[PATCH 0/3] spi: Better specification for rx-sample-delay-ns and core parsing

[Frieder Schrempf]

"rx-sample-delay-ns" has been a generic SPI peripheral property since it
was moved to spi-peripheral-props.yaml, but the core has never looked at
it, and what it is meant to describe has become blurred along the way.

It was introduced in 2015 for Rockchip, to compensate "boards with
high-capacitance SPI lines". The wording now in the generic schema came
later, from a description of a DesignWare register, and lost its
controller-specific qualifier on the way. It says what the value does to
the controller, but not what a board should measure to arrive at it.

Patch 1 says what the property describes. Patch 2 parses it in the core,
so that a controller driver can act on it without open-coding the
property name, and patch 3 converts spi-dw, its only user that reads it
from the peripheral node as the binding intends.

Deliberately not converted: spi-rockchip and spi-mtk-snfi read
"rx-sample-delay-ns" from the *controller* node rather than the
peripheral node, which contradicts the binding but is what their device
trees rely on. Converting them would break those boards, so they keep
their private parsing. The controller-wide default that spi-dw reads
from its own node is left in place for the same reason - it lives in a
different node, which is not what the core parses when it looks at a
peripheral.

The one behavioural corner is in patch 3 and is called out there: an
explicit "rx-sample-delay-ns = <0>" on a peripheral is now
indistinguishable from an absent property, so it no longer overrides a
non-zero controller-level default. There are two users of the property
in the tree and neither does this.

This is groundwork for letting SPI devices declare their datasheet
clock-to-output-valid time so that controllers can move their sampling
point instead of forcing a lower spi-max-frequency, posted as an RFC at

  https://lore.kernel.org/r/20260303-fsl-qspi-rx-sampling-delay-v1-0-9326bbc492d6@xxxxxxxxxx

Nothing in that work is needed to read this series, and nothing here
depends on it: the chip side is a separate quantity that composes with
this one, which is why patch 1 spends a paragraph on keeping them apart.

Assisted-by: Claude:claude-opus-5
Signed-off-by: Frieder Schrempf <frieder.schrempf@xxxxxxxxxx>
---
Frieder Schrempf (3):
      dt-bindings: spi: Clarify what rx-sample-delay-ns describes
      spi: Parse the rx-sample-delay-ns peripheral property in the core
      spi: dw: Use the rx-sample-delay-ns value parsed by the core

 .../devicetree/bindings/spi/spi-peripheral-props.yaml       |  8 ++++++++
 drivers/spi/spi-dw-core.c                                   | 13 +++++++------
 drivers/spi/spi.c                                           |  9 +++++++++
 include/linux/spi/spi.h                                     |  9 +++++++++
 4 files changed, 33 insertions(+), 6 deletions(-)
---
base-commit: 238650ef6c7c7cca08e032527329424c9fbd70e5
change-id: 20260917-spi-sample-delay-cleanup-ab1d03ae02fa

Best regards,
--  
Frieder Schrempf <frieder.schrempf@xxxxxxxxxx>