Re: [PATCH v3 1/6] perf cs-etm: Fix unsigned variable comparison to zero

[Leo Yan]

Hi Mathieu,

On Fri, Oct 11, 2019 at 02:16:06PM -0600, Mathieu Poirier wrote:
> On Sat, Oct 05, 2019 at 05:16:09PM +0800, Leo Yan wrote:
> > If the u64 variable 'offset' is a negative integer, comparison it with
> > bigger than zero is always going to be true because it is unsigned.
> > Fix this by using s64 type for variable 'offset'.
> > 
> > Signed-off-by: Leo Yan <leo.yan@xxxxxxxxxx>
> > ---
> >  tools/perf/util/cs-etm.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
> > index 4ba0f871f086..4bc2d9709d4f 100644
> > --- a/tools/perf/util/cs-etm.c
> > +++ b/tools/perf/util/cs-etm.c
> > @@ -940,7 +940,7 @@ u64 cs_etm__last_executed_instr(const struct cs_etm_packet *packet)
> >  static inline u64 cs_etm__instr_addr(struct cs_etm_queue *etmq,
> >  				     u64 trace_chan_id,
> >  				     const struct cs_etm_packet *packet,
> > -				     u64 offset)
> > +				     s64 offset)
> 
> In Suzuki's reply there was two choices, 1) move the while(offset > 0) to
> while (offset) or change the type of @offset to an s64.  Here we know offset
> can't be negative because of the 
>         tidq->period_instructions >= etm->instructions_sample_period 
> 
> in function cs_etm__sample().  As such I think option #1 is the right way to
> deal with this rather than changing the type of the variable.

I took sometime to use formulas to prove that 'offset' is possible to
be a negative value :)

Just paste the updated commit log at here for review:

  Pi: period_instructions
  Ie: instrs_executed
  Io: instrs_over
  Ip: instructions_sample_period

  Pi' = Pi + Ie   -> New period_instructions equals to the old
                     period_instructions + instrs_executed
  Io  = Pi' - Ip  -> period_instructions - instructions_sample_period

  offset = Ie - Io - 1
         = Ie - (Pi' - Ip) -1
	 = Ie - (Pi + Ie - Ip) -1
	 = Ip - Pi - 1

In theory, if Ip (instructions_sample_period) is small enough and Pi
(period_instructions) is bigger than Ip, then it will lead to the
negative value for 'offset'.

So let's see below command:

  perf inject --itrace=i1il128 -i perf.data -o perf.data.new

With this command, 'offset' is very easily to be a negative value when
handling packets; this is because if use the inject option 'i1', then
instructions_sample_period equals to 1; so:

  offset = 1 - Pi - 1
         = -Pi

Any Pi bigger than zero leads 'offset' to a negative value.

Thanks,
Leo Yan

> >  {
> >  	if (packet->isa == CS_ETM_ISA_T32) {
> >  		u64 addr = packet->start_addr;
> > @@ -1372,7 +1372,7 @@ static int cs_etm__sample(struct cs_etm_queue *etmq,
> >  		 * sample is reported as though instruction has just been
> >  		 * executed, but PC has not advanced to next instruction)
> >  		 */
> > -		u64 offset = (instrs_executed - instrs_over - 1);
> > +		s64 offset = (instrs_executed - instrs_over - 1);
> >  		u64 addr = cs_etm__instr_addr(etmq, trace_chan_id,
> >  					      tidq->packet, offset);
> >  
> > -- 
> > 2.17.1
> >