Re: [PATCH 0/7] Cleanup error handling on probe

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Tue, Feb 04, 2025 at 06:10:48PM +0000, Tvrtko Ursulin wrote:
> 
> On 04/02/2025 14:50, Lucas De Marchi wrote:
> > On Tue, Feb 04, 2025 at 08:58:37AM +0000, Tvrtko Ursulin wrote:
> > > 
> > > On 31/01/2025 22:31, Lucas De Marchi wrote:
> > > > Mixing style with goto and devm_add_action_or_reset() is very confusing
> > > > and error prone. Clean that up. The only missing one is one of the
> > > > display calls, but that can be done in parallel to the review of these
> > > > first patches.
> > > > 
> > > > This received limited load/unload tests on LNL and may very well not be
> > > > 100%.
> > > 
> > > Is this series hinting you are looking into making heavier
> > > refactorings in the init code? Triggered by me recently
> > > discovering
> > 
> > yes, I still need to move more display init around and some other gt
> > parts.
> > 
> > > GT-per-engine workarounds are not working?
> > 
> > it was not what triggered this - it was rather my review of the pmu
> > integration in xe when I noticed we were just continuing on a failed pmu
> > registration:
> > 
> > https://lore.kernel.org/intel-xe/lnvrm7guru3ujx7kt4irhvjb3wc3vcfzzhko3h5zkvgdig6znm@xb56nnaemwi5/
> > 
> > what's the state of your patch to fix engine registers in GT WAs?
> > Did you find a suitable place for that fix?
> 
> Status is that I found a chicken and egg where GuC load needs
> engines and engine init needs GuC. I had some vague ideas that
> something could maybe be done in two passes, but then I spotted you
> were working in init paths so parked it until I see what you were
> up to. :)

Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>

There was some need for SRIOV that forced some refactor there
and I thought there were 2 phases, the initial with mmio only
before we could enable the ct communication. But definitely Michal
is the expert on that. And of course to take care of this sensible
SRIOV case.

> 
> Regards,
> 
> Tvrtko
> 
> > > > Lucas De Marchi (7):
> > > >   drm/xe: Fix xe_display_fini() calls
> > > >   drm/xe: Fix error handling in xe_irq_install()
> > > >   drm/xe: Fix xe_tile_init_noalloc() error propagation
> > > >   drm/xe: Stop ignoring errors from xe_ttm_stolen_mgr_init()
> > > >   drm/xe: Cleanup unwind of gt initialization
> > > >   drm/xe: Cleanup extra calls to  xe_hw_fence_irq_finish()
> > > >   drm/xe: Move oa fini to xe_oa
> > > > 
> > > >  drivers/gpu/drm/xe/display/xe_display.c | 22 +++++-----
> > > >  drivers/gpu/drm/xe/display/xe_display.h |  2 -
> > > >  drivers/gpu/drm/xe/xe_device.c          | 53 ++++++++-----------------
> > > >  drivers/gpu/drm/xe/xe_gt.c              | 52 ++++++++++--------------
> > > >  drivers/gpu/drm/xe/xe_gt.h              |  1 -
> > > >  drivers/gpu/drm/xe/xe_irq.c             | 14 +------
> > > >  drivers/gpu/drm/xe/xe_oa.c              | 48 +++++++++++-----------
> > > >  drivers/gpu/drm/xe/xe_oa.h              |  1 -
> > > >  drivers/gpu/drm/xe/xe_tile.c            |  4 +-
> > > >  drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c  | 17 ++++----
> > > >  drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h  |  2 +-
> > > >  11 files changed, 86 insertions(+), 130 deletions(-)
> > > >