From 0d8385a7ad670e96dadef6e04e7541a64b637406 Mon Sep 17 00:00:00 2001 From: Tvrtko Ursulin Date: Wed, 13 Sep 2017 18:28:24 +0100 Subject: intel-gpu-overlay: Catch-up to new i915 PMU v2: Update for i915 changes. v3: Use 1eN for large numbers. (Chris Wilson) v4: Update for upstream engine class enum. Signed-off-by: Tvrtko Ursulin Reviewed-by: Chris Wilson --- overlay/gem-interrupts.c | 2 +- overlay/gpu-freq.c | 8 +++--- overlay/gpu-top.c | 68 +++++++++++++++++++++++++++--------------------- overlay/power.c | 4 +-- overlay/rc6.c | 20 +++++++------- 5 files changed, 54 insertions(+), 48 deletions(-) (limited to 'overlay') diff --git a/overlay/gem-interrupts.c b/overlay/gem-interrupts.c index 5bd8656e..0233fbb0 100644 --- a/overlay/gem-interrupts.c +++ b/overlay/gem-interrupts.c @@ -113,7 +113,7 @@ int gem_interrupts_init(struct gem_interrupts *irqs) { memset(irqs, 0, sizeof(*irqs)); - irqs->fd = perf_i915_open(I915_PERF_INTERRUPTS); + irqs->fd = perf_i915_open(I915_PMU_INTERRUPTS); if (irqs->fd < 0 && interrupts_read() < 0) irqs->error = ENODEV; diff --git a/overlay/gpu-freq.c b/overlay/gpu-freq.c index 76c5ed9a..0d803259 100644 --- a/overlay/gpu-freq.c +++ b/overlay/gpu-freq.c @@ -37,8 +37,8 @@ static int perf_open(void) { int fd; - fd = perf_i915_open_group(I915_PERF_ACTUAL_FREQUENCY, -1); - if (perf_i915_open_group(I915_PERF_REQUESTED_FREQUENCY, fd) < 0) { + fd = perf_i915_open_group(I915_PMU_ACTUAL_FREQUENCY, -1); + if (perf_i915_open_group(I915_PMU_REQUESTED_FREQUENCY, fd) < 0) { close(fd); fd = -1; } @@ -176,8 +176,8 @@ int gpu_freq_update(struct gpu_freq *gf) return EAGAIN; } - gf->current = (s->act - d->act) / d_time; - gf->request = (s->req - d->req) / d_time; + gf->current = (s->act - d->act) * 1e9 / d_time; + gf->request = (s->req - d->req) * 1e9 / d_time; } return 0; diff --git a/overlay/gpu-top.c b/overlay/gpu-top.c index 812f47d5..61b8f62f 100644 --- a/overlay/gpu-top.c +++ b/overlay/gpu-top.c @@ -43,49 +43,57 @@ #define RING_WAIT (1<<11) #define RING_WAIT_SEMAPHORE (1<<10) -#define __I915_PERF_RING(n) (4*n) -#define I915_PERF_RING_BUSY(n) (__I915_PERF_RING(n) + 0) -#define I915_PERF_RING_WAIT(n) (__I915_PERF_RING(n) + 1) -#define I915_PERF_RING_SEMA(n) (__I915_PERF_RING(n) + 2) - static int perf_init(struct gpu_top *gt) { - const char *names[] = { - "RCS", - "BCS", - "VCS0", - "VCS1", - NULL, + struct engine_desc { + unsigned class, inst; + const char *name; + } *d, engines[] = { + { I915_ENGINE_CLASS_RENDER, 0, "rcs0" }, + { I915_ENGINE_CLASS_COPY, 0, "bcs0" }, + { I915_ENGINE_CLASS_VIDEO, 0, "vcs0" }, + { I915_ENGINE_CLASS_VIDEO, 1, "vcs1" }, + { I915_ENGINE_CLASS_VIDEO_ENHANCE, 0, "vecs0" }, + { 0, 0, NULL } }; - int n; - gt->fd = perf_i915_open_group(I915_PERF_RING_BUSY(0), -1); + d = &engines[0]; + + gt->fd = perf_i915_open_group(I915_PMU_ENGINE_BUSY(d->class, d->inst), + -1); if (gt->fd < 0) return -1; - if (perf_i915_open_group(I915_PERF_RING_WAIT(0), gt->fd) >= 0) + if (perf_i915_open_group(I915_PMU_ENGINE_WAIT(d->class, d->inst), + gt->fd) >= 0) gt->have_wait = 1; - if (perf_i915_open_group(I915_PERF_RING_SEMA(0), gt->fd) >= 0) + if (perf_i915_open_group(I915_PMU_ENGINE_SEMA(d->class, d->inst), + gt->fd) >= 0) gt->have_sema = 1; - gt->ring[0].name = names[0]; + gt->ring[0].name = d->name; gt->num_rings = 1; - for (n = 1; names[n]; n++) { - if (perf_i915_open_group(I915_PERF_RING_BUSY(n), gt->fd) >= 0) { - if (gt->have_wait && - perf_i915_open_group(I915_PERF_RING_WAIT(n), - gt->fd) < 0) - return -1; - - if (gt->have_sema && - perf_i915_open_group(I915_PERF_RING_SEMA(n), - gt->fd) < 0) - return -1; - - gt->ring[gt->num_rings++].name = names[n]; - } + for (d++; d->name; d++) { + if (perf_i915_open_group(I915_PMU_ENGINE_BUSY(d->class, + d->inst), + gt->fd) < 0) + continue; + + if (gt->have_wait && + perf_i915_open_group(I915_PMU_ENGINE_WAIT(d->class, + d->inst), + gt->fd) < 0) + return -1; + + if (gt->have_sema && + perf_i915_open_group(I915_PMU_ENGINE_SEMA(d->class, + d->inst), + gt->fd) < 0) + return -1; + + gt->ring[gt->num_rings++].name = d->name; } return 0; diff --git a/overlay/power.c b/overlay/power.c index dd4aec6b..805f4ca7 100644 --- a/overlay/power.c +++ b/overlay/power.c @@ -45,9 +45,7 @@ int power_init(struct power *power) memset(power, 0, sizeof(*power)); - power->fd = perf_i915_open(I915_PERF_ENERGY); - if (power->fd != -1) - return 0; + power->fd = -1; sprintf(buf, "%s/i915_energy_uJ", debugfs_dri_path); fd = open(buf, 0); diff --git a/overlay/rc6.c b/overlay/rc6.c index 46c975a5..8977f099 100644 --- a/overlay/rc6.c +++ b/overlay/rc6.c @@ -43,15 +43,15 @@ static int perf_open(unsigned *flags) { int fd; - fd = perf_i915_open_group(I915_PERF_RC6_RESIDENCY, -1); + fd = perf_i915_open_group(I915_PMU_RC6_RESIDENCY, -1); if (fd < 0) return -1; *flags |= RC6; - if (perf_i915_open_group(I915_PERF_RC6p_RESIDENCY, fd) >= 0) + if (perf_i915_open_group(I915_PMU_RC6p_RESIDENCY, fd) >= 0) *flags |= RC6p; - if (perf_i915_open_group(I915_PERF_RC6pp_RESIDENCY, fd) >= 0) + if (perf_i915_open_group(I915_PMU_RC6pp_RESIDENCY, fd) >= 0) *flags |= RC6pp; return fd; @@ -132,11 +132,11 @@ int rc6_update(struct rc6 *rc6) len = 2; if (rc6->flags & RC6) - s->rc6_residency = data[len++]; + s->rc6_residency = data[len++] / 1e6; if (rc6->flags & RC6p) - s->rc6p_residency = data[len++]; + s->rc6p_residency = data[len++] / 1e6; if (rc6->flags & RC6pp) - s->rc6pp_residency = data[len++]; + s->rc6pp_residency = data[len++] / 1e6; } if (rc6->count == 1) @@ -149,14 +149,14 @@ int rc6_update(struct rc6 *rc6) } d_rc6 = s->rc6_residency - d->rc6_residency; - rc6->rc6 = (100 * d_rc6 + d_time/2) / d_time; + rc6->rc6 = 100 * d_rc6 / d_time; d_rc6p = s->rc6p_residency - d->rc6p_residency; - rc6->rc6p = (100 * d_rc6p + d_time/2) / d_time; + rc6->rc6p = 100 * d_rc6p / d_time; d_rc6pp = s->rc6pp_residency - d->rc6pp_residency; - rc6->rc6pp = (100 * d_rc6pp + d_time/2) / d_time; + rc6->rc6pp = 100 * d_rc6pp / d_time; - rc6->rc6_combined = (100 * (d_rc6 + d_rc6p + d_rc6pp) + d_time/2) / d_time; + rc6->rc6_combined = 100 * (d_rc6 + d_rc6p + d_rc6pp) / d_time; return 0; } -- cgit v1.2.3