Re: [PATCH 6/6] reliable stack trace support (i386 entry.S annotations)

[Andrew Morton <akpm@osdl.org>]

Andi Kleen <ak@suse.de> wrote:
>
> You probably need newer binutils.

Don't think so.

With this:

diff -puN arch/i386/kernel/entry.S~a arch/i386/kernel/entry.S
--- devel/arch/i386/kernel/entry.S~a	2006-05-26 08:32:04.000000000 -0700
+++ devel-akpm/arch/i386/kernel/entry.S	2006-05-26 08:36:50.000000000 -0700
@@ -733,6 +733,7 @@ nmi_debug_stack_check:
 
 nmi_16bit_stack:
 	/* create the pointer to lss back */
+	CFI_STARTPROC simple
 	pushl %ss
 	pushl %esp
 	movzwl %sp, %esp

That great stream of assembler errors comes down to:

arch/i386/kernel/entry.S: Assembler messages:
arch/i386/kernel/entry.S:858: Error: open CFI at the end of file; missing .cfi_endproc directive
make[1]: *** [arch/i386/kernel/entry.o] Error 1

So there seems to be a missing startproc in there.

With this:

--- devel/arch/i386/kernel/entry.S~a	2006-05-26 08:32:04.000000000 -0700
+++ devel-akpm/arch/i386/kernel/entry.S	2006-05-26 08:38:43.000000000 -0700
@@ -733,6 +733,7 @@ nmi_debug_stack_check:
 
 nmi_16bit_stack:
 	/* create the pointer to lss back */
+	CFI_STARTPROC simple
 	pushl %ss
 	pushl %esp
 	movzwl %sp, %esp
@@ -850,6 +851,8 @@ ENTRY(arch_unwind_init_running)
 ENDPROC(arch_unwind_init_running)
 #endif
 
+	CFI_ENDPROC
+
 .section .rodata,"a"
 #include "syscall_table.S"
 

I get

arch/i386/kernel/entry.S: Assembler messages:
arch/i386/kernel/entry.S:860: Error: invalid sections for operation on `L0' and `L0'

Which is a bit mysterious.

Still, I expect that with a bit of fiddling this patch can be made to work
with older binutils.  I also suspect there's actually something wrong with
it.