This is a cache of https://www.96boards.org/blog/bubblegum96-mainlining-update-part-2/. It is a snapshot of the page at 2024-12-22T03:56:40.421+0000.
Bubblegum96 Mainlining Update - Part 2 - 96Boards

Bubblegum96 Mainlining Update - Part 2

Manivannan Sadhasivam
|

Introduction

Hello and Welcome to the blog on “Bubblegum96 Mainlining Update - Part 2”. This blog provides the update on the Mainlining efforts from 96Boards team for the Bubblegum96 board in Linux kernel.

Reason for Mainlining Bubblegum96

Bubblegum96 board is based on Actions Semiconductor S900 64bit ARM-v8 SoC. SoC has 4 ARM Cortex-A53 cores clocked up to 1.8GHz. The clock speed itself is very unique from other A53 cores. While most of the A53 cores clocks 1.2GHz, this SoC clocks 1.8GHz. The board also has 2GB LPDDR3 RAM, 8GB on board eMMC and Power VR G6230 GPU supporting OpenGL ES 3.1, OpenGL 3.2, DirectX 10, OpenCL 1.2 EP along with other standard peripherals required by 96Boards Consumer Edition Spec.

Eventhough the SoC is powerful, the vendor support is very limited for this board. The vendor kernel is based on legacy 3.10 kernel, which is not very recent LTS kernel. For 96Boards we rely on the board/SoC manufacturer to provide the kernel support their boards. But since we get limited support from the vendor for this board, we decided to take up the upstreaming task in our spare time.

Mainlining Update

After merging the separate pinctrl and gpio drivers in the patchseries: Add Actions Semi S900 pinctrl and gpio support, we worked on adding IRQ support for GPIO driver. During this development, we realised that the pinctrl and gpio subsystems in OWL SoC’s are closely coupled and so their register sets are. Having a separate pinctrl and gpio drivers will make the life harder for future development since configuring mux for individual pins would require to disable GPIO functionality in driver.

There are API’s in kernel to do this in other way! i.e., disabling pinctrl functionality while requesting GPIO (GPIO —> Pinctrl) but no API’s in the way we wanted (Pinctrl —> GPIO). So we decided to merge GPIO functionality into the pinctrl driver and drop the previously merged GPIO driver commits. Since the patches were still in linux-nextat that time, Linus Walleij(Pinctrl/GPIO Maintainer) dropped the patches upon our request comfortably.

Then we submitted a patch series for merging the GPIO functionality into pinctrl driver.

V1 - https://lkml.org/lkml/2018/5/17/1074

As a part of the review, Christian Lamparter pointed out the gpio-ranges issue in the driver. The issue is, gpiochip_add_pin_range function usage in DT based systems are known to create issues with gpio-hog mechanism. Since, DT based pinctrl subsystem looks to add gpio-hog functionality before populating the GPIO ranges through gpiochip_add_pin_range function, it will result in -EPROBE_DEFER as the pin ranges won’t be available at that time. And since gpio-hog gets added in the gpiochip_add function, the error prohibits the board from booting. So he suggested to use gpio-ranges property in DT for specifying the pinctrl/GPIO mapping, which provides the mapping at the time of adding gpio-hog.

Finally, we sent a V2 of the driver incorporating his comments and soon Linus Walleij applied the patch series followed by Acks from Andy and Rob!

V2 - https://lkml.org/lkml/2018/5/20/5

comments powered by Disqus