Re: [PATCH v2 1/5] printk/nmi: Generic solution for safe printk in NMI

[Andrew Morton]

On Fri, 4 Dec 2015 17:57:44 +0100 Petr Mladek <pmladek@xxxxxxxx> wrote:

> On Wed 2015-12-02 00:24:49, Jiri Kosina wrote:
> > On Fri, 27 Nov 2015, Petr Mladek wrote:
> > 
> > > MN10300 has its own implementation for entering and exiting NMI 
> > > handlers. It does not call nmi_enter() and nmi_exit(). Please, find 
> > > below an updated patch that adds printk_nmi_enter() and 
> > > printk_nmi_exit() to the custom entry points. Then we could add HAVE_NMI 
> > > to arch/mn10300/Kconfig and avoid the above warning.
> > 
> > Hmm, so what exactly would go wrong if MN10300 (whatever that architecture 
> > is) would call nmi_enter() and nmi_exit() at the places where it's 
> > starting and finishing NMI handler?
> > 
> > >From a cursory look, it seems like most (if not all) of the things called 
> > from nmi_{enter,exit}() would be nops there anyway.
> 
> Good point. Max mentioned in the other main that the NMI handler
> should follow the NMI ruler. I do not why it could not work.
> In fact, it might improve things, e.g. nmi_enter() blocks
> recursive NMIs.
> 
> I think that it will move it into a separate patch, thought.
> 

I've sort of lost the plot on this patchset.

I know Daniel had concerns (resolved?).  Sergey lost the ability to
perform backtraces and has a proposed fix ("printk/nmi: restore
printk_func in nmi_panic") but that wasn't fully resolved and I didn't
merge anything.  I'm not sure what Jan's thinking is on it all.

So... I'll retain 

printk-nmi-generic-solution-for-safe-printk-in-nmi.patch
printk-nmi-use-irq-work-only-when-ready.patch
printk-nmi-warn-when-some-message-has-been-lost-in-nmi-context.patch
printk-nmi-increase-the-size-of-nmi-buffer-and-make-it-configurable.patch

in -mm for now.  Perhaps I should drop them all and we start again
after -rc1?