[PATCH 0/3] drm/v3d: add multiple in/out syncobjs support

[PATCH 0/3] drm/v3d: add multiple in/out syncobjs support

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 2775 bytes --]

Currently, v3d only supports single in/out syncobj per submission (in
v3d_submit_cl, there are two in_sync, one for bin and another for render
job); however, Vulkan queue submit operations expect multiples wait and
signal semaphores. This series extends v3d interface and job dependency
operations to handle more than one in/out syncobj.

The main difference from the RFC series[1] is that here I already worked
on top of Daniel Vetter's work for drm/scheduler[2] (even if the series
is missing a few acks).

The first patch just decouples the steps to lookup and add job
dependency from the job init code, since the operation repeats for every
syncobj that a job should wait before starting. So, the third patch of
this series will reuse it to handle multiples wait for semaphores. 

The second patch extends our interface by using a generic extension.
This approach was inspired by i915_user_extension[3] and
amd_cs_chunks[4] to give a little more flexibility in adding other
submission features in the future. Therefore, the list of extensions
would work as a hub of features that use an id to determine the
corresponding feature data type.

With this base, the third patch adds multiple wait/signal semaphores
support. For this, we add to the list of the generic extensions a new
data type (drm_v3d_multi_sync) that points to two arrays of syncobjs
(in/out) and also determines (flags) if the dependencies must be added
to the bin job or render job (in the case of v3d_submit_cl). An
auxiliary struct (v3d_submit_ext) is used when parsing submission
extensions. Finally, we reserve some space in the semaphore struct
(drm_v3d_sem) to accommodate timeline semaphores that we aim to add
support soon (same reason for already defining v3d_submit_outsync).

[1] https://patchwork.freedesktop.org/series/93388/
[2] https://patchwork.freedesktop.org/series/93413/
[3] https://cgit.freedesktop.org/drm/drm-misc/commit/drivers/gpu/drm/i915/i915_user_extensions.c?id=9d1305ef80b95dde0337106ed8b826604e2155ad
[4] https://cgit.freedesktop.org/drm/drm-misc/tree/include/uapi/drm/amdgpu_drm.h#n556

In the mesa side, the work related to this series is available at
https://gitlab.freedesktop.org/mwen/mesa/-/commit/77bd2b21f61a9caeced934bd13b99e675a428398
where I checked these changes using Mesa CI.

Melissa Wen (3):
  drm/v3d: decouple adding job dependencies steps from job init
  drm/v3d: add generic ioctl extension
  drm/v3d: add multiple syncobjs support

 drivers/gpu/drm/v3d/v3d_drv.c |   7 +-
 drivers/gpu/drm/v3d/v3d_drv.h |  14 ++
 drivers/gpu/drm/v3d/v3d_gem.c | 303 ++++++++++++++++++++++++++++------
 include/uapi/drm/v3d_drm.h    |  76 ++++++++-
 4 files changed, 350 insertions(+), 50 deletions(-)

-- 
2.30.2


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH 1/3] drm/v3d: decouple adding job dependencies steps from job init

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 2695 bytes --]

Prep work to enable a job to wait for more than one syncobj before
start. Also get rid of old checkpatch warnings in the v3d_gem file.
No functional changes.

Signed-off-by: Melissa Wen <mwen@igalia.com>
---
 drivers/gpu/drm/v3d/v3d_gem.c | 28 ++++++++++++++++++----------
 1 file changed, 18 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index a3529809d547..593ed2206d74 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -416,7 +416,7 @@ v3d_wait_bo_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 
 	ret = drm_gem_dma_resv_wait(file_priv, args->handle,
-					      true, timeout_jiffies);
+				    true, timeout_jiffies);
 
 	/* Decrement the user's timeout, in case we got interrupted
 	 * such that the ioctl will be restarted.
@@ -434,12 +434,25 @@ v3d_wait_bo_ioctl(struct drm_device *dev, void *data,
 	return ret;
 }
 
+static int
+v3d_job_add_deps(struct drm_file *file_priv, struct v3d_job *job,
+		 u32 in_sync, u32 point)
+{
+	struct dma_fence *in_fence = NULL;
+	int ret;
+
+	ret = drm_syncobj_find_fence(file_priv, in_sync, point, 0, &in_fence);
+	if (ret == -EINVAL)
+		return ret;
+
+	return drm_sched_job_add_dependency(&job->base, in_fence);
+}
+
 static int
 v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
 	     struct v3d_job *job, void (*free)(struct kref *ref),
 	     u32 in_sync, enum v3d_queue queue)
 {
-	struct dma_fence *in_fence = NULL;
 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
 	int ret;
 
@@ -455,11 +468,7 @@ v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
 	if (ret)
 		goto fail;
 
-	ret = drm_syncobj_find_fence(file_priv, in_sync, 0, 0, &in_fence);
-	if (ret == -EINVAL)
-		goto fail_job;
-
-	ret = drm_sched_job_add_dependency(&job->base, in_fence);
+	ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
 	if (ret)
 		goto fail_job;
 
@@ -499,7 +508,7 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
 	for (i = 0; i < job->bo_count; i++) {
 		/* XXX: Use shared fences for read-only objects. */
 		dma_resv_add_excl_fence(job->bo[i]->resv,
-						  job->done_fence);
+					job->done_fence);
 	}
 
 	drm_gem_unlock_reservations(job->bo, job->bo_count, acquire_ctx);
@@ -904,8 +913,7 @@ v3d_gem_init(struct drm_device *dev)
 	if (!v3d->pt) {
 		drm_mm_takedown(&v3d->mm);
 		dev_err(v3d->drm.dev,
-			"Failed to allocate page tables. "
-			"Please ensure you have CMA enabled.\n");
+			"Failed to allocate page tables. Please ensure you have CMA enabled.\n");
 		return -ENOMEM;
 	}
 
-- 
2.30.2


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH 2/3] drm/v3d: add generic ioctl extension

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 6928 bytes --]

Add support to attach generic extensions on job submission.
This patch is a second prep work to enable multiple syncobjs on job
submission. With this work, when the job submission interface needs
to be extended to accomodate a new feature, we will use a generic
extension struct where an id determines the data type to be pointed.
The first application is to enable multiples in/out syncobj (next
patch), but the base is already done for future features.

Signed-off-by: Melissa Wen <mwen@igalia.com>
---
 drivers/gpu/drm/v3d/v3d_drv.c |  4 +-
 drivers/gpu/drm/v3d/v3d_gem.c | 80 ++++++++++++++++++++++++++++++++---
 include/uapi/drm/v3d_drm.h    | 38 ++++++++++++++++-
 3 files changed, 113 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 9403c3b36aca..6a0516160bb2 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -83,7 +83,6 @@ static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
 		return 0;
 	}
 
-
 	switch (args->param) {
 	case DRM_V3D_PARAM_SUPPORTS_TFU:
 		args->value = 1;
@@ -147,7 +146,7 @@ v3d_postclose(struct drm_device *dev, struct drm_file *file)
 DEFINE_DRM_GEM_FOPS(v3d_drm_fops);
 
 /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have GMP
- * protection between clients.  Note that render nodes would be be
+ * protection between clients.  Note that render nodes would be
  * able to submit CLs that could access BOs from clients authenticated
  * with the master node.  The TFU doesn't use the GMP, so it would
  * need to stay DRM_AUTH until we do buffer size/offset validation.
@@ -222,7 +221,6 @@ static int v3d_platform_drm_probe(struct platform_device *pdev)
 	u32 mmu_debug;
 	u32 ident1;
 
-
 	v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev, drm);
 	if (IS_ERR(v3d))
 		return PTR_ERR(v3d);
diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index 593ed2206d74..e254919b6c5e 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -521,6 +521,38 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
 	}
 }
 
+static int
+v3d_get_extensions(struct drm_file *file_priv,
+		   u32 ext_count, u64 ext_handles)
+{
+	int i;
+	struct drm_v3d_extension __user *handles;
+
+	if (!ext_count)
+		return 0;
+
+	handles	= u64_to_user_ptr(ext_handles);
+	for (i = 0; i < ext_count; i++) {
+		struct drm_v3d_extension ext;
+
+		if (copy_from_user(&ext, handles, sizeof(ext))) {
+			DRM_DEBUG("Failed to copy submit extension\n");
+			return -EFAULT;
+		}
+
+		switch (ext.id) {
+		case 0:
+		default:
+			DRM_DEBUG_DRIVER("Unknown extension id: %d\n", ext.id);
+			return -EINVAL;
+		}
+
+		handles = u64_to_user_ptr(ext.next);
+	}
+
+	return 0;
+}
+
 /**
  * v3d_submit_cl_ioctl() - Submits a job (frame) to the V3D.
  * @dev: DRM device
@@ -549,15 +581,23 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 
 	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args->rcl_end);
 
-	if (args->pad != 0)
-		return -EINVAL;
-
-	if (args->flags != 0 &&
-	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
+	if (args->flags &&
+	    args->flags & ~(DRM_V3D_SUBMIT_CL_FLUSH_CACHE |
+			    DRM_V3D_SUBMIT_EXTENSION)) {
 		DRM_INFO("invalid flags: %d\n", args->flags);
 		return -EINVAL;
 	}
 
+	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
+		ret = v3d_get_extensions(file_priv,
+					 args->extension_count,
+					 args->extensions);
+		if (ret) {
+			DRM_DEBUG("Failed to get extensions.\n");
+			return ret;
+		}
+	}
+
 	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
 	if (!render)
 		return -ENOMEM;
@@ -711,6 +751,21 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 
 	trace_v3d_submit_tfu_ioctl(&v3d->drm, args->iia);
 
+	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
+		DRM_DEBUG("invalid flags: %d\n", args->flags);
+		return -EINVAL;
+	}
+
+	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
+		ret = v3d_get_extensions(file_priv,
+					 args->extension_count,
+					 args->extensions);
+		if (ret) {
+			DRM_DEBUG("Failed to get extensions.\n");
+			return ret;
+		}
+	}
+
 	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
 	if (!job)
 		return -ENOMEM;
@@ -806,6 +861,21 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
+	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
+		DRM_DEBUG("Invalid flags: %d\n", args->flags);
+		return -EINVAL;
+	}
+
+	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
+		ret = v3d_get_extensions(file_priv,
+					 args->extension_count,
+					 args->extensions);
+		if (ret) {
+			DRM_DEBUG("Failed to get extensions.\n");
+			return ret;
+		}
+	}
+
 	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
 	if (!job)
 		return -ENOMEM;
diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
index 4104f22fb3d3..1f4706010eb5 100644
--- a/include/uapi/drm/v3d_drm.h
+++ b/include/uapi/drm/v3d_drm.h
@@ -58,6 +58,19 @@ extern "C" {
 						   struct drm_v3d_perfmon_get_values)
 
 #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
+#define DRM_V3D_SUBMIT_EXTENSION		  0x02
+
+/* struct drm_v3d_extension - ioctl extensions
+ *
+ * Linked-list of generic extensions where the id identify which struct is
+ * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to identify
+ * the extension type.
+ */
+struct drm_v3d_extension {
+	__u64 next;
+	__u64 ext_data;
+	__u32 id;
+};
 
 /**
  * struct drm_v3d_submit_cl - ioctl argument for submitting commands to the 3D
@@ -135,12 +148,17 @@ struct drm_v3d_submit_cl {
 	/* Number of BO handles passed in (size is that times 4). */
 	__u32 bo_handle_count;
 
+	/* DRM_V3D_SUBMIT_* properties */
 	__u32 flags;
 
 	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
 	__u32 perfmon_id;
 
-	__u32 pad;
+	/* Number of extensions*/
+	__u32 extension_count;
+
+	/* Pointer to a list of ioctl extensions*/
+	__u64 extensions;
 };
 
 /**
@@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
 	__u32 in_sync;
 	/* Sync object to signal when the TFU job is done. */
 	__u32 out_sync;
+
+	/* Number of extensions*/
+	__u32 extension_count;
+
+	/* Pointer to an array of ioctl extensions*/
+	__u64 extensions;
+
+	/* DRM_V3D_SUBMIT_* properties */
+	__u32 flags;
 };
 
 /* Submits a compute shader for dispatch.  This job will block on any
@@ -276,6 +303,15 @@ struct drm_v3d_submit_csd {
 
 	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
 	__u32 perfmon_id;
+
+	/* DRM_V3D_SUBMIT_* properties */
+	__u32 flags;
+
+	/* Number of extensions*/
+	__u32 extension_count;
+
+	/* Pointer to a list of ioctl extensions*/
+	__u64 extensions;
 };
 
 enum {
-- 
2.30.2


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 2/3] drm/v3d: add generic ioctl extension

[Iago Toral]

On Wed, 2021-08-18 at 18:56 +0100, Melissa Wen wrote:
> Add support to attach generic extensions on job submission.
> This patch is a second prep work to enable multiple syncobjs on job
> submission. With this work, when the job submission interface needs
> to be extended to accomodate a new feature, we will use a generic
> extension struct where an id determines the data type to be pointed.
> The first application is to enable multiples in/out syncobj (next
> patch), but the base is already done for future features.
> 
> Signed-off-by: Melissa Wen <mwen@igalia.com>
> ---
>  drivers/gpu/drm/v3d/v3d_drv.c |  4 +-
>  drivers/gpu/drm/v3d/v3d_gem.c | 80 ++++++++++++++++++++++++++++++++-
> --
>  include/uapi/drm/v3d_drm.h    | 38 ++++++++++++++++-
>  3 files changed, 113 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/gpu/drm/v3d/v3d_drv.c
> b/drivers/gpu/drm/v3d/v3d_drv.c
> index 9403c3b36aca..6a0516160bb2 100644
> --- a/drivers/gpu/drm/v3d/v3d_drv.c
> +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> @@ -83,7 +83,6 @@ static int v3d_get_param_ioctl(struct drm_device
> *dev, void *data,
>  		return 0;
>  	}
>  
> -
>  	switch (args->param) {
>  	case DRM_V3D_PARAM_SUPPORTS_TFU:
>  		args->value = 1;
> @@ -147,7 +146,7 @@ v3d_postclose(struct drm_device *dev, struct
> drm_file *file)
>  DEFINE_DRM_GEM_FOPS(v3d_drm_fops);
>  
>  /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have
> GMP
> - * protection between clients.  Note that render nodes would be be
> + * protection between clients.  Note that render nodes would be
>   * able to submit CLs that could access BOs from clients
> authenticated
>   * with the master node.  The TFU doesn't use the GMP, so it would
>   * need to stay DRM_AUTH until we do buffer size/offset validation.
> @@ -222,7 +221,6 @@ static int v3d_platform_drm_probe(struct
> platform_device *pdev)
>  	u32 mmu_debug;
>  	u32 ident1;
>  
> -
>  	v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev,
> drm);
>  	if (IS_ERR(v3d))
>  		return PTR_ERR(v3d);
> diff --git a/drivers/gpu/drm/v3d/v3d_gem.c
> b/drivers/gpu/drm/v3d/v3d_gem.c
> index 593ed2206d74..e254919b6c5e 100644
> --- a/drivers/gpu/drm/v3d/v3d_gem.c
> +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> @@ -521,6 +521,38 @@ v3d_attach_fences_and_unlock_reservation(struct
> drm_file *file_priv,
>  	}
>  }
>  
> +static int
> +v3d_get_extensions(struct drm_file *file_priv,
> +		   u32 ext_count, u64 ext_handles)
> +{
> +	int i;
> +	struct drm_v3d_extension __user *handles;
> +
> +	if (!ext_count)
> +		return 0;
> +
> +	handles	= u64_to_user_ptr(ext_handles);
> +	for (i = 0; i < ext_count; i++) {
> +		struct drm_v3d_extension ext;
> +
> +		if (copy_from_user(&ext, handles, sizeof(ext))) {
> +			DRM_DEBUG("Failed to copy submit extension\n");
> +			return -EFAULT;
> +		}
> +
> +		switch (ext.id) {
> +		case 0:
> +		default:
> +			DRM_DEBUG_DRIVER("Unknown extension id: %d\n",
> ext.id);
> +			return -EINVAL;
> +		}
> +
> +		handles = u64_to_user_ptr(ext.next);
> +	}
> +
> +	return 0;
> +}
> +
>  /**
>   * v3d_submit_cl_ioctl() - Submits a job (frame) to the V3D.
>   * @dev: DRM device
> @@ -549,15 +581,23 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> void *data,
>  
>  	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args-
> >rcl_end);
>  
> -	if (args->pad != 0)
> -		return -EINVAL;
> -
> -	if (args->flags != 0 &&
> -	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
> +	if (args->flags &&
> +	    args->flags & ~(DRM_V3D_SUBMIT_CL_FLUSH_CACHE |
> +			    DRM_V3D_SUBMIT_EXTENSION)) {
>  		DRM_INFO("invalid flags: %d\n", args->flags);
>  		return -EINVAL;
>  	}
>  
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
>  	if (!render)
>  		return -ENOMEM;
> @@ -711,6 +751,21 @@ v3d_submit_tfu_ioctl(struct drm_device *dev,
> void *data,
>  
>  	trace_v3d_submit_tfu_ioctl(&v3d->drm, args->iia);
>  
> +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> +		DRM_DEBUG("invalid flags: %d\n", args->flags);
> +		return -EINVAL;
> +	}
> +
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
>  	if (!job)
>  		return -ENOMEM;
> @@ -806,6 +861,21 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> void *data,
>  		return -EINVAL;
>  	}
>  
> +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> +		DRM_DEBUG("Invalid flags: %d\n", args->flags);
> +		return -EINVAL;
> +	}
> +
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
>  	if (!job)
>  		return -ENOMEM;
> diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> index 4104f22fb3d3..1f4706010eb5 100644
> --- a/include/uapi/drm/v3d_drm.h
> +++ b/include/uapi/drm/v3d_drm.h
> @@ -58,6 +58,19 @@ extern "C" {
>  						   struct
> drm_v3d_perfmon_get_values)
>  
>  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
> +#define DRM_V3D_SUBMIT_EXTENSION		  0x02
> +
> +/* struct drm_v3d_extension - ioctl extensions
> + *
> + * Linked-list of generic extensions where the id identify which
> struct is
> + * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to
> identify
> + * the extension type.
> + */
> +struct drm_v3d_extension {
> +	__u64 next;
> +	__u64 ext_data;
> +	__u32 id;
> +};
>  
>  /**
>   * struct drm_v3d_submit_cl - ioctl argument for submitting commands
> to the 3D
> @@ -135,12 +148,17 @@ struct drm_v3d_submit_cl {
>  	/* Number of BO handles passed in (size is that times 4). */
>  	__u32 bo_handle_count;
>  
> +	/* DRM_V3D_SUBMIT_* properties */
>  	__u32 flags;
>  
>  	/* ID of the perfmon to attach to this job. 0 means no perfmon.
> */
>  	__u32 perfmon_id;
>  
> -	__u32 pad;
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to a list of ioctl extensions*/
> +	__u64 extensions;
>  };
>  
>  /**
> @@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
>  	__u32 in_sync;
>  	/* Sync object to signal when the TFU job is done. */
>  	__u32 out_sync;
> +
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to an array of ioctl extensions*/
> +	__u64 extensions;
> +
> +	/* DRM_V3D_SUBMIT_* properties */
> +	__u32 flags;

A silly nit: maybe put flags before the extension fields above for
consistency with the CSD and CL submission commands.

>  };
>  
>  /* Submits a compute shader for dispatch.  This job will block on
> any
> @@ -276,6 +303,15 @@ struct drm_v3d_submit_csd {
>  
>  	/* ID of the perfmon to attach to this job. 0 means no perfmon.
> */
>  	__u32 perfmon_id;
> +
> +	/* DRM_V3D_SUBMIT_* properties */
> +	__u32 flags;
> +
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to a list of ioctl extensions*/
> +	__u64 extensions;
>  };
>  
>  enum {

Re: [PATCH 2/3] drm/v3d: add generic ioctl extension

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 8353 bytes --]

On 09/15, Iago Toral wrote:
> On Wed, 2021-08-18 at 18:56 +0100, Melissa Wen wrote:
> > Add support to attach generic extensions on job submission.
> > This patch is a second prep work to enable multiple syncobjs on job
> > submission. With this work, when the job submission interface needs
> > to be extended to accomodate a new feature, we will use a generic
> > extension struct where an id determines the data type to be pointed.
> > The first application is to enable multiples in/out syncobj (next
> > patch), but the base is already done for future features.
> > 
> > Signed-off-by: Melissa Wen <mwen@igalia.com>
> > ---
> >  drivers/gpu/drm/v3d/v3d_drv.c |  4 +-
> >  drivers/gpu/drm/v3d/v3d_gem.c | 80 ++++++++++++++++++++++++++++++++-
> > --
> >  include/uapi/drm/v3d_drm.h    | 38 ++++++++++++++++-
> >  3 files changed, 113 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.c
> > b/drivers/gpu/drm/v3d/v3d_drv.c
> > index 9403c3b36aca..6a0516160bb2 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.c
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> > @@ -83,7 +83,6 @@ static int v3d_get_param_ioctl(struct drm_device
> > *dev, void *data,
> >  		return 0;
> >  	}
> >  
> > -
> >  	switch (args->param) {
> >  	case DRM_V3D_PARAM_SUPPORTS_TFU:
> >  		args->value = 1;
> > @@ -147,7 +146,7 @@ v3d_postclose(struct drm_device *dev, struct
> > drm_file *file)
> >  DEFINE_DRM_GEM_FOPS(v3d_drm_fops);
> >  
> >  /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have
> > GMP
> > - * protection between clients.  Note that render nodes would be be
> > + * protection between clients.  Note that render nodes would be
> >   * able to submit CLs that could access BOs from clients
> > authenticated
> >   * with the master node.  The TFU doesn't use the GMP, so it would
> >   * need to stay DRM_AUTH until we do buffer size/offset validation.
> > @@ -222,7 +221,6 @@ static int v3d_platform_drm_probe(struct
> > platform_device *pdev)
> >  	u32 mmu_debug;
> >  	u32 ident1;
> >  
> > -
> >  	v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev,
> > drm);
> >  	if (IS_ERR(v3d))
> >  		return PTR_ERR(v3d);
> > diff --git a/drivers/gpu/drm/v3d/v3d_gem.c
> > b/drivers/gpu/drm/v3d/v3d_gem.c
> > index 593ed2206d74..e254919b6c5e 100644
> > --- a/drivers/gpu/drm/v3d/v3d_gem.c
> > +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> > @@ -521,6 +521,38 @@ v3d_attach_fences_and_unlock_reservation(struct
> > drm_file *file_priv,
> >  	}
> >  }
> >  
> > +static int
> > +v3d_get_extensions(struct drm_file *file_priv,
> > +		   u32 ext_count, u64 ext_handles)
> > +{
> > +	int i;
> > +	struct drm_v3d_extension __user *handles;
> > +
> > +	if (!ext_count)
> > +		return 0;
> > +
> > +	handles	= u64_to_user_ptr(ext_handles);
> > +	for (i = 0; i < ext_count; i++) {
> > +		struct drm_v3d_extension ext;
> > +
> > +		if (copy_from_user(&ext, handles, sizeof(ext))) {
> > +			DRM_DEBUG("Failed to copy submit extension\n");
> > +			return -EFAULT;
> > +		}
> > +
> > +		switch (ext.id) {
> > +		case 0:
> > +		default:
> > +			DRM_DEBUG_DRIVER("Unknown extension id: %d\n",
> > ext.id);
> > +			return -EINVAL;
> > +		}
> > +
> > +		handles = u64_to_user_ptr(ext.next);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >  /**
> >   * v3d_submit_cl_ioctl() - Submits a job (frame) to the V3D.
> >   * @dev: DRM device
> > @@ -549,15 +581,23 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> > void *data,
> >  
> >  	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args-
> > >rcl_end);
> >  
> > -	if (args->pad != 0)
> > -		return -EINVAL;
> > -
> > -	if (args->flags != 0 &&
> > -	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
> > +	if (args->flags &&
> > +	    args->flags & ~(DRM_V3D_SUBMIT_CL_FLUSH_CACHE |
> > +			    DRM_V3D_SUBMIT_EXTENSION)) {
> >  		DRM_INFO("invalid flags: %d\n", args->flags);
> >  		return -EINVAL;
> >  	}
> >  
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
> >  	if (!render)
> >  		return -ENOMEM;
> > @@ -711,6 +751,21 @@ v3d_submit_tfu_ioctl(struct drm_device *dev,
> > void *data,
> >  
> >  	trace_v3d_submit_tfu_ioctl(&v3d->drm, args->iia);
> >  
> > +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> > +		DRM_DEBUG("invalid flags: %d\n", args->flags);
> > +		return -EINVAL;
> > +	}
> > +
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> >  	if (!job)
> >  		return -ENOMEM;
> > @@ -806,6 +861,21 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> > void *data,
> >  		return -EINVAL;
> >  	}
> >  
> > +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> > +		DRM_DEBUG("Invalid flags: %d\n", args->flags);
> > +		return -EINVAL;
> > +	}
> > +
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> >  	if (!job)
> >  		return -ENOMEM;
> > diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> > index 4104f22fb3d3..1f4706010eb5 100644
> > --- a/include/uapi/drm/v3d_drm.h
> > +++ b/include/uapi/drm/v3d_drm.h
> > @@ -58,6 +58,19 @@ extern "C" {
> >  						   struct
> > drm_v3d_perfmon_get_values)
> >  
> >  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
> > +#define DRM_V3D_SUBMIT_EXTENSION		  0x02
> > +
> > +/* struct drm_v3d_extension - ioctl extensions
> > + *
> > + * Linked-list of generic extensions where the id identify which
> > struct is
> > + * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to
> > identify
> > + * the extension type.
> > + */
> > +struct drm_v3d_extension {
> > +	__u64 next;
> > +	__u64 ext_data;
> > +	__u32 id;
> > +};
> >  
> >  /**
> >   * struct drm_v3d_submit_cl - ioctl argument for submitting commands
> > to the 3D
> > @@ -135,12 +148,17 @@ struct drm_v3d_submit_cl {
> >  	/* Number of BO handles passed in (size is that times 4). */
> >  	__u32 bo_handle_count;
> >  
> > +	/* DRM_V3D_SUBMIT_* properties */
> >  	__u32 flags;
> >  
> >  	/* ID of the perfmon to attach to this job. 0 means no perfmon.
> > */
> >  	__u32 perfmon_id;
> >  
> > -	__u32 pad;
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to a list of ioctl extensions*/
> > +	__u64 extensions;
> >  };
> >  
> >  /**
> > @@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
> >  	__u32 in_sync;
> >  	/* Sync object to signal when the TFU job is done. */
> >  	__u32 out_sync;
> > +
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to an array of ioctl extensions*/
> > +	__u64 extensions;
> > +
> > +	/* DRM_V3D_SUBMIT_* properties */
> > +	__u32 flags;
> 
> A silly nit: maybe put flags before the extension fields above for
> consistency with the CSD and CL submission commands.

hmm.. I arranged it that way for alignment reasons (afaiu), but I can
(or should) include a _u32 pad right after out_sync to set these in the
same sequence.

