Re: [libvirt] [PATCH 05/14] qemu: Implement NVDIMM
On 02/23/2017 11:02 AM, Daniel P. Berrange wrote:
On Thu, Feb 23, 2017 at 10:02:48AM +0100, Michal Privoznik wrote:
> So, majority of the code is just ready as-is. Well, with one
> slight change: differentiate between dimm and nvdimm in places
> like device alias generation, generating the command line and so
> on.
>
> Speaking of the command line, we also need to append 'nvdimm=on'
> to the '-machine' argument so that the nvdimm feature is
> advertised in the ACPI tables properly.
>
> Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com>
> + if (prealloc &&
> + virJSONValueObjectAdd(props,
> + "b:prealloc", true,
> + NULL) < 0)
> + goto cleanup;
As discussed on IRC, using prealloc with QEMU causes it to memset()
the first byte of every page of memory to 0. With NVDIMM this is
obviously corrupting application data stored in the NVDIMM.
Obviously this needs fixing in QEMU, but I would think that libvirt
needs to block use of prealloc==true when running against existing
broken QEMU versions, otherwise users are going to be rather upset
to see their data corrupted on every boot
They are, but should we work around broken QEMUs? And if so - how do we
detect whether the qemu we are dealing with is broken or already fixed
in that respect?
On the other hand, we can just not set prealloc true for nvdimm. That
will have one downside though - after qemu mmaps() the file, kernel is
not forced to create a private copy of the pages.
Michal