[Top] [All Lists]

Re: [PATCH V2] xfs: introduce CONFIG_XFS_WARN

To: Chandra Seetharaman <sekharan@xxxxxxxxxx>
Subject: Re: [PATCH V2] xfs: introduce CONFIG_XFS_WARN
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Thu, 25 Apr 2013 08:58:21 +1000
Cc: Brian Foster <bfoster@xxxxxxxxxx>, xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <1366828792.3762.32405.camel@xxxxxxxxxxxxxxxxxx>
References: <1366699138-24055-1-git-send-email-david@xxxxxxxxxxxxx> <5176833C.8070305@xxxxxxxxxx> <20130423210226.GN10481@dastard> <20130424085546.GQ10481@dastard> <1366828792.3762.32405.camel@xxxxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Wed, Apr 24, 2013 at 01:39:52PM -0500, Chandra Seetharaman wrote:
> Hi Dave,
> Since this solution is for production environment, would it be valuable
> to have a sysctl variable to allow enabling/disabling XFS_WARN, as
> opposed to needing to recompile the module afresh ?

The idea is that distros enable it on the debug kernel packages
rather than using CONFIG_XFS_DEBUG=y. Hence if someone has a
problem, they just install the debug kernel and they get all this
additional checking.

There are roughly 1700 ASSERT statements in the XFS code, so
compiling them in unconditionally is a lot of extra code. Adding a
sysctl to make them conditional adds as many branches into the code
than in 99.99999% of calls are never going to evaluate as true. The
convenience of a sysctl is more than outweighed by the additional
overhead for the majortiy of people that don't need to diagnose
problems on their system.

Hence I don't think the overhead of unconditionally compiling in
ASSERT checks is worth while for the majority of users, especially
as most distros ship a debug kernel for exactly this purpose....


Dave Chinner

<Prev in Thread] Current Thread [Next in Thread>