4 * Copyright (c) 2015-2016 Amazon.com, Inc. or its affiliates.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of copyright holder nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <rte_ether.h>
35 #include <rte_ethdev.h>
37 #include <rte_atomic.h>
39 #include <rte_errno.h>
40 #include <rte_version.h>
41 #include <rte_eal_memconfig.h>
43 #include "ena_ethdev.h"
45 #include "ena_platform.h"
47 #include "ena_eth_com.h"
49 #include <ena_common_defs.h>
50 #include <ena_regs_defs.h>
51 #include <ena_admin_defs.h>
52 #include <ena_eth_io_defs.h>
54 #define DRV_MODULE_VER_MAJOR 1
55 #define DRV_MODULE_VER_MINOR 0
56 #define DRV_MODULE_VER_SUBMINOR 0
58 #define ENA_IO_TXQ_IDX(q) (2 * (q))
59 #define ENA_IO_RXQ_IDX(q) (2 * (q) + 1)
60 /*reverse version of ENA_IO_RXQ_IDX*/
61 #define ENA_IO_RXQ_IDX_REV(q) ((q - 1) / 2)
63 /* While processing submitted and completed descriptors (rx and tx path
64 * respectively) in a loop it is desired to:
65 * - perform batch submissions while populating sumbissmion queue
66 * - avoid blocking transmission of other packets during cleanup phase
67 * Hence the utilization ratio of 1/8 of a queue size.
69 #define ENA_RING_DESCS_RATIO(ring_size) (ring_size / 8)
71 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
72 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
74 #define GET_L4_HDR_LEN(mbuf) \
75 ((rte_pktmbuf_mtod_offset(mbuf, struct tcp_hdr *, \
76 mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
78 #define ENA_RX_RSS_TABLE_LOG_SIZE 7
79 #define ENA_RX_RSS_TABLE_SIZE (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
80 #define ENA_HASH_KEY_SIZE 40
81 #define ENA_ETH_SS_STATS 0xFF
82 #define ETH_GSTRING_LEN 32
84 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
86 enum ethtool_stringset {
92 char name[ETH_GSTRING_LEN];
96 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
98 .stat_offset = offsetof(struct ena_com_stats_admin, stat) \
101 #define ENA_STAT_ENTRY(stat, stat_type) { \
103 .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
106 #define ENA_STAT_RX_ENTRY(stat) \
107 ENA_STAT_ENTRY(stat, rx)
109 #define ENA_STAT_TX_ENTRY(stat) \
110 ENA_STAT_ENTRY(stat, tx)
112 #define ENA_STAT_GLOBAL_ENTRY(stat) \
113 ENA_STAT_ENTRY(stat, dev)
115 static const struct ena_stats ena_stats_global_strings[] = {
116 ENA_STAT_GLOBAL_ENTRY(tx_timeout),
117 ENA_STAT_GLOBAL_ENTRY(io_suspend),
118 ENA_STAT_GLOBAL_ENTRY(io_resume),
119 ENA_STAT_GLOBAL_ENTRY(wd_expired),
120 ENA_STAT_GLOBAL_ENTRY(interface_up),
121 ENA_STAT_GLOBAL_ENTRY(interface_down),
122 ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
125 static const struct ena_stats ena_stats_tx_strings[] = {
126 ENA_STAT_TX_ENTRY(cnt),
127 ENA_STAT_TX_ENTRY(bytes),
128 ENA_STAT_TX_ENTRY(queue_stop),
129 ENA_STAT_TX_ENTRY(queue_wakeup),
130 ENA_STAT_TX_ENTRY(dma_mapping_err),
131 ENA_STAT_TX_ENTRY(linearize),
132 ENA_STAT_TX_ENTRY(linearize_failed),
133 ENA_STAT_TX_ENTRY(tx_poll),
134 ENA_STAT_TX_ENTRY(doorbells),
135 ENA_STAT_TX_ENTRY(prepare_ctx_err),
136 ENA_STAT_TX_ENTRY(missing_tx_comp),
137 ENA_STAT_TX_ENTRY(bad_req_id),
140 static const struct ena_stats ena_stats_rx_strings[] = {
141 ENA_STAT_RX_ENTRY(cnt),
142 ENA_STAT_RX_ENTRY(bytes),
143 ENA_STAT_RX_ENTRY(refil_partial),
144 ENA_STAT_RX_ENTRY(bad_csum),
145 ENA_STAT_RX_ENTRY(page_alloc_fail),
146 ENA_STAT_RX_ENTRY(skb_alloc_fail),
147 ENA_STAT_RX_ENTRY(dma_mapping_err),
148 ENA_STAT_RX_ENTRY(bad_desc_num),
149 ENA_STAT_RX_ENTRY(small_copy_len_pkt),
152 static const struct ena_stats ena_stats_ena_com_strings[] = {
153 ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
154 ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
155 ENA_STAT_ENA_COM_ENTRY(completed_cmd),
156 ENA_STAT_ENA_COM_ENTRY(out_of_space),
157 ENA_STAT_ENA_COM_ENTRY(no_completion),
160 #define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
161 #define ENA_STATS_ARRAY_TX ARRAY_SIZE(ena_stats_tx_strings)
162 #define ENA_STATS_ARRAY_RX ARRAY_SIZE(ena_stats_rx_strings)
163 #define ENA_STATS_ARRAY_ENA_COM ARRAY_SIZE(ena_stats_ena_com_strings)
165 /** Vendor ID used by Amazon devices */
166 #define PCI_VENDOR_ID_AMAZON 0x1D0F
167 /** Amazon devices */
168 #define PCI_DEVICE_ID_ENA_VF 0xEC20
169 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
171 static struct rte_pci_id pci_id_ena_map[] = {
172 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
173 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
177 static int ena_device_init(struct ena_com_dev *ena_dev,
178 struct ena_com_dev_get_features_ctx *get_feat_ctx);
179 static int ena_dev_configure(struct rte_eth_dev *dev);
180 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
182 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
183 uint16_t nb_desc, unsigned int socket_id,
184 const struct rte_eth_txconf *tx_conf);
185 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
186 uint16_t nb_desc, unsigned int socket_id,
187 const struct rte_eth_rxconf *rx_conf,
188 struct rte_mempool *mp);
189 static uint16_t eth_ena_recv_pkts(void *rx_queue,
190 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
191 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
192 static void ena_init_rings(struct ena_adapter *adapter);
193 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
194 static int ena_start(struct rte_eth_dev *dev);
195 static void ena_close(struct rte_eth_dev *dev);
196 static void ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
197 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
198 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
199 static void ena_rx_queue_release(void *queue);
200 static void ena_tx_queue_release(void *queue);
201 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
202 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
203 static int ena_link_update(struct rte_eth_dev *dev,
204 __rte_unused int wait_to_complete);
205 static int ena_queue_restart(struct ena_ring *ring);
206 static int ena_queue_restart_all(struct rte_eth_dev *dev,
207 enum ena_ring_type ring_type);
208 static void ena_stats_restart(struct rte_eth_dev *dev);
209 static void ena_infos_get(__rte_unused struct rte_eth_dev *dev,
210 struct rte_eth_dev_info *dev_info);
211 static int ena_rss_reta_update(struct rte_eth_dev *dev,
212 struct rte_eth_rss_reta_entry64 *reta_conf,
214 static int ena_rss_reta_query(struct rte_eth_dev *dev,
215 struct rte_eth_rss_reta_entry64 *reta_conf,
217 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
219 static struct eth_dev_ops ena_dev_ops = {
220 .dev_configure = ena_dev_configure,
221 .dev_infos_get = ena_infos_get,
222 .rx_queue_setup = ena_rx_queue_setup,
223 .tx_queue_setup = ena_tx_queue_setup,
224 .dev_start = ena_start,
225 .link_update = ena_link_update,
226 .stats_get = ena_stats_get,
227 .mtu_set = ena_mtu_set,
228 .rx_queue_release = ena_rx_queue_release,
229 .tx_queue_release = ena_tx_queue_release,
230 .dev_close = ena_close,
231 .reta_update = ena_rss_reta_update,
232 .reta_query = ena_rss_reta_query,
235 #define NUMA_NO_NODE SOCKET_ID_ANY
237 static inline int ena_cpu_to_node(int cpu)
239 struct rte_config *config = rte_eal_get_configuration();
241 if (likely(cpu < RTE_MAX_MEMZONE))
242 return config->mem_config->memzone[cpu].socket_id;
247 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
248 struct ena_com_rx_ctx *ena_rx_ctx)
250 uint64_t ol_flags = 0;
252 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
253 ol_flags |= PKT_TX_TCP_CKSUM;
254 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
255 ol_flags |= PKT_TX_UDP_CKSUM;
257 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
258 ol_flags |= PKT_TX_IPV4;
259 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
260 ol_flags |= PKT_TX_IPV6;
262 if (unlikely(ena_rx_ctx->l4_csum_err))
263 ol_flags |= PKT_RX_L4_CKSUM_BAD;
264 if (unlikely(ena_rx_ctx->l3_csum_err))
265 ol_flags |= PKT_RX_IP_CKSUM_BAD;
267 mbuf->ol_flags = ol_flags;
270 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
271 struct ena_com_tx_ctx *ena_tx_ctx)
273 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
276 (PKT_TX_L4_MASK | PKT_TX_IP_CKSUM | PKT_TX_TCP_SEG)) {
277 /* check if TSO is required */
278 if (mbuf->ol_flags & PKT_TX_TCP_SEG) {
279 ena_tx_ctx->tso_enable = true;
281 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
284 /* check if L3 checksum is needed */
285 if (mbuf->ol_flags & PKT_TX_IP_CKSUM)
286 ena_tx_ctx->l3_csum_enable = true;
288 if (mbuf->ol_flags & PKT_TX_IPV6) {
289 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
291 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
293 /* set don't fragment (DF) flag */
294 if (mbuf->packet_type &
295 (RTE_PTYPE_L4_NONFRAG
296 | RTE_PTYPE_INNER_L4_NONFRAG))
297 ena_tx_ctx->df = true;
300 /* check if L4 checksum is needed */
301 switch (mbuf->ol_flags & PKT_TX_L4_MASK) {
302 case PKT_TX_TCP_CKSUM:
303 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
304 ena_tx_ctx->l4_csum_enable = true;
306 case PKT_TX_UDP_CKSUM:
307 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
308 ena_tx_ctx->l4_csum_enable = true;
311 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
312 ena_tx_ctx->l4_csum_enable = false;
316 ena_meta->mss = mbuf->tso_segsz;
317 ena_meta->l3_hdr_len = mbuf->l3_len;
318 ena_meta->l3_hdr_offset = mbuf->l2_len;
319 /* this param needed only for TSO */
320 ena_meta->l3_outer_hdr_len = 0;
321 ena_meta->l3_outer_hdr_offset = 0;
323 ena_tx_ctx->meta_valid = true;
325 ena_tx_ctx->meta_valid = false;
329 static void ena_config_host_info(struct ena_com_dev *ena_dev)
331 struct ena_admin_host_info *host_info;
334 /* Allocate only the host info */
335 rc = ena_com_allocate_host_info(ena_dev);
337 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
341 host_info = ena_dev->host_attr.host_info;
343 host_info->os_type = ENA_ADMIN_OS_DPDK;
344 host_info->kernel_ver = RTE_VERSION;
345 snprintf((char *)host_info->kernel_ver_str,
346 sizeof(host_info->kernel_ver_str),
347 "%s", rte_version());
348 host_info->os_dist = RTE_VERSION;
349 snprintf((char *)host_info->os_dist_str,
350 sizeof(host_info->os_dist_str),
351 "%s", rte_version());
352 host_info->driver_version =
353 (DRV_MODULE_VER_MAJOR) |
354 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
355 (DRV_MODULE_VER_SUBMINOR <<
356 ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
358 rc = ena_com_set_host_attributes(ena_dev);
361 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
363 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
371 ena_com_delete_host_info(ena_dev);
375 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
377 if (sset != ETH_SS_STATS)
380 /* Workaround for clang:
381 * touch internal structures to prevent
384 ENA_TOUCH(ena_stats_global_strings);
385 ENA_TOUCH(ena_stats_tx_strings);
386 ENA_TOUCH(ena_stats_rx_strings);
387 ENA_TOUCH(ena_stats_ena_com_strings);
389 return dev->data->nb_tx_queues *
390 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
391 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
394 static void ena_config_debug_area(struct ena_adapter *adapter)
399 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
401 RTE_LOG(ERR, PMD, "SS count is negative\n");
405 /* allocate 32 bytes for each string and 64bit for the value */
406 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
408 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
410 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
414 rc = ena_com_set_host_attributes(&adapter->ena_dev);
417 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
419 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
425 ena_com_delete_debug_area(&adapter->ena_dev);
428 static void ena_close(struct rte_eth_dev *dev)
430 struct ena_adapter *adapter =
431 (struct ena_adapter *)(dev->data->dev_private);
433 adapter->state = ENA_ADAPTER_STATE_STOPPED;
435 ena_rx_queue_release_all(dev);
436 ena_tx_queue_release_all(dev);
439 static int ena_rss_reta_update(struct rte_eth_dev *dev,
440 struct rte_eth_rss_reta_entry64 *reta_conf,
443 struct ena_adapter *adapter =
444 (struct ena_adapter *)(dev->data->dev_private);
445 struct ena_com_dev *ena_dev = &adapter->ena_dev;
451 if ((reta_size == 0) || (reta_conf == NULL))
454 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
455 RTE_LOG(WARNING, PMD,
456 "indirection table %d is bigger than supported (%d)\n",
457 reta_size, ENA_RX_RSS_TABLE_SIZE);
462 for (i = 0 ; i < reta_size ; i++) {
463 /* each reta_conf is for 64 entries.
464 * to support 128 we use 2 conf of 64
466 conf_idx = i / RTE_RETA_GROUP_SIZE;
467 idx = i % RTE_RETA_GROUP_SIZE;
468 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
470 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
471 ret = ena_com_indirect_table_fill_entry(ena_dev,
474 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
476 "Cannot fill indirect table\n");
483 ret = ena_com_indirect_table_set(ena_dev);
484 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
485 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
490 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
491 __func__, reta_size, adapter->rte_dev->data->port_id);
496 /* Query redirection table. */
497 static int ena_rss_reta_query(struct rte_eth_dev *dev,
498 struct rte_eth_rss_reta_entry64 *reta_conf,
501 struct ena_adapter *adapter =
502 (struct ena_adapter *)(dev->data->dev_private);
503 struct ena_com_dev *ena_dev = &adapter->ena_dev;
506 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
510 if (reta_size == 0 || reta_conf == NULL ||
511 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
514 ret = ena_com_indirect_table_get(ena_dev, indirect_table);
515 if (unlikely(ret && (ret != ENA_COM_PERMISSION))) {
516 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
521 for (i = 0 ; i < reta_size ; i++) {
522 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
523 reta_idx = i % RTE_RETA_GROUP_SIZE;
524 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
525 reta_conf[reta_conf_idx].reta[reta_idx] =
526 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
532 static int ena_rss_init_default(struct ena_adapter *adapter)
534 struct ena_com_dev *ena_dev = &adapter->ena_dev;
535 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
539 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
541 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
545 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
546 val = i % nb_rx_queues;
547 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
548 ENA_IO_RXQ_IDX(val));
549 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
550 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
555 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
556 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
557 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
558 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
562 rc = ena_com_set_default_hash_ctrl(ena_dev);
563 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
564 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
568 rc = ena_com_indirect_table_set(ena_dev);
569 if (unlikely(rc && (rc != ENA_COM_PERMISSION))) {
570 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
573 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
574 adapter->rte_dev->data->port_id);
579 ena_com_rss_destroy(ena_dev);
585 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
587 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
588 int nb_queues = dev->data->nb_rx_queues;
591 for (i = 0; i < nb_queues; i++)
592 ena_rx_queue_release(queues[i]);
595 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
597 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
598 int nb_queues = dev->data->nb_tx_queues;
601 for (i = 0; i < nb_queues; i++)
602 ena_tx_queue_release(queues[i]);
605 static void ena_rx_queue_release(void *queue)
607 struct ena_ring *ring = (struct ena_ring *)queue;
608 struct ena_adapter *adapter = ring->adapter;
611 ena_assert_msg(ring->configured,
612 "API violation - releasing not configured queue");
613 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
616 /* Destroy HW queue */
617 ena_qid = ENA_IO_RXQ_IDX(ring->id);
618 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
621 ena_rx_queue_release_bufs(ring);
623 /* Free ring resources */
624 if (ring->rx_buffer_info)
625 rte_free(ring->rx_buffer_info);
626 ring->rx_buffer_info = NULL;
628 ring->configured = 0;
630 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
631 ring->port_id, ring->id);
634 static void ena_tx_queue_release(void *queue)
636 struct ena_ring *ring = (struct ena_ring *)queue;
637 struct ena_adapter *adapter = ring->adapter;
640 ena_assert_msg(ring->configured,
641 "API violation. Releasing not configured queue");
642 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
645 /* Destroy HW queue */
646 ena_qid = ENA_IO_TXQ_IDX(ring->id);
647 ena_com_destroy_io_queue(&adapter->ena_dev, ena_qid);
650 ena_tx_queue_release_bufs(ring);
652 /* Free ring resources */
653 if (ring->tx_buffer_info)
654 rte_free(ring->tx_buffer_info);
656 if (ring->empty_tx_reqs)
657 rte_free(ring->empty_tx_reqs);
659 ring->empty_tx_reqs = NULL;
660 ring->tx_buffer_info = NULL;
662 ring->configured = 0;
664 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
665 ring->port_id, ring->id);
668 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
670 unsigned int ring_mask = ring->ring_size - 1;
672 while (ring->next_to_clean != ring->next_to_use) {
674 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
677 __rte_mbuf_raw_free(m);
679 ring->next_to_clean++;
683 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
685 unsigned int ring_mask = ring->ring_size - 1;
687 while (ring->next_to_clean != ring->next_to_use) {
688 struct ena_tx_buffer *tx_buf =
689 &ring->tx_buffer_info[ring->next_to_clean & ring_mask];
692 rte_pktmbuf_free(tx_buf->mbuf);
694 ring->next_to_clean++;
698 static int ena_link_update(struct rte_eth_dev *dev,
699 __rte_unused int wait_to_complete)
701 struct rte_eth_link *link = &dev->data->dev_link;
703 link->link_status = 1;
704 link->link_speed = ETH_SPEED_NUM_10G;
705 link->link_duplex = ETH_LINK_FULL_DUPLEX;
710 static int ena_queue_restart_all(struct rte_eth_dev *dev,
711 enum ena_ring_type ring_type)
713 struct ena_adapter *adapter =
714 (struct ena_adapter *)(dev->data->dev_private);
715 struct ena_ring *queues = NULL;
719 queues = (ring_type == ENA_RING_TYPE_RX) ?
720 adapter->rx_ring : adapter->tx_ring;
722 for (i = 0; i < adapter->num_queues; i++) {
723 if (queues[i].configured) {
724 if (ring_type == ENA_RING_TYPE_RX) {
726 dev->data->rx_queues[i] == &queues[i],
727 "Inconsistent state of rx queues\n");
730 dev->data->tx_queues[i] == &queues[i],
731 "Inconsistent state of tx queues\n");
734 rc = ena_queue_restart(&queues[i]);
738 "failed to restart queue %d type(%d)\n",
748 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
750 uint32_t max_frame_len = adapter->max_mtu;
752 if (adapter->rte_eth_dev_data->dev_conf.rxmode.jumbo_frame == 1)
754 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
756 return max_frame_len;
759 static int ena_check_valid_conf(struct ena_adapter *adapter)
761 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
763 if (max_frame_len > adapter->max_mtu) {
764 PMD_INIT_LOG(ERR, "Unsupported MTU of %d\n", max_frame_len);
772 ena_calc_queue_size(struct ena_com_dev *ena_dev,
773 struct ena_com_dev_get_features_ctx *get_feat_ctx)
775 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
777 queue_size = RTE_MIN(queue_size,
778 get_feat_ctx->max_queues.max_cq_depth);
779 queue_size = RTE_MIN(queue_size,
780 get_feat_ctx->max_queues.max_sq_depth);
782 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
783 queue_size = RTE_MIN(queue_size,
784 get_feat_ctx->max_queues.max_llq_depth);
786 /* Round down to power of 2 */
787 if (!rte_is_power_of_2(queue_size))
788 queue_size = rte_align32pow2(queue_size >> 1);
790 if (queue_size == 0) {
791 PMD_INIT_LOG(ERR, "Invalid queue size\n");
798 static void ena_stats_restart(struct rte_eth_dev *dev)
800 struct ena_adapter *adapter =
801 (struct ena_adapter *)(dev->data->dev_private);
803 rte_atomic64_init(&adapter->drv_stats->ierrors);
804 rte_atomic64_init(&adapter->drv_stats->oerrors);
805 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
808 static void ena_stats_get(struct rte_eth_dev *dev,
809 struct rte_eth_stats *stats)
811 struct ena_admin_basic_stats ena_stats;
812 struct ena_adapter *adapter =
813 (struct ena_adapter *)(dev->data->dev_private);
814 struct ena_com_dev *ena_dev = &adapter->ena_dev;
817 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
820 memset(&ena_stats, 0, sizeof(ena_stats));
821 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
823 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
827 /* Set of basic statistics from ENA */
828 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
829 ena_stats.rx_pkts_low);
830 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
831 ena_stats.tx_pkts_low);
832 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
833 ena_stats.rx_bytes_low);
834 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
835 ena_stats.tx_bytes_low);
836 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
837 ena_stats.rx_drops_low);
839 /* Driver related stats */
840 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
841 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
842 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
845 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
847 struct ena_adapter *adapter;
848 struct ena_com_dev *ena_dev;
851 ena_assert_msg(dev->data != NULL, "Uninitialized device");
852 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
853 adapter = (struct ena_adapter *)(dev->data->dev_private);
855 ena_dev = &adapter->ena_dev;
856 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
858 if (mtu > ena_get_mtu_conf(adapter)) {
860 "Given MTU (%d) exceeds maximum MTU supported (%d)\n",
861 mtu, ena_get_mtu_conf(adapter));
866 rc = ena_com_set_dev_mtu(ena_dev, mtu);
868 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
870 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
876 static int ena_start(struct rte_eth_dev *dev)
878 struct ena_adapter *adapter =
879 (struct ena_adapter *)(dev->data->dev_private);
882 if (!(adapter->state == ENA_ADAPTER_STATE_CONFIG ||
883 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
884 PMD_INIT_LOG(ERR, "API violation");
888 rc = ena_check_valid_conf(adapter);
892 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
896 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
900 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
901 ETH_MQ_RX_RSS_FLAG) {
902 rc = ena_rss_init_default(adapter);
907 ena_stats_restart(dev);
909 adapter->state = ENA_ADAPTER_STATE_RUNNING;
914 static int ena_queue_restart(struct ena_ring *ring)
918 ena_assert_msg(ring->configured == 1,
919 "Trying to restart unconfigured queue\n");
921 ring->next_to_clean = 0;
922 ring->next_to_use = 0;
924 if (ring->type == ENA_RING_TYPE_TX)
927 rc = ena_populate_rx_queue(ring, ring->ring_size);
928 if ((unsigned int)rc != ring->ring_size) {
929 PMD_INIT_LOG(ERR, "Failed to populate rx ring !\n");
936 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
939 __rte_unused unsigned int socket_id,
940 __rte_unused const struct rte_eth_txconf *tx_conf)
942 struct ena_com_create_io_ctx ctx =
943 /* policy set to _HOST just to satisfy icc compiler */
944 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
945 ENA_COM_IO_QUEUE_DIRECTION_TX, 0, 0, 0, 0 };
946 struct ena_ring *txq = NULL;
947 struct ena_adapter *adapter =
948 (struct ena_adapter *)(dev->data->dev_private);
952 struct ena_com_dev *ena_dev = &adapter->ena_dev;
954 txq = &adapter->tx_ring[queue_idx];
956 if (txq->configured) {
958 "API violation. Queue %d is already configured\n",
963 if (!rte_is_power_of_2(nb_desc)) {
965 "Unsupported size of RX queue: %d is not a power of 2.",
970 if (nb_desc > adapter->tx_ring_size) {
972 "Unsupported size of TX queue (max size: %d)\n",
973 adapter->tx_ring_size);
977 ena_qid = ENA_IO_TXQ_IDX(queue_idx);
979 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
981 ctx.msix_vector = -1; /* admin interrupts not used */
982 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
983 ctx.queue_size = adapter->tx_ring_size;
984 ctx.numa_node = ena_cpu_to_node(queue_idx);
986 rc = ena_com_create_io_queue(ena_dev, &ctx);
989 "failed to create io TX queue #%d (qid:%d) rc: %d\n",
990 queue_idx, ena_qid, rc);
992 txq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
993 txq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
995 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
997 &txq->ena_com_io_cq);
1000 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1002 ena_com_destroy_io_queue(ena_dev, ena_qid);
1006 txq->port_id = dev->data->port_id;
1007 txq->next_to_clean = 0;
1008 txq->next_to_use = 0;
1009 txq->ring_size = nb_desc;
1011 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1012 sizeof(struct ena_tx_buffer) *
1014 RTE_CACHE_LINE_SIZE);
1015 if (!txq->tx_buffer_info) {
1016 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1020 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1021 sizeof(u16) * txq->ring_size,
1022 RTE_CACHE_LINE_SIZE);
1023 if (!txq->empty_tx_reqs) {
1024 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1025 rte_free(txq->tx_buffer_info);
1028 for (i = 0; i < txq->ring_size; i++)
1029 txq->empty_tx_reqs[i] = i;
1031 /* Store pointer to this queue in upper layer */
1032 txq->configured = 1;
1033 dev->data->tx_queues[queue_idx] = txq;
1038 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1041 __rte_unused unsigned int socket_id,
1042 __rte_unused const struct rte_eth_rxconf *rx_conf,
1043 struct rte_mempool *mp)
1045 struct ena_com_create_io_ctx ctx =
1046 /* policy set to _HOST just to satisfy icc compiler */
1047 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1048 ENA_COM_IO_QUEUE_DIRECTION_RX, 0, 0, 0, 0 };
1049 struct ena_adapter *adapter =
1050 (struct ena_adapter *)(dev->data->dev_private);
1051 struct ena_ring *rxq = NULL;
1052 uint16_t ena_qid = 0;
1054 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1056 rxq = &adapter->rx_ring[queue_idx];
1057 if (rxq->configured) {
1059 "API violation. Queue %d is already configured\n",
1064 if (!rte_is_power_of_2(nb_desc)) {
1066 "Unsupported size of TX queue: %d is not a power of 2.",
1071 if (nb_desc > adapter->rx_ring_size) {
1073 "Unsupported size of RX queue (max size: %d)\n",
1074 adapter->rx_ring_size);
1078 ena_qid = ENA_IO_RXQ_IDX(queue_idx);
1081 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1082 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1083 ctx.msix_vector = -1; /* admin interrupts not used */
