[Intel-gfx] [PATCH 2/2] drm/i915: Use init power domain during reset

Ville Syrjälä ville.syrjala at linux.intel.com
Tue Jan 19 12:05:41 PST 2016


On Tue, Jan 19, 2016 at 09:50:09PM +0200, Mika Kuoppala wrote:
> If we have driver failure in our power well and/or dc
> state keeping, we might try to reset without powers.
> 
> Evidence shows that resetting the chip with dc6 enabled
> don't lead to desired results. The dmc kept it's enabled
> dc6 state over the reset. On subsequent init the rings
> just hanged right from the start so they didn't reset
> properly or that the dmc interfered with the init.
> 
> Harden the reset by setting power domains to be in
> init mode during reset. This causes the hw state to be
> flushed so dc6 will be forcibly disabled also.

Hmm. Aren't we holding a forcewake around the reset already? Also with
the GT hung it shouldn't really be dropping into RC6 anyway, which
should keep the thing in pc2 (IIRC) or higher. I guess if we do a
simulated hang it might be in rc6, but the forcewake should be enough
protection there.

Deep package C-state is the only link I could think of between the GT
and DMC. Otherwise I can't think of any real link between the two. So
quite baffled by this.

> 
> References: https://bugs.freedesktop.org/show_bug.cgi?id=93768
> Cc: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Patrik Jakobsson <patrik.jakobsson at linux.intel.com>
> Cc: Imre Deak <imre.deak at intel.com>
> Cc: Ville Syrjälä <ville.syrjala at linux.intel.com>
> Signed-off-by: Mika Kuoppala <mika.kuoppala at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_irq.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
> index 25a89373df63..78e242b5c357 100644
> --- a/drivers/gpu/drm/i915/i915_irq.c
> +++ b/drivers/gpu/drm/i915/i915_irq.c
> @@ -2510,6 +2510,14 @@ static void i915_reset_and_wakeup(struct drm_device *dev)
>  		 */
>  		intel_runtime_pm_get(dev_priv);
>  
> +		/* Even if we hold the pm ref, we still might have inconsistent
> +		 * power states due to driver failure. Trying to reset without
> +		 * powers or with wrong dmc firmware state is futile. Flush
> +		 * our power well and dc states ensuring that we reset with
> +		 * powers enabled.
> +		 */
> +		intel_display_set_init_power(dev_priv, true);
> +
>  		intel_prepare_reset(dev);
>  
>  		/*
> @@ -2522,6 +2530,8 @@ static void i915_reset_and_wakeup(struct drm_device *dev)
>  
>  		intel_finish_reset(dev);
>  
> +		intel_display_set_init_power(dev_priv, false);
> +
>  		intel_runtime_pm_put(dev_priv);
>  
>  		if (ret == 0) {
> -- 
> 2.5.0

-- 
Ville Syrjälä
Intel OTC


More information about the Intel-gfx mailing list