Re: [RFC][PATCH v4 -next 1/4] Move kmsg_dump(KMSG_DUMP_PANIC) below smp_send_stop()

[Don Zickus <dzickus@redhat.com>]

On Wed, Jan 11, 2012 at 03:28:11PM +0800, Chen Gong wrote:
> 于 2012/1/11 4:29, Seiji Aguchi 写道:
> >
> >>I agree with you. How about adding macros or something like WARN_ON(XX_ARCH) or
> >>Kconfig to limit its scope?
> >
> >Thank you for giving me your idea.
> >Your suggestions above will work for me because I'm a x86 user.
> >If Tony agrees to it, I can update my patch.
> >
> >But, I'm hesitating to add WARN_ON() or change Kconfig only for specific arch
> >because pstore aims for generic interface and this is related to its design.
> >Also, ramoops is going to use pstore now. It doesn't depend on x86.
> >I'm worried that ramoops users will complain about this change.
> >
> >So, I think a reasonable solution at this time is just adding some explanations
> >about smp_send_stop() to documentation as follows.
> >
> >Users can use pstore with their own responsibility and ask developers
> >if smp_send_stop() is reliable enough in panic situation on architecture they want to run.
> >
> >What do you think?
> >
> >---
> >  Documentation/ABI/testing/pstore |    8 ++++++++
> >  1 files changed, 8 insertions(+), 0 deletions(-)
> >
> >diff --git a/Documentation/ABI/testing/pstore b/Documentation/ABI/testing/pstore
> >index ff1df4e..5583729 100644
> >--- a/Documentation/ABI/testing/pstore
> >+++ b/Documentation/ABI/testing/pstore
> >@@ -11,6 +11,14 @@ Description:	Generic interface to platform dependent persistent storage.
> >  		of the console log is captured, but other interesting
> >  		data can also be saved.
> >
> >+		In case of panic, pstore is invoked after smp_send_stop()
> >+		,a function call stopping other cpus, so that we can get
> >+		logs simpler and cleaner with just one running cpu.
> >+
> >+		As for x86, smp_send_stop() is reliable enough to work in
> >+		panic situation. But we are not guaranteed that it works
> >+		reliably on other architectures.
> >+
> >  		# mount -t pstore -o kmsg_bytes=8000 - /dev/pstore
> >
> >  		$ ls -l /dev/pstore
> 
> The explanation is great. but In my opinion, I still insist that
> a WARN_ON() is necessary. What do you think, Tony and Don?

I guess I still don't understand why.  Who uses kmsg_dump besides x86?  It
seems like there was only 3 or 4 subsystems that were registering.

Cheers,
Don