[Intel-gfx] [PATCH] drm/i915/ilk: special case enabling of PCU_EVENT interrupt

[Daniel Vetter]

On Tue, Aug 26, 2014 at 11:18 PM, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
> Daniel Vetter <daniel at ffwll.ch> wrote:
>
>>
>> On Tue, Aug 26, 2014 at 8:52 PM, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
>> > On Tue, 26 Aug 2014 09:23:54 +0200
>> > Daniel Vetter <daniel at ffwll.ch> wrote:
>> >
>> >> On Mon, Aug 25, 2014 at 04:24:55PM -0700, Jesse Barnes wrote:
>> >> > This happens in irq_postinstall before we've set the pm._irqs_disabled flag,
>> >> > but shouldn't warn.  So add a nowarn variant to allow this to happen w/o
>> >> > a backtrace and keep the rest of the IRQ tracking code happy.
>> >> >
>> >> > Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
>> >>
>> >> Shouldn't we instead just move the pm._irqs_disabled = false in i915_dma.c
>> >> right above the drm_irq_install call? In
>> >> intel_runtime_pm_restore_interrupts we also set it to false before we call
>> >> the various hooks.
>> >
>> > I didn't check on all the paths whether that was safe, we have a lot of
>> > warnings.
>> >
>> > And really this init time thing is a special case, so it made sense to
>> > me.
>>
>> Well I fully agree that your patch is the safe option and much easier to
>> review, too.
>>
>> But driver load/resume are the most fragile paths we have in our codebase
>> - you look at them and it falls apart. And we have absolutely no handle on
>> these issues on a fundamental level at all (compared to other areas where
>> we reworked the code or added enough tests to pretty much kill entire
>> classes of regressions). The only half-assed thing we can do is try to not
>> have too much complexity (so that you can still understand it, we're
>> probably over that already) and lock down the ordering and other
>> constraints with piles of really loud WARN_ON asserts.
>>
>> Your patch both removes WARN_ONs from these codepaths and adds special
>> cases, so falls a bit short on those metrics. And if I'm not mistaken
>> (like I've said the code is too complex by now to really understand) the
>> below change should get us there, too. So I want to see whether that
>> wouldn't work before going with your patch.
>
> Yes, it might work, but if you look through the history, we set this
> field carefully; first to true in the irq_init code, then to false only
> after the irq_install completes.  So I think your fragility arguments
> apply to this change too.

Well we've done it in 4 commits or so, but currently we have:

- Set irqs_disabled to true early in driver load to make sure checks
that. That's done in irq_init, which is totally not the function that
enables interrupts, only the function that initializes all the vtables
and similar things. We actually have a fairly sane naming scheme
nowadays (not fully consistent ofc): _init is sw setup,
_enable/_hw_init is the actual hw setup. That is done in
95f25beddba2ec9510b249740bacc11eca70cf75

- Set irqs_disabled to false right after the irqs are actually
enabled. This is done in ed2e6df18935beb3d63613c50103bf9757b2aa85

So my change should only move the flag change over the ->preinstall
and ->postinstall hooks. I've done a little audit and didn't spot
anything amiss. Furthermore the runtime pm setup already clears
irqs_disabled _before_ calling these two hooks.

So I think overall this is solid. And now I've also written the commit
message for the patch if it tests out on Oliver's box ;-)

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch