x86/ftrace: use uaccess in atomic context
authorFrédéric Weisbecker <fweisbec@gmail.com>
Wed, 24 Sep 2008 15:31:56 +0000 (16:31 +0100)
committerIngo Molnar <mingo@elte.hu>
Tue, 14 Oct 2008 08:38:16 +0000 (10:38 +0200)
With latest -tip I get this bug:

[   49.439988] in_atomic():0, irqs_disabled():1
[   49.440118] INFO: lockdep is turned off.
[   49.440118] Pid: 2814, comm: modprobe Tainted: G        W 2.6.27-rc7 #4
[   49.440118]  [<c01215e1>] __might_sleep+0xe1/0x120
[   49.440118]  [<c01148ea>] ftrace_modify_code+0x2a/0xd0
[   49.440118]  [<c01148a2>] ? ftrace_test_p6nop+0x0/0xa
[   49.440118]  [<c016e80e>] __ftrace_update_code+0xfe/0x2f0
[   49.440118]  [<c01148a2>] ? ftrace_test_p6nop+0x0/0xa
[   49.440118]  [<c016f190>] ftrace_convert_nops+0x50/0x80
[   49.440118]  [<c016f1d6>] ftrace_init_module+0x16/0x20
[   49.440118]  [<c015498b>] load_module+0x185b/0x1d30
[   49.440118]  [<c01767a0>] ? find_get_page+0x0/0xf0
[   49.440118]  [<c02463c0>] ? sprintf+0x0/0x30
[   49.440118]  [<c034e012>] ? mutex_lock_interruptible_nested+0x1f2/0x350
[   49.440118]  [<c0154eb3>] sys_init_module+0x53/0x1b0
[   49.440118]  [<c0352340>] ? do_page_fault+0x0/0x740
[   49.440118]  [<c0104012>] syscall_call+0x7/0xb
[   49.440118]  =======================

It is because ftrace_modify_code() calls copy_to_user and
copy_from_user.
These functions have been inserted after guessing that there
couldn't be any race condition but copy_[to/from]_user might
sleep and __ftrace_update_code is called with local_irq_saved.

These function have been inserted since this commit:
d5e92e8978fd2574e415dc2792c5eb592978243d:
"ftrace: x86 use copy from user function"

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/kernel/ftrace.c

index 66d900248fc2b38ac17c990c119f70ec125c7446..222507e8157be6713850b6bd178f7c35ae0103e7 100644 (file)
@@ -71,13 +71,13 @@ ftrace_modify_code(unsigned long ip, unsigned char *old_code,
         * No real locking needed, this code is run through
         * kstop_machine, or before SMP starts.
         */
-       if (__copy_from_user(replaced, (char __user *)ip, MCOUNT_INSN_SIZE))
+       if (__copy_from_user_inatomic(replaced, (char __user *)ip, MCOUNT_INSN_SIZE))
                return 1;
 
        if (memcmp(replaced, old_code, MCOUNT_INSN_SIZE) != 0)
                return 2;
 
-       WARN_ON_ONCE(__copy_to_user((char __user *)ip, new_code,
+       WARN_ON_ONCE(__copy_to_user_inatomic((char __user *)ip, new_code,
                                    MCOUNT_INSN_SIZE));
 
        sync_core();