Re: [PATCH] ftrace: Allow inline functions not inlined to be traced
From: Mark Rutland <mark.rutland@arm.com>
Date: 2023-06-08 09:51:02
Also in:
live-patching, lkml
On Tue, May 02, 2023 at 04:41:02PM -0400, Steven Rostedt wrote:
From: "Steven Rostedt (Google)" <rostedt@goodmis.org> Over 10 years ago there were many bugs that caused function tracing to crash because some inlined function was not inlined and should not have been traced. This made it hard to debug because when the developer tried to reproduce it, if their compiler still inlined the function, the bug would not trigger. The solution back then was simply to add "notrace" to "inline" which would make sure all functions that are marked inline are never traced even when the compiler decides to not inline them. A lot has changed over the last 10 years. 1) ftrace_test_recursion_trylock() is now used by all ftrace hooks which will prevent the recursive crashes from happening that was caused by inlined functions being traced. 2) noinstr is now used to mark pretty much all functions that would also cause problems if they are traced. Today, it is no longer a problem if an inlined function is not inlined and is traced. Removing notrace from inline has been requested several times over the years. I believe it is now safe to do so. Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Unfortunately, from a quick build-test of arm64 there are a bunch of places that are currently inline that need to be __always_inline for this to be safe. Notably we have a few low-level helpers like is_kernel_in_hyp_mode() that are only inlines, and those get used in the bowels of our entry code before we've restored some HW state (e.g. in arch_nmi_enter()). I'm happy to go audit and fixup arm64, but that will take some work. Maybe it's worth having something like: #ifdef ARCH_CAN_TRACE_INLINE #define __notrace_inline #else #define __notrace_inline notrace #endif ... so that we can opt-in where this is safe, (e.g. on x86)? Thanks, Mark.
quoted hunk ↗ jump to hunk
--- include/linux/compiler_types.h | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h index 547ea1ff806e..c8f23ba1c339 100644 --- a/include/linux/compiler_types.h +++ b/include/linux/compiler_types.h@@ -182,9 +182,8 @@ struct ftrace_likely_data { * externally visible function. This makes extern inline behave as per gnu89 * semantics rather than c99. This prevents multiple symbol definition errors * of extern inline functions at link time. - * A lot of inline functions can cause havoc with function tracing. */ -#define inline inline __gnu_inline __inline_maybe_unused notrace +#define inline inline __gnu_inline __inline_maybe_unused /* * gcc provides both __inline__ and __inline as alternate spellings of@@ -230,7 +229,7 @@ struct ftrace_likely_data { * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67368 * '__maybe_unused' allows us to avoid defined-but-not-used warnings. */ -# define __no_kasan_or_inline __no_sanitize_address notrace __maybe_unused +# define __no_kasan_or_inline __no_sanitize_address __maybe_unused # define __no_sanitize_or_inline __no_kasan_or_inline #else # define __no_kasan_or_inline __always_inline@@ -247,7 +246,7 @@ struct ftrace_likely_data { * disable all instrumentation. See Kconfig.kcsan where this is mandatory. */ # define __no_kcsan __no_sanitize_thread __disable_sanitizer_instrumentation -# define __no_sanitize_or_inline __no_kcsan notrace __maybe_unused +# define __no_sanitize_or_inline __no_kcsan __maybe_unused #else # define __no_kcsan #endif-- 2.39.2