Messages in this thread | | | Date | Sun, 07 Nov 2021 07:57:55 -0300 | From | Arnaldo Carvalho de Melo <> | Subject | Re: [PATCH v3] perf evsel: Fix missing exclude_{host,guest} setting |
| |
On November 7, 2021 7:25:45 AM GMT-03:00, Ravi Bangoria <ravi.bangoria@amd.com> wrote: > > >On 07-Nov-21 12:54 AM, Arnaldo Carvalho de Melo wrote: >> Em Fri, Nov 05, 2021 at 11:00:29AM -0700, Namhyung Kim escreveu: >>> Hello, >>> >>> On Wed, Nov 3, 2021 at 12:22 AM Ravi Bangoria <ravi.bangoria@amd.com> wrote: >>>> >>>>> The current logic for the perf missing feature has a bug that it can >>>>> wrongly clear some modifiers like G or H. Actually some PMUs don't >>>>> support any filtering or exclusion while others do. But we check it >>>>> as a global feature. >>>> >>>> (Sorry to pitch in bit late) >>>> >>>> AMD has one more problem on a similar line. On AMD, non-precise and >>>> precise sampling are provided by core and IBS pmu respectively. Plus, >>>> core pmu has filtering capability but IBS does not. Perf by default >>>> sets precise_ip=3 and exclude_guest=1 and goes on decreasing precise_ip >>>> with exclude_guest set until perf_event_open() succeeds. This is >>>> causing perf to always fallback to core pmu (non-precise mode) even if >>>> it's perfectly feasible to do precise sampling. Do you guys think this >>>> problem should also be addressed while designing solution for Namhyung's >>>> patch or solve it seperately like below patch: >>>> >>>> ---><--- >>>> >>>> From 48808299679199c39ff737a30a7f387669314fd7 Mon Sep 17 00:00:00 2001 >>>> From: Ravi Bangoria <ravi.bangoria@amd.com> >>>> Date: Tue, 2 Nov 2021 11:01:12 +0530 >>>> Subject: [PATCH] perf/amd/ibs: Don't set exclude_guest by default >>>> >>>> Perf tool sets exclude_guest by default while calling perf_event_open(). >>>> Because IBS does not have filtering capability, it always gets rejected >>>> by IBS PMU driver and thus perf falls back to non-precise sampling. Fix >>>> it by not setting exclude_guest by default on AMD. >>>> >>>> Before: >>>> $ sudo ./perf record -C 0 -vvv true |& grep precise >>>> precise_ip 3 >>>> decreasing precise_ip by one (2) >>>> precise_ip 2 >>>> decreasing precise_ip by one (1) >>>> precise_ip 1 >>>> decreasing precise_ip by one (0) >>>> >>>> After: >>>> $ sudo ./perf record -C 0 -vvv true |& grep precise >>>> precise_ip 3 >>>> decreasing precise_ip by one (2) >>>> precise_ip 2 >>>> >>>> Reported-by: Kim Phillips <kim.phillips@amd.com> >>>> Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com> >>> >>> It'd be nice if it can cover explicit -e cycles:pp as well. Anyway, >> >> Ravi, please consider Namhyung's request, a patch on top as I'm adding >> this already. > >For explicit :pp modifier, evsel->precise_max does not get set and thus perf >does not try with different attr->precise_ip values while exclude_guest set. >So no issue with explicit :pp: > > $ sudo ./perf record -C 0 -e cycles:pp -vvv |& grep "precise_ip\|exclude_guest" > precise_ip 2 > exclude_guest 1 > precise_ip 2 > exclude_guest 1 > switching off exclude_guest, exclude_host > precise_ip 2 > ^C > >Also, with :P modifier, evsel->precise_max gets set but exclude_guest does >not and thus :P also works fine: > > $ sudo ./perf record -C 0 -e cycles:P -vvv |& grep "precise_ip\|exclude_guest" > precise_ip 3 > decreasing precise_ip by one (2) > precise_ip 2 > ^C
Perfect, I'll add this analysis to the cset log message.
- Arnaldo > >Thanks, >Ravi
| |