[lustre-devel] [PATCH 1/6] staging: lustre: Correct missing newline for CERROR call in sfw_handle_server_rpc

[Joe Perches]

On Sat, 2016-03-12 at 18:32 +0000, Drokin, Oleg wrote:
> On Mar 12, 2016, at 1:23 PM, Joe Perches wrote:
> > On Sat, 2016-03-12 at 13:00 -0500, James Simmons wrote:
> > > From: James Nunez <james.a.nunez at intel.com>
> > > 
> > > This is one of the fixes broken out of patch 10000 that was
> > > missed in the merger. With this fix the CERROR called in
> > > sfw_handle_server_rpc will print out correctly.
> > Speaking of CERROR and logging, it it really useful
> > for each CERROR use to have 2 static structs?
> > 
> > In CERROR -> CDEBUG_LIMIT there is a:
> > 	static struct cfs_debug_limit_state cdls;
> > 	(12 or 16 bytes depending on 32/64 bit arch)
> > 
> > and in CDEBUG_LIMIT -> _CDEBUG
> > 	static struct libcfs_debug_msg_data msgdata;
> > 	(24 or 36 bytes depending on 32/64 bit arch)
> > 
> > That seems a largish bit of data and code to initialize
> > these structs for over a thousand call sites.
> > 
> > Wouldn't a single static suffice?
> Single static would not work because the code is parallel so it'll
> stomp over each other.

Sure, but would that matter in practice?

net_ratelimit() has similar parallelization and it doesn't
seem to matter there.

>  or do you mean to have a common
> structure for every callsite (but instantiated separately)?

That might help a tiny bit.

Some possibly unnecessary bits:

o .msg_cdls
o __FILE__, __func__ and __LINE__ fields have marginal value
o .msg_subsys seems set only to DEBUG_SUBSYSTEM.

> This used to be a local structure in the past, but that
> caused considerable stack growth for some functions, that added
> up along the call chain, and that was the solution we came up with
> that did help.