Re: [BugReport] 2nd cpu does not work.
Ken-ichi Yaku (yaku@css1.kbnes.nec.co.jp)
Wed, 21 Apr 1999 21:29:36 +0900 (JST)
Ingo Molnar wrote:
>
> On Mon, 19 Apr 1999, Ken-ichi Yaku wrote:
>
> > /proc/cpuinfo said kernel detected both cpus, but 2nd cpu did not work.
> > It means 2nd cpu kept running idle tasks.
> >
> > We think that value of has_cpu flag in idle task is the cause of this
> > problem.
> >
> > Even if 2nd cpu has idle task, has_cpu flag in idle task is not set.
> >
> > Therefore scheduler can't deliver new tasks to 2nd cpu.
> > (keeps running idle task.)
>
> yes but this happens only until the first reschedule happens on the second
> CPU. I suspect all interrupts are delivered to the first CPU on your
> system for some reason, correct? In that case the idle thread on the
> second CPU is never directly rescheduled by an interrupt. [btw., i've
> recently fixed it the same way you did in my tree, and this is not the
> only problem in that area, another problem is that the first time the idle
> thread is started up we might reschedule before it has gotten it's
> 'proper' stack, which can cause a rather subtle and silent memory
> corruption. I suspect could be the reason why certain kernel builds have
> problems booting under SMP. I'll send a full patch for both of these
> problems this week.]
>
> -- mingo
Thank you for your comments. We are looking forward to your full patch.
Let me explain again. All interrupts are delivered to the first CPU only.
Our machine has two IO APICs. Current 2.2.6 SMP kernel does not support
multiple IO APICs yet, so our machine is running with non SMP mode. In
this case that second CPU did not receive interrupts and the secound CPU
will never be rescheduled. We accordingly sent this bug report.
By the way, we will finish soon to create the patch for kernel 2.2.6
which support multiple IO APICs. Now we are testing it on our system. We
will contribute patch to this mailing list, after our testing is completed.
Regards,
--------
Ken-ichi Yaku
NEC Software Kobe, Ltd
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/