Re: [ndctl PATCH 2/2] ndctl/namespace: introduce ndctl_namespace_is_configurable()

["Verma, Vishal L" <vishal.l.verma@intel.com>]

On Mon, 2019-11-04 at 08:35 -0800, Dan Williams wrote:
> On Fri, Nov 1, 2019 at 1:27 PM Vishal Verma <vishal.l.verma@intel.com> wrote:
> > The motivation for this change is that we want to refrain from
> > (re)configuring what appear to be partially configured namespaces.
> > Namespaces may end up in a state that looks partially configured when
> > the kernel isn't able to enable a namespace due to mismatched
> > PAGE_SIZE expectations. In such cases, ndctl should not treat those
> > as unconfigured 'seed' namespaces, and reuse them.
> > 
> > Add a new ndctl_namespace_is_configurable API, whish tests whether a
> > namespaces is active, and whether it is partially configured. If neither
> > are true, then it can be used for (re)configuration. Additionally, deal
> > with the corner case of ND_DEVICE_NAMESPACE_IO (legacy PMEM) namespaces
> > by testing whether the size attribute is read-only (which indicates such
> > a namespace). Legacy namespaces always appear as configured, since their
> > size cannot be changed, but they are also always re-configurable.
> > 
> > Use this API in namespace_reconfig() and namespace_create() to enable
> > this partial configuration detection.
> 
> A couple comments.... I think it's probably sufficient to just check
> for ND_DEVICE_NAMESPACE_IO as I don't anticipate we'll ever have more
> than one namespace type with a read-only size attribute.

Yep I did notice I could do that, but I decided to go to the source of
it instead of adding another ND_DEVICE_NAMESPACE_IO assumption. Also I
had already written sysfs_attr_writable() before I noticed that :)
But there are already assumptions around ND_DEVICE_NAMESPACE_IO, so
adding another one here is fine.

> Also, how about s/ndctl_namespace_is_configurable/ndctl_namespace_is_configuration_idle/?
> Because to me all namespaces are always "configurable", but some may
> have active non-default properties set.

Sounds reasonable, but how about ndctl_namespace_has_partial_config(),
which I think describes the situation better, and makes it
straightforward for a potential future --really-force (or -ff) option
where we might still want to blow away anything on this namespace and
reclaim it.
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org