Re: [PATCH] scsi: qedf: Limit number of CQs

[Chad Dupuis]

On Thu, 13 Jul 2017, 12:19pm, Thomas Bogendoerfer wrote:

> From: Thomas Bogendoerfer <tbogendoerfer@xxxxxxx>
> 
> FCOE offloading on qedf devices fails with:
> 
> [qed_sp_fcoe_func_start:150(sp-0-3b:00.02)]Cannot satisfy CQ amount. CQs
> 		 requested 8, CQs available 6. Aborting function start
> [qed_fcoe_start:821()]Failed to start fcoe
> [__qedf_probe:3041]:6: Cannot start FCoE function.
> 
> The reason is a newly introduced check in the qed main part. This change
> also provides the information about how many CQs are available, so we
> simply limit the number of requested CQs.
> 
> Fixes: 3c5da9427802 qed: Share additional information with qedf
> Signed-off-by: Thomas Bogendoerfer <tbogendoerfer@xxxxxxx>
> ---
>  drivers/scsi/qedf/qedf_main.c | 16 +++++++++-------
>  1 file changed, 9 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/scsi/qedf/qedf_main.c b/drivers/scsi/qedf/qedf_main.c
> index b58bba4604e8..5778218734fa 100644
> --- a/drivers/scsi/qedf/qedf_main.c
> +++ b/drivers/scsi/qedf/qedf_main.c
> @@ -2765,6 +2765,8 @@ static int qedf_set_fcoe_pf_param(struct qedf_ctx *qedf)
>  	 */
>  	qedf->num_queues = min((unsigned int)QEDF_MAX_NUM_CQS,
>  	    num_online_cpus());
> +	/* limit to acutal available CQs */
> +	qedf->num_queues = min(qedf->num_queues, qedf->dev_info.num_cqs);

We do need to account for the case where there are less CPUs than CQs.  
Does this look reasonable:

diff --git a/drivers/scsi/qedf/qedf.h b/drivers/scsi/qedf/qedf.h
index 4d03892..3c1d6fd 100644
--- a/drivers/scsi/qedf/qedf.h
+++ b/drivers/scsi/qedf/qedf.h
@@ -528,7 +528,7 @@ struct fip_vlan {
 #define QEDF_WRITE                    (1 << 0)
 #define MAX_FIBRE_LUNS			0xffffffff
 
-#define QEDF_MAX_NUM_CQS		8
+#define MIN_NUM_CPUS_MSIX(x)	min((u32)x->dev_info.num_cqs, 
num_online_cpus())
 
 /*
  * PCI function probe defines
diff --git a/drivers/scsi/qedf/qedf_main.c b/drivers/scsi/qedf/qedf_main.c
index 937df5b..f4facd3 100644
--- a/drivers/scsi/qedf/qedf_main.c
+++ b/drivers/scsi/qedf/qedf_main.c
@@ -2760,11 +2760,9 @@ static int qedf_set_fcoe_pf_param(struct qedf_ctx 
*qedf)
 	 * we allocation is the minimum off:
 	 *
 	 * Number of CPUs
-	 * Number of MSI-X vectors
-	 * Max number allocated in hardware (QEDF_MAX_NUM_CQS)
+	 * Number allocated by qed for our PCI function
 	 */
-	qedf->num_queues = min((unsigned int)QEDF_MAX_NUM_CQS,
-	    num_online_cpus());
+	qedf->num_queues = MIN_NUM_CPUS_MSIX(qedf);

Otherwise this patch looks good.

>  
>  	QEDF_INFO(&(qedf->dbg_ctx), QEDF_LOG_DISC, "Number of CQs is %d.\n",
>  		   qedf->num_queues);
> @@ -2962,6 +2964,13 @@ static int __qedf_probe(struct pci_dev *pdev, int mode)
>  		goto err1;
>  	}
>  
> +	/* Learn information crucial for qedf to progress */
> +	rc = qed_ops->fill_dev_info(qedf->cdev, &qedf->dev_info);
> +	if (rc) {
> +		QEDF_ERR(&(qedf->dbg_ctx), "Failed to dev info.\n");
> +		goto err1;
> +	}
> +
>  	/* queue allocation code should come here
>  	 * order should be
>  	 * 	slowpath_start
> @@ -2977,13 +2986,6 @@ static int __qedf_probe(struct pci_dev *pdev, int mode)
>  	}
>  	qed_ops->common->update_pf_params(qedf->cdev, &qedf->pf_params);
>  
> -	/* Learn information crucial for qedf to progress */
> -	rc = qed_ops->fill_dev_info(qedf->cdev, &qedf->dev_info);
> -	if (rc) {
> -		QEDF_ERR(&(qedf->dbg_ctx), "Failed to dev info.\n");
> -		goto err1;
> -	}
> -
>  	/* Record BDQ producer doorbell addresses */
>  	qedf->bdq_primary_prod = qedf->dev_info.primary_dbq_rq_addr;
>  	qedf->bdq_secondary_prod = qedf->dev_info.secondary_bdq_rq_addr;
>