| Message ID | E1a24wj-0004iw-Rf@rmk-PC.arm.linux.org.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Nov 26, 2015 at 10:23:21PM +0000, Russell King wrote: > Add documentation for the Marvell clock divider driver, which is used > to source clocks for the AXI bus, video decoder, GPU and LCD blocks. > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > --- > .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ > 1 file changed, 28 insertions(+) > create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > new file mode 100644 > index 000000000000..0c602de279e5 > --- /dev/null > +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > @@ -0,0 +1,28 @@ > +PLl divider based Dove clocks Nit pick. It would be nice if the second L of PLL was also a capital. Acked-by: Andrew Lunn <andrew@lunn.ch> Andrew
On Thu, Nov 26, 2015 at 10:23:21PM +0000, Russell King wrote: > Add documentation for the Marvell clock divider driver, which is used > to source clocks for the AXI bus, video decoder, GPU and LCD blocks. > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > --- > .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ > 1 file changed, 28 insertions(+) > create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > new file mode 100644 > index 000000000000..0c602de279e5 > --- /dev/null > +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > @@ -0,0 +1,28 @@ > +PLl divider based Dove clocks > + > +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide > +high speed clocks for a number of peripherals. These dividers are part of > +the PMU, and thus this node should be a child of the PMU node. It seems a bit strange to just be documenting these clocks. What about the rest of the SOC clocks? > + > +The following clocks are provided: > + > +ID Clock > +------------- > +0 AXI bus clock > +1 GPU clock > +2 VMeta clock > +3 LCD clock > + > +Required properties: > +- compatible : shall be "marvell,dove-divider-clock" > +- reg : shall be the register address of the Core PLL and Clock Divider > + Control 0 register. This will cover that register, as well as the > + Core PLL and Clock Divider Control 1 register. Thus, it will have > + a size of 8. > +- #clock-cells : from common clock binding; shall be set to 1 > + > +divider_clk: core-clock@0064 { > + compatible = "marvell,dove-divider-clock"; > + reg = <0x0064 0x8>; > + #clock-cells = <1>; > +}; > -- > 2.1.0 >
> > +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide > > +high speed clocks for a number of peripherals. These dividers are part of > > +the PMU, and thus this node should be a child of the PMU node. > > It seems a bit strange to just be documenting these clocks. What about > the rest of the SOC clocks? $ ls Documentation/devicetree/bindings/clock/mvebu* Documentation/devicetree/bindings/clock/mvebu-core-clock.txt Documentation/devicetree/bindings/clock/mvebu-corediv-clock.txt Documentation/devicetree/bindings/clock/mvebu-cpu-clock.txt Documentation/devicetree/bindings/clock/mvebu-gated-clock.txt Dove is a member of mvebu, and gets most of its clocks from these drivers. But it has additional clocks which are not shared with other members of mvebu. Andrew
On Fri, Nov 27, 2015 at 02:21:14PM -0600, Rob Herring wrote: > On Thu, Nov 26, 2015 at 10:23:21PM +0000, Russell King wrote: > > Add documentation for the Marvell clock divider driver, which is used > > to source clocks for the AXI bus, video decoder, GPU and LCD blocks. > > > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > > --- > > .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ > > 1 file changed, 28 insertions(+) > > create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > > > diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > new file mode 100644 > > index 000000000000..0c602de279e5 > > --- /dev/null > > +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > @@ -0,0 +1,28 @@ > > +PLl divider based Dove clocks > > + > > +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide > > +high speed clocks for a number of peripherals. These dividers are part of > > +the PMU, and thus this node should be a child of the PMU node. > > It seems a bit strange to just be documenting these clocks. What about > the rest of the SOC clocks? This is all that this pair of registers provide. The SoC has other clocks handled by other DT nodes - for Dove, we now have: gate_clk: clock-gating-ctrl@0038 { compatible = "marvell,dove-gating-clock"; reg = <0x0038 0x4>; clocks = <&core_clk 0>; #clock-cells = <1>; }; divider_clk: core-clock@0064 { compatible = "marvell,dove-divider-clock"; reg = <0x0064 0x8>; #clock-cells = <1>; }; core_clk: core-clocks@0214 { compatible = "marvell,dove-core-clock"; reg = <0x0214 0x4>; #clock-cells = <1>; }; and all three of these are part of the PMU register block. I'm not sure why the mvebu maintainers decided to minutely describe the PMU like this, but unfortunately that's the structure we have. What's more silly is that the "dove-core-clock" appears to disagree in terminology with the manual - there's a "Core PLL" which supplies the dividers in the "divider_clk" block, and is entirely separate from the CPU PLL which "core_clk" is describing. IMHO, it would've been cleaner to have these components registered separately from a central PMU driver (as I'm doing with the power domains, resets and IRQs that are part of the PMU), but my view is limited to Dove and not the other mvebu clocks, so there may be a good reason for it.
On 27.11.2015 21:39, Russell King - ARM Linux wrote: > On Fri, Nov 27, 2015 at 02:21:14PM -0600, Rob Herring wrote: >> On Thu, Nov 26, 2015 at 10:23:21PM +0000, Russell King wrote: >>> Add documentation for the Marvell clock divider driver, which is used >>> to source clocks for the AXI bus, video decoder, GPU and LCD blocks. >>> >>> Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> >>> --- >>> .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ >>> 1 file changed, 28 insertions(+) >>> create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt >>> >>> diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt >>> new file mode 100644 >>> index 000000000000..0c602de279e5 >>> --- /dev/null >>> +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt >>> @@ -0,0 +1,28 @@ >>> +PLl divider based Dove clocks >>> + >>> +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide >>> +high speed clocks for a number of peripherals. These dividers are part of >>> +the PMU, and thus this node should be a child of the PMU node. >> >> It seems a bit strange to just be documenting these clocks. What about >> the rest of the SOC clocks? > > This is all that this pair of registers provide. > > The SoC has other clocks handled by other DT nodes - for Dove, we now > have: > > gate_clk: clock-gating-ctrl@0038 { > compatible = "marvell,dove-gating-clock"; > reg = <0x0038 0x4>; > clocks = <&core_clk 0>; > #clock-cells = <1>; > }; > divider_clk: core-clock@0064 { > compatible = "marvell,dove-divider-clock"; > reg = <0x0064 0x8>; > #clock-cells = <1>; > }; > core_clk: core-clocks@0214 { > compatible = "marvell,dove-core-clock"; > reg = <0x0214 0x4>; > #clock-cells = <1>; > }; > > and all three of these are part of the PMU register block. I'm not sure > why the mvebu maintainers decided to minutely describe the PMU like this, > but unfortunately that's the structure we have. > > What's more silly is that the "dove-core-clock" appears to disagree in > terminology with the manual - there's a "Core PLL" which supplies the > dividers in the "divider_clk" block, and is entirely separate from the > CPU PLL which "core_clk" is describing. > > IMHO, it would've been cleaner to have these components registered > separately from a central PMU driver (as I'm doing with the power > domains, resets and IRQs that are part of the PMU), but my view is > limited to Dove and not the other mvebu clocks, so there may be a good > reason for it. > Rob, Russell, the main reason why dove clocks are the way they are is that back when we thought of the bindings for mvebu SoC clocks, CCF and DT was in its very beginning. Looking back, some decisions shouldn't have been made that way. Also, we tried to describe Dove and the other SoCs similarily, but Dove is always a little odd. From todays point of view, I certainly agree with Russell that all PMU related stuff is best kept in a single PMU node. Thanks for providing the patches, I appreciate the extra work of separating those from your working trees. Acked-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com> Sebastian
On Fri, Nov 27, 2015 at 08:39:32PM +0000, Russell King - ARM Linux wrote: > On Fri, Nov 27, 2015 at 02:21:14PM -0600, Rob Herring wrote: > > On Thu, Nov 26, 2015 at 10:23:21PM +0000, Russell King wrote: > > > Add documentation for the Marvell clock divider driver, which is used > > > to source clocks for the AXI bus, video decoder, GPU and LCD blocks. > > > > > > Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> > > > --- > > > .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ > > > 1 file changed, 28 insertions(+) > > > create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > > > > > diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > > new file mode 100644 > > > index 000000000000..0c602de279e5 > > > --- /dev/null > > > +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt > > > @@ -0,0 +1,28 @@ > > > +PLl divider based Dove clocks > > > + > > > +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide > > > +high speed clocks for a number of peripherals. These dividers are part of > > > +the PMU, and thus this node should be a child of the PMU node. > > > > It seems a bit strange to just be documenting these clocks. What about > > the rest of the SOC clocks? > > This is all that this pair of registers provide. > > The SoC has other clocks handled by other DT nodes - for Dove, we now > have: > > gate_clk: clock-gating-ctrl@0038 { > compatible = "marvell,dove-gating-clock"; > reg = <0x0038 0x4>; > clocks = <&core_clk 0>; > #clock-cells = <1>; > }; > divider_clk: core-clock@0064 { > compatible = "marvell,dove-divider-clock"; > reg = <0x0064 0x8>; > #clock-cells = <1>; > }; > core_clk: core-clocks@0214 { > compatible = "marvell,dove-core-clock"; > reg = <0x0214 0x4>; > #clock-cells = <1>; > }; > > and all three of these are part of the PMU register block. I'm not sure > why the mvebu maintainers decided to minutely describe the PMU like this, > but unfortunately that's the structure we have. Okay, Acked-by: Rob Herring <robh@kernel.org> > IMHO, it would've been cleaner to have these components registered > separately from a central PMU driver (as I'm doing with the power > domains, resets and IRQs that are part of the PMU), but my view is > limited to Dove and not the other mvebu clocks, so there may be a good > reason for it. Yes, we debated which way to go early on and decided generally to go this direction, but it seems Dove was before that. We could try to migrate Dove, but it is probably not worth the pain at this point. Rob
diff --git a/Documentation/devicetree/bindings/clock/dove-divider-clock.txt b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt new file mode 100644 index 000000000000..0c602de279e5 --- /dev/null +++ b/Documentation/devicetree/bindings/clock/dove-divider-clock.txt @@ -0,0 +1,28 @@ +PLl divider based Dove clocks + +Marvell Dove has a 2GHz PLL, which feeds into a set of dividers to provide +high speed clocks for a number of peripherals. These dividers are part of +the PMU, and thus this node should be a child of the PMU node. + +The following clocks are provided: + +ID Clock +------------- +0 AXI bus clock +1 GPU clock +2 VMeta clock +3 LCD clock + +Required properties: +- compatible : shall be "marvell,dove-divider-clock" +- reg : shall be the register address of the Core PLL and Clock Divider + Control 0 register. This will cover that register, as well as the + Core PLL and Clock Divider Control 1 register. Thus, it will have + a size of 8. +- #clock-cells : from common clock binding; shall be set to 1 + +divider_clk: core-clock@0064 { + compatible = "marvell,dove-divider-clock"; + reg = <0x0064 0x8>; + #clock-cells = <1>; +};
Add documentation for the Marvell clock divider driver, which is used to source clocks for the AXI bus, video decoder, GPU and LCD blocks. Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> --- .../bindings/clock/dove-divider-clock.txt | 28 ++++++++++++++++++++++ 1 file changed, 28 insertions(+) create mode 100644 Documentation/devicetree/bindings/clock/dove-divider-clock.txt