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 {

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

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 {