Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma updates from Jason Gunthorpe:
 "This has been a large cycle for RDMA, with several major patch series
  reworking parts of the core code.

   - Rework the so-called 'gid cache' and internal APIs to use a kref'd
     pointer to a struct instead of copying, push this upwards into the
     callers and add more stuff to the struct. The new design avoids
     some ugly races the old one suffered with. This is part of the
     namespace enablement work as the new struct is learning to be
     namespace aware.

   - Various uapi cleanups, moving more stuff to include/uapi and fixing
     some long standing bugs that have recently been discovered.

   - Driver updates for mlx5, mlx4 i40iw, rxe, cxgb4, hfi1, usnic,
     pvrdma, and hns

   - Provide max_send_sge and max_recv_sge attributes to better support
     HW where these values are asymmetric.

   - mlx5 user API 'devx' allows sending commands directly to the device
     FW, instead of trying to cram every wild and niche feature into the
     common API. Sort of like what GPU does.

   - Major write() and ioctl() API rework to cleanly support PCI device
     hot unplug and advance the ioctl conversion work

   - Sparse and compile warning cleanups

   - Add 'const' to the ib_poll_cq() signature, and permit a NULL
     'bad_wr', which is the common use case

   - Various patches to avoid high order allocations across the stack

   - SRQ support for cxgb4, hns and qedr

   - Changes to IPoIB to better follow the netdev model for working with
     struct net_device liftime"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (312 commits)
  Revert "net/smc: Replace ib_query_gid with rdma_get_gid_attr"
  RDMA/hns: Fix usage of bitmap allocation functions return values
  IB/core: Change filter function return type from int to bool
  IB/core: Update GID entries for netdevice whose mac address changes
  IB/core: Add default GIDs of the bond master netdev
  IB/core: Consider adding default GIDs of bond device
  IB/core: Delete lower netdevice default GID entries in bonding scenario
  IB/core: Avoid confusing del_netdev_default_ips
  IB/core: Add comment for change upper netevent handling
  qedr: Add user space support for SRQ
  qedr: Add support for kernel mode SRQ's
  qedr: Add wrapping generic structure for qpidr and adjust idr routines.
  IB/mlx5: Fix leaking stack memory to userspace
  Update the e-mail address of Bart Van Assche
  IB/ucm: Fix compiling ucm.c
  IB/uverbs: Do not check for device disassociation during ioctl
  IB/uverbs: Remove struct uverbs_root_spec and all supporting code
  IB/uverbs: Use uverbs_api to unmarshal ioctl commands
  IB/uverbs: Use uverbs_alloc for allocations
  IB/uverbs: Add a simple allocator to uverbs_attr_bundle
  ...

1 files changed, 549 insertions, 76 deletions

diff --git a/drivers/infiniband/hw/qedr/verbs.c b/drivers/infiniband/hw/qedr/verbs.c
index f07b8df96f43..8cc3df24e04e 100644
--- a/drivers/infiniband/hw/qedr/verbs.c
+++ b/drivers/infiniband/hw/qedr/verbs.c
@@ -51,6 +51,10 @@
 #include <rdma/qedr-abi.h>
 #include "qedr_roce_cm.h"
 
+#define QEDR_SRQ_WQE_ELEM_SIZE	sizeof(union rdma_srq_elm)
+#define	RDMA_MAX_SGE_PER_SRQ	(4)
+#define RDMA_MAX_SRQ_WQE_SIZE	(RDMA_MAX_SGE_PER_SRQ + 1)
+
 #define DB_ADDR_SHIFT(addr)		((addr) << DB_PWM_ADDR_OFFSET_SHIFT)
 
 static inline int qedr_ib_copy_to_udata(struct ib_udata *udata, void *src,
@@ -84,6 +88,19 @@ int qedr_iw_query_gid(struct ib_device *ibdev, u8 port,
 	return 0;
 }
 
+int qedr_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
+{
+	struct qedr_dev *dev = get_qedr_dev(ibsrq->device);
+	struct qedr_device_attr *qattr = &dev->attr;
+	struct qedr_srq *srq = get_qedr_srq(ibsrq);
+
+	srq_attr->srq_limit = srq->srq_limit;
+	srq_attr->max_wr = qattr->max_srq_wr;
+	srq_attr->max_sge = qattr->max_sge;
+
+	return 0;
+}
+
 int qedr_query_device(struct ib_device *ibdev,
 		      struct ib_device_attr *attr, struct ib_udata *udata)
 {
@@ -112,7 +129,8 @@ int qedr_query_device(struct ib_device *ibdev,
 	    IB_DEVICE_RC_RNR_NAK_GEN |
 	    IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_MGT_EXTENSIONS;
 
-	attr->max_sge = qattr->max_sge;
+	attr->max_send_sge = qattr->max_sge;
+	attr->max_recv_sge = qattr->max_sge;
 	attr->max_sge_rd = qattr->max_sge;
 	attr->max_cq = qattr->max_cq;
 	attr->max_cqe = qattr->max_cqe;
@@ -224,7 +242,7 @@ int qedr_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *attr)
 	attr->lmc = 0;
 	attr->sm_lid = 0;
 	attr->sm_sl = 0;
-	attr->port_cap_flags = IB_PORT_IP_BASED_GIDS;
+	attr->ip_gids = true;
 	if (rdma_protocol_iwarp(&dev->ibdev, 1)) {
 		attr->gid_tbl_len = 1;
 		attr->pkey_tbl_len = 1;
@@ -1075,27 +1093,19 @@ static inline int get_gid_info_from_table(struct ib_qp *ibqp,
 					  struct qed_rdma_modify_qp_in_params
 					  *qp_params)
 {
+	const struct ib_gid_attr *gid_attr;
 	enum rdma_network_type nw_type;
-	struct ib_gid_attr gid_attr;
 	const struct ib_global_route *grh = rdma_ah_read_grh(&attr->ah_attr);
-	union ib_gid gid;
 	u32 ipv4_addr;
-	int rc = 0;
 	int i;
 
-	rc = ib_get_cached_gid(ibqp->device,
-			       rdma_ah_get_port_num(&attr->ah_attr),
-			       grh->sgid_index, &gid, &gid_attr);
-	if (rc)
-		return rc;
-
-	qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr.ndev);
+	gid_attr = grh->sgid_attr;
+	qp_params->vlan_id = rdma_vlan_dev_vlan_id(gid_attr->ndev);
 
-	dev_put(gid_attr.ndev);
-	nw_type = ib_gid_to_network_type(gid_attr.gid_type, &gid);
+	nw_type = rdma_gid_attr_network_type(gid_attr);
 	switch (nw_type) {
 	case RDMA_NETWORK_IPV6:
-		memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+		memcpy(&qp_params->sgid.bytes[0], &gid_attr->gid.raw[0],
 		       sizeof(qp_params->sgid));
 		memcpy(&qp_params->dgid.bytes[0],
 		       &grh->dgid,
@@ -1105,7 +1115,7 @@ static inline int get_gid_info_from_table(struct ib_qp *ibqp,
 			  QED_ROCE_MODIFY_QP_VALID_ROCE_MODE, 1);
 		break;
 	case RDMA_NETWORK_IB:
-		memcpy(&qp_params->sgid.bytes[0], &gid.raw[0],
+		memcpy(&qp_params->sgid.bytes[0], &gid_attr->gid.raw[0],
 		       sizeof(qp_params->sgid));
 		memcpy(&qp_params->dgid.bytes[0],
 		       &grh->dgid,
@@ -1115,7 +1125,7 @@ static inline int get_gid_info_from_table(struct ib_qp *ibqp,
 	case RDMA_NETWORK_IPV4:
 		memset(&qp_params->sgid, 0, sizeof(qp_params->sgid));
 		memset(&qp_params->dgid, 0, sizeof(qp_params->dgid));
-		ipv4_addr = qedr_get_ipv4_from_gid(gid.raw);
+		ipv4_addr = qedr_get_ipv4_from_gid(gid_attr->gid.raw);
 		qp_params->sgid.ipv4_addr = ipv4_addr;
 		ipv4_addr =
 		    qedr_get_ipv4_from_gid(grh->dgid.raw);
@@ -1189,6 +1199,21 @@ static int qedr_check_qp_attrs(struct ib_pd *ibpd, struct qedr_dev *dev,
 	return 0;
 }
 
+static int qedr_copy_srq_uresp(struct qedr_dev *dev,
+			       struct qedr_srq *srq, struct ib_udata *udata)
+{
+	struct qedr_create_srq_uresp uresp = {};
+	int rc;
+
+	uresp.srq_id = srq->srq_id;
+
+	rc = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+	if (rc)
+		DP_ERR(dev, "create srq: problem copying data to user space\n");
+
+	return rc;
+}
+
 static void qedr_copy_rq_uresp(struct qedr_dev *dev,
 			       struct qedr_create_qp_uresp *uresp,
 			       struct qedr_qp *qp)
@@ -1255,13 +1280,18 @@ static void qedr_set_common_qp_params(struct qedr_dev *dev,
 	qp->state = QED_ROCE_QP_STATE_RESET;
 	qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false;
 	qp->sq_cq = get_qedr_cq(attrs->send_cq);
-	qp->rq_cq = get_qedr_cq(attrs->recv_cq);
 	qp->dev = dev;
-	qp->rq.max_sges = attrs->cap.max_recv_sge;
 
-	DP_DEBUG(dev, QEDR_MSG_QP,
-		 "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n",
-		 qp->rq.max_sges, qp->rq_cq->icid);
+	if (attrs->srq) {
+		qp->srq = get_qedr_srq(attrs->srq);
+	} else {
+		qp->rq_cq = get_qedr_cq(attrs->recv_cq);
+		qp->rq.max_sges = attrs->cap.max_recv_sge;
+		DP_DEBUG(dev, QEDR_MSG_QP,
+			 "RQ params:\trq_max_sges = %d, rq_cq_id = %d\n",
+			 qp->rq.max_sges, qp->rq_cq->icid);
+	}
+
 	DP_DEBUG(dev, QEDR_MSG_QP,
 		 "QP params:\tpd = %d, qp_type = %d, max_inline_data = %d, state = %d, signaled = %d, use_srq=%d\n",
 		 pd->pd_id, qp->qp_type, qp->max_inline_data,
@@ -1276,9 +1306,303 @@ static void qedr_set_roce_db_info(struct qedr_dev *dev, struct qedr_qp *qp)
 	qp->sq.db = dev->db_addr +
 		    DB_ADDR_SHIFT(DQ_PWM_OFFSET_XCM_RDMA_SQ_PROD);
 	qp->sq.db_data.data.icid = qp->icid + 1;
-	qp->rq.db = dev->db_addr +
-		    DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
-	qp->rq.db_data.data.icid = qp->icid;
+	if (!qp->srq) {
+		qp->rq.db = dev->db_addr +
+			    DB_ADDR_SHIFT(DQ_PWM_OFFSET_TCM_ROCE_RQ_PROD);
+		qp->rq.db_data.data.icid = qp->icid;
+	}
+}
+
+static int qedr_check_srq_params(struct ib_pd *ibpd, struct qedr_dev *dev,
+				 struct ib_srq_init_attr *attrs,
+				 struct ib_udata *udata)
+{
+	struct qedr_device_attr *qattr = &dev->attr;
+
+	if (attrs->attr.max_wr > qattr->max_srq_wr) {
+		DP_ERR(dev,
+		       "create srq: unsupported srq_wr=0x%x requested (max_srq_wr=0x%x)\n",
+		       attrs->attr.max_wr, qattr->max_srq_wr);
+		return -EINVAL;
+	}
+
+	if (attrs->attr.max_sge > qattr->max_sge) {
+		DP_ERR(dev,
+		       "create srq: unsupported sge=0x%x requested (max_srq_sge=0x%x)\n",
+		       attrs->attr.max_sge, qattr->max_sge);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static void qedr_free_srq_user_params(struct qedr_srq *srq)
+{
+	qedr_free_pbl(srq->dev, &srq->usrq.pbl_info, srq->usrq.pbl_tbl);
+	ib_umem_release(srq->usrq.umem);
+	ib_umem_release(srq->prod_umem);
+}
+
+static void qedr_free_srq_kernel_params(struct qedr_srq *srq)
+{
+	struct qedr_srq_hwq_info *hw_srq = &srq->hw_srq;
+	struct qedr_dev *dev = srq->dev;
+
+	dev->ops->common->chain_free(dev->cdev, &hw_srq->pbl);
+
+	dma_free_coherent(&dev->pdev->dev, sizeof(struct rdma_srq_producers),
+			  hw_srq->virt_prod_pair_addr,
+			  hw_srq->phy_prod_pair_addr);
+}
+
+static int qedr_init_srq_user_params(struct ib_ucontext *ib_ctx,
+				     struct qedr_srq *srq,
+				     struct qedr_create_srq_ureq *ureq,
+				     int access, int dmasync)
+{
+	struct scatterlist *sg;
+	int rc;
+
+	rc = qedr_init_user_queue(ib_ctx, srq->dev, &srq->usrq, ureq->srq_addr,
+				  ureq->srq_len, access, dmasync, 1);
+	if (rc)
+		return rc;
+
+	srq->prod_umem = ib_umem_get(ib_ctx, ureq->prod_pair_addr,
+				     sizeof(struct rdma_srq_producers),
+				     access, dmasync);
+	if (IS_ERR(srq->prod_umem)) {
+		qedr_free_pbl(srq->dev, &srq->usrq.pbl_info, srq->usrq.pbl_tbl);
+		ib_umem_release(srq->usrq.umem);
+		DP_ERR(srq->dev,
+		       "create srq: failed ib_umem_get for producer, got %ld\n",
+		       PTR_ERR(srq->prod_umem));
+		return PTR_ERR(srq->prod_umem);
+	}
+
+	sg = srq->prod_umem->sg_head.sgl;
+	srq->hw_srq.phy_prod_pair_addr = sg_dma_address(sg);
+
+	return 0;
+}
+
+static int qedr_alloc_srq_kernel_params(struct qedr_srq *srq,
+					struct qedr_dev *dev,
+					struct ib_srq_init_attr *init_attr)
+{
+	struct qedr_srq_hwq_info *hw_srq = &srq->hw_srq;
+	dma_addr_t phy_prod_pair_addr;
+	u32 num_elems;
+	void *va;
+	int rc;
+
+	va = dma_alloc_coherent(&dev->pdev->dev,
+				sizeof(struct rdma_srq_producers),
+				&phy_prod_pair_addr, GFP_KERNEL);
+	if (!va) {
+		DP_ERR(dev,
+		       "create srq: failed to allocate dma memory for producer\n");
+		return -ENOMEM;
+	}
+
+	hw_srq->phy_prod_pair_addr = phy_prod_pair_addr;
+	hw_srq->virt_prod_pair_addr = va;
+
+	num_elems = init_attr->attr.max_wr * RDMA_MAX_SRQ_WQE_SIZE;
+	rc = dev->ops->common->chain_alloc(dev->cdev,
+					   QED_CHAIN_USE_TO_CONSUME_PRODUCE,
+					   QED_CHAIN_MODE_PBL,
+					   QED_CHAIN_CNT_TYPE_U32,
+					   num_elems,
+					   QEDR_SRQ_WQE_ELEM_SIZE,
+					   &hw_srq->pbl, NULL);
+	if (rc)
+		goto err0;
+
+	hw_srq->num_elems = num_elems;
+
+	return 0;
+
+err0:
+	dma_free_coherent(&dev->pdev->dev, sizeof(struct rdma_srq_producers),
+			  va, phy_prod_pair_addr);
+	return rc;
+}
+
+static int qedr_idr_add(struct qedr_dev *dev, struct qedr_idr *qidr,
+			void *ptr, u32 id);
+static void qedr_idr_remove(struct qedr_dev *dev,
+			    struct qedr_idr *qidr, u32 id);
+
+struct ib_srq *qedr_create_srq(struct ib_pd *ibpd,
+			       struct ib_srq_init_attr *init_attr,
+			       struct ib_udata *udata)
+{
+	struct qed_rdma_destroy_srq_in_params destroy_in_params;
+	struct qed_rdma_create_srq_in_params in_params = {};
+	struct qedr_dev *dev = get_qedr_dev(ibpd->device);
+	struct qed_rdma_create_srq_out_params out_params;
+	struct qedr_pd *pd = get_qedr_pd(ibpd);
+	struct qedr_create_srq_ureq ureq = {};
+	u64 pbl_base_addr, phy_prod_pair_addr;
+	struct ib_ucontext *ib_ctx = NULL;
+	struct qedr_srq_hwq_info *hw_srq;
+	struct qedr_ucontext *ctx = NULL;
+	u32 page_cnt, page_size;
+	struct qedr_srq *srq;
+	int rc = 0;
+
+	DP_DEBUG(dev, QEDR_MSG_QP,
+		 "create SRQ called from %s (pd %p)\n",
+		 (udata) ? "User lib" : "kernel", pd);
+
+	rc = qedr_check_srq_params(ibpd, dev, init_attr, udata);
+	if (rc)
+		return ERR_PTR(-EINVAL);
+
+	srq = kzalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq)
+		return ERR_PTR(-ENOMEM);
+
+	srq->dev = dev;
+	hw_srq = &srq->hw_srq;
+	spin_lock_init(&srq->lock);
+
+	hw_srq->max_wr = init_attr->attr.max_wr;
+	hw_srq->max_sges = init_attr->attr.max_sge;
+
+	if (udata && ibpd->uobject && ibpd->uobject->context) {
+		ib_ctx = ibpd->uobject->context;
+		ctx = get_qedr_ucontext(ib_ctx);
+
+		if (ib_copy_from_udata(&ureq, udata, sizeof(ureq))) {
+			DP_ERR(dev,
+			       "create srq: problem copying data from user space\n");
+			goto err0;
+		}
+
+		rc = qedr_init_srq_user_params(ib_ctx, srq, &ureq, 0, 0);
+		if (rc)
+			goto err0;
+
+		page_cnt = srq->usrq.pbl_info.num_pbes;
+		pbl_base_addr = srq->usrq.pbl_tbl->pa;
+		phy_prod_pair_addr = hw_srq->phy_prod_pair_addr;
+		page_size = BIT(srq->usrq.umem->page_shift);
+	} else {
+		struct qed_chain *pbl;
+
+		rc = qedr_alloc_srq_kernel_params(srq, dev, init_attr);
+		if (rc)
+			goto err0;
+
+		pbl = &hw_srq->pbl;
+		page_cnt = qed_chain_get_page_cnt(pbl);
+		pbl_base_addr = qed_chain_get_pbl_phys(pbl);
+		phy_prod_pair_addr = hw_srq->phy_prod_pair_addr;
+		page_size = QED_CHAIN_PAGE_SIZE;
+	}
+
+	in_params.pd_id = pd->pd_id;
+	in_params.pbl_base_addr = pbl_base_addr;
+	in_params.prod_pair_addr = phy_prod_pair_addr;
+	in_params.num_pages = page_cnt;
+	in_params.page_size = page_size;
+
+	rc = dev->ops->rdma_create_srq(dev->rdma_ctx, &in_params, &out_params);
+	if (rc)
+		goto err1;
+
+	srq->srq_id = out_params.srq_id;
+
+	if (udata) {
+		rc = qedr_copy_srq_uresp(dev, srq, udata);
+		if (rc)
+			goto err2;
+	}
+
+	rc = qedr_idr_add(dev, &dev->srqidr, srq, srq->srq_id);
+	if (rc)
+		goto err2;
+
+	DP_DEBUG(dev, QEDR_MSG_SRQ,
+		 "create srq: created srq with srq_id=0x%0x\n", srq->srq_id);
+	return &srq->ibsrq;
+
+err2:
+	destroy_in_params.srq_id = srq->srq_id;
+
+	dev->ops->rdma_destroy_srq(dev->rdma_ctx, &destroy_in_params);
+err1:
+	if (udata)
+		qedr_free_srq_user_params(srq);
+	else
+		qedr_free_srq_kernel_params(srq);
+err0:
+	kfree(srq);
+
+	return ERR_PTR(-EFAULT);
+}
+
+int qedr_destroy_srq(struct ib_srq *ibsrq)
+{
+	struct qed_rdma_destroy_srq_in_params in_params = {};
+	struct qedr_dev *dev = get_qedr_dev(ibsrq->device);
+	struct qedr_srq *srq = get_qedr_srq(ibsrq);
+
+	qedr_idr_remove(dev, &dev->srqidr, srq->srq_id);
+	in_params.srq_id = srq->srq_id;
+	dev->ops->rdma_destroy_srq(dev->rdma_ctx, &in_params);
+
+	if (ibsrq->pd->uobject)
+		qedr_free_srq_user_params(srq);
+	else
+		qedr_free_srq_kernel_params(srq);
+
+	DP_DEBUG(dev, QEDR_MSG_SRQ,
+		 "destroy srq: destroyed srq with srq_id=0x%0x\n",
+		 srq->srq_id);
+	kfree(srq);
+
+	return 0;
+}
+
+int qedr_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		    enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
+{
+	struct qed_rdma_modify_srq_in_params in_params = {};
+	struct qedr_dev *dev = get_qedr_dev(ibsrq->device);
+	struct qedr_srq *srq = get_qedr_srq(ibsrq);
+	int rc;
+
+	if (attr_mask & IB_SRQ_MAX_WR) {
+		DP_ERR(dev,
+		       "modify srq: invalid attribute mask=0x%x specified for %p\n",
+		       attr_mask, srq);
+		return -EINVAL;
+	}
+
+	if (attr_mask & IB_SRQ_LIMIT) {
+		if (attr->srq_limit >= srq->hw_srq.max_wr) {
+			DP_ERR(dev,
+			       "modify srq: invalid srq_limit=0x%x (max_srq_limit=0x%x)\n",
+			       attr->srq_limit, srq->hw_srq.max_wr);
+			return -EINVAL;
+		}
+
+		in_params.srq_id = srq->srq_id;
+		in_params.wqe_limit = attr->srq_limit;
+		rc = dev->ops->rdma_modify_srq(dev->rdma_ctx, &in_params);
+		if (rc)
+			return rc;
+	}
+
+	srq->srq_limit = attr->srq_limit;
+
+	DP_DEBUG(dev, QEDR_MSG_SRQ,
+		 "modify srq: modified srq with srq_id=0x%0x\n", srq->srq_id);
+
+	return 0;
 }
 
 static inline void
@@ -1299,9 +1623,17 @@ qedr_init_common_qp_in_params(struct qedr_dev *dev,
 	params->dpi = pd->uctx ? pd->uctx->dpi : dev->dpi;
 	params->sq_cq_id = get_qedr_cq(attrs->send_cq)->icid;
 	params->stats_queue = 0;
-	params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid;
 	params->srq_id = 0;
 	params->use_srq = false;
+
+	if (!qp->srq) {
+		params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid;
+
+	} else {
+		params->rq_cq_id = get_qedr_cq(attrs->recv_cq)->icid;
+		params->srq_id = qp->srq->srq_id;
+		params->use_srq = true;
+	}
 }
 
 static inline void qedr_qp_user_print(struct qedr_dev *dev, struct qedr_qp *qp)
@@ -1318,32 +1650,27 @@ static inline void qedr_qp_user_print(struct qedr_dev *dev, struct qedr_qp *qp)
 		 qp->usq.buf_len, qp->urq.buf_addr, qp->urq.buf_len);
 }
 
-static int qedr_idr_add(struct qedr_dev *dev, void *ptr, u32 id)
+static int qedr_idr_add(struct qedr_dev *dev, struct qedr_idr *qidr,
+			void *ptr, u32 id)
 {
 	int rc;
 
-	if (!rdma_protocol_iwarp(&dev->ibdev, 1))
-		return 0;
-
 	idr_preload(GFP_KERNEL);
-	spin_lock_irq(&dev->idr_lock);
+	spin_lock_irq(&qidr->idr_lock);
 
-	rc = idr_alloc(&dev->qpidr, ptr, id, id + 1, GFP_ATOMIC);
+	rc = idr_alloc(&qidr->idr, ptr, id, id + 1, GFP_ATOMIC);
 
-	spin_unlock_irq(&dev->idr_lock);
+	spin_unlock_irq(&qidr->idr_lock);
 	idr_preload_end();
 
 	return rc < 0 ? rc : 0;
 }
 
-static void qedr_idr_remove(struct qedr_dev *dev, u32 id)
+static void qedr_idr_remove(struct qedr_dev *dev, struct qedr_idr *qidr, u32 id)
 {
-	if (!rdma_protocol_iwarp(&dev->ibdev, 1))
-		return;
-
-	spin_lock_irq(&dev->idr_lock);
-	idr_remove(&dev->qpidr, id);
-	spin_unlock_irq(&dev->idr_lock);
+	spin_lock_irq(&qidr->idr_lock);
+	idr_remove(&qidr->idr, id);
+	spin_unlock_irq(&qidr->idr_lock);
 }
 
 static inline void
@@ -1356,9 +1683,10 @@ qedr_iwarp_populate_user_qp(struct qedr_dev *dev,
 
 	qedr_populate_pbls(dev, qp->usq.umem, qp->usq.pbl_tbl,
 			   &qp->usq.pbl_info, FW_PAGE_SHIFT);
-
-	qp->urq.pbl_tbl->va = out_params->rq_pbl_virt;
-	qp->urq.pbl_tbl->pa = out_params->rq_pbl_phys;
+	if (!qp->srq) {
+		qp->urq.pbl_tbl->va = out_params->rq_pbl_virt;
+		qp->urq.pbl_tbl->pa = out_params->rq_pbl_phys;
+	}
 
 	qedr_populate_pbls(dev, qp->urq.umem, qp->urq.pbl_tbl,
 			   &qp->urq.pbl_info, FW_PAGE_SHIFT);
@@ -1404,11 +1732,13 @@ static int qedr_create_user_qp(struct qedr_dev *dev,
 	if (rc)
 		return rc;
 
