[PATCH] powerpc/numa: Handle partially initialized numa nodes

[Oscar Salvador]

On Wed, Mar 30, 2022 at 07:21:23PM +0530, Srikar Dronamraju wrote:
> With commit 09f49dca570a ("mm: handle uninitialized numa nodes
> gracefully") NODE_DATA for even a memoryless/cpuless node is partially
> initialized at boot time.
> 
> Before onlining the node, current Powerpc code checks for NODE_DATA to
> be NULL. However since NODE_DATA is partially initialized, this check
> will end up always being false.
> 
> This causes hotplugging a CPU to a memoryless/cpuless node to fail.
> 
> Before adding CPUs
> $ numactl -H
> available: 1 nodes (4)
> node 4 cpus: 0 1 2 3 4 5 6 7
> node 4 size: 97372 MB
> node 4 free: 95545 MB
> node distances:
> node   4
> 4:  10
> 
> $ lparstat
> System Configuration
> type=Dedicated mode=Capped smt=8 lcpu=1 mem=99709440 kB cpus=0 ent=1.00
> 
> %user  %sys %wait    %idle    physc %entc lbusy   app  vcsw phint
> ----- ----- -----    -----    ----- ----- ----- ----- ----- -----
> 2.66  2.67  0.16    94.51     0.00  0.00  5.33  0.00 67749     0
> 
> After hotplugging 32 cores
> $ numactl -H
> node 4 cpus: 0 1 2 3 4 5 6 7 120 121 122 123 124 125 126 127 128 129 130
> 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148
> 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
> 167 168 169 170 171 172 173 174 175
> node 4 size: 97372 MB
> node 4 free: 93636 MB
> node distances:
> node   4
> 4:  10
> 
> $ lparstat
> System Configuration
> type=Dedicated mode=Capped smt=8 lcpu=33 mem=99709440 kB cpus=0 ent=33.00
> 
> %user  %sys %wait    %idle    physc %entc lbusy   app  vcsw phint
> ----- ----- -----    -----    ----- ----- ----- ----- ----- -----
> 0.04  0.02  0.00    99.94     0.00  0.00  0.06  0.00 1128751     3
> 
> As we can see numactl is listing only 8 cores while lparstat is showing
> 33 cores.
> 
> Also dmesg is showing messages like:
> [ 2261.318350 ] BUG: arch topology borken
> [ 2261.318357 ]      the DIE domain not a subset of the NODE domain
> 
> Fixes: 09f49dca570a ("mm: handle uninitialized numa nodes gracefully")
> Cc: linuxppc-dev at lists.ozlabs.org
> Cc: linux-mm at kvack.org
> Cc: Michal Hocko <mhocko at kernel.org>
> Cc: Michael Ellerman <mpe at ellerman.id.au>
> Reported-by: Geetika Moolchandani <Geetika.Moolchandani1 at ibm.com>
> Signed-off-by: Srikar Dronamraju <srikar at linux.vnet.ibm.com>

Acked-by: Oscar Salvador <osalvador at suse.de>

Thanks Srikar!

> ---
>  arch/powerpc/mm/numa.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
> index b9b7fefbb64b..13022d734951 100644
> --- a/arch/powerpc/mm/numa.c
> +++ b/arch/powerpc/mm/numa.c
> @@ -1436,7 +1436,7 @@ int find_and_online_cpu_nid(int cpu)
>  	if (new_nid < 0 || !node_possible(new_nid))
>  		new_nid = first_online_node;
>  
> -	if (NODE_DATA(new_nid) == NULL) {
> +	if (!node_online(new_nid)) {
>  #ifdef CONFIG_MEMORY_HOTPLUG
>  		/*
>  		 * Need to ensure that NODE_DATA is initialized for a node from
> -- 
> 2.27.0
> 
> 

-- 
Oscar Salvador
SUSE Labs