4 * Copyright (c) 2015-2016 Amazon.com, Inc. or its affiliates.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of copyright holder nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <rte_ether.h>
35 #include <rte_ethdev_driver.h>
36 #include <rte_ethdev_pci.h>
38 #include <rte_atomic.h>
40 #include <rte_errno.h>
41 #include <rte_version.h>
42 #include <rte_eal_memconfig.h>
45 #include "ena_ethdev.h"
47 #include "ena_platform.h"
49 #include "ena_eth_com.h"
51 #include <ena_common_defs.h>
52 #include <ena_regs_defs.h>
53 #include <ena_admin_defs.h>
54 #include <ena_eth_io_defs.h>
56 #define DRV_MODULE_VER_MAJOR 1
57 #define DRV_MODULE_VER_MINOR 0
58 #define DRV_MODULE_VER_SUBMINOR 0
60 #define ENA_IO_TXQ_IDX(q) (2 * (q))
61 #define ENA_IO_RXQ_IDX(q) (2 * (q) + 1)
62 /*reverse version of ENA_IO_RXQ_IDX*/
63 #define ENA_IO_RXQ_IDX_REV(q) ((q - 1) / 2)
65 /* While processing submitted and completed descriptors (rx and tx path
66 * respectively) in a loop it is desired to:
67 * - perform batch submissions while populating sumbissmion queue
68 * - avoid blocking transmission of other packets during cleanup phase
69 * Hence the utilization ratio of 1/8 of a queue size.
71 #define ENA_RING_DESCS_RATIO(ring_size) (ring_size / 8)
73 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
74 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
76 #define GET_L4_HDR_LEN(mbuf) \
77 ((rte_pktmbuf_mtod_offset(mbuf, struct tcp_hdr *, \
78 mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
80 #define ENA_RX_RSS_TABLE_LOG_SIZE 7
81 #define ENA_RX_RSS_TABLE_SIZE (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
82 #define ENA_HASH_KEY_SIZE 40
83 #define ENA_ETH_SS_STATS 0xFF
84 #define ETH_GSTRING_LEN 32
86 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
88 enum ethtool_stringset {
94 char name[ETH_GSTRING_LEN];
98 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
100 .stat_offset = offsetof(struct ena_com_stats_admin, stat) \
103 #define ENA_STAT_ENTRY(stat, stat_type) { \
105 .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
108 #define ENA_STAT_RX_ENTRY(stat) \
109 ENA_STAT_ENTRY(stat, rx)
111 #define ENA_STAT_TX_ENTRY(stat) \
112 ENA_STAT_ENTRY(stat, tx)
114 #define ENA_STAT_GLOBAL_ENTRY(stat) \
115 ENA_STAT_ENTRY(stat, dev)
117 static const struct ena_stats ena_stats_global_strings[] = {
118 ENA_STAT_GLOBAL_ENTRY(tx_timeout),
119 ENA_STAT_GLOBAL_ENTRY(io_suspend),
120 ENA_STAT_GLOBAL_ENTRY(io_resume),
121 ENA_STAT_GLOBAL_ENTRY(wd_expired),
122 ENA_STAT_GLOBAL_ENTRY(interface_up),
123 ENA_STAT_GLOBAL_ENTRY(interface_down),
124 ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
127 static const struct ena_stats ena_stats_tx_strings[] = {
128 ENA_STAT_TX_ENTRY(cnt),
129 ENA_STAT_TX_ENTRY(bytes),
130 ENA_STAT_TX_ENTRY(queue_stop),
131 ENA_STAT_TX_ENTRY(queue_wakeup),
132 ENA_STAT_TX_ENTRY(dma_mapping_err),
133 ENA_STAT_TX_ENTRY(linearize),
134 ENA_STAT_TX_ENTRY(linearize_failed),
135 ENA_STAT_TX_ENTRY(tx_poll),
136 ENA_STAT_TX_ENTRY(doorbells),
137 ENA_STAT_TX_ENTRY(prepare_ctx_err),
138 ENA_STAT_TX_ENTRY(missing_tx_comp),
139 ENA_STAT_TX_ENTRY(bad_req_id),
142 static const struct ena_stats ena_stats_rx_strings[] = {
143 ENA_STAT_RX_ENTRY(cnt),
144 ENA_STAT_RX_ENTRY(bytes),
145 ENA_STAT_RX_ENTRY(refil_partial),
146 ENA_STAT_RX_ENTRY(bad_csum),
147 ENA_STAT_RX_ENTRY(page_alloc_fail),
148 ENA_STAT_RX_ENTRY(skb_alloc_fail),
149 ENA_STAT_RX_ENTRY(dma_mapping_err),
150 ENA_STAT_RX_ENTRY(bad_desc_num),
151 ENA_STAT_RX_ENTRY(small_copy_len_pkt),
154 static const struct ena_stats ena_stats_ena_com_strings[] = {
155 ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
156 ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
157 ENA_STAT_ENA_COM_ENTRY(completed_cmd),
158 ENA_STAT_ENA_COM_ENTRY(out_of_space),
159 ENA_STAT_ENA_COM_ENTRY(no_completion),
162 #define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
163 #define ENA_STATS_ARRAY_TX ARRAY_SIZE(ena_stats_tx_strings)
164 #define ENA_STATS_ARRAY_RX ARRAY_SIZE(ena_stats_rx_strings)
165 #define ENA_STATS_ARRAY_ENA_COM ARRAY_SIZE(ena_stats_ena_com_strings)
167 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
168 DEV_TX_OFFLOAD_UDP_CKSUM |\
169 DEV_TX_OFFLOAD_IPV4_CKSUM |\
170 DEV_TX_OFFLOAD_TCP_TSO)
171 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
175 /** Vendor ID used by Amazon devices */
176 #define PCI_VENDOR_ID_AMAZON 0x1D0F
177 /** Amazon devices */
178 #define PCI_DEVICE_ID_ENA_VF 0xEC20
179 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
181 #define ENA_TX_OFFLOAD_MASK (\
186 #define ENA_TX_OFFLOAD_NOTSUP_MASK \
187 (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
189 int ena_logtype_init;
190 int ena_logtype_driver;
192 static const struct rte_pci_id pci_id_ena_map[] = {
193 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
194 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
198 static int ena_device_init(struct ena_com_dev *ena_dev,
199 struct ena_com_dev_get_features_ctx *get_feat_ctx);
200 static int ena_dev_configure(struct rte_eth_dev *dev);
201 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
203 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
205 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
206 uint16_t nb_desc, unsigned int socket_id,
207 const struct rte_eth_txconf *tx_conf);
208 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
209 uint16_t nb_desc, unsigned int socket_id,
210 const struct rte_eth_rxconf *rx_conf,
211 struct rte_mempool *mp);
212 static uint16_t eth_ena_recv_pkts(void *rx_queue,
213 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
214 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
215 static void ena_init_rings(struct ena_adapter *adapter);
216 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
217 static int ena_start(struct rte_eth_dev *dev);
218 static void ena_close(struct rte_eth_dev *dev);
219 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
220 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
221 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
222 static void ena_rx_queue_release(void *queue);
223 static void ena_tx_queue_release(void *queue);
224 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
225 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