-	/* RQ - read access only (0), dma sync not required (0) */
-	rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq.rq_addr,
-				  ureq.rq_len, 0, 0, alloc_and_init);
-	if (rc)
-		return rc;
+	if (!qp->srq) {
+		/* RQ - read access only (0), dma sync not required (0) */
+		rc = qedr_init_user_queue(ib_ctx, dev, &qp->urq, ureq.rq_addr,
+					  ureq.rq_len, 0, 0, alloc_and_init);
+		if (rc)
+			return rc;
+	}
 
 	memset(&in_params, 0, sizeof(in_params));
 	qedr_init_common_qp_in_params(dev, pd, qp, attrs, false, &in_params);
@@ -1416,8 +1746,10 @@ static int qedr_create_user_qp(struct qedr_dev *dev,
 	in_params.qp_handle_hi = ureq.qp_handle_hi;
 	in_params.sq_num_pages = qp->usq.pbl_info.num_pbes;
 	in_params.sq_pbl_ptr = qp->usq.pbl_tbl->pa;
-	in_params.rq_num_pages = qp->urq.pbl_info.num_pbes;
-	in_params.rq_pbl_ptr = qp->urq.pbl_tbl->pa;
+	if (!qp->srq) {
+		in_params.rq_num_pages = qp->urq.pbl_info.num_pbes;
+		in_params.rq_pbl_ptr = qp->urq.pbl_tbl->pa;
+	}
 
 	qp->qed_qp = dev->ops->rdma_create_qp(dev->rdma_ctx,
 					      &in_params, &out_params);
@@ -1679,16 +2011,13 @@ struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
 	if (rc)
 		return ERR_PTR(rc);
 
-	if (attrs->srq)
-		return ERR_PTR(-EINVAL);
-
 	DP_DEBUG(dev, QEDR_MSG_QP,
 		 "create qp: called from %s, event_handler=%p, eepd=%p sq_cq=%p, sq_icid=%d, rq_cq=%p, rq_icid=%d\n",
 		 udata ? "user library" : "kernel", attrs->event_handler, pd,
 		 get_qedr_cq(attrs->send_cq),
 		 get_qedr_cq(attrs->send_cq)->icid,
 		 get_qedr_cq(attrs->recv_cq),
-		 get_qedr_cq(attrs->recv_cq)->icid);
+		 attrs->recv_cq ? get_qedr_cq(attrs->recv_cq)->icid : 0);
 
 	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
 	if (!qp) {
@@ -1715,9 +2044,11 @@ struct ib_qp *qedr_create_qp(struct ib_pd *ibpd,
 
 	qp->ibqp.qp_num = qp->qp_id;
 
-	rc = qedr_idr_add(dev, qp, qp->qp_id);
-	if (rc)
-		goto err;
+	if (rdma_protocol_iwarp(&dev->ibdev, 1)) {
+		rc = qedr_idr_add(dev, &dev->qpidr, qp, qp->qp_id);
+		if (rc)
+			goto err;
+	}
 
 	return &qp->ibqp;
 
@@ -2289,8 +2620,9 @@ int qedr_destroy_qp(struct ib_qp *ibqp)
 
 	qedr_free_qp_resources(dev, qp);
 
-	if (atomic_dec_and_test(&qp->refcnt)) {
-		qedr_idr_remove(dev, qp->qp_id);
+	if (atomic_dec_and_test(&qp->refcnt) &&
+	    rdma_protocol_iwarp(&dev->ibdev, 1)) {
+		qedr_idr_remove(dev, &dev->qpidr, qp->qp_id);
 		kfree(qp);
 	}
 	return rc;
@@ -2305,7 +2637,7 @@ struct ib_ah *qedr_create_ah(struct ib_pd *ibpd, struct rdma_ah_attr *attr,
 	if (!ah)
 		return ERR_PTR(-ENOMEM);
 
-	ah->attr = *attr;
+	rdma_copy_ah_attr(&ah->attr, attr);
 
 	return &ah->ibah;
 }
@@ -2314,6 +2646,7 @@ int qedr_destroy_ah(struct ib_ah *ibah)
 {
 	struct qedr_ah *ah = get_qedr_ah(ibah);
 
+	rdma_destroy_ah_attr(&ah->attr);
 	kfree(ah);
 	return 0;
 }
@@ -2705,9 +3038,9 @@ static void swap_wqe_data64(u64 *p)
 
 static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev,
 				       struct qedr_qp *qp, u8 *wqe_size,
-				       struct ib_send_wr *wr,
-				       struct ib_send_wr **bad_wr, u8 *bits,
-				       u8 bit)
+				       const struct ib_send_wr *wr,
+				       const struct ib_send_wr **bad_wr,
+				       u8 *bits, u8 bit)
 {
 	u32 data_size = sge_data_len(wr->sg_list, wr->num_sge);
 	char *seg_prt, *wqe;
@@ -2790,7 +3123,7 @@ static u32 qedr_prepare_sq_inline_data(struct qedr_dev *dev,
 	} while (0)
 
 static u32 qedr_prepare_sq_sges(struct qedr_qp *qp, u8 *wqe_size,
-				struct ib_send_wr *wr)
+				const struct ib_send_wr *wr)
 {
 	u32 data_size = 0;
 	int i;
@@ -2814,8 +3147,8 @@ static u32 qedr_prepare_sq_rdma_data(struct qedr_dev *dev,
 				     struct qedr_qp *qp,
 				     struct rdma_sq_rdma_wqe_1st *rwqe,
 				     struct rdma_sq_rdma_wqe_2nd *rwqe2,
-				     struct ib_send_wr *wr,
-				     struct ib_send_wr **bad_wr)
+				     const struct ib_send_wr *wr,
+				     const struct ib_send_wr **bad_wr)
 {
 	rwqe2->r_key = cpu_to_le32(rdma_wr(wr)->rkey);
 	DMA_REGPAIR_LE(rwqe2->remote_va, rdma_wr(wr)->remote_addr);
@@ -2837,8 +3170,8 @@ static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev,
 				     struct qedr_qp *qp,
 				     struct rdma_sq_send_wqe_1st *swqe,
 				     struct rdma_sq_send_wqe_2st *swqe2,
-				     struct ib_send_wr *wr,
-				     struct ib_send_wr **bad_wr)
+				     const struct ib_send_wr *wr,
+				     const struct ib_send_wr **bad_wr)
 {
 	memset(swqe2, 0, sizeof(*swqe2));
 	if (wr->send_flags & IB_SEND_INLINE) {
@@ -2854,7 +3187,7 @@ static u32 qedr_prepare_sq_send_data(struct qedr_dev *dev,
 
 static int qedr_prepare_reg(struct qedr_qp *qp,
 			    struct rdma_sq_fmr_wqe_1st *fwqe1,
-			    struct ib_reg_wr *wr)
+			    const struct ib_reg_wr *wr)
 {
 	struct qedr_mr *mr = get_qedr_mr(wr->mr);
 	struct rdma_sq_fmr_wqe_2nd *fwqe2;
@@ -2916,7 +3249,8 @@ static enum ib_wc_opcode qedr_ib_to_wc_opcode(enum ib_wr_opcode opcode)
 	}
 }
 
-static inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr)
+static inline bool qedr_can_post_send(struct qedr_qp *qp,
+				      const struct ib_send_wr *wr)
 {
 	int wq_is_full, err_wr, pbl_is_full;
 	struct qedr_dev *dev = qp->dev;
@@ -2953,8 +3287,8 @@ static inline bool qedr_can_post_send(struct qedr_qp *qp, struct ib_send_wr *wr)
 	return true;
 }
 
-static int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
-		     struct ib_send_wr **bad_wr)
+static int __qedr_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
+			    const struct ib_send_wr **bad_wr)
 {
 	struct qedr_dev *dev = get_qedr_dev(ibqp->device);
 	struct qedr_qp *qp = get_qedr_qp(ibqp);
@@ -3168,8 +3502,8 @@ static int __qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	return rc;
 }
 
-int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
-		   struct ib_send_wr **bad_wr)
+int qedr_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
+		   const struct ib_send_wr **bad_wr)
 {
 	struct qedr_dev *dev = get_qedr_dev(ibqp->device);
 	struct qedr_qp *qp = get_qedr_qp(ibqp);
@@ -3234,8 +3568,104 @@ int qedr_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	return rc;
 }
 
