[PATCH 0/3] accel: ethosu: Runtime PM refcounting and cmd stream validation fixes

[PATCH 0/3] accel: ethosu: Runtime PM refcounting and cmd stream validation fixes

[Rob Herring (Arm)]

This is series of fixes for issues I found in testing additional models 
and adding more supported operations in mesa.

Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
Rob Herring (Arm) (3):
      accel: ethosu: Fix job submit error clean-up refcount underflows
      accel: ethosu: Fix NPU_OP_ELEMENTWISE validation with scalar
      accel: ethosu: Handle possible underflow in IFM size calculations

 drivers/accel/ethosu/ethosu_gem.c | 12 +++++++++---
 drivers/accel/ethosu/ethosu_job.c | 26 ++++++++++++++++++--------
 2 files changed, 27 insertions(+), 11 deletions(-)
---
base-commit: 05f7e89ab9731565d8a62e3b5d1ec206485eeb0b
change-id: 20260218-ethos-fixes-e3508a579582

Best regards,
--  
Rob Herring (Arm) <robh@kernel.org>

[PATCH 1/3] accel: ethosu: Fix job submit error clean-up refcount underflows

[Rob Herring (Arm)]

If the job submit fails before adding the job to the scheduler queue
such as when the GEM buffer bounds checks fail, then doing a
ethosu_job_put() results in a pm_runtime_put_autosuspend() without the
corresponding pm_runtime_resume_and_get(). The dma_fence_put()'s are
also unnecessary, but seem to be harmless.

Split the ethosu_job_cleanup() function into 2 parts for the before
and after the job is queued.

Fixes: 5a5e9c0228e6 ("accel: Add Arm Ethos-U NPU driver")
Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
 drivers/accel/ethosu/ethosu_job.c | 26 ++++++++++++++++++--------
 1 file changed, 18 insertions(+), 8 deletions(-)

diff --git a/drivers/accel/ethosu/ethosu_job.c b/drivers/accel/ethosu/ethosu_job.c
index 26e7a2f64d71..70a144803b09 100644
--- a/drivers/accel/ethosu/ethosu_job.c
+++ b/drivers/accel/ethosu/ethosu_job.c
@@ -143,23 +143,29 @@ static int ethosu_job_push(struct ethosu_job *job)
 	return ret;
 }
 
+static void ethosu_job_err_cleanup(struct ethosu_job *job)
+{
+	unsigned int i;
+
+	for (i = 0; i < job->region_cnt; i++)
+		drm_gem_object_put(job->region_bo[i]);
+
+	drm_gem_object_put(job->cmd_bo);
+
+	kfree(job);
+}
+
 static void ethosu_job_cleanup(struct kref *ref)
 {
 	struct ethosu_job *job = container_of(ref, struct ethosu_job,
 						refcount);
-	unsigned int i;
 
 	pm_runtime_put_autosuspend(job->dev->base.dev);
 
 	dma_fence_put(job->done_fence);
 	dma_fence_put(job->inference_done_fence);
 
-	for (i = 0; i < job->region_cnt; i++)
-		drm_gem_object_put(job->region_bo[i]);
-
-	drm_gem_object_put(job->cmd_bo);
-
-	kfree(job);
+	ethosu_job_err_cleanup(job);
 }
 
 static void ethosu_job_put(struct ethosu_job *job)
@@ -454,12 +460,16 @@ static int ethosu_ioctl_submit_job(struct drm_device *dev, struct drm_file *file
 		}
 	}
 	ret = ethosu_job_push(ejob);
+	if (!ret) {
+		ethosu_job_put(ejob);
+		return 0;
+	}
 
 out_cleanup_job:
 	if (ret)
 		drm_sched_job_cleanup(&ejob->base);
 out_put_job:
-	ethosu_job_put(ejob);
+	ethosu_job_err_cleanup(ejob);
 
 	return ret;
 }

-- 
2.51.0

[PATCH 2/3] accel: ethosu: Fix NPU_OP_ELEMENTWISE validation with scalar

[Rob Herring (Arm)]

The NPU_OP_ELEMENTWISE instruction uses a scalar value for IFM2 if the
IFM2_BROADCAST "scalar" mode is set. It is a bit (7) on the u65 and
part of a field (bits 3:0) on the u85. The driver was hardcoded to the
u85.

Fixes: 5a5e9c0228e6 ("accel: Add Arm Ethos-U NPU driver")
Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
 drivers/accel/ethosu/ethosu_gem.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/accel/ethosu/ethosu_gem.c b/drivers/accel/ethosu/ethosu_gem.c
index 473b5f5d7514..a735f860a119 100644
--- a/drivers/accel/ethosu/ethosu_gem.c
+++ b/drivers/accel/ethosu/ethosu_gem.c
@@ -417,7 +417,10 @@ static int ethosu_gem_cmdstream_copy_and_validate(struct drm_device *ddev,
 				return ret;
 			break;
 		case NPU_OP_ELEMENTWISE:
-			use_ifm2 = !((st.ifm2.broadcast == 8) || (param == 5) ||
+			use_scale = ethosu_is_u65(edev) ?
+				    (st.ifm2.broadcast & 0x80) :
+				    (st.ifm2.broadcast == 8);
+			use_ifm2 = !(use_scale || (param == 5) ||
 				(param == 6) || (param == 7) || (param == 0x24));
 			use_ifm = st.ifm.broadcast != 8;
 			ret = calc_sizes_elemwise(ddev, info, cmd, &st, use_ifm, use_ifm2);

-- 
2.51.0

[PATCH 3/3] accel: ethosu: Handle possible underflow in IFM size calculations

[Rob Herring (Arm)]

If the command stream has larger padding sizes than the IFM and OFM
diminsions, then the calculations will underflow to a negative value.
The result is a very large region bounds which is caught on submit, but
it's better to catch it earlier.

Current mesa ethosu driver has a signedness bug which resulted in
padding of 127 (the max) and triggers this issue.

Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
---
 drivers/accel/ethosu/ethosu_gem.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/accel/ethosu/ethosu_gem.c b/drivers/accel/ethosu/ethosu_gem.c
index a735f860a119..d1169001c83d 100644
--- a/drivers/accel/ethosu/ethosu_gem.c
+++ b/drivers/accel/ethosu/ethosu_gem.c
@@ -245,11 +245,14 @@ static int calc_sizes(struct drm_device *ddev,
 			((st->ifm.stride_kernel >> 1) & 0x1) + 1;
 		u32 stride_x = ((st->ifm.stride_kernel >> 5) & 0x2) +
 			(st->ifm.stride_kernel & 0x1) + 1;
-		u32 ifm_height = st->ofm.height[2] * stride_y +
+		s32 ifm_height = st->ofm.height[2] * stride_y +
 			st->ifm.height[2] - (st->ifm.pad_top + st->ifm.pad_bottom);
-		u32 ifm_width  = st->ofm.width * stride_x +
+		s32 ifm_width = st->ofm.width * stride_x +
 			st->ifm.width - (st->ifm.pad_left + st->ifm.pad_right);
 
+		if (ifm_height < 0 || ifm_width < 0)
+			return -EINVAL;
+
 		len = feat_matrix_length(info, &st->ifm, ifm_width,
 					 ifm_height, st->ifm.depth);
 		dev_dbg(ddev->dev, "op %d: IFM:%d:0x%llx-0x%llx\n",

-- 
2.51.0

Re: [PATCH 1/3] accel: ethosu: Fix job submit error clean-up refcount underflows

[Anders Roxell]

On Wed, 18 Feb 2026 at 23:22, Rob Herring (Arm) <robh@kernel.org> wrote:
>
> If the job submit fails before adding the job to the scheduler queue
> such as when the GEM buffer bounds checks fail, then doing a
> ethosu_job_put() results in a pm_runtime_put_autosuspend() without the
> corresponding pm_runtime_resume_and_get(). The dma_fence_put()'s are
> also unnecessary, but seem to be harmless.
>
> Split the ethosu_job_cleanup() function into 2 parts for the before
> and after the job is queued.
>
> Fixes: 5a5e9c0228e6 ("accel: Add Arm Ethos-U NPU driver")
> Signed-off-by: Rob Herring (Arm) <robh@kernel.org>

Reviewed-and-Tested-by: Anders Roxell <anders.roxell@linaro.org>

> ---
>  drivers/accel/ethosu/ethosu_job.c | 26 ++++++++++++++++++--------
>  1 file changed, 18 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/accel/ethosu/ethosu_job.c b/drivers/accel/ethosu/ethosu_job.c
> index 26e7a2f64d71..70a144803b09 100644
> --- a/drivers/accel/ethosu/ethosu_job.c
> +++ b/drivers/accel/ethosu/ethosu_job.c
> @@ -143,23 +143,29 @@ static int ethosu_job_push(struct ethosu_job *job)
>         return ret;
>  }
>
> +static void ethosu_job_err_cleanup(struct ethosu_job *job)
> +{
> +       unsigned int i;
> +
> +       for (i = 0; i < job->region_cnt; i++)
> +               drm_gem_object_put(job->region_bo[i]);
> +
> +       drm_gem_object_put(job->cmd_bo);
> +
> +       kfree(job);
> +}
> +
>  static void ethosu_job_cleanup(struct kref *ref)
>  {
>         struct ethosu_job *job = container_of(ref, struct ethosu_job,
>                                                 refcount);
> -       unsigned int i;
>
>         pm_runtime_put_autosuspend(job->dev->base.dev);
>
>         dma_fence_put(job->done_fence);
>         dma_fence_put(job->inference_done_fence);
>
> -       for (i = 0; i < job->region_cnt; i++)
> -               drm_gem_object_put(job->region_bo[i]);
> -
> -       drm_gem_object_put(job->cmd_bo);
> -
> -       kfree(job);
> +       ethosu_job_err_cleanup(job);
>  }
>
>  static void ethosu_job_put(struct ethosu_job *job)
> @@ -454,12 +460,16 @@ static int ethosu_ioctl_submit_job(struct drm_device *dev, struct drm_file *file
>                 }
>         }
>         ret = ethosu_job_push(ejob);
> +       if (!ret) {
> +               ethosu_job_put(ejob);
> +               return 0;
> +       }
>
>  out_cleanup_job:
>         if (ret)
>                 drm_sched_job_cleanup(&ejob->base);
>  out_put_job:
> -       ethosu_job_put(ejob);
> +       ethosu_job_err_cleanup(ejob);
>
>         return ret;
>  }
>
> --
> 2.51.0
>

