The AMD/ATI Bar & Grill

Lockheed · 2013-11-18 11:22:27

Well, I am stuck with my system upgrades because:

$ yaourt -Syua
[sudo] password for juha: 
:: Synchronising package databases...
 xorg112 is up to date
 core is up to date
 extra is up to date
 community is up to date
 multilib is up to date
 archlinuxfr is up to date
error: failed to prepare transaction (could not satisfy dependencies)
:: Starting full system upgrade...
:: catalyst-total-hd234k: requires linux<3.12

Any ways around that? Is the update of the catalyst-total-hd234k expected?

Last edited by Lockheed (2013-11-18 11:23:03)

Vi0L0 · 2013-11-18 12:10:56

https://aur.archlinux.org/packages/cata … al-hd234k/ is saying linux<3.13
So update will fix this, just don't know why yaourt failed here...
May i propose cower + makepkg instead of yaourt? ;P

Lockheed · 2013-11-18 15:13:42

I wasa trying to reinstall this package, but there is a problem with checksum:

==> Validating source files with md5sums...
    amd-driver-installer-catalyst-13.1-legacy-linux-x86.x86_64.zip ... FAILED
    catalyst_build_module ... Passed
    lib32-catalyst.sh ... Passed
    catalyst.sh ... Passed
    atieventsd.sh ... Passed
    atieventsd.service ... Passed
    catalyst.conf ... Passed
    arch-fglrx-authatieventsd.patch ... Passed
    hook-fglrx ... Passed
    ati_make.sh ... Passed
    makefile_compat.patch ... Passed
    catalyst-hook.service ... Passed
    3.5-do_mmap.patch ... Passed
    arch-fglrx-3.7.patch ... Passed
    arch-fglrx-3.8.patch ... Passed
    gentoo_linux-3.10-proc.diff ... Passed
    foutrelis_3.10_fix_for_legacy.patch ... Passed
    lano1106_fglrx_intel_iommu.patch ... Passed
    lano1106_kcl_agp_13_4.patch ... Passed
==> ERROR: One or more files did not pass the validity check!
==> ERROR: Makepkg was unable to build catalyst-total-hd234k.

Vi0L0 · 2013-11-18 20:20:02

yeah, you would find workaround in the comments and PKGBUILD, but you no longer have to do this because thanks to zoopp and his referer we no longer have to download installer manually ... just retry

lano1106 · 2013-11-18 21:21:02

In my quest to use xvba, I'm pretty sure that recent catalyst release just broke the thing. libxvba is communicating with catalyst Xorg video driver by using libxext. Here is the output of my attempt:

xvba_video: FGLRX driver version 13.25.5 detected
xvba_video: FGLRX device ID 0x6798
xvba_video: Evergreen GPU detected
xvba_video: XvBA version 0.77 detected
...
xvba_video: XVBACreateDecode(): context 0x7f91000402a0, size 1920x1088, capability_id XVBA_H264, flags XVBA_H264_HIGH, surface_type NV12
X Error of failed request: BadRequest (invalid request code or no such operation)
Major opcode of failed request: 158 ()
Minor opcode of failed request: 1
Serial number of failed request: 86
Current serial number in output stream: 86

Update:

I have created a bug. You are welcome to follow it to demonstrate to AMD your interest in a fix into that problem

http://ati.cchtml.com/show_bug.cgi?id=950

Last edited by lano1106 (2013-11-18 21:39:24)

ivdok · 2013-11-19 15:40:41

lano1106 wrote:

In my quest to use xvba, I'm pretty sure that recent catalyst release just broke the thing. libxvba is communicating with catalyst Xorg video driver by using libxext. Here is the output of my attempt:
xvba_video: FGLRX driver version 13.25.5 detected
xvba_video: FGLRX device ID 0x6798
xvba_video: Evergreen GPU detected
xvba_video: XvBA version 0.77 detected
...
xvba_video: XVBACreateDecode(): context 0x7f91000402a0, size 1920x1088, capability_id XVBA_H264, flags XVBA_H264_HIGH, surface_type NV12
X Error of failed request: BadRequest (invalid request code or no such operation)
Major opcode of failed request: 158 ()
Minor opcode of failed request: 1
Serial number of failed request: 86
Current serial number in output stream: 86
Update:
I have created a bug. You are welcome to follow it to demonstrate to AMD your interest in a fix into that problem
http://ati.cchtml.com/show_bug.cgi?id=950

XvBA doesn't work for me with 13.7+ drivers at all - it just crashes Xorg, and SLiM pops out, asking to log in. But can you try VAAPI thing? Does it work? Because 0.7 xvba is obsolete, and 0.8 has something trimmed. (@Vi0L0, do you know, what is it?) Besides, both are working for me only under 13.4, which, I believe, sucks for linux-gaming.

middy · 2013-11-19 21:50:52

I just picked up a laptop with a AMD A8-4500M APU that uses a radeon HD 7640G. I'm just curious whether I should use the open source radeon or fgrlx drivers with it.

Right now I'm using the open source radeon drivers but my apu runs very hot with them (much more hot compared to when it had windows 8 installed). I don't think its downclocking its self with them.

Also, what are the radeonsi drivers?

Last edited by middy (2013-11-19 21:55:32)

AuHop · 2013-11-20 02:17:38

@middy
The "radeonsi drivers" refer to the FOSS Radeon stack which is specific to cards of at least the 7xxx(m) series (that aren't rebranded 6xxx(m) series cards). Generally if it was made on a 28nm node, it will be radeonsi when it comes to FOSS.

lano1106 · 2013-11-20 04:00:58

ivdok wrote:

XvBA doesn't work for me with 13.7+ drivers at all - it just crashes Xorg, and SLiM pops out, asking to log in. But can you try VAAPI thing? Does it work? Because 0.7 xvba is obsolete, and 0.8 has something trimmed. (@Vi0L0, do you know, what is it?) Besides, both are working for me only under 13.4, which, I believe, sucks for linux-gaming.

ivdok,

I am not using XvBA directly. I am using it through VAAPI. The way it works is that libva provides 2 API.

1 for the application: vlc, ffmpeg, mplayer, etc
1 for the device driver.

the libva-xvba-driver only serve as an adaptor to be able to plug XvBA API into VAAPI application.

I have read that the closed version 0.7 has more functionnality than the 0.8 open source version. I have tried to figure out what exactly but based on vainfo output both driver version offer the same capability so I prefer the open source one which make it possible to apply potential patches.

VAAPI drivers make possible to decode video and render frames in a single step but most players to do not use this feature (VLC for sure) possibly to make it possible the insertion of video filters between the decoder and the video output module.

At some point VLC had a VAAPI decoder module and a VAAPI video output module and it was using the 1 step decode/render method by passing a VAAPI surface IDs from the decoder module to the video out module. This turns out to be catastrophic because most VLC modules are active with their own thread and VAAPI is not thread safe :-(

Also note that with the vdpau-va-gl driver, it is not very useful with VLC and possible with mplayer as well. The reason why is that VLC has 7-8 different modules of the same type each been given a score. The player will try out all the module from the one having the highest score and stepping down until it find one that will work.

The order looks like that:

glx
Xvideo
Xshm extension
XCB
Plain X11

So VLC does not need a vdpau driver wrapping rendering with gl to use that method. vdpau-va-gl driver is only useful in the context of the flash plugin that either use vdpau for rendering or use very simple way to rendre frames.

That being said, the OSS radeon UVD support appears to be much superior than what is offered by catalyst.

For one,

it is working and it support MPEG4 decoding.

I am kinda stuck with Catalyst for openGL perf and OpenCL but I have a feeling that at some point the OSS radeon driver will match or surpass catalyst.

