[openib-general] fmr question

[Steve Wise]

So this implies that there is really only one mapping outstanding at any
point in time (less the cache issue).  Right?  So why is there a map
count as an fmr attribute?  Its seems like just an arbitrary limit put
on how many times you can map an fmr before unmaping.  -and- once you
unmap you can start mapping again up to the map count... ??




On Thu, 2006-02-23 at 12:39 -0800, Roland Dreier wrote:
>     Steve> Lets say a FMR is allocated with a map count of 4.  And the
>     Steve> FMR is mapped 4 times, then unmapped.  Are the underlying
>     Steve> MRs still valid and setup for RDMA from the HW perspective?
>     Steve> Or are they marked "INVALID" as part of unmapping?  IE: Do
>     Steve> they remain valid up to unmap or up to dealloc of the FMR
> 
> They are not valid after the unmap operation.  In fact, old mappings
> are probably not valid after the FMR is remapped; the issue for
> Mellanox hardware is that the old mapping might still be hanging
> around in an internal cache somewhere.  However the unmap_fmr
> operation flushes all these caches, so all old mappings are gone after
> an FMR is unmapped.
> 
>  - R.