226 static int ena_link_update(struct rte_eth_dev *dev,
227 int wait_to_complete);
228 static int ena_queue_restart(struct ena_ring *ring);
229 static int ena_queue_restart_all(struct rte_eth_dev *dev,
230 enum ena_ring_type ring_type);
231 static void ena_stats_restart(struct rte_eth_dev *dev);
232 static void ena_infos_get(struct rte_eth_dev *dev,
233 struct rte_eth_dev_info *dev_info);
234 static int ena_rss_reta_update(struct rte_eth_dev *dev,
235 struct rte_eth_rss_reta_entry64 *reta_conf,
237 static int ena_rss_reta_query(struct rte_eth_dev *dev,
238 struct rte_eth_rss_reta_entry64 *reta_conf,
240 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
241 static bool ena_are_tx_queue_offloads_allowed(struct ena_adapter *adapter,
243 static bool ena_are_rx_queue_offloads_allowed(struct ena_adapter *adapter,
246 static const struct eth_dev_ops ena_dev_ops = {
247 .dev_configure = ena_dev_configure,
248 .dev_infos_get = ena_infos_get,
249 .rx_queue_setup = ena_rx_queue_setup,
250 .tx_queue_setup = ena_tx_queue_setup,
251 .dev_start = ena_start,
252 .link_update = ena_link_update,
253 .stats_get = ena_stats_get,
254 .mtu_set = ena_mtu_set,
255 .rx_queue_release = ena_rx_queue_release,
256 .tx_queue_release = ena_tx_queue_release,
257 .dev_close = ena_close,
258 .reta_update = ena_rss_reta_update,
259 .reta_query = ena_rss_reta_query,
262 #define NUMA_NO_NODE SOCKET_ID_ANY
264 static inline int ena_cpu_to_node(int cpu)
266 struct rte_config *config = rte_eal_get_configuration();
268 if (likely(cpu < RTE_MAX_MEMZONE))
269 return config->mem_config->memzone[cpu].socket_id;
274 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
275 struct ena_com_rx_ctx *ena_rx_ctx)
277 uint64_t ol_flags = 0;
278 uint32_t packet_type = 0;
280 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
281 packet_type |= RTE_PTYPE_L4_TCP;
282 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
283 packet_type |= RTE_PTYPE_L4_UDP;
285 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
286 packet_type |= RTE_PTYPE_L3_IPV4;
287 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
288 packet_type |= RTE_PTYPE_L3_IPV6;
290 if (unlikely(ena_rx_ctx->l4_csum_err))
291 ol_flags |= PKT_RX_L4_CKSUM_BAD;
292 if (unlikely(ena_rx_ctx->l3_csum_err))
293 ol_flags |= PKT_RX_IP_CKSUM_BAD;
295 mbuf->ol_flags = ol_flags;
296 mbuf->packet_type = packet_type;
299 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
300 struct ena_com_tx_ctx *ena_tx_ctx,
301 uint64_t queue_offloads)
303 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
305 if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
306 (queue_offloads & QUEUE_OFFLOADS)) {
307 /* check if TSO is required */
308 if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
309 (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
310 ena_tx_ctx->tso_enable = true;
312 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
315 /* check if L3 checksum is needed */
316 if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
317 (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
318 ena_tx_ctx->l3_csum_enable = true;
320 if (mbuf->ol_flags & PKT_TX_IPV6) {
321 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
323 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
325 /* set don't fragment (DF) flag */
326 if (mbuf->packet_type &
327 (RTE_PTYPE_L4_NONFRAG
328 | RTE_PTYPE_INNER_L4_NONFRAG))
329 ena_tx_ctx->df = true;
332 /* check if L4 checksum is needed */
333 if ((mbuf->ol_flags & PKT_TX_TCP_CKSUM) &&
334 (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
335 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
336 ena_tx_ctx->l4_csum_enable = true;
337 } else if ((mbuf->ol_flags & PKT_TX_UDP_CKSUM) &&
338 (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
339 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
340 ena_tx_ctx->l4_csum_enable = true;
342 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
343 ena_tx_ctx->l4_csum_enable = false;
346 ena_meta->mss = mbuf->tso_segsz;
347 ena_meta->l3_hdr_len = mbuf->l3_len;
348 ena_meta->l3_hdr_offset = mbuf->l2_len;
349 /* this param needed only for TSO */
350 ena_meta->l3_outer_hdr_len = 0;
351 ena_meta->l3_outer_hdr_offset = 0;
353 ena_tx_ctx->meta_valid = true;
355 ena_tx_ctx->meta_valid = false;
359 static void ena_config_host_info(struct ena_com_dev *ena_dev)
361 struct ena_admin_host_info *host_info;
364 /* Allocate only the host info */
365 rc = ena_com_allocate_host_info(ena_dev);
367 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
371 host_info = ena_dev->host_attr.host_info;
373 host_info->os_type = ENA_ADMIN_OS_DPDK;
374 host_info->kernel_ver = RTE_VERSION;
375 snprintf((char *)host_info->kernel_ver_str,
376 sizeof(host_info->kernel_ver_str),
377 "%s", rte_version());
378 host_info->os_dist = RTE_VERSION;
379 snprintf((char *)host_info->os_dist_str,
380 sizeof(host_info->os_dist_str),
381 "%s", rte_version());
382 host_info->driver_version =
383 (DRV_MODULE_VER_MAJOR) |
384 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
385 (DRV_MODULE_VER_SUBMINOR <<
386 ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
388 rc = ena_com_set_host_attributes(ena_dev);
390 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
398 ena_com_delete_host_info(ena_dev);
402 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
404 if (sset != ETH_SS_STATS)
407 /* Workaround for clang:
408 * touch internal structures to prevent
411 ENA_TOUCH(ena_stats_global_strings);
412 ENA_TOUCH(ena_stats_tx_strings);
413 ENA_TOUCH(ena_stats_rx_strings);
414 ENA_TOUCH(ena_stats_ena_com_strings);
416 return dev->data->nb_tx_queues *
417 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
418 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
421 static void ena_config_debug_area(struct ena_adapter *adapter)
426 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
428 RTE_LOG(ERR, PMD, "SS count is negative\n");
432 /* allocate 32 bytes for each string and 64bit for the value */
433 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
435 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
437 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
441 rc = ena_com_set_host_attributes(&adapter->ena_dev);
443 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
450 ena_com_delete_debug_area(&adapter->ena_dev);
453 static void ena_close(struct rte_eth_dev *dev)
455 struct ena_adapter *adapter =
456 (struct ena_adapter *)(dev->data->dev_private);
458 adapter->state = ENA_ADAPTER_STATE_STOPPED;
460 ena_rx_queue_release_all(dev);
461 ena_tx_queue_release_all(dev);
464 static int ena_rss_reta_update(struct rte_eth_dev *dev,
465 struct rte_eth_rss_reta_entry64 *reta_conf,
468 struct ena_adapter *adapter =
469 (struct ena_adapter *)(dev->data->dev_private);
470 struct ena_com_dev *ena_dev = &adapter->ena_dev;
476 if ((reta_size == 0) || (reta_conf == NULL))
479 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
480 RTE_LOG(WARNING, PMD,
481 "indirection table %d is bigger than supported (%d)\n",
482 reta_size, ENA_RX_RSS_TABLE_SIZE);
487 for (i = 0 ; i < reta_size ; i++) {
488 /* each reta_conf is for 64 entries.
489 * to support 128 we use 2 conf of 64
491 conf_idx = i / RTE_RETA_GROUP_SIZE;
492 idx = i % RTE_RETA_GROUP_SIZE;
493 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
495 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
496 ret = ena_com_indirect_table_fill_entry(ena_dev,
499 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
501 "Cannot fill indirect table\n");
508 ret = ena_com_indirect_table_set(ena_dev);
509 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
510 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
515 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
516 __func__, reta_size, adapter->rte_dev->data->port_id);
521 /* Query redirection table. */
522 static int ena_rss_reta_query(struct rte_eth_dev *dev,
523 struct rte_eth_rss_reta_entry64 *reta_conf,
526 struct ena_adapter *adapter =
527 (struct ena_adapter *)(dev->data->dev_private);
528 struct ena_com_dev *ena_dev = &adapter->ena_dev;
531 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
535 if (reta_size == 0 || reta_conf == NULL ||
536 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
539 ret = ena_com_indirect_table_get(ena_dev, indirect_table);
540 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
541 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
546 for (i = 0 ; i < reta_size ; i++) {
547 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
548 reta_idx = i % RTE_RETA_GROUP_SIZE;
549 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
550 reta_conf[reta_conf_idx].reta[reta_idx] =
551 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
557 static int ena_rss_init_default(struct ena_adapter *adapter)
559 struct ena_com_dev *ena_dev = &adapter->ena_dev;
560 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
564 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
566 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
570 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
571 val = i % nb_rx_queues;
572 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
573 ENA_IO_RXQ_IDX(val));
574 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
575 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
580 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
581 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
582 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
583 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
587 rc = ena_com_set_default_hash_ctrl(ena_dev);
588 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
589 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
593 rc = ena_com_indirect_table_set(ena_dev);
594 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
595 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
598 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
599 adapter->rte_dev->data->port_id);
604 ena_com_rss_destroy(ena_dev);
610 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
612 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
613 int nb_queues = dev->data->nb_rx_queues;
616 for (i = 0; i < nb_queues; i++)
617 ena_rx_queue_release(queues[i]);
620 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
622 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
623 int nb_queues = dev->data->nb_tx_queues;
626 for (i = 0; i < nb_queues; i++)
627 ena_tx_queue_release(queues[i]);
630 static void ena_rx_queue_release(void *queue)
632 struct ena_ring *ring = (struct ena_ring *)queue;
633 struct ena_adapter *adapter = ring->adapter;
636 ena_assert_msg(ring->configured,
637 "API violation - releasing not configured queue");
638 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
641 /* Destroy HW queue */
642 ena_qid = ENA_IO_RXQ_IDX(ring->id);
643 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
646 ena_rx_queue_release_bufs(ring);
648 /* Free ring resources */
649 if (ring->rx_buffer_info)
650 rte_free(ring->rx_buffer_info);
651 ring->rx_buffer_info = NULL;
653 ring->configured = 0;
655 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
656 ring->port_id, ring->id);
659 static void ena_tx_queue_release(void *queue)
661 struct ena_ring *ring = (struct ena_ring *)queue;
662 struct ena_adapter *adapter = ring->adapter;
665 ena_assert_msg(ring->configured,
666 "API violation. Releasing not configured queue");
667 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
670 /* Destroy HW queue */
671 ena_qid = ENA_IO_TXQ_IDX(ring->id);
672 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
675 ena_tx_queue_release_bufs(ring);
677 /* Free ring resources */
678 if (ring->tx_buffer_info)
679 rte_free(ring->tx_buffer_info);
681 if (ring->empty_tx_reqs)
682 rte_free(ring->empty_tx_reqs);
684 ring->empty_tx_reqs = NULL;
685 ring->tx_buffer_info = NULL;
687 ring->configured = 0;
689 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
690 ring->port_id, ring->id);
693 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
695 unsigned int ring_mask = ring->ring_size - 1;
697 while (ring->next_to_clean != ring->next_to_use) {
699 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
702 rte_mbuf_raw_free(m);
704 ring->next_to_clean++;
708 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
712 for (i = 0; i < ring->ring_size; ++i) {
713 struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
716 rte_pktmbuf_free(tx_buf->mbuf);
718 ring->next_to_clean++;
722 static int ena_link_update(struct rte_eth_dev *dev,
723 __rte_unused int wait_to_complete)
725 struct rte_eth_link *link = &dev->data->dev_link;
727 link->link_status = 1;
728 link->link_speed = ETH_SPEED_NUM_10G;
729 link->link_duplex = ETH_LINK_FULL_DUPLEX;
734 static int ena_queue_restart_all(struct rte_eth_dev *dev,
735 enum ena_ring_type ring_type)
737 struct ena_adapter *adapter =
738 (struct ena_adapter *)(dev->data->dev_private);
739 struct ena_ring *queues = NULL;
743 queues = (ring_type == ENA_RING_TYPE_RX) ?
744 adapter->rx_ring : adapter->tx_ring;
746 for (i = 0; i < adapter->num_queues; i++) {
747 if (queues[i].configured) {
748 if (ring_type == ENA_RING_TYPE_RX) {
750 dev->data->rx_queues[i] == &queues[i],
751 "Inconsistent state of rx queues\n");
754 dev->data->tx_queues[i] == &queues[i],
755 "Inconsistent state of tx queues\n");
758 rc = ena_queue_restart(&queues[i]);
762 "failed to restart queue %d type(%d)",
772 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
774 uint32_t max_frame_len = adapter->max_mtu;
776 if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
777 DEV_RX_OFFLOAD_JUMBO_FRAME)
779 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
781 return max_frame_len;
784 static int ena_check_valid_conf(struct ena_adapter *adapter)
786 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
788 if (max_frame_len > adapter->max_mtu) {
789 PMD_INIT_LOG(ERR, "Unsupported MTU of %d", max_frame_len);
797 ena_calc_queue_size(struct ena_com_dev *ena_dev,
798 struct ena_com_dev_get_features_ctx *get_feat_ctx)
800 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
802 queue_size = RTE_MIN(queue_size,
803 get_feat_ctx->max_queues.max_cq_depth);
804 queue_size = RTE_MIN(queue_size,
805 get_feat_ctx->max_queues.max_sq_depth);
807 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
808 queue_size = RTE_MIN(queue_size,
809 get_feat_ctx->max_queues.max_llq_depth);
811 /* Round down to power of 2 */
812 if (!rte_is_power_of_2(queue_size))
813 queue_size = rte_align32pow2(queue_size >> 1);
815 if (queue_size == 0) {
816 PMD_INIT_LOG(ERR, "Invalid queue size");
823 static void ena_stats_restart(struct rte_eth_dev *dev)
825 struct ena_adapter *adapter =
826 (struct ena_adapter *)(dev->data->dev_private);
828 rte_atomic64_init(&adapter->drv_stats->ierrors);
829 rte_atomic64_init(&adapter->drv_stats->oerrors);
830 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
833 static int ena_stats_get(struct rte_eth_dev *dev,
834 struct rte_eth_stats *stats)
836 struct ena_admin_basic_stats ena_stats;
837 struct ena_adapter *adapter =
838 (struct ena_adapter *)(dev->data->dev_private);
839 struct ena_com_dev *ena_dev = &adapter->ena_dev;
842 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
845 memset(&ena_stats, 0, sizeof(ena_stats));
846 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
848 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
852 /* Set of basic statistics from ENA */
853 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
854 ena_stats.rx_pkts_low);
855 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
856 ena_stats.tx_pkts_low);
857 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
858 ena_stats.rx_bytes_low);
859 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
860 ena_stats.tx_bytes_low);
861 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
862 ena_stats.rx_drops_low);
864 /* Driver related stats */
865 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
866 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
867 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
871 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
873 struct ena_adapter *adapter;
874 struct ena_com_dev *ena_dev;
877 ena_assert_msg(dev->data != NULL, "Uninitialized device");
878 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
879 adapter = (struct ena_adapter *)(dev->data->dev_private);
881 ena_dev = &adapter->ena_dev;
882 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
884 if (mtu > ena_get_mtu_conf(adapter)) {
886 "Given MTU (%d) exceeds maximum MTU supported (%d)\n",
887 mtu, ena_get_mtu_conf(adapter));
892 rc = ena_com_set_dev_mtu(ena_dev, mtu);
894 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
896 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
902 static int ena_start(struct rte_eth_dev *dev)
904 struct ena_adapter *adapter =
905 (struct ena_adapter *)(dev->data->dev_private);
908 if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
909 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
910 PMD_INIT_LOG(ERR, "API violation");
914 rc = ena_check_valid_conf(adapter);
918 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
922 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
926 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
927 ETH_MQ_RX_RSS_FLAG) {
928 rc = ena_rss_init_default(adapter);
933 ena_stats_restart(dev);
935 adapter->state = ENA_ADAPTER_STATE_RUNNING;
940 static int ena_queue_restart(struct ena_ring *ring)
944 ena_assert_msg(ring->configured == 1,
945 "Trying to restart unconfigured queue\n");
947 ring->next_to_clean = 0;
948 ring->next_to_use = 0;
950 if (ring->type == ENA_RING_TYPE_TX)
953 bufs_num = ring->ring_size - 1;
954 rc = ena_populate_rx_queue(ring, bufs_num);
955 if (rc != bufs_num) {
956 PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
963 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
966 __rte_unused unsigned int socket_id,
967 const struct rte_eth_txconf *tx_conf)
969 struct ena_com_create_io_ctx ctx =
970 /* policy set to _HOST just to satisfy icc compiler */
971 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
972 ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
973 struct ena_ring *txq = NULL;
974 struct ena_adapter *adapter =
975 (struct ena_adapter *)(dev->data->dev_private);
979 struct ena_com_dev *ena_dev = &adapter->ena_dev;
981 txq = &adapter->tx_ring[queue_idx];
983 if (txq->configured) {
985 "API violation. Queue %d is already configured\n",
990 if (!rte_is_power_of_2(nb_desc)) {
992 "Unsupported size of RX queue: %d is not a power of 2.",
997 if (nb_desc > adapter->tx_ring_size) {
999 "Unsupported size of TX queue (max size: %d)\n",
1000 adapter->tx_ring_size);
1004 if (tx_conf->txq_flags == ETH_TXQ_FLAGS_IGNORE &&
1005 !ena_are_tx_queue_offloads_allowed(adapter, tx_conf->offloads)) {
1006 RTE_LOG(ERR, PMD, "Unsupported queue offloads\n");
1010 ena_qid = ENA_IO_TXQ_IDX(queue_idx);
1012 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1014 ctx.msix_vector = -1; /* admin interrupts not used */
1015 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1016 ctx.queue_size = adapter->tx_ring_size;
1017 ctx.numa_node = ena_cpu_to_node(queue_idx);
1019 rc = ena_com_create_io_queue(ena_dev, &ctx);
1022 "failed to create io TX queue #%d (qid:%d) rc: %d\n",
1023 queue_idx, ena_qid, rc);
1025 txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1026 txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1028 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1029 &txq->ena_com_io_sq,
1030 &txq->ena_com_io_cq);
1033 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1035 ena_com_destroy_io_queue(ena_dev, ena_qid);
1039 txq->port_id = dev->data->port_id;
1040 txq->next_to_clean = 0;
1041 txq->next_to_use = 0;
1042 txq->ring_size = nb_desc;
1044 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1045 sizeof(struct ena_tx_buffer) *
1047 RTE_CACHE_LINE_SIZE);
1048 if (!txq->tx_buffer_info) {
1049 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1053 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1054 sizeof(u16) * txq->ring_size,
1055 RTE_CACHE_LINE_SIZE);
1056 if (!txq->empty_tx_reqs) {
1057 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1058 rte_free(txq->tx_buffer_info);
1061 for (i = 0; i < txq->ring_size; i++)
1062 txq->empty_tx_reqs[i] = i;
1064 txq->offloads = tx_conf->offloads;
1066 /* Store pointer to this queue in upper layer */
1067 txq->configured = 1;
1068 dev->data->tx_queues[queue_idx] = txq;
1073 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1076 __rte_unused unsigned int socket_id,
1077 const struct rte_eth_rxconf *rx_conf,
1078 struct rte_mempool *mp)
1080 struct ena_com_create_io_ctx ctx =
1081 /* policy set to _HOST just to satisfy icc compiler */
1082 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1083 ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1084 struct ena_adapter *adapter =
1085 (struct ena_adapter *)(dev->data->dev_private);
1086 struct ena_ring *rxq = NULL;
1087 uint16_t ena_qid = 0;
1089 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1091 rxq = &adapter->rx_ring[queue_idx];
1092 if (rxq->configured) {
1094 "API violation. Queue %d is already configured\n",
1099 if (!rte_is_power_of_2(nb_desc)) {
1101 "Unsupported size of TX queue: %d is not a power of 2.",
1106 if (nb_desc > adapter->rx_ring_size) {
1108 "Unsupported size of RX queue (max size: %d)\n",
1109 adapter->rx_ring_size);
1113 if (!ena_are_rx_queue_offloads_allowed(adapter, rx_conf->offloads)) {
1114 RTE_LOG(ERR, PMD, "Unsupported queue offloads\n");
1118 ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1121 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1122 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1123 ctx.msix_vector = -1; /* admin interrupts not used */
1124 ctx.queue_size = adapter->rx_ring_size;
1125 ctx.numa_node = ena_cpu_to_node(queue_idx);
1127 rc = ena_com_create_io_queue(ena_dev, &ctx);
1129 RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1132 rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1133 rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1135 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1136 &rxq->ena_com_io_sq,
1137 &rxq->ena_com_io_cq);
1140 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1142 ena_com_destroy_io_queue(ena_dev, ena_qid);
1145 rxq->port_id = dev->data->port_id;
1146 rxq->next_to_clean = 0;
1147 rxq->next_to_use = 0;
1148 rxq->ring_size = nb_desc;
1151 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1152 sizeof(struct rte_mbuf *) * nb_desc,
1153 RTE_CACHE_LINE_SIZE);
1154 if (!rxq->rx_buffer_info) {
1155 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1159 /* Store pointer to this queue in upper layer */
1160 rxq->configured = 1;
1161 dev->data->rx_queues[queue_idx] = rxq;
1166 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1170 uint16_t ring_size = rxq->ring_size;
1171 uint16_t ring_mask = ring_size - 1;
1172 uint16_t next_to_use = rxq->next_to_use;
1174 struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1176 if (unlikely(!count))
1179 in_use = rxq->next_to_use - rxq->next_to_clean;
1180 ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1182 count = RTE_MIN(count,
1183 (uint16_t)(ring_size - (next_to_use & ring_mask)));
1185 /* get resources for incoming packets */
1186 rc = rte_mempool_get_bulk(rxq->mb_pool,
1187 (void **)(&mbufs[next_to_use & ring_mask]),
1189 if (unlikely(rc < 0)) {
1190 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1191 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1195 for (i = 0; i < count; i++) {
1196 uint16_t next_to_use_masked = next_to_use & ring_mask;
1197 struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1198 struct ena_com_buf ebuf;
1200 rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1201 /* prepare physical address for DMA transaction */
1202 ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1203 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1204 /* pass resource to device */
1205 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1206 &ebuf, next_to_use_masked);
1208 rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1210 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1216 /* When we submitted free recources to device... */
1218 /* ...let HW know that it can fill buffers with data */
1220 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1222 rxq->next_to_use = next_to_use;
1228 static int ena_device_init(struct ena_com_dev *ena_dev,
1229 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1232 bool readless_supported;
1234 /* Initialize mmio registers */
1235 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1237 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1241 /* The PCIe configuration space revision id indicate if mmio reg
1244 readless_supported =
1245 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1246 & ENA_MMIO_DISABLE_REG_READ);
1247 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1250 rc = ena_com_dev_reset(ena_dev);
1252 RTE_LOG(ERR, PMD, "cannot reset device\n");
1253 goto err_mmio_read_less;
1256 /* check FW version */
1257 rc = ena_com_validate_version(ena_dev);
1259 RTE_LOG(ERR, PMD, "device version is too low\n");
1260 goto err_mmio_read_less;
1263 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1265 /* ENA device administration layer init */
1266 rc = ena_com_admin_init(ena_dev, NULL, true);
1269 "cannot initialize ena admin queue with device\n");
1270 goto err_mmio_read_less;
1273 /* To enable the msix interrupts the driver needs to know the number
1274 * of queues. So the driver uses polling mode to retrieve this
1277 ena_com_set_admin_polling_mode(ena_dev, true);
1279 ena_config_host_info(ena_dev);
1281 /* Get Device Attributes and features */
1282 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1285 "cannot get attribute for ena device rc= %d\n", rc);
1286 goto err_admin_init;
1292 ena_com_admin_destroy(ena_dev);
1295 ena_com_mmio_reg_read_request_destroy(ena_dev);
1300 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1302 struct rte_pci_device *pci_dev;
1303 struct ena_adapter *adapter =
1304 (struct ena_adapter *)(eth_dev->data->dev_private);
1305 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1306 struct ena_com_dev_get_features_ctx get_feat_ctx;
1309 static int adapters_found;
1311 memset(adapter, 0, sizeof(struct ena_adapter));
1312 ena_dev = &adapter->ena_dev;
1314 eth_dev->dev_ops = &ena_dev_ops;
1315 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1316 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1317 eth_dev->tx_pkt_prepare = ð_ena_prep_pkts;
1318 adapter->rte_eth_dev_data = eth_dev->data;
1319 adapter->rte_dev = eth_dev;
1321 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1324 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1325 adapter->pdev = pci_dev;
1327 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1328 pci_dev->addr.domain,
1330 pci_dev->addr.devid,
1331 pci_dev->addr.function);
1333 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1334 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1336 /* Present ENA_MEM_BAR indicates available LLQ mode.
1337 * Use corresponding policy
1339 if (adapter->dev_mem_base)
1340 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1341 else if (adapter->regs)
1342 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1344 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1347 ena_dev->reg_bar = adapter->regs;
1348 ena_dev->dmadev = adapter->pdev;
1350 adapter->id_number = adapters_found;
1352 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1353 adapter->id_number);
1355 /* device specific initialization routine */
1356 rc = ena_device_init(ena_dev, &get_feat_ctx);
1358 PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1362 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1363 if (get_feat_ctx.max_queues.max_llq_num == 0) {
1365 "Trying to use LLQ but llq_num is 0.\n"
1366 "Fall back into regular queues.");
1367 ena_dev->tx_mem_queue_type =
1368 ENA_ADMIN_PLACEMENT_POLICY_HOST;
1369 adapter->num_queues =
1370 get_feat_ctx.max_queues.max_sq_num;
1372 adapter->num_queues =
1373 get_feat_ctx.max_queues.max_llq_num;
1376 adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1379 queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1380 if ((queue_size <= 0) || (adapter->num_queues <= 0))
1383 adapter->tx_ring_size = queue_size;
1384 adapter->rx_ring_size = queue_size;
1386 /* prepare ring structures */
1387 ena_init_rings(adapter);
1389 ena_config_debug_area(adapter);
1391 /* Set max MTU for this device */
1392 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1394 /* set device support for TSO */
1395 adapter->tso4_supported = get_feat_ctx.offload.tx &
1396 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1398 /* Copy MAC address and point DPDK to it */
1399 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1400 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1401 (struct ether_addr *)adapter->mac_addr);
1403 adapter->drv_stats = rte_zmalloc("adapter stats",
1404 sizeof(*adapter->drv_stats),
1405 RTE_CACHE_LINE_SIZE);
1406 if (!adapter->drv_stats) {
1407 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1412 adapter->state = ENA_ADAPTER_STATE_INIT;
1417 static int ena_dev_configure(struct rte_eth_dev *dev)
1419 struct ena_adapter *adapter =
1420 (struct ena_adapter *)(dev->data->dev_private);
1421 uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
1422 uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1424 if ((tx_offloads & adapter->tx_supported_offloads) != tx_offloads) {
1425 RTE_LOG(ERR, PMD, "Some Tx offloads are not supported "
1426 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
1427 tx_offloads, adapter->tx_supported_offloads);
1431 if ((rx_offloads & adapter->rx_supported_offloads) != rx_offloads) {
1432 RTE_LOG(ERR, PMD, "Some Rx offloads are not supported "
1433 "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
1434 rx_offloads, adapter->rx_supported_offloads);
1438 if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1439 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1440 PMD_INIT_LOG(ERR, "Illegal adapter state: %d",
1445 switch (adapter->state) {
1446 case ENA_ADAPTER_STATE_INIT:
1447 case ENA_ADAPTER_STATE_STOPPED:
1448 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1450 case ENA_ADAPTER_STATE_CONFIG:
1451 RTE_LOG(WARNING, PMD,
1452 "Ivalid driver state while trying to configure device\n");
1458 adapter->tx_selected_offloads = tx_offloads;
1459 adapter->rx_selected_offloads = rx_offloads;
1463 static void ena_init_rings(struct ena_adapter *adapter)
1467 for (i = 0; i < adapter->num_queues; i++) {
1468 struct ena_ring *ring = &adapter->tx_ring[i];
1470 ring->configured = 0;
1471 ring->type = ENA_RING_TYPE_TX;
1472 ring->adapter = adapter;
1474 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1475 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1478 for (i = 0; i < adapter->num_queues; i++) {
1479 struct ena_ring *ring = &adapter->rx_ring[i];
1481 ring->configured = 0;
1482 ring->type = ENA_RING_TYPE_RX;
1483 ring->adapter = adapter;
1488 static bool ena_are_tx_queue_offloads_allowed(struct ena_adapter *adapter,
1491 uint64_t port_offloads = adapter->tx_selected_offloads;
1493 /* Check if port supports all requested offloads.
1494 * True if all offloads selected for queue are set for port.
1496 if ((offloads & port_offloads) != offloads)
1501 static bool ena_are_rx_queue_offloads_allowed(struct ena_adapter *adapter,
1504 uint64_t port_offloads = adapter->rx_selected_offloads;
1506 /* Check if port supports all requested offloads.
1507 * True if all offloads selected for queue are set for port.
1509 if ((offloads & port_offloads) != offloads)
1514 static void ena_infos_get(struct rte_eth_dev *dev,
1515 struct rte_eth_dev_info *dev_info)
1517 struct ena_adapter *adapter;
1518 struct ena_com_dev *ena_dev;
1519 struct ena_com_dev_get_features_ctx feat;
1520 uint64_t rx_feat = 0, tx_feat = 0;
1523 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1524 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1525 adapter = (struct ena_adapter *)(dev->data->dev_private);
1527 ena_dev = &adapter->ena_dev;
1528 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1530 dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1532 dev_info->speed_capa =
1534 ETH_LINK_SPEED_2_5G |
1536 ETH_LINK_SPEED_10G |
1537 ETH_LINK_SPEED_25G |
1538 ETH_LINK_SPEED_40G |
1539 ETH_LINK_SPEED_50G |
1540 ETH_LINK_SPEED_100G;
1542 /* Get supported features from HW */
1543 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1546 "Cannot get attribute for ena device rc= %d\n", rc);
1550 /* Set Tx & Rx features available for device */
1551 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1552 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1554 if (feat.offload.tx &
1555 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1556 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1557 DEV_TX_OFFLOAD_UDP_CKSUM |
1558 DEV_TX_OFFLOAD_TCP_CKSUM;
1560 if (feat.offload.rx_supported &
1561 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1562 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1563 DEV_RX_OFFLOAD_UDP_CKSUM |
1564 DEV_RX_OFFLOAD_TCP_CKSUM;
1566 /* Inform framework about available features */
1567 dev_info->rx_offload_capa = rx_feat;
1568 dev_info->rx_queue_offload_capa = rx_feat;
1569 dev_info->tx_offload_capa = tx_feat;
1570 dev_info->tx_queue_offload_capa = tx_feat;
1572 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1573 dev_info->max_rx_pktlen = adapter->max_mtu;
1574 dev_info->max_mac_addrs = 1;
1576 dev_info->max_rx_queues = adapter->num_queues;
1577 dev_info->max_tx_queues = adapter->num_queues;
1578 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1580 adapter->tx_supported_offloads = tx_feat;
1581 adapter->rx_supported_offloads = rx_feat;
1584 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1587 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1588 unsigned int ring_size = rx_ring->ring_size;
1589 unsigned int ring_mask = ring_size - 1;
1590 uint16_t next_to_clean = rx_ring->next_to_clean;
1591 uint16_t desc_in_use = 0;
1592 unsigned int recv_idx = 0;
1593 struct rte_mbuf *mbuf = NULL;
1594 struct rte_mbuf *mbuf_head = NULL;
1595 struct rte_mbuf *mbuf_prev = NULL;
1596 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1597 unsigned int completed;
1599 struct ena_com_rx_ctx ena_rx_ctx;
1602 /* Check adapter state */
1603 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1605 "Trying to receive pkts while device is NOT running\n");
1609 desc_in_use = rx_ring->next_to_use - next_to_clean;
1610 if (unlikely(nb_pkts > desc_in_use))
1611 nb_pkts = desc_in_use;
1613 for (completed = 0; completed < nb_pkts; completed++) {
1616 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1617 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1618 ena_rx_ctx.descs = 0;
1619 /* receive packet context */
1620 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1621 rx_ring->ena_com_io_sq,
1624 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1628 if (unlikely(ena_rx_ctx.descs == 0))
1631 while (segments < ena_rx_ctx.descs) {
1632 mbuf = rx_buff_info[next_to_clean & ring_mask];
1633 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1634 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1637 if (segments == 0) {
1638 mbuf->nb_segs = ena_rx_ctx.descs;
1639 mbuf->port = rx_ring->port_id;
1643 /* for multi-segment pkts create mbuf chain */
1644 mbuf_prev->next = mbuf;
1646 mbuf_head->pkt_len += mbuf->data_len;
1653 /* fill mbuf attributes if any */
1654 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1655 mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1657 /* pass to DPDK application head mbuf */
1658 rx_pkts[recv_idx] = mbuf_head;
1662 rx_ring->next_to_clean = next_to_clean;
1664 desc_in_use = desc_in_use - completed + 1;
1665 /* Burst refill to save doorbells, memory barriers, const interval */
1666 if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1667 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1673 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1679 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1680 struct ipv4_hdr *ip_hdr;
1682 uint16_t frag_field;
1684 for (i = 0; i != nb_pkts; i++) {
1686 ol_flags = m->ol_flags;
1688 if (!(ol_flags & PKT_TX_IPV4))
1691 /* If there was not L2 header length specified, assume it is
1692 * length of the ethernet header.
1694 if (unlikely(m->l2_len == 0))
1695 m->l2_len = sizeof(struct ether_hdr);
1697 ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1699 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1701 if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1702 m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1704 /* If IPv4 header has DF flag enabled and TSO support is
1705 * disabled, partial chcecksum should not be calculated.
1707 if (!tx_ring->adapter->tso4_supported)
1711 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1712 (ol_flags & PKT_TX_L4_MASK) ==
1713 PKT_TX_SCTP_CKSUM) {
1714 rte_errno = -ENOTSUP;
1718 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1719 ret = rte_validate_tx_offload(m);
1726 /* In case we are supposed to TSO and have DF not set (DF=0)
1727 * hardware must be provided with partial checksum, otherwise
1728 * it will take care of necessary calculations.
1731 ret = rte_net_intel_cksum_flags_prepare(m,
1732 ol_flags & ~PKT_TX_TCP_SEG);
1742 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1745 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1746 uint16_t next_to_use = tx_ring->next_to_use;
1747 uint16_t next_to_clean = tx_ring->next_to_clean;
1748 struct rte_mbuf *mbuf;
1749 unsigned int ring_size = tx_ring->ring_size;
1750 unsigned int ring_mask = ring_size - 1;
1751 struct ena_com_tx_ctx ena_tx_ctx;
1752 struct ena_tx_buffer *tx_info;
1753 struct ena_com_buf *ebuf;
1754 uint16_t rc, req_id, total_tx_descs = 0;
1755 uint16_t sent_idx = 0, empty_tx_reqs;
1758 /* Check adapter state */
1759 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1761 "Trying to xmit pkts while device is NOT running\n");
1765 empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
1766 if (nb_pkts > empty_tx_reqs)
1767 nb_pkts = empty_tx_reqs;
1769 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1770 mbuf = tx_pkts[sent_idx];
1772 req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1773 tx_info = &tx_ring->tx_buffer_info[req_id];
1774 tx_info->mbuf = mbuf;
1775 tx_info->num_of_bufs = 0;
1776 ebuf = tx_info->bufs;
1778 /* Prepare TX context */
1779 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1780 memset(&ena_tx_ctx.ena_meta, 0x0,
1781 sizeof(struct ena_com_tx_meta));
1782 ena_tx_ctx.ena_bufs = ebuf;
1783 ena_tx_ctx.req_id = req_id;
1784 if (tx_ring->tx_mem_queue_type ==
1785 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1786 /* prepare the push buffer with
1787 * virtual address of the data
1789 ena_tx_ctx.header_len =
1790 RTE_MIN(mbuf->data_len,
1791 tx_ring->tx_max_header_size);
1792 ena_tx_ctx.push_header =
1793 (void *)((char *)mbuf->buf_addr +
1795 } /* there's no else as we take advantage of memset zeroing */
1797 /* Set TX offloads flags, if applicable */
1798 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads);
1800 if (unlikely(mbuf->ol_flags &
1801 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1802 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1804 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1806 /* Process first segment taking into
1807 * consideration pushed header
1809 if (mbuf->data_len > ena_tx_ctx.header_len) {
1810 ebuf->paddr = mbuf->buf_iova +
1812 ena_tx_ctx.header_len;
1813 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1815 tx_info->num_of_bufs++;
1818 while ((mbuf = mbuf->next) != NULL) {
1819 ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
1820 ebuf->len = mbuf->data_len;
1822 tx_info->num_of_bufs++;
1825 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1827 /* Write data to device */
1828 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1829 &ena_tx_ctx, &nb_hw_desc);
1833 tx_info->tx_descs = nb_hw_desc;
1838 /* If there are ready packets to be xmitted... */
1840 /* ...let HW do its best :-) */
1842 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1844 tx_ring->next_to_use = next_to_use;
1847 /* Clear complete packets */
1848 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1849 /* Get Tx info & store how many descs were processed */
1850 tx_info = &tx_ring->tx_buffer_info[req_id];
1851 total_tx_descs += tx_info->tx_descs;
1853 /* Free whole mbuf chain */
1854 mbuf = tx_info->mbuf;
1855 rte_pktmbuf_free(mbuf);
1856 tx_info->mbuf = NULL;
1858 /* Put back descriptor to the ring for reuse */
1859 tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1862 /* If too many descs to clean, leave it for another run */
1863 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1867 if (total_tx_descs > 0) {
1868 /* acknowledge completion of sent packets */
1869 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1870 tx_ring->next_to_clean = next_to_clean;
1876 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1877 struct rte_pci_device *pci_dev)
1879 return rte_eth_dev_pci_generic_probe(pci_dev,
1880 sizeof(struct ena_adapter), eth_ena_dev_init);
1883 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
1885 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1888 static struct rte_pci_driver rte_ena_pmd = {
1889 .id_table = pci_id_ena_map,
1890 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1891 .probe = eth_ena_pci_probe,
1892 .remove = eth_ena_pci_remove,
1895 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
1896 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
1897 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
1899 RTE_INIT(ena_init_log);
1903 ena_logtype_init = rte_log_register("pmd.ena.init");
1904 if (ena_logtype_init >= 0)
1905 rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
1906 ena_logtype_driver = rte_log_register("pmd.ena.driver");
1907 if (ena_logtype_driver >= 0)
1908 rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);