Re: [PATCH 2/2] ptrace: fix set_task_blockstep()->update_debugctlmsr() logic

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 08/07/2012 05:15 PM, Oleg Nesterov wrote:
> It turns out, original code is even more buggy than I thought.
>
> Ironically, "task != current" case is more difficult and so far
> I do not see how we can handle this case correctly. I'll return
> to this a bit later, currently I am working on other patches.

maybe you could remove the autodectect mode and add helper for uprobe
which disables it.

>> For uprobes we never set the bit, we only need it cleared.
>
> Yes, at least at first step, and probably we will never need more.
>
>> We get here
>> via int 3 and do_debug() already clears TIF_BLOCKSTEP
>
> No, we get here via do_int3(), TIF_BLOCKSTEP is not cleared,

Yes, Sorry. my fault.

>> because the
>> CPU clears the bit in CPU.
>
> I am not sure. The manual says:
>
> 	 If the BTF flag is set when the processor generates a debug
> 	 exception, the processor clears the BTF flag along with the
> 	 TF flag.
>
> but I am not sure "debug exception" also means "breakpoint exception".
>
>
>
> do_debug() does clear TIF_BLOCKSTEP, and "The processor cleared BTF"
> is true in this case. But it is called after single-step.

I was wrong here in regard to do_debug() since do_int3() is correct.
Anyway, I checked it on real hardware and I saw the CPU in do_int3()
with BTF set after executing int3 with TF flag set and the BTF bit.

>
> Oleg.

Sebastian