Re: [PATCH v2] arm64: ftrace: use function_nocfi for _mcount as well

From: Mark Rutland <mark.rutland@arm.com>
To: Sami Tolvanen <samitolvanen@google.com>
Cc: Sumit Garg <sumit.garg@linaro.org>,
	linux-arm-kernel <linux-arm-kernel@lists.infradead.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Ingo Molnar <mingo@redhat.com>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Will Deacon <will@kernel.org>,
	ben.dai@unisoc.com, Nathan Chancellor <nathan@kernel.org>,
	Kees Cook <keescook@chromium.org>,
	Nick Desaulniers <ndesaulniers@google.com>,
	daniel.thompson@linaro.org, LKML <linux-kernel@vger.kernel.org>,
	llvm@lists.linux.dev
Subject: Re: [PATCH v2] arm64: ftrace: use function_nocfi for _mcount as well
Date: Tue, 5 Oct 2021 16:35:23 +0100	[thread overview]
Message-ID: <20211005153523.GD6678@C02TD0UTHF1T.local> (raw)
In-Reply-To: <CABCJKuesYcGdKLi1YqHP3PU5n6vf-3Q-A+UNyCLzsoJ+0oiKmw@mail.gmail.com>

On Tue, Oct 05, 2021 at 08:20:02AM -0700, Sami Tolvanen wrote:
> Hi Sumit,
> 
> On Tue, Oct 5, 2021 at 5:37 AM Sumit Garg <sumit.garg@linaro.org> wrote:
> >
> > Commit 800618f955a9 ("arm64: ftrace: use function_nocfi for ftrace_call")
> > only fixed address of ftrace_call but address of _mcount needs to be
> > fixed as well. Use function_nocfi() to get the actual address of _mcount
> > function as with CONFIG_CFI_CLANG, the compiler replaces function pointers
> > with jump table addresses which breaks dynamic ftrace as the address of
> > _mcount is replaced with the address of _mcount.cfi_jt.
> >
> > This problem won't apply where the toolchain implements
> > -fpatchable-function-entry as we'll use that in preference to regular -pg,
> > i.e. this won't show up with recent versions of clang.
> >
> > Fixes: 9186ad8e66bab6a1 ("arm64: allow CONFIG_CFI_CLANG to be selected")
> > Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
> > Acked-by: Mark Rutland <mark.rutland@arm.com>
> > ---
> >
> > Changes in v2:
> > - Added fixes tag.
> > - Extended commit description.
> > - Picked up Mark's ack.
> >
> >  arch/arm64/include/asm/ftrace.h | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
> > index 91fa4baa1a93..347b0cc68f07 100644
> > --- a/arch/arm64/include/asm/ftrace.h
> > +++ b/arch/arm64/include/asm/ftrace.h
> > @@ -15,7 +15,7 @@
> >  #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
> >  #define ARCH_SUPPORTS_FTRACE_OPS 1
> >  #else
> > -#define MCOUNT_ADDR            ((unsigned long)_mcount)
> > +#define MCOUNT_ADDR            ((unsigned long)function_nocfi(_mcount))
> >  #endif
> >
> >  /* The BL at the callsite's adjusted rec->ip */
> > --
> > 2.17.1
> >
> 
> Clang >= 10 supports -fpatchable-function-entry and CFI requires Clang
> 12, so I assume this is only an issue if
> CONFIG_DYNAMIC_FTRACE_WITH_REGS is explicitly disabled?

I don't believe it's possible to disable explicitly, since
DYNAMIC_FTRACE_WITH_REGS isn't user selectable, and is def bool y,
depending on HAVE_DYNAMIC_FTRACE_WITH_REGS.

Sumit, have you actually seen a problem, or was this found by
inspection?

If this isn't an issue in practice, we could add the funciton_nocfi()
for consistency, but we should make that clear in the commit message,
and drop the fixes tag.

Thanks,
Mark.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel