[Intel-gfx] [PATCH] drm/i915/pmu: Atomically acquire the gt_pm wakeref

[Tvrtko Ursulin]

On 01/08/2019 15:39, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-08-01 15:21:17)
>>
>> On 01/08/2019 13:20, Chris Wilson wrote:
>>> Currently, we only sample if the intel_gt is awake, but we acquire our
>>> own runtime_pm wakeref. Since intel_gt has transitioned to tracking its
>>> own wakeref, we can atomically test and acquire that wakeref instead.
>>>
>>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>> ---
>>> No automagic drop!
>>> ---
>>>    drivers/gpu/drm/i915/gt/intel_gt_pm.h |  8 +++++++-
>>>    drivers/gpu/drm/i915/i915_pmu.c       | 23 ++++++++---------------
>>>    2 files changed, 15 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
>>> index ba960e1fc209..016298483de7 100644
>>> --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h
>>> +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
>>> @@ -9,7 +9,8 @@
>>>    
>>>    #include <linux/types.h>
>>>    
>>> -struct intel_gt;
>>> +#include "intel_gt_types.h"
>>> +#include "intel_wakeref.h"
>>>    
>>>    enum {
>>>        INTEL_GT_UNPARK,
>>> @@ -19,6 +20,11 @@ enum {
>>>    void intel_gt_pm_get(struct intel_gt *gt);
>>>    void intel_gt_pm_put(struct intel_gt *gt);
>>>    
>>> +static inline bool intel_gt_pm_get_if_awake(struct intel_gt *gt)
>>> +{
>>> +     return intel_wakeref_get_if_active(&gt->wakeref);
>>> +}
>>> +
>>>    void intel_gt_pm_init_early(struct intel_gt *gt);
>>>    
>>>    void intel_gt_sanitize(struct intel_gt *gt, bool force);
>>> diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
>>> index eff86483bec0..e71192804996 100644
>>> --- a/drivers/gpu/drm/i915/i915_pmu.c
>>> +++ b/drivers/gpu/drm/i915/i915_pmu.c
>>> @@ -8,6 +8,7 @@
>>>    #include <linux/pm_runtime.h>
>>>    
>>>    #include "gt/intel_engine.h"
>>> +#include "gt/intel_gt_pm.h"
>>>    
>>>    #include "i915_drv.h"
>>>    #include "i915_pmu.h"
>>> @@ -161,16 +162,12 @@ engines_sample(struct drm_i915_private *dev_priv, unsigned int period_ns)
>>>    {
>>>        struct intel_engine_cs *engine;
>>>        enum intel_engine_id id;
>>> -     intel_wakeref_t wakeref;
>>>        unsigned long flags;
>>>    
>>>        if ((dev_priv->pmu.enable & ENGINE_SAMPLE_MASK) == 0)
>>>                return;
>>>    
>>> -     wakeref = 0;
>>> -     if (READ_ONCE(dev_priv->gt.awake))
>>> -             wakeref = intel_runtime_pm_get_if_in_use(&dev_priv->runtime_pm);
>>> -     if (!wakeref)
>>> +     if (!intel_gt_pm_get_if_awake(&dev_priv->gt))
>>>                return;
>>>    
>>>        spin_lock_irqsave(&dev_priv->uncore.lock, flags);
>>> @@ -205,7 +202,7 @@ engines_sample(struct drm_i915_private *dev_priv, unsigned int period_ns)
>>>        }
>>>        spin_unlock_irqrestore(&dev_priv->uncore.lock, flags);
>>>    
>>> -     intel_runtime_pm_put(&dev_priv->runtime_pm, wakeref);
>>> +     intel_gt_pm_put(&dev_priv->gt);
>>>    }
>>>    
>>>    static void
>>> @@ -222,15 +219,11 @@ frequency_sample(struct drm_i915_private *dev_priv, unsigned int period_ns)
>>>                u32 val;
>>>    
>>>                val = dev_priv->gt_pm.rps.cur_freq;
>>> -             if (dev_priv->gt.awake) {
>>> -                     intel_wakeref_t wakeref;
>>> -
>>> -                     with_intel_runtime_pm_if_in_use(&dev_priv->runtime_pm,
>>> -                                                     wakeref) {
>>> -                             val = intel_uncore_read_notrace(&dev_priv->uncore,
>>> -                                                             GEN6_RPSTAT1);
>>> -                             val = intel_get_cagf(dev_priv, val);
>>> -                     }
>>> +             if (intel_gt_pm_get_if_awake(&dev_priv->gt)) {
>>> +                     val = intel_uncore_read_notrace(&dev_priv->uncore,
>>> +                                                     GEN6_RPSTAT1);
>>> +                     val = intel_get_cagf(dev_priv, val);
>>> +                     intel_gt_pm_put(&dev_priv->gt);
>>>                }
>>>    
>>>                add_sample_mult(&dev_priv->pmu.sample[__I915_SAMPLE_FREQ_ACT],
>>>
>>
>> I guess I'll be rebasing mine, at some point. :)
> 
> I anticipated you merging it at some point.

This patch to my series or what?

> Had a thought, and we don't need gt_pm for the engine sampling, but
> rather intel_engine_pm_get_if_awake.

I thought about it, but since we iterate all engines did not see a real 
benefit. Wouldn't harm either, only some more CPU cycles in the loop, so 
up to you.

Regards,

Tvrtko