Re: [Qemu-devel] [PATCH 1/3] rcu: Add automatically released rcu_read_lock variant

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Daniel P. Berrangé (berrange@redhat.com) wrote:
> On Wed, Sep 11, 2019 at 06:04:23PM +0100, Dr. David Alan Gilbert wrote:
> > * Daniel P. Berrangé (berrange@redhat.com) wrote:
> > > On Wed, Sep 11, 2019 at 05:42:00PM +0100, Dr. David Alan Gilbert (git) wrote:
> > > > From: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
> > > > 
> > > > RCU_READ_LOCK_AUTO takes the rcu_read_lock  and then uses glib's
> > > > g_auto infrastrcture (and thus whatever the compilers hooks are) to
> > > > release it on all exits of the block.
> > > > 
> > > > Note this macro has a variable declaration in, and hence is not in
> > > > a while loop.
> > > > 
> > > > Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
> > > > ---
> > > >  include/qemu/rcu.h | 12 ++++++++++++
> > > >  1 file changed, 12 insertions(+)
> > > > 
> > > > diff --git a/include/qemu/rcu.h b/include/qemu/rcu.h
> > > > index 22876d1428..6a25b27d28 100644
> > > > --- a/include/qemu/rcu.h
> > > > +++ b/include/qemu/rcu.h
> > > > @@ -154,6 +154,18 @@ extern void call_rcu1(struct rcu_head *head, RCUCBFunc *func);
> > > >        }),                                                                \
> > > >        (RCUCBFunc *)g_free);
> > > >  
> > > > +typedef char rcu_read_auto_t;
> > > > +static inline void rcu_read_auto_unlock(rcu_read_auto_t *r)
> > > > +{
> > > > +  rcu_read_unlock();
> > > > +}
> > > > +
> > > > +G_DEFINE_AUTO_CLEANUP_CLEAR_FUNC(rcu_read_auto_t, rcu_read_auto_unlock)
> > > >
> > > > +#define RCU_READ_LOCK_AUTO g_auto(rcu_read_auto_t) \
> > > > +    _rcu_read_auto = 'x'; \
> > > > +    rcu_read_lock();
> > > > +
> > > 
> > > Functionally this works, but my gut feeling would be to follow
> > > the design of GMutexLocker as-is:
> > > 
> > >   https://developer.gnome.org/glib/stable/glib-Threads.html#g-mutex-locker-new
> > > 
> > > so you get a use pattern of
> > > 
> > >   g_autoptr(rcu_read_locker) locker = rcu_read_locker_new();
> > > 
> > > This makes it explicit that the code is creating a variable here, which
> > > in turns means it is clear to force unlock early with
> > > 
> > >   g_clear_pointer(&locker, rcu_read_locker_free)
> > 
> > The difference compared to the g-mutex-locker is that I don't have
> > another object to use as my pointer; that uses the address of the GMutex
> > as the dummy pointer value.  I did try an experiment with g_autoptr
> > and found that it did need to return a non-NULL value for it to work,
> > which then lead me to think what value to use - while it seems to work
> > if I return (void *)1 it makes me nervous.
> 
> Yeah, '(void*)1' would have been what I'd pick. The only thing that the
> value is used for is to pass to the rcu_read_locker_free() function
> which ignores it, which seems safe enough.

glib seems to be at least checking it; if you pass NULL the free'r
doesn't get called; so it worries me that we'd be relying on the current
definition.

Dave

> Regards,
> Daniel
> -- 
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK