Re: x86: warning in unwind_get_return_address

[Dmitry Vyukov]

On Thu, Jan 5, 2017 at 3:49 PM, Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> On Tue, Dec 27, 2016 at 05:38:59PM +0100, Dmitry Vyukov wrote:
>> On Thu, Dec 22, 2016 at 6:17 AM, Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>> > On Wed, Dec 21, 2016 at 01:46:36PM +0100, Andrey Konovalov wrote:
>> >> On Wed, Dec 21, 2016 at 12:36 AM, Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>> >> >
>> >> > Thanks.  Looking at the stack trace, my guess is that an interrupt hit
>> >> > while running in generated BPF code, and the unwinder got confused
>> >> > because regs->ip points to the generated code.  I may need to disable
>> >> > that warning until we figure out a better solution.
>> >> >
>> >> > Can you share your .config file?
>> >>
>> >> Sure, attached.
>> >
>> > Ok, I was able to recreate with your config.  The culprit was generated
>> > code, as I suspected, though it wasn't BPF, it was a kprobe (created by
>> > dccpprobe_init()).
>> >
>> > I'll make a patch to disable the warning.
>>
>> Hi,
>>
>> I am also seeing the following warnings:
>>
>> [  281.889259] WARNING: kernel stack regs at ffff8801c29a7ea8 in
>> syz-executor8:1302 has bad 'bp' value ffff8801c29a7f28
>> [  833.994878] WARNING: kernel stack regs at ffff8801c4e77ea8 in
>> syz-executor1:13094 has bad 'bp' value ffff8801c4e77f28
>>
>> Can it also be caused by bpf/kprobe?
>
> This is a different warning.  I suspect it's due to unwinding the stack
> of another CPU while it's running, which is still possible in a few
> places.  I'm going to have to disable all these warnings for now.


I also have the following diff locally. These loads trigger episodic
KASAN warnings about stack-of-bounds reads on rcu stall warnings when
it does backtrace of all cpus.
If it looks correct to you, can you please also incorporate it into your patch?


diff --git a/arch/x86/include/asm/stacktrace.h
b/arch/x86/include/asm/stacktrace.h
index a3269c897ec5..d8d4fc66ffec 100644
--- a/arch/x86/include/asm/stacktrace.h
+++ b/arch/x86/include/asm/stacktrace.h
@@ -58,7 +58,7 @@ get_frame_pointer(struct task_struct *task, struct
pt_regs *regs)
        if (task == current)
                return __builtin_frame_address(0);

-       return (unsigned long *)((struct inactive_task_frame
*)task->thread.sp)->bp;
+       return (unsigned long *)READ_ONCE_NOCHECK(((struct
inactive_task_frame *)task->thread.sp)->bp);
 }
 #else
 static inline unsigned long *
diff --git a/arch/x86/kernel/unwind_frame.c b/arch/x86/kernel/unwind_frame.c
index 4443e499f279..f3a225ffa231 100644
--- a/arch/x86/kernel/unwind_frame.c
+++ b/arch/x86/kernel/unwind_frame.c
@@ -162,7 +162,7 @@ bool unwind_next_frame(struct unwind_state *state)
        if (state->regs)
                next_bp = (unsigned long *)state->regs->bp;
        else
-               next_bp = (unsigned long *)*state->bp;
+               next_bp = (unsigned long *)READ_ONCE_NOCHECK(*state->bp);

        /* is the next frame pointer an encoded pointer to pt_regs? */
        regs = decode_frame_pointer(next_bp);