On 04/16/2013 02:58 PM, Jan Stancek wrote: > > > > > ----- Original Message ----- >> From: "Zhouping Liu"<[email protected]> >> To: "gmail"<[email protected]> >> Cc: [email protected], "Wanlong >> Gao"<[email protected]>, "Jan Stancek"<[email protected]> >> Sent: Tuesday, 16 April, 2013 6:34:53 AM >> Subject: Re: [LTP] [PATCH] Fix short of nodemask array. >> >> Hi Lans, >> >> ----- Original Message ----- >>> From: "gmail"<[email protected]> >>> To: [email protected] >>> Sent: Monday, March 4, 2013 12:49:45 PM >>> Subject: [LTP] [PATCH] Fix short of nodemask array. >>> >>> In kernel, when the user specified more nodes, e.g, 512 nodes, than >>> supported, e.g, 4 nodes, just check if the non supported part is all >>> zero. If user space just specified the nodemask array with only one >>> element, which makes the check in kernel actually become invalid. >> >> I think the comments here is not clear, you should explain more details >> about the 'one element' or the reason why the checking in kernel will >> be invalid. > > I think what it says is that we are overrunning "nmask", whose length is > shorter than MAXNODES.
Correct. > >> >>> In addition, some test cases in current implement doesn't consider >>> if a node number is greater than sizeof(unsigned long) * 8 - 1. >> >> yes, you are right, we don't care the system which has more than 512 numa >> nodes. >> and the current code only support 64 numa nodes, only one unsigned long node >> mask. >> >> I checked kernel code in RHEL6, CONFIG_NODES_SHIFT=9, so 512 is the biggest >> value to support in RHEL6 for x86_64, and I didn't meet any large NUMA system >> which have 512+ numa nodes so far. >> >>> >>> Signed-off-by: Lans Zhang<[email protected]> >>> --- >>> testcases/kernel/mem/cpuset/cpuset01.c | 6 +++--- >>> testcases/kernel/mem/ksm/ksm02.c | 6 +++--- >>> testcases/kernel/mem/ksm/ksm04.c | 6 +++--- >>> testcases/kernel/mem/lib/mem.c | 6 +++--- >>> 4 files changed, 12 insertions(+), 12 deletions(-) >>> >>> diff --git a/testcases/kernel/mem/cpuset/cpuset01.c >>> b/testcases/kernel/mem/cpuset/cpuset01.c >>> index b473e45..e459306 100644 >>> --- a/testcases/kernel/mem/cpuset/cpuset01.c >>> +++ b/testcases/kernel/mem/cpuset/cpuset01.c >>> @@ -93,7 +93,7 @@ static void testcpuset(void) >>> { >>> int lc; >>> int child, i, status; >>> - unsigned long nmask = 0; >>> + unsigned long nmask[MAXNODES / (8*sizeof(unsigned long))] = { 0 }; >>> char mems[BUFSIZ], buf[BUFSIZ]; >>> >>> read_cpuset_files(CPATH, "cpus", buf); >>> @@ -108,8 +108,8 @@ static void testcpuset(void) >>> tst_brkm(TBROK | TERRNO, cleanup, "fork"); >>> case 0: >>> for (i = 0; i< nnodes; i++) >>> - nmask += 1<< nodes[i]; >>> - if (set_mempolicy(MPOL_BIND,&nmask, MAXNODES) == -1) >>> + nmask[i / (8*sizeof(unsigned long))] |= 1<< (i % >>> (8*sizeof(unsigned >>> long))); >>> + if (set_mempolicy(MPOL_BIND, nmask, MAXNODES) == -1) >>> tst_brkm(TBROK | TERRNO, cleanup, "set_mempolicy"); >>> exit(mem_hog_cpuset(ncpus> 1 ? ncpus : 1)); >>> } >>> diff --git a/testcases/kernel/mem/ksm/ksm02.c >>> b/testcases/kernel/mem/ksm/ksm02.c >>> index be9ff96..9d2d142 100644 >>> --- a/testcases/kernel/mem/ksm/ksm02.c >>> +++ b/testcases/kernel/mem/ksm/ksm02.c >>> @@ -87,7 +87,7 @@ int main(int argc, char *argv[]) >>> int lc; >>> char *msg; >>> int size = 128, num = 3, unit = 1; >>> - unsigned long nmask = 0; >>> + unsigned long nmask[MAXNODES / (8*sizeof(unsigned long))] = { 0 }; >>> unsigned int node; >>> >>> msg = parse_opts(argc, argv, ksm_options, ksm_usage); >>> @@ -95,7 +95,7 @@ int main(int argc, char *argv[]) >>> tst_brkm(TBROK, NULL, "OPTION PARSING ERROR - %s", msg); >>> >>> node = get_a_numa_node(tst_exit); >>> - nmask = 1<< node; >>> + nmask[node / (8*sizeof(unsigned long))] = 1<< (node % >>> (8*sizeof(unsigned >>> long))); >> >> I found '1<< (node % (8*sizeof(unsigned long)))' is used frequently, I >> suggest we >> can create a macro, such as: >> # define GET_NODE_MASK(nid) 1<< (nid % (8*sizeof(unsigned long))) > > I agree, also 'nmask[i / (8*sizeof(unsigned long))]' is there often. > How about something like: > > #define bitsperlong (8 * sizeof(unsigned long)) > void set_bit(unsigned long *b, unsigned int n, unsigned int v) > { > if (v) > b[n / bitsperlong] |= 1UL<< (n % bitsperlong); > else > b[n / bitsperlong]&= ~(1UL<< (n % bitsperlong)); > } > set_bit(nmask, node, 1); Looks better. I will use this style in V2. Thanks, lz > > Regards, > Jan > >> >> also please update the patch, as there's some updates before your patch in >> lib/mem.c, >> which makes some conflict with the latest tree. >> >> Thanks, >> Zhouping >> >>> >>> setup(); >>> >>> @@ -103,7 +103,7 @@ int main(int argc, char *argv[]) >>> Tst_count = 0; >>> check_ksm_options(&size,&num,&unit); >>> >>> - if (set_mempolicy(MPOL_BIND,&nmask, MAXNODES) == -1) { >>> + if (set_mempolicy(MPOL_BIND, nmask, MAXNODES) == -1) { >>> if (errno != ENOSYS) >>> tst_brkm(TBROK | TERRNO, cleanup, >>> "set_mempolicy"); >>> diff --git a/testcases/kernel/mem/ksm/ksm04.c >>> b/testcases/kernel/mem/ksm/ksm04.c >>> index e4aa417..7829afd 100644 >>> --- a/testcases/kernel/mem/ksm/ksm04.c >>> +++ b/testcases/kernel/mem/ksm/ksm04.c >>> @@ -87,7 +87,7 @@ int main(int argc, char *argv[]) >>> int lc; >>> char *msg; >>> int size = 128, num = 3, unit = 1; >>> - unsigned long nmask = 0; >>> + unsigned long nmask[MAXNODES / (8*sizeof(unsigned long))] = { 0 }; >>> unsigned int node; >>> >>> msg = parse_opts(argc, argv, ksm_options, ksm_usage); >>> @@ -95,7 +95,7 @@ int main(int argc, char *argv[]) >>> tst_brkm(TBROK, NULL, "OPTION PARSING ERROR - %s", msg); >>> >>> node = get_a_numa_node(tst_exit); >>> - nmask = 1<< node; >>> + nmask[node / (8*sizeof(unsigned long))] = 1<< (node % >>> (8*sizeof(unsigned >>> long))); >>> >>> setup(); >>> >>> @@ -105,7 +105,7 @@ int main(int argc, char *argv[]) >>> >>> write_memcg(); >>> >>> - if (set_mempolicy(MPOL_BIND,&nmask, MAXNODES) == -1) { >>> + if (set_mempolicy(MPOL_BIND, nmask, MAXNODES) == -1) { >>> if (errno != ENOSYS) >>> tst_brkm(TBROK | TERRNO, cleanup, >>> "set_mempolicy"); >>> diff --git a/testcases/kernel/mem/lib/mem.c >>> b/testcases/kernel/mem/lib/mem.c >>> index f095fe1..abe29fd 100644 >>> --- a/testcases/kernel/mem/lib/mem.c >>> +++ b/testcases/kernel/mem/lib/mem.c >>> @@ -66,12 +66,12 @@ void oom(int testcase, int mempolicy, int lite) >>> int status; >>> #if HAVE_NUMA_H&& HAVE_LINUX_MEMPOLICY_H&& HAVE_NUMAIF_H \ >>> && HAVE_MPOL_CONSTANTS >>> - unsigned long nmask = 0; >>> + unsigned long nmask[MAXNODES / (8*sizeof(unsigned long))] = { 0 }; >>> unsigned int node; >>> >>> if (mempolicy) >>> node = get_a_numa_node(cleanup); >>> - nmask += 1<< node; >>> + nmask[node / (8*sizeof(unsigned long))] = 1<< (node % >>> (8*sizeof(unsigned >>> long))); >>> #endif >>> >>> switch (pid = fork()) { >>> @@ -81,7 +81,7 @@ void oom(int testcase, int mempolicy, int lite) >>> #if HAVE_NUMA_H&& HAVE_LINUX_MEMPOLICY_H&& HAVE_NUMAIF_H \ >>> && HAVE_MPOL_CONSTANTS >>> if (mempolicy) >>> - if (set_mempolicy(MPOL_BIND,&nmask, MAXNODES) == -1) >>> + if (set_mempolicy(MPOL_BIND, nmask, MAXNODES) == -1) >>> tst_brkm(TBROK | TERRNO, cleanup, >>> "set_mempolicy"); >>> #endif >>> -- >>> 1.7.8.110.g4cb5d >>> >>> ------------------------------------------------------------------------------ >>> Minimize network downtime and maximize team effectiveness. >>> Reduce network management and security costs.Learn how to hire >>> the most talented Cisco Certified professionals. Visit the >>> Employer Resources Portal >>> http://www.cisco.com/web/learning/employer_resources/index.html >>> _______________________________________________ >>> Ltp-list mailing list >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/ltp-list >>> >> >> -- >> Thanks, >> Zhouping >> > ------------------------------------------------------------------------------ Try New Relic Now & We'll Send You this Cool Shirt New Relic is the only SaaS-based application performance monitoring service that delivers powerful full stack analytics. Optimize and monitor your browser, app, & servers with just a few lines of code. Try New Relic and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_apr _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
