Re: [RFC][PATCH] kprobes: Add separate preempt_disabling for kprobes

[Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>]

(2011/07/01 0:51), Steven Rostedt wrote:
> Kprobes requires preemption to be disabled as it single steps the code
> it replaced with a breakpoint. But because the code that is single
> stepped could be reading the preempt count, the kprobe disabling of the
> preempt count can cause the wrong value to end up as a result. Here's an
> example:
> 
> If we add a kprobe on a inc_preempt_count() call:

BTW, on my tip tree, add_preempt_count (a.k.a. inc_preempt_count())
is marked as __kprobes, so it can not be probed. Is there any change?

Anyway, I'll send the removing preempt_disable from kprobe patch.

Thank you,

> 
> 	[ preempt_count = 0 ]
> 
> 	ld preempt_count, %eax	<<--- trap
> 
> 		<trap>
> 		preempt_disable();
> 		[ preempt_count = 1]
> 		setup_singlestep();
> 		<trap return>
> 
> 	[ preempt_count = 1 ]
> 
> 	ld preempt_count, %eax
> 
> 	[ %eax = 1 ]
> 
> 		<trap>
> 		post_kprobe_handler()
> 			preempt_enable_no_resched();
> 			[ preempt_count = 0 ]
> 		<trap return>
> 
> 	[ %eax = 1 ]
> 
> 	add %eax,1
> 
> 	[ %eax = 2 ]
> 
> 	st %eax, preempt_count
> 
> 	[ preempt_count = 2 ]
> 
> 
> We just caused preempt count to increment twice when it should have only
> incremented once, and this screws everything else up.
> 
> To solve this, I've added a per_cpu variable called
> kprobe_preempt_disabled, that is set by the kprobe code. If it is set,
> the preempt_schedule() will not preempt the code.
> 

-- 
Masami HIRAMATSU
Software Platform Research Dept. Linux Technology Center
Hitachi, Ltd., Yokohama Research Laboratory
E-mail: masami.hiramatsu.pt@hitachi.com