On Thu, Oct 13, 2016 at 03:52:45PM +0200, Benjamin Tissoires wrote:

quoted

quoted

You could basically export rmi_fn_reset() which would call
rmi_free_function_list(), rmi_scan_pdt (if initial reset),
rmi_probe_interrupts() and rmi_init_functions, and this would allow you
to have all this in f34.

I see what you mean, and I do agree that it would be neater to have all
of this in the f34 code.

However, the problem is that when you call rmi_free_function_list(), the
f34 driver and all the context information attached to it (struct
rmi_function, struct f34_data, and any sysfs attributes in the f34
directory) gets torn down, so you're kind of left without the branch you
were sitting on.

To get around that, I'd have to make f34 a special case anyway. Taking
that into account, the current solution seemed neater to me. I could
possibly cram a little more of it into rmi_f34.c, but I think the
context has to be "struct rmi_driver_data".

If I understand correctly, rmi_firmware_update() is only called through
the sysfs. So how about you export the required functions from core you
are using and export 2 functions from rmi_f34 that will be a special
case: rmi_f34_create_sysfs() and rmi_f34_remove_sysfs() (or any better
names). You could just put your code in rmi_f34, provide noops
declarations if RMI_F34 is not set, and core will have only 2 calls to
rmi_f34.

OK: I will have a go at achieving this over the next few days. I also
have some changes almost ready to add F34 bootloader V7 support.

BTW, I am thinking at carrying in my next RMI4 series your 1/2 patch. I
also want to take Bjorn and Andrew left patches so that we have a common
tree at some point. Any objections? 
Of course, if you resubmit before me, feel free to carry over 1/2.

Sounds like a good idea, it will reduce merging difficulty later.

N