[ofa-general] Re: [GIT PULL] 2.6.22: please pull ~mst/linux-2.6/.git

[Michael S. Tsirkin]

> Quoting Roland Dreier <rdreier at cisco.com>:
> Subject: Re: [GIT PULL] 2.6.22: please pull ~mst/linux-2.6/.git
> 
> thanks...
> 
>  > Michael S. Tsirkin (3):
>  >       IB/mthca: fix posting >255 recv WRs for Tavor
>  >       ipoib/cm: optimize stale connection detection
> 
> I applied this one.
> 
>  >       IB/mthca: fix RESET to ERROR transition
> 
> I will read this over more carefully -- it seems to be a rather big
> patch that adds constification various places etc.

Thanks.

Some explanations:
The only reason for const change is because there's a table of attribute structs that's
always the same so I decided it's nice to have it a global const,
and the change rippled over.

The rest is just splitting up the modify command so that
on RESET->ERROR I can perform 2 commands without code duplication.

>  > Yosef Etigin (2):
>  >       IB/core: add helpers for uncached gid/pkey queries
>  >       IB/ipoib: handle pkey re-shuffling
> 
> I need to catch up on the discussion that I did not read while I was
> traveling last week, so I'll hold these two as well.

Here's a summary:

The last time we all agreed that long term we want to get rid of ib_cache, which
will solve all kind of coherency issues.  So the ipoib is a minimal patch to do
this wrt to pkey, fixing the bug Voltaire is seeing with their partitioning
setup.

The core patch just adds helpers for this bit, but since query_port
can't be chached by provider (it gives e.g. physical port state),
it seemed worth the while to query table lengths at startup only,
rather than have each ib_find_pkey call re-do this.

Yosef also has more patches cooking to remove the rest of the cache
usage and speed up query_gid/query_pkey in providers, and also
clean the pkey polling thread in ipoib, but it seemed like a
good idea to have the bugfix out first.

-- 
MST