Re: [PATCH libata-dev#upstream-fixes] pata_amd: fix an obvious bug in cable detection

[Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>]

On Sunday 25 February 2007, Tejun Heo wrote:
> [cc'ing Mark, Bart and Sergei. Hi]
> 
> Alan wrote:
> >>> This makes unreliable cable detection even more unreliable.  Please
> >>> consider for -stable.  Thanks.
> > 
> > At minimum you also need to stop doing drive side detect for PATA_CBL_80
> > as many controllers don't do both so if you've got host side detect that
> > is what you must trust.
> > 
> > Fixed that in my internal tree and it seems happier
> 
> It seems libata is wrong about device side cable detection.  Device side
> cable detection tries to detect host side capacitor connected to PDIAG-,
> depends on the slave device releasing PDIAG- signal.
> 
> According to Annex B of ATA/ATAPI-5, IDENTIFY should be issued to the
> slave device first to ensure that it releases PDIAG- and then use the
> cable detection result from the master device.  As we IDENTIFY master
> first right after reset, slave if present is driving PDIAG-, so the
> master on IDENTIFY always thinks the capacitor is present and 40c limit
> is always applied.
> 
> IDE is better off because it doesn't reset before IDENTIFYing and it's
> likely that BIOS has issued some commands to the slave device prior to
> passing control to OS thus making it release PDIAG-.
> 
> 1. Device side cable detection can only be used to limit speed because
> if motherboard doesn't have capacitor attached to PDIAG-, the test
> result always indicates 80c.
> 
> 2. Host side detection involves issuing commands to attached devices
> because it depends on the slave device releasing PDIAG-, so most
> controllers do cable detection in BIOS and just put the result in PCI
> config region or wherever.  We dunno how BIOS does it exactly but many
> of them probably consider device side detection result as well.
> 
> So, considering #1 and #2, it might be best to just believe what the
> controller (BIOS) says and not bother about device side detection.  In
> fact, we've been effectively ignoring device side detection result
> before my "fix" for device side cable detection.
> 
> What do you think?

Why can't libata do the right thing and just send IDENTIFY command
to the slave device first?

Bart