- Feb 19, 2017
-
-
Sebastian Sanchez authored
There are some memory allocation calls in hfi1_create_ctxtdata() that do not use the numa function parameter. This can cause cache lines to be filled over QPI. Reviewed-by:
Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by:
Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by:
Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by:
Doug Ledford <dledford@redhat.com>
-
Sebastian Sanchez authored
Having per-CPU reference count for each MR prevents cache-line bouncing across the system. Thus, it prevents bottlenecks. Use per-CPU reference counts per MR. The per-CPU reference count for FMRs is used in atomic mode to allow accurate testing of the busy state. Other MR types run in per-CPU mode MR until they're freed. Reviewed-by:
-
Sebastian Sanchez authored
Receive code paths use the QP's device and port number to access the struct hfi1_ibport. When an instance of struct hfi1_ctxtdata is present, it can be used to access struct hfi1_ibport through a pointer. This makes struct hfi1_ibport lookup time faster as an array doesn't have to be indexed and access fields in other cache-lines. Reviewed-by:
-
Mike Marciniszyn authored
The resource specific wait locking missed correcting the lock for the notify_error_qp() calldown. The code is fixed to correctly use the iowait lock field to protect the head that is protected by that lock. Fixes: Commit 4e045572 ("IB/hfi1: Add unique txwait_lock for txreq events") Reviewed-by:
-
Easwar Hariharan authored
After extended testing, it was found that the previous PCIe Gen 3 recipe, which used adaptive CTLE with Preset 4, could cause an NMI/Surprise Link Down in about 1 in 100 to 1 in 1000 power cycles on some platforms. New EV data combined with extensive empirical data indicates that the new recipe should use static CTLE with Preset 6 for all integrated silicon SKUs. Fixes: c3f8de0b ("IB/hfi1: Add static PCIe Gen3 CTLE tuning") Reviewed-by:
Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by:
-
Mike Marciniszyn authored
The read of s_head in the hfi1_make_rc_req() and qib_make_rc_req() lack the necesary barrier instuctions. Correct other ACCESS_ONCE() warnings in the same file. Reviewed-by:
Ashutosh Dixit <ashutosh.dixit@intel.com> Signed-off-by:
-
Mike Marciniszyn authored
In the event that the IRQ thread is extremely busy, the processing of an rcd wait list can be delayed by quite a bit until the IRQ thread completes its work. The QP reset reference count wait can then appear to be stuck, thus causing up a QP destroy to emit the hung task diagnostic. Fix by processing the qp wait list periodically from the thread. The interval is a multiple (currently 4) of the MAX_PKT_RECV. Also, reduce some of the excessive inlining. The guidelines are per packet is ok inline, otherwise the choice is based on likelyhood of execution. Reviewed-by:
-
Mike Marciniszyn authored
The qp_wait_list processing leaves the defered ack count at its prior value. This can result in a premature send of an ack. Fixed by unconditionally reseting the defered ack count in hfi1_send_rc_ack(). Fixes: Commit 7c091e5c ("staging/rdma/hfi1: add ACK coalescing logic") Signed-off-by:
-
Wei Yongjun authored
Use setup_timer function instead of initializing timer with the function and data fields. Signed-off-by:
Wei Yongjun <weiyongjun1@huawei.com> Reviewed-by:
Leon Romanovsky <leonro@mellanox.com> Signed-off-by:
-
Max Gurtovoy authored
As iser connections can share completion contexts, we need to protect the active_qps update each time we set it. Signed-off-by:
Max Gurtovoy <maxg@mellanox.com> Acked-by:
Sagi Grimberg <sagi@grimberg.me> Signed-off-by:
-
Ganesh Goudar authored
Clean up send_connect() and make use of t6 specific active open request struct. Acked-by:
Steve Wise <swise@opengridcomputing.com> Signed-off-by:
Bharat Teja <bharat@chelsio.com> Signed-off-by:
Ganesh Goudar <ganeshgr@chelsio.com> Signed-off-by:
-
Doug Ledford authored
-
- Feb 15, 2017
-
-
Moni Shoua authored
The destination address in a listening rdma_id does not have an address family. Since address family in both sides of a connection must be the same in rdma_bind_addr() we set the address family of the destination to the address family of the source. This patch serves the logic in cma_port_is_unique() which requires to know if destination address that is associated with a rdma_id is any address (cma_zero_addr() and cma_loopback_addr()). This can happen when port reuse is checked for a port number that is being listened to. Fixes: 19b752a1 ("IB/cma: Allow port reuse for rdma_id") Signed-off-by:
Moni Shoua <monis@mellanox.com> Signed-off-by:
Leon Romanovsky <leon@kernel.org> Signed-off-by:
-
Majd Dibbiny authored
Add new entry to the RDMA-CM configfs that allows users to select default TOS for RDMA-CM QPs. This is useful for users that want to control the TOS for legacy applications without changing their code. Application that sets the TOS explicitly using the rdma_set_option API will continue to work as expected, meaning overriding the configfs value. CC: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
Majd Dibbiny <majd@mellanox.com> Reviewed-by:
-
Parav Pandit authored
This patch avoids unnecessary type casting from void to net_device. CC: Yuval Shaia <yuval.shaia@oracle.com> Signed-off-by:
Parav Pandit <parav@mellanox.com> Signed-off-by:
Yuval Shaia <yuval.shaia@oracle.com> Signed-off-by:
-
Erez Shitrit authored
When sending packet to destination that was not resolved yet via path query, the driver keeps the skb and tries to re-send it again when the path is resolved. But when re-sending via dev_queue_xmit the kernel doesn't call to dev_hard_header, so IPoIB needs to keep 20 bytes in the skb and to put the destination address inside them. In that way the dev_start_xmit will have the correct destination, and the driver won't take the destination from the skb->data, while nothing exists there, which causes to packet be be dropped. The test flow is: 1. Run the SM on remote node, 2. Restart the driver. 4. Ping some destination, 3. Observe that first ICMP request will be dropped. Fixes: fc791b63 ("IB/ipoib: move back IB LL address into the hard header") Cc: <stable@vger.kernel.org> # v4.8+ Signed-off-by:
Erez Shitrit <erezsh@mellanox.com> Signed-off-by:
Noa Osherovich <noaos@mellanox.com> Signed-off-by:
-
- Feb 14, 2017
-
-
Selvin Xavier authored
Makefile and Kconfig changes for enabling bnxt_re compilation Signed-off-by:
Devesh Sharma <devesh.sharma@broadcom.com> Signed-off-by:
Somnath Kotur <somnath.kotur@broadcom.com> Signed-off-by:
Sriharsha Basavapatna <sriharsha.basavapatna@broadcom.com> Signed-off-by:
Selvin Xavier <selvin.xavier@broadcom.com> Reviewed-by:
-
Selvin Xavier authored
This patch introduces the RoCE driver for the Broadcom NetXtreme-E 10/25/40/50G RoCE HCAs. The RoCE driver is a two part driver that relies on the parent bnxt_en NIC driver to operate. The changes needed in the bnxt_en driver have already been incorporated via Dave Miller's net tree into the mainline kernel. The vendor official git repository for this driver is available on github as: https://github.com/Broadcom/linux-rdma-nxt/ Signed-off-by:
Eddie Wai <eddie.wai@broadcom.com> Signed-off-by:
-
- Feb 09, 2017
-
-
Leon Romanovsky authored
Remove references to private kernel header and defines from exported ib_user_verb.h file. The code snippet below is used to reproduce the issue: #include <stdio.h> #include <rdma/ib_user_verb.h> int main(void) { printf("IB_USER_VERBS_ABI_VERSION = %d\n", IB_USER_VERBS_ABI_VERSION); return 0; } It fails during compilation phase with an error: ➜ /tmp gcc main.c main.c:2:31: fatal error: rdma/ib_user_verb.h: No such file or directory #include <rdma/ib_user_verb.h> ^ compilation terminated. Fixes: 189aba99 ("IB/uverbs: Extend modify_qp and support packet pacing") CC: Bodong Wang <bodong@mellanox.com> CC: Matan Barak <matanb@mellanox.com> CC: Christoph Hellwig <hch@infradead.org> Tested-by:Slava Shwartsman <slavash@mellanox.com> Signed-off-by:
-
Eyal Itkin authored
Update the range check to avoid integer-overflow in edge case. Resolves CVE 2016-8636. Signed-off-by:
Eyal Itkin <eyal.itkin@gmail.com> Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Reviewed-by:
-
Eyal Itkin authored
Update the response's resid field when larger than MTU, instead of only updating the local resid variable. Fixes: 8700e3e7 ("Soft RoCE driver") Signed-off-by:
-
- Jan 28, 2017
-
-
Yuval Shaia authored
Signed-off-by:
-
Bart Van Assche authored
Port and ACL information must be configured before an initiator logs in. Make it possible to configure this information before a subnet prefix has been assigned to a port by not only accepting GIDs as target port and initiator port names but by also accepting port GUIDs. Add a 'priv' member to struct se_wwn to allow target drivers to associate their own data with struct se_wwn. Reported-by:
Bart Van Assche <bart.vanassche@sandisk.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Nicholas Bellinger <nab@linux-iscsi.org> Signed-off-by:
-
Christophe Jaillet authored
This test looks reverted. We should log an error message only if 'ib_attach_mcast()' fails. Signed-off-by:
Christophe JAILLET <christophe.jaillet@wanadoo.fr> Signed-off-by:
-
Jack Morgenstein authored
If IPV6 has not been enabled in the underlying kernel, we must avoid calling IPV6 procedures in rdma_cm.ko. This requires using "IS_ENABLED(CONFIG_IPV6)" in "if" statements surrounding any code which calls external IPV6 procedures. In the instance fixed here, procedure cma_bind_addr() called ipv6_addr_type() -- which resulted in calling external procedure __ipv6_addr_type(). Fixes: 6c26a771 ("RDMA/cma: fix IPv6 address resolution") Cc: <stable@vger.kernel.org> # v4.2+ Cc: Spencer Baugh <sbaugh@catern.com> Signed-off-by:
Jack Morgenstein <jackm@dev.mellanox.co.il> Reviewed-by:
-
- Jan 25, 2017
-
-
Zhu Yanjun authored
The function ipoib_mcast_start_thread/ipoib_ib_dev_up always return zero. As such, in the function ipoib_open, err_stop will never be reached. So remove this err_stop and change the return type of the function ipoib_mcast_start_thread/ipoib_ib_dev_up to void. Signed-off-by:
Zhu Yanjun <yanjun.zhu@oracle.com> Reviewed-by:
-
Shiraz Saleem authored
Set maj_err and min_err in i40iw_sc_cqp_create so that it returns correct values for all return cases. This also addresses an uninitialized variable warning for maj_err and min_err in i40iw_create_cqp. Reported-by:
Shiraz Saleem <shiraz.saleem@intel.com> Reviewed-by:
-
Leon Romanovsky authored
Commit f0626710 ("RDMA: Update workqueue usage") removed content of qib_qsfp_deinit(...) and left it empty. This patch deletes all leftovers of that function. Signed-off-by:
-
Jack Wang authored
As Jason suggested, we have 4 elements for per port arrays, it's better to have a separate structure to represent them. It simplifies code a bit, ~ 30 lines of code less :) Signed-off-by:
Jack Wang <jinpu.wang@profitbricks.com> Reviewed-by:
Michael Wang <yun.wang@profitbricks.com> Signed-off-by:
-
Zhu Yanjun authored
The ipoib_ib_dev_down/ipoib_ib_dev_stop return zero unconditionally and the callers never check the returned values, change the return type to void and remove the redundant return values. Reviewed-by:
Shan Hai <shan.hai@oracle.com> Signed-off-by:
-
Dan Carpenter authored
The debug printk was indented more than it should have been and we can remove an unnecessary line break. Signed-off-by:
-
Moni Shoua authored
When allocating a port number for binding to a rdma_id, assuming the allocation is not for a specific port, the rule is to allow only ports that were not in use before by any other rdma_id. This condition is too strong to achieve the goal of a unique 5 tuple rdma_id. Instead, we can compare current rdma_id with other rdma_id for difference in one of destination port, source address and destination address to allow port reuse. Signed-off-by:
Sean Hefty <sean.hefty@intel.com> Signed-off-by:
-
Moni Shoua authored
Print debug messages to the kernel log to add more information about RDMA_CM events that indicate an error. Signed-off-by:
-
Zhu Yanjun authored
In the function ipoib_set_dev_features, the returned value is always 0. As such, it is not necessary to check the returned value. This is not a bug. It is a trivial problem. Reviewed-by:
Guanglei Li <guanglei.li@oracle.com> Signed-off-by:
-
Colin Ian King authored
Trivial fix to spelling mistake in isert_warn message Signed-off-by:
Colin Ian King <colin.king@canonical.com> Reviewed-by:
-
Yonatan Cohen authored
On UD QP completer tasklet is scheduled for each packet sent. If it is followed by a destroy_qp(), the kernel panic will happen as the completer tries to operate on a destroyed QP. Fixes: 8700e3e7 ("Soft RoCE driver") Signed-off-by:
Yonatan Cohen <yonatanc@mellanox.com> Reviewed-by:
-
Maor Gottlieb authored
The first argument of list_add_tail is the new item and the second is the head of the list. Fix the code to pass arguments in the right order, otherwise not all the rxe devices will be removed during teardown. Fixes: 8700e3e7 ('Soft RoCE driver') Signed-off-by:
Maor Gottlieb <maorg@mellanox.com> Reviewed-by:
-
Kenneth Lee authored
1. Release pid before enter odp flow 2. Release pid when fail to allocate memory Fixes: 87773dd5 ("IB: ib_umem_release() should decrement mm->pinned_vm from ib_umem_get") Fixes: 8ada2c1c ("IB/core: Add support for on demand paging regions") Signed-off-by:
Kenneth Lee <liguozhu@hisilicon.com> Reviewed-by:
Haggai Eran <haggaie@mellanox.com> Reviewed-by:
-
Ram Amrani authored
Relying on qede to trigger qedr on startup is problematic. When probing both if qedr loads slowly then qede can assume qedr is missing and not trigger it. This patch adds a triggering from qedr and protects against a race via an atomic bit. Signed-off-by:
Ram Amrani <Ram.Amrani@cavium.com> Signed-off-by:
Ariel Elior <Ariel.Elior@cavium.com> Signed-off-by:
-
Ram Amrani authored
Free the PD if no internal resources were available. Move userspace code under the relevant 'if'. Signed-off-by:
-
