On Fri, Dec 18, 2015 at 10:52 AM, Toshi Kani <toshi.kani(a)hpe.com> wrote:
On Fri, 2015-12-18 at 09:54 -0800, Dan Williams wrote:
> On Fri, Dec 18, 2015 at 7:15 AM, Toshi Kani <toshi.kani(a)hpe.com> wrote:
> > On Fri, 2015-12-18 at 01:34 -0700, Vishal Verma wrote:
> > > Also, I wonder if this problem is solved by using libndctl to manage
> > > BTTs.
> >
> > I have not tested with libndctl yet, but I think our bind/unbind
> > scripts do the same procedures.
>
> We loop through all combinations of sector size in our unit test. If
> you want to change the sector size the expectation is that the
> namespace is destroyed and fully re-created, especially due to the
> fact that changing sector size invalidates all data on the namespace.
>
> See:
https://github.com/pmem/ndctl/blob/master/lib/test-libndctl.c
The parent_uuid is not set on our NVDIMM-N systems. I do not see 'uuid'
file under sysfs 'namespaceX.X' per namespace_visible(), either. This
concept of creating/destroying a namespace is a bit foreign to us since
we've never needed to do. Can you elaborate how it's supposed to work for
NVDIMM-N?
Ugh, yes. An oversight on my part, let me give this some thought.
Whatever we decide, I want the libndctl api to be identical for the
two cases.
I expect the simplest option is to have
ndctl_namespace_disable_invalidate() destroy the btt-info block in the
NVDIMM-N case.