Re: drm/msm: CONFIG_MSM_OCMEM?

Re: drm/msm: CONFIG_MSM_OCMEM?

[Rob Clark]

On Tue, Feb 11, 2014 at 10:39 AM, Paul Bolle <pebolle@tiscali.nl> wrote:
> Rob,
>
> Commit 55459968176f ("drm/msm: add a330/apq8x74") added preprocessor
> checks for CONFIG_MSM_OCMEM. But I couldn't find a Kconfig symbol
> MSM_OCMEM (nor a preprocessor define for a macro of that name). Why were
> those preprocessor checks added?

At the moment, CONFIG_MSM_OCMEM is only in downstream kernel.. not
100% sure about how we'll handle that upstream when there is more than
just drm/msm using OCMEM (which can be shared between gpu, video
enc/dec, etc).

> Please note that lines 197-201 currently read:
>     /* Set the OCMEM base address for A330, etc */
>     if (a3xx_gpu->ocmem_hdl) {
>             gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
>                     (unsigned int)(a3xx_gpu->ocmem_base >> 14));
>     }
>
> ocmem_hdl and ocmem_base are used uninitialized here, aren't they? So it
> seems these lines should be wrapped with a preprocessor check for
> CONFIG_MSM_OCMEM too.

in this case ocmem_hdl will be NULL (since a3xx_gpu is kzalloc'd).
The #ifdef's could be added here, but not required.

BR,
-R

>
> Paul Bolle
>

Re: drm/msm: CONFIG_MSM_OCMEM?

[Paul Bolle]

On Tue, 2014-02-11 at 12:29 -0500, Rob Clark wrote:
> On Tue, Feb 11, 2014 at 10:39 AM, Paul Bolle <pebolle@tiscali.nl> wrote:
> > Commit 55459968176f ("drm/msm: add a330/apq8x74") added preprocessor
> > checks for CONFIG_MSM_OCMEM. But I couldn't find a Kconfig symbol
> > MSM_OCMEM (nor a preprocessor define for a macro of that name). Why were
> > those preprocessor checks added?
> 
> At the moment, CONFIG_MSM_OCMEM is only in downstream kernel.. not
> 100% sure about how we'll handle that upstream when there is more than
> just drm/msm using OCMEM (which can be shared between gpu, video
> enc/dec, etc).

I see. (Wouldn't that be an upstream kernel, actually?)

> > Please note that lines 197-201 currently read:
> >     /* Set the OCMEM base address for A330, etc */
> >     if (a3xx_gpu->ocmem_hdl) {
> >             gpu_write(gpu, REG_A3XX_RB_GMEM_BASE_ADDR,
> >                     (unsigned int)(a3xx_gpu->ocmem_base >> 14));
> >     }
> >
> > ocmem_hdl and ocmem_base are used uninitialized here, aren't they? So it
> > seems these lines should be wrapped with a preprocessor check for
> > CONFIG_MSM_OCMEM too.
> 
> in this case ocmem_hdl will be NULL (since a3xx_gpu is kzalloc'd).
> The #ifdef's could be added here, but not required.

I missed that. So whether or not to #ifdef this would be mostly a matter
of style.

Thanks,


Paul Bolle