[PATCH] Optionally disable QAS in MPT Fusion spi driver

[PATCH] Optionally disable QAS in MPT Fusion spi driver

[Andrew Patterson]

Certain drives report that they support QAS, but it some cases can have
trouble with it causing bus resets.  This patch optionally disables QAS
for the MPT Fusion spi driver.  QAS is disabled using the mpt_qas module
parameter.

Andrew

Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>

-- 
Andrew Patterson
Hewlett-Packard

Index: scsi-misc-2.6/drivers/message/fusion/mptscsih.h
===================================================================
--- scsi-misc-2.6.orig/drivers/message/fusion/mptscsih.h	2006-06-02 12:29:31.000000000 -0600
+++ scsi-misc-2.6/drivers/message/fusion/mptscsih.h	2006-06-02 14:32:07.000000000 -0600
@@ -66,6 +66,7 @@
 #define MPTSCSIH_DOMAIN_VALIDATION      1
 #define MPTSCSIH_MAX_WIDTH              1
 #define MPTSCSIH_MIN_SYNC               0x08
+#define MPTSCSIH_QAS                    1
 #define MPTSCSIH_SAF_TE                 0
 #define MPTSCSIH_PT_CLEAR               0
 
Index: scsi-misc-2.6/drivers/message/fusion/mptspi.c
===================================================================
--- scsi-misc-2.6.orig/drivers/message/fusion/mptspi.c	2006-06-02 12:29:31.000000000 -0600
+++ scsi-misc-2.6/drivers/message/fusion/mptspi.c	2006-06-02 14:32:07.000000000 -0600
@@ -87,6 +87,10 @@ static int mpt_pq_filter = 0;
 module_param(mpt_pq_filter, int, 0);
 MODULE_PARM_DESC(mpt_pq_filter, " Enable peripheral qualifier filter: enable=1  (default=0)");
 
+static int mpt_qas = MPTSCSIH_QAS;
+module_param(mpt_qas, int, 1);
+MODULE_PARM_DESC(mpt_qas, " Quick Arbitration and Selection (QAS) enabled=1, disabled=0 (default=MPTSCSIH_QAS=1)");
+
 static void mptspi_write_offset(struct scsi_target *, int);
 static void mptspi_write_width(struct scsi_target *, int);
 static int mptspi_write_spi_device_pg1(struct scsi_target *,
@@ -607,7 +611,8 @@ static void mptspi_write_qas(struct scsi
 	VirtTarget *vtarget = starget->hostdata;
 	u32 nego;
 
-	if ((vtarget->negoFlags & MPT_TARGET_NO_NEGO_QAS) ||
+	if (!mpt_qas ||
+	    (vtarget->negoFlags & MPT_TARGET_NO_NEGO_QAS) ||
 	    hd->ioc->spi_data.noQas)
 		spi_qas(starget) = 0;
 	else
@@ -989,7 +994,7 @@ mptspi_probe(struct pci_dev *pdev, const
 		ioc->name,
 		mpt_saf_te,
 		mpt_pq_filter));
-	ioc->spi_data.noQas = 0;
+	ioc->spi_data.noQas = mpt_qas ? 0 : MPT_TARGET_NO_NEGO_QAS;
 
 	init_waitqueue_head(&hd->scandv_waitq);
 	hd->scandv_wait_done = 0;

RE: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[Moore, Eric]

On Friday, June 02, 2006 3:14 PM, Andrew Patterson wrote: 
> 
> Certain drives report that they support QAS, but it some 
> cases can have
> trouble with it causing bus resets.  This patch optionally 
> disables QAS
> for the MPT Fusion spi driver.  QAS is disabled using the 
> mpt_qas module
> parameter.
> 
> Andrew
> 
> Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
> 

ACK, James pls apply to your scsi-misc tree.   I validated
the patch here in the labs.

Sorry for the delay. I've been mostly out of the office the
past three weeks.

Eric Moore
LSI Logic

Re: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[Christoph Hellwig]

On Fri, Jun 02, 2006 at 03:14:25PM -0600, Andrew Patterson wrote:
> Certain drives report that they support QAS, but it some cases can have
> trouble with it causing bus resets.  This patch optionally disables QAS
> for the MPT Fusion spi driver.  QAS is disabled using the mpt_qas module
> parameter.

I'm not very happy about making this a module paramater in the LLDD as that
can't scale.  This kind of stuff needs to be done in the generic DV code.

RE: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[Moore, Eric]

On Thursday, June 15, 2006 9:12 AM, Christoph Hellwig wrote: 
> 
> On Fri, Jun 02, 2006 at 03:14:25PM -0600, Andrew Patterson wrote:
> > Certain drives report that they support QAS, but it some 
> cases can have
> > trouble with it causing bus resets.  This patch optionally 
> disables QAS
> > for the MPT Fusion spi driver.  QAS is disabled using the 
> mpt_qas module
> > parameter.
> 
> I'm not very happy about making this a module paramater in 
> the LLDD as that
> can't scale.  This kind of stuff needs to be done in the 
> generic DV code.
> 

We had discussed this with James Bottomley at Storage Summit,
Sorry I wanted to run this by you as well, but forgot. 
The QAS enable/disable feature is per device in spi transport
layer, not per host, which is what we need.

James said it was okay for having this done in lld, instead
of transport.  I'm sure both me and Andrew would be happy 
the same if the transport handled this.  
Is that possible for you to look into this, or James?

Eric Moore

RE: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[James Bottomley]

On Thu, 2006-06-15 at 09:31 -0600, Moore, Eric wrote:
> James said it was okay for having this done in lld, instead
> of transport.  I'm sure both me and Andrew would be happy 
> the same if the transport handled this.  
> Is that possible for you to look into this, or James?

Before we go the lld route, could someone just assure me this is a
fusion specific problem.  Andrew, could you attach the same drives to
another QAS LLD, like aic79xx and verify the problem doesn't recur?

If it's something drive specific, we probably need more of a blacklist
in the spi transport class.

James

RE: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[Andrew Patterson]

On Thu, 2006-06-15 at 09:21 -0700, James Bottomley wrote:
> On Thu, 2006-06-15 at 09:31 -0600, Moore, Eric wrote:
> > James said it was okay for having this done in lld, instead
> > of transport.  I'm sure both me and Andrew would be happy 
> > the same if the transport handled this.  
> > Is that possible for you to look into this, or James?
> 
> Before we go the lld route, could someone just assure me this is a
> fusion specific problem.  Andrew, could you attach the same drives to
> another QAS LLD, like aic79xx and verify the problem doesn't recur?

We can only reproduce this problem at a specific site which we no longer
have access to.  At that site, we had problems with both HP OEM'd Maxtor
and Fujitsu drives.  I also don't have any other u320 cards to try this
on other than Fusion.  Eric? Have you been able to reproduce this
problem on Linux at LSI?

> 
> If it's something drive specific, we probably need more of a blacklist
> in the spi transport class.

My guess is this is a drive/fusion interaction problem. 

Andrew

RE: [PATCH] Optionally disable QAS in MPT Fusion spi driver

[Moore, Eric]

On Thursday, June 15, 2006 11:28 AM, Andrew Patterson wrote:  

> 
> We can only reproduce this problem at a specific site which 
> we no longer
> have access to.  At that site, we had problems with both HP 
> OEM'd Maxtor
> and Fujitsu drives.  I also don't have any other u320 cards 
> to try this
> on other than Fusion.  Eric? Have you been able to reproduce this
> problem on Linux at LSI?
> 

No I haven't.