- 19 Feb, 2017 19 commits
-
-
Venkata Sandeep Dhanalakota authored
Reduce hfi1 code footprint by using the rdmavt timers. Reviewed-by:
Mike Marciniszyn <mike.marciniszyn@intel.com> Reviewed-by:
Brian Welty <brian.welty@intel.com> Signed-off-by:
Venkata Sandeep Dhanalakota <venkata.s.dhanalakota@intel.com> Signed-off-by:
Dennis Dalessandro <dennis.dalessandro@intel.com> Signed-off-by:
Doug Ledford <dledford@redhat.com>
-
Venkata Sandeep Dhanalakota authored
To move common code across target to rdmavt for code reuse. Reviewed-by:
-
Brian Welty authored
Add rvt_compute_aeth() and rvt_get_credit() as shared functions in rdmavt, moved from hfi1/qib logic. Reviewed-by:
-
Brian Welty authored
Add rvt_rc_error() and rvt_comm_est() as shared functions in rdmavt, moved from hfi1/qib logic. Reviewed-by:
-
Sebastian Sanchez authored
The EFI variable that provides board ID is named by the PCI address of the device, which is published in upper-case, while the HFI1 driver reads the EFI variable in lower-case. This prevents returning the correct board id when queried through sysfs. Read EFI variables in upper-case if the lower-case read fails. Reviewed-by:
Easwar Hariharan <easwar.hariharan@intel.com> Signed-off-by:
Sebastian Sanchez <sebastian.sanchez@intel.com> Signed-off-by:
-
Sebastian Sanchez authored
Some fields in struct hfi1_packet are oversized. Reduce them. Reviewed-by:
-
Mike Marciniszyn authored
The r_psn and s_rnr_retry are missing. Add with this patch. Reviewed-by:
Kaike Wan <kaike.wan@intel.com> Signed-off-by:
-
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:
-
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:
-
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
-
- 15 Feb, 2017 4 commits
-
-
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:
-
- 14 Feb, 2017 2 commits
-
-
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:
-
- 08 Feb, 2017 3 commits
-
-
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:
-
- 27 Jan, 2017 4 commits
-
-
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:
-
- 24 Jan, 2017 8 commits
-
-
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:
-
