[PATCH] drm/i915/gt: Reset twice

[Vivi, Rodrigo]

On Tue, 2022-12-13 at 00:08 +0100, Andi Shyti wrote:
> Hi Rodrigo,
> 
> On Mon, Dec 12, 2022 at 11:55:10AM -0500, Rodrigo Vivi wrote:
> > On Mon, Dec 12, 2022 at 05:13:38PM +0100, Andi Shyti wrote:
> > > From: Chris Wilson <chris at chris-wilson.co.uk>
> > > 
> > > After applying an engine reset, on some platforms like
> > > Jasperlake, we
> > > occasionally detect that the engine state is not cleared until
> > > shortly
> > > after the resume. As we try to resume the engine with volatile
> > > internal
> > > state, the first request fails with a spurious CS event (it looks
> > > like
> > > it reports a lite-restore to the hung context, instead of the
> > > expected
> > > idle->active context switch).
> > > 
> > > Signed-off-by: Chris Wilson <hris at chris-wilson.co.uk>
> > 
> > There's a typo in the signature email I'm afraid...
> 
> oh yes, I forgot the 'C' :)

you forgot?
who signed it off?

> 
> > Other than that, have we checked the possibility of using the
> > driver-initiated-flr bit
> > instead of this second loop? That should be the right way to
> > guarantee everything is
> > cleared on gen11+...
> 
> maybe I am misinterpreting it, but is FLR the same as resetting
> hardware domains individually?

No, it is bigger than that... almost the PCI FLR with some exceptions:

https://lists.freedesktop.org/archives/intel-gfx/2022-December/313956.html

> How am I supposed to use driver_initiated_flr() in this context?

Some drivers are not even using this gt reset anymore and going
directly with the driver-initiated FLR in case that guc reset failed.

I believe we can still keep the gt reset in our case as we currently
have, but instead of keep retrying it until it succeeds we probably
should go to the next level and do the driver initiated FLR when the GT
reset failed.

> 
> Thanks,
> Andi
> 
> > > Cc: stable at vger.kernel.org
> > > Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
> > > Signed-off-by: Andi Shyti <andi.shyti at linux.intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/gt/intel_reset.c | 34
> > > ++++++++++++++++++++++-----
> > >  1 file changed, 28 insertions(+), 6 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/i915/gt/intel_reset.c
> > > b/drivers/gpu/drm/i915/gt/intel_reset.c
> > > index ffde89c5835a4..88dfc0c5316ff 100644
> > > --- a/drivers/gpu/drm/i915/gt/intel_reset.c
> > > +++ b/drivers/gpu/drm/i915/gt/intel_reset.c
> > > @@ -268,6 +268,7 @@ static int ilk_do_reset(struct intel_gt *gt,
> > > intel_engine_mask_t engine_mask,
> > >  static int gen6_hw_domain_reset(struct intel_gt *gt, u32
> > > hw_domain_mask)
> > >  {
> > >         struct intel_uncore *uncore = gt->uncore;
> > > +       int loops = 2;
> > >         int err;
> > >  
> > >         /*
> > > @@ -275,18 +276,39 @@ static int gen6_hw_domain_reset(struct
> > > intel_gt *gt, u32 hw_domain_mask)
> > >          * for fifo space for the write or forcewake the chip for
> > >          * the read
> > >          */
> > > -       intel_uncore_write_fw(uncore, GEN6_GDRST,
> > > hw_domain_mask);
> > > +       do {
> > > +               intel_uncore_write_fw(uncore, GEN6_GDRST,
> > > hw_domain_mask);
> > >  
> > > -       /* Wait for the device to ack the reset requests */
> > > -       err = __intel_wait_for_register_fw(uncore,
> > > -                                          GEN6_GDRST,
> > > hw_domain_mask, 0,
> > > -                                          500, 0,
> > > -                                          NULL);
> > > +               /*
> > > +                * Wait for the device to ack the reset requests.
> > > +                *
> > > +                * On some platforms, e.g. Jasperlake, we see see
> > > that the
> > > +                * engine register state is not cleared until
> > > shortly after
> > > +                * GDRST reports completion, causing a failure as
> > > we try
> > > +                * to immediately resume while the internal state
> > > is still
> > > +                * in flux. If we immediately repeat the reset,
> > > the second
> > > +                * reset appears to serialise with the first, and
> > > since
> > > +                * it is a no-op, the registers should retain
> > > their reset
> > > +                * value. However, there is still a concern that
> > > upon
> > > +                * leaving the second reset, the internal engine
> > > state
> > > +                * is still in flux and not ready for resuming.
> > > +                */
> > > +               err = __intel_wait_for_register_fw(uncore,
> > > GEN6_GDRST,
> > > +                                                 
> > > hw_domain_mask, 0,
> > > +                                                  2000, 0,
> > > +                                                  NULL);
> > > +       } while (err == 0 && --loops);
> > >         if (err)
> > >                 GT_TRACE(gt,
> > >                          "Wait for 0x%08x engines reset
> > > failed\n",
> > >                          hw_domain_mask);
> > >  
> > > +       /*
> > > +        * As we have observed that the engine state is still
> > > volatile
> > > +        * after GDRST is acked, impose a small delay to let
> > > everything settle.
> > > +        */
> > > +       udelay(50);
> > > +
> > >         return err;
> > >  }
> > >  
> > > -- 
> > > 2.38.1
> > >