Race or barrier when switching stack

[aik]

I am observing weirdness in PeiCheckAndSwitchStack() when SecCoreData is offset twice. This is to demonstrate the idea:
aik@2e81b55

TRACE1 prints SecCoreData == 81FD20 and TRACE2 prints 7BF55D20 and TRACE3 prints 7BF55D20 (the same as TRACE2).
But if TRACE2 is commented out (as in the patch), TRACE3 prins 0xf768bd20 which is 2*(0x7BF55D20 - 0x81FD20) + 0x81FD20. And the effect is not 100% stable.

The code adjusts SecCoreData before it switches the stack, SecCoreData should be adjusted once but it happens twice. /me brain explodes.

This is AMD EPYC, upstream QEMU, q32 machine, 2GB of RAM, 1 vcpu.

Is some barrier missing here? Thanks,

This is the QEMU cmdline:

/home/aik/pbuild/qemu-snp-localhost-x86_64/qemu-system-x86_64 \
-enable-kvm \
-m 2G \
-smp 1 \
-netdev user,id=USER0,hostfwd=tcp::2223-:22 \
-device virtio-net-pci,id=vnet0,iommu_platform=on,disable-legacy=on,romfile=,netdev=USER0 \
-machine q35 \
-device virtio-scsi-pci,id=vscsi0,iommu_platform=on,disable-modern=off,disable-legacy=on \
-drive id=DRIVE0,if=none,file=img/u2204_128G_aikbook_sev.qcow2,format=qcow2 \
-device scsi-hd,id=scsi-hd0,drive=DRIVE0 \
-drive if=pflash,format=raw,unit=0,file=/home/aik/OVMF_CODE.fd,readonly=on,id=MYPF \
-nographic \
-chardev stdio,id=STDIO0,signal=off,mux=on \
-device isa-serial,id=isa-serial0,chardev=STDIO0 \
-mon id=MON0,chardev=STDIO0,mode=readline \
-kernel /boot/vmlinuz \
-append console=ttyS0,115200n1 earlyprintk root=/dev/sda3 \
-d guest_errors