> Larry McVoy writes:
> > : Does Mr McVoy by any chance happen to have an "is L2 cache shared" detector
> > : lurking in the lmbench and other magic ?
> >
> > Hmm. Not as such. I could probably make one. It would be a lot easier
> > if Linux had a pin() function which pinned a process to a processor.
> > SGI's interface looks like:
> >
> > sysmp(MP_MUSTRUN, cpuNumber);
>
> Digital UNIX spells this
>
> bind_to_cpu(pid_t pid, unsigned long cpu_mask, unsigned long flag);
>
> flag can be set to BIND_NO_INHERIT so that children of the process
> don't inherit the binding. It's a synchronous request: by the time
> it's returned (successfully), the process is running on the designated
> CPU. Oddly, the cpu_mask argument really is a mask: you set a single
> bit corresponding to the CPU you want. I like the idea of the flags
> argument but the mask argument is silly: that API wouldn't even work
> on our 72-CPU O2000. Naughty DEC^H^H^HCompaq.
>
> --Malcolm
Actually, that mask could be useful.. Say you have 64 cpus.. You might
want it to run on procs 1,2,3,4 but not the rest.. Perhaps the reason for
this is that your computer is not 'real' smp but a cluster.. Or perhaps
it's a numa machiene.. Say that 1234,5678,9.. are classic smp computers in
a cluster (or procs in a numa computer shareing the same memory space)
this would prevent migration and keep memory access localized..
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu