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 1
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)
118 * Each rte_memzone should have unique name.
119 * To satisfy it, count number of allocation and add it to name.
121 uint32_t ena_alloc_cnt;
123 static const struct ena_stats ena_stats_global_strings[] = {
124 ENA_STAT_GLOBAL_ENTRY(tx_timeout),
125 ENA_STAT_GLOBAL_ENTRY(io_suspend),
126 ENA_STAT_GLOBAL_ENTRY(io_resume),
127 ENA_STAT_GLOBAL_ENTRY(wd_expired),
128 ENA_STAT_GLOBAL_ENTRY(interface_up),
129 ENA_STAT_GLOBAL_ENTRY(interface_down),
130 ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
133 static const struct ena_stats ena_stats_tx_strings[] = {
134 ENA_STAT_TX_ENTRY(cnt),
135 ENA_STAT_TX_ENTRY(bytes),
136 ENA_STAT_TX_ENTRY(queue_stop),
137 ENA_STAT_TX_ENTRY(queue_wakeup),
138 ENA_STAT_TX_ENTRY(dma_mapping_err),
139 ENA_STAT_TX_ENTRY(linearize),
140 ENA_STAT_TX_ENTRY(linearize_failed),
141 ENA_STAT_TX_ENTRY(tx_poll),
142 ENA_STAT_TX_ENTRY(doorbells),
143 ENA_STAT_TX_ENTRY(prepare_ctx_err),
144 ENA_STAT_TX_ENTRY(missing_tx_comp),
145 ENA_STAT_TX_ENTRY(bad_req_id),
148 static const struct ena_stats ena_stats_rx_strings[] = {
149 ENA_STAT_RX_ENTRY(cnt),
150 ENA_STAT_RX_ENTRY(bytes),
151 ENA_STAT_RX_ENTRY(refil_partial),
152 ENA_STAT_RX_ENTRY(bad_csum),
153 ENA_STAT_RX_ENTRY(page_alloc_fail),
154 ENA_STAT_RX_ENTRY(skb_alloc_fail),
155 ENA_STAT_RX_ENTRY(dma_mapping_err),
156 ENA_STAT_RX_ENTRY(bad_desc_num),
157 ENA_STAT_RX_ENTRY(small_copy_len_pkt),
160 static const struct ena_stats ena_stats_ena_com_strings[] = {
161 ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
162 ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
163 ENA_STAT_ENA_COM_ENTRY(completed_cmd),
164 ENA_STAT_ENA_COM_ENTRY(out_of_space),
165 ENA_STAT_ENA_COM_ENTRY(no_completion),
168 #define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
169 #define ENA_STATS_ARRAY_TX ARRAY_SIZE(ena_stats_tx_strings)
170 #define ENA_STATS_ARRAY_RX ARRAY_SIZE(ena_stats_rx_strings)
171 #define ENA_STATS_ARRAY_ENA_COM ARRAY_SIZE(ena_stats_ena_com_strings)
173 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
174 DEV_TX_OFFLOAD_UDP_CKSUM |\
175 DEV_TX_OFFLOAD_IPV4_CKSUM |\
176 DEV_TX_OFFLOAD_TCP_TSO)
177 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
181 /** Vendor ID used by Amazon devices */
182 #define PCI_VENDOR_ID_AMAZON 0x1D0F
183 /** Amazon devices */
184 #define PCI_DEVICE_ID_ENA_VF 0xEC20
185 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
187 #define ENA_TX_OFFLOAD_MASK (\
192 #define ENA_TX_OFFLOAD_NOTSUP_MASK \
193 (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
195 int ena_logtype_init;
196 int ena_logtype_driver;
198 static const struct rte_pci_id pci_id_ena_map[] = {
199 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
200 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
204 static struct ena_aenq_handlers aenq_handlers;
206 static int ena_device_init(struct ena_com_dev *ena_dev,
207 struct ena_com_dev_get_features_ctx *get_feat_ctx);
208 static int ena_dev_configure(struct rte_eth_dev *dev);
209 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
211 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
213 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
214 uint16_t nb_desc, unsigned int socket_id,
215 const struct rte_eth_txconf *tx_conf);
216 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
217 uint16_t nb_desc, unsigned int socket_id,
218 const struct rte_eth_rxconf *rx_conf,
219 struct rte_mempool *mp);
220 static uint16_t eth_ena_recv_pkts(void *rx_queue,
221 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
222 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
223 static void ena_init_rings(struct ena_adapter *adapter);
224 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
225 static int ena_start(struct rte_eth_dev *dev);
226 static void ena_stop(struct rte_eth_dev *dev);
227 static void ena_close(struct rte_eth_dev *dev);
228 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
229 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
230 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
231 static void ena_rx_queue_release(void *queue);
232 static void ena_tx_queue_release(void *queue);
233 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
234 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
235 static int ena_link_update(struct rte_eth_dev *dev,
236 int wait_to_complete);
237 static int ena_queue_restart(struct ena_ring *ring);
238 static int ena_queue_restart_all(struct rte_eth_dev *dev,
239 enum ena_ring_type ring_type);
240 static void ena_stats_restart(struct rte_eth_dev *dev);
241 static void ena_infos_get(struct rte_eth_dev *dev,
242 struct rte_eth_dev_info *dev_info);
243 static int ena_rss_reta_update(struct rte_eth_dev *dev,
244 struct rte_eth_rss_reta_entry64 *reta_conf,
246 static int ena_rss_reta_query(struct rte_eth_dev *dev,
247 struct rte_eth_rss_reta_entry64 *reta_conf,
249 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
250 static void ena_interrupt_handler_rte(void *cb_arg);
252 static const struct eth_dev_ops ena_dev_ops = {
253 .dev_configure = ena_dev_configure,
254 .dev_infos_get = ena_infos_get,
255 .rx_queue_setup = ena_rx_queue_setup,
256 .tx_queue_setup = ena_tx_queue_setup,
257 .dev_start = ena_start,
258 .dev_stop = ena_stop,
259 .link_update = ena_link_update,
260 .stats_get = ena_stats_get,
261 .mtu_set = ena_mtu_set,
262 .rx_queue_release = ena_rx_queue_release,
263 .tx_queue_release = ena_tx_queue_release,
264 .dev_close = ena_close,
265 .reta_update = ena_rss_reta_update,
266 .reta_query = ena_rss_reta_query,
269 #define NUMA_NO_NODE SOCKET_ID_ANY
271 static inline int ena_cpu_to_node(int cpu)
273 struct rte_config *config = rte_eal_get_configuration();
274 struct rte_fbarray *arr = &config->mem_config->memzones;
275 const struct rte_memzone *mz;
277 if (unlikely(cpu >= RTE_MAX_MEMZONE))
280 mz = rte_fbarray_get(arr, cpu);
282 return mz->socket_id;
285 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
286 struct ena_com_rx_ctx *ena_rx_ctx)
288 uint64_t ol_flags = 0;
289 uint32_t packet_type = 0;
291 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
292 packet_type |= RTE_PTYPE_L4_TCP;
293 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
294 packet_type |= RTE_PTYPE_L4_UDP;
296 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
297 packet_type |= RTE_PTYPE_L3_IPV4;
298 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
299 packet_type |= RTE_PTYPE_L3_IPV6;
301 if (unlikely(ena_rx_ctx->l4_csum_err))
302 ol_flags |= PKT_RX_L4_CKSUM_BAD;
303 if (unlikely(ena_rx_ctx->l3_csum_err))
304 ol_flags |= PKT_RX_IP_CKSUM_BAD;
306 mbuf->ol_flags = ol_flags;
307 mbuf->packet_type = packet_type;
310 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
311 struct ena_com_tx_ctx *ena_tx_ctx,
312 uint64_t queue_offloads)
314 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
316 if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
317 (queue_offloads & QUEUE_OFFLOADS)) {
318 /* check if TSO is required */
319 if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
320 (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
321 ena_tx_ctx->tso_enable = true;
323 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
326 /* check if L3 checksum is needed */
327 if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
328 (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
329 ena_tx_ctx->l3_csum_enable = true;
331 if (mbuf->ol_flags & PKT_TX_IPV6) {
332 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
334 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
336 /* set don't fragment (DF) flag */
337 if (mbuf->packet_type &
338 (RTE_PTYPE_L4_NONFRAG
339 | RTE_PTYPE_INNER_L4_NONFRAG))
340 ena_tx_ctx->df = true;
343 /* check if L4 checksum is needed */
344 if ((mbuf->ol_flags & PKT_TX_TCP_CKSUM) &&
345 (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
346 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
347 ena_tx_ctx->l4_csum_enable = true;
348 } else if ((mbuf->ol_flags & PKT_TX_UDP_CKSUM) &&
349 (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
350 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
351 ena_tx_ctx->l4_csum_enable = true;
353 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
354 ena_tx_ctx->l4_csum_enable = false;
357 ena_meta->mss = mbuf->tso_segsz;
358 ena_meta->l3_hdr_len = mbuf->l3_len;
359 ena_meta->l3_hdr_offset = mbuf->l2_len;
361 ena_tx_ctx->meta_valid = true;
363 ena_tx_ctx->meta_valid = false;
367 static void ena_config_host_info(struct ena_com_dev *ena_dev)
369 struct ena_admin_host_info *host_info;
372 /* Allocate only the host info */
373 rc = ena_com_allocate_host_info(ena_dev);
375 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
379 host_info = ena_dev->host_attr.host_info;
381 host_info->os_type = ENA_ADMIN_OS_DPDK;
382 host_info->kernel_ver = RTE_VERSION;
383 snprintf((char *)host_info->kernel_ver_str,
384 sizeof(host_info->kernel_ver_str),
385 "%s", rte_version());
386 host_info->os_dist = RTE_VERSION;
387 snprintf((char *)host_info->os_dist_str,
388 sizeof(host_info->os_dist_str),
389 "%s", rte_version());
390 host_info->driver_version =
391 (DRV_MODULE_VER_MAJOR) |
392 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
393 (DRV_MODULE_VER_SUBMINOR <<
394 ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
396 rc = ena_com_set_host_attributes(ena_dev);
398 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
399 if (rc != -ENA_COM_UNSUPPORTED)
406 ena_com_delete_host_info(ena_dev);
410 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
412 if (sset != ETH_SS_STATS)
415 /* Workaround for clang:
416 * touch internal structures to prevent
419 ENA_TOUCH(ena_stats_global_strings);
420 ENA_TOUCH(ena_stats_tx_strings);
421 ENA_TOUCH(ena_stats_rx_strings);
422 ENA_TOUCH(ena_stats_ena_com_strings);
424 return dev->data->nb_tx_queues *
425 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
426 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
429 static void ena_config_debug_area(struct ena_adapter *adapter)
434 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
436 RTE_LOG(ERR, PMD, "SS count is negative\n");
440 /* allocate 32 bytes for each string and 64bit for the value */
441 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
443 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
445 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
449 rc = ena_com_set_host_attributes(&adapter->ena_dev);
451 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
452 if (rc != -ENA_COM_UNSUPPORTED)
458 ena_com_delete_debug_area(&adapter->ena_dev);
461 static void ena_close(struct rte_eth_dev *dev)
463 struct ena_adapter *adapter =
464 (struct ena_adapter *)(dev->data->dev_private);
467 adapter->state = ENA_ADAPTER_STATE_CLOSED;
469 ena_rx_queue_release_all(dev);
470 ena_tx_queue_release_all(dev);
473 static int ena_rss_reta_update(struct rte_eth_dev *dev,
474 struct rte_eth_rss_reta_entry64 *reta_conf,
477 struct ena_adapter *adapter =
478 (struct ena_adapter *)(dev->data->dev_private);
479 struct ena_com_dev *ena_dev = &adapter->ena_dev;
485 if ((reta_size == 0) || (reta_conf == NULL))
488 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
489 RTE_LOG(WARNING, PMD,
490 "indirection table %d is bigger than supported (%d)\n",
491 reta_size, ENA_RX_RSS_TABLE_SIZE);
496 for (i = 0 ; i < reta_size ; i++) {
497 /* each reta_conf is for 64 entries.
498 * to support 128 we use 2 conf of 64
500 conf_idx = i / RTE_RETA_GROUP_SIZE;
501 idx = i % RTE_RETA_GROUP_SIZE;
502 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
504 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
505 ret = ena_com_indirect_table_fill_entry(ena_dev,
508 if (unlikely(ret && (ret != ENA_COM_UNSUPPORTED))) {
510 "Cannot fill indirect table\n");
517 ret = ena_com_indirect_table_set(ena_dev);
518 if (unlikely(ret && (ret != ENA_COM_UNSUPPORTED))) {
519 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
524 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
525 __func__, reta_size, adapter->rte_dev->data->port_id);
530 /* Query redirection table. */
531 static int ena_rss_reta_query(struct rte_eth_dev *dev,
532 struct rte_eth_rss_reta_entry64 *reta_conf,
535 struct ena_adapter *adapter =
536 (struct ena_adapter *)(dev->data->dev_private);
537 struct ena_com_dev *ena_dev = &adapter->ena_dev;
540 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
544 if (reta_size == 0 || reta_conf == NULL ||
545 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
548 ret = ena_com_indirect_table_get(ena_dev, indirect_table);
549 if (unlikely(ret && (ret != ENA_COM_UNSUPPORTED))) {
550 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
555 for (i = 0 ; i < reta_size ; i++) {
556 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
557 reta_idx = i % RTE_RETA_GROUP_SIZE;
558 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
559 reta_conf[reta_conf_idx].reta[reta_idx] =
560 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
566 static int ena_rss_init_default(struct ena_adapter *adapter)
568 struct ena_com_dev *ena_dev = &adapter->ena_dev;
569 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
573 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
575 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
579 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
580 val = i % nb_rx_queues;
581 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
582 ENA_IO_RXQ_IDX(val));
583 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
584 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
589 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
590 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
591 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
592 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
596 rc = ena_com_set_default_hash_ctrl(ena_dev);
597 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
598 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
602 rc = ena_com_indirect_table_set(ena_dev);
603 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
604 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
607 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
608 adapter->rte_dev->data->port_id);
613 ena_com_rss_destroy(ena_dev);
619 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
621 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
622 int nb_queues = dev->data->nb_rx_queues;
625 for (i = 0; i < nb_queues; i++)
626 ena_rx_queue_release(queues[i]);
629 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
631 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
632 int nb_queues = dev->data->nb_tx_queues;
635 for (i = 0; i < nb_queues; i++)
636 ena_tx_queue_release(queues[i]);
639 static void ena_rx_queue_release(void *queue)
641 struct ena_ring *ring = (struct ena_ring *)queue;
642 struct ena_adapter *adapter = ring->adapter;
645 ena_assert_msg(ring->configured,
646 "API violation - releasing not configured queue");
647 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
650 /* Destroy HW queue */
651 ena_qid = ENA_IO_RXQ_IDX(ring->id);
652 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
655 ena_rx_queue_release_bufs(ring);
657 /* Free ring resources */
658 if (ring->rx_buffer_info)
659 rte_free(ring->rx_buffer_info);
660 ring->rx_buffer_info = NULL;
662 ring->configured = 0;
664 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
665 ring->port_id, ring->id);
668 static void ena_tx_queue_release(void *queue)
670 struct ena_ring *ring = (struct ena_ring *)queue;
671 struct ena_adapter *adapter = ring->adapter;
674 ena_assert_msg(ring->configured,
675 "API violation. Releasing not configured queue");
676 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
679 /* Destroy HW queue */
680 ena_qid = ENA_IO_TXQ_IDX(ring->id);
681 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
684 ena_tx_queue_release_bufs(ring);
686 /* Free ring resources */
687 if (ring->tx_buffer_info)
688 rte_free(ring->tx_buffer_info);
690 if (ring->empty_tx_reqs)
691 rte_free(ring->empty_tx_reqs);
693 ring->empty_tx_reqs = NULL;
694 ring->tx_buffer_info = NULL;
696 ring->configured = 0;
698 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
699 ring->port_id, ring->id);
702 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
704 unsigned int ring_mask = ring->ring_size - 1;
706 while (ring->next_to_clean != ring->next_to_use) {
708 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
711 rte_mbuf_raw_free(m);
713 ring->next_to_clean++;
717 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
721 for (i = 0; i < ring->ring_size; ++i) {
722 struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
725 rte_pktmbuf_free(tx_buf->mbuf);
727 ring->next_to_clean++;
731 static int ena_link_update(struct rte_eth_dev *dev,
732 __rte_unused int wait_to_complete)
734 struct rte_eth_link *link = &dev->data->dev_link;
735 struct ena_adapter *adapter;
737 adapter = (struct ena_adapter *)(dev->data->dev_private);
739 link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
740 link->link_speed = ETH_SPEED_NUM_10G;
741 link->link_duplex = ETH_LINK_FULL_DUPLEX;
746 static int ena_queue_restart_all(struct rte_eth_dev *dev,
747 enum ena_ring_type ring_type)
749 struct ena_adapter *adapter =
750 (struct ena_adapter *)(dev->data->dev_private);
751 struct ena_ring *queues = NULL;
756 if (ring_type == ENA_RING_TYPE_RX) {
757 queues = adapter->rx_ring;
758 nb_queues = dev->data->nb_rx_queues;
760 queues = adapter->tx_ring;
761 nb_queues = dev->data->nb_tx_queues;
763 for (i = 0; i < nb_queues; i++) {
764 if (queues[i].configured) {
765 if (ring_type == ENA_RING_TYPE_RX) {
767 dev->data->rx_queues[i] == &queues[i],
768 "Inconsistent state of rx queues\n");
771 dev->data->tx_queues[i] == &queues[i],
772 "Inconsistent state of tx queues\n");
775 rc = ena_queue_restart(&queues[i]);
779 "failed to restart queue %d type(%d)",
789 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
791 uint32_t max_frame_len = adapter->max_mtu;
793 if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
794 DEV_RX_OFFLOAD_JUMBO_FRAME)
796 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
798 return max_frame_len;
801 static int ena_check_valid_conf(struct ena_adapter *adapter)
803 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
805 if (max_frame_len > adapter->max_mtu) {
806 PMD_INIT_LOG(ERR, "Unsupported MTU of %d", max_frame_len);
814 ena_calc_queue_size(struct ena_com_dev *ena_dev,
815 struct ena_com_dev_get_features_ctx *get_feat_ctx)
817 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
819 queue_size = RTE_MIN(queue_size,
820 get_feat_ctx->max_queues.max_cq_depth);
821 queue_size = RTE_MIN(queue_size,
822 get_feat_ctx->max_queues.max_sq_depth);
824 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
825 queue_size = RTE_MIN(queue_size,
826 get_feat_ctx->max_queues.max_llq_depth);
828 /* Round down to power of 2 */
829 if (!rte_is_power_of_2(queue_size))
830 queue_size = rte_align32pow2(queue_size >> 1);
832 if (queue_size == 0) {
833 PMD_INIT_LOG(ERR, "Invalid queue size");
840 static void ena_stats_restart(struct rte_eth_dev *dev)
842 struct ena_adapter *adapter =
843 (struct ena_adapter *)(dev->data->dev_private);
845 rte_atomic64_init(&adapter->drv_stats->ierrors);
846 rte_atomic64_init(&adapter->drv_stats->oerrors);
847 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
850 static int ena_stats_get(struct rte_eth_dev *dev,
851 struct rte_eth_stats *stats)
853 struct ena_admin_basic_stats ena_stats;
854 struct ena_adapter *adapter =
855 (struct ena_adapter *)(dev->data->dev_private);
856 struct ena_com_dev *ena_dev = &adapter->ena_dev;
859 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
862 memset(&ena_stats, 0, sizeof(ena_stats));
863 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
865 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
869 /* Set of basic statistics from ENA */
870 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
871 ena_stats.rx_pkts_low);
872 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
873 ena_stats.tx_pkts_low);
874 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
875 ena_stats.rx_bytes_low);
876 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
877 ena_stats.tx_bytes_low);
878 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
879 ena_stats.rx_drops_low);
881 /* Driver related stats */
882 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
883 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
884 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
888 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
890 struct ena_adapter *adapter;
891 struct ena_com_dev *ena_dev;
894 ena_assert_msg(dev->data != NULL, "Uninitialized device");
895 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
896 adapter = (struct ena_adapter *)(dev->data->dev_private);
898 ena_dev = &adapter->ena_dev;
899 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
901 if (mtu > ena_get_mtu_conf(adapter)) {
903 "Given MTU (%d) exceeds maximum MTU supported (%d)\n",
904 mtu, ena_get_mtu_conf(adapter));
909 rc = ena_com_set_dev_mtu(ena_dev, mtu);
911 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
913 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
919 static int ena_start(struct rte_eth_dev *dev)
921 struct ena_adapter *adapter =
922 (struct ena_adapter *)(dev->data->dev_private);
925 rc = ena_check_valid_conf(adapter);
929 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
933 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
937 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
938 ETH_MQ_RX_RSS_FLAG) {
939 rc = ena_rss_init_default(adapter);
944 ena_stats_restart(dev);
946 adapter->state = ENA_ADAPTER_STATE_RUNNING;
951 static void ena_stop(struct rte_eth_dev *dev)
953 struct ena_adapter *adapter =
954 (struct ena_adapter *)(dev->data->dev_private);
956 adapter->state = ENA_ADAPTER_STATE_STOPPED;
959 static int ena_queue_restart(struct ena_ring *ring)
963 ena_assert_msg(ring->configured == 1,
964 "Trying to restart unconfigured queue\n");
966 ring->next_to_clean = 0;
967 ring->next_to_use = 0;
969 if (ring->type == ENA_RING_TYPE_TX)
972 bufs_num = ring->ring_size - 1;
973 rc = ena_populate_rx_queue(ring, bufs_num);
974 if (rc != bufs_num) {
975 PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
982 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
985 __rte_unused unsigned int socket_id,
986 const struct rte_eth_txconf *tx_conf)
988 struct ena_com_create_io_ctx ctx =
989 /* policy set to _HOST just to satisfy icc compiler */
990 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
991 ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
992 struct ena_ring *txq = NULL;
993 struct ena_adapter *adapter =
994 (struct ena_adapter *)(dev->data->dev_private);
998 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1000 txq = &adapter->tx_ring[queue_idx];
1002 if (txq->configured) {
1004 "API violation. Queue %d is already configured\n",
1009 if (!rte_is_power_of_2(nb_desc)) {
1011 "Unsupported size of RX queue: %d is not a power of 2.",
1016 if (nb_desc > adapter->tx_ring_size) {
1018 "Unsupported size of TX queue (max size: %d)\n",
1019 adapter->tx_ring_size);
1023 ena_qid = ENA_IO_TXQ_IDX(queue_idx);
1025 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1027 ctx.msix_vector = -1; /* admin interrupts not used */
1028 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1029 ctx.queue_size = adapter->tx_ring_size;
1030 ctx.numa_node = ena_cpu_to_node(queue_idx);
1032 rc = ena_com_create_io_queue(ena_dev, &ctx);
1035 "failed to create io TX queue #%d (qid:%d) rc: %d\n",
1036 queue_idx, ena_qid, rc);
1038 txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1039 txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1041 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1042 &txq->ena_com_io_sq,
1043 &txq->ena_com_io_cq);
1046 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1048 ena_com_destroy_io_queue(ena_dev, ena_qid);
1052 txq->port_id = dev->data->port_id;
1053 txq->next_to_clean = 0;
1054 txq->next_to_use = 0;
1055 txq->ring_size = nb_desc;
1057 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1058 sizeof(struct ena_tx_buffer) *
1060 RTE_CACHE_LINE_SIZE);
1061 if (!txq->tx_buffer_info) {
1062 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1066 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1067 sizeof(u16) * txq->ring_size,
1068 RTE_CACHE_LINE_SIZE);
1069 if (!txq->empty_tx_reqs) {
1070 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1071 rte_free(txq->tx_buffer_info);
1074 for (i = 0; i < txq->ring_size; i++)
1075 txq->empty_tx_reqs[i] = i;
1077 txq->offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1079 /* Store pointer to this queue in upper layer */
1080 txq->configured = 1;
1081 dev->data->tx_queues[queue_idx] = txq;
1086 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1089 __rte_unused unsigned int socket_id,
1090 __rte_unused const struct rte_eth_rxconf *rx_conf,
1091 struct rte_mempool *mp)
1093 struct ena_com_create_io_ctx ctx =
1094 /* policy set to _HOST just to satisfy icc compiler */
1095 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1096 ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1097 struct ena_adapter *adapter =
1098 (struct ena_adapter *)(dev->data->dev_private);
1099 struct ena_ring *rxq = NULL;
1100 uint16_t ena_qid = 0;
1102 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1104 rxq = &adapter->rx_ring[queue_idx];
1105 if (rxq->configured) {
1107 "API violation. Queue %d is already configured\n",
1112 if (!rte_is_power_of_2(nb_desc)) {
1114 "Unsupported size of TX queue: %d is not a power of 2.",
1119 if (nb_desc > adapter->rx_ring_size) {
1121 "Unsupported size of RX queue (max size: %d)\n",
1122 adapter->rx_ring_size);
1126 ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1129 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1130 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1131 ctx.msix_vector = -1; /* admin interrupts not used */
1132 ctx.queue_size = adapter->rx_ring_size;
1133 ctx.numa_node = ena_cpu_to_node(queue_idx);
1135 rc = ena_com_create_io_queue(ena_dev, &ctx);
1137 RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1140 rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1141 rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1143 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1144 &rxq->ena_com_io_sq,
1145 &rxq->ena_com_io_cq);
1148 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1150 ena_com_destroy_io_queue(ena_dev, ena_qid);
1153 rxq->port_id = dev->data->port_id;
1154 rxq->next_to_clean = 0;
1155 rxq->next_to_use = 0;
1156 rxq->ring_size = nb_desc;
1159 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1160 sizeof(struct rte_mbuf *) * nb_desc,
1161 RTE_CACHE_LINE_SIZE);
1162 if (!rxq->rx_buffer_info) {
1163 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1167 /* Store pointer to this queue in upper layer */
1168 rxq->configured = 1;
1169 dev->data->rx_queues[queue_idx] = rxq;
1174 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1178 uint16_t ring_size = rxq->ring_size;
1179 uint16_t ring_mask = ring_size - 1;
1180 uint16_t next_to_use = rxq->next_to_use;
1182 struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1184 if (unlikely(!count))
1187 in_use = rxq->next_to_use - rxq->next_to_clean;
1188 ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1190 count = RTE_MIN(count,
1191 (uint16_t)(ring_size - (next_to_use & ring_mask)));
1193 /* get resources for incoming packets */
1194 rc = rte_mempool_get_bulk(rxq->mb_pool,
1195 (void **)(&mbufs[next_to_use & ring_mask]),
1197 if (unlikely(rc < 0)) {
1198 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1199 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1203 for (i = 0; i < count; i++) {
1204 uint16_t next_to_use_masked = next_to_use & ring_mask;
1205 struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1206 struct ena_com_buf ebuf;
1208 rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1209 /* prepare physical address for DMA transaction */
1210 ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1211 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1212 /* pass resource to device */
1213 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1214 &ebuf, next_to_use_masked);
1216 rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbuf),
1218 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1224 /* When we submitted free recources to device... */
1226 /* ...let HW know that it can fill buffers with data */
1228 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1230 rxq->next_to_use = next_to_use;
1236 static int ena_device_init(struct ena_com_dev *ena_dev,
1237 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1239 uint32_t aenq_groups;
1241 bool readless_supported;
1243 /* Initialize mmio registers */
1244 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1246 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1250 /* The PCIe configuration space revision id indicate if mmio reg
1253 readless_supported =
1254 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1255 & ENA_MMIO_DISABLE_REG_READ);
1256 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1259 rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1261 RTE_LOG(ERR, PMD, "cannot reset device\n");
1262 goto err_mmio_read_less;
1265 /* check FW version */
1266 rc = ena_com_validate_version(ena_dev);
1268 RTE_LOG(ERR, PMD, "device version is too low\n");
1269 goto err_mmio_read_less;
1272 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1274 /* ENA device administration layer init */
1275 rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
1278 "cannot initialize ena admin queue with device\n");
1279 goto err_mmio_read_less;
1282 /* To enable the msix interrupts the driver needs to know the number
1283 * of queues. So the driver uses polling mode to retrieve this
1286 ena_com_set_admin_polling_mode(ena_dev, true);
1288 ena_config_host_info(ena_dev);
1290 /* Get Device Attributes and features */
1291 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1294 "cannot get attribute for ena device rc= %d\n", rc);
1295 goto err_admin_init;
1298 aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1299 BIT(ENA_ADMIN_NOTIFICATION);
1301 aenq_groups &= get_feat_ctx->aenq.supported_groups;
1302 rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1304 RTE_LOG(ERR, PMD, "Cannot configure aenq groups rc: %d\n", rc);
1305 goto err_admin_init;
1311 ena_com_admin_destroy(ena_dev);
1314 ena_com_mmio_reg_read_request_destroy(ena_dev);
1319 static void ena_interrupt_handler_rte(void *cb_arg)
1321 struct ena_adapter *adapter = (struct ena_adapter *)cb_arg;
1322 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1324 ena_com_admin_q_comp_intr_handler(ena_dev);
1325 if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1326 ena_com_aenq_intr_handler(ena_dev, adapter);
1329 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1331 struct rte_pci_device *pci_dev;
1332 struct rte_intr_handle *intr_handle;
1333 struct ena_adapter *adapter =
1334 (struct ena_adapter *)(eth_dev->data->dev_private);
1335 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1336 struct ena_com_dev_get_features_ctx get_feat_ctx;
1339 static int adapters_found;
1341 memset(adapter, 0, sizeof(struct ena_adapter));
1342 ena_dev = &adapter->ena_dev;
1344 eth_dev->dev_ops = &ena_dev_ops;
1345 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1346 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1347 eth_dev->tx_pkt_prepare = ð_ena_prep_pkts;
1348 adapter->rte_eth_dev_data = eth_dev->data;
1349 adapter->rte_dev = eth_dev;
1351 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1354 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1355 adapter->pdev = pci_dev;
1357 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1358 pci_dev->addr.domain,
1360 pci_dev->addr.devid,
1361 pci_dev->addr.function);
1363 intr_handle = &pci_dev->intr_handle;
1365 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1366 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1368 if (!adapter->regs) {
1369 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1374 ena_dev->reg_bar = adapter->regs;
1375 ena_dev->dmadev = adapter->pdev;
1377 adapter->id_number = adapters_found;
1379 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1380 adapter->id_number);
1382 /* device specific initialization routine */
1383 rc = ena_device_init(ena_dev, &get_feat_ctx);
1385 PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1389 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1390 adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1392 queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1393 if ((queue_size <= 0) || (adapter->num_queues <= 0))
1396 adapter->tx_ring_size = queue_size;
1397 adapter->rx_ring_size = queue_size;
1399 /* prepare ring structures */
1400 ena_init_rings(adapter);
1402 ena_config_debug_area(adapter);
1404 /* Set max MTU for this device */
1405 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1407 /* set device support for TSO */
1408 adapter->tso4_supported = get_feat_ctx.offload.tx &
1409 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1411 /* Copy MAC address and point DPDK to it */
1412 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1413 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1414 (struct ether_addr *)adapter->mac_addr);
1416 adapter->drv_stats = rte_zmalloc("adapter stats",
1417 sizeof(*adapter->drv_stats),
1418 RTE_CACHE_LINE_SIZE);
1419 if (!adapter->drv_stats) {
1420 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1424 rte_intr_callback_register(intr_handle,
1425 ena_interrupt_handler_rte,
1427 rte_intr_enable(intr_handle);
1428 ena_com_set_admin_polling_mode(ena_dev, false);
1429 ena_com_admin_aenq_enable(ena_dev);
1432 adapter->state = ENA_ADAPTER_STATE_INIT;
1437 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1439 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1440 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1441 struct ena_adapter *adapter =
1442 (struct ena_adapter *)(eth_dev->data->dev_private);
1444 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1447 if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1450 eth_dev->dev_ops = NULL;
1451 eth_dev->rx_pkt_burst = NULL;
1452 eth_dev->tx_pkt_burst = NULL;
1453 eth_dev->tx_pkt_prepare = NULL;
1455 rte_free(adapter->drv_stats);
1456 adapter->drv_stats = NULL;
1458 rte_intr_disable(intr_handle);
1459 rte_intr_callback_unregister(intr_handle,
1460 ena_interrupt_handler_rte,
1463 adapter->state = ENA_ADAPTER_STATE_FREE;
1468 static int ena_dev_configure(struct rte_eth_dev *dev)
1470 struct ena_adapter *adapter =
1471 (struct ena_adapter *)(dev->data->dev_private);
1473 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1475 adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1476 adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1480 static void ena_init_rings(struct ena_adapter *adapter)
1484 for (i = 0; i < adapter->num_queues; i++) {
1485 struct ena_ring *ring = &adapter->tx_ring[i];
1487 ring->configured = 0;
1488 ring->type = ENA_RING_TYPE_TX;
1489 ring->adapter = adapter;
1491 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1492 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1495 for (i = 0; i < adapter->num_queues; i++) {
1496 struct ena_ring *ring = &adapter->rx_ring[i];
1498 ring->configured = 0;
1499 ring->type = ENA_RING_TYPE_RX;
1500 ring->adapter = adapter;
1505 static void ena_infos_get(struct rte_eth_dev *dev,
1506 struct rte_eth_dev_info *dev_info)
1508 struct ena_adapter *adapter;
1509 struct ena_com_dev *ena_dev;
1510 struct ena_com_dev_get_features_ctx feat;
1511 uint64_t rx_feat = 0, tx_feat = 0;
1514 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1515 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1516 adapter = (struct ena_adapter *)(dev->data->dev_private);
1518 ena_dev = &adapter->ena_dev;
1519 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1521 dev_info->speed_capa =
1523 ETH_LINK_SPEED_2_5G |
1525 ETH_LINK_SPEED_10G |
1526 ETH_LINK_SPEED_25G |
1527 ETH_LINK_SPEED_40G |
1528 ETH_LINK_SPEED_50G |
1529 ETH_LINK_SPEED_100G;
1531 /* Get supported features from HW */
1532 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1535 "Cannot get attribute for ena device rc= %d\n", rc);
1539 /* Set Tx & Rx features available for device */
1540 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1541 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1543 if (feat.offload.tx &
1544 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1545 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1546 DEV_TX_OFFLOAD_UDP_CKSUM |
1547 DEV_TX_OFFLOAD_TCP_CKSUM;
1549 if (feat.offload.rx_supported &
1550 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1551 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1552 DEV_RX_OFFLOAD_UDP_CKSUM |
1553 DEV_RX_OFFLOAD_TCP_CKSUM;
1555 rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1557 /* Inform framework about available features */
1558 dev_info->rx_offload_capa = rx_feat;
1559 dev_info->rx_queue_offload_capa = rx_feat;
1560 dev_info->tx_offload_capa = tx_feat;
1561 dev_info->tx_queue_offload_capa = tx_feat;
1563 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1564 dev_info->max_rx_pktlen = adapter->max_mtu;
1565 dev_info->max_mac_addrs = 1;
1567 dev_info->max_rx_queues = adapter->num_queues;
1568 dev_info->max_tx_queues = adapter->num_queues;
1569 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1571 adapter->tx_supported_offloads = tx_feat;
1572 adapter->rx_supported_offloads = rx_feat;
1575 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1578 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1579 unsigned int ring_size = rx_ring->ring_size;
1580 unsigned int ring_mask = ring_size - 1;
1581 uint16_t next_to_clean = rx_ring->next_to_clean;
1582 uint16_t desc_in_use = 0;
1583 unsigned int recv_idx = 0;
1584 struct rte_mbuf *mbuf = NULL;
1585 struct rte_mbuf *mbuf_head = NULL;
1586 struct rte_mbuf *mbuf_prev = NULL;
1587 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1588 unsigned int completed;
1590 struct ena_com_rx_ctx ena_rx_ctx;
1593 /* Check adapter state */
1594 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1596 "Trying to receive pkts while device is NOT running\n");
1600 desc_in_use = rx_ring->next_to_use - next_to_clean;
1601 if (unlikely(nb_pkts > desc_in_use))
1602 nb_pkts = desc_in_use;
1604 for (completed = 0; completed < nb_pkts; completed++) {
1607 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1608 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1609 ena_rx_ctx.descs = 0;
1610 /* receive packet context */
1611 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1612 rx_ring->ena_com_io_sq,
1615 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1619 if (unlikely(ena_rx_ctx.descs == 0))
1622 while (segments < ena_rx_ctx.descs) {
1623 mbuf = rx_buff_info[next_to_clean & ring_mask];
1624 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1625 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1628 if (segments == 0) {
1629 mbuf->nb_segs = ena_rx_ctx.descs;
1630 mbuf->port = rx_ring->port_id;
1634 /* for multi-segment pkts create mbuf chain */
1635 mbuf_prev->next = mbuf;
1637 mbuf_head->pkt_len += mbuf->data_len;
1644 /* fill mbuf attributes if any */
1645 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1646 mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1648 /* pass to DPDK application head mbuf */
1649 rx_pkts[recv_idx] = mbuf_head;
1653 rx_ring->next_to_clean = next_to_clean;
1655 desc_in_use = desc_in_use - completed + 1;
1656 /* Burst refill to save doorbells, memory barriers, const interval */
1657 if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1658 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1664 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1670 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1671 struct ipv4_hdr *ip_hdr;
1673 uint16_t frag_field;
1675 for (i = 0; i != nb_pkts; i++) {
1677 ol_flags = m->ol_flags;
1679 if (!(ol_flags & PKT_TX_IPV4))
1682 /* If there was not L2 header length specified, assume it is
1683 * length of the ethernet header.
1685 if (unlikely(m->l2_len == 0))
1686 m->l2_len = sizeof(struct ether_hdr);
1688 ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1690 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1692 if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1693 m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1695 /* If IPv4 header has DF flag enabled and TSO support is
1696 * disabled, partial chcecksum should not be calculated.
1698 if (!tx_ring->adapter->tso4_supported)
1702 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
1703 (ol_flags & PKT_TX_L4_MASK) ==
1704 PKT_TX_SCTP_CKSUM) {
1705 rte_errno = -ENOTSUP;
1709 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
1710 ret = rte_validate_tx_offload(m);
1717 /* In case we are supposed to TSO and have DF not set (DF=0)
1718 * hardware must be provided with partial checksum, otherwise
1719 * it will take care of necessary calculations.
1722 ret = rte_net_intel_cksum_flags_prepare(m,
1723 ol_flags & ~PKT_TX_TCP_SEG);
1733 static void ena_update_hints(struct ena_adapter *adapter,
1734 struct ena_admin_ena_hw_hints *hints)
1736 if (hints->admin_completion_tx_timeout)
1737 adapter->ena_dev.admin_queue.completion_timeout =
1738 hints->admin_completion_tx_timeout * 1000;
1740 if (hints->mmio_read_timeout)
1741 /* convert to usec */
1742 adapter->ena_dev.mmio_read.reg_read_to =
1743 hints->mmio_read_timeout * 1000;
1746 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1749 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1750 uint16_t next_to_use = tx_ring->next_to_use;
1751 uint16_t next_to_clean = tx_ring->next_to_clean;
1752 struct rte_mbuf *mbuf;
1753 unsigned int ring_size = tx_ring->ring_size;
1754 unsigned int ring_mask = ring_size - 1;
1755 struct ena_com_tx_ctx ena_tx_ctx;
1756 struct ena_tx_buffer *tx_info;
1757 struct ena_com_buf *ebuf;
1758 uint16_t rc, req_id, total_tx_descs = 0;
1759 uint16_t sent_idx = 0, empty_tx_reqs;
1762 /* Check adapter state */
1763 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1765 "Trying to xmit pkts while device is NOT running\n");
1769 empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
1770 if (nb_pkts > empty_tx_reqs)
1771 nb_pkts = empty_tx_reqs;
1773 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1774 mbuf = tx_pkts[sent_idx];
1776 req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1777 tx_info = &tx_ring->tx_buffer_info[req_id];
1778 tx_info->mbuf = mbuf;
1779 tx_info->num_of_bufs = 0;
1780 ebuf = tx_info->bufs;
1782 /* Prepare TX context */
1783 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1784 memset(&ena_tx_ctx.ena_meta, 0x0,
1785 sizeof(struct ena_com_tx_meta));
1786 ena_tx_ctx.ena_bufs = ebuf;
1787 ena_tx_ctx.req_id = req_id;
1788 if (tx_ring->tx_mem_queue_type ==
1789 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1790 /* prepare the push buffer with
1791 * virtual address of the data
1793 ena_tx_ctx.header_len =
1794 RTE_MIN(mbuf->data_len,
1795 tx_ring->tx_max_header_size);
1796 ena_tx_ctx.push_header =
1797 (void *)((char *)mbuf->buf_addr +
1799 } /* there's no else as we take advantage of memset zeroing */
1801 /* Set TX offloads flags, if applicable */
1802 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads);
1804 if (unlikely(mbuf->ol_flags &
1805 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1806 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1808 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1810 /* Process first segment taking into
1811 * consideration pushed header
1813 if (mbuf->data_len > ena_tx_ctx.header_len) {
1814 ebuf->paddr = mbuf->buf_iova +
1816 ena_tx_ctx.header_len;
1817 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1819 tx_info->num_of_bufs++;
1822 while ((mbuf = mbuf->next) != NULL) {
1823 ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
1824 ebuf->len = mbuf->data_len;
1826 tx_info->num_of_bufs++;
1829 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1831 /* Write data to device */
1832 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1833 &ena_tx_ctx, &nb_hw_desc);
1837 tx_info->tx_descs = nb_hw_desc;
1842 /* If there are ready packets to be xmitted... */
1844 /* ...let HW do its best :-) */
1846 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1848 tx_ring->next_to_use = next_to_use;
1851 /* Clear complete packets */
1852 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1853 /* Get Tx info & store how many descs were processed */
1854 tx_info = &tx_ring->tx_buffer_info[req_id];
1855 total_tx_descs += tx_info->tx_descs;
1857 /* Free whole mbuf chain */
1858 mbuf = tx_info->mbuf;
1859 rte_pktmbuf_free(mbuf);
1860 tx_info->mbuf = NULL;
1862 /* Put back descriptor to the ring for reuse */
1863 tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1866 /* If too many descs to clean, leave it for another run */
1867 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1871 if (total_tx_descs > 0) {
1872 /* acknowledge completion of sent packets */
1873 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1874 tx_ring->next_to_clean = next_to_clean;
1880 /*********************************************************************
1882 *********************************************************************/
1883 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1884 struct rte_pci_device *pci_dev)
1886 return rte_eth_dev_pci_generic_probe(pci_dev,
1887 sizeof(struct ena_adapter), eth_ena_dev_init);
1890 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
1892 return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
1895 static struct rte_pci_driver rte_ena_pmd = {
1896 .id_table = pci_id_ena_map,
1897 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1898 .probe = eth_ena_pci_probe,
1899 .remove = eth_ena_pci_remove,
1902 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
1903 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
1904 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
1906 RTE_INIT(ena_init_log);
1910 ena_logtype_init = rte_log_register("pmd.net.ena.init");
1911 if (ena_logtype_init >= 0)
1912 rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
1913 ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
1914 if (ena_logtype_driver >= 0)
1915 rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
1918 /******************************************************************************
1919 ******************************** AENQ Handlers *******************************
1920 *****************************************************************************/
1921 static void ena_update_on_link_change(void *adapter_data,
1922 struct ena_admin_aenq_entry *aenq_e)
1924 struct rte_eth_dev *eth_dev;
1925 struct ena_adapter *adapter;
1926 struct ena_admin_aenq_link_change_desc *aenq_link_desc;
1929 adapter = (struct ena_adapter *)adapter_data;
1930 aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
1931 eth_dev = adapter->rte_dev;
1933 status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
1934 adapter->link_status = status;
1936 ena_link_update(eth_dev, 0);
1937 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
1940 static void ena_notification(void *data,
1941 struct ena_admin_aenq_entry *aenq_e)
1943 struct ena_adapter *adapter = (struct ena_adapter *)data;
1944 struct ena_admin_ena_hw_hints *hints;
1946 if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
1947 RTE_LOG(WARNING, PMD, "Invalid group(%x) expected %x\n",
1948 aenq_e->aenq_common_desc.group,
1949 ENA_ADMIN_NOTIFICATION);
1951 switch (aenq_e->aenq_common_desc.syndrom) {
1952 case ENA_ADMIN_UPDATE_HINTS:
1953 hints = (struct ena_admin_ena_hw_hints *)
1954 (&aenq_e->inline_data_w4);
1955 ena_update_hints(adapter, hints);
1958 RTE_LOG(ERR, PMD, "Invalid aenq notification link state %d\n",
1959 aenq_e->aenq_common_desc.syndrom);
1964 * This handler will called for unknown event group or unimplemented handlers
1966 static void unimplemented_aenq_handler(__rte_unused void *data,
1967 __rte_unused struct ena_admin_aenq_entry *aenq_e)
1969 // Unimplemented handler
1972 static struct ena_aenq_handlers aenq_handlers = {
1974 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
1975 [ENA_ADMIN_NOTIFICATION] = ena_notification,
1976 [ENA_ADMIN_KEEP_ALIVE] = unimplemented_aenq_handler
1978 .unimplemented_handler = unimplemented_aenq_handler