On Mon, Sep 20, 2021 at 08:42:42AM +0100, Tvrtko Ursulin wrote:
>
> On 20/09/2021 08:38, Jani Nikula wrote:
> > On Mon, 20 Sep 2021, Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx> wrote:
> > > On 18/09/2021 00:38, Matthew Brost wrote:
> > > > From: Hugh Dickins <hughd@xxxxxxxxxx>
> > > >
> > > > 5.15-rc1 crashes with blank screen when booting up on two ThinkPads
> > > > using i915. Bisections converge convincingly, but arrive at different
> > > > and surprising "culprits", none of them the actual culprit.
> > >
> > > It is certainly surprising this patch crashed SNB and KBL.
> > >
> > > How feasible would it be to make this code just not run when GuC is not
> > > used? Given the field it adds is called ce->guc_blocked it sounds like a
> > > natural and preferable thing to do... if possible.
> > >
> > > > netconsole (with init_netconsole() hacked to call i915_init() when
> > > > logging has started, instead of by module_init()) tells the story:
> > > >
> > > > kernel BUG at drivers/gpu/drm/i915/i915_sw_fence.c:245!
> > > > with RSI: ffffffff814d408b pointing to sw_fence_dummy_notify().
> > > > I've been building with CONFIG_CC_OPTIMIZE_FOR_SIZE=y, and that
> > > > function needs to be 4-byte aligned.
> > > >
> > > > v2:
> > > > (Jani Nikula)
> > > > - Change BUG_ON to WARN_ON
> > >
> > > However in this case the code would then go on and call into a wrong
> > > function offset which may be worse than a BUG_ON, no?
> >
> > So how about just
> >
> > if (WARN_ON(...))
> > return;
I don't think it is quite that simple as if we short circuit this
function fence->flags will be NULL which would be bad too. I'll have
make a few more changes to make this safe.
Matt
> >
> > or whatever is needed to give both the user and the CI a better
> > opportunity to see the error.
>
> Sounds good to me.
>
> Regards,
>
> Tvrtko
>
>
> >
> > BR,
> > Jani
> >
> >
> > >
> > > >
> > > > Fixes: 62eaf0ae217d ("drm/i915/guc: Support request cancellation")
> > > > Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
> > > > Signed-off-by: Matthew Brost <matthew.brost@xxxxxxxxx>
> > > > Reviewed-by: Matthew Brost <matthew.brost@xxxxxxxxx>
> > > > ---
> > > > drivers/gpu/drm/i915/gt/intel_context.c | 1 +
> > > > drivers/gpu/drm/i915/i915_sw_fence.c | 4 +++-
> > > > 2 files changed, 4 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/gpu/drm/i915/gt/intel_context.c b/drivers/gpu/drm/i915/gt/intel_context.c
> > > > index ff637147b1a9..f02c2202da9d 100644
> > > > --- a/drivers/gpu/drm/i915/gt/intel_context.c
> > > > +++ b/drivers/gpu/drm/i915/gt/intel_context.c
> > > > @@ -362,6 +362,7 @@ static int __intel_context_active(struct i915_active *active)
> > > > return 0;
> > > > }
> > > > +__aligned(4) /* Respect the I915_SW_FENCE_MASK */
> > >
> > > Hugh suggested __i915_sw_fence_call which I think would be the right
> > > thing to do.
> > >
> > > Regards,
> > >
> > > Tvrtko
> > >
> > > > static int sw_fence_dummy_notify(struct i915_sw_fence *sf,
> > > > enum i915_sw_fence_notify state)
> > > > {
> > > > diff --git a/drivers/gpu/drm/i915/i915_sw_fence.c b/drivers/gpu/drm/i915/i915_sw_fence.c
> > > > index c589a681da77..1217b124c1d0 100644
> > > > --- a/drivers/gpu/drm/i915/i915_sw_fence.c
> > > > +++ b/drivers/gpu/drm/i915/i915_sw_fence.c
> > > > @@ -14,8 +14,10 @@
> > > > #if IS_ENABLED(CONFIG_DRM_I915_DEBUG)
> > > > #define I915_SW_FENCE_BUG_ON(expr) BUG_ON(expr)
> > > > +#define I915_SW_FENCE_WARN_ON(expr) WARN_ON(expr)
> > > > #else
> > > > #define I915_SW_FENCE_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
> > > > +#define I915_SW_FENCE_WARN_ON(expr) BUILD_BUG_ON_INVALID(expr)
> > > > #endif
> > > > static DEFINE_SPINLOCK(i915_sw_fence_lock);
> > > > @@ -242,7 +244,7 @@ void __i915_sw_fence_init(struct i915_sw_fence *fence,
> > > > const char *name,
> > > > struct lock_class_key *key)
> > > > {
> > > > - BUG_ON(!fn || (unsigned long)fn & ~I915_SW_FENCE_MASK);
> > > > + I915_SW_FENCE_WARN_ON(!fn || (unsigned long)fn & ~I915_SW_FENCE_MASK);
> > > > __init_waitqueue_head(&fence->wait, name, key);
> > > > fence->flags = (unsigned long)fn;
> > > >
> >
