19 Feb 2016 - Ubuntu fglrx Bug

jay
jay
Joined: 25 Jan 07
Posts: 67
Credit: 60,382,120
RAC: 52,413
Topic 198453

Greetings
I was happily crunching... and then...
I was running the Long Term Support version of Ubuntu Mate
( 14.04 (Trusty Tahr))
There was a fglrx driver update for my Radeon/ATI GPU card
and without thinking, installed it.

I get the "no GPU found." message from BOINC
:-(

I back up my /home ;
wipe the disk and go for a newer Ubuntu-Mate. (release: 15.10)
Part of the Config steps for me is to install
fglrx-updates and
fglrx-amdcccle-updates

and reboot.

This time, the whole machine freezes on reboot.
:-(

Looking around, I see:
https://bugs.launchpad.net/ubuntu/+source/fglrx-installer/+bug/1493888

Which describes the symptoms very well.
BUT
It was supposed to be fixed.

Before I waste another two days, I'll ask...
Anyone else at this point and have a successful fix????

I'll try to go into 'launchpad and submit a bug report
with a reference to 1493888.

Thanks in advance, Jay

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513,211,304
RAC: 0

19 Feb 2016 - Ubuntu fglrx Bug

Quote:


Looking around, I see:
https://bugs.launchpad.net/ubuntu/+source/fglrx-installer/+bug/1493888

Which describes the symptoms very well.
BUT
It was supposed to be fixed.

I know this may seem counter-intuitive, but following the latest release can be a little stressful!

These non LTS releases have a very short lifetime and no sooner do you get one bug fixed a new release introduces a new set of bugs. Release 15.10 end of life is July 2016. In your case may i suggest simply reverse to 14.04, and ensure you upgrade boinc to 7.6.22 as the GPU recognition is better behaved.

Ubuntu releases indicates 16.04 LTS is a few weeks away, so that should be the next stable platform to go to.

robl
robl
Joined: 2 Jan 13
Posts: 1,632
Credit: 1,100,847,185
RAC: 685,784

RE: Greetings I was happily

Quote:

Greetings
I was happily crunching... and then...
I was running the Long Term Support version of Ubuntu Mate
( 14.04 (Trusty Tahr))
There was a fglrx driver update for my Radeon/ATI GPU card
and without thinking, installed it.

I get the "no GPU found." message from BOINC
:-(

I back up my /home ;
wipe the disk and go for a newer Ubuntu-Mate. (release: 15.10)
Part of the Config steps for me is to install
fglrx-updates and
fglrx-amdcccle-updates

and reboot.

This time, the whole machine freezes on reboot.
:-(

Looking around, I see:
https://bugs.launchpad.net/ubuntu/+source/fglrx-installer/+bug/1493888

Which describes the symptoms very well.
BUT
It was supposed to be fixed.

Before I waste another two days, I'll ask...
Anyone else at this point and have a successful fix????

I'll try to go into 'launchpad and submit a bug report
with a reference to 1493888.

Thanks in advance, Jay

[EDIT] I just noticed that you have no screen. Can you ssh into the effected box and using "apt-get" install the packages-below?

I am running Ubuntu 14 and had a similar problem. Here is what I installed to get it to recognize the AMD GPU:

1. ATI binary X.Org driver - install it.

2. fglrx-dev - install it (other prerequisites will install) - this seems to be key

3. reboot

robl
robl
Joined: 2 Jan 13
Posts: 1,632
Credit: 1,100,847,185
RAC: 685,784

RE: Ubuntu releases

Quote:

Ubuntu releases indicates 16.04 LTS is a few weeks away, so that should be the next stable platform to go to.

Great to know. I wonder if the "upgrade" option will work. I have never experienced a successful upgrade on any OS until Win10 came along. Ubu had it for jumping to 15 from 14 but it failed miserably for me causing a virgin install.

jay
jay
Joined: 25 Jan 07
Posts: 67
Credit: 60,382,120
RAC: 52,413

partially SOLVED. My

partially SOLVED. My Mistake

After the upgrade, I couldn't get BOINC to see the GPU so I went back to the
previous version.

*THAT* was the version with the bug.
Lesson Learned: Don't after 11PM.

Still need to get BOINC to see the shared object.
Hopefully I can figure that one out.

I did remember to do the sudo aticonfig -initial

So, now the machine boots and has graphics.
Just BOINC can't see the shared library - that makes it 'not' see the GPU.

I'm going for a beer and diner and will revisit with a clear (relatively) mind.

THANKS!!!
Jay

jay
jay
Joined: 25 Jan 07
Posts: 67
Credit: 60,382,120
RAC: 52,413

RE: [EDIT] I just

Quote:

[EDIT] I just noticed that you have no screen. Can you ssh into the effected box and using "apt-get" install the packages-below?

I am running Ubuntu 14 and had a similar problem. Here is what I installed to get it to recognize the AMD GPU:

1. ATI binary X.Org driver - install it.

2. fglrx-dev - install it (other prerequisites will install) - this seems to be key

3. reboot

Hi,
Yes.
I had started with the 14.04 LTS and couldn't get BOINC to see the GPU afterwards.
My Mistake 1.) start over with 15.
My Mistake 2.)When BOINC didn't see the GPU I went to the buggy version of fglrx (problem with GCC in installing)

I had installed the MESA drivers to get 'standard' ati video drivers.
glxgears works - GL is OK.

I had problem with using the plain fglrx driver;
but the "fglrx-updates" and "fglrx-updates"
have worked for me in the past..
Here is what is now installed..

aptitude search fglrx | grep ^i
i   fglrx-amdcccle-updates          - Catalyst Control Center for the AMD graphi
i A fglrx-updates                   - Video driver for the AMD graphics accelera
i A fglrx-updates-core              - Minimal video driver for the AMD graphics 


I will try fglrx-dev tonight!!!

Now I need an openCl benchmark...

THANKS!!!!
Jay

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513,211,304
RAC: 0

RE: RE: Ubuntu releases

Quote:
Quote:

Ubuntu releases indicates 16.04 LTS is a few weeks away, so that should be the next stable platform to go to.

Great to know. I wonder if the "upgrade" option will work. I have never experienced a successful upgrade on any OS until Win10 came along. Ubu had it for jumping to 15 from 14 but it failed miserably for me causing a virgin install.

Been there done that too many times. The 16.04 alpha releases were out a few weeks back, so they seem on time. i guess i'm going to be "systemd". I don't bother with any non LTS releases any more, other than maybe in a VM, and if anyone is reading and thinking about linux / ubuntu for the first time to break free from other places, please start with NN.04 LTS releases where NN is even.

Running an in-place upgrade on any OS is really asking for trouble imho. The way i look at it is, take the opportunity to sweep clean unwanted clutter, look over the synaptic/apt logs to see what you have installed and decide what apts you installed and then decide what you really need. Format the disk of junk and enjoy a fast system and test your backup and restore process, and it really helps to have two computers.

robl
robl
Joined: 2 Jan 13
Posts: 1,632
Credit: 1,100,847,185
RAC: 685,784

RE: Running an in-place

Quote:

Running an in-place upgrade on any OS is really asking for trouble imho. The way i look at it is, take the opportunity to sweep clean unwanted clutter, look over the synaptic/apt logs to see what you have installed and decide what apts you installed and then decide what you really need. Format the disk of junk and enjoy a fast system and test your backup and restore process, and it really helps to have two computers.

I agree. But its one thing to upgrade a laptop/desktop and save your bookmarks and an entirely different scenario to reconfigure, reload various servers, sql databases, etc. Ok. I admit it. I am looking for the easy way out.

robl
robl
Joined: 2 Jan 13
Posts: 1,632
Credit: 1,100,847,185
RAC: 685,784

RE: I will try fglrx-dev

Quote:

I will try fglrx-dev tonight!!!
THANKS!!!!
Jay

No problem. Let us know how it works out. I and others have been down the road before on more than one occassion. It would be so nice to know what the missing piece really is.

