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.
There are just a few patches for the x86/misc pull request for the Linux 6.9 merge window. The mentioned changes are:
It's interesting to note that too many non-maskable interrupt (NMI) messages had the potential to slow down systems. It turned out to happen on some AMD platforms. The patch from an AMD engineer had a patch message:
With Linux 6.9 the NMI messages are now rate-limited to avoid slowing down the system. Presumably the patch will also be back-ported to existing stable series.
There are just a few patches for the x86/misc pull request for the Linux 6.9 merge window. The mentioned changes are:
- Fix a wrong check in the function reporting whether a CPU executes (or not) a NMI handler
- Ratelimit unknown NMIs messages in order to not potentially slow down the machine
- Other fixlets
It's interesting to note that too many non-maskable interrupt (NMI) messages had the potential to slow down systems. It turned out to happen on some AMD platforms. The patch from an AMD engineer had a patch message:
"On some AMD machines, unknown NMI messages were printed on the console continuously when using perf command with IBS. It was reported that it can slow down the kernel. Let's ratelimit the unknown NMI messages."
With Linux 6.9 the NMI messages are now rate-limited to avoid slowing down the system. Presumably the patch will also be back-ported to existing stable series.
11 Comments