/*
* TODO: we could "error proof" these as done in test_hash_perf.c ln 165:
*
* The current setup may give errors if too full in some cases which we check
* for. However, since EFD allows for ~99% capacity, these errors are rare for
* #"KEYS_TO_ADD" which is 75% capacity.
*/
static int
setup_keys_and_data(struct efd_perf_params *params, unsigned int cycle)
{
unsigned int i, j;
int num_duplicates;
params->key_size = hashtest_key_lens[cycle];
params->cycle = cycle;
/* Reset all arrays */
for (i = 0; i < params->key_size; i++)
keys[0][i] = 0;
/* Generate a list of keys, some of which may be duplicates */
for (i = 0; i < KEYS_TO_ADD; i++) {
for (j = 0; j < params->key_size; j++)
keys[i][j] = rte_rand() & 0xFF;
data[i] = rte_rand() & VALUE_BITMASK;
}
/* Remove duplicates from the keys array */
do {
num_duplicates = 0;
/* Sort the list of keys to make it easier to find duplicates */
qsort(keys, KEYS_TO_ADD, MAX_KEYSIZE, key_compare);
/* Sift through the list of keys and look for duplicates */
int num_duplicates = 0;
for (i = 0; i < KEYS_TO_ADD - 1; i++) {
if (memcmp(keys[i], keys[i + 1], params->key_size) == 0) {
/* This key already exists, try again */
num_duplicates++;
for (j = 0; j < params->key_size; j++)
keys[i][j] = rte_rand() & 0xFF;
}
}
} while (num_duplicates != 0);
/* Shuffle the random values again */
shuffle_input_keys(params);
params->efd_table = rte_efd_create("test_efd_perf",
MAX_ENTRIES, params->key_size,
efd_get_all_sockets_bitmask(), test_socket_id);
TEST_ASSERT_NOT_NULL(params->efd_table, "Error creating the efd table\n");
return 0;
}
/* Shuffle the keys that have been added, so lookups will be totally random */
static void
shuffle_input_keys(unsigned table_index)
{
unsigned i;
uint32_t swap_idx;
uint8_t temp_key[MAX_KEYSIZE];
hash_sig_t temp_signature;
int32_t temp_position;
for (i = KEYS_TO_ADD - 1; i > 0; i--) {
swap_idx = rte_rand() % i;
memcpy(temp_key, keys[i], hashtest_key_lens[table_index]);
temp_signature = signatures[i];
temp_position = positions[i];
memcpy(keys[i], keys[swap_idx], hashtest_key_lens[table_index]);
signatures[i] = signatures[swap_idx];
positions[i] = positions[swap_idx];
memcpy(keys[swap_idx], temp_key, hashtest_key_lens[table_index]);
signatures[swap_idx] = temp_signature;
positions[swap_idx] = temp_position;
}
}
static int
avf_init_rss(struct avf_adapter *adapter)
{
struct avf_info *vf = AVF_DEV_PRIVATE_TO_VF(adapter);
struct avf_hw *hw = AVF_DEV_PRIVATE_TO_HW(adapter);
struct rte_eth_rss_conf *rss_conf;
uint8_t i, j, nb_q;
int ret;
rss_conf = &adapter->eth_dev->data->dev_conf.rx_adv_conf.rss_conf;
nb_q = RTE_MIN(adapter->eth_dev->data->nb_rx_queues,
AVF_MAX_NUM_QUEUES);
if (!(vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF)) {
PMD_DRV_LOG(DEBUG, "RSS is not supported");
return -ENOTSUP;
}
if (adapter->eth_dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
PMD_DRV_LOG(WARNING, "RSS is enabled by PF by default");
/* set all lut items to default queue */
for (i = 0; i < vf->vf_res->rss_lut_size; i++)
vf->rss_lut[i] = 0;
ret = avf_configure_rss_lut(adapter);
return ret;
}
/* In AVF, RSS enablement is set by PF driver. It is not supported
* to set based on rss_conf->rss_hf.
*/
/* configure RSS key */
if (!rss_conf->rss_key) {
/* Calculate the default hash key */
for (i = 0; i <= vf->vf_res->rss_key_size; i++)
vf->rss_key[i] = (uint8_t)rte_rand();
} else
rte_memcpy(vf->rss_key, rss_conf->rss_key,
RTE_MIN(rss_conf->rss_key_len,
vf->vf_res->rss_key_size));
/* init RSS LUT table */
for (i = 0, j = 0; i < vf->vf_res->rss_lut_size; i++, j++) {
if (j >= nb_q)
j = 0;
vf->rss_lut[i] = j;
}
/* send virtchnnl ops to configure rss*/
ret = avf_configure_rss_lut(adapter);
if (ret)
return ret;
ret = avf_configure_rss_key(adapter);
if (ret)
return ret;
return 0;
}
static int hw_atl_utils_get_mac_permanent(struct aq_hw_s *self,
u8 *mac)
{
int err = 0;
u32 h = 0U;
u32 l = 0U;
u32 mac_addr[2];
if (!aq_hw_read_reg(self, HW_ATL_UCP_0X370_REG)) {
unsigned int rnd = (uint32_t)rte_rand();
unsigned int ucp_0x370 = 0;
//get_random_bytes(&rnd, sizeof(unsigned int));
ucp_0x370 = 0x02020202 | (0xFEFEFEFE & rnd);
aq_hw_write_reg(self, HW_ATL_UCP_0X370_REG, ucp_0x370);
}
err = hw_atl_utils_fw_downld_dwords(self,
aq_hw_read_reg(self, 0x00000374U) +
(40U * 4U),
mac_addr,
ARRAY_SIZE(mac_addr));
if (err < 0) {
mac_addr[0] = 0U;
mac_addr[1] = 0U;
err = 0;
} else {
mac_addr[0] = rte_constant_bswap32(mac_addr[0]);
mac_addr[1] = rte_constant_bswap32(mac_addr[1]);
}
ether_addr_copy((struct ether_addr *)mac_addr,
(struct ether_addr *)mac);
if ((mac[0] & 0x01U) || ((mac[0] | mac[1] | mac[2]) == 0x00U)) {
/* chip revision */
l = 0xE3000000U
| (0xFFFFU & aq_hw_read_reg(self, HW_ATL_UCP_0X370_REG))
| (0x00 << 16);
h = 0x8001300EU;
mac[5] = (u8)(0xFFU & l);
l >>= 8;
mac[4] = (u8)(0xFFU & l);
l >>= 8;
mac[3] = (u8)(0xFFU & l);
l >>= 8;
mac[2] = (u8)(0xFFU & l);
mac[1] = (u8)(0xFFU & h);
h >>= 8;
mac[0] = (u8)(0xFFU & h);
}
/*
* Do a single performance test, of one type of operation.
*
* @param h
* hash table to run test on
* @param func
* function to call (add, delete or lookup function)
* @param avg_occupancy
* The average number of entries in each bucket of the hash table
* @param invalid_pos_count
* The amount of errors (e.g. due to a full bucket).
* @return
* The average number of ticks per hash function call. A negative number
* signifies failure.
*/
static double
run_single_tbl_perf_test(const struct rte_hash *h, hash_operation func,
const struct tbl_perf_test_params *params, double *avg_occupancy,
uint32_t *invalid_pos_count)
{
uint64_t begin, end, ticks = 0;
uint8_t *key = NULL;
uint32_t *bucket_occupancies = NULL;
uint32_t num_buckets, i, j;
int32_t pos;
/* Initialise */
num_buckets = params->entries / params->bucket_entries;
key = (uint8_t *) rte_zmalloc("hash key",
params->key_len * sizeof(uint8_t), 16);
if (key == NULL)
return -1;
bucket_occupancies = (uint32_t *) rte_zmalloc("bucket occupancies",
num_buckets * sizeof(uint32_t), 16);
if (bucket_occupancies == NULL) {
rte_free(key);
return -1;
}
ticks = 0;
*invalid_pos_count = 0;
for (i = 0; i < params->num_iterations; i++) {
/* Prepare inputs for the current iteration */
for (j = 0; j < params->key_len; j++)
key[j] = (uint8_t) rte_rand();
/* Perform operation, and measure time it takes */
begin = rte_rdtsc();
pos = func(h, key);
end = rte_rdtsc();
ticks += end - begin;
/* Other work per iteration */
if (pos < 0)
*invalid_pos_count += 1;
else
bucket_occupancies[pos / params->bucket_entries]++;
}
*avg_occupancy = get_avg(bucket_occupancies, num_buckets);
rte_free(bucket_occupancies);
rte_free(key);
return (double)ticks / params->num_iterations;
}
/* Initialise data buffers. */
static int
init_buffers(void)
{
unsigned i;
large_buf_read = rte_malloc("memcpy", LARGE_BUFFER_SIZE, ALIGNMENT_UNIT);
if (large_buf_read == NULL)
goto error_large_buf_read;
large_buf_write = rte_malloc("memcpy", LARGE_BUFFER_SIZE, ALIGNMENT_UNIT);
if (large_buf_write == NULL)
goto error_large_buf_write;
small_buf_read = rte_malloc("memcpy", SMALL_BUFFER_SIZE, ALIGNMENT_UNIT);
if (small_buf_read == NULL)
goto error_small_buf_read;
small_buf_write = rte_malloc("memcpy", SMALL_BUFFER_SIZE, ALIGNMENT_UNIT);
if (small_buf_write == NULL)
goto error_small_buf_write;
for (i = 0; i < LARGE_BUFFER_SIZE; i++)
large_buf_read[i] = rte_rand();
for (i = 0; i < SMALL_BUFFER_SIZE; i++)
small_buf_read[i] = rte_rand();
return 0;
error_small_buf_write:
rte_free(small_buf_read);
error_small_buf_read:
rte_free(large_buf_write);
error_large_buf_write:
rte_free(large_buf_read);
error_large_buf_read:
printf("ERROR: not enough memory\n");
return -1;
}
int
rte_red_config_init(struct rte_red_config *red_cfg,
const uint16_t wq_log2,
const uint16_t min_th,
const uint16_t max_th,
const uint16_t maxp_inv)
{
if (red_cfg == NULL) {
return -1;
}
if (max_th > RTE_RED_MAX_TH_MAX) {
return -2;
}
if (min_th >= max_th) {
return -3;
}
if (wq_log2 > RTE_RED_WQ_LOG2_MAX) {
return -4;
}
if (wq_log2 < RTE_RED_WQ_LOG2_MIN) {
return -5;
}
if (maxp_inv < RTE_RED_MAXP_INV_MIN) {
return -6;
}
if (maxp_inv > RTE_RED_MAXP_INV_MAX) {
return -7;
}
/**
* Initialize the RED module if not already done
*/
if (!rte_red_init_done) {
rte_red_rand_seed = rte_rand();
rte_red_rand_val = rte_fast_rand();
__rte_red_init_tables();
rte_red_init_done = 1;
}
red_cfg->min_th = ((uint32_t) min_th) << (wq_log2 + RTE_RED_SCALING);
red_cfg->max_th = ((uint32_t) max_th) << (wq_log2 + RTE_RED_SCALING);
red_cfg->pa_const = (2 * (max_th - min_th) * maxp_inv) << RTE_RED_SCALING;
red_cfg->maxp_inv = maxp_inv;
red_cfg->wq_log2 = wq_log2;
return 0;
}
/* Shuffle the keys that have been added, so lookups will be totally random */
static void
shuffle_input_keys(struct efd_perf_params *params)
{
efd_value_t temp_data;
unsigned int i;
uint32_t swap_idx;
uint8_t temp_key[MAX_KEYSIZE];
for (i = KEYS_TO_ADD - 1; i > 0; i--) {
swap_idx = rte_rand() % i;
memcpy(temp_key, keys[i], hashtest_key_lens[params->cycle]);
temp_data = data[i];
memcpy(keys[i], keys[swap_idx], hashtest_key_lens[params->cycle]);
data[i] = data[swap_idx];
memcpy(keys[swap_idx], temp_key, hashtest_key_lens[params->cycle]);
data[swap_idx] = temp_data;
}
}
/*
* Test a hash function.
*/
static void run_hash_func_test(rte_hash_function f, uint32_t init_val,
uint32_t key_len)
{
static uint8_t key[RTE_HASH_KEY_LENGTH_MAX];
uint64_t ticks = 0, start, end;
unsigned i, j;
for (i = 0; i < HASHTEST_ITERATIONS; i++) {
for (j = 0; j < key_len; j++)
key[j] = (uint8_t) rte_rand();
start = rte_rdtsc();
f(key, key_len, init_val);
end = rte_rdtsc();
ticks += end - start;
}
printf("%-12s, %-18u, %-13u, %.02f\n", get_hash_name(f), (unsigned) key_len,
(unsigned) init_val, (double)ticks / HASHTEST_ITERATIONS);
}
static int hw_atl_utils_init_ucp(struct aq_hw_s *self)
{
int err = 0;
if (!aq_hw_read_reg(self, 0x370U)) {
unsigned int rnd = (uint32_t)rte_rand();
unsigned int ucp_0x370 = 0U;
ucp_0x370 = 0x02020202U | (0xFEFEFEFEU & rnd);
aq_hw_write_reg(self, HW_ATL_UCP_0X370_REG, ucp_0x370);
}
hw_atl_reg_glb_cpu_scratch_scp_set(self, 0x00000000U, 25U);
/* check 10 times by 1ms */
AQ_HW_WAIT_FOR(0U != (self->mbox_addr =
aq_hw_read_reg(self, 0x360U)), 1000U, 10U);
AQ_HW_WAIT_FOR(0U != (self->rpc_addr =
aq_hw_read_reg(self, 0x334U)), 1000U, 100U);
return err;
}
static int
test_memzone_reserve_max_aligned(void)
{
const struct rte_memzone *mz;
const struct rte_config *config;
const struct rte_memseg *ms;
int memseg_idx = 0;
int memzone_idx = 0;
uintptr_t addr_offset;
size_t len = 0;
void* last_addr;
size_t maxlen = 0;
/* random alignment */
rte_srand((unsigned)rte_rdtsc());
const unsigned align = 1 << ((rte_rand() % 8) + 5); /* from 128 up to 4k alignment */
/* get pointer to global configuration */
config = rte_eal_get_configuration();
ms = rte_eal_get_physmem_layout();
addr_offset = 0;
for (memseg_idx = 0; memseg_idx < RTE_MAX_MEMSEG; memseg_idx++){
/* ignore smaller memsegs as they can only get smaller */
if (ms[memseg_idx].len < maxlen)
continue;
/* align everything */
last_addr = RTE_PTR_ALIGN_CEIL(ms[memseg_idx].addr, RTE_CACHE_LINE_SIZE);
len = ms[memseg_idx].len - RTE_PTR_DIFF(last_addr, ms[memseg_idx].addr);
len &= ~((size_t) RTE_CACHE_LINE_MASK);
/* cycle through all memzones */
for (memzone_idx = 0; memzone_idx < RTE_MAX_MEMZONE; memzone_idx++) {
/* stop when reaching last allocated memzone */
if (config->mem_config->memzone[memzone_idx].addr == NULL)
break;
/* check if the memzone is in our memseg and subtract length */
if ((config->mem_config->memzone[memzone_idx].addr >=
ms[memseg_idx].addr) &&
(config->mem_config->memzone[memzone_idx].addr <
(RTE_PTR_ADD(ms[memseg_idx].addr, ms[memseg_idx].len)))) {
/* since the zones can now be aligned and occasionally skip
* some space, we should calculate the length based on
* reported length and start addresses difference.
*/
len -= (uintptr_t) RTE_PTR_SUB(
config->mem_config->memzone[memzone_idx].addr,
(uintptr_t) last_addr);
len -= config->mem_config->memzone[memzone_idx].len;
last_addr =
RTE_PTR_ADD(config->mem_config->memzone[memzone_idx].addr,
(size_t) config->mem_config->memzone[memzone_idx].len);
}
}
/* make sure we get the alignment offset */
if (len > maxlen) {
addr_offset = RTE_PTR_ALIGN_CEIL((uintptr_t) last_addr, align) - (uintptr_t) last_addr;
maxlen = len;
}
}
if (maxlen == 0 || maxlen == addr_offset) {
printf("There is no space left for biggest %u-aligned memzone!\n", align);
return 0;
}
maxlen -= addr_offset;
mz = rte_memzone_reserve_aligned("max_zone_aligned", 0,
SOCKET_ID_ANY, 0, align);
if (mz == NULL){
printf("Failed to reserve a big chunk of memory\n");
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
if (mz->len != maxlen) {
printf("Memzone reserve with 0 size and alignment %u did not return"
" bigest block\n", align);
printf("Expected size = %zu, actual size = %zu\n",
maxlen, mz->len);
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
return 0;
}
static int
fbk_hash_perf_test(void)
{
struct rte_fbk_hash_params params = {
.name = "fbk_hash_test",
.entries = ENTRIES,
.entries_per_bucket = 4,
.socket_id = rte_socket_id(),
};
struct rte_fbk_hash_table *handle = NULL;
uint32_t *keys = NULL;
unsigned indexes[TEST_SIZE];
uint64_t lookup_time = 0;
unsigned added = 0;
unsigned value = 0;
uint32_t key;
uint16_t val;
unsigned i, j;
handle = rte_fbk_hash_create(¶ms);
if (handle == NULL) {
printf("Error creating table\n");
return -1;
}
keys = rte_zmalloc(NULL, ENTRIES * sizeof(*keys), 0);
if (keys == NULL) {
printf("fbk hash: memory allocation for key store failed\n");
return -1;
}
/* Generate random keys and values. */
for (i = 0; i < ENTRIES; i++) {
key = (uint32_t)rte_rand();
key = ((uint64_t)key << 32) | (uint64_t)rte_rand();
val = (uint16_t)rte_rand();
if (rte_fbk_hash_add_key(handle, key, val) == 0) {
keys[added] = key;
added++;
}
if (added > (LOAD_FACTOR * ENTRIES))
break;
}
for (i = 0; i < TEST_ITERATIONS; i++) {
uint64_t begin;
uint64_t end;
/* Generate random indexes into keys[] array. */
for (j = 0; j < TEST_SIZE; j++)
indexes[j] = rte_rand() % added;
begin = rte_rdtsc();
/* Do lookups */
for (j = 0; j < TEST_SIZE; j++)
value += rte_fbk_hash_lookup(handle, keys[indexes[j]]);
end = rte_rdtsc();
lookup_time += (double)(end - begin);
}
printf("\n\n *** FBK Hash function performance test results ***\n");
/*
* The use of the 'value' variable ensures that the hash lookup is not
* being optimised out by the compiler.
