[PATCH]: pcmcia - spot slave decode flaws (for testing)

[PATCH]: pcmcia - spot slave decode flaws (for testing)

[Alan]

If you've got a CF adapter or PCMCIA disc which shows up twice in libata
pata_pcmcia can you try this patch on top of the updates posted. It tries
to spot when the slave is a mirror of the master and to fix up problems
that causes.

Signed-off-by: Alan Cox <alan@redhat.com>

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.20-mm2/drivers/ata/pata_pcmcia.c linux-2.6.20-mm2/drivers/ata/pata_pcmcia.c
--- linux.vanilla-2.6.20-mm2/drivers/ata/pata_pcmcia.c	2007-02-20 13:37:58.000000000 +0000
+++ linux-2.6.20-mm2/drivers/ata/pata_pcmcia.c	2007-02-20 14:28:13.000000000 +0000
@@ -54,6 +54,39 @@
 	dev_node_t	node;
 };
 
+/**
+ *	pcmcia_set_mode	-	PCMCIA specific mode setup
+ *	@ap: Port
+ *	@r_failed_dev: Return pointer for failed device
+ *
+ *	Perform the tuning and setup of the devices and timings, which
+ *	for PCMCIA is the same as any other controller. We wrap it however
+ *	as we need to spot hardware with incorrect or missing master/slave
+ *	decode, which alas is embarrassingly common in the PC world
+ */
+ 
+static int pcmcia_set_mode(struct ata_port *ap, struct ata_device **r_failed_dev)
+{
+	struct ata_device *master = &ap->device[0];
+	struct ata_device *slave = &ap->device[1];
+	
+	if (!ata_dev_enabled(master) || !ata_dev_enabled(slave))
+		return ata_do_set_mode(ap, r_failed_dev);
+		
+	if (memcmp(master->id + ATA_ID_FW_REV,  slave->id + ATA_ID_FW_REV,
+			   ATA_ID_FW_REV_LEN + ATA_ID_PROD_LEN) == 0)
+	{
+		/* Suspicious match, but could be two cards from
+		   the same vendor - check serial */
+		if (memcmp(master->id + ATA_ID_SERNO, slave->id + ATA_ID_SERNO,
+			   ATA_ID_SERNO_LEN) == 0 && master->id[ATA_ID_SERNO] >> 8) {
+			ata_dev_printk(slave, KERN_WARNING, "is a ghost device, ignoring.\n");
+			ata_dev_disable(slave);
+		}	 
+	}
+	return ata_do_set_mode(ap, r_failed_dev);
+}
+
 static struct scsi_host_template pcmcia_sht = {
 	.module			= THIS_MODULE,
 	.name			= DRV_NAME,
@@ -73,6 +106,7 @@
 };
 
 static struct ata_port_operations pcmcia_port_ops = {
+	.set_mode	= pcmcia_set_mode,
 	.port_disable	= ata_port_disable,
 	.tf_load	= ata_tf_load,
 	.tf_read	= ata_tf_read,

Re: [PATCH]: pcmcia - spot slave decode flaws (for testing)

[Jeff Garzik]

Alan wrote:
> If you've got a CF adapter or PCMCIA disc which shows up twice in libata
> pata_pcmcia can you try this patch on top of the updates posted. It tries
> to spot when the slave is a mirror of the master and to fix up problems
> that causes.
> 
> Signed-off-by: Alan Cox <alan@redhat.com>
> 
> diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.20-mm2/drivers/ata/pata_pcmcia.c linux-2.6.20-mm2/drivers/ata/pata_pcmcia.c
> --- linux.vanilla-2.6.20-mm2/drivers/ata/pata_pcmcia.c	2007-02-20 13:37:58.000000000 +0000
> +++ linux-2.6.20-mm2/drivers/ata/pata_pcmcia.c	2007-02-20 14:28:13.000000000 +0000
> @@ -54,6 +54,39 @@
>  	dev_node_t	node;
>  };
>  
> +/**
> + *	pcmcia_set_mode	-	PCMCIA specific mode setup
> + *	@ap: Port
> + *	@r_failed_dev: Return pointer for failed device
> + *
> + *	Perform the tuning and setup of the devices and timings, which
> + *	for PCMCIA is the same as any other controller. We wrap it however
> + *	as we need to spot hardware with incorrect or missing master/slave
> + *	decode, which alas is embarrassingly common in the PC world
> + */
> + 
> +static int pcmcia_set_mode(struct ata_port *ap, struct ata_device **r_failed_dev)
> +{
> +	struct ata_device *master = &ap->device[0];
> +	struct ata_device *slave = &ap->device[1];
> +	
> +	if (!ata_dev_enabled(master) || !ata_dev_enabled(slave))
> +		return ata_do_set_mode(ap, r_failed_dev);
> +		
> +	if (memcmp(master->id + ATA_ID_FW_REV,  slave->id + ATA_ID_FW_REV,
> +			   ATA_ID_FW_REV_LEN + ATA_ID_PROD_LEN) == 0)
> +	{
> +		/* Suspicious match, but could be two cards from
> +		   the same vendor - check serial */
> +		if (memcmp(master->id + ATA_ID_SERNO, slave->id + ATA_ID_SERNO,
> +			   ATA_ID_SERNO_LEN) == 0 && master->id[ATA_ID_SERNO] >> 8) {
> +			ata_dev_printk(slave, KERN_WARNING, "is a ghost device, ignoring.\n");
> +			ata_dev_disable(slave);
> +		}	 
> +	}
> +	return ata_do_set_mode(ap, r_failed_dev);

Code looks OK. Not applied due to "for testing" note.

General comment:  it might be nice to do this in the core, just as a 
sanity check for a variety of problems, past, present and future.

Re: [PATCH]: pcmcia - spot slave decode flaws (for testing)

[Alan]

> Code looks OK. Not applied due to "for testing" note.
> 
> General comment:  it might be nice to do this in the core, just as a 
> sanity check for a variety of problems, past, present and future.

We tried that with old IDE and all hell broke loose. Lots of virtual disk
stuff and raid volumes have non-unique serial numbers. We even found a
case of identically serial numbered Maxtor drives.

It needs to stay in pcmcia, which is the only place we've seen the
duplication.



Alan

Re: [PATCH]: pcmcia - spot slave decode flaws (for testing)

[Eric D. Mudama]

On 2/23/07, Alan <alan@lxorguk.ukuu.org.uk> wrote:
> > Code looks OK. Not applied due to "for testing" note.
> >
> > General comment:  it might be nice to do this in the core, just as a
> > sanity check for a variety of problems, past, present and future.
>
> We tried that with old IDE and all hell broke loose. Lots of virtual disk
> stuff and raid volumes have non-unique serial numbers. We even found a
> case of identically serial numbered Maxtor drives.
>
> It needs to stay in pcmcia, which is the only place we've seen the
> duplication.

I don't think the maxtor drives actually had duplicate serial numbers.
 They were coming back as "M0000000000000000000" or something like
that to the limit of the strlen.  It looked more like buffer
corruption or something, and only happened with one or two people...

--eric