Quoting Tvrtko Ursulin (2018-04-17 09:41:11)
> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
>
> We want to make sure RT tasks which use a lot of CPU times can submit
> batch buffers with roughly the same latency (and certainly not worse)
> compared to normal tasks.
>
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> ---
> tests/gem_exec_latency.c | 176 +++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 176 insertions(+)
>
> diff --git a/tests/gem_exec_latency.c b/tests/gem_exec_latency.c
> index 9498c0921e60..420ede0f83a0 100644
> --- a/tests/gem_exec_latency.c
> +++ b/tests/gem_exec_latency.c
> @@ -36,11 +36,15 @@
> #include <sys/time.h>
> #include <sys/signal.h>
> #include <time.h>
> +#include <sched.h>
>
> #include "drm.h"
>
> #include "igt_sysfs.h"
> #include "igt_vgem.h"
> +#include "igt_dummyload.h"
> +#include "igt_stats.h"
> +
> #include "i915/gem_ring.h"
>
> #define LOCAL_I915_EXEC_NO_RELOC (1<<11)
> @@ -351,6 +355,172 @@ static void latency_from_ring(int fd,
> }
> }
>
> +static void __rearm_spin_batch(igt_spin_t *spin)
> +{
> + const uint32_t mi_arb_chk = 0x5 << 23;
> +
> + *spin->batch = mi_arb_chk;
> + *spin->running = 0;
> + __sync_synchronize();
> +}
> +
> +static void
> +__submit_spin_batch(int fd, igt_spin_t *spin, unsigned int flags)
> +{
> + struct drm_i915_gem_execbuffer2 eb = spin->execbuf;
> +
> + eb.flags &= ~(0x3f | I915_EXEC_BSD_MASK);
> + eb.flags |= flags | I915_EXEC_NO_RELOC;
> +
> + gem_execbuf(fd, &eb);
> +}
> +
> +struct rt_pkt
> +{
> +#define RT_OK (0)
> +#define RT_FAIL (1)
> +#define RT_TIMEOUT (2)
> + int status;
> + struct igt_mean mean;
> + double max;
> +};
> +
> +static void __spin_wait(struct rt_pkt *msg, igt_spin_t *spin, double *t_wait)
> +{
> + struct timespec ts = { };
> + uint64_t t_last = 0;
> +
> + igt_nsec_elapsed(&ts);
> +
> + while (!READ_ONCE(*spin->running)) {
> + uint64_t t = igt_nsec_elapsed(&ts);
> +
> + if ((t - t_last) > 5UL * NSEC_PER_SEC) {
> + /* Absolute timeout to save time. */
> + msg->status = RT_TIMEOUT;
> + } else if ((t - t_last) > NSEC_PER_SEC / 10) {
> + /* Backoff every 100ms to give it chance to complete. */
> + t_last = t;
> + usleep(1);
> + }
> + }
> +
> + *t_wait = igt_nsec_elapsed(&ts) / 1e9;
(I would keep this as nanonseconds in the double until we need to
convert into human readable units.)
> + msg->status = RT_OK;
> +}
> +
> +/*
> + * Test whether RT thread which hogs the CPU a lot can submit work with
> + * reasonable latency.
> + */
> +static void
> +rthog_latency_on_ring(int fd, unsigned int ring, const char *name)
> +{
> + const char *passname[] = { "warmup", "normal", "rt" };
> + struct rt_pkt res[3];
> + unsigned int i;
> + int link[2];
> + int ret;
> +
> + igt_require(gem_can_store_dword(fd, ring));
> +
> + igt_assert(pipe(link) == 0);
> +
> + memset(res, 0, sizeof(res));
> +
> + igt_fork(child, 1) {
> + unsigned int pass = 0; /* Three passes: warmup, normal, rt. */
> +
> + do {
> + struct rt_pkt msg = { };
> + igt_spin_t *spin;
> +
> + igt_mean_init(&msg.mean);
> +
> + if (pass == 2) {
> + struct sched_param rt =
> + { .sched_priority = 99 };
> +
> + ret = sched_setscheduler(0,
> + SCHED_FIFO | SCHED_RESET_ON_FORK,
> + &rt);
> + if (ret) {
> + igt_warn("Failed to set scheduling policy!\n");
> + msg.status = RT_FAIL;
> + write(link[1], &msg, sizeof(msg));
> + exit(1);
> + }
> + }
> +
> + spin = __igt_spin_batch_new_poll(fd, 0, ring);
> + if (!spin) {
> + igt_warn("Failed to create spinner! (%s)\n",
> + passname[pass]);
> + msg.status = RT_FAIL;
> + write(link[1], &msg, sizeof(msg));
> + exit(1);
> + }
> + igt_spin_busywait_until_running(spin);
> +
> + igt_until_timeout(pass > 0 ? 5 : 2) {
> + double t;
> +
> + igt_spin_batch_end(spin);
> + gem_sync(fd, spin->handle);
> +
> + __rearm_spin_batch(spin);
> + __submit_spin_batch(fd, spin, ring);
> +
> + __spin_wait(&msg, spin, &t);
> + if (msg.status != RT_OK) {
> + igt_warn("Wait timeout! (%s)\n",
> + passname[pass]);
> + write(link[1], &msg, sizeof(msg));
> + exit(1);
> + }
> +
> + if (t > msg.max)
> + msg.max = t;
> +
> + igt_mean_add(&msg.mean, t);
I think I would include a background load (normal process, submitting
overlapping nops from alternate contexts, just so that we know that we
should be saturating execlists_submission_tasklet) and then have the RT
process compete with a MAX_PRIO timed request. As I understand the
problem, it is the priority inversion between the RT hog spinning on the
breadcrumb preventing ksoftirqd from running; so the challenge of the
test setup is in tricking ksoftirqd into running on the same CPU as the
hog. I think we need a cpuset to run both RT and background on the same
CPU.
> + }
> +
> + igt_spin_batch_free(fd, spin);
> +
> + igt_info("%10s: mean=%.2fus variance=%.2fus max=%.2fus (n=%lu)\n",
> + passname[pass],
> + igt_mean_get(&msg.mean) * 1e6,
> + igt_mean_get_variance(&msg.mean) * 1e6,
> + msg.max * 1e6,
> + msg.mean.count);
> +
> + write(link[1], &msg, sizeof(msg));
> + } while (++pass < 3);
> +
> + exit(0);
> + }
> +
> + for (i = 0; i < 3; i++) {
> + ret = read(link[0], &res[i], sizeof(res[0]));
> + igt_assert_eq(ret, sizeof(res[0]));
> +
> + igt_assert_eq(res[i].status, RT_OK);
> + }
> +
> + close(link[0]);
> + close(link[1]);
> +
> + igt_waitchildren();
> +
> + /*
> + * Check that the submission latency variance for a task with RT
> + * priority is no larger than three times the same of a normal task.
> + */
> + igt_assert(igt_mean_get_variance(&res[2].mean) <
> + igt_mean_get_variance(&res[1].mean) * 3);
So sample period of 5s, we expect to sample over many many requests, so
mean/variance should be reasonably stable even on a noisy system.
Please make this an assert_f and explain the failure in the error
message. Even a s/2/RT/; s/1/NORMAL/ would help the stringify.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
