----- Original Message ----- > From: "Caspar Zhang" <[email protected]> > To: "Jan Stancek" <[email protected]> > Cc: "LTP List" <[email protected]> > Sent: Wednesday, 1 August, 2012 10:55:47 AM > Subject: Re: [LTP] [PATCH/RFE 2/2] mm: use new numa_helper > > On 07/31/2012 03:38 PM, Jan Stancek wrote: > > Hi Caspar, > > > > ------- snip -------- > > + ret = get_allowed_nodes(NH_MEMS|NH_CPUS, 2, &nd1, &nd2); > > + switch (ret) { > > + case 0: > > + tst_resm(TINFO, "get node%lu.", nd2); > > + return nd2; > > + case -3: > > + /* > > + * for unbalanced NUMA systems, at least 1 available node is > > + * required. > > + */ > > + ret = get_allowed_nodes(NH_MEMS|NH_CPUS, 1, &nd1); > > + switch (ret) { > > + case 0: > > + tst_resm(TINFO, "get node%lu.", nd1); > > + return nd1; > > + case -3: > > + tst_brkm(TCONF, cleanup_fn, "require a NUMA system " > > + "that has at least one node with both " > > + "memory and cpu available."); > > + default: > > + tst_brkm(TBROK|TERRNO, cleanup_fn, > > + "3rd get_allowed_nodes"); > > + } > > + } > > + tst_brkm(TBROK|TERRNO, cleanup_fn, "2nd get_allowed_nodes"); > > ------- snip -------- > > > > I'm not sure I follow this snippet. > > So if there are 2+ nodes, it takes second one. If there is just > > one, it will take that one. > > Can't it take always first one? > > It was the original design. Since a non-NUMA system have 1 node, the > first (and the only) node should have been tested already in > ksm01/oom01/etc cases. To increase test coverage, we chose 2nd node > on > NUMA system. > > As to the fallback to 1 node design, if an unbalanced system only > contains 1 available node, we still want to test NUMA in separate > case, > ksm01/oom01/etc cases would probably fail to cover it. Agreed, having separate testcase for NUMA makes sense.
> > Do you think it will affect test coverage if we always test first > node? I'm leaning towards "no". I'm assuming oom01 is using default mem policy, so it can allocate memory from any node (including second one). Issue 2. -------- I noticed you put call to "get_a_numa_node()" to "oom()" and "testoom()", which is called also from oom01. As I understand from your previous email, this test should not be NUMA-aware, correct? <<<test_start>>> tag=oom01 stime=1343713154 cmdline="oom01" contacts="" analysis=exit <<<test_output>>> oom01 0 TINFO : set overcommit_memory to 2 oom01 1 TCONF : require a NUMA system. oom01 2 TCONF : Remaining cases not appropriate for configuration oom01 0 TINFO : set overcommit_memory to 0 <<<execution_status>>> Regards, Jan > > Thanks, > Caspar > > > > Regards, > > Jan > > > > > > ----- Original Message ----- > >> From: "Caspar Zhang" <[email protected]> > >> To: "LTP List" <[email protected]> > >> Sent: Tuesday, 31 July, 2012 4:57:19 AM > >> Subject: [LTP] [PATCH/RFE 2/2] mm: use new numa_helper > >> > >> > >> This patch makes the tests in mem/ dir use numa_helper in > >> libkerntest. > >> > >> Signed-off-by: Caspar Zhang <[email protected]> > >> --- > >> testcases/kernel/mem/cpuset/Makefile | 1 + > >> testcases/kernel/mem/cpuset/cpuset01.c | 16 +++--- > >> testcases/kernel/mem/hugetlb/Makefile.inc | 1 + > >> testcases/kernel/mem/hugetlb/hugemmap/Makefile | 1 + > >> testcases/kernel/mem/include/mem.h | 2 +- > >> testcases/kernel/mem/ksm/Makefile | 1 + > >> testcases/kernel/mem/ksm/ksm02.c | 9 ++-- > >> testcases/kernel/mem/ksm/ksm04.c | 9 ++-- > >> testcases/kernel/mem/lib/Makefile | 1 + > >> testcases/kernel/mem/lib/mem.c | 61 > >> +++++++++++++++++------ > >> testcases/kernel/mem/oom/Makefile | 1 + > >> testcases/kernel/mem/oom/oom02.c | 4 -- > >> testcases/kernel/mem/oom/oom04.c | 4 -- > >> testcases/kernel/mem/swapping/Makefile | 1 + > >> testcases/kernel/mem/thp/Makefile | 1 + > >> testcases/kernel/mem/tunable/Makefile | 1 + > >> 16 files changed, 71 insertions(+), 43 deletions(-) > >> > >> > >> ------------------------------------------------------------------------------ > >> Live Security Virtual Conference > >> Exclusive live event will cover all the ways today's security and > >> threat landscape has changed and how IT managers can respond. > >> Discussions > >> will include endpoint security, mobile security and the latest in > >> malware > >> threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ > >> _______________________________________________ > >> Ltp-list mailing list > >> [email protected] > >> https://lists.sourceforge.net/lists/listinfo/ltp-list > >> > > > ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
