[haiku-commits] Re: r37094 - haiku/trunk/src/servers/net

  • From: Philippe Houdoin <philippe.houdoin@xxxxxxxxx>
  • To: haiku-commits@xxxxxxxxxxxxx
  • Date: Fri, 11 Jun 2010 12:36:27 +0200

> Shutting it down might just not have been implemented. I did only the
> reader, and that one should be properly shut down.

Indeed. And the device is closed(), which what this disable feature is
all about: don't keep loaded device driver.

> The consumer thread
> was Hugo's idea to separate reading packets from processing it, as well
> as simplifying locking for local protocols (that previously did all
> their processing in the sending thread).
> Due to the added context switches, I'm not yet sure if that's a good
> idea, though, anyway, but I never did any measurements either
> (regarding CPU load, and performance).

IIRC, Ingo did measurements. Can't remember what was the conclusion,
but as his design is still there, I bet it was not worth than the
previous one.

I'll see if I could add consumer shutdown then. One kernel thread less
is one kernel thread less.

> IIRC the device would need to go away to let the network server remove
> the interface. If not even the device disappears, I would start looking
> into the USB stack's unpublishing part.

I'll investigate it as much as I could. Which is not much these days...

Other related posts: