Re: [RFC][PATCHSET] sanitized pathwalk machinery (v3)

[Al Viro]

On Tue, Mar 03, 2020 at 11:23:39PM -0600, Eric W. Biederman wrote:
> Al Viro <viro@xxxxxxxxxxxxxxxxxx> writes:
> 
> > On Tue, Mar 03, 2020 at 05:48:31PM -0600, Eric W. Biederman wrote:
> >
> >> > I hope it gets serious beating, though - it touches pretty much every
> >> > codepath in pathname resolution.  Is there any way to sic the bots on
> >> > a branch, short of "push it into -next and wait for screams"?
> >> 
> >> Last I looked pushing a branch to kernel.org was enough for the
> >> kbuild bots.  Sending patches to LKML is also enough for those bots.
> >> 
> >> I don't know if that kind of bot is what you need testing your code.
> >
> > Build bots are generally nice, but in this case... pretty much all of
> > the changes are in fs/namei.c, which is not all that sensitive to
> > config/architecture/whatnot.  Sure, something like "is audit enabled?"
> > may affect the build problems, but not much beyond that.
> >
> > What was that Intel-run(?) bot that posts "such-and-such metrics has
> > 42% regression on such-and-such commit" from time to time?
> > <checks>
> > Subject: [locking/qspinlock] 7b6da71157: unixbench.score 8.4% improvement
> > seems to be the latest of that sort,
> > From: kernel test robot <rong.a.chen@xxxxxxxxx>
> >
> > Not sure how much of pathwalk-heavy loads is covered by profiling
> > bots of that sort, unfortunately... ;-/
> 
> Do the xfs-tests cover that sort of thing?
> The emphasis is stress testing the filesystem not the VFS but there is a
> lot of overlap between the two.

I do run xfstests.  But "runs in KVM without visible slowdowns" != "won't
cause them on 48-core bare metal".  And this area (especially when it
comes to RCU mode) can be, er, interesting in that respect.

FWIW, I'm putting together some litmus tests for pathwalk semantics -
one of the things I'd like to discuss at LSF; quite a few codepaths
are simply not touched by anything in xfstests.