[PATCH] drm: simple_kms_helper: Handle the vblank events

[Marek Vasut]

On 09/26/2016 11:41 AM, Marek Vasut wrote:
> On 09/25/2016 11:00 PM, Daniel Vetter wrote:
>> On Sun, Sep 25, 2016 at 09:41:58PM +0200, Marek Vasut wrote:
>>> Handle the vblank events in the simple_kms_helper driver, otherwise
>>> the drm_atomic_helper flip_done event never happens.
>>>
>>> Signed-off-by: Marek Vasut <marex at denx.de>
>>> Cc: Noralf Trønnes <noralf at tronnes.org>
>>> Cc: Daniel Vetter <daniel at ffwll.ch>
>>> Cc: David Airlie <airlied at linux.ie>
>>> ---
>>>  drivers/gpu/drm/drm_simple_kms_helper.c | 18 ++++++++++++++++++
>>>  1 file changed, 18 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/drm_simple_kms_helper.c b/drivers/gpu/drm/drm_simple_kms_helper.c
>>> index 7b6d26e..3345b40 100644
>>> --- a/drivers/gpu/drm/drm_simple_kms_helper.c
>>> +++ b/drivers/gpu/drm/drm_simple_kms_helper.c
>>> @@ -34,6 +34,23 @@ static const struct drm_encoder_funcs drm_simple_kms_encoder_funcs = {
>>>  	.destroy = drm_encoder_cleanup,
>>>  };
>>>  
>>> +static void drm_simple_kms_crtc_begin(struct drm_crtc *crtc,
>>> +				      struct drm_crtc_state *state)
>>> +{
>>> +	struct drm_pending_vblank_event *event = crtc->state->event;
>>> +
>>> +	if (event) {
>>> +		crtc->state->event = NULL;
>>> +
>>> +		spin_lock_irq(&crtc->dev->event_lock);
>>> +		if (drm_crtc_vblank_get(crtc) == 0)
>>> +			drm_crtc_arm_vblank_event(crtc, event);
>>> +		else
>>> +			drm_crtc_send_vblank_event(crtc, event);
>>> +		spin_unlock_irq(&crtc->dev->event_lock);
>>> +	}
>>> +}
>>
>> This should be done by drivers, in the ->update hook. At least if we want
>> to pretend that it's semi-accurate and not racy (which the above is).
>> -Daniel
> 
> Got it and wrapped into mxsfb, thanks.

Hrm, while this works most of the time, I managed to get a page_flip
timeout when terminating the kmscube OpenGL demo (see below). I'm not
getting this splat with this patch applied. Any idea what this could
mean ?

 ------------[ cut here ]------------
 WARNING: CPU: 0 PID: 77 at drivers/gpu/drm/drm_atomic_helper.c:1579
drm_atomic_helper_commit_hw_done+0xa8/0xb8
 Modules linked in:
 CPU: 0 PID: 77 Comm: kworker/0:2 Tainted: G        W
4.8.0-rc7-next-20160923-00033-ge3f15d6 #56
 Hardware name: Freescale i.MX6 SoloX (Device Tree)
 Workqueue: events drm_mode_rmfb_work_fn
 Backtrace:
 [<c010c180>] (dump_backtrace) from [<c010c378>] (show_stack+0x18/0x1c)
  r7:00000000 r6:600e0013 r5:00000000 r4:c0e2342c
 [<c010c360>] (show_stack) from [<c03f4ad8>] (dump_stack+0xb4/0xe8)
 [<c03f4a24>] (dump_stack) from [<c0124a90>] (__warn+0xd8/0x104)
  r9:c04bdb60 r8:0000062b r7:00000009 r6:c0c38298 r5:00000000 r4:00000000
 [<c01249b8>] (__warn) from [<c0124b70>] (warn_slowpath_null+0x28/0x30)
  r9:db43639c r8:0000062b r7:c0c38298 r6:da962e40 r5:00000000 r4:00000000
 [<c0124b48>] (warn_slowpath_null) from [<c04bdb60>]
(drm_atomic_helper_commit_hw_done+0xa8/0xb8)
 [<c04bdab8>] (drm_atomic_helper_commit_hw_done) from [<c04c0568>]
(drm_atomic_helper_commit_tail+0x44/0x60)
  r10:c0c3bd84 r9:c0c371cc r8:db770000 r7:db770000 r6:00000000 r5:db770000
  r4:daffa780 r3:daffa3c0
 [<c04c0524>] (drm_atomic_helper_commit_tail) from [<c04c05e8>]
(commit_tail+0x64/0x68)
  r5:00000000 r4:daffa780
 [<c04c0584>] (commit_tail) from [<c04c06b0>]
(drm_atomic_helper_commit+0xac/0xec)
  r5:00000000 r4:daffa780
 [<c04c0604>] (drm_atomic_helper_commit) from [<c04e11b4>]
(drm_atomic_commit+0x58/0x78)
  r7:db436028 r6:db770000 r5:daffa780 r4:00000000
 [<c04e115c>] (drm_atomic_commit) from [<c04c0be4>]
(drm_atomic_helper_set_config+0x80/0xa4)
  r6:dad37e14 r5:00000000 r4:daffa780
 [<c04c0b64>] (drm_atomic_helper_set_config) from [<c04d35fc>]
(drm_mode_set_config_internal+0x6c/0xfc)
  r7:db77048c r6:db770480 r5:db436028 r4:00000000
 [<c04d3590>] (drm_mode_set_config_internal) from [<c04d36c4>]
(drm_crtc_force_disable+0x38/0x40)
  r7:db77048c r6:db770480 r5:da962c00 r4:db436028
 [<c04d368c>] (drm_crtc_force_disable) from [<c04e2b60>]
(drm_framebuffer_remove+0xd8/0x124)
 [<c04e2a88>] (drm_framebuffer_remove) from [<c04e2bec>]
(drm_mode_rmfb_work_fn+0x40/0x50)
  r10:00000001 r9:00000000 r8:dbbb2a00 r7:dad37ea8 r6:dbbaf800 r5:db60be68
  r4:db60be48
 [<c04e2bac>] (drm_mode_rmfb_work_fn) from [<c013fa10>]
(process_one_work+0x24c/0x4e8)
  r5:dad12f00 r4:db60be48
 [<c013f7c4>] (process_one_work) from [<c01408dc>]
(worker_thread+0x40/0x574)
  r10:dad12f00 r9:dad36000 r8:c0e02100 r7:dbbaf834 r6:00000008 r5:dad12f18
  r4:dbbaf800
 [<c014089c>] (worker_thread) from [<c0146908>] (kthread+0xd8/0xf8)
  r10:00000000 r9:00000000 r8:00000000 r7:c014089c r6:dad12f00 r5:dad28500
  r4:00000000
 [<c0146830>] (kthread) from [<c0107cb0>] (ret_from_fork+0x14/0x24)
  r7:00000000 r6:00000000 r5:c0146830 r4:dad28500
 ---[ end trace 2b6e3065f0299ba5 ]---

> But then, I see a few drivers (arm hdlcd, fsl-dcu,...) doing the same
> thing at random callbacks of CRTC . Shouldn't this event handling be
> consolidated into some generic function and those drivers fixed to
> call it from atomic update ?
> 


-- 
Best regards,
Marek Vasut