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

[Jeff Garzik <jgarzik@pobox.com>]

On 11/16/2009 01:14 AM, Tejun Heo wrote:
> Hello,
>
> Arjan van de Ven wrote:
>> sigh. so I moved all the generic logic generic, and left the ahci
>> specific code specific to ahci. I put the logic there where it was
>> easy to implement, and there where the other link power management
>> controls are (in sysfs). If that's not good enough, I'm out of my skills
>> in the libata world to be honest, and would like to ask you to implement
>> that instead. let me know what sysfs looks like and I'll adjust
>> powertop to it....
>
> The reason why we have sysfs attributes which should have been at link
> layer at host was that it originally was for ahci alpm which is host
> specific feature which got extended to something somewhat generic.
> Now another pm feature which should belong to link is added to host
> following the precedence.
>
> Then again, it's also true that nobody really cares about ATA PM
> features enough during past couple of years so I really don't want to
> prevent the feature you're trying to add.  It would be best if there's
> someone who would pick it up and implement proper infrastructure but
> well that doesn't seem to be happening anytime soon.
>
> So, I don't know.  That's the concern I have but I don't want to nack
> your change either.  One thing is at least make those functions take
> ata_link isntead of ata_port as there's nothing port specific about
> those.  Jeff, what do you think?

Well,

- these are link-level features
- libata lacks a link-level sysfs API
- we need a link-level sysfs API (ata_transport, anyone?)

The ugly alternative has always been to hack in something at the host level.

In term of internal data structures, the v3 patch putting the stats into 
struct ata_link is definitely the right thing to do.  I would also put 
the accounting_enabled variable in there, as non-AHCI implementations, 
polling or not, seem highly likely to need that.

	Jeff