Skip to contentSkip to conversation

IRQ sharing: BUG: spinlock lockup on CPU#0

DISCUSSION Chuck Ebbert 2 participants 2 messages Jun 11, 2006

lore
Share & tools
lore
#0 lore raw

In-Reply-To: <[email protected]>

Quoted 5 lines · Keith Chew

We have updated all the 10 PCs kernel to 2.6.16.18. We only have 1 of them running in IOAPIC mode (ie local APIC mode support), the rest are in XT-PIC. The APIC machine crashed after 24 hours of operation. Below is the stack trace. Is this related to the IO APIC, or should we be worried about the XT-PIC machines too?

Something has corrupted memory. thread_info->task points to a task_struct at ed103560, but that task_struct->thread_info contains 00000000, its command-line field (comm) contains junk and its pid is -1. This is very hard to diagnose. Are you using 8K stacks? Stack overflow is one possible cause, the other likely one is random memory scribbles.

I've been playing with this patch but it's only boot tested on one config. It should catch this kind of corruption earlier but I can't easily test that.

Whether IO-APIC caused this bug or not, it's hard to say...

include/asm-i386/current.h +15 -0
@@ -3,6 +3,19 @@
 
 #include <linux/thread_info.h>
 
+#define CONFIG_PARANOID
+#ifdef CONFIG_PARANOID
+
+/* must be a macro or things will get ugly */
+#define get_current()						\
+({								\
+	struct task_struct *task = current_thread_info()->task;	\
+	BUG_ON(task->thread_info != current_thread_info());	\
+	task;							\
+})
+
+#else
+
 struct task_struct;
 
 static __always_inline struct task_struct * get_current(void)
@@ -10,6 +23,8 @@ static __always_inline struct task_struc
 	return current_thread_info()->task;
 }
  
+#endif
+
 #define current get_current()
 
 #endif /* !(_I386_CURRENT_H) */
Signature · 1 line
Chuck
#1 lore raw

Hi Chuck

Quoted 2 lines

Whether IO-APIC caused this bug or not, it's hard to say...

We tested it with pci=noacpi, and it has been stable for 36 hours now. It looks like it has something to do with that.

Regards
Keith