Forgot to mention, this patch and analyze are based on 3.4 rt. On 2014/4/30 12:02, Qiang Huang wrote: > Hi all, > > Ingo's patch: > 00853150572df2 tasklet: Prevent tasklets from going into infinite spin in RT > changed behavior about tasklet in NO-RT scenario, but I found some suspicious > from code and comments. > > Like the below patch says, if a tasklet is already running on another CPU, > then it would have RUN flag and have no SCHED flag, so when we call > tasklet_schedule, then in __tasklet_common_schedule(), tasklet_trylock() > will return false, and the tasklet won't be added to tasklet_vec list, > so there is no reschedule, right? > > Another, the comments said schedule called from tasklet itself would cause > lockup, like it mentioned above, in tasklet function, the tasklet has RUN > flag, and have no SCHED flag, so when it call tasklet_schedule, it will > only meet tasklet_trylock return false, then nothing happened. I think this > should not be any lockup. > > After all, if my understanding is correct, it changed the behavior when > tasklet scheduled itself, this would affect some kernel modules using > tasklet, is this what we want in RT? If so, should we use RT_FULL config > to isolate this affect? > > -------------------- > > Correct the comments as the code really do. > > Signed-off-by: Qiang Huang <h.huangqiang@xxxxxxxxxx> > --- > include/linux/interrupt.h | 5 ++--- > 1 file changed, 2 insertions(+), 3 deletions(-) > > diff --git a/include/linux/interrupt.h b/include/linux/interrupt.h > index fb05761..e9d73e6 100644 > --- a/include/linux/interrupt.h > +++ b/include/linux/interrupt.h > @@ -514,9 +514,8 @@ extern void __send_remote_softirq(struct call_single_data *cp, int cpu, > to be executed on some cpu at least once after this. > * If the tasklet is already scheduled, but its execution is still not > started, it will be executed only once. > - * If this tasklet is already running on another CPU, it is rescheduled > - for later. > - * Schedule must not be called from the tasklet itself (a lockup occurs) > + * If this tasklet is already running on another CPU (or schedule is called > + from tasklet itself), it wound not be rescheduled. > * Tasklet is strictly serialized wrt itself, but not > wrt another tasklets. If client needs some intertask synchronization, > he makes it with spinlocks. > -- > 1.8.3 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html > > . > -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
