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 <m...@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 {
>
signature.asc
Description: PGP signature
