On Tue, Dec 03, 2019 at 04:01:46PM +0800, Rong Chen wrote:
> On 12/2/19 5:43 PM, Ard Biesheuvel wrote:
> > On Sun, 1 Dec 2019 at 16:49, kernel test robot <rong.a.chen@intel.com> wrote:
> > > FYI, we noticed the following commit (built with gcc-7):
> > > 
> > > commit: d2d337b185bd2abff262f3cf7de0080b3888e41c ("[RESEND PATCH v4
> > > 08/10] refcount: Consolidate implementations of refcount_t")
> > > url:
> > > https://github.com/0day-ci/linux/commits/Will-Deacon/Rework-REFCOUNT_FULL-using-atomic_fetch_-operations/20191124-052413
> > > 
> > > 
> > > in testcase: ocfs2test
> > > with following parameters:
> > > 
> > >          disk: 1SSD
> > >          test: test-mkfs
> > > 
> > So we went from a success rate of 0 out of 24 to 0 out of 24 by
> > applying that patch. How on earth is that a result that justifies
> > spamming everybody?
> 
> These failures were triggered by ocfs2test test, and all tests were failed
> include parent commit "2ab80bd4ae".

https://lore.kernel.org/lkml/20191119182745.GA11397@willie-the-truck
https://lore.kernel.org/lkml/20190912105640.2l6mtdjmcyyhmyun@willie-the-truck/

The refcount code is doing its job afaict and its the ocfs2 code at fault.

Will