On Wed, Jun 13, 2018 at 12:11:40PM +0100, Robert Walker wrote:
On 12/06/18 02:02, Leo Yan wrote:
On Mon, Jun 11, 2018 at 11:13:48AM -0600, Mathieu Poirier wrote:
On 10 June 2018 at 02:28, Leo Yan leo.yan@linaro.org wrote:
Hi Mathieu,
On Sun, Jun 10, 2018 at 04:19:05PM +0800, Leo Yan wrote:
CS_ETM_TRACE_ON packet itself can give the info that there have a discontinuity in the trace, this patch is to add branch sample for CS_ETM_TRACE_ON packet if it is inserted in the middle of CS_ETM_RANGE packets; as result we can have hint for the trace discontinuity.
I followed your suggestion to prepare the cs-etm fixing patch series (if I didn't miss anything), but I am struggling to add this patch into the fixing patch series or not, though I understand I should keep the patch series as simple as possible for easier reviewing and merging.
Seems to me, the patch 5/6 'perf cs-etm: Generate branch sample when receiving a CS_ETM_TRACE_ON packet' has inserted partial info of CS_ETM_TRACE_ON for samples, e.g. the CS_ETM_TRACE_ON packet::addr has been used to generate branch sample, if without this patch we will miss the branch sample for CS_ETM_TRACE_ON::ip. So I decided to add this patch into fixing series. So my main reason for adding patch 6/6 is to make the fixing to be more complete.
Please let me know if you agree with this? Or have other suggestion. Thanks in advance.
Good day Leo,
Aside from the comment in 1/6 I think the compartmentalisation of this series is just fine as it is. I am personally fine with it and think it is ready for action on the public mailing list again. As I said earlier, you will need to wait for the merge window to close.
Thanks for reviewing, Mathieu. Will fix 1/6 with suggestion.
In the mean time, Mike and Rob - please review this set and see if anything is missing.
Yeah, It's good to get green light or suggestions from Mike and Rob before I send patch series to public mailing list.
The patches look good to me and the instruction sample / branch stack output for AutoFDO is unchanged, so I'm happy.
Thanks a lot for reviewing and testing, Rob.
Are there any automated regression tests run for perf? It would be useful to have a set of trace captures that can be run through perf inject/report/script and the output checked for expected behaviours.
Good question and I also have same question for this, usually I need to manually run 'perf script' command at my side.
I search a bit, perf command has one sub command 'perf test' for sanity testing; and in the kernel source code tools/perf/tests there have some testing cases. But I don't really dig into the testing code. Let's see if others have better suggestion for this.
Thanks, Leo Yan