[PATCH RT] Don't call mcount from vsyscall_fn's

[PATCH RT] Don't call mcount from vsyscall_fn's

[Steven Rostedt]

This bit me in the butt.

I couldn't understand why my init app was segfaulting, with a kernel
address, but a user RIP and RSP.  Well, the RIP I think was bogus, but
the kernel address was always the start of "mcount".  Looking deeper, I
printed out what was in the RSP (even though it was a user stack).  It
ended up showing me that the calling address was from the VDSO area.
Looking even further, I found the offending culprit, which was
vread_hpet.

Looking at the assembly dump, I saw the vread_hpet was calling mcount,
but I could not see it in the code. Nor could I see it in hpet.i (-E
option of compiling).

Well, I guess Ingo is a magician when it comes to compiler tricks, and
has the mcount being called by "every!!" function, unless you add the
"notrace" option.

This patch adds the notrace to vsyscall_fn, so that we don't have user
land apps calling mcount and crashing!

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>

Index: linux-2.6-rt-test/include/asm-x86_64/vsyscall.h
===================================================================
--- linux-2.6-rt-test.orig/include/asm-x86_64/vsyscall.h
+++ linux-2.6-rt-test/include/asm-x86_64/vsyscall.h
@@ -22,7 +22,7 @@ enum vsyscall_num {
 /* Definitions for CONFIG_GENERIC_TIME definitions */
 #define __section_vsyscall_gtod_data __attribute__ \
 	((unused, __section__ (".vsyscall_gtod_data"),aligned(16)))
-#define __vsyscall_fn __attribute__ ((unused,__section__(".vsyscall_fn")))
+#define __vsyscall_fn __attribute__ ((unused,__section__(".vsyscall_fn"))) notrace
 
 #define VGETCPU_RDTSCP	1
 #define VGETCPU_LSL	2

Re: [PATCH RT] Don't call mcount from vsyscall_fn's

[Ingo Molnar]

* Steven Rostedt <rostedt@goodmis.org> wrote:

> This bit me in the butt.
> 
> I couldn't understand why my init app was segfaulting, with a kernel 
> address, but a user RIP and RSP.  Well, the RIP I think was bogus, but 
> the kernel address was always the start of "mcount".  Looking deeper, 
> I printed out what was in the RSP (even though it was a user stack).  
> It ended up showing me that the calling address was from the VDSO 
> area. Looking even further, I found the offending culprit, which was 
> vread_hpet.
> 
> Looking at the assembly dump, I saw the vread_hpet was calling mcount, 
> but I could not see it in the code. Nor could I see it in hpet.i (-E 
> option of compiling).
> 
> Well, I guess Ingo is a magician when it comes to compiler tricks, and 
> has the mcount being called by "every!!" function, unless you add the 
> "notrace" option.
> 
> This patch adds the notrace to vsyscall_fn, so that we don't have user 
> land apps calling mcount and crashing!

doh - applied. Thanks!

	Ingo