/*
* verify_pages_linear() - verify pages are interleaved
* @pages: array of pages to be verified
* @status: the NUMA node of each page
* @num: the no. of pages
*/
void verify_pages_linear(void **pages, int *status, unsigned int num)
{
#if HAVE_NUMA_H
unsigned int i;
unsigned int n = 0;
int nodes[num];
int num_allowed_nodes;
int *allowed_nodes;
int ret;
ret = get_allowed_nodes_arr(NH_MEMS, &num_allowed_nodes,
&allowed_nodes);
if (ret < 0)
tst_brkm(TBROK | TERRNO, NULL, "get_allowed_nodes(): %d", ret);
for (i = 0; i < num; i++) {
nodes[i] = allowed_nodes[n];
n++;
if (n >= num_allowed_nodes)
n = 0;
}
free(allowed_nodes);
verify_pages_on_nodes(pages, status, num, nodes);
#endif
}
/*
* get_allowed_nodes - convenience function to get fixed number of nodes
* @count: how many nodes to get
* @...: int pointers, where node ids will be stored
* RETURNS:
* 0 on success
* -1 on allocation failure
* -2 on get_mempolicy failure
* -3 on not enough allowed nodes
*/
int get_allowed_nodes(int flag, int count, ...)
{
int ret;
int i, *nodep;
va_list ap;
int num_nodes = 0;
int *nodes = NULL;
ret = get_allowed_nodes_arr(flag, &num_nodes, &nodes);
if (ret < 0)
return ret;
va_start(ap, count);
for (i = 0; i < count; i++) {
nodep = va_arg(ap, int *);
if (i < num_nodes) {
*nodep = nodes[i];
} else {
ret = -3;
errno = EINVAL;
break;
}
}
free(nodes);
va_end(ap);
return ret;
}
/*
* is_numa - judge a system is NUMA system or not
* NOTE: the function is designed to try to find more than
* 1 available node, at least each node contains memory.
* WARN: Don't use this func in child, as it calls tst_brkm()
* RETURNS:
* 0 - it's not a NUMA system
* 1 - it's a NUMA system
*/
int is_numa(void (*cleanup_fn)(void))
{
int ret;
int numa_nodes = 0;
ret = get_allowed_nodes_arr(NH_MEMS, &numa_nodes, NULL);
if (ret < 0)
tst_brkm(TBROK | TERRNO, cleanup_fn, "get_allowed_nodes_arr");
if (numa_nodes > 1)
return 1;
else
return 0;
}
static void print_node_info(int flag)
{
int *allowed_nodes = NULL;
int i, ret, num_nodes;
ret = get_allowed_nodes_arr(flag, &num_nodes, &allowed_nodes);
printf("nodes (flag=%d): ", flag);
if (ret == 0) {
for (i = 0; i < num_nodes; i++)
printf("%d ", allowed_nodes[i]);
printf("\n");
} else
printf("error(%d)\n", ret);
free(allowed_nodes);
}
int main(int argc, char *argv[])
{
tst_parse_opts(argc, argv, NULL, NULL);
ncpus = count_cpu();
if (get_allowed_nodes_arr(NH_MEMS | NH_CPUS, &nnodes, &nodes) < 0)
tst_brkm(TBROK | TERRNO, NULL, "get_allowed_nodes_arr");
if (nnodes <= 1)
tst_brkm(TCONF, NULL, "requires a NUMA system.");
setup();
testcpuset();
cleanup();
tst_exit();
}
static void set_global_mempolicy(int mempolicy)
{
#if HAVE_NUMA_H && HAVE_LINUX_MEMPOLICY_H && HAVE_NUMAIF_H \
&& HAVE_MPOL_CONSTANTS
unsigned long nmask[MAXNODES / BITS_PER_LONG] = { 0 };
int num_nodes, *nodes;
int ret;
if (mempolicy) {
ret = get_allowed_nodes_arr(NH_MEMS|NH_CPUS, &num_nodes, &nodes);
if (ret != 0)
tst_brkm(TBROK|TERRNO, cleanup,
"get_allowed_nodes_arr");
if (num_nodes < 2) {
tst_resm(TINFO, "mempolicy need NUMA system support");
free(nodes);
return;
}
switch(mempolicy) {
case MPOL_BIND:
/* bind the second node */
set_node(nmask, nodes[1]);
break;
case MPOL_INTERLEAVE:
case MPOL_PREFERRED:
if (num_nodes == 2) {
tst_resm(TINFO, "The mempolicy need "
"more than 2 numa nodes");
free(nodes);
return;
} else {
/* Using the 2nd,3rd node */
set_node(nmask, nodes[1]);
set_node(nmask, nodes[2]);
}
break;
default:
tst_brkm(TBROK|TERRNO, cleanup, "Bad mempolicy mode");
}
if (set_mempolicy(mempolicy, nmask, MAXNODES) == -1)
tst_brkm(TBROK|TERRNO, cleanup, "set_mempolicy");
}
#endif
}
/*
* check_config() - check for required configuration
* @min_nodes: the minimum required NUMA nodes
*
* Checks if numa support is availabe, kernel is >= 2.6.18, arch is
* one of the supported architectures.
*/
void check_config(unsigned int min_nodes)
{
#if HAVE_NUMA_H && HAVE_NUMAIF_H
int num_allowed_nodes;
int ret;
ret = get_allowed_nodes_arr(NH_MEMS, &num_allowed_nodes, NULL);
if (ret < 0)
tst_brkm(TBROK | TERRNO, NULL, "get_allowed_nodes(): %d", ret);
if (numa_available() < 0) {
tst_brkm(TCONF, NULL, "NUMA support is not available");
} else if (num_allowed_nodes < min_nodes) {
tst_brkm(TCONF, NULL, "at least %d allowed NUMA nodes"
" are required", min_nodes);
} else if (tst_kvercmp(2, 6, 18) < 0) {
tst_brkm(TCONF, NULL, "2.6.18 or greater kernel required");
}
#else
tst_brkm(TCONF, NULL, "NUMA support not provided");
#endif
}
static void test_invalid_nodes(void)
{
int *nodes;
int num_nodes, ret, i;
int invalid_node = 0;
unsigned long *old_nodes, *new_nodes;
tst_resm(TINFO, "test_invalid_nodes");
ret = get_allowed_nodes_arr(NH_MEMS, &num_nodes, &nodes);
if (ret < 0)
tst_brkm(TBROK | TERRNO, cleanup,
"get_allowed_nodes_arr: %d", ret);
/* get first node which is not in nodes */
for (i = 0; i < num_nodes; i++, invalid_node++)
if (invalid_node != nodes[i])
break;
if (invalid_node < sane_max_node) {
old_nodes = SAFE_MALLOC(NULL, sane_nodemask_size);
new_nodes = SAFE_MALLOC(NULL, sane_nodemask_size);
memcpy(old_nodes, sane_old_nodes, sane_nodemask_size);
memset(new_nodes, 0, sane_nodemask_size);
set_bit(new_nodes, invalid_node, 1);
TEST(ltp_syscall(__NR_migrate_pages, 0, sane_max_node,
old_nodes, new_nodes));
check_ret(-1);
check_errno(EINVAL);
free(old_nodes);
free(new_nodes);
} else {
tst_resm(TCONF, "All possible nodes are present");
}
free(nodes);
}
static void setup(void)
{
int ret, i, j;
int pagesize = getpagesize();
void *p;
tst_require_root(NULL);
TEST(ltp_syscall(__NR_migrate_pages, 0, 0, NULL, NULL));
if (numa_available() == -1)
tst_brkm(TCONF, NULL, "NUMA not available");
ret = get_allowed_nodes_arr(NH_MEMS, &num_nodes, &nodes);
if (ret < 0)
tst_brkm(TBROK | TERRNO, NULL, "get_allowed_nodes(): %d", ret);
if (num_nodes < 2)
tst_brkm(TCONF, NULL, "at least 2 allowed NUMA nodes"
" are required");
else if (tst_kvercmp(2, 6, 18) < 0)
tst_brkm(TCONF, NULL, "2.6.18 or greater kernel required");
/*
* find 2 nodes, which can hold NODE_MIN_FREEMEM bytes
* The reason is that:
* 1. migrate_pages() is expected to succeed
* 2. this test avoids hitting:
* Bug 870326 - migrate_pages() reports success, but pages are
* not moved to desired node
* https://bugzilla.redhat.com/show_bug.cgi?id=870326
*/
nodeA = nodeB = -1;
for (i = 0; i < num_nodes; i++) {
p = numa_alloc_onnode(NODE_MIN_FREEMEM, nodes[i]);
if (p == NULL)
break;
memset(p, 0xff, NODE_MIN_FREEMEM);
j = 0;
while (j < NODE_MIN_FREEMEM) {
if (addr_on_node(p + j) != nodes[i])
break;
j += pagesize;
}
numa_free(p, NODE_MIN_FREEMEM);
if (j >= NODE_MIN_FREEMEM) {
if (nodeA == -1)
nodeA = nodes[i];
else if (nodeB == -1)
nodeB = nodes[i];
else
break;
}
}
if (nodeA == -1 || nodeB == -1)
tst_brkm(TCONF, NULL, "at least 2 NUMA nodes with "
"free mem > %d are needed", NODE_MIN_FREEMEM);
tst_resm(TINFO, "Using nodes: %d %d", nodeA, nodeB);
ltpuser = getpwnam(nobody_uid);
if (ltpuser == NULL)
tst_brkm(TBROK | TERRNO, NULL, "getpwnam failed");
TEST_PAUSE;
}
void test_ksm_merge_across_nodes(unsigned long nr_pages)
{
char **memory;
int i, ret;
int num_nodes, *nodes;
unsigned long length;
unsigned long pagesize;
#if HAVE_NUMA_H && HAVE_LINUX_MEMPOLICY_H && HAVE_NUMAIF_H \
&& HAVE_MPOL_CONSTANTS
unsigned long nmask[MAXNODES / BITS_PER_LONG] = { 0 };
#endif
ret = get_allowed_nodes_arr(NH_MEMS|NH_CPUS, &num_nodes, &nodes);
if (ret != 0)
tst_brkm(TBROK|TERRNO, cleanup, "get_allowed_nodes_arr");
if (num_nodes < 2) {
tst_resm(TINFO, "need NUMA system support");
free(nodes);
return;
}
pagesize = sysconf(_SC_PAGE_SIZE);
length = nr_pages * pagesize;
memory = malloc(num_nodes * sizeof(char *));
for (i = 0; i < num_nodes; i++) {
memory[i] = mmap(NULL, length, PROT_READ|PROT_WRITE,
MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
if (memory[i] == MAP_FAILED)
tst_brkm(TBROK|TERRNO, tst_exit, "mmap");
#ifdef HAVE_MADV_MERGEABLE
if (madvise(memory[i], length, MADV_MERGEABLE) == -1)
tst_brkm(TBROK|TERRNO, tst_exit, "madvise");
#endif
#if HAVE_NUMA_H && HAVE_LINUX_MEMPOLICY_H && HAVE_NUMAIF_H \
&& HAVE_MPOL_CONSTANTS
clean_node(nmask);
set_node(nmask, nodes[i]);
/*
* Use mbind() to make sure each node contains
* length size memory.
*/
ret = mbind(memory[i], length, MPOL_BIND, nmask, MAXNODES, 0);
if (ret == -1)
tst_brkm(TBROK|TERRNO, tst_exit, "mbind");
#endif
memset(memory[i], 10, length);
}
SAFE_FILE_PRINTF(cleanup, PATH_KSM "sleep_millisecs", "0");
SAFE_FILE_PRINTF(cleanup, PATH_KSM "pages_to_scan", "%ld",
nr_pages * num_nodes);
/*
* merge_across_nodes setting can be changed only when there
* are no ksm shared pages in system, so set run 2 to unmerge
* pages first, then to 1 after changing merge_across_nodes,
* to remerge according to the new setting.
*/
SAFE_FILE_PRINTF(cleanup, PATH_KSM "run", "2");
wait_ksmd_done();
tst_resm(TINFO, "Start to test KSM with merge_across_nodes=1");
SAFE_FILE_PRINTF(cleanup, PATH_KSM "merge_across_nodes", "1");
SAFE_FILE_PRINTF(cleanup, PATH_KSM "run", "1");
group_check(1, 1, nr_pages * num_nodes - 1, 0, 0, 0,
nr_pages * num_nodes);
SAFE_FILE_PRINTF(cleanup, PATH_KSM "run", "2");
wait_ksmd_done();
tst_resm(TINFO, "Start to test KSM with merge_across_nodes=0");
SAFE_FILE_PRINTF(cleanup, PATH_KSM "merge_across_nodes", "0");
SAFE_FILE_PRINTF(cleanup, PATH_KSM "run", "1");
group_check(1, num_nodes, nr_pages * num_nodes - num_nodes,
0, 0, 0, nr_pages * num_nodes);
SAFE_FILE_PRINTF(cleanup, PATH_KSM "run", "2");
wait_ksmd_done();
}
