On Monday 09 December 2013, boris brezillon wrote:Hello Arnd,Yes, that is a bit painful, but we really need to ensure that the drivers
On 09/12/2013 16:48, Arnd Bergmann wrote:On Monday 09 December 2013, boris brezillon wrote:We have to keep both clk system working for at91 platform (new CCF basedYes, I understood that the *drivers* use the names, but are they actuallyYou are adding "clock-names" properties in a lot of cases. Are you sure youI rechecked it, and almost all drivers call [devm_]clk_get with a
are using the strings that are documented in the respective device bindings
for each name? In a lot of cases, drivers just want an anonymous clock
and we don't name them.
specific clock
name, and as a result we must specify the "clock-names" property.
The only exceptions I found are the spi and PIT (Periodic Interval
Timer) drivers,
and "clock-names" property is not defined in these nodes.
documented in the device bindings? If not, it might be better to change the
drivers.
clk implementations and old clk implementations) for two reasons:
1) the new clk implemetations are only compatible with DT enabled boards
and some at91 boards are not yet ported to DT.
2) we decided to move to the new clk implementations in multiple steps
(one SoC
family after another) in order to ease PATCH review.
If we change the drivers to avoid specific clock name request
([devm]_clk_get(dev, NULL)),
we will have to change the old static clk lookup tables
(i.e.
http://lxr.free-electrons.com/source/arch/arm/mach-at91/at91sam9g45.c#L226,
http://lxr.free-electrons.com/source/arch/arm/mach-at91/at91sam9260.c#L202,
...).
implement the documented binding. If they don't match we have to either
review and change all the binding documents, or change the static clock
tables and the drivers.
Note that I have not checked your binding documents for the devices in
question. If they already document these names, we don't need to change
anything.
Arnd