Long-lived sessions under high load can accumulate a starvation counter,
and the current implementation does not allow this counter to be reset
during an active session.

If HBA sends correct ATIO IOCB, then it has enough resources to process
commands and we should not call ISP recovery.

Cc: stable@vger.kernel.org
Fixes: ead038556f64 ("qla2xxx: Add Dual mode support in the driver")
Signed-off-by: Anastasia Kovaleva <a.kovaleva@yadro.com>
Reviewed-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
---
 drivers/scsi/qla2xxx/qla_isr.c    |  4 ++++
 drivers/scsi/qla2xxx/qla_target.c | 11 +++++++++++
 2 files changed, 15 insertions(+)

diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
index fe98c76e9be3..5234ce0985e0 100644
--- a/drivers/scsi/qla2xxx/qla_isr.c
+++ b/drivers/scsi/qla2xxx/qla_isr.c
@@ -1959,6 +1959,10 @@ qla2x00_async_event(scsi_qla_host_t *vha, struct rsp_que *rsp, uint16_t *mb)
 		ql_dbg(ql_dbg_async, vha, 0x5091, "Transceiver Removal\n");
 		break;
 
+	case MBA_REJECTED_FCP_CMD:
+		ql_dbg(ql_dbg_async, vha, 0x5092, "LS_RJT was sent. No resources to process the ELS request.\n");
+		break;
+
 	default:
 		ql_dbg(ql_dbg_async, vha, 0x5057,
 		    "Unknown AEN:%04x %04x %04x %04x\n",
diff --git a/drivers/scsi/qla2xxx/qla_target.c b/drivers/scsi/qla2xxx/qla_target.c
index d7551b1443e4..bc7feef6ee79 100644
--- a/drivers/scsi/qla2xxx/qla_target.c
+++ b/drivers/scsi/qla2xxx/qla_target.c
@@ -6801,6 +6801,7 @@ qlt_24xx_process_atio_queue(struct scsi_qla_host *vha, uint8_t ha_locked)
 	struct qla_hw_data *ha = vha->hw;
 	struct atio_from_isp *pkt;
 	int cnt, i;
+	unsigned long flags = 0;
 
 	if (!ha->flags.fw_started)
 		return;
@@ -6826,6 +6827,16 @@ qlt_24xx_process_atio_queue(struct scsi_qla_host *vha, uint8_t ha_locked)
 			qlt_send_term_exchange(ha->base_qpair, NULL, pkt,
 			    ha_locked, 0);
 		} else {
+			/*
+			 * If we get correct ATIO, then HBA had enough memory
+			 * to proceed without reset.
+			 */
+			if (!ha_locked)
+				spin_lock_irqsave(&ha->hardware_lock, flags);
+			vha->hw->exch_starvation = 0;
+			if (!ha_locked)
+				spin_unlock_irqrestore(&ha->hardware_lock, flags);
+
 			qlt_24xx_atio_pkt_all_vps(vha,
 			    (struct atio_from_isp *)pkt, ha_locked);
 		}
-- 
2.40.1

On 10/8/24 15:24, Anastasia Kovaleva wrote:
> Long-lived sessions under high load can accumulate a starvation counter,
> and the current implementation does not allow this counter to be reset
> during an active session.
> 
> If HBA sends correct ATIO IOCB, then it has enough resources to process
> commands and we should not call ISP recovery.
> 
> Cc: stable@vger.kernel.org
> Fixes: ead038556f64 ("qla2xxx: Add Dual mode support in the driver")
> Signed-off-by: Anastasia Kovaleva <a.kovaleva@yadro.com>
> Reviewed-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
> ---
>   drivers/scsi/qla2xxx/qla_isr.c    |  4 ++++
>   drivers/scsi/qla2xxx/qla_target.c | 11 +++++++++++
>   2 files changed, 15 insertions(+)
> 
> diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
> index fe98c76e9be3..5234ce0985e0 100644
> --- a/drivers/scsi/qla2xxx/qla_isr.c
> +++ b/drivers/scsi/qla2xxx/qla_isr.c
> @@ -1959,6 +1959,10 @@ qla2x00_async_event(scsi_qla_host_t *vha, struct rsp_que *rsp, uint16_t *mb)
>   		ql_dbg(ql_dbg_async, vha, 0x5091, "Transceiver Removal\n");
>   		break;
>   
> +	case MBA_REJECTED_FCP_CMD:
> +		ql_dbg(ql_dbg_async, vha, 0x5092, "LS_RJT was sent. No resources to process the ELS request.\n");
> +		break;
> +
>   	default:
>   		ql_dbg(ql_dbg_async, vha, 0x5057,
>   		    "Unknown AEN:%04x %04x %04x %04x\n",
> diff --git a/drivers/scsi/qla2xxx/qla_target.c b/drivers/scsi/qla2xxx/qla_target.c
> index d7551b1443e4..bc7feef6ee79 100644
> --- a/drivers/scsi/qla2xxx/qla_target.c
> +++ b/drivers/scsi/qla2xxx/qla_target.c
> @@ -6801,6 +6801,7 @@ qlt_24xx_process_atio_queue(struct scsi_qla_host *vha, uint8_t ha_locked)
>   	struct qla_hw_data *ha = vha->hw;
>   	struct atio_from_isp *pkt;
>   	int cnt, i;
> +	unsigned long flags = 0;
>   
>   	if (!ha->flags.fw_started)
>   		return;
> @@ -6826,6 +6827,16 @@ qlt_24xx_process_atio_queue(struct scsi_qla_host *vha, uint8_t ha_locked)
>   			qlt_send_term_exchange(ha->base_qpair, NULL, pkt,
>   			    ha_locked, 0);
>   		} else {
> +			/*
> +			 * If we get correct ATIO, then HBA had enough memory
> +			 * to proceed without reset.
> +			 */
> +			if (!ha_locked)
> +				spin_lock_irqsave(&ha->hardware_lock, flags);
> +			vha->hw->exch_starvation = 0;
> +			if (!ha_locked)
> +				spin_unlock_irqrestore(&ha->hardware_lock, flags);
> +
>   			qlt_24xx_atio_pkt_all_vps(vha,
>   			    (struct atio_from_isp *)pkt, ha_locked);
>   		}

Why not just 'WRITE_ONCE()' and drop the spinlock?

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                  Kernel Storage Architect
hare@suse.com                               +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich