Re: RT and Cascade interrupts

[Trond Myklebust <trond.myklebust@fys.uio.no>]

on den 01.06.2005 Klokka 15:20 (-0400) skreiv john cooper:

> You might have missed in my earlier mail as
> this is a not an MP kernel ie: !CONFIG_SMP
> The synchronous timer delete primitives don't
> exist in this configuration:
> 
> include/linux/timer.h:
> 
> #ifdef CONFIG_SMP
>    extern int del_timer_sync(struct timer_list *timer);
>    extern int del_singleshot_timer_sync(struct timer_list *timer);
> #else
> # define del_timer_sync(t) del_timer(t)
> # define del_singleshot_timer_sync(t) del_timer(t)
> #endif


For the RT patched stuff that should read

#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_SOFTIRQS)
  extern int del_timer_sync(struct timer_list *timer);
  extern int del_singleshot_timer_sync(struct timer_list *timer);
#else
# define del_timer_sync(t) del_timer(t)
# define del_singleshot_timer_sync(t) del_timer(t)
#endif


> BTW, I don't know if you happened on the mail I sent
> yesterday.  It details rpc_run_timer() waking up an
> application task blocked in call_transmit().  The
> app task preempts ksoftirqd and eventually does a
> __rpc_sleep_on()/__mod_timer() which requeues the
> timer in the cascade.  When ksoftirqd/rpc_run_timer()
> resumes RPC_TASK_HAS_TIMER is unconditionally cleared
> however the timer is queued in the cascade.  This
> appears to be at least one cause of the timer cascade
> corruption I've seen.

I saw it. Once again, I don't see how that can happen. __rpc_execute()
should be calling rpc_delete_timer() before it calls task->tk_action.

There should be no instances of RPC entering call_transmit() or any
other tk_action callback with a pending timer.

Cheers,
  Trond