Messages in this thread | | | From | Théo Lebrun <> | Subject | [PATCH 00/13] Add Mobileye EyeQ5 support to the Nomadik I2C controller & use hrtimers for timeouts | Date | Thu, 15 Feb 2024 17:52:07 +0100 |
| |
Hi,
This series adds two tangent features to the Nomadik I2C controller:
- Add a new compatible to support Mobileye EyeQ5 which uses the same IP block as Nomadik.
It has two quirks to be handled: - The memory bus only supports 32-bit accesses. A writeb() is used which we avoid. - We must write a value into a shared register region (OLB, "Other Logic Block") depending on the I2C bus speed.
- Allow xfer timeouts below one jiffy by using a workqueue and hrtimers instead of a completion.
The situation to be addressed is: - Many devices on the same I2C bus. - One xfer to each device is sent at regular interval. - One device gets stuck and does not answer. - With long timeouts, following devices won't get their message. A shorter timeout ensures we can still talk to the following devices.
This clashes a bit with the current i2c_adapter timeout field that stores a jiffies amount. We cannot rely on it and therefore we take a value from devicetree as a µs value. If the timeout is less than a jiffy duration, we switch from standard jiffies timeout to hrtimers.
There is one patch targeting a hwmon dt-bindings file: Documentation/devicetree/bindings/hwmon/lm75.yaml. The rest is touching the I2C bus driver, its bindings and platform devicetrees.
About dependencies: - The series is based upon v6.8-rc4. - For testing on EyeQ5 hardware and devicetree patches, we need the base platform series from Grégory [0]. - The last commit (adding DT phandles for resets), we need the syscon series [1] that provides the reset controller node.
I think there are discussions to be had about:
- The handling of timeouts. Having a non-jiffy value is not driver specific. Should this change be done at the subsystem layer? The subsystem could even fetch the value from devicetree and auto-fill timeout, with a default given by the driver. Not many drivers seem to use the i2c_adapter timeout field from my quick grepping.
- The DT prop for timeout. I've picked "timeout-usecs". Some drivers use vendor prefixes, but this is not vendor-specific and only a software implementation detail.
- The shape of this series. Initially it was split in two. However I brought them together as they cannot be applied independently. Please tell me if a better approach is to be preferred.
Those are thoughts, I'm sure people will have feedback on this.
Have a nice day, Théo Lebrun
[0]: https://lore.kernel.org/lkml/20240205153503.574468-1-gregory.clement@bootlin.com/ [1]: https://lore.kernel.org/lkml/20240212-mbly-clk-v6-0-c46fa1f93839@bootlin.com/
Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> --- Théo Lebrun (13): dt-bindings: i2c: nomadik: add timeout-usecs property bindings dt-bindings: i2c: nomadik: add mobileye,eyeq5-i2c bindings and example dt-bindings: hwmon: lm75: add label property i2c: nomadik: rename private struct pointers from dev to priv i2c: nomadik: simplify IRQ masking logic i2c: nomadik: use bitops helpers i2c: nomadik: support short xfer timeouts using waitqueue & hrtimer i2c: nomadik: replace jiffies by ktime for FIFO flushing timeout i2c: nomadik: fetch timeout-usecs property from devicetree i2c: nomadik: support Mobileye EyeQ5 I2C controller MIPS: mobileye: eyeq5: add 5 I2C controller nodes MIPS: mobileye: eyeq5: add evaluation board I2C temp sensor MIPS: mobileye: eyeq5: add resets to I2C controllers
Documentation/devicetree/bindings/hwmon/lm75.yaml | 4 + .../devicetree/bindings/i2c/st,nomadik-i2c.yaml | 49 +- arch/mips/boot/dts/mobileye/eyeq5-epm5.dts | 8 + arch/mips/boot/dts/mobileye/eyeq5.dtsi | 75 +++ drivers/i2c/busses/i2c-nomadik.c | 710 ++++++++++++--------- 5 files changed, 534 insertions(+), 312 deletions(-) --- base-commit: d55aa725e32849f709b61eab3b7a50b810a71a84 change-id: 20231023-mbly-i2c-7c2fbbb1299f
Best regards, -- Théo Lebrun <theo.lebrun@bootlin.com>
| |