On 25/06/2024 11:53, Jerome Brunet wrote:
On Tue 25 Jun 2024 at 11:38, Neil Armstrong <neil.armstrong@xxxxxxxxxx> wrote:
Hi,
[+cc people from linux-msm]
On 24/06/2024 19:31, Jerome Brunet wrote:
Add support for the HW found in most Amlogic SoC dedicated to measure
system clocks.
This drivers aims to replace the one found in
drivers/soc/amlogic/meson-clk-measure.c with following improvements:
* Access to the measurements through the IIO API:
Easier re-use of the results in userspace and other drivers
* Controllable scale with raw measurements
* Higher precision with processed measurements
Jerome Brunet (2):
dt-bindings: iio: frequency: add clock measure support
iio: frequency: add amlogic clock measure support
.../iio/frequency/amlogic,clk-msr-io.yaml | 50 ++
drivers/iio/frequency/Kconfig | 15 +
drivers/iio/frequency/Makefile | 1 +
drivers/iio/frequency/amlogic-clk-msr-io.c | 802 ++++++++++++++++++
4 files changed, 868 insertions(+)
create mode 100644 Documentation/devicetree/bindings/iio/frequency/amlogic,clk-msr-io.yaml
create mode 100644 drivers/iio/frequency/amlogic-clk-msr-io.c
While I really appreciate the effort, and the code looks cool, the clkmsr is really
a debug tool, and I'm not sure IIO is the right place for such debug tool ?
The reason why I went through the trouble of doing an IIO port is
because I need that for other purposes than debug. I need to to be able
to check a frequency from another driver. I don't see a reason to invent
another API when IIO provide a perfectly good one.
The HW does measurements. IIO seems like the best place for it.
For the record, I need this for a eARC support.
eARC has a PLL that locks on incoming stream. eARC registers show wether
the PLL is locked or not, but not at which rate. That information is
needed in ASoC. Fortunately the eARC PLL is one of measured clock, which
is a life saver in that case.
This is a very interesting use-case, and quite weird nothing is provided
on the eARC side.
So yes it's definitely a valid use-case, but:
- we should keep the debugfs interface, perhaps move it in the iio driver ?
- we should keep a single compatible, so simply update the current bindings with iio cells
- for s4 & c3, it's ok to either add a second reg entry in the bindings
Neil
Everything that was available through the old driver still is, with more
precision and more control.
There's almost the same interface on qcom SoCs (https://github.com/linux-msm/debugcc) but
they chose to keep it in userspace until we find an appropriate way to expose
this from the kernel the right way.
If it enabled us to monitor a frequency input for a product use-case, IIO would be
the appropriate interface, but AFAIK it's only internal clocks and thus I'm worried
it's not the best way to expose those clocks.
Neil
