qemu64 CPU model is incorrect

Bug #1860575 reported by The Lemon Man on 2020-01-22
This bug affects 1 person
Affects Status Importance Assigned to Milestone

Bug Description

At the moment the "qemu64" CPU is defined as follows:

        .vendor = CPUID_VENDOR_AMD,
        .family = 6,
        .model = 6,
        .stepping = 3,

According to Wikipedia [1] this means the CPU is defined as part of the
K7 family while the AMD64 ISA was only introduced with the K8 series!

This causes some software such as LLVM to notice the problem (32-bit cpu
with 64-bit capability reported in the cpuid flag) and produce various
error messages.

The simple solution would be to upgrade this definition to use the Sledgehammer
family (15) instead.

[1] https://en.wikipedia.org/wiki/List_of_AMD_CPU_microarchitectures

Daniel Berrange (berrange) wrote :

Your analysis of the problem with family makes sense & we do have mechanism to fix this in QEMU while keeping back compat for existing deployments.

I'm curious as to the actual errors LLVM reports ?

FWIW, even though qemu64 is the default CPU, practically everyone would be better off choosing one of the other CPU models explicitly to better suit their desired use case. There is some guidance here https://qemu.weilnetz.de/doc/qemu-doc.html#cpu_005fmodels

The Lemon Man (lemonboy) wrote :

The error message is a rather cryptic "LLVM ERROR: 64-bit code requested on a subtarget
that doesn't support it!" as it knows Athlon CPUs don't support the AMD64 ISA.

I will relay the tip to the people managing the VMs, I guess this problem went unnoticed
for so long because there are not many `qemu64` users.

I'm available to test a patch whenever it becomes available, I didn't directly send one
because I was afraid of breaking the backward compatibility and some (many?) VMs.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers