Re: [Qemu-devel] [RFC PATCH 4/4] nvdimm acpi: build flush hint address structure if required

[Stefan Hajnoczi <stefanha@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1122 bytes --]

On Thu, Apr 06, 2017 at 06:53:09PM +0800, Haozhong Zhang wrote:
> On 04/06/17 11:13 +0100, Stefan Hajnoczi wrote:
> > On Fri, Mar 31, 2017 at 04:41:47PM +0800, Haozhong Zhang wrote:
> > > 
> > > Add an boolean option 'flush-hint' to device 'nvdimm'. If it's on, a
> > > flush hint address structure will be constructed for each nvdimm
> > > device.
> > 
> > Users should not need to set the flush hint option.  NVDIMM
> > configurations that persist data properly without Flush Hint Addresses
> > shouldn't use them (for best performance).  Configurations that rely on
> > flush hints *must* use them to guarantee data integrity.
> 
> It's for backwards compatibility, i.e. migrating a VM on QEMU w/o
> flush hint support to another one w/ flush hint support. By using a
> flush-hint option and making it disabled by default, users can ensure
> both QEMU provide the same VM configuration.

I think QEMU should play a role in deciding whether to use Address Flush
Hints or not.

We should not require the user to set a sensible value.  If they get it
wrong then they may suffer data loss!

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]