022		pmcstat_pmcinfilter++;
2023		if (pmcstat_pmcinfilter >= pmcstat_npmcs)
2024			pmcstat_pmcinfilter = 0;
2025		pmcstat_changefilter();
2026		wprintw(w, "switching to PMC %s.%d",
2027		    pmcstat_pmcindex_to_name(pmcstat_pmcinfilter),
2028		    pmcstat_pmcinfilter);
2029		break;
2030	case ' ':
2031		pmcstat_pause = !pmcstat_pause;
2032		if (pmcstat_pause)
2033			wprintw(w, "pause => press space again to continue");
2034		break;
2035	case 'q':
2036		wprintw(w, "exiting...");
2037		ret = 1;
2038		break;
2039	default:
2040		if (plugins[args.pa_plugin].pl_topkeypress != NULL)
2041			if (plugins[args.pa_plugin].pl_topkeypress(c, w))
2042				ret = 1;
2043	}
2044
2045	wrefresh(w);
2046	delwin(w);
2047	return ret;
2048}
2049
2050
2051/*
2052 * Top mode display.
2053 */
2054
2055void
2056pmcstat_display_log(void)
2057{
2058
2059	pmcstat_refresh_top();
2060
2061	/* Reset everythings if delta mode. */
2062	if (args.pa_topmode == PMCSTAT_TOP_DELTA) {
2063		if (plugins[args.pa_plugin].pl_shutdown != NULL)
2064			plugins[args.pa_plugin].pl_shutdown(NULL);
2065		pmcstat_stats_reset(0);
2066		if (plugins[args.pa_plugin].pl_init != NULL)
2067			plugins[args.pa_plugin].pl_init();
2068	}
2069
2070}
2071
2072/*
2073 * Configure a plugins.
2074 */
2075
2076void
2077pmcstat_pluginconfigure_log(char *opt)
2078{
2079
2080	if (strncmp(opt, "threshold=", 10) == 0) {
2081		pmcstat_threshold = atof(opt+10);
2082	} else {
2083		if (plugins[args.pa_plugin].pl_configure != NULL) {
2084			if (!plugins[args.pa_plugin].pl_configure(opt))
2085				err(EX_USAGE,
2086				    "ERROR: unknown option <%s>.", opt);
2087		}
2088	}
2089}
2090
2091/*
2092 * Initialize module.
2093 */
2094
2095void
2096pmcstat_initialize_logging(void)
2097{
2098	int i;
2099
2100	/* use a convenient format for 'ldd' output */
2101	if (setenv("LD_TRACE_LOADED_OBJECTS_FMT1","%o \"%p\" %x\n",1) != 0)
2102		err(EX_OSERR, "ERROR: Cannot setenv");
2103
2104	/* Initialize hash tables */
2105	pmcstat_string_initialize();
2106	for (i = 0; i < PMCSTAT_NHASH; i++) {
2107		LIST_INIT(&pmcstat_image_hash[i]);
2108		LIST_INIT(&pmcstat_process_hash[i]);
2109	}
2110
2111	/*
2112	 * Create a fake 'process' entry for the kernel with pid -1.
2113	 * hwpmc(4) will subsequently inform us about where the kernel
2114	 * and any loaded kernel modules are mapped.
2115	 */
2116	if ((pmcstat_kernproc = pmcstat_process_lookup((pid_t) -1,
2117		 PMCSTAT_ALLOCATE)) == NULL)
2118		err(EX_OSERR, "ERROR: Cannot initialize logging");
2119
2120	/* PMC count. */
2121	pmcstat_npmcs = 0;
2122
2123	/* Merge PMC with same name. */
2124	pmcstat_mergepmc = args.pa_mergepmc;
2125
2126	/*
2127	 * Initialize plugins
2128	 */
2129
2130	if (plugins[args.pa_pplugin].pl_init != NULL)
2131		plugins[args.pa_pplugin].pl_init();
2132	if (plugins[args.pa_plugin].pl_init != NULL)
2133		plugins[args.pa_plugin].pl_init();
2134}
2135
2136/*
2137 * Shutdown module.
