Re: [PATCH 3/6] drm/i915: Only track live rings for retiring

[Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>]

On 23/04/2018 11:36, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2018-04-23 11:25:54)
>>
>> On 23/04/2018 11:13, Chris Wilson wrote:
>>> We don't need to track every ring for its lifetime as they are managed
>>> by the contexts/engines. What we do want to track are the live rings so
>>> that we can sporadically clean up requests if userspace falls behind. We
>>> can simply restrict the gt->rings list to being only gt->live_rings.
>>>
>>> Suggested-by: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
>>> Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
>>> Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
>>> ---
>>>    drivers/gpu/drm/i915/i915_drv.h                  | 2 +-
>>>    drivers/gpu/drm/i915/i915_gem.c                  | 3 ++-
>>>    drivers/gpu/drm/i915/i915_request.c              | 6 +++++-
>>>    drivers/gpu/drm/i915/i915_utils.h                | 6 ++++++
>>>    drivers/gpu/drm/i915/intel_ringbuffer.c          | 4 ----
>>>    drivers/gpu/drm/i915/intel_ringbuffer.h          | 2 +-
>>>    drivers/gpu/drm/i915/selftests/mock_engine.c     | 4 ----
>>>    drivers/gpu/drm/i915/selftests/mock_gem_device.c | 2 +-
>>>    8 files changed, 16 insertions(+), 13 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
>>> index 73936be90aed..a7787c2cb53c 100644
>>> --- a/drivers/gpu/drm/i915/i915_drv.h
>>> +++ b/drivers/gpu/drm/i915/i915_drv.h
>>> @@ -2060,7 +2060,7 @@ struct drm_i915_private {
>>>    
>>>                struct i915_gem_timeline global_timeline;
>>>                struct list_head timelines;
>>> -             struct list_head rings;
>>> +             struct list_head live_rings;
>>>                u32 active_requests;
>>>    
>>>                /**
>>> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
>>> index 906e2395c245..0097a77fae8d 100644
>>> --- a/drivers/gpu/drm/i915/i915_gem.c
>>> +++ b/drivers/gpu/drm/i915/i915_gem.c
>>> @@ -141,6 +141,7 @@ static u32 __i915_gem_park(struct drm_i915_private *i915)
>>>    {
>>>        lockdep_assert_held(&i915->drm.struct_mutex);
>>>        GEM_BUG_ON(i915->gt.active_requests);
>>> +     GEM_BUG_ON(!list_empty(&i915->gt.live_rings));
>>>    
>>>        if (!i915->gt.awake)
>>>                return I915_EPOCH_INVALID;
>>> @@ -5600,7 +5601,7 @@ int i915_gem_init_early(struct drm_i915_private *dev_priv)
>>>                goto err_dependencies;
>>>    
>>>        mutex_lock(&dev_priv->drm.struct_mutex);
>>> -     INIT_LIST_HEAD(&dev_priv->gt.rings);
>>> +     INIT_LIST_HEAD(&dev_priv->gt.live_rings);
>>>        INIT_LIST_HEAD(&dev_priv->gt.timelines);
>>>        err = i915_gem_timeline_init__global(dev_priv);
>>>        mutex_unlock(&dev_priv->drm.struct_mutex);
>>> diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
>>> index 0bf949ec9f1a..534b8d684cef 100644
>>> --- a/drivers/gpu/drm/i915/i915_request.c
>>> +++ b/drivers/gpu/drm/i915/i915_request.c
>>> @@ -316,6 +316,7 @@ static void advance_ring(struct i915_request *request)
>>>                 * noops - they are safe to be replayed on a reset.
>>>                 */
>>>                tail = READ_ONCE(request->tail);
>>> +             list_del(&ring->live);
>>>        } else {
>>>                tail = request->postfix;
>>>        }
>>> @@ -1046,6 +1047,8 @@ void __i915_request_add(struct i915_request *request, bool flush_caches)
>>>        i915_gem_active_set(&timeline->last_request, request);
>>>    
>>>        list_add_tail(&request->ring_link, &ring->request_list);
>>> +     if (list_is_first(&request->ring_link, &ring->request_list))
>>> +             list_add(&ring->live, &request->i915->gt.live_rings);
>>
>> If you re-order the two list adds you could use list_empty and wouldn't
>> have to add list_is_first.
> 
> list_is_first tallies nicely with the list_is_last used before the
> corresponding list_del.

Yes but to me that's minor, basically immaterial as argument whether or 
not to add our own list helper.

>>
>>>        request->emitted_jiffies = jiffies;
>>>    
>>>        /*
>>> @@ -1375,7 +1378,8 @@ void i915_retire_requests(struct drm_i915_private *i915)
>>>        if (!i915->gt.active_requests)
>>>                return;
>>>    
>>> -     list_for_each_entry_safe(ring, next, &i915->gt.rings, link)
>>> +     GEM_BUG_ON(list_empty(&i915->gt.live_rings));
>>
>> Maybe blank line here since the assert is not logically associated with
>> the list but with the !i915.active_requests?
> 
> I was thinking list when I wrote it. It's small enough we can argue both
> and both be right.

Hm obviosuly it is not an error to call i915_retire_requests with 
nothing active (early return). So I even briefly wanted to suggest to 
make it 100% explicit and have the assert at the top of the function as:

GEM_BUG_ON(!!i915->gt.active_requests ^ !!list_empty(..));

Unless I messed it up, the idea is to check those two are always in the 
same state.

> 
>>
>>> +     list_for_each_entry_safe(ring, next, &i915->gt.live_rings, live)
>>>                ring_retire_requests(ring);
>>>    }
>>>    
>>> diff --git a/drivers/gpu/drm/i915/i915_utils.h b/drivers/gpu/drm/i915/i915_utils.h
>>> index 0695717522ea..00165ad55fb3 100644
>>> --- a/drivers/gpu/drm/i915/i915_utils.h
>>> +++ b/drivers/gpu/drm/i915/i915_utils.h
>>> @@ -120,6 +120,12 @@ static inline u64 ptr_to_u64(const void *ptr)
>>>    
>>>    #include <linux/list.h>
>>>    
>>> +static inline int list_is_first(const struct list_head *list,
>>> +                             const struct list_head *head)
>>
>> Return bool if you decide you prefer to keep list_is_first?
> 
> Copy'n'paste from list_is_last().
> 
>>
>>> +{
>>> +     return head->next == list;
>>> +}
>>> +
>>>    static inline void __list_del_many(struct list_head *head,
>>>                                   struct list_head *first)
>>>    {
>>> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
>>> index 792a2ca95872..3453e7426f6b 100644
>>> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
>>> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
>>> @@ -1150,8 +1150,6 @@ intel_engine_create_ring(struct intel_engine_cs *engine, int size)
>>>        }
>>>        ring->vma = vma;
>>>    
>>> -     list_add(&ring->link, &engine->i915->gt.rings);
>>> -
>>>        return ring;
>>>    }
>>>    
>>> @@ -1163,8 +1161,6 @@ intel_ring_free(struct intel_ring *ring)
>>>        i915_vma_close(ring->vma);
>>>        __i915_gem_object_release_unless_active(obj);
>>>    
>>> -     list_del(&ring->link);
>>> -
>>>        kfree(ring);
>>>    }
>>>    
>>> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
>>> index d816f8dea245..fd5a6363ab1d 100644
>>> --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
>>> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
>>> @@ -129,7 +129,7 @@ struct intel_ring {
>>>        void *vaddr;
>>>    
>>>        struct list_head request_list;
>>> -     struct list_head link;
>>> +     struct list_head live;
>>
>> live_link?
> 
> live or active.
> 
> active_rings ties in with active_requests, so active_link here.

Fine by me.

Regards,

Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx