However, this behaviour makes me suspect more and more that there is
something broken in the PPP (or serial code) timings,
Well, there is a known problem in the tty layer, which is that upper
layer processing doesn't happen except on the next clock tick. Hence,
there is on average a .005 second delay between when the PPP framing
character comes in and when it actually gets processed.
I've considered putting in a PPP specific hack where when the serial
layer sees the PPP framing character, it immediately forces the serial
processing to happen as soon as the interrupt returns (as a bottom half
routine). It's been a layering violation, though, so I wouldn't really
want to do this unless someone can show that it would really make a big
difference.
- Ted