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 1
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 #define ENA_MAX_RING_DESC ENA_DEFAULT_RING_SIZE
89 #define ENA_MIN_RING_DESC 128
91 enum ethtool_stringset {
97 char name[ETH_GSTRING_LEN];
101 #define ENA_STAT_ENA_COM_ENTRY(stat) { \
103 .stat_offset = offsetof(struct ena_com_stats_admin, stat) \
106 #define ENA_STAT_ENTRY(stat, stat_type) { \
108 .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
111 #define ENA_STAT_RX_ENTRY(stat) \
112 ENA_STAT_ENTRY(stat, rx)
114 #define ENA_STAT_TX_ENTRY(stat) \
115 ENA_STAT_ENTRY(stat, tx)
117 #define ENA_STAT_GLOBAL_ENTRY(stat) \
118 ENA_STAT_ENTRY(stat, dev)
121 * Each rte_memzone should have unique name.
122 * To satisfy it, count number of allocation and add it to name.
124 uint32_t ena_alloc_cnt;
126 static const struct ena_stats ena_stats_global_strings[] = {
127 ENA_STAT_GLOBAL_ENTRY(tx_timeout),
128 ENA_STAT_GLOBAL_ENTRY(io_suspend),
129 ENA_STAT_GLOBAL_ENTRY(io_resume),
130 ENA_STAT_GLOBAL_ENTRY(wd_expired),
131 ENA_STAT_GLOBAL_ENTRY(interface_up),
132 ENA_STAT_GLOBAL_ENTRY(interface_down),
133 ENA_STAT_GLOBAL_ENTRY(admin_q_pause),
136 static const struct ena_stats ena_stats_tx_strings[] = {
137 ENA_STAT_TX_ENTRY(cnt),
138 ENA_STAT_TX_ENTRY(bytes),
139 ENA_STAT_TX_ENTRY(queue_stop),
140 ENA_STAT_TX_ENTRY(queue_wakeup),
141 ENA_STAT_TX_ENTRY(dma_mapping_err),
142 ENA_STAT_TX_ENTRY(linearize),
143 ENA_STAT_TX_ENTRY(linearize_failed),
144 ENA_STAT_TX_ENTRY(tx_poll),
145 ENA_STAT_TX_ENTRY(doorbells),
146 ENA_STAT_TX_ENTRY(prepare_ctx_err),
147 ENA_STAT_TX_ENTRY(missing_tx_comp),
148 ENA_STAT_TX_ENTRY(bad_req_id),
151 static const struct ena_stats ena_stats_rx_strings[] = {
152 ENA_STAT_RX_ENTRY(cnt),
153 ENA_STAT_RX_ENTRY(bytes),
154 ENA_STAT_RX_ENTRY(refil_partial),
155 ENA_STAT_RX_ENTRY(bad_csum),
156 ENA_STAT_RX_ENTRY(page_alloc_fail),
157 ENA_STAT_RX_ENTRY(skb_alloc_fail),
158 ENA_STAT_RX_ENTRY(dma_mapping_err),
159 ENA_STAT_RX_ENTRY(bad_desc_num),
160 ENA_STAT_RX_ENTRY(small_copy_len_pkt),
163 static const struct ena_stats ena_stats_ena_com_strings[] = {
164 ENA_STAT_ENA_COM_ENTRY(aborted_cmd),
165 ENA_STAT_ENA_COM_ENTRY(submitted_cmd),
166 ENA_STAT_ENA_COM_ENTRY(completed_cmd),
167 ENA_STAT_ENA_COM_ENTRY(out_of_space),
168 ENA_STAT_ENA_COM_ENTRY(no_completion),
171 #define ENA_STATS_ARRAY_GLOBAL ARRAY_SIZE(ena_stats_global_strings)
172 #define ENA_STATS_ARRAY_TX ARRAY_SIZE(ena_stats_tx_strings)
173 #define ENA_STATS_ARRAY_RX ARRAY_SIZE(ena_stats_rx_strings)
174 #define ENA_STATS_ARRAY_ENA_COM ARRAY_SIZE(ena_stats_ena_com_strings)
176 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
177 DEV_TX_OFFLOAD_UDP_CKSUM |\
178 DEV_TX_OFFLOAD_IPV4_CKSUM |\
179 DEV_TX_OFFLOAD_TCP_TSO)
180 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
184 /** Vendor ID used by Amazon devices */
185 #define PCI_VENDOR_ID_AMAZON 0x1D0F
186 /** Amazon devices */
187 #define PCI_DEVICE_ID_ENA_VF 0xEC20
188 #define PCI_DEVICE_ID_ENA_LLQ_VF 0xEC21
190 #define ENA_TX_OFFLOAD_MASK (\
195 #define ENA_TX_OFFLOAD_NOTSUP_MASK \
196 (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
198 int ena_logtype_init;
199 int ena_logtype_driver;
201 static const struct rte_pci_id pci_id_ena_map[] = {
202 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
203 { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_LLQ_VF) },
207 static struct ena_aenq_handlers aenq_handlers;
209 static int ena_device_init(struct ena_com_dev *ena_dev,
210 struct ena_com_dev_get_features_ctx *get_feat_ctx,
212 static int ena_dev_configure(struct rte_eth_dev *dev);
213 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
215 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
217 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
218 uint16_t nb_desc, unsigned int socket_id,
219 const struct rte_eth_txconf *tx_conf);
220 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
221 uint16_t nb_desc, unsigned int socket_id,
222 const struct rte_eth_rxconf *rx_conf,
223 struct rte_mempool *mp);
224 static uint16_t eth_ena_recv_pkts(void *rx_queue,
225 struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
226 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
227 static void ena_init_rings(struct ena_adapter *adapter);
228 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
229 static int ena_start(struct rte_eth_dev *dev);
230 static void ena_stop(struct rte_eth_dev *dev);
231 static void ena_close(struct rte_eth_dev *dev);
232 static int ena_dev_reset(struct rte_eth_dev *dev);
233 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
234 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
235 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
236 static void ena_rx_queue_release(void *queue);
237 static void ena_tx_queue_release(void *queue);
238 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
239 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
240 static int ena_link_update(struct rte_eth_dev *dev,
241 int wait_to_complete);
242 static int ena_create_io_queue(struct ena_ring *ring);
243 static void ena_free_io_queues_all(struct ena_adapter *adapter);
244 static int ena_queue_restart(struct ena_ring *ring);
245 static int ena_queue_restart_all(struct rte_eth_dev *dev,
246 enum ena_ring_type ring_type);
247 static void ena_stats_restart(struct rte_eth_dev *dev);
248 static void ena_infos_get(struct rte_eth_dev *dev,
249 struct rte_eth_dev_info *dev_info);
250 static int ena_rss_reta_update(struct rte_eth_dev *dev,
251 struct rte_eth_rss_reta_entry64 *reta_conf,
253 static int ena_rss_reta_query(struct rte_eth_dev *dev,
254 struct rte_eth_rss_reta_entry64 *reta_conf,
256 static int ena_get_sset_count(struct rte_eth_dev *dev, int sset);
257 static void ena_interrupt_handler_rte(void *cb_arg);
258 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
260 static const struct eth_dev_ops ena_dev_ops = {
261 .dev_configure = ena_dev_configure,
262 .dev_infos_get = ena_infos_get,
263 .rx_queue_setup = ena_rx_queue_setup,
264 .tx_queue_setup = ena_tx_queue_setup,
265 .dev_start = ena_start,
266 .dev_stop = ena_stop,
267 .link_update = ena_link_update,
268 .stats_get = ena_stats_get,
269 .mtu_set = ena_mtu_set,
270 .rx_queue_release = ena_rx_queue_release,
271 .tx_queue_release = ena_tx_queue_release,
272 .dev_close = ena_close,
273 .dev_reset = ena_dev_reset,
274 .reta_update = ena_rss_reta_update,
275 .reta_query = ena_rss_reta_query,
278 #define NUMA_NO_NODE SOCKET_ID_ANY
280 static inline int ena_cpu_to_node(int cpu)
282 struct rte_config *config = rte_eal_get_configuration();
283 struct rte_fbarray *arr = &config->mem_config->memzones;
284 const struct rte_memzone *mz;
286 if (unlikely(cpu >= RTE_MAX_MEMZONE))
289 mz = rte_fbarray_get(arr, cpu);
291 return mz->socket_id;
294 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
295 struct ena_com_rx_ctx *ena_rx_ctx)
297 uint64_t ol_flags = 0;
298 uint32_t packet_type = 0;
300 if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
301 packet_type |= RTE_PTYPE_L4_TCP;
302 else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
303 packet_type |= RTE_PTYPE_L4_UDP;
305 if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4)
306 packet_type |= RTE_PTYPE_L3_IPV4;
307 else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6)
308 packet_type |= RTE_PTYPE_L3_IPV6;
310 if (unlikely(ena_rx_ctx->l4_csum_err))
311 ol_flags |= PKT_RX_L4_CKSUM_BAD;
312 if (unlikely(ena_rx_ctx->l3_csum_err))
313 ol_flags |= PKT_RX_IP_CKSUM_BAD;
315 mbuf->ol_flags = ol_flags;
316 mbuf->packet_type = packet_type;
319 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
320 struct ena_com_tx_ctx *ena_tx_ctx,
321 uint64_t queue_offloads)
323 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
325 if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
326 (queue_offloads & QUEUE_OFFLOADS)) {
327 /* check if TSO is required */
328 if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
329 (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
330 ena_tx_ctx->tso_enable = true;
332 ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
335 /* check if L3 checksum is needed */
336 if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
337 (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
338 ena_tx_ctx->l3_csum_enable = true;
340 if (mbuf->ol_flags & PKT_TX_IPV6) {
341 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
343 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
345 /* set don't fragment (DF) flag */
346 if (mbuf->packet_type &
347 (RTE_PTYPE_L4_NONFRAG
348 | RTE_PTYPE_INNER_L4_NONFRAG))
349 ena_tx_ctx->df = true;
352 /* check if L4 checksum is needed */
353 if ((mbuf->ol_flags & PKT_TX_TCP_CKSUM) &&
354 (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
355 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
356 ena_tx_ctx->l4_csum_enable = true;
357 } else if ((mbuf->ol_flags & PKT_TX_UDP_CKSUM) &&
358 (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
359 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
360 ena_tx_ctx->l4_csum_enable = true;
362 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
363 ena_tx_ctx->l4_csum_enable = false;
366 ena_meta->mss = mbuf->tso_segsz;
367 ena_meta->l3_hdr_len = mbuf->l3_len;
368 ena_meta->l3_hdr_offset = mbuf->l2_len;
370 ena_tx_ctx->meta_valid = true;
372 ena_tx_ctx->meta_valid = false;
376 static inline int validate_rx_req_id(struct ena_ring *rx_ring, uint16_t req_id)
378 if (likely(req_id < rx_ring->ring_size))
381 RTE_LOG(ERR, PMD, "Invalid rx req_id: %hu\n", req_id);
383 rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
384 rx_ring->adapter->trigger_reset = true;
389 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
391 struct ena_tx_buffer *tx_info = NULL;
393 if (likely(req_id < tx_ring->ring_size)) {
394 tx_info = &tx_ring->tx_buffer_info[req_id];
395 if (likely(tx_info->mbuf))
400 RTE_LOG(ERR, PMD, "tx_info doesn't have valid mbuf\n");
402 RTE_LOG(ERR, PMD, "Invalid req_id: %hu\n", req_id);
404 /* Trigger device reset */
405 tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
406 tx_ring->adapter->trigger_reset = true;
410 static void ena_config_host_info(struct ena_com_dev *ena_dev)
412 struct ena_admin_host_info *host_info;
415 /* Allocate only the host info */
416 rc = ena_com_allocate_host_info(ena_dev);
418 RTE_LOG(ERR, PMD, "Cannot allocate host info\n");
422 host_info = ena_dev->host_attr.host_info;
424 host_info->os_type = ENA_ADMIN_OS_DPDK;
425 host_info->kernel_ver = RTE_VERSION;
426 snprintf((char *)host_info->kernel_ver_str,
427 sizeof(host_info->kernel_ver_str),
428 "%s", rte_version());
429 host_info->os_dist = RTE_VERSION;
430 snprintf((char *)host_info->os_dist_str,
431 sizeof(host_info->os_dist_str),
432 "%s", rte_version());
433 host_info->driver_version =
434 (DRV_MODULE_VER_MAJOR) |
435 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
436 (DRV_MODULE_VER_SUBMINOR <<
437 ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
439 rc = ena_com_set_host_attributes(ena_dev);
441 if (rc == -ENA_COM_UNSUPPORTED)
442 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
444 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
452 ena_com_delete_host_info(ena_dev);
456 ena_get_sset_count(struct rte_eth_dev *dev, int sset)
458 if (sset != ETH_SS_STATS)
461 /* Workaround for clang:
462 * touch internal structures to prevent
465 ENA_TOUCH(ena_stats_global_strings);
466 ENA_TOUCH(ena_stats_tx_strings);
467 ENA_TOUCH(ena_stats_rx_strings);
468 ENA_TOUCH(ena_stats_ena_com_strings);
470 return dev->data->nb_tx_queues *
471 (ENA_STATS_ARRAY_TX + ENA_STATS_ARRAY_RX) +
472 ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENA_COM;
475 static void ena_config_debug_area(struct ena_adapter *adapter)
480 ss_count = ena_get_sset_count(adapter->rte_dev, ETH_SS_STATS);
482 RTE_LOG(ERR, PMD, "SS count is negative\n");
486 /* allocate 32 bytes for each string and 64bit for the value */
487 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
489 rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
491 RTE_LOG(ERR, PMD, "Cannot allocate debug area\n");
495 rc = ena_com_set_host_attributes(&adapter->ena_dev);
497 if (rc == -ENA_COM_UNSUPPORTED)
498 RTE_LOG(WARNING, PMD, "Cannot set host attributes\n");
500 RTE_LOG(ERR, PMD, "Cannot set host attributes\n");
507 ena_com_delete_debug_area(&adapter->ena_dev);
510 static void ena_close(struct rte_eth_dev *dev)
512 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
513 struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
514 struct ena_adapter *adapter =
515 (struct ena_adapter *)(dev->data->dev_private);
517 if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
519 adapter->state = ENA_ADAPTER_STATE_CLOSED;
521 ena_rx_queue_release_all(dev);
522 ena_tx_queue_release_all(dev);
524 rte_free(adapter->drv_stats);
525 adapter->drv_stats = NULL;
527 rte_intr_disable(intr_handle);
528 rte_intr_callback_unregister(intr_handle,
529 ena_interrupt_handler_rte,
533 * MAC is not allocated dynamically. Setting NULL should prevent from
534 * release of the resource in the rte_eth_dev_release_port().
