Hi Michael, On 04/04/2014 09:18 AM, Michael wang wrote: > Hi, Srivatsa > > Thanks for your reply :) > > On 04/03/2014 04:50 PM, Srivatsa S. Bhat wrote: > [snip] >> >> Now, the interesting thing to note here is that, if CPU0's node was already >> set as node0, *nothing* should go wrong, since its just a redundant update. >> However, if CPU0's original node mapping was something different, or if >> node0 doesn't even exist in the machine, then the system can crash. > > By printk I confirmed all cpus was belong to node 1 at very beginning, > and things become magically after the wrong updating... >
Ok, thanks! >> >> Have you verified that CPU0's node mapping is different from node 0? >> That is, boot the kernel with "numa=debug" in the kernel command line and >> it will print out the cpu-to-node associativity during boot. That way you >> can figure out what was the original associativity that was set. This will >> confirm the theory that the hypervisor sent a redundant update, but because >> of the weird pre-allocation using kzalloc that we do inside >> arch_update_cpu_topology(), we wrongly updated CPU0's mapping as CPU0 <-> >> Node0. > > Associativity should changes, otherwise we won't continue the updating, > and empty updates[] was confirmed to show up inside > arch_update_cpu_topology(). > Ah, ok, that makes it very clear. So, I agree that your patch is correct, but I think the comment in your patch can be enhanced a bit. I'll suggest something if I manage to come up with a better wording. > What I can't make sure is whether this is legal, notify changes but no > changes happen sounds weird...however, even if it's legal, a check in > here still make sense IMHO. > That looks like a bug in the hypervisor/firmware. But the Linux kernel should be able to handle such NULL updates without crashing. So yes, your patch makes sense to me. Thank you! Regards, Srivatsa S. Bhat >> >>> Thus we should stop the updating in such cases, this patch will achieve >>> this and fix the issue. >>> >>> CC: Benjamin Herrenschmidt <b...@kernel.crashing.org> >>> CC: Paul Mackerras <pau...@samba.org> >>> CC: Nathan Fontenot <nf...@linux.vnet.ibm.com> >>> CC: Stephen Rothwell <s...@canb.auug.org.au> >>> CC: Andrew Morton <a...@linux-foundation.org> >>> CC: Robert Jennings <r...@linux.vnet.ibm.com> >>> CC: Jesse Larrew <jlar...@linux.vnet.ibm.com> >>> CC: "Srivatsa S. Bhat" <srivatsa.b...@linux.vnet.ibm.com> >>> CC: Alistair Popple <alist...@popple.id.au> >>> Signed-off-by: Michael Wang <wang...@linux.vnet.ibm.com> >>> --- >>> arch/powerpc/mm/numa.c | 9 +++++++++ >>> 1 file changed, 9 insertions(+) >>> >>> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c >>> index 30a42e2..6757690 100644 >>> --- a/arch/powerpc/mm/numa.c >>> +++ b/arch/powerpc/mm/numa.c >>> @@ -1591,6 +1591,14 @@ int arch_update_cpu_topology(void) >>> cpu = cpu_last_thread_sibling(cpu); >>> } >>> >>> + /* >>> + * The 'cpu_associativity_changes_mask' could be cleared if >>> + * all the cpus it indicates won't change their node, in >>> + * which case the 'updated_cpus' will be empty. >>> + */ >>> + if (!cpumask_weight(&updated_cpus)) >>> + goto out; >>> + >>> stop_machine(update_cpu_topology, &updates[0], &updated_cpus); >>> >>> /* >>> @@ -1612,6 +1620,7 @@ int arch_update_cpu_topology(void) >>> changed = 1; >>> } >>> >>> +out: >>> kfree(updates); >>> return changed; >>> } >>> >> _______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
