On Mon, Nov 26, 2018 at 7:12 AM Dan Williams <dan.j.williams(a)intel.com> wrote:
On Mon, Nov 19, 2018 at 12:11 AM Oliver O'Halloran <oohall(a)gmail.com> wrote:
> When creating an fsdax or devdax namespace we need to verify that the
> seed namespaces exist. This patch reworks the validation so that it's
> done earlier to simplify the subsequent patches in the series.
> No functional changes.
It does appear to have a functional change. do_setup_pfn() supports
the case of statically allocated namespaces in NDCTL_NS_MODE_MEMORY
Hmm, ok. Up until now I had assumed that as far as ndctl was concerned
NS_MODE_MEMORY was synonymous with a pfn namespace.
This is what one gets by default with "legacy" pmem defined
E820-type-12 memory. In that case the kernel assumes that the
resulting memmap is always small enough to be allocated from DRAM and
there is no need to use a dynamic pfn device. So, if I'm not
mistaken, the deletion of do_setup_pfn() loses that special case
From what I see, the main difference is that ndctl would fail
validation when fsdax mode is specified and there's no pfn namespace
support in the kernel. I agree that's not great, but I'm not sure what
we should be doing here. The current behaviour will silently ignore -a
if "-m fsdax -M mem" is specified in the reconfigure case, which
doesn't seem great either.