MFC r290811:

  Fix hwpmc "stalled" behavior

  Currently, there is a single pm_stalled flag that tracks whether a
  performance monitor was "stalled" due to insufficent ring buffer
  space for samples. However, because the same performance monitor
  can run on multiple processes or threads at the same time, a single
  pm_stalled flag that impacts them all seems insufficient.

  In particular, you can hit corner cases where the code fails to stop
  performance monitors during a context switch out, because it thinks
  the performance monitor is already stopped. However, in reality,
  it may be that only the monitor running on a different CPU was stalled.

  This patch attempts to fix that behavior by tracking on a per-CPU basis
  whether a PM desires to run and whether it is "stalled". This lets the
  code make better decisions about when to stop PMs and when to try to
  restart them. Ideally, we should avoid the case where the code fails
  to stop a PM during a context switch out.

MFC r290813:
  Optimizations to the way hwpmc gathers user callchains

  Changes to the code to gather user stacks:
  * Delay setting pmc_cpumask until we actually have the stack.
  * When recording user stack traces, only walk the portion of the ring
    that should have samples for us.

MFC r290929:
  Change the driver stats to what they really are: unsigned values.

  When pmcstat exits after some samples were dropped, give the user an
  idea of how many were lost. (Granted, these are global numbers, but
  they may still help quantify the scope of the loss.)

MFC r290930:
  Improve accuracy of PMC sampling frequency

  The code tracks a counter which is the number of events until the next
  sample. On context switch in, it loads the saved counter. On context
  switch out, it tries to calculate a new saved counter.

  Problems:

  1. The saved counter was shared by all threads in a process. However, this
  means that all threads would be initially loaded with the same saved
  counter. However, that could result in sampling more often than once every
  X number of events.

  2. The calculation to determine a new saved counter was backwards. It
  added when it should have subtracted, and subtracted when it should have
  added. Assume a single-threaded process with a reload count of 1000
  events. Assuming the counter on context switch in was 100 and the counter
  on context switch out was 50 (meaning the thread has "consumed" 50 more
  events), the code would calculate a new saved counter of 150 (instead of
  the proper 50).

  Fix:

  1. As soon as the saved counter is used to initialize a monitor for a
  thread on context switch in, set the saved counter to the reload count.
  That way, subsequent threads to use the saved counter will get the full
  reload count, assuring we sample at least once every X number of events
  (across all threads).

  2. Change the calculation of the saved counter. Due to the change to the
  saved counter in #1, we simply need to add (modulo the reload count) the
  remaining counter time we retrieve from the CPU when a thread is context
  switched out.

MFC r291016:
  Support a wider history counter in pmcstat(8) gmon output

  pmcstat(8) contains an option to output sampling data in a gmon format
  compatible with gprof(1). Currently, it uses the default histcounter,
  which is an (unsigned short). With large sets of sampling data, it
  is possible to overflow the maximum value provided by an (unsigned
  short).

  This change adds the -e argument to pmcstat. If -e and -g are both
  specified, pmcstat will use a histcounter type of uint64_t.

MFC r291017:
  Fix the date on the pmcstat(8) man page from r291016.

4 files changed, 78 insertions, 18 deletions

diff --git a/usr.sbin/pmcstat/pmcpl_gprof.c b/usr.sbin/pmcstat/pmcpl_gprof.c
index 9ff78e8..5fc9b41 100644
--- a/usr.sbin/pmcstat/pmcpl_gprof.c
+++ b/usr.sbin/pmcstat/pmcpl_gprof.c
@@ -74,6 +74,14 @@ __FBSDID("$FreeBSD$");
 #include "pmcpl_callgraph.h"
 #include "pmcpl_gprof.h"
 
+typedef	uint64_t	WIDEHISTCOUNTER;
+
+#define	WIDEHISTCOUNTER_MAX		UINT64_MAX
+#define	HISTCOUNTER_MAX			USHRT_MAX
+#define	WIDEHISTCOUNTER_GMONTYPE	((int) 64)
+#define	HISTCOUNTER_GMONTYPE		((int) 0)
+static int hc_sz=0;
+
 /*
  * struct pmcstat_gmonfile tracks a given 'gmon.out' file.  These
  * files are mmap()'ed in as needed.
@@ -126,11 +134,13 @@ pmcstat_gmon_create_file(struct pmcstat_gmonfile *pgf,
 
 	gm.lpc = image->pi_start;
 	gm.hpc = image->pi_end;
-	gm.ncnt = (pgf->pgf_nbuckets * sizeof(HISTCOUNTER)) +
-	    sizeof(struct gmonhdr);
+	gm.ncnt = (pgf->pgf_nbuckets * hc_sz) + sizeof(struct gmonhdr);
 	gm.version = GMONVERSION;
 	gm.profrate = 0;		/* use ticks */
-	gm.histcounter_type = 0;	/* compatibility with moncontrol() */
+	if (args.pa_flags & FLAG_DO_WIDE_GPROF_HC)
+		gm.histcounter_type = WIDEHISTCOUNTER_GMONTYPE;
+	else
+		gm.histcounter_type = HISTCOUNTER_GMONTYPE;
 	gm.spare[0] = gm.spare[1] = 0;
 
 	/* Write out the gmon header */
@@ -400,6 +410,7 @@ pmcpl_gmon_process(struct pmcstat_process *pp, struct pmcstat_pmcrecord *pmcr,
 	struct pmcstat_gmonfile *pgf;
 	uintfptr_t bucket;
 	HISTCOUNTER *hc;
+	WIDEHISTCOUNTER *whc;
 	pmc_id_t pmcid;
 
 	(void) nsamples; (void) usermode; (void) cpu;
@@ -437,6 +448,14 @@ pmcpl_gmon_process(struct pmcstat_process *pp, struct pmcstat_pmcrecord *pmcr,
 	 */
 	pgf = pmcstat_image_find_gmonfile(image, pmcid);
 	if (pgf == NULL) {
+		if (hc_sz == 0) {
+			/* Determine the correct histcounter size. */
+			if (args.pa_flags & FLAG_DO_WIDE_GPROF_HC)
+				hc_sz = sizeof(WIDEHISTCOUNTER);
+			else
+				hc_sz = sizeof(HISTCOUNTER);
+		}
+
 		if ((pgf = calloc(1, sizeof(*pgf))) == NULL)
 			err(EX_OSERR, "ERROR:");
 
@@ -448,7 +467,7 @@ pmcpl_gmon_process(struct pmcstat_process *pp, struct pmcstat_pmcrecord *pmcr,
 		pgf->pgf_nbuckets = (image->pi_end - image->pi_start) /
 		    FUNCTION_ALIGNMENT;	/* see <machine/profile.h> */
 		pgf->pgf_ndatabytes = sizeof(struct gmonhdr) +
-		    pgf->pgf_nbuckets * sizeof(HISTCOUNTER);
+		    pgf->pgf_nbuckets * hc_sz;
 		pgf->pgf_nsamples = 0;
 		pgf->pgf_file = NULL;
 
@@ -474,14 +493,25 @@ pmcpl_gmon_process(struct pmcstat_process *pp, struct pmcstat_pmcrecord *pmcr,
 
 	assert(bucket < pgf->pgf_nbuckets);
 
-	hc = (HISTCOUNTER *) ((uintptr_t) pgf->pgf_gmondata +
-	    sizeof(struct gmonhdr));
-
-	/* saturating add */
-	if (hc[bucket] < 0xFFFFU)  /* XXX tie this to sizeof(HISTCOUNTER) */
-		hc[bucket]++;
-	else /* mark that an overflow occurred */
-		pgf->pgf_overflow = 1;
+	if (args.pa_flags & FLAG_DO_WIDE_GPROF_HC) {
+		whc = (WIDEHISTCOUNTER *) ((uintptr_t) pgf->pgf_gmondata +
+		    sizeof(struct gmonhdr));
+
+		/* saturating add */
+		if (whc[bucket] < WIDEHISTCOUNTER_MAX)
+			whc[bucket]++;
+		else /* mark that an overflow occurred */
+			pgf->pgf_overflow = 1;
+	} else {
+		hc = (HISTCOUNTER *) ((uintptr_t) pgf->pgf_gmondata +
+		    sizeof(struct gmonhdr));
+
+		/* saturating add */
+		if (hc[bucket] < HISTCOUNTER_MAX)
+			hc[bucket]++;
+		else /* mark that an overflow occurred */
+			pgf->pgf_overflow = 1;
+	}
 
 	pgf->pgf_nsamples++;
 }
diff --git a/usr.sbin/pmcstat/pmcstat.8 b/usr.sbin/pmcstat/pmcstat.8
index 7de335d..31b136b 100644
--- a/usr.sbin/pmcstat/pmcstat.8
+++ b/usr.sbin/pmcstat/pmcstat.8
@@ -25,7 +25,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd May 27, 2015
+.Dd November 18, 2015
 .Dt PMCSTAT 8
 .Os
 .Sh NAME
@@ -49,6 +49,7 @@
 .Op Fl a Ar pathname
 .Op Fl c Ar cpu-spec
 .Op Fl d
+.Op Fl e
 .Op Fl f Ar pluginopt
 .Op Fl g
 .Op Fl k Ar kerneldir
@@ -260,6 +261,12 @@ The default is to measure events for the target process alone.
 .Fl P ,
 or
 .Fl S ) .
+.It Fl e
+Specify that the gprof profile files will use a wide history counter.
+These files are produced in a format compatible with
+.Xr gprof 1 .
+However, other tools that cannot fully parse a BSD-style
+gmon header might be unable to correctly parse these files.
 .It Fl f Ar pluginopt
 Pass option string to the active plugin.
 .br
diff --git a/usr.sbin/pmcstat/pmcstat.c b/usr.sbin/pmcstat/pmcstat.c
index 5ae53aa..481213a 100644
--- a/usr.sbin/pmcstat/pmcstat.c
+++ b/usr.sbin/pmcstat/pmcstat.c
@@ -506,6 +506,7 @@ pmcstat_show_usage(void)
 	    "\t -a <file>\t print sampled PCs and callgraph to \"file\"\n"
 	    "\t -c cpu-list\t set cpus for subsequent system-wide PMCs\n"
 	    "\t -d\t\t (toggle) track descendants\n"
+	    "\t -e\t\t use wide history counter for gprof(1) output\n"
 	    "\t -f spec\t pass \"spec\" to as plugin option\n"
 	    "\t -g\t\t produce gprof(1) compatible profiles\n"
 	    "\t -k dir\t\t set the path to the kernel\n"
@@ -627,7 +628,7 @@ main(int argc, char **argv)
 	CPU_COPY(&rootmask, &cpumask);
 
