i915: Exercise sysfs heartbeat controls

We [will] expose various per-engine scheduling controls. One of which,
'heartbeat_duration_ms', defines how often we send a heartbeat down the
engine to check upon the health of the engine. If a heartbeat does not
complete within the interval (or two), the engine is declared hung.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>

1 files changed, 446 insertions, 0 deletions

diff --git a/tests/i915/sysfs_heartbeat_interval.c b/tests/i915/sysfs_heartbeat_interval.c
new file mode 100644
index 00000000..0ba19b26
--- /dev/null
+++ b/tests/i915/sysfs_heartbeat_interval.c
@@ -0,0 +1,446 @@
+/*
+ * Copyright © 2019 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include <dirent.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <inttypes.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+
+#include "drmtest.h" /* gem_quiescent_gpu()! */
+#include "i915/gem_engine_topology.h"
+#include "igt_dummyload.h"
+#include "igt_sysfs.h"
+#include "ioctl_wrappers.h" /* igt_require_gem()! */
+#include "sw_sync.h"
+
+#include "igt_debugfs.h"
+
+#define ATTR "heartbeat_interval_ms"
+#define RESET_TIMEOUT 20 /* milliseconds, at least one jiffie for kworker */
+
+static bool __enable_hangcheck(int dir, bool state)
+{
+	return igt_sysfs_set(dir, "enable_hangcheck", state ? "1" : "0");
+}
+
+static void enable_hangcheck(int i915, bool state)
+{
+	int dir;
+
+	dir = igt_sysfs_open_parameters(i915);
+	if (dir < 0) /* no parameters, must be default! */
+		return;
+
+	__enable_hangcheck(dir, state);
+	close(dir);
+}
+
+static void set_heartbeat(int engine, unsigned int value)
+{
+	unsigned int delay = ~value;
+
+	igt_debug("set %s:%d\n", ATTR, value);
+	igt_require(igt_sysfs_printf(engine, ATTR, "%u", value) > 0);
+	igt_sysfs_scanf(engine, ATTR, "%u", &delay);
+	igt_assert_eq(delay, value);
+}
+
+static void test_idempotent(int i915, int engine)
+{
+	unsigned int delays[] = { 1, 1000, 5000, 50000, 123456789 };
+	unsigned int saved;
+
+	/* Quick test that the property reports the values we set */
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+
+	for (int i = 0; i < ARRAY_SIZE(delays); i++)
+		set_heartbeat(engine, delays[i]);
+
+	set_heartbeat(engine, saved);
+}
+
+static void test_invalid(int i915, int engine)
+{
+	unsigned int saved, delay;
+
+	/* Quick test that we reject any unrepresentable intervals */
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+
+	igt_sysfs_printf(engine, ATTR, PRIu64, -1);
+	igt_sysfs_scanf(engine, ATTR, "%u", &delay);
+	igt_assert_eq(delay, saved);
+
+	igt_sysfs_printf(engine, ATTR, PRIu64, 10ull << 32);
+	igt_sysfs_scanf(engine, ATTR, "%u", &delay);
+	igt_assert_eq(delay, saved);
+}
+
+static void set_unbannable(int i915, uint32_t ctx)
+{
+	struct drm_i915_gem_context_param p = {
+		.ctx_id = ctx,
+		.param = I915_CONTEXT_PARAM_BANNABLE,
+	};
+
+	gem_context_set_param(i915, &p);
+}
+
+static uint32_t create_context(int i915, unsigned int class, unsigned int inst, int prio)
+{
+	uint32_t ctx;
+
+	ctx = gem_context_create_for_engine(i915, class, inst);
+	set_unbannable(i915, ctx);
+	gem_context_set_priority(i915, ctx, prio);
+
+	return ctx;
+}
+
+static uint64_t __test_timeout(int i915, int engine, unsigned int timeout)
+{
