[PATCH v4 01/11] mfd: add pruss mfd driver.

[Arnd Bergmann]

On Thursday 28 April 2011 09:17:21 Subhasish Ghosh wrote:
> >
> > You can easily do that by adding a small header to the firmware
> > format and interpret that header by the MFD driver. When the name
> > of the subdevice is part of that header, the MFD driver does not
> > need to understand the difference, it can simply pass that on
> > when creating its child devices.
> 
> I don't understand why loading the firmware should be done at the MFD 
> driver.
> The user already specifies the device he/she wants to start on the PRU via 
> modprobe.
> A driver can be inserted, which can download a printer firmware on one PRU 
> and a
> scanner firmware on the other. This way both cores can be used for separate 
> purposes.
> I mean, say in a real MFD controller, that will also have two separate cores 
> running on it,
> just that, the firmware on it would not be downloaded runtime but fused in 
> some non volatile memory. 

Then I must be misreading what your code currently does, because it does not
match your explanations. What I see in the platform code is that you create
MFD cells for specific devices that get automatically created by the MFD
driver. This will cause udev to load the drivers for these devices, which
then load the firmware they need.

Also, I cannot see how the method you describe would make it possible to
the same driver into both units, e.g. when you want to have two serial
ports. The reason is that you currently hardcode the PRU number in the
driver and that you cannot load a single driver twice.

Finally, I'm trying to make sure that whatever solution you come up with
will still work when we migrate the code to using a flattened device tree.
In that case, you would ideally put the device firmware into the device
tree as a property that matches whatever you have connected on the specific
board (at least as an option, you can still fall back to request_firmware).
You definitely want automatic module loading in that case.

Note that using module loading with specific parameters in order to
match the hardware is not a recommended procedure any more. The code
really needs to work the same way when all drivers are built into the
kernel. It should not be hard to use the firmware loading mechanism
in the MFD driver to both load the firmware and configure the devices
appropriately so we always use the right driver for the currently
active devices.

	Arnd

BTW, something is wrong with your email client line wrapping. I've fixed
this up manually before when replying, but please find a way to get this
right in the future.