From 23332c2e9da16b717f18fd95bc90211a663aa6ee Mon Sep 17 00:00:00 2001 From: Tong Li Date: Fri, 3 Feb 2006 03:04:09 -0800 Subject: [PATCH] [PATCH] OProfile: fixed x86_64 incorrect kernel call graphs Fix the problem in kernel 2.6.15.1 (and early versions) that OProfile on x86_64 does not correctly collect the stack traces for kernel functions. The original code in valid_kernel_stack() in arch/i386/oprofile/backtrace.c assumes that the frame pointer (headaddr) should be greater than stack (i.e., regs). This assumption is wrong for x86_64 because NMIs in x86_64 use a seperate stack different from the kernel stack. Therefore, the variable stack now points to some location on the NMI stack, which turns out to be at a higher address than the frame pointer (headaddr) on the kernel stack. The correct comparison here should be between headaddr and regs->rsp for x86_64. Signed-off-by: Tong Li Cc: John Levon Cc: Philippe Elie Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- arch/i386/oprofile/backtrace.c | 19 +++++++++++++++++-- 1 file changed, 17 insertions(+), 2 deletions(-) diff --git a/arch/i386/oprofile/backtrace.c b/arch/i386/oprofile/backtrace.c index 21654be3f73..acc18138fb2 100644 --- a/arch/i386/oprofile/backtrace.c +++ b/arch/i386/oprofile/backtrace.c @@ -49,7 +49,9 @@ dump_backtrace(struct frame_head * head) * | stack | * --------------- saved regs->ebp value if valid (frame_head address) * . . - * --------------- struct pt_regs stored on stack (struct pt_regs *) + * --------------- saved regs->rsp value if x86_64 + * | | + * --------------- struct pt_regs * stored on stack if 32-bit * | | * . . * | | @@ -57,13 +59,26 @@ dump_backtrace(struct frame_head * head) * | | * | | \/ Lower addresses * - * Thus, &pt_regs <-> stack base restricts the valid(ish) ebp values + * Thus, regs (or regs->rsp for x86_64) <-> stack base restricts the + * valid(ish) ebp values. Note: (1) for x86_64, NMI and several other + * exceptions use special stacks, maintained by the interrupt stack table + * (IST). These stacks are set up in trap_init() in + * arch/x86_64/kernel/traps.c. Thus, for x86_64, regs now does not point + * to the kernel stack; instead, it points to some location on the NMI + * stack. On the other hand, regs->rsp is the stack pointer saved when the + * NMI occurred. (2) For 32-bit, regs->esp is not valid because the + * processor does not save %esp on the kernel stack when interrupts occur + * in the kernel mode. */ #ifdef CONFIG_FRAME_POINTER static int valid_kernel_stack(struct frame_head * head, struct pt_regs * regs) { unsigned long headaddr = (unsigned long)head; +#ifdef CONFIG_X86_64 + unsigned long stack = (unsigned long)regs->rsp; +#else unsigned long stack = (unsigned long)regs; +#endif unsigned long stack_base = (stack & ~(THREAD_SIZE - 1)) + THREAD_SIZE; return headaddr > stack && headaddr < stack_base; -- 2.41.1