perf mem/c2c amd: Add ldlat support

'perf mem/c2c' uses IBS Op PMU on AMD platforms.

IBS Op PMU on Zen5 uarch has added support for Load Latency filtering.

Implement 'perf mem/c2c' --ldlat using IBS Op Load Latency filtering
capability.

Some subtle differences between AMD and other arch:

o --ldlat is disabled by default on AMD

o Supported values are 128 to 2048.

Signed-off-by: Ravi Bangoria <[e-mail]>
Cc: Ananth Narayan <[e-mail]>
Cc: Ian Rogers <[e-mail]>
Cc: Ingo Molnar <[e-mail]>
Cc: Jiri Olsa <[e-mail]>
Cc: Joe Mario <[e-mail]>
Cc: Kan Liang <[e-mail]>
Cc: Namhyung Kim <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Cc: Sandipan Das <[e-mail]>
Cc: Santosh Shukla <[e-mail]>
Cc: Stephane Eranian <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Arnaldo Carvalho de Melo <[e-mail]>

tools/perf/Documentation/perf-c2c.txt

···
 
 -l::
 --ldlat::
-	Configure mem-loads latency. Supported on Intel and Arm64 processors
-	only. Ignored on other archs.
+	Configure mem-loads latency. Supported on Intel, Arm64 and some AMD
+	processors. Ignored on other archs.
+
+	On supported AMD processors:
+	- /sys/bus/event_source/devices/ibs_op/caps/ldlat file contains '1'.
+	- Supported latency values are 128 to 2048 (both inclusive).
+	- Latency value which is a multiple of 128 incurs a little less profiling
+	  overhead compared to other values.
+	- Load latency filtering is disabled by default.
 
 -k::
 --all-kernel::

tools/perf/Documentation/perf-mem.txt

···
 Due to the statistical nature of SPE sampling, not every memory operation will
 be sampled.
 
+On AMD this use IBS Op PMU to sample load-store operations.
+
 COMMON OPTIONS
 --------------
 -f::
···
 	Configure all used events to run in user space.
 
 --ldlat <n>::
-	Specify desired latency for loads event. Supported on Intel and Arm64
-	processors only. Ignored on other archs.
+	Specify desired latency for loads event. Supported on Intel, Arm64 and
+	some AMD processors. Ignored on other archs.
+
+	On supported AMD processors:
+	- /sys/bus/event_source/devices/ibs_op/caps/ldlat file contains '1'.
+	- Supported latency values are 128 to 2048 (both inclusive).
+	- Latency value which is a multiple of 128 incurs a little less profiling
+	  overhead compared to other values.
+	- Load latency filtering is disabled by default.
 
 REPORT OPTIONS
 --------------

tools/perf/arch/x86/util/mem-events.c

···
 	E(NULL,		NULL,		NULL,	false,	0),
 	E("mem-ldst",	"%s//",		NULL,	false,	0),
 };
+
+struct perf_mem_event perf_mem_events_amd_ldlat[PERF_MEM_EVENTS__MAX] = {
+	E(NULL,		NULL,		NULL,	false,	0),
+	E(NULL,		NULL,		NULL,	false,	0),
+	E("mem-ldst",	"%s/ldlat=%u/",	NULL,	true,	0),
+};

tools/perf/arch/x86/util/mem-events.h

···
 extern struct perf_mem_event perf_mem_events_intel_aux[PERF_MEM_EVENTS__MAX];
 
 extern struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX];
+extern struct perf_mem_event perf_mem_events_amd_ldlat[PERF_MEM_EVENTS__MAX];
 
 #endif /* _X86_MEM_EVENTS_H */

tools/perf/arch/x86/util/pmu.c

···
 #include "mem-events.h"
 #include "util/env.h"
 
-void perf_pmu__arch_init(struct perf_pmu *pmu __maybe_unused)
+void perf_pmu__arch_init(struct perf_pmu *pmu)
 {
+	struct perf_pmu_caps *ldlat_cap;
+
 #ifdef HAVE_AUXTRACE_SUPPORT
 	if (!strcmp(pmu->name, INTEL_PT_PMU_NAME)) {
 		pmu->auxtrace = true;
···
 #endif
 
 	if (x86__is_amd_cpu()) {
-		if (!strcmp(pmu->name, "ibs_op"))
-			pmu->mem_events = perf_mem_events_amd;
+		if (strcmp(pmu->name, "ibs_op"))
+			return;
+
+		pmu->mem_events = perf_mem_events_amd;
+
+		if (!perf_pmu__caps_parse(pmu))
+			return;
+
+		ldlat_cap = perf_pmu__get_cap(pmu, "ldlat");
+		if (!ldlat_cap || strcmp(ldlat_cap->value, "1"))
+			return;
+
+		perf_mem_events__loads_ldlat = 0;
+		pmu->mem_events = perf_mem_events_amd_ldlat;
 	} else if (pmu->is_core) {
 		if (perf_pmu__have_event(pmu, "mem-loads-aux"))
 			pmu->mem_events = perf_mem_events_intel_aux;

tools/perf/tests/shell/test_data_symbol.sh

···
 
 echo "Recording workload..."
 
-# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't support
-# user/kernel filtering and per-process monitoring, spin program on
-# specific CPU and test in per-CPU mode.
 is_amd=$(grep -E -c 'vendor_id.*AuthenticAMD' /proc/cpuinfo)
 if (($is_amd >= 1)); then
+	mem_events="$(perf mem record -v -e list 2>&1)"
+	if ! [[ "$mem_events" =~ ^mem\-ldst.*ibs_op/(.*)/.*available ]]; then
+		echo "ERROR: mem-ldst event is not matching"
+		exit 1
+	fi
+
+	# --ldlat on AMD:
+	# o Zen4 and earlier uarch does not support ldlat
+	# o Even on supported platforms, it's disabled (--ldlat=0) by default.
+	ldlat=${BASH_REMATCH[1]}
+	if [[ -n $ldlat ]]; then
+		if ! [[ "$ldlat" =~ ldlat=0 ]]; then
+			echo "ERROR: ldlat not initialized to 0?"
+			exit 1
+		fi
+
+		mem_events="$(perf mem record -v --ldlat=150 -e list 2>&1)"
+		if ! [[ "$mem_events" =~ ^mem-ldst.*ibs_op/ldlat=150/.*available ]]; then
+			echo "ERROR: --ldlat not honored?"
+			exit 1
+		fi
+	fi
+
+	# perf mem/c2c internally uses IBS PMU on AMD CPU which doesn't
+	# support user/kernel filtering and per-process monitoring on older
+	# kernels, spin program on specific CPU and test in per-CPU mode.
 	perf mem record -vvv -o ${PERF_DATA} -C 0 -- taskset -c 0 $TEST_PROGRAM 2>"${ERR_FILE}"
 else
 	perf mem record -vvv --all-user -o ${PERF_DATA} -- $TEST_PROGRAM 2>"${ERR_FILE}"

tools/perf/util/pmu.c

···
 	}
 }
 
+struct perf_pmu_caps *perf_pmu__get_cap(struct perf_pmu *pmu, const char *name)
+{
+	struct perf_pmu_caps *caps;
+
+	list_for_each_entry(caps, &pmu->caps, list) {
+		if (!strcmp(caps->name, name))
+			return caps;
+	}
+	return NULL;
+}
+
 /*
  * Reading/parsing the given pmu capabilities, which should be located at:
  * /sys/bus/event_source/devices/<dev>/caps as sysfs group attributes.

tools/perf/util/pmu.h

···
 
 int perf_pmu__convert_scale(const char *scale, char **end, double *sval);
 
+struct perf_pmu_caps *perf_pmu__get_cap(struct perf_pmu *pmu, const char *name);
+
 int perf_pmu__caps_parse(struct perf_pmu *pmu);
 
 void perf_pmu__warn_invalid_config(struct perf_pmu *pmu, __u64 config,