- 16 Feb, 2021 1 commit
-
-
Christoph Lameter authored
When a system receives a REREG event from the SM, then the SM information in the kernel is marked as invalid and a request is sent to the SM to update the information. The SM information is invalid in that time period. However, receiving a REREG also occurs simultaneously in user space applications that are now trying to rejoin the multicast groups. Some of those may be sendonly multicast groups which are then failing. If the SM information is invalid then ib_sa_sendonly_fullmem_support() returns false. That is wrong because it just means that we do not know yet if the potentially new SM supports sendonly joins. Sendonly join was introduced in 2015 and all the Subnet managers have supported it ever since. So there is no point in checking if a subnet manager supports it. Should an old opensm get a request for a sendonly join then the request will fail. The code that is removed here accomodated that situation and fell back to a full join. Falling back to a full join is problematic in itself. The reason to use the sendonly join was to reduce the traffic on the Infiniband fabric otherwise one could have just stayed with the regular join. So this patch may cause users of very old opensms to discover that lots of traffic needlessly crosses their IB fabrics. Link: https://lore.kernel.org/r/alpine.DEB.2.22.394.2101281845160.13303@www.lameter.comSigned-off-by:
Christoph Lameter <cl@linux.com> Signed-off-by:
Jason Gunthorpe <jgg@nvidia.com>
-
- 09 Feb, 2021 18 commits
-
-
Patrisious Haddad authored
Support 400Gbps IB rate in mlx5 driver. Link: https://lore.kernel.org/r/20210209130429.698237-1-leon@kernel.orgSigned-off-by:
Patrisious Haddad <phaddad@nvidia.com> Signed-off-by:
Mark Zhang <markzhang@nvidia.com> Reviewed-by:
Maor Gottlieb <maorg@nvidia.com> Signed-off-by:
Leon Romanovsky <leonro@nvidia.com> Signed-off-by:
-
Sebastian Andrzej Siewior authored
qedr_gsi_post_send() has a debug output which prints the return value of in_irq() and irqs_disabled(). The result of the in_irq(), even if invoked from an interrupt handler, is subject to change depending on the `threadirqs' command line switch. The result of irqs_disabled() is always be 1 because the function acquires spinlock_t with spin_lock_irqsave(). Remove in_irq() and irqs_disabled() from the debug output because it provides little value. Link: https://lore.kernel.org/r/20210208193347.383254-1-bigeasy@linutronix.deSigned-off-by:
Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by:
-
Bob Pearson authored
This patch changes the type of init_send_wqe in rxe_verbs.c to void since it always returns 0. It also separates out the code that copies inline data into the send wqe as copy_inline_data_to_wqe(). Link: https://lore.kernel.org/r/20210206002437.2756-1-rpearson@hpe.comSigned-off-by:
Bob Pearson <rpearson@hpe.com> Signed-off-by:
-
Bob Pearson authored
checkpatch -f found 3 warnings in RDMA/rxe 1. a missing space following switch 2. return followed by else 3. use of strlcpy() instead of strscpy(). This patch fixes each of these. In ... } elseif (...) { ... return 0; } else ... The middle block can be safely moved since it is completely independent of the other code. Link: https://lore.kernel.org/r/20210205230525.49068-1-rpearson@hpe.comSigned-off-by: -
Yishai Hadas authored
Cleanup the synchronize_srcu() from the ODP flow as it was found to be a very heavy time consumer as part of dereg_mr. For example de-registration of 10000 ODP MRs each with size of 2M hugepage took 19.6 sec comparing de-registration of same number of non ODP MRs that took 172 ms. The new locking scheme uses the wait_event() mechanism which follows the use count of the MR instead of using synchronize_srcu(). By that change, the time required for the above test took 95 ms which is even better than the non ODP flow. Once fully dropped the srcu usage, had to come with a lock to protect the XA access. As part of using the above mechanism we could also clean the num_deferred_work stuff and follow the use count instead. Link: https://lore.kernel.org/r/20210202071309.2057998-1-leon@kernel.orgSigned-off-by:
Yishai Hadas <yishaih@nvidia.com> Signed-off-by:
-
Xinhao Liu authored
There is no need to use a for loop to assign values for an array of cmd descriptors which has only two elements. Link: https://lore.kernel.org/r/1612517974-31867-13-git-send-email-liweihang@huawei.comSigned-off-by:
Xinhao Liu <liuxinhao5@hisilicon.com> Signed-off-by:
Weihang Li <liweihang@huawei.com> Signed-off-by:
-
Yixian Liu authored
The hns driver wrap around the consumer index of AEQ and CEQ when they reach to two times of queue entries number for owner mechanism, actually, it is unnecessary to wrap around since the hardware itself will mask it before use. Link: https://lore.kernel.org/r/1612517974-31867-12-git-send-email-liweihang@huawei.comSigned-off-by:
Yixian Liu <liuyixian@huawei.com> Signed-off-by:
-
Lang Cheng authored
All fields of WQE will be rewrote, so the memset is unnecessary. And when SQ is working in OWNER mode, the pipeline may prefetch the WQEs beyond PI, the memset operation may flip the owner bit too early, then the pipeline may get a wrong WQ. Link: https://lore.kernel.org/r/1612517974-31867-11-git-send-email-liweihang@huawei.comSigned-off-by:
Lang Cheng <chenglang@huawei.com> Signed-off-by:
-
Xinhao Liu authored
Use macros instead of magic numbers to represent shift of dma_handle_wqe, dma_handle_idx and UDP destination port number of RoCEv2. Link: https://lore.kernel.org/r/1612517974-31867-10-git-send-email-liweihang@huawei.comSigned-off-by:
-
Lang Cheng authored
hns_roce_device.h is not specific to hardware, some definitions are only used for HIP06, they should be moved into hns_roce_hw_v1.h. Link: https://lore.kernel.org/r/1612517974-31867-9-git-send-email-liweihang@huawei.comSigned-off-by:
-
Lang Cheng authored
Currently, the driver updates doorbell looks like this: post() { wqe.field = 0x111; wmb(); update_wq_db(); } update_wq_db() { db.field = 0x222; __raw_writeq(db, db_reg); } writeq() is a better choice than __raw_writeq() because it calls dma_wmb() to barrier in ARM64, and dma_wmb() is better than wmb() for ROCEE device. This patch removes all wmb() before updating doorbell of SQ/RQ/CQ/SRQ by replacing __raw_writeq() with writeq() to improve performence. The new process looks like this: post() { wqe.field = 0x111; update_wq_db(); } update_wq_db() { db.field = 0x222; writeq(db, db_reg); } Link: https://lore.kernel.org/r/1612517974-31867-8-git-send-email-liweihang@huawei.comSigned-off-by: -
Yixing Liu authored
Since HIP09 does not require this function, it should be masked. Link: https://lore.kernel.org/r/1612517974-31867-7-git-send-email-liweihang@huawei.comSigned-off-by:
Yixing Liu <liuyixing1@huawei.com> Signed-off-by:
-
Lijun Ou authored
This feature should only be enabled by querying capability from firmware. Fixes: ba6bb7e9 ("RDMA/hns: Add interfaces to get pf capabilities from firmware") Link: https://lore.kernel.org/r/1612517974-31867-5-git-send-email-liweihang@huawei.comSigned-off-by:
Lijun Ou <oulijun@huawei.com> Signed-off-by:
-
Xi Wang authored
Add the mapped page count checking flow to avoid invalid page size when creating MTR. Fixes: 38389eaa ("RDMA/hns: Add mtr support for mixed multihop addressing") Link: https://lore.kernel.org/r/1612517974-31867-4-git-send-email-liweihang@huawei.comSigned-off-by:
Xi Wang <wangxi11@huawei.com> Signed-off-by:
-
Weihang Li authored
This bit should be in type of enum ib_sig_type, or there will be a sparse warning. Fixes: bfe86035 ("RDMA/hns: Fix cast from or to restricted __le32 for driver") Link: https://lore.kernel.org/r/1612517974-31867-3-git-send-email-liweihang@huawei.comReported-by:
kernel test robot <lkp@intel.com> Signed-off-by:
-
Weihang Li authored
ULP usually set IB(V)_QP_AV when trying to modify QP to RTR if they want to record sgid index into QPC. For UD QPs, it is useless because it will be included in WQE. For RC QPs, it will be filled in hns_roce_set_path(). So sgid index shouldn't be filled by default. Then hns_get_gid_index() is moved to hns_roce_hw_v1.c because it is only called in it. Fixes: 926a01dc ("RDMA/hns: Add QP operations support for hip08 SoC") Link: https://lore.kernel.org/r/1612517974-31867-2-git-send-email-liweihang@huawei.comSigned-off-by:
-
Yixing Liu authored
Direct wqe is a mechanism to fill wqe directly into the hardware. In the case of light load, the wqe will be filled into pcie bar space of the hardware, this will reduce one memory access operation and therefore reduce the latency. Link: https://lore.kernel.org/r/1611997513-27107-1-git-send-email-liweihang@huawei.comSigned-off-by:
-
Kamal Heib authored
The size of tx_valid_cpus was calculated under the assumption that the numa nodes identifiers are continuous, which is not the case in all archs as this could lead to the following panic when trying to access an invalid tx_valid_cpus index, avoid the following panic by using nr_node_ids instead of num_online_nodes() to allocate the tx_valid_cpus size. Kernel attempted to read user page (8) - exploit attempt? (uid: 0) BUG: Kernel NULL pointer dereference on read at 0x00000008 Faulting instruction address: 0xc0080000081b4a90 Oops: Kernel access of bad area, sig: 11 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: siw(+) rfkill rpcrdma ib_isert iscsi_target_mod ib_iser libiscsi scsi_transport_iscsi ib_srpt target_core_mod ib_srp scsi_transport_srp ib_ipoib rdma_ucm sunrpc ib_umad rdma_cm ib_cm iw_cm i40iw ib_uverbs ib_core i40e ses enclosure scsi_transport_sas ipmi_powernv ibmpowernv at24 ofpart ipmi_devintf regmap_i2c ipmi_msghandler powernv_flash uio_pdrv_genirq uio mtd opal_prd zram ip_tables xfs libcrc32c sd_mod t10_pi ast i2c_algo_bit drm_vram_helper drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops cec drm_ttm_helper ttm drm vmx_crypto aacraid drm_panel_orientation_quirks dm_mod CPU: 40 PID: 3279 Comm: modprobe Tainted: G W X --------- --- 5.11.0-0.rc4.129.eln108.ppc64le #2 NIP: c0080000081b4a90 LR: c0080000081b4a2c CTR: c0000000007ce1c0 REGS: c000000027fa77b0 TRAP: 0300 Tainted: G W X --------- --- (5.11.0-0.rc4.129.eln108.ppc64le) MSR: 9000000002009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE> CR: 44224882 XER: 00000000 CFAR: c0000000007ce200 DAR: 0000000000000008 DSISR: 40000000 IRQMASK: 0 GPR00: c0080000081b4a2c c000000027fa7a50 c0080000081c3900 0000000000000040 GPR04: c000000002023080 c000000012e1c300 000020072ad70000 0000000000000001 GPR08: c000000001726068 0000000000000008 0000000000000008 c0080000081b5758 GPR12: c0000000007ce1c0 c0000007fffc3000 00000001590b1e40 0000000000000000 GPR16: 0000000000000000 0000000000000001 000000011ad68fc8 00007fffcc09c5c8 GPR20: 0000000000000008 0000000000000000 00000001590b2850 00000001590b1d30 GPR24: 0000000000043d68 000000011ad67a80 000000011ad67a80 0000000000100000 GPR28: c000000012e1c300 c0000000020271c8 0000000000000001 c0080000081bf608 NIP [c0080000081b4a90] siw_init_cpulist+0x194/0x214 [siw] LR [c0080000081b4a2c] siw_init_cpulist+0x130/0x214 [siw] Call Trace: [c000000027fa7a50] [c0080000081b4a2c] siw_init_cpulist+0x130/0x214 [siw] (unreliable) [c000000027fa7a90] [c0080000081b4e68] siw_init_module+0x40/0x2a0 [siw] [c000000027fa7b30] [c0000000000124f4] do_one_initcall+0x84/0x2e0 [c000000027fa7c00] [c000000000267ffc] do_init_module+0x7c/0x350 [c000000027fa7c90] [c00000000026a180] __do_sys_init_module+0x210/0x250 [c000000027fa7db0] [c0000000000387e4] system_call_exception+0x134/0x230 [c000000027fa7e10] [c00000000000d660] system_call_common+0xf0/0x27c Instruction dump: 40810044 3d420000 e8bf0000 e88a82d0 3d420000 e90a82c8 792a1f24 7cc4302a 7d2642aa 79291f24 7d25482a 7d295214 <7d4048a8> 7d4a3b78 7d4049ad 40c2fff4 Fixes: bdcf26bf ("rdma/siw: network and RDMA core interface") Link: https://lore.kernel.org/r/20210201112922.141085-1-kamalheib1@gmail.comSigned-off-by:
Kamal Heib <kamalheib1@gmail.com> Reviewed-by:
Bernard Metzler <bmt@zurich.ibm.com> Tested-by:
Yi Zhang <yi.zhang@redhat.com> Signed-off-by:
-
- 08 Feb, 2021 13 commits
-
-
Wenpeng Liang authored
The post_recv only supports QP types of RC, GSI and UD. Link: https://lore.kernel.org/r/1611997090-48820-13-git-send-email-liweihang@huawei.comSigned-off-by:
Wenpeng Liang <liangwenpeng@huawei.com> Signed-off-by:
-
Wenpeng Liang authored
