Re: [RFC PATCH V5] mm readahead: Fix readahead fail for no local memory and limit readahead pages

[Nishanth Aravamudan <nacc@linux.vnet.ibm.com>]

On 17.02.2014 [15:14:06 -0800], David Rientjes wrote:
> On Mon, 17 Feb 2014, Nishanth Aravamudan wrote:
> 
> > Here is what I'm running into now:
> > 
> > setup_arch ->
> > 	do_init_bootmem ->
> > 		cpu_numa_callback ->
> > 			numa_setup_cpu ->
> > 				map_cpu_to_node -> 
> > 					update_numa_cpu_lookup_table
> > 
> > Which current updates the powerpc specific numa_cpu_lookup_table. I
> > would like to update that function to use set_cpu_numa_node() and
> > set_cpu_numa_mem(), but local_memory_node() is not yet functional
> > because build_all_zonelists is called later in start_kernel. Would it
> > make sense for first_zones_zonelist() to return NUMA_NO_NODE if we
> > don't have a zone?
> > 
> 
> Hmm, I don't think we'll want to modify the generic first_zones_zonelist() 
> for a special case that is only true during boot.  Instead, would it make 
> sense to modify numa_setup_cpu() to use the generic cpu_to_node() instead 
> of using a powerpc mapping and then do the set_cpu_numa_mem() after 
> paging_init() when the zonelists will have been built and zones without 
> present pages are properly excluded?

Sorry, I was unclear in my e-mail. I meant to modify
local_memory_node(), not first_zones_zonelist(). Well, it only needs the
following, I think?

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index e3758a0..5de4337 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3650,6 +3650,8 @@ int local_memory_node(int node)
                                   gfp_zone(GFP_KERNEL),
                                   NULL,
                                   &zone);
+       if (!zone)
+               return NUMA_NO_NODE;
        return zone->node;
 }
 #endif

I think that condition should only happen during boot -- maybe even
deserving of an unlikely, but I don't think the above is considered a
hot-path. If the above isn't palatable, I can look into your suggestion
instead.

Thanks,
Nish