On Thu, Apr 13, 2023 at 03:05:26PM -0700, Stephen Boyd wrote:
> Quoting Andrew Halaney (2023-04-13 14:01:27)
> > On Thu, Apr 13, 2023 at 01:47:19PM -0700, Stephen Boyd wrote:
> > > Quoting Andrew Halaney (2023-04-13 12:15:41)
> > > > arch/arm64/boot/dts/qcom/sa8540p-ride.dts | 179 ++++++++++++++++++++++
> > > > 1 file changed, 179 insertions(+)
> > > >
> > > > diff --git a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
> > > > index 40db5aa0803c..650cd54f418e 100644
> > > > --- a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
> > > > +++ b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
> > > > @@ -28,6 +28,65 @@ aliases {
> > > > chosen {
> > > > stdout-path = "serial0:115200n8";
> > > > };
> > > > +
> > > > + mtl_rx_setup: rx-queues-config {
> > >
> > > Is there a reason why this isn't a child of an ethernet node?
> > >
> > >
> >
> > I debated if it was more appropriate to:
> >
> > 1. make a duplicate in each ethernet node (ethernet0/1)
> > 2. Put it in one and reference from both
> > 3. have it floating around independent like this, similar to what is
> > done in sa8155p-adp.dts[0]
> >
> > I chose 3 as it seemed cleanest, but if there's a good argument for a
> > different approach I'm all ears!
>
> I wonder if it allows the binding checker to catch bad properties by
> having it under the ethernet node? That's the only thing I can think of
> that may be improved, but I'll let binding reviewers comment here.
>
Thanks, I was curious so I played around to answer the question via
testing, and you're right... rx-queues-config/tx-queues-config aren't
evaluated unless they sit under the node with the compatible (i.e. it
doesn't just follow the phandle and evaluate). That makes sense to me I
suppose.
So, I guess, would maintainers prefer to see option (1) or (2) above? I
want that thing evaluated.
Option 1., above, has duplicated configuration, but is probably a more accurate
representation of the hardware description.
Option 2., above, doesn't duplicate rx-queues-config/tx-queues-config,
but is a weirder representation of hardware description, and only
complains once (which is fine since it's shared) when the binding checker
runs (i.e. only the etherent parent containing rx-queues-config yells).
In the below example you can see what I mean by the "only complains
once" comment as well as illustration that the patchset as is doesn't
allow rx-queues-config/tx-queues-config to be validated by dt-binding
checks:
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % # Purposely introduce a dt-binding error on top of the current patchset :(
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % git diff :(
diff --git a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
index 650cd54f418e..ecb0000db4e2 100644
--- a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
+++ b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
@@ -54,7 +54,7 @@ queue2 {
queue3 {
snps,avb-algorithm;
- snps,map-to-dma-channel = <0x3>;
+ snps,map-to-dma-channel = "not-correct";
snps,priority = <0xc>;
};
};
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % make CHECK_DTBS=y DT_SCHEMA_FILES=/net/qcom,ethqos.yaml qcom/sa8540p-ride.dtb
DTC_CHK arch/arm64/boot/dts/qcom/sa8540p-ride.dtb
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % # That should have failed
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % # Move the whole node under ethernet0, have ethernet1 reference via phandle only still
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % git diff | cat
diff --git a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
index 650cd54f418e..451246936731 100644
--- a/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
+++ b/arch/arm64/boot/dts/qcom/sa8540p-ride.dts
@@ -29,35 +29,6 @@ chosen {
stdout-path = "serial0:115200n8";
};
- mtl_rx_setup: rx-queues-config {
- snps,rx-queues-to-use = <1>;
- snps,rx-sched-sp;
-
- queue0 {
- snps,dcb-algorithm;
- snps,map-to-dma-channel = <0x0>;
- snps,route-up;
- snps,priority = <0x1>;
- };
-
- queue1 {
- snps,dcb-algorithm;
- snps,map-to-dma-channel = <0x1>;
- snps,route-ptp;
- };
-
- queue2 {
- snps,avb-algorithm;
- snps,map-to-dma-channel = <0x2>;
- snps,route-avcp;
- };
-
- queue3 {
- snps,avb-algorithm;
- snps,map-to-dma-channel = <0x3>;
- snps,priority = <0xc>;
- };
- };
mtl_tx_setup: tx-queues-config {
snps,tx-queues-to-use = <1>;
@@ -223,6 +194,36 @@ ðernet0 {
status = "okay";
+ mtl_rx_setup: rx-queues-config {
+ snps,rx-queues-to-use = <1>;
+ snps,rx-sched-sp;
+
+ queue0 {
+ snps,dcb-algorithm;
+ snps,map-to-dma-channel = <0x0>;
+ snps,route-up;
+ snps,priority = <0x1>;
+ };
+
+ queue1 {
+ snps,dcb-algorithm;
+ snps,map-to-dma-channel = <0x1>;
+ snps,route-ptp;
+ };
+
+ queue2 {
+ snps,avb-algorithm;
+ snps,map-to-dma-channel = <0x2>;
+ snps,route-avcp;
+ };
+
+ queue3 {
+ snps,avb-algorithm;
+ snps,map-to-dma-channel = "not-correct";
+ snps,priority = <0xc>;
+ };
+ };
+
mdio {
compatible = "snps,dwmac-mdio";
#address-cells = <1>;
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % make CHECK_DTBS=y DT_SCHEMA_FILES=/net/qcom,ethqos.yaml qcom/sa8540p-ride.dtb
DTC_CHK arch/arm64/boot/dts/qcom/sa8540p-ride.dtb
/home/ahalaney/git/redhat/stmmac/arch/arm64/boot/dts/qcom/sa8540p-ride.dtb: ethernet@20000: rx-queues-config:queue3:snps,map-to-dma-channel:0: [1852797997, 1668248178, 1701016576] is too long
From schema: /home/ahalaney/git/redhat/stmmac/Documentation/devicetree/bindings/net/qcom,ethqos.yaml
/home/ahalaney/git/redhat/stmmac/arch/arm64/boot/dts/qcom/sa8540p-ride.dtb: ethernet@20000: Unevaluated properties are not allowed ('max-speed', 'mdio', 'phy-handle', 'phy-mode', 'power-domains', 'rx-fifo-depth', 'rx-queues-config', 'snps,mtl-rx-config', 'snps,mtl-tx-config', 'snps,pbl', 'snps,tso', 'tx-fifo-depth' were unexpected)
From schema: /home/ahalaney/git/redhat/stmmac/Documentation/devicetree/bindings/net/qcom,ethqos.yaml
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] % # That warned as expected, since snps,dwmac.yaml failed on rx-queues-config it warns, \
and since part of the schema failed its not inherited (hence the Unevaluated properties warning following) \
also note how only ethernet0 (@20000) is evaluating rx-queues-config since that's where the rx-queues-config node lives
(dtschema) ahalaney@halaney-x13s ~/git/redhat/stmmac (git)-[stmmac] %
Thanks for the review!
- Andrew
