Re: Regression in 2.6.23-rc2-mm2, mounting cpusets causes a hang

[Lee Schermerhorn <Lee.Schermerhorn@hp.com>]

On Mon, 2007-08-13 at 15:12 -0500, Serge E. Hallyn wrote:
> Quoting Dhaval Giani (dhaval@linux.vnet.ibm.com):
> > Hi,
> > 
> > On mounting cpusets using containers, I have been hitting the following
> > bug.
> > 
> > 
> > -----------[ cut here ]------------
> > kernel BUG at kernel/cpuset.c:331!
<snip>
> > CONFIG_HIGHMEM=y
> > CONFIG_X86_PAE=y
> > # CONFIG_NUMA is not set
> > CONFIG_ARCH_POPULATES_NODE_MAP=y
> > CONFIG_SELECT_MEMORY_MODEL=y
> > CONFIG_FLATMEM_MANUAL=y
> > # CONFIG_DISCONTIGMEM_MANUAL is not set
> > # CONFIG_SPARSEMEM_MANUAL is not set
> > CONFIG_FLATMEM=y
> > CONFIG_FLAT_NODE_MEM_MAP=y
> > # CONFIG_SPARSEMEM_STATIC is not set
> > CONFIG_SPLIT_PTLOCK_CPUS=4
> > CONFIG_RESOURCES_64BIT=y
> > CONFIG_ZONE_DMA_FLAG=1
<snip>
> 
> Yeah, I'm seeing the same thing.  Oddly, my node_states[N_NORMAL_MEMORY]
> and node_states[N_HIGH_MEMORY] are empty, while node_states[N_ONLINE]
> contains my single cpu (on i386 kvm image).
> 
> -serge

Yes, you'll definitely hit that BUG if the N_HIGH_MEMORY mask is empty.
So far, I can't see how this could be, tho'.  __build_all_zonelists()
should be called for non-NUMA as well as NUMA.  It iterates over "all
on-line nodes"--the N_ONLINE state, which is initialized to have node 0
on-line.  That should work for a non-NUMA i386, right?  

In the loop over all on-line nodes, __build_all_zonelists() sets
N_HIGH_MEMORY for each [on-line] node with ANY present pages--i.e.,
non-zero pgdat->node_present_pages.  Then, check_for_regular_memory()
will set N_NORMAL_MEMORY for any nodes with non-zero zone->present_pages
for any zone <= ZONE_NORMAL.  

I don't have an i386 test system to verify this, but from cscope, it
looks like the pgdat, including node_present_pages should be setup in
the following call path:

	start_kernel()->setup_arch()->zone_sizes_init()->
	free_area_init_zones()->free_area_init_zone->
	calculate_node_total_pages()

This is called before build_all_zonelists()--also from start_kernel(),
so node_present_pages should be non-zero when we populate the N_*_MEMORY
states...

Any ideas, anyone?

Lee