Re: [RFC][PATCH] printk: Add option to append kernel version to the dict

[Calvin Owens <calvinowens@fb.com>]

On Sunday 05/15 at 00:19 +0200, Richard Weinberger wrote:
> On Fri, May 13, 2016 at 10:58 PM, Calvin Owens <calvinowens@fb.com> wrote:
> > We use netconsole to collect kernel logs from all the servers at
> > Facebook. We use this patch internally so each logline has a record of
> > which kernel version emitted it.
> >
> > At first glance, this might seem lazy: as you would expect, we have a
> > database which records which kernel version a host is currently running.
> > But there are a lot of situations where that database cannot be current:
> > early-ish boot crashes are probably the best example, but even beyond
> > that there are lots of varieties of kernel brokenness that can prevent
> > the database from being updated. Doing it explicitly this way ensures
> > that we always know exactly what version emitted a given message.
> >
> > Doing it in printk() itself rather than extended netconsole ends up
> > being much simpler, and has the advantage that future extended console
> > implementations will be able to benefit from this as well.
> >
> > Signed-off-by: Calvin Owens <calvinowens@fb.com>
> > ---
> >  init/Kconfig           |  8 ++++++++
> >  kernel/printk/printk.c | 14 ++++++++++++++
> >  2 files changed, 22 insertions(+)
> 
> I don't think adding a new config option is appropriate.
> How about adding a log format string tunable to netconsole?

Like a generic way to append arbitrary key/val to the dict? It would still
need to be configured via the kernel cmdline or at compile time: I can't
rely getting to userspace.

I had steered away from something more general like that because it
didn't seem worth the trobule, but I can certainly go that route: what
sort of format specifiers would we want? I only care about UTS_RELEASE,
but I suppose UTS_NODENAME and UTS_VERSION might be handy in some cases?

Thanks,
Calvin

> -- 
> Thanks,
> //richard