On 3/25/21 7:21 AM, David Gibson wrote:
On Wed, Mar 24, 2021 at 09:34:06AM +0530, Aneesh Kumar K.V wrote:
On 3/24/21 8:37 AM, David Gibson wrote:
On Tue, Mar 23, 2021 at 09:47:38AM -0400, Shivaprasad G Bhat wrote:
The patch adds support for the SCM flush hcall for the nvdimm devices.
...
collects all the hcall states from 'completed' list. The necessary
nvdimm flush specific vmstate structures are added to the spapr
machine vmstate.
Signed-off-by: Shivaprasad G Bhat <sbhat@xxxxxxxxxxxxx>
An overal question: surely the same issue must arise on x86 with
file-backed NVDIMMs. How do they handle this case?
On x86 we have different ways nvdimm can be discovered. ACPI NFIT, e820 map
and virtio_pmem. Among these virio_pmem always operated with synchronous dax
disabled and both ACPI and e820 doesn't have the ability to differentiate
support for synchronous dax.
Ok. And for the virtio-pmem case, how are the extra flushes actually
done on x86?
virtio-pmem device has virtqueue with virtio_pmem_flush() as the handler
which gets called for all flush requests from guest. virtio_pmem_flush() is
offloading the flush to thread pool with a worker doing fsync() and the
completion callback notifying the guest with response.
With that I would expect users to use virtio_pmem when using using file
backed NVDIMMS
So... should we prevent advertising an NVDIMM through ACPI or e820 if
it doesn't have sync-dax enabled?
Is it possible to have different defaults for sync-dax based on
architecture ?
The behaviour on x86 is sync-dax=on for nvdimms. So, it would be correct to
have the default as "on" for x86. For pseries - "off" for new machines.
Looking at code, I didnt find much ways to achieve this. Can you suggest
what can be done ?
