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>
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 /** Vendor ID used by Amazon devices */
168 #define PCI_VENDOR_ID_AMAZON 0x1D0F
169 /** Amazon devices */
170 #define PCI_DEVICE_ID_ENA_VF 0xEC20
171 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
173 #define ENA_TX_OFFLOAD_MASK (\
178 #define ENA_TX_OFFLOAD_NOTSUP_MASK \
179 (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
181 static const struct rte_pci_id pci_id_ena_map[] = {
182 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
183 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
187 static int ena_device_init(struct ena_com_dev *ena_dev,
188 struct ena_com_dev_get_features_ctx *get_feat_ctx);
189 static int ena_dev_configure(struct rte_eth_dev *dev);
190 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
192 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
194 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
195 uint16_t nb_desc, unsigned int socket_id,
196 const struct rte_eth_txconf *tx_conf);
197 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
198 uint16_t nb_desc, unsigned int socket_id,
199 const struct rte_eth_rxconf *rx_conf,
200 struct rte_mempool *mp);
201 static uint16_t eth_ena_recv_pkts(void *rx_queue,
202 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
203 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
204 static void ena_init_rings(struct ena_adapter *adapter);
205 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
206 static int ena_start(struct rte_eth_dev *dev);
207 static void ena_close(struct rte_eth_dev *dev);
208 static void ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
209 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
210 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
211 static void ena_rx_queue_release(void *queue);
212 static void ena_tx_queue_release(void *queue);
213 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
214 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
215 static int ena_link_update(struct rte_eth_dev *dev,
216 int wait_to_complete);
217 static int ena_queue_restart(struct ena_ring *ring);
218 static int ena_queue_restart_all(struct rte_eth_dev *dev,
219 enum ena_ring_type ring_type);
220 static void ena_stats_restart(struct rte_eth_dev *dev);
221 static void ena_infos_get(struct rte_eth_dev *dev,
222 struct rte_eth_dev_info *dev_info);
223 static int ena_rss_reta_update(struct rte_eth_dev *dev,
224 struct rte_eth_rss_reta_entry64 *reta_conf,
226 static int ena_rss_reta_query(struct rte_eth_dev *dev,
227 struct rte_eth_rss_reta_entry64 *reta_conf,
229 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
231 static const struct eth_dev_ops ena_dev_ops = {
232 .dev_configure = ena_dev_configure,
233 .dev_infos_get = ena_infos_get,
234 .rx_queue_setup = ena_rx_queue_setup,
235 .tx_queue_setup = ena_tx_queue_setup,
236 .dev_start = ena_start,
237 .link_update = ena_link_update,
238 .stats_get = ena_stats_get,
239 .mtu_set = ena_mtu_set,
240 .rx_queue_release = ena_rx_queue_release,
241 .tx_queue_release = ena_tx_queue_release,
242 .dev_close = ena_close,
243 .reta_update = ena_rss_reta_update,
244 .reta_query = ena_rss_reta_query,
247 #define NUMA_NO_NODE SOCKET_ID_ANY
249 static inline int ena_cpu_to_node(int cpu)
251 struct rte_config *config = rte_eal_get_configuration();
253 if (likely(cpu < RTE_MAX_MEMZONE))
254 return config->mem_config->memzone[cpu].socket_id;
259 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
260 struct ena_com_rx_ctx *ena_rx_ctx)
262 uint64_t ol_flags = 0;
264 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
265 ol_flags |= PKT_TX_TCP_CKSUM;
266 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
267 ol_flags |= PKT_TX_UDP_CKSUM;
269 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
270 ol_flags |= PKT_TX_IPV4;
271 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
272 ol_flags |= PKT_TX_IPV6;
274 if (unlikely(ena_rx_ctx->l4_csum_err))
275 ol_flags |= PKT_RX_L4_CKSUM_BAD;
276 if (unlikely(ena_rx_ctx->l3_csum_err))
277 ol_flags |= PKT_RX_IP_CKSUM_BAD;
279 mbuf->ol_flags = ol_flags;
282 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
283 struct ena_com_tx_ctx *ena_tx_ctx)
285 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
288 (PKT_TX_L4_MASK | PKT_TX_IP_CKSUM | PKT_TX_TCP_SEG)) {
289 /* check if TSO is required */
290 if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
291 ena_tx_ctx->tso_enable = true;
293 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
296 /* check if L3 checksum is needed */
297 if (mbuf->ol_flags & PKT_TX_IP_CKSUM)
298 ena_tx_ctx->l3_csum_enable = true;
300 if (mbuf->ol_flags & PKT_TX_IPV6) {
301 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
303 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
305 /* set don't fragment (DF) flag */
306 if (mbuf->packet_type &
307 (RTE_PTYPE_L4_NONFRAG
308 | RTE_PTYPE_INNER_L4_NONFRAG))
309 ena_tx_ctx->df = true;
312 /* check if L4 checksum is needed */
313 switch (mbuf->ol_flags & PKT_TX_L4_MASK) {
314 case PKT_TX_TCP_CKSUM:
315 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
316 ena_tx_ctx->l4_csum_enable = true;
318 case PKT_TX_UDP_CKSUM:
319 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
320 ena_tx_ctx->l4_csum_enable = true;
323 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
324 ena_tx_ctx->l4_csum_enable = false;
328 ena_meta->mss = mbuf->tso_segsz;
329 ena_meta->l3_hdr_len = mbuf->l3_len;
330 ena_meta->l3_hdr_offset = mbuf->l2_len;
331 /* this param needed only for TSO */
332 ena_meta->l3_outer_hdr_len = 0;
333 ena_meta->l3_outer_hdr_offset = 0;
335 ena_tx_ctx->meta_valid = true;
337 ena_tx_ctx->meta_valid = false;
341 static void ena_config_host_info(struct ena_com_dev *ena_dev)
343 struct ena_admin_host_info *host_info;
346 /* Allocate only the host info */
347 rc = ena_com_allocate_host_info(ena_dev);
349 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
353 host_info = ena_dev->host_attr.host_info;
355 host_info->os_type = ENA_ADMIN_OS_DPDK;
356 host_info->kernel_ver = RTE_VERSION;
357 snprintf((char *)host_info->kernel_ver_str,
358 sizeof(host_info->kernel_ver_str),
359 "%s", rte_version());
360 host_info->os_dist = RTE_VERSION;
361 snprintf((char *)host_info->os_dist_str,
362 sizeof(host_info->os_dist_str),
363 "%s", rte_version());
364 host_info->driver_version =
365 (DRV_MODULE_VER_MAJOR) |
366 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
367 (DRV_MODULE_VER_SUBMINOR <<
368 ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
370 rc = ena_com_set_host_attributes(ena_dev);
372 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
380 ena_com_delete_host_info(ena_dev);
384 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
386 if (sset != ETH_SS_STATS)
389 /* Workaround for clang:
390 * touch internal structures to prevent
393 ENA_TOUCH(ena_stats_global_strings);
394 ENA_TOUCH(ena_stats_tx_strings);
395 ENA_TOUCH(ena_stats_rx_strings);
396 ENA_TOUCH(ena_stats_ena_com_strings);
398 return dev->data->nb_tx_queues *
399 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
400 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
403 static void ena_config_debug_area(struct ena_adapter *adapter)
408 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
410 RTE_LOG(ERR, PMD, "SS count is negative\n");
414 /* allocate 32 bytes for each string and 64bit for the value */
415 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
417 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
419 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
423 rc = ena_com_set_host_attributes(&adapter->ena_dev);
425 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
432 ena_com_delete_debug_area(&adapter->ena_dev);
435 static void ena_close(struct rte_eth_dev *dev)
437 struct ena_adapter *adapter =
438 (struct ena_adapter *)(dev->data->dev_private);
440 adapter->state = ENA_ADAPTER_STATE_STOPPED;
442 ena_rx_queue_release_all(dev);
443 ena_tx_queue_release_all(dev);
446 static int ena_rss_reta_update(struct rte_eth_dev *dev,
447 struct rte_eth_rss_reta_entry64 *reta_conf,
450 struct ena_adapter *adapter =
451 (struct ena_adapter *)(dev->data->dev_private);
452 struct ena_com_dev *ena_dev = &adapter->ena_dev;
458 if ((reta_size == 0) || (reta_conf == NULL))
461 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
462 RTE_LOG(WARNING, PMD,
463 "indirection table %d is bigger than supported (%d)\n",
464 reta_size, ENA_RX_RSS_TABLE_SIZE);
469 for (i = 0 ; i < reta_size ; i++) {
470 /* each reta_conf is for 64 entries.
471 * to support 128 we use 2 conf of 64
473 conf_idx = i / RTE_RETA_GROUP_SIZE;
474 idx = i % RTE_RETA_GROUP_SIZE;
475 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
477 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
478 ret = ena_com_indirect_table_fill_entry(ena_dev,
481 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
483 "Cannot fill indirect table\n");
490 ret = ena_com_indirect_table_set(ena_dev);
491 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
492 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
497 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
498 __func__, reta_size, adapter->rte_dev->data->port_id);
503 /* Query redirection table. */
504 static int ena_rss_reta_query(struct rte_eth_dev *dev,
505 struct rte_eth_rss_reta_entry64 *reta_conf,
508 struct ena_adapter *adapter =
509 (struct ena_adapter *)(dev->data->dev_private);
510 struct ena_com_dev *ena_dev = &adapter->ena_dev;
513 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
517 if (reta_size == 0 || reta_conf == NULL ||
518 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
521 ret = ena_com_indirect_table_get(ena_dev, indirect_table);
522 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
523 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
528 for (i = 0 ; i < reta_size ; i++) {
529 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
530 reta_idx = i % RTE_RETA_GROUP_SIZE;
531 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
532 reta_conf[reta_conf_idx].reta[reta_idx] =
533 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
539 static int ena_rss_init_default(struct ena_adapter *adapter)
541 struct ena_com_dev *ena_dev = &adapter->ena_dev;
542 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
546 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
548 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
552 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
553 val = i % nb_rx_queues;
554 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
555 ENA_IO_RXQ_IDX(val));
556 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
557 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
562 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
563 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
564 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
565 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
569 rc = ena_com_set_default_hash_ctrl(ena_dev);
570 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
571 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
575 rc = ena_com_indirect_table_set(ena_dev);
576 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
577 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
580 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
581 adapter->rte_dev->data->port_id);
586 ena_com_rss_destroy(ena_dev);
592 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
594 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
595 int nb_queues = dev->data->nb_rx_queues;
598 for (i = 0; i < nb_queues; i++)
599 ena_rx_queue_release(queues[i]);
602 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
604 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
605 int nb_queues = dev->data->nb_tx_queues;
608 for (i = 0; i < nb_queues; i++)
609 ena_tx_queue_release(queues[i]);
612 static void ena_rx_queue_release(void *queue)
614 struct ena_ring *ring = (struct ena_ring *)queue;
615 struct ena_adapter *adapter = ring->adapter;
618 ena_assert_msg(ring->configured,
619 "API violation - releasing not configured queue");
620 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
623 /* Destroy HW queue */
624 ena_qid = ENA_IO_RXQ_IDX(ring->id);
625 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
628 ena_rx_queue_release_bufs(ring);
630 /* Free ring resources */
631 if (ring->rx_buffer_info)
632 rte_free(ring->rx_buffer_info);
633 ring->rx_buffer_info = NULL;
635 ring->configured = 0;
637 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
638 ring->port_id, ring->id);
641 static void ena_tx_queue_release(void *queue)
643 struct ena_ring *ring = (struct ena_ring *)queue;
644 struct ena_adapter *adapter = ring->adapter;
647 ena_assert_msg(ring->configured,
648 "API violation. Releasing not configured queue");
649 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
652 /* Destroy HW queue */
653 ena_qid = ENA_IO_TXQ_IDX(ring->id);
654 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
657 ena_tx_queue_release_bufs(ring);
659 /* Free ring resources */
660 if (ring->tx_buffer_info)
661 rte_free(ring->tx_buffer_info);
663 if (ring->empty_tx_reqs)
664 rte_free(ring->empty_tx_reqs);
666 ring->empty_tx_reqs = NULL;
667 ring->tx_buffer_info = NULL;
669 ring->configured = 0;
671 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
672 ring->port_id, ring->id);
675 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
677 unsigned int ring_mask = ring->ring_size - 1;
679 while (ring->next_to_clean != ring->next_to_use) {
681 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
684 rte_mbuf_raw_free(m);
686 ring->next_to_clean++;
690 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
692 unsigned int ring_mask = ring->ring_size - 1;
694 while (ring->next_to_clean != ring->next_to_use) {
695 struct ena_tx_buffer *tx_buf =
696 &ring->tx_buffer_info[ring->next_to_clean & ring_mask];
699 rte_pktmbuf_free(tx_buf->mbuf);
701 ring->next_to_clean++;
705 static int ena_link_update(struct rte_eth_dev *dev,
706 __rte_unused int wait_to_complete)
708 struct rte_eth_link *link = &dev->data->dev_link;
710 link->link_status = 1;
711 link->link_speed = ETH_SPEED_NUM_10G;
712 link->link_duplex = ETH_LINK_FULL_DUPLEX;
717 static int ena_queue_restart_all(struct rte_eth_dev *dev,
718 enum ena_ring_type ring_type)
720 struct ena_adapter *adapter =
721 (struct ena_adapter *)(dev->data->dev_private);
722 struct ena_ring *queues = NULL;
726 queues = (ring_type == ENA_RING_TYPE_RX) ?
