[Nouveau] [PATCH] nv50/disp: Fix modeset on G94

[poma]

On 31.10.2014 11:28, Roy Spliet wrote:
> 
> --- Ursprüngliche Nachricht ---
> Von: Ben Skeggs <skeggsb at gmail.com>
> Datum: 00:52:05 31-10-2014
> An: Ilia Mirkin <imirkin at alum.mit.edu>
> Betreff: Re: [Nouveau] [PATCH] nv50/disp: Fix modeset on G94
> 
>> On Fri, Oct 31, 2014 at 8:00 AM, Ilia Mirkin <imirkin at alum.mit.edu>
>> wrote:
>>> On Thu, Oct 30, 2014 at 5:57 PM, Roy Spliet <rspliet at eclipso.eu>
>> wrote:
>>>> Commit 1dce6264045cd23e9c07574ed0bb31c7dce9354f introduced a regression
>> spotted
>>>> on several G94 (FDObz #85160). This device seems to expect the vblank
>> period to
>>>
>>> I believe that's often done as a
>>>
>>> Bugzilla: https://bugs.freedesktop.org/bla
>>>
>>> annotation
>>>
>>>> be set after setting scale instead of before.
>>>>
>>>> V2: shove this in a separate function
>>>>
>>>> This is a candidate bug-fix for 3.18
>>>>
>>>> Signed-off-by: Roy Spliet <rspliet at eclipso.eu>
>>>> Tested-by: Zlatko Calusic <zcalusic at bitsync.net>
>>>> Tested-by: Michael Riesch <michael at riesch.at>
>>>> Tested-by: "poma" <pomidorabelisima at gmail.com>
>>>> Tested-by: Adam Williamson <adamw at happyassassin.net>
>>>> ---
>>>>  drivers/gpu/drm/nouveau/nv50_display.c | 26 ++++++++++++++++++++++++--
>>
>>>>  1 file changed, 24 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
>>
>>>> index ae873d1..2f24a08 100644
>>>> --- a/drivers/gpu/drm/nouveau/nv50_display.c
>>>> +++ b/drivers/gpu/drm/nouveau/nv50_display.c
>>>> @@ -791,6 +791,23 @@ nv50_crtc_set_scale(struct nouveau_crtc *nv_crtc,
>> bool update)
>>>>  }
>>>>
>>>>  static int
>>>> +nv50_crtc_set_raster_vblank_dmi(struct nouveau_crtc *nv_crtc, u32
>> usec)
>>>
>>> What's "dmi"?
>> SetRasterVertBlankDmi is the name of method 0x828.  I presume it's
>> Display Memory Interface or something to that effect.
>>
>>>
>>>> +{
>>>> +       struct nv50_mast *mast = nv50_mast(nv_crtc->base.dev);
>>
>>>> +       u32 *push;
>>>> +
>>>> +       push = evo_wait(mast, 8);
>>>
>>> Just needs to be 2, no?
>> Yes, doesn't matter too much though.
> If it is, we might need to fix nv50_crtc_mode_set() too; it seems to assume the second parameter in evo_wait() is bytes, not words.
> 
>>
>>>
>>>> +       if (!push)
>>>> +               return -ENOMEM;
>>>> +
>>>> +       evo_mthd(push, 0x0828 + (nv_crtc->index * 0x400), 1);
>>
>>>> +       evo_data(push, usec);
>>>> +       evo_kick(push, mast);
>>>> +
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static int
>>>>  nv50_crtc_set_color_vibrance(struct nouveau_crtc *nv_crtc, bool
>> update)
>>>>  {
>>>>         struct nv50_mast *mast = nv50_mast(nv_crtc->base.dev);
>>
>>>> @@ -1104,14 +1121,14 @@ nv50_crtc_mode_set(struct drm_crtc *crtc,
>> struct drm_display_mode *umode,
>>>>                         evo_mthd(push, 0x0804 + (nv_crtc->index
>> * 0x400), 2);
>>>>                         evo_data(push, 0x00800000 | mode->clock);
>>
>>>>                         evo_data(push, (ilace == 2) ? 2 : 0);
>>>> -                       evo_mthd(push, 0x0810 + (nv_crtc->index
>> * 0x400), 8);
>>>> +                       evo_mthd(push, 0x0810 + (nv_crtc->index
>> * 0x400), 6);
>>>>                         evo_data(push, 0x00000000);
>>>>                         evo_data(push, (vactive << 16) | hactive);
>>
>>>>                         evo_data(push, ( vsynce << 16) | hsynce);
>>
>>>>                         evo_data(push, (vblanke << 16) | hblanke);
>>
>>>>                         evo_data(push, (vblanks << 16) | hblanks);
>>
>>>>                         evo_data(push, (vblan2e << 16) | vblan2s);
>>
>>>> -                       evo_data(push, vblankus);
>>>> +                       evo_mthd(push, 0x082c + (nv_crtc->index
>> * 0x400), 1);
>>>>                         evo_data(push, 0x00000000);
>>>>                         evo_mthd(push, 0x0900 + (nv_crtc->index
>> * 0x400), 2);
>>>>                         evo_data(push, 0x00000311);
>>>> @@ -1141,6 +1158,11 @@ nv50_crtc_mode_set(struct drm_crtc *crtc,
>> struct drm_display_mode *umode,
>>>>         nv_connector = nouveau_crtc_connector_get(nv_crtc);
>>>>         nv50_crtc_set_dither(nv_crtc, false);
>>>>         nv50_crtc_set_scale(nv_crtc, false);
>>>> +
>>>> +       /* G94 only accepts this after setting scale */
>>>> +       if (nv50_vers(mast) < GF110_DISP_CORE_CHANNEL_DMA)
>>>> +               nv50_crtc_set_raster_vblank_dmi(nv_crtc, vblankus);
>>
>>>> +
>>>>         nv50_crtc_set_color_vibrance(nv_crtc, false);
>>>>         nv50_crtc_set_image(nv_crtc, crtc->primary->fb, x,
>> y, false);
>>>>         return 0;
>>>> --
>>>> 2.1.0
>>>>


None of all this&that patches does not work anymore.


poma