Re: [PATCH printk v4 4/6] printk: remove NMI tracking

[Petr Mladek]

On Wed 2021-07-21 14:52:15, John Ogness wrote:
> On 2021-07-21, Petr Mladek <pmladek@xxxxxxxx> wrote:
> >> --- a/kernel/trace/trace.c
> >> +++ b/kernel/trace/trace.c
> >> @@ -9647,7 +9647,7 @@ void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
> >>  	tracing_off();
> >>  
> >>  	local_irq_save(flags);
> >> -	printk_nmi_direct_enter();
> >> +	printk_deferred_enter();
> >
> > I would prefer to do not manipulate the printk context here anymore,
> > as it was done in v3.
> >
> > printk_nmi_direct_enter() was added here by the commit the commit
> > 03fc7f9c99c1e7ae2925d4 ("printk/nmi: Prevent deadlock when
> > accessing the main log buffer in NMI"). It was _not_ about console
> > handling. The reason was to modify the default behavior under NMI
> > and store the messages directly into the main log buffer.
> >
> > When I think about it. The original fix was not correct. We should have
> > modified the context only when ftrace_dump() was really called under NMI:
> >
> > 	if (in_nmi())
> > 		printk_nmi_direct_enter();
> >
> > By other words. We should try to show the messages on the console
> > when ftrace_dump()/panic() is not called from NMI. It will help
> > to see all messages even when the ftrace buffers are bigger
> > than printk() ones.
> >
> > And we do not need any special handling here for NMI. vprintk()
> > in printk/printk_safe.c will do the right thing for us.
> 
> Agreed. We need to mention this behavior change in the commit
> message. Perhaps this as the commit message:
>
> All NMI contexts are handled the same as the safe context: store the
> message and defer printing. There is no need to have special NMI
> context tracking for this. Using in_nmi() is enough.
> 
> There are several parts of the kernel that are manually calling into
> the printk NMI context tracking in order to cause general printk
> deferred printing:
> 
>     arch/arm/kernel/smp.c
>     arch/powerpc/kexec/crash.c
>     kernel/trace/trace.c
> 
> For arm/kernel/smp.c and powerpc/kexec/crash.c, provide a new
> function pair printk_deferred_enter/exit that explicitly achieves the
> same objective.
> 
> For ftrace, remove general printk deferring. This general deferrment
> was added in commit 03fc7f9c99c1 ("printk/nmi: Prevent deadlock when
> accessing the main log buffer in NMI"), but really should have only
> been deferred when in NMI context. Since vprintk() now checks for
> NMI context when deciding to defer, ftrace does not need any special
> handling.

I would make it less focused on the deferring part and try to explain
the original purpose here, something like:

"For ftrace, remove the printk context manipulation completely. It was
added in commit 03fc7f9c99c1 ("printk/nmi: Prevent deadlock when
accessing the main log buffer in NMI"). The purpose was to enforce
storing messages directly into the ring buffer even in NMI context.
It really should have only modified the behavior in NMI context.
There is no need for a special behavior any longer. All messages are
always stored directly now. The console deferring is handled
transparently in vprintk()."

But I do not resist on it. And feel free to make it shorter.

Best Regards,
Petr