On 11/03/2020 11:01, Lokesh Vutla wrote:
Hi Richard,
On 10/03/20 7:02 PM, Richard Cochran wrote:
On Tue, Mar 10, 2020 at 01:17:40PM +0530, Lokesh Vutla wrote:
Timer resolution is entirely dependent on the input clock frequency. 30us
resolution is true only if the input clock frequency of DMTIMER is 32KHz.
AM335x[0] and later TI SoCs provide the following 3 options as the input clocks
to DMTIMERs that we are concerned with:
- Oscillator clock (24MHz in BBB). Resolution is 42ns
That is better. Still a bit disappointing considering that the
resolution is coarser than the expected synchronization accuracy.
Is this information sufficient to consider for a generic user-space solution :)
Or you have other opinions?
It would be useful. I can see several moving parts:
You have got the moving parts right. Let me provide the status of each:
- A device tree snippet that configures the clock tree to select
TIMERx_GCLK source.
PWM omap timer already has this support for selecting the clock[0]
- Somehow arrange for the pulse to appear both as an input to CPTS
EXTTS and also as an output on a GPIO pin (device tree covers this?)
Output of DMTIMER(PWM Pulse) is already connected to both CPTS EXTTS and a GPIO
pin. Just pinmux need to be configured properly to capture the pulse on the
scope. DT captures the pinmux part.
- Kernel driver work to configure the DMTIMER (?) (iirc this is missing from
mainline)
There is already existing pwm omap dmtimer driver. I have recently posted
updates to this driver to be able to change pwm period on the fly [1]. Hoping to
get this merged soon.
- Kernel driver work in CPTS to support PTP_EXTTS_REQUEST (and PTP_PIN_SETFUNC?)
Grygorii (in CC) is already working on that. Patches will be posted soon to ML.
I've posted TI CPSW CPTS updates to enable HW_TS_PUSH events
(PTP_EXTTS_REQUEST).
PTP_PIN_SETFUNC - there are no direct pins which can be manipulated.
- User space program that:
- configures pwm via sysfs
- enables CPTS PHC PIN
- enables and reads PTP_EXTTS_REQUESTs
- feeds the offsets into a servo
- corrects PWM period with servo adjustment via sysfs
This is the userspace program I have worked on to get the synchronization
working. Currently the servo part of the tool is in very hacky shape. Before
cleanup I wanted to get more inputs on where it can be integrated. But anyways
as you said, Ill clean it up and post it as a separate tool. We can then decide
the correct location.
And I'd like to rise additional question here.
The above solution adds PPS generation using PTP_EXTTS events and PIN output,
but, as per my understanding, there are no way now to feed phc2sys with such
PPS.
So question is how PPS generated by PTP_EXTTS events can be wired to phc2sys?
- option1: update phc2sys to accept PTP_EXTTS as PPS
- option2: allow kernel PHC to generate or PTP_CLOCK_EXTTS/or PTP_CLOCK_PPS/or
both
by somehow configuring that PTP_EXTTSx is also PPS.
Challenge with option 2 is that Kernel PPS configuration is part of
ptp_clock_register() and so static.
(please comment/correct the above)
I agree that this would be useful. However, this is really
device-specific stuff and not general purpose. Possibly we could
publish this in linuxptp under contrib/ or something, but my feeling
is that it really belongs in its own little project. Does TI have a
github or similar from which to publish?
TI does host all the public code under git.ti.com.
In any case, please do post what you have got to this list. I am
eager to try it out!
okay. Give me couple of days. Patches will be posted to the list.
[1] https://lkml.org/lkml/2020/3/13/1008
[2] https://patchwork.ozlabs.org/cover/1254708/
--
Best regards,
grygorii
_______________________________________________
Linuxptp-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
