Re: RT and Cascade interrupts

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

lau den 28.05.2005 Klokka 13:48 (-0400) skreiv john cooper:
> The other possibility to fix the original problem within
> the scope of the RPC code was to replace the bit state of
> RPC_TASK_HAS_TIMER with a count so we don't obscure the
> fact the timer was requeued during the preemption window.
> This happens as rpc_run_timer() does an unconditional
> clear_bit(RPC_TASK_HAS_TIMER,..) before returning.

Just to clarify a bit more here. RPC_TASK_HAS_TIMER is definitely _not_
redundantly replicating timer state, as I saw you asserting previously.

There are two distinct cases where timer->base may be NULL. In the first
case, the timer is inactive and was never queued so it is entirely
pointless to be calling del_timer() & friends. In the second case, the
timer was active, but has expired and was picked up by __run_timers().
del_singleshot_timer_sync() fails to distinguish these two cases, and
will therefore end up redundantly calling del_timer_sync() every time a
task loops through __rpc_execute() without setting a timer. Your patch
to remove RPC_TASK_HAS_TIMER will re-introduce that redundant behaviour
but will re-introduce precisely the race that del_timer_sync() is
designed to fix.

So, I repeat: I want to see a case description that clearly demonstrates
a race scenario in the existing code before I'm willing to consider any
patches at all.

Cheers,
  Trond