Re: [PATCH 01/11] pinctrl: mvebu: pinctrl driver core

[Thomas Petazzoni <thomas.petazzoni@free-electrons.com>]

Hello,

Le Mon, 20 Aug 2012 11:46:14 +0200,
Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com> a écrit :

> >> +uart1: serial@12100 {
> >> +       compatible = "ns16550a";
> >> +       reg = <0x12100 0x100>;
> >> +       reg-shift = <2>;
> >> +       interrupts = <7>;
> >> +       clock-frequency = <166666667>;
> >
> > It's got nothing to do with this patch, but getting a clock frequency
> > out of the DT instead of getting it from the clk_get_rate(clk) and
> > the clock tree seems absurd... (But maybe this platform does not
> > even have a clk implementation?)
> 
> It's of_serial's implementation. I patched that once for getting
> frequency out of "clocks" property but then I got busy with
> porting mach-dove and pinctrl.. Marvell SoCs do have a clk
> implementation and as soon as of_serial can handle "clocks"
> property it will be used for sure. I can remove "clock-frequency"
> from the example anyway as it is not really part of pinctrl
> binding documentation.

We are also working on using the clk framework for the 370/XP support
(my colleague Grégory in Cc has started working on this last week), and
we also want to be able to get the serial clock-frequency from the clk
framework instead of an explicit value in the DT node. But that's a
separate topic :)

> > Is it possible to use devm_* managed devm_kzalloc() for this map
> > so you don't need to free it explicitly?
> >
> > (Maybe not, just checking.)
> 
> Hmm, I guess not as I thought I've read not to use devm_kfree when
> you allocate _and_ free stuff on runtime without removing the device
> itself, right?

It is also my understanding that devm_*() functions should be used to
allocate things that should persist until the device is removed. But I
might be wrong here.

> >> +struct mvebu_mpp_ctrl {
> >> +       const char *name;
> >> +       u8 pid;
> >> +       u8 npins;
> >
> > So, there will never be > 256 pins on a Marvell platform?
> 
> Well, with all current platforms we are well below 100. I guess
> 256 max (muxable) pins will be enough.

Agreed, and this structure is completely internal to the kernel, so we
can easily change it in the future if needed.

> >> + * struct mvebu_mpp_ctrl_setting - describe a mpp ctrl setting
> >> + * @val: ctrl setting value
> >
> > It is not obvious to me what this means, it it possible to elaborate
> > on how this member is defined and used?
> 
> Well, I see if I can clarify the description but wrt the datasheet it
> _should_ be quite obvious.

I think the setting/function/group/control terminology would benefit
from an explanation, as it isn't very easy to figure out what all these
words mean in the context of the pinctrl-mvebu driver.

> In some internal review with Andrew I also added a spinlock to
> mvebu_pinconf_get/_set that will protect all calls to generic and specific
> _get/_set register accesses. Moreover, I replaced clk_get_sys in pinctrl-dove
> with the devm_ counterpart and removed the explicit clk_put.

Yes, I had seen this discussion, but I am not sure it is needed: it
seems the pinctrl core calls all the pinconf_set/pinconf_get methods
with the pinctrl_mutex held. When I wrote an initial pinctrl driver for
370/XP I had the same question as Andrew and my conclusion was that the
locking done by the pinctrl subsystem core was sufficient.

Best regards,

Thomas
-- 
Thomas Petazzoni, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com