[Intel-gfx] [PATCH] drm/i915: Move drm_framebuffer_unreference out of struct_mutex for takeover

[Tvrtko Ursulin]

On 04/13/2015 01:09 PM, Jani Nikula wrote:
> On Thu, 26 Mar 2015, Tvrtko Ursulin <tvrtko.ursulin at linux.intel.com> wrote:
>> On 03/26/2015 01:30 PM, Ville Syrjälä wrote:
>>> On Thu, Mar 26, 2015 at 12:39:40PM +0000, Tvrtko Ursulin wrote:
>>>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>>
>>>> intel_user_framebuffer_destroy() requires the struct_mutex for its
>>>> object bookkeeping, so this means that all calls to
>>>> drm_framebuffer_unreference must be held without that lock.
>>>>
>>>> This is a simplified version of the identically named patch by Chris Wilson.
>>>>
>>>> References: https://bugs.freedesktop.org/show_bug.cgi?id=89166
>>>> Cc: Chris Wilson <chris at chris-wilson.co.uk>
>>>> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>>> ---
>>>>    drivers/gpu/drm/i915/intel_display.c | 10 ++++++++++
>>>>    1 file changed, 10 insertions(+)
>>>>
>>>> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
>>>> index cb50854..0788507 100644
>>>> --- a/drivers/gpu/drm/i915/intel_display.c
>>>> +++ b/drivers/gpu/drm/i915/intel_display.c
>>>> @@ -14020,11 +14020,21 @@ void intel_modeset_gem_init(struct drm_device *dev)
>>>>    					       c->primary->fb,
>>>>    					       c->primary->state,
>>>>    					       NULL)) {
>>>> +			/*
>>>> +			 * We must drop struct_mutex when calling
>>>> +			 * drm_framebuffer_unreference and it is safe to do so
>>>> +			 * because it is not needed at this point anyway.
>>>> +			 * At this stage the driver is still single-threaded and
>>>> +			 * we are taking it only to silence a warning in
>>>> +			 * intel_pin_and_fence_fb_obj.
>>>> +			 */
>>>> +			mutex_unlock(&dev->struct_mutex);
>>>>    			DRM_ERROR("failed to pin boot fb on pipe %d\n",
>>>>    				  to_intel_crtc(c)->pipe);
>>>>    			drm_framebuffer_unreference(c->primary->fb);
>>>>    			c->primary->fb = NULL;
>>>>    			update_state_fb(c->primary);
>>>> +			mutex_lock(&dev->struct_mutex);
>>>>    		}
>>>>    	}
>>>>    	mutex_unlock(&dev->struct_mutex);
>>>
>>> Just grab the mutex around the pin_and_fence inside the loop. It doesn't
>>> protect anything else.
>>
>> Well the comment says so, but this way it only grabs and releases it
>> once if there are multiple active crtcs and nothing fails. So I was
>> hoping the comment was enough to explain the reality, even though the
>> other option would be more obvious code strictly speaking.
>
> Tvrtko & Ville, can you reach a solution on this one? Or is there a
> new patch that I may have missed?

It was pretty much bike shedding - I am happy with this version since it 
has a single lock/unlock on the normal path, compared to one pair per 
active display with what Ville wanted.

Either approach makes for unclear code so needs a big comment anyway. 
Which leaves only the exact placement of mutex_lock/unlock under discussion.

If we want to spend this much time on this that is.

Regards,

Tvrtko