- 14 Dec, 2016 40 commits
-
-
Doug Ledford authored
-
Adit Ranadive authored
This patch series adds a driver for a paravirtual RDMA device. The device is developed for VMware's Virtual Machines and allows existing RDMA applications to continue to use existing Verbs API when deployed in VMs on ESXi. We recently did a presentation in the OFA Workshop [1] regarding this device. Description and RDMA Support ============================ The virtual device is exposed as a dual function PCIe device. One part is a virtual network device (VMXNet3) which provides networking properties like MAC, IP addresses to the RDMA part of the device. The networking properties are used to register GIDs required by RDMA applications to communicate. These patches add support and the all required infrastructure for letting applications use such a device. We support the mandatory Verbs API as well as the base memory management extensions (Local Inv, Send with Inv and Fast Register Work Requests). We currently support both Reliable Connected and Unreliable Datagram QPs but do not support Shared Receive Queues (SRQs). Also, we support the following types of Work Requests: o Send/Receive (with or without Immediate Data) o RDMA Write (with or without Immediate Data) o RDMA Read o Local Invalidate o Send with Invalidate o Fast Register Work Requests This version only adds support for version 1 of RoCE. We will add RoCEv2 support in a future patch. We do support registration of both MAC-based and IP-based GIDs. I have also created a git tree for our user-level driver [2]. Testing ======= We have tested this internally for various types of Guest OS - Red Hat, Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12 using backported versions of this driver. The tests included several runs of the performance tests (included with OFED), Intel MPI PingPong benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough to test the backported version of the driver internally on their hardware using a VMware provided ESX build. I have also applied and tested this with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch series should be applied all together. I split out the commits so that it may be easier to review. PVRDMA Resources ================ [1] OFA Workshop Presentation - https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf [2] Libpvrdma User-level library - http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summaryReviewed-by:
Jorgen Hansen <jhansen@vmware.com> Reviewed-by:
George Zhang <georgezhang@vmware.com> Reviewed-by:
Aditya Sarwade <asarwade@vmware.com> Reviewed-by:
Bryan Tan <bryantan@vmware.com> Reviewed-by:
Leon Romanovsky <leonro@mellanox.com> Signed-off-by:
Adit Ranadive <aditr@vmware.com> Signed-off-by:
Doug Ledford <dledford@redhat.com>
-
Doug Ledford authored
-
Doug Ledford authored
-
Doug Ledford authored
-
Doug Ledford authored
-
Pan Bian authored
If uhw->inlen is non-zero, the value of variable err is 0 if the copy succeeds. Then, if kzalloc() or kmalloc() returns a NULL pointer, it will return 0 to the callers. As a result, the callers cannot detect the errors. This patch fixes the bug, assign "-ENOMEM" to err before the NULL pointer checks, and remove the initialization of err at the beginning. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=189031Signed-off-by:
Pan Bian <bianpan2016@163.com> Reviewed-by:
-
Pan Bian authored
In function ocrdma_mbx_create_ah_tbl(), returns the value of status on errors. However, because status is initialized with 0, 0 will be returned even if on error paths. This patch initialize status with "-ENOMEM". Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=188831Signed-off-by:
-
Zhouyi Zhou authored
Return value of skb_linearize should be handled in function nes_netdev_start_xmit. Compiled in x86_64 Signed-off-by:
Zhouyi Zhou <yizhouzhou@ict.ac.cn> Reviewed-by:
Yuval Shaia <yuval.shaia@oracle.com> Reviewed-by:
Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by:
-
Shiraz Saleem authored
Signed-off-by:
Shiraz Saleem <shiraz.saleem@intel.com> Signed-off-by:
Faisal Latif <faisal.latif@intel.com> Signed-off-by:
-
Leon Romanovsky authored
Remove Mitesh Ahuja <mitesh.ahuja@avagotech.com> from maintainers file. This email address seems not active and causes to mail bounces during submissions. Signed-off-by:
Leon Romanovsky <leon@kernel.org> Signed-off-by:
-
Sebastian Ott authored
__ib_umem_release calls dma_unmap_sg with a different number of sg_entries than ib_umem_get uses for dma_map_sg. This might cause trouble for implementations that merge sglist entries and results in the following dma debug complaint: DMA-API: device driver frees DMA sg list with different entry count [map count=2] [unmap count=1] Fix it by using the correct value. Signed-off-by:Sebastian Ott <sebott@linux.vnet.ibm.com> Signed-off-by:
-
Amrani, Ram authored
The recent introduction of qedr driver support in qede causes a GPF when probing the driver in a server without a RoCE enabled QLogic NIC. This fix avoids using an uninitialized pointer in such a case. Caught by the kernel test robot. Signed-off-by:
Ram Amrani <Ram.Amrani@cavium.com> Signed-off-by:
-
Souptick Joarder authored
In mthca_create_ah(), pci_pool_alloc() followed by memset will be replaced by pci_pool_zalloc() Signed-off-by:
Souptick joarder <jrdr.linux@gmail.com> Reviewed-by:
-
Bart Van Assche authored
Make it clear that qp->sq.wqe_cnt is not the number of WQEs. Signed-off-by:
Bart Van Assche <bart.vanassche@sandisk.com> Cc: Eli Cohen <eli@mellanox.com> Acked-by:
-
Bart Van Assche authored
This has been detected by building the mlx5 driver with W=1. Fixes: 1a412fb1 ('net/mlx5: Fixes: 1a412fb1 (IB/mlx5: Modify QP commands via mlx5 ifc') Signed-off-by:
-
Bart Van Assche authored
Detected by sparse. Signed-off-by:
-
Bart Van Assche authored
Avoid that shutdown of srp_daemon is delayed if add_target_mutex is held by another process. Signed-off-by:
-
Bart Van Assche authored
Make it easier to figure out what is going on if memory mapping fails because more memory regions than mr_per_cmd are needed. Signed-off-by:
-
Bart Van Assche authored
If login fails because memory region allocation failed it can be hard to figure out what happened. Make it easier to figure out why login failed by logging a message if ib_alloc_mr() fails. Signed-off-by:
-
Bart Van Assche authored
This patch makes the srp_add_one() code more compact and does not change any functionality. Signed-off-by:
-
Bart Van Assche authored
Avoid that the kernel build fails as follows if dynamic debug support is disabled: drivers/infiniband/ulp/srp/ib_srp.c:2272:3: error: implicit declaration of function 'DEFINE_DYNAMIC_DEBUG_METADATA' drivers/infiniband/ulp/srp/ib_srp.c:2272:33: error: 'ddm' undeclared (first use in this function) drivers/infiniband/ulp/srp/ib_srp.c:2275:39: error: '_DPRINTK_FLAGS_PRINT' undeclared (first use in this function) Signed-off-by:
-
Bart Van Assche authored
This patch avoids that Coverity complains about not checking the ib_find_pkey() return value. Fixes: commit 547af765 ("IB/multicast: Report errors on multicast groups if P_key changes") Signed-off-by:
-
Bart Van Assche authored
This patch avoids that Coverity reports the following: Using uninitialized value port_attr.state when calling printk Fixes: commit 94232d9c ("IPoIB: Start multicast join process only on active ports") Signed-off-by: -
Bart Van Assche authored
The array ib_mad_mgmt_class_table.method_table has MAX_MGMT_CLASS (80) elements. Hence compare the array index with that value instead of with IB_MGMT_MAX_METHODS (128). This patch avoids that Coverity reports the following: Overrunning array class->method_table of 80 8-byte elements at element index 127 (byte offset 1016) using index convert_mgmt_class(mad_hdr->mgmt_class) (which evaluates to 127). Fixes: commit b7ab0b19 ("IB/mad: Verify mgmt class in received MADs") Signed-off-by:
Hal Rosenstock <hal@mellanox.com> Signed-off-by:
-
Bart Van Assche authored
The special QP creation error path relies on offset_of(struct mlx4_ib_sqp, qp) == 0. Remove this assumption because that makes the QP creation code easier to understand. Signed-off-by:
Laurence Oberman <loberman@redhat.com> Reviewed-by:
Yishai Hadas <yishaih@mellanox.com> Signed-off-by:
-
Bart Van Assche authored
Report the following message only once if no ACL has been configured yet for an initiator port: "Rejected login because no ACL has been configured yet for initiator %s.\n" Signed-off-by:
Max Gurtovoy <maxg@mellanox.com> Signed-off-by:
-
Julia Lawall authored
The function usnic_ib_qp_grp_get_chunk only returns an ERR_PTR value or a valid pointer, never NULL. The same is true of get_qp_res_chunk, which just returns the result of calling usnic_ib_qp_grp_get_chunk. Simplify IS_ERR_OR_NULL to IS_ERR in both cases. The semantic patch that makes this change is as follows: (http://coccinelle.lip6.fr/) // <smpl> @@ expression t,e; @@ t = \(usnic_ib_qp_grp_get_chunk(...)\|get_qp_res_chunk(...)\) ... when != t=e - IS_ERR_OR_NULL(t) + IS_ERR(t) @@ expression t,e,e1; @@ t = \(usnic_ib_qp_grp_get_chunk(...)\|get_qp_res_chunk(...)\) ... when != t=e ?- t ? PTR_ERR(t) : e1 + PTR_ERR(t) ... when any // </smpl> Signed-off-by:
Julia Lawall <Julia.Lawall@lip6.fr> Reviewed-by:
-
Hans Westgaard Ry authored
from "InfiBand Architecture Specifications Volume 1": A QP is said to have a stale connection when only one side has connection information. A stale connection may result if the remote CM had dropped the connection and sent a DREQ but the DREQ was never received by the local CM. Alternatively the remote CM may have lost all record of past connections because its node crashed and rebooted, while the local CM did not become aware of the remote node's reboot and therefore did not clean up stale connections. and: A local CM may receive a REQ/REP for a stale connection. It shall abort the connection issuing REJ to the REQ/REP. It shall then issue DREQ with "DREQ:remote QPN” set to the remote QPN from the REQ/REP. This patch solves a problem with reuse of QPN. Current codebase, that is IPoIB, relies on a REAP-mechanism to do cleanup of the structures in CM. A problem with this is the timeconstants governing this mechanism; they are up to 768 seconds and the interface may look inresponsive in that period. Issuing a DREQ (and receiving a DREP) does the necessary cleanup and the interface comes up. Signed-off-by:
Hans Westgaard Ry <hans.westgaard.ry@oracle.com> Reviewed-by:
Håkon Bugge <haakon.bugge@oracle.com> Reviewed-by:
Sagi Grimberg <sagi@grimberg.me> Signed-off-by:
-
Philippe Reynes authored
The ethtool api {get|set}_settings is deprecated. We move this driver to new api {get|set}_link_ksettings. Signed-off-by:Philippe Reynes <tremyfr@gmail.com> Signed-off-by:
-
Alexey Khoroshilov authored
There are several places, where errors in dma_map_single() are ignored. The patch fixes them. Found by Linux Driver Verification project (linuxtesting.org). Signed-off-by:
Alexey Khoroshilov <khoroshilov@ispras.ru> Acked-by:
-
Jim Foraker authored
rvt_create_qp() creates qp->ip only when a qp creation request comes from userspace (udata is not NULL). If we exceed the number of available queue pairs however, the error path always attempts to put a kref to this structure. If the requestor is inside the kernel, this leads to a crash. We fix this by checking that qp->ip is not NULL before caling kref_put(). Signed-off-by:
Jim Foraker <foraker1@llnl.gov> Acked-by:
Dennis Dalessandro <dennis.dalessandro@intel.com> Acked-by:
Jonathan Toppins <jtoppins@redhat.com> Acked-by:
Alex Estrin <alex.estrin@intel.com> Signed-off-by:
-
Petr Mladek authored
Use the new API to create and destroy the cq kthread worker. The API hides some implementation details. In particular, kthread_create_worker() allocates and initializes struct kthread_worker. It runs the kthread the right way and stores task_struct into the worker structure. In addition, the *on_cpu() variant binds the kthread to the given cpu and the related memory node. kthread_destroy_worker() flushes all pending works, stops the kthread and frees the structure. This patch does not change the existing behavior. Note that we must use the on_cpu() variant because the function starts the kthread and it must bind it to the right CPU before waking. The numa node is associated for given CPU as well. Signed-off-by:
Petr Mladek <pmladek@suse.com> Signed-off-by:
-
Petr Mladek authored
The memory barrier is not enough to protect queuing works into a destroyed cq kthread. Just imagine the following situation: CPU1 CPU2 rvt_cq_enter() worker = cq->rdi->worker; rvt_cq_exit() rdi->worker = NULL; smp_wmb(); kthread_flush_worker(worker); kthread_stop(worker->task); kfree(worker); // nothing queued yet => // nothing flushed and // happily stopped and freed if (likely(worker)) { // true => read before CPU2 acted cq->notify = RVT_CQ_NONE; cq->triggered++; kthread_queue_work(worker, &cq->comptask); BANG: worker has been flushed/stopped/freed in the meantime. This patch solves this by protecting the critical sections by rdi->n_cqs_lock. It seems that this lock is not much contended and looks reasonable for this purpose. One catch is that rvt_cq_enter() might be called from IRQ context. Therefore we must always take the lock with IRQs disabled to avoid a possible deadlock. Signed-off-by: -
Arnd Bergmann authored
There is an old warning about mlx4_SW2HW_EQ_wrapper on x86: ethernet/mellanox/mlx4/resource_tracker.c: In function ‘mlx4_SW2HW_EQ_wrapper’: ethernet/mellanox/mlx4/resource_tracker.c:3071:10: error: ‘eq’ may be used uninitialized in this function [-Werror=maybe-uninitialized] The problem here is that gcc won't track the state of the variable across a spin_unlock. Moving the assignment out of the lock is safe here and avoids the warning. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Reviewed-by:
-
Arnd Bergmann authored
We get a false-positive warning in linux-next for the mlx5 driver: infiniband/hw/mlx5/mr.c: In function ‘mlx5_ib_reg_user_mr’: infiniband/hw/mlx5/mr.c:1172:5: error: ‘order’ may be used uninitialized in this function [-Werror=maybe-uninitialized] infiniband/hw/mlx5/mr.c:1161:6: note: ‘order’ was declared here infiniband/hw/mlx5/mr.c:1173:6: error: ‘ncont’ may be used uninitialized in this function [-Werror=maybe-uninitialized] infiniband/hw/mlx5/mr.c:1160:6: note: ‘ncont’ was declared here infiniband/hw/mlx5/mr.c:1173:6: error: ‘page_shift’ may be used uninitialized in this function [-Werror=maybe-uninitialized] infiniband/hw/mlx5/mr.c:1158:6: note: ‘page_shift’ was declared here infiniband/hw/mlx5/mr.c:1143:13: error: ‘npages’ may be used uninitialized in this function [-Werror=maybe-uninitialized] infiniband/hw/mlx5/mr.c:1159:6: note: ‘npages’ was declared here I had a trivial workaround for gcc-5 or higher, but that didn't work on gcc-4.9 unfortunately. The only way I found to avoid the warnings for gcc-4.9, short of initializing each of the arguments first was to change the calling conventions to separate the error code from the umem pointer. This avoids casting the error codes from one pointer to another incompatible pointer, and lets gcc figure out when that the data is actually valid whenever we return successfully. Acked-by:
-
Jason Gunthorpe authored
The kernel side is #ifdef'd to this type, and the UAPI header should use it directly. It has slightly different alignment requirments from the usual user space version. Signed-off-by:
Jason Gunthorpe <jgunthorpe@obsidianresearch.com> Reviewed-by:
-
Steve Wise authored
Acked-by:
Steve Wise <swise@opengridcomputing.com> Reviewed-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
Steve Wise authored
Acked-by:
-
Steve Wise authored
Acked-by:
Santosh Shilimkar <santosh.shilimkar@oracle.com> Signed-off-by:
-