Re: [PATCH 2/3] accel: ethosu: Fix NPU_OP_ELEMENTWISE validation with scalar

[Anders Roxell]

On Wed, 18 Feb 2026 at 23:22, Rob Herring (Arm) <robh@kernel.org> wrote:
>
> The NPU_OP_ELEMENTWISE instruction uses a scalar value for IFM2 if the
> IFM2_BROADCAST "scalar" mode is set. It is a bit (7) on the u65 and
> part of a field (bits 3:0) on the u85. The driver was hardcoded to the
> u85.
>
> Fixes: 5a5e9c0228e6 ("accel: Add Arm Ethos-U NPU driver")
> Signed-off-by: Rob Herring (Arm) <robh@kernel.org>

Reviewed-and-Tested-by: Anders Roxell <anders.roxell@linaro.org>

> ---
>  drivers/accel/ethosu/ethosu_gem.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/accel/ethosu/ethosu_gem.c b/drivers/accel/ethosu/ethosu_gem.c
> index 473b5f5d7514..a735f860a119 100644
> --- a/drivers/accel/ethosu/ethosu_gem.c
> +++ b/drivers/accel/ethosu/ethosu_gem.c
> @@ -417,7 +417,10 @@ static int ethosu_gem_cmdstream_copy_and_validate(struct drm_device *ddev,
>                                 return ret;
>                         break;
>                 case NPU_OP_ELEMENTWISE:
> -                       use_ifm2 = !((st.ifm2.broadcast == 8) || (param == 5) ||
> +                       use_scale = ethosu_is_u65(edev) ?
> +                                   (st.ifm2.broadcast & 0x80) :
> +                                   (st.ifm2.broadcast == 8);
> +                       use_ifm2 = !(use_scale || (param == 5) ||
>                                 (param == 6) || (param == 7) || (param == 0x24));
>                         use_ifm = st.ifm.broadcast != 8;
>                         ret = calc_sizes_elemwise(ddev, info, cmd, &st, use_ifm, use_ifm2);
>
> --
> 2.51.0
>

Re: [PATCH 3/3] accel: ethosu: Handle possible underflow in IFM size calculations

[Anders Roxell]

On Wed, 18 Feb 2026 at 23:22, Rob Herring (Arm) <robh@kernel.org> wrote:
>
> If the command stream has larger padding sizes than the IFM and OFM
> diminsions, then the calculations will underflow to a negative value.

Nit is "diminsions" -> "dimensions"

> The result is a very large region bounds which is caught on submit, but
> it's better to catch it earlier.
>
> Current mesa ethosu driver has a signedness bug which resulted in
> padding of 127 (the max) and triggers this issue.
>
> Signed-off-by: Rob Herring (Arm) <robh@kernel.org>

Reviewed-and-Tested-by: Anders Roxell <anders.roxell@linaro.org>

> ---
>  drivers/accel/ethosu/ethosu_gem.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/accel/ethosu/ethosu_gem.c b/drivers/accel/ethosu/ethosu_gem.c
> index a735f860a119..d1169001c83d 100644
> --- a/drivers/accel/ethosu/ethosu_gem.c
> +++ b/drivers/accel/ethosu/ethosu_gem.c
> @@ -245,11 +245,14 @@ static int calc_sizes(struct drm_device *ddev,
>                         ((st->ifm.stride_kernel >> 1) & 0x1) + 1;
>                 u32 stride_x = ((st->ifm.stride_kernel >> 5) & 0x2) +
>                         (st->ifm.stride_kernel & 0x1) + 1;
> -               u32 ifm_height = st->ofm.height[2] * stride_y +
> +               s32 ifm_height = st->ofm.height[2] * stride_y +
>                         st->ifm.height[2] - (st->ifm.pad_top + st->ifm.pad_bottom);
> -               u32 ifm_width  = st->ofm.width * stride_x +
> +               s32 ifm_width = st->ofm.width * stride_x +
>                         st->ifm.width - (st->ifm.pad_left + st->ifm.pad_right);
>
> +               if (ifm_height < 0 || ifm_width < 0)
> +                       return -EINVAL;
> +
>                 len = feat_matrix_length(info, &st->ifm, ifm_width,
>                                          ifm_height, st->ifm.depth);
>                 dev_dbg(ddev->dev, "op %d: IFM:%d:0x%llx-0x%llx\n",
>
> --
> 2.51.0
>