Re: dmcache RAID1 bug?

[Mike Snitzer <snitzer@redhat.com>]

On Wed, Nov 19 2014 at  4:39pm -0500,
Leonardo Santos <heiligerstein@gmail.com> wrote:

> After run regression tests using 'git bisect', I identify that the error
> was created after commit 8c081b52c6833a30a69ea3bdcef316eccc740c87
> 
> To reproduce this error just:
> - create a raid1 HDDs for origin device;
> - create a cache device with SSDs (could or not be raid)
> - create a metadata device with SSDs (could or not be raid)
> - create a cache device using prior devices
> - kernel crashes.
> 
> I attached the bitseclog and diff file!

Surprising considering I went over that commit multiple times (because
the changes weren't obvious relative to the "No functional change."
assertion in the commit's header).  The __cache_map and cache_map change
to centralize the inc_ds + cell_defer made the __cache_map changes
_seem_ like a difference was being introduced but they still look fine
to me.

Looking again I notice check_for_quiesced_migrations() switched to
an unconditional dm_deferred_entry_dec(); but that is fine because
earlier in check_for_quiesced_migrations() there is an early return if
pb->all_io_entry is NULL.

But could be a more trained eye will see some functional change that may
have been introduced (cc'ing Joe).

Mike