Re: [PATCH v2 6/7] sched: Shard per-LLC shared runqueues

[Peter Zijlstra]

On Wed, Jul 12, 2023 at 03:36:27PM +0530, Gautham R. Shenoy wrote:

> On some Intel servers, it is possible that the CPU numbers are
> interleaved across the two sockets. On my 2 socket, 32Cores per socket
> Ice Lake Server, all the even numbered CPUs are in one socket and all
> the odd numbered CPUs in the other socket.
> 
> The SMT siblings are {0,64}, {2, 66}, .... on one socket and {1, 65},
> {3, 67}, .. on the other.

Yeah, Intel SMT enumeration is a mess. There's a random mix of {n,n+1}
and {0..n-1} {n..2n-1}. And then there's the fun hybrid stuff.  Those
appear to do {n,n+1} for the big cores and then continue with the small
cores in a dense set. My 8+8 ADL, has:

{0,1} {2,3} {4,5} {6,7} {8,9} {10,11} {12,13} {14,15} {16} {17} {18} {19} {20} {21} {22} {23}

I suspect it might be easier to re-number the whole show at boot to fit
a sane pattern rather than trying to match the various random garbage
gifted to us by the BIOS.


I wouldn't worry about it too much at this point.