Re: [PATCH 3/9] uprobes: allow ignoring of probe hits

[Rabin Vincent <rabin@rab.in>]

On Wed, Oct 17, 2012 at 07:35:10PM +0200, Oleg Nesterov wrote:
> On 10/16, Rabin Vincent wrote:
> > 2012/10/15 Oleg Nesterov <oleg@redhat.com>:
> > > Not sure I understand why we shouldn't call handlers in this case,
> > > but OK, I know nothing about arm.
> >
> > This old discussion about kprobes should be useful:
> >
> > http://lists.infradead.org/pipermail/linux-arm-kernel/2011-March/045755.html
> 
> Thanks... Not sure I understand this discussion...
> 
> And, to clarify, I am not arguing. Just curious.
> 
> So, is this like cmov on x86? And this patch allows to not report if
> the condition is not true? Or there are other issues on arm?

Yes, I guess this is like CMOV on x86.  In the ARM instruction set most
instructions can be conditionally executed.

In order to set the probe on a conditional instruction, we use an
undefined instruction with the same condition as the instruction we
replace.  However, it is implementation defined whether an undefined
instruction with a failing condition code will trigger an undefined
instruction exception or just be executed as a NOP.  So for those
processor implementations where we do get the undefined instruction
exception even for a failing condition code, we have to ignore it in
order to provide consistent behaviour.