536 dev->data->mac_addrs = NULL;
540 ena_dev_reset(struct rte_eth_dev *dev)
542 struct rte_mempool *mb_pool_rx[ENA_MAX_NUM_QUEUES];
543 struct rte_eth_dev *eth_dev;
544 struct rte_pci_device *pci_dev;
545 struct rte_intr_handle *intr_handle;
546 struct ena_com_dev *ena_dev;
547 struct ena_com_dev_get_features_ctx get_feat_ctx;
548 struct ena_adapter *adapter;
553 adapter = (struct ena_adapter *)(dev->data->dev_private);
554 ena_dev = &adapter->ena_dev;
555 eth_dev = adapter->rte_dev;
556 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
557 intr_handle = &pci_dev->intr_handle;
558 nb_queues = eth_dev->data->nb_rx_queues;
560 ena_com_set_admin_running_state(ena_dev, false);
562 rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
564 RTE_LOG(ERR, PMD, "Device reset failed\n");
566 for (i = 0; i < nb_queues; i++)
567 mb_pool_rx[i] = adapter->rx_ring[i].mb_pool;
569 ena_rx_queue_release_all(eth_dev);
570 ena_tx_queue_release_all(eth_dev);
572 rte_intr_disable(intr_handle);
574 ena_com_abort_admin_commands(ena_dev);
575 ena_com_wait_for_abort_completion(ena_dev);
576 ena_com_admin_destroy(ena_dev);
577 ena_com_mmio_reg_read_request_destroy(ena_dev);
579 rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
581 PMD_INIT_LOG(CRIT, "Cannot initialize device\n");
584 adapter->wd_state = wd_state;
586 rte_intr_enable(intr_handle);
587 ena_com_set_admin_polling_mode(ena_dev, false);
588 ena_com_admin_aenq_enable(ena_dev);
590 for (i = 0; i < nb_queues; ++i)
591 ena_rx_queue_setup(eth_dev, i, adapter->rx_ring_size, 0, NULL,
594 for (i = 0; i < nb_queues; ++i)
595 ena_tx_queue_setup(eth_dev, i, adapter->tx_ring_size, 0, NULL);
597 adapter->trigger_reset = false;
602 static int ena_rss_reta_update(struct rte_eth_dev *dev,
603 struct rte_eth_rss_reta_entry64 *reta_conf,
606 struct ena_adapter *adapter =
607 (struct ena_adapter *)(dev->data->dev_private);
608 struct ena_com_dev *ena_dev = &adapter->ena_dev;
614 if ((reta_size == 0) || (reta_conf == NULL))
617 if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
618 RTE_LOG(WARNING, PMD,
619 "indirection table %d is bigger than supported (%d)\n",
620 reta_size, ENA_RX_RSS_TABLE_SIZE);
624 for (i = 0 ; i < reta_size ; i++) {
625 /* each reta_conf is for 64 entries.
626 * to support 128 we use 2 conf of 64
628 conf_idx = i / RTE_RETA_GROUP_SIZE;
629 idx = i % RTE_RETA_GROUP_SIZE;
630 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
632 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
634 rc = ena_com_indirect_table_fill_entry(ena_dev,
637 if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
639 "Cannot fill indirect table\n");
645 rc = ena_com_indirect_table_set(ena_dev);
646 if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
647 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
651 RTE_LOG(DEBUG, PMD, "%s(): RSS configured %d entries for port %d\n",
652 __func__, reta_size, adapter->rte_dev->data->port_id);
657 /* Query redirection table. */
658 static int ena_rss_reta_query(struct rte_eth_dev *dev,
659 struct rte_eth_rss_reta_entry64 *reta_conf,
662 struct ena_adapter *adapter =
663 (struct ena_adapter *)(dev->data->dev_private);
664 struct ena_com_dev *ena_dev = &adapter->ena_dev;
667 u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
671 if (reta_size == 0 || reta_conf == NULL ||
672 (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
675 rc = ena_com_indirect_table_get(ena_dev, indirect_table);
676 if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
677 RTE_LOG(ERR, PMD, "cannot get indirect table\n");
681 for (i = 0 ; i < reta_size ; i++) {
682 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
683 reta_idx = i % RTE_RETA_GROUP_SIZE;
684 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
685 reta_conf[reta_conf_idx].reta[reta_idx] =
686 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
692 static int ena_rss_init_default(struct ena_adapter *adapter)
694 struct ena_com_dev *ena_dev = &adapter->ena_dev;
695 uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
699 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
701 RTE_LOG(ERR, PMD, "Cannot init indirect table\n");
705 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
706 val = i % nb_rx_queues;
707 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
708 ENA_IO_RXQ_IDX(val));
709 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
710 RTE_LOG(ERR, PMD, "Cannot fill indirect table\n");
715 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
716 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
717 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
718 RTE_LOG(INFO, PMD, "Cannot fill hash function\n");
722 rc = ena_com_set_default_hash_ctrl(ena_dev);
723 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
724 RTE_LOG(INFO, PMD, "Cannot fill hash control\n");
728 rc = ena_com_indirect_table_set(ena_dev);
729 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
730 RTE_LOG(ERR, PMD, "Cannot flush the indirect table\n");
733 RTE_LOG(DEBUG, PMD, "RSS configured for port %d\n",
734 adapter->rte_dev->data->port_id);
739 ena_com_rss_destroy(ena_dev);
745 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
747 struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
748 int nb_queues = dev->data->nb_rx_queues;
751 for (i = 0; i < nb_queues; i++)
752 ena_rx_queue_release(queues[i]);
755 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
757 struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
758 int nb_queues = dev->data->nb_tx_queues;
761 for (i = 0; i < nb_queues; i++)
762 ena_tx_queue_release(queues[i]);
765 static void ena_rx_queue_release(void *queue)
767 struct ena_ring *ring = (struct ena_ring *)queue;
769 ena_assert_msg(ring->configured,
770 "API violation - releasing not configured queue");
771 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
774 /* Free ring resources */
775 if (ring->rx_buffer_info)
776 rte_free(ring->rx_buffer_info);
777 ring->rx_buffer_info = NULL;
779 if (ring->rx_refill_buffer)
780 rte_free(ring->rx_refill_buffer);
781 ring->rx_refill_buffer = NULL;
783 if (ring->empty_rx_reqs)
784 rte_free(ring->empty_rx_reqs);
785 ring->empty_rx_reqs = NULL;
787 ring->configured = 0;
789 RTE_LOG(NOTICE, PMD, "RX Queue %d:%d released\n",
790 ring->port_id, ring->id);
793 static void ena_tx_queue_release(void *queue)
795 struct ena_ring *ring = (struct ena_ring *)queue;
797 ena_assert_msg(ring->configured,
798 "API violation. Releasing not configured queue");
799 ena_assert_msg(ring->adapter->state != ENA_ADAPTER_STATE_RUNNING,
803 ena_tx_queue_release_bufs(ring);
805 /* Free ring resources */
806 if (ring->tx_buffer_info)
807 rte_free(ring->tx_buffer_info);
809 if (ring->empty_tx_reqs)
810 rte_free(ring->empty_tx_reqs);
812 ring->empty_tx_reqs = NULL;
813 ring->tx_buffer_info = NULL;
815 ring->configured = 0;
817 RTE_LOG(NOTICE, PMD, "TX Queue %d:%d released\n",
818 ring->port_id, ring->id);
821 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
823 unsigned int ring_mask = ring->ring_size - 1;
825 while (ring->next_to_clean != ring->next_to_use) {
827 ring->rx_buffer_info[ring->next_to_clean & ring_mask];
830 rte_mbuf_raw_free(m);
832 ring->next_to_clean++;
836 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
840 for (i = 0; i < ring->ring_size; ++i) {
841 struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
844 rte_pktmbuf_free(tx_buf->mbuf);
846 ring->next_to_clean++;
850 static int ena_link_update(struct rte_eth_dev *dev,
851 __rte_unused int wait_to_complete)
853 struct rte_eth_link *link = &dev->data->dev_link;
854 struct ena_adapter *adapter;
856 adapter = (struct ena_adapter *)(dev->data->dev_private);
858 link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
859 link->link_speed = ETH_SPEED_NUM_NONE;
860 link->link_duplex = ETH_LINK_FULL_DUPLEX;
865 static int ena_queue_restart_all(struct rte_eth_dev *dev,
866 enum ena_ring_type ring_type)
868 struct ena_adapter *adapter =
869 (struct ena_adapter *)(dev->data->dev_private);
870 struct ena_ring *queues = NULL;
875 if (ring_type == ENA_RING_TYPE_RX) {
876 queues = adapter->rx_ring;
877 nb_queues = dev->data->nb_rx_queues;
879 queues = adapter->tx_ring;
880 nb_queues = dev->data->nb_tx_queues;
882 for (i = 0; i < nb_queues; i++) {
883 if (queues[i].configured) {
884 if (ring_type == ENA_RING_TYPE_RX) {
886 dev->data->rx_queues[i] == &queues[i],
887 "Inconsistent state of rx queues\n");
890 dev->data->tx_queues[i] == &queues[i],
891 "Inconsistent state of tx queues\n");
894 rc = ena_queue_restart(&queues[i]);
898 "failed to restart queue %d type(%d)",
908 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
910 uint32_t max_frame_len = adapter->max_mtu;
912 if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
913 DEV_RX_OFFLOAD_JUMBO_FRAME)
915 adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
917 return max_frame_len;
920 static int ena_check_valid_conf(struct ena_adapter *adapter)
922 uint32_t max_frame_len = ena_get_mtu_conf(adapter);
924 if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
925 PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
926 "max mtu: %d, min mtu: %d\n",
927 max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
928 return ENA_COM_UNSUPPORTED;
935 ena_calc_queue_size(struct ena_com_dev *ena_dev,
936 u16 *max_tx_sgl_size,
937 struct ena_com_dev_get_features_ctx *get_feat_ctx)
939 uint32_t queue_size = ENA_DEFAULT_RING_SIZE;
941 queue_size = RTE_MIN(queue_size,
942 get_feat_ctx->max_queues.max_cq_depth);
943 queue_size = RTE_MIN(queue_size,
944 get_feat_ctx->max_queues.max_sq_depth);
946 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
947 queue_size = RTE_MIN(queue_size,
948 get_feat_ctx->max_queues.max_llq_depth);
950 /* Round down to power of 2 */
951 if (!rte_is_power_of_2(queue_size))
952 queue_size = rte_align32pow2(queue_size >> 1);
954 if (unlikely(queue_size == 0)) {
955 PMD_INIT_LOG(ERR, "Invalid queue size");
959 *max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
960 get_feat_ctx->max_queues.max_packet_tx_descs);
965 static void ena_stats_restart(struct rte_eth_dev *dev)
967 struct ena_adapter *adapter =
968 (struct ena_adapter *)(dev->data->dev_private);
970 rte_atomic64_init(&adapter->drv_stats->ierrors);
971 rte_atomic64_init(&adapter->drv_stats->oerrors);
972 rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
975 static int ena_stats_get(struct rte_eth_dev *dev,
976 struct rte_eth_stats *stats)
978 struct ena_admin_basic_stats ena_stats;
979 struct ena_adapter *adapter =
980 (struct ena_adapter *)(dev->data->dev_private);
981 struct ena_com_dev *ena_dev = &adapter->ena_dev;
984 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
987 memset(&ena_stats, 0, sizeof(ena_stats));
988 rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
990 RTE_LOG(ERR, PMD, "Could not retrieve statistics from ENA");
994 /* Set of basic statistics from ENA */
995 stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
996 ena_stats.rx_pkts_low);
997 stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
998 ena_stats.tx_pkts_low);
999 stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
1000 ena_stats.rx_bytes_low);
1001 stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
1002 ena_stats.tx_bytes_low);
1003 stats->imissed = __MERGE_64B_H_L(ena_stats.rx_drops_high,
1004 ena_stats.rx_drops_low);
1006 /* Driver related stats */
1007 stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
1008 stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
1009 stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
1013 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1015 struct ena_adapter *adapter;
1016 struct ena_com_dev *ena_dev;
1019 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1020 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1021 adapter = (struct ena_adapter *)(dev->data->dev_private);
1023 ena_dev = &adapter->ena_dev;
1024 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1026 if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1028 "Invalid MTU setting. new_mtu: %d "
1029 "max mtu: %d min mtu: %d\n",
1030 mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1034 rc = ena_com_set_dev_mtu(ena_dev, mtu);
1036 RTE_LOG(ERR, PMD, "Could not set MTU: %d\n", mtu);
1038 RTE_LOG(NOTICE, PMD, "Set MTU: %d\n", mtu);
1043 static int ena_start(struct rte_eth_dev *dev)
1045 struct ena_adapter *adapter =
1046 (struct ena_adapter *)(dev->data->dev_private);
1050 rc = ena_check_valid_conf(adapter);
1054 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_RX);
1058 rc = ena_queue_restart_all(dev, ENA_RING_TYPE_TX);
1062 if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1063 ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1064 rc = ena_rss_init_default(adapter);
1069 ena_stats_restart(dev);
1071 adapter->timestamp_wd = rte_get_timer_cycles();
1072 adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1074 ticks = rte_get_timer_hz();
1075 rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1076 ena_timer_wd_callback, adapter);
1078 adapter->state = ENA_ADAPTER_STATE_RUNNING;
1083 static void ena_stop(struct rte_eth_dev *dev)
1085 struct ena_adapter *adapter =
1086 (struct ena_adapter *)(dev->data->dev_private);
1088 rte_timer_stop_sync(&adapter->timer_wd);
1089 ena_free_io_queues_all(adapter);
1091 adapter->state = ENA_ADAPTER_STATE_STOPPED;
1094 static int ena_create_io_queue(struct ena_ring *ring)
1096 struct ena_adapter *adapter;
1097 struct ena_com_dev *ena_dev;
1098 struct ena_com_create_io_ctx ctx =
1099 /* policy set to _HOST just to satisfy icc compiler */
1100 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1106 adapter = ring->adapter;
1107 ena_dev = &adapter->ena_dev;
1109 if (ring->type == ENA_RING_TYPE_TX) {
1110 ena_qid = ENA_IO_TXQ_IDX(ring->id);
1111 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1112 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1113 ctx.queue_size = adapter->tx_ring_size;
1114 for (i = 0; i < ring->ring_size; i++)
1115 ring->empty_tx_reqs[i] = i;
1117 ena_qid = ENA_IO_RXQ_IDX(ring->id);
1118 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1119 ctx.queue_size = adapter->rx_ring_size;
1120 for (i = 0; i < ring->ring_size; i++)
1121 ring->empty_rx_reqs[i] = i;
1124 ctx.msix_vector = -1; /* interrupts not used */
1125 ctx.numa_node = ena_cpu_to_node(ring->id);
1127 rc = ena_com_create_io_queue(ena_dev, &ctx);
1130 "failed to create io queue #%d (qid:%d) rc: %d\n",
1131 ring->id, ena_qid, rc);
1135 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1136 &ring->ena_com_io_sq,
1137 &ring->ena_com_io_cq);
1140 "Failed to get io queue handlers. queue num %d rc: %d\n",
1142 ena_com_destroy_io_queue(ena_dev, ena_qid);
