Re: More FC Transport Issues

[Michael Reed <mdr@sgi.com>]

Andrew Vasquez wrote:
> On Thu, 22 Dec 2005, Michael Reed wrote:
> 
>>It's possible the OOPS is related to the qla2xxx serialization issue I reported
>>earlier.
>>
>>http://marc.theaimsgroup.com/?l=linux-scsi&m=113380446825841&w=2
>>http://marc.theaimsgroup.com/?l=linux-scsi&m=113458339109319&w=2
> 
> As far as the synchronization issues, could you try out the following.

As for initial testing w/git3, it's hanging during boot.
I'll try and compile KDB into the kernel to get more info.
Perhaps I can do some more work on this over the weekend.

It would seem as though the driver is thrashing the add/delete
mechanism.  With lots of host ports and lots of targets, every SCN that
comes in when another host port joins the fabric causes the others
to delete and re-add their targets.

Picture 64 host ports and 128 targets on each.  (I only wish this were
my test config.  Maybe Santa will bring me some more fabric!)

Perhaps check to see if target is missing after SCN and just
delete it?

As a sanity check, I verified that the system boots properly
with 2.6.14.3.


Mike



> 
> ---
> 
> diff --git a/drivers/scsi/qla2xxx/qla_def.h b/drivers/scsi/qla2xxx/qla_def.h
> index 7b3efd5..dba240b 100644
> --- a/drivers/scsi/qla2xxx/qla_def.h
> +++ b/drivers/scsi/qla2xxx/qla_def.h
> @@ -1664,7 +1664,7 @@ typedef struct fc_port {
>  	uint8_t mp_byte;		/* multi-path byte (not used) */
>      	uint8_t cur_path;		/* current path id */
>  
> -	struct fc_rport *rport;
> +	struct fc_rport *rport, *drport;
>  	u32 supported_classes;
>  	struct work_struct rport_add_work;
>  	struct work_struct rport_del_work;
> diff --git a/drivers/scsi/qla2xxx/qla_init.c b/drivers/scsi/qla2xxx/qla_init.c
> index c46d246..b99f547 100644
> --- a/drivers/scsi/qla2xxx/qla_init.c
> +++ b/drivers/scsi/qla2xxx/qla_init.c
> @@ -1687,10 +1687,16 @@ static void
>  qla2x00_rport_del(void *data)
>  {
>  	fc_port_t *fcport = data;
> +	struct fc_rport *rport;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&fcport->ha->hardware_lock, flags);
> +	rport = fcport->drport;
> +	fcport->drport = NULL;
> +	spin_unlock_irqrestore(&fcport->ha->hardware_lock, flags);
> +	if (rport)
> +		fc_remote_port_delete(rport);
>  
> -	if (fcport->rport)
> -		fc_remote_port_delete(fcport->rport);
> -	fcport->rport = NULL;
>  }
>  
>  /**
> @@ -2081,10 +2087,10 @@ qla2x00_reg_remote_port(scsi_qla_host_t 
>  	struct fc_rport_identifiers rport_ids;
>  	struct fc_rport *rport;
>  
> -	if (fcport->rport) {
> -		fc_remote_port_delete(fcport->rport);
> -		fcport->rport = NULL;
> -	}
> +	if (fcport->drport)
> +		qla2x00_rport_del(fcport);
> +	if (fcport->rport)
> +		return;
>  
>  	rport_ids.node_name = wwn_to_u64(fcport->node_name);
>  	rport_ids.port_name = wwn_to_u64(fcport->port_name);
> diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
> index c58c9d9..9842fb7 100644
> --- a/drivers/scsi/qla2xxx/qla_os.c
> +++ b/drivers/scsi/qla2xxx/qla_os.c
> @@ -1627,6 +1627,20 @@ qla2x00_free_device(scsi_qla_host_t *ha)
>  	pci_disable_device(ha->pdev);
>  }
>  
> +static inline void
> +qla2x00_schedule_rport_del(struct scsi_qla_host *ha, fc_port_t *fcport)
> +{
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ha->hardware_lock, flags);
> +	if (fcport->rport) {
> +		fcport->drport = fcport->rport;
> +		fcport->rport = NULL;
> +	}
> +	spin_unlock_irqrestore(&ha->hardware_lock, flags);
> +	schedule_work(&fcport->rport_del_work);
> +}
> +
>  /*
>   * qla2x00_mark_device_lost Updates fcport state when device goes offline.
>   *
> @@ -1639,8 +1653,8 @@ qla2x00_free_device(scsi_qla_host_t *ha)
>  void qla2x00_mark_device_lost(scsi_qla_host_t *ha, fc_port_t *fcport,
>      int do_login)
>  {
> -	if (atomic_read(&fcport->state) == FCS_ONLINE && fcport->rport)
> -		schedule_work(&fcport->rport_del_work);
> +	if (atomic_read(&fcport->state) == FCS_ONLINE)
> +		qla2x00_schedule_rport_del(ha, fcport);
>  
>  	/*
>  	 * We may need to retry the login, so don't change the state of the
> @@ -1701,8 +1715,8 @@ qla2x00_mark_all_devices_lost(scsi_qla_h
>  		 */
>  		if (atomic_read(&fcport->state) == FCS_DEVICE_DEAD)
>  			continue;
> -		if (atomic_read(&fcport->state) == FCS_ONLINE && fcport->rport)
> -			schedule_work(&fcport->rport_del_work);
> +		if (atomic_read(&fcport->state) == FCS_ONLINE)
> +			qla2x00_schedule_rport_del(ha, fcport);
>  		atomic_set(&fcport->state, FCS_DEVICE_LOST);
>  	}
>  }
>