Skip to contentSkip to conversation

Venus - change clk enable, disable order and change bw values

PATCH Superseded Mansur Alisha Shaik 3 participants 9 messages Sep 24, 2020

Subsystems (4)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover Venus - change clk enable, disable order and change bw values

The intention of this patchset is to correct clock enable and disable order and vote for venus-ebi and cpucfg paths with average bandwidth instad of peak bandwidth since with current implementation we are seeing clock related warning during XO-SD and suspend device while video playback

Mansur Alisha Shaik (4):
venus: core: change clk enable and disable order in resume and suspend
venus: core: vote for video-mem path
venus: core: vote with average bandwidth and peak bandwidth as zero
venus: put dummy vote on video-mem path after last session release

 drivers/media/platform/qcom/venus/core.c       | 32 ++++++++++++++++++++------
 drivers/media/platform/qcom/venus/pm_helpers.c | 10 ++++++++
 2 files changed, 35 insertions(+), 7 deletions(-)
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member 
of Code Aurora Forum, hosted by The Linux Foundation
#1 lore raw

PATCH v3 1/4 venus: core: change clk enable and disable order in resume and suspend

Currently video driver is voting after clk enable and un voting before clk disable. This is incorrect, video driver should vote before clk enable and unvote after clk disable.

Corrected this by changing the order of clk enable and clk disable.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")
---
 drivers/media/platform/qcom/venus/core.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
drivers/media/platform/qcom/venus/core.c +10 -7
index 6103aaf..52a3886 100644
@@ -355,13 +355,16 @@ static __maybe_unused int venus_runtime_suspend(struct device *dev)
 	if (ret)
 		return ret;
 
+	if (pm_ops->core_power) {
+		ret = pm_ops->core_power(dev, POWER_OFF);
+		if (ret)
+			return ret;
+	}
+
 	ret = icc_set_bw(core->cpucfg_path, 0, 0);
 	if (ret)
 		return ret;
 
-	if (pm_ops->core_power)
-		ret = pm_ops->core_power(dev, POWER_OFF);
-
 	return ret;
 }
 
@@ -371,16 +374,16 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
 	const struct venus_pm_ops *pm_ops = core->pm_ops;
 	int ret;
 
+	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
+	if (ret)
+		return ret;
+
 	if (pm_ops->core_power) {
 		ret = pm_ops->core_power(dev, POWER_ON);
 		if (ret)
 			return ret;
 	}
 
-	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
-	if (ret)
-		return ret;
-
 	return hfi_core_resume(core, false);
 }
 
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member 
of Code Aurora Forum, hosted by The Linux Foundation
#2 lore raw

Re: [PATCH v3 1/4] venus: core: change clk enable and disable order in resume and suspend

Hi Mansur,

Quoted 7 lines · Mansur Alisha Shaik

Currently video driver is voting after clk enable and un voting before clk disable. This is incorrect, video driver should vote before clk enable and unvote after clk disable.

Corrected this by changing the order of clk enable and clk disable.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

The Fixes tag is incorrect. It should be

07f8f22a33a9e ("media: venus: core: remove CNOC voting while device suspend")

Quoted 52 lines
---
 drivers/media/platform/qcom/venus/core.c | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)
drivers/media/platform/qcom/venus/core.c +10 -7
index 6103aaf..52a3886 100644
@@ -355,13 +355,16 @@ static __maybe_unused int venus_runtime_suspend(struct device *dev)
 	if (ret)
 		return ret;
 
+	if (pm_ops->core_power) {
+		ret = pm_ops->core_power(dev, POWER_OFF);
+		if (ret)
+			return ret;
+	}
+
 	ret = icc_set_bw(core->cpucfg_path, 0, 0);
 	if (ret)
 		return ret;
 
-	if (pm_ops->core_power)
-		ret = pm_ops->core_power(dev, POWER_OFF);
-
 	return ret;
 }
 
@@ -371,16 +374,16 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
 	const struct venus_pm_ops *pm_ops = core->pm_ops;
 	int ret;
 
+	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
+	if (ret)
+		return ret;
+
 	if (pm_ops->core_power) {
 		ret = pm_ops->core_power(dev, POWER_ON);
 		if (ret)
 			return ret;
 	}
 
-	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
-	if (ret)
-		return ret;
-
 	return hfi_core_resume(core, false);
 }
 
Signature · 2 lines
regards,
Stan
#3 lore raw

PATCH v3 2/4 venus: core: vote for video-mem path

In reply to Mansur Alisha Shaik

Currently video driver is voting for venus0-ebi path during buffer processing with an average bandwidth of all the instances and unvoting during session release.

While video streaming when we try to do XO-SD using the command "echo mem > /sys/power/state command" , device is not entering to suspend state and from interconnect summary seeing votes for venus0-ebi

Corrected this by voting for venus0-ebi path in venus_runtime_resume() and unvote during venus_runtime_suspend().

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---
Changes in v3:
- Addressed review comments by Stephen Boyd

 drivers/media/platform/qcom/venus/core.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)
drivers/media/platform/qcom/venus/core.c +16 -1
index 52a3886..fa363b8 100644
@@ -363,7 +363,18 @@ static __maybe_unused int venus_runtime_suspend(struct device *dev)
 
 	ret = icc_set_bw(core->cpucfg_path, 0, 0);
 	if (ret)
-		return ret;
+		goto err_cpucfg_path;
+
+	ret = icc_set_bw(core->video_path, 0, 0);
+	if (ret)
+		goto err_video_path;
+
+	return ret;
+
+err_video_path:
+	icc_set_bw(core->cpucfg_path, kbps_to_icc(1000), 0);
+err_cpucfg_path:
+	pm_ops->core_power(dev, POWER_ON);
 
 	return ret;
 }
@@ -374,6 +385,10 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
 	const struct venus_pm_ops *pm_ops = core->pm_ops;
 	int ret;
 
+	ret = icc_set_bw(core->video_path, 0, kbps_to_icc(1000));
+	if (ret)
+		return ret;
+
 	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
 	if (ret)
 		return ret;
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member 
of Code Aurora Forum, hosted by The Linux Foundation
Reviewed-by on v3 2/4
#4 lore raw

Re: [PATCH v3 2/4] venus: core: vote for video-mem path

Quoting Mansur Alisha Shaik (2020-09-23 23:51:04)

Quoted 8 lines

Currently video driver is voting for venus0-ebi path during buffer processing with an average bandwidth of all the instances and unvoting during session release.

While video streaming when we try to do XO-SD using the command "echo mem > /sys/power/state command" , device is not entering to suspend state and from interconnect summary seeing votes for venus0-ebi

Quoted 8 lines

Corrected this by voting for venus0-ebi path in venus_runtime_resume() and unvote during venus_runtime_suspend().

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---
Changes in v3:
- Addressed review comments by Stephen Boyd

Please Cc me on patches I review.

Quoted 29 lines
 drivers/media/platform/qcom/venus/core.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)
drivers/media/platform/qcom/venus/core.c +12 -1
index 52a3886..fa363b8 100644
@@ -363,7 +363,18 @@ static __maybe_unused int venus_runtime_suspend(struct device *dev)
 
        ret = icc_set_bw(core->cpucfg_path, 0, 0);
        if (ret)
-               return ret;
+               goto err_cpucfg_path;
+
+       ret = icc_set_bw(core->video_path, 0, 0);
+       if (ret)
+               goto err_video_path;
+
+       return ret;
+
+err_video_path:
+       icc_set_bw(core->cpucfg_path, kbps_to_icc(1000), 0);
+err_cpucfg_path:
+       pm_ops->core_power(dev, POWER_ON);
 
        return ret;
 }

drivers/media/platform/qcom/venus/core.c static __maybe_unused int venus_runtime_resume(struct device *dev) · line 391

Quoted 7 lines
drivers/media/platform/qcom/venus/core.c +4 -0
@@ -374,6 +385,10 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
        const struct venus_pm_ops *pm_ops = core->pm_ops;
        int ret;
 
+       ret = icc_set_bw(core->video_path, 0, kbps_to_icc(1000));
+       if (ret)
+               return ret;
+

Feels like this patch should come after the next one but OK.

#5 lore raw

PATCH v3 3/4 venus: core: vote with average bandwidth and peak bandwidth as zero

In reply to Mansur Alisha Shaik

As per bandwidth table video driver is voting with average bandwidth for "video-mem" and "cpu-cfg" paths as peak bandwidth is zero in bandwidth table.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---
Changes in v3:
- Added fixes tag

 drivers/media/platform/qcom/venus/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/core.c +2 -2
index fa363b8..d5bfd6f 100644
@@ -385,11 +385,11 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
 	const struct venus_pm_ops *pm_ops = core->pm_ops;
 	int ret;
 
-	ret = icc_set_bw(core->video_path, 0, kbps_to_icc(1000));
+	ret = icc_set_bw(core->video_path, kbps_to_icc(20000), 0);
 	if (ret)
 		return ret;
 
-	ret = icc_set_bw(core->cpucfg_path, 0, kbps_to_icc(1000));
+	ret = icc_set_bw(core->cpucfg_path, kbps_to_icc(1000), 0);
 	if (ret)
 		return ret;
 
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member 
of Code Aurora Forum, hosted by The Linux Foundation
Reviewed-by on v3 3/4
#6 lore raw

Re: [PATCH v3 3/4] venus: core: vote with average bandwidth and peak bandwidth as zero

Quoting Mansur Alisha Shaik (2020-09-23 23:51:05)

Quoted 18 lines

As per bandwidth table video driver is voting with average bandwidth for "video-mem" and "cpu-cfg" paths as peak bandwidth is zero in bandwidth table.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---
Changes in v3:
- Added fixes tag

 drivers/media/platform/qcom/venus/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/media/platform/qcom/venus/core.c +0 -0
index fa363b8..d5bfd6f 100644

drivers/media/platform/qcom/venus/core.c static __maybe_unused int venus_runtime_resume(struct device *dev) · line 388

Quoted 5 lines
drivers/media/platform/qcom/venus/core.c +1 -1
@@ -385,11 +385,11 @@ static __maybe_unused int venus_runtime_resume(struct device *dev)
        const struct venus_pm_ops *pm_ops = core->pm_ops;
        int ret;
 
-       ret = icc_set_bw(core->video_path, 0, kbps_to_icc(1000));
+       ret = icc_set_bw(core->video_path, kbps_to_icc(20000), 0);

This gets added in the previous patch. Why not put this patch before that one?

Anyway..

#7 lore raw

PATCH v3 4/4 venus: put dummy vote on video-mem path after last session release

In reply to Mansur Alisha Shaik

As per current implementation, video driver is unvoting "videom-mem" path for last video session during vdec_session_release().
While video playback when we try to suspend device, we see video clock warnings since votes are already removed during vdec_session_release().

corrected this by putting dummy vote on "video-mem" after last video session release and unvoting it during suspend.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---
Changes in v3:
- Added fixes tag

 drivers/media/platform/qcom/venus/pm_helpers.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/media/platform/qcom/venus/pm_helpers.c +10 -0
index 57877ea..ca09ea8 100644
@@ -212,6 +212,16 @@ static int load_scale_bw(struct venus_core *core)
 	}
 	mutex_unlock(&core->lock);
 
+	/*
+	 * keep minimum bandwidth vote for "video-mem" path,
+	 * so that clks can be disabled during vdec_session_release().
+	 * Actual bandwidth drop will be done during device supend
+	 * so that device can power down without any warnings.
+	 */
+
+	if (!total_avg && !total_peak)
+		total_avg = kbps_to_icc(1000);
+
 	dev_dbg(core->dev, VDBGL "total: avg_bw: %u, peak_bw: %u\n",
 		total_avg, total_peak);
 
Signature · 2 lines
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member 
of Code Aurora Forum, hosted by The Linux Foundation
Reviewed-by on v3 4/4
#8 lore raw

Re: [PATCH v3 4/4] venus: put dummy vote on video-mem path after last session release

Quoting Mansur Alisha Shaik (2020-09-23 23:51:06)

Quoted 11 lines

As per current implementation, video driver is unvoting "videom-mem" path for last video session during vdec_session_release().
While video playback when we try to suspend device, we see video clock warnings since votes are already removed during vdec_session_release().

corrected this by putting dummy vote on "video-mem" after last video session release and unvoting it during suspend.

Fixes 7482a983d ("media: venus: redesign clocks and pm domains control")

---