Hi Doug, others, 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 64278fe). 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=summary --- Changes v4->v5: - PATCH [02/16] - Moved pvrdma_uapi.h and pvrdma_user.h into common UAPI folder. - Renamed to pvrdma-uapi.h and pvrdma-abi.h respectively. - Prefixed unsigned vars with __. - PATCH [03/16] - Removed __ prefix for unsigned vars. - PATCH [04/16] - Update include for headers moved to UAPI. - Removed __ prefix for unsigned vars. - PATCH [05/16] - Update include for headers in UAPI folder. - Removed setting any properties that are reported by device as 0. - Simplified modify_port. - PD should be allocated first in kernel then in device. - Update to pvrdma_cmd_post for creating/destroying PD, Query port/device. - PATCH [06/16] - pvrdma_cmd_post takes the response code. - PATCH [07/16] - Correct var type passed to dma_alloc_coherent. - PATCH [08/16] - Moved the timeout to pvrdma_cmd_recv. - Added additional response code parameter to pvrdma_cmd_post. - PATCH [09/16] - Updated include for headers in UAPI folder. - Changed from EINVAL to ENOMEM if atomic add fails. - Added error code if destroy cq command failed. - Update to pvrdma_cmd_post for creating/destroying CQ. - PATCH [11/16] - Check the access flags correctly for DMA MR. - Update to pvrdma_cmd_post for creating/destroying MRs. - PATCH [12/16] - Updated include for headers in UAPI folder. - Update to pvrdma_cmd_post for creating/destroying/querying/modifying QPs. - Use the pvrdma_sge struct when posting WRs/allocating QP memory. - Removed two set but unused variables. - PATCH [13/16] - Removed two unnecessary lines. - Updated include for headers in UAPI folder. - Update to pvrdma_cmd_post for add/delete GIDs. - Add error code in dev_warn if pvrdma_cmd_post failed. - PATCH [16/16] - Added pvrdma files to common UAPI folder. Changes v3->v4: - Rebased on for-4.9 branch - commit 64278fe89b729 ("Merge branch 'hns-roce' into k.o/for-4.9") - PATCH [01/16] - New in v4 - Moved vmxnet3 id to pci_ids.h - PATCH [02,03/16] - pvrdma_sge was moved into pvrdma_uapi.h - PATCH [04/16] - Removed explicit enum values. - PATCH [05/16] - Renamed priviledged -> privileged. - Added error numbers for command errors. - Removed unnecessary goto in modify_device. - Moved pd allocation to after command execution. - Removed an incorrect atomic_dec. - PATCH [06/16] - Renamed priviledged -> privileged. - Renamed pvrdma_flush_cqe to _pvrdma_flush_cqe since we hold a lock to call it. - Added wrapper functions for writing to UARs for CQ/QP. - The conversion functions are updated as func_name(dst, src) format. - Renamed max_gs to max_sg. - Added work struct for net device events. - PATCH [07/16] - Updated conversion functions to func_name(dst, src) format. - Removed unneeded local variables. - PATCH [08/16] - Removed the min check and added a BUILD_BUG_ON check for size. - PATCH [09/16] - Added a pvrdma_destroy_cq in the error path. - Renamed pvrdma_flush_cqe to _pvrdma_flush_cqe since we need a lock to be held while calling this. - Updated to use wrapper for UAR write for CQ. - Ensure that poll_cq does not return error values. - PATCH [10/16] - Removed an unnecessary comment. - PATCH [11/16] - Changed access flag check for DMA MR to using bit operation. - Removed some local variables. - PATCH [12/16] - Removed an unnecessary switch case. - Unified the returns in pvrdma_create_qp to use one exit point. - Renamed pvrdma_flush_cqe to _pvrdma_flush_cqe since we need a lock to be held when calling this. - Updated to use wrapper for UAR write for QP. - Updated conversion function to func_name(dst, src) format. - Renamed max_gs to max_sg. - Renamed cap variable to req_cap in pvrdma_set_sq/rq_size. - Changed dev_warn to dev_warn_ratelimited in pvrdma_post_send/recv. - Added nesting locking for flushing CQs when destroying/resetting a QP. - Added missing ret value. - PATCH [13/16] - Fixed some checkpatch warnings. - Added support for new get_dev_fw_str API. - Added event workqueue for netdevice events. - Restructured the pvrdma_pci_remove function a little bit. - PATCH [14/16] - Enforced dependency on VMXNet3 module. Changes v2->v3: - I reordered the patches so that the definitions of enums, structures is before their use (suggested by Yuval Shaia) so its easier to review. - Removed an unneccesary bool in pvrdma_cmd_post (suggested by Yuval Shaia). - Made the use of comma at end of enums consistent across files (suggested by Leon Romanovsky). Changes v1->v2: - Patch [07/15] - Addressed Yuval Shaia's comments and 32-bit build errors. --- Adit Ranadive (16): vmxnet3: Move PCI Id to pci_ids.h IB/pvrdma: Add user-level shared functions IB/pvrdma: Add virtual device RDMA structures IB/pvrdma: Add the paravirtual RDMA device specification IB/pvrdma: Add functions for Verbs support IB/pvrdma: Add paravirtual rdma device IB/pvrdma: Add helper functions IB/pvrdma: Add device command support IB/pvrdma: Add support for Completion Queues IB/pvrdma: Add UAR support IB/pvrdma: Add support for memory regions IB/pvrdma: Add Queue Pair support IB/pvrdma: Add the main driver module for PVRDMA IB/pvrdma: Add Kconfig and Makefile IB: Add PVRDMA driver MAINTAINERS: Update for PVRDMA driver MAINTAINERS | 9 + drivers/infiniband/Kconfig | 1 + drivers/infiniband/hw/Makefile | 1 + drivers/infiniband/hw/pvrdma/Kconfig | 7 + drivers/infiniband/hw/pvrdma/Makefile | 3 + drivers/infiniband/hw/pvrdma/pvrdma.h | 473 +++++++++ drivers/infiniband/hw/pvrdma/pvrdma_cmd.c | 117 +++ drivers/infiniband/hw/pvrdma/pvrdma_cq.c | 426 +++++++++ drivers/infiniband/hw/pvrdma/pvrdma_defs.h | 301 ++++++ drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h | 342 +++++++ drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c | 127 +++ drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h | 444 +++++++++ drivers/infiniband/hw/pvrdma/pvrdma_main.c | 1220 ++++++++++++++++++++++++ drivers/infiniband/hw/pvrdma/pvrdma_misc.c | 304 ++++++ drivers/infiniband/hw/pvrdma/pvrdma_mr.c | 334 +++++++ drivers/infiniband/hw/pvrdma/pvrdma_qp.c | 973 +++++++++++++++++++ drivers/infiniband/hw/pvrdma/pvrdma_verbs.c | 577 +++++++++++ drivers/infiniband/hw/pvrdma/pvrdma_verbs.h | 108 +++ drivers/net/vmxnet3/vmxnet3_int.h | 3 +- include/linux/pci_ids.h | 1 + include/uapi/rdma/Kbuild | 2 + include/uapi/rdma/pvrdma-abi.h | 99 ++ include/uapi/rdma/pvrdma-uapi.h | 255 +++++ 23 files changed, 6125 insertions(+), 2 deletions(-) create mode 100644 drivers/infiniband/hw/pvrdma/Kconfig create mode 100644 drivers/infiniband/hw/pvrdma/Makefile create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma.h create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cmd.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_cq.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_defs.h create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_dev_api.h create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_doorbell.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_ib_verbs.h create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_main.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_misc.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_mr.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_qp.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.c create mode 100644 drivers/infiniband/hw/pvrdma/pvrdma_verbs.h create mode 100644 include/uapi/rdma/pvrdma-abi.h create mode 100644 include/uapi/rdma/pvrdma-uapi.h -- 2.7.4 -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
