[PATCH]: add dmesg log symbols to /proc/vmcoreinfo lists

[Neil Horman]

On Tue, Jan 20, 2009 at 09:15:51AM -0500, Vivek Goyal wrote:
> On Mon, Jan 19, 2009 at 04:21:27PM -0500, Neil Horman wrote:
> > Hey all-
> > 	It would be nice to be able to extract the dmesg log from a vmcore file
> > without needing to keep the debug symbols for the running kernel handy all the
> > time.  We have a facility to do this in /proc/vmcore.  This patch adds the
> > log_buf and log_end symbols to the vmcoreinfo area so that tools (like
> > makedumpfile) can easily extract the dmesg logs from a vmcore image.
> > 
> 
> It would be nice to get dmesg log if debug symbols are not around. Can't
> we use System.map for getting symbol addresses? vmcoreinfo had started
> small and seems to be growing now. I am thinking down the line will making
> use of System.map for such cases make sense?
> 

System.map only include exported and public symbols.  log_buf and log_end are
static, and not reported in the System.map.  Its a nice idea, but not
comprehensive for everything in the kernel.

><snip>
> 
> Will it be an issue if we make log_buf and log_end non static and directly
> access these in kexec.c?
> 
I had considered that, but was a bit hesitant to do so, since it exposes the
internal implementation of the dmesg buffer.  In the event someone disables
CONFIG_PRINTK, thats just more code we need to ifdef.  With this implementation
we just stub out the log_buf_setup function, and let that be that.  It seems
more consice to me this way.

Regards
Neil

> Thanks
> Vivek
>