[ATI] X11 broken: display sleeps and won't wake

[falsifian]

When my boot process reaches the starting-X11 phase, both my monitors go to sleep and won't wake up.

In my dmesg output I see many copies of this message:

[  373.284719] [drm:cypress_dpm_set_power_state] *ERROR* rv770_restrict_performance_levels_before_switch failed

This happens as early as 2995072 and as recently as acefc22. I think it was working fine when I rebuild at 4b78eff, but I'm not 100% confident in my archaeology.

My graphics card is this one: http://www.newegg.com/Product/Product.aspx?Item=N82E16814129261 . From lspci:

01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Cedar [Radeon HD 5000/6000/7350/8350 Series]

Unfortunately I might not have time to investigate this for a while, but if anyone has ideas I can try them.

[falsifian]

More notes:

$ uname -a
Linux angel-nixos 3.17.4 #1-NixOS SMP Thu Jan 1 00:00:01 UTC 1970 x86_64 GNU/Linux

I tried adding the kernel parameter radeon.dpm=0 as suggested here: https://bbs.archlinux.org/viewtopic.php?id=178307 . It does not fix the problem, although I don't see the rv770_restrict... messages in dmsg any more. I haven't tried nomodeset yet.

https://www.libreoffice.org/bugzilla/show_bug.cgi?format=multiple&id=69120 mentions the same error message, but not the symptom I have.

I have two monitors connected by DVI. The left one is rotated. xorg.conf:

# Thanks to http://manual.aptosid.com/en/hw-dev-mon-en.htm accessed 20130129.

Section "Monitor"
    Identifier "Monitor0"
    Option "Rotate" "Left"
    Option "PreferredMode" "1680x1050"
EndSection

Section "Monitor"
    Identifier "Monitor1"
    Option "RightOf" "Monitor0"
    Option "PreferredMode" "1920x1200"
EndSection

Section "Device"
    Identifier "TheDevice"
    Option "Monitor-DVI-0" "Monitor0"
    Option "Monitor-DVI-1" "Monitor1"
EndSection

[falsifian]

With both nomodeset and radeon.dpm=0, the monitors don't go to sleep, but I see X try to start several times and am left with text consoles. X.0.log: https://gist.github.com/falsifian/03ed0791529caacb2eac

[falsifian]

configuration.nix: https://gist.github.com/falsifian/b5ca9848e3691fbafadf

(xorg.conf.randr is the xorg.conf I pasted earlier.)

I think I'm using the open source radeon driver, but don't remember off the top of my head how to check that.

[vcunat]

Why don't you define videoDrivers? I thought it was a required thing, at least if one wants to use the ati driver (I guess that's what you wanted).

[vcunat]

BTW, I do test a box with "ati" driver before pushing larger X changes to master. I now again re-tested on ca776cd without any problems. (Except for that #4923 "error". It's HD 3000 series IGP.)

[falsifian]

I just tried two values for videoDrivers:

• If I set it to [ "ati_unfree" ], then I see the log message described in systemd-logind problem? #4923 . I didn't see whether X started
• If I set it to [ "ati" ], I see the problem I described in this bug.

[vcunat]

@falsifian: can you get logs with [ "ati" ], without any kernel options (no nomodeset etc.), and with as newest kernel as possible?

[falsifian]

@vcunat Unfortunately I'm away from that computer until Sunday night and it can't boot unattended. For now, here's one more existing X.0.log log: https://gist.github.com/falsifian/45f66a135d6273011c1b

I think this is with [ "ati" ]. There aren't any kernel parameters (the log file confirms that). It's only for kernel 3.17.4 though. One additional note is that X.0.log.old was generated just 1 second earlier.

[vcunat]

Hmm, that log looks very good to me (I see no critical error). 3.17 is new enough, I think.

But from reading changelog of xserver-1.16.2, I suspect that bump might help. I wanted to get that bugfix into release this week anyway. I'll CC this issue in the commit, so you can find it during weekend.

[falsifian]

On Sunday I tried the staging branch (ea1dd9f) with videoDrivers = [ "ati" ] and saw similar behaviour. (Similar messages in dmesg, and monitors went to sleep.)

[domenkozar]

Seems like we found out the reason in #5266

[falsifian]

Isn't that bug about the unfree driver? My trouble is with the free videoDrivers = [ "ati" ] driver.

[domenkozar]

Yeah. @falsifian could you try using ati_unfree with latest master? Then we can tackle ati driver.

[falsifian]

Sure, I will take a look tonight.

On 12 December 2014 at 10:06, Domen Kožar notifications@github.com wrote:

Yeah. @falsifian https://github.com/falsifian could you try using
ati_unfree with latest master? Then we can tackle ati driver.

—
Reply to this email directly or view it on GitHub
#5101 (comment).

[offlinehacker]

I have totaly different problems. With my switchable graphics, if i enable ati, it won't even get to starting X.

[offlinehacker]

This is the error: Failed to load firmware "radeon/OLAND_pfp.bin"

[domenkozar]

part of non-free firmwares: https://packages.debian.org/sid/firmware-linux-nonfree

[offlinehacker]

Yes, after hardware.enableAllFirmware = true; that error is gone and laptop boots normally with intel GPU, but if i try to switch to ati echo ON > /sys/kernel/debug/vgaswitcheroo/switch nothing happens, any idea?

[offlinehacker]

Actually this is from dmes vga_switcheroo: client 0 refused switch

[offlinehacker]

Ok, looks like i have to stop X before switching. If i do that i get this bug: https://bugzilla.kernel.org/show_bug.cgi?id=55311 screen goes black and this in dmes i915: switched off. If i update to anything newer than stable kernel, firmwares are not found when hardware.enableAllFirmware = true;, any idea?

[jgeerds]

@offlinehacker Did it ever work with NixOS or other distributions?

[falsifian]

ati_unfree doesn't work at master (e1c679f) for me (end of X.0.log: https://gist.github.com/falsifian/e72f24831f4c69966423).

But now I'm trying older system configurations and I can't find one where X works. Could be a hardware problem or could be something else; I will investigate further.

[domenkozar]

Same as #4659.

[domenkozar]

Is this still an issue?

[falsifian]

Just enabled X11 again and everything works. Thanks for poking me :-)

Small configuration difference: I'm using only one monitor now. I'll re-open if I observe the problem with two.