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.h>
37 #include <rte_atomic.h>
39 #include <rte_errno.h>
40 #include <rte_version.h>
42 #include "ena_ethdev.h"
44 #include "ena_platform.h"
46 #include "ena_eth_com.h"
48 #include <ena_common_defs.h>
49 #include <ena_regs_defs.h>
50 #include <ena_admin_defs.h>
51 #include <ena_eth_io_defs.h>
53 #define DRV_MODULE_VER_MAJOR 1
54 #define DRV_MODULE_VER_MINOR 0
55 #define DRV_MODULE_VER_SUBMINOR 0
57 #define ENA_IO_TXQ_IDX(q) (2 * (q))
58 #define ENA_IO_RXQ_IDX(q) (2 * (q) + 1)
59 /*reverse version of ENA_IO_RXQ_IDX*/
60 #define ENA_IO_RXQ_IDX_REV(q) ((q - 1) / 2)
62 /* While processing submitted and completed descriptors (rx and tx path
63 * respectively) in a loop it is desired to:
64 * - perform batch submissions while populating sumbissmion queue
65 * - avoid blocking transmission of other packets during cleanup phase
66 * Hence the utilization ratio of 1/8 of a queue size.
68 #define ENA_RING_DESCS_RATIO(ring_size) (ring_size / 8)
70 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
71 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
73 #define GET_L4_HDR_LEN(mbuf) \
74 ((rte_pktmbuf_mtod_offset(mbuf, struct tcp_hdr *, \
75 mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
77 #define ENA_RX_RSS_TABLE_LOG_SIZE 7
78 #define ENA_RX_RSS_TABLE_SIZE (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
79 #define ENA_HASH_KEY_SIZE 40
80 #define ENA_ETH_SS_STATS 0xFF
81 #define ETH_GSTRING_LEN 32
83 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
85 enum ethtool_stringset {
91 char name[ETH_GSTRING_LEN];
95 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
97 .stat_offset = offsetof(struct ena_com_stats_admin, stat) \
100 #define ENA_STAT_ENTRY(stat, stat_type) { \
102 .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
105 #define ENA_STAT_RX_ENTRY(stat) \
106 ENA_STAT_ENTRY(stat, rx)
108 #define ENA_STAT_TX_ENTRY(stat) \
109 ENA_STAT_ENTRY(stat, tx)
111 #define ENA_STAT_GLOBAL_ENTRY(stat) \
112 ENA_STAT_ENTRY(stat, dev)
114 static const struct ena_stats ena_stats_global_strings[] = {
115 ENA_STAT_GLOBAL_ENTRY(tx_timeout),
116 ENA_STAT_GLOBAL_ENTRY(io_suspend),
117 ENA_STAT_GLOBAL_ENTRY(io_resume),
118 ENA_STAT_GLOBAL_ENTRY(wd_expired),
119 ENA_STAT_GLOBAL_ENTRY(interface_up),
120 ENA_STAT_GLOBAL_ENTRY(interface_down),
121 ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
124 static const struct ena_stats ena_stats_tx_strings[] = {
125 ENA_STAT_TX_ENTRY(cnt),
126 ENA_STAT_TX_ENTRY(bytes),
127 ENA_STAT_TX_ENTRY(queue_stop),
128 ENA_STAT_TX_ENTRY(queue_wakeup),
129 ENA_STAT_TX_ENTRY(dma_mapping_err),
130 ENA_STAT_TX_ENTRY(linearize),
131 ENA_STAT_TX_ENTRY(linearize_failed),
132 ENA_STAT_TX_ENTRY(tx_poll),
133 ENA_STAT_TX_ENTRY(doorbells),
134 ENA_STAT_TX_ENTRY(prepare_ctx_err),
135 ENA_STAT_TX_ENTRY(missing_tx_comp),
136 ENA_STAT_TX_ENTRY(bad_req_id),
139 static const struct ena_stats ena_stats_rx_strings[] = {
140 ENA_STAT_RX_ENTRY(cnt),
141 ENA_STAT_RX_ENTRY(bytes),
142 ENA_STAT_RX_ENTRY(refil_partial),
143 ENA_STAT_RX_ENTRY(bad_csum),
144 ENA_STAT_RX_ENTRY(page_alloc_fail),
145 ENA_STAT_RX_ENTRY(skb_alloc_fail),
146 ENA_STAT_RX_ENTRY(dma_mapping_err),
147 ENA_STAT_RX_ENTRY(bad_desc_num),
148 ENA_STAT_RX_ENTRY(small_copy_len_pkt),
151 static const struct ena_stats ena_stats_ena_com_strings[] = {
152 ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
153 ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
154 ENA_STAT_ENA_COM_ENTRY(completed_cmd),
155 ENA_STAT_ENA_COM_ENTRY(out_of_space),
156 ENA_STAT_ENA_COM_ENTRY(no_completion),
159 #define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
160 #define ENA_STATS_ARRAY_TX ARRAY_SIZE(ena_stats_tx_strings)
161 #define ENA_STATS_ARRAY_RX ARRAY_SIZE(ena_stats_rx_strings)
162 #define ENA_STATS_ARRAY_ENA_COM ARRAY_SIZE(ena_stats_ena_com_strings)
164 /** Vendor ID used by Amazon devices */
165 #define PCI_VENDOR_ID_AMAZON 0x1D0F
166 /** Amazon devices */
167 #define PCI_DEVICE_ID_ENA_VF 0xEC20
168 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
170 static struct rte_pci_id pci_id_ena_map[] = {
171 #define RTE_PCI_DEV_ID_DECL_ENA(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
172 RTE_PCI_DEV_ID_DECL_ENA(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF)
173 RTE_PCI_DEV_ID_DECL_ENA(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF)
177 static int ena_device_init(struct ena_com_dev *ena_dev,
178 struct ena_com_dev_get_features_ctx *get_feat_ctx);
179 static int ena_dev_configure(struct rte_eth_dev *dev);
180 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
182 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
183 uint16_t nb_desc, unsigned int socket_id,
184 const struct rte_eth_txconf *tx_conf);
185 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
186 uint16_t nb_desc, unsigned int socket_id,
187 const struct rte_eth_rxconf *rx_conf,
188 struct rte_mempool *mp);
189 static uint16_t eth_ena_recv_pkts(void *rx_queue,
190 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
191 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
192 static void ena_init_rings(struct ena_adapter *adapter);
193 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
194 static int ena_start(struct rte_eth_dev *dev);
195 static void ena_close(struct rte_eth_dev *dev);
196 static void ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
197 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
198 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
199 static void ena_rx_queue_release(void *queue);
200 static void ena_tx_queue_release(void *queue);
201 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
202 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
203 static int ena_link_update(struct rte_eth_dev *dev,
204 __rte_unused int wait_to_complete);
205 static int ena_queue_restart(struct ena_ring *ring);
206 static int ena_queue_restart_all(struct rte_eth_dev *dev,
207 enum ena_ring_type ring_type);
208 static void ena_stats_restart(struct rte_eth_dev *dev);
209 static void ena_infos_get(__rte_unused struct rte_eth_dev *dev,
210 struct rte_eth_dev_info *dev_info);
211 static int ena_rss_reta_update(struct rte_eth_dev *dev,
212 struct rte_eth_rss_reta_entry64 *reta_conf,
214 static int ena_rss_reta_query(struct rte_eth_dev *dev,
215 struct rte_eth_rss_reta_entry64 *reta_conf,
217 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
219 static struct eth_dev_ops ena_dev_ops = {
220 .dev_configure = ena_dev_configure,
221 .dev_infos_get = ena_infos_get,
222 .rx_queue_setup = ena_rx_queue_setup,
223 .tx_queue_setup = ena_tx_queue_setup,
224 .dev_start = ena_start,
225 .link_update = ena_link_update,
226 .stats_get = ena_stats_get,
227 .mtu_set = ena_mtu_set,
228 .rx_queue_release = ena_rx_queue_release,
229 .tx_queue_release = ena_tx_queue_release,
230 .dev_close = ena_close,
231 .reta_update = ena_rss_reta_update,
232 .reta_query = ena_rss_reta_query,
235 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
236 struct ena_com_rx_ctx *ena_rx_ctx)
238 uint64_t ol_flags = 0;
240 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
241 ol_flags |= PKT_TX_TCP_CKSUM;
242 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
243 ol_flags |= PKT_TX_UDP_CKSUM;
245 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
246 ol_flags |= PKT_TX_IPV4;
247 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
248 ol_flags |= PKT_TX_IPV6;
250 if (unlikely(ena_rx_ctx->l4_csum_err))
251 ol_flags |= PKT_RX_L4_CKSUM_BAD;
252 if (unlikely(ena_rx_ctx->l3_csum_err))
253 ol_flags |= PKT_RX_IP_CKSUM_BAD;
255 mbuf->ol_flags = ol_flags;
258 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
259 struct ena_com_tx_ctx *ena_tx_ctx)
261 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
264 (PKT_TX_L4_MASK | PKT_TX_IP_CKSUM | PKT_TX_TCP_SEG)) {
265 /* check if TSO is required */
266 if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
267 ena_tx_ctx->tso_enable = true;
269 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
272 /* check if L3 checksum is needed */
273 if (mbuf->ol_flags & PKT_TX_IP_CKSUM)
274 ena_tx_ctx->l3_csum_enable = true;
276 if (mbuf->ol_flags & PKT_TX_IPV6) {
277 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
279 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
281 /* set don't fragment (DF) flag */
282 if (mbuf->packet_type &
283 (RTE_PTYPE_L4_NONFRAG
284 | RTE_PTYPE_INNER_L4_NONFRAG))
285 ena_tx_ctx->df = true;
288 /* check if L4 checksum is needed */
289 switch (mbuf->ol_flags & PKT_TX_L4_MASK) {
290 case PKT_TX_TCP_CKSUM:
291 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
292 ena_tx_ctx->l4_csum_enable = true;
294 case PKT_TX_UDP_CKSUM:
295 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
296 ena_tx_ctx->l4_csum_enable = true;
299 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
300 ena_tx_ctx->l4_csum_enable = false;
304 ena_meta->mss = mbuf->tso_segsz;
305 ena_meta->l3_hdr_len = mbuf->l3_len;
306 ena_meta->l3_hdr_offset = mbuf->l2_len;
307 /* this param needed only for TSO */
308 ena_meta->l3_outer_hdr_len = 0;
309 ena_meta->l3_outer_hdr_offset = 0;
311 ena_tx_ctx->meta_valid = true;
313 ena_tx_ctx->meta_valid = false;
317 static void ena_config_host_info(struct ena_com_dev *ena_dev)
319 struct ena_admin_host_info *host_info;
322 /* Allocate only the host info */
323 rc = ena_com_allocate_host_info(ena_dev);
325 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
329 host_info = ena_dev->host_attr.host_info;
331 host_info->os_type = ENA_ADMIN_OS_DPDK;
332 host_info->kernel_ver = RTE_VERSION;
333 strncpy((char *)host_info->kernel_ver_str, rte_version(),
334 strlen(rte_version()));
335 host_info->os_dist = RTE_VERSION;
336 strncpy((char *)host_info->os_dist_str, rte_version(),
337 strlen(rte_version()));
338 host_info->driver_version =
339 (DRV_MODULE_VER_MAJOR) |
340 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
341 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
343 rc = ena_com_set_host_attributes(ena_dev);
346 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
348 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
356 ena_com_delete_host_info(ena_dev);
360 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
362 if (sset != ETH_SS_STATS)
365 /* Workaround for clang:
366 * touch internal structures to prevent
369 ENA_TOUCH(ena_stats_global_strings);
370 ENA_TOUCH(ena_stats_tx_strings);
371 ENA_TOUCH(ena_stats_rx_strings);
372 ENA_TOUCH(ena_stats_ena_com_strings);
374 return dev->data->nb_tx_queues *
375 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
376 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
379 static void ena_config_debug_area(struct ena_adapter *adapter)
384 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
386 RTE_LOG(ERR, PMD, "SS count is negative\n");
390 /* allocate 32 bytes for each string and 64bit for the value */
391 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
393 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
395 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
399 rc = ena_com_set_host_attributes(&adapter->ena_dev);
402 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
404 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
410 ena_com_delete_debug_area(&adapter->ena_dev);
413 static void ena_close(struct rte_eth_dev *dev)
415 struct ena_adapter *adapter =
416 (struct ena_adapter *)(dev->data->dev_private);
418 adapter->state = ENA_ADAPTER_STATE_STOPPED;
420 ena_rx_queue_release_all(dev);
421 ena_tx_queue_release_all(dev);
424 static int ena_rss_reta_update(struct rte_eth_dev *dev,
425 struct rte_eth_rss_reta_entry64 *reta_conf,
428 struct ena_adapter *adapter =
429 (struct ena_adapter *)(dev->data->dev_private);
430 struct ena_com_dev *ena_dev = &adapter->ena_dev;
436 if ((reta_size == 0) || (reta_conf == NULL))
439 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
440 RTE_LOG(WARNING, PMD,
441 "indirection table %d is bigger than supported (%d)\n",
442 reta_size, ENA_RX_RSS_TABLE_SIZE);
447 for (i = 0 ; i < reta_size ; i++) {
448 /* each reta_conf is for 64 entries.
449 * to support 128 we use 2 conf of 64
451 conf_idx = i / RTE_RETA_GROUP_SIZE;
452 idx = i % RTE_RETA_GROUP_SIZE;
453 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
455 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
456 ret = ena_com_indirect_table_fill_entry(ena_dev,
459 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
461 "Cannot fill indirect table\n");
468 ret = ena_com_indirect_table_set(ena_dev);
469 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
470 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
475 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
476 __func__, reta_size, adapter->rte_dev->data->port_id);
481 /* Query redirection table. */
482 static int ena_rss_reta_query(struct rte_eth_dev *dev,
483 struct rte_eth_rss_reta_entry64 *reta_conf,
486 struct ena_adapter *adapter =
487 (struct ena_adapter *)(dev->data->dev_private);
488 struct ena_com_dev *ena_dev = &adapter->ena_dev;
491 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
495 if (reta_size == 0 || reta_conf == NULL ||
496 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
499 ret = ena_com_indirect_table_get(ena_dev, indirect_table);
500 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
501 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
506 for (i = 0 ; i < reta_size ; i++) {
507 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
508 reta_idx = i % RTE_RETA_GROUP_SIZE;
509 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
510 reta_conf[reta_conf_idx].reta[reta_idx] =
511 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
517 static int ena_rss_init_default(struct ena_adapter *adapter)
519 struct ena_com_dev *ena_dev = &adapter->ena_dev;
520 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
524 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
526 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
530 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
531 val = i % nb_rx_queues;
532 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
533 ENA_IO_RXQ_IDX(val));
534 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
535 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
540 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
541 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
542 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
543 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
547 rc = ena_com_set_default_hash_ctrl(ena_dev);
548 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
549 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
553 rc = ena_com_indirect_table_set(ena_dev);
554 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
555 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
558 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
559 adapter->rte_dev->data->port_id);
564 ena_com_rss_destroy(ena_dev);
570 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
572 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
573 int nb_queues = dev->data->nb_rx_queues;
576 for (i = 0; i < nb_queues; i++)
577 ena_rx_queue_release(queues[i]);
580 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
582 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
583 int nb_queues = dev->data->nb_tx_queues;
586 for (i = 0; i < nb_queues; i++)
587 ena_tx_queue_release(queues[i]);
590 static void ena_rx_queue_release(void *queue)
592 struct ena_ring *ring = (struct ena_ring *)queue;
593 struct ena_adapter *adapter = ring->adapter;
596 ena_assert_msg(ring->configured,
