[PATCH v2 0/2] drm/panfrost: Wire cycle counters and timestamp info to userspace

[PATCH v2 0/2] drm/panfrost: Wire cycle counters and timestamp info to userspace

[Mary Guillemard]

Mali has hardware cycle counters and GPU timestamps available for
profiling.

This patch series adds support for cycle counters propagation and
also new timestamp info parameters.

Those new changes to the uAPI will be used in Mesa to implement
timestamp queries for OpenGL and Vulkan.

The Mesa MR using this series is available here [1].

v2:
- Rewrote to use GPU timestamp register
- Add missing include for arch_timer_get_cntfrq
- Squash job requirement uAPI changes and implementation in one patch
- Simplify changes based on Steven Price comments

[1]https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30553

Mary Guillemard (2):
  drm/panfrost: Add SYSTEM_TIMESTAMP and SYSTEM_TIMESTAMP_FREQUENCY
    parameters
  drm/panfrost: Add cycle counter job requirement

 drivers/gpu/drm/panfrost/panfrost_drv.c  | 43 ++++++++++++++++++++++--
 drivers/gpu/drm/panfrost/panfrost_gpu.c  | 12 +++++++
 drivers/gpu/drm/panfrost/panfrost_gpu.h  |  1 +
 drivers/gpu/drm/panfrost/panfrost_job.c  | 28 +++++++++------
 drivers/gpu/drm/panfrost/panfrost_regs.h |  2 ++
 include/uapi/drm/panfrost_drm.h          |  3 ++
 6 files changed, 76 insertions(+), 13 deletions(-)


base-commit: dfa5543193f303a7270ec7c725e656970faf7d57
-- 
2.45.2

[PATCH v2 1/2] drm/panfrost: Add SYSTEM_TIMESTAMP and SYSTEM_TIMESTAMP_FREQUENCY parameters

[Mary Guillemard]

Expose system timestamp and frequency supported by the GPU.

Mali uses an external timer as GPU system time. On ARM, this is wired to
the generic arch timer so we wire cntfrq_el0 as device frequency.

This new uAPI will be used in Mesa to implement timestamp queries and
VK_KHR_calibrated_timestamps.

v2:
- Rewrote to use GPU timestamp register
- Add missing include for arch_timer_get_cntfrq
- Rework commit message

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
---
 drivers/gpu/drm/panfrost/panfrost_drv.c  | 35 ++++++++++++++++++++++++
 drivers/gpu/drm/panfrost/panfrost_gpu.c  | 12 ++++++++
 drivers/gpu/drm/panfrost/panfrost_gpu.h  |  1 +
 drivers/gpu/drm/panfrost/panfrost_regs.h |  2 ++
 include/uapi/drm/panfrost_drm.h          |  2 ++
 5 files changed, 52 insertions(+)

diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
index 671eed4ad890..83696d06d697 100644
--- a/drivers/gpu/drm/panfrost/panfrost_drv.c
+++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
@@ -3,6 +3,10 @@
 /* Copyright 2019 Linaro, Ltd., Rob Herring <robh@kernel.org> */
 /* Copyright 2019 Collabora ltd. */
 
+#ifdef CONFIG_ARM_ARCH_TIMER
+#include <asm/arch_timer.h>
+#endif
+
 #include <linux/module.h>
 #include <linux/of.h>
 #include <linux/pagemap.h>
@@ -24,10 +28,26 @@
 static bool unstable_ioctls;
 module_param_unsafe(unstable_ioctls, bool, 0600);
 
+static int panfrost_ioctl_query_timestamp(struct panfrost_device *pfdev,
+					  u64 *arg)
+{
+	int ret;
+
+	ret = pm_runtime_resume_and_get(pfdev->dev);
+	if (ret)
+		return ret;
+
+	*arg = panfrost_timestamp_read(pfdev);
+
+	pm_runtime_put(pfdev->dev);
+	return 0;
+}
+
 static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct drm_file *file)
 {
 	struct drm_panfrost_get_param *param = data;
 	struct panfrost_device *pfdev = ddev->dev_private;
+	int ret;
 
 	if (param->pad != 0)
 		return -EINVAL;
@@ -69,6 +89,21 @@ static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct
 		PANFROST_FEATURE_ARRAY(JS_FEATURES, js_features, 15);
 		PANFROST_FEATURE(NR_CORE_GROUPS, nr_core_groups);
 		PANFROST_FEATURE(THREAD_TLS_ALLOC, thread_tls_alloc);
+
+	case DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP:
+		ret = panfrost_ioctl_query_timestamp(pfdev, &param->value);
+		if (ret)
+			return ret;
+		break;
+
+	case DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP_FREQUENCY:
+#ifdef CONFIG_ARM_ARCH_TIMER
+		param->value = arch_timer_get_cntfrq();
+#else
+		param->value = 0;
+#endif
+		break;
+
 	default:
 		return -EINVAL;
 	}
diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.c b/drivers/gpu/drm/panfrost/panfrost_gpu.c
index fd8e44992184..92de71725f53 100644
--- a/drivers/gpu/drm/panfrost/panfrost_gpu.c
+++ b/drivers/gpu/drm/panfrost/panfrost_gpu.c
@@ -380,6 +380,18 @@ unsigned long long panfrost_cycle_counter_read(struct panfrost_device *pfdev)
 	return ((u64)hi << 32) | lo;
 }
 
+unsigned long long panfrost_timestamp_read(struct panfrost_device *pfdev)
+{
+	u32 hi, lo;
+
+	panfrost_cycle_counter_get(pfdev);
+	hi = gpu_read(pfdev, GPU_TIMESTAMP_HI);
+	lo = gpu_read(pfdev, GPU_TIMESTAMP_LO);
+	panfrost_cycle_counter_put(pfdev);
+
+	return ((u64)hi << 32) | lo;
+}
+
 static u64 panfrost_get_core_mask(struct panfrost_device *pfdev)
 {
 	u64 core_mask;
diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.h b/drivers/gpu/drm/panfrost/panfrost_gpu.h
index d841b86504ea..b4fef11211d5 100644
--- a/drivers/gpu/drm/panfrost/panfrost_gpu.h
+++ b/drivers/gpu/drm/panfrost/panfrost_gpu.h
@@ -20,6 +20,7 @@ void panfrost_gpu_suspend_irq(struct panfrost_device *pfdev);
 void panfrost_cycle_counter_get(struct panfrost_device *pfdev);
 void panfrost_cycle_counter_put(struct panfrost_device *pfdev);
 unsigned long long panfrost_cycle_counter_read(struct panfrost_device *pfdev);
+unsigned long long panfrost_timestamp_read(struct panfrost_device *pfdev);
 
 void panfrost_gpu_amlogic_quirk(struct panfrost_device *pfdev);
 
diff --git a/drivers/gpu/drm/panfrost/panfrost_regs.h b/drivers/gpu/drm/panfrost/panfrost_regs.h
index c25743b05c55..c7bba476ab3f 100644
--- a/drivers/gpu/drm/panfrost/panfrost_regs.h
+++ b/drivers/gpu/drm/panfrost/panfrost_regs.h
@@ -78,6 +78,8 @@
 
 #define GPU_CYCLE_COUNT_LO		0x90
 #define GPU_CYCLE_COUNT_HI		0x94
+#define GPU_TIMESTAMP_LO		0x98
+#define GPU_TIMESTAMP_HI		0x9C
 
 #define GPU_THREAD_MAX_THREADS		0x0A0	/* (RO) Maximum number of threads per core */
 #define GPU_THREAD_MAX_WORKGROUP_SIZE	0x0A4	/* (RO) Maximum workgroup size */
diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
index 9f231d40a146..52b050e2b660 100644
--- a/include/uapi/drm/panfrost_drm.h
+++ b/include/uapi/drm/panfrost_drm.h
@@ -172,6 +172,8 @@ enum drm_panfrost_param {
 	DRM_PANFROST_PARAM_NR_CORE_GROUPS,
 	DRM_PANFROST_PARAM_THREAD_TLS_ALLOC,
 	DRM_PANFROST_PARAM_AFBC_FEATURES,
+	DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP,
+	DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP_FREQUENCY,
 };
 
 struct drm_panfrost_get_param {
-- 
2.45.2

[PATCH v2 2/2] drm/panfrost: Add cycle counter job requirement

[Mary Guillemard]

Extend the uAPI with a new job requirement flag for cycle
counters. This requirement is used by userland to indicate that a job
requires cycle counters or system timestamp to be propagated. (for use
with write value timestamp jobs)

We cannot enable cycle counters unconditionally as this would result in
an increase of GPU power consumption. As a result, they should be left
off unless required by the application.

If a job requires cycle counters or system timestamps propagation, we
must enable cycle counting before issuing a job and disable it right
after the job completes.

Since this extends the uAPI and because userland needs a way to advertise
features like VK_KHR_shader_clock conditionally, we bumps the driver
minor version.

v2:
- Rework commit message
- Squash uAPI changes and implementation in this commit
- Simplify changes based on Steven Price comments

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
---
 drivers/gpu/drm/panfrost/panfrost_drv.c |  8 +++++--
 drivers/gpu/drm/panfrost/panfrost_job.c | 28 +++++++++++++++----------
 include/uapi/drm/panfrost_drm.h         |  1 +
 3 files changed, 24 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
index 83696d06d697..07a09f32c32e 100644
--- a/drivers/gpu/drm/panfrost/panfrost_drv.c
+++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
@@ -25,6 +25,8 @@
 #include "panfrost_gpu.h"
 #include "panfrost_perfcnt.h"
 
+#define JOB_REQUIREMENTS (PANFROST_JD_REQ_FS | PANFROST_JD_REQ_CYCLE_COUNT)
+
 static bool unstable_ioctls;
 module_param_unsafe(unstable_ioctls, bool, 0600);
 
@@ -280,7 +282,7 @@ static int panfrost_ioctl_submit(struct drm_device *dev, void *data,
 	if (!args->jc)
 		return -EINVAL;
 
-	if (args->requirements && args->requirements != PANFROST_JD_REQ_FS)
+	if (args->requirements & ~JOB_REQUIREMENTS)
 		return -EINVAL;
 
 	if (args->out_sync > 0) {
@@ -619,6 +621,8 @@ static const struct file_operations panfrost_drm_driver_fops = {
  * - 1.0 - initial interface
  * - 1.1 - adds HEAP and NOEXEC flags for CREATE_BO
  * - 1.2 - adds AFBC_FEATURES query
+ * - 1.3 - adds JD_REQ_CYCLE_COUNT job requirement for SUBMIT
+ *       - adds SYSTEM_TIMESTAMP and SYSTEM_TIMESTAMP_FREQUENCY queries
  */
 static const struct drm_driver panfrost_drm_driver = {
 	.driver_features	= DRIVER_RENDER | DRIVER_GEM | DRIVER_SYNCOBJ,
@@ -632,7 +636,7 @@ static const struct drm_driver panfrost_drm_driver = {
 	.desc			= "panfrost DRM",
 	.date			= "20180908",
 	.major			= 1,
-	.minor			= 2,
+	.minor			= 3,
 
 	.gem_create_object	= panfrost_gem_create_object,
 	.gem_prime_import_sg_table = panfrost_gem_prime_import_sg_table,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index df49d37d0e7e..e5e62ee356ef 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -159,16 +159,17 @@ panfrost_dequeue_job(struct panfrost_device *pfdev, int slot)
 	struct panfrost_job *job = pfdev->jobs[slot][0];
 
 	WARN_ON(!job);
-	if (job->is_profiled) {
-		if (job->engine_usage) {
-			job->engine_usage->elapsed_ns[slot] +=
-				ktime_to_ns(ktime_sub(ktime_get(), job->start_time));
-			job->engine_usage->cycles[slot] +=
-				panfrost_cycle_counter_read(pfdev) - job->start_cycles;
-		}
-		panfrost_cycle_counter_put(job->pfdev);
+
+	if (job->is_profiled && job->engine_usage) {
+		job->engine_usage->elapsed_ns[slot] +=
+			ktime_to_ns(ktime_sub(ktime_get(), job->start_time));
+		job->engine_usage->cycles[slot] +=
+			panfrost_cycle_counter_read(pfdev) - job->start_cycles;
 	}
 
+	if (job->requirements & PANFROST_JD_REQ_CYCLE_COUNT || job->is_profiled)
+		panfrost_cycle_counter_put(pfdev);
+
 	pfdev->jobs[slot][0] = pfdev->jobs[slot][1];
 	pfdev->jobs[slot][1] = NULL;
 
@@ -243,9 +244,13 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js)
 	subslot = panfrost_enqueue_job(pfdev, js, job);
 	/* Don't queue the job if a reset is in progress */
 	if (!atomic_read(&pfdev->reset.pending)) {
-		if (pfdev->profile_mode) {
+		job->is_profiled = pfdev->profile_mode;
+
+		if (job->requirements & PANFROST_JD_REQ_CYCLE_COUNT ||
+		    job->is_profiled)
 			panfrost_cycle_counter_get(pfdev);
-			job->is_profiled = true;
+
+		if (job->is_profiled) {
 			job->start_time = ktime_get();
 			job->start_cycles = panfrost_cycle_counter_read(pfdev);
 		}
@@ -693,7 +698,8 @@ panfrost_reset(struct panfrost_device *pfdev,
 	spin_lock(&pfdev->js->job_lock);
 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
 		for (j = 0; j < ARRAY_SIZE(pfdev->jobs[0]) && pfdev->jobs[i][j]; j++) {
-			if (pfdev->jobs[i][j]->is_profiled)
+			if (pfdev->jobs[i][j]->requirements & PANFROST_JD_REQ_CYCLE_COUNT ||
+				pfdev->jobs[i][j]->is_profiled)
 				panfrost_cycle_counter_put(pfdev->jobs[i][j]->pfdev);
 			pm_runtime_put_noidle(pfdev->dev);
 			panfrost_devfreq_record_idle(&pfdev->pfdevfreq);
diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
index 52b050e2b660..568724be6628 100644
--- a/include/uapi/drm/panfrost_drm.h
+++ b/include/uapi/drm/panfrost_drm.h
@@ -40,6 +40,7 @@ extern "C" {
 #define DRM_IOCTL_PANFROST_PERFCNT_DUMP		DRM_IOW(DRM_COMMAND_BASE + DRM_PANFROST_PERFCNT_DUMP, struct drm_panfrost_perfcnt_dump)
 
 #define PANFROST_JD_REQ_FS (1 << 0)
+#define PANFROST_JD_REQ_CYCLE_COUNT (1 << 1)
 /**
  * struct drm_panfrost_submit - ioctl argument for submitting commands to the 3D
  * engine.
-- 
2.45.2

Re: [PATCH v2 1/2] drm/panfrost: Add SYSTEM_TIMESTAMP and SYSTEM_TIMESTAMP_FREQUENCY parameters

[Steven Price]

On 14/08/2024 12:21, Mary Guillemard wrote:
> Expose system timestamp and frequency supported by the GPU.
> 
> Mali uses an external timer as GPU system time. On ARM, this is wired to
> the generic arch timer so we wire cntfrq_el0 as device frequency.
> 
> This new uAPI will be used in Mesa to implement timestamp queries and
> VK_KHR_calibrated_timestamps.
> 
> v2:
> - Rewrote to use GPU timestamp register
> - Add missing include for arch_timer_get_cntfrq
> - Rework commit message
> 
> Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
> ---
>  drivers/gpu/drm/panfrost/panfrost_drv.c  | 35 ++++++++++++++++++++++++
>  drivers/gpu/drm/panfrost/panfrost_gpu.c  | 12 ++++++++
>  drivers/gpu/drm/panfrost/panfrost_gpu.h  |  1 +
>  drivers/gpu/drm/panfrost/panfrost_regs.h |  2 ++
>  include/uapi/drm/panfrost_drm.h          |  2 ++
>  5 files changed, 52 insertions(+)
> 
> diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
> index 671eed4ad890..83696d06d697 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_drv.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
> @@ -3,6 +3,10 @@
>  /* Copyright 2019 Linaro, Ltd., Rob Herring <robh@kernel.org> */
>  /* Copyright 2019 Collabora ltd. */
>  
> +#ifdef CONFIG_ARM_ARCH_TIMER
> +#include <asm/arch_timer.h>
> +#endif
> +
>  #include <linux/module.h>
>  #include <linux/of.h>
>  #include <linux/pagemap.h>
> @@ -24,10 +28,26 @@
>  static bool unstable_ioctls;
>  module_param_unsafe(unstable_ioctls, bool, 0600);
>  
> +static int panfrost_ioctl_query_timestamp(struct panfrost_device *pfdev,
> +					  u64 *arg)
> +{
> +	int ret;
> +
> +	ret = pm_runtime_resume_and_get(pfdev->dev);
> +	if (ret)
> +		return ret;
> +
> +	*arg = panfrost_timestamp_read(pfdev);
> +
> +	pm_runtime_put(pfdev->dev);
> +	return 0;
> +}
> +
>  static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct drm_file *file)
>  {
>  	struct drm_panfrost_get_param *param = data;
>  	struct panfrost_device *pfdev = ddev->dev_private;
> +	int ret;
>  
>  	if (param->pad != 0)
>  		return -EINVAL;
> @@ -69,6 +89,21 @@ static int panfrost_ioctl_get_param(struct drm_device *ddev, void *data, struct
>  		PANFROST_FEATURE_ARRAY(JS_FEATURES, js_features, 15);
>  		PANFROST_FEATURE(NR_CORE_GROUPS, nr_core_groups);
>  		PANFROST_FEATURE(THREAD_TLS_ALLOC, thread_tls_alloc);
> +
> +	case DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP:
> +		ret = panfrost_ioctl_query_timestamp(pfdev, &param->value);
> +		if (ret)
> +			return ret;
> +		break;
> +
> +	case DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP_FREQUENCY:
> +#ifdef CONFIG_ARM_ARCH_TIMER
> +		param->value = arch_timer_get_cntfrq();
> +#else
> +		param->value = 0;
> +#endif
> +		break;
> +
>  	default:
>  		return -EINVAL;
>  	}
> diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.c b/drivers/gpu/drm/panfrost/panfrost_gpu.c
> index fd8e44992184..92de71725f53 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_gpu.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_gpu.c
> @@ -380,6 +380,18 @@ unsigned long long panfrost_cycle_counter_read(struct panfrost_device *pfdev)
>  	return ((u64)hi << 32) | lo;
>  }
>  
> +unsigned long long panfrost_timestamp_read(struct panfrost_device *pfdev)
> +{
> +	u32 hi, lo;
> +
> +	panfrost_cycle_counter_get(pfdev);
> +	hi = gpu_read(pfdev, GPU_TIMESTAMP_HI);
> +	lo = gpu_read(pfdev, GPU_TIMESTAMP_LO);

To get a reliable timestamp you need to use the hi-lo-hi trick:

do {
	hi = gpu_read(pfdev, GPU_TIMESTAMP_HI);
	lo = gpu_read(pfdev, GPU_TIMESTAMP_LO);
} while (hi != gpu_read(pfdev, GPU_TIMESTAMP_HI);

Otherwise if you read around the point of overflowing from LO to HI
you'll get a value which is 2^32 out - see panfrost_cycle_counter_read().

I'm also a little concerned that we have a different API here -
panfrost_cycle_counter_read() assumes that the cycle counter has been
enabled by the caller, but this doesn't. It would be nice to keep these
two functions consistent (i.e. move the get/put into
panfrost_ioctl_query_timestamp()).

Steve

> +	panfrost_cycle_counter_put(pfdev);
> +
> +	return ((u64)hi << 32) | lo;
> +}
> +
>  static u64 panfrost_get_core_mask(struct panfrost_device *pfdev)
>  {
>  	u64 core_mask;
> diff --git a/drivers/gpu/drm/panfrost/panfrost_gpu.h b/drivers/gpu/drm/panfrost/panfrost_gpu.h
> index d841b86504ea..b4fef11211d5 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_gpu.h
> +++ b/drivers/gpu/drm/panfrost/panfrost_gpu.h
> @@ -20,6 +20,7 @@ void panfrost_gpu_suspend_irq(struct panfrost_device *pfdev);
>  void panfrost_cycle_counter_get(struct panfrost_device *pfdev);
>  void panfrost_cycle_counter_put(struct panfrost_device *pfdev);
>  unsigned long long panfrost_cycle_counter_read(struct panfrost_device *pfdev);
> +unsigned long long panfrost_timestamp_read(struct panfrost_device *pfdev);
>  
>  void panfrost_gpu_amlogic_quirk(struct panfrost_device *pfdev);
>  
> diff --git a/drivers/gpu/drm/panfrost/panfrost_regs.h b/drivers/gpu/drm/panfrost/panfrost_regs.h
> index c25743b05c55..c7bba476ab3f 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_regs.h
> +++ b/drivers/gpu/drm/panfrost/panfrost_regs.h
> @@ -78,6 +78,8 @@
>  
>  #define GPU_CYCLE_COUNT_LO		0x90
>  #define GPU_CYCLE_COUNT_HI		0x94
> +#define GPU_TIMESTAMP_LO		0x98
> +#define GPU_TIMESTAMP_HI		0x9C
>  
>  #define GPU_THREAD_MAX_THREADS		0x0A0	/* (RO) Maximum number of threads per core */
>  #define GPU_THREAD_MAX_WORKGROUP_SIZE	0x0A4	/* (RO) Maximum workgroup size */
> diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
> index 9f231d40a146..52b050e2b660 100644
> --- a/include/uapi/drm/panfrost_drm.h
> +++ b/include/uapi/drm/panfrost_drm.h
> @@ -172,6 +172,8 @@ enum drm_panfrost_param {
>  	DRM_PANFROST_PARAM_NR_CORE_GROUPS,
>  	DRM_PANFROST_PARAM_THREAD_TLS_ALLOC,
>  	DRM_PANFROST_PARAM_AFBC_FEATURES,
> +	DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP,
> +	DRM_PANFROST_PARAM_SYSTEM_TIMESTAMP_FREQUENCY,
>  };
>  
>  struct drm_panfrost_get_param {

Re: [PATCH v2 2/2] drm/panfrost: Add cycle counter job requirement

[Steven Price]

On 14/08/2024 12:21, Mary Guillemard wrote:
> Extend the uAPI with a new job requirement flag for cycle
> counters. This requirement is used by userland to indicate that a job
> requires cycle counters or system timestamp to be propagated. (for use
> with write value timestamp jobs)
> 
> We cannot enable cycle counters unconditionally as this would result in
> an increase of GPU power consumption. As a result, they should be left
> off unless required by the application.
> 
> If a job requires cycle counters or system timestamps propagation, we
> must enable cycle counting before issuing a job and disable it right
> after the job completes.
> 
> Since this extends the uAPI and because userland needs a way to advertise
> features like VK_KHR_shader_clock conditionally, we bumps the driver
> minor version.
> 
> v2:
> - Rework commit message
> - Squash uAPI changes and implementation in this commit
> - Simplify changes based on Steven Price comments
> 
> Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>

Reviewed-by: Steven Price <steven.price@arm.com>

Thanks,
Steve

> ---
>  drivers/gpu/drm/panfrost/panfrost_drv.c |  8 +++++--
>  drivers/gpu/drm/panfrost/panfrost_job.c | 28 +++++++++++++++----------
>  include/uapi/drm/panfrost_drm.h         |  1 +
>  3 files changed, 24 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
> index 83696d06d697..07a09f32c32e 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_drv.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
> @@ -25,6 +25,8 @@
>  #include "panfrost_gpu.h"
>  #include "panfrost_perfcnt.h"
>  
> +#define JOB_REQUIREMENTS (PANFROST_JD_REQ_FS | PANFROST_JD_REQ_CYCLE_COUNT)
> +
>  static bool unstable_ioctls;
>  module_param_unsafe(unstable_ioctls, bool, 0600);
>  
> @@ -280,7 +282,7 @@ static int panfrost_ioctl_submit(struct drm_device *dev, void *data,
>  	if (!args->jc)
>  		return -EINVAL;
>  
> -	if (args->requirements && args->requirements != PANFROST_JD_REQ_FS)
> +	if (args->requirements & ~JOB_REQUIREMENTS)
>  		return -EINVAL;
>  
>  	if (args->out_sync > 0) {
> @@ -619,6 +621,8 @@ static const struct file_operations panfrost_drm_driver_fops = {
>   * - 1.0 - initial interface
>   * - 1.1 - adds HEAP and NOEXEC flags for CREATE_BO
>   * - 1.2 - adds AFBC_FEATURES query
> + * - 1.3 - adds JD_REQ_CYCLE_COUNT job requirement for SUBMIT
> + *       - adds SYSTEM_TIMESTAMP and SYSTEM_TIMESTAMP_FREQUENCY queries
>   */
>  static const struct drm_driver panfrost_drm_driver = {
>  	.driver_features	= DRIVER_RENDER | DRIVER_GEM | DRIVER_SYNCOBJ,
> @@ -632,7 +636,7 @@ static const struct drm_driver panfrost_drm_driver = {
>  	.desc			= "panfrost DRM",
>  	.date			= "20180908",
>  	.major			= 1,
> -	.minor			= 2,
> +	.minor			= 3,
>  
>  	.gem_create_object	= panfrost_gem_create_object,
>  	.gem_prime_import_sg_table = panfrost_gem_prime_import_sg_table,
> diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
> index df49d37d0e7e..e5e62ee356ef 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_job.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_job.c
> @@ -159,16 +159,17 @@ panfrost_dequeue_job(struct panfrost_device *pfdev, int slot)
>  	struct panfrost_job *job = pfdev->jobs[slot][0];
>  
>  	WARN_ON(!job);
> -	if (job->is_profiled) {
> -		if (job->engine_usage) {
> -			job->engine_usage->elapsed_ns[slot] +=
> -				ktime_to_ns(ktime_sub(ktime_get(), job->start_time));
> -			job->engine_usage->cycles[slot] +=
> -				panfrost_cycle_counter_read(pfdev) - job->start_cycles;
> -		}
> -		panfrost_cycle_counter_put(job->pfdev);
> +
> +	if (job->is_profiled && job->engine_usage) {
> +		job->engine_usage->elapsed_ns[slot] +=
> +			ktime_to_ns(ktime_sub(ktime_get(), job->start_time));
> +		job->engine_usage->cycles[slot] +=
> +			panfrost_cycle_counter_read(pfdev) - job->start_cycles;
>  	}
>  
> +	if (job->requirements & PANFROST_JD_REQ_CYCLE_COUNT || job->is_profiled)
> +		panfrost_cycle_counter_put(pfdev);
> +
>  	pfdev->jobs[slot][0] = pfdev->jobs[slot][1];
>  	pfdev->jobs[slot][1] = NULL;
>  
> @@ -243,9 +244,13 @@ static void panfrost_job_hw_submit(struct panfrost_job *job, int js)
>  	subslot = panfrost_enqueue_job(pfdev, js, job);
>  	/* Don't queue the job if a reset is in progress */
>  	if (!atomic_read(&pfdev->reset.pending)) {
> -		if (pfdev->profile_mode) {
> +		job->is_profiled = pfdev->profile_mode;
> +
> +		if (job->requirements & PANFROST_JD_REQ_CYCLE_COUNT ||
> +		    job->is_profiled)
>  			panfrost_cycle_counter_get(pfdev);
> -			job->is_profiled = true;
> +
> +		if (job->is_profiled) {
>  			job->start_time = ktime_get();
>  			job->start_cycles = panfrost_cycle_counter_read(pfdev);
>  		}
> @@ -693,7 +698,8 @@ panfrost_reset(struct panfrost_device *pfdev,
>  	spin_lock(&pfdev->js->job_lock);
>  	for (i = 0; i < NUM_JOB_SLOTS; i++) {
>  		for (j = 0; j < ARRAY_SIZE(pfdev->jobs[0]) && pfdev->jobs[i][j]; j++) {
> -			if (pfdev->jobs[i][j]->is_profiled)
> +			if (pfdev->jobs[i][j]->requirements & PANFROST_JD_REQ_CYCLE_COUNT ||
> +				pfdev->jobs[i][j]->is_profiled)
>  				panfrost_cycle_counter_put(pfdev->jobs[i][j]->pfdev);
>  			pm_runtime_put_noidle(pfdev->dev);
>  			panfrost_devfreq_record_idle(&pfdev->pfdevfreq);
> diff --git a/include/uapi/drm/panfrost_drm.h b/include/uapi/drm/panfrost_drm.h
> index 52b050e2b660..568724be6628 100644
> --- a/include/uapi/drm/panfrost_drm.h
> +++ b/include/uapi/drm/panfrost_drm.h
> @@ -40,6 +40,7 @@ extern "C" {
>  #define DRM_IOCTL_PANFROST_PERFCNT_DUMP		DRM_IOW(DRM_COMMAND_BASE + DRM_PANFROST_PERFCNT_DUMP, struct drm_panfrost_perfcnt_dump)
>  
>  #define PANFROST_JD_REQ_FS (1 << 0)
> +#define PANFROST_JD_REQ_CYCLE_COUNT (1 << 1)
>  /**
>   * struct drm_panfrost_submit - ioctl argument for submitting commands to the 3D
>   * engine.