[PATCH] drm/msm: fix race in Adreno header generation

[PATCH] drm/msm: fix race in Adreno header generation

[Alex Robinson]

Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
before adreno_common.xml.h is generated in trees that generate Adreno
headers at build time. Make msm-y depend on the generated headers,
removing the race.

Signed-off-by: Alex Robinson <alex@ironrobin.net>
---
 drivers/gpu/drm/msm/Makefile | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
index 0c0dfb25f01b..1a918d44ac48 100644
--- a/drivers/gpu/drm/msm/Makefile
+++ b/drivers/gpu/drm/msm/Makefile
@@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
 	generated/sfpb.xml.h
 
 $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
+$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
 $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
 
 targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
-- 
2.50.1

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Rob Clark]

On Fri, Aug 22, 2025 at 7:09 PM Alex Robinson <alex@ironrobin.net> wrote:
>
> Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
> before adreno_common.xml.h is generated in trees that generate Adreno
> headers at build time. Make msm-y depend on the generated headers,
> removing the race.
>
> Signed-off-by: Alex Robinson <alex@ironrobin.net>

looks like the issue is introduced by:
https://patchwork.freedesktop.org/patch/664829/?series=151864&rev=1

BR,
-R

> ---
>  drivers/gpu/drm/msm/Makefile | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
> index 0c0dfb25f01b..1a918d44ac48 100644
> --- a/drivers/gpu/drm/msm/Makefile
> +++ b/drivers/gpu/drm/msm/Makefile
> @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
>         generated/sfpb.xml.h
>
>  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
>  $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
>
>  targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
> --
> 2.50.1
>
>

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Dmitry Baryshkov]

On Sat, Aug 23, 2025 at 02:09:39AM +0000, Alex Robinson wrote:
> Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
> before adreno_common.xml.h is generated in trees that generate Adreno
> headers at build time. Make msm-y depend on the generated headers,
> removing the race.
> 
> Signed-off-by: Alex Robinson <alex@ironrobin.net>
> ---
>  drivers/gpu/drm/msm/Makefile | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
> index 0c0dfb25f01b..1a918d44ac48 100644
> --- a/drivers/gpu/drm/msm/Makefile
> +++ b/drivers/gpu/drm/msm/Makefile
> @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
>  	generated/sfpb.xml.h
>  
>  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))

I'd say, no. The idea was that msm-y should not depend on
ADRENO_HEADERS. If there is any dependency, please move it to adreno-y.

>  $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
>  
>  targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
> -- 
> 2.50.1
> 
> 

-- 
With best wishes
Dmitry

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Rob Clark]

On Sat, Aug 23, 2025 at 6:49 AM Dmitry Baryshkov
<dmitry.baryshkov@oss.qualcomm.com> wrote:
>
> On Sat, Aug 23, 2025 at 02:09:39AM +0000, Alex Robinson wrote:
> > Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
> > before adreno_common.xml.h is generated in trees that generate Adreno
> > headers at build time. Make msm-y depend on the generated headers,
> > removing the race.
> >
> > Signed-off-by: Alex Robinson <alex@ironrobin.net>
> > ---
> >  drivers/gpu/drm/msm/Makefile | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
> > index 0c0dfb25f01b..1a918d44ac48 100644
> > --- a/drivers/gpu/drm/msm/Makefile
> > +++ b/drivers/gpu/drm/msm/Makefile
> > @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
> >       generated/sfpb.xml.h
> >
> >  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> > +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
>
> I'd say, no. The idea was that msm-y should not depend on
> ADRENO_HEADERS. If there is any dependency, please move it to adreno-y.

probably we should s/adreno/gpu/ and move all the msm_gpu*.* there..

In the mean time, I think we were going to drop this patch from the IFPC series

BR,
-R

> >  $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
> >
> >  targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
> > --
> > 2.50.1
> >
> >
>
> --
> With best wishes
> Dmitry

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Akhil P Oommen]

On 8/23/2025 7:46 PM, Rob Clark wrote:
> On Sat, Aug 23, 2025 at 6:49 AM Dmitry Baryshkov
> <dmitry.baryshkov@oss.qualcomm.com> wrote:
>>
>> On Sat, Aug 23, 2025 at 02:09:39AM +0000, Alex Robinson wrote:
>>> Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
>>> before adreno_common.xml.h is generated in trees that generate Adreno
>>> headers at build time. Make msm-y depend on the generated headers,
>>> removing the race.
>>>
>>> Signed-off-by: Alex Robinson <alex@ironrobin.net>
>>> ---
>>>  drivers/gpu/drm/msm/Makefile | 1 +
>>>  1 file changed, 1 insertion(+)
>>>
>>> diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
>>> index 0c0dfb25f01b..1a918d44ac48 100644
>>> --- a/drivers/gpu/drm/msm/Makefile
>>> +++ b/drivers/gpu/drm/msm/Makefile
>>> @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
>>>       generated/sfpb.xml.h
>>>
>>>  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
>>> +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
>>
>> I'd say, no. The idea was that msm-y should not depend on
>> ADRENO_HEADERS. If there is any dependency, please move it to adreno-y.
> 
> probably we should s/adreno/gpu/ and move all the msm_gpu*.* there..
> 
> In the mean time, I think we were going to drop this patch from the IFPC series

Yeah. I will drop that patch.

Btw, was my usage of adreno_gpu symbol in msm_gpu* files incorrect?

