1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
3 * Copyright 2015-2020 Amazon.com, Inc. or its affiliates. All rights reserved.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #ifdef CONFIG_RFS_ACCEL
9 #include <linux/cpu_rmap.h>
10 #endif /* CONFIG_RFS_ACCEL */
11 #include <linux/ethtool.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/numa.h>
15 #include <linux/pci.h>
16 #include <linux/utsname.h>
17 #include <linux/version.h>
18 #include <linux/vmalloc.h>
21 #include "ena_netdev.h"
22 #include <linux/bpf_trace.h>
23 #include "ena_pci_id_tbl.h"
25 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
26 MODULE_DESCRIPTION(DEVICE_NAME);
27 MODULE_LICENSE("GPL");
29 /* Time in jiffies before concluding the transmitter is hung. */
30 #define TX_TIMEOUT (5 * HZ)
32 #define ENA_MAX_RINGS min_t(unsigned int, ENA_MAX_NUM_IO_QUEUES, num_possible_cpus())
34 #define ENA_NAPI_BUDGET 64
36 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
37 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
38 static int debug = -1;
39 module_param(debug, int, 0);
40 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
42 static struct ena_aenq_handlers aenq_handlers;
44 static struct workqueue_struct *ena_wq;
46 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
48 static int ena_rss_init_default(struct ena_adapter *adapter);
49 static void check_for_admin_com_state(struct ena_adapter *adapter);
50 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
51 static int ena_restore_device(struct ena_adapter *adapter);
53 static void ena_init_io_rings(struct ena_adapter *adapter,
54 int first_index, int count);
55 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
57 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
59 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
60 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
63 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
64 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
65 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
66 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
67 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
68 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
69 int first_index, int count);
70 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
71 int first_index, int count);
72 static int ena_up(struct ena_adapter *adapter);
73 static void ena_down(struct ena_adapter *adapter);
74 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
75 struct ena_ring *rx_ring);
76 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
77 struct ena_ring *rx_ring);
78 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
79 struct ena_tx_buffer *tx_info);
80 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
81 int first_index, int count);
83 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
84 static void ena_increase_stat(u64 *statp, u64 cnt,
85 struct u64_stats_sync *syncp)
87 u64_stats_update_begin(syncp);
89 u64_stats_update_end(syncp);
92 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
94 struct ena_adapter *adapter = netdev_priv(dev);
96 /* Change the state of the device to trigger reset
97 * Check that we are not in the middle or a trigger already
100 if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
103 adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
104 ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
106 netif_err(adapter, tx_err, dev, "Transmit time out\n");
109 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
113 for (i = 0; i < adapter->num_io_queues; i++)
114 adapter->rx_ring[i].mtu = mtu;
117 static int ena_change_mtu(struct net_device *dev, int new_mtu)
119 struct ena_adapter *adapter = netdev_priv(dev);
122 ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
124 netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
125 update_rx_ring_mtu(adapter, new_mtu);
128 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
135 static int ena_xmit_common(struct net_device *dev,
136 struct ena_ring *ring,
137 struct ena_tx_buffer *tx_info,
138 struct ena_com_tx_ctx *ena_tx_ctx,
142 struct ena_adapter *adapter = netdev_priv(dev);
145 if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
147 netif_dbg(adapter, tx_queued, dev,
148 "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
150 ena_com_write_sq_doorbell(ring->ena_com_io_sq);
153 /* prepare the packet's descriptors to dma engine */
154 rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
157 /* In case there isn't enough space in the queue for the packet,
158 * we simply drop it. All other failure reasons of
159 * ena_com_prepare_tx() are fatal and therefore require a device reset.
162 netif_err(adapter, tx_queued, dev,
163 "Failed to prepare tx bufs\n");
164 ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
167 adapter->reset_reason =
168 ENA_REGS_RESET_DRIVER_INVALID_STATE;
169 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
174 u64_stats_update_begin(&ring->syncp);
175 ring->tx_stats.cnt++;
176 ring->tx_stats.bytes += bytes;
177 u64_stats_update_end(&ring->syncp);
179 tx_info->tx_descs = nb_hw_desc;
180 tx_info->last_jiffies = jiffies;
181 tx_info->print_once = 0;
183 ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
188 /* This is the XDP napi callback. XDP queues use a separate napi callback
191 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
193 struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
194 u32 xdp_work_done, xdp_budget;
195 struct ena_ring *xdp_ring;
196 int napi_comp_call = 0;
199 xdp_ring = ena_napi->xdp_ring;
200 xdp_ring->first_interrupt = ena_napi->first_interrupt;
204 if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
205 test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
206 napi_complete_done(napi, 0);
210 xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
212 /* If the device is about to reset or down, avoid unmask
213 * the interrupt and return 0 so NAPI won't reschedule
215 if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
216 napi_complete_done(napi, 0);
218 } else if (xdp_budget > xdp_work_done) {
220 if (napi_complete_done(napi, xdp_work_done))
221 ena_unmask_interrupt(xdp_ring, NULL);
222 ena_update_ring_numa_node(xdp_ring, NULL);
228 u64_stats_update_begin(&xdp_ring->syncp);
229 xdp_ring->tx_stats.napi_comp += napi_comp_call;
230 xdp_ring->tx_stats.tx_poll++;
231 u64_stats_update_end(&xdp_ring->syncp);
236 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
237 struct ena_tx_buffer *tx_info,
238 struct xdp_frame *xdpf,
239 struct ena_com_tx_ctx *ena_tx_ctx)
241 struct ena_adapter *adapter = xdp_ring->adapter;
242 struct ena_com_buf *ena_buf;
248 tx_info->xdpf = xdpf;
249 data = tx_info->xdpf->data;
250 size = tx_info->xdpf->len;
252 if (xdp_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
253 /* Designate part of the packet for LLQ */
254 push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
256 ena_tx_ctx->push_header = data;
262 ena_tx_ctx->header_len = push_len;
265 dma = dma_map_single(xdp_ring->dev,
269 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
270 goto error_report_dma_error;
272 tx_info->map_linear_data = 0;
274 ena_buf = tx_info->bufs;
275 ena_buf->paddr = dma;
278 ena_tx_ctx->ena_bufs = ena_buf;
279 ena_tx_ctx->num_bufs = tx_info->num_of_bufs = 1;
284 error_report_dma_error:
285 ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
287 netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
292 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
293 struct net_device *dev,
294 struct xdp_frame *xdpf,
297 struct ena_com_tx_ctx ena_tx_ctx = {};
298 struct ena_tx_buffer *tx_info;
299 u16 next_to_use, req_id;
302 next_to_use = xdp_ring->next_to_use;
303 req_id = xdp_ring->free_ids[next_to_use];
304 tx_info = &xdp_ring->tx_buffer_info[req_id];
305 tx_info->num_of_bufs = 0;
307 rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &ena_tx_ctx);
311 ena_tx_ctx.req_id = req_id;
313 rc = ena_xmit_common(dev,
320 goto error_unmap_dma;
321 /* trigger the dma engine. ena_com_write_sq_doorbell()
324 if (flags & XDP_XMIT_FLUSH) {
325 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
326 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
333 ena_unmap_tx_buff(xdp_ring, tx_info);
334 tx_info->xdpf = NULL;
338 static int ena_xdp_xmit(struct net_device *dev, int n,
339 struct xdp_frame **frames, u32 flags)
341 struct ena_adapter *adapter = netdev_priv(dev);
342 struct ena_ring *xdp_ring;
343 int qid, i, nxmit = 0;
345 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
348 if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
351 /* We assume that all rings have the same XDP program */
352 if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
355 qid = smp_processor_id() % adapter->xdp_num_queues;
356 qid += adapter->xdp_first_ring;
357 xdp_ring = &adapter->tx_ring[qid];
359 /* Other CPU ids might try to send thorugh this queue */
360 spin_lock(&xdp_ring->xdp_tx_lock);
362 for (i = 0; i < n; i++) {
363 if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
368 /* Ring doorbell to make device aware of the packets */
369 if (flags & XDP_XMIT_FLUSH) {
370 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
371 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
375 spin_unlock(&xdp_ring->xdp_tx_lock);
377 /* Return number of packets sent */
381 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
383 struct bpf_prog *xdp_prog;
384 struct ena_ring *xdp_ring;
385 u32 verdict = XDP_PASS;
386 struct xdp_frame *xdpf;
391 xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
396 verdict = bpf_prog_run_xdp(xdp_prog, xdp);
400 xdpf = xdp_convert_buff_to_frame(xdp);
401 if (unlikely(!xdpf)) {
402 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
403 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
404 verdict = XDP_ABORTED;
408 /* Find xmit queue */
409 qid = rx_ring->qid + rx_ring->adapter->num_io_queues;
410 xdp_ring = &rx_ring->adapter->tx_ring[qid];
412 /* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
413 spin_lock(&xdp_ring->xdp_tx_lock);
415 if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
417 xdp_return_frame(xdpf);
419 spin_unlock(&xdp_ring->xdp_tx_lock);
420 xdp_stat = &rx_ring->rx_stats.xdp_tx;
423 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
424 xdp_stat = &rx_ring->rx_stats.xdp_redirect;
427 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
428 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
429 verdict = XDP_ABORTED;
432 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
433 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
436 xdp_stat = &rx_ring->rx_stats.xdp_drop;
439 xdp_stat = &rx_ring->rx_stats.xdp_pass;
442 bpf_warn_invalid_xdp_action(verdict);
443 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
446 ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
453 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
455 adapter->xdp_first_ring = adapter->num_io_queues;
456 adapter->xdp_num_queues = adapter->num_io_queues;
458 ena_init_io_rings(adapter,
459 adapter->xdp_first_ring,
460 adapter->xdp_num_queues);
463 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
467 rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
468 adapter->xdp_num_queues);
472 rc = ena_create_io_tx_queues_in_range(adapter,
473 adapter->xdp_first_ring,
474 adapter->xdp_num_queues);
481 ena_free_all_io_tx_resources(adapter);
486 /* Provides a way for both kernel and bpf-prog to know
487 * more about the RX-queue a given XDP frame arrived on.
489 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
493 rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
496 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
497 "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
502 rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
506 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
507 "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
509 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
516 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
518 xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
519 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
522 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
523 struct bpf_prog *prog,
524 int first, int count)
526 struct ena_ring *rx_ring;
529 for (i = first; i < count; i++) {
530 rx_ring = &adapter->rx_ring[i];
531 xchg(&rx_ring->xdp_bpf_prog, prog);
533 ena_xdp_register_rxq_info(rx_ring);
534 rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
536 ena_xdp_unregister_rxq_info(rx_ring);
537 rx_ring->rx_headroom = 0;
542 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
543 struct bpf_prog *prog)
545 struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
547 ena_xdp_exchange_program_rx_in_range(adapter,
550 adapter->num_io_queues);
553 bpf_prog_put(old_bpf_prog);
556 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
561 was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
566 adapter->xdp_first_ring = 0;
567 adapter->xdp_num_queues = 0;
568 ena_xdp_exchange_program(adapter, NULL);
570 rc = ena_up(adapter);
577 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
579 struct ena_adapter *adapter = netdev_priv(netdev);
580 struct bpf_prog *prog = bpf->prog;
581 struct bpf_prog *old_bpf_prog;
585 is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
586 rc = ena_xdp_allowed(adapter);
587 if (rc == ENA_XDP_ALLOWED) {
588 old_bpf_prog = adapter->xdp_bpf_prog;
591 ena_init_all_xdp_queues(adapter);
592 } else if (!old_bpf_prog) {
594 ena_init_all_xdp_queues(adapter);
596 ena_xdp_exchange_program(adapter, prog);
598 if (is_up && !old_bpf_prog) {
599 rc = ena_up(adapter);
603 } else if (old_bpf_prog) {
604 rc = ena_destroy_and_free_all_xdp_queues(adapter);
609 prev_mtu = netdev->max_mtu;
610 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
613 netif_info(adapter, drv, adapter->netdev,
614 "XDP program is set, changing the max_mtu from %d to %d",
615 prev_mtu, netdev->max_mtu);
617 } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
618 netif_err(adapter, drv, adapter->netdev,
619 "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
620 netdev->mtu, ENA_XDP_MAX_MTU);
621 NL_SET_ERR_MSG_MOD(bpf->extack,
622 "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
624 } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
625 netif_err(adapter, drv, adapter->netdev,
626 "Failed to set xdp program, the Rx/Tx channel count should be at most half of the maximum allowed channel count. The current queue count (%d), the maximal queue count (%d)\n",
627 adapter->num_io_queues, adapter->max_num_io_queues);
628 NL_SET_ERR_MSG_MOD(bpf->extack,
629 "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
636 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
637 * program as well as to query the current xdp program id.
639 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
641 switch (bpf->command) {
643 return ena_xdp_set(netdev, bpf);
650 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
652 #ifdef CONFIG_RFS_ACCEL
656 adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
657 if (!adapter->netdev->rx_cpu_rmap)
659 for (i = 0; i < adapter->num_io_queues; i++) {
660 int irq_idx = ENA_IO_IRQ_IDX(i);
662 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
663 pci_irq_vector(adapter->pdev, irq_idx));
665 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
666 adapter->netdev->rx_cpu_rmap = NULL;
670 #endif /* CONFIG_RFS_ACCEL */
674 static void ena_init_io_rings_common(struct ena_adapter *adapter,
675 struct ena_ring *ring, u16 qid)
678 ring->pdev = adapter->pdev;
679 ring->dev = &adapter->pdev->dev;
680 ring->netdev = adapter->netdev;
681 ring->napi = &adapter->ena_napi[qid].napi;
682 ring->adapter = adapter;
683 ring->ena_dev = adapter->ena_dev;
684 ring->per_napi_packets = 0;
686 ring->first_interrupt = false;
687 ring->no_interrupt_event_cnt = 0;
688 u64_stats_init(&ring->syncp);
691 static void ena_init_io_rings(struct ena_adapter *adapter,
692 int first_index, int count)
694 struct ena_com_dev *ena_dev;
695 struct ena_ring *txr, *rxr;
698 ena_dev = adapter->ena_dev;
700 for (i = first_index; i < first_index + count; i++) {
701 txr = &adapter->tx_ring[i];
702 rxr = &adapter->rx_ring[i];
704 /* TX common ring state */
705 ena_init_io_rings_common(adapter, txr, i);
707 /* TX specific ring state */
708 txr->ring_size = adapter->requested_tx_ring_size;
709 txr->tx_max_header_size = ena_dev->tx_max_header_size;
710 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
711 txr->sgl_size = adapter->max_tx_sgl_size;
712 txr->smoothed_interval =
713 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
714 txr->disable_meta_caching = adapter->disable_meta_caching;
715 spin_lock_init(&txr->xdp_tx_lock);
717 /* Don't init RX queues for xdp queues */
718 if (!ENA_IS_XDP_INDEX(adapter, i)) {
719 /* RX common ring state */
720 ena_init_io_rings_common(adapter, rxr, i);
722 /* RX specific ring state */
723 rxr->ring_size = adapter->requested_rx_ring_size;
724 rxr->rx_copybreak = adapter->rx_copybreak;
725 rxr->sgl_size = adapter->max_rx_sgl_size;
726 rxr->smoothed_interval =
727 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
728 rxr->empty_rx_queue = 0;
729 adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
734 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
735 * @adapter: network interface device structure
738 * Return 0 on success, negative on failure
740 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
742 struct ena_ring *tx_ring = &adapter->tx_ring[qid];
743 struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
746 if (tx_ring->tx_buffer_info) {
747 netif_err(adapter, ifup,
748 adapter->netdev, "tx_buffer_info info is not NULL");
752 size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
753 node = cpu_to_node(ena_irq->cpu);
755 tx_ring->tx_buffer_info = vzalloc_node(size, node);
756 if (!tx_ring->tx_buffer_info) {
757 tx_ring->tx_buffer_info = vzalloc(size);
758 if (!tx_ring->tx_buffer_info)
759 goto err_tx_buffer_info;
762 size = sizeof(u16) * tx_ring->ring_size;
763 tx_ring->free_ids = vzalloc_node(size, node);
764 if (!tx_ring->free_ids) {
765 tx_ring->free_ids = vzalloc(size);
766 if (!tx_ring->free_ids)
767 goto err_tx_free_ids;
770 size = tx_ring->tx_max_header_size;
771 tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
772 if (!tx_ring->push_buf_intermediate_buf) {
773 tx_ring->push_buf_intermediate_buf = vzalloc(size);
774 if (!tx_ring->push_buf_intermediate_buf)
775 goto err_push_buf_intermediate_buf;
778 /* Req id ring for TX out of order completions */
779 for (i = 0; i < tx_ring->ring_size; i++)
780 tx_ring->free_ids[i] = i;
782 /* Reset tx statistics */
783 memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
785 tx_ring->next_to_use = 0;
786 tx_ring->next_to_clean = 0;
787 tx_ring->cpu = ena_irq->cpu;
790 err_push_buf_intermediate_buf:
791 vfree(tx_ring->free_ids);
792 tx_ring->free_ids = NULL;
794 vfree(tx_ring->tx_buffer_info);
795 tx_ring->tx_buffer_info = NULL;
800 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
801 * @adapter: network interface device structure
804 * Free all transmit software resources
806 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
808 struct ena_ring *tx_ring = &adapter->tx_ring[qid];
810 vfree(tx_ring->tx_buffer_info);
811 tx_ring->tx_buffer_info = NULL;
813 vfree(tx_ring->free_ids);
814 tx_ring->free_ids = NULL;
816 vfree(tx_ring->push_buf_intermediate_buf);
817 tx_ring->push_buf_intermediate_buf = NULL;
820 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
826 for (i = first_index; i < first_index + count; i++) {
827 rc = ena_setup_tx_resources(adapter, i);
836 netif_err(adapter, ifup, adapter->netdev,
837 "Tx queue %d: allocation failed\n", i);
839 /* rewind the index freeing the rings as we go */
840 while (first_index < i--)
841 ena_free_tx_resources(adapter, i);
845 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
846 int first_index, int count)
850 for (i = first_index; i < first_index + count; i++)
851 ena_free_tx_resources(adapter, i);
854 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
855 * @adapter: board private structure
857 * Free all transmit software resources
859 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
861 ena_free_all_io_tx_resources_in_range(adapter,
863 adapter->xdp_num_queues +
864 adapter->num_io_queues);
867 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
868 * @adapter: network interface device structure
871 * Returns 0 on success, negative on failure
873 static int ena_setup_rx_resources(struct ena_adapter *adapter,
876 struct ena_ring *rx_ring = &adapter->rx_ring[qid];
877 struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
880 if (rx_ring->rx_buffer_info) {
881 netif_err(adapter, ifup, adapter->netdev,
882 "rx_buffer_info is not NULL");
886 /* alloc extra element so in rx path
887 * we can always prefetch rx_info + 1
889 size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
890 node = cpu_to_node(ena_irq->cpu);
892 rx_ring->rx_buffer_info = vzalloc_node(size, node);
893 if (!rx_ring->rx_buffer_info) {
894 rx_ring->rx_buffer_info = vzalloc(size);
895 if (!rx_ring->rx_buffer_info)
899 size = sizeof(u16) * rx_ring->ring_size;
900 rx_ring->free_ids = vzalloc_node(size, node);
901 if (!rx_ring->free_ids) {
902 rx_ring->free_ids = vzalloc(size);
903 if (!rx_ring->free_ids) {
904 vfree(rx_ring->rx_buffer_info);
905 rx_ring->rx_buffer_info = NULL;
910 /* Req id ring for receiving RX pkts out of order */
911 for (i = 0; i < rx_ring->ring_size; i++)
912 rx_ring->free_ids[i] = i;
914 /* Reset rx statistics */
915 memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
917 rx_ring->next_to_clean = 0;
918 rx_ring->next_to_use = 0;
919 rx_ring->cpu = ena_irq->cpu;
924 /* ena_free_rx_resources - Free I/O Rx Resources
925 * @adapter: network interface device structure
928 * Free all receive software resources
930 static void ena_free_rx_resources(struct ena_adapter *adapter,
933 struct ena_ring *rx_ring = &adapter->rx_ring[qid];
935 vfree(rx_ring->rx_buffer_info);
936 rx_ring->rx_buffer_info = NULL;
938 vfree(rx_ring->free_ids);
939 rx_ring->free_ids = NULL;
942 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
943 * @adapter: board private structure
945 * Return 0 on success, negative on failure
947 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
951 for (i = 0; i < adapter->num_io_queues; i++) {
952 rc = ena_setup_rx_resources(adapter, i);
961 netif_err(adapter, ifup, adapter->netdev,
962 "Rx queue %d: allocation failed\n", i);
964 /* rewind the index freeing the rings as we go */
966 ena_free_rx_resources(adapter, i);
970 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
971 * @adapter: board private structure
973 * Free all receive software resources
975 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
979 for (i = 0; i < adapter->num_io_queues; i++)
980 ena_free_rx_resources(adapter, i);
983 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
984 struct ena_rx_buffer *rx_info, gfp_t gfp)
986 int headroom = rx_ring->rx_headroom;
987 struct ena_com_buf *ena_buf;
991 /* restore page offset value in case it has been changed by device */
992 rx_info->page_offset = headroom;
994 /* if previous allocated page is not used */
995 if (unlikely(rx_info->page))
998 page = alloc_page(gfp);
999 if (unlikely(!page)) {
1000 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
1005 /* To enable NIC-side port-mirroring, AKA SPAN port,
1006 * we make the buffer readable from the nic as well
1008 dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
1010 if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
1011 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
1017 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1018 "Allocate page %p, rx_info %p\n", page, rx_info);
1020 rx_info->page = page;
1021 ena_buf = &rx_info->ena_buf;
1022 ena_buf->paddr = dma + headroom;
1023 ena_buf->len = ENA_PAGE_SIZE - headroom;
1028 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1029 struct ena_rx_buffer *rx_info)
1031 struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1033 dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1038 static void ena_free_rx_page(struct ena_ring *rx_ring,
1039 struct ena_rx_buffer *rx_info)
1041 struct page *page = rx_info->page;
1043 if (unlikely(!page)) {
1044 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1045 "Trying to free unallocated buffer\n");
1049 ena_unmap_rx_buff(rx_ring, rx_info);
1052 rx_info->page = NULL;
1055 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1057 u16 next_to_use, req_id;
1061 next_to_use = rx_ring->next_to_use;
1063 for (i = 0; i < num; i++) {
1064 struct ena_rx_buffer *rx_info;
1066 req_id = rx_ring->free_ids[next_to_use];
1068 rx_info = &rx_ring->rx_buffer_info[req_id];
1070 rc = ena_alloc_rx_page(rx_ring, rx_info,
1071 GFP_ATOMIC | __GFP_COMP);
1072 if (unlikely(rc < 0)) {
1073 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1074 "Failed to allocate buffer for rx queue %d\n",
1078 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1082 netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1083 "Failed to add buffer for rx queue %d\n",
1087 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1088 rx_ring->ring_size);
1091 if (unlikely(i < num)) {
1092 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1094 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1095 "Refilled rx qid %d with only %d buffers (from %d)\n",
1096 rx_ring->qid, i, num);
1099 /* ena_com_write_sq_doorbell issues a wmb() */
1101 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1103 rx_ring->next_to_use = next_to_use;
1108 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1111 struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1114 for (i = 0; i < rx_ring->ring_size; i++) {
1115 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1118 ena_free_rx_page(rx_ring, rx_info);
1122 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1123 * @adapter: board private structure
1125 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1127 struct ena_ring *rx_ring;
1128 int i, rc, bufs_num;
1130 for (i = 0; i < adapter->num_io_queues; i++) {
1131 rx_ring = &adapter->rx_ring[i];
1132 bufs_num = rx_ring->ring_size - 1;
1133 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1135 if (unlikely(rc != bufs_num))
1136 netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1137 "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1142 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1146 for (i = 0; i < adapter->num_io_queues; i++)
1147 ena_free_rx_bufs(adapter, i);
1150 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1151 struct ena_tx_buffer *tx_info)
1153 struct ena_com_buf *ena_buf;
1157 ena_buf = tx_info->bufs;
1158 cnt = tx_info->num_of_bufs;
1163 if (tx_info->map_linear_data) {
1164 dma_unmap_single(tx_ring->dev,
1165 dma_unmap_addr(ena_buf, paddr),
1166 dma_unmap_len(ena_buf, len),
1172 /* unmap remaining mapped pages */
1173 for (i = 0; i < cnt; i++) {
1174 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1175 dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1180 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1181 * @tx_ring: TX ring for which buffers be freed
1183 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1185 bool print_once = true;
1188 for (i = 0; i < tx_ring->ring_size; i++) {
1189 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1195 netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1196 "Free uncompleted tx skb qid %d idx 0x%x\n",
1200 netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1201 "Free uncompleted tx skb qid %d idx 0x%x\n",
1205 ena_unmap_tx_buff(tx_ring, tx_info);
1207 dev_kfree_skb_any(tx_info->skb);
1209 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1213 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1215 struct ena_ring *tx_ring;
1218 for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1219 tx_ring = &adapter->tx_ring[i];
1220 ena_free_tx_bufs(tx_ring);
1224 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1229 for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1230 ena_qid = ENA_IO_TXQ_IDX(i);
1231 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1235 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1240 for (i = 0; i < adapter->num_io_queues; i++) {
1241 ena_qid = ENA_IO_RXQ_IDX(i);
1242 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1243 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1247 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1249 ena_destroy_all_tx_queues(adapter);
1250 ena_destroy_all_rx_queues(adapter);
1253 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1254 struct ena_tx_buffer *tx_info, bool is_xdp)
1257 netif_err(ring->adapter,
1260 "tx_info doesn't have valid %s",
1261 is_xdp ? "xdp frame" : "skb");
1263 netif_err(ring->adapter,
1266 "Invalid req_id: %hu\n",
1269 ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1271 /* Trigger device reset */
1272 ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1273 set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1277 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1279 struct ena_tx_buffer *tx_info = NULL;
1281 if (likely(req_id < tx_ring->ring_size)) {
1282 tx_info = &tx_ring->tx_buffer_info[req_id];
1283 if (likely(tx_info->skb))
1287 return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1290 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1292 struct ena_tx_buffer *tx_info = NULL;
1294 if (likely(req_id < xdp_ring->ring_size)) {
1295 tx_info = &xdp_ring->tx_buffer_info[req_id];
1296 if (likely(tx_info->xdpf))
1300 return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1303 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1305 struct netdev_queue *txq;
1314 next_to_clean = tx_ring->next_to_clean;
1315 txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1317 while (tx_pkts < budget) {
1318 struct ena_tx_buffer *tx_info;
1319 struct sk_buff *skb;
1321 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1326 rc = validate_tx_req_id(tx_ring, req_id);
1330 tx_info = &tx_ring->tx_buffer_info[req_id];
1333 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1334 prefetch(&skb->end);
1336 tx_info->skb = NULL;
1337 tx_info->last_jiffies = 0;
1339 ena_unmap_tx_buff(tx_ring, tx_info);
1341 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1342 "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1345 tx_bytes += skb->len;
1348 total_done += tx_info->tx_descs;
1350 tx_ring->free_ids[next_to_clean] = req_id;
1351 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1352 tx_ring->ring_size);
1355 tx_ring->next_to_clean = next_to_clean;
1356 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1357 ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1359 netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1361 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1362 "tx_poll: q %d done. total pkts: %d\n",
1363 tx_ring->qid, tx_pkts);
1365 /* need to make the rings circular update visible to
1366 * ena_start_xmit() before checking for netif_queue_stopped().
1370 above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1371 ENA_TX_WAKEUP_THRESH);
1372 if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1373 __netif_tx_lock(txq, smp_processor_id());
1375 ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1376 ENA_TX_WAKEUP_THRESH);
1377 if (netif_tx_queue_stopped(txq) && above_thresh &&
1378 test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1379 netif_tx_wake_queue(txq);
1380 ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1383 __netif_tx_unlock(txq);
1389 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
1391 struct sk_buff *skb;
1394 skb = napi_get_frags(rx_ring->napi);
1396 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1397 rx_ring->rx_copybreak);
1399 if (unlikely(!skb)) {
1400 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1402 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1403 "Failed to allocate skb. frags: %d\n", frags);
1410 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1411 struct ena_com_rx_buf_info *ena_bufs,
1415 struct sk_buff *skb;
1416 struct ena_rx_buffer *rx_info;
1417 u16 len, req_id, buf = 0;
1420 len = ena_bufs[buf].len;
1421 req_id = ena_bufs[buf].req_id;
1423 rx_info = &rx_ring->rx_buffer_info[req_id];
1425 if (unlikely(!rx_info->page)) {
1426 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1431 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1432 "rx_info %p page %p\n",
1433 rx_info, rx_info->page);
1435 /* save virt address of first buffer */
1436 va = page_address(rx_info->page) + rx_info->page_offset;
1440 if (len <= rx_ring->rx_copybreak) {
1441 skb = ena_alloc_skb(rx_ring, false);
1445 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1446 "RX allocated small packet. len %d. data_len %d\n",
1447 skb->len, skb->data_len);
1449 /* sync this buffer for CPU use */
1450 dma_sync_single_for_cpu(rx_ring->dev,
1451 dma_unmap_addr(&rx_info->ena_buf, paddr),
1454 skb_copy_to_linear_data(skb, va, len);
1455 dma_sync_single_for_device(rx_ring->dev,
1456 dma_unmap_addr(&rx_info->ena_buf, paddr),
1461 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1462 rx_ring->free_ids[*next_to_clean] = req_id;
1463 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1464 rx_ring->ring_size);
1468 skb = ena_alloc_skb(rx_ring, true);
1473 ena_unmap_rx_buff(rx_ring, rx_info);
1475 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1476 rx_info->page_offset, len, ENA_PAGE_SIZE);
1478 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1479 "RX skb updated. len %d. data_len %d\n",
1480 skb->len, skb->data_len);
1482 rx_info->page = NULL;
1484 rx_ring->free_ids[*next_to_clean] = req_id;
1486 ENA_RX_RING_IDX_NEXT(*next_to_clean,
1487 rx_ring->ring_size);
1488 if (likely(--descs == 0))
1492 len = ena_bufs[buf].len;
1493 req_id = ena_bufs[buf].req_id;
1495 rx_info = &rx_ring->rx_buffer_info[req_id];
1501 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1502 * @adapter: structure containing adapter specific data
1503 * @ena_rx_ctx: received packet context/metadata
1504 * @skb: skb currently being received and modified
1506 static void ena_rx_checksum(struct ena_ring *rx_ring,
1507 struct ena_com_rx_ctx *ena_rx_ctx,
1508 struct sk_buff *skb)
1510 /* Rx csum disabled */
1511 if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1512 skb->ip_summed = CHECKSUM_NONE;
1516 /* For fragmented packets the checksum isn't valid */
1517 if (ena_rx_ctx->frag) {
1518 skb->ip_summed = CHECKSUM_NONE;
1522 /* if IP and error */
1523 if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1524 (ena_rx_ctx->l3_csum_err))) {
1525 /* ipv4 checksum error */
1526 skb->ip_summed = CHECKSUM_NONE;
1527 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1529 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1530 "RX IPv4 header checksum error\n");
1535 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1536 (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1537 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1538 /* TCP/UDP checksum error */
1539 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1541 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1542 "RX L4 checksum error\n");
1543 skb->ip_summed = CHECKSUM_NONE;
1547 if (likely(ena_rx_ctx->l4_csum_checked)) {
1548 skb->ip_summed = CHECKSUM_UNNECESSARY;
1549 ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1552 ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1554 skb->ip_summed = CHECKSUM_NONE;
1557 skb->ip_summed = CHECKSUM_NONE;
1563 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1564 struct ena_com_rx_ctx *ena_rx_ctx,
1565 struct sk_buff *skb)
1567 enum pkt_hash_types hash_type;
1569 if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1570 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1571 (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1573 hash_type = PKT_HASH_TYPE_L4;
1575 hash_type = PKT_HASH_TYPE_NONE;
1577 /* Override hash type if the packet is fragmented */
1578 if (ena_rx_ctx->frag)
1579 hash_type = PKT_HASH_TYPE_NONE;
1581 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1585 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1587 struct ena_rx_buffer *rx_info;
1590 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1591 xdp_prepare_buff(xdp, page_address(rx_info->page),
1592 rx_info->page_offset,
1593 rx_ring->ena_bufs[0].len, false);
1594 /* If for some reason we received a bigger packet than
1595 * we expect, then we simply drop it
1597 if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1600 ret = ena_xdp_execute(rx_ring, xdp);
1602 /* The xdp program might expand the headers */
1603 if (ret == XDP_PASS) {
1604 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1605 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1610 /* ena_clean_rx_irq - Cleanup RX irq
1611 * @rx_ring: RX ring to clean
1612 * @napi: napi handler
1613 * @budget: how many packets driver is allowed to clean
1615 * Returns the number of cleaned buffers.
1617 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1620 u16 next_to_clean = rx_ring->next_to_clean;
1621 struct ena_com_rx_ctx ena_rx_ctx;
1622 struct ena_rx_buffer *rx_info;
1623 struct ena_adapter *adapter;
1624 u32 res_budget, work_done;
1625 int rx_copybreak_pkt = 0;
1626 int refill_threshold;
1627 struct sk_buff *skb;
1628 int refill_required;
1629 struct xdp_buff xdp;
1636 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1637 "%s qid %d\n", __func__, rx_ring->qid);
1638 res_budget = budget;
1639 xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1642 xdp_verdict = XDP_PASS;
1644 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1645 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1646 ena_rx_ctx.descs = 0;
1647 ena_rx_ctx.pkt_offset = 0;
1648 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1649 rx_ring->ena_com_io_sq,
1654 if (unlikely(ena_rx_ctx.descs == 0))
1657 /* First descriptor might have an offset set by the device */
1658 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1659 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1661 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1662 "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1663 rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1664 ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1666 if (ena_xdp_present_ring(rx_ring))
1667 xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1669 /* allocate skb and fill it */
1670 if (xdp_verdict == XDP_PASS)
1671 skb = ena_rx_skb(rx_ring,
1676 if (unlikely(!skb)) {
1677 for (i = 0; i < ena_rx_ctx.descs; i++) {
1678 int req_id = rx_ring->ena_bufs[i].req_id;
1680 rx_ring->free_ids[next_to_clean] = req_id;
1682 ENA_RX_RING_IDX_NEXT(next_to_clean,
1683 rx_ring->ring_size);
1685 /* Packets was passed for transmission, unmap it
1688 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1689 ena_unmap_rx_buff(rx_ring,
1690 &rx_ring->rx_buffer_info[req_id]);
1691 rx_ring->rx_buffer_info[req_id].page = NULL;
1694 if (xdp_verdict != XDP_PASS) {
1695 xdp_flags |= xdp_verdict;
1702 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1704 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1706 skb_record_rx_queue(skb, rx_ring->qid);
1708 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1709 total_len += rx_ring->ena_bufs[0].len;
1711 napi_gro_receive(napi, skb);
1713 total_len += skb->len;
1714 napi_gro_frags(napi);
1718 } while (likely(res_budget));
1720 work_done = budget - res_budget;
1721 rx_ring->per_napi_packets += work_done;
1722 u64_stats_update_begin(&rx_ring->syncp);
1723 rx_ring->rx_stats.bytes += total_len;
1724 rx_ring->rx_stats.cnt += work_done;
1725 rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1726 u64_stats_update_end(&rx_ring->syncp);
1728 rx_ring->next_to_clean = next_to_clean;
1730 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1732 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1733 ENA_RX_REFILL_THRESH_PACKET);
1735 /* Optimization, try to batch new rx buffers */
1736 if (refill_required > refill_threshold) {
1737 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1738 ena_refill_rx_bufs(rx_ring, refill_required);
1741 if (xdp_flags & XDP_REDIRECT)
1747 adapter = netdev_priv(rx_ring->netdev);
1749 if (rc == -ENOSPC) {
1750 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1752 adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1754 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1756 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1759 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1764 static void ena_dim_work(struct work_struct *w)
1766 struct dim *dim = container_of(w, struct dim, work);
1767 struct dim_cq_moder cur_moder =
1768 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1769 struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1771 ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1772 dim->state = DIM_START_MEASURE;
1775 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1777 struct dim_sample dim_sample;
1778 struct ena_ring *rx_ring = ena_napi->rx_ring;
1780 if (!rx_ring->per_napi_packets)
1783 rx_ring->non_empty_napi_events++;
1785 dim_update_sample(rx_ring->non_empty_napi_events,
1786 rx_ring->rx_stats.cnt,
1787 rx_ring->rx_stats.bytes,
1790 net_dim(&ena_napi->dim, dim_sample);
1792 rx_ring->per_napi_packets = 0;
1795 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1796 struct ena_ring *rx_ring)
1798 struct ena_eth_io_intr_reg intr_reg;
1799 u32 rx_interval = 0;
1800 /* Rx ring can be NULL when for XDP tx queues which don't have an
1801 * accompanying rx_ring pair.
1804 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1805 rx_ring->smoothed_interval :
1806 ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1808 /* Update intr register: rx intr delay,
1809 * tx intr delay and interrupt unmask
1811 ena_com_update_intr_reg(&intr_reg,
1813 tx_ring->smoothed_interval,
1816 ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1819 /* It is a shared MSI-X.
1820 * Tx and Rx CQ have pointer to it.
1821 * So we use one of them to reach the intr reg
1822 * The Tx ring is used because the rx_ring is NULL for XDP queues
1824 ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1827 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1828 struct ena_ring *rx_ring)
1830 int cpu = get_cpu();
1833 /* Check only one ring since the 2 rings are running on the same cpu */
1834 if (likely(tx_ring->cpu == cpu))
1837 numa_node = cpu_to_node(cpu);
1840 if (numa_node != NUMA_NO_NODE) {
1841 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1843 ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1856 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1865 if (unlikely(!xdp_ring))
1867 next_to_clean = xdp_ring->next_to_clean;
1869 while (tx_pkts < budget) {
1870 struct ena_tx_buffer *tx_info;
1871 struct xdp_frame *xdpf;
1873 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1878 rc = validate_xdp_req_id(xdp_ring, req_id);
1882 tx_info = &xdp_ring->tx_buffer_info[req_id];
1883 xdpf = tx_info->xdpf;
1885 tx_info->xdpf = NULL;
1886 tx_info->last_jiffies = 0;
1887 ena_unmap_tx_buff(xdp_ring, tx_info);
1889 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1890 "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1893 tx_bytes += xdpf->len;
1895 total_done += tx_info->tx_descs;
1897 xdp_return_frame(xdpf);
1898 xdp_ring->free_ids[next_to_clean] = req_id;
1899 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1900 xdp_ring->ring_size);
1903 xdp_ring->next_to_clean = next_to_clean;
1904 ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1905 ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1907 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1908 "tx_poll: q %d done. total pkts: %d\n",
1909 xdp_ring->qid, tx_pkts);
1914 static int ena_io_poll(struct napi_struct *napi, int budget)
1916 struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1917 struct ena_ring *tx_ring, *rx_ring;
1919 int rx_work_done = 0;
1921 int napi_comp_call = 0;
1924 tx_ring = ena_napi->tx_ring;
1925 rx_ring = ena_napi->rx_ring;
1927 tx_ring->first_interrupt = ena_napi->first_interrupt;
1928 rx_ring->first_interrupt = ena_napi->first_interrupt;
1930 tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1932 if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1933 test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1934 napi_complete_done(napi, 0);
1938 tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1939 /* On netpoll the budget is zero and the handler should only clean the
1943 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1945 /* If the device is about to reset or down, avoid unmask
1946 * the interrupt and return 0 so NAPI won't reschedule
1948 if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1949 test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1950 napi_complete_done(napi, 0);
1953 } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1956 /* Update numa and unmask the interrupt only when schedule
1957 * from the interrupt context (vs from sk_busy_loop)
1959 if (napi_complete_done(napi, rx_work_done) &&
1960 READ_ONCE(ena_napi->interrupts_masked)) {
1961 smp_rmb(); /* make sure interrupts_masked is read */
1962 WRITE_ONCE(ena_napi->interrupts_masked, false);
1963 /* We apply adaptive moderation on Rx path only.
1964 * Tx uses static interrupt moderation.
1966 if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1967 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1969 ena_unmask_interrupt(tx_ring, rx_ring);
1972 ena_update_ring_numa_node(tx_ring, rx_ring);
1979 u64_stats_update_begin(&tx_ring->syncp);
1980 tx_ring->tx_stats.napi_comp += napi_comp_call;
1981 tx_ring->tx_stats.tx_poll++;
1982 u64_stats_update_end(&tx_ring->syncp);
1987 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1989 struct ena_adapter *adapter = (struct ena_adapter *)data;
1991 ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1993 /* Don't call the aenq handler before probe is done */
1994 if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1995 ena_com_aenq_intr_handler(adapter->ena_dev, data);
2000 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2001 * @irq: interrupt number
2002 * @data: pointer to a network interface private napi device structure
2004 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2006 struct ena_napi *ena_napi = data;
2008 ena_napi->first_interrupt = true;
2010 WRITE_ONCE(ena_napi->interrupts_masked, true);
2011 smp_wmb(); /* write interrupts_masked before calling napi */
2013 napi_schedule_irqoff(&ena_napi->napi);
2018 /* Reserve a single MSI-X vector for management (admin + aenq).
2019 * plus reserve one vector for each potential io queue.
2020 * the number of potential io queues is the minimum of what the device
2021 * supports and the number of vCPUs.
2023 static int ena_enable_msix(struct ena_adapter *adapter)
2025 int msix_vecs, irq_cnt;
2027 if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2028 netif_err(adapter, probe, adapter->netdev,
2029 "Error, MSI-X is already enabled\n");
2033 /* Reserved the max msix vectors we might need */
2034 msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2035 netif_dbg(adapter, probe, adapter->netdev,
2036 "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2038 irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2039 msix_vecs, PCI_IRQ_MSIX);
2042 netif_err(adapter, probe, adapter->netdev,
2043 "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2047 if (irq_cnt != msix_vecs) {
2048 netif_notice(adapter, probe, adapter->netdev,
2049 "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2050 irq_cnt, msix_vecs);
2051 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2054 if (ena_init_rx_cpu_rmap(adapter))
2055 netif_warn(adapter, probe, adapter->netdev,
2056 "Failed to map IRQs to CPUs\n");
2058 adapter->msix_vecs = irq_cnt;
2059 set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2064 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2068 snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2069 ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2070 pci_name(adapter->pdev));
2071 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2072 ena_intr_msix_mgmnt;
2073 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2074 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2075 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2076 cpu = cpumask_first(cpu_online_mask);
2077 adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2078 cpumask_set_cpu(cpu,
2079 &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2082 static void ena_setup_io_intr(struct ena_adapter *adapter)
2084 struct net_device *netdev;
2085 int irq_idx, i, cpu;
2088 netdev = adapter->netdev;
2089 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2091 for (i = 0; i < io_queue_count; i++) {
2092 irq_idx = ENA_IO_IRQ_IDX(i);
2093 cpu = i % num_online_cpus();
2095 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2096 "%s-Tx-Rx-%d", netdev->name, i);
2097 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2098 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2099 adapter->irq_tbl[irq_idx].vector =
2100 pci_irq_vector(adapter->pdev, irq_idx);
2101 adapter->irq_tbl[irq_idx].cpu = cpu;
2103 cpumask_set_cpu(cpu,
2104 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2108 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2110 unsigned long flags = 0;
2111 struct ena_irq *irq;
2114 irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2115 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2118 netif_err(adapter, probe, adapter->netdev,
2119 "Failed to request admin irq\n");
2123 netif_dbg(adapter, probe, adapter->netdev,
2124 "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2125 irq->affinity_hint_mask.bits[0], irq->vector);
2127 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2132 static int ena_request_io_irq(struct ena_adapter *adapter)
2134 u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2135 unsigned long flags = 0;
2136 struct ena_irq *irq;
2139 if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2140 netif_err(adapter, ifup, adapter->netdev,
2141 "Failed to request I/O IRQ: MSI-X is not enabled\n");
2145 for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2146 irq = &adapter->irq_tbl[i];
2147 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2150 netif_err(adapter, ifup, adapter->netdev,
2151 "Failed to request I/O IRQ. index %d rc %d\n",
2156 netif_dbg(adapter, ifup, adapter->netdev,
2157 "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2158 i, irq->affinity_hint_mask.bits[0], irq->vector);
2160 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2166 for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2167 irq = &adapter->irq_tbl[k];
2168 free_irq(irq->vector, irq->data);
2174 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2176 struct ena_irq *irq;
2178 irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2179 synchronize_irq(irq->vector);
2180 irq_set_affinity_hint(irq->vector, NULL);
2181 free_irq(irq->vector, irq->data);
2184 static void ena_free_io_irq(struct ena_adapter *adapter)
2186 u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2187 struct ena_irq *irq;
2190 #ifdef CONFIG_RFS_ACCEL
2191 if (adapter->msix_vecs >= 1) {
2192 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2193 adapter->netdev->rx_cpu_rmap = NULL;
2195 #endif /* CONFIG_RFS_ACCEL */
2197 for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2198 irq = &adapter->irq_tbl[i];
2199 irq_set_affinity_hint(irq->vector, NULL);
2200 free_irq(irq->vector, irq->data);
2204 static void ena_disable_msix(struct ena_adapter *adapter)
2206 if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2207 pci_free_irq_vectors(adapter->pdev);
2210 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2212 u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2215 if (!netif_running(adapter->netdev))
2218 for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2219 synchronize_irq(adapter->irq_tbl[i].vector);
2222 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2228 for (i = first_index; i < first_index + count; i++) {
2229 netif_napi_del(&adapter->ena_napi[i].napi);
2231 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2232 adapter->ena_napi[i].xdp_ring);
2236 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2237 int first_index, int count)
2241 for (i = first_index; i < first_index + count; i++) {
2242 struct ena_napi *napi = &adapter->ena_napi[i];
2244 netif_napi_add(adapter->netdev,
2246 ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2249 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2250 napi->rx_ring = &adapter->rx_ring[i];
2251 napi->tx_ring = &adapter->tx_ring[i];
2253 napi->xdp_ring = &adapter->tx_ring[i];
2259 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2265 for (i = first_index; i < first_index + count; i++)
2266 napi_disable(&adapter->ena_napi[i].napi);
2269 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2275 for (i = first_index; i < first_index + count; i++)
2276 napi_enable(&adapter->ena_napi[i].napi);
2279 /* Configure the Rx forwarding */
2280 static int ena_rss_configure(struct ena_adapter *adapter)
2282 struct ena_com_dev *ena_dev = adapter->ena_dev;
2285 /* In case the RSS table wasn't initialized by probe */
2286 if (!ena_dev->rss.tbl_log_size) {
2287 rc = ena_rss_init_default(adapter);
2288 if (rc && (rc != -EOPNOTSUPP)) {
2289 netif_err(adapter, ifup, adapter->netdev,
2290 "Failed to init RSS rc: %d\n", rc);
2295 /* Set indirect table */
2296 rc = ena_com_indirect_table_set(ena_dev);
2297 if (unlikely(rc && rc != -EOPNOTSUPP))
2300 /* Configure hash function (if supported) */
2301 rc = ena_com_set_hash_function(ena_dev);
2302 if (unlikely(rc && (rc != -EOPNOTSUPP)))
2305 /* Configure hash inputs (if supported) */
2306 rc = ena_com_set_hash_ctrl(ena_dev);
2307 if (unlikely(rc && (rc != -EOPNOTSUPP)))
2313 static int ena_up_complete(struct ena_adapter *adapter)
2317 rc = ena_rss_configure(adapter);
2321 ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2323 ena_refill_all_rx_bufs(adapter);
2325 /* enable transmits */
2326 netif_tx_start_all_queues(adapter->netdev);
2328 ena_napi_enable_in_range(adapter,
2330 adapter->xdp_num_queues + adapter->num_io_queues);
2335 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2337 struct ena_com_create_io_ctx ctx;
2338 struct ena_com_dev *ena_dev;
2339 struct ena_ring *tx_ring;
2344 ena_dev = adapter->ena_dev;
2346 tx_ring = &adapter->tx_ring[qid];
2347 msix_vector = ENA_IO_IRQ_IDX(qid);
2348 ena_qid = ENA_IO_TXQ_IDX(qid);
2350 memset(&ctx, 0x0, sizeof(ctx));
2352 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2354 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2355 ctx.msix_vector = msix_vector;
2356 ctx.queue_size = tx_ring->ring_size;
2357 ctx.numa_node = cpu_to_node(tx_ring->cpu);
2359 rc = ena_com_create_io_queue(ena_dev, &ctx);
2361 netif_err(adapter, ifup, adapter->netdev,
2362 "Failed to create I/O TX queue num %d rc: %d\n",
2367 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2368 &tx_ring->ena_com_io_sq,
2369 &tx_ring->ena_com_io_cq);
2371 netif_err(adapter, ifup, adapter->netdev,
2372 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2374 ena_com_destroy_io_queue(ena_dev, ena_qid);
2378 ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2382 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2383 int first_index, int count)
2385 struct ena_com_dev *ena_dev = adapter->ena_dev;
2388 for (i = first_index; i < first_index + count; i++) {
2389 rc = ena_create_io_tx_queue(adapter, i);
2397 while (i-- > first_index)
2398 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2403 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2405 struct ena_com_dev *ena_dev;
2406 struct ena_com_create_io_ctx ctx;
2407 struct ena_ring *rx_ring;
2412 ena_dev = adapter->ena_dev;
2414 rx_ring = &adapter->rx_ring[qid];
2415 msix_vector = ENA_IO_IRQ_IDX(qid);
2416 ena_qid = ENA_IO_RXQ_IDX(qid);
2418 memset(&ctx, 0x0, sizeof(ctx));
2421 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2422 ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2423 ctx.msix_vector = msix_vector;
2424 ctx.queue_size = rx_ring->ring_size;
2425 ctx.numa_node = cpu_to_node(rx_ring->cpu);
2427 rc = ena_com_create_io_queue(ena_dev, &ctx);
2429 netif_err(adapter, ifup, adapter->netdev,
2430 "Failed to create I/O RX queue num %d rc: %d\n",
2435 rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2436 &rx_ring->ena_com_io_sq,
2437 &rx_ring->ena_com_io_cq);
2439 netif_err(adapter, ifup, adapter->netdev,
2440 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2445 ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2449 ena_com_destroy_io_queue(ena_dev, ena_qid);
2453 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2455 struct ena_com_dev *ena_dev = adapter->ena_dev;
2458 for (i = 0; i < adapter->num_io_queues; i++) {
2459 rc = ena_create_io_rx_queue(adapter, i);
2462 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2469 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2470 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2476 static void set_io_rings_size(struct ena_adapter *adapter,
2482 for (i = 0; i < adapter->num_io_queues; i++) {
2483 adapter->tx_ring[i].ring_size = new_tx_size;
2484 adapter->rx_ring[i].ring_size = new_rx_size;
2488 /* This function allows queue allocation to backoff when the system is
2489 * low on memory. If there is not enough memory to allocate io queues
2490 * the driver will try to allocate smaller queues.
2492 * The backoff algorithm is as follows:
2493 * 1. Try to allocate TX and RX and if successful.
2494 * 1.1. return success
2496 * 2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2498 * 3. If TX or RX is smaller than 256
2499 * 3.1. return failure.
2501 * 4.1. go back to 1.
2503 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2505 int rc, cur_rx_ring_size, cur_tx_ring_size;
2506 int new_rx_ring_size, new_tx_ring_size;
2508 /* current queue sizes might be set to smaller than the requested
2509 * ones due to past queue allocation failures.
2511 set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2512 adapter->requested_rx_ring_size);
2515 if (ena_xdp_present(adapter)) {
2516 rc = ena_setup_and_create_all_xdp_queues(adapter);
2521 rc = ena_setup_tx_resources_in_range(adapter,
2523 adapter->num_io_queues);
2527 rc = ena_create_io_tx_queues_in_range(adapter,
2529 adapter->num_io_queues);
2531 goto err_create_tx_queues;
2533 rc = ena_setup_all_rx_resources(adapter);
2537 rc = ena_create_all_io_rx_queues(adapter);
2539 goto err_create_rx_queues;
2543 err_create_rx_queues:
2544 ena_free_all_io_rx_resources(adapter);
2546 ena_destroy_all_tx_queues(adapter);
2547 err_create_tx_queues:
2548 ena_free_all_io_tx_resources(adapter);
2550 if (rc != -ENOMEM) {
2551 netif_err(adapter, ifup, adapter->netdev,
2552 "Queue creation failed with error code %d\n",
2557 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2558 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2560 netif_err(adapter, ifup, adapter->netdev,
2561 "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2562 cur_tx_ring_size, cur_rx_ring_size);
2564 new_tx_ring_size = cur_tx_ring_size;
2565 new_rx_ring_size = cur_rx_ring_size;
2567 /* Decrease the size of the larger queue, or
2568 * decrease both if they are the same size.
2570 if (cur_rx_ring_size <= cur_tx_ring_size)
2571 new_tx_ring_size = cur_tx_ring_size / 2;
2572 if (cur_rx_ring_size >= cur_tx_ring_size)
2573 new_rx_ring_size = cur_rx_ring_size / 2;
2575 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2576 new_rx_ring_size < ENA_MIN_RING_SIZE) {
2577 netif_err(adapter, ifup, adapter->netdev,
2578 "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2583 netif_err(adapter, ifup, adapter->netdev,
2584 "Retrying queue creation with sizes TX=%d, RX=%d\n",
2588 set_io_rings_size(adapter, new_tx_ring_size,
2593 static int ena_up(struct ena_adapter *adapter)
2595 int io_queue_count, rc, i;
2597 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2599 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2600 ena_setup_io_intr(adapter);
2602 /* napi poll functions should be initialized before running
2603 * request_irq(), to handle a rare condition where there is a pending
2604 * interrupt, causing the ISR to fire immediately while the poll
2605 * function wasn't set yet, causing a null dereference
2607 ena_init_napi_in_range(adapter, 0, io_queue_count);
2609 rc = ena_request_io_irq(adapter);
2613 rc = create_queues_with_size_backoff(adapter);
2615 goto err_create_queues_with_backoff;
2617 rc = ena_up_complete(adapter);
2621 if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2622 netif_carrier_on(adapter->netdev);
2624 ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2627 set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2629 /* Enable completion queues interrupt */
2630 for (i = 0; i < adapter->num_io_queues; i++)
2631 ena_unmask_interrupt(&adapter->tx_ring[i],
2632 &adapter->rx_ring[i]);
2634 /* schedule napi in case we had pending packets
2635 * from the last time we disable napi
2637 for (i = 0; i < io_queue_count; i++)
2638 napi_schedule(&adapter->ena_napi[i].napi);
2643 ena_destroy_all_tx_queues(adapter);
2644 ena_free_all_io_tx_resources(adapter);
2645 ena_destroy_all_rx_queues(adapter);
2646 ena_free_all_io_rx_resources(adapter);
2647 err_create_queues_with_backoff:
2648 ena_free_io_irq(adapter);
2650 ena_del_napi_in_range(adapter, 0, io_queue_count);
2655 static void ena_down(struct ena_adapter *adapter)
2657 int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2659 netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2661 clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2663 ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2666 netif_carrier_off(adapter->netdev);
2667 netif_tx_disable(adapter->netdev);
2669 /* After this point the napi handler won't enable the tx queue */
2670 ena_napi_disable_in_range(adapter, 0, io_queue_count);
2672 /* After destroy the queue there won't be any new interrupts */
2674 if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2677 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2679 netif_err(adapter, ifdown, adapter->netdev,
2680 "Device reset failed\n");
2681 /* stop submitting admin commands on a device that was reset */
2682 ena_com_set_admin_running_state(adapter->ena_dev, false);
2685 ena_destroy_all_io_queues(adapter);
2687 ena_disable_io_intr_sync(adapter);
2688 ena_free_io_irq(adapter);
2689 ena_del_napi_in_range(adapter, 0, io_queue_count);
2691 ena_free_all_tx_bufs(adapter);
2692 ena_free_all_rx_bufs(adapter);
2693 ena_free_all_io_tx_resources(adapter);
2694 ena_free_all_io_rx_resources(adapter);
2697 /* ena_open - Called when a network interface is made active
2698 * @netdev: network interface device structure
2700 * Returns 0 on success, negative value on failure
2702 * The open entry point is called when a network interface is made
2703 * active by the system (IFF_UP). At this point all resources needed
2704 * for transmit and receive operations are allocated, the interrupt
2705 * handler is registered with the OS, the watchdog timer is started,
2706 * and the stack is notified that the interface is ready.
2708 static int ena_open(struct net_device *netdev)
2710 struct ena_adapter *adapter = netdev_priv(netdev);
2713 /* Notify the stack of the actual queue counts. */
2714 rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2716 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2720 rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2722 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2726 rc = ena_up(adapter);
2733 /* ena_close - Disables a network interface
2734 * @netdev: network interface device structure
2736 * Returns 0, this is not allowed to fail
2738 * The close entry point is called when an interface is de-activated
2739 * by the OS. The hardware is still under the drivers control, but
2740 * needs to be disabled. A global MAC reset is issued to stop the
2741 * hardware, and all transmit and receive resources are freed.
2743 static int ena_close(struct net_device *netdev)
2745 struct ena_adapter *adapter = netdev_priv(netdev);
2747 netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2749 if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2752 if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2755 /* Check for device status and issue reset if needed*/
2756 check_for_admin_com_state(adapter);
2757 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2758 netif_err(adapter, ifdown, adapter->netdev,
2759 "Destroy failure, restarting device\n");
2760 ena_dump_stats_to_dmesg(adapter);
2761 /* rtnl lock already obtained in dev_ioctl() layer */
2762 ena_destroy_device(adapter, false);
2763 ena_restore_device(adapter);
2769 int ena_update_queue_sizes(struct ena_adapter *adapter,
2775 dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2776 ena_close(adapter->netdev);
2777 adapter->requested_tx_ring_size = new_tx_size;
2778 adapter->requested_rx_ring_size = new_rx_size;
2779 ena_init_io_rings(adapter,
2781 adapter->xdp_num_queues +
2782 adapter->num_io_queues);
2783 return dev_was_up ? ena_up(adapter) : 0;
2786 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2788 struct ena_com_dev *ena_dev = adapter->ena_dev;
2789 int prev_channel_count;
2792 dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2793 ena_close(adapter->netdev);
2794 prev_channel_count = adapter->num_io_queues;
2795 adapter->num_io_queues = new_channel_count;
2796 if (ena_xdp_present(adapter) &&
2797 ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2798 adapter->xdp_first_ring = new_channel_count;
2799 adapter->xdp_num_queues = new_channel_count;
2800 if (prev_channel_count > new_channel_count)
2801 ena_xdp_exchange_program_rx_in_range(adapter,
2804 prev_channel_count);
2806 ena_xdp_exchange_program_rx_in_range(adapter,
2807 adapter->xdp_bpf_prog,
2812 /* We need to destroy the rss table so that the indirection
2813 * table will be reinitialized by ena_up()
2815 ena_com_rss_destroy(ena_dev);
2816 ena_init_io_rings(adapter,
2818 adapter->xdp_num_queues +
2819 adapter->num_io_queues);
2820 return dev_was_up ? ena_open(adapter->netdev) : 0;
2823 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2824 struct sk_buff *skb,
2825 bool disable_meta_caching)
2827 u32 mss = skb_shinfo(skb)->gso_size;
2828 struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2831 if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2832 ena_tx_ctx->l4_csum_enable = 1;
2834 ena_tx_ctx->tso_enable = 1;
2835 ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2836 ena_tx_ctx->l4_csum_partial = 0;
2838 ena_tx_ctx->tso_enable = 0;
2839 ena_meta->l4_hdr_len = 0;
2840 ena_tx_ctx->l4_csum_partial = 1;
2843 switch (ip_hdr(skb)->version) {
2845 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2846 if (ip_hdr(skb)->frag_off & htons(IP_DF))
2849 ena_tx_ctx->l3_csum_enable = 1;
2850 l4_protocol = ip_hdr(skb)->protocol;
2853 ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2854 l4_protocol = ipv6_hdr(skb)->nexthdr;
2860 if (l4_protocol == IPPROTO_TCP)
2861 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2863 ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2865 ena_meta->mss = mss;
2866 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2867 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2868 ena_tx_ctx->meta_valid = 1;
2869 } else if (disable_meta_caching) {
2870 memset(ena_meta, 0, sizeof(*ena_meta));
2871 ena_tx_ctx->meta_valid = 1;
2873 ena_tx_ctx->meta_valid = 0;
2877 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2878 struct sk_buff *skb)
2880 int num_frags, header_len, rc;
2882 num_frags = skb_shinfo(skb)->nr_frags;
2883 header_len = skb_headlen(skb);
2885 if (num_frags < tx_ring->sgl_size)
2888 if ((num_frags == tx_ring->sgl_size) &&
2889 (header_len < tx_ring->tx_max_header_size))
2892 ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2894 rc = skb_linearize(skb);
2896 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2903 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2904 struct ena_tx_buffer *tx_info,
2905 struct sk_buff *skb,
2909 struct ena_adapter *adapter = tx_ring->adapter;
2910 struct ena_com_buf *ena_buf;
2912 u32 skb_head_len, frag_len, last_frag;
2917 skb_head_len = skb_headlen(skb);
2919 ena_buf = tx_info->bufs;
2921 if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2922 /* When the device is LLQ mode, the driver will copy
2923 * the header into the device memory space.
2924 * the ena_com layer assume the header is in a linear
2926 * This assumption might be wrong since part of the header
2927 * can be in the fragmented buffers.
2928 * Use skb_header_pointer to make sure the header is in a
2929 * linear memory space.
2932 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2933 *push_hdr = skb_header_pointer(skb, 0, push_len,
2934 tx_ring->push_buf_intermediate_buf);
2935 *header_len = push_len;
2936 if (unlikely(skb->data != *push_hdr)) {
2937 ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2940 delta = push_len - skb_head_len;
2944 *header_len = min_t(u32, skb_head_len,
2945 tx_ring->tx_max_header_size);
2948 netif_dbg(adapter, tx_queued, adapter->netdev,
2949 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2950 *push_hdr, push_len);
2952 if (skb_head_len > push_len) {
2953 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2954 skb_head_len - push_len, DMA_TO_DEVICE);
2955 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2956 goto error_report_dma_error;
2958 ena_buf->paddr = dma;
2959 ena_buf->len = skb_head_len - push_len;
2962 tx_info->num_of_bufs++;
2963 tx_info->map_linear_data = 1;
2965 tx_info->map_linear_data = 0;
2968 last_frag = skb_shinfo(skb)->nr_frags;
2970 for (i = 0; i < last_frag; i++) {
2971 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2973 frag_len = skb_frag_size(frag);
2975 if (unlikely(delta >= frag_len)) {
2980 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2981 frag_len - delta, DMA_TO_DEVICE);
2982 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2983 goto error_report_dma_error;
2985 ena_buf->paddr = dma;
2986 ena_buf->len = frag_len - delta;
2988 tx_info->num_of_bufs++;
2994 error_report_dma_error:
2995 ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
2997 netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
2999 tx_info->skb = NULL;
3001 tx_info->num_of_bufs += i;
3002 ena_unmap_tx_buff(tx_ring, tx_info);
3007 /* Called with netif_tx_lock. */
3008 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3010 struct ena_adapter *adapter = netdev_priv(dev);
3011 struct ena_tx_buffer *tx_info;
3012 struct ena_com_tx_ctx ena_tx_ctx;
3013 struct ena_ring *tx_ring;
3014 struct netdev_queue *txq;
3016 u16 next_to_use, req_id, header_len;
3019 netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3020 /* Determine which tx ring we will be placed on */
3021 qid = skb_get_queue_mapping(skb);
3022 tx_ring = &adapter->tx_ring[qid];
3023 txq = netdev_get_tx_queue(dev, qid);
3025 rc = ena_check_and_linearize_skb(tx_ring, skb);
3027 goto error_drop_packet;
3029 skb_tx_timestamp(skb);
3031 next_to_use = tx_ring->next_to_use;
3032 req_id = tx_ring->free_ids[next_to_use];
3033 tx_info = &tx_ring->tx_buffer_info[req_id];
3034 tx_info->num_of_bufs = 0;
3036 WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3038 rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3040 goto error_drop_packet;
3042 memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3043 ena_tx_ctx.ena_bufs = tx_info->bufs;
3044 ena_tx_ctx.push_header = push_hdr;
3045 ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3046 ena_tx_ctx.req_id = req_id;
3047 ena_tx_ctx.header_len = header_len;
3049 /* set flags and meta data */
3050 ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3052 rc = ena_xmit_common(dev,
3059 goto error_unmap_dma;
3061 netdev_tx_sent_queue(txq, skb->len);
3063 /* stop the queue when no more space available, the packet can have up
3064 * to sgl_size + 2. one for the meta descriptor and one for header
3065 * (if the header is larger than tx_max_header_size).
3067 if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3068 tx_ring->sgl_size + 2))) {
3069 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3072 netif_tx_stop_queue(txq);
3073 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3076 /* There is a rare condition where this function decide to
3077 * stop the queue but meanwhile clean_tx_irq updates
3078 * next_to_completion and terminates.
3079 * The queue will remain stopped forever.
3080 * To solve this issue add a mb() to make sure that
3081 * netif_tx_stop_queue() write is vissible before checking if
3082 * there is additional space in the queue.
3086 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3087 ENA_TX_WAKEUP_THRESH)) {
3088 netif_tx_wake_queue(txq);
3089 ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3094 if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3095 /* trigger the dma engine. ena_com_write_sq_doorbell()
3098 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3099 ena_increase_stat(&tx_ring->tx_stats.doorbells, 1,
3103 return NETDEV_TX_OK;
3106 ena_unmap_tx_buff(tx_ring, tx_info);
3107 tx_info->skb = NULL;
3111 return NETDEV_TX_OK;
3114 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3115 struct net_device *sb_dev)
3118 /* we suspect that this is good for in--kernel network services that
3119 * want to loop incoming skb rx to tx in normal user generated traffic,
3120 * most probably we will not get to this
3122 if (skb_rx_queue_recorded(skb))
3123 qid = skb_get_rx_queue(skb);
3125 qid = netdev_pick_tx(dev, skb, NULL);
3130 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3132 struct device *dev = &pdev->dev;
3133 struct ena_admin_host_info *host_info;
3136 /* Allocate only the host info */
3137 rc = ena_com_allocate_host_info(ena_dev);
3139 dev_err(dev, "Cannot allocate host info\n");
3143 host_info = ena_dev->host_attr.host_info;
3145 host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3146 host_info->os_type = ENA_ADMIN_OS_LINUX;
3147 host_info->kernel_ver = LINUX_VERSION_CODE;
3148 strlcpy(host_info->kernel_ver_str, utsname()->version,
3149 sizeof(host_info->kernel_ver_str) - 1);
3150 host_info->os_dist = 0;
3151 strncpy(host_info->os_dist_str, utsname()->release,
3152 sizeof(host_info->os_dist_str) - 1);
3153 host_info->driver_version =
3154 (DRV_MODULE_GEN_MAJOR) |
3155 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3156 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3157 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3158 host_info->num_cpus = num_online_cpus();
3160 host_info->driver_supported_features =
3161 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3162 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3163 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3164 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3166 rc = ena_com_set_host_attributes(ena_dev);
3168 if (rc == -EOPNOTSUPP)
3169 dev_warn(dev, "Cannot set host attributes\n");
3171 dev_err(dev, "Cannot set host attributes\n");
3179 ena_com_delete_host_info(ena_dev);
3182 static void ena_config_debug_area(struct ena_adapter *adapter)
3184 u32 debug_area_size;
3187 ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3188 if (ss_count <= 0) {
3189 netif_err(adapter, drv, adapter->netdev,
3190 "SS count is negative\n");
3194 /* allocate 32 bytes for each string and 64bit for the value */
3195 debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3197 rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3199 netif_err(adapter, drv, adapter->netdev,
3200 "Cannot allocate debug area\n");
3204 rc = ena_com_set_host_attributes(adapter->ena_dev);
3206 if (rc == -EOPNOTSUPP)
3207 netif_warn(adapter, drv, adapter->netdev,
3208 "Cannot set host attributes\n");
3210 netif_err(adapter, drv, adapter->netdev,
3211 "Cannot set host attributes\n");
3217 ena_com_delete_debug_area(adapter->ena_dev);
3220 int ena_update_hw_stats(struct ena_adapter *adapter)
3224 rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3226 dev_info_once(&adapter->pdev->dev, "Failed to get ENI stats\n");
3233 static void ena_get_stats64(struct net_device *netdev,
3234 struct rtnl_link_stats64 *stats)
3236 struct ena_adapter *adapter = netdev_priv(netdev);
3237 struct ena_ring *rx_ring, *tx_ring;
3243 if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3246 for (i = 0; i < adapter->num_io_queues; i++) {
3249 tx_ring = &adapter->tx_ring[i];
3252 start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3253 packets = tx_ring->tx_stats.cnt;
3254 bytes = tx_ring->tx_stats.bytes;
3255 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3257 stats->tx_packets += packets;
3258 stats->tx_bytes += bytes;
3260 rx_ring = &adapter->rx_ring[i];
3263 start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3264 packets = rx_ring->rx_stats.cnt;
3265 bytes = rx_ring->rx_stats.bytes;
3266 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3268 stats->rx_packets += packets;
3269 stats->rx_bytes += bytes;
3273 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3274 rx_drops = adapter->dev_stats.rx_drops;
3275 tx_drops = adapter->dev_stats.tx_drops;
3276 } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3278 stats->rx_dropped = rx_drops;
3279 stats->tx_dropped = tx_drops;
3281 stats->multicast = 0;
3282 stats->collisions = 0;
3284 stats->rx_length_errors = 0;
3285 stats->rx_crc_errors = 0;
3286 stats->rx_frame_errors = 0;
3287 stats->rx_fifo_errors = 0;
3288 stats->rx_missed_errors = 0;
3289 stats->tx_window_errors = 0;
3291 stats->rx_errors = 0;
3292 stats->tx_errors = 0;
3295 static const struct net_device_ops ena_netdev_ops = {
3296 .ndo_open = ena_open,
3297 .ndo_stop = ena_close,
3298 .ndo_start_xmit = ena_start_xmit,
3299 .ndo_select_queue = ena_select_queue,
3300 .ndo_get_stats64 = ena_get_stats64,
3301 .ndo_tx_timeout = ena_tx_timeout,
3302 .ndo_change_mtu = ena_change_mtu,
3303 .ndo_set_mac_address = NULL,
3304 .ndo_validate_addr = eth_validate_addr,
3306 .ndo_xdp_xmit = ena_xdp_xmit,
3309 static int ena_device_validate_params(struct ena_adapter *adapter,
3310 struct ena_com_dev_get_features_ctx *get_feat_ctx)
3312 struct net_device *netdev = adapter->netdev;
3315 rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3318 netif_err(adapter, drv, netdev,
3319 "Error, mac address are different\n");
3323 if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3324 netif_err(adapter, drv, netdev,
3325 "Error, device max mtu is smaller than netdev MTU\n");
3332 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3334 llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3335 llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3336 llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3337 llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3338 llq_config->llq_ring_entry_size_value = 128;
3341 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3342 struct ena_com_dev *ena_dev,
3343 struct ena_admin_feature_llq_desc *llq,
3344 struct ena_llq_configurations *llq_default_configurations)
3347 u32 llq_feature_mask;
3349 llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3350 if (!(ena_dev->supported_features & llq_feature_mask)) {
3352 "LLQ is not supported Fallback to host mode policy.\n");
3353 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3357 rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3360 "Failed to configure the device mode. Fallback to host mode policy.\n");
3361 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3367 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3370 bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3373 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3375 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3376 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3382 ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3383 pci_resource_start(pdev, ENA_MEM_BAR),
3384 pci_resource_len(pdev, ENA_MEM_BAR));
3386 if (!ena_dev->mem_bar)
3392 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3393 struct ena_com_dev_get_features_ctx *get_feat_ctx,
3396 struct ena_llq_configurations llq_config;
3397 struct device *dev = &pdev->dev;
3398 bool readless_supported;
3403 rc = ena_com_mmio_reg_read_request_init(ena_dev);
3405 dev_err(dev, "Failed to init mmio read less\n");
3409 /* The PCIe configuration space revision id indicate if mmio reg
3412 readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3413 ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3415 rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3417 dev_err(dev, "Can not reset device\n");
3418 goto err_mmio_read_less;
3421 rc = ena_com_validate_version(ena_dev);
3423 dev_err(dev, "Device version is too low\n");
3424 goto err_mmio_read_less;
3427 dma_width = ena_com_get_dma_width(ena_dev);
3428 if (dma_width < 0) {
3429 dev_err(dev, "Invalid dma width value %d", dma_width);
3431 goto err_mmio_read_less;
3434 rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3436 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3437 goto err_mmio_read_less;
3440 /* ENA admin level init */
3441 rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3444 "Can not initialize ena admin queue with device\n");
3445 goto err_mmio_read_less;
3448 /* To enable the msix interrupts the driver needs to know the number
3449 * of queues. So the driver uses polling mode to retrieve this
3452 ena_com_set_admin_polling_mode(ena_dev, true);
3454 ena_config_host_info(ena_dev, pdev);
3456 /* Get Device Attributes*/
3457 rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3459 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3460 goto err_admin_init;
3463 /* Try to turn all the available aenq groups */
3464 aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3465 BIT(ENA_ADMIN_FATAL_ERROR) |
3466 BIT(ENA_ADMIN_WARNING) |
3467 BIT(ENA_ADMIN_NOTIFICATION) |
3468 BIT(ENA_ADMIN_KEEP_ALIVE);
3470 aenq_groups &= get_feat_ctx->aenq.supported_groups;
3472 rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3474 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3475 goto err_admin_init;
3478 *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3480 set_default_llq_configurations(&llq_config);
3482 rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3485 dev_err(dev, "ENA device init failed\n");
3486 goto err_admin_init;
3492 ena_com_delete_host_info(ena_dev);
3493 ena_com_admin_destroy(ena_dev);
3495 ena_com_mmio_reg_read_request_destroy(ena_dev);
3500 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3502 struct ena_com_dev *ena_dev = adapter->ena_dev;
3503 struct device *dev = &adapter->pdev->dev;
3506 rc = ena_enable_msix(adapter);
3508 dev_err(dev, "Can not reserve msix vectors\n");
3512 ena_setup_mgmnt_intr(adapter);
3514 rc = ena_request_mgmnt_irq(adapter);
3516 dev_err(dev, "Can not setup management interrupts\n");
3517 goto err_disable_msix;
3520 ena_com_set_admin_polling_mode(ena_dev, false);
3522 ena_com_admin_aenq_enable(ena_dev);
3527 ena_disable_msix(adapter);
3532 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3534 struct net_device *netdev = adapter->netdev;
3535 struct ena_com_dev *ena_dev = adapter->ena_dev;
3538 if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3541 netif_carrier_off(netdev);
3543 del_timer_sync(&adapter->timer_service);
3545 dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3546 adapter->dev_up_before_reset = dev_up;
3548 ena_com_set_admin_running_state(ena_dev, false);
3550 if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3553 /* Stop the device from sending AENQ events (in case reset flag is set
3554 * and device is up, ena_down() already reset the device.
