msperl

@notro: Any recommendation for devices with which to test this?

actually (and again as per official documentation) you can only do DMA transfers of 0-65535 bytes (16 bits) - at least that is the way I read it. But if...

See what I have found and measured so far - the wiki contains all of it. These scheduling delays may be a major headache, but we have shaved of about...

The more I read of LoSSI the more I come to the conclusion that with the code now it can not work. The question is: is it needed by anyone?

Some more thought on this gives me the impression that the "read" operations are not implemented at all (as you said: read would require knowledge of the last command and...

Next thing I will do is enable the realtime priority with the bcm2835 driver and put up the differences on the wiki including screenshots...

There have been now a lot of fixes already applied - some of which I can not really test due to missing HW. See the other open issues... Now I...

Actually I have found a quick workaround for my issues by implementing polling for situations where we run the interrupt which then wakes up the worker thread for requests that...

We could probably modify the code to use DMA for anything that lies beyond this. This would mean minimal overheads/penalties for most cases. The only thing that is open to...

reducing the time to 10us will mean an interrupt getting scheduled and that means a penalty of 36us. The data has been added and I will include the graphs in...