]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
ftrace/sysprof: don't trace the user stack if we are a kernel thread.
authorSoeren Sandmann <sandmann@daimi.au.dk>
Mon, 12 May 2008 03:28:50 +0000 (05:28 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Fri, 23 May 2008 21:59:12 +0000 (23:59 +0200)
Check that current->mm is non-NULL before attempting to trace the user
stack.

Also take depth of the kernel stack into account when comparing
against sample_max_depth.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
kernel/trace/trace_sysprof.c

index fe23d6dba7f1dad2c8fa059d93a0bb79285353de..2301e1e7c60648d7376f430b32e771257973be8a 100644 (file)
@@ -95,13 +95,12 @@ const static struct stacktrace_ops backtrace_ops = {
        .address                = backtrace_address,
 };
 
-static struct pt_regs *
+static int
 trace_kernel(struct pt_regs *regs, struct trace_array *tr,
             struct trace_array_cpu *data)
 {
        struct backtrace_info info;
        unsigned long bp;
-       char *user_stack;
        char *stack;
 
        info.tr = tr;
@@ -119,10 +118,7 @@ trace_kernel(struct pt_regs *regs, struct trace_array *tr,
 
        dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info);
 
-       /* Now trace the user stack */
-       user_stack = ((char *)current->thread.sp0 - sizeof(struct pt_regs));
-
-       return (struct pt_regs *)user_stack;
+       return info.pos;
 }
 
 static void timer_notify(struct pt_regs *regs, int cpu)
@@ -150,32 +146,44 @@ static void timer_notify(struct pt_regs *regs, int cpu)
        __trace_special(tr, data, 0, 0, current->pid);
 
        if (!is_user)
-               regs = trace_kernel(regs, tr, data);
+               i = trace_kernel(regs, tr, data);
+       else
+               i = 0;
 
-       fp = (void __user *)regs->bp;
+       /*
+        * Trace user stack if we are not a kernel thread
+        */
+       if (current->mm && i < sample_max_depth) {
+               regs = (struct pt_regs *)current->thread.sp0 - 1;
 
-       __trace_special(tr, data, 2, regs->ip, 0);
+               fp = (void __user *)regs->bp;
 
-       for (i = 0; i < sample_max_depth; i++) {
-               frame.next_fp = 0;
-               frame.return_address = 0;
-               if (!copy_stack_frame(fp, &frame))
-                       break;
-               if ((unsigned long)fp < regs->sp)
-                       break;
+               __trace_special(tr, data, 2, regs->ip, 0);
 
-               __trace_special(tr, data, 2, frame.return_address,
-                             (unsigned long)fp);
-               fp = frame.next_fp;
-       }
+               while (i < sample_max_depth) {
+                       frame.next_fp = 0;
+                       frame.return_address = 0;
+                       if (!copy_stack_frame(fp, &frame))
+                               break;
+                       if ((unsigned long)fp < regs->sp)
+                               break;
 
-       __trace_special(tr, data, 3, current->pid, i);
+                       __trace_special(tr, data, 2, frame.return_address,
+                                       (unsigned long)fp);
+                       fp = frame.next_fp;
+
+                       i++;
+               }
+
+       }
 
        /*
         * Special trace entry if we overflow the max depth:
         */
        if (i == sample_max_depth)
                __trace_special(tr, data, -1, -1, -1);
+
+       __trace_special(tr, data, 3, current->pid, i);
 }
 
 static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer)