On Wed, Apr 16, 2008 at 11:02:51AM -0700, Bart Smaalders wrote: > > What's the size of the relevant TLBs? With text, stack and heap > mappings for all > threads, this result isn't terribly surprising. > Huh, how do I find the size of TLBs? :/ Anyway, stack and heap mappings should not affect the instruction TLB at all. And, as already said, all threads execute the same code, so the number of threads should not affect the number of ITLB misses.
> > etc, aren't fixed. The kernel mapping are global if the CPU supports that. > Yes, I was asking about kernel mappings. (I posted another post to the -code forum. After browsing through the kernel code, it seems that CR3 is unconditionally reloaded in hat_swtch, regardless of whether the new thread belongs to the current process or not. Maybe I have looked at the wrong place, or maybe the check isn't obvious.) _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
