Re: [patch] proposed fix for ptrace() SMP race

[Manfred Spraul]

>
> BTW, checking this stuff I found two bugs, one is the check for
> cpus_allowed before calling reschedule_idle,

The call in linux/kernel.c, around line 348?
348 if (!synchronous || !(p->cpus_allowed & (1 << smp_processor_id())))
349       reschedule_idle(p);

The test was added for wake_up_{,interruptible_}sync(): if the woken up
task is not permitted to run on the current cpu, reschedule() is
necessary, otherwise skip the reschedule.
If ptrace sets cpus_allowed to 0 between wake_up_sync() and schedule(),
the reschedule is lost. But always rescheduling would defeat the purpose
of wake_up_sync().

--
	Manfred