Hi, Stefan,
today I made a stupid patch for testing only: I instructed the wm8731
codec's probe function to wait for si5351's clock output and return
EPROBE_DEFER until it is ready. This delays any I2C bus access from
wm8731 driver until si5351 initialization is complete. It worked like a
charm.
Now I have no idea what type of issue we are dealing with. I can't
believe that I2C bus should allow only one device at a time.
Needless to say if only one of the drivers is loaded(either wm8731 or
si5351) then I2C bus works with no issue.
Sergey
On 12/7/18 6:42 PM, Stefan Wahren wrote:
Hi Sergey,
[add Noralf and drop the Broadcom guys from CC since this specific to
the RPi]
Am 07.12.18 um 14:36 schrieb Sergey Suloev:
Greetings,
I am dealing with a custom hardware based on RPI3B running kernel 4.19
(or 4.20).
My device has 2 slaves on I2C bus: codec WM8731 (on address 0x1a) and
clock generator si5351a (on 0x60).
It has been a lot of time for me fighting with an I2C issue which does
not allow the two I2C devices work correctly together: I2C bus hangs
up at kernel boot.
Could you give some kernel message or some measurements of the I2C line
states?
The issue happens in 50-70% boot cases and the fail percentage
noticeably depends on the bus clock set in DTS: the more frequency
set the less fail percentage.
I assume you don't use the downstream cpufreq driver, which could
generate interferences on the clock.
The kernel itself loads successfully, I can ssh to the RPI in order to
issue "i2cdetect -y 0" and make sure that I2C bus is locked completely
and not responding.
AFAIK this I2C0 is shared with the VideoCore. Did you already tried I2C1
which should be exclusive for the ARM core(s)?
Stefan
So far I have found the only way to make it work : to compile si5351a
driver into kernel and use wm8731 driver as a module.
This probably makes the clock driver load 100% before the codec and
I2C bus does not hang up.
Since it can be fixed this way the problem really looks like a "race
condition" in I2C bus/driver.
Would be really nice to know your opinion and possible ways to fix this.
Best regards,
Sergey
