When a FMR is unmapped, mlx4 resets the map count to 0, and clears the
upper part of the R_Key which is used as the sequence counter.

This poses a problem for RDS, which uses ib_fmr_unmap as a fence
operation.  RDS assumes that after issuing an unmap, the old R_Keys
will be invalid for a "reasonable" period of time. For instance,
Oracle processes uses shared memory buffers allocated from a pool of
buffers.  When a process dies, we want to reclaim these buffers -- but
we must make sure there are no pending RDMA operations to/from those
buffers.  The only way to achieve that is by using unmap and sync the
TPT.

However, when the sequence count is reset on unmap, there is a high
likelihood that a new mapping will be given the same R_Key that was
issued a few milliseconds ago.

To prevent this, don't reset the sequence count when unmapping a FMR.
Signed-off-by: NOlaf Kirch <olaf.kirch@oracle.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

mlx4_table_find (for FMR MPTs) requires that ICM memory already be
mapped.  Before this fix, FMR allocation depended on ICM memory
already being mapped for the MPT entry.  If all currently mapped
entries are taken, the find operation fails (even if the MPT ICM table
still had more entries, which were just not mapped yet).

This fix moves the mpt find operation to fmr_enable, to guarantee that
any required ICM memory mapping has already occurred.

Found by Oren Duer of Mellanox.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Now that struct mlx4_buf.u is a struct instead of a union because of
the vmap() changes, there's no point in having a struct at all.  So
move .direct and .page_list directly into struct mlx4_buf and get rid
of a bunch of unnecessary ".u"s.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

    
Commit 3d73c288 ("mlx4_core: Fix section mismatches") fixed some of
the section mismatches introduced when error recovery was added, but
there were still more cases of errory recovery code calling into
__devinit code from regular .text.  Fix this by getting rid of the
now-incorrect __devinit annotations.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

    
Commit ee49bd93 ("mlx4_core: Reset device when internal error is
detected") introduced some section mismatch problems when
CONFIG_HOTPLUG=n, because the error recovery code tears down and
reinitializes the device after everything is loaded, which ends up
calling into lots of code marked __devinit and __devexit from regular
.text.  Fix this by getting rid of these now-incorrect section
markers.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Implement FMRs for mlx4.  This is an adaptation of code from mthca.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Write MTT entries directly to ICM from the driver (eliminating use of
WRITE_MTT command).  This reduces the number of FW commands needed to
register an MR by at least a factor of 2 and speeds up memory
registration significantly.  This code will also be used to implement
FMRs.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Taking ilog2(dev->caps.reserved_mtts) to find out the order to pass to
the MTT buddy allocator will do the wrong thing if reserved_mtts is ever
not a power of 2.  Be safe and use fls(dev->caps.reserved_mtts - 1).
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

mlx4_mr_alloc() doesn't actually allocate mr (it just initializes the
pointer that the caller passes in), so it shouldn't free it if an
error occurs.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

If a dMPT entry has the PA flag (direct physical address) set, then
the (unused) MTT base address field has to be set to 0.
Signed-off-by: NJack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add an InfiniBand driver for Mellanox ConnectX adapters.  Because
these adapters can also be used as ethernet NICs and Fibre Channel 
HBAs, the driver is split into two modules: 
 
  mlx4_core: Handles low-level things like device initialization and 
    processing firmware commands.  Also controls resource allocation 
    so that the InfiniBand, ethernet and FC functions can share a 
    device without stepping on each other. 
 
  mlx4_ib: Handles InfiniBand-specific things; plugs into the 
    InfiniBand midlayer. 
Signed-off-by: NRoland Dreier <rolandd@cisco.com>