This patch optimizes pio buffer allocation in the kernel.

For qib, kernel pio buffers are used for sending acks.  The code to
allocate the buffer would always start at 0 until it found a buffer.

This means that an average of 64 comparisions were done on each
allocate, since the busy bit won't be cleared until the bits are
refreshed when buffers are exhausted.

This patch adds two new fields in the devdata struct, last_pio and
min_kernel_pio.  last_pio is the last buffer that was allocated.
min_kernel_pio is the lowest potential available buffer.

min_kernel_pio is modifed as contexts are allocated and deallocted.
Reviewed-by: NRamkrishna Vepa <ramkrishna.vepa@intel.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Add a prefetch call when a packet has been stored.  The nature of the
prefetch is correctly determined by the alternatives mechanism.
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

If the call to mlx4_MAD_IFC() fails in ib_link_query_port() we will
currently do 'return err;' which will leak 'in_mad' and 'out_mad'.  We
should instead do 'goto out;' where we'll properly free the memory we
previously allocated.
Signed-off-by: NJesper Juhl <jj@chaosbits.net>
Acked-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

When the IB port is down, the active_speed value returned by the
MAD_IFC command is seven (7) which isn't among the defined IB speeds
in enum ib_port_speed, and this invalid speed value is passed up to
higher layers or applications who do port query.

Fix that by setting the speed to be SDR -- the lowest possible -- when
the port is down.
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Remove all #inclusions of asm/system.h preparatory to splitting and killing
it.  Performed with the following command:

perl -p -i -e 's!^#\s*include\s*<asm/system[.]h>.*\n!!' `grep -Irl '^#\s*include\s*<asm/system[.]h>' *`
Signed-off-by: NDavid Howells <dhowells@redhat.com>

To issue a port query, use the QUERY_(Ethernet)_PORT command instead
of the MAD_IFC command, since MAD_IFC attempts to query the firmware
IB SMA, which is irrelevant for IBoE ports.

This allows us to handle both 10Gb/s and 40Gb/s rates (e.g in sysfs),
using QDR speed (10Gb/s) and width of 1X or 4X.
Signed-off-by: NDotan Barak <dotanb@mellanox.com>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

If an erroneous CQE is polled in the first iteration (i.e. npolled ==
0), we don't update the consumer index and hence the hardware could
get a wrong notion of how many CQEs software polled.  Fix this by
unconditionally updating the doorbell record.  We could change the
check to be something like

	if (npolled || err != -EAGAIN)
		...

but it does not seem worth the effort since a posted write to memory
should not cost too much.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Use a bit in wc_flags rather then a whole integer to hold the
"checksum OK" flag.  By itself, this change doesn't reduce the size of
struct ib_wc on 64bit machines -- it stays on 56 bytes because of
padding.  However, it will allow to add more fields in the future
without enlarging the struct.  Also, it will let us have a unified
approach with future libibverbs checksum offload reporting, because a
bit flag doesn't break the library ABI.

This patch was suggested during conversation with Liran Liss
<liranl@mellanox.com>.
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: NSean Hefty <sean.hefty@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Since flush_qp() is always called with irqs disabled, all the locking
inside flush_qp() and __flush_qp() doesn't need irq save/restore.

Further, passing the flag variable from iwch_modify_qp() is just wrong
and causes a WARN_ON() in local_bh_enable().
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

While doing the work for commit a6f7feae ("IB/mlx4: pass SMP
vendor-specific attribute MADs to firmware") we realized that the
firmware would respond on all sorts of vendor-specific MADs.
Therefore commit 97285b78 ("mlx4_core: Add extended port
capabilities support") adds redundant code into the driver, since
there's no real reaon to maintain the extended capabilities of the
port, as they can be queried on demand (e.g the FDR10 capability).

This patch reverts commit 97285b78 and removes the check for
extended caps from the mlx4_ib driver port query flow.
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

I'm getting compile failures building this driver, which I narrowed
down to the ilog2 call in ehca_get_max_hwpage_size...

    ERROR: ".____ilog2_NaN" [drivers/infiniband/hw/ehca/ib_ehca.ko]
    undefined!
    make[1]: *** [__modpost] Error 1
    make: *** [modules] Error 2

The use of shca->hca_cap_mr_pgsize is confusing the compiler, and
resulting in the __builtin_constant_p in ilog2 going insane.

I tried making it take the u32 pgsize as an argument and the expansion
of shca->_pgsize in the caller, but that failed as well.

With this patch in place, the driver compiles on my GCC 4.6.2 here.
Suggested-by: NRoland Dreier <roland@purestorage.com>
Signed-off-by: NKyle McMartin <kmcmarti@redhat.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The kernel IB stack uses one enumeration for IB speed, which wasn't
explicitly specified in the verbs header file.  Add that enum, and use
it all over the code.

The IB speed/width notation is also used by iWARP and IBoE HW drivers,
which use the convention of rate = speed * width to advertise their
port link rate.
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

ConnectX devices have a limit on the number of mappings that can be
done on an FMR before having to call sync_tpt.  The current
mlx4_ib driver reports the limit correctly in max_map_per_fmr in
.query_device(), but mlx4_core doesn't check it when actually
allocating FMRs.

