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 slightly more active cycle than normal with ongoing
  core changes and quite a lot of collected driver updates.

   - Various driver fixes for bnxt_re, cxgb4, hns, mlx5, pvrdma, rxe

   - A new data transfer mode for HFI1 giving higher performance

   - Significant functional and bug fix update to the mlx5
     On-Demand-Paging MR feature

   - A chip hang reset recovery system for hns

   - Change mm->pinned_vm to an atomic64

   - Update bnxt_re to support a new 57500 chip

   - A sane netlink 'rdma link add' method for creating rxe devices and
     fixing the various unregistration race conditions in rxe's
     unregister flow

   - Allow lookup up objects by an ID over netlink

   - Various reworking of the core to driver interface:
       - drivers should not assume umem SGLs are in PAGE_SIZE chunks
       - ucontext is accessed via udata not other means
       - start to make the core code responsible for object memory
         allocation
       - drivers should convert struct device to struct ib_device via a
         helper
       - drivers have more tools to avoid use after unregister problems"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (280 commits)
  net/mlx5: ODP support for XRC transport is not enabled by default in FW
  IB/hfi1: Close race condition on user context disable and close
  RDMA/umem: Revert broken 'off by one' fix
  RDMA/umem: minor bug fix in error handling path
  RDMA/hns: Use GFP_ATOMIC in hns_roce_v2_modify_qp
  cxgb4: kfree mhp after the debug print
  IB/rdmavt: Fix concurrency panics in QP post_send and modify to error
  IB/rdmavt: Fix loopback send with invalidate ordering
  IB/iser: Fix dma_nents type definition
  IB/mlx5: Set correct write permissions for implicit ODP MR
  bnxt_re: Clean cq for kernel consumers only
  RDMA/uverbs: Don't do double free of allocated PD
  RDMA: Handle ucontext allocations by IB/core
  RDMA/core: Fix a WARN() message
  bnxt_re: fix the regression due to changes in alloc_pbl
  IB/mlx4: Increase the timeout for CM cache
  IB/core: Abort page fault handler silently during owning process exit
  IB/mlx5: Validate correct PD before prefetch MR
  IB/mlx5: Protect against prefetch of invalid MR
  RDMA/uverbs: Store PR pointer before it is overwritten
  ...

drivers/infiniband/hw/cxgb4/Makefile

@@ -1,5 +1,5 @@
-ccflags-y := -Idrivers/net/ethernet/chelsio/cxgb4
-ccflags-y += -Idrivers/net/ethernet/chelsio/libcxgb
+ccflags-y := -I $(srctree)/drivers/net/ethernet/chelsio/cxgb4
+ccflags-y += -I $(srctree)/drivers/net/ethernet/chelsio/libcxgb
 
 obj-$(CONFIG_INFINIBAND_CXGB4) += iw_cxgb4.o
 

drivers/infiniband/hw/cxgb4/cm.c

@@ -655,7 +655,33 @@ static int send_halfclose(struct c4iw_ep *ep)
 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
 }
 
-static int send_abort(struct c4iw_ep *ep)
+static void read_tcb(struct c4iw_ep *ep)
+{
+	struct sk_buff *skb;
+	struct cpl_get_tcb *req;
+	int wrlen = roundup(sizeof(*req), 16);
+
+	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
+	if (WARN_ON(!skb))
+		return;
+
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
+	req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	INIT_TP_WR(req, ep->hwtid);
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
+	req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
+
+	/*
+	 * keep a ref on the ep so the tcb is not unlocked before this
+	 * cpl completes. The ref is released in read_tcb_rpl().
+	 */
+	c4iw_get_ep(&ep->com);
+	if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
+		c4iw_put_ep(&ep->com);
+}
+
+static int send_abort_req(struct c4iw_ep *ep)
 {
 	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
 	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
@@ -670,6 +696,17 @@ static int send_abort(struct c4iw_ep *ep)
 	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
 }
 
+static int send_abort(struct c4iw_ep *ep)
+{
+	if (!ep->com.qp || !ep->com.qp->srq) {
+		send_abort_req(ep);
+		return 0;
+	}
+	set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
+	read_tcb(ep);
+	return 0;
+}
+
 static int send_connect(struct c4iw_ep *ep)
 {
 	struct cpl_act_open_req *req = NULL;
@@ -1851,14 +1888,11 @@ static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
 	return 0;
 }
 
-static void complete_cached_srq_buffers(struct c4iw_ep *ep,
-					__be32 srqidx_status)
+static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
 {
 	enum chip_type adapter_type;
-	u32 srqidx;
 
 	adapter_type = ep->com.dev->rdev.lldi.adapter_type;
-	srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(srqidx_status));
 
 	/*
 	 * If this TCB had a srq buffer cached, then we must complete
@@ -1876,6 +1910,7 @@ static void complete_cached_srq_buffers(struct c4iw_ep *ep,
 
 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
 {
+	u32 srqidx;
 	struct c4iw_ep *ep;
 	struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
 	int release = 0;
@@ -1887,7 +1922,10 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
 		return 0;
 	}
 
-	complete_cached_srq_buffers(ep, rpl->srqidx_status);
+	if (ep->com.qp && ep->com.qp->srq) {
+		srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
+		complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
+	}
 
 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
 	mutex_lock(&ep->com.mutex);
@@ -1903,8 +1941,10 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
 	}
 	mutex_unlock(&ep->com.mutex);
 
-	if (release)
+	if (release) {
+		close_complete_upcall(ep, -ECONNRESET);
 		release_ep_resources(ep);
+	}
 	c4iw_put_ep(&ep->com);
 	return 0;
 }
@@ -2072,7 +2112,7 @@ static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
 	} else {
 		pdev = get_real_dev(n->dev);
 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
-					n, pdev, 0);
+					n, pdev, rt_tos2priority(tos));
 		if (!ep->l2t)
 			goto out;
 		ep->mtu = dst_mtu(dst);
@@ -2161,7 +2201,8 @@ static int c4iw_reconnect(struct c4iw_ep *ep)
 					   laddr6->sin6_addr.s6_addr,
 					   raddr6->sin6_addr.s6_addr,
 					   laddr6->sin6_port,
-					   raddr6->sin6_port, 0,
+					   raddr6->sin6_port,
+					   ep->com.cm_id->tos,
 					   raddr6->sin6_scope_id);
 		iptype = 6;
 		ra = (__u8 *)&raddr6->sin6_addr;
@@ -2476,7 +2517,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 	u16 peer_mss = ntohs(req->tcpopt.mss);
 	int iptype;
 	unsigned short hdrs;
-	u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
+	u8 tos;
 
 	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
 	if (!parent_ep) {
@@ -2490,6 +2531,11 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 		goto reject;
 	}
 
+	if (parent_ep->com.cm_id->tos_set)
+		tos = parent_ep->com.cm_id->tos;
+	else
+		tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
+
 	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
 			&iptype, local_ip, peer_ip, &local_port, &peer_port);
 
@@ -2509,7 +2555,7 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
 			 ntohs(peer_port), peer_mss);
 		dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
 				local_ip, peer_ip, local_port, peer_port,
-				PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
+				tos,
 				((struct sockaddr_in6 *)
 				 &parent_ep->com.local_addr)->sin6_scope_id);
 	}
@@ -2740,6 +2786,21 @@ static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
 	return 0;
 }
 
+static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
+{
+	complete_cached_srq_buffers(ep, ep->srqe_idx);
+	if (ep->com.cm_id && ep->com.qp) {
+		struct c4iw_qp_attributes attrs;
+
+		attrs.next_state = C4IW_QP_STATE_ERROR;
+		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+			       C4IW_QP_ATTR_NEXT_STATE,	&attrs, 1);
+	}
+	peer_abort_upcall(ep);
+	release_ep_resources(ep);
+	c4iw_put_ep(&ep->com);
+}
+
 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 {
 	struct cpl_abort_req_rss6 *req = cplhdr(skb);
@@ -2750,6 +2811,7 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 	int release = 0;
 	unsigned int tid = GET_TID(req);
 	u8 status;
+	u32 srqidx;
 
 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
 
@@ -2769,8 +2831,6 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 		goto deref_ep;
 	}
 
-	complete_cached_srq_buffers(ep, req->srqidx_status);
-
 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
 		 ep->com.state);
 	set_bit(PEER_ABORT, &ep->com.history);
@@ -2819,6 +2879,23 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 		stop_ep_timer(ep);
 		/*FALLTHROUGH*/
 	case FPDU_MODE:
+		if (ep->com.qp && ep->com.qp->srq) {
+			srqidx = ABORT_RSS_SRQIDX_G(
+					be32_to_cpu(req->srqidx_status));
+			if (srqidx) {
+				complete_cached_srq_buffers(ep,
+							    req->srqidx_status);
+			} else {
+				/* Hold ep ref until finish_peer_abort() */
+				c4iw_get_ep(&ep->com);
+				__state_set(&ep->com, ABORTING);
+				set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
+				read_tcb(ep);
+				break;
+
+			}
+		}
+
 		if (ep->com.cm_id && ep->com.qp) {
 			attrs.next_state = C4IW_QP_STATE_ERROR;
 			ret = c4iw_modify_qp(ep->com.qp->rhp,
@@ -2942,15 +3019,18 @@ static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
 
 	ep = get_ep_from_tid(dev, tid);
 
-	if (ep && ep->com.qp) {
-		pr_warn("TERM received tid %u qpid %u\n",
-			tid, ep->com.qp->wq.sq.qid);
-		attrs.next_state = C4IW_QP_STATE_TERMINATE;
-		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
-			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+	if (ep) {
+		if (ep->com.qp) {
+			pr_warn("TERM received tid %u qpid %u\n", tid,
+				ep->com.qp->wq.sq.qid);
+			attrs.next_state = C4IW_QP_STATE_TERMINATE;
+			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		}
+
+		c4iw_put_ep(&ep->com);
 	} else
 		pr_warn("TERM received tid %u no ep/qp\n", tid);
-	c4iw_put_ep(&ep->com);
 
 	return 0;
 }
@@ -3318,7 +3398,7 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 					   laddr6->sin6_addr.s6_addr,
 					   raddr6->sin6_addr.s6_addr,
 					   laddr6->sin6_port,
-					   raddr6->sin6_port, 0,
+					   raddr6->sin6_port, cm_id->tos,
 					   raddr6->sin6_scope_id);
 	}
 	if (!ep->dst) {
@@ -3606,7 +3686,6 @@ int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
 	if (close) {
 		if (abrupt) {
 			set_bit(EP_DISC_ABORT, &ep->com.history);
-			close_complete_upcall(ep, -ECONNRESET);
 			ret = send_abort(ep);
 		} else {
 			set_bit(EP_DISC_CLOSE, &ep->com.history);
@@ -3717,6 +3796,80 @@ static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
 	return;
 }
 
+static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
+{
+	u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
+	u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
+	u64 t;
+	u32 shift = 32;
+
+	t = (thi << shift) | (tlo >> shift);
+
+	return t;
+}
+
+static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
+{
+	u32 v;
+	u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
+
+	if (word & 0x1)
+		shift += 32;
+	v = (t >> shift) & mask;
+	return v;
+}
+
+static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
+	__be64 *tcb = (__be64 *)(rpl + 1);
+	unsigned int tid = GET_TID(rpl);
+	struct c4iw_ep *ep;
+	u64 t_flags_64;
+	u32 rx_pdu_out;
+
+	ep = get_ep_from_tid(dev, tid);
+	if (!ep)
+		return 0;
+	/* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
+	 * determine if there's a rx PDU feedback event pending.
+	 *
+	 * If that bit is set, it means we'll need to re-read the TCB's
+	 * rq_start value. The final value is the one present in a TCB
+	 * with the TF_RX_PDU_OUT bit cleared.
+	 */
+
+	t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
+	rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
+
+	c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
+	c4iw_put_ep(&ep->com); /* from read_tcb() */
+
+	/* If TF_RX_PDU_OUT bit is set, re-read the TCB */
+	if (rx_pdu_out) {
+		if (++ep->rx_pdu_out_cnt >= 2) {
+			WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
+			goto cleanup;
+		}
+		read_tcb(ep);
+		return 0;
+	}
+
+	ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_W,
+			TCB_RQ_START_S);
+cleanup:
+	pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
+
+	if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
+		finish_peer_abort(dev, ep);
+	else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
+		send_abort_req(ep);
+	else
+		WARN_ONCE(1, "unexpected state!");
+
+	return 0;
+}
+
 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
 {
 	struct cpl_fw6_msg *rpl = cplhdr(skb);
@@ -4037,6 +4190,7 @@ static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
 	[CPL_RDMA_TERMINATE] = terminate,
 	[CPL_FW4_ACK] = fw4_ack,
+	[CPL_GET_TCB_RPL] = read_tcb_rpl,
 	[CPL_FW6_MSG] = deferred_fw6_msg,
 	[CPL_RX_PKT] = rx_pkt,
 	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
@@ -4268,6 +4422,7 @@ c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
 	[CPL_RDMA_TERMINATE] = sched,
 	[CPL_FW4_ACK] = sched,
 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
+	[CPL_GET_TCB_RPL] = sched,
 	[CPL_FW6_MSG] = fw6_msg,
 	[CPL_RX_PKT] = sched
 };

drivers/infiniband/hw/cxgb4/device.c

@@ -720,11 +720,8 @@ static const struct file_operations ep_debugfs_fops = {
 	.read    = debugfs_read,
 };
 
-static int setup_debugfs(struct c4iw_dev *devp)
+static void setup_debugfs(struct c4iw_dev *devp)
 {
-	if (!devp->debugfs_root)
-		return -1;
-
 	debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
 				 (void *)devp, &qp_debugfs_fops, 4096);
 
@@ -740,7 +737,6 @@ static int setup_debugfs(struct c4iw_dev *devp)
 	if (c4iw_wr_log)
 		debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
 					 (void *)devp, &wr_log_debugfs_fops, 4096);
-	return 0;
 }
 
 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
@@ -981,7 +977,7 @@ static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
 		pr_info("%s: On-Chip Queues not supported on this device\n",
 			pci_name(infop->pdev));
 
-	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
+	devp = ib_alloc_device(c4iw_dev, ibdev);
 	if (!devp) {
 		pr_err("Cannot allocate ib device\n");
 		return ERR_PTR(-ENOMEM);
@@ -1564,8 +1560,6 @@ static int __init c4iw_init_module(void)
 		return err;
 
 	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
-	if (!c4iw_debugfs_root)
-		pr_warn("could not create debugfs entry, continuing\n");
 
 	reg_workq = create_singlethread_workqueue("Register_iWARP_device");
 	if (!reg_workq) {

drivers/infiniband/hw/cxgb4/iw_cxgb4.h

@@ -589,7 +589,6 @@ struct c4iw_ucontext {
 	u32 key;
 	spinlock_t mmap_lock;
 	struct list_head mmaps;
-	struct kref kref;
 	bool is_32b_cqe;
 };
 
@@ -598,18 +597,6 @@ static inline struct c4iw_ucontext *to_c4iw_ucontext(struct ib_ucontext *c)
 	return container_of(c, struct c4iw_ucontext, ibucontext);
 }
 
-void _c4iw_free_ucontext(struct kref *kref);
-
-static inline void c4iw_put_ucontext(struct c4iw_ucontext *ucontext)
-{
-	kref_put(&ucontext->kref, _c4iw_free_ucontext);
-}
-
-static inline void c4iw_get_ucontext(struct c4iw_ucontext *ucontext)
-{
-	kref_get(&ucontext->kref);
-}
-
 struct c4iw_mm_entry {
 	struct list_head entry;
 	u64 addr;
@@ -982,6 +969,9 @@ struct c4iw_ep {
 	int rcv_win;
 	u32 snd_wscale;
 	struct c4iw_ep_stats stats;
+	u32 srqe_idx;
+	u32 rx_pdu_out_cnt;
+	struct sk_buff *peer_abort_skb;
 };
 
 static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id)

drivers/infiniband/hw/cxgb4/mem.c

@@ -502,10 +502,9 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			       u64 virt, int acc, struct ib_udata *udata)
 {
 	__be64 *pages;
-	int shift, n, len;
-	int i, k, entry;
+	int shift, n, i;
 	int err = -ENOMEM;
-	struct scatterlist *sg;
+	struct sg_dma_page_iter sg_iter;
 	struct c4iw_dev *rhp;
 	struct c4iw_pd *php;
 	struct c4iw_mr *mhp;
@@ -537,11 +536,11 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	mhp->rhp = rhp;
 
-	mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
+	mhp->umem = ib_umem_get(udata, start, length, acc, 0);
 	if (IS_ERR(mhp->umem))
 		goto err_free_skb;
 
-	shift = mhp->umem->page_shift;
+	shift = PAGE_SHIFT;
 
 	n = mhp->umem->nmap;
 	err = alloc_pbl(mhp, n);
@@ -556,21 +555,16 @@ struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	i = n = 0;
 
-	for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
-		len = sg_dma_len(sg) >> shift;
-		for (k = 0; k < len; ++k) {
-			pages[i++] = cpu_to_be64(sg_dma_address(sg) +
-						 (k << shift));
-			if (i == PAGE_SIZE / sizeof *pages) {
-				err = write_pbl(&mhp->rhp->rdev,
-				      pages,
-				      mhp->attr.pbl_addr + (n << 3), i,
-				      mhp->wr_waitp);
-				if (err)
-					goto pbl_done;
-				n += i;
-				i = 0;
-			}
+	for_each_sg_dma_page(mhp->umem->sg_head.sgl, &sg_iter, mhp->umem->nmap, 0) {
+		pages[i++] = cpu_to_be64(sg_page_iter_dma_address(&sg_iter));
+		if (i == PAGE_SIZE / sizeof(*pages)) {
+			err = write_pbl(&mhp->rhp->rdev, pages,
+					mhp->attr.pbl_addr + (n << 3), i,
+					mhp->wr_waitp);
+			if (err)
+				goto pbl_done;
+			n += i;
+			i = 0;
 		}
 	}
 
@@ -684,8 +678,8 @@ int c4iw_dealloc_mw(struct ib_mw *mw)
 			  mhp->wr_waitp);
 	kfree_skb(mhp->dereg_skb);
 	c4iw_put_wr_wait(mhp->wr_waitp);
-	kfree(mhp);
 	pr_debug("ib_mw %p mmid 0x%x ptr %p\n", mw, mmid, mhp);
+	kfree(mhp);
 	return 0;
 }
 

drivers/infiniband/hw/cxgb4/provider.c

@@ -58,51 +58,34 @@ static int fastreg_support = 1;
 module_param(fastreg_support, int, 0644);
 MODULE_PARM_DESC(fastreg_support, "Advertise fastreg support (default=1)");
 
-void _c4iw_free_ucontext(struct kref *kref)
+static void c4iw_dealloc_ucontext(struct ib_ucontext *context)
 {
-	struct c4iw_ucontext *ucontext;
+	struct c4iw_ucontext *ucontext = to_c4iw_ucontext(context);
 	struct c4iw_dev *rhp;
 	struct c4iw_mm_entry *mm, *tmp;
 
-	ucontext = container_of(kref, struct c4iw_ucontext, kref);
+	pr_debug("context %p\n", context);
 	rhp = to_c4iw_dev(ucontext->ibucontext.device);
 
-	pr_debug("ucontext %p\n", ucontext);
 	list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
 		kfree(mm);
 	c4iw_release_dev_ucontext(&rhp->rdev, &ucontext->uctx);
-	kfree(ucontext);
 }
 
-static int c4iw_dealloc_ucontext(struct ib_ucontext *context)
+static int c4iw_alloc_ucontext(struct ib_ucontext *ucontext,
+			       struct ib_udata *udata)
 {
-	struct c4iw_ucontext *ucontext = to_c4iw_ucontext(context);
-
-	pr_debug("context %p\n", context);
-	c4iw_put_ucontext(ucontext);
-	return 0;
-}
-
-static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev,
-					       struct ib_udata *udata)
-{
-	struct c4iw_ucontext *context;
+	struct ib_device *ibdev = ucontext->device;
+	struct c4iw_ucontext *context = to_c4iw_ucontext(ucontext);
 	struct c4iw_dev *rhp = to_c4iw_dev(ibdev);
 	struct c4iw_alloc_ucontext_resp uresp;
 	int ret = 0;
 	struct c4iw_mm_entry *mm = NULL;
 
 	pr_debug("ibdev %p\n", ibdev);
-	context = kzalloc(sizeof(*context), GFP_KERNEL);
-	if (!context) {
-		ret = -ENOMEM;
-		goto err;
-	}
-
 	c4iw_init_dev_ucontext(&rhp->rdev, &context->uctx);
 	INIT_LIST_HEAD(&context->mmaps);
 	spin_lock_init(&context->mmap_lock);
-	kref_init(&context->kref);
 
 	if (udata->outlen < sizeof(uresp) - sizeof(uresp.reserved)) {
 		pr_err_once("Warning - downlevel libcxgb4 (non-fatal), device status page disabled\n");
@@ -111,7 +94,7 @@ static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev,
 		mm = kmalloc(sizeof(*mm), GFP_KERNEL);
 		if (!mm) {
 			ret = -ENOMEM;
-			goto err_free;
+			goto err;
 		}
 
 		uresp.status_page_size = PAGE_SIZE;
@@ -131,13 +114,11 @@ static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev,
 		mm->len = PAGE_SIZE;
 		insert_mmap(context, mm);
 	}
-	return &context->ibucontext;
+	return 0;
 err_mm:
 	kfree(mm);
-err_free:
-	kfree(context);
 err:
-	return ERR_PTR(ret);
+	return ret;
 }
 
 static int c4iw_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
@@ -209,7 +190,7 @@ static int c4iw_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 	return ret;
 }
 
-static int c4iw_deallocate_pd(struct ib_pd *pd)
+static void c4iw_deallocate_pd(struct ib_pd *pd)
 {
 	struct c4iw_dev *rhp;
 	struct c4iw_pd *php;
@@ -221,15 +202,13 @@ static int c4iw_deallocate_pd(struct ib_pd *pd)
 	mutex_lock(&rhp->rdev.stats.lock);
 	rhp->rdev.stats.pd.cur--;
 	mutex_unlock(&rhp->rdev.stats.lock);
-	kfree(php);
-	return 0;
 }
 
-static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev,
-				      struct ib_ucontext *context,
-				      struct ib_udata *udata)
+static int c4iw_allocate_pd(struct ib_pd *pd, struct ib_ucontext *context,
+			    struct ib_udata *udata)
 {
-	struct c4iw_pd *php;
+	struct c4iw_pd *php = to_c4iw_pd(pd);
+	struct ib_device *ibdev = pd->device;
 	u32 pdid;
 	struct c4iw_dev *rhp;
 
@@ -237,12 +216,8 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev,
 	rhp = (struct c4iw_dev *) ibdev;
 	pdid =  c4iw_get_resource(&rhp->rdev.resource.pdid_table);
 	if (!pdid)
-		return ERR_PTR(-EINVAL);
-	php = kzalloc(sizeof(*php), GFP_KERNEL);
-	if (!php) {
-		c4iw_put_resource(&rhp->rdev.resource.pdid_table, pdid);
-		return ERR_PTR(-ENOMEM);
-	}
+		return -EINVAL;
+
 	php->pdid = pdid;
 	php->rhp = rhp;
 	if (context) {
@@ -250,7 +225,7 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev,
 
 		if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) {
 			c4iw_deallocate_pd(&php->ibpd);
-			return ERR_PTR(-EFAULT);
+			return -EFAULT;
 		}
 	}
 	mutex_lock(&rhp->rdev.stats.lock);
@@ -259,7 +234,7 @@ static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev,
 		rhp->rdev.stats.pd.max = rhp->rdev.stats.pd.cur;
 	mutex_unlock(&rhp->rdev.stats.lock);
 	pr_debug("pdid 0x%0x ptr 0x%p\n", pdid, php);
-	return &php->ibpd;
+	return 0;
 }
 
 static int c4iw_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
@@ -376,8 +351,9 @@ static int c4iw_query_port(struct ib_device *ibdev, u8 port,
 static ssize_t hw_rev_show(struct device *dev,
 			   struct device_attribute *attr, char *buf)
 {
-	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
-						 ibdev.dev);
+	struct c4iw_dev *c4iw_dev =
+			rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev);
+
 	pr_debug("dev 0x%p\n", dev);
 	return sprintf(buf, "%d\n",
 		       CHELSIO_CHIP_RELEASE(c4iw_dev->rdev.lldi.adapter_type));
@@ -387,8 +363,8 @@ static DEVICE_ATTR_RO(hw_rev);
 static ssize_t hca_type_show(struct device *dev,
 			     struct device_attribute *attr, char *buf)
 {
-	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
-						 ibdev.dev);
+	struct c4iw_dev *c4iw_dev =
+			rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev);
 	struct ethtool_drvinfo info;
 	struct net_device *lldev = c4iw_dev->rdev.lldi.ports[0];
 
@@ -401,8 +377,9 @@ static DEVICE_ATTR_RO(hca_type);
 static ssize_t board_id_show(struct device *dev, struct device_attribute *attr,
 			     char *buf)
 {
-	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
-						 ibdev.dev);
+	struct c4iw_dev *c4iw_dev =
+			rdma_device_to_drv_device(dev, struct c4iw_dev, ibdev);
+
 	pr_debug("dev 0x%p\n", dev);
 	return sprintf(buf, "%x.%x\n", c4iw_dev->rdev.lldi.pdev->vendor,
 		       c4iw_dev->rdev.lldi.pdev->device);
@@ -547,6 +524,7 @@ static const struct ib_device_ops c4iw_dev_ops = {
 	.destroy_cq = c4iw_destroy_cq,
 	.destroy_qp = c4iw_destroy_qp,
 	.destroy_srq = c4iw_destroy_srq,
+	.fill_res_entry = fill_res_entry,
 	.get_dev_fw_str = get_dev_fw_str,
 	.get_dma_mr = c4iw_get_dma_mr,
 	.get_hw_stats = c4iw_get_mib,
@@ -567,6 +545,8 @@ static const struct ib_device_ops c4iw_dev_ops = {
 	.query_qp = c4iw_ib_query_qp,
 	.reg_user_mr = c4iw_reg_user_mr,
 	.req_notify_cq = c4iw_arm_cq,
+	INIT_RDMA_OBJ_SIZE(ib_pd, c4iw_pd, ibpd),
+	INIT_RDMA_OBJ_SIZE(ib_ucontext, c4iw_ucontext, ibucontext),
 };
 
 void c4iw_register_device(struct work_struct *work)
@@ -613,7 +593,7 @@ void c4iw_register_device(struct work_struct *work)
 	dev->ibdev.dev.parent = &dev->rdev.lldi.pdev->dev;
 	dev->ibdev.uverbs_abi_ver = C4IW_UVERBS_ABI_VERSION;
 
-	dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
+	dev->ibdev.iwcm = kzalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
 	if (!dev->ibdev.iwcm) {
 		ret = -ENOMEM;
 		goto err_dealloc_ctx;
@@ -627,14 +607,13 @@ void c4iw_register_device(struct work_struct *work)
 	dev->ibdev.iwcm->add_ref = c4iw_qp_add_ref;
 	dev->ibdev.iwcm->rem_ref = c4iw_qp_rem_ref;
 	dev->ibdev.iwcm->get_qp = c4iw_get_qp;
-	dev->ibdev.res.fill_res_entry = fill_res_entry;
 	memcpy(dev->ibdev.iwcm->ifname, dev->rdev.lldi.ports[0]->name,
 	       sizeof(dev->ibdev.iwcm->ifname));
 
 	rdma_set_device_sysfs_group(&dev->ibdev, &c4iw_attr_group);
 	dev->ibdev.driver_id = RDMA_DRIVER_CXGB4;
 	ib_set_device_ops(&dev->ibdev, &c4iw_dev_ops);
-	ret = ib_register_device(&dev->ibdev, "cxgb4_%d", NULL);
+	ret = ib_register_device(&dev->ibdev, "cxgb4_%d");
 	if (ret)
 		goto err_kfree_iwcm;
 	return;

drivers/infiniband/hw/cxgb4/qp.c

@@ -31,6 +31,7 @@
  */
 
 #include <linux/module.h>
+#include <rdma/uverbs_ioctl.h>
 
 #include "iw_cxgb4.h"
 
@@ -632,7 +633,10 @@ static void build_rdma_write_cmpl(struct t4_sq *sq,
 
 	wcwr->stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
 	wcwr->to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
-	wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey);
+	if (wr->next->opcode == IB_WR_SEND)
+		wcwr->stag_inv = 0;
+	else
+		wcwr->stag_inv = cpu_to_be32(wr->next->ex.invalidate_rkey);
 	wcwr->r2 = 0;
 	wcwr->r3 = 0;
 
@@ -726,7 +730,10 @@ static void post_write_cmpl(struct c4iw_qp *qhp, const struct ib_send_wr *wr)
 
 	/* SEND_WITH_INV swsqe */
 	swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
-	swsqe->opcode = FW_RI_SEND_WITH_INV;
+	if (wr->next->opcode == IB_WR_SEND)
+		swsqe->opcode = FW_RI_SEND;
+	else
+		swsqe->opcode = FW_RI_SEND_WITH_INV;
 	swsqe->idx = qhp->wq.sq.pidx;
 	swsqe->complete = 0;
 	swsqe->signaled = send_signaled;
@@ -897,8 +904,6 @@ static void free_qp_work(struct work_struct *work)
 	destroy_qp(&rhp->rdev, &qhp->wq,
 		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx, !qhp->srq);
 
-	if (ucontext)
-		c4iw_put_ucontext(ucontext);
 	c4iw_put_wr_wait(qhp->wr_waitp);
 	kfree(qhp);
 }
@@ -1133,9 +1138,9 @@ int c4iw_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 	/*
 	 * Fastpath for NVMe-oF target WRITE + SEND_WITH_INV wr chain which is
 	 * the response for small NVMEe-oF READ requests.  If the chain is
-	 * exactly a WRITE->SEND_WITH_INV and the sgl depths and lengths
-	 * meet the requirements of the fw_ri_write_cmpl_wr work request,
-	 * then build and post the write_cmpl WR.  If any of the tests
+	 * exactly a WRITE->SEND_WITH_INV or a WRITE->SEND and the sgl depths
+	 * and lengths meet the requirements of the fw_ri_write_cmpl_wr work
+	 * request, then build and post the write_cmpl WR. If any of the tests
 	 * below are not true, then we continue on with the tradtional WRITE
 	 * and SEND WRs.
 	 */
@@ -1145,7 +1150,8 @@ int c4iw_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
 	    wr && wr->next && !wr->next->next &&
 	    wr->opcode == IB_WR_RDMA_WRITE &&
 	    wr->sg_list[0].length && wr->num_sge <= T4_WRITE_CMPL_MAX_SGL &&
-	    wr->next->opcode == IB_WR_SEND_WITH_INV &&
+	    (wr->next->opcode == IB_WR_SEND ||
+	    wr->next->opcode == IB_WR_SEND_WITH_INV) &&
 	    wr->next->sg_list[0].length == T4_WRITE_CMPL_MAX_CQE &&
 	    wr->next->num_sge == 1 && num_wrs >= 2) {
 		post_write_cmpl(qhp, wr);
@@ -2129,7 +2135,8 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
 	struct c4iw_cq *rchp;
 	struct c4iw_create_qp_resp uresp;
 	unsigned int sqsize, rqsize = 0;
-	struct c4iw_ucontext *ucontext;
+	struct c4iw_ucontext *ucontext = rdma_udata_to_drv_context(
+		udata, struct c4iw_ucontext, ibucontext);
 	int ret;
 	struct c4iw_mm_entry *sq_key_mm, *rq_key_mm = NULL, *sq_db_key_mm;
 	struct c4iw_mm_entry *rq_db_key_mm = NULL, *ma_sync_key_mm = NULL;
@@ -2163,8 +2170,6 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
 	if (sqsize < 8)
 		sqsize = 8;
 
-	ucontext = udata ? to_c4iw_ucontext(pd->uobject->context) : NULL;
-
 	qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
 	if (!qhp)
 		return ERR_PTR(-ENOMEM);
@@ -2331,7 +2336,6 @@ struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
 			insert_mmap(ucontext, ma_sync_key_mm);
 		}
 
-		c4iw_get_ucontext(ucontext);
 		qhp->ucontext = ucontext;
 	}
 	if (!attrs->srq) {
@@ -2589,7 +2593,7 @@ static int alloc_srq_queue(struct c4iw_srq *srq, struct c4iw_dev_ucontext *uctx,
 	/* build fw_ri_res_wr */
 	wr_len = sizeof(*res_wr) + sizeof(*res);
 
-	skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
+	skb = alloc_skb(wr_len, GFP_KERNEL);
 	if (!skb)
 		goto err_free_queue;
 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
@@ -2711,7 +2715,8 @@ struct ib_srq *c4iw_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *attrs,
 	rqsize = attrs->attr.max_wr + 1;
 	rqsize = roundup_pow_of_two(max_t(u16, rqsize, 16));
 
-	ucontext = udata ? to_c4iw_ucontext(pd->uobject->context) : NULL;
+	ucontext = rdma_udata_to_drv_context(udata, struct c4iw_ucontext,
+					     ibucontext);
 
 	srq = kzalloc(sizeof(*srq), GFP_KERNEL);
 	if (!srq)

drivers/infiniband/hw/cxgb4/t4.h

@@ -35,6 +35,7 @@
 #include "t4_regs.h"
 #include "t4_values.h"
 #include "t4_msg.h"
+#include "t4_tcb.h"
 #include "t4fw_ri_api.h"
 
 #define T4_MAX_NUM_PD 65536