Upstream: inspect ops() abstraction, move "private" functions

[plbossart]

Copy/paste from emails:

>> +       /* doorbell */
>> +       .irq_handler    = byt_irq_handler,
>> +       .irq_thread     = byt_irq_thread,
>> +
> What is the reason for having irq_handler/irq_thread functions inside the
> snd_sof_dsp_ops structure?
>
> These functions are never used outside via sdev->ops pointer.

Good point indeed, thanks for raising it. We were in the middle of
tagging which ops are required/optional (feedback from Mark) but we
started from the core and should have looked at the structure definition.

Most drivers are "self-contained" and can reference the irq_thread and
irq_handler directly.

The exception where the abstraction is used is internal to the HDaudio
stuff:

intel/hda.c:    ret = request_threaded_irq(sdev->ipc_irq,
hda_dsp_ipc_irq_handler,
intel/hda.c: sof_ops(sdev)->irq_thread, IRQF_SHARED,

That's useful since there a minor variations between hardware
generations and you want to hide the hardware-specific parts.

But as you point out, it's a "private" use of ops callbacks, the core
doesn't touch this.

I have no explanation other than legacy/historical reasons or a shortcut
to make one's life easier during development.  Liam and Keyon might know?

We could try and move this to a more "private" structure, the
"chip_info" part might be more suitable for this?

[stripes416]

Now that I'm back from leave, I'll be picking this back up.