Anyway this was a long post, I am currently intrumenting xorg-server to trace what it receive from a XvBA app. I'll report back when I have new info.

thanks,

lano1106 · 2013-11-20 05:48:13

My Xorg-server instrumentation has paid off!

AMDXVBA and AMDXVOPL Xextensions are registered twice to the server by the catalyst driver.

Visit the bug page for the nittty gritty technical details:

http://ati.cchtml.com/show_bug.cgi?id=950

dejavu · 2013-11-20 13:57:51

Hey, found a simple workaround for the problem with Catalyst legacy and Gnome Shell 3.10:

/org/gnome/settings-daemon/plugins/cursor/active has to be set to false in dconf.
That has to be done for every user and also for the gdm user. Just open a terminal and open the dconf-editor with following commands:
For any user:

dconf-editor

For gdm user:

xhost +

sudo -u gdm dbus-launch dconf-editor

Thanks to the information of https://bugs.archlinux.org/task/34901 I have found that workaround.

pirateserb · 2013-11-20 16:24:12

I have a problem with amd proprietary drivers. I installed Arch on my laptop by following the wiki. When I tried startx it just give me black screen. At a certain angle I can see 3 windows open, it is just like there is no screen brightness at all. Laptop is K52JT with radeon 6370m gpu.
Here is my Xorg.0.log - http://pastebin.com/FJ1RS7X6

clfarron4 · 2013-11-20 19:37:36

pirateserb wrote:

I have a problem with amd proprietary drivers. I installed Arch on my laptop by following the wiki. When I tried startx it just give me black screen. At a certain angle I can see 3 windows open, it is just like there is no screen brightness at all. Laptop is K52JT with radeon 6370m gpu.
Here is my Xorg.0.log - http://pastebin.com/FJ1RS7X6

Which environment are you running (GNOME, LXDE, something else)?

Unrelated - @Vi0L0: I've noticed with Catalyst 13.11 that I no longer need to use the nomodeset kernel parameter for my card to load an X environment with my AMD Radeon 6480G (though it works perfectly fine with, though it returns the famed drm_radeon_init error with nomodeset). AMD seems to be making progress again for my card.

pirateserb · 2013-11-20 21:30:12

clfarron4 wrote:

Which environment are you running (GNOME, LXDE, something else)?

I tried testing X without installing any DE, later I tried xfce. I can see that xfce has started but the screen brightness can`t be turned on. On my desktop I also have arch installed and AMD gpu with proprietary drivers and everything is fine. This is pissing me off.

clfarron4 · 2013-11-20 22:43:07

pirateserb wrote:

clfarron4 wrote:
Which environment are you running (GNOME, LXDE, something else)?
I tried testing X without installing any DE, later I tried xfce. I can see that xfce has started but the screen brightness can`t be turned on. On my desktop I also have arch installed and AMD gpu with proprietary drivers and everything is fine. This is pissing me off.

That's... wierd. I could understand if the screen was just black, but having the screen working with the backlight OFF without doing anything is just wrong... I take it the function buttons to bump the brightness/turn the backlight on/off (you might not have this second button) are doing nothing to help the situation.

BitPuffin · 2013-11-22 13:48:45

Vi0L0 wrote:

BitPuffin wrote:
Okay so apparently, by adding IDs from the legacy common/lib/modules/fglrx/build_mod/fglrxko_pci_ids.h to the latest non legacy driver the non legacy driver might work on older hardware as well. I haven't tested this yet (as I can currently not reboot because I'll lose my REPL session ), but if this is true it would save Vi0L0 a lot of work. Might be worth investigating, plus we could run on newer X11 versions making our systems more secure. And also perhaps Wayland one day! But let's not get our hopes up just yet
It will let you to compile fglrx module successfully but won't let catalyst to startx, because still it will complain about missing adapter.
If you will also copy amdpcsdb.default and control from legacy's /etc/ati (maybe legacy's aticonfig also, but i don't remember) it will let you to generate xorg.conf.
Then I was trying to copy uki and others legacy's libs, but i'm not sure if it was a good idea - it was some time ago but i never found enough time (and will) to finish/figure it out.
Maybe you will?