AgentB
AgentB
Joined: 17 Mar 12
Posts: 915
Credit: 513,211,304
RAC: 0

RE: No problem. Let us

Quote:
No problem. Let us know how it works out. I and others have been down the road before on more than one occassion. It would be so nice to know what the missing piece really is.

There are at least two pieces to the puzzle, that i know of...

Assuming a Debian based linux distro such as ubuntu.

The first piece is the OpenCL library which boinc tries to dynamically load during its "GPU detection" at start-up.

The current or correct name for this library is libOpenCL.so.1 and fgrlx installs it as /usr/lib/libOpenCL.so.1 and has done for a long time.

boinc looks for a library named libOpenCL.so and so it fails to find the library needed, and so does not detect the card.

This was fixed at boinc 7.6.22 (thanks Juha), where it looks for first libOpenCL.so, and if not found tries libOpenCL.so.1 (and will also report some helpful diagnostics if the right coproc flag is set).

Prior to 7.6.22 - the workaround is create a symbolic link to the library.

cd /usr/lib
sudo ln -s libOpenCL.so.1 libOpenCL.so

or

fgrlx-dev adds the (not needed) development files and also creates the required link.

The second piece is the standard boinc debian install being "as a service". Basically a user "boinc" runs the boinc client and the science applications, and is started automatically during system bootup.

This works fine for CPU only tasks (and NVidia GPUs).

In short it's broken for AMD-GPU and the fix is

login and from a terminal window

sudo /etc/init.d/boinc-client restart

and the GPU should start.

The long story - Personally i like this boinc user autostart approach, but now there are complexities (especially AMD) relating to how X windows system manages the GPU drivers and all those processes and users needing it.

With the standard install and the link above, the GPU recognition now fails at startup due to permissions. (it restarts ok after you login and do the restart above)

boinc in theory should be able to be granted these needed X permissions via "xhost" but I have not been able to get boinc started with an xhost before running an X session. Fixing this is probably something for AMD to do, or beyond my X understanding, however there is a workaround to this - execute the xhost command from the login-greeter (the very first X-window which asks for login)

This varies depends on what display manager you use, i'm running lightdm and the file is /etc/lightdm/lightdm.conf

[SeatDefaults]
display-setup-script=/usr/bin/xhost +SI:localuser:boinc

good luck and bear with it...

Edit: and there is more reading at the boinc GPU forum

jay
jay
Joined: 25 Jan 07
Posts: 67
Credit: 60,382,120
RAC: 52,413

Thank you all for your good

Thank you all for your good wishes and support.

I'm embarrassed to think how many times I've hit this.

I'm trying the "make-a-small-change;reboot;test" approach.

Currently, I'm running with Locutus's ppa.
(( sudo add-apt-repository ppa:costamagnagianfranco/boinc ))
I did a git of the BOINC source and was puzzling over the dlopen() routine.
I admit I'll have to play around with that later.

Still, no gpu found.

My libOpenCL.so was in /usr/lib32/
I made a link...
lrwxrwxrwx 1 0 0 14 Feb 23 09:30 libOpenCL.so -> libOpenCL.so.1
-rw-r--r-- 1 0 0 29156 Oct 28 12:25 libOpenCL.so.1

Still , no GPU found.

I went back through my notes and saw that I set xhost last year.

Hurray! That was it. GPU found

This was:
xhost +si:localuser:boinc.

My current plan is to let all my WU complete;
do a purge on Boinc
remove the ppa
remove the link to libOpenCL.so.1
get the out-of-the box boinc from ubuntu
try the xhost

and see if that one change was all it took.

May get pizza & beer to celebrate. Wish you were here....

I'll do another post later and share results of further tests...

Jay

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.