Re: atomisp kernel driver(s)

[Andy Shevchenko <andy.shevchenko@gmail.com>]

On Sat, May 2, 2020 at 8:04 PM Mauro Carvalho Chehab <mchehab@kernel.org> wrote:
> Em Sat, 2 May 2020 19:08:36 +0300
> Andy Shevchenko <andy.shevchenko@gmail.com> escreveu:
>
> > I follow your attempts to enable that driver (I, myself, spent a lot
> > of time to an attempt of getting this driver in a shape). However, I
> > guess you started from a wrong side. Even with access to internal tree
> > for Android firmware we didn't manage to find a proper one to whatever
> > has been published in drivers/staging. So, to get it done, one should
> > first to find a *working* Android for the certain device. Without that
> > it will be a journey of wasted time and big disappointment.
>
> I found a driver that is probably working:
>         https://github.com/intel-aero/meta-intel-aero.git
>
> It is for an Intel Mobile Aero AUV platform. I suspect that his one
> should likely work.

So, have anybody tried to build and get a picture (raw is very good
start for it) out of that sources and respective firmware?

> > My achievements end with no getting IRQ from the driver (and I was
> > experimenting on MRD-7 CRB).
> >
> > P.S. I also have some (semi-) debug patches I can share.
> > Perhaps they will give some more ideas.
>
> Anything you can share will be welcomed.

https://paste.debian.net/1144410/

No SoB from me (despite patches have them).

> > Btw, based on this discussion I think that
> > it can be power issues with sensors that possible affect IRQ
> > generation inside SiliconHive vector processor.
>
> Yeah, the current issue sounds simple to solve, but I need to understand
> how an ACPI-based device would be calling regulator_register(). Using
> regulators with ACPI is new to me. I suspect that this should be done
> by ./arch/x86/platform/intel-mid, with of course doesn't have the needed
> bits for this board. Also, there is a dummy regulator driver for atomisp
> based boards (drivers/platform/x86/intel_atomisp2_pm.c). This one could
> be causing some issues too.
>
> The atomisp driver uses regulator_get() to turn on the sensors.

It should use PMIC to get them.

> In order for regulator_get() to work, regulator_dev_lookup() should
> be capable of finding a regulator either via DT or via the
> regulator_map_list.
>
> The contents of the regulator_map_list should visible on a configfs
> node (/sys/kernel/debug/regulator/supply_map).
>
> Yet, those aren't visible (probably because the ACPI data was written
> for Windows). So, the atomisp code should very likely call
> regulator_register() for the regulators with the atomisp driver
> need, in order to setup the regulator list.
>
>
> > In IPU3 the dedicated
> > PMIC is used for camera devices, and I have no idea of the design for
> > old ones.
>
> I have here a Dell notebook with IPU3 on it (marketed for MS windows):
>
>         ipu3_imgu: module is from the staging directory, the quality is unknown, you have been warned.
>         ipu3-imgu 0000:00:05.0: enabling device (0000 -> 0002)
>         ipu3-imgu 0000:00:05.0: device 0x1919 (rev: 0x1)
>         ipu3-imgu 0000:00:05.0: physical base address 0x00000000ec000000, 4194304 bytes
>         ipu3-imgu 0000:00:05.0: loaded firmware version irci_irci_ecr-master_20161208_0213_20170112_1500, 17 binaries, 1212984 bytes
>         ipu3-cio2 0000:00:14.3: enabling device (0000 -> 0002)
>         ipu3-cio2 0000:00:14.3: device 0x9d32 (rev: 0x1)
>
> This command:
>
>         # cat /sys/kernel/debug/regulator/supply_map
>
> Returns nothing. So, it seems that the very same issue may also be
> happening on IPU3-based laptops that won't have BIOSes designed to
> work on Linux.

Because you have to have an OpRegion in ACPI for the camera PMIC (see
this driver
https://elixir.bootlin.com/linux/latest/source/drivers/acpi/pmic/tps68470_pmic.c).

AFAIU ISPv2 uses the generic Atom PMIC (see other drivers in the above
mentioned folder).

-- 
With Best Regards,
Andy Shevchenko