Ryzen Stability Issues Are Still Affecting Some FreeBSD Users

[phoronix]

Phoronix: Ryzen Stability Issues Are Still Affecting Some FreeBSD Users

While in recent months there have been some improvements to FreeBSD that have helped yield greater reliability in running AMD Ryzen processors on this BSD operating system, some users are still reporting hard to diagnose stability problems on FreeBSD...

http://www.phoronix.com/scan.php?page=news_item&px=Ryzen-FreeBSD-Stability-Spring

[shmerl]

For some, FreeBSD on Ryzen is still leading to lock-ups

With Ryzen 2 (2600/2700[X])? That would be quite annoying.

[Chewi]

Although it hasn't worked for everyone, most people report success when selecting the new "Typical Current Idle" option in the BIOS. It has certainly worked for me. AMD haven't explained what this does but it appears to disable the C6 Package state, using around 1-4W more on idle. I can live with that. If the option appears ineffective or you don't have it in your BIOS then you can use a patched zenstates.py (basically just modifies an MSR) to do the same thing. While no one seems to know what the real issue is, most agree that it is not an OS issue so there is little point in discussing FreeBSD specifics.

[Almindor]

It's the dreaded freezup. I had this on Linux in the early days with my R7 1700. Not sure what changed but it "got fixed" for me. I upped BIOS to latest beta from MSI, overclocked to 3.7Ghz and XAMP profile 1 for RAM and turned off the "quiet" feature. I suspect one of those things might have turned off the C6 as well.

[shmerl]

Although it hasn't worked for everyone, most people report success when selecting the new "Typical Current Idle" option in the BIOS. It has certainly worked for me. AMD haven't explained what this does but it appears to disable the C6 Package state, using around 1-4W more on idle. I can live with that. If the option appears ineffective or you don't have it in your BIOS then you can use a patched zenstates.py (basically just modifies an MSR) to do the same thing. While no one seems to know what the real issue is, most agree that it is not an OS issue so there is little point in discussing FreeBSD specifics.

Yep. That disables package C6 state and fixes freezes with older Ryzens. I just got the new 2700X and will test whether it's still affected by this issue or package C6 can be enabled normally.

[Djhg2000]

My 2400G still locks up on Linux. The most reliable trigger seems to be running Far Cry 4 in WINE with DXVK. It can happen at idle as well, but it can survive days or weeks at idle while the above case usually triggers a crash within seconds.

Curiously all of the traces I've been able to capture points to a general protection fault in syscall 64 (semget). Anyone with similar logs? I used pstore to have a persistent kernel log area in RAM which usually survives a hard reset.

[VikingGe]

The most reliable trigger seems to be running Far Cry 4 in WINE with DXVK.

You sure it's actually your CPU locking up in this particular case and not your GPU? Because that sadly happens a lot with DXVK on RADV.

That said, I got my 2700X today, no issues compiling stuff or anything, I really hope it's stable.

[shmerl]

That said, I got my 2700X today, no issues compiling stuff or anything, I really hope it's stable.

Congrats with CPU upgrade

I hope it's stable too and won't be heating up too much.

[ElmerFudd]

Although it hasn't worked for everyone, most people report success when selecting the new "Typical Current Idle" option in the BIOS. It has certainly worked for me. AMD haven't explained what this does but it appears to disable the C6 Package state, using around 1-4W more on idle. I can live with that. If the option appears ineffective or you don't have it in your BIOS then you can use a patched zenstates.py (basically just modifies an MSR) to do the same thing. While no one seems to know what the real issue is, most agree that it is not an OS issue so there is little point in discussing FreeBSD specifics.

From what I understand, "Typical Current Idle" makes the CPU use a little power always, as some PSU/Mobo combinations doesn't handle the CPU using 0 amps. I can't remember where I read it, but it was a link from the kernel bugzilla entry I think.
My PSU doesn't handle it, and my mobo finally got the option in the BIOS a few days ago - so I can use C6 state again.

[torsionbar28]

Hope they get these squashed soon, I've got racks full of production RHEL servers at work on AMD Opteron 6000 series. I'd like to replace them with Epyc, but will wait it out a little longer. RHEL on Opteron is about as rock solid stable as it gets! Dont want any early adopter headaches when its time to upgrade.