+	unsigned int class, inst;
+	struct timespec ts = {};
+	igt_spin_t *spin[2];
+	uint64_t elapsed;
+	uint32_t ctx[2];
+
+	igt_assert(igt_sysfs_scanf(engine, "class", "%u", &class) == 1);
+	igt_assert(igt_sysfs_scanf(engine, "instance", "%u", &inst) == 1);
+
+	set_heartbeat(engine, timeout);
+
+	ctx[0] = create_context(i915, class, inst, 1023);
+	spin[0] = igt_spin_new(i915, ctx[0],
+			       .flags = (IGT_SPIN_NO_PREEMPTION |
+					 IGT_SPIN_POLL_RUN |
+					 IGT_SPIN_FENCE_OUT));
+	igt_spin_busywait_until_started(spin[0]);
+
+	ctx[1] = create_context(i915, class, inst, -1023);
+	igt_nsec_elapsed(&ts);
+	spin[1] = igt_spin_new(i915, ctx[1], .flags = IGT_SPIN_POLL_RUN);
+	igt_spin_busywait_until_started(spin[1]);
+	elapsed = igt_nsec_elapsed(&ts);
+
+	igt_spin_free(i915, spin[1]);
+
+	igt_assert_eq(sync_fence_wait(spin[0]->out_fence, RESET_TIMEOUT), 0);
+	igt_assert_eq(sync_fence_status(spin[0]->out_fence), -EIO);
+
+	igt_spin_free(i915, spin[0]);
+
+	gem_context_destroy(i915, ctx[1]);
+	gem_context_destroy(i915, ctx[0]);
+	gem_quiescent_gpu(i915);
+
+	return elapsed;
+}
+
+static void test_precise(int i915, int engine)
+{
+	int delays[] = { 1, 50, 100, 500 };
+	unsigned int saved;
+
+	/*
+	 * The heartbeat interval defines how long the kernel waits between
+	 * checking on the status of the engines. It first sends down a
+	 * heartbeat pulse, waits the interval and sees if the system managed
+	 * to complete the pulse. If not, it gives a priority bump to the pulse
+	 * and waits again. This is repeated until the priority cannot be bumped
+	 * any more, and the system is declared hung.
+	 *
+	 * If we combine the preemptive pulse with forced preemption, we instead
+	 * get a much faster hang detection. Thus in combination we can measure
+	 * the system response time to reseting a hog as a measure of the
+	 * heartbeat interval, and so confirm it matches our specification.
+	 */
+
+	igt_require(igt_sysfs_printf(engine, "preempt_timeout_ms", "%u", 1) == 1);
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+	gem_quiescent_gpu(i915);
+
+	for (int i = 0; i < ARRAY_SIZE(delays); i++) {
+		uint64_t elapsed;
+
+		elapsed = __test_timeout(i915, engine, delays[i]);
+		igt_info("%s:%d, elapsed=%.3fms[%d]\n", ATTR,
+			 delays[i], elapsed * 1e-6,
+			 (int)(elapsed / 1000 / 1000));
+
+		/*
+		 * It takes a couple of missed heartbeats before we start
+		 * terminating hogs, and a little bit of jiffie slack for
+		 * scheduling at each step. 150ms should cover all of our
+		 * sins and be useful tolerance.
+		 */
+		igt_assert_f(elapsed / 1000 / 1000 < 3 * delays[i] + 150,
+			     "Heartbeat interval (and CPR) exceeded request!\n");
+	}
+
+	gem_quiescent_gpu(i915);
+	set_heartbeat(engine, saved);
+}
+
+static void test_nopreempt(int i915, int engine)
+{
+	int delays[] = { 1, 50, 100, 500 };
+	unsigned int saved;
+
+	/*
+	 * The same principle as test_precise(), except that forced preemption
+	 * is disabled (or simply not supported by the platform). This time,
+	 * it waits until the system misses a few heartbeat before doing a
+	 * per-engine/full-gpu reset. As such it is less precise, but we
+	 * can still estimate an upper bound for our specified heartbeat
+	 * interval and verify the system conforms.
+	 */
+
+	/* Test heartbeats with forced preemption  disabled */
+	igt_sysfs_printf(engine, "preempt_timeout_ms", "%u", 0);
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+	gem_quiescent_gpu(i915);
+
+	for (int i = 0; i < ARRAY_SIZE(delays); i++) {
+		uint64_t elapsed;
+
+		elapsed = __test_timeout(i915, engine, delays[i]);
+		igt_info("%s:%d, elapsed=%.3fms[%d]\n", ATTR,
+			 delays[i], elapsed * 1e-6,
+			 (int)(elapsed / 1000 / 1000));
+
+		/*
+		 * It takes a few missed heartbeats before we start
+		 * terminating hogs, and a little bit of jiffie slack for
+		 * scheduling at each step. 250ms should cover all of our
+		 * sins and be useful tolerance.
+		 */
+		igt_assert_f(elapsed / 1000 / 1000 < 5 * delays[i] + 250,
+			     "Heartbeat interval (and CPR) exceeded request!\n");
+	}
+
+	gem_quiescent_gpu(i915);
+	set_heartbeat(engine, saved);
+}
+
+static void client(int i915, int engine, int *ctl, int duration, int expect)
+{
+	unsigned int class, inst;
+	unsigned long count = 0;
+	uint32_t ctx;
+
+	igt_assert(igt_sysfs_scanf(engine, "class", "%u", &class) == 1);
+	igt_assert(igt_sysfs_scanf(engine, "instance", "%u", &inst) == 1);
+
+	ctx = create_context(i915, class, inst, 0);
+
+	while (!READ_ONCE(*ctl)) {
+		igt_spin_t *spin;
+
+		spin = igt_spin_new(i915, ctx,
+				    .flags = (IGT_SPIN_NO_PREEMPTION |
+					      IGT_SPIN_POLL_RUN |
+					      IGT_SPIN_FENCE_OUT));
+		igt_spin_busywait_until_started(spin);
+
+		igt_spin_set_timeout(spin, (uint64_t)duration * 1000 * 1000);
+		sync_fence_wait(spin->out_fence, -1);
+
+		igt_assert_eq(sync_fence_status(spin->out_fence), expect);
+		count++;
+	}
+
+	gem_context_destroy(i915, ctx);
+	igt_info("%s client completed %lu spins\n",
+		 expect < 0 ? "Bad" : "Good", count);
+}
+
+static void __test_mixed(int i915, int engine,
+			 int heartbeat,
+			 int good,
+			 int bad,
+			 int duration)
+{
+	unsigned int saved;
+	int *shared;
+
+	/*
+	 * Given two clients of which one is a hog, be sure we cleanly
+	 * terminate the hog leaving the good client to run.
+	 */
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+	gem_quiescent_gpu(i915);
+
+	shared = mmap(NULL, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
+	igt_assert(shared != MAP_FAILED);
+
+	set_heartbeat(engine, heartbeat);
+
+	igt_fork(child, 1) /* good client */
+		client(i915, engine, shared, good, 1);
+	igt_fork(child, 1) /* bad client */
+		client(i915, engine, shared, bad, -EIO);
+
+	sleep(duration);
+
+	*shared = true;
+	igt_waitchildren();
+	munmap(shared, 4096);
+
+	gem_quiescent_gpu(i915);
+	set_heartbeat(engine, saved);
+}
+
+static void test_mixed(int i915, int engine)
+{
+	/*
+	 * Hogs rarely run alone. Our hang detection must carefully wean
+	 * out the hogs from the innocent clients. Thus we run a mixed workload
+	 * with non-preemptable hogs that exceed the heartbeat, and quicker
+	 * innocents. We inspect the fence status of each to verify that
+	 * only the hogs are reset.
+	 */
+	igt_require(igt_sysfs_printf(engine, "preempt_timeout_ms", "%u", 1) == 1);
+	__test_mixed(i915, engine, 10, 10, 100, 5);
+}
+
+static void test_long(int i915, int engine)
+{
+	/*
+	 * Some clients relish being hogs, and demand that the system
+	 * never do hangchecking. Never is hard to test, so instead we
+	 * run over a day and verify that only the super hogs are reset.
+	 */
+	igt_sysfs_printf(engine, "preempt_timeout_ms", "%u", 0);
+	__test_mixed(i915, engine,
+		     60 * 1000, /* 60s */
+		     60 * 1000, /* 60s */
+		     300 * 1000, /* 5min */
+		     24 * 3600 /* 24hours */);
+}
+
+static void test_off(int i915, int engine)
+{
+	unsigned int class, inst;
+	unsigned int saved;
+	igt_spin_t *spin;
+	uint32_t ctx;
+
+	/*
+	 * Some other clients request that there is never any interruption
+	 * or jitter in their workload and so demand that the kernel never
+	 * sends a heartbeat to steal precious cycles from their workload.
+	 * Turn off the heartbeat and check that the workload is uninterrupted
+	 * for 150s.
+	 */
+
+	igt_assert(igt_sysfs_scanf(engine, ATTR, "%u", &saved) == 1);
+	igt_debug("Initial %s:%u\n", ATTR, saved);
+	gem_quiescent_gpu(i915);
+
+	igt_assert(igt_sysfs_scanf(engine, "class", "%u", &class) == 1);
+	igt_assert(igt_sysfs_scanf(engine, "instance", "%u", &inst) == 1);
+
+	set_heartbeat(engine, 0);
+
+	ctx = create_context(i915, class, inst, 0);
+
+	spin = igt_spin_new(i915, ctx,
+			    .flags = (IGT_SPIN_POLL_RUN |
+				      IGT_SPIN_NO_PREEMPTION |
+				      IGT_SPIN_FENCE_OUT));
+	igt_spin_busywait_until_started(spin);
+
+	for (int i = 0; i < 150; i++) {
+		igt_assert_eq(sync_fence_status(spin->out_fence), 0);
+		sleep(1);
+	}
+
+	set_heartbeat(engine, 1);
+
+	igt_assert_eq(sync_fence_wait(spin->out_fence, 250), 0);
+	igt_assert_eq(sync_fence_status(spin->out_fence), -EIO);
+
+	igt_spin_free(i915, spin);
+
+	gem_quiescent_gpu(i915);
+	set_heartbeat(engine, saved);
+}
+
+igt_main
+{
+	static const struct {
+		const char *name;
+		void (*fn)(int, int);
+	} tests[] = {
+		{ "idempotent", test_idempotent },
+		{ "invalid", test_invalid },
+		{ "precise", test_precise },
+		{ "nopreempt", test_nopreempt },
+		{ "mixed", test_mixed },
+		{ "off", test_off },
+		{ "long", test_long },
+		{ }
+	};
+	int i915 = -1, engines = -1;
+
+	igt_fixture {
+		int sys;
+
+		i915 = drm_open_driver(DRIVER_INTEL);
+		igt_require_gem(i915);
+		igt_allow_hang(i915, 0, 0);
+
+		sys = igt_sysfs_open(i915);
+		igt_require(sys != -1);
+
+		engines = openat(sys, "engine", O_RDONLY);
+		igt_require(engines != -1);
+		close(sys);
+
+		enable_hangcheck(i915, true);
+	}
+
+	for (typeof(*tests) *t = tests; t->name; t++)
+		igt_subtest_with_dynamic(t->name)
+			dyn_sysfs_engines(i915, engines, ATTR, t->fn);
+
+	igt_fixture {
+		close(engines);
+		close(i915);
+	}
+}