On Fri, 4 Dec 2015 17:57:44 +0100 Petr Mladek [off-list ref] wrote:

On Wed 2015-12-02 00:24:49, Jiri Kosina wrote:

quoted

On Fri, 27 Nov 2015, Petr Mladek wrote:

quoted

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?

quoted

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?