> -----Original Message-----
> From: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx>
> Sent: Wednesday, September 14, 2022 8:13 PM
> To: Vivi, Rodrigo <rodrigo.vivi@xxxxxxxxx>
> Cc: Gupta, Anshuman <anshuman.gupta@xxxxxxxxx>; intel-
> gfx@xxxxxxxxxxxxxxxxxxxxx; joonas.lahtinen@xxxxxxxxxxxxxxx; Ewins, Jon
> <jon.ewins@xxxxxxxxx>; andi.shyti@xxxxxxxxxxxxxxx; Auld, Matthew
> <matthew.auld@xxxxxxxxx>
> Subject: Re: [PATCH] drm/i915/DG{1,2}: FIXME Temporary hammer to disable
> rpm
>
> Hi Anshuman,
>
> On Wed, Sep 14, 2022 at 10:33:15AM -0400, Rodrigo Vivi wrote:
> > On Wed, Sep 14, 2022 at 07:45:53PM +0530, Anshuman Gupta wrote:
> > > DG1 and DG2 has lmem, and cpu can access the lmem objects via mmap
> > > and i915 internal i915_gem_object_pin_map() for
> > > i915 own usages. Both of these methods has pre-requisite requirement
> > > to keep GFX PCI endpoint in D0 for a supported iomem transaction
> > > over PCI link. (Refer PCIe specs 5.3.1.4.1)
> > >
> > > TODO:
> > > With respect to i915_gem_object_pin_map(), every caller has to grab
> > > a wakeref if gem object lies in lmem.
> > >
> > > Till we fix all issues related to runtime PM, we need to keep
> > > runtime PM disable on both DG1 and DG2.
> > >
> > > Cc: Matthew Auld <matthew.auld@xxxxxxxxx>
> > > Cc: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx>
> > > Signed-off-by: Anshuman Gupta <anshuman.gupta@xxxxxxxxx>
> > > ---
> > > drivers/gpu/drm/i915/i915_pci.c | 21 +++++++++++++++++++++
> > > 1 file changed, 21 insertions(+)
> > >
> > > diff --git a/drivers/gpu/drm/i915/i915_pci.c
> > > b/drivers/gpu/drm/i915/i915_pci.c index 77e7df21f539..f31d7f5399cc
> > > 100644
> > > --- a/drivers/gpu/drm/i915/i915_pci.c
> > > +++ b/drivers/gpu/drm/i915/i915_pci.c
> > > @@ -931,6 +931,26 @@ static const struct intel_device_info dg1_info = {
> > > BIT(VCS0) | BIT(VCS2),
> > > /* Wa_16011227922 */
> > > .__runtime.ppgtt_size = 47,
> > > +
> > > + /*
> > > + * FIXME: Temporary hammer to disable rpm.
> > > + * As per PCIe specs 5.3.1.4.1, all iomem read write request over a PCIe
> > > + * function will be unsupported in case PCIe endpoint function is in D3.
> > > + * But both DG1/DG2 has a hardware bug that violates the PCIe
> > > +specs
>
> /has/have/
>
> > > + * and supports the iomem read write transaction over PCIe bus
> > > +despite
>
> /supports/support/
>
> > > + * endpoint is D3 state.
> > > + * Due to above H/W bug, we had never observed any issue with i915
> runtime
> > > + * PM versus lmem access.
> > > + * But this issue gets discover when PCIe gfx endpoint's upstream
>
> /gets discover/becomes visible/
>
> > > + * bridge enters to D3, at this point any lmem read/write access will be
> > > + * returned as unsupported request. But again this issue is not observed
> > > + * on every platform because it has been observed on few host
> machines
> > > + * DG1/DG2 endpoint's upstream bridge does not binds with pcieport
> driver.
>
> /binds/bind/
>
> > > + * which really disables the PCIe power savings and leaves the bridge to
> D0
> > > + * state.
> > > + * Let's disable i915 rpm till we fix all known issue with lmem access in
> D3.
> > > + */
> > > + .has_runtime_pm = 0,
> > > };
> > >
> > > static const struct intel_device_info adl_s_info = { @@ -1076,6
> > > +1096,7 @@ static const struct intel_device_info dg2_info = {
> > > XE_LPD_FEATURES,
> > > .__runtime.cpu_transcoder_mask = BIT(TRANSCODER_A) |
> BIT(TRANSCODER_B) |
> > > BIT(TRANSCODER_C) | BIT(TRANSCODER_D),
> > > + .has_runtime_pm = 0,
> >
> > The FIXME msg can be smaller, but it also needs to be here.
>
> I actually like the comment, is very clear and helps understanding the issue :)
Shall I move the comment to commit log , and keep a smaller comment for both DG1 and DG2 ?
With that I can address your comment and Rodrigo comment as well.
Keeping such a big comment at two places will not make any sense.
Thanks,
Anshuman Gupta.
>
> Thanks again for addressing the issue and with the hope to see the proper fix
> soon:
>
> Reviewed-by: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx>
>
> Thanks,
> Andi
>
> > With this in place fell free to use:
> >
> > Reviewed-by: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx>
> >
> > Since the proper solution might take a while let's protect from this
> > case, regardless of any other on going discussion about the force_probe
> protection.
> >
> >
> > > .require_force_probe = 1,
> > > };
> > >
> > > --
> > > 2.26.2
> > >
