On Wed, Oct 30, 2019 at 9:10 PM Aneesh Kumar K.V
<aneesh.kumar@linux.ibm.com> wrote:
[..]
> >
>
> Can you share the call stack? we clear bad blocks on write isn't? and we
> do only read while probing?
>
> We are still working on enabling `ndctl test` to run on POWER. Hence was
> not able capture these issues. Will try to get that resolved soon.

Here's the splat. I've also attached a reworked version of the patch
that has code organization fixups. It might be the case that it needs
to fully move to a model where each nd_*_probe() routine does its own
management of devm_namespace_enable() based on the sizes it needs to
complete the probe process.

nd_pmem namespace3.0: request out of range
WARNING: CPU: 16 PID: 1551 at
tools/testing/nvdimm/../../../drivers/nvdimm/claim.c:264
nsio_rw_bytes+0x14c/0x260 [libnvdimm]
RIP: 0010:nsio_rw_bytes+0x14c/0x260 [libnvdimm]
Call Trace:
 nd_pfn_validate+0x3d9/0x560 [libnvdimm]
 nd_pfn_probe+0xc0/0x150 [libnvdimm]
 ? nd_btt_probe+0x1e/0x270 [libnvdimm]
 nd_pmem_probe+0x63/0xc0 [nd_pmem]
 nvdimm_bus_probe+0x91/0x310 [libnvdimm]

That nd_pfn_validate+0x3d9 corresponds to the nvdimm_write_bytes() in
nd_pfn_clear_memmap_errors().