Re: Q: user_enable_single_step() && update_debugctlmsr()

From: Sebastian Andrzej Siewior
Date: Wed Aug 01 2012 - 10:51:38 EST


On 08/01/2012 04:31 PM, Oleg Nesterov wrote:
And I think you missed my point. I'll try again.
Okay.

We have the GDB process and the (stopped) tracee T. And we have
another task X which have TIF_SINGLESTEP but not TIF_BLOCKSTEP.
To simplify, suppose that X is already TASK_RUNNING but not on rq.

GDB does ptrace(PTRACE_SINGLEBLOCK, T). This sets X->TIF_BLOCKSTEP.
Now suppose that GDB is preempted right after it does
update_debugctlmsr(), and the scheduler choses X as the next task.

Both GDB and X do not have TIF_BLOCKSTEP, so __switch_to_extra()
does not update DEBUGCTLMSR_BTF.

X returns to the user-mode with TIF_SINGLESTEP and TIF_BLOCKSTEP,
the latter is wrong.

No?

Yup, correct, you are right sir. Thank you for trying so hard to make
me see this :)

So a patch like
--- a/arch/x86/kernel/step.c
+++ b/arch/x86/kernel/step.c
@@ -173,8 +173,8 @@ static void enable_step(struct task_struct *child, bool block)
unsigned long debugctl = get_debugctlmsr();

debugctl |= DEBUGCTLMSR_BTF;
- update_debugctlmsr(debugctl);
set_tsk_thread_flag(child, TIF_BLOCKSTEP);
+ update_debugctlmsr(debugctl);
} else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP)) {
unsigned long debugctl = get_debugctlmsr();

should fix the race and _yes_ I also would follow your suggestion to
remove this update_debugctlmsr() here since switch_to() should do this.


Oleg.


Sebastian
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/