[LTP] [PATCH] ltp/numa: add new test11
Cyril Hrubis
chrubis@suse.cz
Tue Jan 24 20:33:07 CET 2017
Hi!
> +# Verification of hugepage memory allocated on a node
> +test11()
> +{
> + Mem_huge=0
> +
> + if [ ! -d "/sys/kernel/mm/hugepages/" ]; then
> + tst_brk TCONF "hugepage is not supported"
This should be tst_res followed by a return on a next line. The
tst_brk will exit the whole test just because hugepages are not
supported.
> + fi
> +
> + for node in $nodes_list; do
> + Ori_hpgs=$(cat /sys/devices/system/node/node${node}/hugepages/hugepages-${HPAGE_SIZE}kB/nr_hugepages)
> + echo 1 >/sys/devices/system/node/node${node}/hugepages/hugepages-${HPAGE_SIZE}kB/nr_hugepages
Uh, so this sysfs knob controlls how much hugepages can be allocated on
a node right? What if it was set to nonzero and some hugepages are
allocated already? Shouldn't we rather set it to current value + 1 ?
> + numactl --cpunodebind=$node --membind=$node support_numa $HUGE_PAGE &
> + pid=$!
> + wait_for_support_numa $pid
> +
> + Mem_huge=$(echo $(numastat -p $pid |grep '^Huge' |awk '{print $'$((node+2))'}'))
> + Mem_huge=$(echo "$Mem_huge * 1024" |bc)
Eh, why not just $((Mem_huge * 1024)) ?
> + if [ $(echo "$Mem_huge < $HPAGE_SIZE" | bc) -eq 1 ]; then
Here as well, why not just if [ "$Mem_huge" -lt "$HPAGE_SIZE" ]; then
> + tst_res TFAIL \
> + "NUMA memory allocated in node$node is less than expected"
> + return
> + fi
> +
> + kill -CONT $pid >/dev/null 2>&1
> + echo $Ori_hpgs >/sys/devices/system/node/node${node}/hugepages/hugepages-${HPAGE_SIZE}kB/nr_hugepages
> + done
> +
> + tst_res TPASS "NUMA local node hugepage memory allocated"
> +}
> +
> tst_run
> diff --git a/testcases/kernel/numa/support_numa.c b/testcases/kernel/numa/support_numa.c
> index eaf63e3..0241a19 100644
> --- a/testcases/kernel/numa/support_numa.c
> +++ b/testcases/kernel/numa/support_numa.c
> @@ -22,7 +22,7 @@
> /* */
> /* File: support_numa.c */
> /* */
> -/* Description: Allocates 1MB of memory and touches it to verify numa */
> +/* Description: Allocates memory and touches it to verify numa */
> /* */
> /* Author: Sivakumar Chinnaiah Sivakumar.C@in.ibm.com */
> /* */
> @@ -52,16 +52,43 @@ static void help(void)
> printf("Input: Describe input arguments to this program\n");
> printf(" argv[1] == 1 then allocate 1MB of memory\n");
> printf(" argv[1] == 2 then allocate 1MB of share memory\n");
> - printf(" argv[1] == 3 then pause the program to catch sigint\n");
> + printf(" argv[1] == 3 then allocate 1HUGE PAGE SIZE of memory\n");
> + printf(" argv[1] == 4 then pause the program to catch sigint\n");
> printf("Exit: On failure - Exits with non-zero value\n");
> printf(" On success - exits with 0 exit value\n");
>
> exit(1);
> }
>
> +static int read_hugepagesize(void)
> +{
> + FILE *fp;
> + char line[BUFSIZ], buf[BUFSIZ];
> + int val;
> +
> + fp = fopen("/proc/meminfo", "r");
> + if (fp == NULL) {
> + fprintf(stderr, "Failed to open /proc/meminfo");
> + return 1;
> + }
> +
> + while (fgets(line, BUFSIZ, fp) != NULL) {
> + if (sscanf(line, "%64s %d", buf, &val) == 2)
> + if (strcmp(buf, "Hugepagesize:") == 0) {
> + fclose(fp);
> + return 1024 * val;
> + }
> + }
> +
> + fclose(fp);
> + fprintf(stderr, "can't find \"%s\" in %s", "Hugepagesize:", "/proc/meminfo");
> +
> + return 1;
> +}
We should rather return 0 on a failure so that we can easily check
hpsz != 0 in the main()
> int main(int argc, char *argv[])
> {
> - int i, fd, rc;
> + int i, fd, rc, hpsz;
> char *buf = NULL;
> struct stat sb;
>
> @@ -114,6 +141,24 @@ int main(int argc, char *argv[])
> remove(TEST_SFILE);
> break;
> case 3:
> + hpsz = read_hugepagesize();
Eh, we do not check for a failure and gracefully try to allocate 1 byte
here? That does not sound right.
> + buf = mmap(NULL, hpsz, PROT_READ | PROT_WRITE,
> + MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB,
> + -1, 0);
> +
> + if (buf == MAP_FAILED){
> + fprintf(stderr, "mmap failed\n");
What about errno? We should rather use perror() here. Also coding style
is wrong, you should use checkpatch.pl to check for common mistakes.
> + exit(1);
> + }
> +
> + memset(buf, 'a', hpsz);
> +
> + raise(SIGSTOP);
> +
> + munmap(buf, hpsz);
> + break;
> + case 4:
> raise(SIGSTOP);
> break;
> default:
> --
> 1.8.3.1
>
--
Cyril Hrubis
chrubis@suse.cz
More information about the ltp
mailing list