[Server-devel] Suggested SATA change by Marvell

[David Woodhouse]

I've added Jeff Garzik, the Linux SATA maintainer, to Cc. He's best
positioned to opine on the software front.

Jeff, in case it wasn't obvious, this is about the OLPC school server.

On Fri, 2007-06-29 at 14:26 -0400, John Watlington wrote:
> A
>     Marvell has been pushing hard for a change in the SATA subsystem.
> 
> Currently, this is planned to be:
> 
> A.)  A SATA interface on the Discovery 6 (MV-64660), and
> a SATA interface on an 88SE6111, connected to the Discovery 6 through  
> PCI-E.
> 
> The proposal from Marvell is to move to either:
> 
> B.)  A Port Multiplier (88SM4140) connected to the SATA interface on  
> the Discovery 6, which supports up to four SATA ports,  or
> 
> C.) Two SATA interfaces on an 88SE6121, connected to the Discovery 6  
> through PCI-E
> 
> The main reason seems to be to move us to a single driver.
> Pricing on either of these solutions would be higher (but not in our  
> case: Marvell will
> sell us any of these three solution for the same price.)
> Power will be roughly equivalent, as the PCI-E PHY will always be  
> turned on in solutions
> A and C, and B requires two SATA hops.
> 
> So it comes down to software/driver costs.
> 
> Work is being done by Marvell to ensure 88SE61xx family support in  
> libata.   The Port Multiplier isn't well supported by libata at this time.
> 
> The SATA interface on the Discovery 6 is very similar to the Hercules
> SATA chipset, currently supported by the open source mrvlsat driver
> (?) and is also being included in libata (but not as fast as the
> 88SE61xx support).
> 
> An internal driver for the Discovery 6 SATA, including PM support,  
> has been shipping for over three years.   We will have that code
> dropped off this afternoon.
> 
> Comments ?  Suggestions ?
> wad

In general I would suggest that you ignore any code which comes from
Marvell. In my experience it'll all need to be rewritten from scratch
anyway. We want to use something which is already properly supported,
unless we're prepared to do the gruntwork of writing a decent driver.

-- 
dwmw2