Re: [PATCH] i386: Do backtrace fallback too

Re: [PATCH] i386: Do backtrace fallback too

[Dave Jones]

On Sat, Jul 29, 2006 at 03:00:06AM +0000, Linux Kernel wrote:
 > commit c97d20a6c51067a38f53680d9609b4cf2867d077
 > tree 59867ac01d1b752ba7e520e33f9f84cade6d024e
 > parent b783fd925cdd56d24d164e5bdcb072f2a67aedf4
 > author Andi Kleen <ak@suse.de> Fri, 28 Jul 2006 14:44:57 +0200
 > committer Linus Torvalds <torvalds@g5.osdl.org> Sat, 29 Jul 2006 09:28:00 -0700
 > 
 > [PATCH] i386: Do backtrace fallback too
 > 
 > Similar patch to earlier x86-64 patch. When the dwarf2 unwinder fails
 > dump the left over stack with the old unwinder.
 > 
 > Also some clarifications in the headers.
 > 
 > Signed-off-by: Andi Kleen <ak@suse.de>
 > Signed-off-by: Linus Torvalds <torvalds@osdl.org>
 > 
 >  arch/i386/kernel/traps.c |   17 ++++++++++++++---
 >  1 files changed, 14 insertions(+), 3 deletions(-)

Hmm, this breaks the build for me..

arch/i386/kernel/traps.c: In function 'show_trace_log_lvl':
arch/i386/kernel/traps.c:195: error: invalid type argument of '->'
arch/i386/kernel/traps.c:198: error: invalid type argument of '->'
arch/i386/kernel/traps.c:199: error: invalid type argument of '->'
make[1]: *** [arch/i386/kernel/traps.o] Error 1

(The line numbers are different to mainline due to some unrelated
patches, they point to the UNW_PC/UNW_SP usages),


Also, shouldn't this..

	print_symbol("DWARF2 unwinder stuck at %s\n",
		UNW_PC(info.regs));

be using %p ?

		Dave

-- 
http://www.codemonkey.org.uk

Re: [PATCH] i386: Do backtrace fallback too

[Erik Mouw]

On Sat, Jul 29, 2006 at 03:54:14AM -0400, Dave Jones wrote:
> On Sat, Jul 29, 2006 at 03:00:06AM +0000, Linux Kernel wrote:
>  > commit c97d20a6c51067a38f53680d9609b4cf2867d077

[...]

> Hmm, this breaks the build for me..
> 
> arch/i386/kernel/traps.c: In function 'show_trace_log_lvl':
> arch/i386/kernel/traps.c:195: error: invalid type argument of '->'
> arch/i386/kernel/traps.c:198: error: invalid type argument of '->'
> arch/i386/kernel/traps.c:199: error: invalid type argument of '->'
> make[1]: *** [arch/i386/kernel/traps.o] Error 1
> 
> (The line numbers are different to mainline due to some unrelated
> patches, they point to the UNW_PC/UNW_SP usages),
> 
> 
> Also, shouldn't this..
> 
> 	print_symbol("DWARF2 unwinder stuck at %s\n",
> 		UNW_PC(info.regs));
> 
> be using %p ?

It should be UNWP_PC(&info), just like in the functions above. Here's a
patch:

Signed-off-by: Erik Mouw <erik@harddisk-recovery.com>

diff --git a/arch/i386/kernel/traps.c b/arch/i386/kernel/traps.c
index 3facc8f..017c015 100644
--- a/arch/i386/kernel/traps.c
+++ b/arch/i386/kernel/traps.c
@@ -190,11 +190,11 @@ static void show_trace_log_lvl(struct ta
 		if (unw_ret > 0 && !arch_unw_user_mode(&info)) {
 #ifdef CONFIG_STACK_UNWIND
 			print_symbol("DWARF2 unwinder stuck at %s\n",
-				     UNW_PC(info.regs));
+				     UNW_PC(&info));
 			if (call_trace == 1) {
 				printk("Leftover inexact backtrace:\n");
-				if (UNW_SP(info.regs))
-					stack = (void *)UNW_SP(info.regs);
+				if (UNW_SP(&info))
+					stack = (void *)UNW_SP(&info);
 			} else if (call_trace > 1)
 				return;
 			else


Erik

-- 
+-- Erik Mouw -- www.harddisk-recovery.com -- +31 70 370 12 90 --
| Lab address: Delftechpark 26, 2628 XH, Delft, The Netherlands

Re: [PATCH] i386: Do backtrace fallback too

[Andi Kleen]

On Saturday 29 July 2006 09:54, Dave Jones wrote:
> On Sat, Jul 29, 2006 at 03:00:06AM +0000, Linux Kernel wrote:
>  > commit c97d20a6c51067a38f53680d9609b4cf2867d077
>  > tree 59867ac01d1b752ba7e520e33f9f84cade6d024e
>  > parent b783fd925cdd56d24d164e5bdcb072f2a67aedf4
>  > author Andi Kleen <ak@suse.de> Fri, 28 Jul 2006 14:44:57 +0200
>  > committer Linus Torvalds <torvalds@g5.osdl.org> Sat, 29 Jul 2006
>  > 09:28:00 -0700
>  >
>  > [PATCH] i386: Do backtrace fallback too
>  >
>  > Similar patch to earlier x86-64 patch. When the dwarf2 unwinder fails
>  > dump the left over stack with the old unwinder.
>  >
>  > Also some clarifications in the headers.
>  >
>  > Signed-off-by: Andi Kleen <ak@suse.de>
>  > Signed-off-by: Linus Torvalds <torvalds@osdl.org>
>  >
>  >  arch/i386/kernel/traps.c |   17 ++++++++++++++---
>  >  1 files changed, 14 insertions(+), 3 deletions(-)
>
> Hmm, this breaks the build for me..

Hmm, it definitely builds here. Ah do you have UNWIND_INFO
disabled? 


> 	print_symbol("DWARF2 unwinder stuck at %s\n",
> 		UNW_PC(info.regs));
>
> be using %p ?

Yes good catch.

-Andi

Re: [PATCH] i386: Do backtrace fallback too

[Dave Jones]

On Sat, Jul 29, 2006 at 06:35:54PM +0200, Andi Kleen wrote:

 > >  >  arch/i386/kernel/traps.c |   17 ++++++++++++++---
 > >  >  1 files changed, 14 insertions(+), 3 deletions(-)
 > >
 > > Hmm, this breaks the build for me..
 > 
 > Hmm, it definitely builds here. Ah do you have UNWIND_INFO
 > disabled? 

That was with it enabled iirc, I'll double check and do another
build later (though you may want to look at Erik Mouw's reply)
 
 > > 	print_symbol("DWARF2 unwinder stuck at %s\n",
 > > 		UNW_PC(info.regs));
 > >
 > > be using %p ?
 > 
 > Yes good catch.

The x86-64 equivalent also has an instance of the same bug.

		Dave

-- 
http://www.codemonkey.org.uk

Re: [PATCH] i386: Do backtrace fallback too

[Andi Kleen]

On Saturday 29 July 2006 18:42, Dave Jones wrote:
> On Sat, Jul 29, 2006 at 06:35:54PM +0200, Andi Kleen wrote:
>  > >  >  arch/i386/kernel/traps.c |   17 ++++++++++++++---
>  > >  >  1 files changed, 14 insertions(+), 3 deletions(-)
>  > >
>  > > Hmm, this breaks the build for me..
>  >
>  > Hmm, it definitely builds here. Ah do you have UNWIND_INFO
>  > disabled?
>
> That was with it enabled iirc, I'll double check and do another
> build later (though you may want to look at Erik Mouw's reply)

Ok I fixed it now. Patch with some other patches will come soon.

-Andi

Re: [PATCH] i386: Do backtrace fallback too

[Andi Kleen]

>  > > 	print_symbol("DWARF2 unwinder stuck at %s\n",
>  > > 		UNW_PC(info.regs));
>  > >
>  > > be using %p ?
>  >
>  > Yes good catch.
>
> The x86-64 equivalent also has an instance of the same bug.

Actually on double checking the %s is correct because it's print_symbol

-Andi

Re: [PATCH] i386: Do backtrace fallback too

[Dave Jones]

On Sat, Jul 29, 2006 at 07:03:10PM +0200, Andi Kleen wrote:
 > 
 > >  > > 	print_symbol("DWARF2 unwinder stuck at %s\n",
 > >  > > 		UNW_PC(info.regs));
 > >  > >
 > >  > > be using %p ?
 > >  >
 > >  > Yes good catch.
 > >
 > > The x86-64 equivalent also has an instance of the same bug.
 > 
 > Actually on double checking the %s is correct because it's print_symbol

Ah, of course. Somehow I missed that on two separate readings of the code.

		Dave

-- 
http://www.codemonkey.org.uk