RDMA/bnxt_re: synchronize the qp-handle table array

There is a race between the CREQ tasklet and destroy qp when accessing the
qp-handle table. There is a chance of reading a valid qp-handle in the
CREQ tasklet handler while the QP is already moving ahead with the
destruction.

Fixing this race by implementing a table-lock to synchronize the access.

Fixes: f218d67 ("RDMA/bnxt_re: Allow posting when QPs are in error")
Fixes: 84cf229 ("RDMA/bnxt_re: Fix the qp table indexing")
Link: https://patch.msgid.link/r/[email protected]
Signed-off-by: Kalesh AP <[email protected]>
Signed-off-by: Selvin Xavier <[email protected]>
Signed-off-by: Jason Gunthorpe <[email protected]>

Author: selvintxavier

drivers/infiniband/hw/bnxt_re/qplib_fp.c

@@ -1532,9 +1532,11 @@ int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
 	u32 tbl_indx;
 	int rc;
 
+	spin_lock_bh(&rcfw->tbl_lock);
 	tbl_indx = map_qp_id_to_tbl_indx(qp->id, rcfw);
 	rcfw->qp_tbl[tbl_indx].qp_id = BNXT_QPLIB_QP_ID_INVALID;
 	rcfw->qp_tbl[tbl_indx].qp_handle = NULL;
+	spin_unlock_bh(&rcfw->tbl_lock);
 
 	bnxt_qplib_rcfw_cmd_prep((struct cmdq_base *)&req,
 				 CMDQ_BASE_OPCODE_DESTROY_QP,
@@ -1545,8 +1547,10 @@ int bnxt_qplib_destroy_qp(struct bnxt_qplib_res *res,
 				sizeof(resp), 0);
 	rc = bnxt_qplib_rcfw_send_message(rcfw, &msg);
 	if (rc) {
+		spin_lock_bh(&rcfw->tbl_lock);
 		rcfw->qp_tbl[tbl_indx].qp_id = qp->id;
 		rcfw->qp_tbl[tbl_indx].qp_handle = qp;
+		spin_unlock_bh(&rcfw->tbl_lock);
 		return rc;
 	}
 

drivers/infiniband/hw/bnxt_re/qplib_rcfw.c

@@ -634,17 +634,21 @@ static int bnxt_qplib_process_qp_event(struct bnxt_qplib_rcfw *rcfw,
 	case CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION:
 		err_event = (struct creq_qp_error_notification *)qp_event;
 		qp_id = le32_to_cpu(err_event->xid);
+		spin_lock(&rcfw->tbl_lock);
 		tbl_indx = map_qp_id_to_tbl_indx(qp_id, rcfw);
 		qp = rcfw->qp_tbl[tbl_indx].qp_handle;
+		if (!qp) {
+			spin_unlock(&rcfw->tbl_lock);
+			break;
+		}
+		bnxt_qplib_mark_qp_error(qp);
+		rc = rcfw->creq.aeq_handler(rcfw, qp_event, qp);
+		spin_unlock(&rcfw->tbl_lock);
 		dev_dbg(&pdev->dev, "Received QP error notification\n");
 		dev_dbg(&pdev->dev,
 			"qpid 0x%x, req_err=0x%x, resp_err=0x%x\n",
 			qp_id, err_event->req_err_state_reason,
 			err_event->res_err_state_reason);
-		if (!qp)
-			break;
-		bnxt_qplib_mark_qp_error(qp);
-		rc = rcfw->creq.aeq_handler(rcfw, qp_event, qp);
 		break;
 	default:
 		/*
@@ -973,6 +977,7 @@ int bnxt_qplib_alloc_rcfw_channel(struct bnxt_qplib_res *res,
 			       GFP_KERNEL);
 	if (!rcfw->qp_tbl)
 		goto fail;
+	spin_lock_init(&rcfw->tbl_lock);
 
 	rcfw->max_timeout = res->cctx->hwrm_cmd_max_timeout;
 

drivers/infiniband/hw/bnxt_re/qplib_rcfw.h

@@ -224,6 +224,8 @@ struct bnxt_qplib_rcfw {
 	struct bnxt_qplib_crsqe		*crsqe_tbl;
 	int qp_tbl_size;
 	struct bnxt_qplib_qp_node *qp_tbl;
+	/* To synchronize the qp-handle hash table */
+	spinlock_t			tbl_lock;
 	u64 oos_prev;
 	u32 init_oos_stats;
 	u32 cmdq_depth;