[RFA] testsuite tweak for gdb.arch/i386-signal.c

[RFA] testsuite tweak for gdb.arch/i386-signal.c

[Michael Snyder]

[-- Attachment #1: Type: text/plain, Size: 134 bytes --]

Tweak this test so that the new i386 epilogue unwinder won't be
tripped up by a function consisting of the single instruction 'ret'.


[-- Attachment #2: nop.txt --]
[-- Type: text/plain, Size: 1015 bytes --]

2009-07-13  Michael Snyder  <msnyder@vmware.com>

	* gdb.arch/i386-signal.c (sigframe): Add a nop to avoid
	confusing the i386 epilogue unwinder.

Index: gdb.arch/i386-signal.c
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-signal.c,v
retrieving revision 1.4
diff -u -p -r1.4 i386-signal.c
--- gdb.arch/i386-signal.c	3 Jan 2009 05:58:03 -0000	1.4
+++ gdb.arch/i386-signal.c	13 Jul 2009 20:22:52 -0000
@@ -37,6 +37,10 @@ main (void)
    The default CIE created by gas suffices to unwind from an empty
    function.  */
 
+/* Note: to make sure that the Dwarf unwinder gets to handle
+   the frame, we add an extra 'nop' after the label.  Otherwise,
+   the new epilogue unwinder will see the 'ret' and grab the frame.  */
+
 asm(".text\n"
     "    .align 8\n"
     "    .globl setup\n"
@@ -49,5 +53,6 @@ asm(".text\n"
     "    nop\n"
     "    .globl sigframe\n"
     "sigframe:\n"
+    "    nop\n"
     "    ret\n"
     "    .cfi_endproc");

Re: [RFA] testsuite tweak for gdb.arch/i386-signal.c

[Mark Kettenis]

> Date: Mon, 13 Jul 2009 13:23:39 -0700
> From: Michael Snyder <msnyder@vmware.com>
> 
> Index: gdb.arch/i386-signal.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-signal.c,v
> retrieving revision 1.4
> diff -u -p -r1.4 i386-signal.c
> --- gdb.arch/i386-signal.c	3 Jan 2009 05:58:03 -0000	1.4
> +++ gdb.arch/i386-signal.c	13 Jul 2009 20:22:52 -0000
> @@ -37,6 +37,10 @@ main (void)
>     The default CIE created by gas suffices to unwind from an empty
>     function.  */
>  
> +/* Note: to make sure that the Dwarf unwinder gets to handle
> +   the frame, we add an extra 'nop' after the label.  Otherwise,
> +   the new epilogue unwinder will see the 'ret' and grab the frame.  */

Hmm, at some point in the future this epilogue unwinder will no longer
be new.

Re: [RFA] testsuite tweak for gdb.arch/i386-signal.c

[Michael Snyder]

Mark Kettenis wrote:
>> Date: Mon, 13 Jul 2009 13:23:39 -0700
>> From: Michael Snyder <msnyder@vmware.com>
>>
>> Index: gdb.arch/i386-signal.c
>> ===================================================================
>> RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-signal.c,v
>> retrieving revision 1.4
>> diff -u -p -r1.4 i386-signal.c
>> --- gdb.arch/i386-signal.c	3 Jan 2009 05:58:03 -0000	1.4
>> +++ gdb.arch/i386-signal.c	13 Jul 2009 20:22:52 -0000
>> @@ -37,6 +37,10 @@ main (void)
>>     The default CIE created by gas suffices to unwind from an empty
>>     function.  */
>>  
>> +/* Note: to make sure that the Dwarf unwinder gets to handle
>> +   the frame, we add an extra 'nop' after the label.  Otherwise,
>> +   the new epilogue unwinder will see the 'ret' and grab the frame.  */
> 
> Hmm, at some point in the future this epilogue unwinder will no longer
> be new.

Well it'll still be newer than the others, but ok, I'll take out the 
word 'new'.   ;-)

Re: [RFA] testsuite tweak for gdb.arch/i386-signal.c

[Michael Snyder]

Mark Kettenis wrote:
>> Date: Mon, 13 Jul 2009 13:23:39 -0700
>> From: Michael Snyder <msnyder@vmware.com>
>>
>> Index: gdb.arch/i386-signal.c
>> ===================================================================
>> RCS file: /cvs/src/src/gdb/testsuite/gdb.arch/i386-signal.c,v
>> retrieving revision 1.4
>> diff -u -p -r1.4 i386-signal.c
>> --- gdb.arch/i386-signal.c	3 Jan 2009 05:58:03 -0000	1.4
>> +++ gdb.arch/i386-signal.c	13 Jul 2009 20:22:52 -0000
>> @@ -37,6 +37,10 @@ main (void)
>>     The default CIE created by gas suffices to unwind from an empty
>>     function.  */
>>  
>> +/* Note: to make sure that the Dwarf unwinder gets to handle
>> +   the frame, we add an extra 'nop' after the label.  Otherwise,
>> +   the new epilogue unwinder will see the 'ret' and grab the frame.  */
> 
> Hmm, at some point in the future this epilogue unwinder will no longer
> be new.

Checked in with the word "new" removed.