[Intel-gfx] [PATCH 1/3] drm/i915: Use intel_engine_stop_cs when stopping ringbuffer

[Chris Wilson]

Quoting Mika Kuoppala (2019-02-28 16:53:46)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > Quoting Mika Kuoppala (2019-02-28 16:14:11)
> >> We have an exported function for stopping the engine before
> >> disabling a ringbuffer. Take it into use.
> >> 
> >> v2: use fw on empty check
> >> v3: tail is tail
> >> 
> >> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> >> Signed-off-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> >> Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>
> >> ---
> >>  drivers/gpu/drm/i915/intel_engine_cs.c  |  3 ++
> >>  drivers/gpu/drm/i915/intel_ringbuffer.c | 41 ++++++++++++++-----------
> >>  2 files changed, 26 insertions(+), 18 deletions(-)
> >> 
> >> diff --git a/drivers/gpu/drm/i915/intel_engine_cs.c b/drivers/gpu/drm/i915/intel_engine_cs.c
> >> index df8f88142f1d..e35dc0386bf6 100644
> >> --- a/drivers/gpu/drm/i915/intel_engine_cs.c
> >> +++ b/drivers/gpu/drm/i915/intel_engine_cs.c
> >> @@ -856,6 +856,9 @@ void intel_engine_cancel_stop_cs(struct intel_engine_cs *engine)
> >>  {
> >>         struct drm_i915_private *dev_priv = engine->i915;
> >>  
> >> +       if (INTEL_GEN(dev_priv) < 3)
> >> +               return;
> >> +
> >>         GEM_TRACE("%s\n", engine->name);
> >>  
> >>         I915_WRITE_FW(RING_MI_MODE(engine->mmio_base),
> >> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> >> index 1b96b0960adc..5fe28d9087b7 100644
> >> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> >> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> >> @@ -604,26 +604,32 @@ static void ring_setup_status_page(struct intel_engine_cs *engine)
> >>         flush_cs_tlb(engine);
> >>  }
> >>  
> >> +static bool ring_is_empty(struct intel_engine_cs *engine)
> >> +{
> >> +       struct drm_i915_private *dev_priv = engine->i915;
> >> +       const u32 base = engine->mmio_base;
> >> +
> >> +       return (I915_READ_FW(RING_HEAD(base)) & HEAD_ADDR) ==
> >> +               (I915_READ_FW(RING_TAIL(base)) & TAIL_ADDR);
> >> +}
> >> +
> >>  static bool stop_ring(struct intel_engine_cs *engine)
> >>  {
> >>         struct drm_i915_private *dev_priv = engine->i915;
> >> +       int ret;
> >>  
> >> -       if (INTEL_GEN(dev_priv) > 2) {
> >> -               I915_WRITE_MODE(engine, _MASKED_BIT_ENABLE(STOP_RING));
> >> -               if (intel_wait_for_register(dev_priv,
> >> -                                           RING_MI_MODE(engine->mmio_base),
> >> -                                           MODE_IDLE,
> >> -                                           MODE_IDLE,
> >> -                                           1000)) {
> >> -                       DRM_ERROR("%s : timed out trying to stop ring\n",
> >> -                                 engine->name);
> >> -                       /* Sometimes we observe that the idle flag is not
> >> -                        * set even though the ring is empty. So double
> >> -                        * check before giving up.
> >> -                        */
> >> -                       if (I915_READ_HEAD(engine) != I915_READ_TAIL(engine))
> >> -                               return false;
> >> -               }
> >> +       ret = intel_engine_stop_cs(engine);
> >> +       if (ret == -ENODEV)
> >> +               ret = 0;
> >> +
> >> +       if (ret) {
> >> +               /*
> >> +                * Sometimes we observe that the idle flag is not
> >> +                * set even though the ring is empty. So double
> >> +                * check before giving up.
> >> +                */
> >> +               if (!ring_is_empty(engine))
> >> +                       return false;
> >
> > Hmm, thinking more about this, shouldn't we push this down to stop_cs()?
> >
> > If that's reporting an error in a situation where we can determine that
> > the ring is idle anyway, we can report the stop_cs succeeded.
> 
> Makes sense, I will take a look.
> 
> I felt small urge to deflate the 'stop'.
> 
> Would it be confusing if we just did
> intel_engine_start|stop instead of stop_cs and
> cancel_stop_cs?
> 
> So hmm:
> 
> intel_engine_start()
> intel_engine_stop()
> 
> these would only toggle the STOP_RING
> 
> and for replacing stop_ring with:
> intel_engine_empty_ring()

intel_engine_clear_ring() / reset_ring(). Hmm, clear_ring of those two.
> 
> for zeroing the heads.
> 
> one 'stop' to rule the (ring)world!?

The counter argument is that _start() is a little too broad. The appeal
of stop_cs() is that it describing what it is doing, poking at the bit
to stop the CS advancing and nothing more. It frequently doesn't
succeed...

So I think it's not a worthy change, but I never did feel totally
satisfied with stop_cs -- cs is too short, but we do have usage with
xCS.

intel_engine_stop_command_streamer,
intel_engine_halt_command_streamer,
intel_engine_pause_command_streamer,
?

But intel_engine_clear_ring() I could be sold on.
-Chris