On Mon, Apr 08, 2019 at 10:52:13AM -0700, Paul E. McKenney wrote:
> On Sat, Apr 06, 2019 at 02:17:05AM +0000, Joel Fernandes wrote:
> > On Thu, Apr 04, 2019 at 01:10:39PM -0700, Paul E. McKenney wrote:
> > > On Fri, Mar 29, 2019 at 10:05:55AM -0400, Joel Fernandes (Google) wrote:
> > > > Document similar real world examples in the kernel corresponding to the
> > > > second and third code snippets. Also correct an issue in
> > > > release_referenced() in the code snippet example.
> > > >
> > > > Cc: oleg@xxxxxxxxxx
> > > > Cc: jannh@xxxxxxxxxx
> > > > Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > >
> > > Good catch, thank you!
> > >
> > > As usual, I could not resist doing a bit of wordsmithing. Please let me
> > > know if I messed anything up in the version shown below.
> > >
> > > Thanx, Paul
> > >
> > > ------------------------------------------------------------------------
> > >
> > > commit adcd92c0ab303b57b28a3cd097bd9ece824c14f6
> > > Author: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > > Date: Fri Mar 29 10:05:55 2019 -0400
> > >
> > > doc/rcuref: Document real world examples in kernel
> > >
> > > Document similar real world examples in the kernel corresponding to the
> > > second and third code snippets. Also correct an issue in
> > > release_referenced() in the code snippet example.
> > >
> > > Cc: oleg@xxxxxxxxxx
> > > Cc: jannh@xxxxxxxxxx
> > > Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > > [ paulmck: Do a bit of wordsmithing. ]
> > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxx>
> > >
> > > diff --git a/Documentation/RCU/rcuref.txt b/Documentation/RCU/rcuref.txt
> > > index 613033ff2b9b..c0bab7fb57e7 100644
> > > --- a/Documentation/RCU/rcuref.txt
> > > +++ b/Documentation/RCU/rcuref.txt
> > > @@ -12,6 +12,7 @@ please read on.
> > > Reference counting on elements of lists which are protected by traditional
> > > reader/writer spinlocks or semaphores are straightforward:
> > >
> > > +CODE LISTING A:
> > > 1. 2.
> > > add() search_and_reference()
> > > { {
> > > @@ -28,7 +29,8 @@ add() search_and_reference()
> > > release_referenced() delete()
> > > { {
> > > ... write_lock(&list_lock);
> > > - atomic_dec(&el->rc, relfunc) ...
> > > + if(atomic_dec_and_test(&el->rc)) ...
> > > + kfree(el);
> > > ... remove_element
> > > } write_unlock(&list_lock);
> > > ...
> > > @@ -44,6 +46,7 @@ search_and_reference() could potentially hold reference to an element which
> > > has already been deleted from the list/array. Use atomic_inc_not_zero()
> > > in this scenario as follows:
> > >
> > > +CODE LISTING B:
> > > 1. 2.
> > > add() search_and_reference()
> > > { {
> > > @@ -79,6 +82,7 @@ search_and_reference() code path. In such cases, the
> > > atomic_dec_and_test() may be moved from delete() to el_free()
> > > as follows:
> > >
> > > +CODE LISTING C:
> > > 1. 2.
> > > add() search_and_reference()
> > > { {
> > > @@ -114,6 +118,16 @@ element can therefore safely be freed. This in turn guarantees that if
> > > any reader finds the element, that reader may safely acquire a reference
> > > without checking the value of the reference counter.
> > >
> > > +A clear advantage of the RCU-based pattern in listing C over the one
> > > +in listing B is that any call to search_and_reference() that locates
> > > +a given object will succeed in obtaining a reference to that object,
> > > +even given a concurrent invocation of delete() for that same object.
> >
> > This part sounds good to me.
> >
> > > +Similarly, a call to delete() is not delayed even if there are an
> > > +arbitrarily large number of calls to search_and_reference() searching
> > > +for the same object that delete() was invoked on. Instead, all that is
> > > +delayed is the eventual invocation of kfree(), which is usually not a
> > > +problem on modern computer systems, even the small ones.
> > > +
> >
> > small nit:
> > This part is common to both listing B and C right? The delete() is never
> > delayed due to the search_and_reference in either case, and the kfree is what
> > is delayed. My patch was highlighting the difference between the 2
> > listings, but this text says what is common between both listings.
> >
> > As such I am Ok with the changes you made, and thanks for this document in
> > the first place.
>
> Good point! How about the following patch to be merged into the current
> patch?
>
> Thanx, Paul
>
> ------------------------------------------------------------------------
>
> diff --git a/Documentation/RCU/rcuref.txt b/Documentation/RCU/rcuref.txt
> index c0bab7fb57e7..5e6429d66c24 100644
> --- a/Documentation/RCU/rcuref.txt
> +++ b/Documentation/RCU/rcuref.txt
> @@ -122,11 +122,12 @@ A clear advantage of the RCU-based pattern in listing C over the one
> in listing B is that any call to search_and_reference() that locates
> a given object will succeed in obtaining a reference to that object,
> even given a concurrent invocation of delete() for that same object.
> -Similarly, a call to delete() is not delayed even if there are an
> -arbitrarily large number of calls to search_and_reference() searching
> -for the same object that delete() was invoked on. Instead, all that is
> -delayed is the eventual invocation of kfree(), which is usually not a
> -problem on modern computer systems, even the small ones.
> +Similarly, a clear advantage of both listings B and C over listing A is
> +that a call to delete() is not delayed even if there are an arbitrarily
> +large number of calls to search_and_reference() searching for the same
> +object that delete() was invoked on. Instead, all that is delayed is
> +the eventual invocation of kfree(), which is usually not a problem on
> +modern computer systems, even the small ones.
>
> In cases where delete() can sleep, synchronize_rcu() can be called from
> delete(), so that el_free() can be subsumed into delete as follows:
>
This one looks better to me, thanks a lot!
Reviewed-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
thanks,
- Joel
