Re: [patch 1/8] Kprobes - do not use kprobes mutex in arch code

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Ananth N Mavinakayanahalli (ananth@in.ibm.com) wrote:
> On Sat, Jul 14, 2007 at 03:20:02PM -0400, Mathieu Desnoyers wrote:
> > * Christoph Hellwig (hch@infradead.org) wrote:
> > > On Fri, Jul 13, 2007 at 09:21:34PM -0400, Mathieu Desnoyers wrote:
> > > > Remove the kprobes mutex from kprobes.h, since it does not belong there. Also
> > > > remove all use of this mutex in the architecture specific code, replacing it by
> > > > a proper mutex lock/unlock in the architecture agnostic code.
> > > 
> > > This is not very nice for avr32/sparc64 which have a noop arch_remove_kprobe
> > > and now need to take a mutex to do nothing.  Maybe you can find a nice
> > > way to avoid that?
> > > 
> > > Except for this issue making kprobes_mutex static to kprobes.c sounds like
> > > a good improvement.
> > > 
> > 
> > Since only unregister_kprobe() calls arch_remove_kprobe(), and only
> > after having removed the struct kprobe from the kprobes list (while the
> > kprobes mutex is held), I wonder if there is any need to hold the
> > kprobes mutex at all when calling arch_remove_kprobe(). It turns out
> > that only get_insn_slot()/free_insn_slot() (which is in
> > kernel/kprobes.c, but called from arch specific code) seems to really
> > use protection of this mutex.
> 
> Right.
> 
> > Would it make sense to protect the kprobe_insn_pages list with a
> > new kprobe_insn_mutex, nestable in the kprobe_mutex ?
> 
> Do you think it is required after your change to make kprobe_mutex
> static? But yes, for architectures that don't need a arch_remove_kprobe,
> the situation is a bit odd... a mutex to do nothing. IIRC, that was the
> primary reason why we made the mutex visible outside of kernel/kprobes.c
> 

After making the kprobe_mutex static, the only alternative we have is to
protect the arch_remove_kprobe (empty on some architectures) call with
kprobe_mutex, which, as Christoph pointed out, is not so great. Besides,
I think the kprobe_insn_mutex would make more sense than taking a mutex
in the arch-specific arch_remove_kprobe() for a resource that is not
clearly identified.

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68