Add a max_fmr_maps field to struct mlx4_caps and enforce this maximum
value on FMR allocations.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

If the opcode of a work request exceeds the range of valid opcodes,
return the pointer to the offending work request.
Signed-off-by: NEli Cohen <eli@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Since create_comp_task() creates percpu kthread, it makes sense to use
kthread_create_on_node() to get proper NUMA affinity for kthread
stack.
Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com>
Acked-by: NHoang-Nam Nguyen <hnguyen@de.ibm.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The current get portinfo handling filters the LID being sent,
changing zero to 0xffff.

This causes OpenSM to log excessive warning messages.
Reviewed-by: NEdward Mascarenhas <edward.mascarenhas@qlogic.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Call irq_set_affinity_hint() to give userspace programs such as
irqbalance the information to be able to distribute qib interrupts
appropriately.

The logic allocates all non-receive interrupts to the first CPU local
to the HCA.  Receive interrupts are allocated round robin starting
with the second CPU local to the HCA with potential wrap back to the
second CPU.

This patch also adds a refinement to the name registered for MSI-X
interrupts so that user level scripts can determine the device
associated with the IRQs when there are multiple HCAs with a
potentially different set of local CPUs.
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Fix endianness problems detect by sparse, introduced with the enhanced
MPA patch.
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Don't worry about p2p_type if peer2peer itself is not requested in the
first place.
Signed-off-by: NKumar Sanghvi <kumaras@chelsio.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Update copyright information in the source files.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

In the current code, vendor-specific MADs (e.g with the FDR-10
attribute) are silently dropped by the driver, resulting in timeouts
at the sending side and inability to query/configure the relevant
feature.  However, the ConnectX firmware is able to handle such MADs.
For unsupported attributes, the firmware returns a GET_RESPONSE MAD
containing an error status.

For example, for a FDR-10 node with LID 11:

    # ibstat mlx4_0 1

    CA: 'mlx4_0'
    Port 1:
    State: Active
    Physical state: LinkUp
    Rate: 40 (FDR10)
    Base lid: 11
    LMC: 0
    SM lid: 24
    Capability mask: 0x02514868
    Port GUID: 0x0002c903002e65d1
    Link layer: InfiniBand

Extended Port Query (EPI) vendor mad timeouts before the patch:

    # smpquery MEPI 11 -d

    ibwarn: [4196] smp_query_via: attr 0xff90 mod 0x0 route Lid 11
    ibwarn: [4196] _do_madrpc: retry 1 (timeout 1000 ms)
    ibwarn: [4196] _do_madrpc: retry 2 (timeout 1000 ms)
    ibwarn: [4196] _do_madrpc: timeout after 3 retries, 3000 ms
    ibwarn: [4196] mad_rpc: _do_madrpc failed; dport (Lid 11)
    smpquery: iberror: [pid 4196] main: failed: operation EPI: ext port info query failed

EPI query works OK with the patch:

    # smpquery MEPI 11 -d

    ibwarn: [6548] smp_query_via: attr 0xff90 mod 0x0 route Lid 11
    ibwarn: [6548] mad_rpc: data offs 64 sz 64
    mad data
    0000 0000 0000 0001 0000 0001 0000 0001
    0000 0000 0000 0000 0000 0000 0000 0000
    0000 0000 0000 0000 0000 0000 0000 0000
    0000 0000 0000 0000 0000 0000 0000 0000
    # Ext Port info: Lid 11 port 0
    StateChangeEnable:...............0x00
    LinkSpeedSupported:..............0x01
    LinkSpeedEnabled:................0x01
    LinkSpeedActive:.................0x01
Signed-off-by: NJack Morgenstein <jackm@mellanox.com>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Acked-by: NIra Weiny <weiny2@llnl.gov>
Cc: <stable@vger.kernel.org>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Fix fast memory registration opcode in local invalidate completion.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NDonald Wood <Donald.E.Wood@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Zero high order word of fast memory registration (FMR) length field.
FMR length field is 32 bits, so high word should always be zero.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NDonald Wood <Donald.E.Wood@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Commit 8d4548f2 ("IB/qib: Default some module parameters optimally")
introduced an issue with older root complexes.  They cannot handle the
pcie_caps of 0x51 (MaxReadReq 4096, MaxPayload=256).

A typical diagnostic in this situation reported by syslog contains
the text:

  [PCIe Poisoned TLP][Send DMA memory read]

Restore the module paramter default to zero with will avoid any
changes in the root complex.
Reviewed-by: NMark Debbage <mark.debbage@qlogic.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

alloc_dummy_hdrq() is called with locks held and thus should not use
GFP_KERNEL.

