On Wed, 10 Jun 2015 07:09:41 -0400, Peter Hurley wrote: > On 06/08/2015 05:48 PM, Steven Walter wrote: > > On Mon, Jun 8, 2015 at 4:15 PM, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote: > >> On Mon, Jun 08, 2015 at 03:28:21PM -0400, Steven Walter wrote: > >>> Ports that are marked low-latency get their own kthread that is > >>> high-priority. > >>> > >>> Signed-off-by: Steven Walter <stevenrwalter@xxxxxxxxx> > >>> --- > >>> drivers/tty/tty_buffer.c | 14 +++++++++++++- > >>> 1 file changed, 13 insertions(+), 1 deletion(-) > >> > >> What kind of measurements have you done to show that these changes > >> actually matter? > > > > The application that benefits from this is driving a motor controller > > card over an RS-232 link. The userspace application that communicates > > over the TTY needs to respond to received bytes within about 1ms. > > Without this patch, I have observed (via ftrace) the kworker threads > > get delayed by over 100ms, > > Which is what happens when a high-prio RT thread starves the system > (which is effectively a priority inversion with the i/o workers). > > On a vanilla preemptive kernel, <10us worker latency is typical, and rarely > over 250us (though not bounded). Could you explain why do you expect non-RT task like kworker to have such a low latency? If the system is fully loaded with other non-rt tasks, preemptive kernel or not, I kworkers will wait for every other process to consume their time slice. Is there anything special about scheduling worker threads? Thanks, Kuba -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
