Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[david.hagood]

I'm trying to build the TI supplied Linux kernel driver modules for the
PowerVR accelerator chip in the OMAP3. I have a GIT tree pulled from the
linux-omap-2.6 tree, and a set of cross-compiler tools built and
installed.

I can build the kernel itself, but when I try to build the TI supplied code:
1) It references "asm/semaphore.h" rather than "linux/semaphore.h" - as
asm/semaphore no longer exists in the code the build fails. I can work
around that by creating a symlink to linux/semaphore.h.
2) It then references "asm/resources.h", which is ALSO gone. I'm guessing
it needs "asm-generic/resources.h".
3) However, at that point the compile REALLY dies, with the code looking
for "asm/arch/resources.h". If I link "arch/arm/include/asm" to
"asm/arch", then I get a large number of errors:

  CC [M] 
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysconfig.o
  CC [M] 
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysutils.o
In file included from
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysutils.c:28:
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysutils_linux.c:154:
error: variable 'cnstr_id_vdd2' has initializer but incomplete type
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysutils_linux.c:155:
error: unknown field 'type' specified in initializer
/space/src/OMAP35x_Graphics_SDC_3_00_00_06/GFX_Linux_KM/services4/srvkm/env/linux/kbuild/../../../../../services4/system/omap3430/sysutils_linux.c:155:
error: 'RES_OPP_CO' undeclared here (not in a function)

So does anybody have an good advice on this?

Re: Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[Koen Kooi]

[-- Attachment #1: Type: text/plain, Size: 470 bytes --]


Op 2 mrt 2009, om 21:56 heeft david.hagood@gmail.com het volgende  
geschreven:

> I'm trying to build the TI supplied Linux kernel driver modules for  
> the
> PowerVR accelerator chip in the OMAP3. I have a GIT tree pulled from  
> the
> linux-omap-2.6 tree, and a set of cross-compiler tools built and
> installed.

In OpenEmbedded we have patches to a) get it to compile b) pick up  
symbols so they actually insert and c) build against Tomi's DSS2

regards,

Koen

[-- Attachment #2: Dit deel van het bericht is digitaal ondertekend --]
[-- Type: application/pgp-signature, Size: 186 bytes --]

Re: Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[David Hagood]

On Tue, 2009-03-03 at 08:32 +0100, Koen Kooi wrote:
> In OpenEmbedded we have patches to a) get it to compile b) pick up  
> symbols so they actually insert and c) build against Tomi's DSS2
> 
> regards,

For various reasons I am not using OpenEmbedded, and while I had looked
at the OpenEmbedded pages I didn't immediately see how to apply what OE
had done to my setup.

It would be nice to have an explanation of what is needed that was a
little less "apply this opaque magic" and more "these are the steps
needed, in our environment this automates those steps."

I would also assert that it would be better in the long term if the
upstream maintainers (in this case TI) would correct the problems
themselves rather than requiring somebody like OE to work around them.

Re: Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[Koen Kooi]

[-- Attachment #1: Type: text/plain, Size: 349 bytes --]


Op 3 mrt 2009, om 13:27 heeft David Hagood het volgende geschreven:
> I would also assert that it would be better in the long term if the
> upstream maintainers (in this case TI) would correct the problems
> themselves rather than requiring somebody like OE to work around them.

Well, TI is using OE to build their new PSPs, your point is moot :)

[-- Attachment #2: Dit deel van het bericht is digitaal ondertekend --]
[-- Type: application/pgp-signature, Size: 186 bytes --]

Re: Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[David Hagood]

On Tue, 2009-03-03 at 13:30 +0100, Koen Kooi wrote:

OK, I apologize in advance, but that statement has pricked the festering
boil of my feelings about this matter, and I feel that I must address
it.

> Well, TI is using OE to build their new PSPs, your point is moot :)

Not really - that assumes that OE becomes the new standard for all
builds.

