Application didn't request lockless (MT_UNSAFE) output queues
in scheduler mode, even when the pktio had enough queues (one
tx queue per worker). Updated tx side queue configuration code
to utilize lockless queues also in sched mode. Also updated
queue binding print out, so that it's easy to check which
queues workers are using (only tx in sched mode).
Signed-off-by: Petri Savolainen <petri.savolai...@nokia.com>
---
test/common_plat/performance/odp_l2fwd.c | 221 ++++++++++++++++---------------
1 file changed, 112 insertions(+), 109 deletions(-)
diff --git a/test/common_plat/performance/odp_l2fwd.c
b/test/common_plat/performance/odp_l2fwd.c
index a656547..90e3b88 100644
--- a/test/common_plat/performance/odp_l2fwd.c
+++ b/test/common_plat/performance/odp_l2fwd.c
@@ -104,6 +104,7 @@ typedef struct {
int dst_change; /**< Change destination eth addresses */
int src_change; /**< Change source eth addresses */
int error_check; /**< Check packet errors */
+ int sched_mode; /**< Scheduler mode */
} appl_args_t;
static int exit_threads; /**< Break workers loop if set to 1 */
@@ -132,8 +133,6 @@ typedef struct thread_args_t {
int num_pktio;
struct {
- odp_pktio_t rx_pktio;
- odp_pktio_t tx_pktio;
odp_pktin_queue_t pktin;
odp_pktout_queue_t pktout;
odp_queue_t rx_queue;
@@ -290,8 +289,8 @@ static int run_worker_sched_mode(void *arg)
int pkts;
int thr;
int dst_idx;
- int thr_idx;
int i;
+ int pktio, num_pktio;
odp_pktout_queue_t pktout[MAX_PKTIOS];
odp_queue_t tx_queue[MAX_PKTIOS];
thread_args_t *thr_args = arg;
@@ -300,24 +299,17 @@ static int run_worker_sched_mode(void *arg)
pktin_mode_t in_mode = gbl_args->appl.in_mode;
thr = odp_thread_id();
- thr_idx = thr_args->thr_idx;
- memset(pktout, 0, sizeof(pktout));
+ num_pktio = thr_args->num_pktio;
- for (i = 0; i < MAX_PKTIOS; i++)
- tx_queue[i] = ODP_QUEUE_INVALID;
+ if (num_pktio > MAX_PKTIOS) {
+ LOG_ERR("Too many pktios %i\n", num_pktio);
+ return -1;
+ }
- for (i = 0; i < gbl_args->appl.if_count; i++) {
- if (gbl_args->pktios[i].num_tx_queue ==
- gbl_args->appl.num_workers) {
- pktout[i] = gbl_args->pktios[i].pktout[thr_idx];
- tx_queue[i] = gbl_args->pktios[i].tx_q[thr_idx];
- } else if (gbl_args->pktios[i].num_tx_queue == 1) {
- pktout[i] = gbl_args->pktios[i].pktout[0];
- tx_queue[i] = gbl_args->pktios[i].tx_q[0];
- } else {
- LOG_ABORT("Bad number of output queues %i\n", i);
- }
+ for (pktio = 0; pktio < num_pktio; pktio++) {
+ tx_queue[pktio] = thr_args->pktio[pktio].tx_queue;
+ pktout[pktio] = thr_args->pktio[pktio].pktout;
}
printf("[%02i] PKTIN_SCHED_%s, %s\n", thr,
@@ -605,13 +597,12 @@ static int create_pktio(const char *dev, int idx, int
num_rx, int num_tx,
odp_pktio_op_mode_t mode_rx;
odp_pktio_op_mode_t mode_tx;
pktin_mode_t in_mode = gbl_args->appl.in_mode;
- int num_tx_shared;
odp_pktio_param_init(&pktio_param);
- if (gbl_args->appl.in_mode == PLAIN_QUEUE)
+ if (in_mode == PLAIN_QUEUE)
pktio_param.in_mode = ODP_PKTIN_MODE_QUEUE;
- else if (gbl_args->appl.in_mode != DIRECT_RECV) /* pktin_mode SCHED_* */
+ else if (in_mode != DIRECT_RECV) /* pktin_mode SCHED_* */
pktio_param.in_mode = ODP_PKTIN_MODE_SCHED;
if (gbl_args->appl.out_mode != PKTOUT_DIRECT)
@@ -634,11 +625,12 @@ static int create_pktio(const char *dev, int idx, int
num_rx, int num_tx,
odp_pktin_queue_param_init(&pktin_param);
odp_pktout_queue_param_init(&pktout_param);
- if (sched_mode(in_mode)) {
- num_tx_shared = 1;
- mode_tx = ODP_PKTIO_OP_MT;
- mode_rx = ODP_PKTIO_OP_MT;
+ /* By default use a queue per worker. Sched mode ignores rx side
+ * setting. */
+ mode_rx = ODP_PKTIO_OP_MT_UNSAFE;
+ mode_tx = ODP_PKTIO_OP_MT_UNSAFE;
+ if (gbl_args->appl.sched_mode) {
if (gbl_args->appl.in_mode == SCHED_ATOMIC)
sync_mode = ODP_SCHED_SYNC_ATOMIC;
else if (gbl_args->appl.in_mode == SCHED_ORDERED)
@@ -649,10 +641,6 @@ static int create_pktio(const char *dev, int idx, int
num_rx, int num_tx,
pktin_param.queue_param.sched.prio = ODP_SCHED_PRIO_DEFAULT;
pktin_param.queue_param.sched.sync = sync_mode;
pktin_param.queue_param.sched.group = ODP_SCHED_GROUP_ALL;
- } else {
- num_tx_shared = capa.max_output_queues;
- mode_tx = ODP_PKTIO_OP_MT_UNSAFE;
- mode_rx = ODP_PKTIO_OP_MT_UNSAFE;
}
if (num_rx > (int)capa.max_input_queues) {
@@ -664,8 +652,8 @@ static int create_pktio(const char *dev, int idx, int
num_rx, int num_tx,
if (num_tx > (int)capa.max_output_queues) {
printf("Sharing %i output queues between %i workers\n",
- num_tx_shared, num_tx);
- num_tx = num_tx_shared;
+ capa.max_output_queues, num_tx);
+ num_tx = capa.max_output_queues;
mode_tx = ODP_PKTIO_OP_MT;
}
@@ -795,31 +783,10 @@ static int print_speed_stats(int num_workers, stats_t
*thr_stats,
static void print_port_mapping(void)
{
- int if_count, num_workers;
- int thr, pktio;
+ int if_count;
+ int pktio;
if_count = gbl_args->appl.if_count;
- num_workers = gbl_args->appl.num_workers;
-
- printf("\nWorker mapping table (port[queue])\n--------------------\n");
-
- for (thr = 0; thr < num_workers; thr++) {
- int rx_idx, tx_idx;
- int rx_queue_idx, tx_queue_idx;
- thread_args_t *thr_args = &gbl_args->thread[thr];
- int num = thr_args->num_pktio;
-
- printf("Worker %i\n", thr);
-
- for (pktio = 0; pktio < num; pktio++) {
- rx_idx = thr_args->pktio[pktio].rx_idx;
- tx_idx = thr_args->pktio[pktio].tx_idx;
- rx_queue_idx = thr_args->pktio[pktio].rx_queue_idx;
- tx_queue_idx = thr_args->pktio[pktio].tx_queue_idx;
- printf(" %i[%i] -> %i[%i]\n",
- rx_idx, rx_queue_idx, tx_idx, tx_queue_idx);
- }
- }
printf("\nPort config\n--------------------\n");
@@ -875,51 +842,77 @@ static int find_dest_port(int port)
static void bind_workers(void)
{
int if_count, num_workers;
- int rx_idx, tx_idx, thr, pktio;
+ int rx_idx, tx_idx, thr, pktio, i;
thread_args_t *thr_args;
if_count = gbl_args->appl.if_count;
num_workers = gbl_args->appl.num_workers;
- /* initialize port forwarding table */
- for (rx_idx = 0; rx_idx < if_count; rx_idx++)
- gbl_args->dst_port[rx_idx] = find_dest_port(rx_idx);
-
- if (if_count > num_workers) {
- thr = 0;
-
- for (rx_idx = 0; rx_idx < if_count; rx_idx++) {
- thr_args = &gbl_args->thread[thr];
- pktio = thr_args->num_pktio;
- tx_idx = gbl_args->dst_port[rx_idx];
- thr_args->pktio[pktio].rx_idx = rx_idx;
- thr_args->pktio[pktio].tx_idx = tx_idx;
- thr_args->num_pktio++;
-
- gbl_args->pktios[rx_idx].num_rx_thr++;
- gbl_args->pktios[tx_idx].num_tx_thr++;
-
- thr++;
- if (thr >= num_workers)
- thr = 0;
+ if (gbl_args->appl.sched_mode) {
+ /* all threads receive and send on all pktios */
+ for (i = 0; i < if_count; i++) {
+ gbl_args->pktios[i].num_rx_thr = num_workers;
+ gbl_args->pktios[i].num_tx_thr = num_workers;
}
- } else {
- rx_idx = 0;
for (thr = 0; thr < num_workers; thr++) {
thr_args = &gbl_args->thread[thr];
- pktio = thr_args->num_pktio;
- tx_idx = gbl_args->dst_port[rx_idx];
- thr_args->pktio[pktio].rx_idx = rx_idx;
- thr_args->pktio[pktio].tx_idx = tx_idx;
- thr_args->num_pktio++;
-
- gbl_args->pktios[rx_idx].num_rx_thr++;
- gbl_args->pktios[tx_idx].num_tx_thr++;
-
- rx_idx++;
- if (rx_idx >= if_count)
- rx_idx = 0;
+ thr_args->num_pktio = if_count;
+
+ /* In sched mode, pktios are not cross connected with
+ * local pktio indexes */
+ for (i = 0; i < if_count; i++) {
+ thr_args->pktio[i].rx_idx = i;
+ thr_args->pktio[i].tx_idx = i;
+ }
+ }
+ } else {
+ /* initialize port forwarding table */
+ for (rx_idx = 0; rx_idx < if_count; rx_idx++)
+ gbl_args->dst_port[rx_idx] = find_dest_port(rx_idx);
+
+ if (if_count > num_workers) {
+ /* Less workers than pktios. Assign single worker per
+ * pktio. */
+ thr = 0;
+
+ for (rx_idx = 0; rx_idx < if_count; rx_idx++) {
+ thr_args = &gbl_args->thread[thr];
+ pktio = thr_args->num_pktio;
+ /* Cross connect rx to tx */
+ tx_idx = gbl_args->dst_port[rx_idx];
+ thr_args->pktio[pktio].rx_idx = rx_idx;
+ thr_args->pktio[pktio].tx_idx = tx_idx;
+ thr_args->num_pktio++;
+
+ gbl_args->pktios[rx_idx].num_rx_thr++;
+ gbl_args->pktios[tx_idx].num_tx_thr++;
+
+ thr++;
+ if (thr >= num_workers)
+ thr = 0;
+ }
+ } else {
+ /* More workers than pktios. Assign at least one worker
+ * per pktio. */
+ rx_idx = 0;
+
+ for (thr = 0; thr < num_workers; thr++) {
+ thr_args = &gbl_args->thread[thr];
+ pktio = thr_args->num_pktio;
+ /* Cross connect rx to tx */
+ tx_idx = gbl_args->dst_port[rx_idx];
+ thr_args->pktio[pktio].rx_idx = rx_idx;
+ thr_args->pktio[pktio].tx_idx = tx_idx;
+ thr_args->num_pktio++;
+
+ gbl_args->pktios[rx_idx].num_rx_thr++;
+ gbl_args->pktios[tx_idx].num_tx_thr++;
+
+ rx_idx++;
+ if (rx_idx >= if_count)
+ rx_idx = 0;
+ }
}
}
}
@@ -930,37 +923,44 @@ static void bind_workers(void)
static void bind_queues(void)
{
int num_workers;
- int thr, pktio;
+ int thr, i;
num_workers = gbl_args->appl.num_workers;
+ printf("\nQueue binding (indexes)\n-----------------------\n");
+
for (thr = 0; thr < num_workers; thr++) {
int rx_idx, tx_idx;
thread_args_t *thr_args = &gbl_args->thread[thr];
int num = thr_args->num_pktio;
- for (pktio = 0; pktio < num; pktio++) {
+ printf("worker %i\n", thr);
+
+ for (i = 0; i < num; i++) {
int rx_queue, tx_queue;
- rx_idx = thr_args->pktio[pktio].rx_idx;
- tx_idx = thr_args->pktio[pktio].tx_idx;
+ rx_idx = thr_args->pktio[i].rx_idx;
+ tx_idx = thr_args->pktio[i].tx_idx;
rx_queue = gbl_args->pktios[rx_idx].next_rx_queue;
tx_queue = gbl_args->pktios[tx_idx].next_tx_queue;
- thr_args->pktio[pktio].rx_queue_idx = rx_queue;
- thr_args->pktio[pktio].tx_queue_idx = tx_queue;
- thr_args->pktio[pktio].pktin =
+ thr_args->pktio[i].rx_queue_idx = rx_queue;
+ thr_args->pktio[i].tx_queue_idx = tx_queue;
+ thr_args->pktio[i].pktin =
gbl_args->pktios[rx_idx].pktin[rx_queue];
- thr_args->pktio[pktio].pktout =
- gbl_args->pktios[tx_idx].pktout[tx_queue];
- thr_args->pktio[pktio].rx_queue =
+ thr_args->pktio[i].rx_queue =
gbl_args->pktios[rx_idx].rx_q[rx_queue];
- thr_args->pktio[pktio].tx_queue =
+ thr_args->pktio[i].pktout =
+ gbl_args->pktios[tx_idx].pktout[tx_queue];
+ thr_args->pktio[i].tx_queue =
gbl_args->pktios[tx_idx].tx_q[tx_queue];
- thr_args->pktio[pktio].rx_pktio =
- gbl_args->pktios[rx_idx].pktio;
- thr_args->pktio[pktio].tx_pktio =
- gbl_args->pktios[tx_idx].pktio;
+
+ if (!gbl_args->appl.sched_mode)
+ printf(" rx: pktio %i, queue %i\n",
+ rx_idx, rx_queue);
+
+ printf(" tx: pktio %i, queue %i\n",
+ tx_idx, tx_queue);
rx_queue++;
tx_queue++;
@@ -974,6 +974,8 @@ static void bind_queues(void)
gbl_args->pktios[tx_idx].next_tx_queue = tx_queue;
}
}
+
+ printf("\n");
}
static void init_port_lookup_tbl(void)
@@ -1340,6 +1342,9 @@ int main(int argc, char *argv[])
/* Parse and store the application arguments */
parse_args(argc, argv, &gbl_args->appl);
+ if (sched_mode(gbl_args->appl.in_mode))
+ gbl_args->appl.sched_mode = 1;
+
/* Print both system and application information */
print_info(NO_PATH(argv[0]), &gbl_args->appl);
@@ -1392,8 +1397,7 @@ int main(int argc, char *argv[])
num_rx = num_workers;
num_tx = num_workers;
- if (gbl_args->appl.in_mode == DIRECT_RECV ||
- gbl_args->appl.in_mode == PLAIN_QUEUE) {
+ if (!gbl_args->appl.sched_mode) {
/* A queue per assigned worker */
num_rx = gbl_args->pktios[i].num_rx_thr;
num_tx = gbl_args->pktios[i].num_tx_thr;
@@ -1431,8 +1435,7 @@ int main(int argc, char *argv[])
init_port_lookup_tbl();
- if (gbl_args->appl.in_mode == DIRECT_RECV ||
- gbl_args->appl.in_mode == PLAIN_QUEUE)
+ if (!gbl_args->appl.sched_mode)
print_port_mapping();
memset(thread_tbl, 0, sizeof(thread_tbl));
--
2.8.1
