On Tue, Jul 12, 2022 at 02:13:42PM -0600, Rob Herring wrote:

On Thu, Jul 07, 2022 at 06:25:39PM +0300, Serge Semin wrote:

quoted

On Wed, Jul 06, 2022 at 04:36:42PM -0600, Rob Herring wrote:

quoted

On Fri, Jun 17, 2022 at 10:37:44PM +0300, Serge Semin wrote:

quoted

On Tue, Jun 14, 2022 at 04:27:54PM -0600, Rob Herring wrote:

quoted

On Fri, Jun 10, 2022 at 11:17:55AM +0300, Serge Semin wrote:

quoted

Synopsys AHCI SATA controller is mainly compatible with the generic AHCI
SATA controller except a few peculiarities and the platform environment
requirements. In particular it can have one or two reference clocks to
feed up its AXI/AHB interface and SATA PHYs domain and at least one reset
control for the application clock domain. In addition to that the DMA
interface of each port can be tuned up to work with the predefined maximum
data chunk size. Note unlike generic AHCI controller DWC AHCI can't have
more than 8 ports. All of that is reflected in the new DWC AHCI SATA
device DT binding.

Note the DWC AHCI SATA controller DT-schema has been created in a way so
to be reused for the vendor-specific DT-schemas (see for example the
"snps,dwc-ahci" compatible string binding). One of which we are about to
introduce.

Signed-off-by: Serge Semin <redacted>

---

Changelog v2:
- Replace min/max constraints of the snps,{tx,rx}-ts-max property with
  enum [ 1, 2, 4, ..., 1024 ]. (@Rob)

Changelog v4:
- Decrease the "additionalProperties" property identation otherwise it's
  percieved as the node property instead of the key one. (@Rob)
- Use the ahci-port properties definition from the AHCI common schema
  in order to extend it with DWC AHCI SATA port properties. (@Rob)
- Remove the Hannes' rb tag since the patch content has changed.
---
 .../bindings/ata/ahci-platform.yaml           |   8 --
 .../bindings/ata/snps,dwc-ahci.yaml           | 129 ++++++++++++++++++
 2 files changed, 129 insertions(+), 8 deletions(-)
 create mode 100644 Documentation/devicetree/bindings/ata/snps,dwc-ahci.yaml

diff --git a/Documentation/devicetree/bindings/ata/ahci-platform.yaml b/Documentation/devicetree/bindings/ata/ahci-platform.yaml
index e19cf9828e68..7dc2a2e8f598 100644
--- a/Documentation/devicetree/bindings/ata/ahci-platform.yaml
+++ b/Documentation/devicetree/bindings/ata/ahci-platform.yaml

@@ -30,8 +30,6 @@ select:
           - marvell,armada-3700-ahci
           - marvell,armada-8k-ahci
           - marvell,berlin2q-ahci
-          - snps,dwc-ahci
-          - snps,spear-ahci
   required:
     - compatible
 

@@ -48,17 +46,11 @@ properties:
               - marvell,berlin2-ahci
               - marvell,berlin2q-ahci
           - const: generic-ahci
-      - items:
-          - enum:
-              - rockchip,rk3568-dwc-ahci
-          - const: snps,dwc-ahci
       - enum:
           - cavium,octeon-7130-ahci
           - hisilicon,hisi-ahci
           - ibm,476gtr-ahci
           - marvell,armada-3700-ahci
-          - snps,dwc-ahci
-          - snps,spear-ahci
 
   reg:
     minItems: 1

diff --git a/Documentation/devicetree/bindings/ata/snps,dwc-ahci.yaml b/Documentation/devicetree/bindings/ata/snps,dwc-ahci.yaml
new file mode 100644
index 000000000000..af78f6c9b857
--- /dev/null
+++ b/Documentation/devicetree/bindings/ata/snps,dwc-ahci.yaml

@@ -0,0 +1,129 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
+%YAML 1.2
+---
+$id: http://devicetree.org/schemas/ata/snps,dwc-ahci.yaml#
+$schema: http://devicetree.org/meta-schemas/core.yaml#
+
+title: Synopsys DWC AHCI SATA controller
+
+maintainers:
+  - Serge Semin <fancer.lancer@gmail.com>
+
+description:
+  This document defines device tree bindings for the Synopsys DWC
+  implementation of the AHCI SATA controller.
+
+allOf:
+  - $ref: ahci-common.yaml#
+
+properties:
+  compatible:
+    oneOf:
+      - description: Synopsys AHCI SATA-compatible devices
+        contains:
+          const: snps,dwc-ahci
+      - description: SPEAr1340 AHCI SATA device
+        const: snps,spear-ahci
+      - description: Rockhip RK3568 ahci controller
+        const: rockchip,rk3568-dwc-ahci

quoted

This is never true because there is a fallback. We should keep what we 
had before.

Could you be more specific what you meant? I don't see
"snps,spear-ahci" and "rockchip,rk3568-dwc-ahci" used with the fallback
string so modification is correct in that case.

quoted

Spear does not, just rockchip:

arch/arm64/boot/dts/rockchip/rk3568.dtsi:               compatible = "rockchip,rk3568-dwc-ahci", "snps,dwc-ahci";
arch/arm64/boot/dts/rockchip/rk356x.dtsi:               compatible = "rockchip,rk3568-dwc-ahci", "snps,dwc-ahci";
arch/arm64/boot/dts/rockchip/rk356x.dtsi:               compatible = "rockchip,rk3568-dwc-ahci", "snps,dwc-ahci";

So the 3rd entry is never true.

Then I'll have to split the schema up into two bindings:
1. snps,dwc-ahci-common.yaml: generic DW SATA AHCI properties and no "compatible"
property constraint since you said fallback was useless.
2. snps,dwc-ahci.yaml: generic DW SATA AHCI DT-schema with
competibles: ("snps,dwc-ahci"), ("snps,spear-ahci"),
("rockchip,rk3568-dwc-ahci","snps,dwc-ahci").

Are you ok with this?

Yes.

quoted

BTW if we had the fallback required the splitting up couldn't have
been needed.

We generally end up needing a split like this anyways.

Ok. I'll split it up into two schemas then.

quoted

quoted

quoted

My idea was to have the compatible strings with the required generic
fallback "snps,dwc-ahci" for all new devices thus identifying the
controller IP-core origin. But later you said "The generic IP block
fallbacks have proven to be useless." I do agree that functionally it
isn't that often used, but in some cases it can be handy for instance
to implement quirks in the generic code or use the fallback as an
additional info regarding the IP-core origin/version. So if I were you
I wouldn't be that strict about dropping the generic IP-core fallback
identifier. It's much easier to have it specified from the very
beginning than adding it after it has been declared as not required.

I wish they were useful, but experience has shown they are not.

So what to do with the generic fallback compatibles then? Please
answer to the next questions so I would correct all my currently
stashed patches in accordance with it.

1) Do you want all the new DT-binding schemas refusing to have the
fallback compatibles except for the nodes which bindings have already
been defined that way?

Yes. I wouldn't go quite as far as 'refusing'. I'm okay with a fallback 
in cases that are simple enough to actually work without platform 
specific code. As soon as the clocks, resets, phys, etc. aren't 
standard, that goes out the window. Based on experience, that pretty 
much never happens except on the IP vendor's FPGA.

quoted

2) What if a device IP-core has some versioning, but it's either
not auto-detectable at runtime or can be auto-detected but starting
from some IP-core version? Do we need it being specified in addition
to the vendor-specific compatible string?

By the time you are probing the device, you know the specific SoC and 
can just set a version variable easily. Why have a string to parse that 
doesn't work for version comparisons (e.g GT/GE/LT).

Also, what if you don't know the exact IP version? Maybe you can guess 
that it is at least at certain version based on knowing the features, so 
you set that version. Would you really want to put that guess in DT when 
later on you might need to change it?

quoted

3) The same as 2), but shall it have a generic version-less fallback
compatible string too?

If the device can function without the version specific compatible.

quoted

4) The same as 2), but what if it concerns a device which driver
relies on the versioning?

5) The same as 2), but what if it concerns the device which currently
doesn't have a driver relying on the IP-core version?

Again, let the driver set the version based on the platform specific 
compatible.

quoted

6) What if we don't have the generic fallback compatible string
required, but at some point a kernel would need it to
implement a version/IP-core-specific quirk? If we had the generic
fallback specified in dts the older systems would have been supported
out-of-box, otherwise the firmware update would also needed.

Again, when you start probing the device, you already know the specific 
platform implementation. From that, you can easily imply the IP vendor 
and version. No DT change needed.

quoted

IMO having the IP-core version + generic compatibles give many
benefits and it's much easier to have them required from the very
beginning instead of adding afterwards when then a need arises.

Certainly adding afterwards is broken. That's why we insist on SoC 
specific compatibles. Adding them when we have some platform specific 
quirk doesn't work.

Got it. Thanks for the very thorough clarification. I'll fix my patches
in accordance with the described requirements.

-Serge(y)

Rob