Re: + kbuild-disable-wformat-truncation-warnings-by-default.patch added to -mm tree

[Arnd Bergmann]

On Sun, Aug 20, 2017 at 3:19 PM, Masahiro Yamada
<yamada.masahiro@xxxxxxxxxxxxx> wrote:
> 2017-07-20 16:24 GMT+09:00 Arnd Bergmann <arnd@xxxxxxxx>:
>> - enable all three warnings with "make W=1" in 4.13, but leave them
>>   disabled by default.
>> - backport Linus' patch, plus the follow-up for W=1 to stable kernels,
>>   to allow stable kernels to build cleanly
>> - backport the patches that address any other gcc-7 warnings, as
>>   well as those that are not obvious false-positives to stable kernels
>> - In 4.14+, use my version above and address all int-in-bool-context
>>   and format-overflow warnings, but only use -Wformat-truncation
>>   with make W=1.
>>
>
> Talking about 4.14+, shall we move -Wformat-truncation
> from the top Makefile (always disable) to
> Makefile.extrawarn (enable with W=1) ?

I dropped the ball on this one, sorry. I think we should do this for
all three warnings (format-overflow, format-truncation and
int-in-bool-context) for the time being.

In case of format-truncation, there are countless warnings,
most of them false-postives, so we simply can't enable them
by default.

For -Wformat-overflow, there is one patch that I need to
rewrite, all my other patches are pending for 4.14, see
https://patchwork.kernel.org/patch/9840801/ for the missing
one. This should be trivial to fix. However, enabling
CONFIG_UBSAN_SANITIZE_ALL results in seven additional
false positives. I created an patch for this in
https://pastebin.com/CD7nhRNp but can't submit that as it's
obviously bogus. I reported the gcc bug as
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81592

What we could do there is to disable the warning if
CONFIG_UBSAN_SANITIZE_ALL is turned on (like
we do for -Wmaybe-uninitialized in
CONFIG_UBSAN_SANITIZE_ALL) but leave it on otherwise.

I submitted patches for all -Wint-in-bool-context in arm/arm64/x86
randconfig builds, but there are still six known warnings for which
my patches did not get queued for 4.14.
I have to revisit those all to decide whether we can find an
acceptable workaround in the kernel and enable the warning again
by default, or leave it in W=1 until gcc improves enough.

      Arnd