When qemu aarch64 is configured as NUMA, the EDK2 boot fail.
Below is my test command:
```
qemu-system-aarch64 \
-nographic \
-serial tcp:localhost:54320 -serial tcp:localhost:54321 \
-smp 4 \
-s -S -machine virt,secure=on,mte=off,gic-version=3,virtualization=false \
-cpu max,sve=off \
-d unimp -semihosting-config enable=on,target=native \
-m 2048 \
-bios bl1.bin\
-initrd rootfs.cpio.gz \
-kernel Image -no-acpi \
-append 'console=ttyAMA0,38400 keep_bootcon root=/dev/vda2 ' \
-machine virt,iommu=smmuv3 -device
virtio-scsi-pci,disable-legacy=on,id=scsi0,iommu_platform=on,addr=0x2 \
-object memory-backend-ram,id=mem0,size=1G \
-object memory-backend-ram,id=mem1,size=1G \
-numa node,memdev=mem0,cpus=0-1,nodeid=0 \
-numa node,memdev=mem1,cpus=2-3,nodeid=1
```
Then I found that ArmVirtPkg always take the first memory node as system memory,
but when I configure qemu, the memory node in qemu dtb is as following.
```
memory@80000000 {
numa-node-id = <0x01>;
reg = <0x00 0x80000000 0x00 0x40000000>;
device_type = "memory";
};
memory@40000000 {
numa-node-id = <0x00>;
reg = <0x00 0x40000000 0x00 0x40000000>;
device_type = "memory";
};
```
Then edk2 use 'memory@80000000' as system memory which is NUMA node 1.
And the memory node order is implement in qemu: arm_load_dtb() which add memory
from node 0 to N.
https://gitlab.com/qemu-project/qemu/-/blob/master/hw/arm/boot.c#L618
As I understand it, loader should take node 0.
Should I modify ArmVirtPkg/PrePi/FdtParser.c: FindMemnode() for the NUMA case ?
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#91074): https://edk2.groups.io/g/devel/message/91074
Mute This Topic: https://groups.io/mt/92198841/21656
Group Owner: devel+ow...@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