-int qedr_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
-		   struct ib_recv_wr **bad_wr)
+static u32 qedr_srq_elem_left(struct qedr_srq_hwq_info *hw_srq)
+{
+	u32 used;
+
+	/* Calculate number of elements used based on producer
+	 * count and consumer count and subtract it from max
+	 * work request supported so that we get elements left.
+	 */
+	used = hw_srq->wr_prod_cnt - hw_srq->wr_cons_cnt;
+
+	return hw_srq->max_wr - used;
+}
+
+int qedr_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
+		       const struct ib_recv_wr **bad_wr)
+{
+	struct qedr_srq *srq = get_qedr_srq(ibsrq);
+	struct qedr_srq_hwq_info *hw_srq;
+	struct qedr_dev *dev = srq->dev;
+	struct qed_chain *pbl;
+	unsigned long flags;
+	int status = 0;
+	u32 num_sge;
+	u32 offset;
+
+	spin_lock_irqsave(&srq->lock, flags);
+
+	hw_srq = &srq->hw_srq;
+	pbl = &srq->hw_srq.pbl;
+	while (wr) {
+		struct rdma_srq_wqe_header *hdr;
+		int i;
+
+		if (!qedr_srq_elem_left(hw_srq) ||
+		    wr->num_sge > srq->hw_srq.max_sges) {
+			DP_ERR(dev, "Can't post WR  (%d,%d) || (%d > %d)\n",
+			       hw_srq->wr_prod_cnt, hw_srq->wr_cons_cnt,
+			       wr->num_sge, srq->hw_srq.max_sges);
+			status = -ENOMEM;
+			*bad_wr = wr;
+			break;
+		}
+
+		hdr = qed_chain_produce(pbl);
+		num_sge = wr->num_sge;
+		/* Set number of sge and work request id in header */
+		SRQ_HDR_SET(hdr, wr->wr_id, num_sge);
+
+		srq->hw_srq.wr_prod_cnt++;
+		hw_srq->wqe_prod++;
+		hw_srq->sge_prod++;
+
+		DP_DEBUG(dev, QEDR_MSG_SRQ,
+			 "SRQ WR: SGEs: %d with wr_id[%d] = %llx\n",
+			 wr->num_sge, hw_srq->wqe_prod, wr->wr_id);
+
+		for (i = 0; i < wr->num_sge; i++) {
+			struct rdma_srq_sge *srq_sge = qed_chain_produce(pbl);
+
+			/* Set SGE length, lkey and address */
+			SRQ_SGE_SET(srq_sge, wr->sg_list[i].addr,
+				    wr->sg_list[i].length, wr->sg_list[i].lkey);
+
+			DP_DEBUG(dev, QEDR_MSG_SRQ,
+				 "[%d]: len %d key %x addr %x:%x\n",
+				 i, srq_sge->length, srq_sge->l_key,
+				 srq_sge->addr.hi, srq_sge->addr.lo);
+			hw_srq->sge_prod++;
+		}
+
+		/* Flush WQE and SGE information before
+		 * updating producer.
+		 */
+		wmb();
+
+		/* SRQ producer is 8 bytes. Need to update SGE producer index
+		 * in first 4 bytes and need to update WQE producer in
+		 * next 4 bytes.
+		 */
+		*srq->hw_srq.virt_prod_pair_addr = hw_srq->sge_prod;
+		offset = offsetof(struct rdma_srq_producers, wqe_prod);
+		*((u8 *)srq->hw_srq.virt_prod_pair_addr + offset) =
+			hw_srq->wqe_prod;
+
+		/* Flush producer after updating it. */
+		wmb();
+		wr = wr->next;
+	}
+
+	DP_DEBUG(dev, QEDR_MSG_SRQ, "POST: Elements in S-RQ: %d\n",
+		 qed_chain_get_elem_left(pbl));
+	spin_unlock_irqrestore(&srq->lock, flags);
+
+	return status;
+}
+
+int qedr_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
+		   const struct ib_recv_wr **bad_wr)
 {
 	struct qedr_qp *qp = get_qedr_qp(ibqp);
 	struct qedr_dev *dev = qp->dev;
@@ -3625,6 +4055,31 @@ static void __process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
 	wc->wr_id = wr_id;
 }
 
+static int process_resp_one_srq(struct qedr_dev *dev, struct qedr_qp *qp,
+				struct qedr_cq *cq, struct ib_wc *wc,
+				struct rdma_cqe_responder *resp)
+{
+	struct qedr_srq *srq = qp->srq;
+	u64 wr_id;
+
+	wr_id = HILO_GEN(le32_to_cpu(resp->srq_wr_id.hi),
+			 le32_to_cpu(resp->srq_wr_id.lo), u64);
+
+	if (resp->status == RDMA_CQE_RESP_STS_WORK_REQUEST_FLUSHED_ERR) {
+		wc->status = IB_WC_WR_FLUSH_ERR;
+		wc->vendor_err = 0;
+		wc->wr_id = wr_id;
+		wc->byte_len = 0;
+		wc->src_qp = qp->id;
+		wc->qp = &qp->ibqp;
+		wc->wr_id = wr_id;
+	} else {
+		__process_resp_one(dev, qp, cq, wc, resp, wr_id);
+	}
+	srq->hw_srq.wr_cons_cnt++;
+
+	return 1;
+}
 static int process_resp_one(struct qedr_dev *dev, struct qedr_qp *qp,
 			    struct qedr_cq *cq, struct ib_wc *wc,
 			    struct rdma_cqe_responder *resp)
@@ -3674,6 +4129,19 @@ static void try_consume_resp_cqe(struct qedr_cq *cq, struct qedr_qp *qp,
 	}
 }
 
+static int qedr_poll_cq_resp_srq(struct qedr_dev *dev, struct qedr_qp *qp,
+				 struct qedr_cq *cq, int num_entries,
+				 struct ib_wc *wc,
+				 struct rdma_cqe_responder *resp)
+{
+	int cnt;
+
+	cnt = process_resp_one_srq(dev, qp, cq, wc, resp);
+	consume_cqe(cq);
+
+	return cnt;
+}
+
 static int qedr_poll_cq_resp(struct qedr_dev *dev, struct qedr_qp *qp,
 			     struct qedr_cq *cq, int num_entries,
 			     struct ib_wc *wc, struct rdma_cqe_responder *resp,
@@ -3751,6 +4219,11 @@ int qedr_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
 			cnt = qedr_poll_cq_resp(dev, qp, cq, num_entries, wc,
 						&cqe->resp, &update);
 			break;
+		case RDMA_CQE_TYPE_RESPONDER_SRQ:
+			cnt = qedr_poll_cq_resp_srq(dev, qp, cq, num_entries,
+						    wc, &cqe->resp);
+			update = 1;
+			break;
 		case RDMA_CQE_TYPE_INVALID:
 		default:
 			DP_ERR(dev, "Error: invalid CQE type = %d\n",