Re: [PATCH 4/5] perf intel-pt: Insert callchain context into synthesized callchains

[Arnaldo Carvalho de Melo]

Em Wed, Oct 31, 2018 at 04:15:26PM +0200, Adrian Hunter escreveu:
> On 31/10/18 3:28 PM, Arnaldo Carvalho de Melo wrote:
> > Em Wed, Oct 31, 2018 at 11:10:42AM +0200, Adrian Hunter escreveu:
> >> In the absence of a fallback, callchains must encode also the callchain
> >> context. Do that now there is no fallback.
> > 
> > So, this one is independent of the first 3 patches, right?
> 
> Yes.  I was just going to test it separately when I noticed I had
> screwed up my earlier testing.  When I re-tested I discovered this patch
> has an off-by-one error:
> 
> diff --git a/tools/perf/util/thread-stack.c b/tools/perf/util/thread-stack.c
> index afdf36852ac8..61a4286a74dc 100644
> --- a/tools/perf/util/thread-stack.c
> +++ b/tools/perf/util/thread-stack.c
> @@ -337,7 +337,7 @@ void thread_stack__sample(struct thread *thread, struct ip_callchain *chain,
>  
>  	last_context = context;
>  
> -	for (i = 2, j = 0; i < sz && j < thread->ts->cnt; i++, j++) {
> +	for (i = 2, j = 1; i < sz && j <= thread->ts->cnt; i++, j++) {
>  		ip = thread->ts->stack[thread->ts->cnt - j].ret_addr;
>  		context = callchain_context(ip, kernel_start);
>  		if (context != last_context) {
> 
> Shall I send V2?

No need, I can fix this here
 
> > So, this one is independent of the first 3 patches, right? Ok, applying
> > it first, I'll relook the first ones next.
> > 
> > - Arnaldo
> >  
> >> Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> >> Cc: stable@xxxxxxxxxxxxxxx # 4.19
> >> ---
> >>  tools/perf/util/intel-pt.c     |  6 +++--
> >>  tools/perf/util/thread-stack.c | 44 +++++++++++++++++++++++++++-------
> >>  tools/perf/util/thread-stack.h |  2 +-
> >>  3 files changed, 40 insertions(+), 12 deletions(-)
> >>
> >> diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
> >> index ffa385a029b3..60732213d16a 100644
> >> --- a/tools/perf/util/intel-pt.c
> >> +++ b/tools/perf/util/intel-pt.c
> >> @@ -759,7 +759,8 @@ static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt,
> >>  	if (pt->synth_opts.callchain) {
> >>  		size_t sz = sizeof(struct ip_callchain);
> >>  
> >> -		sz += pt->synth_opts.callchain_sz * sizeof(u64);
> >> +		/* Add 1 to callchain_sz for callchain context */
> >> +		sz += (pt->synth_opts.callchain_sz + 1) * sizeof(u64);
> >>  		ptq->chain = zalloc(sz);
> >>  		if (!ptq->chain)
> >>  			goto out_free;
> >> @@ -1160,7 +1161,8 @@ static void intel_pt_prep_sample(struct intel_pt *pt,
> >>  
> >>  	if (pt->synth_opts.callchain) {
> >>  		thread_stack__sample(ptq->thread, ptq->chain,
> >> -				     pt->synth_opts.callchain_sz, sample->ip);
> >> +				     pt->synth_opts.callchain_sz + 1,
> >> +				     sample->ip, pt->kernel_start);
> >>  		sample->callchain = ptq->chain;
> >>  	}
> >>  
> >> diff --git a/tools/perf/util/thread-stack.c b/tools/perf/util/thread-stack.c
> >> index c091635bf7dc..afdf36852ac8 100644
> >> --- a/tools/perf/util/thread-stack.c
> >> +++ b/tools/perf/util/thread-stack.c
> >> @@ -310,20 +310,46 @@ void thread_stack__free(struct thread *thread)
> >>  	}
> >>  }
> >>  
> >> +static inline u64 callchain_context(u64 ip, u64 kernel_start)
> >> +{
> >> +	return ip < kernel_start ? PERF_CONTEXT_USER : PERF_CONTEXT_KERNEL;
> >> +}
> >> +
> >>  void thread_stack__sample(struct thread *thread, struct ip_callchain *chain,
> >> -			  size_t sz, u64 ip)
> >> +			  size_t sz, u64 ip, u64 kernel_start)
> >>  {
> >> -	size_t i;
> >> +	u64 context = callchain_context(ip, kernel_start);
> >> +	u64 last_context;
> >> +	size_t i, j;
> >>  
> >> -	if (!thread || !thread->ts)
> >> -		chain->nr = 1;
> >> -	else
> >> -		chain->nr = min(sz, thread->ts->cnt + 1);
> >> +	if (sz < 2) {
> >> +		chain->nr = 0;
> >> +		return;
> >> +	}
> >>  
> >> -	chain->ips[0] = ip;
> >> +	chain->ips[0] = context;
> >> +	chain->ips[1] = ip;
> >> +
> >> +	if (!thread || !thread->ts) {
> >> +		chain->nr = 2;
> >> +		return;
> >> +	}
> >> +
> >> +	last_context = context;
> >> +
> >> +	for (i = 2, j = 0; i < sz && j < thread->ts->cnt; i++, j++) {
> >> +		ip = thread->ts->stack[thread->ts->cnt - j].ret_addr;
> >> +		context = callchain_context(ip, kernel_start);
> >> +		if (context != last_context) {
> >> +			if (i >= sz - 1)
> >> +				break;
> >> +			chain->ips[i++] = context;
> >> +			last_context = context;
> >> +		}
> >> +		chain->ips[i] = ip;
> >> +	}
> >>  
> >> -	for (i = 1; i < chain->nr; i++)
> >> -		chain->ips[i] = thread->ts->stack[thread->ts->cnt - i].ret_addr;
> >> +	chain->nr = i;
> >>  }
> >>  
> >>  struct call_return_processor *
> >> diff --git a/tools/perf/util/thread-stack.h b/tools/perf/util/thread-stack.h
> >> index b7e41c4ebfdd..f97c00a8c251 100644
> >> --- a/tools/perf/util/thread-stack.h
> >> +++ b/tools/perf/util/thread-stack.h
> >> @@ -84,7 +84,7 @@ int thread_stack__event(struct thread *thread, u32 flags, u64 from_ip,
> >>  			u64 to_ip, u16 insn_len, u64 trace_nr);
> >>  void thread_stack__set_trace_nr(struct thread *thread, u64 trace_nr);
> >>  void thread_stack__sample(struct thread *thread, struct ip_callchain *chain,
> >> -			  size_t sz, u64 ip);
> >> +			  size_t sz, u64 ip, u64 kernel_start);
> >>  int thread_stack__flush(struct thread *thread);
> >>  void thread_stack__free(struct thread *thread);
> >>  size_t thread_stack__depth(struct thread *thread);
> >> -- 
> >> 2.17.1
> >