Re: IBM HDAPS things are looking up (was: Re: [Hdaps-devel] Re: [ltp] IBM HDAPS Someone interested? (Accelerometer))

[Jens Axboe <axboe@suse.de>]

On Thu, Jul 07 2005, Matthew Garrett wrote:
> Jens Axboe <axboe@suse.de> wrote:
> 
> > What is needed is to flesh out what the kernel interface should looke
> > like. I suggested a sysfs file for suspending and resuming access to the
> > device, if people have other ideas they should voice them.
> 
> That sounds quite reasonable. Does it need to do anything other than
> park the head and suspend the command queue for that device?

Not really, no. Someone mentioned a timeout for this as well, but I
think that should just be done in user space.

> (On an only slightly related note, for full ACPI support of PATA, we're
> supposed to use the _GTF interface. This returns a set of taskfile
> commands that are then supposed to be executed by the host. However, at
> the point where we want to do this, the IDE queues haven't been
> restarted. Is the best solution here just to add a trivial and stupid
> IDE driver for managing the disks when we don't want userspace doing
> anything with them?)

Either that, or always allow non-fs commands to go through a frozen
queue. Or flag those commands as ok for a frozen queue. The advantage of
either of those approaches, is that we probably don't have to add any
kernel support then and the generic block device freezing/unfreezing can
be used.

-- 
Jens Axboe