Hi Brice, I have discussed the issue internally and we can confirm that CPU topology for POWER7 as reported by 2.6.32 kernel (in RHEL6.0) is indeed wrong.
I have submitted BZ against RHEL6.0 https://bugzilla.redhat.com/show_bug.cgi?id=621546 to get it fixed. Let's see if the fix can be back-ported from 2.6.34 kernel. Thanks Jirka On Thursday, August 05, 2010 07:08:09 am Brice Goglin wrote: > Power7 topology isn't properly reported by old kernels. We've been said > that it works fine with 2.6.34. I am not sure which commit fixed this. I > don't see many commits talk about Power7 topology between 2.6.32 and > 2.6.34, so it may be this one (from 2.6.34): > > commit 4b83c330b4d38e869111bda6e9077d4f61ed974a > Author: Anton Blanchard <an...@samba.org> > Date: Wed Apr 7 15:33:44 2010 +0000 > > powerpc/numa: Add form 1 NUMA affinity > > Firmware changed the way it represents memory and cpu affinity on > POWER7. Unfortunately the old method now caps the topology to work around > issues with legacy operating systems. For Linux to get the correct > topology we need to use the new form 1 affinity information. > > We set the form 1 field in the client architecture, and if we see "1" > in the ibm,associativity-form property firmware supports form 1 affinity > and we should look at the first field in the > ibm,associativity-reference-points array. If not we use the second field > as we always have. > > Signed-off-by: Anton Blanchard <an...@samba.org> > Signed-off-by: Benjamin Herrenschmidt <b...@kernel.crashing.org> > > > Maybe some distros should backport it... > > Brice > > Le 05/08/2010 00:19, Jirka Hladky a écrit : > > Hi, > > > > I just got access to one POWER7 box. Both hwloc 1.0.2 and latest hwloc > > 1.1 seem to have problems to detect the sockets correctly. > > > > It's IBM 8406-71Y server equipped with only one socket. On the chip there > > are 8 cores. lstopo seems to be wrong here, reporting 8 sockets. > > > > I'm not sure if this problem arose in linux kernel or it's bug in lstopo > > code. Does anybody else tried to run hwloc on POWER7? > > > > Thanks > > Jirka > > > > > > Kernel: 2.6.32-54.el6.ppc64 > > > > r...@ibm-js24-01.lab.bos.redhat.com: /tmp/hwloc-1.1a1r2387/utils > > $ ./lstopo > > Machine (7616MB) > > > > Socket #0 + L3 #0 (4096KB) + L2 #0 (256KB) + L1 #0 (32KB) + Core #0 > > > > PU #0 (phys=0) > > PU #1 (phys=1) > > PU #2 (phys=2) > > PU #3 (phys=3) > > > > Socket #1 + L3 #1 (4096KB) + L2 #1 (256KB) + L1 #1 (32KB) + Core #1 > > > > PU #4 (phys=4) > > PU #5 (phys=5) > > PU #6 (phys=6) > > PU #7 (phys=7) > > > > Socket #2 + L3 #2 (4096KB) + L2 #2 (256KB) + L1 #2 (32KB) + Core #2 > > > > PU #8 (phys=8) > > PU #9 (phys=9) > > PU #10 (phys=10) > > PU #11 (phys=11) > > > > Socket #3 + L3 #3 (4096KB) + L2 #3 (256KB) + L1 #3 (32KB) + Core #3 > > > > PU #12 (phys=12) > > PU #13 (phys=13) > > PU #14 (phys=14) > > PU #15 (phys=15) > > > > Socket #4 + L3 #4 (4096KB) + L2 #4 (256KB) + L1 #4 (32KB) + Core #4 > > > > PU #16 (phys=16) > > PU #17 (phys=17) > > PU #18 (phys=18) > > PU #19 (phys=19) > > > > Socket #5 + L3 #5 (4096KB) + L2 #5 (256KB) + L1 #5 (32KB) + Core #5 > > > > PU #20 (phys=20) > > PU #21 (phys=21) > > PU #22 (phys=22) > > PU #23 (phys=23) > > > > Socket #6 + L3 #6 (4096KB) + L2 #6 (256KB) + L1 #6 (32KB) + Core #6 > > > > PU #24 (phys=24) > > PU #25 (phys=25) > > PU #26 (phys=26) > > PU #27 (phys=27) > > > > Socket #7 + L3 #7 (4096KB) + L2 #7 (256KB) + L1 #7 (32KB) + Core #7 > > > > PU #28 (phys=28) > > PU #29 (phys=29) > > PU #30 (phys=30) > > PU #31 (phys=31) > > > > _______________________________________________ > > hwloc-devel mailing list > > hwloc-de...@open-mpi.org > > http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel > > _______________________________________________ > hwloc-devel mailing list > hwloc-de...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel
