Re: [PATCH] pmem: report error on clear poison failure

From: Dan Williams
Date: Thu Oct 13 2016 - 15:24:40 EST


On Thu, Oct 13, 2016 at 11:16 AM, Kani, Toshimitsu <toshi.kani@xxxxxxx> wrote:
> On Thu, 2016-10-13 at 10:22 -0700, Dan Williams wrote:
>> On Thu, Oct 13, 2016 at 9:08 AM, Kani, Toshimitsu <toshi.kani@xxxxxxx
>> > wrote:
>> >
>> > On Thu, 2016-10-13 at 09:01 -0700, Dan Williams wrote:
>> > >
>> > > On Thu, Oct 13, 2016 at 8:54 AM, Toshi Kani <toshi.kani@xxxxxxx>
>> > > wrote:
>> > > >
>> > > >
>> > > > ACPI Clear Uncorrectable Error DSM function may fail or may be
>> > > > unsupported on a platform. pmem_clear_poison() returns without
>> > > > clearing badblocks in such cases, which leads to a silent data
>> > > > corruption.
>> > > >
>> > > > Change pmem_do_bvec() and pmem_clear_poison() to return -EIO
>> > > > so that filesystem can log an error message.
>> > >
>> > > What's the silent data corruption scenario? If the clear poison
>> > > fails I'm assuming that the poison will still be notified on the
>> > > next
>> > > read.
>> >
>> > I agree that the data is eventually read, but there is no guranteed
>> > that when it is read soon enough, i.e. user might not access to the
>> > data for a long time.
>>
>> ...but that's the same behavior for errors that we don't yet know
>> about. That said, we indeed know that the write failed. I'd feel
>> better about this patch if the justification / impact was clearer in
>> the changelog, because "silent data corruption" is not the impact.
>
> Agreed. How about the following descritpion?
>
> ===
> ACPI Clear Uncorrectable Error DSM function may fail or may be
> unsupported on a platform. pmem_clear_poison() returns without
> clearing badblocks in such cases. This failure is detected at
> the next read (-EIO).
>
> This behavior can lead to an issue when user keeps writing but
> does not read immedicately. For instance, flight recorder file
> may be only read when it is necessary for troubleshooting.
>
> Change pmem_do_bvec() and pmem_clear_poison() to return -EIO
> so that filesystem can log an error message on a write error.
> ===

Looks good, thanks Toshi. I'll update the nvdimm.git branches after
-rc1 is out.