 	while ((option = getopt(argc, argv,
-	    "CD:EF:G:M:NO:P:R:S:TWa:c:df:gk:l:m:n:o:p:qr:s:t:vw:z:")) != -1)
+	    "CD:EF:G:M:NO:P:R:S:TWa:c:def:gk:l:m:n:o:p:qr:s:t:vw:z:")) != -1)
 		switch (option) {
 		case 'a':	/* Annotate + callgraph */
 			args.pa_flags |= FLAG_DO_ANNOTATE;
@@ -668,6 +669,10 @@ main(int argc, char **argv)
 			args.pa_required |= FLAG_HAS_PROCESS_PMCS;
 			break;
 
+		case 'e':	/* wide gprof metrics */
+			args.pa_flags |= FLAG_DO_WIDE_GPROF_HC;
+			break;
+
 		case 'F':	/* produce a system-wide calltree */
 			args.pa_flags |= FLAG_DO_CALLGRAPHS;
 			args.pa_plugin = PMCSTAT_PL_CALLTREE;
@@ -1022,6 +1027,13 @@ main(int argc, char **argv)
 "ERROR: options -g/-G/-m/-T require sampling PMCs or -R to be specified."
 		    );
 
+	/* check if -e was specified without -g */
+	if ((args.pa_flags & FLAG_DO_WIDE_GPROF_HC) &&
+	    !(args.pa_flags & FLAG_DO_GPROF))
+		errx(EX_USAGE,
+"ERROR: option -e requires gprof mode to be specified."
+		    );
+
 	/* check if -O was spuriously specified */
 	if ((args.pa_flags & FLAG_HAS_OUTPUT_LOGFILE) &&
 	    (args.pa_required & FLAG_HAS_OUTPUT_LOGFILE) == 0)
@@ -1500,14 +1512,24 @@ main(int argc, char **argv)
 			    "ERROR: Cannot retrieve driver statistics");
 		if (ds_start.pm_intr_bufferfull != ds_end.pm_intr_bufferfull &&
 		    args.pa_verbosity > 0)
-			warnx("WARNING: some samples were dropped.\n"
-"Please consider tuning the \"kern.hwpmc.nsamples\" tunable."
+			warnx(
+"WARNING: sampling was paused at least %u time%s.\n"
+"Please consider tuning the \"kern.hwpmc.nsamples\" tunable.",
+			    ds_end.pm_intr_bufferfull -
+			    ds_start.pm_intr_bufferfull,
+			    ((ds_end.pm_intr_bufferfull -
+			    ds_start.pm_intr_bufferfull) != 1) ? "s" : ""
 			    );
 		if (ds_start.pm_buffer_requests_failed !=
 		    ds_end.pm_buffer_requests_failed &&
 		    args.pa_verbosity > 0)
-			warnx("WARNING: some events were discarded.\n"
-"Please consider tuning the \"kern.hwpmc.nbuffers\" tunable."
+			warnx(
+"WARNING: at least %u event%s were discarded while running.\n"
+"Please consider tuning the \"kern.hwpmc.nbuffers\" tunable.",
+	 		    ds_end.pm_buffer_requests_failed -
+			    ds_start.pm_buffer_requests_failed,
+			    ((ds_end.pm_buffer_requests_failed -
+			    ds_start.pm_buffer_requests_failed) != 1) ? "s" : ""
 			    );
 	}
 
diff --git a/usr.sbin/pmcstat/pmcstat.h b/usr.sbin/pmcstat/pmcstat.h
index 29dfeb7..5b1d3d9 100644
--- a/usr.sbin/pmcstat/pmcstat.h
+++ b/usr.sbin/pmcstat/pmcstat.h
@@ -55,6 +55,7 @@
 #define	FLAG_DO_ANALYSIS		0x00020000	/* -g or -G or -m or -T */
 #define	FLAGS_HAS_CPUMASK		0x00040000	/* -c */
 #define	FLAG_HAS_DURATION		0x00080000	/* -l secs */
+#define	FLAG_DO_WIDE_GPROF_HC		0x00100000	/* -e */
 
 #define	DEFAULT_SAMPLE_COUNT		65536
 #define	DEFAULT_WAIT_INTERVAL		5.0