2 * Copyright 2015 Amazon.com, Inc. or its affiliates.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #ifdef CONFIG_RFS_ACCEL
36 #include <linux/cpu_rmap.h>
37 #endif /* CONFIG_RFS_ACCEL */
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/kernel.h>
41 #include <linux/module.h>
42 #include <linux/moduleparam.h>
43 #include <linux/numa.h>
44 #include <linux/pci.h>
45 #include <linux/utsname.h>
46 #include <linux/version.h>
47 #include <linux/vmalloc.h>
50 #include "ena_netdev.h"
51 #include "ena_pci_id_tbl.h"
53 static char version
[] = DEVICE_NAME
" v" DRV_MODULE_VERSION
"\n";
55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
56 MODULE_DESCRIPTION(DEVICE_NAME
);
57 MODULE_LICENSE("GPL");
58 MODULE_VERSION(DRV_MODULE_VERSION
);
60 /* Time in jiffies before concluding the transmitter is hung. */
61 #define TX_TIMEOUT (5 * HZ)
63 #define ENA_NAPI_BUDGET 64
65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
66 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
67 static int debug
= -1;
68 module_param(debug
, int, 0);
69 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
71 static struct ena_aenq_handlers aenq_handlers
;
73 static struct workqueue_struct
*ena_wq
;
75 MODULE_DEVICE_TABLE(pci
, ena_pci_tbl
);
77 static int ena_rss_init_default(struct ena_adapter
*adapter
);
79 static void ena_tx_timeout(struct net_device
*dev
)
81 struct ena_adapter
*adapter
= netdev_priv(dev
);
83 u64_stats_update_begin(&adapter
->syncp
);
84 adapter
->dev_stats
.tx_timeout
++;
85 u64_stats_update_end(&adapter
->syncp
);
87 netif_err(adapter
, tx_err
, dev
, "Transmit time out\n");
89 /* Change the state of the device to trigger reset */
90 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
93 static void update_rx_ring_mtu(struct ena_adapter
*adapter
, int mtu
)
97 for (i
= 0; i
< adapter
->num_queues
; i
++)
98 adapter
->rx_ring
[i
].mtu
= mtu
;
101 static int ena_change_mtu(struct net_device
*dev
, int new_mtu
)
103 struct ena_adapter
*adapter
= netdev_priv(dev
);
106 if ((new_mtu
> adapter
->max_mtu
) || (new_mtu
< ENA_MIN_MTU
)) {
107 netif_err(adapter
, drv
, dev
,
108 "Invalid MTU setting. new_mtu: %d\n", new_mtu
);
113 ret
= ena_com_set_dev_mtu(adapter
->ena_dev
, new_mtu
);
115 netif_dbg(adapter
, drv
, dev
, "set MTU to %d\n", new_mtu
);
116 update_rx_ring_mtu(adapter
, new_mtu
);
119 netif_err(adapter
, drv
, dev
, "Failed to set MTU to %d\n",
126 static int ena_init_rx_cpu_rmap(struct ena_adapter
*adapter
)
128 #ifdef CONFIG_RFS_ACCEL
132 adapter
->netdev
->rx_cpu_rmap
= alloc_irq_cpu_rmap(adapter
->num_queues
);
133 if (!adapter
->netdev
->rx_cpu_rmap
)
135 for (i
= 0; i
< adapter
->num_queues
; i
++) {
136 int irq_idx
= ENA_IO_IRQ_IDX(i
);
138 rc
= irq_cpu_rmap_add(adapter
->netdev
->rx_cpu_rmap
,
139 adapter
->msix_entries
[irq_idx
].vector
);
141 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
142 adapter
->netdev
->rx_cpu_rmap
= NULL
;
146 #endif /* CONFIG_RFS_ACCEL */
150 static void ena_init_io_rings_common(struct ena_adapter
*adapter
,
151 struct ena_ring
*ring
, u16 qid
)
154 ring
->pdev
= adapter
->pdev
;
155 ring
->dev
= &adapter
->pdev
->dev
;
156 ring
->netdev
= adapter
->netdev
;
157 ring
->napi
= &adapter
->ena_napi
[qid
].napi
;
158 ring
->adapter
= adapter
;
159 ring
->ena_dev
= adapter
->ena_dev
;
160 ring
->per_napi_packets
= 0;
161 ring
->per_napi_bytes
= 0;
163 u64_stats_init(&ring
->syncp
);
166 static void ena_init_io_rings(struct ena_adapter
*adapter
)
168 struct ena_com_dev
*ena_dev
;
169 struct ena_ring
*txr
, *rxr
;
172 ena_dev
= adapter
->ena_dev
;
174 for (i
= 0; i
< adapter
->num_queues
; i
++) {
175 txr
= &adapter
->tx_ring
[i
];
176 rxr
= &adapter
->rx_ring
[i
];
178 /* TX/RX common ring state */
179 ena_init_io_rings_common(adapter
, txr
, i
);
180 ena_init_io_rings_common(adapter
, rxr
, i
);
182 /* TX specific ring state */
183 txr
->ring_size
= adapter
->tx_ring_size
;
184 txr
->tx_max_header_size
= ena_dev
->tx_max_header_size
;
185 txr
->tx_mem_queue_type
= ena_dev
->tx_mem_queue_type
;
186 txr
->sgl_size
= adapter
->max_tx_sgl_size
;
187 txr
->smoothed_interval
=
188 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev
);
190 /* RX specific ring state */
191 rxr
->ring_size
= adapter
->rx_ring_size
;
192 rxr
->rx_copybreak
= adapter
->rx_copybreak
;
193 rxr
->sgl_size
= adapter
->max_rx_sgl_size
;
194 rxr
->smoothed_interval
=
195 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev
);
199 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
200 * @adapter: network interface device structure
203 * Return 0 on success, negative on failure
205 static int ena_setup_tx_resources(struct ena_adapter
*adapter
, int qid
)
207 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
208 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
211 if (tx_ring
->tx_buffer_info
) {
212 netif_err(adapter
, ifup
,
213 adapter
->netdev
, "tx_buffer_info info is not NULL");
217 size
= sizeof(struct ena_tx_buffer
) * tx_ring
->ring_size
;
218 node
= cpu_to_node(ena_irq
->cpu
);
220 tx_ring
->tx_buffer_info
= vzalloc_node(size
, node
);
221 if (!tx_ring
->tx_buffer_info
) {
222 tx_ring
->tx_buffer_info
= vzalloc(size
);
223 if (!tx_ring
->tx_buffer_info
)
227 size
= sizeof(u16
) * tx_ring
->ring_size
;
228 tx_ring
->free_tx_ids
= vzalloc_node(size
, node
);
229 if (!tx_ring
->free_tx_ids
) {
230 tx_ring
->free_tx_ids
= vzalloc(size
);
231 if (!tx_ring
->free_tx_ids
) {
232 vfree(tx_ring
->tx_buffer_info
);
237 /* Req id ring for TX out of order completions */
238 for (i
= 0; i
< tx_ring
->ring_size
; i
++)
239 tx_ring
->free_tx_ids
[i
] = i
;
241 /* Reset tx statistics */
242 memset(&tx_ring
->tx_stats
, 0x0, sizeof(tx_ring
->tx_stats
));
244 tx_ring
->next_to_use
= 0;
245 tx_ring
->next_to_clean
= 0;
246 tx_ring
->cpu
= ena_irq
->cpu
;
250 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
251 * @adapter: network interface device structure
254 * Free all transmit software resources
256 static void ena_free_tx_resources(struct ena_adapter
*adapter
, int qid
)
258 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
260 vfree(tx_ring
->tx_buffer_info
);
261 tx_ring
->tx_buffer_info
= NULL
;
263 vfree(tx_ring
->free_tx_ids
);
264 tx_ring
->free_tx_ids
= NULL
;
267 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
268 * @adapter: private structure
270 * Return 0 on success, negative on failure
272 static int ena_setup_all_tx_resources(struct ena_adapter
*adapter
)
276 for (i
= 0; i
< adapter
->num_queues
; i
++) {
277 rc
= ena_setup_tx_resources(adapter
, i
);
286 netif_err(adapter
, ifup
, adapter
->netdev
,
287 "Tx queue %d: allocation failed\n", i
);
289 /* rewind the index freeing the rings as we go */
291 ena_free_tx_resources(adapter
, i
);
295 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
296 * @adapter: board private structure
298 * Free all transmit software resources
300 static void ena_free_all_io_tx_resources(struct ena_adapter
*adapter
)
304 for (i
= 0; i
< adapter
->num_queues
; i
++)
305 ena_free_tx_resources(adapter
, i
);
308 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
309 * @adapter: network interface device structure
312 * Returns 0 on success, negative on failure
314 static int ena_setup_rx_resources(struct ena_adapter
*adapter
,
317 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
318 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
321 if (rx_ring
->rx_buffer_info
) {
322 netif_err(adapter
, ifup
, adapter
->netdev
,
323 "rx_buffer_info is not NULL");
327 /* alloc extra element so in rx path
328 * we can always prefetch rx_info + 1
330 size
= sizeof(struct ena_rx_buffer
) * (rx_ring
->ring_size
+ 1);
331 node
= cpu_to_node(ena_irq
->cpu
);
333 rx_ring
->rx_buffer_info
= vzalloc_node(size
, node
);
334 if (!rx_ring
->rx_buffer_info
) {
335 rx_ring
->rx_buffer_info
= vzalloc(size
);
336 if (!rx_ring
->rx_buffer_info
)
340 /* Reset rx statistics */
341 memset(&rx_ring
->rx_stats
, 0x0, sizeof(rx_ring
->rx_stats
));
343 rx_ring
->next_to_clean
= 0;
344 rx_ring
->next_to_use
= 0;
345 rx_ring
->cpu
= ena_irq
->cpu
;
350 /* ena_free_rx_resources - Free I/O Rx Resources
351 * @adapter: network interface device structure
354 * Free all receive software resources
356 static void ena_free_rx_resources(struct ena_adapter
*adapter
,
359 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
361 vfree(rx_ring
->rx_buffer_info
);
362 rx_ring
->rx_buffer_info
= NULL
;
365 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
366 * @adapter: board private structure
368 * Return 0 on success, negative on failure
370 static int ena_setup_all_rx_resources(struct ena_adapter
*adapter
)
374 for (i
= 0; i
< adapter
->num_queues
; i
++) {
375 rc
= ena_setup_rx_resources(adapter
, i
);
384 netif_err(adapter
, ifup
, adapter
->netdev
,
385 "Rx queue %d: allocation failed\n", i
);
387 /* rewind the index freeing the rings as we go */
389 ena_free_rx_resources(adapter
, i
);
393 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
394 * @adapter: board private structure
396 * Free all receive software resources
398 static void ena_free_all_io_rx_resources(struct ena_adapter
*adapter
)
402 for (i
= 0; i
< adapter
->num_queues
; i
++)
403 ena_free_rx_resources(adapter
, i
);
406 static inline int ena_alloc_rx_page(struct ena_ring
*rx_ring
,
407 struct ena_rx_buffer
*rx_info
, gfp_t gfp
)
409 struct ena_com_buf
*ena_buf
;
413 /* if previous allocated page is not used */
414 if (unlikely(rx_info
->page
))
417 page
= alloc_page(gfp
);
418 if (unlikely(!page
)) {
419 u64_stats_update_begin(&rx_ring
->syncp
);
420 rx_ring
->rx_stats
.page_alloc_fail
++;
421 u64_stats_update_end(&rx_ring
->syncp
);
425 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
,
427 if (unlikely(dma_mapping_error(rx_ring
->dev
, dma
))) {
428 u64_stats_update_begin(&rx_ring
->syncp
);
429 rx_ring
->rx_stats
.dma_mapping_err
++;
430 u64_stats_update_end(&rx_ring
->syncp
);
435 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
436 "alloc page %p, rx_info %p\n", page
, rx_info
);
438 rx_info
->page
= page
;
439 rx_info
->page_offset
= 0;
440 ena_buf
= &rx_info
->ena_buf
;
441 ena_buf
->paddr
= dma
;
442 ena_buf
->len
= PAGE_SIZE
;
447 static void ena_free_rx_page(struct ena_ring
*rx_ring
,
448 struct ena_rx_buffer
*rx_info
)
450 struct page
*page
= rx_info
->page
;
451 struct ena_com_buf
*ena_buf
= &rx_info
->ena_buf
;
453 if (unlikely(!page
)) {
454 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
455 "Trying to free unallocated buffer\n");
459 dma_unmap_page(rx_ring
->dev
, ena_buf
->paddr
, PAGE_SIZE
,
463 rx_info
->page
= NULL
;
466 static int ena_refill_rx_bufs(struct ena_ring
*rx_ring
, u32 num
)
472 next_to_use
= rx_ring
->next_to_use
;
474 for (i
= 0; i
< num
; i
++) {
475 struct ena_rx_buffer
*rx_info
=
476 &rx_ring
->rx_buffer_info
[next_to_use
];
478 rc
= ena_alloc_rx_page(rx_ring
, rx_info
,
479 __GFP_COLD
| GFP_ATOMIC
| __GFP_COMP
);
480 if (unlikely(rc
< 0)) {
481 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
482 "failed to alloc buffer for rx queue %d\n",
486 rc
= ena_com_add_single_rx_desc(rx_ring
->ena_com_io_sq
,
490 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
491 "failed to add buffer for rx queue %d\n",
495 next_to_use
= ENA_RX_RING_IDX_NEXT(next_to_use
,
499 if (unlikely(i
< num
)) {
500 u64_stats_update_begin(&rx_ring
->syncp
);
501 rx_ring
->rx_stats
.refil_partial
++;
502 u64_stats_update_end(&rx_ring
->syncp
);
503 netdev_warn(rx_ring
->netdev
,
504 "refilled rx qid %d with only %d buffers (from %d)\n",
505 rx_ring
->qid
, i
, num
);
509 /* Add memory barrier to make sure the desc were written before
513 ena_com_write_sq_doorbell(rx_ring
->ena_com_io_sq
);
516 rx_ring
->next_to_use
= next_to_use
;
521 static void ena_free_rx_bufs(struct ena_adapter
*adapter
,
524 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
527 for (i
= 0; i
< rx_ring
->ring_size
; i
++) {
528 struct ena_rx_buffer
*rx_info
= &rx_ring
->rx_buffer_info
[i
];
531 ena_free_rx_page(rx_ring
, rx_info
);
535 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
536 * @adapter: board private structure
539 static void ena_refill_all_rx_bufs(struct ena_adapter
*adapter
)
541 struct ena_ring
*rx_ring
;
544 for (i
= 0; i
< adapter
->num_queues
; i
++) {
545 rx_ring
= &adapter
->rx_ring
[i
];
546 bufs_num
= rx_ring
->ring_size
- 1;
547 rc
= ena_refill_rx_bufs(rx_ring
, bufs_num
);
549 if (unlikely(rc
!= bufs_num
))
550 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
551 "refilling Queue %d failed. allocated %d buffers from: %d\n",
556 static void ena_free_all_rx_bufs(struct ena_adapter
*adapter
)
560 for (i
= 0; i
< adapter
->num_queues
; i
++)
561 ena_free_rx_bufs(adapter
, i
);
564 /* ena_free_tx_bufs - Free Tx Buffers per Queue
565 * @tx_ring: TX ring for which buffers be freed
567 static void ena_free_tx_bufs(struct ena_ring
*tx_ring
)
571 for (i
= 0; i
< tx_ring
->ring_size
; i
++) {
572 struct ena_tx_buffer
*tx_info
= &tx_ring
->tx_buffer_info
[i
];
573 struct ena_com_buf
*ena_buf
;
580 netdev_notice(tx_ring
->netdev
,
581 "free uncompleted tx skb qid %d idx 0x%x\n",
584 ena_buf
= tx_info
->bufs
;
585 dma_unmap_single(tx_ring
->dev
,
590 /* unmap remaining mapped pages */
591 nr_frags
= tx_info
->num_of_bufs
- 1;
592 for (j
= 0; j
< nr_frags
; j
++) {
594 dma_unmap_page(tx_ring
->dev
,
600 dev_kfree_skb_any(tx_info
->skb
);
602 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring
->netdev
,
606 static void ena_free_all_tx_bufs(struct ena_adapter
*adapter
)
608 struct ena_ring
*tx_ring
;
611 for (i
= 0; i
< adapter
->num_queues
; i
++) {
612 tx_ring
= &adapter
->tx_ring
[i
];
613 ena_free_tx_bufs(tx_ring
);
617 static void ena_destroy_all_tx_queues(struct ena_adapter
*adapter
)
622 for (i
= 0; i
< adapter
->num_queues
; i
++) {
623 ena_qid
= ENA_IO_TXQ_IDX(i
);
624 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
628 static void ena_destroy_all_rx_queues(struct ena_adapter
*adapter
)
633 for (i
= 0; i
< adapter
->num_queues
; i
++) {
634 ena_qid
= ENA_IO_RXQ_IDX(i
);
635 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
639 static void ena_destroy_all_io_queues(struct ena_adapter
*adapter
)
641 ena_destroy_all_tx_queues(adapter
);
642 ena_destroy_all_rx_queues(adapter
);
645 static int validate_tx_req_id(struct ena_ring
*tx_ring
, u16 req_id
)
647 struct ena_tx_buffer
*tx_info
= NULL
;
649 if (likely(req_id
< tx_ring
->ring_size
)) {
650 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
651 if (likely(tx_info
->skb
))
656 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
657 "tx_info doesn't have valid skb\n");
659 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
660 "Invalid req_id: %hu\n", req_id
);
662 u64_stats_update_begin(&tx_ring
->syncp
);
663 tx_ring
->tx_stats
.bad_req_id
++;
664 u64_stats_update_end(&tx_ring
->syncp
);
666 /* Trigger device reset */
667 set_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
);
671 static int ena_clean_tx_irq(struct ena_ring
*tx_ring
, u32 budget
)
673 struct netdev_queue
*txq
;
682 next_to_clean
= tx_ring
->next_to_clean
;
683 txq
= netdev_get_tx_queue(tx_ring
->netdev
, tx_ring
->qid
);
685 while (tx_pkts
< budget
) {
686 struct ena_tx_buffer
*tx_info
;
688 struct ena_com_buf
*ena_buf
;
691 rc
= ena_com_tx_comp_req_id_get(tx_ring
->ena_com_io_cq
,
696 rc
= validate_tx_req_id(tx_ring
, req_id
);
700 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
703 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
707 tx_info
->last_jiffies
= 0;
709 if (likely(tx_info
->num_of_bufs
!= 0)) {
710 ena_buf
= tx_info
->bufs
;
712 dma_unmap_single(tx_ring
->dev
,
713 dma_unmap_addr(ena_buf
, paddr
),
714 dma_unmap_len(ena_buf
, len
),
717 /* unmap remaining mapped pages */
718 nr_frags
= tx_info
->num_of_bufs
- 1;
719 for (i
= 0; i
< nr_frags
; i
++) {
721 dma_unmap_page(tx_ring
->dev
,
722 dma_unmap_addr(ena_buf
, paddr
),
723 dma_unmap_len(ena_buf
, len
),
728 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
729 "tx_poll: q %d skb %p completed\n", tx_ring
->qid
,
732 tx_bytes
+= skb
->len
;
735 total_done
+= tx_info
->tx_descs
;
737 tx_ring
->free_tx_ids
[next_to_clean
] = req_id
;
738 next_to_clean
= ENA_TX_RING_IDX_NEXT(next_to_clean
,
742 tx_ring
->next_to_clean
= next_to_clean
;
743 ena_com_comp_ack(tx_ring
->ena_com_io_sq
, total_done
);
744 ena_com_update_dev_comp_head(tx_ring
->ena_com_io_cq
);
746 netdev_tx_completed_queue(txq
, tx_pkts
, tx_bytes
);
748 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
749 "tx_poll: q %d done. total pkts: %d\n",
750 tx_ring
->qid
, tx_pkts
);
752 /* need to make the rings circular update visible to
753 * ena_start_xmit() before checking for netif_queue_stopped().