*/
if (value != 0)
printf("Number of ticks per lookup = %g\n",
(double)lookup_time /
((double)TEST_ITERATIONS * (double)TEST_SIZE));
rte_fbk_hash_free(handle);
return 0;
}
static int
test_hash_perf(void)
{
unsigned with_pushes;
for (with_pushes = 0; with_pushes <= 1; with_pushes++) {
if (with_pushes == 0)
printf("\nALL ELEMENTS IN PRIMARY LOCATION\n");
else
printf("\nELEMENTS IN PRIMARY OR SECONDARY LOCATION\n");
if (run_all_tbl_perf_tests(with_pushes) < 0)
return -1;
}
if (fbk_hash_perf_test() < 0)
return -1;
return 0;
}
static int
cperf_initialize_cryptodev(struct cperf_options *opts, uint8_t *enabled_cdevs)
{
uint8_t enabled_cdev_count = 0, nb_lcores, cdev_id;
uint32_t sessions_needed = 0;
unsigned int i, j;
int ret;
enabled_cdev_count = rte_cryptodev_devices_get(opts->device_type,
enabled_cdevs, RTE_CRYPTO_MAX_DEVS);
if (enabled_cdev_count == 0) {
printf("No crypto devices type %s available\n",
opts->device_type);
return -EINVAL;
}
nb_lcores = rte_lcore_count() - 1;
if (nb_lcores < 1) {
RTE_LOG(ERR, USER1,
"Number of enabled cores need to be higher than 1\n");
return -EINVAL;
}
/*
* Use less number of devices,
* if there are more available than cores.
*/
if (enabled_cdev_count > nb_lcores)
enabled_cdev_count = nb_lcores;
/* Create a mempool shared by all the devices */
uint32_t max_sess_size = 0, sess_size;
for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
sess_size = rte_cryptodev_sym_get_private_session_size(cdev_id);
if (sess_size > max_sess_size)
max_sess_size = sess_size;
}
/*
* Calculate number of needed queue pairs, based on the amount
* of available number of logical cores and crypto devices.
* For instance, if there are 4 cores and 2 crypto devices,
* 2 queue pairs will be set up per device.
*/
opts->nb_qps = (nb_lcores % enabled_cdev_count) ?
(nb_lcores / enabled_cdev_count) + 1 :
nb_lcores / enabled_cdev_count;
for (i = 0; i < enabled_cdev_count &&
i < RTE_CRYPTO_MAX_DEVS; i++) {
cdev_id = enabled_cdevs[i];
#ifdef RTE_LIBRTE_PMD_CRYPTO_SCHEDULER
/*
* If multi-core scheduler is used, limit the number
* of queue pairs to 1, as there is no way to know
* how many cores are being used by the PMD, and
* how many will be available for the application.
*/
if (!strcmp((const char *)opts->device_type, "crypto_scheduler") &&
rte_cryptodev_scheduler_mode_get(cdev_id) ==
CDEV_SCHED_MODE_MULTICORE)
opts->nb_qps = 1;
#endif
struct rte_cryptodev_info cdev_info;
uint8_t socket_id = rte_cryptodev_socket_id(cdev_id);
rte_cryptodev_info_get(cdev_id, &cdev_info);
if (opts->nb_qps > cdev_info.max_nb_queue_pairs) {
printf("Number of needed queue pairs is higher "
"than the maximum number of queue pairs "
"per device.\n");
printf("Lower the number of cores or increase "
"the number of crypto devices\n");
return -EINVAL;
}
struct rte_cryptodev_config conf = {
.nb_queue_pairs = opts->nb_qps,
.socket_id = socket_id
};
struct rte_cryptodev_qp_conf qp_conf = {
.nb_descriptors = opts->nb_descriptors
};
/**
* Device info specifies the min headroom and tailroom
* requirement for the crypto PMD. This need to be honoured
* by the application, while creating mbuf.