> 
> >  };
> >  
> >  /* Submits a compute shader for dispatch.  This job will block on
> > any
> > @@ -276,6 +303,15 @@ struct drm_v3d_submit_csd {
> >  
> >  	/* ID of the perfmon to attach to this job. 0 means no perfmon.
> > */
> >  	__u32 perfmon_id;
> > +
> > +	/* DRM_V3D_SUBMIT_* properties */
> > +	__u32 flags;
> > +
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to a list of ioctl extensions*/
> > +	__u64 extensions;
> >  };
> >  
> >  enum {
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 2/3] drm/v3d: add generic ioctl extension

[Iago Toral]

On Wed, 2021-09-15 at 17:28 +0100, Melissa Wen wrote:
> On 09/15, Iago Toral wrote:
> > On Wed, 2021-08-18 at 18:56 +0100, Melissa Wen wrote:
(...)
> > > 
> > >  /**
> > > @@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
> > >  	__u32 in_sync;
> > >  	/* Sync object to signal when the TFU job is done. */
> > >  	__u32 out_sync;
> > > +
> > > +	/* Number of extensions*/
> > > +	__u32 extension_count;
> > > +
> > > +	/* Pointer to an array of ioctl extensions*/
> > > +	__u64 extensions;
> > > +
> > > +	/* DRM_V3D_SUBMIT_* properties */
> > > +	__u32 flags;
> > 
> > A silly nit: maybe put flags before the extension fields above for
> > consistency with the CSD and CL submission commands.
> 
> hmm.. I arranged it that way for alignment reasons (afaiu), but I can
> (or should) include a _u32 pad right after out_sync to set these in
> the
> same sequence.

Ah, that's fine, my suggestion was just for style, let's keep it as is.

Iago

Re: [PATCH 2/3] drm/v3d: add generic ioctl extension

[Daniel Vetter]

On Wed, Aug 18, 2021 at 06:56:41PM +0100, Melissa Wen wrote:
> Add support to attach generic extensions on job submission.
> This patch is a second prep work to enable multiple syncobjs on job
> submission. With this work, when the job submission interface needs
> to be extended to accomodate a new feature, we will use a generic
> extension struct where an id determines the data type to be pointed.
> The first application is to enable multiples in/out syncobj (next
> patch), but the base is already done for future features.
> 
> Signed-off-by: Melissa Wen <mwen@igalia.com>
> ---
>  drivers/gpu/drm/v3d/v3d_drv.c |  4 +-
>  drivers/gpu/drm/v3d/v3d_gem.c | 80 ++++++++++++++++++++++++++++++++---
>  include/uapi/drm/v3d_drm.h    | 38 ++++++++++++++++-
>  3 files changed, 113 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
> index 9403c3b36aca..6a0516160bb2 100644
> --- a/drivers/gpu/drm/v3d/v3d_drv.c
> +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> @@ -83,7 +83,6 @@ static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
>  		return 0;
>  	}
>  
> -
>  	switch (args->param) {
>  	case DRM_V3D_PARAM_SUPPORTS_TFU:
>  		args->value = 1;
> @@ -147,7 +146,7 @@ v3d_postclose(struct drm_device *dev, struct drm_file *file)
>  DEFINE_DRM_GEM_FOPS(v3d_drm_fops);
>  
>  /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have GMP
> - * protection between clients.  Note that render nodes would be be
> + * protection between clients.  Note that render nodes would be
>   * able to submit CLs that could access BOs from clients authenticated
>   * with the master node.  The TFU doesn't use the GMP, so it would
>   * need to stay DRM_AUTH until we do buffer size/offset validation.
> @@ -222,7 +221,6 @@ static int v3d_platform_drm_probe(struct platform_device *pdev)
>  	u32 mmu_debug;
>  	u32 ident1;
>  
> -
>  	v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev, drm);
>  	if (IS_ERR(v3d))
>  		return PTR_ERR(v3d);
> diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
> index 593ed2206d74..e254919b6c5e 100644
> --- a/drivers/gpu/drm/v3d/v3d_gem.c
> +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> @@ -521,6 +521,38 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
>  	}
>  }
>  
> +static int
> +v3d_get_extensions(struct drm_file *file_priv,
> +		   u32 ext_count, u64 ext_handles)
> +{
> +	int i;
> +	struct drm_v3d_extension __user *handles;
> +
> +	if (!ext_count)
> +		return 0;
> +
> +	handles	= u64_to_user_ptr(ext_handles);
> +	for (i = 0; i < ext_count; i++) {
> +		struct drm_v3d_extension ext;
> +
> +		if (copy_from_user(&ext, handles, sizeof(ext))) {
> +			DRM_DEBUG("Failed to copy submit extension\n");
> +			return -EFAULT;
> +		}
> +
> +		switch (ext.id) {
> +		case 0:
> +		default:
> +			DRM_DEBUG_DRIVER("Unknown extension id: %d\n", ext.id);
> +			return -EINVAL;
> +		}
> +
> +		handles = u64_to_user_ptr(ext.next);
> +	}
> +
> +	return 0;
> +}
> +
>  /**
>   * v3d_submit_cl_ioctl() - Submits a job (frame) to the V3D.
>   * @dev: DRM device
> @@ -549,15 +581,23 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
>  
>  	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args->rcl_end);
>  
> -	if (args->pad != 0)
> -		return -EINVAL;
> -
> -	if (args->flags != 0 &&
> -	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
> +	if (args->flags &&
> +	    args->flags & ~(DRM_V3D_SUBMIT_CL_FLUSH_CACHE |
> +			    DRM_V3D_SUBMIT_EXTENSION)) {
>  		DRM_INFO("invalid flags: %d\n", args->flags);
>  		return -EINVAL;
>  	}
>  
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
>  	if (!render)
>  		return -ENOMEM;
> @@ -711,6 +751,21 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
>  
>  	trace_v3d_submit_tfu_ioctl(&v3d->drm, args->iia);
>  
> +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> +		DRM_DEBUG("invalid flags: %d\n", args->flags);
> +		return -EINVAL;
> +	}
> +
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
>  	if (!job)
>  		return -ENOMEM;
> @@ -806,6 +861,21 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
>  		return -EINVAL;
>  	}
>  
> +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> +		DRM_DEBUG("Invalid flags: %d\n", args->flags);
> +		return -EINVAL;
> +	}
> +
> +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> +		ret = v3d_get_extensions(file_priv,
> +					 args->extension_count,
> +					 args->extensions);
> +		if (ret) {
> +			DRM_DEBUG("Failed to get extensions.\n");
> +			return ret;
> +		}
> +	}
> +
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
>  	if (!job)
>  		return -ENOMEM;
> diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> index 4104f22fb3d3..1f4706010eb5 100644
> --- a/include/uapi/drm/v3d_drm.h
> +++ b/include/uapi/drm/v3d_drm.h
> @@ -58,6 +58,19 @@ extern "C" {
>  						   struct drm_v3d_perfmon_get_values)
>  
>  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
> +#define DRM_V3D_SUBMIT_EXTENSION		  0x02
> +
> +/* struct drm_v3d_extension - ioctl extensions
> + *
> + * Linked-list of generic extensions where the id identify which struct is
> + * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to identify
> + * the extension type.
> + */
> +struct drm_v3d_extension {
> +	__u64 next;

Why do you both need a next pointer here and extension_count everywhere?
That seems one too much.

> +	__u64 ext_data;

This isn't needed if you link them. Instead each extension can subclass
this struct here, and add whatever parameter they need there. Some
extensions could be just a flag which only needs to be the extension
present. Maybe what you want here is a __u32 for flags? Solves also the
aligning.

> +	__u32 id;

Align to 64bit just to be save.

One thing I wondered is whether we shouldn't lift this to be a drm thing.
i915 has something similar with i915_user_extension.

That way we could share some helpers for parsing these, and people would
do extensible drm ioctls all the same way?
-Daniel



> +};
>  
>  /**
>   * struct drm_v3d_submit_cl - ioctl argument for submitting commands to the 3D
> @@ -135,12 +148,17 @@ struct drm_v3d_submit_cl {
>  	/* Number of BO handles passed in (size is that times 4). */
>  	__u32 bo_handle_count;
>  
> +	/* DRM_V3D_SUBMIT_* properties */
>  	__u32 flags;
>  
>  	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
>  	__u32 perfmon_id;
>  
> -	__u32 pad;
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to a list of ioctl extensions*/
> +	__u64 extensions;
>  };
>  
>  /**
> @@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
>  	__u32 in_sync;
>  	/* Sync object to signal when the TFU job is done. */
>  	__u32 out_sync;
> +
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to an array of ioctl extensions*/
> +	__u64 extensions;
> +
> +	/* DRM_V3D_SUBMIT_* properties */
> +	__u32 flags;
>  };
>  
>  /* Submits a compute shader for dispatch.  This job will block on any
> @@ -276,6 +303,15 @@ struct drm_v3d_submit_csd {
>  
>  	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
>  	__u32 perfmon_id;
> +
> +	/* DRM_V3D_SUBMIT_* properties */
> +	__u32 flags;
> +
> +	/* Number of extensions*/
> +	__u32 extension_count;
> +
> +	/* Pointer to a list of ioctl extensions*/
> +	__u64 extensions;
>  };
>  
>  enum {
> -- 
> 2.30.2
> 



-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Re: [PATCH 2/3] drm/v3d: add generic ioctl extension

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 9202 bytes --]

On 09/16, Daniel Vetter wrote:
> On Wed, Aug 18, 2021 at 06:56:41PM +0100, Melissa Wen wrote:
> > Add support to attach generic extensions on job submission.
> > This patch is a second prep work to enable multiple syncobjs on job
> > submission. With this work, when the job submission interface needs
> > to be extended to accomodate a new feature, we will use a generic
> > extension struct where an id determines the data type to be pointed.
> > The first application is to enable multiples in/out syncobj (next
> > patch), but the base is already done for future features.
> > 
> > Signed-off-by: Melissa Wen <mwen@igalia.com>
> > ---
> >  drivers/gpu/drm/v3d/v3d_drv.c |  4 +-
> >  drivers/gpu/drm/v3d/v3d_gem.c | 80 ++++++++++++++++++++++++++++++++---
> >  include/uapi/drm/v3d_drm.h    | 38 ++++++++++++++++-
> >  3 files changed, 113 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
> > index 9403c3b36aca..6a0516160bb2 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.c
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> > @@ -83,7 +83,6 @@ static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
> >  		return 0;
> >  	}
> >  
> > -
> >  	switch (args->param) {
> >  	case DRM_V3D_PARAM_SUPPORTS_TFU:
> >  		args->value = 1;
> > @@ -147,7 +146,7 @@ v3d_postclose(struct drm_device *dev, struct drm_file *file)
> >  DEFINE_DRM_GEM_FOPS(v3d_drm_fops);
> >  
> >  /* DRM_AUTH is required on SUBMIT_CL for now, while we don't have GMP
> > - * protection between clients.  Note that render nodes would be be
> > + * protection between clients.  Note that render nodes would be
> >   * able to submit CLs that could access BOs from clients authenticated
> >   * with the master node.  The TFU doesn't use the GMP, so it would
> >   * need to stay DRM_AUTH until we do buffer size/offset validation.
> > @@ -222,7 +221,6 @@ static int v3d_platform_drm_probe(struct platform_device *pdev)
> >  	u32 mmu_debug;
> >  	u32 ident1;
> >  
> > -
> >  	v3d = devm_drm_dev_alloc(dev, &v3d_drm_driver, struct v3d_dev, drm);
> >  	if (IS_ERR(v3d))
> >  		return PTR_ERR(v3d);
> > diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
> > index 593ed2206d74..e254919b6c5e 100644
> > --- a/drivers/gpu/drm/v3d/v3d_gem.c
> > +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> > @@ -521,6 +521,38 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
> >  	}
> >  }
> >  
> > +static int
> > +v3d_get_extensions(struct drm_file *file_priv,
> > +		   u32 ext_count, u64 ext_handles)
> > +{
> > +	int i;
> > +	struct drm_v3d_extension __user *handles;
> > +
> > +	if (!ext_count)
> > +		return 0;
> > +
> > +	handles	= u64_to_user_ptr(ext_handles);
> > +	for (i = 0; i < ext_count; i++) {
> > +		struct drm_v3d_extension ext;
> > +
> > +		if (copy_from_user(&ext, handles, sizeof(ext))) {
> > +			DRM_DEBUG("Failed to copy submit extension\n");
> > +			return -EFAULT;
> > +		}
> > +
> > +		switch (ext.id) {
> > +		case 0:
> > +		default:
> > +			DRM_DEBUG_DRIVER("Unknown extension id: %d\n", ext.id);
> > +			return -EINVAL;
> > +		}
> > +
> > +		handles = u64_to_user_ptr(ext.next);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> >  /**
> >   * v3d_submit_cl_ioctl() - Submits a job (frame) to the V3D.
> >   * @dev: DRM device
> > @@ -549,15 +581,23 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
> >  
> >  	trace_v3d_submit_cl_ioctl(&v3d->drm, args->rcl_start, args->rcl_end);
> >  
> > -	if (args->pad != 0)
> > -		return -EINVAL;
> > -
> > -	if (args->flags != 0 &&
> > -	    args->flags != DRM_V3D_SUBMIT_CL_FLUSH_CACHE) {
> > +	if (args->flags &&
> > +	    args->flags & ~(DRM_V3D_SUBMIT_CL_FLUSH_CACHE |
> > +			    DRM_V3D_SUBMIT_EXTENSION)) {
> >  		DRM_INFO("invalid flags: %d\n", args->flags);
> >  		return -EINVAL;
> >  	}
> >  
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
> >  	if (!render)
> >  		return -ENOMEM;
> > @@ -711,6 +751,21 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
> >  
> >  	trace_v3d_submit_tfu_ioctl(&v3d->drm, args->iia);
> >  
> > +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> > +		DRM_DEBUG("invalid flags: %d\n", args->flags);
> > +		return -EINVAL;
> > +	}
> > +
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> >  	if (!job)
> >  		return -ENOMEM;
> > @@ -806,6 +861,21 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
> >  		return -EINVAL;
> >  	}
> >  
> > +	if (args->flags && !(args->flags & DRM_V3D_SUBMIT_EXTENSION)) {
> > +		DRM_DEBUG("Invalid flags: %d\n", args->flags);
> > +		return -EINVAL;
> > +	}
> > +
> > +	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > +		ret = v3d_get_extensions(file_priv,
> > +					 args->extension_count,
> > +					 args->extensions);
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to get extensions.\n");
> > +			return ret;
> > +		}
> > +	}
> > +
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> >  	if (!job)
> >  		return -ENOMEM;
> > diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> > index 4104f22fb3d3..1f4706010eb5 100644
> > --- a/include/uapi/drm/v3d_drm.h
> > +++ b/include/uapi/drm/v3d_drm.h
> > @@ -58,6 +58,19 @@ extern "C" {
> >  						   struct drm_v3d_perfmon_get_values)
> >  
> >  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
> > +#define DRM_V3D_SUBMIT_EXTENSION		  0x02
> > +
> > +/* struct drm_v3d_extension - ioctl extensions
> > + *
> > + * Linked-list of generic extensions where the id identify which struct is
> > + * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to identify
> > + * the extension type.
> > + */
> > +struct drm_v3d_extension {
> > +	__u64 next;
> 
> Why do you both need a next pointer here and extension_count everywhere?
> That seems one too much.
> 
> > +	__u64 ext_data;
> 
> This isn't needed if you link them. Instead each extension can subclass
> this struct here, and add whatever parameter they need there. Some
> extensions could be just a flag which only needs to be the extension
> present. Maybe what you want here is a __u32 for flags? Solves also the
> aligning.
> 
> > +	__u32 id;
> 
> Align to 64bit just to be save.

Hi Daniel,

Thanks for suggesting these improvements. I'll polish it for the next version.

> 
> One thing I wondered is whether we shouldn't lift this to be a drm thing.
> i915 has something similar with i915_user_extension.
> 
> That way we could share some helpers for parsing these, and people would
> do extensible drm ioctls all the same way?

I think so. I've based on solutions from i915 and amd drivers for ioctl
extension. The motivations presented in the commit message for
i915_user_extension (vulkan) are similar to ours, and maybe being in drm
will make it simpler for the drivers to adopt too.

Melissa

> -Daniel
> 
> 
> 
> > +};
> >  
> >  /**
> >   * struct drm_v3d_submit_cl - ioctl argument for submitting commands to the 3D
> > @@ -135,12 +148,17 @@ struct drm_v3d_submit_cl {
> >  	/* Number of BO handles passed in (size is that times 4). */
> >  	__u32 bo_handle_count;
> >  
> > +	/* DRM_V3D_SUBMIT_* properties */
> >  	__u32 flags;
> >  
> >  	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
> >  	__u32 perfmon_id;
> >  
> > -	__u32 pad;
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to a list of ioctl extensions*/
> > +	__u64 extensions;
> >  };
> >  
> >  /**
> > @@ -248,6 +266,15 @@ struct drm_v3d_submit_tfu {
> >  	__u32 in_sync;
> >  	/* Sync object to signal when the TFU job is done. */
> >  	__u32 out_sync;
> > +
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to an array of ioctl extensions*/
> > +	__u64 extensions;
> > +
> > +	/* DRM_V3D_SUBMIT_* properties */
> > +	__u32 flags;
> >  };
> >  
> >  /* Submits a compute shader for dispatch.  This job will block on any
> > @@ -276,6 +303,15 @@ struct drm_v3d_submit_csd {
> >  
> >  	/* ID of the perfmon to attach to this job. 0 means no perfmon. */
> >  	__u32 perfmon_id;
> > +
> > +	/* DRM_V3D_SUBMIT_* properties */
> > +	__u32 flags;
> > +
> > +	/* Number of extensions*/
> > +	__u32 extension_count;
> > +
> > +	/* Pointer to a list of ioctl extensions*/
> > +	__u64 extensions;
> >  };
> >  
> >  enum {
> > -- 
> > 2.30.2
> > 
> 
> 
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH 3/3] drm/v3d: add multiple syncobjs support

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 15769 bytes --]

Using the generic extension support set in the previous patch, this
patch enables more than one in/out binary syncobj per job submission.
Arrays of syncobjs are set in a specific extension type (multisync)
that also cares of determining the stage for sync (bin/render)
through a flag - when this is the case.

Signed-off-by: Melissa Wen <mwen@igalia.com>
---
 drivers/gpu/drm/v3d/v3d_drv.c |   3 +
 drivers/gpu/drm/v3d/v3d_drv.h |  14 +++
 drivers/gpu/drm/v3d/v3d_gem.c | 209 +++++++++++++++++++++++++++-------
 include/uapi/drm/v3d_drm.h    |  38 +++++++
 4 files changed, 226 insertions(+), 38 deletions(-)

diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 6a0516160bb2..939ca8c833f5 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -96,6 +96,9 @@ static int v3d_get_param_ioctl(struct drm_device *dev, void *data,
 	case DRM_V3D_PARAM_SUPPORTS_PERFMON:
 		args->value = (v3d->ver >= 40);
 		return 0;
+	case DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT:
+		args->value = 1;
+		return 0;
 	default:
 		DRM_DEBUG("Unknown parameter %d\n", args->param);
 		return -EINVAL;
diff --git a/drivers/gpu/drm/v3d/v3d_drv.h b/drivers/gpu/drm/v3d/v3d_drv.h
index b900a050d5e2..544c60404a0f 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.h
+++ b/drivers/gpu/drm/v3d/v3d_drv.h
@@ -294,6 +294,20 @@ struct v3d_csd_job {
 	struct drm_v3d_submit_csd args;
 };
 
+struct v3d_submit_outsync {
+	struct drm_syncobj *syncobj;
+};
+
+struct v3d_submit_ext {
+	u32 flags;
+
+	u32 in_sync_count;
+	u64 in_syncs;
+
+	u32 out_sync_count;
+	struct v3d_submit_outsync *out_syncs;
+};
+
 /**
  * __wait_for - magic wait macro
  *
diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index e254919b6c5e..e7aabe1a0e11 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -392,6 +392,9 @@ v3d_render_job_free(struct kref *ref)
 
 void v3d_job_cleanup(struct v3d_job *job)
 {
+	if (!job)
+		return;
+
 	drm_sched_job_cleanup(&job->base);
 	v3d_job_put(job);
 }
@@ -451,10 +454,11 @@ v3d_job_add_deps(struct drm_file *file_priv, struct v3d_job *job,
 static int
 v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
 	     struct v3d_job *job, void (*free)(struct kref *ref),
-	     u32 in_sync, enum v3d_queue queue)
+	     u32 in_sync, struct v3d_submit_ext *se, enum v3d_queue queue)
 {
 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
-	int ret;
+	bool has_multinsync = (se && se->in_sync_count);
+	int ret, i;
 
 	job->v3d = v3d;
 	job->free = free;
@@ -463,14 +467,30 @@ v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
 	if (ret < 0)
 		return ret;
 
-	ret = drm_sched_job_init(&job->base, &v3d_priv->sched_entity[queue],
-				 v3d_priv);
+	ret = drm_sched_job_init(&job->base, &v3d_priv->sched_entity[queue], v3d_priv);
 	if (ret)
 		goto fail;
 
-	ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
-	if (ret)
-		goto fail_job;
+	if (has_multinsync && (se->flags == (queue == V3D_RENDER))) {
+		struct drm_v3d_sem __user *handle = u64_to_user_ptr(se->in_syncs);
+
+		for (i = 0; i < se->in_sync_count; i++) {
+			struct drm_v3d_sem in;
+
+			ret = copy_from_user(&in, handle++, sizeof(in));
+			if (ret) {
+				DRM_DEBUG("Failed to copy wait dep handle.\n");
+				goto fail_job;
+			}
+			ret = v3d_job_add_deps(file_priv, job, in.handle, 0);
+			if (ret)
+				goto fail_job;
+		}
+	} else if (!has_multinsync) {
+		ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
+		if (ret)
+			goto fail_job;
+	}
 
 	kref_init(&job->refcount);
 
@@ -500,6 +520,7 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
 					 struct v3d_job *job,
 					 struct ww_acquire_ctx *acquire_ctx,
 					 u32 out_sync,
+					 struct v3d_submit_ext *se,
 					 struct dma_fence *done_fence)
 {
 	struct drm_syncobj *sync_out;
@@ -514,6 +535,18 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
 	drm_gem_unlock_reservations(job->bo, job->bo_count, acquire_ctx);
 
 	/* Update the return sync object for the job */
+	/* If multiples semaphores is supported */
+	if (se && se->out_sync_count) {
+		for (i = 0; i < se->out_sync_count; i++) {
+			drm_syncobj_replace_fence(se->out_syncs[i].syncobj,
+						  done_fence);
+			drm_syncobj_put(se->out_syncs[i].syncobj);
+		}
+		kvfree(se->out_syncs);
+		return;
+	}
+
+	/* Single signal semaphore */
 	sync_out = drm_syncobj_find(file_priv, out_sync);
 	if (sync_out) {
 		drm_syncobj_replace_fence(sync_out, done_fence);
@@ -521,11 +554,93 @@ v3d_attach_fences_and_unlock_reservation(struct drm_file *file_priv,
 	}
 }
 
+static void
+v3d_put_multisync_post_deps(struct v3d_submit_ext *se)
+{
+	unsigned int i;
+
+	for (i = 0; i < se->out_sync_count; i++)
+		drm_syncobj_put(se->out_syncs[i].syncobj);
+	kvfree(se->out_syncs);
+}
+
+static int
+v3d_get_multisync_post_deps(struct drm_file *file_priv,
+			    struct v3d_submit_ext *se,
+			    u32 count, u64 handles)
+{
+	struct drm_v3d_sem __user *post_deps;
+	int i, ret;
+
+	se->out_syncs = (struct v3d_submit_outsync *)
+			kvmalloc_array(count,
+				       sizeof(struct v3d_submit_outsync),
+				       GFP_KERNEL);
+	if (!se->out_syncs)
+		return -ENOMEM;
+
+	post_deps = u64_to_user_ptr(handles);
+
+	for (i = 0; i < count; i++) {
+		struct drm_v3d_sem out;
+
+		ret = copy_from_user(&out, post_deps++, sizeof(out));
+		if (ret) {
+			DRM_DEBUG("Failed to copy post dep handles\n");
+			goto fail;
+		}
+
+		se->out_syncs[i].syncobj = drm_syncobj_find(file_priv,
+							    out.handle);
+		if (!se->out_syncs[i].syncobj) {
+			ret = -EINVAL;
+			goto fail;
+		}
+	}
+	se->out_sync_count = count;
+
+	return 0;
+
+fail:
+	for (i--; i >= 0; i--)
+		drm_syncobj_put(se->out_syncs[i].syncobj);
+	kvfree(se->out_syncs);
+
+	return ret;
+}
+
+static int
+v3d_get_multisync_submit_deps(struct drm_file *file_priv,
+			      struct v3d_submit_ext *se,
+			      u64 ext_data)
+{
+	struct drm_v3d_multi_sync multisync = {0};
+	int ret;
+
+	ret = copy_from_user(&multisync, u64_to_user_ptr(ext_data),
+			     sizeof(multisync));
+	if (ret)
+		return ret;
+
+	ret = v3d_get_multisync_post_deps(file_priv, se, multisync.out_sync_count,
+					  multisync.out_syncs);
+	if (ret)
+		return ret;
+
+	se->in_sync_count = multisync.in_sync_count;
+	se->in_syncs = multisync.in_syncs;
+
+	se->flags = multisync.flags;
+
+	return 0;
+}
+
 static int
 v3d_get_extensions(struct drm_file *file_priv,
+		   struct v3d_submit_ext *se,
 		   u32 ext_count, u64 ext_handles)
 {
-	int i;
+	int i, ret;
 	struct drm_v3d_extension __user *handles;
 
 	if (!ext_count)
@@ -541,7 +656,12 @@ v3d_get_extensions(struct drm_file *file_priv,
 		}
 
 		switch (ext.id) {
-		case 0:
+		case DRM_V3D_EXT_ID_MULTI_SYNC:
+			ret = v3d_get_multisync_submit_deps(file_priv, se,
+							    ext.ext_data);
+			if (ret)
+				return ret;
+			break;
 		default:
 			DRM_DEBUG_DRIVER("Unknown extension id: %d\n", ext.id);
 			return -EINVAL;
@@ -572,6 +692,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
 	struct drm_v3d_submit_cl *args = data;
+	struct v3d_submit_ext se = {0};
 	struct v3d_bin_job *bin = NULL;
 	struct v3d_render_job *render;
 	struct v3d_job *clean_job = NULL;
@@ -589,7 +710,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	}
 
 	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
-		ret = v3d_get_extensions(file_priv,
+		ret = v3d_get_extensions(file_priv, &se,
 					 args->extension_count,
 					 args->extensions);
 		if (ret) {
@@ -599,33 +720,35 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	}
 
 	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
-	if (!render)
+	if (!render) {
+		v3d_put_multisync_post_deps(&se);
 		return -ENOMEM;
+	}
 
 	render->start = args->rcl_start;
 	render->end = args->rcl_end;
 	INIT_LIST_HEAD(&render->unref_list);
 
-	ret = v3d_job_init(v3d, file_priv, &render->base,
-			   v3d_render_job_free, args->in_sync_rcl, V3D_RENDER);
+	ret = v3d_job_init(v3d, file_priv, &render->base, v3d_render_job_free,
+			   args->in_sync_rcl, &se, V3D_RENDER);
 	if (ret) {
 		kfree(render);
+		v3d_put_multisync_post_deps(&se);
 		return ret;
 	}
 
 	if (args->bcl_start != args->bcl_end) {
 		bin = kcalloc(1, sizeof(*bin), GFP_KERNEL);
 		if (!bin) {
-			v3d_job_put(&render->base);
-			return -ENOMEM;
+			ret = -ENOMEM;
+			goto fail;
 		}
 
 		ret = v3d_job_init(v3d, file_priv, &bin->base,
-				   v3d_job_free, args->in_sync_bcl, V3D_BIN);
+				   v3d_job_free, args->in_sync_bcl, &se, V3D_BIN);
 		if (ret) {
-			v3d_job_put(&render->base);
 			kfree(bin);
-			return ret;
+			goto fail;
 		}
 
 		bin->start = args->bcl_start;
@@ -643,7 +766,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 			goto fail;
 		}
 
-		ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0, V3D_CACHE_CLEAN);
+		ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0, 0, V3D_CACHE_CLEAN);
 		if (ret) {
 			kfree(clean_job);
 			clean_job = NULL;
@@ -706,6 +829,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 						 last_job,
 						 &acquire_ctx,
 						 args->out_sync,
+						 &se,
 						 last_job->done_fence);
 
 	if (bin)
@@ -721,11 +845,10 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	drm_gem_unlock_reservations(last_job->bo,
 				    last_job->bo_count, &acquire_ctx);
 fail:
-	if (bin)
-		v3d_job_cleanup(&bin->base);
+	v3d_job_cleanup(&bin->base);
 	v3d_job_cleanup(&render->base);
-	if (clean_job)
-		v3d_job_cleanup(clean_job);
+	v3d_job_cleanup(clean_job);
+	v3d_put_multisync_post_deps(&se);
 
 	return ret;
 }
@@ -745,6 +868,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct drm_v3d_submit_tfu *args = data;
+	struct v3d_submit_ext se = {0};
 	struct v3d_tfu_job *job;
 	struct ww_acquire_ctx acquire_ctx;
 	int ret = 0;
@@ -757,7 +881,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 	}
 
 	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
-		ret = v3d_get_extensions(file_priv,
+		ret = v3d_get_extensions(file_priv, &se,
 					 args->extension_count,
 					 args->extensions);
 		if (ret) {
@@ -767,21 +891,24 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 	}
 
 	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
-	if (!job)
+	if (!job) {
+		v3d_put_multisync_post_deps(&se);
 		return -ENOMEM;
+	}
 
 	ret = v3d_job_init(v3d, file_priv, &job->base,
-			   v3d_job_free, args->in_sync, V3D_TFU);
+			   v3d_job_free, args->in_sync, &se, V3D_TFU);
 	if (ret) {
 		kfree(job);
+		v3d_put_multisync_post_deps(&se);
 		return ret;
 	}
 
 	job->base.bo = kcalloc(ARRAY_SIZE(args->bo_handles),
 			       sizeof(*job->base.bo), GFP_KERNEL);
 	if (!job->base.bo) {
-		v3d_job_put(&job->base);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto fail;
 	}
 
 	job->args = *args;
@@ -821,6 +948,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 	v3d_attach_fences_and_unlock_reservation(file_priv,
 						 &job->base, &acquire_ctx,
 						 args->out_sync,
+						 &se,
 						 job->base.done_fence);
 
 	v3d_job_put(&job->base);
@@ -829,6 +957,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
 
 fail:
 	v3d_job_cleanup(&job->base);
+	v3d_put_multisync_post_deps(&se);
 
 	return ret;
 }
@@ -849,8 +978,9 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
 	struct drm_v3d_submit_csd *args = data;
+	struct v3d_submit_ext se = {0};
 	struct v3d_csd_job *job;
-	struct v3d_job *clean_job;
+	struct v3d_job *clean_job = NULL;
 	struct ww_acquire_ctx acquire_ctx;
 	int ret;
 
@@ -867,7 +997,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 	}
 
 	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
-		ret = v3d_get_extensions(file_priv,
+		ret = v3d_get_extensions(file_priv, &se,
 					 args->extension_count,
 					 args->extensions);
 		if (ret) {
@@ -877,28 +1007,29 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 	}
 
 	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
-	if (!job)
+	if (!job) {
+		v3d_put_multisync_post_deps(&se);
 		return -ENOMEM;
+	}
 
 	ret = v3d_job_init(v3d, file_priv, &job->base,
-			   v3d_job_free, args->in_sync, V3D_CSD);
+			   v3d_job_free, args->in_sync, &se, V3D_CSD);
 	if (ret) {
 		kfree(job);
+		v3d_put_multisync_post_deps(&se);
 		return ret;
 	}
 
 	clean_job = kcalloc(1, sizeof(*clean_job), GFP_KERNEL);
 	if (!clean_job) {
-		v3d_job_put(&job->base);
-		kfree(job);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto fail;
 	}
 
-	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0, V3D_CACHE_CLEAN);
+	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0, 0, V3D_CACHE_CLEAN);
 	if (ret) {
-		v3d_job_put(&job->base);
 		kfree(clean_job);
-		return ret;
+		goto fail;
 	}
 
 	job->args = *args;
@@ -936,6 +1067,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 						 clean_job,
 						 &acquire_ctx,
 						 args->out_sync,
+						 &se,
 						 clean_job->done_fence);
 
 	v3d_job_put(&job->base);
@@ -950,6 +1082,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void *data,
 fail:
 	v3d_job_cleanup(&job->base);
 	v3d_job_cleanup(clean_job);
+	v3d_put_multisync_post_deps(&se);
 
 	return ret;
 }
diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
index 1f4706010eb5..bbb904c521b4 100644
--- a/include/uapi/drm/v3d_drm.h
+++ b/include/uapi/drm/v3d_drm.h
@@ -60,6 +60,42 @@ extern "C" {
 #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
 #define DRM_V3D_SUBMIT_EXTENSION		  0x02
 
+/* struct drm_v3d_sem - wait/signal semaphore
+ *
+ * If binary semaphore, it only takes syncobj handle and ignores flags and
+ * point fields. Point is defined for timeline syncobj feature.
+ */
+struct drm_v3d_sem {
+	__u32 handle; /* syncobj */
+	/* rsv below, for future uses */
+	__u32 flags;
+	__u64 point;  /* for timeline sem support */
+	__u64 mbz[2]; /* must be zero, rsv */
+};
+
+/**
+ * struct drm_v3d_multi_sync - ioctl extension to add support multiples
+ * syncobjs for commands submission.
+ *
+ * When an extension of DRM_V3D_EXT_ID_MULTI_SYNC id is defined, it points to
+ * this extension to define wait and signal dependencies, instead of single
+ * in/out sync entries on submitting commands. The field flags is used to
+ * determine the stage to set wait dependencies.
+ */
+struct drm_v3d_multi_sync {
+	/* Array of wait and signal semaphores */
+	__u64 in_syncs;
+	__u64 out_syncs;
+
+	/* Number of entries */
+	__u32 in_sync_count;
+	__u32 out_sync_count;
+
+	/* in_sync on render stage */
+	__u32 flags;
+#define DRM_V3D_IN_SYNC_RCL			0x01
+};
+
 /* struct drm_v3d_extension - ioctl extensions
  *
  * Linked-list of generic extensions where the id identify which struct is
@@ -70,6 +106,7 @@ struct drm_v3d_extension {
 	__u64 next;
 	__u64 ext_data;
 	__u32 id;
+#define DRM_V3D_EXT_ID_MULTI_SYNC		0x01
 };
 
 /**
@@ -228,6 +265,7 @@ enum drm_v3d_param {
 	DRM_V3D_PARAM_SUPPORTS_CSD,
 	DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH,
 	DRM_V3D_PARAM_SUPPORTS_PERFMON,
+	DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT,
 };
 
 struct drm_v3d_get_param {
-- 
2.30.2


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 3/3] drm/v3d: add multiple syncobjs support

[Iago Toral]

I think this looks good overall, I have a few questions/commments
below:

On Wed, 2021-08-18 at 18:57 +0100, Melissa Wen wrote:
> Using the generic extension support set in the previous patch, this
> patch enables more than one in/out binary syncobj per job submission.
> Arrays of syncobjs are set in a specific extension type (multisync)
> that also cares of determining the stage for sync (bin/render)
> through a flag - when this is the case.
> 
> Signed-off-by: Melissa Wen <mwen@igalia.com>
> ---
>  drivers/gpu/drm/v3d/v3d_drv.c |   3 +
>  drivers/gpu/drm/v3d/v3d_drv.h |  14 +++
>  drivers/gpu/drm/v3d/v3d_gem.c | 209 +++++++++++++++++++++++++++-----
> --
>  include/uapi/drm/v3d_drm.h    |  38 +++++++
>  4 files changed, 226 insertions(+), 38 deletions(-)
> 
> diff --git a/drivers/gpu/drm/v3d/v3d_drv.c
> b/drivers/gpu/drm/v3d/v3d_drv.c
> index 6a0516160bb2..939ca8c833f5 100644
> --- a/drivers/gpu/drm/v3d/v3d_drv.c
> +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> @@ -96,6 +96,9 @@ static int v3d_get_param_ioctl(struct drm_device
> *dev, void *data,
>  	case DRM_V3D_PARAM_SUPPORTS_PERFMON:
>  		args->value = (v3d->ver >= 40);
>  		return 0;
> +	case DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT:
> +		args->value = 1;
> +		return 0;
>  	default:
>  		DRM_DEBUG("Unknown parameter %d\n", args->param);
>  		return -EINVAL;
> diff --git a/drivers/gpu/drm/v3d/v3d_drv.h
> b/drivers/gpu/drm/v3d/v3d_drv.h
> index b900a050d5e2..544c60404a0f 100644
> --- a/drivers/gpu/drm/v3d/v3d_drv.h
> +++ b/drivers/gpu/drm/v3d/v3d_drv.h
> @@ -294,6 +294,20 @@ struct v3d_csd_job {
>  	struct drm_v3d_submit_csd args;
>  };
>  
> +struct v3d_submit_outsync {
> +	struct drm_syncobj *syncobj;
> +};
> +
> +struct v3d_submit_ext {
> +	u32 flags;
> +
> +	u32 in_sync_count;
> +	u64 in_syncs;
> +
> +	u32 out_sync_count;
> +	struct v3d_submit_outsync *out_syncs;
> +};
> +
>  /**
>   * __wait_for - magic wait macro
>   *
> diff --git a/drivers/gpu/drm/v3d/v3d_gem.c
> b/drivers/gpu/drm/v3d/v3d_gem.c
> index e254919b6c5e..e7aabe1a0e11 100644
> --- a/drivers/gpu/drm/v3d/v3d_gem.c
> +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> @@ -392,6 +392,9 @@ v3d_render_job_free(struct kref *ref)
>  
>  void v3d_job_cleanup(struct v3d_job *job)
>  {
> +	if (!job)
> +		return;
> +
>  	drm_sched_job_cleanup(&job->base);
>  	v3d_job_put(job);
>  }
> @@ -451,10 +454,11 @@ v3d_job_add_deps(struct drm_file *file_priv,
> struct v3d_job *job,
>  static int
>  v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
>  	     struct v3d_job *job, void (*free)(struct kref *ref),
> -	     u32 in_sync, enum v3d_queue queue)
> +	     u32 in_sync, struct v3d_submit_ext *se, enum v3d_queue
> queue)
>  {
>  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
> -	int ret;
> +	bool has_multinsync = (se && se->in_sync_count);
> +	int ret, i;
>  
>  	job->v3d = v3d;
>  	job->free = free;
> @@ -463,14 +467,30 @@ v3d_job_init(struct v3d_dev *v3d, struct
> drm_file *file_priv,
>  	if (ret < 0)
>  		return ret;
>  
> -	ret = drm_sched_job_init(&job->base, &v3d_priv-
> >sched_entity[queue],
> -				 v3d_priv);
> +	ret = drm_sched_job_init(&job->base, &v3d_priv-
> >sched_entity[queue], v3d_priv);
>  	if (ret)
>  		goto fail;
>  
> -	ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
> -	if (ret)
> -		goto fail_job;
> +	if (has_multinsync && (se->flags == (queue == V3D_RENDER))) {

I think this is unnecessarily difficult to understand, I'd suggest to
code the condition more explicitly:

if (has_multisync &&
    (see->flags & DRM_V3D_IN_SYNC_RCL) &&
    queue ==  V3D_RENDER)

unless I am missing the point here :)

> +		struct drm_v3d_sem __user *handle = u64_to_user_ptr(se-
> >in_syncs);
> +
> +		for (i = 0; i < se->in_sync_count; i++) {
> +			struct drm_v3d_sem in;
> +
> +			ret = copy_from_user(&in, handle++, sizeof(
> in));
> +			if (ret) {
> +				DRM_DEBUG("Failed to copy wait dep
> handle.\n");
> +				goto fail_job;
> +			}
> +			ret = v3d_job_add_deps(file_priv, job,
> in.handle, 0);
> +			if (ret)
> +				goto fail_job;
> +		}
> +	} else if (!has_multinsync) {
> +		ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
> +		if (ret)
> +			goto fail_job;
> +	}
>  
>  	kref_init(&job->refcount);
>  
> @@ -500,6 +520,7 @@ v3d_attach_fences_and_unlock_reservation(struct
> drm_file *file_priv,
>  					 struct v3d_job *job,
>  					 struct ww_acquire_ctx
> *acquire_ctx,
>  					 u32 out_sync,
> +					 struct v3d_submit_ext *se,
>  					 struct dma_fence *done_fence)
>  {
>  	struct drm_syncobj *sync_out;
> @@ -514,6 +535,18 @@ v3d_attach_fences_and_unlock_reservation(struct
> drm_file *file_priv,
>  	drm_gem_unlock_reservations(job->bo, job->bo_count,
> acquire_ctx);
>  
>  	/* Update the return sync object for the job */
> +	/* If multiples semaphores is supported */
> +	if (se && se->out_sync_count) {
> +		for (i = 0; i < se->out_sync_count; i++) {
> +			drm_syncobj_replace_fence(se-
> >out_syncs[i].syncobj,
> +						  done_fence);
> +			drm_syncobj_put(se->out_syncs[i].syncobj);
> +		}
> +		kvfree(se->out_syncs);
> +		return;
> +	}
> +
> +	/* Single signal semaphore */
>  	sync_out = drm_syncobj_find(file_priv, out_sync);
>  	if (sync_out) {
>  		drm_syncobj_replace_fence(sync_out, done_fence);
> @@ -521,11 +554,93 @@ v3d_attach_fences_and_unlock_reservation(struct
> drm_file *file_priv,
>  	}
>  }
>  
> +static void
> +v3d_put_multisync_post_deps(struct v3d_submit_ext *se)
> +{
> +	unsigned int i;
> +
> +	for (i = 0; i < se->out_sync_count; i++)
> +		drm_syncobj_put(se->out_syncs[i].syncobj);
> +	kvfree(se->out_syncs);
> +}
> +
> +static int
> +v3d_get_multisync_post_deps(struct drm_file *file_priv,
> +			    struct v3d_submit_ext *se,
> +			    u32 count, u64 handles)
> +{
> +	struct drm_v3d_sem __user *post_deps;
> +	int i, ret;
> +
> +	se->out_syncs = (struct v3d_submit_outsync *)
> +			kvmalloc_array(count,
> +				       sizeof(struct
> v3d_submit_outsync),
> +				       GFP_KERNEL);
> +	if (!se->out_syncs)
> +		return -ENOMEM;
> +
> +	post_deps = u64_to_user_ptr(handles);
> +
> +	for (i = 0; i < count; i++) {
> +		struct drm_v3d_sem out;
> +
> +		ret = copy_from_user(&out, post_deps++, sizeof(out));
> +		if (ret) {
> +			DRM_DEBUG("Failed to copy post dep handles\n");
> +			goto fail;
> +		}
> +
> +		se->out_syncs[i].syncobj = drm_syncobj_find(file_priv,
> +							    out.handle)
> ;
> +		if (!se->out_syncs[i].syncobj) {
> +			ret = -EINVAL;
> +			goto fail;
> +		}
> +	}
> +	se->out_sync_count = count;
> +
> +	return 0;
> +
> +fail:
> +	for (i--; i >= 0; i--)
> +		drm_syncobj_put(se->out_syncs[i].syncobj);
> +	kvfree(se->out_syncs);
> +
> +	return ret;
> +}
> +
> +static int
> +v3d_get_multisync_submit_deps(struct drm_file *file_priv,
> +			      struct v3d_submit_ext *se,
> +			      u64 ext_data)
> +{
> +	struct drm_v3d_multi_sync multisync = {0};
> +	int ret;
> +
> +	ret = copy_from_user(&multisync, u64_to_user_ptr(ext_data),
> +			     sizeof(multisync));
> +	if (ret)
> +		return ret;
> +
> +	ret = v3d_get_multisync_post_deps(file_priv, se,
> multisync.out_sync_count,
> +					  multisync.out_syncs);
> +	if (ret)
> +		return ret;
> +
> +	se->in_sync_count = multisync.in_sync_count;
> +	se->in_syncs = multisync.in_syncs;
> +
> +	se->flags = multisync.flags;
> +
> +	return 0;
> +}
> +
>  static int
>  v3d_get_extensions(struct drm_file *file_priv,
> +		   struct v3d_submit_ext *se,
>  		   u32 ext_count, u64 ext_handles)
>  {
> -	int i;
> +	int i, ret;
>  	struct drm_v3d_extension __user *handles;
>  
>  	if (!ext_count)
> @@ -541,7 +656,12 @@ v3d_get_extensions(struct drm_file *file_priv,
>  		}
>  
>  		switch (ext.id) {
> -		case 0:
> +		case DRM_V3D_EXT_ID_MULTI_SYNC:
> +			ret = v3d_get_multisync_submit_deps(file_priv,
> se,
> +							    ext.ext_dat
> a);
> +			if (ret)
> +				return ret;
> +			break;
>  		default:
>  			DRM_DEBUG_DRIVER("Unknown extension id: %d\n",
> ext.id);
>  			return -EINVAL;
> @@ -572,6 +692,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> *data,
>  	struct v3d_dev *v3d = to_v3d_dev(dev);
>  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
>  	struct drm_v3d_submit_cl *args = data;
> +	struct v3d_submit_ext se = {0};
>  	struct v3d_bin_job *bin = NULL;
>  	struct v3d_render_job *render;
>  	struct v3d_job *clean_job = NULL;
> @@ -589,7 +710,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> *data,
>  	}
>  
>  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> -		ret = v3d_get_extensions(file_priv,
> +		ret = v3d_get_extensions(file_priv, &se,
>  					 args->extension_count,
>  					 args->extensions);
>  		if (ret) {
> @@ -599,33 +720,35 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> void *data,
>  	}
>  
>  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
> -	if (!render)
> +	if (!render) {
> +		v3d_put_multisync_post_deps(&se);
>  		return -ENOMEM;
> +	}
>  
>  	render->start = args->rcl_start;
>  	render->end = args->rcl_end;
>  	INIT_LIST_HEAD(&render->unref_list);
>  
> -	ret = v3d_job_init(v3d, file_priv, &render->base,
> -			   v3d_render_job_free, args->in_sync_rcl,
> V3D_RENDER);
> +	ret = v3d_job_init(v3d, file_priv, &render->base,
> v3d_render_job_free,
> +			   args->in_sync_rcl, &se, V3D_RENDER);
>  	if (ret) {
>  		kfree(render);
> +		v3d_put_multisync_post_deps(&se);
>  		return ret;
>  	}
>  
>  	if (args->bcl_start != args->bcl_end) {
>  		bin = kcalloc(1, sizeof(*bin), GFP_KERNEL);
>  		if (!bin) {
> -			v3d_job_put(&render->base);
> -			return -ENOMEM;
> +			ret = -ENOMEM;
> +			goto fail;

We are now calling v3d_job_cleanup where we would call v3d_job_put
before for error codepaths. Is it safe to call this if we fail job
creation before calling drm_sched_job_init like here?

The documentations says that this cleans up the resources allocated
with drm_sched_job_init, which he haven't called yet.

>  		}
>  
>  		ret = v3d_job_init(v3d, file_priv, &bin->base,
> -				   v3d_job_free, args->in_sync_bcl,
> V3D_BIN);
> +				   v3d_job_free, args->in_sync_bcl,
> &se, V3D_BIN);
>  		if (ret) {
> -			v3d_job_put(&render->base);
>  			kfree(bin);
> -			return ret;
> +			goto fail;
>  		}
>  
>  		bin->start = args->bcl_start;
> @@ -643,7 +766,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> *data,
>  			goto fail;
>  		}
>  
> -		ret = v3d_job_init(v3d, file_priv, clean_job,
> v3d_job_free, 0, V3D_CACHE_CLEAN);
> +		ret = v3d_job_init(v3d, file_priv, clean_job,
> v3d_job_free, 0, 0, V3D_CACHE_CLEAN);
>  		if (ret) {
>  			kfree(clean_job);
>  			clean_job = NULL;
> @@ -706,6 +829,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> *data,
>  						 last_job,
>  						 &acquire_ctx,
>  						 args->out_sync,
> +						 &se,
>  						 last_job->done_fence);
>  
>  	if (bin)
> @@ -721,11 +845,10 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> void *data,
>  	drm_gem_unlock_reservations(last_job->bo,
>  				    last_job->bo_count, &acquire_ctx);
>  fail:
> -	if (bin)
> -		v3d_job_cleanup(&bin->base);
> +	v3d_job_cleanup(&bin->base);
>  	v3d_job_cleanup(&render->base);
> -	if (clean_job)
> -		v3d_job_cleanup(clean_job);
> +	v3d_job_cleanup(clean_job);
> +	v3d_put_multisync_post_deps(&se);
>  
>  	return ret;
>  }
> @@ -745,6 +868,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> *data,
>  {
>  	struct v3d_dev *v3d = to_v3d_dev(dev);
>  	struct drm_v3d_submit_tfu *args = data;
> +	struct v3d_submit_ext se = {0};
>  	struct v3d_tfu_job *job;
>  	struct ww_acquire_ctx acquire_ctx;
>  	int ret = 0;
> @@ -757,7 +881,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> *data,
>  	}
>  
>  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> -		ret = v3d_get_extensions(file_priv,
> +		ret = v3d_get_extensions(file_priv, &se,
>  					 args->extension_count,
>  					 args->extensions);
>  		if (ret) {
> @@ -767,21 +891,24 @@ v3d_submit_tfu_ioctl(struct drm_device *dev,
> void *data,
>  	}
>  
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> -	if (!job)
> +	if (!job) {
> +		v3d_put_multisync_post_deps(&se);
>  		return -ENOMEM;
> +	}
>  
>  	ret = v3d_job_init(v3d, file_priv, &job->base,
> -			   v3d_job_free, args->in_sync, V3D_TFU);
> +			   v3d_job_free, args->in_sync, &se, V3D_TFU);
>  	if (ret) {
>  		kfree(job);
> +		v3d_put_multisync_post_deps(&se);
>  		return ret;
>  	}
>  
>  	job->base.bo = kcalloc(ARRAY_SIZE(args->bo_handles),
>  			       sizeof(*job->base.bo), GFP_KERNEL);
>  	if (!job->base.bo) {
> -		v3d_job_put(&job->base);
> -		return -ENOMEM;
> +		ret = -ENOMEM;
> +		goto fail;
>  	}
>  
>  	job->args = *args;
> @@ -821,6 +948,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> *data,
>  	v3d_attach_fences_and_unlock_reservation(file_priv,
>  						 &job->base,
> &acquire_ctx,
>  						 args->out_sync,
> +						 &se,
>  						 job->base.done_fence);
>  
>  	v3d_job_put(&job->base);
> @@ -829,6 +957,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> *data,
>  
>  fail:
>  	v3d_job_cleanup(&job->base);
> +	v3d_put_multisync_post_deps(&se);
>  
>  	return ret;
>  }
> @@ -849,8 +978,9 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void
> *data,
>  	struct v3d_dev *v3d = to_v3d_dev(dev);
>  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
>  	struct drm_v3d_submit_csd *args = data;
> +	struct v3d_submit_ext se = {0};
>  	struct v3d_csd_job *job;
> -	struct v3d_job *clean_job;
> +	struct v3d_job *clean_job = NULL;
>  	struct ww_acquire_ctx acquire_ctx;
>  	int ret;
>  
> @@ -867,7 +997,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void
> *data,
>  	}
>  
>  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> -		ret = v3d_get_extensions(file_priv,
> +		ret = v3d_get_extensions(file_priv, &se,
>  					 args->extension_count,
>  					 args->extensions);
>  		if (ret) {
> @@ -877,28 +1007,29 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> void *data,
>  	}
>  
>  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> -	if (!job)
> +	if (!job) {
> +		v3d_put_multisync_post_deps(&se);
>  		return -ENOMEM;
> +	}
>  
>  	ret = v3d_job_init(v3d, file_priv, &job->base,
> -			   v3d_job_free, args->in_sync, V3D_CSD);
> +			   v3d_job_free, args->in_sync, &se, V3D_CSD);
>  	if (ret) {
>  		kfree(job);
> +		v3d_put_multisync_post_deps(&se);
>  		return ret;
>  	}
>  
>  	clean_job = kcalloc(1, sizeof(*clean_job), GFP_KERNEL);
>  	if (!clean_job) {
> -		v3d_job_put(&job->base);
> -		kfree(job);
> -		return -ENOMEM;
> +		ret = -ENOMEM;
> +		goto fail;
>  	}
>  
> -	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0,
> V3D_CACHE_CLEAN);
> +	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0,
> 0, V3D_CACHE_CLEAN);
>  	if (ret) {
> -		v3d_job_put(&job->base);
>  		kfree(clean_job);
> -		return ret;
> +		goto fail;
>  	}
>  
>  	job->args = *args;
> @@ -936,6 +1067,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> void *data,
>  						 clean_job,
>  						 &acquire_ctx,
>  						 args->out_sync,
> +						 &se,
>  						 clean_job-
> >done_fence);
>  
>  	v3d_job_put(&job->base);
> @@ -950,6 +1082,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> void *data,
>  fail:
>  	v3d_job_cleanup(&job->base);
>  	v3d_job_cleanup(clean_job);
> +	v3d_put_multisync_post_deps(&se);
>  
>  	return ret;
>  }
> diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> index 1f4706010eb5..bbb904c521b4 100644
> --- a/include/uapi/drm/v3d_drm.h
> +++ b/include/uapi/drm/v3d_drm.h
> @@ -60,6 +60,42 @@ extern "C" {
>  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
>  #define DRM_V3D_SUBMIT_EXTENSION		  0x02
>  
> +/* struct drm_v3d_sem - wait/signal semaphore
> + *
> + * If binary semaphore, it only takes syncobj handle and ignores
> flags and
> + * point fields. Point is defined for timeline syncobj feature.
> + */
> +struct drm_v3d_sem {
> +	__u32 handle; /* syncobj */
> +	/* rsv below, for future uses */
> +	__u32 flags;
> +	__u64 point;  /* for timeline sem support */
> +	__u64 mbz[2]; /* must be zero, rsv */
> +};

I guess the idea here would be that we would check handle and/or point
for whether they have a valid value to decide what type of semaphore
this is, right?

> +
> +/**
> + * struct drm_v3d_multi_sync - ioctl extension to add support
> multiples
> + * syncobjs for commands submission.
> + *
> + * When an extension of DRM_V3D_EXT_ID_MULTI_SYNC id is defined, it
> points to
> + * this extension to define wait and signal dependencies, instead of
> single
> + * in/out sync entries on submitting commands. The field flags is
> used to
> + * determine the stage to set wait dependencies.
> + */
> +struct drm_v3d_multi_sync {
> +	/* Array of wait and signal semaphores */
> +	__u64 in_syncs;
> +	__u64 out_syncs;
> +
> +	/* Number of entries */
> +	__u32 in_sync_count;
> +	__u32 out_sync_count;
> +
> +	/* in_sync on render stage */
> +	__u32 flags;
> +#define DRM_V3D_IN_SYNC_RCL			0x01
> +};
> +
>  /* struct drm_v3d_extension - ioctl extensions
>   *
>   * Linked-list of generic extensions where the id identify which
> struct is
> @@ -70,6 +106,7 @@ struct drm_v3d_extension {
>  	__u64 next;
>  	__u64 ext_data;
>  	__u32 id;
> +#define DRM_V3D_EXT_ID_MULTI_SYNC		0x01
>  };
>  
>  /**
> @@ -228,6 +265,7 @@ enum drm_v3d_param {
>  	DRM_V3D_PARAM_SUPPORTS_CSD,
>  	DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH,
>  	DRM_V3D_PARAM_SUPPORTS_PERFMON,
> +	DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT,
>  };
>  
>  struct drm_v3d_get_param {

Re: [PATCH 3/3] drm/v3d: add multiple syncobjs support

[Melissa Wen]

[-- Attachment #1: Type: text/plain, Size: 20330 bytes --]

On 09/16, Iago Toral wrote:
> I think this looks good overall, I have a few questions/commments
> below:
> 
> On Wed, 2021-08-18 at 18:57 +0100, Melissa Wen wrote:
> > Using the generic extension support set in the previous patch, this
> > patch enables more than one in/out binary syncobj per job submission.
> > Arrays of syncobjs are set in a specific extension type (multisync)
> > that also cares of determining the stage for sync (bin/render)
> > through a flag - when this is the case.
> > 
> > Signed-off-by: Melissa Wen <mwen@igalia.com>
> > ---
> >  drivers/gpu/drm/v3d/v3d_drv.c |   3 +
> >  drivers/gpu/drm/v3d/v3d_drv.h |  14 +++
> >  drivers/gpu/drm/v3d/v3d_gem.c | 209 +++++++++++++++++++++++++++-----
> > --
> >  include/uapi/drm/v3d_drm.h    |  38 +++++++
> >  4 files changed, 226 insertions(+), 38 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.c
> > b/drivers/gpu/drm/v3d/v3d_drv.c
> > index 6a0516160bb2..939ca8c833f5 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.c
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.c
> > @@ -96,6 +96,9 @@ static int v3d_get_param_ioctl(struct drm_device
> > *dev, void *data,
> >  	case DRM_V3D_PARAM_SUPPORTS_PERFMON:
> >  		args->value = (v3d->ver >= 40);
> >  		return 0;
> > +	case DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT:
> > +		args->value = 1;
> > +		return 0;
> >  	default:
> >  		DRM_DEBUG("Unknown parameter %d\n", args->param);
> >  		return -EINVAL;
> > diff --git a/drivers/gpu/drm/v3d/v3d_drv.h
> > b/drivers/gpu/drm/v3d/v3d_drv.h
> > index b900a050d5e2..544c60404a0f 100644
> > --- a/drivers/gpu/drm/v3d/v3d_drv.h
> > +++ b/drivers/gpu/drm/v3d/v3d_drv.h
> > @@ -294,6 +294,20 @@ struct v3d_csd_job {
> >  	struct drm_v3d_submit_csd args;
> >  };
> >  
> > +struct v3d_submit_outsync {
> > +	struct drm_syncobj *syncobj;
> > +};
> > +
> > +struct v3d_submit_ext {
> > +	u32 flags;
> > +
> > +	u32 in_sync_count;
> > +	u64 in_syncs;
> > +
> > +	u32 out_sync_count;
> > +	struct v3d_submit_outsync *out_syncs;
> > +};
> > +
> >  /**
> >   * __wait_for - magic wait macro
> >   *
> > diff --git a/drivers/gpu/drm/v3d/v3d_gem.c
> > b/drivers/gpu/drm/v3d/v3d_gem.c
> > index e254919b6c5e..e7aabe1a0e11 100644
> > --- a/drivers/gpu/drm/v3d/v3d_gem.c
> > +++ b/drivers/gpu/drm/v3d/v3d_gem.c
> > @@ -392,6 +392,9 @@ v3d_render_job_free(struct kref *ref)
> >  
> >  void v3d_job_cleanup(struct v3d_job *job)
> >  {
> > +	if (!job)
> > +		return;
> > +
> >  	drm_sched_job_cleanup(&job->base);
> >  	v3d_job_put(job);
> >  }
> > @@ -451,10 +454,11 @@ v3d_job_add_deps(struct drm_file *file_priv,
> > struct v3d_job *job,
> >  static int
> >  v3d_job_init(struct v3d_dev *v3d, struct drm_file *file_priv,
> >  	     struct v3d_job *job, void (*free)(struct kref *ref),
> > -	     u32 in_sync, enum v3d_queue queue)
> > +	     u32 in_sync, struct v3d_submit_ext *se, enum v3d_queue
> > queue)
> >  {
> >  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
> > -	int ret;
> > +	bool has_multinsync = (se && se->in_sync_count);
> > +	int ret, i;
> >  
> >  	job->v3d = v3d;
> >  	job->free = free;
> > @@ -463,14 +467,30 @@ v3d_job_init(struct v3d_dev *v3d, struct
> > drm_file *file_priv,
> >  	if (ret < 0)
> >  		return ret;
> >  
> > -	ret = drm_sched_job_init(&job->base, &v3d_priv-
> > >sched_entity[queue],
> > -				 v3d_priv);
> > +	ret = drm_sched_job_init(&job->base, &v3d_priv-
> > >sched_entity[queue], v3d_priv);
> >  	if (ret)
> >  		goto fail;
> >  
> > -	ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
> > -	if (ret)
> > -		goto fail_job;
> > +	if (has_multinsync && (se->flags == (queue == V3D_RENDER))) {
> 
> I think this is unnecessarily difficult to understand, I'd suggest to
> code the condition more explicitly:
> 
> if (has_multisync &&
>     (see->flags & DRM_V3D_IN_SYNC_RCL) &&
>     queue ==  V3D_RENDER)
> 
> unless I am missing the point here :)

more or less it.. the condition should be true if (IN_SYNC_RCL flag and
queue is V3_RENDER) and also if (no IN_SYNC_RCL and queue is not
V3_RENDER).

anyway, you are right it needs to be improved and more understable, I
will work on it for the next version.

> 
> > +		struct drm_v3d_sem __user *handle = u64_to_user_ptr(se-
> > >in_syncs);
> > +
> > +		for (i = 0; i < se->in_sync_count; i++) {
> > +			struct drm_v3d_sem in;
> > +
> > +			ret = copy_from_user(&in, handle++, sizeof(
> > in));
> > +			if (ret) {
> > +				DRM_DEBUG("Failed to copy wait dep
> > handle.\n");
> > +				goto fail_job;
> > +			}
> > +			ret = v3d_job_add_deps(file_priv, job,
> > in.handle, 0);
> > +			if (ret)
> > +				goto fail_job;
> > +		}
> > +	} else if (!has_multinsync) {
> > +		ret = v3d_job_add_deps(file_priv, job, in_sync, 0);
> > +		if (ret)
> > +			goto fail_job;
> > +	}
> >  
> >  	kref_init(&job->refcount);
> >  
> > @@ -500,6 +520,7 @@ v3d_attach_fences_and_unlock_reservation(struct
> > drm_file *file_priv,
> >  					 struct v3d_job *job,
> >  					 struct ww_acquire_ctx
> > *acquire_ctx,
> >  					 u32 out_sync,
> > +					 struct v3d_submit_ext *se,
> >  					 struct dma_fence *done_fence)
> >  {
> >  	struct drm_syncobj *sync_out;
> > @@ -514,6 +535,18 @@ v3d_attach_fences_and_unlock_reservation(struct
> > drm_file *file_priv,
> >  	drm_gem_unlock_reservations(job->bo, job->bo_count,
> > acquire_ctx);
> >  
> >  	/* Update the return sync object for the job */
> > +	/* If multiples semaphores is supported */
> > +	if (se && se->out_sync_count) {
> > +		for (i = 0; i < se->out_sync_count; i++) {
> > +			drm_syncobj_replace_fence(se-
> > >out_syncs[i].syncobj,
> > +						  done_fence);
> > +			drm_syncobj_put(se->out_syncs[i].syncobj);
> > +		}
> > +		kvfree(se->out_syncs);
> > +		return;
> > +	}
> > +
> > +	/* Single signal semaphore */
> >  	sync_out = drm_syncobj_find(file_priv, out_sync);
> >  	if (sync_out) {
> >  		drm_syncobj_replace_fence(sync_out, done_fence);
> > @@ -521,11 +554,93 @@ v3d_attach_fences_and_unlock_reservation(struct
> > drm_file *file_priv,
> >  	}
> >  }
> >  
> > +static void
> > +v3d_put_multisync_post_deps(struct v3d_submit_ext *se)
> > +{
> > +	unsigned int i;
> > +
> > +	for (i = 0; i < se->out_sync_count; i++)
> > +		drm_syncobj_put(se->out_syncs[i].syncobj);
> > +	kvfree(se->out_syncs);
> > +}
> > +
> > +static int
> > +v3d_get_multisync_post_deps(struct drm_file *file_priv,
> > +			    struct v3d_submit_ext *se,
> > +			    u32 count, u64 handles)
> > +{
> > +	struct drm_v3d_sem __user *post_deps;
> > +	int i, ret;
> > +
> > +	se->out_syncs = (struct v3d_submit_outsync *)
> > +			kvmalloc_array(count,
> > +				       sizeof(struct
> > v3d_submit_outsync),
> > +				       GFP_KERNEL);
> > +	if (!se->out_syncs)
> > +		return -ENOMEM;
> > +
> > +	post_deps = u64_to_user_ptr(handles);
> > +
> > +	for (i = 0; i < count; i++) {
> > +		struct drm_v3d_sem out;
> > +
> > +		ret = copy_from_user(&out, post_deps++, sizeof(out));
> > +		if (ret) {
> > +			DRM_DEBUG("Failed to copy post dep handles\n");
> > +			goto fail;
> > +		}
> > +
> > +		se->out_syncs[i].syncobj = drm_syncobj_find(file_priv,
> > +							    out.handle)
> > ;
> > +		if (!se->out_syncs[i].syncobj) {
> > +			ret = -EINVAL;
> > +			goto fail;
> > +		}
> > +	}
> > +	se->out_sync_count = count;
> > +
> > +	return 0;
> > +
> > +fail:
> > +	for (i--; i >= 0; i--)
> > +		drm_syncobj_put(se->out_syncs[i].syncobj);
> > +	kvfree(se->out_syncs);
> > +
> > +	return ret;
> > +}
> > +
> > +static int
> > +v3d_get_multisync_submit_deps(struct drm_file *file_priv,
> > +			      struct v3d_submit_ext *se,
> > +			      u64 ext_data)
> > +{
> > +	struct drm_v3d_multi_sync multisync = {0};
> > +	int ret;
> > +
> > +	ret = copy_from_user(&multisync, u64_to_user_ptr(ext_data),
> > +			     sizeof(multisync));
> > +	if (ret)
> > +		return ret;
> > +
> > +	ret = v3d_get_multisync_post_deps(file_priv, se,
> > multisync.out_sync_count,
> > +					  multisync.out_syncs);
> > +	if (ret)
> > +		return ret;
> > +
> > +	se->in_sync_count = multisync.in_sync_count;
> > +	se->in_syncs = multisync.in_syncs;
> > +
> > +	se->flags = multisync.flags;
> > +
> > +	return 0;
> > +}
> > +
> >  static int
> >  v3d_get_extensions(struct drm_file *file_priv,
> > +		   struct v3d_submit_ext *se,
> >  		   u32 ext_count, u64 ext_handles)
> >  {
> > -	int i;
> > +	int i, ret;
> >  	struct drm_v3d_extension __user *handles;
> >  
> >  	if (!ext_count)
> > @@ -541,7 +656,12 @@ v3d_get_extensions(struct drm_file *file_priv,
> >  		}
> >  
> >  		switch (ext.id) {
> > -		case 0:
> > +		case DRM_V3D_EXT_ID_MULTI_SYNC:
> > +			ret = v3d_get_multisync_submit_deps(file_priv,
> > se,
> > +							    ext.ext_dat
> > a);
> > +			if (ret)
> > +				return ret;
> > +			break;
> >  		default:
> >  			DRM_DEBUG_DRIVER("Unknown extension id: %d\n",
> > ext.id);
> >  			return -EINVAL;
> > @@ -572,6 +692,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> > *data,
> >  	struct v3d_dev *v3d = to_v3d_dev(dev);
> >  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
> >  	struct drm_v3d_submit_cl *args = data;
> > +	struct v3d_submit_ext se = {0};
> >  	struct v3d_bin_job *bin = NULL;
> >  	struct v3d_render_job *render;
> >  	struct v3d_job *clean_job = NULL;
> > @@ -589,7 +710,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> > *data,
> >  	}
> >  
> >  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > -		ret = v3d_get_extensions(file_priv,
> > +		ret = v3d_get_extensions(file_priv, &se,
> >  					 args->extension_count,
> >  					 args->extensions);
> >  		if (ret) {
> > @@ -599,33 +720,35 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> > void *data,
> >  	}
> >  
> >  	render = kcalloc(1, sizeof(*render), GFP_KERNEL);
> > -	if (!render)
> > +	if (!render) {
> > +		v3d_put_multisync_post_deps(&se);
> >  		return -ENOMEM;
> > +	}
> >  
> >  	render->start = args->rcl_start;
> >  	render->end = args->rcl_end;
> >  	INIT_LIST_HEAD(&render->unref_list);
> >  
> > -	ret = v3d_job_init(v3d, file_priv, &render->base,
> > -			   v3d_render_job_free, args->in_sync_rcl,
> > V3D_RENDER);
> > +	ret = v3d_job_init(v3d, file_priv, &render->base,
> > v3d_render_job_free,
> > +			   args->in_sync_rcl, &se, V3D_RENDER);
> >  	if (ret) {
> >  		kfree(render);
> > +		v3d_put_multisync_post_deps(&se);
> >  		return ret;
> >  	}
> >  
> >  	if (args->bcl_start != args->bcl_end) {
> >  		bin = kcalloc(1, sizeof(*bin), GFP_KERNEL);
> >  		if (!bin) {
> > -			v3d_job_put(&render->base);
> > -			return -ENOMEM;
> > +			ret = -ENOMEM;
> > +			goto fail;
> 
> We are now calling v3d_job_cleanup where we would call v3d_job_put
> before for error codepaths. Is it safe to call this if we fail job
> creation before calling drm_sched_job_init like here?
> 
> The documentations says that this cleans up the resources allocated
> with drm_sched_job_init, which he haven't called yet.

Thanks for pointing it out. It took me a while to find the loose end,
but I just realized the problem is in the current implementation.

At this point, bin is null, but render is between drm_sched_job_init and
drm_sched_job_arm; so, if the bin job allocation fails, we still need to
cleanup resources already allocated to the render job, because render
was already initialized (v3d_job_init -> drm_sched_job_init).
I'll send a fix for it.

For the bin case, I've added a check in v3d_cleanup_job() to return if
the @job is null (no cleanup or put for this job). But I should also
check if bin is null in the 'fail:' section (as it was previously) to
avoid unexpected behavior.

> 
> >  		}
> >  
> >  		ret = v3d_job_init(v3d, file_priv, &bin->base,
> > -				   v3d_job_free, args->in_sync_bcl,
> > V3D_BIN);
> > +				   v3d_job_free, args->in_sync_bcl,
> > &se, V3D_BIN);
> >  		if (ret) {
> > -			v3d_job_put(&render->base);
> >  			kfree(bin);
> > -			return ret;
> > +			goto fail;
> >  		}
> >  
> >  		bin->start = args->bcl_start;
> > @@ -643,7 +766,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> > *data,
> >  			goto fail;
> >  		}
> >  
> > -		ret = v3d_job_init(v3d, file_priv, clean_job,
> > v3d_job_free, 0, V3D_CACHE_CLEAN);
> > +		ret = v3d_job_init(v3d, file_priv, clean_job,
> > v3d_job_free, 0, 0, V3D_CACHE_CLEAN);
> >  		if (ret) {
> >  			kfree(clean_job);
> >  			clean_job = NULL;
> > @@ -706,6 +829,7 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void
> > *data,
> >  						 last_job,
> >  						 &acquire_ctx,
> >  						 args->out_sync,
> > +						 &se,
> >  						 last_job->done_fence);
> >  
> >  	if (bin)
> > @@ -721,11 +845,10 @@ v3d_submit_cl_ioctl(struct drm_device *dev,
> > void *data,
> >  	drm_gem_unlock_reservations(last_job->bo,
> >  				    last_job->bo_count, &acquire_ctx);
> >  fail:
> > -	if (bin)
> > -		v3d_job_cleanup(&bin->base);
> > +	v3d_job_cleanup(&bin->base);
> >  	v3d_job_cleanup(&render->base);
> > -	if (clean_job)
> > -		v3d_job_cleanup(clean_job);
> > +	v3d_job_cleanup(clean_job);
> > +	v3d_put_multisync_post_deps(&se);
> >  
> >  	return ret;
> >  }
> > @@ -745,6 +868,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> > *data,
> >  {
> >  	struct v3d_dev *v3d = to_v3d_dev(dev);
> >  	struct drm_v3d_submit_tfu *args = data;
> > +	struct v3d_submit_ext se = {0};
> >  	struct v3d_tfu_job *job;
> >  	struct ww_acquire_ctx acquire_ctx;
> >  	int ret = 0;
> > @@ -757,7 +881,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> > *data,
> >  	}
> >  
> >  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > -		ret = v3d_get_extensions(file_priv,
> > +		ret = v3d_get_extensions(file_priv, &se,
> >  					 args->extension_count,
> >  					 args->extensions);
> >  		if (ret) {
> > @@ -767,21 +891,24 @@ v3d_submit_tfu_ioctl(struct drm_device *dev,
> > void *data,
> >  	}
> >  
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> > -	if (!job)
> > +	if (!job) {
> > +		v3d_put_multisync_post_deps(&se);
> >  		return -ENOMEM;
> > +	}
> >  
> >  	ret = v3d_job_init(v3d, file_priv, &job->base,
> > -			   v3d_job_free, args->in_sync, V3D_TFU);
> > +			   v3d_job_free, args->in_sync, &se, V3D_TFU);
> >  	if (ret) {
> >  		kfree(job);
> > +		v3d_put_multisync_post_deps(&se);
> >  		return ret;
> >  	}
> >  
> >  	job->base.bo = kcalloc(ARRAY_SIZE(args->bo_handles),
> >  			       sizeof(*job->base.bo), GFP_KERNEL);
> >  	if (!job->base.bo) {
> > -		v3d_job_put(&job->base);
> > -		return -ENOMEM;
> > +		ret = -ENOMEM;
> > +		goto fail;
> >  	}
> >  
> >  	job->args = *args;
> > @@ -821,6 +948,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> > *data,
> >  	v3d_attach_fences_and_unlock_reservation(file_priv,
> >  						 &job->base,
> > &acquire_ctx,
> >  						 args->out_sync,
> > +						 &se,
> >  						 job->base.done_fence);
> >  
> >  	v3d_job_put(&job->base);
> > @@ -829,6 +957,7 @@ v3d_submit_tfu_ioctl(struct drm_device *dev, void
> > *data,
> >  
> >  fail:
> >  	v3d_job_cleanup(&job->base);
> > +	v3d_put_multisync_post_deps(&se);
> >  
> >  	return ret;
> >  }
> > @@ -849,8 +978,9 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void
> > *data,
> >  	struct v3d_dev *v3d = to_v3d_dev(dev);
> >  	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
> >  	struct drm_v3d_submit_csd *args = data;
> > +	struct v3d_submit_ext se = {0};
> >  	struct v3d_csd_job *job;
> > -	struct v3d_job *clean_job;
> > +	struct v3d_job *clean_job = NULL;
> >  	struct ww_acquire_ctx acquire_ctx;
> >  	int ret;
> >  
> > @@ -867,7 +997,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev, void
> > *data,
> >  	}
> >  
> >  	if (args->flags & DRM_V3D_SUBMIT_EXTENSION) {
> > -		ret = v3d_get_extensions(file_priv,
> > +		ret = v3d_get_extensions(file_priv, &se,
> >  					 args->extension_count,
> >  					 args->extensions);
> >  		if (ret) {
> > @@ -877,28 +1007,29 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> > void *data,
> >  	}
> >  
> >  	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
> > -	if (!job)
> > +	if (!job) {
> > +		v3d_put_multisync_post_deps(&se);
> >  		return -ENOMEM;
> > +	}
> >  
> >  	ret = v3d_job_init(v3d, file_priv, &job->base,
> > -			   v3d_job_free, args->in_sync, V3D_CSD);
> > +			   v3d_job_free, args->in_sync, &se, V3D_CSD);
> >  	if (ret) {
> >  		kfree(job);
> > +		v3d_put_multisync_post_deps(&se);
> >  		return ret;
> >  	}
> >  
> >  	clean_job = kcalloc(1, sizeof(*clean_job), GFP_KERNEL);
> >  	if (!clean_job) {
> > -		v3d_job_put(&job->base);
> > -		kfree(job);
> > -		return -ENOMEM;
> > +		ret = -ENOMEM;
> > +		goto fail;
> >  	}
> >  
> > -	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0,
> > V3D_CACHE_CLEAN);
> > +	ret = v3d_job_init(v3d, file_priv, clean_job, v3d_job_free, 0,
> > 0, V3D_CACHE_CLEAN);
> >  	if (ret) {
> > -		v3d_job_put(&job->base);
> >  		kfree(clean_job);
> > -		return ret;
> > +		goto fail;
> >  	}
> >  
> >  	job->args = *args;
> > @@ -936,6 +1067,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> > void *data,
> >  						 clean_job,
> >  						 &acquire_ctx,
> >  						 args->out_sync,
> > +						 &se,
> >  						 clean_job-
> > >done_fence);
> >  
> >  	v3d_job_put(&job->base);
> > @@ -950,6 +1082,7 @@ v3d_submit_csd_ioctl(struct drm_device *dev,
> > void *data,
> >  fail:
> >  	v3d_job_cleanup(&job->base);
> >  	v3d_job_cleanup(clean_job);
> > +	v3d_put_multisync_post_deps(&se);
> >  
> >  	return ret;
> >  }
> > diff --git a/include/uapi/drm/v3d_drm.h b/include/uapi/drm/v3d_drm.h
> > index 1f4706010eb5..bbb904c521b4 100644
> > --- a/include/uapi/drm/v3d_drm.h
> > +++ b/include/uapi/drm/v3d_drm.h
> > @@ -60,6 +60,42 @@ extern "C" {
> >  #define DRM_V3D_SUBMIT_CL_FLUSH_CACHE             0x01
> >  #define DRM_V3D_SUBMIT_EXTENSION		  0x02
> >  
> > +/* struct drm_v3d_sem - wait/signal semaphore
> > + *
> > + * If binary semaphore, it only takes syncobj handle and ignores
> > flags and
> > + * point fields. Point is defined for timeline syncobj feature.
> > + */
> > +struct drm_v3d_sem {
> > +	__u32 handle; /* syncobj */
> > +	/* rsv below, for future uses */
> > +	__u32 flags;
> > +	__u64 point;  /* for timeline sem support */
> > +	__u64 mbz[2]; /* must be zero, rsv */
> > +};
> 
> I guess the idea here would be that we would check handle and/or point
> for whether they have a valid value to decide what type of semaphore
> this is, right?

Yes, when syncobj timeline support is enabled, point will distinguish
whether it is a timeline or binary semaphore.

Thanks for reviewing, I'll address those points.

Melissa

> 
> > +
> > +/**
> > + * struct drm_v3d_multi_sync - ioctl extension to add support
> > multiples
> > + * syncobjs for commands submission.
> > + *
> > + * When an extension of DRM_V3D_EXT_ID_MULTI_SYNC id is defined, it
> > points to
> > + * this extension to define wait and signal dependencies, instead of
> > single
> > + * in/out sync entries on submitting commands. The field flags is
> > used to
> > + * determine the stage to set wait dependencies.
> > + */
> > +struct drm_v3d_multi_sync {
> > +	/* Array of wait and signal semaphores */
> > +	__u64 in_syncs;
> > +	__u64 out_syncs;
> > +
> > +	/* Number of entries */
> > +	__u32 in_sync_count;
> > +	__u32 out_sync_count;
> > +
> > +	/* in_sync on render stage */
> > +	__u32 flags;
> > +#define DRM_V3D_IN_SYNC_RCL			0x01
> > +};
> > +
> >  /* struct drm_v3d_extension - ioctl extensions
> >   *
> >   * Linked-list of generic extensions where the id identify which
> > struct is
> > @@ -70,6 +106,7 @@ struct drm_v3d_extension {
> >  	__u64 next;
> >  	__u64 ext_data;
> >  	__u32 id;
> > +#define DRM_V3D_EXT_ID_MULTI_SYNC		0x01
> >  };
> >  
> >  /**
> > @@ -228,6 +265,7 @@ enum drm_v3d_param {
> >  	DRM_V3D_PARAM_SUPPORTS_CSD,
> >  	DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH,
> >  	DRM_V3D_PARAM_SUPPORTS_PERFMON,
> > +	DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT,
> >  };
> >  
> >  struct drm_v3d_get_param {
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]