Re: [PATCH] drm/i915: Implement SINGLE_TIMELINE with a syncobj (v2)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Thu, Mar 25, 2021 at 4:21 PM Matthew Brost <matthew.brost@xxxxxxxxx> wrote:
>
> On Tue, Mar 23, 2021 at 12:51:49PM -0500, Jason Ekstrand wrote:
> > This API is entirely unnecessary and I'd love to get rid of it.  If
> > userspace wants a single timeline across multiple contexts, they can
> > either use implicit synchronization or a syncobj, both of which existed
> > at the time this feature landed.  The justification given at the time
> > was that it would help GL drivers which are inherently single-timeline.
> > However, neither of our GL drivers actually wanted the feature.  i965
> > was already in maintenance mode at the time and iris uses syncobj for
> > everything.
> >
> > Unfortunately, as much as I'd love to get rid of it, it is used by the
> > media driver so we can't do that.  We can, however, do the next-best
> > thing which is to embed a syncobj in the context and do exactly what
> > we'd expect from userspace internally.  This isn't an entirely identical
> > implementation because it's no longer atomic if userspace races with
> > itself by calling execbuffer2 twice simultaneously from different
> > threads.  It won't crash in that case; it just doesn't guarantee any
> > ordering between those two submits.
> >
> > Moving SINGLE_TIMELINE to a syncobj emulation has a couple of technical
> > advantages beyond mere annoyance.  One is that intel_timeline is no
> > longer an api-visible object and can remain entirely an implementation
> > detail.  This may be advantageous as we make scheduler changes going
> > forward.  Second is that, together with deleting the CLONE_CONTEXT API,
> > we should now have a 1:1 mapping between intel_context and
> > intel_timeline which may help us reduce locking.
> >
>
> I'm not going to dive into the philosophical debate that this patch has
> pigeonholed into, but a 1:1 mapping between intel_context and
> intel_timeline is huge win IMO - these are the types of cleanups I can
> get really get behind.
>
> > v2 (Jason Ekstrand):
> >  - Update the comment on i915_gem_context::syncobj to mention that it's
> >    an emulation and the possible race if userspace calls execbuffer2
> >    twice on the same context concurrently.
> >  - Wrap the checks for eb.gem_context->syncobj in unlikely()
> >  - Drop the dma_fence reference
> >  - Improved commit message
> >
> > Signed-off-by: Jason Ekstrand <jason@xxxxxxxxxxxxxx>
> > Cc: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx>
> > Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
> > ---
> >  drivers/gpu/drm/i915/gem/i915_gem_context.c   | 47 ++++---------------
> >  .../gpu/drm/i915/gem/i915_gem_context_types.h | 14 +++++-
> >  .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 16 +++++++
> >  3 files changed, 39 insertions(+), 38 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > index f88bac19333ec..e094f4a1ca4cd 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
> > @@ -67,6 +67,8 @@
> >  #include <linux/log2.h>
> >  #include <linux/nospec.h>
> >
> > +#include <drm/drm_syncobj.h>
> > +
> >  #include "gt/gen6_ppgtt.h"
> >  #include "gt/intel_context.h"
> >  #include "gt/intel_engine_heartbeat.h"
> > @@ -224,10 +226,6 @@ static void intel_context_set_gem(struct intel_context *ce,
> >               ce->vm = vm;
> >       }
> >
> > -     GEM_BUG_ON(ce->timeline);
> > -     if (ctx->timeline)
> > -             ce->timeline = intel_timeline_get(ctx->timeline);
> > -
> >       if (ctx->sched.priority >= I915_PRIORITY_NORMAL &&
> >           intel_engine_has_timeslices(ce->engine))
> >               __set_bit(CONTEXT_USE_SEMAPHORES, &ce->flags);
> > @@ -344,8 +342,8 @@ void i915_gem_context_release(struct kref *ref)
> >       mutex_destroy(&ctx->engines_mutex);
> >       mutex_destroy(&ctx->lut_mutex);
> >
> > -     if (ctx->timeline)
> > -             intel_timeline_put(ctx->timeline);
> > +     if (ctx->syncobj)
> > +             drm_syncobj_put(ctx->syncobj);
> >
> >       put_pid(ctx->pid);
> >       mutex_destroy(&ctx->mutex);
> > @@ -790,33 +788,11 @@ static void __assign_ppgtt(struct i915_gem_context *ctx,
> >               i915_vm_close(vm);
> >  }
> >
> > -static void __set_timeline(struct intel_timeline **dst,
> > -                        struct intel_timeline *src)
> > -{
> > -     struct intel_timeline *old = *dst;
> > -
> > -     *dst = src ? intel_timeline_get(src) : NULL;
