Re: Reset implementation for Zynq

[Moritz Fischer]

Philip,

On Mon, Oct 24, 2016 at 11:00:05AM +0200, Philipp Zabel wrote:
> Hi Iztok, Moritz,
> 
> Am Freitag, den 21.10.2016, 10:04 -0700 schrieb Moritz Fischer:
> > Iztok,
> > 
> > On Fri, Oct 21, 2016 at 03:08:47AM -0700, iztok.jeras@xxxxxxxxxxxxx wrote:
> > > Hi Moritz,
> > >  
> > > I was looking at your reset implementation for Zynq:
> > > https://github.com/Xilinx/linux-xlnx/blob/629041605b93343ad2e8971ceaac3edcef0b043b/drivers/reset/reset-zynq.c
> > > I went through related mailing list posts (including earlier versions of the patch) so I kind of understand what to change in the device tree.
> > 
> > Please look at the upstream kernel sources and use the mailing list
> > (lkml) if you want to report bugs. Xilinx' vendor tree might or might
> > not be up to date.
> > 
> > > I would like to use this driver to reset the Zynq I2C controller, since we have trouble with it getting into a lock up state.
> > > I plan to use function device_reset_optional() from:
> > > https://github.com/Xilinx/linux-xlnx/blob/629041605b93343ad2e8971ceaac3edcef0b043b/include/linux/reset.h
> > >  
> > > But this function is calling the reset function pointer from the reset_control_ops structure.
> > > For the zynq driver this function pointer is not defined, only assert, deassert and status are.
> > >  
> > > Is this a missing implementation, or is there a default implementation (I did not find one) which which performs an assert+deassert,
> > > or is there another set of reset APIs I should use inside the kernel.
> > 
> > You could just call reset_control_assert() and reset_control_deassert().
> > You're right there is currently no implementation for the 'reset' function for
> > zynq (and most of the other SoCs). I'll need to see if it makes sense at
> > all.
> 
> The implementation of reset_control_reset in software really only makes
> sense if the reset provider driver knows about the necessary delays for
> all reset consumers.

That's what I meant by needing to see if it makes sense at all; it makes
no sense to have a 'reset' if you don't know how long it needs to be
asserted for, since that's obviously a consumer property.

Thanks for clarifying,

Moritz