[elrepo] how to get the amd driver working on RHEL 7 ?

[Manuel Wolfshant]

On 04/04/2016 01:56 PM, jacques klein wrote:
> Well, first thing I have to say: I did a "full installation", that is, 
> during installation I selected "Developper ....." at left of screen 
> and added quite everything from right.
>
> Second, before installing the elrepo's fglrx rpms, X11 works fine, and 
> also after I do a rsync from my original and untouched CentOS 7.
In this case... I do not know. If you can pinpoint a bug in the fglrx 
packages, I will be glad to fix them ( if the bug pertains to my 
packaging, that is ). My real problem here is that I have no hardware 
where to test the packages on EL7, everything I have is based on EL6 ( 
and I currently use the EL6 packages on dozens of systems, including the 
very one I am typing from ). Fortunately over time various people helped 
me in correcting the packages built for EL7 by taking the time to test 
locally on their systems and providing feedback. The packages are still 
in elrepo-testing exactly because I am not yet satisfied with the amount 
of QA done to these packages.


>
> The Xpointer is "invisible" only with the fglrx rpms installed and 
> used, and as what I saw, the only diffs. are
> 1. kernel module radeon replaced by fglrx
that is correct


> 2. fglrx driver now used by X server ( because of the new file 
> /etc/X11/xorg.conf )
that is also correct


> 3. one new rpm installed, named acpid.x86_64 ( what is this for ? ).
Description :
acpid is a daemon that dispatches ACPI events to user-space programs.

Most useful for laptops, helps them react properly to lid closure and 
similar events. According to posts I found while trying to learn what 
the heck the drivers do, the authentication daemon installed by ATI 
seems to benefit from its presence, although originally it does not 
mention any need for it. You can use "rpm -e --nodeps" to remove the 
package using a [large] hammer.

>
> So, what can I do ?, 
I am afraid that I cannot answer. So far you are the first with a 
functional C7 clean , full ( as in "not upgraded from minimal") 
installation with this problem. In the tests I did with a borrowed 
workstation ( meanwhile reinstalled with C6 ), the switch from the stock 
video driver to the proprietary one went fine ( modulo some issues - 
such as conflicts between stock glx.so and AMD glx.so - identified and 
fixed over time ).
Late idea: ask google for "invisible cursor", "cursor not displayed" and 
the such. I remember there was a bug in a far far past which required 
something like "HWAccel off" ( or similar, I do not remember exactly )


> I posted at "Apr 1" (not a joke) a message on one of AMD's Support 
> Forums;  "Not Answered" up to now .....
>
> I also wonder why the AMD's built rpms are just "not installable", 
They are installable but they are a mess. And the same rpm is used for 
both EL6 and EL7.

> and what you do at elrepo to make "not complaining" rpms.
Since there is no hidden magic, you can look at our specs at 
https://github.com/elrepo/packages/tree/master/fglrx-kmod and 
https://github.com/elrepo/packages/tree/master/fglrx-x11-drv . 
Everything I do is public.


> Do you patch the amd .so files to get rid of the "Failed dependencies" 
> ? or is there more involved ?.
Not at all. Basically we just repackage the files shipped by AMD in a 
sane manner and provide the kernel module in such a way as to take 
advantage of the stable ABI provided by RHEL kernels, thus avoiding the 
need to recompile the module at each kernel upgrade.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.elrepo.org/pipermail/elrepo/attachments/20160404/6283ddb8/attachment-0001.html>