client coredumping a lot about corefreq HOT 6 CLOSED

Ok sorry i was too fast, but the issue still occuring ! coredumping again and again !

from corefreq.

Hello netmonk

I need more details to help you solve the issue :
1- I believe you have already disabled the nmi_watchdog ?
2- You are using recent and baseline kernel + gcc + libc ?
3- You don't have modified the compiler flags ? No optimization O1, O2 or O3 which could involve unexpected sse instructions ?

To investigate the issue, feel free to post details of your setup : Processor , Kernel distro and so on

Cyril

from corefreq.

Hello cyril,

Only the client is coredumping, do you think nmi_watchdog can have something to do with the client (kernel modul is ok and the daemon too).

Im using archlinux with gcc 6.3.1 with kernel 4.9.11-1, standard make without any optims

from corefreq.

Yes it has a huge impact for a simple reason :

For drawing performance and cpu overhead, corefreq-cli does not check any buffer overflow. It's an implementation choice.

Programmatically speaking, it means no ' if ' condition is surrounding ascii buffers read/write because values have to be well defined by the driver, then the daemon.

When nmi_watchdog is enabled, it conflicts with CoreFreq driver whose values are corrupted and buffer overflows happen in client.

So what could be solutions :
1- Test any value (thus cpu overhead)
2- Respect the prerequisites and disable nmi_watchdog (less cpu overhead)
3- others ?

So yes, please disable nmi_watchdog, and let me know if it runs better ?

Sincerely,
Cyril

from corefreq.

from corefreq.

Disabling nmi_watchdog at cmdline seems to solve the issue !
No more coredump after reboot ! :) great

from corefreq.