Re: [XEN PATCH 6/9] x86/shutdown: protect against recurrent machine_restart()

[Jan Beulich <jbeulich@xxxxxxxx>]

On 12.03.2024 17:05, Krystian Hebel wrote:
> 
> On 8.02.2024 12:30, Jan Beulich wrote:
>> On 14.11.2023 18:50, Krystian Hebel wrote:
>>> If multiple CPUs called machine_restart() before actual restart took
>>> place, but after boot CPU declared itself not online,
>> Can you help me please in identifying where this operation is? I can see
>> two places where a CPU is removed from cpu_online_map, yet neither
>> __stop_this_cpu() nor __cpu_disable() ought to be coming into play here.
>> In fact I didn't think CPU0 was ever marked not-online. Except perhaps
>> if we came through machine_crash_shutdown() -> nmi_shootdown_cpus(), but
>> I'm sure you would have mentioned such a further dependency.
>>
> BUG_ON() in cpu_notifier_call_chain() (I've been playing with some of
> the notifiers and one of them eventually failed) resulted in panic()
> around the same time AP did in pm_idle() due to inconsistent settings
> between BSP and AP for MWAIT/MONITOR support after TXT dynamic
> launch. There is 5s delay between smp_send_stop() and actual reboot,
> during that time AP spammed the output so the original reason for
> panic() was visible only after unreasonable amount of scrolling.
> 
> Adding TXT support is the reason why I even started making AP bring-up
> parallel. Problem with MWAIT doesn't happen in current code or changes
> in this patchset, but handling of such error is related to SMP so I've 
> included it.

If you mean to address a latent problem, then you want to say so and you
want to make sure you include enough detail on the (future) conditions
under which the problem may happen. Otherwise anything you say wants to
match present code / behavior.

Jan