3556 if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3557 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3559 ena_free_mgmnt_irq(adapter);
3561 ena_disable_msix(adapter);
3563 ena_com_abort_admin_commands(ena_dev);
3565 ena_com_wait_for_abort_completion(ena_dev);
3567 ena_com_admin_destroy(ena_dev);
3569 ena_com_mmio_reg_read_request_destroy(ena_dev);
3571 /* return reset reason to default value */
3572 adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3574 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3575 clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3578 static int ena_restore_device(struct ena_adapter *adapter)
3580 struct ena_com_dev_get_features_ctx get_feat_ctx;
3581 struct ena_com_dev *ena_dev = adapter->ena_dev;
3582 struct pci_dev *pdev = adapter->pdev;
3586 set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3587 rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3589 dev_err(&pdev->dev, "Can not initialize device\n");
3592 adapter->wd_state = wd_state;
3594 rc = ena_device_validate_params(adapter, &get_feat_ctx);
3596 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3597 goto err_device_destroy;
3600 rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3602 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3603 goto err_device_destroy;
3605 /* If the interface was up before the reset bring it up */
3606 if (adapter->dev_up_before_reset) {
3607 rc = ena_up(adapter);
3609 dev_err(&pdev->dev, "Failed to create I/O queues\n");
3610 goto err_disable_msix;
3614 set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3616 clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3617 if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3618 netif_carrier_on(adapter->netdev);
3620 mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3621 adapter->last_keep_alive_jiffies = jiffies;
3623 dev_err(&pdev->dev, "Device reset completed successfully\n");
3627 ena_free_mgmnt_irq(adapter);
3628 ena_disable_msix(adapter);
3630 ena_com_abort_admin_commands(ena_dev);
3631 ena_com_wait_for_abort_completion(ena_dev);
3632 ena_com_admin_destroy(ena_dev);
3633 ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3634 ena_com_mmio_reg_read_request_destroy(ena_dev);
3636 clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3637 clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3639 "Reset attempt failed. Can not reset the device\n");
3644 static void ena_fw_reset_device(struct work_struct *work)
3646 struct ena_adapter *adapter =
3647 container_of(work, struct ena_adapter, reset_task);
3651 if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3652 ena_destroy_device(adapter, false);
3653 ena_restore_device(adapter);
3659 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3660 struct ena_ring *rx_ring)
3662 if (likely(rx_ring->first_interrupt))
3665 if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3668 rx_ring->no_interrupt_event_cnt++;
3670 if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3671 netif_err(adapter, rx_err, adapter->netdev,
3672 "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3674 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3675 smp_mb__before_atomic();
3676 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3683 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3684 struct ena_ring *tx_ring)
3686 struct ena_tx_buffer *tx_buf;
3687 unsigned long last_jiffies;
3691 for (i = 0; i < tx_ring->ring_size; i++) {
3692 tx_buf = &tx_ring->tx_buffer_info[i];
3693 last_jiffies = tx_buf->last_jiffies;
3695 if (last_jiffies == 0)
3696 /* no pending Tx at this location */
3699 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3700 2 * adapter->missing_tx_completion_to))) {
3701 /* If after graceful period interrupt is still not
3702 * received, we schedule a reset
3704 netif_err(adapter, tx_err, adapter->netdev,
3705 "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3707 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3708 smp_mb__before_atomic();
3709 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3713 if (unlikely(time_is_before_jiffies(last_jiffies +
3714 adapter->missing_tx_completion_to))) {
3715 if (!tx_buf->print_once)
3716 netif_notice(adapter, tx_err, adapter->netdev,
3717 "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3720 tx_buf->print_once = 1;
3725 if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3726 netif_err(adapter, tx_err, adapter->netdev,
3727 "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3729 adapter->missing_tx_completion_threshold);
3730 adapter->reset_reason =
3731 ENA_REGS_RESET_MISS_TX_CMPL;
3732 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3736 ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3742 static void check_for_missing_completions(struct ena_adapter *adapter)
3744 struct ena_ring *tx_ring;
3745 struct ena_ring *rx_ring;
3749 io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3750 /* Make sure the driver doesn't turn the device in other process */
3753 if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3756 if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3759 if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3762 budget = ENA_MONITORED_TX_QUEUES;
3764 for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3765 tx_ring = &adapter->tx_ring[i];
3766 rx_ring = &adapter->rx_ring[i];
3768 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3772 rc = !ENA_IS_XDP_INDEX(adapter, i) ?
3773 check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3782 adapter->last_monitored_tx_qid = i % io_queue_count;
3785 /* trigger napi schedule after 2 consecutive detections */
3786 #define EMPTY_RX_REFILL 2
3787 /* For the rare case where the device runs out of Rx descriptors and the
3788 * napi handler failed to refill new Rx descriptors (due to a lack of memory
3790 * This case will lead to a deadlock:
3791 * The device won't send interrupts since all the new Rx packets will be dropped
3792 * The napi handler won't allocate new Rx descriptors so the device will be
3793 * able to send new packets.
3795 * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3796 * It is recommended to have at least 512MB, with a minimum of 128MB for
3797 * constrained environment).
3799 * When such a situation is detected - Reschedule napi
3801 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3803 struct ena_ring *rx_ring;
3804 int i, refill_required;
3806 if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3809 if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3812 for (i = 0; i < adapter->num_io_queues; i++) {
3813 rx_ring = &adapter->rx_ring[i];
3815 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3816 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3817 rx_ring->empty_rx_queue++;
3819 if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3820 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3823 netif_err(adapter, drv, adapter->netdev,
3824 "Trigger refill for ring %d\n", i);
3826 napi_schedule(rx_ring->napi);
3827 rx_ring->empty_rx_queue = 0;
3830 rx_ring->empty_rx_queue = 0;
3835 /* Check for keep alive expiration */
3836 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3838 unsigned long keep_alive_expired;
3840 if (!adapter->wd_state)
3843 if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3846 keep_alive_expired = adapter->last_keep_alive_jiffies +
3847 adapter->keep_alive_timeout;
3848 if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3849 netif_err(adapter, drv, adapter->netdev,
3850 "Keep alive watchdog timeout.\n");
3851 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3853 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3854 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3858 static void check_for_admin_com_state(struct ena_adapter *adapter)
3860 if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3861 netif_err(adapter, drv, adapter->netdev,
3862 "ENA admin queue is not in running state!\n");
3863 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3865 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3866 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3870 static void ena_update_hints(struct ena_adapter *adapter,
3871 struct ena_admin_ena_hw_hints *hints)
3873 struct net_device *netdev = adapter->netdev;
3875 if (hints->admin_completion_tx_timeout)
3876 adapter->ena_dev->admin_queue.completion_timeout =
3877 hints->admin_completion_tx_timeout * 1000;
3879 if (hints->mmio_read_timeout)
3880 /* convert to usec */
3881 adapter->ena_dev->mmio_read.reg_read_to =
3882 hints->mmio_read_timeout * 1000;
3884 if (hints->missed_tx_completion_count_threshold_to_reset)
3885 adapter->missing_tx_completion_threshold =
3886 hints->missed_tx_completion_count_threshold_to_reset;
3888 if (hints->missing_tx_completion_timeout) {
3889 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3890 adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3892 adapter->missing_tx_completion_to =
3893 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3896 if (hints->netdev_wd_timeout)
3897 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3899 if (hints->driver_watchdog_timeout) {
3900 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3901 adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3903 adapter->keep_alive_timeout =
3904 msecs_to_jiffies(hints->driver_watchdog_timeout);
3908 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3909 struct net_device *netdev)
3911 host_info->supported_network_features[0] =
3912 netdev->features & GENMASK_ULL(31, 0);
3913 host_info->supported_network_features[1] =
3914 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3917 static void ena_timer_service(struct timer_list *t)
3919 struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3920 u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3921 struct ena_admin_host_info *host_info =
3922 adapter->ena_dev->host_attr.host_info;
3924 check_for_missing_keep_alive(adapter);
3926 check_for_admin_com_state(adapter);
3928 check_for_missing_completions(adapter);
3930 check_for_empty_rx_ring(adapter);
3933 ena_dump_stats_to_buf(adapter, debug_area);
3936 ena_update_host_info(host_info, adapter->netdev);
3938 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3939 netif_err(adapter, drv, adapter->netdev,
3940 "Trigger reset is on\n");
3941 ena_dump_stats_to_dmesg(adapter);
3942 queue_work(ena_wq, &adapter->reset_task);
3946 /* Reset the timer */
3947 mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3950 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3951 struct ena_com_dev *ena_dev,
3952 struct ena_com_dev_get_features_ctx *get_feat_ctx)
3954 u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3956 if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3957 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3958 &get_feat_ctx->max_queue_ext.max_queue_ext;
3959 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3960 max_queue_ext->max_rx_cq_num);
3962 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3963 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3965 struct ena_admin_queue_feature_desc *max_queues =
3966 &get_feat_ctx->max_queues;
3967 io_tx_sq_num = max_queues->max_sq_num;
3968 io_tx_cq_num = max_queues->max_cq_num;
3969 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3972 /* In case of LLQ use the llq fields for the tx SQ/CQ */
3973 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3974 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3976 max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3977 max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3978 max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3979 max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3980 /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
3981 max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3982 if (unlikely(!max_num_io_queues)) {
3983 dev_err(&pdev->dev, "The device doesn't have io queues\n");
3987 return max_num_io_queues;
3990 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3991 struct net_device *netdev)
3993 netdev_features_t dev_features = 0;
3995 /* Set offload features */
3996 if (feat->offload.tx &
3997 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3998 dev_features |= NETIF_F_IP_CSUM;
4000 if (feat->offload.tx &
4001 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4002 dev_features |= NETIF_F_IPV6_CSUM;
4004 if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4005 dev_features |= NETIF_F_TSO;
4007 if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4008 dev_features |= NETIF_F_TSO6;
4010 if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4011 dev_features |= NETIF_F_TSO_ECN;
4013 if (feat->offload.rx_supported &
4014 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4015 dev_features |= NETIF_F_RXCSUM;
4017 if (feat->offload.rx_supported &
4018 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4019 dev_features |= NETIF_F_RXCSUM;
4027 netdev->hw_features |= netdev->features;
4028 netdev->vlan_features |= netdev->features;
4031 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4032 struct ena_com_dev_get_features_ctx *feat)
4034 struct net_device *netdev = adapter->netdev;
4036 /* Copy mac address */
4037 if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4038 eth_hw_addr_random(netdev);
4039 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4041 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4042 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4045 /* Set offload features */
4046 ena_set_dev_offloads(feat, netdev);
4048 adapter->max_mtu = feat->dev_attr.max_mtu;
4049 netdev->max_mtu = adapter->max_mtu;
4050 netdev->min_mtu = ENA_MIN_MTU;
4053 static int ena_rss_init_default(struct ena_adapter *adapter)
4055 struct ena_com_dev *ena_dev = adapter->ena_dev;
4056 struct device *dev = &adapter->pdev->dev;
4060 rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4062 dev_err(dev, "Cannot init indirect table\n");
4066 for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4067 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4068 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4069 ENA_IO_RXQ_IDX(val));
4070 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4071 dev_err(dev, "Cannot fill indirect table\n");
4072 goto err_fill_indir;
4076 rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4077 ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4078 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4079 dev_err(dev, "Cannot fill hash function\n");
4080 goto err_fill_indir;
4083 rc = ena_com_set_default_hash_ctrl(ena_dev);
4084 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4085 dev_err(dev, "Cannot fill hash control\n");
4086 goto err_fill_indir;
4092 ena_com_rss_destroy(ena_dev);
4098 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4100 int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4102 pci_release_selected_regions(pdev, release_bars);
4106 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4108 struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4109 struct ena_com_dev *ena_dev = ctx->ena_dev;
4110 u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4111 u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4112 u32 max_tx_queue_size;
4113 u32 max_rx_queue_size;
4115 if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4116 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4117 &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4118 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4119 max_queue_ext->max_rx_sq_depth);
4120 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4122 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4123 max_tx_queue_size = min_t(u32, max_tx_queue_size,
4124 llq->max_llq_depth);
4126 max_tx_queue_size = min_t(u32, max_tx_queue_size,
4127 max_queue_ext->max_tx_sq_depth);
4129 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4130 max_queue_ext->max_per_packet_tx_descs);
4131 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4132 max_queue_ext->max_per_packet_rx_descs);
4134 struct ena_admin_queue_feature_desc *max_queues =
4135 &ctx->get_feat_ctx->max_queues;
4136 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4137 max_queues->max_sq_depth);
4138 max_tx_queue_size = max_queues->max_cq_depth;
4140 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4141 max_tx_queue_size = min_t(u32, max_tx_queue_size,
4142 llq->max_llq_depth);
4144 max_tx_queue_size = min_t(u32, max_tx_queue_size,
4145 max_queues->max_sq_depth);
4147 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4148 max_queues->max_packet_tx_descs);
4149 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4150 max_queues->max_packet_rx_descs);
4153 max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4154 max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4156 tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4158 rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4161 tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4162 rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4164 ctx->max_tx_queue_size = max_tx_queue_size;
4165 ctx->max_rx_queue_size = max_rx_queue_size;
4166 ctx->tx_queue_size = tx_queue_size;
4167 ctx->rx_queue_size = rx_queue_size;
4172 /* ena_probe - Device Initialization Routine
4173 * @pdev: PCI device information struct
4174 * @ent: entry in ena_pci_tbl
4176 * Returns 0 on success, negative on failure
4178 * ena_probe initializes an adapter identified by a pci_dev structure.
4179 * The OS initialization, configuring of the adapter private structure,
4180 * and a hardware reset occur.
4182 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4184 struct ena_calc_queue_size_ctx calc_queue_ctx = {};
4185 struct ena_com_dev_get_features_ctx get_feat_ctx;
4186 struct ena_com_dev *ena_dev = NULL;
4187 struct ena_adapter *adapter;
4188 struct net_device *netdev;
4189 static int adapters_found;
4190 u32 max_num_io_queues;
4194 dev_dbg(&pdev->dev, "%s\n", __func__);
4196 rc = pci_enable_device_mem(pdev);
4198 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4202 rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4204 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4205 goto err_disable_device;
4208 pci_set_master(pdev);
4210 ena_dev = vzalloc(sizeof(*ena_dev));
4213 goto err_disable_device;
4216 bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4217 rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4219 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4221 goto err_free_ena_dev;
4224 ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4225 pci_resource_start(pdev, ENA_REG_BAR),
4226 pci_resource_len(pdev, ENA_REG_BAR));
4227 if (!ena_dev->reg_bar) {
4228 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4230 goto err_free_region;
4233 ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4235 ena_dev->dmadev = &pdev->dev;
4237 netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4239 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4241 goto err_free_region;
4244 SET_NETDEV_DEV(netdev, &pdev->dev);
4245 adapter = netdev_priv(netdev);
4246 adapter->ena_dev = ena_dev;
4247 adapter->netdev = netdev;
4248 adapter->pdev = pdev;
4249 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4251 ena_dev->net_device = netdev;
4253 pci_set_drvdata(pdev, adapter);
4255 rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4257 dev_err(&pdev->dev, "ENA device init failed\n");
4260 goto err_netdev_destroy;
4263 rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4265 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4266 goto err_device_destroy;
4269 calc_queue_ctx.ena_dev = ena_dev;
4270 calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4271 calc_queue_ctx.pdev = pdev;
4273 /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4274 * Updated during device initialization with the real granularity
4276 ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4277 ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4278 ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4279 max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4280 rc = ena_calc_io_queue_size(&calc_queue_ctx);
4281 if (rc || !max_num_io_queues) {
4283 goto err_device_destroy;
4286 ena_set_conf_feat_params(adapter, &get_feat_ctx);
4288 adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4290 adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4291 adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4292 adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4293 adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4294 adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4295 adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4297 adapter->num_io_queues = max_num_io_queues;
4298 adapter->max_num_io_queues = max_num_io_queues;
4299 adapter->last_monitored_tx_qid = 0;
4301 adapter->xdp_first_ring = 0;
4302 adapter->xdp_num_queues = 0;
4304 adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4305 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4306 adapter->disable_meta_caching =
4307 !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4308 BIT(ENA_ADMIN_DISABLE_META_CACHING));
4310 adapter->wd_state = wd_state;
4312 snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4314 rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4317 "Failed to query interrupt moderation feature\n");
4318 goto err_device_destroy;
4320 ena_init_io_rings(adapter,
4322 adapter->xdp_num_queues +
4323 adapter->num_io_queues);
4325 netdev->netdev_ops = &ena_netdev_ops;
4326 netdev->watchdog_timeo = TX_TIMEOUT;
4327 ena_set_ethtool_ops(netdev);
4329 netdev->priv_flags |= IFF_UNICAST_FLT;
4331 u64_stats_init(&adapter->syncp);
4333 rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4336 "Failed to enable and set the admin interrupts\n");
4337 goto err_worker_destroy;
4339 rc = ena_rss_init_default(adapter);
4340 if (rc && (rc != -EOPNOTSUPP)) {
4341 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4345 ena_config_debug_area(adapter);
4347 if (!ena_update_hw_stats(adapter))
4348 adapter->eni_stats_supported = true;
4350 adapter->eni_stats_supported = false;
4352 memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4354 netif_carrier_off(netdev);
4356 rc = register_netdev(netdev);
4358 dev_err(&pdev->dev, "Cannot register net device\n");
4362 INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4364 adapter->last_keep_alive_jiffies = jiffies;
4365 adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4366 adapter->missing_tx_completion_to = TX_TIMEOUT;
4367 adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4369 ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4371 timer_setup(&adapter->timer_service, ena_timer_service, 0);
4372 mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4374 dev_info(&pdev->dev,
4375 "%s found at mem %lx, mac addr %pM\n",
4376 DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4379 set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4386 ena_com_delete_debug_area(ena_dev);
4387 ena_com_rss_destroy(ena_dev);
4389 ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4390 /* stop submitting admin commands on a device that was reset */
4391 ena_com_set_admin_running_state(ena_dev, false);
4392 ena_free_mgmnt_irq(adapter);
4393 ena_disable_msix(adapter);
4395 del_timer(&adapter->timer_service);
4397 ena_com_delete_host_info(ena_dev);
4398 ena_com_admin_destroy(ena_dev);
4400 free_netdev(netdev);
4402 ena_release_bars(ena_dev, pdev);
4406 pci_disable_device(pdev);
4410 /*****************************************************************************/
4412 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4413 * @pdev: PCI device information struct
4414 * @shutdown: Is it a shutdown operation? If false, means it is a removal
4416 * __ena_shutoff is a helper routine that does the real work on shutdown and
4417 * removal paths; the difference between those paths is with regards to whether
4418 * dettach or unregister the netdevice.
4420 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4422 struct ena_adapter *adapter = pci_get_drvdata(pdev);
4423 struct ena_com_dev *ena_dev;
4424 struct net_device *netdev;
4426 ena_dev = adapter->ena_dev;
4427 netdev = adapter->netdev;
4429 #ifdef CONFIG_RFS_ACCEL
4430 if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4431 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4432 netdev->rx_cpu_rmap = NULL;
4434 #endif /* CONFIG_RFS_ACCEL */
4436 /* Make sure timer and reset routine won't be called after
4437 * freeing device resources.
4439 del_timer_sync(&adapter->timer_service);
4440 cancel_work_sync(&adapter->reset_task);
4442 rtnl_lock(); /* lock released inside the below if-else block */
4443 adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4444 ena_destroy_device(adapter, true);
4446 netif_device_detach(netdev);
4451 unregister_netdev(netdev);
4452 free_netdev(netdev);
4455 ena_com_rss_destroy(ena_dev);
4457 ena_com_delete_debug_area(ena_dev);
4459 ena_com_delete_host_info(ena_dev);
4461 ena_release_bars(ena_dev, pdev);
4463 pci_disable_device(pdev);
4468 /* ena_remove - Device Removal Routine
4469 * @pdev: PCI device information struct
4471 * ena_remove is called by the PCI subsystem to alert the driver
4472 * that it should release a PCI device.
4475 static void ena_remove(struct pci_dev *pdev)
4477 __ena_shutoff(pdev, false);
4480 /* ena_shutdown - Device Shutdown Routine
4481 * @pdev: PCI device information struct
4483 * ena_shutdown is called by the PCI subsystem to alert the driver that
4484 * a shutdown/reboot (or kexec) is happening and device must be disabled.
4487 static void ena_shutdown(struct pci_dev *pdev)
4489 __ena_shutoff(pdev, true);
4492 /* ena_suspend - PM suspend callback
4493 * @dev_d: Device information struct
4495 static int __maybe_unused ena_suspend(struct device *dev_d)
4497 struct pci_dev *pdev = to_pci_dev(dev_d);
4498 struct ena_adapter *adapter = pci_get_drvdata(pdev);
4500 ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4503 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4505 "Ignoring device reset request as the device is being suspended\n");
4506 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4508 ena_destroy_device(adapter, true);
4513 /* ena_resume - PM resume callback
4514 * @dev_d: Device information struct
4516 static int __maybe_unused ena_resume(struct device *dev_d)
4518 struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4521 ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4524 rc = ena_restore_device(adapter);
4529 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4531 static struct pci_driver ena_pci_driver = {
4532 .name = DRV_MODULE_NAME,
4533 .id_table = ena_pci_tbl,
4535 .remove = ena_remove,
4536 .shutdown = ena_shutdown,
4537 .driver.pm = &ena_pm_ops,
4538 .sriov_configure = pci_sriov_configure_simple,
4541 static int __init ena_init(void)
4543 ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4545 pr_err("Failed to create workqueue\n");
4549 return pci_register_driver(&ena_pci_driver);
4552 static void __exit ena_cleanup(void)
4554 pci_unregister_driver(&ena_pci_driver);
4557 destroy_workqueue(ena_wq);
4562 /******************************************************************************
4563 ******************************** AENQ Handlers *******************************
4564 *****************************************************************************/
4565 /* ena_update_on_link_change:
4566 * Notify the network interface about the change in link status
4568 static void ena_update_on_link_change(void *adapter_data,
4569 struct ena_admin_aenq_entry *aenq_e)
4571 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4572 struct ena_admin_aenq_link_change_desc *aenq_desc =
4573 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4574 int status = aenq_desc->flags &
4575 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4578 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4579 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4580 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4581 netif_carrier_on(adapter->netdev);
4583 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4584 netif_carrier_off(adapter->netdev);
4588 static void ena_keep_alive_wd(void *adapter_data,
4589 struct ena_admin_aenq_entry *aenq_e)
4591 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4592 struct ena_admin_aenq_keep_alive_desc *desc;
4596 desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4597 adapter->last_keep_alive_jiffies = jiffies;
4599 rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4600 tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4602 u64_stats_update_begin(&adapter->syncp);
4603 /* These stats are accumulated by the device, so the counters indicate
4604 * all drops since last reset.
4606 adapter->dev_stats.rx_drops = rx_drops;
4607 adapter->dev_stats.tx_drops = tx_drops;
4608 u64_stats_update_end(&adapter->syncp);
4611 static void ena_notification(void *adapter_data,
4612 struct ena_admin_aenq_entry *aenq_e)
4614 struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4615 struct ena_admin_ena_hw_hints *hints;
4617 WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4618 "Invalid group(%x) expected %x\n",
4619 aenq_e->aenq_common_desc.group,
4620 ENA_ADMIN_NOTIFICATION);
4622 switch (aenq_e->aenq_common_desc.syndrome) {
4623 case ENA_ADMIN_UPDATE_HINTS:
4624 hints = (struct ena_admin_ena_hw_hints *)
4625 (&aenq_e->inline_data_w4);
4626 ena_update_hints(adapter, hints);
4629 netif_err(adapter, drv, adapter->netdev,
4630 "Invalid aenq notification link state %d\n",
4631 aenq_e->aenq_common_desc.syndrome);
4635 /* This handler will called for unknown event group or unimplemented handlers*/
4636 static void unimplemented_aenq_handler(void *data,
4637 struct ena_admin_aenq_entry *aenq_e)
4639 struct ena_adapter *adapter = (struct ena_adapter *)data;
4641 netif_err(adapter, drv, adapter->netdev,
4642 "Unknown event was received or event with unimplemented handler\n");
4645 static struct ena_aenq_handlers aenq_handlers = {
4647 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4648 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4649 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4651 .unimplemented_handler = unimplemented_aenq_handler
4654 module_init(ena_init);
4655 module_exit(ena_cleanup);