KVM regression on 64-bit big endian in 5.4

KVM regression on 64-bit big endian in 5.4

[Marcus Comstedt]

Hi.

I discovered after upgrading to kernel 5.4 on my Talos II (Power9
Nimbus DD2.2) that KVM didn't work anymore.  The guest would crash
already in SLOF.

After bisecting, I found that the culprit was this commit:

| From 6c85b7bc637b64e681760f62c0eafba2f56745c6 Mon Sep 17 00:00:00 2001
| From: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
| Date: Thu, 22 Aug 2019 00:48:38 -0300
| Subject: powerpc/kvm: Use UV_RETURN ucall to return to ultravisor
| 
| When an SVM makes an hypercall or incurs some other exception, the
| Ultravisor usually forwards (a.k.a. reflects) the exceptions to the
| Hypervisor. After processing the exception, Hypervisor uses the
| UV_RETURN ultracall to return control back to the SVM.

Looking at the diff, this change caught my eye:

- ld r6, VCPU_CR(r4)
+ lwz r0, VCPU_CR(r4)

So the loading of VPU_CR has changed from ld to lwz.  This seems
wrong.  ccr is an unsigned long, and the store to VCPU_CR in the same
file uses std, not stw.  So loading VCPU_CR must be done with ld,
otherwise the result will be wrong (truncated on LE, truncated and
left shifted by 16 bits on BE)...


  // Marcus