Re: [PATCH] libata: Add ALPM power state accounting to the AHCI driver

[Tejun Heo <tj@kernel.org>]

Alan Cox wrote:
>> This is true for all link functions, so I'm suggesting the stats
>> helpers to follow the same convention.  This will also make slave link
>> configuration work properly (controllers which present two SATA links
>> as master/slave of the same port but still provide access to separate
>> SCR registers).  You know, it's a link function, make it take a link
>> as all other stuff is designed that way.
> 
> I don't see how you can do per link reporting on AHCI.
> 
> Sure maybe some devices should expose it per link, but it's so expensive
> and it'll get really ugly fast.

Not for ahci but ata_piix, sis and via already do it.

> I'd say it has to go in at host level for now. If some super whizzo
> future silicon exposes it per link without being beaten around the head
> by the software stack then it's easy enough to add per link data and a
> generic helper which provides a summary set of numbers at the host level
> for hardware that can report link data.
> 
> That gives us a useful API now and the ability to fix it in the future,
> not that I anticipate anyone ever needing to. AHCI (or AHCI-ish) seems to
> be what everyone is working to nowdays with little obvious pressure to
> move anywhere else.

Yeap, agreed.  It just is so ugly (not Arjan's fault).

Thanks.

-- 
tejun