On Thu, Jun 02, 2005 at 09:43:46PM -0400, Stuart MacDonald wrote: > The serial sub-drivers (whiteheat.c is one) have to call > tty_flip_buffer_push() somehow. Queueing the task means it runs from > (timer) interrupt context. Calling it directly means a) low_latency is > set; runs from sub-driver rx-side interrupt context or b) low_latency > is *not* set and it's queued, see above. > > So I don't see a way to not have flush_to_ldisc() eventually run in > interrupt context. If there's a way, please let me know, because that > would fix this issue right up. > > So, flush_to_ldisc() is running in interrupt context. It can call back > into usbserial.c in a number of ways, specifically serial_throttle. > Which tries to acquire a semaphore, which sometimes sleeps, resulting > in a "scheduling in interrupt" crash. (In 2.6 the semaphore has been > replaced with a work queue it looks like.)
In 2.6, throttle can be called in interrupt context too. See the recent visor fix that was required for this. > So it seems to me that usbserial is fundamentally broken. Am I > correct? Is there any way to clean this up? Yes I'm volunteering. Probably. Backport the 2.6 changes :) Does anyone still use the newer 2.4 kernel releases in a distro anymore? thanks, greg k-h ------------------------------------------------------- This SF.Net email is sponsored by Yahoo. Introducing Yahoo! Search Developer Network - Create apps using Yahoo! Search APIs Find out how you can build Yahoo! directly into your own Applications - visit http://developer.yahoo.net/?fr=offad-ysdn-ostg-q22005 _______________________________________________ linux-usb-devel@lists.sourceforge.net To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
