Re: REGRESSION: boot stalls on several old dual core Intel CPUs

[Peter Zijlstra <peterz@infradead.org>]

On Sat, Sep 01, 2018 at 11:51:26AM +0930, Kevin Shanahan wrote:
> On Thu, Aug 30, 2018 at 03:04:39PM +0200, Peter Zijlstra wrote:
> > On Thu, Aug 30, 2018 at 12:55:30PM +0200, Siegfried Metz wrote:
> > > Dear kernel developers,
> > > 
> > > since mainline kernel 4.18 (up to the latest mainline kernel 4.18.5)
> > > Intel Core 2 Duo processors are affected by boot stalling early in the
> > > boot process. As it is so early there is no dmesg output (or any log).
> > > 
> > > A few users in the Arch Linux community used git bisect and tracked the
> > > issue down to this the bad commit:
> > > 7197e77abcb65a71d0b21d67beb24f153a96055e clocksource: Remove kthread
> > 
> > I just dug out my core2duo laptop (Lenovo T500) and build a tip/master
> > kernel for it (x86_64 debian distro .config).
> > 
> > Seems to boot just fine.. 3/3 so far.
> > 
> > Any other clues?
> 
> One additional data point, my affected system is a Dell Latitude E6400
> laptop which has a P8400 CPU:
> 
>   vendor_id     : GenuineIntel
>   cpu family    : 6
>   model         : 23
>   model name    : Intel(R) Core(TM)2 Duo CPU     P8400  @ 2.26GHz
>   stepping      : 6
>   microcode     : 0x610
> 
> Judging from what is being discussed in the Arch forums, it does seem
> to related to the CPU having unstable TSC and transitioning to another
> clock source.

Yes; Core2 doesn't have stable TSC.

> Workarounds that seem to be reliable are either booting
> with clocksource=<something_not_tsc> or with nosmp.

nosmp is weird; because even on UP TSC should stop in C state.

processor_idle (acpi_idle) should mark the TSC as unstable on Core2 when
it loads (does so on my T500).

> One person did point out that the commit that introduced the kthread
> did so to remove a deadlock - is the circular locking dependency
> mentioned in that commit still relevant?
> 
> commit 01548f4d3e8e94caf323a4f664eb347fd34a34ab
> Author: Martin Schwidefsky <schwidefsky@de.ibm.com>
> Date:   Tue Aug 18 17:09:42 2009 +0200
> 
>     clocksource: Avoid clocksource watchdog circular locking dependency
> 
>     stop_machine from a multithreaded workqueue is not allowed because
>     of a circular locking dependency between cpu_down and the workqueue
>     execution. Use a kernel thread to do the clocksource downgrade.

I cannot find stop_machine usage there; either it went away or I need to
like wake up.