Re: [PATCH 1/2] mm/memcontrol: do not tweak node in mem_cgroup_init()

From: Haifeng Xu
Date: Fri Jun 16 2023 - 04:28:55 EST




On 2023/6/15 16:14, Michal Hocko wrote:
> On Thu 15-06-23 07:32:25, Haifeng Xu wrote:
>> mem_cgroup_init() request for allocations from each possible node, and
>> it's used to be a problem because NODE_DATA is not allocated for offline
>> node. Things have already changed since commit 09f49dca570a9 ("mm: handle
>> uninitialized numa nodes gracefully"), so it's unnecessary to check for
>> !node_online nodes here.
>
> How have you tested this patch?

Start with one empty node:

qemu-system-x86_64 \
-kernel vmlinux \
-initrd full.rootfs.cpio.gz \
-append "console=ttyS0,115200 root=/dev/ram0 nokaslr earlyprintk=serial oops=panic panic_on_warn" \
-drive format=qcow2,file=vm_disk.qcow2,media=disk,if=ide \
-enable-kvm \
-cpu host \
-m 8G,slots=2,maxmem=16G \
-smp cores=4,threads=1,sockets=2 \
-object memory-backend-ram,id=mem0,size=4G \
-object memory-backend-ram,id=mem1,size=4G \
-numa node,memdev=mem0,cpus=0-3,nodeid=0 \
-numa node,memdev=mem1,cpus=4-7,nodeid=1 \
-numa node,nodeid=2 \
-net nic,model=virtio,macaddr=52:54:00:12:34:58 \
-net user \
-nographic \
-rtc base=localtime \
-gdb tcp::6000

Guest state when booting:
[ 0.048881] NUMA: Node 0 [mem 0x00000000-0x0009ffff] + [mem 0x00100000-0xbfffffff] -> [mem 0x00000000-0xbfffffff]
[ 0.050489] NUMA: Node 0 [mem 0x00000000-0xbfffffff] + [mem 0x100000000-0x13fffffff] -> [mem 0x00000000-0x13fffffff]
[ 0.052173] NODE_DATA(0) allocated [mem 0x13fffc000-0x13fffffff]
[ 0.053164] NODE_DATA(1) allocated [mem 0x23fffa000-0x23fffdfff]
[ 0.054187] Zone ranges:
[ 0.054587] DMA [mem 0x0000000000001000-0x0000000000ffffff]
[ 0.055551] DMA32 [mem 0x0000000001000000-0x00000000ffffffff]
[ 0.056515] Normal [mem 0x0000000100000000-0x000000023fffffff]
[ 0.057484] Movable zone start for each node
[ 0.058149] Early memory node ranges
[ 0.058705] node 0: [mem 0x0000000000001000-0x000000000009efff]
[ 0.059679] node 0: [mem 0x0000000000100000-0x00000000bffdffff]
[ 0.060659] node 0: [mem 0x0000000100000000-0x000000013fffffff]
[ 0.061649] node 1: [mem 0x0000000140000000-0x000000023fffffff]
[ 0.062638] Initmem setup node 0 [mem 0x0000000000001000-0x000000013fffffff]
[ 0.063745] Initmem setup node 1 [mem 0x0000000140000000-0x000000023fffffff]
[ 0.064855] DMA zone: 158 reserved pages exceeds freesize 0
[ 0.065746] Initializing node 2 as memoryless
[ 0.066437] Initmem setup node 2 as memoryless
[ 0.067132] DMA zone: 158 reserved pages exceeds freesize 0
[ 0.068037] On node 0, zone DMA: 1 pages in unavailable ranges
[ 0.068265] On node 0, zone DMA: 97 pages in unavailable ranges
[ 0.124755] On node 0, zone Normal: 32 pages in unavailable ranges


cat /sys/devices/system/node/online
0-1
cat /sys/devices/system/node/possible
0-2

In addition, I add a debug meesage:

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 7ebf64e48b25..3d786281377d 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -7424,7 +7424,7 @@ static int __init mem_cgroup_init(void)
rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node);
if (!rtpn)
continue;
-
+ pr_info("allocate rtpn node %d.\n", node);
rtpn->rb_root = RB_ROOT;
rtpn->rb_rightmost = NULL;
spin_lock_init(&rtpn->lock);


[ 0.561420] allocate rtpn node 0.
[ 0.562324] allocate rtpn node 1.
[ 0.563322] allocate rtpn node 2.


>
> I am not saying it is wrong and it looks like the right thing to do. But
> the early init code has proven to be more subtle than expected so it is
> definitely good to know that this has been tested on memory less setup
> and passed.
>
>> Signed-off-by: Haifeng Xu <haifeng.xu@xxxxxxxxxx>
>> ---
>> mm/memcontrol.c | 3 +--
>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index 4b27e245a055..c73c5fb33f65 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -7421,8 +7421,7 @@ static int __init mem_cgroup_init(void)
>> for_each_node(node) {
>> struct mem_cgroup_tree_per_node *rtpn;
>>
>> - rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL,
>> - node_online(node) ? node : NUMA_NO_NODE);
>> + rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, node);
>>
>> rtpn->rb_root = RB_ROOT;
>> rtpn->rb_rightmost = NULL;
>> --
>> 2.25.1
>