[RFC 0/3] iommu/samsung: Introduce Exynos sysmmu-v8 driver

From: Sam Protsenko
Date: Thu Jan 20 2022 - 15:20:18 EST


This is a draft of a new IOMMU driver used in modern Exynos SoCs (like
Exynos850) and Google's GS101 SoC (used in Pixel 6 phone). Most of its
code were taken from GS101 downstream kernel [1], with some extra
patches on top (fixes from Exynos850 downstream kernel and some porting
changes to adapt it to the mainline kernel). All development history can
be found at [2].

Similarities with existing exynos-iommu.c is minimal. I did some
analysis using similarity-tester tool:

8<-------------------------------------------------------------------->8
$ sim_c -peu -S exynos-iommu.c "|" samsung-*

exynos-iommu.c consists for 15 % of samsung-iommu.c material
exynos-iommu.c consists for 1 % of samsung-iommu-fault.c material
exynos-iommu.c consists for 3 % of samsung-iommu.h material
8<-------------------------------------------------------------------->8

So the similarity is very low, most of that code is some boilerplate
that shouldn't be extracted to common code (like allocating the memory
and requesting clocks/interrupts in probe function).

It was tested on v5.4 Android kernel on Exynos850 (E850-96 board) with
DPU use-case (displaying some graphics to the screen). Also it
apparently works fine on v5.10 GS101 kernel (on Pixel 6). On mainline
kernel I managed to build, match and bind the driver. No real world test
was done, but the changes from v5.10 (where it works fine) are minimal
(see [2] for details). So I'm pretty sure the driver is functional.

For this patch series I'd like to receive some high-level review for
driver's design and architecture. Coding style and API issues I can fix
later, when sending real (not RFC) series. Particularly I'd like to hear
some opinions about:
- namings: Kconfig option, file names, module name, compatible, etc
- modularity: should this driver be a different platform driver (like
in this series), or should it be integrated into existing
exynos-iommu.c driver somehow
- dt-bindings: does it look ok as it is, or some interface changes are
needed
- internal driver architecture: approach seems to be similar to
exynos-iommu.c, but any comments are welcome
- ongoing work: please let me know if you're aware of some efforts to
upstream this driver by some other party (e.g. Google engineers
might be working on something similar)

Basically, I want to figure out what should be changed/fixed in this
driver (on a high level), so it can be considered "upstreamable".

[1] https://android.googlesource.com/kernel/gs/
[2] https://github.com/joe-skb7/linux/commits/iommu-exynos850-dev

Sam Protsenko (3):
dt-bindings: iommu: Add bindings for samsung,sysmmu-v8
iommu/samsung: Introduce Exynos sysmmu-v8 driver
arm64: defconfig: Enable sysmmu-v8 IOMMU

.../bindings/iommu/samsung,sysmmu-v8.txt | 31 +
arch/arm64/configs/defconfig | 2 +
drivers/iommu/Kconfig | 13 +
drivers/iommu/Makefile | 3 +
drivers/iommu/samsung-iommu-fault.c | 617 +++++++
drivers/iommu/samsung-iommu-group.c | 50 +
drivers/iommu/samsung-iommu.c | 1521 +++++++++++++++++
drivers/iommu/samsung-iommu.h | 216 +++
include/dt-bindings/soc/samsung,sysmmu-v8.h | 43 +
9 files changed, 2496 insertions(+)
create mode 100644 Documentation/devicetree/bindings/iommu/samsung,sysmmu-v8.txt
create mode 100644 drivers/iommu/samsung-iommu-fault.c
create mode 100644 drivers/iommu/samsung-iommu-group.c
create mode 100644 drivers/iommu/samsung-iommu.c
create mode 100644 drivers/iommu/samsung-iommu.h
create mode 100644 include/dt-bindings/soc/samsung,sysmmu-v8.h

--
2.30.2