> > -
> > -     if (old)
> > -             intel_timeline_put(old);
> > -}
> > -
> > -static void __apply_timeline(struct intel_context *ce, void *timeline)
> > -{
> > -     __set_timeline(&ce->timeline, timeline);
> > -}
> > -
> > -static void __assign_timeline(struct i915_gem_context *ctx,
> > -                           struct intel_timeline *timeline)
> > -{
> > -     __set_timeline(&ctx->timeline, timeline);
> > -     context_apply_all(ctx, __apply_timeline, timeline);
> > -}
> > -
> >  static struct i915_gem_context *
> >  i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
> >  {
> >       struct i915_gem_context *ctx;
> > +     int ret;
> >
> >       if (flags & I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE &&
> >           !HAS_EXECLISTS(i915))
> > @@ -845,16 +821,13 @@ i915_gem_create_context(struct drm_i915_private *i915, unsigned int flags)
> >       }
> >
> >       if (flags & I915_CONTEXT_CREATE_FLAGS_SINGLE_TIMELINE) {
> > -             struct intel_timeline *timeline;
> > -
> > -             timeline = intel_timeline_create(&i915->gt);
> > -             if (IS_ERR(timeline)) {
> > +             ret = drm_syncobj_create(&ctx->syncobj,
> > +                                      DRM_SYNCOBJ_CREATE_SIGNALED,
> > +                                      NULL);
> > +             if (ret) {
> >                       context_close(ctx);
> > -                     return ERR_CAST(timeline);
> > +                     return ERR_PTR(ret);
> >               }
> > -
> > -             __assign_timeline(ctx, timeline);
> > -             intel_timeline_put(timeline);
> >       }
> >
> >       trace_i915_context_create(ctx);
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
> > index 676592e27e7d2..df76767f0c41b 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_context_types.h
> > @@ -83,7 +83,19 @@ struct i915_gem_context {
> >       struct i915_gem_engines __rcu *engines;
> >       struct mutex engines_mutex; /* guards writes to engines */
> >
> > -     struct intel_timeline *timeline;
> > +     /**
> > +      * @syncobj: Shared timeline syncobj
> > +      *
> > +      * When the SHARED_TIMELINE flag is set on context creation, we
> > +      * emulate a single timeline across all engines using this syncobj.
> > +      * For every execbuffer2 call, this syncobj is used as both an in-
> > +      * and out-fence.  Unlike the real intel_timeline, this doesn't
> > +      * provide perfect atomic in-order guarantees if the client races
> > +      * with itself by calling execbuffer2 twice concurrently.  However,
> > +      * if userspace races with itself, that's not likely to yield well-
> > +      * defined results anyway so we choose to not care.
> > +      */
> > +     struct drm_syncobj *syncobj;
> >
> >       /**
> >        * @vm: unique address space (GTT)
> > diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> > index 96403130a373d..2e9748c1edddf 100644
> > --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> > +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> > @@ -3295,6 +3295,16 @@ i915_gem_do_execbuffer(struct drm_device *dev,
> >               goto err_vma;
> >       }
> >
> > +     if (unlikely(eb.gem_context->syncobj)) {
> > +             struct dma_fence *fence;
> > +
> > +             fence = drm_syncobj_fence_get(eb.gem_context->syncobj);
> > +             err = i915_request_await_dma_fence(eb.request, fence);
> > +             if (err)
> > +                     goto err_ext;
> > +             dma_fence_put(fence);
>
> Also think this needs to be moved above the error checking, as this put
> corresponds to 'drm_syncobj_fence_get', right?

Already done.  Let me send v3.

> Matt
>
> > +     }
> > +
> >       if (in_fence) {
> >               if (args->flags & I915_EXEC_FENCE_SUBMIT)
> >                       err = i915_request_await_execution(eb.request,
> > @@ -3351,6 +3361,12 @@ i915_gem_do_execbuffer(struct drm_device *dev,
> >                       fput(out_fence->file);
> >               }
> >       }
> > +
> > +     if (unlikely(eb.gem_context->syncobj)) {
> > +             drm_syncobj_replace_fence(eb.gem_context->syncobj,
> > +                                       &eb.request->fence);
> > +     }
> > +
> >       i915_request_put(eb.request);
> >
> >  err_vma:
> > --
> > 2.29.2
> >
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel



[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux