[PATCH v2 2/3] drm/msm/gpu: Bypass PM QoS constraint for idle clamp
Rob Clark
robdclark at gmail.com
Tue Jan 10 23:14:43 UTC 2023
From: Rob Clark <robdclark at chromium.org>
Change idle freq clamping back to the direct method, bypassing PM QoS
requests. The problem with using PM QoS requests is they call
(indirectly) the governors ->get_target_freq() which goes thru a
get_dev_status() cycle. The problem comes when the GPU becomes active
again and we remove the idle-clamp request, we go through another
get_dev_status() cycle for the period that the GPU has been idle, which
triggers the governor to lower the target freq excessively.
This partially reverts commit 7c0ffcd40b16 ("drm/msm/gpu: Respect PM QoS
constraints"), but preserves the use of boost QoS request, so that it
will continue to play nicely with other QoS requests such as a cooling
device. This also mostly undoes commit 78f815c1cf8f ("drm/msm: return the
average load over the polling period")
Signed-off-by: Rob Clark <robdclark at chromium.org>
---
drivers/gpu/drm/msm/msm_gpu.h | 12 ++-
drivers/gpu/drm/msm/msm_gpu_devfreq.c | 135 +++++++++++---------------
2 files changed, 65 insertions(+), 82 deletions(-)
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index ab110c377916..fc1c0d8611a8 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -109,11 +109,15 @@ struct msm_gpu_devfreq {
struct mutex lock;
/**
- * idle_constraint:
+ * idle_freq:
*
- * A PM QoS constraint to limit max freq while the GPU is idle.
+ * Shadow frequency used while the GPU is idle. From the PoV of
+ * the devfreq governor, we are continuing to sample busyness and
+ * adjust frequency while the GPU is idle, but we use this shadow
+ * value as the GPU is actually clamped to minimum frequency while
+ * it is inactive.
*/
- struct dev_pm_qos_request idle_freq;
+ unsigned long idle_freq;
/**
* boost_constraint:
@@ -135,8 +139,6 @@ struct msm_gpu_devfreq {
/** idle_time: Time of last transition to idle: */
ktime_t idle_time;
- struct devfreq_dev_status average_status;
-
/**
* idle_work:
*
diff --git a/drivers/gpu/drm/msm/msm_gpu_devfreq.c b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
index 0d7ff7ddc029..e578d74d402f 100644
--- a/drivers/gpu/drm/msm/msm_gpu_devfreq.c
+++ b/drivers/gpu/drm/msm/msm_gpu_devfreq.c
@@ -33,6 +33,16 @@ static int msm_devfreq_target(struct device *dev, unsigned long *freq,
trace_msm_gpu_freq_change(dev_pm_opp_get_freq(opp));
+ /*
+ * If the GPU is idle, devfreq is not aware, so just stash
+ * the new target freq (to use when we return to active)
+ */
+ if (df->idle_freq) {
+ df->idle_freq = *freq;
+ dev_pm_opp_put(opp);
+ return 0;
+ }
+
if (gpu->funcs->gpu_set_freq) {
mutex_lock(&df->lock);
gpu->funcs->gpu_set_freq(gpu, opp, df->suspended);
@@ -48,15 +58,26 @@ static int msm_devfreq_target(struct device *dev, unsigned long *freq,
static unsigned long get_freq(struct msm_gpu *gpu)
{
+ struct msm_gpu_devfreq *df = &gpu->devfreq;
+
+ /*
+ * If the GPU is idle, use the shadow/saved freq to avoid
+ * confusing devfreq (which is unaware that we are switching
+ * to lowest freq until the device is active again)
+ */
+ if (df->idle_freq)
+ return df->idle_freq;
+
if (gpu->funcs->gpu_get_freq)
return gpu->funcs->gpu_get_freq(gpu);
return clk_get_rate(gpu->core_clk);
}
-static void get_raw_dev_status(struct msm_gpu *gpu,
+static int msm_devfreq_get_dev_status(struct device *dev,
struct devfreq_dev_status *status)
{
+ struct msm_gpu *gpu = dev_to_gpu(dev);
struct msm_gpu_devfreq *df = &gpu->devfreq;
u64 busy_cycles, busy_time;
unsigned long sample_rate;
@@ -72,7 +93,7 @@ static void get_raw_dev_status(struct msm_gpu *gpu,
if (df->suspended) {
mutex_unlock(&df->lock);
status->busy_time = 0;
- return;
+ return 0;
}
busy_cycles = gpu->funcs->gpu_busy(gpu, &sample_rate);
@@ -87,71 +108,6 @@ static void get_raw_dev_status(struct msm_gpu *gpu,
busy_time = ~0LU;
status->busy_time = busy_time;
-}
-
-static void update_average_dev_status(struct msm_gpu *gpu,
- const struct devfreq_dev_status *raw)
-{
- struct msm_gpu_devfreq *df = &gpu->devfreq;
- const u32 polling_ms = df->devfreq->profile->polling_ms;
- const u32 max_history_ms = polling_ms * 11 / 10;
- struct devfreq_dev_status *avg = &df->average_status;
- u64 avg_freq;
-
- /* simple_ondemand governor interacts poorly with gpu->clamp_to_idle.
- * When we enforce the constraint on idle, it calls get_dev_status
- * which would normally reset the stats. When we remove the
- * constraint on active, it calls get_dev_status again where busy_time
- * would be 0.
- *
- * To remedy this, we always return the average load over the past
- * polling_ms.