597 "API violation - releasing not configured queue");
598 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
601 /* Destroy HW queue */
602 ena_qid = ENA_IO_RXQ_IDX(ring->id);
603 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
606 ena_rx_queue_release_bufs(ring);
608 /* Free ring resources */
609 if (ring->rx_buffer_info)
610 rte_free(ring->rx_buffer_info);
611 ring->rx_buffer_info = NULL;
613 ring->configured = 0;
615 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
616 ring->port_id, ring->id);
619 static void ena_tx_queue_release(void *queue)
621 struct ena_ring *ring = (struct ena_ring *)queue;
622 struct ena_adapter *adapter = ring->adapter;
625 ena_assert_msg(ring->configured,
626 "API violation. Releasing not configured queue");
627 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
630 /* Destroy HW queue */
631 ena_qid = ENA_IO_TXQ_IDX(ring->id);
632 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
635 ena_tx_queue_release_bufs(ring);
637 /* Free ring resources */
638 if (ring->tx_buffer_info)
639 rte_free(ring->tx_buffer_info);
641 if (ring->empty_tx_reqs)
642 rte_free(ring->empty_tx_reqs);
644 ring->empty_tx_reqs = NULL;
645 ring->tx_buffer_info = NULL;
647 ring->configured = 0;
649 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
650 ring->port_id, ring->id);
653 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
655 unsigned int ring_mask = ring->ring_size - 1;
657 while (ring->next_to_clean != ring->next_to_use) {
659 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
662 __rte_mbuf_raw_free(m);
664 ring->next_to_clean =
665 ENA_CIRC_INC(ring->next_to_clean, 1, ring->ring_size);
669 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
671 unsigned int ring_mask = ring->ring_size - 1;
673 while (ring->next_to_clean != ring->next_to_use) {
674 struct ena_tx_buffer *tx_buf =
675 &ring->tx_buffer_info[ring->next_to_clean & ring_mask];
678 rte_pktmbuf_free(tx_buf->mbuf);
680 ring->next_to_clean =
681 ENA_CIRC_INC(ring->next_to_clean, 1, ring->ring_size);
685 static int ena_link_update(struct rte_eth_dev *dev,
686 __rte_unused int wait_to_complete)
688 struct rte_eth_link *link = &dev->data->dev_link;
690 link->link_status = 1;
691 link->link_speed = ETH_SPEED_NUM_10G;
692 link->link_duplex = ETH_LINK_FULL_DUPLEX;
697 static int ena_queue_restart_all(struct rte_eth_dev *dev,
698 enum ena_ring_type ring_type)
700 struct ena_adapter *adapter =
701 (struct ena_adapter *)(dev->data->dev_private);
702 struct ena_ring *queues = NULL;
706 queues = (ring_type == ENA_RING_TYPE_RX) ?
707 adapter->rx_ring : adapter->tx_ring;
709 for (i = 0; i < adapter->num_queues; i++) {
710 if (queues[i].configured) {
711 if (ring_type == ENA_RING_TYPE_RX) {
713 dev->data->rx_queues[i] == &queues[i],
714 "Inconsistent state of rx queues\n");
717 dev->data->tx_queues[i] == &queues[i],
718 "Inconsistent state of tx queues\n");
721 rc = ena_queue_restart(&queues[i]);
725 "failed to restart queue %d type(%d)\n",
735 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
737 uint32_t max_frame_len = adapter->max_mtu;
739 if (adapter->rte_eth_dev_data->dev_conf.rxmode.jumbo_frame == 1)
741 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
743 return max_frame_len;
746 static int ena_check_valid_conf(struct ena_adapter *adapter)
748 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
750 if (max_frame_len > adapter->max_mtu) {
751 PMD_INIT_LOG(ERR, "Unsupported MTU of %d\n", max_frame_len);
759 ena_calc_queue_size(struct ena_com_dev *ena_dev,
760 struct ena_com_dev_get_features_ctx *get_feat_ctx)
762 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
764 queue_size = RTE_MIN(queue_size,
765 get_feat_ctx->max_queues.max_cq_depth);
766 queue_size = RTE_MIN(queue_size,
767 get_feat_ctx->max_queues.max_sq_depth);
769 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
770 queue_size = RTE_MIN(queue_size,
771 get_feat_ctx->max_queues.max_llq_depth);
773 /* Round down to power of 2 */
774 if (!rte_is_power_of_2(queue_size))
775 queue_size = rte_align32pow2(queue_size >> 1);
777 if (queue_size == 0) {
778 PMD_INIT_LOG(ERR, "Invalid queue size\n");
785 static void ena_stats_restart(struct rte_eth_dev *dev)
787 struct ena_adapter *adapter =
788 (struct ena_adapter *)(dev->data->dev_private);
790 rte_atomic64_init(&adapter->drv_stats->ierrors);
791 rte_atomic64_init(&adapter->drv_stats->oerrors);
792 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
795 static void ena_stats_get(struct rte_eth_dev *dev,
796 struct rte_eth_stats *stats)
798 struct ena_admin_basic_stats ena_stats;
799 struct ena_adapter *adapter =
800 (struct ena_adapter *)(dev->data->dev_private);
801 struct ena_com_dev *ena_dev = &adapter->ena_dev;
804 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
807 memset(&ena_stats, 0, sizeof(ena_stats));
808 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
810 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
814 /* Set of basic statistics from ENA */
815 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
816 ena_stats.rx_pkts_low);
817 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
818 ena_stats.tx_pkts_low);
819 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
820 ena_stats.rx_bytes_low);
821 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
822 ena_stats.tx_bytes_low);
823 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
824 ena_stats.rx_drops_low);
826 /* Driver related stats */
827 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
828 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
829 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
832 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
834 struct ena_adapter *adapter;
835 struct ena_com_dev *ena_dev;
838 ena_assert_msg(dev->data != NULL, "Uninitialized device");
839 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
840 adapter = (struct ena_adapter *)(dev->data->dev_private);
842 ena_dev = &adapter->ena_dev;
843 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
845 if (mtu > ena_get_mtu_conf(adapter)) {
847 "Given MTU (%d) exceeds maximum MTU supported (%d)\n",
848 mtu, ena_get_mtu_conf(adapter));
853 rc = ena_com_set_dev_mtu(ena_dev, mtu);
855 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
857 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
863 static int ena_start(struct rte_eth_dev *dev)
865 struct ena_adapter *adapter =
866 (struct ena_adapter *)(dev->data->dev_private);
869 if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
870 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
871 PMD_INIT_LOG(ERR, "API violation");
875 rc = ena_check_valid_conf(adapter);
879 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
883 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
887 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
888 ETH_MQ_RX_RSS_FLAG) {
889 rc = ena_rss_init_default(adapter);
894 ena_stats_restart(dev);
896 adapter->state = ENA_ADAPTER_STATE_RUNNING;
901 static int ena_queue_restart(struct ena_ring *ring)
905 ena_assert_msg(ring->configured == 1,
906 "Trying to restart unconfigured queue\n");
908 ring->next_to_clean = 0;
909 ring->next_to_use = 0;
911 if (ring->type == ENA_RING_TYPE_TX)
914 rc = ena_populate_rx_queue(ring, ring->ring_size - 1);
915 if ((unsigned int)rc != ring->ring_size - 1) {
916 PMD_INIT_LOG(ERR, "Failed to populate rx ring !\n");
923 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
926 __rte_unused unsigned int socket_id,
927 __rte_unused const struct rte_eth_txconf *tx_conf)
929 struct ena_com_create_io_ctx ctx =
930 /* policy set to _HOST just to satisfy icc compiler */
931 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
932 ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
933 struct ena_ring *txq = NULL;
934 struct ena_adapter *adapter =
935 (struct ena_adapter *)(dev->data->dev_private);
939 struct ena_com_dev *ena_dev = &adapter->ena_dev;
941 txq = &adapter->tx_ring[queue_idx];
943 if (txq->configured) {
945 "API violation. Queue %d is already configured\n",
950 if (nb_desc > adapter->tx_ring_size) {
952 "Unsupported size of TX queue (max size: %d)\n",
953 adapter->tx_ring_size);
957 ena_qid = ENA_IO_TXQ_IDX(queue_idx);
959 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
961 ctx.msix_vector = -1; /* admin interrupts not used */
962 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
963 ctx.queue_size = adapter->tx_ring_size;
965 rc = ena_com_create_io_queue(ena_dev, &ctx);
968 "failed to create io TX queue #%d (qid:%d) rc: %d\n",
969 queue_idx, ena_qid, rc);
971 txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
972 txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
974 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
976 &txq->ena_com_io_cq);
979 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
981 ena_com_destroy_io_queue(ena_dev, ena_qid);
985 txq->port_id = dev->data->port_id;
986 txq->next_to_clean = 0;
987 txq->next_to_use = 0;
988 txq->ring_size = nb_desc;
990 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
991 sizeof(struct ena_tx_buffer) *
993 RTE_CACHE_LINE_SIZE);
994 if (!txq->tx_buffer_info) {
995 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
999 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1000 sizeof(u16) * txq->ring_size,
1001 RTE_CACHE_LINE_SIZE);
1002 if (!txq->empty_tx_reqs) {
1003 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1004 rte_free(txq->tx_buffer_info);
1007 for (i = 0; i < txq->ring_size; i++)
1008 txq->empty_tx_reqs[i] = i;
1010 /* Store pointer to this queue in upper layer */
1011 txq->configured = 1;
1012 dev->data->tx_queues[queue_idx] = txq;
1017 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1020 __rte_unused unsigned int socket_id,
1021 __rte_unused const struct rte_eth_rxconf *rx_conf,
1022 struct rte_mempool *mp)
1024 struct ena_com_create_io_ctx ctx =
1025 /* policy set to _HOST just to satisfy icc compiler */
1026 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1027 ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1028 struct ena_adapter *adapter =
1029 (struct ena_adapter *)(dev->data->dev_private);
1030 struct ena_ring *rxq = NULL;
1031 uint16_t ena_qid = 0;
1033 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1035 rxq = &adapter->rx_ring[queue_idx];
1036 if (rxq->configured) {
1038 "API violation. Queue %d is already configured\n",
1043 if (nb_desc > adapter->rx_ring_size) {
1045 "Unsupported size of RX queue (max size: %d)\n",
1046 adapter->rx_ring_size);
1050 ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1053 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1054 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1055 ctx.msix_vector = -1; /* admin interrupts not used */
1056 ctx.queue_size = adapter->rx_ring_size;
1058 rc = ena_com_create_io_queue(ena_dev, &ctx);
1060 RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1063 rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1064 rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1066 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1067 &rxq->ena_com_io_sq,
1068 &rxq->ena_com_io_cq);
1071 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1073 ena_com_destroy_io_queue(ena_dev, ena_qid);
1076 rxq->port_id = dev->data->port_id;
1077 rxq->next_to_clean = 0;
1078 rxq->next_to_use = 0;
1079 rxq->ring_size = nb_desc;
1082 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1083 sizeof(struct rte_mbuf *) * nb_desc,
1084 RTE_CACHE_LINE_SIZE);
1085 if (!rxq->rx_buffer_info) {
1086 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1090 /* Store pointer to this queue in upper layer */
1091 rxq->configured = 1;
1092 dev->data->rx_queues[queue_idx] = rxq;
1097 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1101 unsigned int ring_size = rxq->ring_size;
1102 unsigned int ring_mask = ring_size - 1;
1103 int next_to_use = rxq->next_to_use & ring_mask;
1104 struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1106 if (unlikely(!count))
1109 ena_assert_msg((((ENA_CIRC_COUNT(rxq->next_to_use, rxq->next_to_clean,
1111 count) < rxq->ring_size), "bad ring state");
1113 count = RTE_MIN(count, ring_size - next_to_use);
1115 /* get resources for incoming packets */
1116 rc = rte_mempool_get_bulk(rxq->mb_pool,
1117 (void **)(&mbufs[next_to_use]), count);
1118 if (unlikely(rc < 0)) {
1119 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1120 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1124 for (i = 0; i < count; i++) {
1125 struct rte_mbuf *mbuf = mbufs[next_to_use];
1126 struct ena_com_buf ebuf;
1128 rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1129 /* prepare physical address for DMA transaction */
1130 ebuf.paddr = mbuf->buf_physaddr + RTE_PKTMBUF_HEADROOM;
1131 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1132 /* pass resource to device */
1133 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1134 &ebuf, next_to_use);
1136 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1139 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use, ring_size);
1143 rxq->next_to_use = next_to_use;
1144 /* let HW know that it can fill buffers with data */
1145 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1150 static int ena_device_init(struct ena_com_dev *ena_dev,
1151 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1155 /* Initialize mmio registers */
1156 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1158 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1163 rc = ena_com_dev_reset(ena_dev);
1165 RTE_LOG(ERR, PMD, "cannot reset device\n");
1166 goto err_mmio_read_less;
1169 /* check FW version */
1170 rc = ena_com_validate_version(ena_dev);
1172 RTE_LOG(ERR, PMD, "device version is too low\n");
1173 goto err_mmio_read_less;
1176 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1178 /* ENA device administration layer init */
1179 rc = ena_com_admin_init(ena_dev, NULL, true);
1182 "cannot initialize ena admin queue with device\n");
1183 goto err_mmio_read_less;
1186 ena_config_host_info(ena_dev);
1188 /* To enable the msix interrupts the driver needs to know the number
1189 * of queues. So the driver uses polling mode to retrieve this
1192 ena_com_set_admin_polling_mode(ena_dev, true);
1194 /* Get Device Attributes and features */
1195 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1198 "cannot get attribute for ena device rc= %d\n", rc);
1199 goto err_admin_init;
1205 ena_com_admin_destroy(ena_dev);
1208 ena_com_mmio_reg_read_request_destroy(ena_dev);
1213 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1215 struct rte_pci_device *pci_dev;
1216 struct ena_adapter *adapter =
1217 (struct ena_adapter *)(eth_dev->data->dev_private);
1218 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1219 struct ena_com_dev_get_features_ctx get_feat_ctx;
1222 static int adapters_found;
1224 memset(adapter, 0, sizeof(struct ena_adapter));
1225 ena_dev = &adapter->ena_dev;
1227 eth_dev->dev_ops = &ena_dev_ops;
1228 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1229 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1230 adapter->rte_eth_dev_data = eth_dev->data;
1231 adapter->rte_dev = eth_dev;
1233 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1236 pci_dev = eth_dev->pci_dev;
1237 adapter->pdev = pci_dev;
1239 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d\n",
1240 pci_dev->addr.domain,
1242 pci_dev->addr.devid,
1243 pci_dev->addr.function);
1245 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1246 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1248 /* Present ENA_MEM_BAR indicates available LLQ mode.
1249 * Use corresponding policy
1251 if (adapter->dev_mem_base)
1252 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1253 else if (adapter->regs)
1254 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1256 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)\n",
1259 ena_dev->reg_bar = adapter->regs;
1260 ena_dev->dmadev = adapter->pdev;
1262 adapter->id_number = adapters_found;
1264 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1265 adapter->id_number);
1267 /* device specific initialization routine */
1268 rc = ena_device_init(ena_dev, &get_feat_ctx);
1270 PMD_INIT_LOG(CRIT, "Failed to init ENA device\n");
1274 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1275 if (get_feat_ctx.max_queues.max_llq_num == 0) {
1277 "Trying to use LLQ but llq_num is 0.\n"
1278 "Fall back into regular queues.\n");
1279 ena_dev->tx_mem_queue_type =
1280 ENA_ADMIN_PLACEMENT_POLICY_HOST;
1281 adapter->num_queues =
1282 get_feat_ctx.max_queues.max_sq_num;
1284 adapter->num_queues =
1285 get_feat_ctx.max_queues.max_llq_num;
1288 adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1291 queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1292 if ((queue_size <= 0) || (adapter->num_queues <= 0))
1295 adapter->tx_ring_size = queue_size;
1296 adapter->rx_ring_size = queue_size;
1298 /* prepare ring structures */
1299 ena_init_rings(adapter);
1301 ena_config_debug_area(adapter);
1303 /* Set max MTU for this device */
1304 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1306 /* Copy MAC address and point DPDK to it */
1307 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1308 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1309 (struct ether_addr *)adapter->mac_addr);
1311 adapter->drv_stats = rte_zmalloc("adapter stats",
1312 sizeof(*adapter->drv_stats),
1313 RTE_CACHE_LINE_SIZE);
1314 if (!adapter->drv_stats) {
1315 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1320 adapter->state = ENA_ADAPTER_STATE_INIT;
1325 static int ena_dev_configure(struct rte_eth_dev *dev)
1327 struct ena_adapter *adapter =
1328 (struct ena_adapter *)(dev->data->dev_private);
1330 if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1331 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1332 PMD_INIT_LOG(ERR, "Illegal adapter state: %d\n",
1337 switch (adapter->state) {
1338 case ENA_ADAPTER_STATE_INIT:
1339 case ENA_ADAPTER_STATE_STOPPED:
1340 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1342 case ENA_ADAPTER_STATE_CONFIG:
1343 RTE_LOG(WARNING, PMD,
1344 "Ivalid driver state while trying to configure device\n");
1353 static void ena_init_rings(struct ena_adapter *adapter)
1357 for (i = 0; i < adapter->num_queues; i++) {
1358 struct ena_ring *ring = &adapter->tx_ring[i];
1360 ring->configured = 0;
1361 ring->type = ENA_RING_TYPE_TX;
1362 ring->adapter = adapter;
1364 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1365 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1368 for (i = 0; i < adapter->num_queues; i++) {
1369 struct ena_ring *ring = &adapter->rx_ring[i];
1371 ring->configured = 0;
1372 ring->type = ENA_RING_TYPE_RX;
1373 ring->adapter = adapter;
1378 static void ena_infos_get(struct rte_eth_dev *dev,
1379 struct rte_eth_dev_info *dev_info)
1381 struct ena_adapter *adapter;
1382 struct ena_com_dev *ena_dev;
1383 struct ena_com_dev_get_features_ctx feat;
1384 uint32_t rx_feat = 0, tx_feat = 0;
1387 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1388 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1389 adapter = (struct ena_adapter *)(dev->data->dev_private);
1391 ena_dev = &adapter->ena_dev;
1392 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1394 dev_info->speed_capa =
1396 ETH_LINK_SPEED_2_5G |
1398 ETH_LINK_SPEED_10G |
1399 ETH_LINK_SPEED_25G |
1400 ETH_LINK_SPEED_40G |
1401 ETH_LINK_SPEED_50G |
1402 ETH_LINK_SPEED_100G;
1404 /* Get supported features from HW */
1405 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1408 "Cannot get attribute for ena device rc= %d\n", rc);
1412 /* Set Tx & Rx features available for device */
1413 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1414 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1416 if (feat.offload.tx &
1417 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1418 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1419 DEV_TX_OFFLOAD_UDP_CKSUM |
1420 DEV_TX_OFFLOAD_TCP_CKSUM;
1422 if (feat.offload.tx &
1423 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1424 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1425 DEV_RX_OFFLOAD_UDP_CKSUM |
1426 DEV_RX_OFFLOAD_TCP_CKSUM;
1428 /* Inform framework about available features */
1429 dev_info->rx_offload_capa = rx_feat;
1430 dev_info->tx_offload_capa = tx_feat;
1432 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1433 dev_info->max_rx_pktlen = adapter->max_mtu;
1434 dev_info->max_mac_addrs = 1;
1436 dev_info->max_rx_queues = adapter->num_queues;
1437 dev_info->max_tx_queues = adapter->num_queues;
1438 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1441 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1444 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1445 unsigned int ring_size = rx_ring->ring_size;
1446 unsigned int ring_mask = ring_size - 1;
1447 uint16_t next_to_clean = rx_ring->next_to_clean;
1448 int desc_in_use = 0;
1449 unsigned int recv_idx = 0;
1450 struct rte_mbuf *mbuf = NULL;
1451 struct rte_mbuf *mbuf_head = NULL;
1452 struct rte_mbuf *mbuf_prev = NULL;
1453 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1454 unsigned int completed;
1456 struct ena_com_rx_ctx ena_rx_ctx;
1459 /* Check adapter state */
1460 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1462 "Trying to receive pkts while device is NOT running\n");
1466 desc_in_use = ENA_CIRC_COUNT(rx_ring->next_to_use,
1467 next_to_clean, ring_size);
1468 if (unlikely(nb_pkts > desc_in_use))
1469 nb_pkts = desc_in_use;
1471 for (completed = 0; completed < nb_pkts; completed++) {
1474 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1475 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1476 ena_rx_ctx.descs = 0;
1477 /* receive packet context */
1478 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1479 rx_ring->ena_com_io_sq,
1482 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1486 if (unlikely(ena_rx_ctx.descs == 0))
1489 while (segments < ena_rx_ctx.descs) {
1490 mbuf = rx_buff_info[next_to_clean & ring_mask];
1491 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1492 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1495 if (segments == 0) {
1496 mbuf->nb_segs = ena_rx_ctx.descs;
1497 mbuf->port = rx_ring->port_id;
1501 /* for multi-segment pkts create mbuf chain */
1502 mbuf_prev->next = mbuf;
1504 mbuf_head->pkt_len += mbuf->data_len;
1509 ENA_RX_RING_IDX_NEXT(next_to_clean, ring_size);
1512 /* fill mbuf attributes if any */
1513 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1514 mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1516 /* pass to DPDK application head mbuf */
1517 rx_pkts[recv_idx] = mbuf_head;
1521 /* Burst refill to save doorbells, memory barriers, const interval */
1522 if (ring_size - desc_in_use - 1 > ENA_RING_DESCS_RATIO(ring_size))
1523 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use - 1);
1525 rx_ring->next_to_clean = next_to_clean & ring_mask;
1530 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1533 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1534 unsigned int next_to_use = tx_ring->next_to_use;
1535 struct rte_mbuf *mbuf;
1536 unsigned int ring_size = tx_ring->ring_size;
1537 unsigned int ring_mask = ring_size - 1;
1538 struct ena_com_tx_ctx ena_tx_ctx;
1539 struct ena_tx_buffer *tx_info;
1540 struct ena_com_buf *ebuf;
1541 uint16_t rc, req_id, total_tx_descs = 0;
1545 /* Check adapter state */
1546 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1548 "Trying to xmit pkts while device is NOT running\n");
1552 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1553 mbuf = tx_pkts[sent_idx];
1555 req_id = tx_ring->empty_tx_reqs[next_to_use];
1556 tx_info = &tx_ring->tx_buffer_info[req_id];
1557 tx_info->mbuf = mbuf;
1558 tx_info->num_of_bufs = 0;
1559 ebuf = tx_info->bufs;
1561 /* Prepare TX context */
1562 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1563 memset(&ena_tx_ctx.ena_meta, 0x0,
1564 sizeof(struct ena_com_tx_meta));
1565 ena_tx_ctx.ena_bufs = ebuf;
1566 ena_tx_ctx.req_id = req_id;
1567 if (tx_ring->tx_mem_queue_type ==
1568 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1569 /* prepare the push buffer with
1570 * virtual address of the data
1572 ena_tx_ctx.header_len =
1573 RTE_MIN(mbuf->data_len,
1574 tx_ring->tx_max_header_size);
1575 ena_tx_ctx.push_header =
1576 (void *)((char *)mbuf->buf_addr +
1578 } /* there's no else as we take advantage of memset zeroing */
1580 /* Set TX offloads flags, if applicable */
1581 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx);
1583 if (unlikely(mbuf->ol_flags &
1584 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1585 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1587 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1589 /* Process first segment taking into
1590 * consideration pushed header
1592 if (mbuf->data_len > ena_tx_ctx.header_len) {
1593 ebuf->paddr = mbuf->buf_physaddr +
1595 ena_tx_ctx.header_len;
1596 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1598 tx_info->num_of_bufs++;
1601 while ((mbuf = mbuf->next) != NULL) {
1602 ebuf->paddr = mbuf->buf_physaddr + mbuf->data_off;
1603 ebuf->len = mbuf->data_len;
1605 tx_info->num_of_bufs++;
1608 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1610 /* Write data to device */
1611 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1612 &ena_tx_ctx, &nb_hw_desc);
1616 tx_info->tx_descs = nb_hw_desc;
1618 next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use, ring_size);
1621 /* Let HW do it's best :-) */
1623 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1625 /* Clear complete packets */
1626 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1627 /* Get Tx info & store how many descs were processed */
1628 tx_info = &tx_ring->tx_buffer_info[req_id];
1629 total_tx_descs += tx_info->tx_descs;
1631 /* Free whole mbuf chain */
1632 mbuf = tx_info->mbuf;
1633 rte_pktmbuf_free(mbuf);
1635 /* Put back descriptor to the ring for reuse */
1636 tx_ring->empty_tx_reqs[tx_ring->next_to_clean] = req_id;
1637 tx_ring->next_to_clean =
1638 ENA_TX_RING_IDX_NEXT(tx_ring->next_to_clean,
1639 tx_ring->ring_size);
1641 /* If too many descs to clean, leave it for another run */
1642 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1646 /* acknowledge completion of sent packets */
1647 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1648 tx_ring->next_to_use = next_to_use;
1652 static struct eth_driver rte_ena_pmd = {
1654 .name = "rte_ena_pmd",
1655 .id_table = pci_id_ena_map,
1656 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1658 .eth_dev_init = eth_ena_dev_init,
1659 .dev_private_size = sizeof(struct ena_adapter),
1663 rte_ena_pmd_init(const char *name __rte_unused,
1664 const char *params __rte_unused)
1666 rte_eth_driver_register(&rte_ena_pmd);
1670 struct rte_driver ena_pmd_drv = {
1672 .init = rte_ena_pmd_init,
1675 PMD_REGISTER_DRIVER(ena_pmd_drv, ena);
1676 DRIVER_REGISTER_PCI_TABLE(ena, pci_id_ena_map);