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

Pull rdma fixes from Jason Gunthorpe:
 "Several smaller driver fixes and a core RDMA CM regression fix:

   - Fix improperly accepting flags from userspace in mlx4

   - Add missing DMA barriers for irdma

   - Fix two kcsan warnings in irdma

   - Report the correct CQ op code to userspace in irdma

   - Report the correct MW bind error code for irdma

   - Load the destination address in RDMA CM to resolve a recent
     regression

   - Fix a QP regression in mthca

   - Remove a race processing completions in bnxt_re resulting in a
     crash

   - Fix driver unloading races with interrupts and tasklets in bnxt_re

   - Fix missing error unwind in rxe"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/irdma: Report correct WC error
  RDMA/irdma: Fix op_type reporting in CQEs
  RDMA/rxe: Fix an error handling path in rxe_bind_mw()
  RDMA/bnxt_re: Fix hang during driver unload
  RDMA/bnxt_re: Prevent handling any completions after qp destroy
  RDMA/mthca: Fix crash when polling CQ for shared QPs
  RDMA/core: Update CMA destination address on rdma_resolve_addr
  RDMA/irdma: Fix data race on CQP request done
  RDMA/irdma: Fix data race on CQP completion stats
  RDMA/irdma: Add missing read barriers
  RDMA/mlx4: Make check for invalid flags stricter

drivers/infiniband/core/cma.c

@@ -4062,6 +4062,8 @@ static int resolve_prepare_src(struct rdma_id_private *id_priv,
 					   RDMA_CM_ADDR_QUERY)))
 			return -EINVAL;
 
+	} else {
+		memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
 	}
 
 	if (cma_family(id_priv) != dst_addr->sa_family) {

drivers/infiniband/hw/bnxt_re/ib_verbs.c

@@ -869,7 +869,10 @@ static int bnxt_re_destroy_gsi_sqp(struct bnxt_re_qp *qp)
 int bnxt_re_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
 {
 	struct bnxt_re_qp *qp = container_of(ib_qp, struct bnxt_re_qp, ib_qp);
+	struct bnxt_qplib_qp *qplib_qp = &qp->qplib_qp;
 	struct bnxt_re_dev *rdev = qp->rdev;
+	struct bnxt_qplib_nq *scq_nq = NULL;
+	struct bnxt_qplib_nq *rcq_nq = NULL;
 	unsigned int flags;
 	int rc;
 
@@ -903,6 +906,15 @@ int bnxt_re_destroy_qp(struct ib_qp *ib_qp, struct ib_udata *udata)
 	ib_umem_release(qp->rumem);
 	ib_umem_release(qp->sumem);
 
+	/* Flush all the entries of notification queue associated with
+	 * given qp.
+	 */
+	scq_nq = qplib_qp->scq->nq;
+	rcq_nq = qplib_qp->rcq->nq;
+	bnxt_re_synchronize_nq(scq_nq);
+	if (scq_nq != rcq_nq)
+		bnxt_re_synchronize_nq(rcq_nq);
+
 	return 0;
 }
 

drivers/infiniband/hw/bnxt_re/qplib_fp.c

@@ -381,6 +381,24 @@ static void bnxt_qplib_service_nq(struct tasklet_struct *t)
 	spin_unlock_bh(&hwq->lock);
 }
 
+/* bnxt_re_synchronize_nq - self polling notification queue.
+ * @nq      -     notification queue pointer
+ *
+ * This function will start polling entries of a given notification queue
+ * for all pending  entries.
+ * This function is useful to synchronize notification entries while resources
+ * are going away.
+ */
+
+void bnxt_re_synchronize_nq(struct bnxt_qplib_nq *nq)
+{
+	int budget = nq->budget;
+
+	nq->budget = nq->hwq.max_elements;
+	bnxt_qplib_service_nq(&nq->nq_tasklet);
+	nq->budget = budget;
+}
+
 static irqreturn_t bnxt_qplib_nq_irq(int irq, void *dev_instance)
 {
 	struct bnxt_qplib_nq *nq = dev_instance;
@@ -402,19 +420,19 @@ void bnxt_qplib_nq_stop_irq(struct bnxt_qplib_nq *nq, bool kill)
 	if (!nq->requested)
 		return;
 
-	tasklet_disable(&nq->nq_tasklet);
+	nq->requested = false;
 	/* Mask h/w interrupt */
 	bnxt_qplib_ring_nq_db(&nq->nq_db.dbinfo, nq->res->cctx, false);
 	/* Sync with last running IRQ handler */
 	synchronize_irq(nq->msix_vec);
-	if (kill)
-		tasklet_kill(&nq->nq_tasklet);
-
 	irq_set_affinity_hint(nq->msix_vec, NULL);
 	free_irq(nq->msix_vec, nq);
 	kfree(nq->name);
 	nq->name = NULL;
-	nq->requested = false;
+
+	if (kill)
+		tasklet_kill(&nq->nq_tasklet);
+	tasklet_disable(&nq->nq_tasklet);
 }
 
 void bnxt_qplib_disable_nq(struct bnxt_qplib_nq *nq)

drivers/infiniband/hw/bnxt_re/qplib_fp.h

@@ -553,6 +553,7 @@ int bnxt_qplib_process_flush_list(struct bnxt_qplib_cq *cq,
 				  struct bnxt_qplib_cqe *cqe,
 				  int num_cqes);
 void bnxt_qplib_flush_cqn_wq(struct bnxt_qplib_qp *qp);
+void bnxt_re_synchronize_nq(struct bnxt_qplib_nq *nq);
 
 static inline void *bnxt_qplib_get_swqe(struct bnxt_qplib_q *que, u32 *swq_idx)
 {

drivers/infiniband/hw/bnxt_re/qplib_rcfw.c

@@ -989,19 +989,18 @@ void bnxt_qplib_rcfw_stop_irq(struct bnxt_qplib_rcfw *rcfw, bool kill)
 	if (!creq->requested)
 		return;
 
-	tasklet_disable(&creq->creq_tasklet);
+	creq->requested = false;
 	/* Mask h/w interrupts */
 	bnxt_qplib_ring_nq_db(&creq->creq_db.dbinfo, rcfw->res->cctx, false);
 	/* Sync with last running IRQ-handler */
 	synchronize_irq(creq->msix_vec);
-	if (kill)
-		tasklet_kill(&creq->creq_tasklet);
-
 	free_irq(creq->msix_vec, rcfw);
 	kfree(creq->irq_name);
 	creq->irq_name = NULL;
-	creq->requested = false;
 	atomic_set(&rcfw->rcfw_intr_enabled, 0);
+	if (kill)
+		tasklet_kill(&creq->creq_tasklet);
+	tasklet_disable(&creq->creq_tasklet);
 }
 
 void bnxt_qplib_disable_rcfw_channel(struct bnxt_qplib_rcfw *rcfw)

drivers/infiniband/hw/irdma/ctrl.c

@@ -2712,13 +2712,13 @@ static int irdma_sc_cq_modify(struct irdma_sc_cq *cq,
  */
 void irdma_check_cqp_progress(struct irdma_cqp_timeout *timeout, struct irdma_sc_dev *dev)
 {
-	if (timeout->compl_cqp_cmds != dev->cqp_cmd_stats[IRDMA_OP_CMPL_CMDS]) {
-		timeout->compl_cqp_cmds = dev->cqp_cmd_stats[IRDMA_OP_CMPL_CMDS];
+	u64 completed_ops = atomic64_read(&dev->cqp->completed_ops);
+
+	if (timeout->compl_cqp_cmds != completed_ops) {
+		timeout->compl_cqp_cmds = completed_ops;
 		timeout->count = 0;
-	} else {
-		if (dev->cqp_cmd_stats[IRDMA_OP_REQ_CMDS] !=
-		    timeout->compl_cqp_cmds)
-			timeout->count++;
+	} else if (timeout->compl_cqp_cmds != dev->cqp->requested_ops) {
+		timeout->count++;
 	}
 }
 
@@ -2761,7 +2761,7 @@ static int irdma_cqp_poll_registers(struct irdma_sc_cqp *cqp, u32 tail,
 		if (newtail != tail) {
 			/* SUCCESS */
 			IRDMA_RING_MOVE_TAIL(cqp->sq_ring);
-			cqp->dev->cqp_cmd_stats[IRDMA_OP_CMPL_CMDS]++;
+			atomic64_inc(&cqp->completed_ops);
 			return 0;
 		}
 		udelay(cqp->dev->hw_attrs.max_sleep_count);
@@ -3121,8 +3121,8 @@ int irdma_sc_cqp_init(struct irdma_sc_cqp *cqp,
 	info->dev->cqp = cqp;
 
 	IRDMA_RING_INIT(cqp->sq_ring, cqp->sq_size);
-	cqp->dev->cqp_cmd_stats[IRDMA_OP_REQ_CMDS] = 0;
-	cqp->dev->cqp_cmd_stats[IRDMA_OP_CMPL_CMDS] = 0;
+	cqp->requested_ops = 0;
+	atomic64_set(&cqp->completed_ops, 0);
 	/* for the cqp commands backlog. */
 	INIT_LIST_HEAD(&cqp->dev->cqp_cmd_head);
 
@@ -3274,7 +3274,7 @@ __le64 *irdma_sc_cqp_get_next_send_wqe_idx(struct irdma_sc_cqp *cqp, u64 scratch
 	if (ret_code)
 		return NULL;
 
-	cqp->dev->cqp_cmd_stats[IRDMA_OP_REQ_CMDS]++;
+	cqp->requested_ops++;
 	if (!*wqe_idx)
 		cqp->polarity = !cqp->polarity;
 	wqe = cqp->sq_base[*wqe_idx].elem;
@@ -3363,6 +3363,9 @@ int irdma_sc_ccq_get_cqe_info(struct irdma_sc_cq *ccq,
 	if (polarity != ccq->cq_uk.polarity)
 		return -ENOENT;
 
+	/* Ensure CEQE contents are read after valid bit is checked */
+	dma_rmb();
+
 	get_64bit_val(cqe, 8, &qp_ctx);
 	cqp = (struct irdma_sc_cqp *)(unsigned long)qp_ctx;
 	info->error = (bool)FIELD_GET(IRDMA_CQ_ERROR, temp);
@@ -3397,7 +3400,7 @@ int irdma_sc_ccq_get_cqe_info(struct irdma_sc_cq *ccq,
 	dma_wmb(); /* make sure shadow area is updated before moving tail */
 
 	IRDMA_RING_MOVE_TAIL(cqp->sq_ring);
-	ccq->dev->cqp_cmd_stats[IRDMA_OP_CMPL_CMDS]++;
+	atomic64_inc(&cqp->completed_ops);
 
 	return ret_code;
 }
@@ -4009,13 +4012,17 @@ int irdma_sc_get_next_aeqe(struct irdma_sc_aeq *aeq,
 	u8 polarity;
 
 	aeqe = IRDMA_GET_CURRENT_AEQ_ELEM(aeq);
-	get_64bit_val(aeqe, 0, &compl_ctx);
 	get_64bit_val(aeqe, 8, &temp);
 	polarity = (u8)FIELD_GET(IRDMA_AEQE_VALID, temp);
 
 	if (aeq->polarity != polarity)
 		return -ENOENT;
 
+	/* Ensure AEQE contents are read after valid bit is checked */
+	dma_rmb();
+
+	get_64bit_val(aeqe, 0, &compl_ctx);
+
 	print_hex_dump_debug("WQE: AEQ_ENTRY WQE", DUMP_PREFIX_OFFSET, 16, 8,
 			     aeqe, 16, false);
 

drivers/infiniband/hw/irdma/defs.h

@@ -191,32 +191,30 @@ enum irdma_cqp_op_type {
 	IRDMA_OP_MANAGE_VF_PBLE_BP		= 25,
 	IRDMA_OP_QUERY_FPM_VAL			= 26,
 	IRDMA_OP_COMMIT_FPM_VAL			= 27,
-	IRDMA_OP_REQ_CMDS			= 28,
-	IRDMA_OP_CMPL_CMDS			= 29,
-	IRDMA_OP_AH_CREATE			= 30,
-	IRDMA_OP_AH_MODIFY			= 31,
-	IRDMA_OP_AH_DESTROY			= 32,
-	IRDMA_OP_MC_CREATE			= 33,
-	IRDMA_OP_MC_DESTROY			= 34,
-	IRDMA_OP_MC_MODIFY			= 35,
-	IRDMA_OP_STATS_ALLOCATE			= 36,
-	IRDMA_OP_STATS_FREE			= 37,
-	IRDMA_OP_STATS_GATHER			= 38,
-	IRDMA_OP_WS_ADD_NODE			= 39,
-	IRDMA_OP_WS_MODIFY_NODE			= 40,
-	IRDMA_OP_WS_DELETE_NODE			= 41,
-	IRDMA_OP_WS_FAILOVER_START		= 42,
-	IRDMA_OP_WS_FAILOVER_COMPLETE		= 43,
-	IRDMA_OP_SET_UP_MAP			= 44,
-	IRDMA_OP_GEN_AE				= 45,
-	IRDMA_OP_QUERY_RDMA_FEATURES		= 46,
-	IRDMA_OP_ALLOC_LOCAL_MAC_ENTRY		= 47,
-	IRDMA_OP_ADD_LOCAL_MAC_ENTRY		= 48,
-	IRDMA_OP_DELETE_LOCAL_MAC_ENTRY		= 49,
-	IRDMA_OP_CQ_MODIFY			= 50,
+	IRDMA_OP_AH_CREATE			= 28,
+	IRDMA_OP_AH_MODIFY			= 29,
+	IRDMA_OP_AH_DESTROY			= 30,
+	IRDMA_OP_MC_CREATE			= 31,
+	IRDMA_OP_MC_DESTROY			= 32,
+	IRDMA_OP_MC_MODIFY			= 33,
+	IRDMA_OP_STATS_ALLOCATE			= 34,
+	IRDMA_OP_STATS_FREE			= 35,
+	IRDMA_OP_STATS_GATHER			= 36,
+	IRDMA_OP_WS_ADD_NODE			= 37,
+	IRDMA_OP_WS_MODIFY_NODE			= 38,
+	IRDMA_OP_WS_DELETE_NODE			= 39,
+	IRDMA_OP_WS_FAILOVER_START		= 40,
+	IRDMA_OP_WS_FAILOVER_COMPLETE		= 41,
+	IRDMA_OP_SET_UP_MAP			= 42,
+	IRDMA_OP_GEN_AE				= 43,
+	IRDMA_OP_QUERY_RDMA_FEATURES		= 44,
+	IRDMA_OP_ALLOC_LOCAL_MAC_ENTRY		= 45,
+	IRDMA_OP_ADD_LOCAL_MAC_ENTRY		= 46,
+	IRDMA_OP_DELETE_LOCAL_MAC_ENTRY		= 47,
+	IRDMA_OP_CQ_MODIFY			= 48,
 
 	/* Must be last entry*/
-	IRDMA_MAX_CQP_OPS			= 51,
+	IRDMA_MAX_CQP_OPS			= 49,
 };
 
 /* CQP SQ WQES */

drivers/infiniband/hw/irdma/hw.c

@@ -191,6 +191,7 @@ static void irdma_set_flush_fields(struct irdma_sc_qp *qp,
 	case IRDMA_AE_AMP_MWBIND_INVALID_RIGHTS:
 	case IRDMA_AE_AMP_MWBIND_BIND_DISABLED:
 	case IRDMA_AE_AMP_MWBIND_INVALID_BOUNDS:
+	case IRDMA_AE_AMP_MWBIND_VALID_STAG:
 		qp->flush_code = FLUSH_MW_BIND_ERR;
 		qp->event_type = IRDMA_QP_EVENT_ACCESS_ERR;
 		break;
@@ -2075,7 +2076,7 @@ void irdma_cqp_ce_handler(struct irdma_pci_f *rf, struct irdma_sc_cq *cq)
 			cqp_request->compl_info.error = info.error;
 
 			if (cqp_request->waiting) {
-				cqp_request->request_done = true;
+				WRITE_ONCE(cqp_request->request_done, true);
 				wake_up(&cqp_request->waitq);
 				irdma_put_cqp_request(&rf->cqp, cqp_request);
 			} else {

drivers/infiniband/hw/irdma/main.h

@@ -161,8 +161,8 @@ struct irdma_cqp_request {
 	void (*callback_fcn)(struct irdma_cqp_request *cqp_request);
 	void *param;
 	struct irdma_cqp_compl_info compl_info;
+	bool request_done; /* READ/WRITE_ONCE macros operate on it */
 	bool waiting:1;
-	bool request_done:1;
 	bool dynamic:1;
 };
 

drivers/infiniband/hw/irdma/puda.c

@@ -230,6 +230,9 @@ static int irdma_puda_poll_info(struct irdma_sc_cq *cq,
 	if (valid_bit != cq_uk->polarity)
 		return -ENOENT;
 
+	/* Ensure CQE contents are read after valid bit is checked */
+	dma_rmb();
+
 	if (cq->dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
 		ext_valid = (bool)FIELD_GET(IRDMA_CQ_EXTCQE, qword3);
 
@@ -243,6 +246,9 @@ static int irdma_puda_poll_info(struct irdma_sc_cq *cq,
 		if (polarity != cq_uk->polarity)
 			return -ENOENT;
 
+		/* Ensure ext CQE contents are read after ext valid bit is checked */
+		dma_rmb();
+
 		IRDMA_RING_MOVE_HEAD_NOCHECK(cq_uk->cq_ring);
 		if (!IRDMA_RING_CURRENT_HEAD(cq_uk->cq_ring))
 			cq_uk->polarity = !cq_uk->polarity;

drivers/infiniband/hw/irdma/type.h

@@ -365,6 +365,8 @@ struct irdma_sc_cqp {
 	struct irdma_dcqcn_cc_params dcqcn_params;
 	__le64 *host_ctx;
 	u64 *scratch_array;
+	u64 requested_ops;
+	atomic64_t completed_ops;
 	u32 cqp_id;
 	u32 sq_size;
 	u32 hw_sq_size;

drivers/infiniband/hw/irdma/uk.c

@@ -1161,7 +1161,7 @@ int irdma_uk_cq_poll_cmpl(struct irdma_cq_uk *cq,
 	}
 	wqe_idx = (u32)FIELD_GET(IRDMA_CQ_WQEIDX, qword3);
 	info->qp_handle = (irdma_qp_handle)(unsigned long)qp;
-	info->op_type = (u8)FIELD_GET(IRDMA_CQ_SQ, qword3);
+	info->op_type = (u8)FIELD_GET(IRDMACQ_OP, qword3);
 
 	if (info->q_type == IRDMA_CQE_QTYPE_RQ) {
 		u32 array_idx;
@@ -1527,6 +1527,9 @@ void irdma_uk_clean_cq(void *q, struct irdma_cq_uk *cq)
 		if (polarity != temp)
 			break;
 
+		/* Ensure CQE contents are read after valid bit is checked */
+		dma_rmb();
+
 		get_64bit_val(cqe, 8, &comp_ctx);
 		if ((void *)(unsigned long)comp_ctx == q)
 			set_64bit_val(cqe, 8, 0);

drivers/infiniband/hw/irdma/utils.c

@@ -481,7 +481,7 @@ void irdma_free_cqp_request(struct irdma_cqp *cqp,
 	if (cqp_request->dynamic) {
 		kfree(cqp_request);
 	} else {
-		cqp_request->request_done = false;
+		WRITE_ONCE(cqp_request->request_done, false);
 		cqp_request->callback_fcn = NULL;
 		cqp_request->waiting = false;
 
@@ -515,7 +515,7 @@ irdma_free_pending_cqp_request(struct irdma_cqp *cqp,
 {
 	if (cqp_request->waiting) {
 		cqp_request->compl_info.error = true;
-		cqp_request->request_done = true;
+		WRITE_ONCE(cqp_request->request_done, true);
 		wake_up(&cqp_request->waitq);
 	}
 	wait_event_timeout(cqp->remove_wq,
@@ -567,11 +567,11 @@ static int irdma_wait_event(struct irdma_pci_f *rf,
 	bool cqp_error = false;
 	int err_code = 0;
 
-	cqp_timeout.compl_cqp_cmds = rf->sc_dev.cqp_cmd_stats[IRDMA_OP_CMPL_CMDS];
+	cqp_timeout.compl_cqp_cmds = atomic64_read(&rf->sc_dev.cqp->completed_ops);
 	do {
 		irdma_cqp_ce_handler(rf, &rf->ccq.sc_cq);
 		if (wait_event_timeout(cqp_request->waitq,
-				       cqp_request->request_done,
+				       READ_ONCE(cqp_request->request_done),
 				       msecs_to_jiffies(CQP_COMPL_WAIT_TIME_MS)))
 			break;
 

drivers/infiniband/hw/mlx4/qp.c

@@ -565,15 +565,15 @@ static int set_qp_rss(struct mlx4_ib_dev *dev, struct mlx4_ib_rss *rss_ctx,
 		return (-EOPNOTSUPP);
 	}
 
-	if (ucmd->rx_hash_fields_mask & ~(MLX4_IB_RX_HASH_SRC_IPV4	|
-					  MLX4_IB_RX_HASH_DST_IPV4	|
-					  MLX4_IB_RX_HASH_SRC_IPV6	|
-					  MLX4_IB_RX_HASH_DST_IPV6	|
-					  MLX4_IB_RX_HASH_SRC_PORT_TCP	|
-					  MLX4_IB_RX_HASH_DST_PORT_TCP	|
-					  MLX4_IB_RX_HASH_SRC_PORT_UDP	|
-					  MLX4_IB_RX_HASH_DST_PORT_UDP  |
-					  MLX4_IB_RX_HASH_INNER)) {
+	if (ucmd->rx_hash_fields_mask & ~(u64)(MLX4_IB_RX_HASH_SRC_IPV4	|
+					       MLX4_IB_RX_HASH_DST_IPV4	|
+					       MLX4_IB_RX_HASH_SRC_IPV6	|
+					       MLX4_IB_RX_HASH_DST_IPV6	|
+					       MLX4_IB_RX_HASH_SRC_PORT_TCP |
+					       MLX4_IB_RX_HASH_DST_PORT_TCP |
+					       MLX4_IB_RX_HASH_SRC_PORT_UDP |
+					       MLX4_IB_RX_HASH_DST_PORT_UDP |
+					       MLX4_IB_RX_HASH_INNER)) {
 		pr_debug("RX Hash fields_mask has unsupported mask (0x%llx)\n",
 			 ucmd->rx_hash_fields_mask);
 		return (-EOPNOTSUPP);

drivers/infiniband/hw/mthca/mthca_qp.c

@@ -1393,7 +1393,7 @@ int mthca_alloc_sqp(struct mthca_dev *dev,
 	if (mthca_array_get(&dev->qp_table.qp, mqpn))
 		err = -EBUSY;
 	else
-		mthca_array_set(&dev->qp_table.qp, mqpn, qp->sqp);
+		mthca_array_set(&dev->qp_table.qp, mqpn, qp);
 	spin_unlock_irq(&dev->qp_table.lock);
 
 	if (err)

drivers/infiniband/sw/rxe/rxe_mw.c

@@ -199,7 +199,8 @@ int rxe_bind_mw(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
 
 	if (access & ~RXE_ACCESS_SUPPORTED_MW) {
 		rxe_err_mw(mw, "access %#x not supported", access);
-		return -EOPNOTSUPP;
+		ret = -EOPNOTSUPP;
+		goto err_drop_mr;
 	}
 
 	spin_lock_bh(&mw->lock);