do the reasonable thing and enable multi-frag only on non-HIGHMEM platforms,
where there's no kmap-deadlock risk.

*   Made openib not use the subnet manager to discover connection parameters

    + Openib listens on a TCP/IP port for service queries and responds with the
      service id, port GID and pkey.

    + Openib peer table entries have become (NID, IP, port) tuples, where IP
      and port connect to the peer's service query server.

    + /proc interfaces for port, listener timeout and backlog (restarts kernel
      listener on update).

    + lmc/lconf support for new openib peers, including only running acceptor
      if net == 'tcp'

    + Changed connection daemon to conduct both sides of the service query as
      well as IB connection establishment.  Spawning several since tcp/ip reads
      can block for a timeout.

    + Added a reaper thread to do connection cleanup and timeout checks (the
      single connection daemon used to do that too).

*   Removed some unused lconf default constants

*   Better openib automagic compilation check.

*   Consistent usage (flipping + optional checksum) of all openib messages
    (service query, connection requests and "normal" message flow).

*   Exhaustive openib "are-you-still-the-same-person" checks

*   Fixed bug which stopped idle persistent peers from getting removed from the
    peer table.

*   Fixed some ranal bugs which became obvious when similar problems were
    debugged in openibnal

    and infinicon NALs

r=adilger

Remove full paths of files in debug logs.

*   better checking/error messages on OpenIB automagic

    *  OBD_FL_DEBUG_CHECK for consistent obdecho end-to-end checking

    *  struct echo_obd cleanup

    *  obdecho support for "difficult" reply testing (test_setattr)

    *  replaced cowardly eap_from_cookie() usage with EAP_FROM_COOKIE()
       which asserts it is being called correctly

    *  bracketed echo_client_iocontrol() guts with unlock/lock_kernel()
       for increased concurrency during echo ioctl execution

    *  better thread control and stats in obd.c

    *  n-threads-over-m-objects test_brw support (required for
       obdfilter-survey)

* openibnal changes

    *  re-enabled openib on 2.5/2.6 builds
       configure --with-openib=no can be used to disable openib explicitly if
       required.

    *  fixed some 64 bit casts

    *  just use kernel's page_to_phys()

* fixed CERROR format problem in libcfs/module.c

    kernel comms API.

error if ever undefined).

- keepalives enabled; starting after 30 seconds, and sending up to
  10 keepalives, every 2 seconds
- enable multi-frag I/O to improve IP-over-IB performance, with the
  vanishingly small risk of kmap-induced deadlock

*   changed device choice to ensure all used everywhere

- remove ancient Sandia copyright statements from Portals
  (our tree has been rewritten so many times that it's not even
   recognizably the same software)
- Remove the LLNL copyright from the qswnal (for the same reason)
- Remove our copyright from header files which only implement
  structures and prototypes, which are widely recognized as
  non-copyrightable

*   Removed S_???NAL in favour of a single S_NAL

*   NB liblustre needs attention

- don't generate a temp filename for a raw dump
- don't unlink non-regular files (e.g. /dev/null)
- don't report an error if kernel doesn't dump anything (fix mismerge?)

    This could crash a qswnal client engaged in bulk READS with a crashing server

- added ldiskfs module info generation to portals/utils/debug.c
- added packing/unpacking for mea in lmv_obd.c
- small cleanups.

r=adilger
Update printing of pid in portals_id2str to print in correct place with correct
limit. Also updated format string to match comments elsewhere.