On 21 May 2015 at 23:17, Michael Turquette <mturquette@xxxxxxxxxx> wrote: > Quoting Joachim Eastwood (2015-05-13 00:57:18) >> On 13 May 2015 at 00:12, Michael Turquette <mturquette@xxxxxxxxxx> wrote: >> > >> > Why do you need to use clock-indices? >> >> Since the CGU can have up to 27 clock lines, but in this device not >> all are used. So I use clock-indices so I don't need to have empty >> entries in the clock-output-names array. I thought this was the >> intended usage(?) >> >> > Why do you need to use clock-output-names? If all of your clock >> > consumers have nodes in DT then you can skip this and name them on the >> > consuming side. See a further explanation here: >> > http://lkml.kernel.org/r/<20150416192014.19585.9663@quantum> >> >> Most of the clocks from the CGU goes to a CCU, but the CCU isn't the >> real consumer. It's more like a clock router with a bunch of gates. >> >> If you take a close look at the CCU driver you will notice that it >> doesn't do clk_get on any of the clocks. My reason for doing this is >> to not increase the usage counter on the CGU clock so that if there >> are no consumer after a CCU the entire clock line up to the CGU can be >> disabled. So this is the reason that the CCU driver uses a "special" >> way to grab the clock output names from the CGU. > > A call to clk_get will not enable/ungate the clock signal. It seems that > there might be some confusion around that. It is always OK to get the > clk. This will not increase the enable_count or prepare_count values. > > If the CCU is a separate IP block then it might make sense for it to > call clk_get on the CGU clocks. On the other hand it might make more > sense for the CCU clocks to simply use the CGU clocks as parent clock > signal inputs. v3, that was sent a couple of days ago, uses of_clk_get/clk_get_name to set the CCU clock parents. See v3 cover letter for a short explanation. > None of the points above have anything to do with my original comment. > The use of the clock-output-names property is unnecessary if all of the > consumer devices have DT nodes of their own. Please look at the thread I > linked to and let me know what you think about getting rid of the > clock-output-names property. v3 removes the property. Please take a look at it when you get the time. regards, Joachim Eastwood -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html