2138 */
2139
2140void
2141pmcstat_shutdown_logging(void)
2142{
2143	int i;
2144	FILE *mf;
2145	struct pmcstat_image *pi, *pitmp;
2146	struct pmcstat_process *pp, *pptmp;
2147	struct pmcstat_pcmap *ppm, *ppmtmp;
2148
2149	/* determine where to send the map file */
2150	mf = NULL;
2151	if (args.pa_mapfilename != NULL)
2152		mf = (strcmp(args.pa_mapfilename, "-") == 0) ?
2153		    args.pa_printfile : fopen(args.pa_mapfilename, "w");
2154
2155	if (mf == NULL && args.pa_flags & FLAG_DO_GPROF &&
2156	    args.pa_verbosity >= 2)
2157		mf = args.pa_printfile;
2158
2159	if (mf)
2160		(void) fprintf(mf, "MAP:\n");
2161
2162	/*
2163	 * Shutdown the plugins
2164	 */
2165
2166	if (plugins[args.pa_plugin].pl_shutdown != NULL)
2167		plugins[args.pa_plugin].pl_shutdown(mf);
2168	if (plugins[args.pa_pplugin].pl_shutdown != NULL)
2169		plugins[args.pa_pplugin].pl_shutdown(mf);
2170
2171	for (i = 0; i < PMCSTAT_NHASH; i++) {
2172		LIST_FOREACH_SAFE(pi, &pmcstat_image_hash[i], pi_next,
2173		    pitmp) {
2174			if (plugins[args.pa_plugin].pl_shutdownimage != NULL)
2175				plugins[args.pa_plugin].pl_shutdownimage(pi);
2176			if (plugins[args.pa_pplugin].pl_shutdownimage != NULL)
2177				plugins[args.pa_pplugin].pl_shutdownimage(pi);
2178
2179			free(pi->pi_symbols);
2180			if (pi->pi_addr2line != NULL)
2181				pclose(pi->pi_addr2line);
2182			LIST_REMOVE(pi, pi_next);
2183			free(pi);
2184		}
2185
2186		LIST_FOREACH_SAFE(pp, &pmcstat_process_hash[i], pp_next,
2187		    pptmp) {
2188			TAILQ_FOREACH_SAFE(ppm, &pp->pp_map, ppm_next, ppmtmp) {
2189				TAILQ_REMOVE(&pp->pp_map, ppm, ppm_next);
2190				free(ppm);
2191			}
2192			LIST_REMOVE(pp, pp_next);
2193			free(pp);
2194		}
2195	}
2196
2197	pmcstat_string_shutdown();
2198
2199	/*
2200	 * Print errors unless -q was specified.  Print all statistics
2201	 * if verbosity > 1.
2202	 */
2203#define	PRINT(N,V) do {							\
2204		if (pmcstat_stats.ps_##V || args.pa_verbosity >= 2)	\
2205			(void) fprintf(args.pa_printfile, " %-40s %d\n",\
2206			    N, pmcstat_stats.ps_##V);			\
2207	} while (0)
2208
2209	if (args.pa_verbosity >= 1 && (args.pa_flags & FLAG_DO_ANALYSIS)) {
2210		(void) fprintf(args.pa_printfile, "CONVERSION STATISTICS:\n");
2211		PRINT("#exec/a.out", exec_aout);
2212		PRINT("#exec/elf", exec_elf);
2213		PRINT("#exec/unknown", exec_indeterminable);
2214		PRINT("#exec handling errors", exec_errors);
2215		PRINT("#samples/total", samples_total);
2216		PRINT("#samples/unclaimed", samples_unknown_offset);
2217		PRINT("#samples/unknown-object", samples_indeterminable);
2218		PRINT("#samples/unknown-function", samples_unknown_function);
2219		PRINT("#callchain/dubious-frames", callchain_dubious_frames);
2220	}
2221
2222	if (mf)
2223		(void) fclose(mf);
2224}
2225