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

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>

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>

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>

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>

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>

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>

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>

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>

Our hardware supports generating an event when the number of receives
posted to a shared receive queue (SRQ) falls below a user-specified
limit.  Implement mthca_modify_srq() to arm the limit, and add code to
handle dispatching SRQ events when they occur.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add code to fill in the bad_pkey_cntr, max_mtu, active_mtu and
subnet_timeout fields in mthca_query_port().
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Add abi_version attribute to uverbs class devices to allow for
ABI versioning of device-specific interfaces.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Return correct atomic capability flag from mthca query function.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Remember to free the multicast group context memory table.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The IB spec defines the field to be 32 bits, not 16 bits.
Signed-off-by: NJack Morgenstein <jackm@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When allocating a table for mem-free HCA context, don't assume that
obj_size * nobj is an even multiple of MTHCA_TABLE_CHUNK_SIZE.  In
particular, make sure we allocate at least one slot even if the table
is smaller than MTHCA_TABLE_CHUNK_SIZE.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The loop in mthca_map_cmd() would fill one entry past the end of the
mailbox buffer before calling the firmware command.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

We should use the first word of the clear interrupt register if
the bit we're after is < 32, not < 31.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

If we allocate a bunch of doorbell records and then free them, we'll
end up with completely empty pages, which we then free.  However, when
we come back to allocate more doorbell pages, we have to reallocate
those empty pages rather than always trying to take a slot that we've
never used.  If we don't, we eventually use up every slot and fail to
allocate a doorbell record, even though we have plenty of free space.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Clean up QP table array on device removal.
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Userspace SRQs don't have a buffer allocated for them in the kernel, so
it doesn't make sense to set srq->last during initialization.  In fact,
this can crash trying to follow a nonexistent buffer pointer.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The error handling paths in mthca_tavor_post_srq_recv() and
mthca_arbel_post_srq_recv() are quite bogus, the result of a
screwed up merge.  Fix them so they work as intended.

Pointed out by Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

In mthca_create_eq(), we call get_eqe() before setting eq->nent.  This
is wrong, because get_eqe() uses eq->nent.  Fix this, and clean up the
code a little while we're at it.  (We got lucky with the current code,
because eq->nent was cleared to 0, which get_eqe() made happen to do
the right thing)

Pointed out by Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Fix posting first WQE for mem-free HCAs: we need to link to previous
WQE even in that case.  While we're at it, simplify code for
Tavor-mode HCAs.  We don't really need the conditional test there
either; we can similarly always link to the previous WQE.

Based on Michael S. Tsirkin's analogous fix for userspace libmthca.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

The hardware reads the ACK timeout field from the most significant 5
bits of struct mthca_qp_path's ackto field, not the least significant
bits.  This fix has the driver put the timeout in the right place.
Without this, we get a timeout that is 2^8 times too small.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

This patch removes CONFIG_PCI_NAMES.
Signed-off-by: NAdrian Bunk <bunk@stusta.de>
Signed-off-by: NGreg Kroah-Hartman <gregkh@suse.de>

Add missing call to init_waitqueue_head().
Signed-off-by: NMichael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When we modify a QP to the RESET state, completely clean up the QP
so that it is really and truly reset.
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>

Add mthca support for shared receive queues (SRQs),
including userspace SRQs.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When creating a table in context memory where the table is smaller
than our chunk size, we don't want to allocate and map a full chunk.
Instead, allocate just enough memory to cover the table.

This can be pretty simple because all tables are a power-of-2 size, so
either the table is a multiple of the chunk size, or it's smaller than
one chunk.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Move the definitions of the WQE structures from mthca_qp.c into
mthca_wqe.h, so that we'll be able to share them when we add the
SRQ code in mthca_srq.c.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Mem-free HCAs never generate error CQEs that complete multiple WQEs,
so just skip the call to mthca_free_err_wqe() for them rather than
having logic to handle the mem-free case in mthca_free_err_wqe().
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Clean up the allocation of memory for queues by factoring out the
common code into mthca_buf_alloc() and mthca_buf_free().  Now CQs and
QPs share the same queue allocation code, which we'll also use for SRQs.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

Set the max_msg_sz port property correctly in mthca's port_query
function.  Also zero out the attr struct so that we don't leave
any other members uninitialized.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>

When we call the INIT_IB firmware command to bring up a port, use
the actual port width capability returned by the QUERY_DEV_LIM
command instead of always trying to enable both 1X and 4X.  This
fixes breakage seen when the firmware is build to allow 4X only.
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>