Skip to contentSkip to conversation

clk: dt: bindings for mux-clock

DISCUSSION Sergej Sawazki 2 participants 4 messages Mar 19, 2015 – Apr 1, 2015

lore
Share & tools
lore
#0 lore raw

Hi Mike,

I came across your "[PATCH v2 0/5] clk: dt: bindings for mux, divider & gate clocks" email from 16 Jun 2013. The DT bindings for simple clock multiplexers would be very helpful for a board I am working on. Do you see any chance to get it into mainline?

Many thanks in advance!
Regards,
Sergej

#1 lore raw

Quoting Sergej Sawazki (2015-03-19 14:50:50)

Quoted 6 lines

Hi Mike,

I came across your "[PATCH v2 0/5] clk: dt: bindings for mux, divider & gate clocks" email from 16 Jun 2013. The DT bindings for simple clock multiplexers would be very helpful for a board I am working on. Do you see any chance to get it into mainline?

Hi Sergej,

I abandoned those binding a while back. The reason is that those are one-node-per-clock bindings, which are unpopular with the DT crowd. Instead most bindings today use a single node to represent a clock provider, which maps onto a clock driver in Linux.

Is your clock provider made up of only a single clock? If so then the bindings you mentioned above may be appropriate. But if you have a clock controller IP block that manages several clocks then it is better for you to follow the clock provider binding style. There is no shortage of good examples on how to do this. See the QCOM, Samsung and Nvidia bindings for ideas.

Regards,
Mike

Quoted 4 lines

Many thanks in advance!
Regards,
Sergej

#2 lore raw

Am 22.03.2015 um 18:10 schrieb Michael Turquette:

Quoted 22 lines

Quoting Sergej Sawazki (2015-03-19 14:50:50)

Quoted 6 lines

Hi Mike,

I came across your "[PATCH v2 0/5] clk: dt: bindings for mux, divider & gate clocks" email from 16 Jun 2013. The DT bindings for simple clock multiplexers would be very helpful for a board I am working on. Do you see any chance to get it into mainline?

Hi Sergej,

I abandoned those binding a while back. The reason is that those are one-node-per-clock bindings, which are unpopular with the DT crowd. Instead most bindings today use a single node to represent a clock provider, which maps onto a clock driver in Linux.

Is your clock provider made up of only a single clock? If so then the bindings you mentioned above may be appropriate. But if you have a clock controller IP block that manages several clocks then it is better for you to follow the clock provider binding style. There is no shortage of good examples on how to do this. See the QCOM, Samsung and Nvidia bindings for ideas.

Quoted 7 lines

Regards,
Mike

Quoted 4 lines

Many thanks in advance!
Regards,
Sergej

Hi Mike, many thanks for your answer.

My clock provider is made up of two external oscillators and an external clock multiplexer. The clock multiplexer has two inputs and one output. See IDT 853S01I for example. The oscillators are connected to the multiplexer inputs. The clock consumer is connected to the output of the multiplexer. The multiplexer is controlled by a gpio to select one of the oscillators.

Based on clk-gpio-gate.c, I am considering to develop a driver for a gpio controlled clock multiplexer. Do you think it makes sense? Or should I choose a different approach?

Regards,
Sergej

#3 lore raw

Quoting Sergej Sawazki (2015-03-25 12:19:42)

Quoted 39 lines

Am 22.03.2015 um 18:10 schrieb Michael Turquette:

Quoted 22 lines

Quoting Sergej Sawazki (2015-03-19 14:50:50)

Quoted 6 lines

Hi Mike,

I came across your "[PATCH v2 0/5] clk: dt: bindings for mux, divider & gate clocks" email from 16 Jun 2013. The DT bindings for simple clock multiplexers would be very helpful for a board I am working on. Do you see any chance to get it into mainline?

Hi Sergej,

I abandoned those binding a while back. The reason is that those are one-node-per-clock bindings, which are unpopular with the DT crowd. Instead most bindings today use a single node to represent a clock provider, which maps onto a clock driver in Linux.

Is your clock provider made up of only a single clock? If so then the bindings you mentioned above may be appropriate. But if you have a clock controller IP block that manages several clocks then it is better for you to follow the clock provider binding style. There is no shortage of good examples on how to do this. See the QCOM, Samsung and Nvidia bindings for ideas.

Quoted 7 lines

Regards,
Mike

Quoted 4 lines

Many thanks in advance!
Regards,
Sergej

Hi Mike, many thanks for your answer.

My clock provider is made up of two external oscillators and an external clock multiplexer. The clock multiplexer has two inputs and one output. See IDT 853S01I for example. The oscillators are connected to the multiplexer inputs. The clock consumer is connected to the output of the multiplexer. The multiplexer is controlled by a gpio to select one of the oscillators.

Quoted 3 lines

Based on clk-gpio-gate.c, I am considering to develop a driver for a gpio controlled clock multiplexer. Do you think it makes sense? Or should I choose a different approach?

It makes enough sense to try. Hopefully there will be others on the list that can use it. Be sure to reference the hardware you are using in your commitlog as that tends to help.

Also I haven't spent much time looking at the gpio gate stuff, but you might consider whether your mux code should be combined with that somehow.

Regards,
Mike

Quoted 4 lines

Regards,
Sergej