The SRQ in the hns driver consists of the following four parts: * wqe buf: the buffer to store WQE. * wqe_idx buf: the cqe of SRQ may be not generated in the order of wqe, so the wqe_idx corresponding to the idle WQE needs to be pushed into the index queue which is a FIFO, then it instructs the hardware to obtain the corresponding WQE. * bitmap: bitmap is used to generate and release wqe_idx. When the user has a new WR, the driver finds the idx of the idle wqe in bitmap. When the CQE of wqe is generated, the driver will release the idx. * wr_id buf: wr_id buf is used to store the user's wr_id, then return it to the user when poll_cq verb is invoked. The process of post SRQ recv is refactored to make preceding code clearer. Link: https://lore.kernel.org/r/1611997090-48820-12-git-send-email-liweihang@huawei.comSigned-off-by:
-
Xi Wang authored
The HIP09 requires the driver to clear the unused data segments in wqe buffer to make the hns ROCEE stop reading the remaining invalid sges for RQ. Link: https://lore.kernel.org/r/1611997090-48820-11-git-send-email-liweihang@huawei.comSigned-off-by:
-
Xi Wang authored
Refactor post recv flow by removing unnecessary checking and removing duplicated code. Link: https://lore.kernel.org/r/1611997090-48820-10-git-send-email-liweihang@huawei.comSigned-off-by:
-
Lang Cheng authored
Use new register operation interfaces to simplify the process of write SRQ Context. Link: https://lore.kernel.org/r/1611997090-48820-9-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
Reduce parameter numbers of write_srqc() and move some related code into it from alloc_srqc(). Link: https://lore.kernel.org/r/1611997090-48820-8-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
Split the SRQ creation process into multiple steps and encapsulate them into functions. Link: https://lore.kernel.org/r/1611997090-48820-7-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
Each SRQs contain an reserved WQE, it is inappropriate and should be removed. Fixes: c7bcb134 ("RDMA/hns: Add SRQ support for hip08 kernel mode") Link: https://lore.kernel.org/r/1611997090-48820-6-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
When an error occurs, the qp_table must be cleared, regardless of whether the SRQ feature is enabled. Fixes: 5c1f167a ("RDMA/hns: Init SRQ table for hip08") Link: https://lore.kernel.org/r/1611997090-48820-5-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
According to the IB Specification, srq_limit shouldn't be configured during SRQ creation. If a user set srq_limit at this time, the driver should forced it to zero, or the result of creating SRQ will conflict with the result of querying SRQ. Fixes: c7bcb134 ("RDMA/hns: Add SRQ support for hip08 kernel mode") Link: https://lore.kernel.org/r/1611997090-48820-4-git-send-email-liweihang@huawei.comSigned-off-by:
-
Wenpeng Liang authored
If a user posts WR by wr_list, the head pointer of idx_queue won't be updated until all wqes are filled, so the judgment of whether head equals to tail will get a wrong result. Fix above issue and move the head and tail pointer from the srq structure into the idx_queue structure. After idx_queue is filled with wqe idx, the head pointer of it will increase. Fixes: c7bcb134 ("RDMA/hns: Add SRQ support for hip08 kernel mode") Link: https://lore.kernel.org/r/1611997090-48820-3-git-send-email-liweihang@huawei.comSigned-off-by:
-
Lang Cheng authored
The RQ/SRQ of HIP08 needs one special sge to stop receive reliably. So the driver needs to allocate at least one SGE when creating RQ/SRQ and ensure that at least one SGE is filled with the special value during post_recv. Besides, the kernel driver should only do this for kernel ULP. For userspace ULP, the userspace driver will allocate the reserved SGE in buffer, and the kernel driver just needs to pin the corresponding size of memory based on the userspace driver's requirements. Link: https://lore.kernel.org/r/1611997090-48820-2-git-send-email-liweihang@huawei.comSigned-off-by:
-
Bob Pearson authored
rxe_udp_encap_recv() drops the reference to rxe->ib_dev taken by rxe_get_dev_from_net() which should be held until each received skb is freed. This patch moves the calls to ib_device_put() to each place a received skb is freed. It also takes references to the ib_device for each cloned skb created to process received multicast packets. Fixes: 4c173f59 ("RDMA/rxe: Use ib_device_get_by_netdev() instead of open coding") Link: https://lore.kernel.org/r/20210128233318.2591-1-rpearson@hpe.comSigned-off-by:
-
- 05 Feb, 2021 8 commits
-
-
Bob Pearson authored