- */
-
- /* raw is longer than polling_ms or avg has no history */
- if (div_u64(raw->total_time, USEC_PER_MSEC) >= polling_ms ||
- !avg->total_time) {
- *avg = *raw;
- return;
- }
-
- /* Truncate the oldest history first.
- *
- * Because we keep the history with a single devfreq_dev_status,
- * rather than a list of devfreq_dev_status, we have to assume freq
- * and load are the same over avg->total_time. We can scale down
- * avg->busy_time and avg->total_time by the same factor to drop
- * history.
- */
- if (div_u64(avg->total_time + raw->total_time, USEC_PER_MSEC) >=
- max_history_ms) {
- const u32 new_total_time = polling_ms * USEC_PER_MSEC -
- raw->total_time;
- avg->busy_time = div_u64(
- mul_u32_u32(avg->busy_time, new_total_time),
- avg->total_time);
- avg->total_time = new_total_time;
- }
-
- /* compute the average freq over avg->total_time + raw->total_time */
- avg_freq = mul_u32_u32(avg->current_frequency, avg->total_time);
- avg_freq += mul_u32_u32(raw->current_frequency, raw->total_time);
- do_div(avg_freq, avg->total_time + raw->total_time);
-
- avg->current_frequency = avg_freq;
- avg->busy_time += raw->busy_time;
- avg->total_time += raw->total_time;
-}
-
-static int msm_devfreq_get_dev_status(struct device *dev,
- struct devfreq_dev_status *status)
-{
- struct msm_gpu *gpu = dev_to_gpu(dev);
- struct devfreq_dev_status raw;
-
- get_raw_dev_status(gpu, &raw);
- update_average_dev_status(gpu, &raw);
- *status = gpu->devfreq.average_status;
return 0;
}
@@ -191,9 +147,6 @@ void msm_devfreq_init(struct msm_gpu *gpu)
mutex_init(&df->lock);
- dev_pm_qos_add_request(&gpu->pdev->dev, &df->idle_freq,
- DEV_PM_QOS_MAX_FREQUENCY,
- PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE);
dev_pm_qos_add_request(&gpu->pdev->dev, &df->boost_freq,
DEV_PM_QOS_MIN_FREQUENCY, 0);
@@ -214,7 +167,6 @@ void msm_devfreq_init(struct msm_gpu *gpu)
if (IS_ERR(df->devfreq)) {
DRM_DEV_ERROR(&gpu->pdev->dev, "Couldn't initialize GPU devfreq\n");
- dev_pm_qos_remove_request(&df->idle_freq);
dev_pm_qos_remove_request(&df->boost_freq);
df->devfreq = NULL;
return;
@@ -256,7 +208,6 @@ void msm_devfreq_cleanup(struct msm_gpu *gpu)
devfreq_cooling_unregister(gpu->cooling);
dev_pm_qos_remove_request(&df->boost_freq);
- dev_pm_qos_remove_request(&df->idle_freq);
}
void msm_devfreq_resume(struct msm_gpu *gpu)
@@ -329,6 +280,7 @@ void msm_devfreq_active(struct msm_gpu *gpu)
{
struct msm_gpu_devfreq *df = &gpu->devfreq;
unsigned int idle_time;
+ unsigned long target_freq;
if (!has_devfreq(gpu))
return;
@@ -338,8 +290,28 @@ void msm_devfreq_active(struct msm_gpu *gpu)
*/
cancel_idle_work(df);
+ /*
+ * Hold devfreq lock to synchronize with get_dev_status()/
+ * target() callbacks
+ */
+ mutex_lock(&df->devfreq->lock);
+
+ target_freq = df->idle_freq;
+
idle_time = ktime_to_ms(ktime_sub(ktime_get(), df->idle_time));
+ df->idle_freq = 0;
+
+ /*
+ * We could have become active again before the idle work had a
+ * chance to run, in which case the df->idle_freq would have
+ * still been zero. In this case, no need to change freq.
+ */
+ if (target_freq)
+ msm_devfreq_target(&gpu->pdev->dev, &target_freq, 0);
+
+ mutex_unlock(&df->devfreq->lock);
+
/*
* If we've been idle for a significant fraction of a polling
* interval, then we won't meet the threshold of busyness for
@@ -348,9 +320,6 @@ void msm_devfreq_active(struct msm_gpu *gpu)
if (idle_time > msm_devfreq_profile.polling_ms) {
msm_devfreq_boost(gpu, 2);
}
-
- dev_pm_qos_update_request(&df->idle_freq,
- PM_QOS_MAX_FREQUENCY_DEFAULT_VALUE);
}
@@ -360,11 +329,23 @@ static void msm_devfreq_idle_work(struct kthread_work *work)
struct msm_gpu_devfreq, idle_work.work);
struct msm_gpu *gpu = container_of(df, struct msm_gpu, devfreq);
struct msm_drm_private *priv = gpu->dev->dev_private;
+ unsigned long idle_freq, target_freq = 0;
- df->idle_time = ktime_get();
+ /*
+ * Hold devfreq lock to synchronize with get_dev_status()/
+ * target() callbacks
+ */
+ mutex_lock(&df->devfreq->lock);
+
+ idle_freq = get_freq(gpu);
if (priv->gpu_clamp_to_idle)
- dev_pm_qos_update_request(&df->idle_freq, 0);
+ msm_devfreq_target(&gpu->pdev->dev, &target_freq, 0);
+
+ df->idle_time = ktime_get();
+ df->idle_freq = idle_freq;
+
+ mutex_unlock(&df->devfreq->lock);
}
void msm_devfreq_idle(struct msm_gpu *gpu)
--
2.38.1
More information about the dri-devel
mailing list