[RFC 1/1] orc: mark it as reliable

[RFC 1/1] orc: mark it as reliable

[Jiri Slaby]

We need a reliable stack unwinder for kernel live patching, but we do
not want to enable frame pointers for performance reasons. So let ORC be
a reliable stack unwinder on x86 as it performs nicely wrt reliability
of traces.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: x86@kernel.org
---

I am sending this as an RFC. Do you still consider ORC to be not-enough
reliable?

 arch/x86/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 063f1e0d51aa..7403267407fc 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -172,7 +172,7 @@ config X86
 	select HAVE_PERF_USER_STACK_DUMP
 	select HAVE_RCU_TABLE_FREE
 	select HAVE_REGS_AND_STACK_ACCESS_API
-	select HAVE_RELIABLE_STACKTRACE		if X86_64 && FRAME_POINTER_UNWINDER && STACK_VALIDATION
+	select HAVE_RELIABLE_STACKTRACE		if X86_64 && (FRAME_POINTER_UNWINDER || ORC_UNWINDER) && STACK_VALIDATION
 	select HAVE_STACK_VALIDATION		if X86_64
 	select HAVE_SYSCALL_TRACEPOINTS
 	select HAVE_UNSTABLE_SCHED_CLOCK
-- 
2.14.2

Re: [RFC 1/1] orc: mark it as reliable

[Josh Poimboeuf]

On Wed, Oct 04, 2017 at 11:23:15AM +0200, Jiri Slaby wrote:
> We need a reliable stack unwinder for kernel live patching, but we do
> not want to enable frame pointers for performance reasons. So let ORC be
> a reliable stack unwinder on x86 as it performs nicely wrt reliability
> of traces.
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Cc: Josh Poimboeuf <jpoimboe@redhat.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: x86@kernel.org
> ---
> 
> I am sending this as an RFC. Do you still consider ORC to be not-enough
> reliable?

Off the top of my head, at least the following is missing:

- save_stack_trace_reliable() assumes that kernel mode pt_regs on the
  stack make the stack trace unreliable.  This is an FP-specific
  assumption which no longer applies for ORC.

- The ORC unwinder needs to set unwind_state.error if it doesn't reach
  all the way to the end (user pt_regs).

-- 
Josh