ipoib_mcast_stop_thread currently tests mcast->query and if it is
NULL, does not perform wait_for_completion on the mcast and frees the
mcast object directly.

However, since both operations are done without locking, it is
possible that ipoib_mcast_join_complete is in progress on this mcast
object and has set mcast->query to NULL already.

Solve this by:
- taking priv->lock before we change mcast->query in ipoib_mcast_join_complete,
  and keeping it until we no longer need the mcast object
- taking priv->lock around mcast->query test in ipoib_mcast_stop_thread
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

If posting receives in ipoib_ib_dev_open() fails, call
ipoib_ib_dev_stop() to move the device's QP back to the RESET state so
that we can try again later.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

ipoib_mcast_send() tests mcast->ah twice.  If this value is changed
between these two points, we leak an skb.  However,
ipoib_mcast_join_finish() sets mcast->ah with no locking, so it could
race against ipoib_mcast_send().

As a solution, take priv->lock around assignment to mcast->ah thus
making sure ipoib_mcast_send() (which also takes priv->lock) is not in
flight.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Cosmetic change: make alignment explicit in to_ipoib_neigh.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Even after the last fix, it's still possible for a send-only join to
start before the join for the broadcast group has finished.  This
could cause us to create a multicast group using attributes from the
broadcast group that haven't been initialized yet, so we would use
garbage for the Q_Key, etc.  Fix this by waiting until the broadcast
group's attached flag is set before starting send-only joins.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Further, there's an additional issue that I saw in testing:
ipoib_mcast_send may get called when priv->broadcast is NULL (e.g. if
the device was downed and then upped internally because of a port
event).

If this happends and the send-only join request gets completed before
priv->broadcast is set, we get an oops.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix the following race scenario:
  - Device is up.
  - Port event or set mcast list triggers ipoib_mcast_stop_thread,
    this cancels the query and waits on mcast "done" completion.
  - Completion is called and "done" is set.
  - Meanwhile, ipoib_mcast_send arrives and starts a new query,
    re-initializing "done".

Fix this by adding a "multicast started" bit and checking it before
starting a send-only join.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Avoid corrupting mcast->pkt_queue by serializing access with
priv->tx_lock.  Also, update dropped packet statistics to count
multicast packets removed from pkt_queue as dropped.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The SA path record query completion can initialize path->pathrec.dlid
before IPoIB's callback runs and initializes path->ah, so we must test
ah rather than dlid.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

semaphore to mutex conversion by Ingo and Arjan's script.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
[ Sanity-checked on real IB hardware ]
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The current handling of multicast groups in IPoIB ends up never
freeing send-only multicast groups.  It turns out the logic was much
more complicated than it needed to be; we can fix this bug and
completely kill ipoib_mcast_dev_down() at the same time.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

dev->mc_list accesses must be protected by dev->xmit_lock.
Found by Eli Cohen <eli@mellanox.co.il>.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Multiple ipoib_neigh structures on mcast->neigh_list may point to the
same ah.  This means that ipoib_mcast_free() can't just make a list of
ah structs to free, since this might end up trying to add the same ah
to the list more than once.  Handle this in ipoib_multicast.c in the
same way as it is handled in ipoib_main.c for struct ipoib_path.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Don't leak memory on allocation failure for broadcast mcast group.
Also, print a warning to match handling for other mcast groups.
Signed-off-by: NEli Cohen <eli@mellanox.co.il>
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

To help in reducing the number of include dependencies, several files were
touched as they were getting needed headers indirectly for stuff they use.

Thanks also to Alan Menegotto for pointing out that net/dccp/proto.c had
linux/dccp.h include twice.
Signed-off-by: NArnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

If ipoib_ib_dev_up() fails after ipoib_ib_dev_open() is called, then
ipoib_ib_dev_stop() needs to be called to clean up.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

race condition: ipoib_ib_dev_flush is accessing child list without locks.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

ipoib_mcast_alloc() uses kzalloc(), so there's no need to zero out
members of the mcast struct after it's allocated.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Make sure mcast->done is initialized to uncompleted value before we
submit a new query, so that it's safe to wait on.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Always set path->query to NULL when the SA path record query
completes, rather than only when we don't have an address handle.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

It's possible that IPoIB will issue multiple SA queries for the same
path struct.  Therefore the struct's completion needs to be
initialized for each query rather than only once when the struct is
allocated, or else we might not wait long enough for later queries to
finish and free the path struct too soon.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

For cut-and-paste reasons, the IPoIB driver was setting skb->dev right
before calling dev_kfree_skb_any().  Get rid of this.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add ibX_path files to debugfs that contain information about the IPoIB
path cache.  IPoIB ARP only gives GIDs, which the IPoIB driver must
resolve to real IB paths through the ib_sa module.  For debugging,
when the ARP table looks OK but traffic isn't flowing, it's useful to
be able to see if the resolution from GID to path worked.

Also clean up the formatting of the existing _mcg debugfs files.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Don't build ipoib_mcast_iter_ functions if CONFIG_INFINIBAND_IPOIB_DEBUG
is not enabled -- their only callers will not be built either.

Also move the prototype for ipoib_open() to ipoib.h to fix a sparse warning.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Shrink our source and .text a little by removing a few assignments of
NULL and 0 to memory that is already cleared as part of the allocation.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Replace kmalloc()+memset(,0,) with kzalloc(), for a net savings of 35
source lines and about 500 bytes of text.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Minor cleanups: fix a misleading comment, and get rid of attr_mask
variables that are only used to hold constants (just use the constants
directly).
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Use spin_trylock_irqsave() in ipoib_start_xmit() instead of
reinventing it out of local_irq_save(), spin_trylock() and
local_irq_restore().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Change the way IPoIB handles RX packets when it can't allocate a new
receive skbuff.  If the allocation of a new receive skb fails, we now
drop the packet we just received and repost the original receive skb.
This means that the receive ring always stays full and we don't have
to monkey around with trying to schedule a refill task for later.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Use jiffies_to_msecs() so we print a human-readable time so
we don't have to worry about what HZ is configured to, and
print out a few values to make post-mortem analysis easier.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

ipoib_create_qp() no longer creates IPoIB's QP, so it shouldn't
destroy the QP on failure -- that unwinding happens elsewhere, so the
current code can cause a double free.  While we're at it, the
function's name should match what it actually does, so rename it to
ipoib_init_qp().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Rename IPoIB driver's path_lookup() to ipoib_path_lookup() to avoid a
clashes with the kernel global path_lookup().  We don't hit this with
the current kernel source, but some external patches seem to trigger
this, and it's cleaner to avoid clashing with global names anyway.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>
refs/heads/for-linus

ipoib_mcast_restart_task() is always called from within the
single-threaded IPoIB workqueue, so flushing the workqueue from within
the function can lead to a recursion overflow.  But since we're
running in a single-threaded workqueue, we're already synchronized
against other items in the workqueue, so just get rid of the flush in
ipoib_mcast_restart_task().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

We got a little mixed up with what the backoff member holds in the
IPoIB multicast group structure: sometimes it was used as a number of
seconds, and sometimes it was used as a number of jiffies.  Fix the
code so that backoff is always in seconds.
Signed-off-by: NHal Rosenstock <halr@voltaire.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Since ipoib uses queue_delayed_work to run flush task on port state events,
it must flush scheduled work after unregistering the event handler.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix IPoIB memory leak on device removal.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Move the InfiniBand headers from drivers/infiniband/include to include/rdma.
This allows InfiniBand-using code to live elsewhere, and lets us remove the
ugly EXTRA_CFLAGS include path from the InfiniBand Makefiles.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Currently we may have work scheduled in default kernel workqueue when
the device is going down.  The device could get freed before this
workqueue gets serviced.  I am actually seeing this causing system
hangs.

The following patch fixes this by using ipoib_workqueue which gets
flushed when the device is going down.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Always make sure that the full membership bit is set in the P_Keys
that IPoIB uses.  This makes sure that all hosts join the correct
multicast groups so that hosts that are partial partition members
can talk to the rest of the network.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

changing CONFIG_LOCALVERSION rebuilds too much, for no appearent reason.
Remove unneeded includes of <linux/version.h>.
Signed-off-by: NOlaf Hering <olh@suse.de>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>