Re: fs/dlm/midcomms.c:913:22: sparse: sparse: restricted __le32 degrades to integer

[Alexander Aring]

Hi,

On Tue, Jan 25, 2022 at 5:35 PM Alexander Aring <aahringo@xxxxxxxxxx> wrote:
>
> Hi,
>
> On Mon, Jan 24, 2022 at 5:14 PM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> >
> > On Mon, Jan 24, 2022 at 04:36:55PM -0500, Alexander Aring wrote:
> > > Hi,
> > >
> > > On Mon, Jan 24, 2022 at 3:23 PM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > >
> > > > On Mon, Jan 24, 2022 at 12:41:04PM -0500, Alexander Aring wrote:
> > > > > Hi,
> > > > >
> > > > > On Mon, Jan 24, 2022 at 12:36 PM Alexander Aring <aahringo@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > Hi,
> > > > > >
> > > > > > On Mon, Jan 24, 2022 at 12:21 PM Alexander Aring <aahringo@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > Hi,
> > > > > > >
> > > > > > > On Mon, Jan 24, 2022 at 7:46 AM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > On Sun, Jan 23, 2022 at 01:41:52PM -0500, Alexander Aring wrote:
> > > > > > > > >
> > > > > > > > > I see also:
> > > > > > > > >
> > > > > > > > > fs/dlm/midcomms.c:213:1: sparse: sparse: symbol
> > > > > > > > > '__srcu_struct_nodes_srcu' was not declared. Should it be static?
> > > > > > > > >
> > > > > > > >
> > > > > > > > Why not just do this?  (Untested.  Maybe I don't understand?)
> > > > > > > >
> > > > > > > > diff --git a/include/linux/srcutree.h b/include/linux/srcutree.h
> > > > > > > > index cb1f4351e8ba..a164089abec4 100644
> > > > > > > > --- a/include/linux/srcutree.h
> > > > > > > > +++ b/include/linux/srcutree.h
> > > > > > > > @@ -121,7 +121,7 @@ struct srcu_struct {
> > > > > > > >  #ifdef MODULE
> > > > > > > >  # define __DEFINE_SRCU(name, is_static)                                        \
> > > > > > > >         is_static struct srcu_struct name;                              \
> > > > > > > > -       struct srcu_struct * const __srcu_struct_##name                 \
> > > > > > > > +       is_static struct srcu_struct * const __srcu_struct_##name       \
> > > > > > > >                 __section("___srcu_struct_ptrs") = &name
> > > > > > > >  #else
> > > > > > > >  # define __DEFINE_SRCU(name, is_static)                                        \
> > > > > > > >
> > > > > > >
> > > > > > > I tried it and yes it will fix the issue and introduce another one
> > > > > > > about "is_static struct srcu_struct * const __srcu_struct_##name" is
> > > > > > > unused ("-Wunused-const-variable").
> > > > > > > I added a __maybe_unused after the introduced is_static and it seems
> > > > > > > to fix the introduced issue, now it compiles and sparse is happy. I am
> > > > > > > not sure if this is the right fix?
> > > > > >
> > > > > > it is obviously unused, but it has something to do with
> > > > > > "__section("___srcu_struct_ptrs")" and during module loading it, I
> > > > > > suppose, srcu tries to access it to find whatever needs to be
> > > > > > registered?
> > > > >
> > > > > Sorry, but if this is true then it can't be declared as static... and
> > > > > we are at the beginning again.
> > > >
> > > > Welcome to my world!!!  ;-)
> > > >
> > > > More seriously, thank you for chasing this down.  But would it work to
> > > > add a declaration just before?
> > > >
> > >
> > > only if I add an "extern" in front of the declaration before, so it looks like:
> > >
> > > extern struct srcu_struct * const __srcu_struct_##name;
> > >
> > > (compile and sparse tested only)
> >
> > If that works for everyone, it seems worth persuing.
> >
> > One way to test this is as follows:
> >
> > 1.      Build a kernel with CONFIG_RCU_TORTURE_TEST=m.  Boot this and
> >         type "modprobe rcutorture torture_type=srcu".
> >
> >         If you want to stop the torture test, type "rmmod rcutorture".
> >
> >         This will test DEFINE_SRCU() for the module case.
> >
>
> I tested this case, I still need to do the 2. case. Sorry I am quite
> busy with something else, but I am still working on it.
>

I did the 2. test successful...:
"tools/testing/selftests/rcutorture/bin/kvm.sh --allcpus --duration 3
--configs 'SRCU-N' --trust-make"
but I was required to hack it because my "qemu-system-x86_64" did not
exist, it was "/usr/libexec/qemu-kvm" (it was able to run by just
doing a symlink). I think I already cc'd the right people to report
this issue (rcu subsystem)...

I will send a patch for the export declaration soon as possible.

- Alex