[PATCH 1/2 v2] perf stat: Rework x86 transaction counter handling

[PATCH 1/2 v2] perf stat: Rework x86 transaction counter handling

[Thomas Richter]

Command 'perf stat -T -- sleep 2' displays transaction
counters.  Right now there is only hard coded support
for x86.

This patch introduces architecture specific counter
tables for x86. The architecture is queried
and the event string for transaction counters is constructed
depending on the architecture and the CPU measurement
facility counter list. This scheme can be easily extended
to support other architectures.

Output does not change.

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
---
 tools/perf/builtin-stat.c | 130 ++++++++++++++++++++++++++++++++++++----------
 1 file changed, 102 insertions(+), 28 deletions(-)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 86c8c8a9229c..8d7b97c708e4 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -95,22 +95,8 @@ static const char *transaction_attrs = {
 	"task-clock,"
 	"{"
 	"instructions,"
-	"cycles,"
-	"cpu/cycles-t/,"
-	"cpu/tx-start/,"
-	"cpu/el-start/,"
-	"cpu/cycles-ct/"
-	"}"
-};
-
-/* More limited version when the CPU does not have all events. */
-static const char * transaction_limited_attrs = {
-	"task-clock,"
-	"{"
-	"instructions,"
-	"cycles,"
-	"cpu/cycles-t/,"
-	"cpu/tx-start/"
+	"cycles"
+	"%s"
 	"}"
 };
 
@@ -2149,13 +2135,93 @@ __weak void arch_topdown_group_warn(void)
 {
 }
 
+struct pmu_tx_events {			/* Define transaction counters */
+	const char *pmu;		/* PMU name */
+	const char *name;		/* Counter name */
+};
+
+static struct pmu_tx_events x86_tx_events[] = {/* x86 transaction counters */
+	{
+		.pmu = "cpu",
+		.name = "cycles-t",
+	},
+	{
+		.pmu = "cpu",
+		.name = "tx-start",
+	},
+	{
+		.pmu = "cpu",
+		.name = "el-start",
+	},
+	{
+		.pmu = "cpu",
+		.name = "cycles-ct",
+	},
+	{
+		.pmu = 0
+	}
+};
+
+struct arch_pmu_tx_events {
+	const char *archname;		/* Architecture name */
+	struct pmu_tx_events *tx;	/* Architecture specific counters */
+} arch_pmu_tx_events[] = {
+	{
+		.archname = "x86",
+		.tx = x86_tx_events
+	},
+	{
+		.archname = 0
+	}
+};
+
+static struct pmu_tx_events *pmu_tx_find_arch(const char *name)
+{
+	struct arch_pmu_tx_events *p = arch_pmu_tx_events;
+
+	for (; p->archname; ++p)
+		if (!strcmp(name, p->archname))
+			return p->tx;
+	return NULL;
+}
+
+/* Search the list of transaction events and test if they are valid for
+ * this PMU. The events are named cpu/cycles-t/ or cpum_cf/TX_NC_TEND/
+ * that is the PMU name followed by the event name surrounded by slashes.
+ *
+ * The function returns a string which contains the tested (and existing)
+ * PMU transaction events. The caller must free this string.
+ *
+ * If no PMU transaction events are found, a NULL pointer is returned.
+ */
+static char *build_tx_string(const char *fmt, struct pmu_tx_events *txp)
+{
+	struct pmu_tx_events *p = txp;
+	char buffer[512], *result;
+	int rc, i = 0;
+
+	memset(buffer, 0, sizeof(buffer));
+	for (p = txp; p->pmu; ++p) {
+		if (pmu_have_event(p->pmu, p->name)) {
+			rc = scnprintf(buffer + i, sizeof(buffer) - i,
+				       ",%s/%s/", p->pmu, p->name);
+			i += rc;
+			if (i >= (int)sizeof(buffer))
+				break;
+		}
+	}
+	if (i)			/* Transaction counters found */
+		return asprintf(&result, fmt, buffer) < 0 ? NULL : result;
+	return NULL;
+}
+
 /*
  * Add default attributes, if there were no attributes specified or
  * if -d/--detailed, -d -d or -d -d -d is used:
  */
 static int add_default_attributes(void)
 {
-	int err;
+	int err = -1;
 	struct perf_event_attr default_attrs0[] = {
 
   { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK		},
@@ -2274,21 +2340,29 @@ static int add_default_attributes(void)
 		return 0;
 
 	if (transaction_run) {
-		struct parse_events_error errinfo;
+		struct parse_events_error errinfo = { .idx = 0 };
+		char *tx_str;
+		struct pmu_tx_events *tx_archp;
 
-		if (pmu_have_event("cpu", "cycles-ct") &&
-		    pmu_have_event("cpu", "el-start"))
-			err = parse_events(evsel_list, transaction_attrs,
-					   &errinfo);
-		else
-			err = parse_events(evsel_list,
-					   transaction_limited_attrs,
-					   &errinfo);
-		if (err) {
+		/* Find architecture transaction counter string table */
+		tx_archp = pmu_tx_find_arch(perf_env__arch(NULL));
+		if (!tx_archp) {
 			fprintf(stderr, "Cannot set up transaction events\n");
 			return -1;
 		}
-		return 0;
+
+		/* Build architecture transaction counter string */
+		tx_str = build_tx_string(transaction_attrs, tx_archp);
+		if (!tx_str) {
+			fprintf(stderr, "Cannot set up transaction events\n");
+			return -1;
+		}
+
+		err = parse_events(evsel_list, tx_str, &errinfo);
+		if (err)
+			parse_events_print_error(&errinfo, tx_str);
+		free(tx_str);
+		return err;
 	}
 
 	if (smi_cost) {
-- 
2.14.3

[PATCH 2/2 v2] perf stat: Add support for flag T on s390

[Thomas Richter]

This patch adds support for command 'perf stat -T' on s390.
It adds s390 specfic counter table for transactions.
The architecture is queried and the event string for
transaction counters is constructed
depending on the s390 type and model CPU measurement
facility counter list.

Output Before:
[root@s35lp76 perf]# perf stat -T -- sleep 1
Cannot set up transaction events
[root@s35lp76 perf]#

Output after:
[root@s35lp76 perf]# ./perf stat -T -- sleep 1

 Performance counter stats for 'sleep 1':

   0.939985      task-clock (msec)         #    0.001 CPUs utilized
  2,557,145      instructions              #    0.53  insn per cycle
  4,785,929      cycles                    #    5.091 GHz
          0      cpum_cf/TX_C_TABORT_NO_SPECIAL/ #    0.000 K/sec
          0      cpum_cf/TX_C_TABORT_SPECIAL/ #    0.000 K/sec
          0      cpum_cf/TX_C_TEND/        #    0.000 K/sec
          0      cpum_cf/TX_NC_TABORT/     #    0.000 K/sec
          0      cpum_cf/TX_NC_TEND/       #    0.000 K/sec

  1.001934710 seconds time elapsed

[root@s35lp76 perf]#

Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com>
Reviewed-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
---
 tools/perf/builtin-stat.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 8d7b97c708e4..943ed3648049 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -2162,10 +2162,40 @@ static struct pmu_tx_events x86_tx_events[] = {/* x86 transaction counters */
 	}
 };
 
+static struct pmu_tx_events s390_tx_events[] = {/* s390 transaction counters */
+	{
+		.pmu = "cpum_cf",
+		.name = "TX_C_TABORT_NO_SPECIAL",
+	},
+	{
+		.pmu = "cpum_cf",
+		.name = "TX_C_TABORT_SPECIAL",
+	},
+	{
+		.pmu = "cpum_cf",
+		.name = "TX_C_TEND",
+	},
+	{
+		.pmu = "cpum_cf",
+		.name = "TX_NC_TABORT",
+	},
+	{
+		.pmu = "cpum_cf",
+		.name = "TX_NC_TEND",
+	},
+	{
+		.pmu = 0
+	}
+};
+
 struct arch_pmu_tx_events {
 	const char *archname;		/* Architecture name */
 	struct pmu_tx_events *tx;	/* Architecture specific counters */
 } arch_pmu_tx_events[] = {
+	{
+		.archname = "s390",
+		.tx = s390_tx_events
+	},
 	{
 		.archname = "x86",
 		.tx = x86_tx_events
-- 
2.14.3