Re: [GIT PULL] x86/apic changes for v2.6.38

[Randy Dunlap <randy.dunlap@oracle.com>]

On Thu, 6 Jan 2011 16:17:10 -0800 Yinghai Lu wrote:

> On Thu, Jan 6, 2011 at 3:53 PM, Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> > On Thu, Jan 6, 2011 at 3:48 PM, Luck, Tony <tony.luck@intel.com> wrote:
> >> Untested on X86.  Builds and boots on ia64 (both normally and with
> >> maxcpus=8 to limit the number of cpus).
> >>
> >> Signed-off-by: Tony Luck <tony.luck@intel.com>
> >
> > Looks much better to me, and gets rid of that silly #ifdef that I reacted to.
> >
> > But no hurry, this looks low-priority. So let's make this go through
> > the x86 tree and get merged later, after it has passed whatever normal
> > tests that the -tip tree goes through. Ok?
> 
> yes. x86 is safe. We have apicid_to_node[] array, and already check
> apic id with MAX_LOCAL_APIC
> in x86 version acpi_numa_processor_affinity_init().
> 
> looks ia64 is using struct array with cpu idx for mapping
> 
>         node_cpuid[srat_num_cpus].phys_id =
>             (pa->apic_id << 8) | (pa->local_sapic_eid);
>         /* nid should be overridden as logical node id later */
>         node_cpuid[srat_num_cpus].nid = pxm;
>         cpu_set(srat_num_cpus, early_cpu_possible_map);
>         srat_num_cpus++;
> 
> but it does not check the boundary of that array...
> 
> struct node_cpuid_s node_cpuid[NR_CPUS]..
> 
> so if some one try to boot kernel with small NR_CPUS on bigger IA64
> system, could get some variables node_cpuid[]...
> 
> Assume Tony will have another patch for IA64 to check that before
> apply this patch.
> 
> sth like
> 
> Index: linux-2.6/arch/ia64/kernel/acpi.c
> ===================================================================
> --- linux-2.6.orig/arch/ia64/kernel/acpi.c
> +++ linux-2.6/arch/ia64/kernel/acpi.c
> @@ -477,6 +479,11 @@ acpi_numa_processor_affinity_init(struct
>  	if (!(pa->flags & ACPI_SRAT_CPU_ENABLED))
>  		return;
> 
> +	if (srat_num_cpus >= NR_CPUS) {
> +		printk(KERN_WARNING "NR_CPUS (%d) is too small, not all cpus used.
> %d\n", NR_CPUS);

		                                                        CPUs used.

Don't need %d 2 times.


> +		return;
> +	}
> +
>  	pxm = get_processor_proximity_domain(pa);
> 
>  	/* record this node in proximity bitmap */


---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***