Skip to contentSkip to conversation

Fix x86 32 bit FRAME_POINTER chasing code

PATCH Applied Arjan van de Ven 4 participants 5 messages Jan 10, 2008 – Jan 14, 2008 x86

lore
Share & tools
lore
#0 lore raw

This patch is simple; I don't know if it's .24 candidate; the bug is pretty bad but not a recent regression, and there is obviously some risk with touching this code.

Subject: Fix x86 32 bit FRAME_POINTER chasing code
From: Arjan van de Ven <[email protected]>

The current x86 32 bit FRAME_POINTER chasing code has a nasty bug in that the EBP tracer doesn't actually update the value of EBP it is tracing, so that the code doesn't actually switch to the irq stack properly.

The result is a truncated backtrace:

WARNING: at timeroops.c:8 kerneloops_regression_test() (Not tainted)
Pid: 0, comm: swapper Not tainted 2.6.24-0.77.rc4.git4.fc9 #1
 [<c040649a>] show_trace_log_lvl+0x1a/0x2f
 [<c0406d41>] show_trace+0x12/0x14
 [<c0407061>] dump_stack+0x6c/0x72
 [<e0258049>] kerneloops_regression_test+0x44/0x46 [timeroops]
 [<c04371ac>] run_timer_softirq+0x127/0x18f
 [<c0434685>] __do_softirq+0x78/0xff
 [<c0407759>] do_softirq+0x74/0xf7
 =======================

This patch fixes the code to update EBP properly, and to check the EIP before printing (as the non-framepointer backtracer does) so that the same test backtrace now looks like this:

WARNING: at timeroops.c:8 kerneloops_regression_test()
Pid: 0, comm: swapper Not tainted 2.6.24-rc7 #4
 [<c0405d17>] show_trace_log_lvl+0x1a/0x2f
 [<c0406681>] show_trace+0x12/0x14
 [<c0406ef2>] dump_stack+0x6a/0x70
 [<e01f6040>] kerneloops_regression_test+0x3b/0x3d [timeroops]
 [<c0426f07>] run_timer_softirq+0x11b/0x17c
 [<c04243ac>] __do_softirq+0x42/0x94
 [<c040704c>] do_softirq+0x50/0xb6
 [<c04242a9>] irq_exit+0x37/0x67
 [<c040714c>] do_IRQ+0x9a/0xaf
 [<c04057da>] common_interrupt+0x2e/0x34
 [<c05807fe>] cpuidle_idle_call+0x52/0x78
 [<c04034f3>] cpu_idle+0x46/0x60
 [<c05fbbd3>] rest_init+0x43/0x45
 [<c070aa3d>] start_kernel+0x279/0x27f
 =======================

This shows that the backtrace goes all the way down to user context now. This bug was found during the port to 64 bit of the frame pointer backtracer.

---
 arch/x86/kernel/traps_32.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/kernel/traps_32.c +3 -1
Index: linux-2.6.24-rc7/arch/x86/kernel/traps_32.c
===================================================================
@@ -124,7 +124,8 @@ static inline unsigned long print_contex
 		unsigned long addr;
 
 		addr = frame->return_address;
-		ops->address(data, addr);
+		if (__kernel_text_address(addr))
+			ops->address(data, addr);
 		/*
 		 * break out of recursive entries (such as
 		 * end_of_stack_stop_unwind_function). Also,
@@ -132,6 +133,7 @@ static inline unsigned long print_contex
 		 * move downwards!
 		 */
 		next = frame->next_frame;
+		ebp = (unsigned long) next;
 		if (next <= frame)
 			break;
 		frame = next;
Signature · 3 lines
If you want to reach me at my work email, use [email protected]
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
Acked-by Applied
#1 lore raw
Quoted 12 lines · Arjan van de Ven

+++ linux-2.6.24-rc7/arch/x86/kernel/traps_32.c

arch/x86/kernel/traps_32.c +2 -1
@@ -124,7 +124,8 @@ static inline unsigned long print_contex
 		unsigned long addr;
 
 		addr = frame->return_address;
-		ops->address(data, addr);
+		if (__kernel_text_address(addr))
+			ops->address(data, addr);
 		/*
 		 * break out of recursive entries (such as
 		 * end_of_stack_stop_unwind_function). Also,

arch/x86/kernel/traps_32.c static inline unsigned long print_contex · line 137

Quoted 5 lines
arch/x86/kernel/traps_32.c +1 -0
@@ -132,6 +133,7 @@ static inline unsigned long print_contex
 		 * move downwards!
 		 */
 		next = frame->next_frame;
+		ebp = (unsigned long) next;
 		if (next <= frame)

thanks, applied. Nice catch!

Quoted 3 lines

This patch is simple; I don't know if it's .24 candidate; the bug is pretty bad but not a recent regression, and there is obviously some risk with touching this code.

it's a 2.6.24.1 candidate i believe. We trigger plenty of various crashes during x86.git maintenance and others hit various crashes in -mm, so by the time .1 is released we'll have it in .25 and can backport it. Most folks/distros will update to 2.6.24.1 very quickly so there's no risk of months loss of quality to kerneloops.org data either.

if there's more than 1-2 weeks to the v2.6.24 release we could merge it right now as well:

because in a week we'll trigger plenty of crashes in -git based x86 trees and will know about any regressions and will be able to reasonably trust it.

Ingo

#2 lore raw
Quoted 7 lines · Ingo Molnar

...
it's a 2.6.24.1 candidate i believe. We trigger plenty of various crashes during x86.git maintenance and others hit various crashes in -mm, so by the time .1 is released we'll have it in .25 and can backport it. Most folks/distros will update to 2.6.24.1 very quickly so there's no risk of months loss of quality to kerneloops.org data either.
...

-stable should not introduce additional regressions, and my personal impression is that it already introduces too many regressions.

Either this patch is considered both important and safe enough for getting into Linus' tree now or it's simply too late for 2.6.24*

Quoted 1 line

Ingo

cu
Adrian

Signature · 4 lines
       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed
#3 lore raw
In reply to Ingo Molnar
Quoted 26 lines · Ingo Molnar
Quoted 12 lines · Arjan van de Ven

+++ linux-2.6.24-rc7/arch/x86/kernel/traps_32.c

arch/x86/kernel/traps_32.c +2 -1
@@ -124,7 +124,8 @@ static inline unsigned long print_contex
 		unsigned long addr;
 
 		addr = frame->return_address;
-		ops->address(data, addr);
+		if (__kernel_text_address(addr))
+			ops->address(data, addr);
 		/*
 		 * break out of recursive entries (such as
 		 * end_of_stack_stop_unwind_function). Also,

arch/x86/kernel/traps_32.c static inline unsigned long print_contex · line 137

Quoted 5 lines
arch/x86/kernel/traps_32.c +1 -0
@@ -132,6 +133,7 @@ static inline unsigned long print_contex
 		 * move downwards!
 		 */
 		next = frame->next_frame;
+		ebp = (unsigned long) next;
 		if (next <= frame)

thanks, applied. Nice catch!

Quoted 3 lines

This patch is simple; I don't know if it's .24 candidate; the bug is pretty bad but not a recent regression, and there is obviously some risk with touching this code.

Quoted 6 lines

it's a 2.6.24.1 candidate i believe. We trigger plenty of various crashes during x86.git maintenance and others hit various crashes in -mm, so by the time .1 is released we'll have it in .25 and can backport it. Most folks/distros will update to 2.6.24.1 very quickly so there's no risk of months loss of quality to kerneloops.org data either.

Using the same logic, why not put it in 2.6.24 and then remove it in 2.6.24.1 if it's broken?

#4 lore raw
Quoted 9 lines · Chuck Ebbert
Quoted 6 lines

it's a 2.6.24.1 candidate i believe. We trigger plenty of various crashes during x86.git maintenance and others hit various crashes in -mm, so by the time .1 is released we'll have it in .25 and can backport it. Most folks/distros will update to 2.6.24.1 very quickly so there's no risk of months loss of quality to kerneloops.org data either.

Using the same logic, why not put it in 2.6.24 and then remove it in 2.6.24.1 if it's broken?

hm, did you understood my .25 reference to mean 2.6.25-final? I meant 2.6.25-rc1. Or if there's a 2.6.24-rc8 planned then we could put it in right now [with the bisectability fixes, i.e. not the original series]. I.e. IMO what we shouldnt do is to put in these fixes without having had _some_ -rc release inbetween.

Ingo