[SCSI] bnx2fc: Avoid holding cq_lock when iounmap() is called

With kernel debugging enabled, holding cq_lock when calling
bnx2fc_free_session_resc() which calls iounmap() leads to a warning
stack trace [INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected].
iounmap() grabs a HARDIRQ-unsafe vmlist lock, so holding
spin_lock_bh(cq_lock) when calling iounmap() will trigger the LOCKDEP
warning.  Since cq_lock is required only to guard against deletion, hold
the lock just before freeing the cq.

Signed-off-by: Bhanu Prakash Gollapudi <bprakash@broadcom.com>
Signed-off-by: Nithin Nayak Sujir <nsujir@broadcom.com>
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>

1 files changed, 2 insertions, 2 deletions

diff --git a/drivers/scsi/bnx2fc/bnx2fc_tgt.c b/drivers/scsi/bnx2fc/bnx2fc_tgt.c
index 7ea93af60260..7cc05e4e82d5 100644
--- a/drivers/scsi/bnx2fc/bnx2fc_tgt.c
+++ b/drivers/scsi/bnx2fc/bnx2fc_tgt.c
@@ -304,10 +304,8 @@ static void bnx2fc_upload_session(struct fcoe_port *port,
                                 " not sent to FW\n");
 
         /* Free session resources */
-        spin_lock_bh(&tgt->cq_lock);
         bnx2fc_free_session_resc(hba, tgt);
         bnx2fc_free_conn_id(hba, tgt->fcoe_conn_id);
-        spin_unlock_bh(&tgt->cq_lock);
 }
 
 static int bnx2fc_init_tgt(struct bnx2fc_rport *tgt,
@@ -830,11 +828,13 @@ static void bnx2fc_free_session_resc(struct bnx2fc_hba *hba,
                 tgt->rq = NULL;
         }
         /* Free CQ */
+        spin_lock_bh(&tgt->cq_lock);
         if (tgt->cq) {
                 dma_free_coherent(&hba->pcidev->dev, tgt->cq_mem_size,
                                     tgt->cq, tgt->cq_dma);
                 tgt->cq = NULL;
         }
+        spin_unlock_bh(&tgt->cq_lock);
         /* Free SQ */
         if (tgt->sq) {
                 dma_free_coherent(&hba->pcidev->dev, tgt->sq_mem_size,