System hangs randomly but only when using amdgpu [solved]

[Amity88]

This is a fresh Linux system, the screen randomly fills with a color (blue/white/yellow etc) and is rendered unusable till a restart. I'm not even sure if this is a hang because sometimes when this happens, I can still hear the audio from the youtube video.

It's weird as it never happens in Windows 8.1 but randomly hits me when I use Gentoo or SysRescueCD or SuSe or Mint of FreeBSD. In short it happens on any non-Windows OS.

I don't think it's a video issue because it happens even in pure CLI. The kern log or dmesg file doesn't indicate any error at the time of the hang. Do you guys have any suggestions on what else I could check to fix this?

Here's the output of lspci, this is an ASUS H81M-CS motherboard:

[Jaglover]

Try swapping RAM modules if you have more than one. (Maybe Microsoft dream is fulfilled finally, a computer that runs Windows only.)
_________________
My Gentoo installation notes.
Please learn how to denote units correctly!

[Amity88]

I tried running memtest86 over the past 24 hours and didn't get any errors or blanks screens.

Currently, I suspect that it's the AMG GPU driver (amdgpu R7 250, Southern Islands GCN 1.0) that is causing the issue. For the purpose of debug, I'm gonna try the following:

1. Try using the older radeon driver and see if the issue persists.
2. If that doesn't work, I'll try using the onboard Intel GPU.
_________________

[Amity88]

(changing the subject to better reflect the actual issue)

So, I was able to narrow down the issue to the gprahics driver.

1. The screen blanks out randomly when I used AMDGPU drivers.
2. It's a lot worse when I used Radion.
3. The only thing that worked in the past was the old fglrx driver a year ago. Can't use this anymore though cause they're dropped support

4. The onboard Intel GPU driver is stable. This is what I'm using now.

Not sure how I can fix this. If you guys get the AMD R7 250 (Southern Islands) working without random hangs, please let me know.
_________________

[Zucca]

Do you get anything in dmesg/logs?

Have you tried other kernel versions?

I have

[Amity88]

I didn't find anything in the logs/dmesg when booted into SysRescueCD after an incident.

About the other kernel version. This system used to work fine with the fglrx drivers. Things just got messy after AMD moved over to the amdgpu drivers.

Also, it's good to know that you actually have something close in design. I think mine is GCN 1.1 and your is probably GCN 1.2

I haven't really started using this build so I'm willing to experiment if you have anything you want me to try..
_________________

[NeddySeagoon]

Amity88,

I have a similar issue with an R450.

I've tried different motherboard slots, the old and new amdgpu drivers, turning off Message Signalled IRQs. (its a command line option)
Memtest finds nothing and there is nothing in kernel logs.

The incident halts the CPU, as it won't even respond to the reset button, which is probably why there is nothing in the logs.
After a power cycle, the system often restarts with one core missing.

A restart can take a couple of hours too. Its left my raid5 'dirty' a few times so it does a resync.

So far, I've only tried the two amdgpu drivers but a few other non accelerated drives should work.
That may help determine if its a hardware or a software problem.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[Amity88]

Hey there Neddy

I've ruled out any hardware issues because I dual boot with Windows 8.1 and it runs pretty stable.

The symptoms on Linux are very similar to what you experience. Doesn't respond to reset key combinations, the actual reset button doesn't work at times. Restarts don't take much time though.

The non-accelerated drivers would do software rendering right? As crappy as it is, I figured that the Intel GPU is better than software rendering. Maybe I should try pulling in fglrx or amdgpu-pro
_________________

[NeddySeagoon]

Amity88,

Yes - there would be no acceleration at all. I had in mind vesa or fbdev.
The GPU does nothing and the CPU does all the drawing. Performance will be terrible.

I've gone back to my 9 year old nVidia card meanwhile as I need to address Meltdown/Spectre eveywhere and having random lock ups doesn't help.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[thumper]

Have you checked your logs for kernel crash dumps?

I had these:

[PrSo]

thumper,
those messages in log are totally harmless and shouldn't be the reason of hard locking, please see this bug report, and this patch on LKML, so this _is_ a coincidence, although this could be a symptom.

Amity88,
is there any special reason that you are on 4.9.76-gentoo-r1 kernel?

[Amity88]

[Mimamau]

As in my other thread, there seems to be problems with southern islands gpus.
I only get a slow 2d desktop, everything else gives me a blank screen or crashes the system completely.

Even the amdgpu-pro drivers don't work on supported distributions. AMD support wrote:

"I apologize for the delay. I was waiting for feedback from the subject matter experts.
Unfortunately, it appears the HD7870 series has not been qualified with our latest drivers.
The recommendation is to use the inbox drivers or an open source driver, available here: https://www.x.org/wiki/RadeonFeature/#index10h2
If you experience issues with the open source drivers, please file a report at the link above. I have been informed that our engineers monitor and investigate reports listed there.
In order to update this service request, please respond, leaving the service request reference intact.
Best regards,
AMD Global Customer Care"

[NeddySeagoon]

Amity88,

There is a new amdgpu driver is the 4.15 kernel.
Its worth a try.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[Tony0945]

[PrSo]

[NeddySeagoon]

Having done the Spectre updates, I've gone back to my RX450 card.
As the 4.15 kernel didn't fix my lockups, I'm trying 4.16-rc1

Watch this space.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

[Zucca]

[gcyoung]

I am also getting intermittent screen and wireless keyboard freezes. While it works, the amdgpu module seems better than the radeonsi. I don't know if it is connected, but my login dmesg output contains a message [[Firmware Bug:] ACPI MWAIT C-state 0x0 not supported by hw].

I note that the Arch web site also contains referenced to problems with the combination of amdgpu and ryzen processor.

I have ssh'd (without X) into the computer from another machine, and find it is still responding normally to commands.
It's a pity, since I like the result before it freezes!

PS: I am using kernel-4.17.6 which is not listed as stable, but I found the same problem with an earlier stable kernel

[Goverp]

This is probably no help, but I'm running an hp laptop with /proc/cpuinfo model name: "AMD A9-9420 RADEON R5, 5 COMPUTE CORES 2C+3G". It's a STONEY graphics thingy.
It also has an rtl8723de modem, which meant I need a very later kernel (and an external module), so I've been running kernel 4.16 originally, 4.17.1 now.
Never had any problems like described in this thread, nor any issues from using a late kernel.
AFAIK (I read Phoronix summaries) AMDGPU support features regularly in the kernel change logs.

I currently have:

[gcyoung]

It may be of interest to others with this problem to know that since my last posting I have followed a suggestion given under the heading dpm' on https://wiki.gentoo.org/wiki/AMDGPU#Hardware_detection.

[echo performance > /sys/class/drm/card0/device/power_dpm_state]
and:-
[echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level]

Since making these settings I have had no further "freezes", except when I made only the first setting. Since making the settings I have used the computer, including one five hour mythtv frontend performance, for about twenty hours. Previously, I failed regularly to complete a fairly standard viewing of a film --say about two hours, without needing a reboot.

Unfortunately the settings disappear when I log out, although I suppose I can write a small script to run these settings on login. If there is any way to include the settings as options to the module, I'd be glad to hear of it:-- or possibly there might be kernel setting which would do the trick.

If I don't return with a message that I've had another "freezup", then it can be assumed that these settings have solved, at least my difficulty, although it might not work in other cases.

[Amity88]

@gcyoung,
You have solved it I think! This is the same solution that worked for me as well. I came back here to updated it. Basically we need to disable the dynamic power management (dpm) of this gpu.
_________________

[Zucca]

Thanks. Gotta poke those settings too.
Sadly, it looks like power consumption will increase. :|
_________________
..: Zucca :..
Gentoo IRC channels reside on Libera.Chat.
--