*/
if (opts->headroom_sz < cdev_info.min_mbuf_headroom_req) {
/* Update headroom */
opts->headroom_sz = cdev_info.min_mbuf_headroom_req;
}
if (opts->tailroom_sz < cdev_info.min_mbuf_tailroom_req) {
/* Update tailroom */
opts->tailroom_sz = cdev_info.min_mbuf_tailroom_req;
}
/* Update segment size to include headroom & tailroom */
opts->segment_sz += (opts->headroom_sz + opts->tailroom_sz);
uint32_t dev_max_nb_sess = cdev_info.sym.max_nb_sessions;
/*
* Two sessions objects are required for each session
* (one for the header, one for the private data)
*/
if (!strcmp((const char *)opts->device_type,
"crypto_scheduler")) {
#ifdef RTE_LIBRTE_PMD_CRYPTO_SCHEDULER
uint32_t nb_slaves =
rte_cryptodev_scheduler_slaves_get(cdev_id,
NULL);
sessions_needed = enabled_cdev_count *
opts->nb_qps * nb_slaves;
#endif
} else
sessions_needed = enabled_cdev_count *
opts->nb_qps;
/*
* A single session is required per queue pair
* in each device
*/
if (dev_max_nb_sess != 0 && dev_max_nb_sess < opts->nb_qps) {
RTE_LOG(ERR, USER1,
"Device does not support at least "
"%u sessions\n", opts->nb_qps);
return -ENOTSUP;
}
ret = fill_session_pool_socket(socket_id, max_sess_size,
sessions_needed);
if (ret < 0)
return ret;
qp_conf.mp_session = session_pool_socket[socket_id].sess_mp;
qp_conf.mp_session_private =
session_pool_socket[socket_id].priv_mp;
ret = rte_cryptodev_configure(cdev_id, &conf);
if (ret < 0) {
printf("Failed to configure cryptodev %u", cdev_id);
return -EINVAL;
}
for (j = 0; j < opts->nb_qps; j++) {
ret = rte_cryptodev_queue_pair_setup(cdev_id, j,
&qp_conf, socket_id);
if (ret < 0) {
printf("Failed to setup queue pair %u on "
"cryptodev %u", j, cdev_id);
return -EINVAL;
}
}
ret = rte_cryptodev_start(cdev_id);
if (ret < 0) {
printf("Failed to start device %u: error %d\n",
cdev_id, ret);
return -EPERM;
}
}
return enabled_cdev_count;
}
static int
cperf_verify_devices_capabilities(struct cperf_options *opts,
uint8_t *enabled_cdevs, uint8_t nb_cryptodevs)
{
struct rte_cryptodev_sym_capability_idx cap_idx;
const struct rte_cryptodev_symmetric_capability *capability;
uint8_t i, cdev_id;
int ret;
for (i = 0; i < nb_cryptodevs; i++) {
cdev_id = enabled_cdevs[i];
if (opts->op_type == CPERF_AUTH_ONLY ||
opts->op_type == CPERF_CIPHER_THEN_AUTH ||
opts->op_type == CPERF_AUTH_THEN_CIPHER) {
cap_idx.type = RTE_CRYPTO_SYM_XFORM_AUTH;
cap_idx.algo.auth = opts->auth_algo;
capability = rte_cryptodev_sym_capability_get(cdev_id,
&cap_idx);
if (capability == NULL)
return -1;
ret = rte_cryptodev_sym_capability_check_auth(
capability,
opts->auth_key_sz,
opts->digest_sz,
opts->auth_iv_sz);
if (ret != 0)
return ret;
}
if (opts->op_type == CPERF_CIPHER_ONLY ||
opts->op_type == CPERF_CIPHER_THEN_AUTH ||
opts->op_type == CPERF_AUTH_THEN_CIPHER) {
cap_idx.type = RTE_CRYPTO_SYM_XFORM_CIPHER;
cap_idx.algo.cipher = opts->cipher_algo;
capability = rte_cryptodev_sym_capability_get(cdev_id,
&cap_idx);
if (capability == NULL)
return -1;
ret = rte_cryptodev_sym_capability_check_cipher(
capability,
opts->cipher_key_sz,
opts->cipher_iv_sz);
if (ret != 0)
return ret;
}
if (opts->op_type == CPERF_AEAD) {
cap_idx.type = RTE_CRYPTO_SYM_XFORM_AEAD;
cap_idx.algo.aead = opts->aead_algo;
capability = rte_cryptodev_sym_capability_get(cdev_id,
&cap_idx);
if (capability == NULL)
return -1;
ret = rte_cryptodev_sym_capability_check_aead(
capability,
opts->aead_key_sz,
opts->digest_sz,
opts->aead_aad_sz,
opts->aead_iv_sz);
if (ret != 0)
return ret;
}
}
return 0;
}
static int
cperf_check_test_vector(struct cperf_options *opts,
struct cperf_test_vector *test_vec)
{
if (opts->op_type == CPERF_CIPHER_ONLY) {
if (opts->cipher_algo == RTE_CRYPTO_CIPHER_NULL) {
if (test_vec->plaintext.data == NULL)
return -1;
} else if (opts->cipher_algo != RTE_CRYPTO_CIPHER_NULL) {
if (test_vec->plaintext.data == NULL)
return -1;
if (test_vec->plaintext.length < opts->max_buffer_size)
return -1;
if (test_vec->ciphertext.data == NULL)
return -1;
if (test_vec->ciphertext.length < opts->max_buffer_size)
return -1;
/* Cipher IV is only required for some algorithms */
if (opts->cipher_iv_sz &&
test_vec->cipher_iv.data == NULL)
return -1;
if (test_vec->cipher_iv.length != opts->cipher_iv_sz)
return -1;
if (test_vec->cipher_key.data == NULL)
return -1;
if (test_vec->cipher_key.length != opts->cipher_key_sz)
return -1;
}
} else if (opts->op_type == CPERF_AUTH_ONLY) {
if (opts->auth_algo != RTE_CRYPTO_AUTH_NULL) {
if (test_vec->plaintext.data == NULL)
return -1;
if (test_vec->plaintext.length < opts->max_buffer_size)
return -1;
/* Auth key is only required for some algorithms */
if (opts->auth_key_sz &&
test_vec->auth_key.data == NULL)
return -1;
if (test_vec->auth_key.length != opts->auth_key_sz)
return -1;
if (test_vec->auth_iv.length != opts->auth_iv_sz)
return -1;
/* Auth IV is only required for some algorithms */
if (opts->auth_iv_sz && test_vec->auth_iv.data == NULL)
return -1;
if (test_vec->digest.data == NULL)
return -1;
if (test_vec->digest.length < opts->digest_sz)
return -1;
}
} else if (opts->op_type == CPERF_CIPHER_THEN_AUTH ||
opts->op_type == CPERF_AUTH_THEN_CIPHER) {
if (opts->cipher_algo == RTE_CRYPTO_CIPHER_NULL) {
if (test_vec->plaintext.data == NULL)
return -1;
if (test_vec->plaintext.length < opts->max_buffer_size)
return -1;
} else if (opts->cipher_algo != RTE_CRYPTO_CIPHER_NULL) {
if (test_vec->plaintext.data == NULL)
return -1;
if (test_vec->plaintext.length < opts->max_buffer_size)
return -1;
if (test_vec->ciphertext.data == NULL)
return -1;
if (test_vec->ciphertext.length < opts->max_buffer_size)
return -1;
if (test_vec->cipher_iv.data == NULL)
return -1;
if (test_vec->cipher_iv.length != opts->cipher_iv_sz)
return -1;
if (test_vec->cipher_key.data == NULL)
return -1;
if (test_vec->cipher_key.length != opts->cipher_key_sz)
return -1;
}
if (opts->auth_algo != RTE_CRYPTO_AUTH_NULL) {
if (test_vec->auth_key.data == NULL)
return -1;
if (test_vec->auth_key.length != opts->auth_key_sz)
return -1;
if (test_vec->auth_iv.length != opts->auth_iv_sz)
return -1;
/* Auth IV is only required for some algorithms */
if (opts->auth_iv_sz && test_vec->auth_iv.data == NULL)
return -1;
if (test_vec->digest.data == NULL)
return -1;
if (test_vec->digest.length < opts->digest_sz)
return -1;
}
} else if (opts->op_type == CPERF_AEAD) {
if (test_vec->plaintext.data == NULL)
return -1;
if (test_vec->plaintext.length < opts->max_buffer_size)
return -1;
if (test_vec->ciphertext.data == NULL)
return -1;
if (test_vec->ciphertext.length < opts->max_buffer_size)
return -1;
if (test_vec->aead_key.data == NULL)
return -1;
if (test_vec->aead_key.length != opts->aead_key_sz)
return -1;
if (test_vec->aead_iv.data == NULL)
return -1;
if (test_vec->aead_iv.length != opts->aead_iv_sz)
return -1;
if (test_vec->aad.data == NULL)
return -1;
if (test_vec->aad.length != opts->aead_aad_sz)
return -1;
if (test_vec->digest.data == NULL)
return -1;
if (test_vec->digest.length < opts->digest_sz)
return -1;
}
return 0;
}
int
main(int argc, char **argv)
{
struct cperf_options opts = {0};
struct cperf_test_vector *t_vec = NULL;
struct cperf_op_fns op_fns;
void *ctx[RTE_MAX_LCORE] = { };
int nb_cryptodevs = 0;
uint16_t total_nb_qps = 0;
uint8_t cdev_id, i;
uint8_t enabled_cdevs[RTE_CRYPTO_MAX_DEVS] = { 0 };
uint8_t buffer_size_idx = 0;
int ret;
uint32_t lcore_id;
/* Initialise DPDK EAL */
ret = rte_eal_init(argc, argv);
if (ret < 0)
rte_exit(EXIT_FAILURE, "Invalid EAL arguments!\n");
argc -= ret;
argv += ret;
cperf_options_default(&opts);
ret = cperf_options_parse(&opts, argc, argv);
if (ret) {
RTE_LOG(ERR, USER1, "Parsing on or more user options failed\n");
goto err;
}
ret = cperf_options_check(&opts);
if (ret) {
RTE_LOG(ERR, USER1,
"Checking on or more user options failed\n");
goto err;
}
nb_cryptodevs = cperf_initialize_cryptodev(&opts, enabled_cdevs);
if (!opts.silent)
cperf_options_dump(&opts);
if (nb_cryptodevs < 1) {
RTE_LOG(ERR, USER1, "Failed to initialise requested crypto "
"device type\n");
nb_cryptodevs = 0;
goto err;
}
ret = cperf_verify_devices_capabilities(&opts, enabled_cdevs,
nb_cryptodevs);
if (ret) {
RTE_LOG(ERR, USER1, "Crypto device type does not support "
"capabilities requested\n");
goto err;
}
if (opts.test_file != NULL) {
t_vec = cperf_test_vector_get_from_file(&opts);
if (t_vec == NULL) {
RTE_LOG(ERR, USER1,
"Failed to create test vector for"
" specified file\n");
goto err;
}
if (cperf_check_test_vector(&opts, t_vec)) {
RTE_LOG(ERR, USER1, "Incomplete necessary test vectors"
"\n");
goto err;
}
} else {
t_vec = cperf_test_vector_get_dummy(&opts);
if (t_vec == NULL) {
RTE_LOG(ERR, USER1,
"Failed to create test vector for"
" specified algorithms\n");
goto err;
}
}
ret = cperf_get_op_functions(&opts, &op_fns);
if (ret) {
RTE_LOG(ERR, USER1, "Failed to find function ops set for "
"specified algorithms combination\n");
goto err;
}
if (!opts.silent)
show_test_vector(t_vec);
total_nb_qps = nb_cryptodevs * opts.nb_qps;
i = 0;
uint8_t qp_id = 0, cdev_index = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
cdev_id = enabled_cdevs[cdev_index];
uint8_t socket_id = rte_cryptodev_socket_id(cdev_id);
ctx[i] = cperf_testmap[opts.test].constructor(
session_pool_socket[socket_id].sess_mp,
session_pool_socket[socket_id].priv_mp,
cdev_id, qp_id,
&opts, t_vec, &op_fns);
if (ctx[i] == NULL) {
RTE_LOG(ERR, USER1, "Test run constructor failed\n");
goto err;
}
qp_id = (qp_id + 1) % opts.nb_qps;
if (qp_id == 0)
cdev_index++;
i++;
}
if (opts.imix_distribution_count != 0) {
uint8_t buffer_size_count = opts.buffer_size_count;
uint16_t distribution_total[buffer_size_count];
uint32_t op_idx;
uint32_t test_average_size = 0;
const uint32_t *buffer_size_list = opts.buffer_size_list;
const uint32_t *imix_distribution_list = opts.imix_distribution_list;
opts.imix_buffer_sizes = rte_malloc(NULL,
sizeof(uint32_t) * opts.pool_sz,
0);
/*
* Calculate accumulated distribution of
* probabilities per packet size
*/
distribution_total[0] = imix_distribution_list[0];
for (i = 1; i < buffer_size_count; i++)
distribution_total[i] = imix_distribution_list[i] +
distribution_total[i-1];
/* Calculate a random sequence of packet sizes, based on distribution */
for (op_idx = 0; op_idx < opts.pool_sz; op_idx++) {
uint16_t random_number = rte_rand() %
distribution_total[buffer_size_count - 1];
for (i = 0; i < buffer_size_count; i++)
if (random_number < distribution_total[i])
break;
opts.imix_buffer_sizes[op_idx] = buffer_size_list[i];
}
/* Calculate average buffer size for the IMIX distribution */
for (i = 0; i < buffer_size_count; i++)
test_average_size += buffer_size_list[i] *
imix_distribution_list[i];
opts.test_buffer_size = test_average_size /
distribution_total[buffer_size_count - 1];
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
rte_eal_remote_launch(cperf_testmap[opts.test].runner,
ctx[i], lcore_id);
i++;
}
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
rte_eal_wait_lcore(lcore_id);
i++;
}
} else {
/* Get next size from range or list */
if (opts.inc_buffer_size != 0)
opts.test_buffer_size = opts.min_buffer_size;
else
opts.test_buffer_size = opts.buffer_size_list[0];
while (opts.test_buffer_size <= opts.max_buffer_size) {
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
rte_eal_remote_launch(cperf_testmap[opts.test].runner,
ctx[i], lcore_id);
i++;
}
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
rte_eal_wait_lcore(lcore_id);
i++;
}
/* Get next size from range or list */
if (opts.inc_buffer_size != 0)
opts.test_buffer_size += opts.inc_buffer_size;
else {
if (++buffer_size_idx == opts.buffer_size_count)
break;
opts.test_buffer_size =
opts.buffer_size_list[buffer_size_idx];
}
}
}
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
cperf_testmap[opts.test].destructor(ctx[i]);
i++;
}
for (i = 0; i < nb_cryptodevs &&
i < RTE_CRYPTO_MAX_DEVS; i++)
rte_cryptodev_stop(enabled_cdevs[i]);
free_test_vector(t_vec, &opts);
printf("\n");
return EXIT_SUCCESS;
err:
i = 0;
RTE_LCORE_FOREACH_SLAVE(lcore_id) {
if (i == total_nb_qps)
break;
if (ctx[i] && cperf_testmap[opts.test].destructor)
cperf_testmap[opts.test].destructor(ctx[i]);
i++;
}
for (i = 0; i < nb_cryptodevs &&
i < RTE_CRYPTO_MAX_DEVS; i++)
rte_cryptodev_stop(enabled_cdevs[i]);
rte_free(opts.imix_buffer_sizes);
free_test_vector(t_vec, &opts);
printf("\n");
return EXIT_FAILURE;
}
static int
fbk_hash_perf_test(void)
{
struct rte_fbk_hash_params params = {
.name = "fbk_hash_test",
.entries = ENTRIES,
.entries_per_bucket = 4,
.socket_id = rte_socket_id(),
};
struct rte_fbk_hash_table *handle;
uint32_t keys[ENTRIES] = {0};
unsigned indexes[TEST_SIZE];
uint64_t lookup_time = 0;
unsigned added = 0;
unsigned value = 0;
unsigned i, j;
handle = rte_fbk_hash_create(¶ms);
RETURN_IF_ERROR_FBK(handle == NULL, "fbk hash creation failed");
/* Generate random keys and values. */
for (i = 0; i < ENTRIES; i++) {
uint32_t key = (uint32_t)rte_rand();
key = ((uint64_t)key << 32) | (uint64_t)rte_rand();
uint16_t val = (uint16_t)rte_rand();
if (rte_fbk_hash_add_key(handle, key, val) == 0) {
keys[added] = key;
added++;
}
if (added > (LOAD_FACTOR * ENTRIES)) {
break;
}
}
for (i = 0; i < TEST_ITERATIONS; i++) {
uint64_t begin;
uint64_t end;
/* Generate random indexes into keys[] array. */
for (j = 0; j < TEST_SIZE; j++) {
indexes[j] = rte_rand() % added;
}
begin = rte_rdtsc();
/* Do lookups */
for (j = 0; j < TEST_SIZE; j++) {
value += rte_fbk_hash_lookup(handle, keys[indexes[j]]);
}
end = rte_rdtsc();
lookup_time += (double)(end - begin);
}
printf("\n\n *** FBK Hash function performance test results ***\n");
/*
* The use of the 'value' variable ensures that the hash lookup is not
* being optimised out by the compiler.
*/
if (value != 0)
printf("Number of ticks per lookup = %g\n",
(double)lookup_time /
((double)TEST_ITERATIONS * (double)TEST_SIZE));
rte_fbk_hash_free(handle);
return 0;
}
/*
* Do all unit and performance tests.
