Re: refcount_warn_saturate WARNING (was: Re: cls_flower: Fix the behavior using port ranges with hw-offload)

[Cong Wang]

On Thu, Dec 19, 2019 at 1:01 PM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
>
> Hi Cong,
>
> On Thu, Dec 19, 2019 at 9:50 PM Cong Wang <xiyou.wangcong@xxxxxxxxx> wrote:
> > On Thu, Dec 19, 2019 at 2:12 AM Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > > I still see the below warning on m68k/ARAnyM during boot with v5.5-rc2
> > > and next-20191219.
> > > Reverting commit 8ffb055beae58574 ("cls_flower: Fix the behavior using
> > > port ranges with hw-offload") fixes that.
> > >
> > > As this is networking, perhaps this is seen on big-endian only?
> > > Or !CONFIG_SMP?
> > >
> > > Do you have a clue?
> > > I'm especially worried as this commit is already being backported to stable.
> > > Thanks!
> >
> > I did a quick look at the offending commit, I can't even connect it to
> > any dst refcnt.
> >
> > Do you have any more information? Like what happened before the
> > warning? Does your system use cls_flower filters at all? If so, please
> > share your tc configurations.
>
> No, I don't use clf_flower filters.  This is just a normal old Debian boot,
> where the root file system is being remounted, followed by the warning.
>
> To me, it also looks very strange.  But it's 100% reproducible for me.
> Git bisect pointed to this commit, and reverting it fixes the issue.

Hmm, does the attached patch have any luck to fix it?

Thanks.

Attachment: flow_dissector.diff
Description: Binary data