This option is used to initialize the ovs_numa module with a fake
configuration and to avoid pthread_setaffinity_np() calls. It will be
useful to test dpif-netdev with pmd threads.
Since it is only used for testing it is not documented in the man pages.
Signed-off-by: Daniele Di Proietto <diproiet...@vmware.com>
---
lib/ovs-numa.c | 166 ++++++++++++++++++++++++++++++++++++++++--------
lib/ovs-numa.h | 1 +
vswitchd/ovs-vswitchd.c | 6 ++
3 files changed, 147 insertions(+), 26 deletions(-)
diff --git a/lib/ovs-numa.c b/lib/ovs-numa.c
index 7bb2bcc..cba08d0 100644
--- a/lib/ovs-numa.c
+++ b/lib/ovs-numa.c
@@ -80,6 +80,11 @@ static struct hmap all_numa_nodes =
HMAP_INITIALIZER(&all_numa_nodes);
static struct hmap all_cpu_cores = HMAP_INITIALIZER(&all_cpu_cores);
/* True if numa node and core info are correctly extracted. */
static bool found_numa_and_core;
+/* True if the module was initialized with dummy options. In this case, the
+ * module must not interact with the actual cpus/nodes in the system. */
+static bool dummy_numa = false;
+
+static struct numa_node* get_numa_by_numa_id(int numa_id);
#ifdef __linux__
/* Returns true if 'str' contains all digits. Returns false otherwise. */
@@ -90,13 +95,89 @@ contain_all_digits(const char *str)
}
#endif /* __linux__ */
+static struct numa_node *
+insert_new_numa_node(int numa_id)
+{
+ struct numa_node *n = xzalloc(sizeof *n);
+
+ hmap_insert(&all_numa_nodes, &n->hmap_node, hash_int(numa_id, 0));
+ ovs_list_init(&n->cores);
+ n->numa_id = numa_id;
+
+ return n;
+}
+
+static struct cpu_core *
+insert_new_cpu_core(struct numa_node *n, unsigned core_id)
+{
+ struct cpu_core *c = xzalloc(sizeof *c);
+
+ hmap_insert(&all_cpu_cores, &c->hmap_node, hash_int(core_id, 0));
+ ovs_list_insert(&n->cores, &c->list_node);
+ c->core_id = core_id;
+ c->numa = n;
+ c->available = true;
+
+ return c;
+}
+
+/* Has the same effect as discover_numa_and_core(), but instead of reading
+ * sysfs entries, extracts the info from 'dummy_config'.
+ *
+ * 'dummy_config' lists the numa_ids of each CPU separated by a comma, e.g.
+ * - "0,0,0,0": four cores on numa socket 0.
+ * - "0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1": 16 cores on two numa sockets.
+ * - "0,0,0,0,1,1,1,1": 8 cores on two numa sockets.
+ *
+ * The different numa ids must be consecutives or the function will abort. */
+static void
+discover_numa_and_core_dummy(const char *dummy_config)
+{
+ char *conf = xstrdup(dummy_config);
+ char *id, *saveptr = NULL;
+ unsigned i = 0;
+ long max_numa_id = 0;
+
+ for (id = strtok_r(conf, ",", &saveptr); id;
+ id = strtok_r(NULL, ",", &saveptr)) {
+ struct hmap_node *hnode;
+ struct numa_node *n;
+ long numa_id;
+
+ numa_id = strtol(id, NULL, 10);
+ if (numa_id < 0 || numa_id >= MAX_NUMA_NODES) {
+ VLOG_WARN("Invalid numa node %ld", numa_id);
+ continue;
+ }
+
+ max_numa_id = MAX(max_numa_id, numa_id);
+
+ hnode = hmap_first_with_hash(&all_numa_nodes, hash_int(numa_id, 0));
+
+ if (hnode) {
+ n = CONTAINER_OF(hnode, struct numa_node, hmap_node);
+ } else {
+ n = insert_new_numa_node(numa_id);
+ }
+
+ insert_new_cpu_core(n, i);
+
+ i++;
+ }
+
+ free(conf);
+
+ if (max_numa_id + 1 != hmap_count(&all_numa_nodes)) {
+ ovs_fatal(0, "dummy numa contains non consecutive numa ids");
+ }
+}
+
/* Discovers all numa nodes and the corresponding cpu cores.
* Constructs the 'struct numa_node' and 'struct cpu_core'. */
static void
discover_numa_and_core(void)
{
#ifdef __linux__
- int n_cpus = 0;
int i;
DIR *dir;
bool numa_supported = true;
@@ -125,31 +206,20 @@ discover_numa_and_core(void)
/* Creates 'struct numa_node' if the 'dir' is non-null. */
if (dir) {
- struct numa_node *n = xzalloc(sizeof *n);
+ struct numa_node *n;
struct dirent *subdir;
- hmap_insert(&all_numa_nodes, &n->hmap_node, hash_int(i, 0));
- ovs_list_init(&n->cores);
- n->numa_id = i;
+ n = insert_new_numa_node(i);
while ((subdir = readdir(dir)) != NULL) {
if (!strncmp(subdir->d_name, "cpu", 3)
- && contain_all_digits(subdir->d_name + 3)){
- struct cpu_core *c = xzalloc(sizeof *c);
+ && contain_all_digits(subdir->d_name + 3)) {
unsigned core_id;
core_id = strtoul(subdir->d_name + 3, NULL, 10);
- hmap_insert(&all_cpu_cores, &c->hmap_node,
- hash_int(core_id, 0));
- ovs_list_insert(&n->cores, &c->list_node);
- c->core_id = core_id;
- c->numa = n;
- c->available = true;
- n_cpus++;
+ insert_new_cpu_core(n, core_id);
}
}
- VLOG_INFO("Discovered %"PRIuSIZE" CPU cores on NUMA node %d",
- ovs_list_size(&n->cores), n->numa_id);
closedir(dir);
} else if (errno != ENOENT) {
VLOG_WARN("opendir(%s) failed (%s)", path,
@@ -161,12 +231,6 @@ discover_numa_and_core(void)
break;
}
}
-
- VLOG_INFO("Discovered %"PRIuSIZE" NUMA nodes and %d CPU cores",
- hmap_count(&all_numa_nodes), n_cpus);
- if (hmap_count(&all_numa_nodes) && hmap_count(&all_cpu_cores)) {
- found_numa_and_core = true;
- }
#endif /* __linux__ */
}
@@ -201,18 +265,63 @@ get_numa_by_numa_id(int numa_id)
}
-/* Extracts the numa node and core info from the 'sysfs'. */
-void
-ovs_numa_init(void)
+
+static bool
+ovs_numa_init__(const char *dummy_config)
{
static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
if (ovsthread_once_start(&once)) {
- discover_numa_and_core();
+ const struct numa_node *n;
+
+ if (!dummy_config) {
+ discover_numa_and_core();
+ } else {
+ discover_numa_and_core_dummy(dummy_config);
+ }
+
+ HMAP_FOR_EACH(n, hmap_node, &all_numa_nodes) {
+ VLOG_INFO("Discovered %"PRIuSIZE" CPU cores on NUMA node %d",
+ ovs_list_size(&n->cores), n->numa_id);
+ }
+
+ VLOG_INFO("Discovered %"PRIuSIZE" NUMA nodes and %"PRIuSIZE" CPU
cores",
+ hmap_count(&all_numa_nodes), hmap_count(&all_cpu_cores));
+
+ if (hmap_count(&all_numa_nodes) && hmap_count(&all_cpu_cores)) {
+ found_numa_and_core = true;
+ }
+
ovsthread_once_done(&once);
+
+ return true;
+ } else {
+ return false;
}
}
+/* Extracts the numa node and core info from the 'dummy_config'. This is
+ * useful for testing purposes. The function must be called only once,
+ * before ovs_numa_init(). Every subsequent call to ovs_numa_init() will
+ * have no effect.
+ *
+ * The format of 'dummy_config' is explained in the comment above
+ * discover_numa_and_core_dummy().*/
+void
+ovs_numa_init_dummy(const char *dummy_config)
+{
+ dummy_numa = true;
+ ovs_assert(dummy_config);
+ ovs_assert(ovs_numa_init__(dummy_config));
+}
+
+/* Extracts the numa node and core info from sysfs. */
+void
+ovs_numa_init(void)
+{
+ ovs_numa_init__(NULL);
+}
+
bool
ovs_numa_numa_id_is_valid(int numa_id)
{
@@ -476,6 +585,11 @@ ovs_numa_set_cpu_mask(const char *cmask)
int ovs_numa_thread_setaffinity_core(unsigned core_id)
{
+ if (dummy_numa) {
+ /* Nothing to do */
+ return 0;
+ }
+
#ifdef __linux__
cpu_set_t cpuset;
int err;
diff --git a/lib/ovs-numa.h b/lib/ovs-numa.h
index 5b3444b..9a434df 100644
--- a/lib/ovs-numa.h
+++ b/lib/ovs-numa.h
@@ -39,6 +39,7 @@ struct ovs_numa_info {
};
void ovs_numa_init(void);
+void ovs_numa_init_dummy(const char *dummy_config);
bool ovs_numa_numa_id_is_valid(int numa_id);
bool ovs_numa_core_id_is_valid(unsigned core_id);
bool ovs_numa_core_is_pinned(unsigned core_id);
diff --git a/vswitchd/ovs-vswitchd.c b/vswitchd/ovs-vswitchd.c
index 7d467a1..12a3db1 100644
--- a/vswitchd/ovs-vswitchd.c
+++ b/vswitchd/ovs-vswitchd.c
@@ -145,6 +145,7 @@ parse_options(int argc, char *argv[], char **unixctl_pathp)
OPT_DISABLE_SYSTEM,
DAEMON_OPTION_ENUMS,
OPT_DPDK,
+ OPT_DUMMY_NUMA,
};
static const struct option long_options[] = {
{"help", no_argument, NULL, 'h'},
@@ -159,6 +160,7 @@ parse_options(int argc, char *argv[], char **unixctl_pathp)
{"enable-dummy", optional_argument, NULL, OPT_ENABLE_DUMMY},
{"disable-system", no_argument, NULL, OPT_DISABLE_SYSTEM},
{"dpdk", optional_argument, NULL, OPT_DPDK},
+ {"dummy-numa", required_argument, NULL, OPT_DUMMY_NUMA},
{NULL, 0, NULL, 0},
};
char *short_options = ovs_cmdl_long_options_to_short_options(long_options);
@@ -214,6 +216,10 @@ parse_options(int argc, char *argv[], char **unixctl_pathp)
ovs_fatal(0, "Using --dpdk to configure DPDK is not supported.");
break;
+ case OPT_DUMMY_NUMA:
+ ovs_numa_init_dummy(optarg);
+ break;
+
default:
abort();
}
--
2.8.1
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
