Re: [PATCH -next] mm/hotplug: skip bad PFNs from pfn_to_online_page()

[Dan Williams <dan.j.williams@intel.com>]

On Fri, Jun 14, 2019 at 10:09 AM Jeff Moyer <jmoyer@redhat.com> wrote:
>
> "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> writes:
>
> > On 6/14/19 10:06 PM, Dan Williams wrote:
> >> On Fri, Jun 14, 2019 at 9:26 AM Aneesh Kumar K.V
> >> <aneesh.kumar@linux.ibm.com> wrote:
> >
> >>> Why not let the arch
> >>> arch decide the SUBSECTION_SHIFT and default to one subsection per
> >>> section if arch is not enabled to work with subsection.
> >>
> >> Because that keeps the implementation from ever reaching a point where
> >> a namespace might be able to be moved from one arch to another. If we
> >> can squash these arch differences then we can have a common tool to
> >> initialize namespaces outside of the kernel. The one wrinkle is
> >> device-dax that wants to enforce the mapping size,
> >
> > The fsdax have a much bigger issue right? The file system block size
> > is the same as PAGE_SIZE and we can't make it portable across archs
> > that support different PAGE_SIZE?
>
> File system blocks are not tied to page size.  They can't be *bigger*
> than the page size currently, but they can be smaller.
>
> Still, I don't see that as an arugment against trying to make the
> namespaces work across architectures.  Consider a user who only has
> sector mode namespaces.  We'd like that to work if at all possible.

Even with fsdax namespaces I don't see the concern. Yes, DAX might be
disabled if the filesystem on the namespace has a block size that is
smaller than the current system PAGE_SIZE, but the filesystem will
still work. I.e. it's fine to put a 512 byte block size filesystem on
a system that has a 4K PAGE_SIZE, you only lose DAX operations, not
your data access.