Hi Chris, > > My understand is that an ordinary I2C device would just use normal (and > > sleepable) I2C transfers while the device is in use. > > You are spot-on here :-) Now I use IMX 290 and 462. > > OTOH I wonder if such issues are limited to those sensors only. Hmm, yes. I know no other I2C device that has these timeout issues. (*) > The problem is I use Sony IMX290 and IMX462 image sensors, and they have > an apparently hard-coded timeout of about 2^18 their master clock cycles > (= ca. 7 ms with my setup). After the timeout they simply disconnect > from the I2C bus. Of course, this isn't mentioned in the docs. hmm. I have no idea about this sensor and your setup. So I can just give hints: This timeout seems strange. If this 7 ms timeout is required, it would mean that I2C masters require to fullfill real-time/deadline requirements. For "small" I2C master in microcontrolles this seems ok-ish, but for general operating systems real-time requirements are hard. The real-time patches for linux just landed recently and it still requires fine tuning the system for the required deadlines. Maybe you just hit a corner case or a bug, that can be avoid, in the I2C device. Maybe check with the manufacturer directly? > Unfortunately, "normal" I2C accesses take frequently more than those > 7 ms (mostly due to scheduling when all CPU cores are in use). Yes, correctly. There are multiple cases in which I2C transactions to the same device can be preempted/delayed: A busy system, as you said, or when some other driver in the kernel accesses another I2C device on the same bus. This will lock the bus/I2C adapter for the duration of its transfer. Do you know the I2C repeated start feature [1]? This allows to batch together multiple I2C read/writes in a single transfer. And in the best case, this transfer is executed in one go without a delay in between. At least in the kernel it's guaranteed that no other driver can go in between with another transfer. Kind regards, Stefan [1]: https://www.i2c-bus.org/repeated-start-condition/ (*) Fun answer: Actually external watchdogs have timeouts. But the timeout duration is in the range of seconds, not milliseconds. And timeout expiration is expected (in error cases ;-).
