[PATCH v6 0/7] MediaTek DVFSRC Bus Bandwidth and Regulator knobs
Rob Herring
robh at kernel.org
Fri Oct 11 15:15:05 PDT 2024
On Mon, Jun 10, 2024 at 3:57 AM AngeloGioacchino Del Regno
<angelogioacchino.delregno at collabora.com> wrote:
>
> Changes in v6:
> - Fixed build with clang (thanks Nathan!)
> - Removed unused mtk_rmw() macro in mtk-dvfsrc.c
> - Added MODULE_DESCRIPTION() to mtk-dvfsrc-regulator.c
>
> Changes in v5:
> - Fixed Kconfig dependencies in interconnect
> - Fixed module build for dvfsrc and interconnect
>
> Changes in v4:
> - Updated patch [3/7] to actually remove address/size cells
> as the old version got unexpectedly pushed in v3.
>
> Changes in v3:
> - Removed examples from interconnect and regulator bindings
> and kept example node with interconnect and regulator in
> the main DVFSRC binding as suggested
> - Removed 'reg' from interconnect and regulator, removed both
> address and size cells from the main DVFSRC binding as that
> was not really needed
> - Added anyOf-required entries in the regulator binding as it
> doesn't make sense to probe it without any regulator subnode
>
> Changes in v2:
> - Fixed issues with regulator binding about useless quotes and
> wrong binding path (oops)
> - Removed useless 'items' from DVFSRC main binding
> - Allowed address/size cells to DVFSRC main binding to resolve
> validation issues on the regulator and interconnect bindings
> - Changed dvfsrc node name to `system-controller`, as the DVFSRC
> is actually able to control multiple system components.
> - Added a commit to remove mtk-dvfs-regulator.c before adding the
> new, refactored regulator driver
>
>
> This series adds support for the MediaTek Dynamic Voltage and Frequency
> Scaling Resource Controller (DVFSRC), found on many MediaTek SoCs.
>
> This hardware collects requests from both software and the various remote
> processors embededd into the SoC, and decides about a minimum operating
> voltage and a minimum DRAM frequency to fulfill those requests, in an
> effort to provide the best achievable performance per watt.
>
> Such hardware IP is capable of transparently performing direct register
> R/W on all of the DVFSRC-controlled regulators and SoC bandwidth knobs.
>
> Summarizing how the DVFSRC works for Interconnect:
>
> ICC provider ICC Nodes
> ---- ----
> _________ |CPU | |--- |VPU |
> _____ | |----- ---- | ----
> | |->| DRAM | ---- | ----
> |DRAM |->|scheduler|----- |GPU | |--- |DISP|
> | |->| (EMI) | ---- | ----
> |_____|->|_________|---. ----- | ----
> /|\ `-|MMSYS|--|--- |VDEC|
> | ----- | ----
> | | ----
> | change DRAM freq |--- |VENC|
> -------- | ----
> SMC --> | DVFSRC | | ----
> -------- |--- |IMG |
> | ----
> | ----
> |--- |CAM |
> ----
>
> ...and for regulators, it's simply...
> SMC -> DVFSRC -> Regulator voltage decider -> (vreg) Registers R/W
>
> Please note that this series is based on an old (abandoned) series from
> MediaTek [1], and reuses some parts of the code found in that.
>
> Besides, included in this series, there's also a refactoring of the
> mtk-dvfsrc-regulator driver, which never got compiled at all, and would
> not build anyway because of missing headers and typos: that commit did
> not get any Fixes tag because, well, backporting makes no sense at all
> as the DVFSRC support - which is critical for that driver to work - is
> introduced with *this series*! :-)
>
> P.S.: The DVFSRC regulator is a requirement for the MediaTek UFSHCI
> controller's crypto boost feature, which is already upstream but
> lacking the actual regulator to work....... :-)
>
> [1]: https://lore.kernel.org/all/20210812085846.2628-1-dawei.chien@mediatek.com/
>
> AngeloGioacchino Del Regno (7):
> dt-bindings: regulator: Add bindings for MediaTek DVFSRC Regulators
> dt-bindings: interconnect: Add MediaTek EMI Interconnect bindings
> dt-bindings: soc: mediatek: Add DVFSRC bindings for MT8183 and MT8195
> soc: mediatek: Add MediaTek DVFS Resource Collector (DVFSRC) driver
Looks like the driver got picked up, but not the binding.
mediatek,mt8183-dvfsrc and mediatek,mt8195-dvfsrc show up in next as
undocumented.
Rob
More information about the Linux-mediatek
mailing list