1146 if (ring->type == ENA_RING_TYPE_TX)
1147 ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1152 static void ena_free_io_queues_all(struct ena_adapter *adapter)
1154 struct rte_eth_dev *eth_dev = adapter->rte_dev;
1155 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1158 uint16_t nb_rxq = eth_dev->data->nb_rx_queues;
1159 uint16_t nb_txq = eth_dev->data->nb_tx_queues;
1161 for (i = 0; i < nb_txq; ++i) {
1162 ena_qid = ENA_IO_TXQ_IDX(i);
1163 ena_com_destroy_io_queue(ena_dev, ena_qid);
1165 ena_tx_queue_release_bufs(&adapter->tx_ring[i]);
1168 for (i = 0; i < nb_rxq; ++i) {
1169 ena_qid = ENA_IO_RXQ_IDX(i);
1170 ena_com_destroy_io_queue(ena_dev, ena_qid);
1172 ena_rx_queue_release_bufs(&adapter->rx_ring[i]);
1176 static int ena_queue_restart(struct ena_ring *ring)
1180 ena_assert_msg(ring->configured == 1,
1181 "Trying to restart unconfigured queue\n");
1183 rc = ena_create_io_queue(ring);
1185 PMD_INIT_LOG(ERR, "Failed to create IO queue!\n");
1189 ring->next_to_clean = 0;
1190 ring->next_to_use = 0;
1192 if (ring->type == ENA_RING_TYPE_TX)
1195 bufs_num = ring->ring_size - 1;
1196 rc = ena_populate_rx_queue(ring, bufs_num);
1197 if (rc != bufs_num) {
1198 PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1199 return ENA_COM_FAULT;
1205 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1208 __rte_unused unsigned int socket_id,
1209 const struct rte_eth_txconf *tx_conf)
1211 struct ena_ring *txq = NULL;
1212 struct ena_adapter *adapter =
1213 (struct ena_adapter *)(dev->data->dev_private);
1216 txq = &adapter->tx_ring[queue_idx];
1218 if (txq->configured) {
1220 "API violation. Queue %d is already configured\n",
1222 return ENA_COM_FAULT;
1225 if (!rte_is_power_of_2(nb_desc)) {
1227 "Unsupported size of TX queue: %d is not a power of 2.",
1232 if (nb_desc > adapter->tx_ring_size) {
1234 "Unsupported size of TX queue (max size: %d)\n",
1235 adapter->tx_ring_size);
1239 txq->port_id = dev->data->port_id;
1240 txq->next_to_clean = 0;
1241 txq->next_to_use = 0;
1242 txq->ring_size = nb_desc;
1244 txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1245 sizeof(struct ena_tx_buffer) *
1247 RTE_CACHE_LINE_SIZE);
1248 if (!txq->tx_buffer_info) {
1249 RTE_LOG(ERR, PMD, "failed to alloc mem for tx buffer info\n");
1253 txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1254 sizeof(u16) * txq->ring_size,
1255 RTE_CACHE_LINE_SIZE);
1256 if (!txq->empty_tx_reqs) {
1257 RTE_LOG(ERR, PMD, "failed to alloc mem for tx reqs\n");
1258 rte_free(txq->tx_buffer_info);
1262 for (i = 0; i < txq->ring_size; i++)
1263 txq->empty_tx_reqs[i] = i;
1265 if (tx_conf != NULL) {
1267 tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1270 /* Store pointer to this queue in upper layer */
1271 txq->configured = 1;
1272 dev->data->tx_queues[queue_idx] = txq;
1277 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1280 __rte_unused unsigned int socket_id,
1281 __rte_unused const struct rte_eth_rxconf *rx_conf,
1282 struct rte_mempool *mp)
1284 struct ena_adapter *adapter =
1285 (struct ena_adapter *)(dev->data->dev_private);
1286 struct ena_ring *rxq = NULL;
1289 rxq = &adapter->rx_ring[queue_idx];
1290 if (rxq->configured) {
1292 "API violation. Queue %d is already configured\n",
1294 return ENA_COM_FAULT;
1297 if (!rte_is_power_of_2(nb_desc)) {
1299 "Unsupported size of RX queue: %d is not a power of 2.",
1304 if (nb_desc > adapter->rx_ring_size) {
1306 "Unsupported size of RX queue (max size: %d)\n",
1307 adapter->rx_ring_size);
1311 rxq->port_id = dev->data->port_id;
1312 rxq->next_to_clean = 0;
1313 rxq->next_to_use = 0;
1314 rxq->ring_size = nb_desc;
1317 rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1318 sizeof(struct rte_mbuf *) * nb_desc,
1319 RTE_CACHE_LINE_SIZE);
1320 if (!rxq->rx_buffer_info) {
1321 RTE_LOG(ERR, PMD, "failed to alloc mem for rx buffer info\n");
1325 rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1326 sizeof(struct rte_mbuf *) * nb_desc,
1327 RTE_CACHE_LINE_SIZE);
1329 if (!rxq->rx_refill_buffer) {
1330 RTE_LOG(ERR, PMD, "failed to alloc mem for rx refill buffer\n");
1331 rte_free(rxq->rx_buffer_info);
1332 rxq->rx_buffer_info = NULL;
1336 rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1337 sizeof(uint16_t) * nb_desc,
1338 RTE_CACHE_LINE_SIZE);
1339 if (!rxq->empty_rx_reqs) {
1340 RTE_LOG(ERR, PMD, "failed to alloc mem for empty rx reqs\n");
1341 rte_free(rxq->rx_buffer_info);
1342 rxq->rx_buffer_info = NULL;
1343 rte_free(rxq->rx_refill_buffer);
1344 rxq->rx_refill_buffer = NULL;
1348 for (i = 0; i < nb_desc; i++)
1349 rxq->empty_tx_reqs[i] = i;
1351 /* Store pointer to this queue in upper layer */
1352 rxq->configured = 1;
1353 dev->data->rx_queues[queue_idx] = rxq;
1358 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1362 uint16_t ring_size = rxq->ring_size;
1363 uint16_t ring_mask = ring_size - 1;
1364 uint16_t next_to_use = rxq->next_to_use;
1365 uint16_t in_use, req_id;
1366 struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1368 if (unlikely(!count))
1371 in_use = rxq->next_to_use - rxq->next_to_clean;
1372 ena_assert_msg(((in_use + count) < ring_size), "bad ring state");
1374 /* get resources for incoming packets */
1375 rc = rte_mempool_get_bulk(rxq->mb_pool, (void **)mbufs, count);
1376 if (unlikely(rc < 0)) {
1377 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1378 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1382 for (i = 0; i < count; i++) {
1383 uint16_t next_to_use_masked = next_to_use & ring_mask;
1384 struct rte_mbuf *mbuf = mbufs[i];
1385 struct ena_com_buf ebuf;
1387 if (likely((i + 4) < count))
1388 rte_prefetch0(mbufs[i + 4]);
1390 req_id = rxq->empty_rx_reqs[next_to_use_masked];
1391 rc = validate_rx_req_id(rxq, req_id);
1392 if (unlikely(rc < 0))
1394 rxq->rx_buffer_info[req_id] = mbuf;
1396 /* prepare physical address for DMA transaction */
1397 ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1398 ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1399 /* pass resource to device */
1400 rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
1403 RTE_LOG(WARNING, PMD, "failed adding rx desc\n");
1404 rxq->rx_buffer_info[req_id] = NULL;
1410 if (unlikely(i < count)) {
1411 RTE_LOG(WARNING, PMD, "refilled rx qid %d with only %d "
1412 "buffers (from %d)\n", rxq->id, i, count);
1413 rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbufs[i]),
1417 /* When we submitted free recources to device... */
1418 if (likely(i > 0)) {
1419 /* ...let HW know that it can fill buffers with data
1421 * Add memory barrier to make sure the desc were written before
1425 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1427 rxq->next_to_use = next_to_use;
1433 static int ena_device_init(struct ena_com_dev *ena_dev,
1434 struct ena_com_dev_get_features_ctx *get_feat_ctx,
1437 uint32_t aenq_groups;
1439 bool readless_supported;
1441 /* Initialize mmio registers */
1442 rc = ena_com_mmio_reg_read_request_init(ena_dev);
1444 RTE_LOG(ERR, PMD, "failed to init mmio read less\n");
1448 /* The PCIe configuration space revision id indicate if mmio reg
1451 readless_supported =
1452 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1453 & ENA_MMIO_DISABLE_REG_READ);
1454 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1457 rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1459 RTE_LOG(ERR, PMD, "cannot reset device\n");
1460 goto err_mmio_read_less;
1463 /* check FW version */
1464 rc = ena_com_validate_version(ena_dev);
1466 RTE_LOG(ERR, PMD, "device version is too low\n");
1467 goto err_mmio_read_less;
1470 ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1472 /* ENA device administration layer init */
1473 rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
1476 "cannot initialize ena admin queue with device\n");
1477 goto err_mmio_read_less;
1480 /* To enable the msix interrupts the driver needs to know the number
1481 * of queues. So the driver uses polling mode to retrieve this
1484 ena_com_set_admin_polling_mode(ena_dev, true);
1486 ena_config_host_info(ena_dev);
1488 /* Get Device Attributes and features */
1489 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1492 "cannot get attribute for ena device rc= %d\n", rc);
1493 goto err_admin_init;
1496 aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1497 BIT(ENA_ADMIN_NOTIFICATION) |
1498 BIT(ENA_ADMIN_KEEP_ALIVE) |
1499 BIT(ENA_ADMIN_FATAL_ERROR) |
1500 BIT(ENA_ADMIN_WARNING);
1502 aenq_groups &= get_feat_ctx->aenq.supported_groups;
1503 rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1505 RTE_LOG(ERR, PMD, "Cannot configure aenq groups rc: %d\n", rc);
1506 goto err_admin_init;
1509 *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1514 ena_com_admin_destroy(ena_dev);
1517 ena_com_mmio_reg_read_request_destroy(ena_dev);
1522 static void ena_interrupt_handler_rte(void *cb_arg)
1524 struct ena_adapter *adapter = (struct ena_adapter *)cb_arg;
1525 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1527 ena_com_admin_q_comp_intr_handler(ena_dev);
1528 if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1529 ena_com_aenq_intr_handler(ena_dev, adapter);
1532 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1534 if (!adapter->wd_state)
1537 if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1540 if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1541 adapter->keep_alive_timeout)) {
1542 RTE_LOG(ERR, PMD, "Keep alive timeout\n");
1543 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1544 adapter->trigger_reset = true;
1548 /* Check if admin queue is enabled */
1549 static void check_for_admin_com_state(struct ena_adapter *adapter)
1551 if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1552 RTE_LOG(ERR, PMD, "ENA admin queue is not in running state!\n");
1553 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1554 adapter->trigger_reset = true;
1558 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1561 struct ena_adapter *adapter = (struct ena_adapter *)arg;
1562 struct rte_eth_dev *dev = adapter->rte_dev;
1564 check_for_missing_keep_alive(adapter);
1565 check_for_admin_com_state(adapter);
1567 if (unlikely(adapter->trigger_reset)) {
1568 RTE_LOG(ERR, PMD, "Trigger reset is on\n");
1569 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1574 static int ena_calc_io_queue_num(__rte_unused struct ena_com_dev *ena_dev,
1575 struct ena_com_dev_get_features_ctx *get_feat_ctx)
1577 int io_sq_num, io_cq_num, io_queue_num;
1579 io_sq_num = get_feat_ctx->max_queues.max_sq_num;
1580 io_cq_num = get_feat_ctx->max_queues.max_cq_num;
1582 io_queue_num = RTE_MIN(io_sq_num, io_cq_num);
1584 if (unlikely(io_queue_num == 0)) {
1585 RTE_LOG(ERR, PMD, "Number of IO queues should not be 0\n");
1589 return io_queue_num;
1592 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1594 struct rte_pci_device *pci_dev;
1595 struct rte_intr_handle *intr_handle;
1596 struct ena_adapter *adapter =
1597 (struct ena_adapter *)(eth_dev->data->dev_private);
1598 struct ena_com_dev *ena_dev = &adapter->ena_dev;
1599 struct ena_com_dev_get_features_ctx get_feat_ctx;
1601 u16 tx_sgl_size = 0;
1603 static int adapters_found;
1606 memset(adapter, 0, sizeof(struct ena_adapter));
1607 ena_dev = &adapter->ena_dev;
1609 eth_dev->dev_ops = &ena_dev_ops;
1610 eth_dev->rx_pkt_burst = ð_ena_recv_pkts;
1611 eth_dev->tx_pkt_burst = ð_ena_xmit_pkts;
1612 eth_dev->tx_pkt_prepare = ð_ena_prep_pkts;
1613 adapter->rte_eth_dev_data = eth_dev->data;
1614 adapter->rte_dev = eth_dev;
1616 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1619 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1620 adapter->pdev = pci_dev;
1622 PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1623 pci_dev->addr.domain,
1625 pci_dev->addr.devid,
1626 pci_dev->addr.function);
1628 intr_handle = &pci_dev->intr_handle;
1630 adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1631 adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1633 if (!adapter->regs) {
1634 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1639 ena_dev->reg_bar = adapter->regs;
1640 ena_dev->dmadev = adapter->pdev;
1642 adapter->id_number = adapters_found;
1644 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1645 adapter->id_number);
1647 /* device specific initialization routine */
1648 rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1650 PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1653 adapter->wd_state = wd_state;
1655 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1656 adapter->num_queues = ena_calc_io_queue_num(ena_dev,
1659 queue_size = ena_calc_queue_size(ena_dev, &tx_sgl_size, &get_feat_ctx);
1660 if (queue_size <= 0 || adapter->num_queues <= 0) {
1662 goto err_device_destroy;
1665 adapter->tx_ring_size = queue_size;
1666 adapter->rx_ring_size = queue_size;
1668 adapter->max_tx_sgl_size = tx_sgl_size;
1670 /* prepare ring structures */
1671 ena_init_rings(adapter);
1673 ena_config_debug_area(adapter);
1675 /* Set max MTU for this device */
1676 adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1678 /* set device support for TSO */
1679 adapter->tso4_supported = get_feat_ctx.offload.tx &
1680 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK;
1682 /* Copy MAC address and point DPDK to it */
1683 eth_dev->data->mac_addrs = (struct ether_addr *)adapter->mac_addr;
1684 ether_addr_copy((struct ether_addr *)get_feat_ctx.dev_attr.mac_addr,
1685 (struct ether_addr *)adapter->mac_addr);
1688 * Pass the information to the rte_eth_dev_close() that it should also
1689 * release the private port resources.
1691 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1693 adapter->drv_stats = rte_zmalloc("adapter stats",
1694 sizeof(*adapter->drv_stats),
1695 RTE_CACHE_LINE_SIZE);
1696 if (!adapter->drv_stats) {
1697 RTE_LOG(ERR, PMD, "failed to alloc mem for adapter stats\n");
1699 goto err_delete_debug_area;
1702 rte_intr_callback_register(intr_handle,
1703 ena_interrupt_handler_rte,
1705 rte_intr_enable(intr_handle);
1706 ena_com_set_admin_polling_mode(ena_dev, false);
1707 ena_com_admin_aenq_enable(ena_dev);
1709 if (adapters_found == 0)
1710 rte_timer_subsystem_init();
1711 rte_timer_init(&adapter->timer_wd);
1714 adapter->state = ENA_ADAPTER_STATE_INIT;
1718 err_delete_debug_area:
1719 ena_com_delete_debug_area(ena_dev);
1722 ena_com_delete_host_info(ena_dev);
1723 ena_com_admin_destroy(ena_dev);
1729 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1731 struct ena_adapter *adapter =
1732 (struct ena_adapter *)(eth_dev->data->dev_private);
1734 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1737 if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1740 eth_dev->dev_ops = NULL;
1741 eth_dev->rx_pkt_burst = NULL;
1742 eth_dev->tx_pkt_burst = NULL;
1743 eth_dev->tx_pkt_prepare = NULL;
1745 adapter->state = ENA_ADAPTER_STATE_FREE;
1750 static int ena_dev_configure(struct rte_eth_dev *dev)
1752 struct ena_adapter *adapter =
1753 (struct ena_adapter *)(dev->data->dev_private);
1755 adapter->state = ENA_ADAPTER_STATE_CONFIG;
1757 adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1758 adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1762 static void ena_init_rings(struct ena_adapter *adapter)
1766 for (i = 0; i < adapter->num_queues; i++) {
1767 struct ena_ring *ring = &adapter->tx_ring[i];
1769 ring->configured = 0;
1770 ring->type = ENA_RING_TYPE_TX;
1771 ring->adapter = adapter;
1773 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1774 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1775 ring->sgl_size = adapter->max_tx_sgl_size;
1778 for (i = 0; i < adapter->num_queues; i++) {
1779 struct ena_ring *ring = &adapter->rx_ring[i];
1781 ring->configured = 0;
1782 ring->type = ENA_RING_TYPE_RX;
1783 ring->adapter = adapter;
1788 static void ena_infos_get(struct rte_eth_dev *dev,
1789 struct rte_eth_dev_info *dev_info)
1791 struct ena_adapter *adapter;
1792 struct ena_com_dev *ena_dev;
1793 struct ena_com_dev_get_features_ctx feat;
1794 uint64_t rx_feat = 0, tx_feat = 0;
1797 ena_assert_msg(dev->data != NULL, "Uninitialized device");
1798 ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device");
1799 adapter = (struct ena_adapter *)(dev->data->dev_private);
1801 ena_dev = &adapter->ena_dev;
1802 ena_assert_msg(ena_dev != NULL, "Uninitialized device");
1804 dev_info->speed_capa =
1806 ETH_LINK_SPEED_2_5G |
1808 ETH_LINK_SPEED_10G |
1809 ETH_LINK_SPEED_25G |
1810 ETH_LINK_SPEED_40G |
1811 ETH_LINK_SPEED_50G |
1812 ETH_LINK_SPEED_100G;
1814 /* Get supported features from HW */
1815 rc = ena_com_get_dev_attr_feat(ena_dev, &feat);
1818 "Cannot get attribute for ena device rc= %d\n", rc);
1822 /* Set Tx & Rx features available for device */
1823 if (feat.offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
1824 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
1826 if (feat.offload.tx &
1827 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
1828 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
1829 DEV_TX_OFFLOAD_UDP_CKSUM |
1830 DEV_TX_OFFLOAD_TCP_CKSUM;
1832 if (feat.offload.rx_supported &
1833 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
1834 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
1835 DEV_RX_OFFLOAD_UDP_CKSUM |
1836 DEV_RX_OFFLOAD_TCP_CKSUM;
1838 rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1840 /* Inform framework about available features */
1841 dev_info->rx_offload_capa = rx_feat;
1842 dev_info->rx_queue_offload_capa = rx_feat;
1843 dev_info->tx_offload_capa = tx_feat;
1844 dev_info->tx_queue_offload_capa = tx_feat;
1846 dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
1847 dev_info->max_rx_pktlen = adapter->max_mtu;
1848 dev_info->max_mac_addrs = 1;
1850 dev_info->max_rx_queues = adapter->num_queues;
1851 dev_info->max_tx_queues = adapter->num_queues;
1852 dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
1854 adapter->tx_supported_offloads = tx_feat;
1855 adapter->rx_supported_offloads = rx_feat;
1857 dev_info->rx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1858 dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1860 dev_info->tx_desc_lim.nb_max = ENA_MAX_RING_DESC;
1861 dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
1862 dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1863 feat.max_queues.max_packet_tx_descs);
1864 dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
1865 feat.max_queues.max_packet_tx_descs);
1868 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
1871 struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
1872 unsigned int ring_size = rx_ring->ring_size;
1873 unsigned int ring_mask = ring_size - 1;
1874 uint16_t next_to_clean = rx_ring->next_to_clean;
1875 uint16_t desc_in_use = 0;
1877 unsigned int recv_idx = 0;
1878 struct rte_mbuf *mbuf = NULL;
1879 struct rte_mbuf *mbuf_head = NULL;
1880 struct rte_mbuf *mbuf_prev = NULL;
1881 struct rte_mbuf **rx_buff_info = rx_ring->rx_buffer_info;
1882 unsigned int completed;
1884 struct ena_com_rx_ctx ena_rx_ctx;
1887 /* Check adapter state */
1888 if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
1890 "Trying to receive pkts while device is NOT running\n");
1894 desc_in_use = rx_ring->next_to_use - next_to_clean;
1895 if (unlikely(nb_pkts > desc_in_use))
1896 nb_pkts = desc_in_use;
1898 for (completed = 0; completed < nb_pkts; completed++) {
1901 ena_rx_ctx.max_bufs = rx_ring->ring_size;
1902 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1903 ena_rx_ctx.descs = 0;
1904 /* receive packet context */
1905 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1906 rx_ring->ena_com_io_sq,
1909 RTE_LOG(ERR, PMD, "ena_com_rx_pkt error %d\n", rc);
1910 rx_ring->adapter->trigger_reset = true;
1914 if (unlikely(ena_rx_ctx.descs == 0))
1917 while (segments < ena_rx_ctx.descs) {
1918 req_id = ena_rx_ctx.ena_bufs[segments].req_id;
1919 rc = validate_rx_req_id(rx_ring, req_id);
1923 mbuf = rx_buff_info[req_id];
1924 mbuf->data_len = ena_rx_ctx.ena_bufs[segments].len;
1925 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
1928 if (unlikely(segments == 0)) {
1929 mbuf->nb_segs = ena_rx_ctx.descs;
1930 mbuf->port = rx_ring->port_id;
1934 /* for multi-segment pkts create mbuf chain */
1935 mbuf_prev->next = mbuf;
1937 mbuf_head->pkt_len += mbuf->data_len;
1940 rx_ring->empty_rx_reqs[next_to_clean & ring_mask] =
1946 /* fill mbuf attributes if any */
1947 ena_rx_mbuf_prepare(mbuf_head, &ena_rx_ctx);
1948 mbuf_head->hash.rss = ena_rx_ctx.hash;
1950 /* pass to DPDK application head mbuf */
1951 rx_pkts[recv_idx] = mbuf_head;
1955 rx_ring->next_to_clean = next_to_clean;
1957 desc_in_use = desc_in_use - completed + 1;
1958 /* Burst refill to save doorbells, memory barriers, const interval */
1959 if (ring_size - desc_in_use > ENA_RING_DESCS_RATIO(ring_size))
1960 ena_populate_rx_queue(rx_ring, ring_size - desc_in_use);
1966 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
1972 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
1973 struct ipv4_hdr *ip_hdr;
1975 uint16_t frag_field;
1977 for (i = 0; i != nb_pkts; i++) {
1979 ol_flags = m->ol_flags;
1981 if (!(ol_flags & PKT_TX_IPV4))
1984 /* If there was not L2 header length specified, assume it is
1985 * length of the ethernet header.
1987 if (unlikely(m->l2_len == 0))
1988 m->l2_len = sizeof(struct ether_hdr);
1990 ip_hdr = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
1992 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
1994 if ((frag_field & IPV4_HDR_DF_FLAG) != 0) {
1995 m->packet_type |= RTE_PTYPE_L4_NONFRAG;
1997 /* If IPv4 header has DF flag enabled and TSO support is
1998 * disabled, partial chcecksum should not be calculated.
2000 if (!tx_ring->adapter->tso4_supported)
2004 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2005 (ol_flags & PKT_TX_L4_MASK) ==
2006 PKT_TX_SCTP_CKSUM) {
2007 rte_errno = -ENOTSUP;
2011 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2012 ret = rte_validate_tx_offload(m);
2019 /* In case we are supposed to TSO and have DF not set (DF=0)
2020 * hardware must be provided with partial checksum, otherwise
2021 * it will take care of necessary calculations.
