Enable ath10k wcn3990 wifi driver support on sdm845

Add missing optional properties in WCN3990 wifi node.

Signed-off-by Govind Singh <[email protected]>

---
 .../bindings/net/wireless/qcom,ath10k.txt          | 28 ++++++++++++++++------
 1 file changed, 21 insertions(+), 7 deletions(-)

Add missing optional properties in WCN3990 wifi node.

Signed-off-by Govind Singh <[email protected]>

---
 .../bindings/net/wireless/qcom,ath10k.txt          | 28 ++++++++++++++++------
 1 file changed, 21 insertions(+), 7 deletions(-)

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

index 7fd4e8c..f831bb1 100644

Hi,

Quoted 8 lines · Govind Singh

Add missing optional properties in WCN3990 wifi node.

Signed-off-by Govind Singh <[email protected]>

---
 .../bindings/net/wireless/qcom,ath10k.txt          | 28 
++++++++++++++++------
 1 file changed, 21 insertions(+), 7 deletions(-)

Point of order: please CC LKML on _all_ your patches. Yes, it's a firehose. CCing LKML allows your patches to be found on lore.kernel.org's patchwork and also allows people to find your patches via <https://lkml.kernel.org/r/MSG_ID> links.

Quoted 14 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

index 7fd4e8c..f831bb1 100644
@@ -37,12 +37,20 @@ Optional properties:
 - clocks: List of clock specifiers, must contain an entry for each 

required
entry in clock-names.
- clock-names: Should contain the clock names "wifi_wcss_cmd",
"wifi_wcss_ref",

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +1 -3

-               "wifi_wcss_rtc".
-- interrupts: List of interrupt lines. Must contain an entry
-             for each entry in the interrupt-names property.
+              "wifi_wcss_rtc" for "qcom,ipq4019-wifi" compatible 

Quoted 4 lines

target and
+ "cxo_ref_clk_pin", "smmu_aggre2_noc_clk" for
"qcom,wcn3990-wifi"
+ compatible target.

I always get confused with these big bindings patches that hide everything under a big "Optional properties" section to avoid specifying which properties are actually optional and which ones are required.

After your patch and thinking about "qcom,wcn3990-wifi" in particular, it's unclear to me which of the following is true (or maybe something totally different I didn't think of)

A) On wcn3990-wifi these clocks should be a required property but it's only listed under "Optional" because it's not used for some of the different WiFi drivers using this same bindings.

B) On wcn3990-wifi these clocks should either both be there or neither.

C) On wcm3990-wifi you can specify zero, either, or both of these clocks. AKA they are independently optional.

It might make sense to reorganize this bindings to make this clearer? ...not just for clock but for interrupts / regulators as well. Maybe you need to break this down into sections per class of compatible string, or add a list per compatible string down below?

Also: even stranger is that even though you list two clocks here the current driver I see in linuxnext only has "cxo_ref_clk_pin".

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt Optional properties: · line 48

Quoted 9 lines

+- interrupts: reference to the list of 17 interrupt no's for "qcom,ipq4019-wifi"

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +2 -0

+             compatible target.
+             reference to the list of 12 interrupt no's for 

"qcom,wcn3990-wifi"

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +3 -0

+             compatible target.
+             Must contain interrupt-names property per entry for
+             "qcom,ath10k", "qcom,ipq4019-wifi" compatible targets.

...and just to add some credence to my concerns above, "interrupts" are currently listed under "Optional" properties but I don't think that the wcn3990 driver will actually work if you don't specify any of the interrupts, right? AKA for wcn3990 they are _not_ optional and you must have exactly 12 interrupts.

One separate issue I have is with your example, which you didn't change in this patch. You should fix the example with the same feedback that I had to your patch ("dts: arm64/sdm845: Add WCN3990 WLAN module device node").

-Doug

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

index 7fd4e8c..f831bb1 100644

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +8 -1

+             compatible target.
+             reference to the list of 12 interrupt no's for "qcom,wcn3990-wifi"
+             compatible target.
+             Must contain interrupt-names property per entry for
+             "qcom,ath10k", "qcom,ipq4019-wifi" compatible targets.
+
 - interrupt-names: Must include the entries for MSI interrupt
                   names ("msi0" to "msi15") and legacy interrupt
-                  name ("legacy"),
+                  name ("legacy") for "qcom,ath10k", "qcom,ipq4019-wifi"
+                  compatible targets.
 - qcom,msi_addr: MSI interrupt address.
 - qcom,msi_base: Base value to add before writing MSI data into
                MSI address register.

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +4 -2

 Example (to supply the calibration data alone):
 

@@ -133,8 +142,10 @@ wifi@18000000 {
                compatible = "qcom,wcn3990-wifi";
                reg = <0x18800000 0x800000>;
                reg-names = "membase";
-               clocks = <&clock_gcc clk_aggre2_noc_clk>;
-               clock-names = "smmu_aggre2_noc_clk"
+               clocks = <&clock_gcc clk_aggre2_noc_clk>,
+                        <&clock_gcc clk_rf_clk2_pin>;
+               clock-names = "smmu_aggre2_noc_clk",
+                             "cxo_ref_clk_pin";
                interrupts =
                           <0 130 0 /* CE0 */ >,
                           <0 131 0 /* CE1 */ >,

Quoting Govind Singh (2018-10-10 04:52:54)

Quoted 2 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt Optional properties: · line 67

Quoted 7 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +2 -1

@@ -55,7 +63,8 @@ Optional properties:
 - qcom,ath10k-pre-calibration-data : pre calibration data as an array,
                                     the length can vary between hw versions.
 - <supply-name>-supply: handle to the regulator device tree node
-                          optional "supply-name" is "vdd-0.8-cx-mx".
+                          optional "supply-name" are "vdd-0.8-cx-mx",
+                          "vdd-1.8-xo", "vdd-1.3-rfa" and "vdd-3.3-ch0".

Why can't the wifi firmware manage these regulators itself?

And these names don't seem to match any sort of schematic or really

describe what this device is. From what I can tell, we've combined the

off-SoC wifi module with the on-SoC wifi I/O space into one DT node here and then relied on that node to make some driver probe in the kernel to do wifi stuff. Can we model this properly by actually showing that

there's something in the SoC, and there's something outside the SoC, and these two things are connected by having two nodes and a phandle between them?

Quoted 16 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +4 -2

 Example (to supply the calibration data alone):
 

@@ -133,8 +142,10 @@ wifi@18000000 {
                compatible = "qcom,wcn3990-wifi";
                reg = <0x18800000 0x800000>;
                reg-names = "membase";
-               clocks = <&clock_gcc clk_aggre2_noc_clk>;
-               clock-names = "smmu_aggre2_noc_clk"
+               clocks = <&clock_gcc clk_aggre2_noc_clk>,
+                        <&clock_gcc clk_rf_clk2_pin>;
+               clock-names = "smmu_aggre2_noc_clk",
+                             "cxo_ref_clk_pin";
                interrupts =
                           <0 130 0 /* CE0 */ >,
                           <0 131 0 /* CE1 */ >,

Hi Stephen and Govind,

I was chatting with Govind, and he seemed to be a bit stalled on this. I'd encourage him to reply with whatever knowledge he has, because it's a bit hard to give definitive answers when I don't know all the inner workings here. (In fact, you, Stephen, probably know more than me about how Qualcomm MSM chips work.)

First of all, I'll explain the limited bits I do know, and see how they fit in below. I may be wrong.

WCN3990 is an external module, for supporting BT and Wifi RF components. It has a host interface for the Wifi, but it's not something the host knows how to talk to directly at all -- it's an "Analog IQ" interface, between an internal SoC MAC and PHY processor. Instead, we talk to this module via the System NOC (?). So while there are basically 2 distinct hardware components (on-SoC coprocessors, various internal communication buses, various shared memory regions, etc.; and the external WCN3990 module), there is almost no way for the main processor to "talk" to the WCN3990 directly.

Another data point to throw into the mix: WCN3990 can apparently be used on multiple different SoCs -- I see public announcments about SDM835 (and 660?), and I'm currently playing with it on SDM845. So perhaps there is some value in understanding "WCN3990" as being distinct from "WiFi MAC/PHY hardware and communication logic found on a MSM SoC." But they do seem to be very tightly coupled...

Side note: there is also a BT component on the WCN3990 module, and we *can* talk to that directly over UART. There's a separate binding going in for that, and it's a much clearer separation to divide "UART controller" and "BT/UART interface on WCN3990."

Quoted 4 lines

Quoting Govind Singh (2018-10-10 04:52:54)

Quoted 2 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

Quoted 9 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt Optional properties: · line 67

Quoted 7 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +2 -1

@@ -55,7 +63,8 @@ Optional properties:
 - qcom,ath10k-pre-calibration-data : pre calibration data as an array,
                                     the length can vary between hw versions.
 - <supply-name>-supply: handle to the regulator device tree node
-                          optional "supply-name" is "vdd-0.8-cx-mx".
+                          optional "supply-name" are "vdd-0.8-cx-mx",
+                          "vdd-1.8-xo", "vdd-1.3-rfa" and "vdd-3.3-ch0".

Why can't the wifi firmware manage these regulators itself?

In my understanding:
At least 3 of those (all the supplies Govind is adding) are external pins on the RF module. Why do you assume these are something the firmware can control? In the schematics I'm looking at, this seems to be connected to a PMIC. While it's certainly possible this is something the Q6 processor (running modem and Wifi firmware) can control, it doesn't seem obvious to me that they *must* be able to.

So I guess I'd say: why not represent these regulators in the device tree? It seems like a valid hardware description (like I said, 3 power rail pins on an external module).

Now, your *next* paragraph might have hairier points :)

Quoted 1 line

And these names don't seem to match any sort of schematic or really

Which names? I see these pin names on the WCN3990 datasheets I see:

VDD18_IO
VDD18_XO
VDD33_CH0
VDD33_CH1
VDD13_RFA

3 of those match the 3 Govind is adding, and they appear to line up with what I see in schematics.

Quoted 2 lines

describe what this device is. From what I can tell, we've combined the

I've described "waht the device is" above to the best of my knowledge. If you're just looking at schematics, you might possibly be thrown by the fact that WCN3990 is packaged in a module provided by other vendors, so it might be labeled by that vendor on the schematic, not by the Qualcomm WCN3990 name.

Quoted 4 lines

there's something in the SoC, and there's something outside the SoC, and these two things are connected by having two nodes and a phandle between them?

Why phandle? Under what bus would you place the WCN3990? As per my description above, there is really no way to talk to it directly. So if anything, it seems like it would be a subnode of the node we're describing here.

In favor of your separation though: there are many ways to utilize WCN3990 apparently, and I'd imagine the binding might change a bit depending on the SoC (e.g., different clocks?). So there might be value in describing the SDM{835,845,...whatever}-wifi-soc-components vs. external WCN3990. Additionally, I don't know if it's even possible to utilize a different RF module with the same SoC (is there a possibility of a, e.g., WCN3991 that can use the same SoC logic?).

But I'm still not totally convinced that splitting this up really makes much sense. Is there other precedent for splitting out a separate node for something that we don't talk to at all (no digital interface)? Or do we just need a more accurate compatible property, that describes the fact that this is a SDM845+WCN3990 combination? The only thing that software would ever do with the separate node is look it up to find the regulators to power it on.

Quoting Brian Norris (2018-11-02 11:43:17)

Quoted 28 lines

Hi Stephen and Govind,

I was chatting with Govind, and he seemed to be a bit stalled on this. I'd encourage him to reply with whatever knowledge he has, because it's a bit hard to give definitive answers when I don't know all the inner workings here. (In fact, you, Stephen, probably know more than me about how Qualcomm MSM chips work.)

First of all, I'll explain the limited bits I do know, and see how they fit in below. I may be wrong.

WCN3990 is an external module, for supporting BT and Wifi RF components. It has a host interface for the Wifi, but it's not something the host knows how to talk to directly at all -- it's an "Analog IQ" interface, between an internal SoC MAC and PHY processor. Instead, we talk to this module via the System NOC (?). So while there are basically 2 distinct hardware components (on-SoC coprocessors, various internal communication buses, various shared memory regions, etc.; and the external WCN3990 module), there is almost no way for the main processor to "talk" to the WCN3990 directly.

Another data point to throw into the mix: WCN3990 can apparently be used on multiple different SoCs -- I see public announcments about SDM835 (and 660?), and I'm currently playing with it on SDM845. So perhaps there is some value in understanding "WCN3990" as being distinct from "WiFi MAC/PHY hardware and communication logic found on a MSM SoC." But they do seem to be very tightly coupled...

Quoted 4 lines

Side note: there is also a BT component on the WCN3990 module, and we *can* talk to that directly over UART. There's a separate binding going in for that, and it's a much clearer separation to divide "UART controller" and "BT/UART interface on WCN3990."

Thanks for the background. I wasn't aware of much about this driver but taking this information and skimming the driver makes my mental model believe that the register space here is a custom I/O region in the SoC used to read/write to the BT/WiFi chip that's outside the SoC. Similar to i2c, SPI, or even PCI, in the way that the I/O region in the SoC is essentially the controller that is connected to the external chip. It's like the hardware engineers stuck the PCI hardware blob inside the SoC and then made special pins and wires for the thing the PCI device used to do internally. Or is that completely wrong still?

Quoted 23 lines · Stephen Boyd

Quoted 4 lines

Quoting Govind Singh (2018-10-10 04:52:54)

Quoted 2 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +0 -0

Quoted 9 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt Optional properties: · line 67

Quoted 7 lines

Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt +2 -1

@@ -55,7 +63,8 @@ Optional properties:
 - qcom,ath10k-pre-calibration-data : pre calibration data as an array,
                                     the length can vary between hw versions.
 - <supply-name>-supply: handle to the regulator device tree node
-                          optional "supply-name" is "vdd-0.8-cx-mx".
+                          optional "supply-name" are "vdd-0.8-cx-mx",
+                          "vdd-1.8-xo", "vdd-1.3-rfa" and "vdd-3.3-ch0".

Why can't the wifi firmware manage these regulators itself?

In my understanding:
At least 3 of those (all the supplies Govind is adding) are external pins on the RF module. Why do you assume these are something the firmware can control? In the schematics I'm looking at, this seems to be connected to a PMIC. While it's certainly possible this is something the Q6 processor (running modem and Wifi firmware) can control, it doesn't seem obvious to me that they *must* be able to.

Quoted 3 lines

So I guess I'd say: why not represent these regulators in the device tree? It seems like a valid hardware description (like I said, 3 power rail pins on an external module).

Agreed. I want to specify the regulators in DT. I'm mostly asking if there is firmware that runs and can control these regulators itself. If so, then I'm lost why that firmware can't manage these itself and let us ignore these requirements and never specify them in DT. Presumably there isn't firmware that can manage it?

Quoted 12 lines

Now, your *next* paragraph might have hairier points :)

Quoted 1 line

And these names don't seem to match any sort of schematic or really

Which names? I see these pin names on the WCN3990 datasheets I see:

VDD18_IO
VDD18_XO
VDD33_CH0
VDD33_CH1
VDD13_RFA

Quoted 2 lines

3 of those match the 3 Govind is adding, and they appear to line up with what I see in schematics.

Yes those three match up, but the other one, vdd-0.8-cx-mx, looks like the CX and MX power supplies that are typical of Qualcomm designs so maybe those regulators are used for the I/O region inside the SoC instead of the off-SoC chip?

Quoted 7 lines

Quoted 1 line

describe what this device is. From what I can tell, we've combined the

I've described "waht the device is" above to the best of my knowledge. If you're just looking at schematics, you might possibly be thrown by the fact that WCN3990 is packaged in a module provided by other vendors, so it might be labeled by that vendor on the schematic, not by the Qualcomm WCN3990 name.

Thanks!

Quoted 5 lines

Quoted 4 lines

there's something in the SoC, and there's something outside the SoC, and these two things are connected by having two nodes and a phandle between them?

Quoted 4 lines

Why phandle? Under what bus would you place the WCN3990? As per my description above, there is really no way to talk to it directly. So if anything, it seems like it would be a subnode of the node we're describing here.

I'm not tied to having a phandle at all. I'm mostly trying to make the following distinction in DT. If a node is under the soc node, it has a reg property and represents a hardware block inside the SoC. Any

regulators or clocks that the node uses should also either be provided inside the SoC, or be pins on the SoC that the device can be connected to. Otherwise, if those regulators aren't going to pins for the SoC,

then it means we have a mash-up of two devices in one place in the DT hierarchy. This is what doesn't look proper to me, and it's why we would want to split the node into two nodes, the SoC part and the module part for the off-SoC WCN chip.

And yes, from what you've told me here it would make sense to make the WCN chip a subnode of this SoC node instead of a phandle connecting the two.

Quoted 7 lines

In favor of your separation though: there are many ways to utilize WCN3990 apparently, and I'd imagine the binding might change a bit depending on the SoC (e.g., different clocks?). So there might be value in describing the SDM{835,845,...whatever}-wifi-soc-components vs. external WCN3990. Additionally, I don't know if it's even possible to utilize a different RF module with the same SoC (is there a possibility of a, e.g., WCN3991 that can use the same SoC logic?).

Sure. Does it matter though? Even one-off solutions would be better off if we described what's going on at the board-level so that it isn't confusing to readers of the schematic and the dts file.

Plus, it would
allow the module creator to deliver a dts file for the module without having to involve the SoC bits and clearly split things so that the SoC dts file can be written without board assumptions.

Quoted 8 lines

But I'm still not totally convinced that splitting this up really makes much sense. Is there other precedent for splitting out a separate node for something that we don't talk to at all (no digital interface)? Or do we just need a more accurate compatible property, that describes the fact that this is a SDM845+WCN3990 combination? The only thing that software would ever do with the separate node is look it up to find the regulators to power it on.

Agreed, it may seem like overkill, but I'll argue that this part allows us to easily see where the division of clocks and regulators is, instead of having to mentally untangle the external module from the internal SoC bits. I started to compare the AHB and the SNOC bus types in the ath10k driver but then I decided they were just a little bit different from each other to where this split wouldn't help that code.

So like you say, if in the future the same SNOC hardware block will be used with a different WCN chip that has different clock and power requirements it would be very easy to integrate that by writing a new sub-device node and driver and doing this split. I admit that there would be some new work in the ath10k driver to support sub-device drivers that the bus layer communicates with and it may conflict with the way the PCI designs are implemented, but I would still argue this is better from a DT implementer perspective because we can see what things are on the board and what things are in the SoC.

Quoted 3 lines · Stephen Boyd

And yes, from what you've told me here it would make sense to make the WCN chip a subnode of this SoC node instead of a phandle connecting the two.

I could begrudgingly agree with that.

	wifi: wifi@18800000 {
		compatible = "qcom,sdm845-wifi";
		reg = <...>
		clocks = <...>
		vdd-0.8-cx-mx-supply = <...>
		... interrupts, etc. ...

		rf { // I don't know what to call this node. Suggestions
		     // welcome.
			compatible = "qcom,wcn3990-wifi";
			vdd-1.8-xo-supply = <...>;
			vdd-1.3-rfa-supply = <...>;
			vdd-3.3-ch0-supply = <...>;
		};
	};

+ wlan_msa_mem: reserved-memory@96700000 {

arch/arm64/boot/dts/qcom/sdm845-mtp.dts +11 -0

+               };
        };
 
        cpus {

@@ -1403,5 +1408,26 @@
                                status = "disabled";
                        };
                };
+
+               wifi: wifi@18800000 {
+                       compatible = "qcom,wcn3990-wifi";
+                       status = "disabled";
+                       reg = <0x18800000 0x800000>;
+                       reg-names = "membase";
+                       memory-region = <&wlan_msa_mem>;
+                       interrupts =
+                               <0 413 0 /* CE0 */ >,
+                               <0 414 0 /* CE1 */ >,

arch/arm64/boot/dts/qcom/sdm845.dtsi +11 -0

+                               <0 414 0 /* CE1 */ >,
+                               <0 415 0 /* CE2 */ >,
+                               <0 416 0 /* CE3 */ >,
+                               <0 417 0 /* CE4 */ >,
+                               <0 418 0 /* CE5 */ >,
+                               <0 420 0 /* CE6 */ >,
+                               <0 421 0 /* CE7 */ >,
+                               <0 422 0 /* CE8 */ >,
+                               <0 423 0 /* CE9 */ >,
+                               <0 424 0 /* CE10 */ >,
+                               <0 425 0 /* CE11 */ >;

WCN3990 wifi module can optionally make use of the IOMMU. Add binding documentation for phandle to the IOMMU and the stream id of wifi iommu block.

Signed-off-by Govind Singh <[email protected]>

---
 Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt | 5 +++++
 1 file changed, 5 insertions(+)

This series enables ath10k wifi driver support for WCN3990 target on sdm845 SOC. This series also updates the missing dt binding documentation and adds optional iommu property.