Re: [Intel-gfx] [RFC PATCH 2/2] drm/i915/gt: Use spin_lock_irq() instead of local_irq_disable() + spin_lock()

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2021-09-16 11:38:55 [+0200], Maarten Lankhorst wrote:
> Patches look good.
Thank you for looking.

> For both patches:
> 
> Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> 
> I've been looking at running i915 with the -rt patch series, and
> noticed i915_request_submit fails with GEM_BUG_ON(!irqs_disabled());
> presumably same failure exists for i915_request_unsubmit().
> 
> Might be worth removing those checks as well? Seems double with
> lockdep_assert_held on an irq lock anyway.

yes, let me prepare something in a few.

> I've also noticed the local_irq_disable/enable is removed from
> intel_pipe_update_(start/end) in the rt series. It might make sense
> from a -rt point of view, but that code needs to run without
> interruptions, or i915 may show visual glitches or even locks up the
> system.
>
> It should just be a set of registers hammered in, but the code might
> needs to be fixed to take the mmio lock as outer lock, and become a
> strict set of register read/writes only.

Let me see. So Anton Lundin (Cc:) reported glitches due to _this_ patch
on -RT. I have just a Sandybridge around with a i915 and it does not get
near that code here. 

> ~Maarten

Sebastian