Skip to contentSkip to conversation

i386: break out of recursion in stackframe walk

PATCH In review Ingo Molnar 2 participants 3 messages May 2, 2006 – May 3, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 i386: break out of recursion in stackframe walk

if CONFIG_FRAME_POINTERS is enabled, and one does a dump_stack() during early SMP init, an infinite stackdump and a bootup hang happens:

 [<c0104e7f>] show_trace+0xd/0xf
 [<c0104e96>] dump_stack+0x15/0x17
 [<c01440df>] save_trace+0xc3/0xce
 [<c014527d>] mark_lock+0x8c/0x4fe
 [<c0145df5>] __lockdep_acquire+0x44e/0xaa5
 [<c0146798>] lockdep_acquire+0x68/0x84
 [<c1048699>] _spin_lock+0x21/0x2f
 [<c010d918>] prepare_set+0xd/0x5d
 [<c010daa8>] generic_set_all+0x1d/0x201
 [<c010ca9a>] mtrr_ap_init+0x23/0x3b
 [<c010ada8>] identify_cpu+0x2a7/0x2af
 [<c01192a7>] smp_store_cpu_info+0x2f/0xb4
 [<c01197d0>] start_secondary+0xb5/0x3ec
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
 [...]

due to "end_of_stack_stop_unwind_function" recursing back to itself in the EBP stackframe-walker. So avoid this type of recursion when walking the stack .

arch/i386/kernel/traps.c +6 -0
Index: linux/arch/i386/kernel/traps.c
===================================================================
@@ -150,6 +150,12 @@ static inline unsigned long print_contex
 	while (valid_stack_ptr(tinfo, (void *)ebp)) {
 		addr = *(unsigned long *)(ebp + 4);
 		printed = print_addr_and_symbol(addr, log_lvl, printed);
+		/*
+		 * break out of recursive entries (such as
+		 * end_of_stack_stop_unwind_function):
+	 	 */
+		if (ebp == *(unsigned long *)ebp)
+			break;
 		ebp = *(unsigned long *)ebp;
 	}
 #else
#1 lore raw
Quoted 39 lines · Ingo Molnar (on Tue, 2 May 2006 11:50:34 +0200) wrote:

if CONFIG_FRAME_POINTERS is enabled, and one does a dump_stack() during early SMP init, an infinite stackdump and a bootup hang happens:

[<c0104e7f>] show_trace+0xd/0xf
[<c0104e96>] dump_stack+0x15/0x17
[<c01440df>] save_trace+0xc3/0xce
[<c014527d>] mark_lock+0x8c/0x4fe
[<c0145df5>] __lockdep_acquire+0x44e/0xaa5
[<c0146798>] lockdep_acquire+0x68/0x84
[<c1048699>] _spin_lock+0x21/0x2f
[<c010d918>] prepare_set+0xd/0x5d
[<c010daa8>] generic_set_all+0x1d/0x201
[<c010ca9a>] mtrr_ap_init+0x23/0x3b
[<c010ada8>] identify_cpu+0x2a7/0x2af
[<c01192a7>] smp_store_cpu_info+0x2f/0xb4
[<c01197d0>] start_secondary+0xb5/0x3ec
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4
[<c104ec11>] end_of_stack_stop_unwind_function+0x1/0x4

[...]

due to "end_of_stack_stop_unwind_function" recursing back to itself in the EBP stackframe-walker. So avoid this type of recursion when walking the stack .

Index: linux/arch/i386/kernel/traps.c
===================================================================

arch/i386/kernel/traps.c +0 -0
@@ -150,6 +150,12 @@ static inline unsigned long print_contex
 	while (valid_stack_ptr(tinfo, (void *)ebp)) {
 		addr = *(unsigned long *)(ebp + 4);

arch/i386/kernel/traps.c static inline unsigned long print_contex · line 161

Quoted 10 lines
arch/i386/kernel/traps.c +6 -0
 		printed = print_addr_and_symbol(addr, log_lvl, printed);
+		/*
+		 * break out of recursive entries (such as
+		 * end_of_stack_stop_unwind_function):
+	 	 */
+		if (ebp == *(unsigned long *)ebp)
+			break;
 		ebp = *(unsigned long *)ebp;
 	}
 #else

KDB just limits kernel traces to a maximum of 200 entries, which catches direct as well as indirect recursion. IA64 is notorious for getting loops in its unwind data, sometime looping over three or four functions. Checking for a maximum number of entries is a simple and architecture independent check.

#2 lore raw
Quoted 5 lines · Keith Owens

KDB just limits kernel traces to a maximum of 200 entries, which catches direct as well as indirect recursion. IA64 is notorious for getting loops in its unwind data, sometime looping over three or four functions. Checking for a maximum number of entries is a simple and architecture independent check.

you are right, but in this particular case this doesnt seem to be 'wrong' unwind data, it's more of a special marker of the end of the frame (if i understood it correctly). If it's wrong unwind data then that data should be fixed.

I also agree with adding a limit to catch buggy cases of recursion, as a separate mechanism, independently of this particular bug.

Ingo