| Message ID | 20171030022037.GA20100@bryantan-devbox.prom.eng.vmware.com.prom.eng.vmware.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On Sun, Oct 29, 2017 at 07:20:43PM -0700, Bryan Tan wrote: > Add the required functions needed to support SRQs. Currently, kernel > clients are not supported. SRQs will only be available in userspace. > > Reviewed-by: Adit Ranadive <aditr@vmware.com> > Reviewed-by: Aditya Sarwade <asarwade@vmware.com> > Reviewed-by: Jorgen Hansen <jhansen@vmware.com> > Reviewed-by: Nitish Bhat <bnitish@vmware.com> > Signed-off-by: Bryan Tan <bryantan@vmware.com> > --- > drivers/infiniband/hw/vmw_pvrdma/Makefile | 2 +- > drivers/infiniband/hw/vmw_pvrdma/pvrdma.h | 25 ++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h | 54 ++++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c | 54 +++- > drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c | 57 +++- > drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c | 311 ++++++++++++++++++++++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c | 3 + > drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h | 18 ++ > include/uapi/rdma/vmw_pvrdma-abi.h | 2 + > 9 files changed, 512 insertions(+), 14 deletions(-) > create mode 100644 drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > > diff --git a/drivers/infiniband/hw/vmw_pvrdma/Makefile b/drivers/infiniband/hw/vmw_pvrdma/Makefile > index 0194ed1..2f52e0a 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/Makefile > +++ b/drivers/infiniband/hw/vmw_pvrdma/Makefile > @@ -1,3 +1,3 @@ > obj-$(CONFIG_INFINIBAND_VMWARE_PVRDMA) += vmw_pvrdma.o > > -vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_verbs.o > +vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_srq.o pvrdma_verbs.o > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > index 984aa34..f53a446 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > @@ -162,6 +162,22 @@ struct pvrdma_ah { > struct pvrdma_av av; > }; > > +struct pvrdma_srq { > + struct ib_srq ibsrq; > + int offset; > + spinlock_t lock; /* SRQ lock. */ > + int wqe_cnt; > + int wqe_size; > + int max_gs; > + struct ib_umem *umem; > + struct pvrdma_ring_state *ring; > + struct pvrdma_page_dir pdir; > + u32 srq_handle; > + int npages; > + atomic_t refcnt; I think that new fashion is to use refcnt_t for such counts. > + wait_queue_head_t wait; > +}; > + > struct pvrdma_qp { > struct ib_qp ibqp; > u32 qp_handle; > @@ -171,6 +187,7 @@ struct pvrdma_qp { > struct ib_umem *rumem; > struct ib_umem *sumem; > struct pvrdma_page_dir pdir; > + struct pvrdma_srq *srq; > int npages; > int npages_send; > int npages_recv; > @@ -210,6 +227,8 @@ struct pvrdma_dev { > struct pvrdma_page_dir cq_pdir; > struct pvrdma_cq **cq_tbl; > spinlock_t cq_tbl_lock; > + struct pvrdma_srq **srq_tbl; > + spinlock_t srq_tbl_lock; > struct pvrdma_qp **qp_tbl; > spinlock_t qp_tbl_lock; > struct pvrdma_uar_table uar_table; > @@ -221,6 +240,7 @@ struct pvrdma_dev { > bool ib_active; > atomic_t num_qps; > atomic_t num_cqs; > + atomic_t num_srqs; > atomic_t num_pds; > atomic_t num_ahs; > > @@ -256,6 +276,11 @@ static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq) > return container_of(ibcq, struct pvrdma_cq, ibcq); > } > > +static inline struct pvrdma_srq *to_vsrq(struct ib_srq *ibsrq) > +{ > + return container_of(ibsrq, struct pvrdma_srq, ibsrq); > +} > + > static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr) > { > return container_of(ibmr, struct pvrdma_user_mr, ibmr); > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > index df0a6b5..6fd5a8f 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > @@ -339,6 +339,10 @@ enum { > PVRDMA_CMD_DESTROY_UC, > PVRDMA_CMD_CREATE_BIND, > PVRDMA_CMD_DESTROY_BIND, > + PVRDMA_CMD_CREATE_SRQ, > + PVRDMA_CMD_MODIFY_SRQ, > + PVRDMA_CMD_QUERY_SRQ, > + PVRDMA_CMD_DESTROY_SRQ, > PVRDMA_CMD_MAX, > }; > > @@ -361,6 +365,10 @@ enum { > PVRDMA_CMD_DESTROY_UC_RESP_NOOP, > PVRDMA_CMD_CREATE_BIND_RESP_NOOP, > PVRDMA_CMD_DESTROY_BIND_RESP_NOOP, > + PVRDMA_CMD_CREATE_SRQ_RESP, > + PVRDMA_CMD_MODIFY_SRQ_RESP, > + PVRDMA_CMD_QUERY_SRQ_RESP, > + PVRDMA_CMD_DESTROY_SRQ_RESP, > PVRDMA_CMD_MAX_RESP, > }; > > @@ -495,6 +503,46 @@ struct pvrdma_cmd_destroy_cq { > u8 reserved[4]; > }; > > +struct pvrdma_cmd_create_srq { > + struct pvrdma_cmd_hdr hdr; > + u64 pdir_dma; > + u32 pd_handle; > + u32 nchunks; > + struct pvrdma_srq_attr attrs; > + u8 srq_type; > + u8 reserved[7]; > +}; > + > +struct pvrdma_cmd_create_srq_resp { > + struct pvrdma_cmd_resp_hdr hdr; > + u32 srqn; > + u8 reserved[4]; > +}; > + > +struct pvrdma_cmd_modify_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u32 attr_mask; > + struct pvrdma_srq_attr attrs; > +}; > + > +struct pvrdma_cmd_query_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u8 reserved[4]; > +}; > + > +struct pvrdma_cmd_query_srq_resp { > + struct pvrdma_cmd_resp_hdr hdr; > + struct pvrdma_srq_attr attrs; > +}; > + > +struct pvrdma_cmd_destroy_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u8 reserved[4]; > +}; > + > struct pvrdma_cmd_create_qp { > struct pvrdma_cmd_hdr hdr; > u64 pdir_dma; > @@ -594,6 +642,10 @@ struct pvrdma_cmd_destroy_bind { > struct pvrdma_cmd_destroy_qp destroy_qp; > struct pvrdma_cmd_create_bind create_bind; > struct pvrdma_cmd_destroy_bind destroy_bind; > + struct pvrdma_cmd_create_srq create_srq; > + struct pvrdma_cmd_modify_srq modify_srq; > + struct pvrdma_cmd_query_srq query_srq; > + struct pvrdma_cmd_destroy_srq destroy_srq; > }; > > union pvrdma_cmd_resp { > @@ -608,6 +660,8 @@ struct pvrdma_cmd_destroy_bind { > struct pvrdma_cmd_create_qp_resp create_qp_resp; > struct pvrdma_cmd_query_qp_resp query_qp_resp; > struct pvrdma_cmd_destroy_qp_resp destroy_qp_resp; > + struct pvrdma_cmd_create_srq_resp create_srq_resp; > + struct pvrdma_cmd_query_srq_resp query_srq_resp; > }; > > #endif /* __PVRDMA_DEV_API_H__ */ > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > index 6ce709a..7ce6687 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > @@ -118,6 +118,7 @@ static int pvrdma_init_device(struct pvrdma_dev *dev) > spin_lock_init(&dev->cmd_lock); > sema_init(&dev->cmd_sema, 1); > atomic_set(&dev->num_qps, 0); > + atomic_set(&dev->num_srqs, 0); > atomic_set(&dev->num_cqs, 0); > atomic_set(&dev->num_pds, 0); > atomic_set(&dev->num_ahs, 0); > @@ -195,6 +196,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | > (1ull << IB_USER_VERBS_CMD_QUERY_QP) | > (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | > + (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | > + (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV) | > (1ull << IB_USER_VERBS_CMD_POST_SEND) | > (1ull << IB_USER_VERBS_CMD_POST_RECV) | > (1ull << IB_USER_VERBS_CMD_CREATE_AH) | > @@ -227,6 +233,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > dev->ib_dev.destroy_cq = pvrdma_destroy_cq; > dev->ib_dev.poll_cq = pvrdma_poll_cq; > dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq; > + dev->ib_dev.create_srq = pvrdma_create_srq; > + dev->ib_dev.modify_srq = pvrdma_modify_srq; > + dev->ib_dev.query_srq = pvrdma_query_srq; > + dev->ib_dev.destroy_srq = pvrdma_destroy_srq; > + dev->ib_dev.post_srq_recv = pvrdma_post_srq_recv; > dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr; > dev->ib_dev.reg_user_mr = pvrdma_reg_user_mr; > dev->ib_dev.dereg_mr = pvrdma_dereg_mr; > @@ -254,9 +265,20 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > goto err_cq_free; > spin_lock_init(&dev->qp_tbl_lock); > > + /* Check if SRQ is supported by backend */ > + if (dev->dsr->caps.max_srq > 0) { > + dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq, sizeof(void *), > + GFP_KERNEL); Shouldn't the "sizeof(void *)" need to be "sizeof(struct pvrdma_srq *)" > + if (!dev->srq_tbl) > + goto err_qp_free; > + } else { > + dev->srq_tbl = NULL; dev is allocated with ib_alloc_device() and srq_tbl will be zero by default. > + } > + spin_lock_init(&dev->srq_tbl_lock); > + > ret = ib_register_device(&dev->ib_dev, NULL); > if (ret) > - goto err_qp_free; > + goto err_srq_free; > > for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) { > ret = device_create_file(&dev->ib_dev.dev, > @@ -271,6 +293,8 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > > err_class: > ib_unregister_device(&dev->ib_dev); > +err_srq_free: > + kfree(dev->srq_tbl); > err_qp_free: > kfree(dev->qp_tbl); > err_cq_free: > @@ -353,6 +377,32 @@ static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type) > } > } > > +static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type) > +{ > + struct pvrdma_srq *srq; > + unsigned long flags; > + > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq]; Are you sure that "dev->srq_tb != NULL && dev->dsr->caps.max_srq > 0" always here? > + if (srq) > + atomic_inc(&srq->refcnt); > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + if (srq && srq->ibsrq.event_handler) { > + struct ib_srq *ibsrq = &srq->ibsrq; > + struct ib_event e; > + > + e.device = ibsrq->device; > + e.element.srq = ibsrq; > + e.event = type; /* 1:1 mapping for now. */ > + ibsrq->event_handler(&e, ibsrq->srq_context); > + } > + if (srq) { > + if (atomic_dec_and_test(&srq->refcnt)) > + wake_up(&srq->wait); > + } > +} > + > static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port, > enum ib_event_type event) > { > @@ -423,6 +473,7 @@ static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id) > > case PVRDMA_EVENT_SRQ_ERR: > case PVRDMA_EVENT_SRQ_LIMIT_REACHED: > + pvrdma_srq_event(dev, eqe->info, eqe->type); > break; > > case PVRDMA_EVENT_PORT_ACTIVE: > @@ -1059,6 +1110,7 @@ static void pvrdma_pci_remove(struct pci_dev *pdev) > iounmap(dev->regs); > kfree(dev->sgid_tbl); > kfree(dev->cq_tbl); > + kfree(dev->srq_tbl); > kfree(dev->qp_tbl); > pvrdma_uar_table_cleanup(dev); > iounmap(dev->driver_uar.map); > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > index ed34d5a..4c4a9d1 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > @@ -198,6 +198,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > struct pvrdma_create_qp ucmd; > unsigned long flags; > int ret; > + bool is_srq = !!init_attr->srq; > > if (init_attr->create_flags) { > dev_warn(&dev->pdev->dev, > @@ -214,6 +215,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > return ERR_PTR(-EINVAL); > } > > + if (is_srq) { > + if (dev->dsr->caps.max_srq == 0) { > + dev_warn(&dev->pdev->dev, > + "SRQs not supported by device\n"); > + return ERR_PTR(-EINVAL); > + } > + } > + > if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp)) > return ERR_PTR(-ENOMEM); > > @@ -252,26 +261,36 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > goto err_qp; > } > > - /* set qp->sq.wqe_cnt, shift, buf_size.. */ > - qp->rumem = ib_umem_get(pd->uobject->context, > - ucmd.rbuf_addr, > - ucmd.rbuf_size, 0, 0); > - if (IS_ERR(qp->rumem)) { > - ret = PTR_ERR(qp->rumem); > - goto err_qp; > + if (!is_srq) { > + /* set qp->sq.wqe_cnt, shift, buf_size.. */ > + qp->rumem = ib_umem_get(pd->uobject->context, > + ucmd.rbuf_addr, > + ucmd.rbuf_size, 0, 0); > + if (IS_ERR(qp->rumem)) { > + ret = PTR_ERR(qp->rumem); > + goto err_qp; > + } > + qp->srq = NULL; > + } else { > + qp->rumem = NULL; > + qp->srq = to_vsrq(init_attr->srq); > } > > qp->sumem = ib_umem_get(pd->uobject->context, > ucmd.sbuf_addr, > ucmd.sbuf_size, 0, 0); > if (IS_ERR(qp->sumem)) { > - ib_umem_release(qp->rumem); > + if (!is_srq) > + ib_umem_release(qp->rumem); > ret = PTR_ERR(qp->sumem); > goto err_qp; > } > > qp->npages_send = ib_umem_page_count(qp->sumem); > - qp->npages_recv = ib_umem_page_count(qp->rumem); > + if (!is_srq) > + qp->npages_recv = ib_umem_page_count(qp->rumem); > + else > + qp->npages_recv = 0; > qp->npages = qp->npages_send + qp->npages_recv; > } else { > qp->is_kernel = true; > @@ -312,12 +331,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > > if (!qp->is_kernel) { > pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0); > - pvrdma_page_dir_insert_umem(&qp->pdir, qp->rumem, > - qp->npages_send); > + if (!is_srq) > + pvrdma_page_dir_insert_umem(&qp->pdir, > + qp->rumem, > + qp->npages_send); > } else { > /* Ring state is always the first page. */ > qp->sq.ring = qp->pdir.pages[0]; > - qp->rq.ring = &qp->sq.ring[1]; > + qp->rq.ring = is_srq ? NULL : &qp->sq.ring[1]; > } > break; > default: > @@ -333,6 +354,10 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > cmd->pd_handle = to_vpd(pd)->pd_handle; > cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle; > cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle; > + if (is_srq) > + cmd->srq_handle = to_vsrq(init_attr->srq)->srq_handle; > + else > + cmd->srq_handle = 0; > cmd->max_send_wr = init_attr->cap.max_send_wr; > cmd->max_recv_wr = init_attr->cap.max_recv_wr; > cmd->max_send_sge = init_attr->cap.max_send_sge; > @@ -340,6 +365,8 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > cmd->max_inline_data = init_attr->cap.max_inline_data; > cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0; > cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type); > + cmd->is_srq = is_srq; > + cmd->lkey = 0; > cmd->access_flags = IB_ACCESS_LOCAL_WRITE; > cmd->total_chunks = qp->npages; > cmd->send_chunks = qp->npages_send - PVRDMA_QP_NUM_HEADER_PAGES; > @@ -815,6 +842,12 @@ int pvrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, > return -EINVAL; > } > > + if (qp->srq) { > + dev_warn(&dev->pdev->dev, "QP associated with SRQ\n"); > + *bad_wr = wr; > + return -EINVAL; > + } > + > spin_lock_irqsave(&qp->rq.lock, flags); > > while (wr) { > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > new file mode 100644 > index 0000000..59d4595 > --- /dev/null > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > @@ -0,0 +1,311 @@ > +/* > + * Copyright (c) 2016-2017 VMware, Inc. All rights reserved. > + * > + * This program is free software; you can redistribute it and/or > + * modify it under the terms of EITHER the GNU General Public License > + * version 2 as published by the Free Software Foundation or the BSD > + * 2-Clause License. This program is distributed in the hope that it > + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED > + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. > + * See the GNU General Public License version 2 for more details at > + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html. > + * > + * You should have received a copy of the GNU General Public License > + * along with this program available in the file COPYING in the main > + * directory of this source tree. > + * > + * The BSD 2-Clause License > + * > + * Redistribution and use in source and binary forms, with or > + * without modification, are permitted provided that the following > + * conditions are met: > + * > + * - Redistributions of source code must retain the above > + * copyright notice, this list of conditions and the following > + * disclaimer. > + * > + * - Redistributions in binary form must reproduce the above > + * copyright notice, this list of conditions and the following > + * disclaimer in the documentation and/or other materials > + * provided with the distribution. > + * > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS > + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE > + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, > + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES > + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR > + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) > + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, > + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) > + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED > + * OF THE POSSIBILITY OF SUCH DAMAGE. > + */ > + > +#include <asm/page.h> > +#include <linux/io.h> > +#include <linux/wait.h> > +#include <rdma/ib_addr.h> > +#include <rdma/ib_smi.h> > +#include <rdma/ib_user_verbs.h> > + > +#include "pvrdma.h" > + > +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, > + struct ib_recv_wr **bad_wr) > +{ > + /* No support for kernel clients. */ > + return -EOPNOTSUPP; > +} > + > +/** > + * pvrdma_query_srq - query shared receive queue > + * @ibsrq: the shared receive queue to query > + * @srq_attr: attributes to query and return to client > + * > + * @return: 0 for success, otherwise returns an errno. > + */ > +int pvrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr) > +{ > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > + struct pvrdma_srq *srq = to_vsrq(ibsrq); > + union pvrdma_cmd_req req; > + union pvrdma_cmd_resp rsp; > + struct pvrdma_cmd_query_srq *cmd = &req.query_srq; > + struct pvrdma_cmd_query_srq_resp *resp = &rsp.query_srq_resp; > + int ret; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_QUERY_SRQ; > + cmd->srq_handle = srq->srq_handle; > + > + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_SRQ_RESP); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not query shared receive queue, error: %d\n", > + ret); > + return -EINVAL; > + } > + > + srq_attr->srq_limit = resp->attrs.srq_limit; > + srq_attr->max_wr = resp->attrs.max_wr; > + srq_attr->max_sge = resp->attrs.max_sge; > + > + return 0; > +} > + > +/** > + * pvrdma_create_srq - create shared receive queue > + * @pd: protection domain > + * @init_attr: shared receive queue attributes > + * @udata: user data > + * > + * @return: the ib_srq pointer on success, otherwise returns an errno. > + */ > +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, > + struct ib_srq_init_attr *init_attr, > + struct ib_udata *udata) > +{ > + struct pvrdma_srq *srq = NULL; > + struct pvrdma_dev *dev = to_vdev(pd->device); > + union pvrdma_cmd_req req; > + union pvrdma_cmd_resp rsp; > + struct pvrdma_cmd_create_srq *cmd = &req.create_srq; > + struct pvrdma_cmd_create_srq_resp *resp = &rsp.create_srq_resp; > + struct pvrdma_create_srq ucmd; > + unsigned long flags; > + int ret; > + > + if (init_attr->srq_type != IB_SRQT_BASIC) > + return ERR_PTR(-EINVAL); > + > + if (init_attr->attr.max_wr > dev->dsr->caps.max_srq_wr || > + init_attr->attr.max_sge > dev->dsr->caps.max_srq_sge) > + return ERR_PTR(-EINVAL); > + > + if (!atomic_add_unless(&dev->num_srqs, 1, dev->dsr->caps.max_srq)) > + return ERR_PTR(-ENOMEM); > + > + srq = kmalloc(sizeof(*srq), GFP_KERNEL); > + if (!srq) { > + ret = -ENOMEM; > + goto err_srq; > + } > + > + spin_lock_init(&srq->lock); > + atomic_set(&srq->refcnt, 1); > + init_waitqueue_head(&srq->wait); > + > + if (!(pd->uobject && udata)) { > + /* No support for kernel clients. */ > + ret = -EOPNOTSUPP; > + goto err_srq; > + } > + > + dev_dbg(&dev->pdev->dev, > + "create shared receive queue from user space\n"); > + > + if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) { > + ret = -EFAULT; > + goto err_srq; > + } > + > + srq->umem = ib_umem_get(pd->uobject->context, > + ucmd.buf_addr, > + ucmd.buf_size, 0, 0); > + if (IS_ERR(srq->umem)) { > + ret = PTR_ERR(srq->umem); > + goto err_srq; > + } > + > + srq->npages = ib_umem_page_count(srq->umem); > + > + if (srq->npages < 0 || srq->npages > PVRDMA_PAGE_DIR_MAX_PAGES) { > + dev_warn(&dev->pdev->dev, > + "overflow pages in shared receive queue\n"); > + ret = -EINVAL; > + goto err_umem; > + } > + > + ret = pvrdma_page_dir_init(dev, &srq->pdir, srq->npages, false); > + if (ret) { > + dev_warn(&dev->pdev->dev, > + "could not allocate page directory\n"); > + goto err_umem; > + } > + > + pvrdma_page_dir_insert_umem(&srq->pdir, srq->umem, 0); > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_CREATE_SRQ; > + cmd->srq_type = init_attr->srq_type; > + cmd->nchunks = srq->npages; > + cmd->pd_handle = to_vpd(pd)->pd_handle; > + cmd->attrs.max_wr = init_attr->attr.max_wr; > + cmd->attrs.max_sge = init_attr->attr.max_sge; > + cmd->attrs.srq_limit = init_attr->attr.srq_limit; > + cmd->pdir_dma = srq->pdir.dir_dma; > + > + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_SRQ_RESP); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not create shared receive queue, error: %d\n", > + ret); > + goto err_page_dir; > + } > + > + srq->srq_handle = resp->srqn; > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + dev->srq_tbl[srq->srq_handle % dev->dsr->caps.max_srq] = srq; > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + /* Copy udata back. */ > + if (ib_copy_to_udata(udata, &srq->srq_handle, sizeof(__u32))) { > + dev_warn(&dev->pdev->dev, "failed to copy back udata\n"); > + pvrdma_destroy_srq(&srq->ibsrq); > + return ERR_PTR(-EINVAL); > + } > + > + return &srq->ibsrq; > + > +err_page_dir: > + pvrdma_page_dir_cleanup(dev, &srq->pdir); > +err_umem: > + ib_umem_release(srq->umem); > +err_srq: > + kfree(srq); > + atomic_dec(&dev->num_srqs); > + > + return ERR_PTR(ret); > +} > + > +static void pvrdma_free_srq(struct pvrdma_dev *dev, struct pvrdma_srq *srq) > +{ > + unsigned long flags; > + > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + dev->srq_tbl[srq->srq_handle] = NULL; Again, are you sure that dev->srq_tbl exist? > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + atomic_dec(&srq->refcnt); > + wait_event(srq->wait, !atomic_read(&srq->refcnt)); > + > + /* There is no support for kernel clients, so this is safe. */ > + ib_umem_release(srq->umem); > + > + pvrdma_page_dir_cleanup(dev, &srq->pdir); > + > + kfree(srq); > + > + atomic_dec(&dev->num_srqs); > +} > + > +/** > + * pvrdma_destroy_srq - destroy shared receive queue > + * @srq: the shared receive queue to destroy > + * > + * @return: 0 for success. > + */ > +int pvrdma_destroy_srq(struct ib_srq *srq) > +{ > + struct pvrdma_srq *vsrq = to_vsrq(srq); > + union pvrdma_cmd_req req; > + struct pvrdma_cmd_destroy_srq *cmd = &req.destroy_srq; > + struct pvrdma_dev *dev = to_vdev(srq->device); > + int ret; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_DESTROY_SRQ; > + cmd->srq_handle = vsrq->srq_handle; > + > + ret = pvrdma_cmd_post(dev, &req, NULL, 0); > + if (ret < 0) > + dev_warn(&dev->pdev->dev, > + "destroy shared receive queue failed, error: %d\n", > + ret); > + > + pvrdma_free_srq(dev, vsrq); > + > + return 0; > +} > + > +/** > + * pvrdma_modify_srq - modify shared receive queue attributes > + * @ibsrq: the shared receive queue to modify > + * @attr: the shared receive queue's new attributes > + * @attr_mask: attributes mask > + * @udata: user data > + * > + * @returns 0 on success, otherwise returns an errno. > + */ > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > +{ > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > + union pvrdma_cmd_req req; > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > + int ret; > + > + /* No support for resizing SRQs. */ Better to write that you support, so future additions to ib_srq_attr_mask won't need to change your code. > + if (attr_mask & IB_SRQ_MAX_WR) > + return -EINVAL; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_MODIFY_SRQ; > + cmd->srq_handle = vsrq->srq_handle; > + cmd->attrs.srq_limit = attr->srq_limit; > + cmd->attr_mask = attr_mask; > + > + ret = pvrdma_cmd_post(dev, &req, NULL, 0); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not modify shared receive queue, error: %d\n", > + ret); > + > + return -EINVAL; > + } > + > + return ret; > +} > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > index 48776f5..16b9661 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > @@ -85,6 +85,9 @@ int pvrdma_query_device(struct ib_device *ibdev, > props->max_sge = dev->dsr->caps.max_sge; > props->max_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge, > dev->dsr->caps.max_sge_rd); > + props->max_srq = dev->dsr->caps.max_srq; > + props->max_srq_wr = dev->dsr->caps.max_srq_wr; > + props->max_srq_sge = dev->dsr->caps.max_srq_sge; > props->max_cq = dev->dsr->caps.max_cq; > props->max_cqe = dev->dsr->caps.max_cqe; > props->max_mr = dev->dsr->caps.max_mr; > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > index 002a9b0..b7b2572 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > @@ -324,6 +324,13 @@ enum pvrdma_mw_type { > PVRDMA_MW_TYPE_2 = 2, > }; > > +struct pvrdma_srq_attr { > + u32 max_wr; > + u32 max_sge; > + u32 srq_limit; > + u32 reserved; > +}; > + > struct pvrdma_qp_attr { > enum pvrdma_qp_state qp_state; > enum pvrdma_qp_state cur_qp_state; > @@ -420,6 +427,17 @@ int pvrdma_resize_cq(struct ib_cq *ibcq, int entries, > struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, > struct ib_udata *udata); > int pvrdma_destroy_ah(struct ib_ah *ah); > + > +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, > + struct ib_srq_init_attr *init_attr, > + struct ib_udata *udata); > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata); > +int pvrdma_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr); > +int pvrdma_destroy_srq(struct ib_srq *srq); > +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, > + struct ib_recv_wr **bad_wr); > + > struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > struct ib_qp_init_attr *init_attr, > struct ib_udata *udata); > diff --git a/include/uapi/rdma/vmw_pvrdma-abi.h b/include/uapi/rdma/vmw_pvrdma-abi.h > index c6569b0..846c6f4 100644 > --- a/include/uapi/rdma/vmw_pvrdma-abi.h > +++ b/include/uapi/rdma/vmw_pvrdma-abi.h > @@ -158,6 +158,8 @@ struct pvrdma_resize_cq { > > struct pvrdma_create_srq { > __u64 buf_addr; > + __u32 buf_size; > + __u32 reserved; > }; > > struct pvrdma_create_srq_resp { > -- > 1.8.5.6 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, Oct 29, 2017 at 07:20:43PM -0700, Bryan Tan wrote: > Add the required functions needed to support SRQs. Currently, kernel > clients are not supported. SRQs will only be available in userspace. > > Reviewed-by: Adit Ranadive <aditr@vmware.com> > Reviewed-by: Aditya Sarwade <asarwade@vmware.com> > Reviewed-by: Jorgen Hansen <jhansen@vmware.com> > Reviewed-by: Nitish Bhat <bnitish@vmware.com> > Signed-off-by: Bryan Tan <bryantan@vmware.com> > --- > drivers/infiniband/hw/vmw_pvrdma/Makefile | 2 +- > drivers/infiniband/hw/vmw_pvrdma/pvrdma.h | 25 ++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h | 54 ++++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c | 54 +++- > drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c | 57 +++- > drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c | 311 ++++++++++++++++++++++ > drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c | 3 + > drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h | 18 ++ > include/uapi/rdma/vmw_pvrdma-abi.h | 2 + > 9 files changed, 512 insertions(+), 14 deletions(-) > create mode 100644 drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > > diff --git a/drivers/infiniband/hw/vmw_pvrdma/Makefile b/drivers/infiniband/hw/vmw_pvrdma/Makefile > index 0194ed1..2f52e0a 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/Makefile > +++ b/drivers/infiniband/hw/vmw_pvrdma/Makefile > @@ -1,3 +1,3 @@ > obj-$(CONFIG_INFINIBAND_VMWARE_PVRDMA) += vmw_pvrdma.o > > -vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_verbs.o > +vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_srq.o pvrdma_verbs.o > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > index 984aa34..f53a446 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h > @@ -162,6 +162,22 @@ struct pvrdma_ah { > struct pvrdma_av av; > }; > > +struct pvrdma_srq { > + struct ib_srq ibsrq; > + int offset; > + spinlock_t lock; /* SRQ lock. */ > + int wqe_cnt; > + int wqe_size; > + int max_gs; > + struct ib_umem *umem; > + struct pvrdma_ring_state *ring; > + struct pvrdma_page_dir pdir; > + u32 srq_handle; > + int npages; > + atomic_t refcnt; > + wait_queue_head_t wait; > +}; > + > struct pvrdma_qp { > struct ib_qp ibqp; > u32 qp_handle; > @@ -171,6 +187,7 @@ struct pvrdma_qp { > struct ib_umem *rumem; > struct ib_umem *sumem; > struct pvrdma_page_dir pdir; > + struct pvrdma_srq *srq; > int npages; > int npages_send; > int npages_recv; > @@ -210,6 +227,8 @@ struct pvrdma_dev { > struct pvrdma_page_dir cq_pdir; > struct pvrdma_cq **cq_tbl; > spinlock_t cq_tbl_lock; > + struct pvrdma_srq **srq_tbl; > + spinlock_t srq_tbl_lock; > struct pvrdma_qp **qp_tbl; > spinlock_t qp_tbl_lock; > struct pvrdma_uar_table uar_table; > @@ -221,6 +240,7 @@ struct pvrdma_dev { > bool ib_active; > atomic_t num_qps; > atomic_t num_cqs; > + atomic_t num_srqs; > atomic_t num_pds; > atomic_t num_ahs; > > @@ -256,6 +276,11 @@ static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq) > return container_of(ibcq, struct pvrdma_cq, ibcq); > } > > +static inline struct pvrdma_srq *to_vsrq(struct ib_srq *ibsrq) > +{ > + return container_of(ibsrq, struct pvrdma_srq, ibsrq); > +} > + > static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr) > { > return container_of(ibmr, struct pvrdma_user_mr, ibmr); > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > index df0a6b5..6fd5a8f 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h > @@ -339,6 +339,10 @@ enum { > PVRDMA_CMD_DESTROY_UC, > PVRDMA_CMD_CREATE_BIND, > PVRDMA_CMD_DESTROY_BIND, > + PVRDMA_CMD_CREATE_SRQ, > + PVRDMA_CMD_MODIFY_SRQ, > + PVRDMA_CMD_QUERY_SRQ, > + PVRDMA_CMD_DESTROY_SRQ, > PVRDMA_CMD_MAX, > }; > > @@ -361,6 +365,10 @@ enum { > PVRDMA_CMD_DESTROY_UC_RESP_NOOP, > PVRDMA_CMD_CREATE_BIND_RESP_NOOP, > PVRDMA_CMD_DESTROY_BIND_RESP_NOOP, > + PVRDMA_CMD_CREATE_SRQ_RESP, > + PVRDMA_CMD_MODIFY_SRQ_RESP, > + PVRDMA_CMD_QUERY_SRQ_RESP, > + PVRDMA_CMD_DESTROY_SRQ_RESP, > PVRDMA_CMD_MAX_RESP, > }; > > @@ -495,6 +503,46 @@ struct pvrdma_cmd_destroy_cq { > u8 reserved[4]; > }; > > +struct pvrdma_cmd_create_srq { > + struct pvrdma_cmd_hdr hdr; > + u64 pdir_dma; > + u32 pd_handle; > + u32 nchunks; > + struct pvrdma_srq_attr attrs; > + u8 srq_type; > + u8 reserved[7]; > +}; > + > +struct pvrdma_cmd_create_srq_resp { > + struct pvrdma_cmd_resp_hdr hdr; > + u32 srqn; > + u8 reserved[4]; > +}; > + > +struct pvrdma_cmd_modify_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u32 attr_mask; > + struct pvrdma_srq_attr attrs; > +}; > + > +struct pvrdma_cmd_query_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u8 reserved[4]; > +}; > + > +struct pvrdma_cmd_query_srq_resp { > + struct pvrdma_cmd_resp_hdr hdr; > + struct pvrdma_srq_attr attrs; > +}; > + > +struct pvrdma_cmd_destroy_srq { > + struct pvrdma_cmd_hdr hdr; > + u32 srq_handle; > + u8 reserved[4]; > +}; > + > struct pvrdma_cmd_create_qp { > struct pvrdma_cmd_hdr hdr; > u64 pdir_dma; > @@ -594,6 +642,10 @@ struct pvrdma_cmd_destroy_bind { > struct pvrdma_cmd_destroy_qp destroy_qp; > struct pvrdma_cmd_create_bind create_bind; > struct pvrdma_cmd_destroy_bind destroy_bind; > + struct pvrdma_cmd_create_srq create_srq; > + struct pvrdma_cmd_modify_srq modify_srq; > + struct pvrdma_cmd_query_srq query_srq; > + struct pvrdma_cmd_destroy_srq destroy_srq; > }; > > union pvrdma_cmd_resp { > @@ -608,6 +660,8 @@ struct pvrdma_cmd_destroy_bind { > struct pvrdma_cmd_create_qp_resp create_qp_resp; > struct pvrdma_cmd_query_qp_resp query_qp_resp; > struct pvrdma_cmd_destroy_qp_resp destroy_qp_resp; > + struct pvrdma_cmd_create_srq_resp create_srq_resp; > + struct pvrdma_cmd_query_srq_resp query_srq_resp; > }; > > #endif /* __PVRDMA_DEV_API_H__ */ > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > index 6ce709a..7ce6687 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c > @@ -118,6 +118,7 @@ static int pvrdma_init_device(struct pvrdma_dev *dev) > spin_lock_init(&dev->cmd_lock); > sema_init(&dev->cmd_sema, 1); > atomic_set(&dev->num_qps, 0); > + atomic_set(&dev->num_srqs, 0); > atomic_set(&dev->num_cqs, 0); > atomic_set(&dev->num_pds, 0); > atomic_set(&dev->num_ahs, 0); > @@ -195,6 +196,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | > (1ull << IB_USER_VERBS_CMD_QUERY_QP) | > (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | > + (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | > + (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | > + (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV) | > (1ull << IB_USER_VERBS_CMD_POST_SEND) | > (1ull << IB_USER_VERBS_CMD_POST_RECV) | > (1ull << IB_USER_VERBS_CMD_CREATE_AH) | > @@ -227,6 +233,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > dev->ib_dev.destroy_cq = pvrdma_destroy_cq; > dev->ib_dev.poll_cq = pvrdma_poll_cq; > dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq; > + dev->ib_dev.create_srq = pvrdma_create_srq; > + dev->ib_dev.modify_srq = pvrdma_modify_srq; > + dev->ib_dev.query_srq = pvrdma_query_srq; > + dev->ib_dev.destroy_srq = pvrdma_destroy_srq; > + dev->ib_dev.post_srq_recv = pvrdma_post_srq_recv; > dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr; > dev->ib_dev.reg_user_mr = pvrdma_reg_user_mr; > dev->ib_dev.dereg_mr = pvrdma_dereg_mr; > @@ -254,9 +265,20 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > goto err_cq_free; > spin_lock_init(&dev->qp_tbl_lock); > > + /* Check if SRQ is supported by backend */ > + if (dev->dsr->caps.max_srq > 0) { > + dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq, sizeof(void *), > + GFP_KERNEL); > + if (!dev->srq_tbl) > + goto err_qp_free; > + } else { > + dev->srq_tbl = NULL; > + } > + spin_lock_init(&dev->srq_tbl_lock); > + > ret = ib_register_device(&dev->ib_dev, NULL); > if (ret) > - goto err_qp_free; > + goto err_srq_free; > > for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) { > ret = device_create_file(&dev->ib_dev.dev, > @@ -271,6 +293,8 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) > > err_class: > ib_unregister_device(&dev->ib_dev); > +err_srq_free: > + kfree(dev->srq_tbl); > err_qp_free: > kfree(dev->qp_tbl); > err_cq_free: > @@ -353,6 +377,32 @@ static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type) > } > } > > +static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type) > +{ > + struct pvrdma_srq *srq; > + unsigned long flags; > + > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq]; > + if (srq) > + atomic_inc(&srq->refcnt); > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + if (srq && srq->ibsrq.event_handler) { > + struct ib_srq *ibsrq = &srq->ibsrq; > + struct ib_event e; > + > + e.device = ibsrq->device; > + e.element.srq = ibsrq; > + e.event = type; /* 1:1 mapping for now. */ > + ibsrq->event_handler(&e, ibsrq->srq_context); > + } > + if (srq) { > + if (atomic_dec_and_test(&srq->refcnt)) > + wake_up(&srq->wait); > + } > +} > + > static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port, > enum ib_event_type event) > { > @@ -423,6 +473,7 @@ static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id) > > case PVRDMA_EVENT_SRQ_ERR: > case PVRDMA_EVENT_SRQ_LIMIT_REACHED: > + pvrdma_srq_event(dev, eqe->info, eqe->type); > break; > > case PVRDMA_EVENT_PORT_ACTIVE: > @@ -1059,6 +1110,7 @@ static void pvrdma_pci_remove(struct pci_dev *pdev) > iounmap(dev->regs); > kfree(dev->sgid_tbl); > kfree(dev->cq_tbl); > + kfree(dev->srq_tbl); > kfree(dev->qp_tbl); > pvrdma_uar_table_cleanup(dev); > iounmap(dev->driver_uar.map); > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > index ed34d5a..4c4a9d1 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c > @@ -198,6 +198,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > struct pvrdma_create_qp ucmd; > unsigned long flags; > int ret; > + bool is_srq = !!init_attr->srq; > > if (init_attr->create_flags) { > dev_warn(&dev->pdev->dev, > @@ -214,6 +215,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > return ERR_PTR(-EINVAL); > } > > + if (is_srq) { > + if (dev->dsr->caps.max_srq == 0) { Let us be consist, so if in function pvrdma_init_device the assumption is that max_srq can have negative value then also check it here. Or on the other hands if not (and i assume it is not) then modify the check on pvrdma_init_device. Also, suggesting: if (is_srq && !dev->dsr->caps.max_srq) Or (if you do not accept my comment): if (is_srq && dev->dsr->caps.max_srq == 0) > + dev_warn(&dev->pdev->dev, > + "SRQs not supported by device\n"); > + return ERR_PTR(-EINVAL); > + } > + } > + > if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp)) > return ERR_PTR(-ENOMEM); > > @@ -252,26 +261,36 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > goto err_qp; > } > > - /* set qp->sq.wqe_cnt, shift, buf_size.. */ > - qp->rumem = ib_umem_get(pd->uobject->context, > - ucmd.rbuf_addr, > - ucmd.rbuf_size, 0, 0); > - if (IS_ERR(qp->rumem)) { > - ret = PTR_ERR(qp->rumem); > - goto err_qp; > + if (!is_srq) { > + /* set qp->sq.wqe_cnt, shift, buf_size.. */ > + qp->rumem = ib_umem_get(pd->uobject->context, > + ucmd.rbuf_addr, > + ucmd.rbuf_size, 0, 0); > + if (IS_ERR(qp->rumem)) { > + ret = PTR_ERR(qp->rumem); > + goto err_qp; > + } > + qp->srq = NULL; > + } else { > + qp->rumem = NULL; > + qp->srq = to_vsrq(init_attr->srq); > } > > qp->sumem = ib_umem_get(pd->uobject->context, > ucmd.sbuf_addr, > ucmd.sbuf_size, 0, 0); > if (IS_ERR(qp->sumem)) { > - ib_umem_release(qp->rumem); > + if (!is_srq) > + ib_umem_release(qp->rumem); > ret = PTR_ERR(qp->sumem); > goto err_qp; > } > > qp->npages_send = ib_umem_page_count(qp->sumem); > - qp->npages_recv = ib_umem_page_count(qp->rumem); > + if (!is_srq) > + qp->npages_recv = ib_umem_page_count(qp->rumem); > + else > + qp->npages_recv = 0; > qp->npages = qp->npages_send + qp->npages_recv; > } else { > qp->is_kernel = true; > @@ -312,12 +331,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > > if (!qp->is_kernel) { > pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0); > - pvrdma_page_dir_insert_umem(&qp->pdir, qp->rumem, > - qp->npages_send); > + if (!is_srq) > + pvrdma_page_dir_insert_umem(&qp->pdir, > + qp->rumem, > + qp->npages_send); > } else { > /* Ring state is always the first page. */ > qp->sq.ring = qp->pdir.pages[0]; > - qp->rq.ring = &qp->sq.ring[1]; > + qp->rq.ring = is_srq ? NULL : &qp->sq.ring[1]; > } > break; > default: > @@ -333,6 +354,10 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > cmd->pd_handle = to_vpd(pd)->pd_handle; > cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle; > cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle; > + if (is_srq) > + cmd->srq_handle = to_vsrq(init_attr->srq)->srq_handle; > + else > + cmd->srq_handle = 0; > cmd->max_send_wr = init_attr->cap.max_send_wr; > cmd->max_recv_wr = init_attr->cap.max_recv_wr; > cmd->max_send_sge = init_attr->cap.max_send_sge; > @@ -340,6 +365,8 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > cmd->max_inline_data = init_attr->cap.max_inline_data; > cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0; > cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type); > + cmd->is_srq = is_srq; > + cmd->lkey = 0; > cmd->access_flags = IB_ACCESS_LOCAL_WRITE; > cmd->total_chunks = qp->npages; > cmd->send_chunks = qp->npages_send - PVRDMA_QP_NUM_HEADER_PAGES; > @@ -815,6 +842,12 @@ int pvrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, > return -EINVAL; > } > > + if (qp->srq) { > + dev_warn(&dev->pdev->dev, "QP associated with SRQ\n"); > + *bad_wr = wr; > + return -EINVAL; > + } > + > spin_lock_irqsave(&qp->rq.lock, flags); > > while (wr) { > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > new file mode 100644 > index 0000000..59d4595 > --- /dev/null > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c > @@ -0,0 +1,311 @@ > +/* > + * Copyright (c) 2016-2017 VMware, Inc. All rights reserved. > + * > + * This program is free software; you can redistribute it and/or > + * modify it under the terms of EITHER the GNU General Public License > + * version 2 as published by the Free Software Foundation or the BSD > + * 2-Clause License. This program is distributed in the hope that it > + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED > + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. > + * See the GNU General Public License version 2 for more details at > + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html. > + * > + * You should have received a copy of the GNU General Public License > + * along with this program available in the file COPYING in the main > + * directory of this source tree. > + * > + * The BSD 2-Clause License > + * > + * Redistribution and use in source and binary forms, with or > + * without modification, are permitted provided that the following > + * conditions are met: > + * > + * - Redistributions of source code must retain the above > + * copyright notice, this list of conditions and the following > + * disclaimer. > + * > + * - Redistributions in binary form must reproduce the above > + * copyright notice, this list of conditions and the following > + * disclaimer in the documentation and/or other materials > + * provided with the distribution. > + * > + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS > + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT > + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS > + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE > + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, > + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES > + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR > + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) > + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, > + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) > + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED > + * OF THE POSSIBILITY OF SUCH DAMAGE. > + */ > + > +#include <asm/page.h> > +#include <linux/io.h> > +#include <linux/wait.h> > +#include <rdma/ib_addr.h> > +#include <rdma/ib_smi.h> > +#include <rdma/ib_user_verbs.h> > + > +#include "pvrdma.h" > + > +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, > + struct ib_recv_wr **bad_wr) > +{ > + /* No support for kernel clients. */ > + return -EOPNOTSUPP; > +} > + > +/** > + * pvrdma_query_srq - query shared receive queue > + * @ibsrq: the shared receive queue to query > + * @srq_attr: attributes to query and return to client > + * > + * @return: 0 for success, otherwise returns an errno. > + */ > +int pvrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr) > +{ > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > + struct pvrdma_srq *srq = to_vsrq(ibsrq); > + union pvrdma_cmd_req req; > + union pvrdma_cmd_resp rsp; > + struct pvrdma_cmd_query_srq *cmd = &req.query_srq; > + struct pvrdma_cmd_query_srq_resp *resp = &rsp.query_srq_resp; > + int ret; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_QUERY_SRQ; > + cmd->srq_handle = srq->srq_handle; > + > + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_SRQ_RESP); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not query shared receive queue, error: %d\n", > + ret); > + return -EINVAL; > + } > + > + srq_attr->srq_limit = resp->attrs.srq_limit; > + srq_attr->max_wr = resp->attrs.max_wr; > + srq_attr->max_sge = resp->attrs.max_sge; > + > + return 0; > +} > + > +/** > + * pvrdma_create_srq - create shared receive queue > + * @pd: protection domain > + * @init_attr: shared receive queue attributes > + * @udata: user data > + * > + * @return: the ib_srq pointer on success, otherwise returns an errno. > + */ > +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, > + struct ib_srq_init_attr *init_attr, > + struct ib_udata *udata) > +{ > + struct pvrdma_srq *srq = NULL; > + struct pvrdma_dev *dev = to_vdev(pd->device); > + union pvrdma_cmd_req req; > + union pvrdma_cmd_resp rsp; > + struct pvrdma_cmd_create_srq *cmd = &req.create_srq; > + struct pvrdma_cmd_create_srq_resp *resp = &rsp.create_srq_resp; > + struct pvrdma_create_srq ucmd; > + unsigned long flags; > + int ret; > + > + if (init_attr->srq_type != IB_SRQT_BASIC) > + return ERR_PTR(-EINVAL); > + > + if (init_attr->attr.max_wr > dev->dsr->caps.max_srq_wr || > + init_attr->attr.max_sge > dev->dsr->caps.max_srq_sge) Again - let us be consist, if we got warning in pvrdma_create_qp we should get one here as well. > + return ERR_PTR(-EINVAL); > + > + if (!atomic_add_unless(&dev->num_srqs, 1, dev->dsr->caps.max_srq)) > + return ERR_PTR(-ENOMEM); > + > + srq = kmalloc(sizeof(*srq), GFP_KERNEL); > + if (!srq) { > + ret = -ENOMEM; > + goto err_srq; > + } > + > + spin_lock_init(&srq->lock); > + atomic_set(&srq->refcnt, 1); > + init_waitqueue_head(&srq->wait); > + > + if (!(pd->uobject && udata)) { > + /* No support for kernel clients. */ > + ret = -EOPNOTSUPP; > + goto err_srq; > + } Why this check cannot be performed before the above initialization?? Also, suggesting a debug message here. > + > + dev_dbg(&dev->pdev->dev, > + "create shared receive queue from user space\n"); > + > + if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) { > + ret = -EFAULT; > + goto err_srq; > + } > + > + srq->umem = ib_umem_get(pd->uobject->context, > + ucmd.buf_addr, > + ucmd.buf_size, 0, 0); > + if (IS_ERR(srq->umem)) { > + ret = PTR_ERR(srq->umem); > + goto err_srq; > + } > + > + srq->npages = ib_umem_page_count(srq->umem); > + > + if (srq->npages < 0 || srq->npages > PVRDMA_PAGE_DIR_MAX_PAGES) { > + dev_warn(&dev->pdev->dev, > + "overflow pages in shared receive queue\n"); > + ret = -EINVAL; > + goto err_umem; > + } > + > + ret = pvrdma_page_dir_init(dev, &srq->pdir, srq->npages, false); > + if (ret) { > + dev_warn(&dev->pdev->dev, > + "could not allocate page directory\n"); > + goto err_umem; > + } > + > + pvrdma_page_dir_insert_umem(&srq->pdir, srq->umem, 0); > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_CREATE_SRQ; > + cmd->srq_type = init_attr->srq_type; > + cmd->nchunks = srq->npages; > + cmd->pd_handle = to_vpd(pd)->pd_handle; > + cmd->attrs.max_wr = init_attr->attr.max_wr; > + cmd->attrs.max_sge = init_attr->attr.max_sge; > + cmd->attrs.srq_limit = init_attr->attr.srq_limit; > + cmd->pdir_dma = srq->pdir.dir_dma; > + > + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_SRQ_RESP); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not create shared receive queue, error: %d\n", > + ret); > + goto err_page_dir; > + } > + > + srq->srq_handle = resp->srqn; > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + dev->srq_tbl[srq->srq_handle % dev->dsr->caps.max_srq] = srq; > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + /* Copy udata back. */ > + if (ib_copy_to_udata(udata, &srq->srq_handle, sizeof(__u32))) { > + dev_warn(&dev->pdev->dev, "failed to copy back udata\n"); > + pvrdma_destroy_srq(&srq->ibsrq); > + return ERR_PTR(-EINVAL); > + } > + > + return &srq->ibsrq; > + > +err_page_dir: > + pvrdma_page_dir_cleanup(dev, &srq->pdir); > +err_umem: > + ib_umem_release(srq->umem); > +err_srq: > + kfree(srq); > + atomic_dec(&dev->num_srqs); > + > + return ERR_PTR(ret); > +} > + > +static void pvrdma_free_srq(struct pvrdma_dev *dev, struct pvrdma_srq *srq) > +{ > + unsigned long flags; > + > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > + dev->srq_tbl[srq->srq_handle] = NULL; > + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); > + > + atomic_dec(&srq->refcnt); > + wait_event(srq->wait, !atomic_read(&srq->refcnt)); > + > + /* There is no support for kernel clients, so this is safe. */ > + ib_umem_release(srq->umem); > + > + pvrdma_page_dir_cleanup(dev, &srq->pdir); > + > + kfree(srq); > + > + atomic_dec(&dev->num_srqs); > +} > + > +/** > + * pvrdma_destroy_srq - destroy shared receive queue > + * @srq: the shared receive queue to destroy > + * > + * @return: 0 for success. > + */ > +int pvrdma_destroy_srq(struct ib_srq *srq) > +{ > + struct pvrdma_srq *vsrq = to_vsrq(srq); > + union pvrdma_cmd_req req; > + struct pvrdma_cmd_destroy_srq *cmd = &req.destroy_srq; > + struct pvrdma_dev *dev = to_vdev(srq->device); > + int ret; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_DESTROY_SRQ; > + cmd->srq_handle = vsrq->srq_handle; > + > + ret = pvrdma_cmd_post(dev, &req, NULL, 0); > + if (ret < 0) > + dev_warn(&dev->pdev->dev, > + "destroy shared receive queue failed, error: %d\n", > + ret); > + > + pvrdma_free_srq(dev, vsrq); > + > + return 0; > +} > + > +/** > + * pvrdma_modify_srq - modify shared receive queue attributes > + * @ibsrq: the shared receive queue to modify > + * @attr: the shared receive queue's new attributes > + * @attr_mask: attributes mask > + * @udata: user data > + * > + * @returns 0 on success, otherwise returns an errno. > + */ > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > +{ > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > + union pvrdma_cmd_req req; > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > + int ret; > + > + /* No support for resizing SRQs. */ > + if (attr_mask & IB_SRQ_MAX_WR) > + return -EINVAL; > + > + memset(cmd, 0, sizeof(*cmd)); > + cmd->hdr.cmd = PVRDMA_CMD_MODIFY_SRQ; > + cmd->srq_handle = vsrq->srq_handle; > + cmd->attrs.srq_limit = attr->srq_limit; > + cmd->attr_mask = attr_mask; > + > + ret = pvrdma_cmd_post(dev, &req, NULL, 0); > + if (ret < 0) { > + dev_warn(&dev->pdev->dev, > + "could not modify shared receive queue, error: %d\n", > + ret); > + > + return -EINVAL; > + } > + > + return ret; > +} > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > index 48776f5..16b9661 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c > @@ -85,6 +85,9 @@ int pvrdma_query_device(struct ib_device *ibdev, > props->max_sge = dev->dsr->caps.max_sge; > props->max_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge, > dev->dsr->caps.max_sge_rd); > + props->max_srq = dev->dsr->caps.max_srq; > + props->max_srq_wr = dev->dsr->caps.max_srq_wr; > + props->max_srq_sge = dev->dsr->caps.max_srq_sge; > props->max_cq = dev->dsr->caps.max_cq; > props->max_cqe = dev->dsr->caps.max_cqe; > props->max_mr = dev->dsr->caps.max_mr; > diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > index 002a9b0..b7b2572 100644 > --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h > @@ -324,6 +324,13 @@ enum pvrdma_mw_type { > PVRDMA_MW_TYPE_2 = 2, > }; > > +struct pvrdma_srq_attr { > + u32 max_wr; > + u32 max_sge; > + u32 srq_limit; > + u32 reserved; > +}; > + > struct pvrdma_qp_attr { > enum pvrdma_qp_state qp_state; > enum pvrdma_qp_state cur_qp_state; > @@ -420,6 +427,17 @@ int pvrdma_resize_cq(struct ib_cq *ibcq, int entries, > struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, > struct ib_udata *udata); > int pvrdma_destroy_ah(struct ib_ah *ah); > + > +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, > + struct ib_srq_init_attr *init_attr, > + struct ib_udata *udata); > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata); > +int pvrdma_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr); > +int pvrdma_destroy_srq(struct ib_srq *srq); > +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, > + struct ib_recv_wr **bad_wr); > + > struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, > struct ib_qp_init_attr *init_attr, > struct ib_udata *udata); > diff --git a/include/uapi/rdma/vmw_pvrdma-abi.h b/include/uapi/rdma/vmw_pvrdma-abi.h > index c6569b0..846c6f4 100644 > --- a/include/uapi/rdma/vmw_pvrdma-abi.h > +++ b/include/uapi/rdma/vmw_pvrdma-abi.h > @@ -158,6 +158,8 @@ struct pvrdma_resize_cq { > > struct pvrdma_create_srq { > __u64 buf_addr; > + __u32 buf_size; > + __u32 reserved; > }; > > struct pvrdma_create_srq_resp { > -- > 1.8.5.6 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Oct 30, 2017 at 08:03:17AM +0200, Leon Romanovsky wrote: > > +struct pvrdma_srq { > > + struct ib_srq ibsrq; > > + int offset; > > + spinlock_t lock; /* SRQ lock. */ > > + int wqe_cnt; > > + int wqe_size; > > + int max_gs; > > + struct ib_umem *umem; > > + struct pvrdma_ring_state *ring; > > + struct pvrdma_page_dir pdir; > > + u32 srq_handle; > > + int npages; > > + atomic_t refcnt; > > I think that new fashion is to use refcnt_t for such counts. > Sure, I'll change this to refcount_t. > > + /* Check if SRQ is supported by backend */ > > + if (dev->dsr->caps.max_srq > 0) { > > + dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq, sizeof(void *), > > + GFP_KERNEL); > > Shouldn't the "sizeof(void *)" need to be "sizeof(struct pvrdma_srq *)" Fixed. > > + if (!dev->srq_tbl) > > + goto err_qp_free; > > + } else { > > + dev->srq_tbl = NULL; > > dev is allocated with ib_alloc_device() and srq_tbl will be zero by default. > Removed. > > +static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type) > > +{ > > + struct pvrdma_srq *srq; > > + unsigned long flags; > > + > > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > > + srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq]; > > Are you sure that "dev->srq_tb != NULL && dev->dsr->caps.max_srq > 0" always here? > Added a check for srq_tbl. > > +static void pvrdma_free_srq(struct pvrdma_dev *dev, struct pvrdma_srq *srq) > > +{ > > + unsigned long flags; > > + > > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > > + dev->srq_tbl[srq->srq_handle] = NULL; > > Again, are you sure that dev->srq_tbl exist? > Here we assumed that the verbs layer will only call down to us to destroy a valid SRQ. If an SRQ has been created, then the srq_tbl should exist. I can add a check here if that's not behaviour we should rely on. > > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > > +{ > > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > > + union pvrdma_cmd_req req; > > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > > + int ret; > > + > > + /* No support for resizing SRQs. */ > > Better to write that you support, so future additions to ib_srq_attr_mask > won't need to change your code. > I'm a little unclear on what you mean here. Do you mean that we should explicitly check for the only attribute we support (IB_SRQ_LIMIT) and fail otherwise? > > + if (attr_mask & IB_SRQ_MAX_WR) > > + return -EINVAL; Thanks for the review! Bryan -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Nov 02, 2017 at 10:52:25AM -0700, Bryan Tan wrote: > On Mon, Oct 30, 2017 at 08:03:17AM +0200, Leon Romanovsky wrote: > > > +struct pvrdma_srq { > > > + struct ib_srq ibsrq; > > > + int offset; > > > + spinlock_t lock; /* SRQ lock. */ > > > + int wqe_cnt; > > > + int wqe_size; > > > + int max_gs; > > > + struct ib_umem *umem; > > > + struct pvrdma_ring_state *ring; > > > + struct pvrdma_page_dir pdir; > > > + u32 srq_handle; > > > + int npages; > > > + atomic_t refcnt; > > > > I think that new fashion is to use refcnt_t for such counts. > > > > Sure, I'll change this to refcount_t. > > > > + /* Check if SRQ is supported by backend */ > > > + if (dev->dsr->caps.max_srq > 0) { > > > + dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq, sizeof(void *), > > > + GFP_KERNEL); > > > > Shouldn't the "sizeof(void *)" need to be "sizeof(struct pvrdma_srq *)" > > Fixed. > > > > + if (!dev->srq_tbl) > > > + goto err_qp_free; > > > + } else { > > > + dev->srq_tbl = NULL; > > > > dev is allocated with ib_alloc_device() and srq_tbl will be zero by default. > > > > Removed. > > > > +static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type) > > > +{ > > > + struct pvrdma_srq *srq; > > > + unsigned long flags; > > > + > > > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > > > + srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq]; > > > > Are you sure that "dev->srq_tb != NULL && dev->dsr->caps.max_srq > 0" always here? > > > > Added a check for srq_tbl. > > > > +static void pvrdma_free_srq(struct pvrdma_dev *dev, struct pvrdma_srq *srq) > > > +{ > > > + unsigned long flags; > > > + > > > + spin_lock_irqsave(&dev->srq_tbl_lock, flags); > > > + dev->srq_tbl[srq->srq_handle] = NULL; > > > > Again, are you sure that dev->srq_tbl exist? > > > > Here we assumed that the verbs layer will only call down to us to > destroy a valid SRQ. If an SRQ has been created, then the srq_tbl > should exist. I can add a check here if that's not behaviour we > should rely on. By saying "verbs layer", are you referring to libibverbs or IB/core? > > > > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > > > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > > > +{ > > > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > > > + union pvrdma_cmd_req req; > > > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > > > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > > > + int ret; > > > + > > > + /* No support for resizing SRQs. */ > > > > Better to write that you support, so future additions to ib_srq_attr_mask > > won't need to change your code. > > > > I'm a little unclear on what you mean here. Do you mean that we > should explicitly check for the only attribute we support > (IB_SRQ_LIMIT) and fail otherwise? Yes, it will ensure that your driver doesn't get untested input. > > > > + if (attr_mask & IB_SRQ_MAX_WR) > > > + return -EINVAL; > > > Thanks for the review! > Bryan > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Nov 02, 2017 at 08:02:47PM +0200, Leon Romanovsky wrote: > > Here we assumed that the verbs layer will only call down to us to > > destroy a valid SRQ. If an SRQ has been created, then the srq_tbl > > should exist. I can add a check here if that's not behaviour we > > should rely on. > > By saying "verbs layer", are you referring to libibverbs or IB/core? > IB/core. Looking at core/verbs.c, it seems like the parameters are assumed to be good. I'm unclear on who is responsible for ensuring or checking for good inputs, kernel clients, IB/core, or the driver. > > > > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > > > > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > > > > +{ > > > > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > > > > + union pvrdma_cmd_req req; > > > > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > > > > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > > > > + int ret; > > > > + > > > > + /* No support for resizing SRQs. */ > > > > > > Better to write that you support, so future additions to ib_srq_attr_mask > > > won't need to change your code. > > > > > > > I'm a little unclear on what you mean here. Do you mean that we > > should explicitly check for the only attribute we support > > (IB_SRQ_LIMIT) and fail otherwise? > > Yes, it will ensure that your driver doesn't get untested input. > Sure, I will do that then. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Nov 02, 2017 at 11:23:25AM -0700, Bryan Tan wrote: > On Thu, Nov 02, 2017 at 08:02:47PM +0200, Leon Romanovsky wrote: > > > Here we assumed that the verbs layer will only call down to us to > > > destroy a valid SRQ. If an SRQ has been created, then the srq_tbl > > > should exist. I can add a check here if that's not behaviour we > > > should rely on. > > > > By saying "verbs layer", are you referring to libibverbs or IB/core? > > > > IB/core. Looking at core/verbs.c, it seems like the parameters are > assumed to be good. I'm unclear on who is responsible for ensuring > or checking for good inputs, kernel clients, IB/core, or the driver. The first one who can recognize the wrong parameters - fail early. So if IB/core provides to you clean input, there is no need to check it again. > > > > > > +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, > > > > > + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) > > > > > +{ > > > > > + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); > > > > > + union pvrdma_cmd_req req; > > > > > + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; > > > > > + struct pvrdma_dev *dev = to_vdev(ibsrq->device); > > > > > + int ret; > > > > > + > > > > > + /* No support for resizing SRQs. */ > > > > > > > > Better to write that you support, so future additions to ib_srq_attr_mask > > > > won't need to change your code. > > > > > > > > > > I'm a little unclear on what you mean here. Do you mean that we > > > should explicitly check for the only attribute we support > > > (IB_SRQ_LIMIT) and fail otherwise? > > > > Yes, it will ensure that your driver doesn't get untested input. > > > > Sure, I will do that then. > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rdma" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Oct 31, 2017 at 04:02:07PM +0200, Yuval Shaia wrote: > > + if (is_srq) { > > + if (dev->dsr->caps.max_srq == 0) { > > Let us be consist, so if in function pvrdma_init_device the assumption is > that max_srq can have negative value then also check it here. Or on the > other hands if not (and i assume it is not) then modify the check on > pvrdma_init_device. > > Also, suggesting: > if (is_srq && !dev->dsr->caps.max_srq) > Or (if you do not accept my comment): > if (is_srq && dev->dsr->caps.max_srq == 0) > Sure. I'll take your comment and fix the check in pvrdma_register_device as well. > > + if (init_attr->srq_type != IB_SRQT_BASIC) > > + return ERR_PTR(-EINVAL); > > + > > + if (init_attr->attr.max_wr > dev->dsr->caps.max_srq_wr || > > + init_attr->attr.max_sge > dev->dsr->caps.max_srq_sge) > > Again - let us be consist, if we got warning in pvrdma_create_qp we should > get one here as well. > Added a warning here, as well as in the srq_type check above. > > + if (!(pd->uobject && udata)) { > > + /* No support for kernel clients. */ > > + ret = -EOPNOTSUPP; > > + goto err_srq; > > + } > > Why this check cannot be performed before the above initialization?? > Also, suggesting a debug message here. > Moved check to beginning, and added a warning here. Thanks for the review! Bryan -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Nov 02, 2017 at 08:29:32PM +0200, Leon Romanovsky wrote: > On Thu, Nov 02, 2017 at 11:23:25AM -0700, Bryan Tan wrote: > > On Thu, Nov 02, 2017 at 08:02:47PM +0200, Leon Romanovsky wrote: > > > > Here we assumed that the verbs layer will only call down to us to > > > > destroy a valid SRQ. If an SRQ has been created, then the srq_tbl > > > > should exist. I can add a check here if that's not behaviour we > > > > should rely on. > > > > > > By saying "verbs layer", are you referring to libibverbs or IB/core? > > > > > > > IB/core. Looking at core/verbs.c, it seems like the parameters are > > assumed to be good. I'm unclear on who is responsible for ensuring > > or checking for good inputs, kernel clients, IB/core, or the driver. > > The first one who can recognize the wrong parameters - fail early. > So if IB/core provides to you clean input, there is no need to check it > again. > I'll add the check then, given that IB/core doesn't make any guarantees. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/infiniband/hw/vmw_pvrdma/Makefile b/drivers/infiniband/hw/vmw_pvrdma/Makefile index 0194ed1..2f52e0a 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/Makefile +++ b/drivers/infiniband/hw/vmw_pvrdma/Makefile @@ -1,3 +1,3 @@ obj-$(CONFIG_INFINIBAND_VMWARE_PVRDMA) += vmw_pvrdma.o -vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_verbs.o +vmw_pvrdma-y := pvrdma_cmd.o pvrdma_cq.o pvrdma_doorbell.o pvrdma_main.o pvrdma_misc.o pvrdma_mr.o pvrdma_qp.o pvrdma_srq.o pvrdma_verbs.o diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h index 984aa34..f53a446 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma.h @@ -162,6 +162,22 @@ struct pvrdma_ah { struct pvrdma_av av; }; +struct pvrdma_srq { + struct ib_srq ibsrq; + int offset; + spinlock_t lock; /* SRQ lock. */ + int wqe_cnt; + int wqe_size; + int max_gs; + struct ib_umem *umem; + struct pvrdma_ring_state *ring; + struct pvrdma_page_dir pdir; + u32 srq_handle; + int npages; + atomic_t refcnt; + wait_queue_head_t wait; +}; + struct pvrdma_qp { struct ib_qp ibqp; u32 qp_handle; @@ -171,6 +187,7 @@ struct pvrdma_qp { struct ib_umem *rumem; struct ib_umem *sumem; struct pvrdma_page_dir pdir; + struct pvrdma_srq *srq; int npages; int npages_send; int npages_recv; @@ -210,6 +227,8 @@ struct pvrdma_dev { struct pvrdma_page_dir cq_pdir; struct pvrdma_cq **cq_tbl; spinlock_t cq_tbl_lock; + struct pvrdma_srq **srq_tbl; + spinlock_t srq_tbl_lock; struct pvrdma_qp **qp_tbl; spinlock_t qp_tbl_lock; struct pvrdma_uar_table uar_table; @@ -221,6 +240,7 @@ struct pvrdma_dev { bool ib_active; atomic_t num_qps; atomic_t num_cqs; + atomic_t num_srqs; atomic_t num_pds; atomic_t num_ahs; @@ -256,6 +276,11 @@ static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq) return container_of(ibcq, struct pvrdma_cq, ibcq); } +static inline struct pvrdma_srq *to_vsrq(struct ib_srq *ibsrq) +{ + return container_of(ibsrq, struct pvrdma_srq, ibsrq); +} + static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr) { return container_of(ibmr, struct pvrdma_user_mr, ibmr); diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h index df0a6b5..6fd5a8f 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h @@ -339,6 +339,10 @@ enum { PVRDMA_CMD_DESTROY_UC, PVRDMA_CMD_CREATE_BIND, PVRDMA_CMD_DESTROY_BIND, + PVRDMA_CMD_CREATE_SRQ, + PVRDMA_CMD_MODIFY_SRQ, + PVRDMA_CMD_QUERY_SRQ, + PVRDMA_CMD_DESTROY_SRQ, PVRDMA_CMD_MAX, }; @@ -361,6 +365,10 @@ enum { PVRDMA_CMD_DESTROY_UC_RESP_NOOP, PVRDMA_CMD_CREATE_BIND_RESP_NOOP, PVRDMA_CMD_DESTROY_BIND_RESP_NOOP, + PVRDMA_CMD_CREATE_SRQ_RESP, + PVRDMA_CMD_MODIFY_SRQ_RESP, + PVRDMA_CMD_QUERY_SRQ_RESP, + PVRDMA_CMD_DESTROY_SRQ_RESP, PVRDMA_CMD_MAX_RESP, }; @@ -495,6 +503,46 @@ struct pvrdma_cmd_destroy_cq { u8 reserved[4]; }; +struct pvrdma_cmd_create_srq { + struct pvrdma_cmd_hdr hdr; + u64 pdir_dma; + u32 pd_handle; + u32 nchunks; + struct pvrdma_srq_attr attrs; + u8 srq_type; + u8 reserved[7]; +}; + +struct pvrdma_cmd_create_srq_resp { + struct pvrdma_cmd_resp_hdr hdr; + u32 srqn; + u8 reserved[4]; +}; + +struct pvrdma_cmd_modify_srq { + struct pvrdma_cmd_hdr hdr; + u32 srq_handle; + u32 attr_mask; + struct pvrdma_srq_attr attrs; +}; + +struct pvrdma_cmd_query_srq { + struct pvrdma_cmd_hdr hdr; + u32 srq_handle; + u8 reserved[4]; +}; + +struct pvrdma_cmd_query_srq_resp { + struct pvrdma_cmd_resp_hdr hdr; + struct pvrdma_srq_attr attrs; +}; + +struct pvrdma_cmd_destroy_srq { + struct pvrdma_cmd_hdr hdr; + u32 srq_handle; + u8 reserved[4]; +}; + struct pvrdma_cmd_create_qp { struct pvrdma_cmd_hdr hdr; u64 pdir_dma; @@ -594,6 +642,10 @@ struct pvrdma_cmd_destroy_bind { struct pvrdma_cmd_destroy_qp destroy_qp; struct pvrdma_cmd_create_bind create_bind; struct pvrdma_cmd_destroy_bind destroy_bind; + struct pvrdma_cmd_create_srq create_srq; + struct pvrdma_cmd_modify_srq modify_srq; + struct pvrdma_cmd_query_srq query_srq; + struct pvrdma_cmd_destroy_srq destroy_srq; }; union pvrdma_cmd_resp { @@ -608,6 +660,8 @@ struct pvrdma_cmd_destroy_bind { struct pvrdma_cmd_create_qp_resp create_qp_resp; struct pvrdma_cmd_query_qp_resp query_qp_resp; struct pvrdma_cmd_destroy_qp_resp destroy_qp_resp; + struct pvrdma_cmd_create_srq_resp create_srq_resp; + struct pvrdma_cmd_query_srq_resp query_srq_resp; }; #endif /* __PVRDMA_DEV_API_H__ */ diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c index 6ce709a..7ce6687 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c @@ -118,6 +118,7 @@ static int pvrdma_init_device(struct pvrdma_dev *dev) spin_lock_init(&dev->cmd_lock); sema_init(&dev->cmd_sema, 1); atomic_set(&dev->num_qps, 0); + atomic_set(&dev->num_srqs, 0); atomic_set(&dev->num_cqs, 0); atomic_set(&dev->num_pds, 0); atomic_set(&dev->num_ahs, 0); @@ -195,6 +196,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | (1ull << IB_USER_VERBS_CMD_QUERY_QP) | (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | + (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | + (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | + (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | + (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | + (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV) | (1ull << IB_USER_VERBS_CMD_POST_SEND) | (1ull << IB_USER_VERBS_CMD_POST_RECV) | (1ull << IB_USER_VERBS_CMD_CREATE_AH) | @@ -227,6 +233,11 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) dev->ib_dev.destroy_cq = pvrdma_destroy_cq; dev->ib_dev.poll_cq = pvrdma_poll_cq; dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq; + dev->ib_dev.create_srq = pvrdma_create_srq; + dev->ib_dev.modify_srq = pvrdma_modify_srq; + dev->ib_dev.query_srq = pvrdma_query_srq; + dev->ib_dev.destroy_srq = pvrdma_destroy_srq; + dev->ib_dev.post_srq_recv = pvrdma_post_srq_recv; dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr; dev->ib_dev.reg_user_mr = pvrdma_reg_user_mr; dev->ib_dev.dereg_mr = pvrdma_dereg_mr; @@ -254,9 +265,20 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) goto err_cq_free; spin_lock_init(&dev->qp_tbl_lock); + /* Check if SRQ is supported by backend */ + if (dev->dsr->caps.max_srq > 0) { + dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq, sizeof(void *), + GFP_KERNEL); + if (!dev->srq_tbl) + goto err_qp_free; + } else { + dev->srq_tbl = NULL; + } + spin_lock_init(&dev->srq_tbl_lock); + ret = ib_register_device(&dev->ib_dev, NULL); if (ret) - goto err_qp_free; + goto err_srq_free; for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) { ret = device_create_file(&dev->ib_dev.dev, @@ -271,6 +293,8 @@ static int pvrdma_register_device(struct pvrdma_dev *dev) err_class: ib_unregister_device(&dev->ib_dev); +err_srq_free: + kfree(dev->srq_tbl); err_qp_free: kfree(dev->qp_tbl); err_cq_free: @@ -353,6 +377,32 @@ static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type) } } +static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type) +{ + struct pvrdma_srq *srq; + unsigned long flags; + + spin_lock_irqsave(&dev->srq_tbl_lock, flags); + srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq]; + if (srq) + atomic_inc(&srq->refcnt); + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); + + if (srq && srq->ibsrq.event_handler) { + struct ib_srq *ibsrq = &srq->ibsrq; + struct ib_event e; + + e.device = ibsrq->device; + e.element.srq = ibsrq; + e.event = type; /* 1:1 mapping for now. */ + ibsrq->event_handler(&e, ibsrq->srq_context); + } + if (srq) { + if (atomic_dec_and_test(&srq->refcnt)) + wake_up(&srq->wait); + } +} + static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port, enum ib_event_type event) { @@ -423,6 +473,7 @@ static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id) case PVRDMA_EVENT_SRQ_ERR: case PVRDMA_EVENT_SRQ_LIMIT_REACHED: + pvrdma_srq_event(dev, eqe->info, eqe->type); break; case PVRDMA_EVENT_PORT_ACTIVE: @@ -1059,6 +1110,7 @@ static void pvrdma_pci_remove(struct pci_dev *pdev) iounmap(dev->regs); kfree(dev->sgid_tbl); kfree(dev->cq_tbl); + kfree(dev->srq_tbl); kfree(dev->qp_tbl); pvrdma_uar_table_cleanup(dev); iounmap(dev->driver_uar.map); diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c index ed34d5a..4c4a9d1 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c @@ -198,6 +198,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, struct pvrdma_create_qp ucmd; unsigned long flags; int ret; + bool is_srq = !!init_attr->srq; if (init_attr->create_flags) { dev_warn(&dev->pdev->dev, @@ -214,6 +215,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, return ERR_PTR(-EINVAL); } + if (is_srq) { + if (dev->dsr->caps.max_srq == 0) { + dev_warn(&dev->pdev->dev, + "SRQs not supported by device\n"); + return ERR_PTR(-EINVAL); + } + } + if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp)) return ERR_PTR(-ENOMEM); @@ -252,26 +261,36 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, goto err_qp; } - /* set qp->sq.wqe_cnt, shift, buf_size.. */ - qp->rumem = ib_umem_get(pd->uobject->context, - ucmd.rbuf_addr, - ucmd.rbuf_size, 0, 0); - if (IS_ERR(qp->rumem)) { - ret = PTR_ERR(qp->rumem); - goto err_qp; + if (!is_srq) { + /* set qp->sq.wqe_cnt, shift, buf_size.. */ + qp->rumem = ib_umem_get(pd->uobject->context, + ucmd.rbuf_addr, + ucmd.rbuf_size, 0, 0); + if (IS_ERR(qp->rumem)) { + ret = PTR_ERR(qp->rumem); + goto err_qp; + } + qp->srq = NULL; + } else { + qp->rumem = NULL; + qp->srq = to_vsrq(init_attr->srq); } qp->sumem = ib_umem_get(pd->uobject->context, ucmd.sbuf_addr, ucmd.sbuf_size, 0, 0); if (IS_ERR(qp->sumem)) { - ib_umem_release(qp->rumem); + if (!is_srq) + ib_umem_release(qp->rumem); ret = PTR_ERR(qp->sumem); goto err_qp; } qp->npages_send = ib_umem_page_count(qp->sumem); - qp->npages_recv = ib_umem_page_count(qp->rumem); + if (!is_srq) + qp->npages_recv = ib_umem_page_count(qp->rumem); + else + qp->npages_recv = 0; qp->npages = qp->npages_send + qp->npages_recv; } else { qp->is_kernel = true; @@ -312,12 +331,14 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, if (!qp->is_kernel) { pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0); - pvrdma_page_dir_insert_umem(&qp->pdir, qp->rumem, - qp->npages_send); + if (!is_srq) + pvrdma_page_dir_insert_umem(&qp->pdir, + qp->rumem, + qp->npages_send); } else { /* Ring state is always the first page. */ qp->sq.ring = qp->pdir.pages[0]; - qp->rq.ring = &qp->sq.ring[1]; + qp->rq.ring = is_srq ? NULL : &qp->sq.ring[1]; } break; default: @@ -333,6 +354,10 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, cmd->pd_handle = to_vpd(pd)->pd_handle; cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle; cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle; + if (is_srq) + cmd->srq_handle = to_vsrq(init_attr->srq)->srq_handle; + else + cmd->srq_handle = 0; cmd->max_send_wr = init_attr->cap.max_send_wr; cmd->max_recv_wr = init_attr->cap.max_recv_wr; cmd->max_send_sge = init_attr->cap.max_send_sge; @@ -340,6 +365,8 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, cmd->max_inline_data = init_attr->cap.max_inline_data; cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0; cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type); + cmd->is_srq = is_srq; + cmd->lkey = 0; cmd->access_flags = IB_ACCESS_LOCAL_WRITE; cmd->total_chunks = qp->npages; cmd->send_chunks = qp->npages_send - PVRDMA_QP_NUM_HEADER_PAGES; @@ -815,6 +842,12 @@ int pvrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr, return -EINVAL; } + if (qp->srq) { + dev_warn(&dev->pdev->dev, "QP associated with SRQ\n"); + *bad_wr = wr; + return -EINVAL; + } + spin_lock_irqsave(&qp->rq.lock, flags); while (wr) { diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c new file mode 100644 index 0000000..59d4595 --- /dev/null +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_srq.c @@ -0,0 +1,311 @@ +/* + * Copyright (c) 2016-2017 VMware, Inc. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of EITHER the GNU General Public License + * version 2 as published by the Free Software Foundation or the BSD + * 2-Clause License. This program is distributed in the hope that it + * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED + * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. + * See the GNU General Public License version 2 for more details at + * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html. + * + * You should have received a copy of the GNU General Public License + * along with this program available in the file COPYING in the main + * directory of this source tree. + * + * The BSD 2-Clause License + * + * Redistribution and use in source and binary forms, with or + * without modification, are permitted provided that the following + * conditions are met: + * + * - Redistributions of source code must retain the above + * copyright notice, this list of conditions and the following + * disclaimer. + * + * - Redistributions in binary form must reproduce the above + * copyright notice, this list of conditions and the following + * disclaimer in the documentation and/or other materials + * provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, + * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED + * OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include <asm/page.h> +#include <linux/io.h> +#include <linux/wait.h> +#include <rdma/ib_addr.h> +#include <rdma/ib_smi.h> +#include <rdma/ib_user_verbs.h> + +#include "pvrdma.h" + +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr) +{ + /* No support for kernel clients. */ + return -EOPNOTSUPP; +} + +/** + * pvrdma_query_srq - query shared receive queue + * @ibsrq: the shared receive queue to query + * @srq_attr: attributes to query and return to client + * + * @return: 0 for success, otherwise returns an errno. + */ +int pvrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr) +{ + struct pvrdma_dev *dev = to_vdev(ibsrq->device); + struct pvrdma_srq *srq = to_vsrq(ibsrq); + union pvrdma_cmd_req req; + union pvrdma_cmd_resp rsp; + struct pvrdma_cmd_query_srq *cmd = &req.query_srq; + struct pvrdma_cmd_query_srq_resp *resp = &rsp.query_srq_resp; + int ret; + + memset(cmd, 0, sizeof(*cmd)); + cmd->hdr.cmd = PVRDMA_CMD_QUERY_SRQ; + cmd->srq_handle = srq->srq_handle; + + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_SRQ_RESP); + if (ret < 0) { + dev_warn(&dev->pdev->dev, + "could not query shared receive queue, error: %d\n", + ret); + return -EINVAL; + } + + srq_attr->srq_limit = resp->attrs.srq_limit; + srq_attr->max_wr = resp->attrs.max_wr; + srq_attr->max_sge = resp->attrs.max_sge; + + return 0; +} + +/** + * pvrdma_create_srq - create shared receive queue + * @pd: protection domain + * @init_attr: shared receive queue attributes + * @udata: user data + * + * @return: the ib_srq pointer on success, otherwise returns an errno. + */ +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, + struct ib_srq_init_attr *init_attr, + struct ib_udata *udata) +{ + struct pvrdma_srq *srq = NULL; + struct pvrdma_dev *dev = to_vdev(pd->device); + union pvrdma_cmd_req req; + union pvrdma_cmd_resp rsp; + struct pvrdma_cmd_create_srq *cmd = &req.create_srq; + struct pvrdma_cmd_create_srq_resp *resp = &rsp.create_srq_resp; + struct pvrdma_create_srq ucmd; + unsigned long flags; + int ret; + + if (init_attr->srq_type != IB_SRQT_BASIC) + return ERR_PTR(-EINVAL); + + if (init_attr->attr.max_wr > dev->dsr->caps.max_srq_wr || + init_attr->attr.max_sge > dev->dsr->caps.max_srq_sge) + return ERR_PTR(-EINVAL); + + if (!atomic_add_unless(&dev->num_srqs, 1, dev->dsr->caps.max_srq)) + return ERR_PTR(-ENOMEM); + + srq = kmalloc(sizeof(*srq), GFP_KERNEL); + if (!srq) { + ret = -ENOMEM; + goto err_srq; + } + + spin_lock_init(&srq->lock); + atomic_set(&srq->refcnt, 1); + init_waitqueue_head(&srq->wait); + + if (!(pd->uobject && udata)) { + /* No support for kernel clients. */ + ret = -EOPNOTSUPP; + goto err_srq; + } + + dev_dbg(&dev->pdev->dev, + "create shared receive queue from user space\n"); + + if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) { + ret = -EFAULT; + goto err_srq; + } + + srq->umem = ib_umem_get(pd->uobject->context, + ucmd.buf_addr, + ucmd.buf_size, 0, 0); + if (IS_ERR(srq->umem)) { + ret = PTR_ERR(srq->umem); + goto err_srq; + } + + srq->npages = ib_umem_page_count(srq->umem); + + if (srq->npages < 0 || srq->npages > PVRDMA_PAGE_DIR_MAX_PAGES) { + dev_warn(&dev->pdev->dev, + "overflow pages in shared receive queue\n"); + ret = -EINVAL; + goto err_umem; + } + + ret = pvrdma_page_dir_init(dev, &srq->pdir, srq->npages, false); + if (ret) { + dev_warn(&dev->pdev->dev, + "could not allocate page directory\n"); + goto err_umem; + } + + pvrdma_page_dir_insert_umem(&srq->pdir, srq->umem, 0); + + memset(cmd, 0, sizeof(*cmd)); + cmd->hdr.cmd = PVRDMA_CMD_CREATE_SRQ; + cmd->srq_type = init_attr->srq_type; + cmd->nchunks = srq->npages; + cmd->pd_handle = to_vpd(pd)->pd_handle; + cmd->attrs.max_wr = init_attr->attr.max_wr; + cmd->attrs.max_sge = init_attr->attr.max_sge; + cmd->attrs.srq_limit = init_attr->attr.srq_limit; + cmd->pdir_dma = srq->pdir.dir_dma; + + ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_SRQ_RESP); + if (ret < 0) { + dev_warn(&dev->pdev->dev, + "could not create shared receive queue, error: %d\n", + ret); + goto err_page_dir; + } + + srq->srq_handle = resp->srqn; + spin_lock_irqsave(&dev->srq_tbl_lock, flags); + dev->srq_tbl[srq->srq_handle % dev->dsr->caps.max_srq] = srq; + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); + + /* Copy udata back. */ + if (ib_copy_to_udata(udata, &srq->srq_handle, sizeof(__u32))) { + dev_warn(&dev->pdev->dev, "failed to copy back udata\n"); + pvrdma_destroy_srq(&srq->ibsrq); + return ERR_PTR(-EINVAL); + } + + return &srq->ibsrq; + +err_page_dir: + pvrdma_page_dir_cleanup(dev, &srq->pdir); +err_umem: + ib_umem_release(srq->umem); +err_srq: + kfree(srq); + atomic_dec(&dev->num_srqs); + + return ERR_PTR(ret); +} + +static void pvrdma_free_srq(struct pvrdma_dev *dev, struct pvrdma_srq *srq) +{ + unsigned long flags; + + spin_lock_irqsave(&dev->srq_tbl_lock, flags); + dev->srq_tbl[srq->srq_handle] = NULL; + spin_unlock_irqrestore(&dev->srq_tbl_lock, flags); + + atomic_dec(&srq->refcnt); + wait_event(srq->wait, !atomic_read(&srq->refcnt)); + + /* There is no support for kernel clients, so this is safe. */ + ib_umem_release(srq->umem); + + pvrdma_page_dir_cleanup(dev, &srq->pdir); + + kfree(srq); + + atomic_dec(&dev->num_srqs); +} + +/** + * pvrdma_destroy_srq - destroy shared receive queue + * @srq: the shared receive queue to destroy + * + * @return: 0 for success. + */ +int pvrdma_destroy_srq(struct ib_srq *srq) +{ + struct pvrdma_srq *vsrq = to_vsrq(srq); + union pvrdma_cmd_req req; + struct pvrdma_cmd_destroy_srq *cmd = &req.destroy_srq; + struct pvrdma_dev *dev = to_vdev(srq->device); + int ret; + + memset(cmd, 0, sizeof(*cmd)); + cmd->hdr.cmd = PVRDMA_CMD_DESTROY_SRQ; + cmd->srq_handle = vsrq->srq_handle; + + ret = pvrdma_cmd_post(dev, &req, NULL, 0); + if (ret < 0) + dev_warn(&dev->pdev->dev, + "destroy shared receive queue failed, error: %d\n", + ret); + + pvrdma_free_srq(dev, vsrq); + + return 0; +} + +/** + * pvrdma_modify_srq - modify shared receive queue attributes + * @ibsrq: the shared receive queue to modify + * @attr: the shared receive queue's new attributes + * @attr_mask: attributes mask + * @udata: user data + * + * @returns 0 on success, otherwise returns an errno. + */ +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata) +{ + struct pvrdma_srq *vsrq = to_vsrq(ibsrq); + union pvrdma_cmd_req req; + struct pvrdma_cmd_modify_srq *cmd = &req.modify_srq; + struct pvrdma_dev *dev = to_vdev(ibsrq->device); + int ret; + + /* No support for resizing SRQs. */ + if (attr_mask & IB_SRQ_MAX_WR) + return -EINVAL; + + memset(cmd, 0, sizeof(*cmd)); + cmd->hdr.cmd = PVRDMA_CMD_MODIFY_SRQ; + cmd->srq_handle = vsrq->srq_handle; + cmd->attrs.srq_limit = attr->srq_limit; + cmd->attr_mask = attr_mask; + + ret = pvrdma_cmd_post(dev, &req, NULL, 0); + if (ret < 0) { + dev_warn(&dev->pdev->dev, + "could not modify shared receive queue, error: %d\n", + ret); + + return -EINVAL; + } + + return ret; +} diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c index 48776f5..16b9661 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.c @@ -85,6 +85,9 @@ int pvrdma_query_device(struct ib_device *ibdev, props->max_sge = dev->dsr->caps.max_sge; props->max_sge_rd = PVRDMA_GET_CAP(dev, dev->dsr->caps.max_sge, dev->dsr->caps.max_sge_rd); + props->max_srq = dev->dsr->caps.max_srq; + props->max_srq_wr = dev->dsr->caps.max_srq_wr; + props->max_srq_sge = dev->dsr->caps.max_srq_sge; props->max_cq = dev->dsr->caps.max_cq; props->max_cqe = dev->dsr->caps.max_cqe; props->max_mr = dev->dsr->caps.max_mr; diff --git a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h index 002a9b0..b7b2572 100644 --- a/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h +++ b/drivers/infiniband/hw/vmw_pvrdma/pvrdma_verbs.h @@ -324,6 +324,13 @@ enum pvrdma_mw_type { PVRDMA_MW_TYPE_2 = 2, }; +struct pvrdma_srq_attr { + u32 max_wr; + u32 max_sge; + u32 srq_limit; + u32 reserved; +}; + struct pvrdma_qp_attr { enum pvrdma_qp_state qp_state; enum pvrdma_qp_state cur_qp_state; @@ -420,6 +427,17 @@ int pvrdma_resize_cq(struct ib_cq *ibcq, int entries, struct ib_ah *pvrdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, struct ib_udata *udata); int pvrdma_destroy_ah(struct ib_ah *ah); + +struct ib_srq *pvrdma_create_srq(struct ib_pd *pd, + struct ib_srq_init_attr *init_attr, + struct ib_udata *udata); +int pvrdma_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, + enum ib_srq_attr_mask attr_mask, struct ib_udata *udata); +int pvrdma_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr); +int pvrdma_destroy_srq(struct ib_srq *srq); +int pvrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr, + struct ib_recv_wr **bad_wr); + struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata); diff --git a/include/uapi/rdma/vmw_pvrdma-abi.h b/include/uapi/rdma/vmw_pvrdma-abi.h index c6569b0..846c6f4 100644 --- a/include/uapi/rdma/vmw_pvrdma-abi.h +++ b/include/uapi/rdma/vmw_pvrdma-abi.h @@ -158,6 +158,8 @@ struct pvrdma_resize_cq { struct pvrdma_create_srq { __u64 buf_addr; + __u32 buf_size; + __u32 reserved; }; struct pvrdma_create_srq_resp {