Messages in this thread |  | | | Date | Fri, 22 May 2009 19:29:31 +1000 | | From | Paul Mackerras <> | | Subject | Re: [PATCH 2/2] perf_counter: optimize context switch between identical inherited contexts |
| |
Peter Zijlstra writes:
> Suppose someone writes a malicious proglet that inherits the counters, > puts the child to sleep, does 2^32 mods on the counter set, and then > wakes up the child. > > Would that merely corrupt the results, or make the kernel explode?
You'd have to do something like create some counters on yourself, fork, do the 2^32 counter creations and deletions, then do another fork. All that would happen is that some of the counters would count on one of the child processes when they should be counting on the other.
But it would be easy and cheap to make the generation count be 64 bits, and then it won't overflow in my lifetime at least, and after that I don't care. :) So I agree with Ingo that we should just do that.
Paul.
|  |