The reason I don't use OE is that I could never get to where I could
regenerate anything non-trivial (like Angstrom) due to errors in Bitbake
and the various OE black magic^W^W"recipes". Attempting to work through
the problems with the OE devs touched upon just about every pet peeve I
have with Free Software projects:

Me: "I cannot get this to build, here's what I am doing, here's the
script I use to set the environment variables needed for OE, here's the
errors I am getting."
Them: "I don't like the fact that you are using this temporary
environment variable to set these other variables."
Me: "OOOKAAAY, but that variable is a temporary and has nothing to do
with anything. It gets unset at the end of the script. As you can see,
the variables OE needs are set correctly."
Them: "I am going to fixate upon your use of this environment variable
and not address the problem you are having, for I don't really
understand how to fix your problem."
Me: "GRRR. OK. Fine. I set all the variables BY HAND. Here is a
transcript of my exact session. Here is THE SAME DAMN PROBLEM
happening."
Them #1: "I still don't like that you were using that temporary
variable, and will ignore the fact that you aren't using it now."
Them #2: "Oh, you are using the released version of Bitbake. Silly boy,
that version is horribly busted. Go get the newest version from
Subversion, because we feel that actually releasing fixed code is
unnecessary if you have public access to your source code control
system."
Me: "OK, I removed the released version my distro vendor has, and
installed the version from Subversion. Now it doesn't even run at all."
Them: <crickets chirp.wav>

I am sorry, but I have tried, step by step, to follow the instructions
on the OE site to reproduce Angstrom. They don't work for me. They are
incomplete and incorrect, with many of the examples being demonstrably
WRONG.

I have tried to follow the instructions on the Angstrom site (which,
BTW, are very different than the OE instructions). They, too, are
incomplete and demonstrably wrong in many cases.

Sorry, but I am NOT going to base MY project on something that I cannot
get working after several WEEKS of effort to get working.

Yes, I know "it works for me" for certain values of "me", but I cannot
get it working under bog-standard Debian, so that doesn't help me.

Don't get me wrong: I *love* Free Software. I've had far worse
experiences with commercial software like pSOS, VxWorks, and <shudder>
Windows than this. But I do feel the "it's fixed in CVS, so it's FIXED:
full stop" attitude is WRONG - if it isn't released it isn't fixed yet.
I do feel that the various Rainman-like folks who will completely ignore
the real problem to pick at an unrelated piece of lint (and thus lower
the signal to noise ratio of the channel below usability) aren't
helping. I feel the "this is broken in upstream, so we'll break out the
duct tape here rather than trying to fix the upstream" mindset is
antithetical to the idea of "Let's all SHARE the improvements".

I am just trying to help improve things so that Free Software becomes
better every day. The idea that I have to use a specific "vendor" for my
project - be that vendor Microsoft or be that vendor OpenEmbedded - just
because obvious issues aren't being fixed in a truly portable fashion is
ABSURD.

Re: Anyone built the kernel drivers for the PowerVR under linux-omap-2.6?

[Felipe Contreras]

On Tue, Mar 3, 2009 at 3:07 PM, David Hagood <david.hagood@gmail.com> wrote:
> On Tue, 2009-03-03 at 13:30 +0100, Koen Kooi wrote:
>
> OK, I apologize in advance, but that statement has pricked the festering
> boil of my feelings about this matter, and I feel that I must address
> it.

<amusing OE failure story/>

> Sorry, but I am NOT going to base MY project on something that I cannot
> get working after several WEEKS of effort to get working.

Me neither, and I *did* try hard, and I even fixed some issues in
bitbake AFAICR, but they weren't even interested.

IMHO if something works on OE and it doesn't work on upstream that
just means somebody hasn't done their job thoroughly; things should go
into upstream.

I had that issue with ALSA, so I took the matter in my own hands and
pushed a patch into linux-omap, even if I had no idea of what I was
doing at least it was a start for the ones that did know.

Cheers.

-- 
Felipe Contreras