rxe_net.c sends packets at the IP layer with skb->data pointing at the IP header but receives packets from a UDP tunnel with skb->data pointing at the UDP header. On the loopback path this was not correctly accounted for. This patch corrects for this by using sbk_pull() to strip the IP header from the skb on received packets. Fixes: 8700e3e7 ("Soft RoCE driver") Link: https://lore.kernel.org/r/20210128182301.16859-1-rpearson@hpe.comSigned-off-by:
-
Bob Pearson authored
rxe_rcv_mcast_pkt() in rxe_recv.c can leak SKBs in error path code. The loop over the QPs attached to a multicast group creates new cloned SKBs for all but the last QP in the list and passes the SKB and its clones to rxe_rcv_pkt() for further processing. Any QPs that do not pass some checks are skipped. If the last QP in the list fails the tests the SKB is leaked. This patch checks if the SKB for the last QP was used and if not frees it. Also removes a redundant loop invariant assignment. Fixes: 8700e3e7 ("Soft RoCE driver") Fixes: 71abf20b ("RDMA/rxe: Handle skb_clone() failure in rxe_recv.c") Link: https://lore.kernel.org/r/20210128174752.16128-1-rpearson@hpe.comSigned-off-by:
-
Bob Pearson authored
In check_keys() in rxe_recv.c if ((...) && pkt->mask) { ... } always has pkt->mask non zero since in rxe_udp_encap_recv() pkt->mask is always set to RXE_GRH_MASK (!= 0). There is no obvious reason for this additional test and the original intent is lost. This patch simplifies the expression. Fixes: 8b7b59d0 ("IB/rxe: remove redudant qpn check") Link: https://lore.kernel.org/r/20210127224203.2812-1-rpearson@hpe.comSigned-off-by: -
Bob Pearson authored
check_type_state() in rxe_recv.c is written as if the type bits in the packet opcode were a bit mask which is not correct. This patch corrects this code to compare all 3 type bits to the required type. Fixes: 8700e3e7 ("Soft RoCE driver") Link: https://lore.kernel.org/r/20210127214500.3707-1-rpearson@hpe.comSigned-off-by:
-
Parav Pandit authored
Instead of open coding the loop for port iteration, use rdma_for_each_port macro provided by core. To use such macro, early initialization of phys_port_cnt is needed. Hence, initialize such constant early enough in the init stage. Whichever functions are called with port using rdma_for_each_port(), convert their port type from u8 to unsigned int to match the core API. Link: https://lore.kernel.org/r/20210203130133.4057329-6-leon@kernel.orgSigned-off-by:
Parav Pandit <parav@nvidia.com> Signed-off-by:
-
Parav Pandit authored
Currently mlx5 driver caches port GID table length for 2 ports. It is also cached by IB core as port immutable data. When mlx5 representor ports are present, which are usually more than 2, invalid access to port_caps array can happen while validating the GID table length which is only for 2 ports. To avoid this, take help of the IB cores port immutable data by exposing an API to read the port immutable fields. Remove mlx5 driver's internal cache, thereby reduce code and data. Link: https://lore.kernel.org/r/20210203130133.4057329-5-leon@kernel.orgSigned-off-by:
-
Parav Pandit authored
Improve query port functionality for representor port as below. 1. RoCE Qkey violation counters are not applicable for representor port. 2. Avoid setting gid_tbl_len twice for representor port. 3. Avoid setting ip_gids and IB_PORT_CM_SUP property for representor port as GID table is empty and CM support is not present in representor mode. Link: https://lore.kernel.org/r/20210203130133.4057329-4-leon@kernel.orgSigned-off-by:
-
Parav Pandit authored
Pkey table length for all the ports of the device is the same. Currently get_ports_cap() reads and stores it for each port by querying the device which reads more than just pkey table length. For representor device ports which can be in range of hundreds, it queries is for each such port and end up returning same value for all the ports. When in representor mode, modify QP accesses pkey port caps for a port index that can be outside of the port_caps table. Hence, simplify the logic to query the max pkey table length only once during device initialization sequence. Link: https://lore.kernel.org/r/20210203130133.4057329-3-leon@kernel.orgSigned-off-by:
-
