As mentioned in the recent discussion on NFS trace points on the NFS &
SystemTap mailing lists. This patch allows field support staff and
kernel developers debug kernel problems, by enabling them to treat
dprintks as precise trace points rather than syslog spamming tools.
This is a forward ported (from 2.6.16), updated, and split version
of a patch that has been used in SGI's internal development tree for
the last few months. The very first version of this was used about
eighteen months ago when debugging NFS/RDMA, which has an enormous
number of dprintks and no other way to debug it.
Jason Baron suggested I post it here for review and contrast with
his dynamic dprintk feature.
--
Greg Banks, P.Engineer, SGI Australian Software Group.
the brightly coloured sporks of revolution.
I don't speak for SGI.
On Tue, Jan 20, 2009 at 12:29:30PM +1100, Greg Banks wrote:
>
> As mentioned in the recent discussion on NFS trace points on the NFS &
> SystemTap mailing lists. This patch allows field support staff and
> kernel developers debug kernel problems, by enabling them to treat
> dprintks as precise trace points rather than syslog spamming tools.
>
> This is a forward ported (from 2.6.16), updated, and split version
> of a patch that has been used in SGI's internal development tree for
> the last few months. The very first version of this was used about
> eighteen months ago when debugging NFS/RDMA, which has an enormous
> number of dprintks and no other way to debug it.
>
> Jason Baron suggested I post it here for review and contrast with
> his dynamic dprintk feature.
>
yes, these two patch sets are very similar in the problem that they are
addressing. For me, one of the core differences, is that 'dprintk' has
per-debug statement control, while my solution, 'dynamic debug' has a
more per-module focused control. 'dprintk' thus checks a different
variable per-debug line to see if its enabled. On the other hand
'dynamic debug' can check 1 global variable (in the most common cases),
to see if its enabled or not. I think we can layer per-line check on top
of the 1 global variable check and have a more efficient solution that
still allows for fine-grained debugging.
'dprintk' also has a richer user interface, which allows for file, line,
module, and statement control.
Thus, I think Greg and I can work together and combine the best features
of both patches. We will re-post a combined solution.
thanks,
-Jason