| Message ID | 20180721192846.18811-2-martin.blumenstingl@googlemail.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | Meson8/Meson8b: introduce a HHI syscon node | expand |
On Sat, Jul 21, 2018 at 09:28:44PM +0200, Martin Blumenstingl wrote: > The clock controller on Meson8/Meson8m2 and Meson8b is part of a > register region called "HHI". This register area contains more > functionality than just a clock controller: > - the clock controller > - some reset controller bits > - temperature sensor calibration data (on Meson8b and Meson8m2 only) > - HDMI controller > > The HHI register area may be accessed concurrently. Allow this by using > a "system controller" as parent node. Why? A single node can be a provider of multiple things. Maybe the HDMI should be a child since it will involve graph nodes, but the rest can be one node. There should be numerous examples of blocks that are clock and reset controllers. > > Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> > --- > .../bindings/clock/amlogic,meson8b-clkc.txt | 13 +++++-------- > 1 file changed, 5 insertions(+), 8 deletions(-) > > diff --git a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > index b455c5aa9139..38fb979210d3 100644 > --- a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > +++ b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > @@ -9,15 +9,13 @@ Required Properties: > - "amlogic,meson8-clkc" for Meson8 (S802) SoCs > - "amlogic,meson8b-clkc" for Meson8 (S805) SoCs > - "amlogic,meson8m2-clkc" for Meson8m2 (S812) SoCs > -- reg: it must be composed by two tuples: > - 0) physical base address of the xtal register and length of memory > - mapped region. > - 1) physical base address of the clock controller and length of memory > - mapped region. > - > - #clock-cells: should be 1. > - #reset-cells: should be 1. > > +Parent node should have the following properties : > +- compatible: "syscon", "simple-mfd" These 2 compatibles alone are not valid. > +- reg: base address and size of the HHI system control register space. > + > Each clock is assigned an identifier and client nodes can use this identifier > to specify the clock which they consume. All available clocks are defined as > preprocessor macros in the dt-bindings/clock/meson8b-clkc.h header and can be > @@ -30,9 +28,8 @@ device tree sources). > > Example: Clock controller node: > > - clkc: clock-controller@c1104000 { > + clkc: clock-controller { > compatible = "amlogic,meson8b-clkc"; > - reg = <0xc1108000 0x4>, <0xc1104000 0x460>; > #clock-cells = <1>; > #reset-cells = <1>; > }; > -- > 2.18.0 >
Hi Rob, On Wed, Jul 25, 2018 at 10:07 PM Rob Herring <robh@kernel.org> wrote: > > On Sat, Jul 21, 2018 at 09:28:44PM +0200, Martin Blumenstingl wrote: > > The clock controller on Meson8/Meson8m2 and Meson8b is part of a > > register region called "HHI". This register area contains more > > functionality than just a clock controller: > > - the clock controller > > - some reset controller bits > > - temperature sensor calibration data (on Meson8b and Meson8m2 only) > > - HDMI controller > > > > The HHI register area may be accessed concurrently. Allow this by using > > a "system controller" as parent node. > > Why? A single node can be a provider of multiple things. Maybe the HDMI > should be a child since it will involve graph nodes, but the rest can be > one node. There should be numerous examples of blocks that are clock and > reset controllers. I understand that a node can provide multiple "things" - currently it's a clock controller and a reset controller the HDMI controller could also be integrated in a similar way however, I do not know how to access the temperature sensor calibration data there is an ADC - one of it's channel has access to a temperature sensor this ADC is located at CBUS offset 0x8680 and we already have a driver for it (meson-saradc) my problem is that the temperature sensor has to be calibrated - this is done by: - read data from efuse - write 4 bits of calibration data to some register in the ADC's register space - write a 5th bit of calibration data to a (seemingly random) register in the HHI register space (if one of the 5 bits is not written to it's correct location then the temperature sensor reads bogus values) I am not sure how to handle this without passing the HHI region to the meson-saradc driver and letting that initialize all the temperature calibration data bits (in it's own register space as well as the HHI register space) do you have any suggestion here? > > > > Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> > > --- > > .../bindings/clock/amlogic,meson8b-clkc.txt | 13 +++++-------- > > 1 file changed, 5 insertions(+), 8 deletions(-) > > > > diff --git a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > > index b455c5aa9139..38fb979210d3 100644 > > --- a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > > +++ b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > > @@ -9,15 +9,13 @@ Required Properties: > > - "amlogic,meson8-clkc" for Meson8 (S802) SoCs > > - "amlogic,meson8b-clkc" for Meson8 (S805) SoCs > > - "amlogic,meson8m2-clkc" for Meson8m2 (S812) SoCs > > -- reg: it must be composed by two tuples: > > - 0) physical base address of the xtal register and length of memory > > - mapped region. > > - 1) physical base address of the clock controller and length of memory > > - mapped region. > > - > > - #clock-cells: should be 1. > > - #reset-cells: should be 1. > > > > +Parent node should have the following properties : > > +- compatible: "syscon", "simple-mfd" > > These 2 compatibles alone are not valid. so I should add (for example) "amlogic,meson8b-hhi-sysctrl" as first compatible? so the result would be: compatible = "amlogic,meson8b-hhi-sysctrl", "syscon", "simple-mfd"; Regards Martin
Hi Rob, Martin, On 25/07/2018 23:16, Martin Blumenstingl wrote: > Hi Rob, > > On Wed, Jul 25, 2018 at 10:07 PM Rob Herring <robh@kernel.org> wrote: >> >> On Sat, Jul 21, 2018 at 09:28:44PM +0200, Martin Blumenstingl wrote: >>> The clock controller on Meson8/Meson8m2 and Meson8b is part of a >>> register region called "HHI". This register area contains more >>> functionality than just a clock controller: >>> - the clock controller >>> - some reset controller bits >>> - temperature sensor calibration data (on Meson8b and Meson8m2 only) >>> - HDMI controller >>> >>> The HHI register area may be accessed concurrently. Allow this by using >>> a "system controller" as parent node. >> >> Why? A single node can be a provider of multiple things. Maybe the HDMI >> should be a child since it will involve graph nodes, but the rest can be >> one node. There should be numerous examples of blocks that are clock and >> reset controllers. > I understand that a node can provide multiple "things" - currently > it's a clock controller and a reset controller > the HDMI controller could also be integrated in a similar way > > however, I do not know how to access the temperature sensor calibration data > there is an ADC - one of it's channel has access to a temperature sensor > this ADC is located at CBUS offset 0x8680 and we already have a driver > for it (meson-saradc) > my problem is that the temperature sensor has to be calibrated - this > is done by: > - read data from efuse > - write 4 bits of calibration data to some register in the ADC's register space > - write a 5th bit of calibration data to a (seemingly random) register > in the HHI register space > (if one of the 5 bits is not written to it's correct location then the > temperature sensor reads bogus values) > > I am not sure how to handle this without passing the HHI region to the > meson-saradc driver and letting that initialize all the temperature > calibration data bits (in it's own register space as well as the HHI > register space) > do you have any suggestion here? The clock controller node was badly represented from the beginning, the HHI register zone *is* a "system control" zone with a lot of different registers controller the system, including all the EE clocks, *some* HDMI PHY settings, *some* Video DAC settings, *some* memory power control... Thus, the clock controller *is* entirely in the HHI registers space, but the HHI register space should be used by other nodes for control some part of the SoCs. This is why we decided to model the HHI zone as a "syscon" to permit other node to have access to these registers and "simple-mfd" to expose the *big* feature of the HHI : the clock controller. We are migrating the Meson8 because we already migrated the GX and because the AXG family was pushed like this (and acked like this), so it's common sense to align the 3 families because they share the same characteristics here. Neil > >>> >>> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> >>> --- >>> .../bindings/clock/amlogic,meson8b-clkc.txt | 13 +++++-------- >>> 1 file changed, 5 insertions(+), 8 deletions(-) >>> >>> diff --git a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt >>> index b455c5aa9139..38fb979210d3 100644 >>> --- a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt >>> +++ b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt >>> @@ -9,15 +9,13 @@ Required Properties: >>> - "amlogic,meson8-clkc" for Meson8 (S802) SoCs >>> - "amlogic,meson8b-clkc" for Meson8 (S805) SoCs >>> - "amlogic,meson8m2-clkc" for Meson8m2 (S812) SoCs >>> -- reg: it must be composed by two tuples: >>> - 0) physical base address of the xtal register and length of memory >>> - mapped region. >>> - 1) physical base address of the clock controller and length of memory >>> - mapped region. >>> - >>> - #clock-cells: should be 1. >>> - #reset-cells: should be 1. >>> >>> +Parent node should have the following properties : >>> +- compatible: "syscon", "simple-mfd" >> >> These 2 compatibles alone are not valid. > so I should add (for example) "amlogic,meson8b-hhi-sysctrl" as first compatible? > so the result would be: compatible = "amlogic,meson8b-hhi-sysctrl", > "syscon", "simple-mfd"; yes, we used a specific compatible for GX and AXG like this. Question, what should be the order between "syscon" and "simple-mfd" ? > > > Regards > Martin >
Hi Rob, On Thu, Jul 26, 2018 at 9:32 AM Neil Armstrong <narmstrong@baylibre.com> wrote: > > Hi Rob, Martin, > > On 25/07/2018 23:16, Martin Blumenstingl wrote: > > Hi Rob, > > > > On Wed, Jul 25, 2018 at 10:07 PM Rob Herring <robh@kernel.org> wrote: > >> > >> On Sat, Jul 21, 2018 at 09:28:44PM +0200, Martin Blumenstingl wrote: > >>> The clock controller on Meson8/Meson8m2 and Meson8b is part of a > >>> register region called "HHI". This register area contains more > >>> functionality than just a clock controller: > >>> - the clock controller > >>> - some reset controller bits > >>> - temperature sensor calibration data (on Meson8b and Meson8m2 only) > >>> - HDMI controller > >>> > >>> The HHI register area may be accessed concurrently. Allow this by using > >>> a "system controller" as parent node. > >> > >> Why? A single node can be a provider of multiple things. Maybe the HDMI > >> should be a child since it will involve graph nodes, but the rest can be > >> one node. There should be numerous examples of blocks that are clock and > >> reset controllers. > > I understand that a node can provide multiple "things" - currently > > it's a clock controller and a reset controller > > the HDMI controller could also be integrated in a similar way > > > > however, I do not know how to access the temperature sensor calibration data > > there is an ADC - one of it's channel has access to a temperature sensor > > this ADC is located at CBUS offset 0x8680 and we already have a driver > > for it (meson-saradc) > > my problem is that the temperature sensor has to be calibrated - this > > is done by: > > - read data from efuse > > - write 4 bits of calibration data to some register in the ADC's register space > > - write a 5th bit of calibration data to a (seemingly random) register > > in the HHI register space > > (if one of the 5 bits is not written to it's correct location then the > > temperature sensor reads bogus values) > > > > I am not sure how to handle this without passing the HHI region to the > > meson-saradc driver and letting that initialize all the temperature > > calibration data bits (in it's own register space as well as the HHI > > register space) > > do you have any suggestion here? > > The clock controller node was badly represented from the beginning, the > HHI register zone *is* a "system control" zone with a lot of different > registers controller the system, including all the EE clocks, *some* HDMI > PHY settings, *some* Video DAC settings, *some* memory power control... > > Thus, the clock controller *is* entirely in the HHI registers space, but > the HHI register space should be used by other nodes for control some > part of the SoCs. This is why we decided to model the HHI zone as a > "syscon" to permit other node to have access to these registers and > "simple-mfd" to expose the *big* feature of the HHI : the clock controller. does our (Neil's and mine) explanation make sense to you? I understand your point that a DT node can provide multiple "things". this is the case on the Meson HHI region, but in addition to that the HHI region contains additional bits for other IP blocks. so my understanding is that using a syscon here is fine > We are migrating the Meson8 because we already migrated the GX and because the > AXG family was pushed like this (and acked like this), so it's common sense > to align the 3 families because they share the same characteristics here. > > Neil > > > > >>> > >>> Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> > >>> --- > >>> .../bindings/clock/amlogic,meson8b-clkc.txt | 13 +++++-------- > >>> 1 file changed, 5 insertions(+), 8 deletions(-) > >>> > >>> diff --git a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > >>> index b455c5aa9139..38fb979210d3 100644 > >>> --- a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > >>> +++ b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt > >>> @@ -9,15 +9,13 @@ Required Properties: > >>> - "amlogic,meson8-clkc" for Meson8 (S802) SoCs > >>> - "amlogic,meson8b-clkc" for Meson8 (S805) SoCs > >>> - "amlogic,meson8m2-clkc" for Meson8m2 (S812) SoCs > >>> -- reg: it must be composed by two tuples: > >>> - 0) physical base address of the xtal register and length of memory > >>> - mapped region. > >>> - 1) physical base address of the clock controller and length of memory > >>> - mapped region. > >>> - > >>> - #clock-cells: should be 1. > >>> - #reset-cells: should be 1. > >>> > >>> +Parent node should have the following properties : > >>> +- compatible: "syscon", "simple-mfd" > >> > >> These 2 compatibles alone are not valid. > > so I should add (for example) "amlogic,meson8b-hhi-sysctrl" as first compatible? > > so the result would be: compatible = "amlogic,meson8b-hhi-sysctrl", > > "syscon", "simple-mfd"; > > yes, we used a specific compatible for GX and AXG like this. > > Question, what should be the order between "syscon" and "simple-mfd" ? for consistency I will go with whatever the GX and AXG dt-bindings use Regards Martin
Quoting Martin Blumenstingl (2018-08-12 11:35:17) > Hi Rob, > > On Thu, Jul 26, 2018 at 9:32 AM Neil Armstrong <narmstrong@baylibre.com> wrote: > > > > Hi Rob, Martin, > > > > On 25/07/2018 23:16, Martin Blumenstingl wrote: > > > Hi Rob, > > > > > > On Wed, Jul 25, 2018 at 10:07 PM Rob Herring <robh@kernel.org> wrote: > > >> > > >> On Sat, Jul 21, 2018 at 09:28:44PM +0200, Martin Blumenstingl wrote: > > >>> The clock controller on Meson8/Meson8m2 and Meson8b is part of a > > >>> register region called "HHI". This register area contains more > > >>> functionality than just a clock controller: > > >>> - the clock controller > > >>> - some reset controller bits > > >>> - temperature sensor calibration data (on Meson8b and Meson8m2 only) > > >>> - HDMI controller > > >>> > > >>> The HHI register area may be accessed concurrently. Allow this by using > > >>> a "system controller" as parent node. > > >> > > >> Why? A single node can be a provider of multiple things. Maybe the HDMI > > >> should be a child since it will involve graph nodes, but the rest can be > > >> one node. There should be numerous examples of blocks that are clock and > > >> reset controllers. > > > I understand that a node can provide multiple "things" - currently > > > it's a clock controller and a reset controller > > > the HDMI controller could also be integrated in a similar way > > > > > > however, I do not know how to access the temperature sensor calibration data > > > there is an ADC - one of it's channel has access to a temperature sensor > > > this ADC is located at CBUS offset 0x8680 and we already have a driver > > > for it (meson-saradc) > > > my problem is that the temperature sensor has to be calibrated - this > > > is done by: > > > - read data from efuse > > > - write 4 bits of calibration data to some register in the ADC's register space > > > - write a 5th bit of calibration data to a (seemingly random) register > > > in the HHI register space > > > (if one of the 5 bits is not written to it's correct location then the > > > temperature sensor reads bogus values) > > > > > > I am not sure how to handle this without passing the HHI region to the > > > meson-saradc driver and letting that initialize all the temperature > > > calibration data bits (in it's own register space as well as the HHI > > > register space) > > > do you have any suggestion here? > > > > The clock controller node was badly represented from the beginning, the > > HHI register zone *is* a "system control" zone with a lot of different > > registers controller the system, including all the EE clocks, *some* HDMI > > PHY settings, *some* Video DAC settings, *some* memory power control... > > > > Thus, the clock controller *is* entirely in the HHI registers space, but > > the HHI register space should be used by other nodes for control some > > part of the SoCs. This is why we decided to model the HHI zone as a > > "syscon" to permit other node to have access to these registers and > > "simple-mfd" to expose the *big* feature of the HHI : the clock controller. > does our (Neil's and mine) explanation make sense to you? > I understand your point that a DT node can provide multiple "things". > this is the case on the Meson HHI region, but in addition to that the > HHI region contains additional bits for other IP blocks. so my > understanding is that using a syscon here is fine > I'm dropping this from my review queue because nobody has replied in many weeks. I think Rob wants people to move away from syscon so if you can't do that then we need good reasons. I recommend putting those reasons in the commit text and resending this series.
On Fri, 2018-08-31 at 10:39 -0700, Stephen Boyd wrote: > > > The clock controller node was badly represented from the beginning, the > > > HHI register zone *is* a "system control" zone with a lot of different > > > registers controller the system, including all the EE clocks, *some* HDMI > > > PHY settings, *some* Video DAC settings, *some* memory power control... > > > > > > Thus, the clock controller *is* entirely in the HHI registers space, but > > > the HHI register space should be used by other nodes for control some > > > part of the SoCs. This is why we decided to model the HHI zone as a > > > "syscon" to permit other node to have access to these registers and > > > "simple-mfd" to expose the *big* feature of the HHI : the clock controller. > > > > does our (Neil's and mine) explanation make sense to you? > > I understand your point that a DT node can provide multiple "things". > > this is the case on the Meson HHI region, but in addition to that the > > HHI region contains additional bits for other IP blocks. so my > > understanding is that using a syscon here is fine > > > > I'm dropping this from my review queue because nobody has replied in > many weeks. I think Rob wants people to move away from syscon so if you > can't do that then we need good reasons. I recommend putting those > reasons in the commit text and resending this series. Doubling down on what Neil and Martin already reported: When this clock controller what initially added, It was thought that the HHI register space only contained clocks and resets, which we know how to handle in a single driver and does not mandate a 'syscon' Since then, we learn that HHI has more than just clocks and reset. This is the case on every amlogic platform, which is why we already moved gxbb/gxl to syscon. axg arrived later and used this model directly. Meson8 did not mandate the change so far. We tried to avoid this mess as long as it was not absolutely necessary. This particular platform (meson8) has an ADC block with is own registers space. The ADC comes with temperature sensor which has a few compensation values. 4 of these compensation values are in the ADC register space, the 5th is apparently in the HHI... HW is what it is ... Bottom line is, HHI is a typical system controller which need to be access by several devices. We are not particularly fond of syscon either but there are still some situations were this is cleanest solution, AFAIK. Jerome
diff --git a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt index b455c5aa9139..38fb979210d3 100644 --- a/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt +++ b/Documentation/devicetree/bindings/clock/amlogic,meson8b-clkc.txt @@ -9,15 +9,13 @@ Required Properties: - "amlogic,meson8-clkc" for Meson8 (S802) SoCs - "amlogic,meson8b-clkc" for Meson8 (S805) SoCs - "amlogic,meson8m2-clkc" for Meson8m2 (S812) SoCs -- reg: it must be composed by two tuples: - 0) physical base address of the xtal register and length of memory - mapped region. - 1) physical base address of the clock controller and length of memory - mapped region. - - #clock-cells: should be 1. - #reset-cells: should be 1. +Parent node should have the following properties : +- compatible: "syscon", "simple-mfd" +- reg: base address and size of the HHI system control register space. + Each clock is assigned an identifier and client nodes can use this identifier to specify the clock which they consume. All available clocks are defined as preprocessor macros in the dt-bindings/clock/meson8b-clkc.h header and can be @@ -30,9 +28,8 @@ device tree sources). Example: Clock controller node: - clkc: clock-controller@c1104000 { + clkc: clock-controller { compatible = "amlogic,meson8b-clkc"; - reg = <0xc1108000 0x4>, <0xc1104000 0x460>; #clock-cells = <1>; #reset-cells = <1>; };
The clock controller on Meson8/Meson8m2 and Meson8b is part of a register region called "HHI". This register area contains more functionality than just a clock controller: - the clock controller - some reset controller bits - temperature sensor calibration data (on Meson8b and Meson8m2 only) - HDMI controller The HHI register area may be accessed concurrently. Allow this by using a "system controller" as parent node. Signed-off-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com> --- .../bindings/clock/amlogic,meson8b-clkc.txt | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-)