no boot on latest kernel drm-radeon-testing/add NI pci ids

[Death Knight]

Fixed via reverting:

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=86f5c9edbb3bac37cc8cee6528a929005ba72aad

Little offtopic but,
But I can't enable 3D with r600g. Using mesa trunk & xf86-video-ati trunk...
Doesn't it support 3D yet?

glxinfo gives segmentation fault and gmesg has this lines:

glxinfo[16659]: segfault at 1e0bbca48 ip 00007f6fddfa2b15 sp 00007fff806e7f60 error 6 in r600_dri.so[7f6fddcda000+dc2000]
glxgears[16672]: segfault at 1e0bbda48 ip 0007f0f3a1d6b15 sp 00007fff6401f120 error 6 in r600_dri.so[7f0f39f0e000+dc2000]

Am I doing something wrong?

[bridgman]

You said the following earlier :

I am not using KMS booting but fglrx. I get into shell via "3" as boot parameter at grub. When I try to load radeon kernel with KMS computer became BSOD... CTRL+ALT+DEL doesn't working yet. But modprobing okey if I made modeset=0.

Does that mean you have fglrx installed as well ?

[Death Knight]

You said the following earlier :
Does that mean you have fglrx installed as well ?

Yes, with patched kernel module code to make it work with 2.6.38.
But when I trying that, fglrx module doesn't loaded in memory. I also tried without fglrx but that not related with BSOD error...

Now I am using my pc with fglrx uninstalled to run Mesa Gallium3D stack but it doesn't working. I have just 2D. /var/log/Xorg.0.log looks OK. No problem appears... But getting segmentation faults when I try to run any 3D app.

[bridgman]

Yes, with patched kernel module code to make it work with 2.6.38.
But when I trying that, fglrx module doesn't loaded in memory. I also tried without fglrx but that not related with BSOD error...

Now I am using my pc with fglrx uninstalled to run Mesa Gallium3D stack but it doesn't working. I have just 2D. /var/log/Xorg.0.log looks OK. No problem appears... But getting segmentation faults when I try to run any 3D app.

Installing fglrx over-writes some 3d-related files and I'm not sure if just building and installing mesa is enough to get them all back. Anyone ?

[Death Knight]

Fixed.
my older config for mesa is :
./autogen.sh TALLOC_LIBS=-ltalloc TALLOC_CFLAGS=-I/usr/include/ --with-dri-drivers=r600,swrast --enable-gallium-{r600,llvm} --prefix=/usr --libdir=/usr/lib64 --enable-xcb --enable-gl-osmesa --enable-egl --enable-openvg --with-state-trackers=egl,glx,dri,xorg,vega --enable-gles1 --enable-gles2

after make make distclean just used this config for mesa.
./autogen.sh TALLOC_LIBS=-ltalloc TALLOC_CFLAGS=-I/usr/include/ --with-dri-drivers=swrast --enable-gallium-r600 --prefix=/usr --libdir=/usr/lib64

Don't know exactly but I am suspecting about dri (not gallium-dri) driver might generate problem since it's not compatible with NI.

[Death Knight]

Now re-test my first config. It's working too. So "make distclean" makes the magic.

Problems:
PC now consume extra ~90W!! than fglrx driver.
I think it doesn't use GPU power save properties properly.

[bridgman]

Problems:
PC now consume extra ~90W!! than fglrx driver.
I think it doesn't use GPU power save properties properly.

Are you using profiles (see RadeonFeature), and if so which profile have you selected ?

[darkbasic]

PC now consume extra ~90W!! than fglrx driver.
I think it doesn't use GPU power save properties properly.

echo profile > /sys/class/drm/card-0/device/power_method
echo low > /sys/class/drm/card1/device/power_profile

[charlie]

Now re-test my first config. It's working too. So "make distclean" makes the magic.

I have found that make clean and make distclean does not totally restore the xorg git to original state. So issues may occur when compiling for different directories or architectures. This restores git to original state:

git reset --hard ; git clean -dxf

[Death Knight]

Are you using profiles (see RadeonFeature), and if so which profile have you selected ?

I wasn't aware from that property.
It's using "default" profile, that consumes extra 90W.
Now using "low" profile that only consumes extra 9W than fglrx, which I believe will be fade to zero over time.
Don't understand why "default" state don't lowers CPU speed.
Thanks.

echo profile > /sys/class/drm/card-0/device/power_method
echo low > /sys/class/drm/card1/device/power_profile

Thanks, you saved me to research that