[Intel-gfx] [PATCH 3/3] drm/i915: Flush the workqueue before draining

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> Quoting Mika Kuoppala (2019-07-04 11:22:17)
>> Chris Wilson <chris at chris-wilson.co.uk> writes:
>> 
>> > Trying to drain a workqueue while we may still be adding to it from
>> > background tasks is, according to kernel/workqueue.c, verboten. So, add
>> > a flush_workqueue() at the start of our cleanup procedure.
>> 
>> I don't get it. drain_workqueue does it's own flushing.
>
> Ordering is important here. The problem with drain_workqueue() is that
> is forbids us from adding more tasks into the workqueue as it drains, so
> before we drain we must plug.
>
> It's just adding more hammers. Eventually it'll break.

If so, then we just increase passes? :)

Ok, I was going to say we don't add to free work from
any of it's handlers.

But then realized this is not about the free list handling,
even tho the freed objects is drained along.

And yes, drain only handles flushing for it's chain.

Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>

> -Chris