[Intel-gfx] [PATCH 3.0-rc3] i915: Fix gen6 (SNB) GPU stalling

[Daniel J Blueman]

On 15 June 2011 23:16, Ben Widawsky <ben at bwidawsk.net> wrote:
> On Wed, Jun 15, 2011 at 01:04:51PM +0800, Daniel J Blueman wrote:
>> On 15 June 2011 12:43, Ben Widawsky <ben at bwidawsk.net> wrote:
>> > On Wed, Jun 15, 2011 at 12:51:47AM +0800, Daniel J Blueman wrote:
>> >> On 14 June 2011 13:23, Eric Anholt <eric at anholt.net> wrote:
>> >> > On Tue, 14 Jun 2011 12:18:36 +0800, Daniel J Blueman <daniel.blueman at gmail.com> wrote:
>> >> >> Hi Eric,
>> >> >>
>> >> >> The frequent ~1.5s pauses I hit with SNB hardware in the gnome3 UI (eg
>> >> >> whenever you hit the top-left of the screen to show all windows) are
>> >> >> nicely addressed by your recent wake patch [1] (ported to -rc3). Thus
>> >> >> I see no 'missed IRQ' kernel messages.
>> >> >>
>> >> >> As this addresses a significant usability regression, are you happy to
>> >> >> add it to the 3.0-rc queue? I think it has very good value in -stable
>> >> >> also (assuming correctness). What do you think?
>> >> >
>> >> > This one had significant performance impacts, and later hacks in this
>> >> > series worked around the problem to approximately the same level of
>> >> > success with less impact, and we don't actually have a justification of
>> >> > why any of them work. ?We were still hoping to come up with some clue,
>> >> > and haven't yet.
>> >>
>> >> True; that is quite heavy handed delay looping.
>> >>
>> >> It's a pity the usual Intel font didn't make it to the programmer's
>> >> reference manuals. Anyway, unmasking the blitter user interrupt in the hardware
>> >> status mask register addresses the root cause. Out of reset it's FFFFFFFFh,
>> >> so we don't need to read it here.
>> >>
>> >> It would be good to get this into -rc4. -stable probably needs some additional
>> >> tweaks.
>> >>
>> >> Signed-off-by: Daniel J Blueman <daniel.blueman at gmail.com>
>> >> ---
>> >> ?drivers/gpu/drm/i915/i915_irq.c | ? ?6 ++++++
>> >> ?1 files changed, 6 insertions(+), 0 deletions(-)
>> >>
>> >> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
>> >> index b9fafe3..9a98c1b 100644
>> >> --- a/drivers/gpu/drm/i915/i915_irq.c
>> >> +++ b/drivers/gpu/drm/i915/i915_irq.c
>> >> @@ -1827,6 +1827,12 @@ int ironlake_irq_postinstall(struct drm_device *dev)
>> >> ? ? ? ? ? ? ? ironlake_enable_display_irq(dev_priv, DE_PCU_EVENT);
>> >> ? ? ? }
>> >>
>> >> + ? ? if (IS_GEN6(dev))
>> >> + ? ? ? ? ? ? /* allow blitter user interrupt to generate a MSI write from
>> >> + ? ? ? ? ? ? ? ?the ISR */
>> >> + ? ? ? ? ? ? I915_WRITE(GEN6_BLITTER_HWSTAM,
>> >> + ? ? ? ? ? ? ? ? ? ? 0xffffffff & ~GEN6_BLITTER_USER_INTERRUPT);
>> >> +
>> >> ? ? ? return 0;
>> >> ?}
>> >>
>> >
>> > I wish the docs said that that this hwstam unmasked MI_USER_INTERRUPTS
>> > parsed by the Blitter Command Parser, instead of the Render Command
>> > parser.
>> >
>> > I was just about to write an email about how this is just making the
>> > same interrupt happen twice, when I realized that the docs make no
>> > sense, and this must be a copy-paste doc bug.
>> >
>> > This patch sounds good to me. Two small comments:
>> > 1. The HWSTAM is touched in preinstall already, why not move this there?
>> > 2. I'd prefer you read the register even though as you say it isn't
>> > necessary. It just makes the code self-documented by doing it that way.
>>
>> The render HWSTAM is tweaked in preinstall, but we need to tweak the
>> blitter HWSTAM (new to gen6).
>>
>> To me, it makes sense to reset the blitter HWSTAM register to what the
>> driver expects, in case anything before the i915 module loads and
>> adjusts it for a particular purpose (including debug); the render
>> HWSTAM is set this way too. I could add a comment to both perhaps?
>
> Well on that note, the docs clearly state only 1 bit can be unmasked at
> a time. Not sure if that applies to masking as well, but if it does,
> that would be not good.

When a bit is unmasked, logic in the silicon will generate a write
from a particular functional block; multiple writes in the same cycles
may generate undefined behaviour (though would need multiple pending
interrupts). Masking would be safe, since there is no side-effect, and
this is typical.

> I'm fine with a comment. Seeing the current masking doesn't make it
> clear to me what is happening/why. Not sure I understand the reason for
> saving the read is in this case.
>
>>
>> Updating the blitter HWSTAM in the postinstall was a marginally safer
>> choice, as there'll not be any potential race with a blitter user
>> interrupt getting emitted before we're ready (which wouldn't have been
>> tested), but the risk is probably so low that it could just go into
>> the preinstall.
>>
>> What do you think?
>
> I think there is no risk since this command could only be executed if
> the driver was up. I'd just like all HWSTAM writes in a single place. I
> don't have any preference which place.

Ok, I'll prepare a patch that will read-modify-write the register and
place with the render HWSTAM write. Note that the render HWSTAM
unmasks multiple bits with impunity (probably fine if no pending
interrupts), but fixing that is outside the scope of what I want to
get into 3.0-rc4.

Thanks,
  Daniel
-- 
Daniel J Blueman