Re: libata SATA support for Sil 3726 port multiplier

[Tejun Heo <htejun@gmail.com>]

Hello, Fredrik.

Fredrik Sjoholm wrote:
> 
> Hi,
> 
> The 3726 multiplier is in fact built into an external box:
> 
>     http://www.datoptic.com/cgi-bin/web.cgi?product=Sbox-P&detail=yes
> 
> I connected the box with the 3726 multiplier to a 3132 based pci-x card 
> via an eSATA cable.
> 
> There is one remaining quirk which i didn't notice before emailing you 
> the mini-fix: When booting cold, the spin-up of each drive happens 
> sequentially and the 5 second retry isn't enough for each drive to get 
> ready. This wouldn't have been a problem except for the fact that even 
> the latest version (7.3.13) of the Sil3132 BIOS doesn't bother detecting 
> and spinning-up anything except the 1st drive on the multiplier. On 
> subsequent bootups, with drives already spinning, it work right away.
> 
> I guess it would be nice to spin-up all drives in rapid succession, and 
> then retry them all at once every few seconds, but all i did was to 
> increase the timeout to 15 seconds which works too.

Yes, we have yet to implement staggered spinup support with someway to 
control the number of concurrently spun up devices.  It's on my agenda. 
  SATA staggered spinup is enabled by loading certain pin in the power 
connector during power-up.  So, it is the enclosure which determines 
whether staggered spin-up is enabled or disabled.  Your enclosure may 
have a dip switch or something to control that.

> here are the dmesg outputs:
> 
> 1. warm start, everything ok:
> 
>     http://www.sjoholm.com/sata/dmesg.warmspin-ok
> 
> 2. cold start, 5 second timeouts. only 1st drive usable:
> 
>     http://www.sjoholm.com/sata/dmesg.spinup5-fail
> 
> 3. cold start, ssleep(15)'s in libata-eh.c, all drives ok:
> 
>     http://www.sjoholm.com/sata/dmesg.spinup15-ok

Can you try new version and see how it acts?

http://home-tj.org/files/libata-tj-stable/libata-tj-2.6.17.4-20060710.tar.bz2

It probably won't fix the problem but hardreset implementation has 
changed and it would help me understanding how spinning up device acts. 
  Also, turning on CONFIG_PRINTK_TIME (Kernel Hacking -> Show timing 
information on printks) will be nice.

Thanks.

-- 
tejun