[RFC] libperf: Add support for user space counter access

* [RFC] libperf: Add support for user space counter access
@ 2020-08-07 23:05 Rob Herring
  2020-08-08 10:22 ` Jiri Olsa
  2020-08-10 20:24 ` Peter Zijlstra
  0 siblings, 2 replies; 8+ messages in thread
From: Rob Herring @ 2020-08-07 23:05 UTC (permalink / raw)
  To: linux-kernel
  Cc: Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
	Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim

x86 and arm64 can both support direct access of event counters in
userspace. The access sequence is less than trivial and currently exists
in perf test code (tools/perf/arch/x86/tests/rdpmc.c) with copies in
projects such as PAPI and libpfm4.

Patches to add arm64 userspace support are pending[1].

For this RFC, looking for a yes, seems like a good idea, or no, go away we
don't want this in libperf.

TODO:
- Handle userspace access not enabled.
- Handle pmc_width and cap_user_time_short in read loop.
- Move existing rdpmc test to libperf based test?
- Abstract out rdtsc/rdpmc (currently only builds on x86 and need to add
  Arm versions)

[1] https://lore.kernel.org/r/20200707205333.624938-1-robh@kernel.org/

Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Rob Herring <robh@kernel.org>
---
 .../perf/Documentation/examples/user_read.c   | 103 ++++++++++++++++++
 tools/lib/perf/include/perf/mmap.h            |   2 +
 tools/lib/perf/libperf.map                    |   1 +
 tools/lib/perf/mmap.c                         |  72 ++++++++++++
 4 files changed, 178 insertions(+)
 create mode 100644 tools/lib/perf/Documentation/examples/user_read.c

diff --git a/tools/lib/perf/Documentation/examples/user_read.c b/tools/lib/perf/Documentation/examples/user_read.c
new file mode 100644
index 000000000000..47d5f1935861
--- /dev/null
+++ b/tools/lib/perf/Documentation/examples/user_read.c
@@ -0,0 +1,103 @@
+#include <linux/perf_event.h>
+#include <perf/evlist.h>
+#include <perf/evsel.h>
+#include <perf/cpumap.h>
+#include <perf/threadmap.h>
+#include <perf/mmap.h>
+#include <perf/core.h>
+#include <perf/event.h>
+#include <stdio.h>
+#include <unistd.h>
+
+static int libperf_print(enum libperf_print_level level,
+			 const char *fmt, va_list ap)
+{
+	return vfprintf(stderr, fmt, ap);
+}
+
+static unsigned long long read_counter(struct perf_evlist *evlist)
+{
+	struct perf_mmap *map;
+	unsigned long long cnt;
+
+	perf_evlist__for_each_mmap(evlist, map, false) {
+		while ((cnt = perf_mmap__read_self(map)) == ~0ULL) ;
+		return cnt;
+	}
+	return 0;
+}
+
+int main(int argc, char **argv)
+{
+	int count = 100000, err = 0;
+	unsigned long long start, end;
+	struct perf_evlist *evlist;
+	struct perf_evsel *evsel;
+	struct perf_thread_map *threads;
+	struct perf_cpu_map *cpus;
+	struct perf_counts_values counts;
+
+	struct perf_event_attr attr1 = {
+		.type        = PERF_TYPE_HARDWARE,
+		.config      = PERF_COUNT_HW_CPU_CYCLES,
+		.disabled    = 1,
+		.exclude_kernel = 1,
+	};
+
+	libperf_init(libperf_print);
+
+	evlist = perf_evlist__new();
+	if (!evlist) {
+		fprintf(stderr, "failed to create evlist\n");
+		goto out_threads;
+	}
+	evsel = perf_evsel__new(&attr1);
+	if (!evsel) {
+		fprintf(stderr, "failed to create evsel1\n");
+		goto out_evlist;
+	}
+	perf_evlist__add(evlist, evsel);
+
+	threads = perf_thread_map__new_dummy();
+	if (!threads) {
+		fprintf(stderr, "failed to create threads\n");
+		return -1;
+	}
+	perf_thread_map__set_pid(threads, 0, 0);
+	cpus = perf_cpu_map__dummy_new();
+	if (!cpus) {
+		fprintf(stderr, "failed to create cpus\n");
+		return -1;
+	}
+	perf_evlist__set_maps(evlist, cpus, threads);
+
+	err = perf_evlist__open(evlist);
+	if (err) {
+		fprintf(stderr, "failed to open evsel\n");
+		goto out_evlist;
+	}
+
+	err = perf_evlist__mmap(evlist, 0);
+	if (err) {
+		fprintf(stderr, "failed to mmap evlist\n");
+		goto out_evlist;
+	}
+
+	perf_evlist__enable(evlist);
+
+	start = read_counter(evlist);
+	while (count--);
+	end = read_counter(evlist);
+
+	perf_evlist__disable(evlist);
+
+	fprintf(stdout, "start %llu, end %llu, delta %llu\n",
+			start, end, end-start);
+
+	perf_evlist__close(evlist);
+out_evlist:
+	perf_evlist__delete(evlist);
+out_threads:
+	perf_thread_map__put(threads);
+	return err;
+}
diff --git a/tools/lib/perf/include/perf/mmap.h b/tools/lib/perf/include/perf/mmap.h
index 9508ad90d8b9..2d0a7e8b13db 100644
--- a/tools/lib/perf/include/perf/mmap.h
+++ b/tools/lib/perf/include/perf/mmap.h
@@ -12,4 +12,6 @@ LIBPERF_API int perf_mmap__read_init(struct perf_mmap *map);
 LIBPERF_API void perf_mmap__read_done(struct perf_mmap *map);
 LIBPERF_API union perf_event *perf_mmap__read_event(struct perf_mmap *map);
 
+LIBPERF_API __u64 perf_mmap__read_self(struct perf_mmap *map);
+
 #endif /* __LIBPERF_MMAP_H */
diff --git a/tools/lib/perf/libperf.map b/tools/lib/perf/libperf.map
index 7be1af8a546c..676a73300add 100644
--- a/tools/lib/perf/libperf.map
+++ b/tools/lib/perf/libperf.map
@@ -46,6 +46,7 @@ LIBPERF_0.0.1 {
 		perf_mmap__read_init;
 		perf_mmap__read_done;
 		perf_mmap__read_event;
+		perf_mmap__read_self;
 	local:
 		*;
 };
diff --git a/tools/lib/perf/mmap.c b/tools/lib/perf/mmap.c
index 79d5ed6c38cc..5a167618f4c5 100644
--- a/tools/lib/perf/mmap.c
+++ b/tools/lib/perf/mmap.c
@@ -273,3 +273,75 @@ union perf_event *perf_mmap__read_event(struct perf_mmap *map)
 
 	return event;
 }
+
+static u64 rdpmc(unsigned int counter)
+{
+	unsigned int low, high;
+
+	asm volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (counter));
+
+	return low | ((u64)high) << 32;
+}
+
+static u64 rdtsc(void)
+{
+	unsigned int low, high;
+
+	asm volatile("rdtsc" : "=a" (low), "=d" (high));
+
+	return low | ((u64)high) << 32;
+}
+
+__u64 perf_mmap__read_self(struct perf_mmap *map)
+{
+	struct perf_event_mmap_page *pc = map->base;
+	u32 seq, idx, time_mult = 0, time_shift = 0;
+	u64 count, cyc = 0, time_offset = 0, enabled, running, delta;
+
+	/*
+	 * Check if event was unmapped due to a POLLHUP/POLLERR.
+	 */
+	if (!refcount_read(&map->refcnt))
+		return ~0;
+
+	do {
+		seq = pc->lock;
+		barrier();
+
+		enabled = pc->time_enabled;
+		running = pc->time_running;
+
+		if (enabled != running) {
+			cyc = rdtsc();
+			time_mult = pc->time_mult;
+			time_shift = pc->time_shift;
+			time_offset = pc->time_offset;
+		}
+
+		idx = pc->index;
+		count = pc->offset;
+		if (idx)
+			count += rdpmc(idx - 1);
+
+		barrier();
+	} while (pc->lock != seq);
+
+	if (enabled != running) {
+		u64 quot, rem;
+
+		quot = (cyc >> time_shift);
+		rem = cyc & (((u64)1 << time_shift) - 1);
+		delta = time_offset + quot * time_mult +
+			((rem * time_mult) >> time_shift);
+
+		enabled += delta;
+		if (idx)
+			running += delta;
+
+		quot = count / running;
+		rem = count % running;
+		count = quot * enabled + (rem * enabled) / running;
+	}
+
+	return count;
+}
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 8+ messages in thread