*/
int test_hash_perf(void)
{
if (run_all_tbl_perf_tests() < 0)
return -1;
run_hash_func_tests();
if (fbk_hash_perf_test() < 0)
return -1;
return 0;
}
#else /* RTE_LIBRTE_HASH */
int
test_hash_perf(void)
{
printf("The Hash library is not included in this build\n");
return 0;
}
static int
test_reciprocal(void)
{
int result = 0;
uint32_t divisor_u32 = 0;
uint32_t dividend_u32;
uint32_t nresult_u32;
uint32_t rresult_u32;
uint64_t i, j;
uint64_t divisor_u64 = 0;
uint64_t dividend_u64;
uint64_t nresult_u64;
uint64_t rresult_u64;
struct rte_reciprocal reci_u32 = {0};
struct rte_reciprocal_u64 reci_u64 = {0};
rte_srand(rte_rdtsc());
printf("Validating unsigned 32bit division.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u32 = rte_rand();
reci_u32 = rte_reciprocal_value(divisor_u32);
}
dividend_u32 = rte_rand();
nresult_u32 = dividend_u32 / divisor_u32;
rresult_u32 = rte_reciprocal_divide(dividend_u32,
reci_u32);
if (nresult_u32 != rresult_u32) {
printf("Division failed, %"PRIu32"/%"PRIu32" = "
"expected %"PRIu32" result %"PRIu32"\n",
dividend_u32, divisor_u32,
nresult_u32, rresult_u32);
result = 1;
break;
}
}
printf("Validating unsigned 64bit division.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u64 = rte_rand();
reci_u64 = rte_reciprocal_value_u64(divisor_u64);
}
dividend_u64 = rte_rand();
nresult_u64 = dividend_u64 / divisor_u64;
rresult_u64 = rte_reciprocal_divide_u64(dividend_u64,
&reci_u64);
if (nresult_u64 != rresult_u64) {
printf("Division failed, %"PRIu64"/%"PRIu64" = "
"expected %"PRIu64" result %"PRIu64"\n",
dividend_u64, divisor_u64,
nresult_u64, rresult_u64);
result = 1;
break;
}
}
printf("Validating unsigned 64bit division with 32bit divisor.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u64 = rte_rand() >> 32;
reci_u64 = rte_reciprocal_value_u64(divisor_u64);
}
dividend_u64 = rte_rand();
nresult_u64 = dividend_u64 / divisor_u64;
rresult_u64 = rte_reciprocal_divide_u64(dividend_u64,
&reci_u64);
if (nresult_u64 != rresult_u64) {
printf("Division failed, %"PRIu64"/%"PRIu64" = "
"expected %"PRIu64" result %"PRIu64"\n",
dividend_u64, divisor_u64,
nresult_u64, rresult_u64);
result = 1;
break;
}
}
int32_t
perf_test(void)
{
struct rte_lpm *lpm = NULL;
uint64_t begin, total_time, lpm_used_entries = 0;
unsigned i, j;
uint8_t next_hop_add = 0xAA, next_hop_return = 0;
int status = 0;
uint64_t cache_line_counter = 0;
int64_t count = 0;
rte_srand(rte_rdtsc());
printf("No. routes = %u\n", (unsigned) NUM_ROUTE_ENTRIES);
print_route_distribution(large_route_table, (uint32_t) NUM_ROUTE_ENTRIES);
lpm = rte_lpm_create(__func__, SOCKET_ID_ANY, 1000000, 0);
TEST_LPM_ASSERT(lpm != NULL);
/* Measue add. */
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
if (rte_lpm_add(lpm, large_route_table[i].ip,
large_route_table[i].depth, next_hop_add) == 0)
status++;
}
/* End Timer. */
total_time = rte_rdtsc() - begin;
printf("Unique added entries = %d\n", status);
/* Obtain add statistics. */
for (i = 0; i < RTE_LPM_TBL24_NUM_ENTRIES; i++) {
if (lpm->tbl24[i].valid)
lpm_used_entries++;
if (i % 32 == 0){
if ((uint64_t)count < lpm_used_entries) {
cache_line_counter++;
count = lpm_used_entries;
}
}
}
printf("Used table 24 entries = %u (%g%%)\n",
(unsigned) lpm_used_entries,
(lpm_used_entries * 100.0) / RTE_LPM_TBL24_NUM_ENTRIES);
printf("64 byte Cache entries used = %u (%u bytes)\n",
(unsigned) cache_line_counter, (unsigned) cache_line_counter * 64);
printf("Average LPM Add: %g cycles\n", (double)total_time / NUM_ROUTE_ENTRIES);
/* Measure single Lookup */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i ++) {
static uint32_t ip_batch[BATCH_SIZE];
for (j = 0; j < BATCH_SIZE; j ++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j ++) {
if (rte_lpm_lookup(lpm, ip_batch[j], &next_hop_return) != 0)
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("Average LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Measure bulk Lookup */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i ++) {
static uint32_t ip_batch[BATCH_SIZE];
uint16_t next_hops[BULK_SIZE];
/* Create array of random IP addresses */
for (j = 0; j < BATCH_SIZE; j ++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j += BULK_SIZE) {
unsigned k;
rte_lpm_lookup_bulk(lpm, &ip_batch[j], next_hops, BULK_SIZE);
for (k = 0; k < BULK_SIZE; k++)
if (unlikely(!(next_hops[k] & RTE_LPM_LOOKUP_SUCCESS)))
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("BULK LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Measure LookupX4 */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i++) {
static uint32_t ip_batch[BATCH_SIZE];
uint16_t next_hops[4];
/* Create array of random IP addresses */
for (j = 0; j < BATCH_SIZE; j++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j += RTE_DIM(next_hops)) {
unsigned k;
__m128i ipx4;
ipx4 = _mm_loadu_si128((__m128i *)(ip_batch + j));
ipx4 = *(__m128i *)(ip_batch + j);
rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT16_MAX);
for (k = 0; k < RTE_DIM(next_hops); k++)
if (unlikely(next_hops[k] == UINT16_MAX))
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("LPM LookupX4: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Delete */
status = 0;
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
/* rte_lpm_delete(lpm, ip, depth) */
status += rte_lpm_delete(lpm, large_route_table[i].ip,
large_route_table[i].depth);
}
total_time += rte_rdtsc() - begin;
printf("Average LPM Delete: %g cycles\n",
(double)total_time / NUM_ROUTE_ENTRIES);
rte_lpm_delete_all(lpm);
rte_lpm_free(lpm);
return PASS;
}
/*
* Get a random offset into large array, with enough space needed to perform
* max copy size. Offset is aligned.
*/
static inline size_t
get_rand_offset(void)
{
return ((rte_rand() % (LARGE_BUFFER_SIZE - SMALL_BUFFER_SIZE)) &
~(ALIGNMENT_UNIT - 1));
}