CPU index (6) should be smaller than maxcpus (4)

[alex.tls]

I tried to make CPU affinity to work on my VM, but I can't.
This is my configure for my vm:

root@pve:/etc/pve/qemu-server# cat 600.conf
agent: 1
boot: order=scsi0;net0
cores: 4
hotplug: disk,network,usb
machine: pc-i440fx-6.0
memory: 4096
name: win10-1
net0: virtio=5A:DD:1C:E8:A3:85,bridge=vmbr0,firewall=1
numa: 1
numa1: cpus=6-9,hostnodes=1,memory=4096,policy=bind
ostype: win10
scsi0: vms:vm-600-disk-0,size=60G
scsihw: virtio-scsi-pci
smbios1: uuid=6177b506-9db7-455f-9beb-3e382c3c152a
sockets: 1
vmgenid: 0e72f937-4e16-4fd8-80a3-4de7c70c3201

This is my host CPU topology:

root@pve:/etc/pve/qemu-server# numactl --hardware
available: 2 nodes (0-1)
node 0 cpus: 0 1 2 3 4 5 12 13 14 15 16 17
node 0 size: 32169 MB
node 0 free: 22073 MB
node 1 cpus: 6 7 8 9 10 11 18 19 20 21 22 23
node 1 size: 32218 MB
node 1 free: 24006 MB
node distances:
node   0   1
  0:  10  20
  1:  20  10

When I start my VM , it showed me bellow error:

kvm: -numa node,nodeid=1,cpus=6-9,memdev=ram-node1: CPU index (6) should be smaller than maxcpus (4)
TASK ERROR: start failed: QEMU exited with code 1

What's wrong on my configuration ? anyone can help ?

 

[alex.tls]

If I disable NUMA by set it to 0 numa: 0 , then I can start my VM , but I found CPU affinity didn't work via

root@pve:/etc/pve/qemu-server# taskset --all-tasks --pid "$(< /run/qemu-server/600.pid)"
pid 1190387's current affinity mask: ffffff
pid 1190388's current affinity mask: ffffff
pid 1190427's current affinity mask: ffffff
pid 1190428's current affinity mask: ffffff
pid 1190429's current affinity mask: ffffff
pid 1190430's current affinity mask: ffffff
pid 1190433's current affinity mask: ffffff

The mask ffffff (0xFFFFFFFF) means all processors which is from #0 through #last. so my VM still on running on all CPU , not just I defined cpus 6-9.

So weird!

 

[leesteken]

Looking at the documentation and the error message, I would say that you specify which cores are in which NUMA group. Therefore, cpus must be a subset of 0-3 (the 4 virtual cores) and you specify to which NUMA node you want them to belong via hostnodes.
I don't think this supports CPU pinning (mapping virtual to physical cores), only specifying which virtual cores should run on which physical NUMA node.

 

[alex.tls]

The document doesn't mention the parameter of cpus= is not mapping virtual cores to physical cores , if it does , then my configuration should be no problem.
If it's not mapping virtual cores to physical cores , then the numactl --hardware lost its meaning.

 

[Catwoolfii]

You should use a range of values for "cpus" less than the value for "cores".
For example:

cores: 4
numa: 1
numa1: cpus=0-3,hostnodes=1,memory=4096,policy=bind