727 adapter->rx_ring : adapter->tx_ring;
729 for (i = 0; i < adapter->num_queues; i++) {
730 if (queues[i].configured) {
731 if (ring_type == ENA_RING_TYPE_RX) {
733 dev->data->rx_queues[i] == &queues[i],
734 "Inconsistent state of rx queues\n");
737 dev->data->tx_queues[i] == &queues[i],
738 "Inconsistent state of tx queues\n");
741 rc = ena_queue_restart(&queues[i]);
745 "failed to restart queue %d type(%d)",
755 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
757 uint32_t max_frame_len = adapter->max_mtu;
759 if (adapter->rte_eth_dev_data->dev_conf.rxmode.jumbo_frame == 1)
761 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
763 return max_frame_len;
766 static int ena_check_valid_conf(struct ena_adapter *adapter)
768 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
770 if (max_frame_len > adapter->max_mtu) {
771 PMD_INIT_LOG(ERR, "Unsupported MTU of %d", max_frame_len);
779 ena_calc_queue_size(struct ena_com_dev *ena_dev,
780 struct ena_com_dev_get_features_ctx *get_feat_ctx)
782 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
784 queue_size = RTE_MIN(queue_size,
785 get_feat_ctx->max_queues.max_cq_depth);
786 queue_size = RTE_MIN(queue_size,
787 get_feat_ctx->max_queues.max_sq_depth);
789 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
790 queue_size = RTE_MIN(queue_size,
791 get_feat_ctx->max_queues.max_llq_depth);
793 /* Round down to power of 2 */
794 if (!rte_is_power_of_2(queue_size))
795 queue_size = rte_align32pow2(queue_size >> 1);
797 if (queue_size == 0) {
798 PMD_INIT_LOG(ERR, "Invalid queue size");
805 static void ena_stats_restart(struct rte_eth_dev *dev)
807 struct ena_adapter *adapter =
808 (struct ena_adapter *)(dev->data->dev_private);
810 rte_atomic64_init(&adapter->drv_stats->ierrors);
811 rte_atomic64_init(&adapter->drv_stats->oerrors);
812 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
815 static void ena_stats_get(struct rte_eth_dev *dev,
816 struct rte_eth_stats *stats)
818 struct ena_admin_basic_stats ena_stats;
819 struct ena_adapter *adapter =
820 (struct ena_adapter *)(dev->data->dev_private);
821 struct ena_com_dev *ena_dev = &adapter->ena_dev;
824 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
827 memset(&ena_stats, 0, sizeof(ena_stats));
828 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
830 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
834 /* Set of basic statistics from ENA */
835 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
836 ena_stats.rx_pkts_low);
837 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
838 ena_stats.tx_pkts_low);
839 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
840 ena_stats.rx_bytes_low);
841 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
842 ena_stats.tx_bytes_low);
843 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
844 ena_stats.rx_drops_low);
846 /* Driver related stats */
847 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
848 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
849 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
852 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
854 struct ena_adapter *adapter;
855 struct ena_com_dev *ena_dev;
858 ena_assert_msg(dev->data != NULL, "Uninitialized device");
859 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
860 adapter = (struct ena_adapter *)(dev->data->dev_private);
862 ena_dev = &adapter->ena_dev;
863 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
865 if (mtu > ena_get_mtu_conf(adapter)) {
867 "Given MTU (%d) exceeds maximum MTU supported (%d)\n",
868 mtu, ena_get_mtu_conf(adapter));
873 rc = ena_com_set_dev_mtu(ena_dev, mtu);
875 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
877 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
883 static int ena_start(struct rte_eth_dev *dev)
885 struct ena_adapter *adapter =
886 (struct ena_adapter *)(dev->data->dev_private);
889 if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
890 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
891 PMD_INIT_LOG(ERR, "API violation");
895 rc = ena_check_valid_conf(adapter);
899 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
903 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
907 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
908 ETH_MQ_RX_RSS_FLAG) {
909 rc = ena_rss_init_default(adapter);
914 ena_stats_restart(dev);
916 adapter->state = ENA_ADAPTER_STATE_RUNNING;
921 static int ena_queue_restart(struct ena_ring *ring)
925 ena_assert_msg(ring->configured == 1,
926 "Trying to restart unconfigured queue\n");
928 ring->next_to_clean = 0;
929 ring->next_to_use = 0;
931 if (ring->type == ENA_RING_TYPE_TX)
934 bufs_num = ring->ring_size - 1;
935 rc = ena_populate_rx_queue(ring, bufs_num);
936 if (rc != bufs_num) {
937 PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
944 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
947 __rte_unused unsigned int socket_id,
948 __rte_unused const struct rte_eth_txconf *tx_conf)
950 struct ena_com_create_io_ctx ctx =
951 /* policy set to _HOST just to satisfy icc compiler */
952 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
953 ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
954 struct ena_ring *txq = NULL;
955 struct ena_adapter *adapter =
956 (struct ena_adapter *)(dev->data->dev_private);
960 struct ena_com_dev *ena_dev = &adapter->ena_dev;
962 txq = &adapter->tx_ring[queue_idx];
964 if (txq->configured) {
966 "API violation. Queue %d is already configured\n",
971 if (!rte_is_power_of_2(nb_desc)) {
973 "Unsupported size of RX queue: %d is not a power of 2.",
978 if (nb_desc > adapter->tx_ring_size) {
980 "Unsupported size of TX queue (max size: %d)\n",
981 adapter->tx_ring_size);
985 ena_qid = ENA_IO_TXQ_IDX(queue_idx);
987 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
989 ctx.msix_vector = -1; /* admin interrupts not used */
990 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
991 ctx.queue_size = adapter->tx_ring_size;
992 ctx.numa_node = ena_cpu_to_node(queue_idx);
994 rc = ena_com_create_io_queue(ena_dev, &ctx);
997 "failed to create io TX queue #%d (qid:%d) rc: %d\n",
998 queue_idx, ena_qid, rc);
1000 txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1001 txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1003 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1004 &txq->ena_com_io_sq,
1005 &txq->ena_com_io_cq);
1008 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1010 ena_com_destroy_io_queue(ena_dev, ena_qid);
1014 txq->port_id = dev->data->port_id;
1015 txq->next_to_clean = 0;
1016 txq->next_to_use = 0;
1017 txq->ring_size = nb_desc;
1019 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1020 sizeof(struct ena_tx_buffer) *
1022 RTE_CACHE_LINE_SIZE);
1023 if (!txq->tx_buffer_info) {
1024 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1028 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1029 sizeof(u16) * txq->ring_size,
1030 RTE_CACHE_LINE_SIZE);
1031 if (!txq->empty_tx_reqs) {
1032 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1033 rte_free(txq->tx_buffer_info);
1036 for (i = 0; i < txq->ring_size; i++)
1037 txq->empty_tx_reqs[i] = i;
1039 /* Store pointer to this queue in upper layer */
1040 txq->configured = 1;
1041 dev->data->tx_queues[queue_idx] = txq;
1046 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1049 __rte_unused unsigned int socket_id,
1050 __rte_unused const struct rte_eth_rxconf *rx_conf,
1051 struct rte_mempool *mp)
1053 struct ena_com_create_io_ctx ctx =
1054 /* policy set to _HOST just to satisfy icc compiler */
1055 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1056 ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1057 struct ena_adapter *adapter =
1058 (struct ena_adapter *)(dev->data->dev_private);
1059 struct ena_ring *rxq = NULL;
1060 uint16_t ena_qid = 0;
1062 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1064 rxq = &adapter->rx_ring[queue_idx];
1065 if (rxq->configured) {
1067 "API violation. Queue %d is already configured\n",
1072 if (!rte_is_power_of_2(nb_desc)) {
1074 "Unsupported size of TX queue: %d is not a power of 2.",
1079 if (nb_desc > adapter->rx_ring_size) {
1081 "Unsupported size of RX queue (max size: %d)\n",
1082 adapter->rx_ring_size);
1086 ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1089 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1090 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1091 ctx.msix_vector = -1; /* admin interrupts not used */
1092 ctx.queue_size = adapter->rx_ring_size;
1093 ctx.numa_node = ena_cpu_to_node(queue_idx);
1095 rc = ena_com_create_io_queue(ena_dev, &ctx);
1097 RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1100 rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1101 rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1103 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1104 &rxq->ena_com_io_sq,
1105 &rxq->ena_com_io_cq);
1108 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1110 ena_com_destroy_io_queue(ena_dev, ena_qid);
1113 rxq->port_id = dev->data->port_id;
1114 rxq->next_to_clean = 0;
1115 rxq->next_to_use = 0;
1116 rxq->ring_size = nb_desc;
1119 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1120 sizeof(struct rte_mbuf *) * nb_desc,
1121 RTE_CACHE_LINE_SIZE);
1122 if (!rxq->rx_buffer_info) {
1123 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1127 /* Store pointer to this queue in upper layer */
1128 rxq->configured = 1;
1129 dev->data->rx_queues[queue_idx] = rxq;
1134 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1138 uint16_t ring_size = rxq->ring_size;
1139 uint16_t ring_mask = ring_size - 1;
1140 uint16_t next_to_use = rxq->next_to_use;
1142 struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1144 if (unlikely(!count))
1147 in_use = rxq->next_to_use - rxq->next_to_clean;
1148 ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1150 count = RTE_MIN(count,
1151 (uint16_t)(ring_size - (next_to_use & ring_mask)));
1153 /* get resources for incoming packets */
1154 rc = rte_mempool_get_bulk(rxq->mb_pool,
1155 (void **)(&mbufs[next_to_use & ring_mask]),
1157 if (unlikely(rc < 0)) {
1158 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1159 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1163 for (i = 0; i < count; i++) {
1164 uint16_t next_to_use_masked = next_to_use & ring_mask;
1165 struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1166 struct ena_com_buf ebuf;
1168 rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1169 /* prepare physical address for DMA transaction */
1170 ebuf.paddr = mbuf->buf_physaddr + RTE_PKTMBUF_HEADROOM;
1171 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1172 /* pass resource to device */
1173 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1174 &ebuf, next_to_use_masked);
1176 rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1178 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1184 /* When we submitted free recources to device... */
1186 /* ...let HW know that it can fill buffers with data */
1188 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1190 rxq->next_to_use = next_to_use;
1196 static int ena_device_init(struct ena_com_dev *ena_dev,
1197 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1200 bool readless_supported;
1202 /* Initialize mmio registers */
1203 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1205 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1209 /* The PCIe configuration space revision id indicate if mmio reg
1212 readless_supported =
1213 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1214 & ENA_MMIO_DISABLE_REG_READ);
1215 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1218 rc = ena_com_dev_reset(ena_dev);
1220 RTE_LOG(ERR, PMD, "cannot reset device\n");
1221 goto err_mmio_read_less;
1224 /* check FW version */
1225 rc = ena_com_validate_version(ena_dev);
1227 RTE_LOG(ERR, PMD, "device version is too low\n");
1228 goto err_mmio_read_less;
1231 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1233 /* ENA device administration layer init */
1234 rc = ena_com_admin_init(ena_dev, NULL, true);
1237 "cannot initialize ena admin queue with device\n");
1238 goto err_mmio_read_less;
1241 /* To enable the msix interrupts the driver needs to know the number
1242 * of queues. So the driver uses polling mode to retrieve this
1245 ena_com_set_admin_polling_mode(ena_dev, true);
1247 ena_config_host_info(ena_dev);
1249 /* Get Device Attributes and features */
1250 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1253 "cannot get attribute for ena device rc= %d\n", rc);
1254 goto err_admin_init;
1260 ena_com_admin_destroy(ena_dev);
1263 ena_com_mmio_reg_read_request_destroy(ena_dev);
1268 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1270 struct rte_pci_device *pci_dev;
1271 struct ena_adapter *adapter =
1272 (struct ena_adapter *)(eth_dev->data->dev_private);
1273 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1274 struct ena_com_dev_get_features_ctx get_feat_ctx;
1277 static int adapters_found;
1279 memset(adapter, 0, sizeof(struct ena_adapter));
1280 ena_dev = &adapter->ena_dev;
1282 eth_dev->dev_ops = &ena_dev_ops;
1283 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1284 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1285 eth_dev->tx_pkt_prepare = ð_ena_prep_pkts;
1286 adapter->rte_eth_dev_data = eth_dev->data;
1287 adapter->rte_dev = eth_dev;
1289 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1292 pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
1293 adapter->pdev = pci_dev;
1295 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1296 pci_dev->addr.domain,
1298 pci_dev->addr.devid,
1299 pci_dev->addr.function);
1301 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1302 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1304 /* Present ENA_MEM_BAR indicates available LLQ mode.
1305 * Use corresponding policy
1307 if (adapter->dev_mem_base)
1308 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1309 else if (adapter->regs)
1310 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1312 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1315 ena_dev->reg_bar = adapter->regs;
1316 ena_dev->dmadev = adapter->pdev;
1318 adapter->id_number = adapters_found;
1320 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1321 adapter->id_number);
1323 /* device specific initialization routine */
1324 rc = ena_device_init(ena_dev, &get_feat_ctx);
1326 PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1330 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1331 if (get_feat_ctx.max_queues.max_llq_num == 0) {
1333 "Trying to use LLQ but llq_num is 0.\n"
1334 "Fall back into regular queues.");
1335 ena_dev->tx_mem_queue_type =
1336 ENA_ADMIN_PLACEMENT_POLICY_HOST;
1337 adapter->num_queues =
1338 get_feat_ctx.max_queues.max_sq_num;
1340 adapter->num_queues =
1341 get_feat_ctx.max_queues.max_llq_num;
1344 adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1347 queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1348 if ((queue_size <= 0) || (adapter->num_queues <= 0))
1351 adapter->tx_ring_size = queue_size;
1352 adapter->rx_ring_size = queue_size;
1354 /* prepare ring structures */
1355 ena_init_rings(adapter);
1357 ena_config_debug_area(adapter);
1359 /* Set max MTU for this device */
1360 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1362 /* set device support for TSO */
1363 adapter->tso4_supported = get_feat_ctx.offload.tx &
1364 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1366 /* Copy MAC address and point DPDK to it */
1367 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1368 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1369 (struct ether_addr *)adapter->mac_addr);
1371 adapter->drv_stats = rte_zmalloc("adapter stats",
1372 sizeof(*adapter->drv_stats),
1373 RTE_CACHE_LINE_SIZE);
1374 if (!adapter->drv_stats) {
1375 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1380 adapter->state = ENA_ADAPTER_STATE_INIT;
1385 static int ena_dev_configure(struct rte_eth_dev *dev)
1387 struct ena_adapter *adapter =
1388 (struct ena_adapter *)(dev->data->dev_private);
1390 if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1391 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1392 PMD_INIT_LOG(ERR, "Illegal adapter state: %d",
1397 switch (adapter->state) {
1398 case ENA_ADAPTER_STATE_INIT:
1399 case ENA_ADAPTER_STATE_STOPPED:
1400 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1402 case ENA_ADAPTER_STATE_CONFIG:
1403 RTE_LOG(WARNING, PMD,
1404 "Ivalid driver state while trying to configure device\n");
1413 static void ena_init_rings(struct ena_adapter *adapter)
1417 for (i = 0; i < adapter->num_queues; i++) {
1418 struct ena_ring *ring = &adapter->tx_ring[i];
1420 ring->configured = 0;
1421 ring->type = ENA_RING_TYPE_TX;
1422 ring->adapter = adapter;
1424 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1425 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1428 for (i = 0; i < adapter->num_queues; i++) {
1429 struct ena_ring *ring = &adapter->rx_ring[i];
1431 ring->configured = 0;
1432 ring->type = ENA_RING_TYPE_RX;
1433 ring->adapter = adapter;
1438 static void ena_infos_get(struct rte_eth_dev *dev,
1439 struct rte_eth_dev_info *dev_info)
1441 struct ena_adapter *adapter;
1442 struct ena_com_dev *ena_dev;
1443 struct ena_com_dev_get_features_ctx feat;
1444 uint32_t rx_feat = 0, tx_feat = 0;
1447 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1448 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1449 adapter = (struct ena_adapter *)(dev->data->dev_private);
1451 ena_dev = &adapter->ena_dev;
1452 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1454 dev_info->pci_dev = RTE_DEV_TO_PCI(dev->device);
1456 dev_info->speed_capa =
1458 ETH_LINK_SPEED_2_5G |
1460 ETH_LINK_SPEED_10G |
1461 ETH_LINK_SPEED_25G |
1462 ETH_LINK_SPEED_40G |
1463 ETH_LINK_SPEED_50G |
1464 ETH_LINK_SPEED_100G;
1466 /* Get supported features from HW */
1467 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1470 "Cannot get attribute for ena device rc= %d\n", rc);
1474 /* Set Tx & Rx features available for device */
1475 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1476 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1478 if (feat.offload.tx &
1479 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1480 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1481 DEV_TX_OFFLOAD_UDP_CKSUM |
1482 DEV_TX_OFFLOAD_TCP_CKSUM;
1484 if (feat.offload.rx_supported &
1485 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1486 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1487 DEV_RX_OFFLOAD_UDP_CKSUM |
1488 DEV_RX_OFFLOAD_TCP_CKSUM;
1490 /* Inform framework about available features */
1491 dev_info->rx_offload_capa = rx_feat;
1492 dev_info->tx_offload_capa = tx_feat;
1494 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1495 dev_info->max_rx_pktlen = adapter->max_mtu;
1496 dev_info->max_mac_addrs = 1;
1498 dev_info->max_rx_queues = adapter->num_queues;
1499 dev_info->max_tx_queues = adapter->num_queues;
1500 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1503 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1506 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1507 unsigned int ring_size = rx_ring->ring_size;
1508 unsigned int ring_mask = ring_size - 1;
1509 uint16_t next_to_clean = rx_ring->next_to_clean;
1510 uint16_t desc_in_use = 0;
1511 unsigned int recv_idx = 0;
1512 struct rte_mbuf *mbuf = NULL;
1513 struct rte_mbuf *mbuf_head = NULL;
1514 struct rte_mbuf *mbuf_prev = NULL;
1515 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1516 unsigned int completed;
1518 struct ena_com_rx_ctx ena_rx_ctx;
1521 /* Check adapter state */
1522 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1524 "Trying to receive pkts while device is NOT running\n");
1528 desc_in_use = rx_ring->next_to_use - next_to_clean;
1529 if (unlikely(nb_pkts > desc_in_use))
1530 nb_pkts = desc_in_use;
1532 for (completed = 0; completed < nb_pkts; completed++) {
1535 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1536 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1537 ena_rx_ctx.descs = 0;
1538 /* receive packet context */
1539 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1540 rx_ring->ena_com_io_sq,
1543 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1547 if (unlikely(ena_rx_ctx.descs == 0))
1550 while (segments < ena_rx_ctx.descs) {
1551 mbuf = rx_buff_info[next_to_clean & ring_mask];
1552 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1553 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1556 if (segments == 0) {
1557 mbuf->nb_segs = ena_rx_ctx.descs;
1558 mbuf->port = rx_ring->port_id;
1562 /* for multi-segment pkts create mbuf chain */
1563 mbuf_prev->next = mbuf;
1565 mbuf_head->pkt_len += mbuf->data_len;
1572 /* fill mbuf attributes if any */
1573 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1574 mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1576 /* pass to DPDK application head mbuf */
1577 rx_pkts[recv_idx] = mbuf_head;
1581 rx_ring->next_to_clean = next_to_clean;
1583 desc_in_use = desc_in_use - completed + 1;
1584 /* Burst refill to save doorbells, memory barriers, const interval */
1585 if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1586 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1592 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1598 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1599 struct ipv4_hdr *ip_hdr;
1601 uint16_t frag_field;
1603 for (i = 0; i != nb_pkts; i++) {
1605 ol_flags = m->ol_flags;
1607 if (!(ol_flags & PKT_TX_IPV4))
1610 /* If there was not L2 header length specified, assume it is
1611 * length of the ethernet header.
