On 24 March 2017 at 16:05, Kim Phillips kim.phillips@arm.com wrote:
On Fri, 24 Mar 2017 15:44:43 -0600 Mathieu Poirier mathieu.poirier@linaro.org wrote:
On 24 March 2017 at 15:37, Kim Phillips kim.phillips@arm.com wrote:
On Fri, 24 Mar 2017 13:21:41 -0600 Mathieu Poirier mathieu.poirier@linaro.org wrote:
On 24 March 2017 at 13:04, Kim Phillips kim.phillips@arm.com wrote:
and searching for etr in the current documentation, I assume you mean I should execute perf like this?:
root@juno:~# taskset -c 2 /home/kim/git/OpenCSD/tools/perf/perf record -o perf-etm.data -e cs_etm/@20070000.etr/u --per-thread taskset -c 2 ./sort-O3 failed to mmap with 12 (Cannot allocate memory) root@juno:~#
<snip> > > This is using Linus' ToT kernel - the one that doesn't lock up: > > There lies the problem - in mainline the ETR driver doesn't have > support for the perf API. Coincidentally Mike is working on this.
OK. Returning -ENOTSUPP with wording to such effect on dmesg would be more helpful than 'failed to mmap' (although it's admittedly better than a stack trace).
The perf-opencsd-4.11-rc1 has an ancient (and inefficient) ETR driver that does support the perf API.
Odd, because I was told to use ETR for its superior efficiency.
The IP block itself is much better than anything else, that is very true but the current driver implementation isn't. On the flip side what is currently there in perf-opencsd-4.11-rc1 allows people to use the ETR with perf while Mike is working away.
perf-opencsd-4.11-rc1 locks up my machine though:
That is very very odd. I have a Juno R2 coming my way but won't have the opportunity to fire it up until (at least) April 17th.
[ perf record: Woken up 9 times to write data ] Warning: AUX data lost 8 times out of 9!
[ perf record: Captured and wrote
<hard lockup>
Mathieu
Kim