Re: [PATCH v8 04/23] scsi: sd: Differentiate system and runtime start/stop management

[Phillip Susi <phill@thesusis.net>]

Damien Le Moal <dlemoal@kernel.org> writes:

> With the device links in place between port and scsi devices, we should be OK.
> But still need to check that we do not need runtime_get/put calls added.
> Ideally, we should have the chain:
>
> scsi disk -> scsi target -> scsi host -> ata port

It looks to me like there is an additional generic block device that
sits on top and that is what actually has the idle timeout.  Or maybe
that's the scsi disk, since it's name incldues the SCSI LUN, but in the
structure, its called sdev_gendev.  But then there's also sdev_dev, and
sdev_target.

> for runtime suspend, and the reverse for runtime resume. If there is a system
> suspend/resume between runtime suspend/resume, the port should not be resumed if
> it is runtime suspended.

I'm not sure about it.  The port has to be resumed so that we can
attempt to revalidate the devices on it.  For disks that have spun up on
their own, we should not leave then marked as runtime suspended, but
really they are spinning.  I suppose we could put them to sleep, though
I was leaning to just marking them as active, and leaving the runtime pm
timer to put them to sleep later, which then could allow the port to
suspend again.