1613 if (unlikely(m->l2_len == 0))
1614 m->l2_len = sizeof(struct ether_hdr);
1616 ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1618 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1620 if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1621 m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1623 /* If IPv4 header has DF flag enabled and TSO support is
1624 * disabled, partial chcecksum should not be calculated.
1626 if (!tx_ring->adapter->tso4_supported)
1630 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1631 (ol_flags & PKT_TX_L4_MASK) ==
1632 PKT_TX_SCTP_CKSUM) {
1633 rte_errno = -ENOTSUP;
1637 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1638 ret = rte_validate_tx_offload(m);
1645 /* In case we are supposed to TSO and have DF not set (DF=0)
1646 * hardware must be provided with partial checksum, otherwise
1647 * it will take care of necessary calculations.
1650 ret = rte_net_intel_cksum_flags_prepare(m,
1651 ol_flags & ~PKT_TX_TCP_SEG);
1661 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1664 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1665 uint16_t next_to_use = tx_ring->next_to_use;
1666 uint16_t next_to_clean = tx_ring->next_to_clean;
1667 struct rte_mbuf *mbuf;
1668 unsigned int ring_size = tx_ring->ring_size;
1669 unsigned int ring_mask = ring_size - 1;
1670 struct ena_com_tx_ctx ena_tx_ctx;
1671 struct ena_tx_buffer *tx_info;
1672 struct ena_com_buf *ebuf;
1673 uint16_t rc, req_id, total_tx_descs = 0;
1674 uint16_t sent_idx = 0, empty_tx_reqs;
1677 /* Check adapter state */
1678 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1680 "Trying to xmit pkts while device is NOT running\n");
1684 empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
1685 if (nb_pkts > empty_tx_reqs)
1686 nb_pkts = empty_tx_reqs;
1688 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1689 mbuf = tx_pkts[sent_idx];
1691 req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1692 tx_info = &tx_ring->tx_buffer_info[req_id];
1693 tx_info->mbuf = mbuf;
1694 tx_info->num_of_bufs = 0;
1695 ebuf = tx_info->bufs;
1697 /* Prepare TX context */
1698 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1699 memset(&ena_tx_ctx.ena_meta, 0x0,
1700 sizeof(struct ena_com_tx_meta));
1701 ena_tx_ctx.ena_bufs = ebuf;
1702 ena_tx_ctx.req_id = req_id;
1703 if (tx_ring->tx_mem_queue_type ==
1704 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1705 /* prepare the push buffer with
1706 * virtual address of the data
1708 ena_tx_ctx.header_len =
1709 RTE_MIN(mbuf->data_len,
1710 tx_ring->tx_max_header_size);
1711 ena_tx_ctx.push_header =
1712 (void *)((char *)mbuf->buf_addr +
1714 } /* there's no else as we take advantage of memset zeroing */
1716 /* Set TX offloads flags, if applicable */
1717 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx);
1719 if (unlikely(mbuf->ol_flags &
1720 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1721 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1723 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1725 /* Process first segment taking into
1726 * consideration pushed header
1728 if (mbuf->data_len > ena_tx_ctx.header_len) {
1729 ebuf->paddr = mbuf->buf_physaddr +
1731 ena_tx_ctx.header_len;
1732 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1734 tx_info->num_of_bufs++;
1737 while ((mbuf = mbuf->next) != NULL) {
1738 ebuf->paddr = mbuf->buf_physaddr + mbuf->data_off;
1739 ebuf->len = mbuf->data_len;
1741 tx_info->num_of_bufs++;
1744 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1746 /* Write data to device */
1747 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1748 &ena_tx_ctx, &nb_hw_desc);
1752 tx_info->tx_descs = nb_hw_desc;
1757 /* If there are ready packets to be xmitted... */
1759 /* ...let HW do its best :-) */
1761 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1763 tx_ring->next_to_use = next_to_use;
1766 /* Clear complete packets */
1767 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1768 /* Get Tx info & store how many descs were processed */
1769 tx_info = &tx_ring->tx_buffer_info[req_id];
1770 total_tx_descs += tx_info->tx_descs;
1772 /* Free whole mbuf chain */
1773 mbuf = tx_info->mbuf;
1774 rte_pktmbuf_free(mbuf);
1776 /* Put back descriptor to the ring for reuse */
1777 tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1780 /* If too many descs to clean, leave it for another run */
1781 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1785 if (total_tx_descs > 0) {
1786 /* acknowledge completion of sent packets */
1787 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1788 tx_ring->next_to_clean = next_to_clean;
1794 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1795 struct rte_pci_device *pci_dev)
1797 return rte_eth_dev_pci_generic_probe(pci_dev,
1798 sizeof(struct ena_adapter), eth_ena_dev_init);
1801 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
1803 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1806 static struct rte_pci_driver rte_ena_pmd = {
1807 .id_table = pci_id_ena_map,
1808 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1809 .probe = eth_ena_pci_probe,
1810 .remove = eth_ena_pci_remove,
1813 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
1814 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
1815 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");