ixgbe_tx_free_bufs(struct ixgbe_tx_queue *txq)
{
struct ixgbe_tx_entry_v *txep;
uint32_t status;
uint32_t n;
uint32_t i;
int nb_free = 0;
struct rte_mbuf *m, *free[RTE_IXGBE_TX_MAX_FREE_BUF_SZ];
/* check DD bit on threshold descriptor */
status = txq->tx_ring[txq->tx_next_dd].wb.status;
if (!(status & IXGBE_ADVTXD_STAT_DD))
return 0;
n = txq->tx_rs_thresh;
/*
* first buffer to free from S/W ring is at index
* tx_next_dd - (tx_rs_thresh-1)
*/
txep = &((struct ixgbe_tx_entry_v *)txq->sw_ring)[txq->tx_next_dd -
(n - 1)];
m = __rte_pktmbuf_prefree_seg(txep[0].mbuf);
if (likely(m != NULL)) {
free[0] = m;
nb_free = 1;
for (i = 1; i < n; i++) {
m = __rte_pktmbuf_prefree_seg(txep[i].mbuf);
if (likely(m != NULL)) {
if (likely(m->pool == free[0]->pool))
free[nb_free++] = m;
else {
rte_mempool_put_bulk(free[0]->pool,
(void *)free, nb_free);
free[0] = m;
nb_free = 1;
}
}
}
rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
} else {
for (i = 1; i < n; i++) {
m = __rte_pktmbuf_prefree_seg(txep[i].mbuf);
if (m != NULL)
rte_mempool_put(m->pool, m);
}
}
/* buffers were freed, update counters */
txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
if (txq->tx_next_dd >= txq->nb_tx_desc)
txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
return txq->tx_rs_thresh;
}
static __rte_always_inline int
fm10k_tx_free_bufs(struct fm10k_tx_queue *txq)
{
struct rte_mbuf **txep;
uint8_t flags;
uint32_t n;
uint32_t i;
int nb_free = 0;
struct rte_mbuf *m, *free[RTE_FM10K_TX_MAX_FREE_BUF_SZ];
/* check DD bit on threshold descriptor */
flags = txq->hw_ring[txq->next_dd].flags;
if (!(flags & FM10K_TXD_FLAG_DONE))
return 0;
n = txq->rs_thresh;
/* First buffer to free from S/W ring is at index
* next_dd - (rs_thresh-1)
*/
txep = &txq->sw_ring[txq->next_dd - (n - 1)];
m = rte_pktmbuf_prefree_seg(txep[0]);
if (likely(m != NULL)) {
free[0] = m;
nb_free = 1;
for (i = 1; i < n; i++) {
m = rte_pktmbuf_prefree_seg(txep[i]);
if (likely(m != NULL)) {
if (likely(m->pool == free[0]->pool))
free[nb_free++] = m;
else {
rte_mempool_put_bulk(free[0]->pool,
(void *)free, nb_free);
free[0] = m;
nb_free = 1;
}
}
}
rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
} else {
for (i = 1; i < n; i++) {
m = rte_pktmbuf_prefree_seg(txep[i]);
if (m != NULL)
rte_mempool_put(m->pool, m);
}
}
/* buffers were freed, update counters */
txq->nb_free = (uint16_t)(txq->nb_free + txq->rs_thresh);
txq->next_dd = (uint16_t)(txq->next_dd + txq->rs_thresh);
if (txq->next_dd >= txq->nb_desc)
txq->next_dd = (uint16_t)(txq->rs_thresh - 1);
return txq->rs_thresh;
}
i40e_tx_free_bufs(struct i40e_tx_queue *txq)
{
struct i40e_tx_entry *txep;
uint32_t n;
uint32_t i;
int nb_free = 0;
struct rte_mbuf *m, *free[RTE_I40E_TX_MAX_FREE_BUF_SZ];
/* check DD bits on threshold descriptor */
if ((txq->tx_ring[txq->tx_next_dd].cmd_type_offset_bsz &
rte_cpu_to_le_64(I40E_TXD_QW1_DTYPE_MASK)) !=
rte_cpu_to_le_64(I40E_TX_DESC_DTYPE_DESC_DONE))
return 0;
n = txq->tx_rs_thresh;
/* first buffer to free from S/W ring is at index
* tx_next_dd - (tx_rs_thresh-1)
*/
txep = &txq->sw_ring[txq->tx_next_dd - (n - 1)];
m = __rte_pktmbuf_prefree_seg(txep[0].mbuf);
if (likely(m != NULL)) {
free[0] = m;
nb_free = 1;
for (i = 1; i < n; i++) {
m = __rte_pktmbuf_prefree_seg(txep[i].mbuf);
if (likely(m != NULL)) {
if (likely(m->pool == free[0]->pool)) {
free[nb_free++] = m;
} else {
rte_mempool_put_bulk(free[0]->pool,
(void *)free,
nb_free);
free[0] = m;
nb_free = 1;
}
}
}
rte_mempool_put_bulk(free[0]->pool, (void **)free, nb_free);
} else {
for (i = 1; i < n; i++) {
m = __rte_pktmbuf_prefree_seg(txep[i].mbuf);
if (m != NULL)
rte_mempool_put(m->pool, m);
}
}
/* buffers were freed, update counters */
txq->nb_tx_free = (uint16_t)(txq->nb_tx_free + txq->tx_rs_thresh);
txq->tx_next_dd = (uint16_t)(txq->tx_next_dd + txq->tx_rs_thresh);
if (txq->tx_next_dd >= txq->nb_tx_desc)
txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
return txq->tx_rs_thresh;
}
/* benchmark alloc-build-free of ops */
static inline int
pmd_cyclecount_bench_ops(struct pmd_cyclecount_state *state, uint32_t cur_op,
uint16_t test_burst_size)
{
uint32_t iter_ops_left = state->opts->total_ops - cur_op;
uint32_t iter_ops_needed =
RTE_MIN(state->opts->nb_descriptors, iter_ops_left);
uint32_t cur_iter_op;
uint32_t imix_idx = 0;
for (cur_iter_op = 0; cur_iter_op < iter_ops_needed;
cur_iter_op += test_burst_size) {
uint32_t burst_size = RTE_MIN(state->opts->total_ops - cur_op,
test_burst_size);
struct rte_crypto_op **ops = &state->ctx->ops[cur_iter_op];
/* Allocate objects containing crypto operations and mbufs */
if (rte_mempool_get_bulk(state->ctx->pool, (void **)ops,
burst_size) != 0) {
RTE_LOG(ERR, USER1,
"Failed to allocate more crypto operations "
"from the crypto operation pool.\n"
"Consider increasing the pool size "
"with --pool-sz\n");
return -1;
}
/* Setup crypto op, attach mbuf etc */
(state->ctx->populate_ops)(ops,
state->ctx->src_buf_offset,
state->ctx->dst_buf_offset,
burst_size,
state->ctx->sess, state->opts,
state->ctx->test_vector, iv_offset,
&imix_idx);
#ifdef CPERF_LINEARIZATION_ENABLE
/* Check if source mbufs require coalescing */
if (state->linearize) {
uint8_t i;
for (i = 0; i < burst_size; i++) {
struct rte_mbuf *src = ops[i]->sym->m_src;
rte_pktmbuf_linearize(src);
}
}
#endif /* CPERF_LINEARIZATION_ENABLE */
rte_mempool_put_bulk(state->ctx->pool, (void **)ops,
burst_size);
}
return 0;
}
/*
* This basic performance test just repeatedly sends in 32 packets at a time
* to the distributor and verifies at the end that we got them all in the worker
* threads and finally how long per packet the processing took.
*/
static inline int
perf_test(struct rte_distributor *d, struct rte_mempool *p)
{
unsigned int i;
uint64_t start, end;
struct rte_mbuf *bufs[BURST];
clear_packet_count();
if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
printf("Error getting mbufs from pool\n");
return -1;
}
/* ensure we have different hash value for each pkt */
for (i = 0; i < BURST; i++)
bufs[i]->hash.usr = i;
start = rte_rdtsc();
for (i = 0; i < (1<<ITER_POWER); i++)
rte_distributor_process(d, bufs, BURST);
end = rte_rdtsc();
do {
usleep(100);
rte_distributor_process(d, NULL, 0);
} while (total_packet_count() < (BURST << ITER_POWER));
rte_distributor_clear_returns(d);
printf("Time per burst: %"PRIu64"\n", (end - start) >> ITER_POWER);
printf("Time per packet: %"PRIu64"\n\n",
((end - start) >> ITER_POWER)/BURST);
rte_mempool_put_bulk(p, (void *)bufs, BURST);
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Total packets: %u (%x)\n", total_packet_count(),
total_packet_count());
printf("=== Perf test done ===\n\n");
return 0;
}
/* Useful function which ensures that all worker functions terminate */
static void
quit_workers(struct rte_distributor *d, struct rte_mempool *p)
{
const unsigned num_workers = rte_lcore_count() - 1;
unsigned i;
struct rte_mbuf *bufs[RTE_MAX_LCORE];
rte_mempool_get_bulk(p, (void *)bufs, num_workers);
quit = 1;
for (i = 0; i < num_workers; i++)
bufs[i]->hash.usr = i << 1;
rte_distributor_process(d, bufs, num_workers);
rte_mempool_put_bulk(p, (void *)bufs, num_workers);
rte_distributor_process(d, NULL, 0);
rte_eal_mp_wait_lcore();
quit = 0;
worker_idx = 0;
}
static void shuffle_mempool(struct rte_mempool* mempool, uint32_t nb_mbuf)
{
struct rte_mbuf** pkts = prox_zmalloc(nb_mbuf * sizeof(*pkts), rte_socket_id());
uint64_t got = 0;
while (rte_mempool_get_bulk(mempool, (void**)(pkts + got), 1) == 0)
++got;
while (got) {
int idx;
do {
idx = rand() % nb_mbuf - 1;
} while (pkts[idx] == 0);
rte_mempool_put_bulk(mempool, (void**)&pkts[idx], 1);
pkts[idx] = 0;
--got;
};
prox_free(pkts);
}
void
spdk_mempool_put_bulk(struct spdk_mempool *mp, void *const *ele_arr, size_t count)
{
rte_mempool_put_bulk((struct rte_mempool *)mp, ele_arr, count);
}
/* do basic sanity testing of the distributor. This test tests the following:
* - send 32 packets through distributor with the same tag and ensure they
* all go to the one worker
* - send 32 packets throught the distributor with two different tags and
* verify that they go equally to two different workers.
* - send 32 packets with different tags through the distributors and
* just verify we get all packets back.
* - send 1024 packets through the distributor, gathering the returned packets
* as we go. Then verify that we correctly got all 1024 pointers back again,
* not necessarily in the same order (as different flows).
*/
static int
sanity_test(struct rte_distributor *d, struct rte_mempool *p)
{
struct rte_mbuf *bufs[BURST];
unsigned i;
printf("=== Basic distributor sanity tests ===\n");
clear_packet_count();
if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
printf("line %d: Error getting mbufs from pool\n", __LINE__);
return -1;
}
/* now set all hash values in all buffers to zero, so all pkts go to the
* one worker thread */
for (i = 0; i < BURST; i++)
bufs[i]->pkt.hash.rss = 0;
rte_distributor_process(d, bufs, BURST);
rte_distributor_flush(d);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with all zero hashes done.\n");
if (worker_stats[0].handled_packets != BURST)
return -1;
/* pick two flows and check they go correctly */
if (rte_lcore_count() >= 3) {
clear_packet_count();
for (i = 0; i < BURST; i++)
bufs[i]->pkt.hash.rss = (i & 1) << 8;
rte_distributor_process(d, bufs, BURST);
rte_distributor_flush(d);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with two hash values done\n");
if (worker_stats[0].handled_packets != 16 ||
worker_stats[1].handled_packets != 16)
return -1;
}
/* give a different hash value to each packet,
* so load gets distributed */
clear_packet_count();
for (i = 0; i < BURST; i++)
bufs[i]->pkt.hash.rss = i;
rte_distributor_process(d, bufs, BURST);
rte_distributor_flush(d);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with non-zero hashes done\n");
rte_mempool_put_bulk(p, (void *)bufs, BURST);
/* sanity test with BIG_BATCH packets to ensure they all arrived back
* from the returned packets function */
clear_packet_count();
struct rte_mbuf *many_bufs[BIG_BATCH], *return_bufs[BIG_BATCH];
unsigned num_returned = 0;
/* flush out any remaining packets */
rte_distributor_flush(d);
rte_distributor_clear_returns(d);
if (rte_mempool_get_bulk(p, (void *)many_bufs, BIG_BATCH) != 0) {
printf("line %d: Error getting mbufs from pool\n", __LINE__);
return -1;
}
for (i = 0; i < BIG_BATCH; i++)
many_bufs[i]->pkt.hash.rss = i << 2;
for (i = 0; i < BIG_BATCH/BURST; i++) {
rte_distributor_process(d, &many_bufs[i*BURST], BURST);
num_returned += rte_distributor_returned_pkts(d,
&return_bufs[num_returned],
BIG_BATCH - num_returned);
}
rte_distributor_flush(d);
num_returned += rte_distributor_returned_pkts(d,
&return_bufs[num_returned], BIG_BATCH - num_returned);
if (num_returned != BIG_BATCH) {
printf("line %d: Number returned is not the same as "
"number sent\n", __LINE__);
return -1;
}
/* big check - make sure all packets made it back!! */
for (i = 0; i < BIG_BATCH; i++) {
unsigned j;
struct rte_mbuf *src = many_bufs[i];
for (j = 0; j < BIG_BATCH; j++)
if (return_bufs[j] == src)
break;
if (j == BIG_BATCH) {
printf("Error: could not find source packet #%u\n", i);
return -1;
}
}
printf("Sanity test of returned packets done\n");
rte_mempool_put_bulk(p, (void *)many_bufs, BIG_BATCH);
printf("\n");
return 0;
}
/* run benchmark per burst size */
static inline int
pmd_cyclecount_bench_burst_sz(
struct pmd_cyclecount_state *state, uint16_t test_burst_size)
{
uint64_t tsc_start;
uint64_t tsc_end;
uint64_t tsc_op;
uint64_t tsc_enq;
uint64_t tsc_deq;
uint32_t cur_op;
/* reset all counters */
tsc_enq = 0;
tsc_deq = 0;
state->ops_enqd = 0;
state->ops_enq_retries = 0;
state->ops_deqd = 0;
state->ops_deq_retries = 0;
/*
* Benchmark crypto op alloc-build-free separately.
*/
tsc_start = rte_rdtsc_precise();
for (cur_op = 0; cur_op < state->opts->total_ops;
cur_op += state->opts->nb_descriptors) {
if (unlikely(pmd_cyclecount_bench_ops(
state, cur_op, test_burst_size)))
return -1;
}
tsc_end = rte_rdtsc_precise();
tsc_op = tsc_end - tsc_start;
/*
* Hardware acceleration cyclecount benchmarking loop.
*
* We're benchmarking raw enq/deq performance by filling up the device
* queue, so we never get any failed enqs unless the driver won't accept
* the exact number of descriptors we requested, or the driver won't
* wrap around the end of the TX ring. However, since we're only
* dequeueing once we've filled up the queue, we have to benchmark it
* piecemeal and then average out the results.
*/
cur_op = 0;
while (cur_op < state->opts->total_ops) {
uint32_t iter_ops_left = state->opts->total_ops - cur_op;
uint32_t iter_ops_needed = RTE_MIN(
state->opts->nb_descriptors, iter_ops_left);
uint32_t iter_ops_allocd = iter_ops_needed;
/* allocate and build ops */
if (unlikely(pmd_cyclecount_build_ops(state, iter_ops_needed,
test_burst_size)))
return -1;
tsc_start = rte_rdtsc_precise();
/* fill up TX ring */
iter_ops_needed = pmd_cyclecount_bench_enq(state,
iter_ops_needed, test_burst_size);
tsc_end = rte_rdtsc_precise();
tsc_enq += tsc_end - tsc_start;
/* allow for HW to catch up */
if (state->delay)
rte_delay_us_block(state->delay);
tsc_start = rte_rdtsc_precise();
/* drain RX ring */
pmd_cyclecount_bench_deq(state, iter_ops_needed,
test_burst_size);
tsc_end = rte_rdtsc_precise();
tsc_deq += tsc_end - tsc_start;
cur_op += iter_ops_needed;
/*
* we may not have processed all ops that we allocated, so
* free everything we've allocated.
*/
rte_mempool_put_bulk(state->ctx->pool,
(void **)state->ctx->ops, iter_ops_allocd);
}
state->cycles_per_build = (double)tsc_op / state->opts->total_ops;
state->cycles_per_enq = (double)tsc_enq / state->ops_enqd;
state->cycles_per_deq = (double)tsc_deq / state->ops_deqd;
return 0;
}
/* do basic sanity testing of the distributor. This test tests the following:
* - send 32 packets through distributor with the same tag and ensure they
* all go to the one worker
* - send 32 packets through the distributor with two different tags and
* verify that they go equally to two different workers.
* - send 32 packets with different tags through the distributors and
* just verify we get all packets back.
* - send 1024 packets through the distributor, gathering the returned packets
* as we go. Then verify that we correctly got all 1024 pointers back again,
* not necessarily in the same order (as different flows).
*/
static int
sanity_test(struct worker_params *wp, struct rte_mempool *p)
{
struct rte_distributor *db = wp->dist;
struct rte_mbuf *bufs[BURST];
struct rte_mbuf *returns[BURST*2];
unsigned int i, count;
unsigned int retries;
printf("=== Basic distributor sanity tests ===\n");
clear_packet_count();
if (rte_mempool_get_bulk(p, (void *)bufs, BURST) != 0) {
printf("line %d: Error getting mbufs from pool\n", __LINE__);
return -1;
}
/* now set all hash values in all buffers to zero, so all pkts go to the
* one worker thread */
for (i = 0; i < BURST; i++)
bufs[i]->hash.usr = 0;
rte_distributor_process(db, bufs, BURST);
count = 0;
do {
rte_distributor_flush(db);
count += rte_distributor_returned_pkts(db,
returns, BURST*2);
} while (count < BURST);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with all zero hashes done.\n");
/* pick two flows and check they go correctly */
if (rte_lcore_count() >= 3) {
clear_packet_count();
for (i = 0; i < BURST; i++)
bufs[i]->hash.usr = (i & 1) << 8;
rte_distributor_process(db, bufs, BURST);
count = 0;
do {
rte_distributor_flush(db);
count += rte_distributor_returned_pkts(db,
returns, BURST*2);
} while (count < BURST);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with two hash values done\n");
}
/* give a different hash value to each packet,
* so load gets distributed */
clear_packet_count();
for (i = 0; i < BURST; i++)
bufs[i]->hash.usr = i+1;
rte_distributor_process(db, bufs, BURST);
count = 0;
do {
rte_distributor_flush(db);
count += rte_distributor_returned_pkts(db,
returns, BURST*2);
} while (count < BURST);
if (total_packet_count() != BURST) {
printf("Line %d: Error, not all packets flushed. "
"Expected %u, got %u\n",
__LINE__, BURST, total_packet_count());
return -1;
}
for (i = 0; i < rte_lcore_count() - 1; i++)
printf("Worker %u handled %u packets\n", i,
worker_stats[i].handled_packets);
printf("Sanity test with non-zero hashes done\n");
rte_mempool_put_bulk(p, (void *)bufs, BURST);
/* sanity test with BIG_BATCH packets to ensure they all arrived back
* from the returned packets function */
clear_packet_count();
struct rte_mbuf *many_bufs[BIG_BATCH], *return_bufs[BIG_BATCH];
unsigned num_returned = 0;
/* flush out any remaining packets */
rte_distributor_flush(db);
rte_distributor_clear_returns(db);
if (rte_mempool_get_bulk(p, (void *)many_bufs, BIG_BATCH) != 0) {
printf("line %d: Error getting mbufs from pool\n", __LINE__);
return -1;
}
for (i = 0; i < BIG_BATCH; i++)
many_bufs[i]->hash.usr = i << 2;
printf("=== testing big burst (%s) ===\n", wp->name);
for (i = 0; i < BIG_BATCH/BURST; i++) {
rte_distributor_process(db,
&many_bufs[i*BURST], BURST);
count = rte_distributor_returned_pkts(db,
&return_bufs[num_returned],
BIG_BATCH - num_returned);
num_returned += count;
}
rte_distributor_flush(db);
count = rte_distributor_returned_pkts(db,
&return_bufs[num_returned],
BIG_BATCH - num_returned);
num_returned += count;
retries = 0;
do {
rte_distributor_flush(db);
count = rte_distributor_returned_pkts(db,
&return_bufs[num_returned],
BIG_BATCH - num_returned);
num_returned += count;
retries++;
} while ((num_returned < BIG_BATCH) && (retries < 100));
if (num_returned != BIG_BATCH) {
printf("line %d: Missing packets, expected %d\n",
__LINE__, num_returned);
return -1;
}
/* big check - make sure all packets made it back!! */
for (i = 0; i < BIG_BATCH; i++) {
unsigned j;
struct rte_mbuf *src = many_bufs[i];
for (j = 0; j < BIG_BATCH; j++) {
if (return_bufs[j] == src)
break;
}
if (j == BIG_BATCH) {
printf("Error: could not find source packet #%u\n", i);
return -1;
}
}
printf("Sanity test of returned packets done\n");
rte_mempool_put_bulk(p, (void *)many_bufs, BIG_BATCH);
printf("\n");
return 0;
}
int
cperf_verify_test_runner(void *test_ctx)
{
struct cperf_verify_ctx *ctx = test_ctx;
uint64_t ops_enqd = 0, ops_enqd_total = 0, ops_enqd_failed = 0;
uint64_t ops_deqd = 0, ops_deqd_total = 0, ops_deqd_failed = 0;
uint64_t ops_failed = 0;
static int only_once;
uint64_t i;
uint16_t ops_unused = 0;
struct rte_crypto_op *ops[ctx->options->max_burst_size];
struct rte_crypto_op *ops_processed[ctx->options->max_burst_size];
uint32_t lcore = rte_lcore_id();
#ifdef CPERF_LINEARIZATION_ENABLE
struct rte_cryptodev_info dev_info;
int linearize = 0;
/* Check if source mbufs require coalescing */
if (ctx->options->segment_sz < ctx->options->max_buffer_size) {
rte_cryptodev_info_get(ctx->dev_id, &dev_info);
if ((dev_info.feature_flags &
RTE_CRYPTODEV_FF_MBUF_SCATTER_GATHER) == 0)
linearize = 1;
}
#endif /* CPERF_LINEARIZATION_ENABLE */
ctx->lcore_id = lcore;
if (!ctx->options->csv)
printf("\n# Running verify test on device: %u, lcore: %u\n",
ctx->dev_id, lcore);
uint16_t iv_offset = sizeof(struct rte_crypto_op) +
sizeof(struct rte_crypto_sym_op);
while (ops_enqd_total < ctx->options->total_ops) {
uint16_t burst_size = ((ops_enqd_total + ctx->options->max_burst_size)
<= ctx->options->total_ops) ?
ctx->options->max_burst_size :
ctx->options->total_ops -
ops_enqd_total;
uint16_t ops_needed = burst_size - ops_unused;
/* Allocate objects containing crypto operations and mbufs */
if (rte_mempool_get_bulk(ctx->pool, (void **)ops,
ops_needed) != 0) {
RTE_LOG(ERR, USER1,
"Failed to allocate more crypto operations "
"from the the crypto operation pool.\n"
"Consider increasing the pool size "
"with --pool-sz\n");
return -1;
}
/* Setup crypto op, attach mbuf etc */
(ctx->populate_ops)(ops, ctx->src_buf_offset,
ctx->dst_buf_offset,
ops_needed, ctx->sess, ctx->options,
ctx->test_vector, iv_offset);
/* Populate the mbuf with the test vector, for verification */
for (i = 0; i < ops_needed; i++)
cperf_mbuf_set(ops[i]->sym->m_src,
ctx->options,
ctx->test_vector);
#ifdef CPERF_LINEARIZATION_ENABLE
if (linearize) {
/* PMD doesn't support scatter-gather and source buffer
* is segmented.
* We need to linearize it before enqueuing.
*/
for (i = 0; i < burst_size; i++)
rte_pktmbuf_linearize(ops[i]->sym->m_src);
}
#endif /* CPERF_LINEARIZATION_ENABLE */
/* Enqueue burst of ops on crypto device */
ops_enqd = rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id,
ops, burst_size);
if (ops_enqd < burst_size)
ops_enqd_failed++;
/**
* Calculate number of ops not enqueued (mainly for hw
* accelerators whose ingress queue can fill up).
*/
ops_unused = burst_size - ops_enqd;
ops_enqd_total += ops_enqd;
/* Dequeue processed burst of ops from crypto device */
ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id,
ops_processed, ctx->options->max_burst_size);
if (ops_deqd == 0) {
/**
* Count dequeue polls which didn't return any
* processed operations. This statistic is mainly
* relevant to hw accelerators.
*/
ops_deqd_failed++;
continue;
}
for (i = 0; i < ops_deqd; i++) {
if (cperf_verify_op(ops_processed[i], ctx->options,
ctx->test_vector))
ops_failed++;
}
/* Free crypto ops so they can be reused. */
rte_mempool_put_bulk(ctx->pool,
(void **)ops_processed, ops_deqd);
ops_deqd_total += ops_deqd;
}
/* Dequeue any operations still in the crypto device */
while (ops_deqd_total < ctx->options->total_ops) {
/* Sending 0 length burst to flush sw crypto device */
rte_cryptodev_enqueue_burst(ctx->dev_id, ctx->qp_id, NULL, 0);
/* dequeue burst */
ops_deqd = rte_cryptodev_dequeue_burst(ctx->dev_id, ctx->qp_id,
ops_processed, ctx->options->max_burst_size);
if (ops_deqd == 0) {
ops_deqd_failed++;
continue;
}
for (i = 0; i < ops_deqd; i++) {
if (cperf_verify_op(ops_processed[i], ctx->options,
ctx->test_vector))
ops_failed++;
}
/* Free crypto ops so they can be reused. */
rte_mempool_put_bulk(ctx->pool,
(void **)ops_processed, ops_deqd);
ops_deqd_total += ops_deqd;
}
if (!ctx->options->csv) {
if (!only_once)
printf("%12s%12s%12s%12s%12s%12s%12s%12s\n\n",
"lcore id", "Buf Size", "Burst size",
"Enqueued", "Dequeued", "Failed Enq",
"Failed Deq", "Failed Ops");
only_once = 1;
printf("%12u%12u%12u%12"PRIu64"%12"PRIu64"%12"PRIu64
"%12"PRIu64"%12"PRIu64"\n",
ctx->lcore_id,
ctx->options->max_buffer_size,
ctx->options->max_burst_size,
ops_enqd_total,
ops_deqd_total,
ops_enqd_failed,
ops_deqd_failed,
ops_failed);
} else {
if (!only_once)
printf("\n# lcore id, Buffer Size(B), "
"Burst Size,Enqueued,Dequeued,Failed Enq,"
"Failed Deq,Failed Ops\n");
only_once = 1;
printf("%10u;%10u;%u;%"PRIu64";%"PRIu64";%"PRIu64";%"PRIu64";"
"%"PRIu64"\n",
ctx->lcore_id,
ctx->options->max_buffer_size,
ctx->options->max_burst_size,
ops_enqd_total,
ops_deqd_total,
ops_enqd_failed,
ops_deqd_failed,
ops_failed);
}
return 0;
}
