As John Ogness commented, I also found it would hang if sleep when >= SYSTEM_SCHEDULING. And as in commit https://lore.kernel.org/lkml/20170516184736.272225698@xxxxxxxxxxxxx/, it should enable sleep right when the scheduler starts working(>= SYSTEM_RUNNING). -Chao. -----Original Message----- From: John Ogness <john.ogness@xxxxxxxxxxxxx> Sent: Friday, July 30, 2021 10:47 PM To: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> Cc: Qin, Chao <chao.qin@xxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; linux-rt-users@xxxxxxxxxxxxxxx; tglx@xxxxxxxxxxxxx; rostedt@xxxxxxxxxxx; mgross@xxxxxxxxxxxxxxx; Mei, Paul <paul.mei@xxxxxxxxx>; Li, Lili <lili.li@xxxxxxxxx> Subject: Re: [PREEMPT_RT][PATCH] printk: Enhance the condition check of msleep in pr_flush() On 2021-07-30, Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote: > On 2021-07-19 17:01:51 [+0206], John Ogness wrote: >> On 2021-07-19, chao.qin@xxxxxxxxx wrote: >> > --- a/kernel/printk/printk.c >> > +++ b/kernel/printk/printk.c >> > @@ -3620,7 +3620,8 @@ bool pr_flush(int timeout_ms, bool reset_on_progress) >> > u64 diff; >> > u64 seq; >> > >> > - may_sleep = (preemptible() && !in_softirq()); >> > + may_sleep = (preemptible() && !in_softirq() >> > + && (system_state >= SYSTEM_RUNNING)); > > I don't have more context but scheduling should work starting with > SYSTEM_SCHEDULING. I also thought this, but a quick test shows that is not the case. For example, init/main.c:kernel_init() is called in preemptible context, but msleep() will hang if called at the beginning of that function. John Ogness
