Hi Arnd,
On Sat, 25 Aug 2018 21:53:17 +0200
Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Sat, Aug 25, 2018 at 8:53 PM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> > On Fri, 24 Aug 2018 14:46:16 +0200
> > Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > > On Fri, Aug 24, 2018 at 10:23 AM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> > > >
> > > > On Fri, 24 Aug 2018 13:32:06 +1000
> > > > Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
> > > >
> > > > > Hi all,
> > > > >
> > > > > After merging the origin tree, today's linux-next build (powerpc
> > > > > allyesconfig) produced these warnings:
> > > > >
> > > > > Maybe introduced by commit
> > > > >
> > > > > 6b7dca401cb1 ("tracing: Allow gcov profiling on only ftrace subsystem")
> > > > >
> > > > > I am guessing, but that is the only new thing that affects all of
> > > > > kernel/trace ...
> > > >
> > > > Yes, I agree. But I just followed Documentation/dev-tools/gcov.rst
> > > > to enable profiling in kernel/trace. Hmm, doesn't ppc64 support
> > > > GCOV_PROFILE? (But as far as I can see, ARCH_HAS_GCOV_PROFILE_ALL
> > > > is enabled in arch/powerpc/Kconfig)
> > > >
> > > > Anyway, I'll try to reproduce it.
> > >
> > > Thje same commit causes a link failure on ARM with a randconfig
> > > kernel (see https://pastebin.com/KspjpyKG for the .config):
> >
> > OK, I confirmed that this happened with both gcc-7.3.0 and gcc-8.1.0
> > with your config.
> > I also confirmed that CONFIG_GCOV_PROFILE_ALL causes same issue.
> > (Note that CONFIG_GCOV_PROFILE_ALL depends on !CONFIG_COMPILE_TEST,
> > so allyesconfig disables it always)
> >
> > Hmm, now gcov kernel itself might have a problem on arm and
> > powerpc(on gcc-8)?
>
> Yes, that is very possible. I always force CONFIG_COMPILE_TEST=y
> in my randconfig tests, so I miss that kind of problem.
Finally, I've found that this issue (on arm) can be reproduced
with a combination of config options.
When CONFIG_BUG=n and CONFIG_GCOV_PROFILE_ALL|FTRACE=y on arm,
this happened. If I set CONFIG_BUG=y, I could build the kernel
without these errors.
I also built on x86-64 with CONFIG_BUG=n & CONFIG_GCOV_PROFILE_FTRACE=y,
but it succeeded. I guess arm kernel need to embed a counter near the
code section. So if it is in .text.exit, it caused the error.
On arch/arm/kernel/vmlinux.lds.h:
#if (defined(CONFIG_SMP_ON_UP) && !defined(CONFIG_DEBUG_SPINLOCK)) || \
defined(CONFIG_GENERIC_BUG) || defined(CONFIG_JUMP_LABEL)
#define ARM_EXIT_KEEP(x) x
#define ARM_EXIT_DISCARD(x)
#else
#define ARM_EXIT_KEEP(x)
#define ARM_EXIT_DISCARD(x) x
#endif
So I think we have to add CONFIG_GCOV_KERNEL to there.
Thank you,
--
Masami Hiramatsu <mhiramat@xxxxxxxxxx>