The semantic patch that makes this report is available in
scripts/coccinelle/locks/call_kern.cocci.
Signed-off-by: NJulia Lawall <julia.lawall@lip6.fr>
Acked-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Make sure all exit paths from this function unlock everything.
Reported-by: NBart Van Assche <bvanassche@acm.org>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Set a reject flag, when sending MPA reject message to inform the peer
that the application has rejected the connection.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

"dentry" is a valid pointer.  "*dentry" was intended.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Acked-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Now we must provide the IP destination address, and a reference has
to be dropped when we're done with the entry.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Now we must provide the IP destination address, and a reference has
to be dropped when we're done with the entry.
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
Signed-off-by: NSteve Wise <swise@opengridcomputing.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

module_param(bool) used to counter-intuitively take an int.  In
fddd5201 (mid-2009) we allowed bool or int/unsigned int using a messy
trick.

It's time to remove the int/unsigned int option.  For this version
it'll simply give a warning, but it'll break next kernel version.
Acked-by: NMauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

For historical reasons, we allow module_param(bool) to take an int (or
an unsigned int).  That's going away.

A few drivers really want an int: they set it to -1 and a parameter
will set it to 0 or 1.  This sucks: reading them from sysfs will give
'Y' for both -1 and 1, but if we change it to an int, then the users
might be broken (if they did "param" instead of "param=1").

Use a new 'bint' parser for them.

(ntfs has a different problem: it needs an int for debug_msgs because
it's also exposed via sysctl.)

Cc: Steve Glendinning <steve.glendinning@smsc.com>
Cc: Jean Delvare <khali@linux-fr.org>
Cc: Guenter Roeck <guenter.roeck@ericsson.com>
Cc: Hoang-Nam Nguyen <hnguyen@de.ibm.com>
Cc: Christoph Raisch <raisch@de.ibm.com>
Cc: Roland Dreier <roland@kernel.org>
Cc: Sean Hefty <sean.hefty@intel.com>
Cc: Hal Rosenstock <hal.rosenstock@gmail.com>
Cc: linux390@de.ibm.com
Cc: Anton Altaparmakov <anton@tuxera.com>
Cc: Jaroslav Kysela <perex@perex.cz>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: lm-sensors@lm-sensors.org
Cc: linux-rdma@vger.kernel.org
Cc: linux-s390@vger.kernel.org
Cc: linux-ntfs-dev@lists.sourceforge.net
Cc: alsa-devel@alsa-project.org
Acked-by: Takashi Iwai <tiwai@suse.de> (For the sound part)
Acked-by: Guenter Roeck <guenter.roeck@ericsson.com> (For the hwmon driver)
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Fix for reset which happens right after sending a terminate message.
Terminate timer is not deleted when the connection is closed.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Warned about by sparse.
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Change the PHY clock divisor to make the MDIO clock 2.5MHz, instead of
3.5MHz (which is out of spec).
Signed-off-by: NTatyana Nikolova <Tatyana.E.Nikolova@intel.com>
Signed-off-by: NFaisal Latif <Faisal.Latif@intel.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

For IBoE, SLs 0-7 are mapped to Ethernet 802.1Q user priority bits
(pbits) which are part of the VLAN tag, SLs 8-15 are reserved.

Under Ethernet, the ConnectX firmware treats (decode/encode) the four
bit SL field in various constructs such as QPC / UD WQE / CQE as PPP0
and not as 0PPP. This correlates well to the fact that within the
vlan tag the pbits are located in bits 15-13 and not 12-14.

The current code wasn't consistent around that area - the
encoding was correct for the IBoE QPC.path.schedule_queue field,
but was wrong for IBoE CQEs and when MLX header was built.

These inconsistencies resulted in wrong SL <--> wire 802.1Q pbits
mapping, which is fixed by using SL <--> PPP0 all around the place.
Signed-off-by: NOren Duer <oren@mellanox.co.il>
Signed-off-by: NOr Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Minimize the need for users to have to set module parameters to get
good performance.

The following two parameters are changed:
 - rcvhdrcnt to twice the rcvegrcnt
 - pcie_caps=0x51

The rcvhdrcnt at twice the egrcount allows the preemptive NAK code
during reception to function in 100% of the cases rather than a sender
jiffies-based timeout.

The pcie_caps default of 0x51 will set the proposed MaxPayload and
MaxReceiveReqest to 256 and 4096 respectively.  The capabilities on
the root complex will be used to limit those values.
Reviewed-by: NRam Vepa <ram.vepa@qlogic.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The current code locks the QP s_lock, followed by the pending_lock, I
guess to to protect against the allocate failing.

This patch only locks the pending_lock, assuming that the empty case
is an exeception, in which case the pending_lock is dropped, and the
original code is executed.  This will save a lock of s_lock in the
normal case.

The observation is that the sdma descriptors will deplete at twice the
rate of txreq's, so this should be rare.
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

Prevent a receive data corruption by ensuring that the write to update
the rcvhdrheadn register to generate an interrupt is at the very end
of the receive processing.
Signed-off-by: NRamkrishna Vepa <ram.vepa@qlogic.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Cc: <stable@kernel.org>
Signed-off-by: NRoland Dreier <roland@purestorage.com>

The qib driver makes use of the the 64-bit jiffies API.

Code inspection reveals that that version of the API is not really
required.  This patch converts to use the "normal" jiffies.
Reviewed-by: NRam Vepa <ram.vepa@qlogic.com>
Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com>
Signed-off-by: NRoland Dreier <roland@purestorage.com>