1084 ctx.queue_size = adapter->rx_ring_size;
1085 ctx.numa_node = ena_cpu_to_node(queue_idx);
1087 rc = ena_com_create_io_queue(ena_dev, &ctx);
1089 RTE_LOG(ERR, PMD, "failed to create io RX queue #%d rc: %d\n",
1092 rxq->ena_com_io_cq = &ena_dev->io_cq_queues[ena_qid];
1093 rxq->ena_com_io_sq = &ena_dev->io_sq_queues[ena_qid];
1095 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1096 &rxq->ena_com_io_sq,
1097 &rxq->ena_com_io_cq);
1100 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1102 ena_com_destroy_io_queue(ena_dev, ena_qid);
1105 rxq->port_id = dev->data->port_id;
1106 rxq->next_to_clean = 0;
1107 rxq->next_to_use = 0;
1108 rxq->ring_size = nb_desc;
1111 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1112 sizeof(struct rte_mbuf *) * nb_desc,
1113 RTE_CACHE_LINE_SIZE);
1114 if (!rxq->rx_buffer_info) {
1115 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1119 /* Store pointer to this queue in upper layer */
1120 rxq->configured = 1;
1121 dev->data->rx_queues[queue_idx] = rxq;
1126 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1130 uint16_t ring_size = rxq->ring_size;
1131 uint16_t ring_mask = ring_size - 1;
1132 uint16_t next_to_use = rxq->next_to_use;
1134 struct rte_mbuf **mbufs = &rxq->rx_buffer_info[0];
1136 if (unlikely(!count))
1139 in_use = rxq->next_to_use - rxq->next_to_clean;
1140 ena_assert_msg(((in_use + count) <= ring_size), "bad ring state");
1142 count = RTE_MIN(count,
1143 (uint16_t)(ring_size - (next_to_use & ring_mask)));
1145 /* get resources for incoming packets */
1146 rc = rte_mempool_get_bulk(rxq->mb_pool,
1147 (void **)(&mbufs[next_to_use & ring_mask]),
1149 if (unlikely(rc < 0)) {
1150 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1151 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1155 for (i = 0; i < count; i++) {
1156 uint16_t next_to_use_masked = next_to_use & ring_mask;
1157 struct rte_mbuf *mbuf = mbufs[next_to_use_masked];
1158 struct ena_com_buf ebuf;
1160 rte_prefetch0(mbufs[((next_to_use + 4) & ring_mask)]);
1161 /* prepare physical address for DMA transaction */
1162 ebuf.paddr = mbuf->buf_physaddr + RTE_PKTMBUF_HEADROOM;
1163 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1164 /* pass resource to device */
1165 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1166 &ebuf, next_to_use_masked);
1168 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1174 /* When we submitted free recources to device... */
1176 /* ...let HW know that it can fill buffers with data */
1178 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1180 rxq->next_to_use = next_to_use;
1186 static int ena_device_init(struct ena_com_dev *ena_dev,
1187 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1190 bool readless_supported;
1192 /* Initialize mmio registers */
1193 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1195 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1199 /* The PCIe configuration space revision id indicate if mmio reg
1202 readless_supported =
1203 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1204 & ENA_MMIO_DISABLE_REG_READ);
1205 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1208 rc = ena_com_dev_reset(ena_dev);
1210 RTE_LOG(ERR, PMD, "cannot reset device\n");
1211 goto err_mmio_read_less;
1214 /* check FW version */
1215 rc = ena_com_validate_version(ena_dev);
1217 RTE_LOG(ERR, PMD, "device version is too low\n");
1218 goto err_mmio_read_less;
1221 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1223 /* ENA device administration layer init */
1224 rc = ena_com_admin_init(ena_dev, NULL, true);
1227 "cannot initialize ena admin queue with device\n");
1228 goto err_mmio_read_less;
1231 ena_config_host_info(ena_dev);
1233 /* To enable the msix interrupts the driver needs to know the number
1234 * of queues. So the driver uses polling mode to retrieve this
1237 ena_com_set_admin_polling_mode(ena_dev, true);
1239 /* Get Device Attributes and features */
1240 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1243 "cannot get attribute for ena device rc= %d\n", rc);
1244 goto err_admin_init;
1250 ena_com_admin_destroy(ena_dev);
1253 ena_com_mmio_reg_read_request_destroy(ena_dev);
1258 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1260 struct rte_pci_device *pci_dev;
1261 struct ena_adapter *adapter =
1262 (struct ena_adapter *)(eth_dev->data->dev_private);
1263 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1264 struct ena_com_dev_get_features_ctx get_feat_ctx;
1267 static int adapters_found;
1269 memset(adapter, 0, sizeof(struct ena_adapter));
1270 ena_dev = &adapter->ena_dev;
1272 eth_dev->dev_ops = &ena_dev_ops;
1273 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1274 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1275 adapter->rte_eth_dev_data = eth_dev->data;
1276 adapter->rte_dev = eth_dev;
1278 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1281 pci_dev = eth_dev->pci_dev;
1282 adapter->pdev = pci_dev;
1284 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d\n",
1285 pci_dev->addr.domain,
1287 pci_dev->addr.devid,
1288 pci_dev->addr.function);
1290 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1291 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1293 /* Present ENA_MEM_BAR indicates available LLQ mode.
1294 * Use corresponding policy
1296 if (adapter->dev_mem_base)
1297 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
1298 else if (adapter->regs)
1299 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1301 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)\n",
1304 ena_dev->reg_bar = adapter->regs;
1305 ena_dev->dmadev = adapter->pdev;
1307 adapter->id_number = adapters_found;
1309 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1310 adapter->id_number);
1312 /* device specific initialization routine */
1313 rc = ena_device_init(ena_dev, &get_feat_ctx);
1315 PMD_INIT_LOG(CRIT, "Failed to init ENA device\n");
1319 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1320 if (get_feat_ctx.max_queues.max_llq_num == 0) {
1322 "Trying to use LLQ but llq_num is 0.\n"
1323 "Fall back into regular queues.\n");
1324 ena_dev->tx_mem_queue_type =
1325 ENA_ADMIN_PLACEMENT_POLICY_HOST;
1326 adapter->num_queues =
1327 get_feat_ctx.max_queues.max_sq_num;
1329 adapter->num_queues =
1330 get_feat_ctx.max_queues.max_llq_num;
1333 adapter->num_queues = get_feat_ctx.max_queues.max_sq_num;
1336 queue_size = ena_calc_queue_size(ena_dev, &get_feat_ctx);
1337 if ((queue_size <= 0) || (adapter->num_queues <= 0))
1340 adapter->tx_ring_size = queue_size;
1341 adapter->rx_ring_size = queue_size;
1343 /* prepare ring structures */
1344 ena_init_rings(adapter);
1346 ena_config_debug_area(adapter);
1348 /* Set max MTU for this device */
1349 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1351 /* Copy MAC address and point DPDK to it */
1352 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1353 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1354 (struct ether_addr *)adapter->mac_addr);
1356 adapter->drv_stats = rte_zmalloc("adapter stats",
1357 sizeof(*adapter->drv_stats),
1358 RTE_CACHE_LINE_SIZE);
1359 if (!adapter->drv_stats) {
1360 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1365 adapter->state = ENA_ADAPTER_STATE_INIT;
1370 static int ena_dev_configure(struct rte_eth_dev *dev)
1372 struct ena_adapter *adapter =
1373 (struct ena_adapter *)(dev->data->dev_private);
1375 if (!(adapter->state == ENA_ADAPTER_STATE_INIT ||
1376 adapter->state == ENA_ADAPTER_STATE_STOPPED)) {
1377 PMD_INIT_LOG(ERR, "Illegal adapter state: %d\n",
1382 switch (adapter->state) {
1383 case ENA_ADAPTER_STATE_INIT:
1384 case ENA_ADAPTER_STATE_STOPPED:
1385 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1387 case ENA_ADAPTER_STATE_CONFIG:
1388 RTE_LOG(WARNING, PMD,
1389 "Ivalid driver state while trying to configure device\n");
1398 static void ena_init_rings(struct ena_adapter *adapter)
1402 for (i = 0; i < adapter->num_queues; i++) {
1403 struct ena_ring *ring = &adapter->tx_ring[i];
1405 ring->configured = 0;
1406 ring->type = ENA_RING_TYPE_TX;
1407 ring->adapter = adapter;
1409 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1410 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1413 for (i = 0; i < adapter->num_queues; i++) {
1414 struct ena_ring *ring = &adapter->rx_ring[i];
1416 ring->configured = 0;
1417 ring->type = ENA_RING_TYPE_RX;
1418 ring->adapter = adapter;
1423 static void ena_infos_get(struct rte_eth_dev *dev,
1424 struct rte_eth_dev_info *dev_info)
1426 struct ena_adapter *adapter;
1427 struct ena_com_dev *ena_dev;
1428 struct ena_com_dev_get_features_ctx feat;
1429 uint32_t rx_feat = 0, tx_feat = 0;
1432 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1433 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1434 adapter = (struct ena_adapter *)(dev->data->dev_private);
1436 ena_dev = &adapter->ena_dev;
1437 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1439 dev_info->speed_capa =
1441 ETH_LINK_SPEED_2_5G |
1443 ETH_LINK_SPEED_10G |
1444 ETH_LINK_SPEED_25G |
1445 ETH_LINK_SPEED_40G |
1446 ETH_LINK_SPEED_50G |
1447 ETH_LINK_SPEED_100G;
1449 /* Get supported features from HW */
1450 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1453 "Cannot get attribute for ena device rc= %d\n", rc);
1457 /* Set Tx & Rx features available for device */
1458 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1459 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1461 if (feat.offload.tx &
1462 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1463 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1464 DEV_TX_OFFLOAD_UDP_CKSUM |
1465 DEV_TX_OFFLOAD_TCP_CKSUM;
1467 if (feat.offload.tx &
1468 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1469 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1470 DEV_RX_OFFLOAD_UDP_CKSUM |
1471 DEV_RX_OFFLOAD_TCP_CKSUM;
1473 /* Inform framework about available features */
1474 dev_info->rx_offload_capa = rx_feat;
1475 dev_info->tx_offload_capa = tx_feat;
1477 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1478 dev_info->max_rx_pktlen = adapter->max_mtu;
1479 dev_info->max_mac_addrs = 1;
1481 dev_info->max_rx_queues = adapter->num_queues;
1482 dev_info->max_tx_queues = adapter->num_queues;
1483 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1486 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1489 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1490 unsigned int ring_size = rx_ring->ring_size;
1491 unsigned int ring_mask = ring_size - 1;
1492 uint16_t next_to_clean = rx_ring->next_to_clean;
1493 uint16_t desc_in_use = 0;
1494 unsigned int recv_idx = 0;
1495 struct rte_mbuf *mbuf = NULL;
1496 struct rte_mbuf *mbuf_head = NULL;
1497 struct rte_mbuf *mbuf_prev = NULL;
1498 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1499 unsigned int completed;
1501 struct ena_com_rx_ctx ena_rx_ctx;
1504 /* Check adapter state */
1505 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1507 "Trying to receive pkts while device is NOT running\n");
1511 desc_in_use = rx_ring->next_to_use - next_to_clean;
1512 if (unlikely(nb_pkts > desc_in_use))
1513 nb_pkts = desc_in_use;
1515 for (completed = 0; completed < nb_pkts; completed++) {
1518 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1519 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1520 ena_rx_ctx.descs = 0;
1521 /* receive packet context */
1522 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1523 rx_ring->ena_com_io_sq,
1526 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1530 if (unlikely(ena_rx_ctx.descs == 0))
1533 while (segments < ena_rx_ctx.descs) {
1534 mbuf = rx_buff_info[next_to_clean & ring_mask];
1535 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1536 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1539 if (segments == 0) {
1540 mbuf->nb_segs = ena_rx_ctx.descs;
1541 mbuf->port = rx_ring->port_id;
1545 /* for multi-segment pkts create mbuf chain */
1546 mbuf_prev->next = mbuf;
1548 mbuf_head->pkt_len += mbuf->data_len;
1555 /* fill mbuf attributes if any */
1556 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1557 mbuf_head->hash.rss = (uint32_t)rx_ring->id;
1559 /* pass to DPDK application head mbuf */
1560 rx_pkts[recv_idx] = mbuf_head;
1564 /* Burst refill to save doorbells, memory barriers, const interval */
1565 if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1566 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1568 rx_ring->next_to_clean = next_to_clean;
1573 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1576 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1577 uint16_t next_to_use = tx_ring->next_to_use;
1578 uint16_t next_to_clean = tx_ring->next_to_clean;
1579 struct rte_mbuf *mbuf;
1580 unsigned int ring_size = tx_ring->ring_size;
1581 unsigned int ring_mask = ring_size - 1;
1582 struct ena_com_tx_ctx ena_tx_ctx;
1583 struct ena_tx_buffer *tx_info;
1584 struct ena_com_buf *ebuf;
1585 uint16_t rc, req_id, total_tx_descs = 0;
1586 uint16_t sent_idx = 0;
1589 /* Check adapter state */
1590 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1592 "Trying to xmit pkts while device is NOT running\n");
1596 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
1597 mbuf = tx_pkts[sent_idx];
1599 req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
1600 tx_info = &tx_ring->tx_buffer_info[req_id];
1601 tx_info->mbuf = mbuf;
1602 tx_info->num_of_bufs = 0;
1603 ebuf = tx_info->bufs;
1605 /* Prepare TX context */
1606 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1607 memset(&ena_tx_ctx.ena_meta, 0x0,
1608 sizeof(struct ena_com_tx_meta));
1609 ena_tx_ctx.ena_bufs = ebuf;
1610 ena_tx_ctx.req_id = req_id;
1611 if (tx_ring->tx_mem_queue_type ==
1612 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1613 /* prepare the push buffer with
1614 * virtual address of the data
1616 ena_tx_ctx.header_len =
1617 RTE_MIN(mbuf->data_len,
1618 tx_ring->tx_max_header_size);
1619 ena_tx_ctx.push_header =
1620 (void *)((char *)mbuf->buf_addr +
1622 } /* there's no else as we take advantage of memset zeroing */
1624 /* Set TX offloads flags, if applicable */
1625 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx);
1627 if (unlikely(mbuf->ol_flags &
1628 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
1629 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
1631 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
1633 /* Process first segment taking into
1634 * consideration pushed header
1636 if (mbuf->data_len > ena_tx_ctx.header_len) {
1637 ebuf->paddr = mbuf->buf_physaddr +
1639 ena_tx_ctx.header_len;
1640 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
1642 tx_info->num_of_bufs++;
1645 while ((mbuf = mbuf->next) != NULL) {
1646 ebuf->paddr = mbuf->buf_physaddr + mbuf->data_off;
1647 ebuf->len = mbuf->data_len;
1649 tx_info->num_of_bufs++;
1652 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1654 /* Write data to device */
1655 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
1656 &ena_tx_ctx, &nb_hw_desc);
1660 tx_info->tx_descs = nb_hw_desc;
1665 /* If there are ready packets to be xmitted... */
1667 /* ...let HW do its best :-) */
1669 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
1671 tx_ring->next_to_use = next_to_use;
1674 /* Clear complete packets */
1675 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
1676 /* Get Tx info & store how many descs were processed */
1677 tx_info = &tx_ring->tx_buffer_info[req_id];
1678 total_tx_descs += tx_info->tx_descs;
1680 /* Free whole mbuf chain */
1681 mbuf = tx_info->mbuf;
1682 rte_pktmbuf_free(mbuf);
1684 /* Put back descriptor to the ring for reuse */
1685 tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
1688 /* If too many descs to clean, leave it for another run */
1689 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
1693 if (total_tx_descs > 0) {
1694 /* acknowledge completion of sent packets */
1695 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
1696 tx_ring->next_to_clean = next_to_clean;
1702 static struct eth_driver rte_ena_pmd = {
1704 .id_table = pci_id_ena_map,
1705 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1706 .probe = rte_eth_dev_pci_probe,
1707 .remove = rte_eth_dev_pci_remove,
1709 .eth_dev_init = eth_ena_dev_init,
1710 .dev_private_size = sizeof(struct ena_adapter),
1713 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd.pci_drv);
1714 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);