On Wed, Oct 16, 2019 at 10:22:24AM +0200, Jiri Olsa wrote: > hi, > Andi reported that maps cloning is eating lot of memory and > it's probably unnecessary, because they keep the same data. > > This 'maps sharing' seems to save lot of heap for reports with > many forks/cloned mmaps (over 60% in example below). > > Profile kernel build: > > $ perf record make -j 40 > > Get heap profile (tools/perf directory): > > $ <install gperftools> > $ make TCMALLOC=1 > $ HEAPPROFILE=/tmp/heapprof ./perf report -i perf.data --stdio > out > $ pprof ./perf /tmp/heapprof.000* > > Before: > > (pprof) top > Total: 2335.5 MB > 1735.1 74.3% 74.3% 1735.1 74.3% memdup > 402.0 17.2% 91.5% 402.0 17.2% zalloc > 140.2 6.0% 97.5% 145.8 6.2% map__new > 33.6 1.4% 98.9% 33.6 1.4% symbol__new > 12.4 0.5% 99.5% 12.4 0.5% alloc_event > 6.2 0.3% 99.7% 6.2 0.3% nsinfo__new > 5.5 0.2% 100.0% 5.5 0.2% nsinfo__copy > 0.3 0.0% 100.0% 0.3 0.0% dso__new > 0.1 0.0% 100.0% 0.1 0.0% do_read_string > 0.0 0.0% 100.0% 0.0 0.0% __GI__IO_file_doallocate > > After: > > (pprof) top > Total: 784.5 MB > 385.8 49.2% 49.2% 385.8 49.2% memdup > 285.8 36.4% 85.6% 285.8 36.4% zalloc > 80.4 10.3% 95.9% 83.7 10.7% map__new > 19.1 2.4% 98.3% 19.1 2.4% symbol__new > 6.2 0.8% 99.1% 6.2 0.8% alloc_event > 3.6 0.5% 99.6% 3.6 0.5% nsinfo__new > 3.2 0.4% 100.0% 3.2 0.4% nsinfo__copy > 0.2 0.0% 100.0% 0.2 0.0% dso__new > 0.0 0.0% 100.0% 0.0 0.0% do_read_string > 0.0 0.0% 100.0% 0.0 0.0% elf_fill > > v2 changes: > - rebased to Arnaldo's perf/core > - patch 1 already taken > > Also available in here: > git://git.kernel.org/pub/scm/linux/kernel/git/jolsa/perf.git > perf/map_shared
I rebased to latest perf/core and pushed the branch out jirka