I wish I had the time to figure it out. It would be really nice if it worked!

If you (or someone else) decide to try figuring it out and don't know if it is a good idea to have it replace the legacy driver, maybe it is better to put it in a separate package like catalyst-legacy-hack or whatever!
It would be great to be able to move off of old XServer, and also maybe it would fix some issues with the current driver, and surely introduce new issues .

ivdok · 2013-11-22 16:20:58

lano1106 wrote:

ivdok wrote:
XvBA doesn't work for me with 13.7+ drivers at all - it just crashes Xorg, and SLiM pops out, asking to log in. But can you try VAAPI thing? Does it work? Because 0.7 xvba is obsolete, and 0.8 has something trimmed. (@Vi0L0, do you know, what is it?) Besides, both are working for me only under 13.4, which, I believe, sucks for linux-gaming.
ivdok,
I am not using XvBA directly. I am using it through VAAPI. The way it works is that libva provides 2 API.
1 for the application: vlc, ffmpeg, mplayer, etc
1 for the device driver.
the libva-xvba-driver only serve as an adaptor to be able to plug XvBA API into VAAPI application.
I have read that the closed version 0.7 has more functionnality than the 0.8 open source version. I have tried to figure out what exactly but based on vainfo output both driver version offer the same capability so I prefer the open source one which make it possible to apply potential patches.
VAAPI drivers make possible to decode video and render frames in a single step but most players to do not use this feature (VLC for sure) possibly to make it possible the insertion of video filters between the decoder and the video output module.
At some point VLC had a VAAPI decoder module and a VAAPI video output module and it was using the 1 step decode/render method by passing a VAAPI surface IDs from the decoder module to the video out module. This turns out to be catastrophic because most VLC modules are active with their own thread and VAAPI is not thread safe :-(
Also note that with the vdpau-va-gl driver, it is not very useful with VLC and possible with mplayer as well. The reason why is that VLC has 7-8 different modules of the same type each been given a score. The player will try out all the module from the one having the highest score and stepping down until it find one that will work.
The order looks like that:
glx
Xvideo
Xshm extension
XCB
Plain X11
So VLC does not need a vdpau driver wrapping rendering with gl to use that method. vdpau-va-gl driver is only useful in the context of the flash plugin that either use vdpau for rendering or use very simple way to rendre frames.
That being said, the OSS radeon UVD support appears to be much superior than what is offered by catalyst.
For one,
it is working and it support MPEG4 decoding.
I am kinda stuck with Catalyst for openGL perf and OpenCL but I have a feeling that at some point the OSS radeon driver will match or surpass catalyst.
Anyway this was a long post, I am currently intrumenting xorg-server to trace what it receive from a XvBA app. I'll report back when I have new info.
thanks,

Wow, a great source of information. Thanks! I used VLC because XBMC is an overkill for me, but now I don't think it's a good solution anymore.

And a short error trace from Xorg. Nothing visible or sensible, it's just what I find it curious.

clfarron4 wrote:

pirateserb wrote:
I have a problem with amd proprietary drivers. I installed Arch on my laptop by following the wiki. When I tried startx it just give me black screen. At a certain angle I can see 3 windows open, it is just like there is no screen brightness at all. Laptop is K52JT with radeon 6370m gpu.
Here is my Xorg.0.log - http://pastebin.com/FJ1RS7X6
Which environment are you running (GNOME, LXDE, something else)?
Unrelated - @Vi0L0: I've noticed with Catalyst 13.11 that I no longer need to use the nomodeset kernel parameter for my card to load an X environment with my AMD Radeon 6480G (though it works perfectly fine with, though it returns the famed drm_radeon_init error with nomodeset). AMD seems to be making progress again for my card.

Didn't care for this output, but thanks - I will try it when the time will come.
UPD: If I remember correctly, nomodeset was mandatory for plymouth, but it wasn't for a typical installation (vanilla framebuffer), because we blacklisted radeon earlier. So... Maybe it was introduced earlier than 13.11? Please, correct me, if I'm wrong.

clfarron4 wrote:

pirateserb wrote:
clfarron4 wrote:
Which environment are you running (GNOME, LXDE, something else)?
I tried testing X without installing any DE, later I tried xfce. I can see that xfce has started but the screen brightness can`t be turned on. On my desktop I also have arch installed and AMD gpu with proprietary drivers and everything is fine. This is pissing me off.
That's... wierd. I could understand if the screen was just black, but having the screen working with the backlight OFF without doing anything is just wrong... I take it the function buttons to bump the brightness/turn the backlight on/off (you might not have this second button) are doing nothing to help the situation.

Probably ACPI error. I heard what new laptops/ultrabooks for Win8+ can have we-e-e-eird ACPI table. For example, one Acer/Lenovo (can't remember which one) model can only brighten display up to 50% off the normal value. Tech support didn't provide Linux support either.
UPD: I remembered - it was one of the Acer Aspire One '12 models. I believe it was ultrabook.

Last edited by ivdok (2013-11-22 20:33:50)

ringtail · 2013-11-23 10:31:30

On my old system: mobo: EPOX-8rda+; AMD-Barton 2600+; 1,75 gb DDR1; Saphirre Radeon 4650HD; with arch , lts-kernel,
I tried xorg 11/12 with catalyst-hd234k (catalyst-hook +catalyst-utils), catalyst-hd234k-12.4, with nomodeset in grub-bios, with loading fglrx at boot, blacklisting radeon; fglrx module is created, loaded (modprobe fglrx return nothing) but startx gives me : Illegal instruction at adress 03f.....xserver:connection refused ; argument list too long, closing log file. And /var/Xorg/...log file shows no EE but Backtrace with point 4. ?? (nill).

[   318.471] 3: /usr/lib/dri/fglrx_dri.so (0xb1ab9000+0x1c3c0d7) [0xb36f50d7]
[   318.471] 4: ?? [(nil)]
[   318.471] Illegal instruction at address 0xb36f50d7

and Xserver aborted.
BUT
removing /usr/lib/dri/fglrx_dri.so permits Xserver start with "revert to software rendering" and another "illegal instruction " when OpenGL tries to "dlopen /usr/lib/xorg/modules/dri/fglrx_dri.so".
THE problem seems related to a faulty fglrx_dri.so. How can I do to generate a "good" fglrx_dri.so?

Last edited by ringtail (2013-11-24 08:18:55)

Shinto · 2013-11-23 11:23:56

Can't wait for the new driver (beta) update.

Last edited by Shinto (2013-11-23 11:24:14)

Kuci · 2013-11-23 11:32:45

Vi0L0: We cannot wait for 13.11 Beta v9.4

Vi0L0 · 2013-11-23 20:38:49

ivdok wrote:

XvBA doesn't work for me with 13.7+ drivers at all - it just crashes Xorg, and SLiM pops out, asking to log in. But can you try VAAPI thing? Does it work? Because 0.7 xvba is obsolete, and 0.8 has something trimmed. (@Vi0L0, do you know, what is it?).

clear, open 0.8 was crippled but afaik h264-level51.patch is making it almost (if not completely) as good as closed 0.7.8

Only xbmc is using xvba directly, any other player is using vaapi (for me mplayer was always working better than vlc).

+@lano1106:
this post of fritsch, xbmc's developer, seems to be related: http://phoronix.com/forums/showthread.p … post358371

clfarron4 wrote:

@Vi0L0: I've noticed with Catalyst 13.11 that I no longer need to use the nomodeset kernel parameter for my card to load an X environment with my AMD Radeon 6480G (though it works perfectly fine with, though it returns the famed drm_radeon_init error with nomodeset). AMD seems to be making progress again for my card.

Still it's good to use nomodeset or radeon.modeset=0 to save some time on unnecessary loading of radeon and drm modules.

Shinto wrote:

Can't wait for the new driver (beta) update.

Kuci wrote:

Vi0L0: We cannot wait for 13.11 Beta v9.4

Done, please let me know if it helped for brightness adjustment issues.
Older 13.11 beta 6 repo is available as [catalyst-13.11-beta6]

Last edited by Vi0L0 (2013-11-23 20:42:14)

Kuci · 2013-11-23 21:23:28

Vi0L0 wrote:

Done, please let me know if it helped for brightness adjustment issues.

No, it didn't. I was not even able to start X with my original xorg.conf.

ChemBro · 2013-11-23 21:47:45

Vi0L0 wrote:

Done, please let me know if it helped for brightness adjustment issues.

Nope, still doesn't work. Tested it in Amnesia: A Machine for Pigs and Metro: Last Light.

Edit: Oh, that is something other, not the gamma thing... Um... I need to go to bed.

Last edited by ChemBro (2013-11-23 22:32:37)

Kuci · 2013-11-23 22:26:03

WHOA!

$ aticonfig --acpi-services=on

And brightness adjustment works once again !

ringtail · 2013-11-24 08:22:03

Has anybody an idea for my problem?
Is it OK that fglrx_dri.so opened with vi starts with "^?ELF@@@@" in /usr/lib/dri location and with "ELF@@@@" in /usr/lib/xorg/modules/dri/ ?

Arch Linux

#5876 2013-11-18 11:22:27

Re: The AMD/ATI Bar & Grill

#5877 2013-11-18 12:10:56

Re: The AMD/ATI Bar & Grill

#5878 2013-11-18 15:13:42

Re: The AMD/ATI Bar & Grill

#5879 2013-11-18 20:20:02

Re: The AMD/ATI Bar & Grill

#5880 2013-11-18 21:21:02

Re: The AMD/ATI Bar & Grill

#5881 2013-11-19 15:40:41

Re: The AMD/ATI Bar & Grill

#5882 2013-11-19 21:50:52

Re: The AMD/ATI Bar & Grill

#5883 2013-11-20 02:17:38

Re: The AMD/ATI Bar & Grill

#5884 2013-11-20 04:00:58

Re: The AMD/ATI Bar & Grill

#5885 2013-11-20 05:48:13

Re: The AMD/ATI Bar & Grill

#5886 2013-11-20 13:57:51

Re: The AMD/ATI Bar & Grill

#5887 2013-11-20 16:24:12

Re: The AMD/ATI Bar & Grill

#5888 2013-11-20 19:37:36

Re: The AMD/ATI Bar & Grill

#5889 2013-11-20 21:30:12

Re: The AMD/ATI Bar & Grill

#5890 2013-11-20 22:43:07

Re: The AMD/ATI Bar & Grill

#5891 2013-11-22 13:48:45

Re: The AMD/ATI Bar & Grill

#5892 2013-11-22 16:20:58

Re: The AMD/ATI Bar & Grill

#5893 2013-11-23 10:31:30

Re: The AMD/ATI Bar & Grill

#5894 2013-11-23 11:23:56

Re: The AMD/ATI Bar & Grill

#5895 2013-11-23 11:32:45

Re: The AMD/ATI Bar & Grill

#5896 2013-11-23 20:38:49

Re: The AMD/ATI Bar & Grill

#5897 2013-11-23 21:23:28

Re: The AMD/ATI Bar & Grill

#5898 2013-11-23 21:47:45

Re: The AMD/ATI Bar & Grill

#5899 2013-11-23 22:26:03

Re: The AMD/ATI Bar & Grill

#5900 2013-11-24 08:22:03

Re: The AMD/ATI Bar & Grill

Board footer