[Intel-gfx] [PATCH] drm/i915: fix blank screen booting crashes

[Lucas De Marchi]

On Tue, Sep 21, 2021 at 03:55:15PM -0700, Matthew Brost wrote:
>On Tue, Sep 21, 2021 at 11:46:37AM -0700, Lucas De Marchi wrote:
>> On Tue, Sep 21, 2021 at 10:43:32AM -0700, Matthew Brost wrote:
>> > From: Hugh Dickins <hughd at google.com>
>> >
>> > 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.
>> >
>> > 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
>> > v3:
>> > (Jani / Tvrtko)
>> >  - Short circuit __i915_sw_fence_init on WARN_ON
>> >
>> > Fixes: 62eaf0ae217d ("drm/i915/guc: Support request cancellation")
>> > Signed-off-by: Hugh Dickins <hughd at google.com>
>> > Signed-off-by: Matthew Brost <matthew.brost at intel.com>
>> > Reviewed-by: Matthew Brost <matthew.brost at intel.com>
>> > ---
>> > drivers/gpu/drm/i915/gt/intel_context.c |  4 ++--
>> > drivers/gpu/drm/i915/i915_sw_fence.c    | 17 ++++++++++-------
>> > 2 files changed, 12 insertions(+), 9 deletions(-)
>> >
>> > diff --git a/drivers/gpu/drm/i915/gt/intel_context.c b/drivers/gpu/drm/i915/gt/intel_context.c
>> > index ff637147b1a9..e7f78bc7ebfc 100644
>> > --- a/drivers/gpu/drm/i915/gt/intel_context.c
>> > +++ b/drivers/gpu/drm/i915/gt/intel_context.c
>> > @@ -362,8 +362,8 @@ static int __intel_context_active(struct i915_active *active)
>> > 	return 0;
>> > }
>> >
>>
>> > -static int sw_fence_dummy_notify(struct i915_sw_fence *sf,
>> > -				 enum i915_sw_fence_notify state)
>> > +static int __i915_sw_fence_call
>> > +sw_fence_dummy_notify(struct i915_sw_fence *sf, enum i915_sw_fence_notify state)
>> > {
>> > 	return NOTIFY_DONE;
>> > }
>> > diff --git a/drivers/gpu/drm/i915/i915_sw_fence.c b/drivers/gpu/drm/i915/i915_sw_fence.c
>> > index c589a681da77..08cea73264e7 100644
>> > --- a/drivers/gpu/drm/i915/i915_sw_fence.c
>> > +++ b/drivers/gpu/drm/i915/i915_sw_fence.c
>> > @@ -13,9 +13,9 @@
>> > #include "i915_selftest.h"
>> >
>> > #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);
>> > @@ -129,7 +129,10 @@ static int __i915_sw_fence_notify(struct i915_sw_fence *fence,
>> > 	i915_sw_fence_notify_t fn;
>> >
>> > 	fn = (i915_sw_fence_notify_t)(fence->flags & I915_SW_FENCE_MASK);
>> > -	return fn(fence, state);
>> > +	if (likely(fn))
>> > +		return fn(fence, state);
>> > +	else
>> > +		return 0;
>>
>> since the knowledge for these being NULL (or with the wrong alignment)
>> are in the init/reinit functions,  wouldn't it be better to just add a
>> fence_nop() and assign it there instead this likely() here?
>>
>
>Maybe? I prefer the way it is.
>
>> > }
>> >
>> > #ifdef CONFIG_DRM_I915_SW_FENCE_DEBUG_OBJECTS
>> > @@ -242,9 +245,9 @@ 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);
>> > -
>> > 	__init_waitqueue_head(&fence->wait, name, key);
>> > +	if (WARN_ON(!fn || (unsigned long)fn & ~I915_SW_FENCE_MASK))
>> > +		return;
>>
>> like:
>> 	if (WARN_ON(!fn || (unsigned long)fn & ~I915_SW_FENCE_MASK))
>> 		fence->flags = (unsigned long)sw_fence_dummy_notify;
>> 	else
>> 		fence->flags = (unsigned long)fn;
>>
>>
>> f you return here instead of calling i915_sw_fence_reinit(), aren't you
>> just going to use uninitialized memory later? At least in the selftests,
>> which allocate it with kmalloc()... I didn't check others.
>>
>
>I don't think so, maybe the fence won't work but it won't blow up
>either.
>
>>
>> For the bug fix we could just add the __aligned(4) and leave the rest to a
>> separate patch.
>>
>
>The bug was sw_fence_dummy_notify in gt/intel_context.c was not 4 byte
>align which triggered a BUG_ON during boot which blank screened a
>laptop. Jani / Tvrtko suggested that we make the BUG_ON to WARN_ONs so
>if someone makes this mistake in the future kernel should boot albiet
>with a WARNING.

yes, I understood. But afaics with WARN_ON you are allowing it to
continue and may be using uninitialized memory later, just causing other
problems down the line, which may be equally difficult to debug.

what I suggested is that there is the easy fix to apply to the current
rcX kernel, adding __aligned(4) to sw_fence_dummy_notify() (patch 1).
And there is the additional protection being added here (patch 2) which
is subject to the debate.

>
>The long term fix is just pull out the I915_SW_FENCE_MASK (stealing bits
>from a poitner) and we don't have to worry any of this.

Patch 2 may not even be needed if you're going that route. But we are
not only protecting against unaligned, but also from code calling
i915_sw_fence_init() with a NULL fn. 

Lucas De Marchi

>
>Matt
>
>>
>> Lucas De Marchi
>>
>> > 	fence->flags = (unsigned long)fn;
>> >
>> > 	i915_sw_fence_reinit(fence);
>> > @@ -257,8 +260,8 @@ void i915_sw_fence_reinit(struct i915_sw_fence *fence)
>> > 	atomic_set(&fence->pending, 1);
>> > 	fence->error = 0;
>> >
>> > -	I915_SW_FENCE_BUG_ON(!fence->flags);
>> > -	I915_SW_FENCE_BUG_ON(!list_empty(&fence->wait.head));
>> > +	I915_SW_FENCE_WARN_ON(!fence->flags);
>> > +	I915_SW_FENCE_WARN_ON(!list_empty(&fence->wait.head));
>> > }
>> >
>> > void i915_sw_fence_commit(struct i915_sw_fence *fence)
>> > --
>> > 2.32.0
>> >