When both MSI-X and legacy INTx fail to generate an interrupt, the
driver frees the MSI-X interrupts twice. Fix this by clearing the
have_irq flag for the MSI-X interrupts when they are freed the first
time.
Reported-by: Yinghai Lu <yhlu.kernel@gmail.com>
Tested-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The ConnectX Programmer's Reference Manual states that the "SO" bit
must be set when posting Fast Register and Local Invalidate send work
requests. When this bit is set, the work request will be executed
only after all previous work requests on the send queue have been
executed. (If the bit is not set, Fast Register and Local Invalidate
WQEs may begin execution too early, which violates the defined
semantics for these operations)
This fixes the issue with NFS/RDMA reported in
<http://lists.openfabrics.org/pipermail/general/2009-April/059253.html>
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Cc: <stable@kernel.org>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
All the fields in the control block are nicely right-aligned, so no
masking is necessary.
Signed-off-by: Joachim Fenkes <fenkes@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
T3 firmware only supports one WRs worth of page list for fast register
work requests. The driver currently allows 2 WRs worth, which
doesn't work for T3, so reduce the limit in the driver.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The current MTT allocator uses kmalloc() to allocate a buffer for its
buddy allocator, and thus is limited in the amount of MTT segments
that it can control. As a result, the size of memory that can be
registered is limited too. This patch uses a module parameter to
control the number of MTT entries that each segment represents,
allowing more memory to be registered with the same number of
segments.
Signed-off-by: Eli Cohen <eli@mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
With a postfix increment, i is incremented one past 10K/5K before the
loop ends, so the error messages will be displayed too soon if the
test succeeds on the last iteration. Fix the comparisons to be >
instead of >=.
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The queue map for flush completion circumvention is only used for
kernel space queue pairs. This patch skips the allocation of the
queue maps in case the QP is created for userspace. In addition, this
patch does not iomap the galpas for kernel usage if the queue pair is
only used in userspace. These changes will improve the performance of
creation of userspace queue pairs.
Signed-off-by: Stefan Roscher <stefan.roscher@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
In case of large queue pairs there is the possibillity of allocation
failures due to memory fragmentation when using kmalloc(). To ensure
the memory is allocated even if kmalloc() can not find chunks which
are big enough, we fall back to allocating the memory with vmalloc().
Signed-off-by: Stefan Roscher <stefan.roscher@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
To improve performance of driver resource allocation, replace
vmalloc() calls with kmalloc().
Signed-off-by: Stefan Roscher <stefan.roscher@de.ibm.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
IB/mlx4: Don't overwrite fast registration page list when posting work request
RDMA/cxgb3: Don't complete flushed send work requests twice
The low-level mlx4 driver modified the page-list addresses for fast
register work requests post send to big-endian, and set a "present"
bit. This caused problems later when the consumer attempted to unmap
the pages using the page-list (using the list addresses which were
assumed to be still in CPU-endian order). Fix the mlx4 driver to
allocate two buffers and use a private buffer for the hardware-format
bus addresses.
This patch fixes <https://bugs.openfabrics.org/show_bug.cgi?id=1571>,
an NFS/RDMA server crash. The cause of the crash was found by Vu Pham
of Mellanox. The fix is along the lines suggested by Steve Wise in
comment #21 in bug 1571.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When the SQ is flushed, mark the flushed entries as not signaled so
the poll logic doesn't re-insert the CQ entry thinking its an out of
order completion.
The bug can cause the NFS/RDMA server to crash due to processing the
same completed work request twice.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
If reg_phys_mem() fails, we need to free memory allocated for MPA
frame with private data before returning the error. Also move
nes_add_ref() after the reg_phys_mem() is successful.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Running large cluster setup, we are hanging after many hours of
testing. Fixing this required going over the code and making sure the
rexmit entry was properly removed based on the cm_node's state and
packet received. Also when receiving a FIN packet, check seq# and
make sure there were no errors before calling handle_fin().
Following are the changes done in nes_cm.c:
* handle_ack_pkt() needs to return error value, so in case of error,
handle_fin() is not called. Some cleanup done while going over the code.
* handle_rst_pkt(), handling of cm_node's NES_CM_STATE_LAST_ACK is missing.
* process_packet(), in case of FIN only packet is received, call
check_seq() before processing.
* in handle_fin_pkt(), we are calling cleanup_retrans_entry() for all
conditions, even if the packets need to be dropped.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Under heavy load with large cluster testing, it may take longer to
receive a response to MPA requests. Change the driver to wait longer
after each rexmit to max time value.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
check_seq() was not checking if the seq#s have wrapped. Fix it.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
When a connect request comes, apbvt should only be set for
non-loopback connections.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Remove the NES_DEBUG that is causing the compile warning about an
unused variable when INFINIBAND_NES_DEBUG is not enabled.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
/sys/class/infiniband/nes?/fw_ver is not displaying firmware version
properly (it shows 0.0.0 with the current code). Fill in the correct
firmware version number.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
With updated PHY firmware for SFP_D, setting the trace length to 1
inch for SFP_D provides a more stable link.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Enable repause timer for port 1. Without this setting, under stress,
the chip may misbehave.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
In commit 1b949324 ("RDMA/nes: Fix SFP+ PHY initialization") there is
a mistake in the clean up code that removed port 1 CDR loop filter
settings for 10G cards other than CX4. Put the correct setting back
for appropriate PHY types.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Change thermo mitigation code to flip the SerDes1 reference clock to
internal, to match the change in commit a4849fc1 ("RDMA/nes: Add
wide_ppm_offset parm for switch compatibility").
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
In error paths where a CQ is not created, pbl is not freeed properly.
In nes_destroy_cq(), add the corresponding check for nescq->mcrqf to
not call nes_free_resource() when it is already done in nes_create_cq().
Signed-off-by: Miroslaw Walukiewicz <miroslaw.walukiewicz@intel.com>
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Commands INIT_HCA, CLOSE_HCA, SYS_EN, SYS_DIS, and CLOSE_IB all have 1
second timeouts. For INIT_HCA this causes problems when had more than
2^18 are QPs configured, since the command takes more than 1 second to
complete.
All other commands have 60-second timeouts. This patch makes the
above commands consistent with the rest of the commands (and with the
chip documentation).
This patch is an expansion of a patch from Arthur Kepner
<akepner@sgi.com> fixing just the INIT_HCA timeout.
Signed-off-by: Jack Morgenstein <jackm@dev.mellanox.co.il>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
QP attributes must stay initialized when moving back to IDLE. Zeroing
them will crash the system in _flush_qp() if the QP is subsequently
moved to ERROR and back to IDLE.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The code incorrectly failed memory registration if the buffer was not
page aligned. Also, the length field is mangled causing the hardware
to think the registration is much larger than it really is.
The fix is to remove the page alignment restriction as well the
incorrect length adjustment. Also make sure that all buffers after
the first start at a page boundary, and all buffers except the last
end on a page boundary.
Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Initialize pbl_count_256 to 0 to get rid of the warning:
drivers/infiniband/hw/nes/nes_verbs.c: In function 'nes_reg_mr':
drivers/infiniband/hw/nes/nes_verbs.c:1955: warning: 'pbl_count_256' may be used uninitialized in this function
Reported-by: Roland Dreier <rdreier@cisco.com>
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
NFS/RDMA currently fails to set up connections if peer2peer is on.
This is due to the fact that the NFS/RDMA client sets its ORD to 0.
If peer2peer is set, make sure the active side ORD is >= 1 and the
passive side IRD is >=1.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband:
RDMA/nes: Add support for new SFP+ PHY
RDMA/nes: Add wide_ppm_offset parm for switch compatibility
RDMA/nes: Fix SFP+ PHY initialization
RDMA/nes: Fix nes_nic_cm_xmit() error handling
RDMA/nes: Fix error handling issues
RDMA/nes: Fix incorrect casts on 32-bit architectures
IPoIB: Document newish features
RDMA/cma: Create cm id even when IB port is down
RDMA/cma: Use rate from IPoIB broadcast when joining IPoIB multicast groups
IPoIB: Avoid free_netdev() BUG when destroying a child interface
mlx4_core: Don't leak mailbox for SET_PORT on Ethernet ports
RDMA/cxgb3: Release dependent resources only when endpoint memory is freed.
RDMA/cxgb3: Handle EEH events
IB/mlx4: Use pgprot_writecombine() for BlueFlame pages
Add new register settings for new SFP+ PHY/firmware.
Add new PHY to to nes_netdev_get/set_settings.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
We have observed unstable link with a new BNT switch.
Add wide_ppm_offset parameter to allow the user to control the clock
ppm offset on the CX4 interface for better compatibility. Default is
100ppm, setting it to 1 will increase it to 300ppm. Change default
SerDes1 reference clock to external source.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
SFP+ PHY initialization has very long delays, incorrect settings for
direct attach copper cables, and inconsistent link detection.
Adjust delays to the minimum required by the PHY. Worst case is now
less than 4 seconds. Add new register settings for direct attach
cables. Change link detection logic to use two new registers for more
consistent link state detection. Reorganize code to shorten line
length.
Signed-off-by: Chien Tung <chien.tin.tung@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
We are getting crash or hung situation when we are running network
cable pull tests during RDMA traffic.
In schedule_nes_timer(), we return an error if nes_nic_cm_xmit()
returns failure. This is changed to success as skb is being put on
the timer routines to be processed later. In send_syn() case, we are
indicating connect failure once from nes_connect() and the other when
the rexmit retries expires.
The other issue is skb->users which we are incrementing before calling
nes_nic_cm_xmit() which calls dev_queue_xmit() but in case of failure
we are decrementing the skb->users at the same time putting the skb on
the rexmit path. Even if dev_queue_xmit() fails, the skb->users is
decremented already. We are removing the decrement of skb->users in
case of failure from both schedule_nes_timer() as well as from
nes_cm_timer_tick().
There is also extra check in nes_cm_timer_tick() for rexmit failure
which does a break from the loop is removed. This causes problem as
the other nodes have their cm_node->ref_count incremented and are not
processed.
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Fix issues found by static code analysis:
(1) Check if cm_node was successfully created for loopback connection.
(2) schedule_nes_timer() does not free up allocated memory after
encountering an error. There is a WARN_ON() for this condition.
(3) there is a cm_node->freed flag which is set but not used.
Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Faisal Latif <faisal.latif@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
The were some incorrect casts to unsigned long that caused 64-bit values
to be truncated on 32-bit architectures and made the driver pass invalid
adresses and lengths to the hardware. The problems were primarily seen
with kernels with highmem configured but some could show up in
non-highmem kernels, too.
Signed-off-by: Don Wood <donald.e.wood@intel.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Replace all DMA_32BIT_MASK macro with DMA_BIT_MASK(32)
Signed-off-by: Yang Hongyang<yanghy@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Replace all DMA_64BIT_MASK macro with DMA_BIT_MASK(64)
Signed-off-by: Yang Hongyang<yanghy@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The cxgb3 l2t entry, hwtid, and dst entry were being released before
all the iwch_ep references were released. This can cause a crash in
t3_l2t_send_slow() and other places where the l2t entry is used.
The fix is to defer releasing these resources until all endpoint
references are gone.
Details:
- move flags field to the iwch_ep_common struct.
- add a flag indicating resources are to be released.
- release resources at endpoint free time instead of close/abort time.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
- wrap calls into cxgb3 and fail them if we're in the middle
of a PCI EEH event.
- correctly unwind and release endpoint and other resources when
we are in an EEH event.
- dispatch IB_EVENT_DEVICE_FATAL event when cxgb3 notifies iw_cxgb3 of
a fatal error.
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
Roland Dreier