Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 8b10ba78 authored by Bryan Tan's avatar Bryan Tan Committed by Doug Ledford
Browse files

RDMA/vmw_pvrdma: Add shared receive queue support 



Add the required functions needed to support SRQs. Currently, kernel
clients are not supported. SRQs will only be available in userspace.

Reviewed-by: default avatarAdit Ranadive <aditr@vmware.com>
Reviewed-by: default avatarAditya Sarwade <asarwade@vmware.com>
Reviewed-by: default avatarJorgen Hansen <jhansen@vmware.com>
Reviewed-by: default avatarNitish Bhat <bnitish@vmware.com>
Signed-off-by: default avatarBryan Tan <bryantan@vmware.com>
Reviewed-by: default avatarYuval Shaia <yuval.shaia@oracle.com>
Signed-off-by: default avatarDoug Ledford <dledford@redhat.com>
parent cb9fd89f

drivers/infiniband/hw/vmw_pvrdma/Makefile

+1 −1
Original line number Diff line number Diff line 
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

drivers/infiniband/hw/vmw_pvrdma/pvrdma.h

+25 −0
Original line number Diff line number Diff line
@@ -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;

	refcount_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);

drivers/infiniband/hw/vmw_pvrdma/pvrdma_dev_api.h

+54 −0
Original line number Diff line number Diff line
@@ -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 @@ union pvrdma_cmd_req { 
	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 @@ union pvrdma_cmd_resp { 
	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__ */

drivers/infiniband/hw/vmw_pvrdma/pvrdma_main.c

+58 −1
Original line number Diff line number Diff line
@@ -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);
@@ -254,9 +255,32 @@ 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) {

		dev->ib_dev.uverbs_cmd_mask |=

			(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);



		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->srq_tbl = kcalloc(dev->dsr->caps.max_srq,

				       sizeof(struct pvrdma_srq *),

				       GFP_KERNEL);

		if (!dev->srq_tbl)

			goto err_qp_free;

	}

	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 +295,8 @@ 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 +379,35 @@ 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);

	if (dev->srq_tbl)

		srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq];

	else

		srq = NULL;

	if (srq)

		refcount_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 (refcount_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 +478,7 @@ 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 +1115,7 @@ 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);

drivers/infiniband/hw/vmw_pvrdma/pvrdma_qp.c

+43 −12
Original line number Diff line number Diff line
@@ -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 +215,12 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, 
		return ERR_PTR(-EINVAL);

	}



	if (is_srq && !dev->dsr->caps.max_srq) {

		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);
@@ -252,6 +259,7 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, 
				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,
@@ -260,18 +268,27 @@ struct ib_qp *pvrdma_create_qp(struct ib_pd *pd, 
					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)) {

				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);

			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 +329,14 @@ 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,

			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 +352,10 @@ 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 +363,8 @@ 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 +840,12 @@ 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) {