2024 ret = rte_net_intel_cksum_flags_prepare(m,
2025 ol_flags & ~PKT_TX_TCP_SEG);
2035 static void ena_update_hints(struct ena_adapter *adapter,
2036 struct ena_admin_ena_hw_hints *hints)
2038 if (hints->admin_completion_tx_timeout)
2039 adapter->ena_dev.admin_queue.completion_timeout =
2040 hints->admin_completion_tx_timeout * 1000;
2042 if (hints->mmio_read_timeout)
2043 /* convert to usec */
2044 adapter->ena_dev.mmio_read.reg_read_to =
2045 hints->mmio_read_timeout * 1000;
2047 if (hints->driver_watchdog_timeout) {
2048 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2049 adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2051 // Convert msecs to ticks
2052 adapter->keep_alive_timeout =
2053 (hints->driver_watchdog_timeout *
2054 rte_get_timer_hz()) / 1000;
2058 static int ena_check_and_linearize_mbuf(struct ena_ring *tx_ring,
2059 struct rte_mbuf *mbuf)
2061 int num_segments, rc;
2063 num_segments = mbuf->nb_segs;
2065 if (likely(num_segments < tx_ring->sgl_size))
2068 rc = rte_pktmbuf_linearize(mbuf);
2070 RTE_LOG(WARNING, PMD, "Mbuf linearize failed\n");
2075 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2078 struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2079 uint16_t next_to_use = tx_ring->next_to_use;
2080 uint16_t next_to_clean = tx_ring->next_to_clean;
2081 struct rte_mbuf *mbuf;
2082 unsigned int ring_size = tx_ring->ring_size;
2083 unsigned int ring_mask = ring_size - 1;
2084 struct ena_com_tx_ctx ena_tx_ctx;
2085 struct ena_tx_buffer *tx_info;
2086 struct ena_com_buf *ebuf;
2087 uint16_t rc, req_id, total_tx_descs = 0;
2088 uint16_t sent_idx = 0, empty_tx_reqs;
2091 /* Check adapter state */
2092 if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2094 "Trying to xmit pkts while device is NOT running\n");
2098 empty_tx_reqs = ring_size - (next_to_use - next_to_clean);
2099 if (nb_pkts > empty_tx_reqs)
2100 nb_pkts = empty_tx_reqs;
2102 for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2103 mbuf = tx_pkts[sent_idx];
2105 rc = ena_check_and_linearize_mbuf(tx_ring, mbuf);
2109 req_id = tx_ring->empty_tx_reqs[next_to_use & ring_mask];
2110 tx_info = &tx_ring->tx_buffer_info[req_id];
2111 tx_info->mbuf = mbuf;
2112 tx_info->num_of_bufs = 0;
2113 ebuf = tx_info->bufs;
2115 /* Prepare TX context */
2116 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2117 memset(&ena_tx_ctx.ena_meta, 0x0,
2118 sizeof(struct ena_com_tx_meta));
2119 ena_tx_ctx.ena_bufs = ebuf;
2120 ena_tx_ctx.req_id = req_id;
2121 if (tx_ring->tx_mem_queue_type ==
2122 ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2123 /* prepare the push buffer with
2124 * virtual address of the data
2126 ena_tx_ctx.header_len =
2127 RTE_MIN(mbuf->data_len,
2128 tx_ring->tx_max_header_size);
2129 ena_tx_ctx.push_header =
2130 (void *)((char *)mbuf->buf_addr +
2132 } /* there's no else as we take advantage of memset zeroing */
2134 /* Set TX offloads flags, if applicable */
2135 ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads);
2137 if (unlikely(mbuf->ol_flags &
2138 (PKT_RX_L4_CKSUM_BAD | PKT_RX_IP_CKSUM_BAD)))
2139 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2141 rte_prefetch0(tx_pkts[(sent_idx + 4) & ring_mask]);
2143 /* Process first segment taking into
2144 * consideration pushed header
2146 if (mbuf->data_len > ena_tx_ctx.header_len) {
2147 ebuf->paddr = mbuf->buf_iova +
2149 ena_tx_ctx.header_len;
2150 ebuf->len = mbuf->data_len - ena_tx_ctx.header_len;
2152 tx_info->num_of_bufs++;
2155 while ((mbuf = mbuf->next) != NULL) {
2156 ebuf->paddr = mbuf->buf_iova + mbuf->data_off;
2157 ebuf->len = mbuf->data_len;
2159 tx_info->num_of_bufs++;
2162 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2164 /* Write data to device */
2165 rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq,
2166 &ena_tx_ctx, &nb_hw_desc);
2170 tx_info->tx_descs = nb_hw_desc;
2175 /* If there are ready packets to be xmitted... */
2177 /* ...let HW do its best :-) */
2179 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2181 tx_ring->next_to_use = next_to_use;
2184 /* Clear complete packets */
2185 while (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) >= 0) {
2186 rc = validate_tx_req_id(tx_ring, req_id);
2190 /* Get Tx info & store how many descs were processed */
2191 tx_info = &tx_ring->tx_buffer_info[req_id];
2192 total_tx_descs += tx_info->tx_descs;
2194 /* Free whole mbuf chain */
2195 mbuf = tx_info->mbuf;
2196 rte_pktmbuf_free(mbuf);
2197 tx_info->mbuf = NULL;
2199 /* Put back descriptor to the ring for reuse */
2200 tx_ring->empty_tx_reqs[next_to_clean & ring_mask] = req_id;
2203 /* If too many descs to clean, leave it for another run */
2204 if (unlikely(total_tx_descs > ENA_RING_DESCS_RATIO(ring_size)))
2208 if (total_tx_descs > 0) {
2209 /* acknowledge completion of sent packets */
2210 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2211 tx_ring->next_to_clean = next_to_clean;
2217 /*********************************************************************
2219 *********************************************************************/
2220 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2221 struct rte_pci_device *pci_dev)
2223 return rte_eth_dev_pci_generic_probe(pci_dev,
2224 sizeof(struct ena_adapter), eth_ena_dev_init);
2227 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2229 return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2232 static struct rte_pci_driver rte_ena_pmd = {
2233 .id_table = pci_id_ena_map,
2234 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2235 RTE_PCI_DRV_WC_ACTIVATE,
2236 .probe = eth_ena_pci_probe,
2237 .remove = eth_ena_pci_remove,
2240 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2241 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2242 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2244 RTE_INIT(ena_init_log)
2246 ena_logtype_init = rte_log_register("pmd.net.ena.init");
2247 if (ena_logtype_init >= 0)
2248 rte_log_set_level(ena_logtype_init, RTE_LOG_NOTICE);
2249 ena_logtype_driver = rte_log_register("pmd.net.ena.driver");
2250 if (ena_logtype_driver >= 0)
2251 rte_log_set_level(ena_logtype_driver, RTE_LOG_NOTICE);
2254 /******************************************************************************
2255 ******************************** AENQ Handlers *******************************
2256 *****************************************************************************/
2257 static void ena_update_on_link_change(void *adapter_data,
2258 struct ena_admin_aenq_entry *aenq_e)
2260 struct rte_eth_dev *eth_dev;
2261 struct ena_adapter *adapter;
2262 struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2265 adapter = (struct ena_adapter *)adapter_data;
2266 aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2267 eth_dev = adapter->rte_dev;
2269 status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2270 adapter->link_status = status;
2272 ena_link_update(eth_dev, 0);
2273 _rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2276 static void ena_notification(void *data,
2277 struct ena_admin_aenq_entry *aenq_e)
2279 struct ena_adapter *adapter = (struct ena_adapter *)data;
2280 struct ena_admin_ena_hw_hints *hints;
2282 if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2283 RTE_LOG(WARNING, PMD, "Invalid group(%x) expected %x\n",
2284 aenq_e->aenq_common_desc.group,
2285 ENA_ADMIN_NOTIFICATION);
2287 switch (aenq_e->aenq_common_desc.syndrom) {
2288 case ENA_ADMIN_UPDATE_HINTS:
2289 hints = (struct ena_admin_ena_hw_hints *)
2290 (&aenq_e->inline_data_w4);
2291 ena_update_hints(adapter, hints);
2294 RTE_LOG(ERR, PMD, "Invalid aenq notification link state %d\n",
2295 aenq_e->aenq_common_desc.syndrom);
2299 static void ena_keep_alive(void *adapter_data,
2300 __rte_unused struct ena_admin_aenq_entry *aenq_e)
2302 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
2304 adapter->timestamp_wd = rte_get_timer_cycles();
2308 * This handler will called for unknown event group or unimplemented handlers
2310 static void unimplemented_aenq_handler(__rte_unused void *data,
2311 __rte_unused struct ena_admin_aenq_entry *aenq_e)
2313 RTE_LOG(ERR, PMD, "Unknown event was received or event with "
2314 "unimplemented handler\n");
2317 static struct ena_aenq_handlers aenq_handlers = {
2319 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
2320 [ENA_ADMIN_NOTIFICATION] = ena_notification,
2321 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
2323 .unimplemented_handler = unimplemented_aenq_handler