|
|
Log in / Subscribe / Register

An unexpected perf feature

An unexpected perf feature

Posted May 22, 2013 1:29 UTC (Wed) by deater (subscriber, #11746)
In reply to: An unexpected perf feature by gerdesj
Parent article: An unexpected perf feature

> On reflection and now I've bothered to check I DO have perf enabled,
> what about treating kernel modules in a similar way to say PHP extras
> like cURL in Ubuntu? (I only say Ubuntu because I have some experience
> there - I'm sure others have similar policies)

People keep going on about disabling perf_event or using it as a module.

perf_event *cannot* be configured as a module, and it *cannot* be disabled on X86. It's always there. And you can't disable it after boot either.
It looks like this was introduced explicitly by cc2067a51424dd25 in Nov 2010 but the dependency existed before because HAVE_HW_BREAKPOINT (default on x86) depends on PERF_EVENTS.

Maybe it's time for a /proc/sys/perf_event_paranoid value of "3" meaning no perf_events at all.

I must admit I'm of two minds of this. As a HPC researcher the main positive feature of perf_event was the fact that it was available everywhere by default, negating the previous hassle of having to get sysadmin intervention to do performance analysis. It would be sad to return to those days, but I guess security trumps convenience.


The LWN site is currently under high scraper load, so comment display has been suppressed for anonymous users. If you are a human, you may read the comments by clicking the button below:

Note: you can avoid this step in the future by logging into your LWN account.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds