Re: [PATCH igt v2] tests: Add a random load generator

[Mika Kuoppala <mika.kuoppala@linux.intel.com>]

Chris Wilson <chris@chris-wilson.co.uk> writes:

> Apply a random load to one or all engines in order to apply stress to
> RPS as it tries to constantly adjust the GPU frequency to meet the
> changing workload.
>
> This can be used to reproduce the byt (j1900) system hang.
>
> v2: igt_until_timeout
>

References: https://bugzilla.kernel.org/show_bug.cgi?id=109051

> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
> ---
>  tests/Makefile.sources |   1 +
>  tests/gem_exec_load.c  | 176 +++++++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 177 insertions(+)
>  create mode 100644 tests/gem_exec_load.c
>
> diff --git a/tests/Makefile.sources b/tests/Makefile.sources
> index 870c90935..8b5a18680 100644
> --- a/tests/Makefile.sources
> +++ b/tests/Makefile.sources
> @@ -79,6 +79,7 @@ TESTS_progs = \
>  	gem_exec_gttfill \
>  	gem_exec_latency \
>  	gem_exec_lut_handle \
> +	gem_exec_load \
>  	gem_exec_nop \
>  	gem_exec_parallel \
>  	gem_exec_params \
> diff --git a/tests/gem_exec_load.c b/tests/gem_exec_load.c
> new file mode 100644
> index 000000000..ba88466a9
> --- /dev/null
> +++ b/tests/gem_exec_load.c
> @@ -0,0 +1,176 @@
> +/*
> + * Copyright © 2018 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 "igt.h"
> +#include "igt_rand.h"
> +#include "igt_sysfs.h"
> +
> +IGT_TEST_DESCRIPTION("Apply a random load to each engine");
> +
> +static inline uint32_t randmax(uint32_t *state, uint32_t ep_ro)
> +{
> +	return (uint64_t)hars_petruska_f54_1_random(state) * ep_ro >> 32;
> +}
> +
> +static void one(int fd, unsigned int engine, int scale, unsigned int timeout)
> +{
> +	uint32_t prng = engine;
> +	igt_spin_t *spin[2] = {};
> +	unsigned int max;
> +	int sysfs;
> +
> +	sysfs = igt_sysfs_open(fd, NULL);
> +
> +	max = sysfs < 0 ? 1 : igt_sysfs_get_u32(sysfs, "gt_max_freq_mhz");
> +
> +	igt_until_timeout(timeout) {
> +		/*
> +		 * For every second, we randomly assign a desire % busyness
> +		 * and the frequency with which to submit a batch, defining
> +		 * a pulse-width modulation (pwm).
> +		 */
> +		unsigned int pwm = randmax(&prng, 1000);
> +		unsigned int load = randmax(&prng, pwm / scale);
> +		const unsigned int interval = 1e6 / pwm;
> +		unsigned int cur =
> +			max > 1 ? igt_sysfs_get_u32(sysfs, "gt_cur_freq_mhz") : 1;
> +
> +		while (pwm--) {
> +			if (randmax(&prng, pwm * cur) <= load * max) {
> +				spin[1] = __igt_spin_batch_new(fd, 0, engine, 0);
> +				load--;
> +			}
> +			igt_spin_batch_free(fd, spin[0]);
> +			spin[0] = spin[1];
> +			spin[1] = NULL;
> +
> +			usleep(interval);
> +		}
> +	}
> +	igt_spin_batch_free(fd, spin[0]);
> +
> +	close(sysfs);
> +}
> +
> +static void all(int fd, unsigned int timeout)
> +{
> +	unsigned int engines[16];
> +	unsigned int nengine;
> +	unsigned int engine;
> +
> +	nengine = 0;
> +	for_each_engine(fd, engine)
> +		engines[nengine++] = engine;
> +	igt_require(nengine > 1);
> +
> +	for (unsigned int n = 0; n < nengine; n++)
> +		igt_fork(child, 1)
> +			one(fd, engines[n], nengine/2, timeout);

nengine/2 a little unexpected. You want a more load on
multiengine tests?

> +	igt_waitchildren();
> +}
> +
> +static void pulse(int fd, unsigned int timeout)
> +{
> +	const uint32_t bbe = MI_BATCH_BUFFER_END;
> +	struct drm_i915_gem_exec_object2 obj[2] = {
> +		{ .flags = EXEC_OBJECT_WRITE },
> +	};
> +	unsigned int engines[16];
> +	unsigned int nengine;
> +	unsigned int engine;
> +
> +	nengine = 0;
> +	for_each_engine(fd, engine) {
> +		if (engine == 0)
> +			continue;
> +
> +		engines[nengine++] = engine;
> +	}
> +	igt_require(nengine);
> +
> +	obj[1].handle = gem_create(fd, 4096);
> +	gem_write(fd, obj[1].handle, 0, &bbe, sizeof(bbe));
> +
> +	igt_until_timeout(timeout) {
> +		for (unsigned int s = 0; s < nengine; s++) {
> +			int64_t boost = 10e6; /* 10ms, long enough to sleep */

..long enough to let cpu sleep

> +			igt_spin_t *spin;
> +
> +			spin = __igt_spin_batch_new(fd, 0, engines[s], 0);
> +			obj[0].handle = spin->handle;
> +			for (unsigned int n = 0; n < nengine; n++) {
> +				struct drm_i915_gem_execbuffer2 eb = {
> +					.buffer_count = 2,
> +					.buffers_ptr = to_user_pointer(obj),
> +					.flags = engines[n],
> +				};
> +				for (unsigned int repeat = 0;
> +				     repeat < 64;
> +				     repeat++)
> +					gem_execbuf(fd, &eb);

As discussed in irc, there is not builtin randomness here. The
question is:  should there be? Regardless, this test did manage
to system hang j1900 so there is great value in here,
which should be preserved. Perhaps by casting this pattern to bronze
and doing improvements on another subtest until it proves
to be more effective on hanging :)

Thanks for the effort to make a more deterministic way
of bringing this nasty bugger closer to daylight.

Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>

> +			}
> +			gem_wait(fd, spin->handle, &boost);
> +			igt_spin_batch_free(fd, spin);
> +
> +			gem_quiescent_gpu(fd);
> +		}
> +	}
> +
> +	gem_close(fd, obj[1].handle);
> +}
> +
> +igt_main
> +{
> +	const struct intel_execution_engine *e;
> +	int fd = -1;
> +
> +	igt_skip_on_simulation();
> +
> +	igt_fixture {
> +		fd = drm_open_driver(DRIVER_INTEL);
> +		igt_require_gem(fd);
> +		igt_fork_hang_detector(fd);
> +	}
> +
> +	for (e = intel_execution_engines; e->name; e++) {
> +		/* default exec-id is purely symbolic */
> +		if (e->exec_id == 0)
> +			continue;
> +
> +		igt_subtest_f("%s", e->name) {
> +			gem_require_ring(fd, e->exec_id | e->flags);
> +			one(fd, e->exec_id | e->flags, 1, 150);
> +		}
> +	}
> +
> +	igt_subtest("all")
> +		all(fd, 900);
> +
> +	igt_subtest("pulse")
> +		pulse(fd, 900);
> +
> +	igt_fixture {
> +		igt_stop_hang_detector();
> +		close(fd);
> +	}
> +}
> -- 
> 2.15.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx