On Thu, 2 May 2019 19:31:29 -0400 Steven Rostedt rostedt@goodmis.org wrote:
Digging a little further, I pinpointed it out to being kretprobes. The problem I believe is the use of kernel_stack_pointer() which does some magic on x86_32. kretprobes uses this to hijack the return address of the function (much like the function graph tracer does). I do have code that would allow kretprobes to use the function graph tracer instead, but that's still in progress (almost done!). But still, we should not have this break the use of kernel_stack_pointer() either.
Adding some printks in that code, it looks to be returning "®s->sp" which I think we changed.
This appears to fix it!
-- Steve
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index 4b8ee05dd6ad..600ead178bf4 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -171,8 +171,12 @@ unsigned long kernel_stack_pointer(struct pt_regs *regs) unsigned long sp = (unsigned long)®s->sp; u32 *prev_esp;
- if (context == (sp & ~(THREAD_SIZE - 1))) + if (context == (sp & ~(THREAD_SIZE - 1))) { + /* int3 code adds a gap */ + if (sp == regs->sp - 5*4) + return regs->sp; return sp; + }
prev_esp = (u32 *)(context); if (*prev_esp)