Per ATA host sysfs attributes?

Per ATA host sysfs attributes?

[Mark Lord]

Guys,

I have added IRQ coalescing to sata_mv, but need a way to activate/tune it.
There are two simple parameters:  an IRQ count threshold, and a time threshold.

Whenever either the count or the time threshold is exceeded,
the chip will generate an interrupt (assuming at least one completed event).

These two parameters are per-chip (actually, per group of four ports within the chip).
How can I export these for setting/query withing sysfs ?

It is all rather easy to create per-SCSI-host attrs in sysfs,
but I don't see any obvious way to do it per-chip. 

Suggestions?

For now, I'm doing it with a module parameter,
but that's really rather stone-age in the sysfs era.

Thanks.

Re: Per ATA host sysfs attributes?

[Jeff Garzik]

Mark Lord wrote:
> Guys,
> 
> I have added IRQ coalescing to sata_mv, but need a way to activate/tune it.
> There are two simple parameters:  an IRQ count threshold, and a time 
> threshold.
> 
> Whenever either the count or the time threshold is exceeded,
> the chip will generate an interrupt (assuming at least one completed 
> event).
> 
> These two parameters are per-chip (actually, per group of four ports 
> within the chip).
> How can I export these for setting/query withing sysfs ?
> 
> It is all rather easy to create per-SCSI-host attrs in sysfs,
> but I don't see any obvious way to do it per-chip.
> Suggestions?

The absence of a kobject associated with ata_host is largely the root 
cause of this problem you've noted.  Long term, we want to move in a 
direction where we have a "storage_host" container object, containing 
one or more port-style objects that can be ATA or SCSI.  In that 
scenario (and what we want to move towards, over the years), the sysfs 
attachment points should be fairly obvious.

Lacking a better suggestion, I suppose the answer is exporting per-chip 
variables multiple times, once per scsi-host.  Just get the locking 
right, since multiple sysfs attributes access the same underlying 
per-chip variable.


> For now, I'm doing it with a module parameter,
> but that's really rather stone-age in the sysfs era.

Indeed!  :)

	Jeff