-Akhil

> 
> BR,
> -R
> 
>>>  $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
>>>
>>>  targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
>>> --
>>> 2.50.1
>>>
>>>
>>
>> --
>> With best wishes
>> Dmitry

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Rob Clark]

On Mon, Aug 25, 2025 at 9:23 AM Akhil P Oommen <akhilpo@oss.qualcomm.com> wrote:
>
> On 8/23/2025 7:46 PM, Rob Clark wrote:
> > On Sat, Aug 23, 2025 at 6:49 AM Dmitry Baryshkov
> > <dmitry.baryshkov@oss.qualcomm.com> wrote:
> >>
> >> On Sat, Aug 23, 2025 at 02:09:39AM +0000, Alex Robinson wrote:
> >>> Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
> >>> before adreno_common.xml.h is generated in trees that generate Adreno
> >>> headers at build time. Make msm-y depend on the generated headers,
> >>> removing the race.
> >>>
> >>> Signed-off-by: Alex Robinson <alex@ironrobin.net>
> >>> ---
> >>>  drivers/gpu/drm/msm/Makefile | 1 +
> >>>  1 file changed, 1 insertion(+)
> >>>
> >>> diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
> >>> index 0c0dfb25f01b..1a918d44ac48 100644
> >>> --- a/drivers/gpu/drm/msm/Makefile
> >>> +++ b/drivers/gpu/drm/msm/Makefile
> >>> @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
> >>>       generated/sfpb.xml.h
> >>>
> >>>  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> >>> +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> >>
> >> I'd say, no. The idea was that msm-y should not depend on
> >> ADRENO_HEADERS. If there is any dependency, please move it to adreno-y.
> >
> > probably we should s/adreno/gpu/ and move all the msm_gpu*.* there..
> >
> > In the mean time, I think we were going to drop this patch from the IFPC series
>
> Yeah. I will drop that patch.
>
> Btw, was my usage of adreno_gpu symbol in msm_gpu* files incorrect?

I suppose _technically_ it is, but the split btwn msm_gpu and adreno
really only made sense for a2xx, and only if we tried to upstream the
old 2d core, which never happened

BR,
-R

> -Akhil
>
> >
> > BR,
> > -R
> >
> >>>  $(addprefix $(obj)/,$(msm-display-y)): $(addprefix $(obj)/,$(DISPLAY_HEADERS))
> >>>
> >>>  targets += $(ADRENO_HEADERS) $(DISPLAY_HEADERS)
> >>> --
> >>> 2.50.1
> >>>
> >>>
> >>
> >> --
> >> With best wishes
> >> Dmitry
>

Re: [PATCH] drm/msm: fix race in Adreno header generation

[Dmitry Baryshkov]

On Mon, Aug 25, 2025 at 10:48:06AM -0700, Rob Clark wrote:
> On Mon, Aug 25, 2025 at 9:23 AM Akhil P Oommen <akhilpo@oss.qualcomm.com> wrote:
> >
> > On 8/23/2025 7:46 PM, Rob Clark wrote:
> > > On Sat, Aug 23, 2025 at 6:49 AM Dmitry Baryshkov
> > > <dmitry.baryshkov@oss.qualcomm.com> wrote:
> > >>
> > >> On Sat, Aug 23, 2025 at 02:09:39AM +0000, Alex Robinson wrote:
> > >>> Builds can compile msm-y objects (e.g. msm_gpu_devfreq.o)
> > >>> before adreno_common.xml.h is generated in trees that generate Adreno
> > >>> headers at build time. Make msm-y depend on the generated headers,
> > >>> removing the race.
> > >>>
> > >>> Signed-off-by: Alex Robinson <alex@ironrobin.net>
> > >>> ---
> > >>>  drivers/gpu/drm/msm/Makefile | 1 +
> > >>>  1 file changed, 1 insertion(+)
> > >>>
> > >>> diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile
> > >>> index 0c0dfb25f01b..1a918d44ac48 100644
> > >>> --- a/drivers/gpu/drm/msm/Makefile
> > >>> +++ b/drivers/gpu/drm/msm/Makefile
> > >>> @@ -221,6 +221,7 @@ DISPLAY_HEADERS = \
> > >>>       generated/sfpb.xml.h
> > >>>
> > >>>  $(addprefix $(obj)/,$(adreno-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> > >>> +$(addprefix $(obj)/,$(msm-y)): $(addprefix $(obj)/,$(ADRENO_HEADERS))
> > >>
> > >> I'd say, no. The idea was that msm-y should not depend on
> > >> ADRENO_HEADERS. If there is any dependency, please move it to adreno-y.
> > >
> > > probably we should s/adreno/gpu/ and move all the msm_gpu*.* there..
> > >
> > > In the mean time, I think we were going to drop this patch from the IFPC series
> >
> > Yeah. I will drop that patch.
> >
> > Btw, was my usage of adreno_gpu symbol in msm_gpu* files incorrect?
> 
> I suppose _technically_ it is, but the split btwn msm_gpu and adreno
> really only made sense for a2xx, and only if we tried to upstream the
> old 2d core, which never happened

Granted that there seem to be nobody interested in OpenVG anymore, I
suggest we forget that Z180 existed. I think it would be wise to
simplify things for the GPU cores (merging more of the msm_gpu_* into
adreno/).

-- 
With best wishes
Dmitry