pata_isapnp: Don't use invalid I/O ports

On Friday 13 September 2013 20:54:37 Ondrej Zary wrote:
> On Thursday 12 September 2013 23:01:29 Ondrej Zary wrote:
> > The test for 2nd I/O port validity is broken (reversed):
> > On devices with no control port, the driver attempts to use invalid port
> > 0, resulting in logs full of bad_io_access errors.
> > On devices with control port, the driver does not use it.
> >
> > Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
> > ---
> >  drivers/ata/pata_isapnp.c |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> >
> > diff --git a/drivers/ata/pata_isapnp.c b/drivers/ata/pata_isapnp.c
> > index 4bceb88..b33d1f9 100644
> > --- a/drivers/ata/pata_isapnp.c
> > +++ b/drivers/ata/pata_isapnp.c
> > @@ -78,7 +78,7 @@ static int isapnp_init_one(struct pnp_dev *idev, const
> > struct pnp_device_id *dev
> >
> >  	ap->ioaddr.cmd_addr = cmd_addr;
> >
> > -	if (pnp_port_valid(idev, 1) == 0) {
> > +	if (pnp_port_valid(idev, 1)) {
> >  		ctl_addr = devm_ioport_map(&idev->dev,
> >  					   pnp_port_start(idev, 1), 1);
> >  		ap->ioaddr.altstatus_addr = ctl_addr;
>
> With this patch, the ATA port works fine if there is a device connected to
> it. However, it takes ages to boot if there are no devices connected. Looks
> like the driver retries IDENTIFY command for both slave and master drives
> in a hope that the devices are there.
>
> log:
>
> [    7.692344] pata_isapnp 01:01.02: [io  0x01e8-0x01ef]
> [    7.692474] pata_isapnp 01:01.02: [io  0x0168-0x016f]
> [    7.692644] pata_isapnp 01:01.02: [irq 10]
> [    7.695012] pata_isapnp 01:01.02: activated
> [    7.751153] scsi2 : pata_isapnp
> [    7.751781] ata3: PATA max PIO0 cmd 0x168 ctl 0x0 irq 10
> [   12.751446] ata3.01: qc timeout (cmd 0xec)
> [   12.751571] ata3.01: failed to IDENTIFY (I/O error, err_mask=0x5)
> [   22.752270] ata3.01: qc timeout (cmd 0xec)
> [   22.752396] ata3.01: failed to IDENTIFY (I/O error, err_mask=0x5)
> [   52.754737] ata3.01: qc timeout (cmd 0xec)
> [   52.754861] ata3.01: failed to IDENTIFY (I/O error, err_mask=0x5)
> [   57.755149] ata3.00: qc timeout (cmd 0xec)
> [   57.755275] ata3.00: failed to IDENTIFY (I/O error, err_mask=0x5)
> [   67.755974] ata3.00: qc timeout (cmd 0xec)
> [   67.756099] ata3.00: failed to IDENTIFY (I/O error, err_mask=0x5)

This is caused by skipping ata_sff_softreset() when ap->ioaddr.ctl_addr is
unset so ata_devchk() is never called. This patch seems to fix the problem.
Is it OK or can it cause more problems?

Hello, Ondrej.

On Fri, Sep 13, 2013 at 11:44:27PM +0200, Ondrej Zary wrote:
> This is caused by skipping ata_sff_softreset() when ap->ioaddr.ctl_addr is
> unset so ata_devchk() is never called. This patch seems to fix the problem.
> Is it OK or can it cause more problems?
> 
> --- a/drivers/ata/libata-sff.c
> +++ b/drivers/ata/libata-sff.c
> @@ -2008,13 +2008,15 @@ static int ata_bus_softreset(struct ata_port *ap, unsigned int devmask,
>  
>  	DPRINTK("ata%u: bus reset via SRST\n", ap->print_id);
>  
> -	/* software reset.  causes dev0 to be selected */
> -	iowrite8(ap->ctl, ioaddr->ctl_addr);
> -	udelay(20);	/* FIXME: flush */
> -	iowrite8(ap->ctl | ATA_SRST, ioaddr->ctl_addr);
> -	udelay(20);	/* FIXME: flush */
> -	iowrite8(ap->ctl, ioaddr->ctl_addr);
> -	ap->last_ctl = ap->ctl;
> +	if (ap->ioaddr.ctl_addr) {
> +		/* software reset.  causes dev0 to be selected */
> +		iowrite8(ap->ctl, ioaddr->ctl_addr);
> +		udelay(20);	/* FIXME: flush */
> +		iowrite8(ap->ctl | ATA_SRST, ioaddr->ctl_addr);
> +		udelay(20);	/* FIXME: flush */
> +		iowrite8(ap->ctl, ioaddr->ctl_addr);
> +		ap->last_ctl = ap->ctl;
> +	}
>  
>  	/* wait the port to become ready */
>  	return ata_sff_wait_after_reset(&ap->link, devmask, deadline);
> @@ -2215,10 +2217,6 @@ void ata_sff_error_handler(struct ata_port *ap)
>  
>  	spin_unlock_irqrestore(ap->lock, flags);
>  
> -	/* ignore ata_sff_softreset if ctl isn't accessible */
> -	if (softreset == ata_sff_softreset && !ap->ioaddr.ctl_addr)
> -		softreset = NULL;
> -
>  	/* ignore built-in hardresets if SCR access is not available */
>  	if ((hardreset == sata_std_hardreset ||
>  	     hardreset == sata_sff_hardreset) && !sata_scr_valid(&ap->link))

Hmm... I'm unsure what to do with these patches.  It's a mostly dead
driver, so I don't wanna mess with it too much unless it's an obvious
fix.  What hardware are you playing with?

Thanks.

On Sunday 22 September 2013 18:49:42 Tejun Heo wrote:
> Hello, Ondrej.
>
> On Fri, Sep 13, 2013 at 11:44:27PM +0200, Ondrej Zary wrote:
> > This is caused by skipping ata_sff_softreset() when ap->ioaddr.ctl_addr
> > is unset so ata_devchk() is never called. This patch seems to fix the
> > problem. Is it OK or can it cause more problems?
> >
> > --- a/drivers/ata/libata-sff.c
> > +++ b/drivers/ata/libata-sff.c
> > @@ -2008,13 +2008,15 @@ static int ata_bus_softreset(struct ata_port *ap,
> > unsigned int devmask,
> >
> >  	DPRINTK("ata%u: bus reset via SRST\n", ap->print_id);
> >
> > -	/* software reset.  causes dev0 to be selected */
> > -	iowrite8(ap->ctl, ioaddr->ctl_addr);
> > -	udelay(20);	/* FIXME: flush */
> > -	iowrite8(ap->ctl | ATA_SRST, ioaddr->ctl_addr);
> > -	udelay(20);	/* FIXME: flush */
> > -	iowrite8(ap->ctl, ioaddr->ctl_addr);
> > -	ap->last_ctl = ap->ctl;
> > +	if (ap->ioaddr.ctl_addr) {
> > +		/* software reset.  causes dev0 to be selected */
> > +		iowrite8(ap->ctl, ioaddr->ctl_addr);
> > +		udelay(20);	/* FIXME: flush */
> > +		iowrite8(ap->ctl | ATA_SRST, ioaddr->ctl_addr);
> > +		udelay(20);	/* FIXME: flush */
> > +		iowrite8(ap->ctl, ioaddr->ctl_addr);
> > +		ap->last_ctl = ap->ctl;
> > +	}
> >
> >  	/* wait the port to become ready */
> >  	return ata_sff_wait_after_reset(&ap->link, devmask, deadline);
> > @@ -2215,10 +2217,6 @@ void ata_sff_error_handler(struct ata_port *ap)
> >
> >  	spin_unlock_irqrestore(ap->lock, flags);
> >
> > -	/* ignore ata_sff_softreset if ctl isn't accessible */
> > -	if (softreset == ata_sff_softreset && !ap->ioaddr.ctl_addr)
> > -		softreset = NULL;
> > -
> >  	/* ignore built-in hardresets if SCR access is not available */
> >  	if ((hardreset == sata_std_hardreset ||
> >  	     hardreset == sata_sff_hardreset) && !sata_scr_valid(&ap->link))
>
> Hmm... I'm unsure what to do with these patches.  It's a mostly dead
> driver, so I don't wanna mess with it too much unless it's an obvious
> fix.  What hardware are you playing with?

IMHO, the first patch (invalid I/O ports) is an obvious fix. That bug causes 
problems with ES968 sound cards.

The second patch is not-so-obvious. Only few ATA controllers don't have 
control register, but they're all broken now. I'm not sure if the patch is 
100% correct but it seems to work fine and it does not affect most 
controllers.