[v3,2/2] drm/virtio: Support sync objects

Message ID 20230323190340.950875-3-dmitry.osipenko@collabora.com
State New
Headers
Series Add sync object UAPI support to VirtIO-GPU driver |

Commit Message

Dmitry Osipenko March 23, 2023, 7:03 p.m. UTC
  Add sync object DRM UAPI support to VirtIO-GPU driver. It's required
for enabling a full-featured Vulkan fencing by Venus and native context
VirtIO-GPU Mesa drivers.

Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
---
 drivers/gpu/drm/virtio/virtgpu_drv.c    |   3 +-
 drivers/gpu/drm/virtio/virtgpu_submit.c | 211 ++++++++++++++++++++++++
 include/uapi/drm/virtgpu_drm.h          |  16 +-
 3 files changed, 228 insertions(+), 2 deletions(-)
  

Comments

Rob Clark March 23, 2023, 9:18 p.m. UTC | #1
On Thu, Mar 23, 2023 at 12:05 PM Dmitry Osipenko
<dmitry.osipenko@collabora.com> wrote:
>
> Add sync object DRM UAPI support to VirtIO-GPU driver. It's required
> for enabling a full-featured Vulkan fencing by Venus and native context
> VirtIO-GPU Mesa drivers.
>
> Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
> ---
>  drivers/gpu/drm/virtio/virtgpu_drv.c    |   3 +-
>  drivers/gpu/drm/virtio/virtgpu_submit.c | 211 ++++++++++++++++++++++++
>  include/uapi/drm/virtgpu_drm.h          |  16 +-
>  3 files changed, 228 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/virtio/virtgpu_drv.c b/drivers/gpu/drm/virtio/virtgpu_drv.c
> index add075681e18..a22155577152 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_drv.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_drv.c
> @@ -176,7 +176,8 @@ static const struct drm_driver driver = {
>          * If KMS is disabled DRIVER_MODESET and DRIVER_ATOMIC are masked
>          * out via drm_device::driver_features:
>          */
> -       .driver_features = DRIVER_MODESET | DRIVER_GEM | DRIVER_RENDER | DRIVER_ATOMIC,
> +       .driver_features = DRIVER_MODESET | DRIVER_GEM | DRIVER_RENDER | DRIVER_ATOMIC |
> +                          DRIVER_SYNCOBJ | DRIVER_SYNCOBJ_TIMELINE,
>         .open = virtio_gpu_driver_open,
>         .postclose = virtio_gpu_driver_postclose,
>
> diff --git a/drivers/gpu/drm/virtio/virtgpu_submit.c b/drivers/gpu/drm/virtio/virtgpu_submit.c
> index 2ce2459c6bc2..9ea4390948bf 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_submit.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_submit.c
> @@ -14,11 +14,26 @@
>  #include <linux/uaccess.h>
>
>  #include <drm/drm_file.h>
> +#include <drm/drm_syncobj.h>
>  #include <drm/virtgpu_drm.h>
>
>  #include "virtgpu_drv.h"
>
> +struct virtio_gpu_submit_post_dep {
> +       struct drm_syncobj *syncobj;
> +       struct dma_fence_chain *chain;
> +       uint64_t point;
> +};
> +
>  struct virtio_gpu_submit {
> +       struct virtio_gpu_submit_post_dep *post_deps;
> +       unsigned int num_out_syncobjs;
> +
> +       struct drm_syncobj **in_syncobjs;
> +       unsigned int num_in_syncobjs;
> +       uint64_t *in_fence_ids;
> +       unsigned int num_in_fence_ids;
> +
>         struct virtio_gpu_object_array *buflist;
>         struct drm_virtgpu_execbuffer *exbuf;
>         struct virtio_gpu_fence *out_fence;
> @@ -58,6 +73,186 @@ static int virtio_gpu_dma_fence_wait(struct virtio_gpu_submit *submit,
>         return 0;
>  }
>
> +static void virtio_gpu_free_syncobjs(struct drm_syncobj **syncobjs,
> +                                    uint32_t nr_syncobjs)
> +{
> +       uint32_t i = nr_syncobjs;
> +
> +       while (syncobjs && i--) {

Checking syncobjs!=NULL here does at least look a bit funny, as the
condition doesn't change in the loop body.  It is not incorrect, it
protects you against the cleanup path where submit->in_syncobjs is
NULL.  But if (!syncobjs)\nreturn seems a bit more clear


> +               if (syncobjs[i])
> +                       drm_syncobj_put(syncobjs[i]);
> +       }
> +
> +       kfree(syncobjs);
> +}
> +
> +static int
> +virtio_gpu_parse_deps(struct virtio_gpu_submit *submit)
> +{
> +       struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
> +       struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
> +       size_t syncobj_stride = exbuf->syncobj_stride;
> +       struct drm_syncobj **syncobjs;
> +       int ret = 0, i;
> +
> +       if (!submit->num_in_syncobjs)
> +               return 0;
> +
> +       syncobjs = kcalloc(submit->num_in_syncobjs, sizeof(*syncobjs),
> +                          GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);

I *think*, assuming I'm reading where this is called correctly (kinda
wish git would show more lines of context by default) that these don't
need to be NOWARN|NORETRY (same for post_deps).  I guess you inherited
this from drm/msm, where I appear to have forgotten to update the
syncobj path in commit f0de40a131d9 ("drm/msm: Reorder lock vs submit
alloc").  I don't see anything obvious that would require NORETRY, but
lockdep should be able to tell you otherwise if needed..

BR,
-R

> +       if (!syncobjs)
> +               return -ENOMEM;
> +
> +       for (i = 0; i < submit->num_in_syncobjs; i++) {
> +               uint64_t address = exbuf->in_syncobjs + i * syncobj_stride;
> +               struct dma_fence *fence;
> +
> +               if (copy_from_user(&syncobj_desc,
> +                                  u64_to_user_ptr(address),
> +                                  min(syncobj_stride, sizeof(syncobj_desc)))) {
> +                       ret = -EFAULT;
> +                       break;
> +               }
> +
> +               if (syncobj_desc.flags & ~VIRTGPU_EXECBUF_SYNCOBJ_FLAGS) {
> +                       ret = -EINVAL;
> +                       break;
> +               }
> +
> +               ret = drm_syncobj_find_fence(submit->file, syncobj_desc.handle,
> +                                            syncobj_desc.point, 0, &fence);
> +               if (ret)
> +                       break;
> +
> +               ret = virtio_gpu_dma_fence_wait(submit, fence);
> +
> +               dma_fence_put(fence);
> +               if (ret)
> +                       break;
> +
> +               if (syncobj_desc.flags & VIRTGPU_EXECBUF_SYNCOBJ_RESET) {
> +                       syncobjs[i] =
> +                               drm_syncobj_find(submit->file, syncobj_desc.handle);
> +                       if (!syncobjs[i]) {
> +                               ret = -EINVAL;
> +                               break;
> +                       }
> +               }
> +       }
> +
> +       if (ret) {
> +               virtio_gpu_free_syncobjs(syncobjs, i);
> +               return ret;
> +       }
> +
> +       submit->in_syncobjs = syncobjs;
> +
> +       return ret;
> +}
> +
> +static void virtio_gpu_reset_syncobjs(struct drm_syncobj **syncobjs,
> +                                     uint32_t nr_syncobjs)
> +{
> +       uint32_t i;
> +
> +       for (i = 0; syncobjs && i < nr_syncobjs; i++) {
> +               if (syncobjs[i])
> +                       drm_syncobj_replace_fence(syncobjs[i], NULL);
> +       }
> +}
> +
> +static void
> +virtio_gpu_free_post_deps(struct virtio_gpu_submit_post_dep *post_deps,
> +                         uint32_t nr_syncobjs)
> +{
> +       uint32_t i = nr_syncobjs;
> +
> +       while (post_deps && i--) {
> +               kfree(post_deps[i].chain);
> +               drm_syncobj_put(post_deps[i].syncobj);
> +       }
> +
> +       kfree(post_deps);
> +}
> +
> +static int virtio_gpu_parse_post_deps(struct virtio_gpu_submit *submit)
> +{
> +       struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
> +       struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
> +       struct virtio_gpu_submit_post_dep *post_deps;
> +       size_t syncobj_stride = exbuf->syncobj_stride;
> +       int ret = 0, i;
> +
> +       if (!submit->num_out_syncobjs)
> +               return 0;
> +
> +       post_deps = kcalloc(submit->num_out_syncobjs, sizeof(*post_deps),
> +                           GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
> +       if (!post_deps)
> +               return -ENOMEM;
> +
> +       for (i = 0; i < submit->num_out_syncobjs; i++) {
> +               uint64_t address = exbuf->out_syncobjs + i * syncobj_stride;
> +
> +               if (copy_from_user(&syncobj_desc,
> +                                  u64_to_user_ptr(address),
> +                                  min(syncobj_stride, sizeof(syncobj_desc)))) {
> +                       ret = -EFAULT;
> +                       break;
> +               }
> +
> +               post_deps[i].point = syncobj_desc.point;
> +
> +               if (syncobj_desc.flags) {
> +                       ret = -EINVAL;
> +                       break;
> +               }
> +
> +               if (syncobj_desc.point) {
> +                       post_deps[i].chain = dma_fence_chain_alloc();
> +                       if (!post_deps[i].chain) {
> +                               ret = -ENOMEM;
> +                               break;
> +                       }
> +               }
> +
> +               post_deps[i].syncobj =
> +                       drm_syncobj_find(submit->file, syncobj_desc.handle);
> +               if (!post_deps[i].syncobj) {
> +                       ret = -EINVAL;
> +                       break;
> +               }
> +       }
> +
> +       if (ret) {
> +               virtio_gpu_free_post_deps(post_deps, i);
> +               return ret;
> +       }
> +
> +       submit->post_deps = post_deps;
> +
> +       return 0;
> +}
> +
> +static void
> +virtio_gpu_process_post_deps(struct virtio_gpu_submit *submit)
> +{
> +       struct virtio_gpu_submit_post_dep *post_deps = submit->post_deps;
> +       struct dma_fence *fence = &submit->out_fence->f;
> +       uint32_t i;
> +
> +       for (i = 0; post_deps && i < submit->num_out_syncobjs; i++) {
> +               if (post_deps[i].chain) {
> +                       drm_syncobj_add_point(post_deps[i].syncobj,
> +                                             post_deps[i].chain,
> +                                             fence, post_deps[i].point);
> +                       post_deps[i].chain = NULL;
> +               } else {
> +                       drm_syncobj_replace_fence(post_deps[i].syncobj, fence);
> +               }
> +       }
> +}
> +
>  static int virtio_gpu_fence_event_create(struct drm_device *dev,
>                                          struct drm_file *file,
>                                          struct virtio_gpu_fence *fence,
> @@ -121,6 +316,11 @@ static int virtio_gpu_init_submit_buflist(struct virtio_gpu_submit *submit)
>
>  static void virtio_gpu_cleanup_submit(struct virtio_gpu_submit *submit)
>  {
> +       virtio_gpu_reset_syncobjs(submit->in_syncobjs, submit->num_in_syncobjs);
> +       virtio_gpu_free_syncobjs(submit->in_syncobjs, submit->num_in_syncobjs);
> +       virtio_gpu_free_post_deps(submit->post_deps, submit->num_out_syncobjs);
> +       kfree(submit->in_fence_ids);
> +
>         if (!IS_ERR(submit->buf))
>                 kvfree(submit->buf);
>
> @@ -173,6 +373,8 @@ static int virtio_gpu_init_submit(struct virtio_gpu_submit *submit,
>                 return err;
>         }
>
> +       submit->num_out_syncobjs = exbuf->num_out_syncobjs;
> +       submit->num_in_syncobjs = exbuf->num_in_syncobjs;
>         submit->out_fence = out_fence;
>         submit->fence_ctx = fence_ctx;
>         submit->ring_idx = ring_idx;
> @@ -285,6 +487,14 @@ int virtio_gpu_execbuffer_ioctl(struct drm_device *dev, void *data,
>         if (ret)
>                 goto cleanup;
>
> +       ret = virtio_gpu_parse_deps(&submit);
> +       if (ret)
> +               goto cleanup;
> +
> +       ret = virtio_gpu_parse_post_deps(&submit);
> +       if (ret)
> +               goto cleanup;
> +
>         ret = virtio_gpu_install_out_fence_fd(&submit);
>         if (ret)
>                 goto cleanup;
> @@ -294,6 +504,7 @@ int virtio_gpu_execbuffer_ioctl(struct drm_device *dev, void *data,
>                 goto cleanup;
>
>         virtio_gpu_submit(&submit);
> +       virtio_gpu_process_post_deps(&submit);
>         virtio_gpu_complete_submit(&submit);
>  cleanup:
>         virtio_gpu_cleanup_submit(&submit);
> diff --git a/include/uapi/drm/virtgpu_drm.h b/include/uapi/drm/virtgpu_drm.h
> index 7b158fcb02b4..ce4948aacafd 100644
> --- a/include/uapi/drm/virtgpu_drm.h
> +++ b/include/uapi/drm/virtgpu_drm.h
> @@ -64,6 +64,16 @@ struct drm_virtgpu_map {
>         __u32 pad;
>  };
>
> +#define VIRTGPU_EXECBUF_SYNCOBJ_RESET          0x01
> +#define VIRTGPU_EXECBUF_SYNCOBJ_FLAGS ( \
> +               VIRTGPU_EXECBUF_SYNCOBJ_RESET | \
> +               0)
> +struct drm_virtgpu_execbuffer_syncobj {
> +       __u32 handle;
> +       __u32 flags;
> +       __u64 point;
> +};
> +
>  /* fence_fd is modified on success if VIRTGPU_EXECBUF_FENCE_FD_OUT flag is set. */
>  struct drm_virtgpu_execbuffer {
>         __u32 flags;
> @@ -73,7 +83,11 @@ struct drm_virtgpu_execbuffer {
>         __u32 num_bo_handles;
>         __s32 fence_fd; /* in/out fence fd (see VIRTGPU_EXECBUF_FENCE_FD_IN/OUT) */
>         __u32 ring_idx; /* command ring index (see VIRTGPU_EXECBUF_RING_IDX) */
> -       __u32 pad;
> +       __u32 syncobj_stride;
> +       __u32 num_in_syncobjs;
> +       __u32 num_out_syncobjs;
> +       __u64 in_syncobjs;
> +       __u64 out_syncobjs;
>  };
>
>  #define VIRTGPU_PARAM_3D_FEATURES 1 /* do we have 3D features in the hw */
> --
> 2.39.2
>
  
Dmitry Osipenko March 23, 2023, 9:51 p.m. UTC | #2
On 3/24/23 00:18, Rob Clark wrote:
...
>> +static int
>> +virtio_gpu_parse_deps(struct virtio_gpu_submit *submit)
>> +{
>> +       struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
>> +       struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
>> +       size_t syncobj_stride = exbuf->syncobj_stride;
>> +       struct drm_syncobj **syncobjs;
>> +       int ret = 0, i;
>> +
>> +       if (!submit->num_in_syncobjs)
>> +               return 0;
>> +
>> +       syncobjs = kcalloc(submit->num_in_syncobjs, sizeof(*syncobjs),
>> +                          GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
> I *think*, assuming I'm reading where this is called correctly (kinda
> wish git would show more lines of context by default) that these don't
> need to be NOWARN|NORETRY (same for post_deps).  I guess you inherited
> this from drm/msm, where I appear to have forgotten to update the
> syncobj path in commit f0de40a131d9 ("drm/msm: Reorder lock vs submit
> alloc").  I don't see anything obvious that would require NORETRY, but
> lockdep should be able to tell you otherwise if needed.

The NORETRY should prevent waking up OOM killer, it shouldn't help with
lockdep. Nothing prevents userspace from giving a big number of
num_in_syncobjs. But perhaps indeed not very practical to care about
this case, given that other similar memalloc paces of execbuffer_ioctl()
aren't using NORETRY. Alright, let's drop it in v4.
  
Dmitry Osipenko March 23, 2023, 9:59 p.m. UTC | #3
On 3/23/23 22:03, Dmitry Osipenko wrote:
>  struct virtio_gpu_submit {
> +	struct virtio_gpu_submit_post_dep *post_deps;
> +	unsigned int num_out_syncobjs;
> +
> +	struct drm_syncobj **in_syncobjs;
> +	unsigned int num_in_syncobjs;
> +	uint64_t *in_fence_ids;
> +	unsigned int num_in_fence_ids;

The in_fence_ids actually doesn't belong to this patch. Missed to remove
it on a rebase, will fix in v4.
  
Dmitry Osipenko March 23, 2023, 10:19 p.m. UTC | #4
On 3/24/23 00:51, Dmitry Osipenko wrote:
> On 3/24/23 00:18, Rob Clark wrote:
> ...
>>> +static int
>>> +virtio_gpu_parse_deps(struct virtio_gpu_submit *submit)
>>> +{
>>> +       struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
>>> +       struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
>>> +       size_t syncobj_stride = exbuf->syncobj_stride;
>>> +       struct drm_syncobj **syncobjs;
>>> +       int ret = 0, i;
>>> +
>>> +       if (!submit->num_in_syncobjs)
>>> +               return 0;
>>> +
>>> +       syncobjs = kcalloc(submit->num_in_syncobjs, sizeof(*syncobjs),
>>> +                          GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
>> I *think*, assuming I'm reading where this is called correctly (kinda
>> wish git would show more lines of context by default) that these don't
>> need to be NOWARN|NORETRY (same for post_deps).  I guess you inherited
>> this from drm/msm, where I appear to have forgotten to update the
>> syncobj path in commit f0de40a131d9 ("drm/msm: Reorder lock vs submit
>> alloc").  I don't see anything obvious that would require NORETRY, but
>> lockdep should be able to tell you otherwise if needed.
> 
> The NORETRY should prevent waking up OOM killer, it shouldn't help with
> lockdep. Nothing prevents userspace from giving a big number of
> num_in_syncobjs. But perhaps indeed not very practical to care about
> this case, given that other similar memalloc paces of execbuffer_ioctl()
> aren't using NORETRY. Alright, let's drop it in v4.
> 

Although no, there is only a kvmalloc_array() in the code and vmalloc
uses NOWARN and NORETRY flags implicitly. May be better switch to use
kvmalloc everywhere, for consistency. Technically, vmalloc shouldn't
ever be needed for a submit code path and kmalloc should be enough. On
the other hand, vmalloc acts like kmalloc until there is no enough
contig memory.
  

Patch

diff --git a/drivers/gpu/drm/virtio/virtgpu_drv.c b/drivers/gpu/drm/virtio/virtgpu_drv.c
index add075681e18..a22155577152 100644
--- a/drivers/gpu/drm/virtio/virtgpu_drv.c
+++ b/drivers/gpu/drm/virtio/virtgpu_drv.c
@@ -176,7 +176,8 @@  static const struct drm_driver driver = {
 	 * If KMS is disabled DRIVER_MODESET and DRIVER_ATOMIC are masked
 	 * out via drm_device::driver_features:
 	 */
-	.driver_features = DRIVER_MODESET | DRIVER_GEM | DRIVER_RENDER | DRIVER_ATOMIC,
+	.driver_features = DRIVER_MODESET | DRIVER_GEM | DRIVER_RENDER | DRIVER_ATOMIC |
+			   DRIVER_SYNCOBJ | DRIVER_SYNCOBJ_TIMELINE,
 	.open = virtio_gpu_driver_open,
 	.postclose = virtio_gpu_driver_postclose,
 
diff --git a/drivers/gpu/drm/virtio/virtgpu_submit.c b/drivers/gpu/drm/virtio/virtgpu_submit.c
index 2ce2459c6bc2..9ea4390948bf 100644
--- a/drivers/gpu/drm/virtio/virtgpu_submit.c
+++ b/drivers/gpu/drm/virtio/virtgpu_submit.c
@@ -14,11 +14,26 @@ 
 #include <linux/uaccess.h>
 
 #include <drm/drm_file.h>
+#include <drm/drm_syncobj.h>
 #include <drm/virtgpu_drm.h>
 
 #include "virtgpu_drv.h"
 
+struct virtio_gpu_submit_post_dep {
+	struct drm_syncobj *syncobj;
+	struct dma_fence_chain *chain;
+	uint64_t point;
+};
+
 struct virtio_gpu_submit {
+	struct virtio_gpu_submit_post_dep *post_deps;
+	unsigned int num_out_syncobjs;
+
+	struct drm_syncobj **in_syncobjs;
+	unsigned int num_in_syncobjs;
+	uint64_t *in_fence_ids;
+	unsigned int num_in_fence_ids;
+
 	struct virtio_gpu_object_array *buflist;
 	struct drm_virtgpu_execbuffer *exbuf;
 	struct virtio_gpu_fence *out_fence;
@@ -58,6 +73,186 @@  static int virtio_gpu_dma_fence_wait(struct virtio_gpu_submit *submit,
 	return 0;
 }
 
