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>

Fix wqe_to_link() to use a structure field that we know is definitely
always unused for receive work requests, so that it really avoids the
free list corruption bug that the comment claims it does.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Userspace CQs that have no completion event channel attached end up
with their cq_context set to NULL.  However, asynchronous events like
"CQ overrun" can still occur on such CQs, so add a uverbs_file member
to struct ib_ucq_object that we can follow to deliver these events.
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>

Implement reporting asynchronous CQ events in Mellanox HCA driver.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix class_device_create() calls to match the new prototype which
takes a parent device pointer.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Move ib_uverbs module to using cdev_alloc() and class_device_create()
so that we can handle device lifetime properly.  Now we can make sure
we keep all of our data structures around until the last way to reach
them is gone.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Move ib_umad module to using cdev_alloc() and class_device_create() so
that we can handle device lifetime properly.  Now we can make sure we
keep all of our data structures around until the last way to reach
them is gone.
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>

The previous patch adding the ability to nest struct class_device
changed the paramaters to the call class_device_create().  This patch
fixes up all in-kernel users of the function.
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Simplify user_mad.c code in a few places, and convert from kmalloc() +
memset() to kzalloc().  This also fixes a theoretical race window by
not accessing packet->length after posting the send buffer (the send
could complete and packet could be freed before we get to the return
statement at the end of ib_umad_write()).
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The conversion of user_mad.c to the new MAD send API was slightly off:
in a few places, we used packet->msg instead of packet->msg->mad when
referring to the actual data buffer, which ended up corrupting the
underlying data structure and crashing when we free an invalid pointer.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add some initial support for detecting and reporting catastrophic
errors reported by Mellanox HCAs.  We start a periodic timer which
polls the catastrophic error reporting buffer in device memory.  If an
error is detected, we dump the contents of the buffer for port-mortem
debugging, and report a fatal asynchronous error to higher levels.

In the future we can try to recover from these errors by resetting the
device, but this will require some work in higher-level code as well.
Let's get this in now, so that we at least get catastrophic errors
reported in logs.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Change alloc_response_msg() in mad_rmpp.c to return the struct
it allocates directly (or an error code a la ERR_PTR), rather than
returning a status and passing the struct back in a pointer param.
This simplifies the code and gets rid of warnings like

    drivers/infiniband/core/mad_rmpp.c: In function nack_recv:
    drivers/infiniband/core/mad_rmpp.c:192: warning: msg may be used uninitialized in this function

with newer versions of gcc.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The UC transport does not support RDMA reads or atomic operations, so
we shouldn't require or even allow the consumer to set attributes
relating to these operations for UC QPs.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The MAD layer was violating the DMA API by touching data buffers used
for sends after the DMA mapping was done.  This causes problems on
non-cache-coherent architectures, because the device doing DMA won't
see updates to the payload buffers that exist only in the CPU cache.

Fix this by having all MAD consumers use ib_create_send_mad() to
allocate their send buffers, and moving the DMA mapping into the MAD
layer so it can be done just before calling send (and after any
modifications of the send buffer by the MAD layer).

Tested on a non-cache-coherent PowerPC 440SPe system.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix cm_init_qp_init_attr(), cm_init_qp_rtr_attr() and cm_init_qp_rts_attr()
so that they correctly handle the differences between UC and RC QPs.  This
fixes problems with setting up UC QPs through the CM.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add idr_destroy() calls to the module_exit() functions of the four IB
driver modules that use idrs, so we don't leak idr_layer_cache objects
when these modules are unloaded.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

We should always re-arm an event queue's interrupt in
mthca_tavor_interrupt() if the corresponding bit is set in the event cause
register (ECR), even if we didn't find any entries in the EQ.  If we don't,
then there's a window where we miss an EQ entry and then get stuck because
we don't get another EQ event.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

We should always re-arm an event queue's interrupt in
mthca_tavor_interrupt() if the corresponding bit is set in the event
cause register (ECR), even if we didn't find any entries in the EQ.
If we don't, then there's a window where we miss an EQ entry and then
get stuck because we don't get another EQ event.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Use devt member of struct class_device so that we don't have to create
our own "dev" file in sysfs.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add spaces after "sizeof" operator to match the rest of file.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add missing "break" in switch statement.  Without the break, the
CM ended up always falling through and setting every connection
request to use RC transport, which meant that UC connections
didn't work.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Avoid entering a QP as member of a multicast group multiple times.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

We keep IB device structures around until the last sysfs reference is
gone, but we shouldn't ask the low-level driver to do anything after
the LLD unregisters the device.  To handle this, check the reg_state
field and just fail sysfs show() requests if the device has already
been unregistered.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Set mthca_driver.owner to THIS_MODULE.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Make the type parameter of mthca_alloc_db() be an enum mthca_db_type
instead of an int.  This doesn't have any practical effect but
documents the functions a little better.
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>

Make ctx_id_mutex and ctx_id_table static to quiet sparse warnings.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Bind communication identifiers to a device to support device removal.
Export per HCA CM devices to userspace.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>

Eliminate ucm.h.  Replace ucm_dbg with direct call to printk KERN_ERR.
Signed-off-by: NSean Hefty <sean.hefty@intel.com>

Add kernel support for userspace calling poll CQ, request CQ
notification, post send, post receive, post SRQ receive, create AH and
destroy AH commands.  These commands allow us to support userspace
verbs for devices that can't perform these operations directly from
userspace (eg the PathScale HCA).
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Give each device a uverbs_cmd_mask, so that a low-level driver can
control which methods may be called on behalf of userspace.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

If the SA query module's initialization fails for a device, then that
device won't have a struct ib_sa_device associated.  We should fail SA
queries in that case, rather than blindly dereferencing the NULL
pointer we get back from ib_get_client_data().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

A couple of functions were missing spin_unlock calls in error paths.
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>

Check the sizes of CQs, QPs and SRQs when creating objects, and fail
instead of creating too-big queues.  Also return real limits instead
of just plausible-sounding values from mthca_query_device().
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

There is a bug in ib_mad_init_device(): if ib_agent_port_open() fails
for a given port, then the current code doesn't call ib_mad_port_close()
for that port.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The hardware relies on us keeping one extra work request that never
gets used in SRQs.  Add checks to the SRQ work request posting
functions so that they fail when someone is about to use up that extra
work request, rather than when someone uses the very last work request.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>