On Thu, Oct 06, 2016 at 03:14:19PM +0200, Michał Winiarski wrote: > We're reserving space in the workqueue early during request allocation. > This can be problematic if we get interrupted later in the process, > because we can end up in a state where GuC workqueue is seemingly full, > while the HW is idle and no work has been submitted. > Let's delay the reservation to avoid such scenario. > > Testcase: igt/gem_ringfill/*-interruptible > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97978 > Cc: Michel Thierry <michel.thie...@intel.com> > Cc: Jeff Mcgee <jeff.mc...@intel.com> > Signed-off-by: Michał Winiarski <michal.winiar...@intel.com> > --- > drivers/gpu/drm/i915/i915_gem_execbuffer.c | 11 +++++++++++ > drivers/gpu/drm/i915/intel_lrc.c | 11 ----------- > 2 files changed, 11 insertions(+), 11 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c > b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > index e88786e..37204db 100644 > --- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c > +++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c > @@ -1529,6 +1529,17 @@ execbuf_submit(struct i915_execbuffer_params *params, > if (ret) > return ret; > > + if (i915.enable_guc_submission) {
No. This logic does not deserve to be at this level. Please try fixing the underlying problem, which would be that we don't cancel the reserved space. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
