Page MenuHomeFreeBSD
Files F160478875

D57637.diff
No OneTemporary
Actions

D57637.diff
View Options

diff --git a/usr.sbin/pmcstat/pmcstat.8 b/usr.sbin/pmcstat/pmcstat.8
--- a/usr.sbin/pmcstat/pmcstat.8
+++ b/usr.sbin/pmcstat/pmcstat.8
@@ -309,6 +309,36 @@
saved with the
.Fl O
option.
+.It Fl b
+Enable PMU event grouping.
+When this flag is set, an argument to
+.Fl p ,
+.Fl P ,
+.Fl s
+or
+.Fl S
+may be a brace-list of the form
+.Sq Brq Ar ev1 , Ar ev2 , Ar ev3 :
+all listed events are programmed atomically as a single PMU group, so
+the counter values for siblings are read at consistent program points.
+Every brace-mode group leader is allocated with
+.Dv PMC_F_GROUP_MUX
+set, which has no effect on groups that fit in hardware but lets the
+kernel transparently time-multiplex any group that overflows the
+.Em remaining
+hardware counter budget.
+For example, on a 6-counter Zen3 the invocation
+.Bd -literal -offset indent
+pmcstat -b -p '{instructions,unhalted-cycles}' \\
+ -p '{ls_alloc_mab_count,ls_not_halted_cyc,ls_dispatch.all}' \\
+ -p '{ls_smi_rx,ls_int_taken,instructions}' sleep 60
+.Ed
+.Pp
+attaches three groups (2+3+3 = 8 events) to one target.
+The first two groups consume 5 of 6 counters, leaving 1.
+The third group is then committed in multiplex mode so its 3 events
+rotate through the remaining slot rather than failing with
+.Er ENOSPC .
.It Fl c Ar cpu-spec
Set the cpus for subsequent system mode PMCs specified on the
command line to
diff --git a/usr.sbin/pmcstat/pmcstat.c b/usr.sbin/pmcstat/pmcstat.c
--- a/usr.sbin/pmcstat/pmcstat.c
+++ b/usr.sbin/pmcstat/pmcstat.c
@@ -115,6 +115,7 @@
static struct kinfo_proc *pmcstat_plist;
struct pmcstat_args args;
static bool libpmc_initialized = false;
+static int next_syntactic_gid = 1;
static void
pmcstat_get_cpumask(const char *cpuspec, cpuset_t *cpumask)
@@ -519,8 +520,12 @@
CPU_COPY(&rootmask, &cpumask);
while ((option = getopt(argc, argv,
- "ACD:EF:G:ILM:NO:P:R:S:TUWZa:c:def:gi:l:m:n:o:p:qr:s:t:u:vw:z:")) != -1)
+ "ACD:EF:G:ILM:NO:P:R:S:TUWZa:bc:def:gi:l:m:n:o:p:qr:s:t:u:vw:z:")) != -1)
switch (option) {
+ case 'b': /* batch grouping: {a,b,c} forms a PMU group */
+ args.pa_flags |= FLAG_DO_GROUPING;
+ break;
+
case 'A':
args.pa_flags |= FLAG_SKIP_TOP_FN_RES;
break;
@@ -639,6 +644,30 @@
case 's': /* system-wide counting PMC */
case 'P': /* process virtual sampling PMC */
case 'S': /* system-wide sampling PMC */
+ if ((args.pa_flags & FLAG_DO_GROUPING) != 0 &&
+ optarg != NULL && optarg[0] == '{') {
+ char **siblings = NULL;
+ size_t si, nsib = 0;
+ int gid, rv;
+
+ rv = pmcstat_parse_event_group(optarg,
+ &siblings, &nsib);
+ if (rv < 0)
+ errx(EX_USAGE,
+ "ERROR: Bad group spec \"%s\"",
+ optarg);
+ if (rv == 0) {
+ gid = next_syntactic_gid++;
+ for (si = 0; si < nsib; si++)
+ pmcstat_add_one_event(option,
+ siblings[si], &args, gid,
+ si == 0);
+ pmcstat_free_event_group(siblings,
+ nsib);
+ break;
+ }
+ /* not really a group; fall through */
+ }
caps = 0;
if ((ev = malloc(sizeof(*ev))) == NULL)
errx(EX_SOFTWARE, "ERROR: Out of memory.");
@@ -1139,13 +1168,43 @@
*/
STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
- if (pmc_allocate(ev->ev_spec, ev->ev_mode,
- ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
- ev->ev_count) < 0)
+ int rc;
+
+ /*
+ * In brace-grouping mode (-b) we cannot tell at parse time
+ * whether the combined event count of every group attached
+ * to one target will fit in the available HW counters. Two
+ * groups that each fit on their own can still collectively
+ * over-subscribe the PMU (e.g. 2+3+3 events on a 6-counter
+ * Zen3). The kernel only consults PMC_F_GROUP_MUX on the
+ * leader's allocation and only acts on it when nevents
+ * exceeds the *remaining* HW slot budget at commit time, so
+ * setting it unconditionally on every leader is a no-op for
+ * groups that fit and a graceful fallback for groups that
+ * would otherwise hit ENOSPC at pmc_group_commit.
+ */
+ if (ev->ev_groupid > 0 && ev->ev_is_leader)
+ ev->ev_flags |= PMC_F_GROUP_MUX;
+
+ if (ev->ev_groupid > 0)
+ rc = pmc_allocate_group(ev->ev_spec, ev->ev_mode,
+ ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
+ ev->ev_count);
+ else
+ rc = pmc_allocate(ev->ev_spec, ev->ev_mode,
+ ev->ev_flags, ev->ev_cpu, &ev->ev_pmcid,
+ ev->ev_count);
+ if (rc < 0)
err(EX_OSERR,
"ERROR: Cannot allocate %s-mode pmc with specification \"%s\"",
PMC_IS_SYSTEM_MODE(ev->ev_mode) ?
"system" : "process", ev->ev_spec);
+ if (args.pa_verbosity > 0 || ev->ev_groupid > 0)
+ fprintf(stderr,
+ "pmcstat: alloc spec=\"%s\" groupid=%d "
+ "leader=%d -> pmcid=0x%jx\n",
+ ev->ev_spec, ev->ev_groupid,
+ ev->ev_is_leader, (uintmax_t)ev->ev_pmcid);
if (PMC_IS_SAMPLING_MODE(ev->ev_mode) &&
pmc_set(ev->ev_pmcid, ev->ev_count) < 0)
@@ -1154,6 +1213,40 @@
ev->ev_name);
}
+ if ((args.pa_flags & FLAG_DO_GROUPING) != 0) {
+ struct pmcstat_ev *ev2;
+ int gid_seen;
+ uint32_t real_gid;
+
+ for (gid_seen = 1; gid_seen < next_syntactic_gid;
+ gid_seen++) {
+ real_gid = 0;
+ if (pmc_group_create(&real_gid) < 0)
+ err(EX_OSERR, "ERROR: pmc_group_create");
+ fprintf(stderr,
+ "pmcstat: created kernel gid=%u (syntactic=%d)\n",
+ real_gid, gid_seen);
+ STAILQ_FOREACH(ev2, &args.pa_events, ev_next) {
+ if (ev2->ev_groupid != gid_seen)
+ continue;
+ fprintf(stderr,
+ "pmcstat: group_add gid=%u pmcid=0x%jx "
+ "spec=\"%s\" leader=%d\n",
+ real_gid, (uintmax_t)ev2->ev_pmcid,
+ ev2->ev_spec, ev2->ev_is_leader);
+ if (pmc_group_add(real_gid, ev2->ev_pmcid,
+ ev2->ev_is_leader) < 0)
+ err(EX_OSERR,
+ "ERROR: pmc_group_add gid=%u",
+ real_gid);
+ }
+ if (pmc_group_commit(real_gid) < 0)
+ err(EX_OSERR,
+ "ERROR: pmc_group_commit gid=%u",
+ real_gid);
+ }
+ }
+
/* compute printout widths */
STAILQ_FOREACH(ev, &args.pa_events, ev_next) {
int counter_width;

File Metadata

Mime Type
text/plain
Expires
Thu, Jun 25, 10:46 PM (12 h, 39 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
34295589
Default Alt Text
D57637.diff (5 KB)

Event Timeline