+static void virtio_gpu_free_syncobjs(struct drm_syncobj **syncobjs,
+				     uint32_t nr_syncobjs)
+{
+	uint32_t i = nr_syncobjs;
+
+	while (syncobjs && i--) {
+		if (syncobjs[i])
+			drm_syncobj_put(syncobjs[i]);
+	}
+
+	kfree(syncobjs);
+}
+
+static int
+virtio_gpu_parse_deps(struct virtio_gpu_submit *submit)
+{
+	struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
+	struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
+	size_t syncobj_stride = exbuf->syncobj_stride;
+	struct drm_syncobj **syncobjs;
+	int ret = 0, i;
+
+	if (!submit->num_in_syncobjs)
+		return 0;
+
+	syncobjs = kcalloc(submit->num_in_syncobjs, sizeof(*syncobjs),
+			   GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
+	if (!syncobjs)
+		return -ENOMEM;
+
+	for (i = 0; i < submit->num_in_syncobjs; i++) {
+		uint64_t address = exbuf->in_syncobjs + i * syncobj_stride;
+		struct dma_fence *fence;
+
+		if (copy_from_user(&syncobj_desc,
+				   u64_to_user_ptr(address),
+				   min(syncobj_stride, sizeof(syncobj_desc)))) {
+			ret = -EFAULT;
+			break;
+		}
+
+		if (syncobj_desc.flags & ~VIRTGPU_EXECBUF_SYNCOBJ_FLAGS) {
+			ret = -EINVAL;
+			break;
+		}
+
+		ret = drm_syncobj_find_fence(submit->file, syncobj_desc.handle,
+					     syncobj_desc.point, 0, &fence);
+		if (ret)
+			break;
+
+		ret = virtio_gpu_dma_fence_wait(submit, fence);
+
+		dma_fence_put(fence);
+		if (ret)
+			break;
+
+		if (syncobj_desc.flags & VIRTGPU_EXECBUF_SYNCOBJ_RESET) {
+			syncobjs[i] =
+				drm_syncobj_find(submit->file, syncobj_desc.handle);
+			if (!syncobjs[i]) {
+				ret = -EINVAL;
+				break;
+			}
+		}
+	}
+
+	if (ret) {
+		virtio_gpu_free_syncobjs(syncobjs, i);
+		return ret;
+	}
+
+	submit->in_syncobjs = syncobjs;
+
+	return ret;
+}
+
+static void virtio_gpu_reset_syncobjs(struct drm_syncobj **syncobjs,
+				      uint32_t nr_syncobjs)
+{
+	uint32_t i;
+
+	for (i = 0; syncobjs && i < nr_syncobjs; i++) {
+		if (syncobjs[i])
+			drm_syncobj_replace_fence(syncobjs[i], NULL);
+	}
+}
+
+static void
+virtio_gpu_free_post_deps(struct virtio_gpu_submit_post_dep *post_deps,
+			  uint32_t nr_syncobjs)
+{
+	uint32_t i = nr_syncobjs;
+
+	while (post_deps && i--) {
+		kfree(post_deps[i].chain);
+		drm_syncobj_put(post_deps[i].syncobj);
+	}
+
+	kfree(post_deps);
+}
+
+static int virtio_gpu_parse_post_deps(struct virtio_gpu_submit *submit)
+{
+	struct drm_virtgpu_execbuffer *exbuf = submit->exbuf;
+	struct drm_virtgpu_execbuffer_syncobj syncobj_desc;
+	struct virtio_gpu_submit_post_dep *post_deps;
+	size_t syncobj_stride = exbuf->syncobj_stride;
+	int ret = 0, i;
+
+	if (!submit->num_out_syncobjs)
+		return 0;
+
+	post_deps = kcalloc(submit->num_out_syncobjs, sizeof(*post_deps),
+			    GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
+	if (!post_deps)
+		return -ENOMEM;
+
+	for (i = 0; i < submit->num_out_syncobjs; i++) {
+		uint64_t address = exbuf->out_syncobjs + i * syncobj_stride;
+
+		if (copy_from_user(&syncobj_desc,
+				   u64_to_user_ptr(address),
+				   min(syncobj_stride, sizeof(syncobj_desc)))) {
+			ret = -EFAULT;
+			break;
+		}
+
+		post_deps[i].point = syncobj_desc.point;
+
+		if (syncobj_desc.flags) {
+			ret = -EINVAL;
+			break;
+		}
+
+		if (syncobj_desc.point) {
+			post_deps[i].chain = dma_fence_chain_alloc();
+			if (!post_deps[i].chain) {
+				ret = -ENOMEM;
+				break;
+			}
+		}
+
+		post_deps[i].syncobj =
+			drm_syncobj_find(submit->file, syncobj_desc.handle);
+		if (!post_deps[i].syncobj) {
+			ret = -EINVAL;
+			break;
+		}
+	}
+
+	if (ret) {
+		virtio_gpu_free_post_deps(post_deps, i);
+		return ret;
+	}
+
+	submit->post_deps = post_deps;
+
+	return 0;
+}
+
+static void
+virtio_gpu_process_post_deps(struct virtio_gpu_submit *submit)
+{
+	struct virtio_gpu_submit_post_dep *post_deps = submit->post_deps;
+	struct dma_fence *fence = &submit->out_fence->f;
+	uint32_t i;
+
+	for (i = 0; post_deps && i < submit->num_out_syncobjs; i++) {
+		if (post_deps[i].chain) {
+			drm_syncobj_add_point(post_deps[i].syncobj,
+					      post_deps[i].chain,
+					      fence, post_deps[i].point);
+			post_deps[i].chain = NULL;
+		} else {
+			drm_syncobj_replace_fence(post_deps[i].syncobj, fence);
+		}
+	}
+}
+
 static int virtio_gpu_fence_event_create(struct drm_device *dev,
 					 struct drm_file *file,
 					 struct virtio_gpu_fence *fence,
@@ -121,6 +316,11 @@  static int virtio_gpu_init_submit_buflist(struct virtio_gpu_submit *submit)
 
 static void virtio_gpu_cleanup_submit(struct virtio_gpu_submit *submit)
 {
+	virtio_gpu_reset_syncobjs(submit->in_syncobjs, submit->num_in_syncobjs);
+	virtio_gpu_free_syncobjs(submit->in_syncobjs, submit->num_in_syncobjs);
+	virtio_gpu_free_post_deps(submit->post_deps, submit->num_out_syncobjs);
+	kfree(submit->in_fence_ids);
+
 	if (!IS_ERR(submit->buf))
 		kvfree(submit->buf);
 
@@ -173,6 +373,8 @@  static int virtio_gpu_init_submit(struct virtio_gpu_submit *submit,
 		return err;
 	}
 
+	submit->num_out_syncobjs = exbuf->num_out_syncobjs;
+	submit->num_in_syncobjs = exbuf->num_in_syncobjs;
 	submit->out_fence = out_fence;
 	submit->fence_ctx = fence_ctx;
 	submit->ring_idx = ring_idx;
@@ -285,6 +487,14 @@  int virtio_gpu_execbuffer_ioctl(struct drm_device *dev, void *data,
 	if (ret)
 		goto cleanup;
 
+	ret = virtio_gpu_parse_deps(&submit);
+	if (ret)
+		goto cleanup;
+
+	ret = virtio_gpu_parse_post_deps(&submit);
+	if (ret)
+		goto cleanup;
+
 	ret = virtio_gpu_install_out_fence_fd(&submit);
 	if (ret)
 		goto cleanup;
@@ -294,6 +504,7 @@  int virtio_gpu_execbuffer_ioctl(struct drm_device *dev, void *data,
 		goto cleanup;
 
 	virtio_gpu_submit(&submit);
+	virtio_gpu_process_post_deps(&submit);
 	virtio_gpu_complete_submit(&submit);
 cleanup:
 	virtio_gpu_cleanup_submit(&submit);
diff --git a/include/uapi/drm/virtgpu_drm.h b/include/uapi/drm/virtgpu_drm.h
index 7b158fcb02b4..ce4948aacafd 100644
--- a/include/uapi/drm/virtgpu_drm.h
+++ b/include/uapi/drm/virtgpu_drm.h
@@ -64,6 +64,16 @@  struct drm_virtgpu_map {
 	__u32 pad;
 };
 
+#define VIRTGPU_EXECBUF_SYNCOBJ_RESET		0x01
+#define VIRTGPU_EXECBUF_SYNCOBJ_FLAGS ( \
+		VIRTGPU_EXECBUF_SYNCOBJ_RESET | \
+		0)
+struct drm_virtgpu_execbuffer_syncobj {
+	__u32 handle;
+	__u32 flags;
+	__u64 point;
+};
+
 /* fence_fd is modified on success if VIRTGPU_EXECBUF_FENCE_FD_OUT flag is set. */
 struct drm_virtgpu_execbuffer {
 	__u32 flags;
@@ -73,7 +83,11 @@  struct drm_virtgpu_execbuffer {
 	__u32 num_bo_handles;
 	__s32 fence_fd; /* in/out fence fd (see VIRTGPU_EXECBUF_FENCE_FD_IN/OUT) */
 	__u32 ring_idx; /* command ring index (see VIRTGPU_EXECBUF_RING_IDX) */
-	__u32 pad;
+	__u32 syncobj_stride;
+	__u32 num_in_syncobjs;
+	__u32 num_out_syncobjs;
+	__u64 in_syncobjs;
+	__u64 out_syncobjs;
 };
 
 #define VIRTGPU_PARAM_3D_FEATURES 1 /* do we have 3D features in the hw */