Too Many Debug Messages Were Reportedly Slowing Down Some AMD Linux Systems

[phoronix]

Phoronix: Too Many Debug Messages Were Reportedly Slowing Down Some AMD Linux Systems

An interesting anecdote was mentioned as part of the x86/misc changes queued for the Linux 6.9 kernel: on some unnamed AMD systems, NMI debug messages were too excessive that they actually slowed down the systems...

https://www.phoronix.com/news/AMD-Linux-Too-Much-NMI

[stiiixy]

AMD's been going way to fast anyway

And also, isn't "verbose" the generally accepted and industry utilised term for everything this article is about?

[timofonic]

Why so many NMI errors? Maybe there's some unresolved issues or unimplemented features...

[stiiixy]

Hence the patch?

[mathieu.cossette]

At the end of this news article, there is a good news! It will be on 6.8.

[user556]

NMI's shouldn't exist at all. They seriously break determinism in real-time systems. They're probably doing something nutty beyond the OSes control like background monitoring by the OEM firmware.

[NSLW]

Phoronix: Too Many Debug Messages Were Reportedly Slowing Down Some AMD Linux Systems

An interesting anecdote was mentioned as part of the x86/misc changes queued for the Linux 6.9 kernel: on some unnamed AMD systems, NMI debug messages were too excessive that they actually slowed down the systems...

https://www.phoronix.com/news/AMD-Linux-Too-Much-NMI

Just curious. Shouldn't it be "as a part of the x86/misc changes"?

[Emmanuel Deloget]

NMI's shouldn't exist at all. They seriously break determinism in real-time systems. They're probably doing something nutty beyond the OSes control like background monitoring by the OEM firmware.

NMI have existed for decades, way before we got what I'll gently call 'smart cores'. The goal of an NMI is exactly what its name implies, to have a high priority generated interrup that cannot be masked by the system developper. Not having them means that you would face major issues in some cases, up to the destruction of some hardware parts in some extreme cases. They are very likely to be generated only for extremely urgent issues (memory corruption, chipset error...). I'm not aware that they are used by OEM firmwares (it would not make any sense, as they are normally very rare; that would not allow much monitoring...).

Why so many NMI errors? Maybe there's some unresolved issues or unimplemented features...

The article mention that a check that was wrongly telling the system it was executing an NMI handler. So maybe there is no NMI error after all.

[user556]

I said what I said exactly because NMIs have always existed - in the PC architecture at least. They very much do get used as a cheap side stepping of the OS.

[onlyLinuxLuvUBack]

wait until they rewrite it in 1000% safe rust