Skip to contentSkip to conversation

qedf: Added the synchronization between IO completions and abort.

PATCH Applied Saurav Kashyap 2 participants 3 messages Sep 1, 2023 – Sep 14, 2023

Subsystems (2)
lore
Share & tools
lore

Saurav Kashyap

author maintainer
#0 lore raw

PATCH 1/1 qedf: Added the synchronization between IO completions and abort.

From: Javed Hasan <[email protected]>

This fix is added to avoid the race condition between IO completion and abort process by protecting the cmd_type with the lock.

---
 drivers/scsi/qedf/qedf_io.c   | 10 ++++++++--
 drivers/scsi/qedf/qedf_main.c |  7 ++++++-
 2 files changed, 14 insertions(+), 3 deletions(-)
drivers/scsi/qedf/qedf_io.c +8 -2
index 4750ec5789a8..10fe3383855c 100644
@@ -1904,6 +1904,7 @@ int qedf_initiate_abts(struct qedf_ioreq *io_req, bool return_scsi_cmd_on_abts)
 		goto drop_rdata_kref;
 	}
 
+	spin_lock_irqsave(&fcport->rport_lock, flags);
 	if (!test_bit(QEDF_CMD_OUTSTANDING, &io_req->flags) ||
 	    test_bit(QEDF_CMD_IN_CLEANUP, &io_req->flags) ||
 	    test_bit(QEDF_CMD_IN_ABORT, &io_req->flags)) {
@@ -1911,17 +1912,20 @@ int qedf_initiate_abts(struct qedf_ioreq *io_req, bool return_scsi_cmd_on_abts)
 			 "io_req xid=0x%x sc_cmd=%p already in cleanup or abort processing or already completed.\n",
 			 io_req->xid, io_req->sc_cmd);
 		rc = 1;
+		spin_unlock_irqrestore(&fcport->rport_lock, flags);
 		goto drop_rdata_kref;
 	}
 
+	/* Set the command type to abort */
+	io_req->cmd_type = QEDF_ABTS;
+	spin_unlock_irqrestore(&fcport->rport_lock, flags);
+
 	kref_get(&io_req->refcount);
 
 	xid = io_req->xid;
 	qedf->control_requests++;
 	qedf->packet_aborts++;
 
-	/* Set the command type to abort */
-	io_req->cmd_type = QEDF_ABTS;
 	io_req->return_scsi_cmd_on_abts = return_scsi_cmd_on_abts;
 
 	set_bit(QEDF_CMD_IN_ABORT, &io_req->flags);
@@ -2210,7 +2214,9 @@ int qedf_initiate_cleanup(struct qedf_ioreq *io_req,
 		  refcount, fcport, fcport->rdata->ids.port_id);
 
 	/* Cleanup cmds re-use the same TID as the original I/O */
+	spin_lock_irqsave(&fcport->rport_lock, flags);
 	io_req->cmd_type = QEDF_CLEANUP;
+	spin_unlock_irqrestore(&fcport->rport_lock, flags);
 	io_req->return_scsi_cmd_on_abts = return_scsi_cmd_on_abts;
 
 	init_completion(&io_req->cleanup_done);
drivers/scsi/qedf/qedf_main.c +6 -1
index 2a31ddc99dde..b5202026b0b9 100644
@@ -2804,6 +2804,8 @@ void qedf_process_cqe(struct qedf_ctx *qedf, struct fcoe_cqe *cqe)
 	struct qedf_ioreq *io_req;
 	struct qedf_rport *fcport;
 	u32 comp_type;
+	u8 io_comp_type;
+	unsigned long flags;
 
 	comp_type = (cqe->cqe_data >> FCOE_CQE_CQE_TYPE_SHIFT) &
 	    FCOE_CQE_CQE_TYPE_MASK;
@@ -2837,11 +2839,14 @@ void qedf_process_cqe(struct qedf_ctx *qedf, struct fcoe_cqe *cqe)
 		return;
 	}
 
+	spin_lock_irqsave(&fcport->rport_lock, flags);
+	io_comp_type = io_req->cmd_type;
+	spin_unlock_irqrestore(&fcport->rport_lock, flags);
 
 	switch (comp_type) {
 	case FCOE_GOOD_COMPLETION_CQE_TYPE:
 		atomic_inc(&fcport->free_sqes);
-		switch (io_req->cmd_type) {
+		switch (io_comp_type) {
 		case QEDF_SCSI_CMD:
 			qedf_scsi_completion(qedf, cqe, io_req);
 			break;
Signature · 1 line
2.23.1
#1 lore raw

Saurav,

Quoted 2 lines

This fix is added to avoid the race condition between IO completion and abort process by protecting the cmd_type with the lock.

Applied to 6.6/scsi-staging, thanks!

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering