Re: [PATCH] mm/filemap: add static for function __add_to_page_cache_locked

[Alexei Starovoitov]

On Mon, Dec 7, 2020 at 2:53 PM Michal Kubecek <mkubecek@xxxxxxx> wrote:
>
> On Mon, Dec 07, 2020 at 02:44:22PM -0800, Alexei Starovoitov wrote:
> > On Mon, Dec 7, 2020 at 10:36 AM Justin Forbes <jmforbes@xxxxxxxxxxx> wrote:
> > >
> > > On Mon, Dec 7, 2020 at 2:16 AM Michal Kubecek <mkubecek@xxxxxxx> wrote:
> > > >
> > > > On Thu, Nov 12, 2020 at 08:18:57AM +0800, Alex Shi wrote:
> > > > >
> > > > >
> > > > > 在 2020/11/11 上午3:50, Andrew Morton 写道:
> > > > > > On Tue, 10 Nov 2020 08:39:24 +0530 Souptick Joarder <jrdr.linux@xxxxxxxxx> wrote:
> > > > > >
> > > > > >> On Fri, Nov 6, 2020 at 4:55 PM Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx> wrote:
> > > > > >>>
> > > > > >>> Otherwise it cause gcc warning:
> > > > > >>>           ^~~~~~~~~~~~~~~
> > > > > >>> ../mm/filemap.c:830:14: warning: no previous prototype for
> > > > > >>> ‘__add_to_page_cache_locked’ [-Wmissing-prototypes]
> > > > > >>>  noinline int __add_to_page_cache_locked(struct page *page,
> > > > > >>>               ^~~~~~~~~~~~~~~~~~~~~~~~~~
> > > > > >>
> > > > > >> Is CONFIG_DEBUG_INFO_BTF enabled in your .config ?
> > > > > >
> > > > > > hm, yes.
> > > > >
> > > > > When the config enabled, compiling looks good untill pahole tool
> > > > > used to get BTF info, but I still failed on a right version pahole
> > > > > > 1.16. Sorry.
> > > > >
> > > > > >
> > > > > >>>
> > > > > >>> Signed-off-by: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
> > > > > >>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > > > > >>> Cc: linux-mm@xxxxxxxxx
> > > > > >>> Cc: linux-kernel@xxxxxxxxxxxxxxx
> > > > > >>> ---
> > > > > >>>  mm/filemap.c | 2 +-
> > > > > >>>  1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > >>>
> > > > > >>> diff --git a/mm/filemap.c b/mm/filemap.c
> > > > > >>> index d90614f501da..249cf489f5df 100644
> > > > > >>> --- a/mm/filemap.c
> > > > > >>> +++ b/mm/filemap.c
> > > > > >>> @@ -827,7 +827,7 @@ int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
> > > > > >>>  }
> > > > > >>>  EXPORT_SYMBOL_GPL(replace_page_cache_page);
> > > > > >>>
> > > > > >>> -noinline int __add_to_page_cache_locked(struct page *page,
> > > > > >>> +static noinline int __add_to_page_cache_locked(struct page *page,
> > > > > >>>                                         struct address_space *mapping,
> > > > > >>>                                         pgoff_t offset, gfp_t gfp,
> > > > > >>>                                         void **shadowp)
> > > > > >
> > > > > > It's unclear to me whether BTF_ID() requires that the target symbol be
> > > > > > non-static.  It doesn't actually reference the symbol:
> > > > > >
> > > > > > #define BTF_ID(prefix, name) \
> > > > > >         __BTF_ID(__ID(__BTF_ID__##prefix##__##name##__))
> > > > > >
> > > > >
> > > > > The above usage make me thought BTF don't require the symbol, though
> > > > > the symbol still exist in vmlinux with 'static'.
> > > > >
> > > > > So any comments of this, Alexei?
> >
> > Sorry. I've completely missed this thread.
> > Now I have a hard time finding context in archives.
> > If I understood what's going on the removal of "static" cases issues?
> > Yes. That's expected.
> > noinline alone is not enough to work reliably.
>
> Not removal, commit 3351b16af494 ("mm/filemap: add static for function
> __add_to_page_cache_locked") made the function static which breaks the
> build in btfids phase - but it seems to happen only on some
> architectures. In our case, ppc64, ppc64le and riscv64 are broken,
> x86_64, i586 and s390x succeed. (I made a mistake above, aarch64 did not
> fail - but only because it was not built at all.)
>
> The thread starts with
> http://lkml.kernel.org/r/1604661895-5495-1-git-send-email-alex.shi@xxxxxxxxxxxxxxxxx

Got it. So the above commit is wrong.
The addition of "static" is incorrect here.
Regardless of btf_id generation.
"static noinline" means that the error injection in that spot is unreliable.
Even when bpf is completely compiled out of the kernel.