Re: [Qemu-devel] About vNVDIMM question in TCG

[Haozhong Zhang]

On 11/20/16 10:58 -0700, Lin Ma wrote:
> Hi Guangrong,
>
> I'm interested in vNVDIMM on qemu. I'd like to give it a try on my laptop.
> Because I dont have skylake server hardware, So I want to expose cpu flags
> clflushopt, clwb and pcommit to guest for NVDIMM drivers on TCG.
>
> I saw clflushopt and clwb feature flags in cpuinfo of guest, but no pcommit.
> Could you please tell me is there anyhing wrong in my qemu CLI or anything
> I misunderstand?

pcommit had been deprecated (see KVM commit
dfa169bbee00671288df25f8ef8a2f6e13fe2581), so it's correct to not have
it in guest.

> qemu-system-x86_64 \
> -cpu Skylake-Client,+pcommit,+clflushopt,+clwb \
> -machine pc-i440fx-2.8,nvdimm=on,accel=tcg \
> -m 4G,maxmem=... \
> -object memory-backend-file,... \
> -device nvdimm,... \
>
> BTW, In kvm mode, Even if I used '-cpu qemu64' without flags pcommit,
> clflushopt and clwb, The block device /dev/pmem0 is still there and can
> be used successfully. Does it make sense? Doesn't nvdimm kernel driver
> need pcommit, clflushopt and clwb to work?

If my memory is correct, Linux NVDIMM driver will fallback to clflush
when neither clflushopt nor clwb is available.

Haozhong