* [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo
@ 2026-11-20 14:12 Sidong Yang
2026-11-20 14:12 ` [PATCH v6 1/2] drm/sched: Add drm_sched_entity_get_runtime_ns() Sidong Yang
` (2 more replies)
0 siblings, 3 replies; 4+ messages in thread
From: Sidong Yang @ 2026-11-20 14:12 UTC (permalink / raw)
To: Tomeu Vizoso, Oded Gabbay, Philipp Stanner, Danilo Krummrich,
Matthew Brost, ckoenig.leichtzumerken
Cc: Tvrtko Ursulin, mcanal, Heiko Stuebner, Gyeyoung Baek, dri-devel,
linux-doc, linux-kernel, Sidong Yang
Report how much of the NPU each client uses through fdinfo. Patch 1
exports the runtime the scheduler already accounts per entity, and patch
2 uses it for rocket's drm-engine-npu key.
v6: Rename the getter to drm_sched_entity_get_runtime_ns() (Maíra
Canal). Report rdev->num_cores, the cores that are bound, as the
capacity instead of rdev->max_cores, and skip the key once no core
is left as a capacity of zero is not allowed (Maíra Canal). Add
tags.
v5: Use the entity runtime from the new drm_sched_entity_get_runtime()
instead of a counter in rocket (Tvrtko Ursulin). Reviewed-by tags
are not carried as the accounting changed. Based on drm-misc-next.
v4: Single drm-engine-npu key with drm-engine-capacity-npu, and
documentation (Tomeu Vizoso).
v3: Add a rocket_engine_usage_create() helper (Gyeyoung Baek).
v2: Size the per-core counters by rdev->max_cores.
Link to v5: https://lore.kernel.org/all/20261112171144.3151554-1-sidong.yang@furiosa.ai/
Link to v4: https://lore.kernel.org/all/20261103155338.1949294-1-sidong.yang@furiosa.ai/
Link to v3: https://lore.kernel.org/all/20261103104804.1887489-1-sidong.yang@furiosa.ai/
Link to v2: https://lore.kernel.org/all/20261101152723.1430312-1-sidong.yang@furiosa.ai/
Link to v1: https://lore.kernel.org/all/20261031083642.1076356-1-sidong.yang@furiosa.ai/
Sidong Yang (2):
drm/sched: Add drm_sched_entity_get_runtime_ns()
accel/rocket: Report NPU usage through fdinfo
Documentation/accel/rocket/index.rst | 29 ++++++++++++++++++++++++
Documentation/gpu/drm-usage-stats.rst | 1 +
drivers/accel/rocket/rocket_drv.c | 25 +++++++++++++++++++-
drivers/gpu/drm/scheduler/sched_entity.c | 25 ++++++++++++++++++++
include/drm/gpu_scheduler.h | 1 +
5 files changed, 80 insertions(+), 1 deletion(-)
base-commit: c408466a4dfaee4b12587e926c5b85e68c8b3c40
--
2.53.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v6 1/2] drm/sched: Add drm_sched_entity_get_runtime_ns()
2026-11-20 14:12 [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
@ 2026-11-20 14:12 ` Sidong Yang
2026-11-20 14:12 ` [PATCH v6 2/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
2026-11-20 15:31 ` [PATCH v6 0/2] " Maíra Canal
2 siblings, 0 replies; 4+ messages in thread
From: Sidong Yang @ 2026-11-20 14:12 UTC (permalink / raw)
To: Tomeu Vizoso, Oded Gabbay, Philipp Stanner, Danilo Krummrich,
Matthew Brost, ckoenig.leichtzumerken
Cc: Tvrtko Ursulin, mcanal, Heiko Stuebner, Gyeyoung Baek, dri-devel,
linux-doc, linux-kernel, Sidong Yang, Tvrtko Ursulin
The scheduler accumulates the time each entity's jobs spend on the
hardware for the fair policy, but drivers cannot read it. Export it so
drivers can report per-client engine time through fdinfo without
keeping their own counter.
The stats object is refcounted and held by the entity and its jobs, so
unlike commit df622729ddbf ("drm/scheduler: track GPU active time per
entity"), which was reverted, accounting a job does not touch the
entity.
Suggested-by: Tvrtko Ursulin <tursulin@ursulin.net>
Signed-off-by: Sidong Yang <sidong.yang@furiosa.ai>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Acked-by: Philipp Stanner <phasta@kernel.org>
---
drivers/gpu/drm/scheduler/sched_entity.c | 25 ++++++++++++++++++++++++
include/drm/gpu_scheduler.h | 1 +
2 files changed, 26 insertions(+)
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 03e68b4102d5..9b059914556a 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -238,6 +238,31 @@ int drm_sched_entity_error(struct drm_sched_entity *entity)
}
EXPORT_SYMBOL(drm_sched_entity_error);
+/**
+ * drm_sched_entity_get_runtime_ns - Get the time an entity's jobs have run
+ * @entity: scheduler entity
+ *
+ * Returns the time the jobs of @entity have spent on the hardware, measured
+ * from the scheduled to the finished fence of each job. A job is accounted
+ * when the scheduler frees it.
+ *
+ * The caller must ensure @entity has not been finalized.
+ *
+ * Return: Accumulated runtime in nanoseconds.
+ */
+u64 drm_sched_entity_get_runtime_ns(struct drm_sched_entity *entity)
+{
+ struct drm_sched_entity_stats *stats = entity->stats;
+ ktime_t runtime;
+
+ spin_lock(&stats->lock);
+ runtime = stats->runtime;
+ spin_unlock(&stats->lock);
+
+ return ktime_to_ns(runtime);
+}
+EXPORT_SYMBOL(drm_sched_entity_get_runtime_ns);
+
static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
struct dma_fence_cb *cb);
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 3b0978fa8cf0..ac632e272499 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -730,6 +730,7 @@ void drm_sched_entity_destroy(struct drm_sched_entity *entity);
void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
enum drm_sched_priority priority);
int drm_sched_entity_error(struct drm_sched_entity *entity);
+u64 drm_sched_entity_get_runtime_ns(struct drm_sched_entity *entity);
void drm_sched_entity_modify_sched(struct drm_sched_entity *entity,
struct drm_gpu_scheduler **sched_list,
unsigned int num_sched_list);
--
2.53.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v6 2/2] accel/rocket: Report NPU usage through fdinfo
2026-11-20 14:12 [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
2026-11-20 14:12 ` [PATCH v6 1/2] drm/sched: Add drm_sched_entity_get_runtime_ns() Sidong Yang
@ 2026-11-20 14:12 ` Sidong Yang
2026-11-20 15:31 ` [PATCH v6 0/2] " Maíra Canal
2 siblings, 0 replies; 4+ messages in thread
From: Sidong Yang @ 2026-11-20 14:12 UTC (permalink / raw)
To: Tomeu Vizoso, Oded Gabbay, Philipp Stanner, Danilo Krummrich,
Matthew Brost, ckoenig.leichtzumerken
Cc: Tvrtko Ursulin, mcanal, Heiko Stuebner, Gyeyoung Baek, dri-devel,
linux-doc, linux-kernel, Sidong Yang, Tvrtko Ursulin
There is no way to tell how much of the NPU a client uses. Implement the
DRM usage stats: report the runtime of the client's scheduler entity as
drm-engine-npu, along with drm_show_memory_stats() for the BOs.
The cores are identical, so they are reported as a single engine with
drm-engine-capacity-npu set to rdev->num_cores, the number of cores
bound to the driver.
Document the keys in the rocket driver documentation.
Signed-off-by: Sidong Yang <sidong.yang@furiosa.ai>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
---
Documentation/accel/rocket/index.rst | 29 +++++++++++++++++++++++++++
Documentation/gpu/drm-usage-stats.rst | 1 +
drivers/accel/rocket/rocket_drv.c | 25 ++++++++++++++++++++++-
3 files changed, 54 insertions(+), 1 deletion(-)
diff --git a/Documentation/accel/rocket/index.rst b/Documentation/accel/rocket/index.rst
index 70f97bccf100..d633502c8f15 100644
--- a/Documentation/accel/rocket/index.rst
+++ b/Documentation/accel/rocket/index.rst
@@ -17,3 +17,32 @@ project.
Hardware currently supported:
* RK3588
+
+.. _rocket-usage-stats:
+
+Rocket DRM client usage stats implementation
+============================================
+
+The accel/rocket driver implements the DRM client usage stats specification as
+documented in :ref:`drm-client-usage-stats`.
+
+The NPU cores are reported as a single engine, npu, with a capacity equal to the
+number of cores bound to the driver. The time is summed over all cores and is
+accounted by the DRM GPU scheduler, from when a job is handed to a core until
+its fence signals.
+
+Example of the output showing the implemented key value pairs:
+
+::
+
+ pos: 0
+ flags: 02400002
+ mnt_id: 33
+ ino: 755
+ drm-driver: rocket
+ drm-client-id: 263
+ drm-engine-npu: 2384122043 ns
+ drm-engine-capacity-npu: 3
+ drm-total-memory: 24216 KiB
+ drm-shared-memory: 0
+ drm-active-memory: 3000 KiB
diff --git a/Documentation/gpu/drm-usage-stats.rst b/Documentation/gpu/drm-usage-stats.rst
index 70b7cfcc194f..288764730d23 100644
--- a/Documentation/gpu/drm-usage-stats.rst
+++ b/Documentation/gpu/drm-usage-stats.rst
@@ -218,3 +218,4 @@ Driver specific implementations
* :ref:`panthor-usage-stats`
* :ref:`xe-usage-stats`
* :ref:`amdxdna-usage-stats`
+* :ref:`rocket-usage-stats`
diff --git a/drivers/accel/rocket/rocket_drv.c b/drivers/accel/rocket/rocket_drv.c
index 5d7805c41dc3..b5fd800f500f 100644
--- a/drivers/accel/rocket/rocket_drv.c
+++ b/drivers/accel/rocket/rocket_drv.c
@@ -3,8 +3,10 @@
#include <drm/drm_accel.h>
#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
#include <drm/drm_gem.h>
#include <drm/drm_ioctl.h>
+#include <drm/drm_print.h>
#include <drm/rocket_accel.h>
#include <linux/clk.h>
#include <linux/err.h>
@@ -139,7 +141,27 @@ static const struct drm_ioctl_desc rocket_drm_driver_ioctls[] = {
ROCKET_IOCTL(FINI_BO, fini_bo),
};
-DEFINE_DRM_ACCEL_FOPS(rocket_accel_driver_fops);
+static void rocket_show_fdinfo(struct drm_printer *p, struct drm_file *file)
+{
+ struct rocket_device *rdev = to_rocket_device(file->minor->dev);
+ struct rocket_file_priv *rocket_priv = file->driver_priv;
+ unsigned int num_cores = rdev->num_cores;
+
+ drm_printf(p, "drm-engine-npu:\t%llu ns\n",
+ drm_sched_entity_get_runtime_ns(&rocket_priv->sched_entity));
+
+ /* A capacity of zero is not allowed, skip it once all cores are gone. */
+ if (num_cores)
+ drm_printf(p, "drm-engine-capacity-npu:\t%u\n", num_cores);
+
+ drm_show_memory_stats(p, file);
+}
+
+static const struct file_operations rocket_accel_driver_fops = {
+ .owner = THIS_MODULE,
+ DRM_ACCEL_FOPS,
+ .show_fdinfo = drm_show_fdinfo,
+};
/*
* Rocket driver version:
@@ -150,6 +172,7 @@ static const struct drm_driver rocket_drm_driver = {
.open = rocket_open,
.postclose = rocket_postclose,
.gem_create_object = rocket_gem_create_object,
+ .show_fdinfo = rocket_show_fdinfo,
.ioctls = rocket_drm_driver_ioctls,
.num_ioctls = ARRAY_SIZE(rocket_drm_driver_ioctls),
.fops = &rocket_accel_driver_fops,
--
2.53.0
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo
2026-11-20 14:12 [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
2026-11-20 14:12 ` [PATCH v6 1/2] drm/sched: Add drm_sched_entity_get_runtime_ns() Sidong Yang
2026-11-20 14:12 ` [PATCH v6 2/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
@ 2026-11-20 15:31 ` Maíra Canal
2 siblings, 0 replies; 4+ messages in thread
From: Maíra Canal @ 2026-11-20 15:31 UTC (permalink / raw)
To: Sidong Yang, Tomeu Vizoso, Oded Gabbay, Philipp Stanner,
Danilo Krummrich, Matthew Brost, ckoenig.leichtzumerken
Cc: Tvrtko Ursulin, Heiko Stuebner, Gyeyoung Baek, dri-devel,
linux-doc, linux-kernel
Hi Sidong,
On 20/11/26 11:12, Sidong Yang wrote:
> Report how much of the NPU each client uses through fdinfo. Patch 1
> exports the runtime the scheduler already accounts per entity, and patch
> 2 uses it for rocket's drm-engine-npu key.
>
> v6: Rename the getter to drm_sched_entity_get_runtime_ns() (Maíra
> Canal). Report rdev->num_cores, the cores that are bound, as the
> capacity instead of rdev->max_cores, and skip the key once no core
> is left as a capacity of zero is not allowed (Maíra Canal). Add
> tags.
> v5: Use the entity runtime from the new drm_sched_entity_get_runtime()
> instead of a counter in rocket (Tvrtko Ursulin). Reviewed-by tags
> are not carried as the accounting changed. Based on drm-misc-next.
> v4: Single drm-engine-npu key with drm-engine-capacity-npu, and
> documentation (Tomeu Vizoso).
> v3: Add a rocket_engine_usage_create() helper (Gyeyoung Baek).
> v2: Size the per-core counters by rdev->max_cores.
For both patches,
Reviewed-by: Maíra Canal <mcanal@igalia.com>
Best regards,
- Maíra
>
> Link to v5: https://lore.kernel.org/all/20261112171144.3151554-1-sidong.yang@furiosa.ai/
> Link to v4: https://lore.kernel.org/all/20261103155338.1949294-1-sidong.yang@furiosa.ai/
> Link to v3: https://lore.kernel.org/all/20261103104804.1887489-1-sidong.yang@furiosa.ai/
> Link to v2: https://lore.kernel.org/all/20261101152723.1430312-1-sidong.yang@furiosa.ai/
> Link to v1: https://lore.kernel.org/all/20261031083642.1076356-1-sidong.yang@furiosa.ai/
>
> Sidong Yang (2):
> drm/sched: Add drm_sched_entity_get_runtime_ns()
> accel/rocket: Report NPU usage through fdinfo
>
> Documentation/accel/rocket/index.rst | 29 ++++++++++++++++++++++++
> Documentation/gpu/drm-usage-stats.rst | 1 +
> drivers/accel/rocket/rocket_drv.c | 25 +++++++++++++++++++-
> drivers/gpu/drm/scheduler/sched_entity.c | 25 ++++++++++++++++++++
> include/drm/gpu_scheduler.h | 1 +
> 5 files changed, 80 insertions(+), 1 deletion(-)
>
>
> base-commit: c408466a4dfaee4b12587e926c5b85e68c8b3c40
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-11-20 15:31 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-20 14:12 [PATCH v6 0/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
2026-11-20 14:12 ` [PATCH v6 1/2] drm/sched: Add drm_sched_entity_get_runtime_ns() Sidong Yang
2026-11-20 14:12 ` [PATCH v6 2/2] accel/rocket: Report NPU usage through fdinfo Sidong Yang
2026-11-20 15:31 ` [PATCH v6 0/2] " Maíra Canal
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®