| Message ID | 1530783438-9860-1-git-send-email-yangx.jy@cn.fujitsu.com |
|---|---|
| State | Accepted |
| Delegated to: | Petr Vorel |
| Headers | show |
| Series | memcg/memcg_subgroup_charge: Fix out of memory | expand |
Hi Xiao, > Since commit f0b9d18, memcg_process maps 33 pages for a process, but > memory.limit_in_bytes is still restricted to 1 page, so that moving > the process into group or subgroup results in oom. > Fixes: f0b9d18 ("memcg/functional: 4.16 kernel updates stat counter in a batch of 33 pages") > Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com> > +++ b/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh ... > testcase_1() > { > - test_subgroup $PAGESIZE $((2*PAGESIZE)) > + test_subgroup $PAGESIZES $((2*PAGESIZES)) > } > testcase_2() > { > - test_subgroup $PAGESIZE $PAGESIZE > + test_subgroup $PAGESIZES $PAGESIZES > } > testcase_3() > { > - test_subgroup $PAGESIZE 0 > + test_subgroup $PAGESIZES 0 > } > run_tests Thanks for your fix, pushed. BTW some of these tests fail on recent kernel even after these 2 fixes. They work well on 3.10, but on 4.x fail at least. Some of the issues has been reported: * memcg_stat_test.sh memcg_stat_test 6 TFAIL: hierarchical_memory_limit is 4096, 8192 expected memcg_stat_test 2 TFAIL: mapped_file is 0, 4096 expected https://github.com/linux-test-project/ltp/issues/326 * memcg_usage_in_bytes_test.sh memcg_usage_in_bytes_test 1 TFAIL: memory.usage_in_bytes is 4325376, 4194304 expected https://github.com/linux-test-project/ltp/issues/326 * memcg_use_hierarchy_test.sh memcg_use_hierarchy_test 2 TFAIL: echo 1 > memory.use_hierarchy passed unexpectedly https://github.com/linux-test-project/ltp/issues/93 Kind regards, Petr
On 7/4/18 7:07 AM, Petr Vorel wrote: > Hi Xiao, > >> Since commit f0b9d18, memcg_process maps 33 pages for a process, but >> memory.limit_in_bytes is still restricted to 1 page, so that moving >> the process into group or subgroup results in oom. >> Fixes: f0b9d18 ("memcg/functional: 4.16 kernel updates stat counter in a batch of 33 pages") >> Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com> >> +++ b/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh > ... >> testcase_1() >> { >> - test_subgroup $PAGESIZE $((2*PAGESIZE)) >> + test_subgroup $PAGESIZES $((2*PAGESIZES)) >> } >> testcase_2() >> { >> - test_subgroup $PAGESIZE $PAGESIZE >> + test_subgroup $PAGESIZES $PAGESIZES >> } >> testcase_3() >> { >> - test_subgroup $PAGESIZE 0 >> + test_subgroup $PAGESIZES 0 >> } >> run_tests > Thanks for your fix, pushed. > > BTW some of these tests fail on recent kernel even after these 2 fixes. > They work well on 3.10, but on 4.x fail at least. > > Some of the issues has been reported: > * memcg_stat_test.sh > memcg_stat_test 6 TFAIL: hierarchical_memory_limit is 4096, 8192 expected > memcg_stat_test 2 TFAIL: mapped_file is 0, 4096 expected > https://github.com/linux-test-project/ltp/issues/326 > > * memcg_usage_in_bytes_test.sh > memcg_usage_in_bytes_test 1 TFAIL: memory.usage_in_bytes is 4325376, 4194304 expected > https://github.com/linux-test-project/ltp/issues/326 > > * memcg_use_hierarchy_test.sh > memcg_use_hierarchy_test 2 TFAIL: echo 1 > memory.use_hierarchy passed unexpectedly > https://github.com/linux-test-project/ltp/issues/93 The last one sounds unrelated to those two fixes. Yang > > > Kind regards, > Petr >
Hi Yang, [snip] > > BTW some of these tests fail on recent kernel even after these 2 fixes. > > They work well on 3.10, but on 4.x fail at least. > > Some of the issues has been reported: > > * memcg_stat_test.sh > > memcg_stat_test 6 TFAIL: hierarchical_memory_limit is 4096, 8192 expected > > memcg_stat_test 2 TFAIL: mapped_file is 0, 4096 expected > > https://github.com/linux-test-project/ltp/issues/326 > > * memcg_usage_in_bytes_test.sh > > memcg_usage_in_bytes_test 1 TFAIL: memory.usage_in_bytes is 4325376, 4194304 expected > > https://github.com/linux-test-project/ltp/issues/326 > > * memcg_use_hierarchy_test.sh > > memcg_use_hierarchy_test 2 TFAIL: echo 1 > memory.use_hierarchy passed unexpectedly > > https://github.com/linux-test-project/ltp/issues/93 > The last one sounds unrelated to those two fixes. Yes, it's just related to memcg_* tests. > Yang Kind regards, Petr
diff --git a/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh b/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh index 6b97407..9b11f7b 100755 --- a/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh +++ b/testcases/kernel/controllers/memcg/functional/memcg_subgroup_charge.sh @@ -33,17 +33,17 @@ TST_TOTAL=3 # Test that group and subgroup have no relationship testcase_1() { - test_subgroup $PAGESIZE $((2*PAGESIZE)) + test_subgroup $PAGESIZES $((2*PAGESIZES)) } testcase_2() { - test_subgroup $PAGESIZE $PAGESIZE + test_subgroup $PAGESIZES $PAGESIZES } testcase_3() { - test_subgroup $PAGESIZE 0 + test_subgroup $PAGESIZES 0 } run_tests
Since commit f0b9d18, memcg_process maps 33 pages for a process, but memory.limit_in_bytes is still restricted to 1 page, so that moving the process into group or subgroup results in oom. Fixes: f0b9d18 ("memcg/functional: 4.16 kernel updates stat counter in a batch of 33 pages") Signed-off-by: Xiao Yang <yangx.jy@cn.fujitsu.com> --- .../kernel/controllers/memcg/functional/memcg_subgroup_charge.sh | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)