757 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
758 ENA_TX_WAKEUP_THRESH
;
759 if (unlikely(netif_tx_queue_stopped(txq
) && above_thresh
)) {
760 __netif_tx_lock(txq
, smp_processor_id());
761 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
762 ENA_TX_WAKEUP_THRESH
;
763 if (netif_tx_queue_stopped(txq
) && above_thresh
) {
764 netif_tx_wake_queue(txq
);
765 u64_stats_update_begin(&tx_ring
->syncp
);
766 tx_ring
->tx_stats
.queue_wakeup
++;
767 u64_stats_update_end(&tx_ring
->syncp
);
769 __netif_tx_unlock(txq
);
772 tx_ring
->per_napi_bytes
+= tx_bytes
;
773 tx_ring
->per_napi_packets
+= tx_pkts
;
778 static struct sk_buff
*ena_rx_skb(struct ena_ring
*rx_ring
,
779 struct ena_com_rx_buf_info
*ena_bufs
,
784 struct ena_rx_buffer
*rx_info
=
785 &rx_ring
->rx_buffer_info
[*next_to_clean
];
790 len
= ena_bufs
[0].len
;
791 if (unlikely(!rx_info
->page
)) {
792 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
797 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
798 "rx_info %p page %p\n",
799 rx_info
, rx_info
->page
);
801 /* save virt address of first buffer */
802 va
= page_address(rx_info
->page
) + rx_info
->page_offset
;
803 prefetch(va
+ NET_IP_ALIGN
);
805 if (len
<= rx_ring
->rx_copybreak
) {
806 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
807 rx_ring
->rx_copybreak
);
808 if (unlikely(!skb
)) {
809 u64_stats_update_begin(&rx_ring
->syncp
);
810 rx_ring
->rx_stats
.skb_alloc_fail
++;
811 u64_stats_update_end(&rx_ring
->syncp
);
812 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
813 "Failed to allocate skb\n");
817 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
818 "rx allocated small packet. len %d. data_len %d\n",
819 skb
->len
, skb
->data_len
);
821 /* sync this buffer for CPU use */
822 dma_sync_single_for_cpu(rx_ring
->dev
,
823 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
826 skb_copy_to_linear_data(skb
, va
, len
);
827 dma_sync_single_for_device(rx_ring
->dev
,
828 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
833 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
834 *next_to_clean
= ENA_RX_RING_IDX_ADD(*next_to_clean
, descs
,
839 skb
= napi_get_frags(rx_ring
->napi
);
840 if (unlikely(!skb
)) {
841 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
842 "Failed allocating skb\n");
843 u64_stats_update_begin(&rx_ring
->syncp
);
844 rx_ring
->rx_stats
.skb_alloc_fail
++;
845 u64_stats_update_end(&rx_ring
->syncp
);
850 dma_unmap_page(rx_ring
->dev
,
851 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
852 PAGE_SIZE
, DMA_FROM_DEVICE
);
854 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, rx_info
->page
,
855 rx_info
->page_offset
, len
, PAGE_SIZE
);
857 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
858 "rx skb updated. len %d. data_len %d\n",
859 skb
->len
, skb
->data_len
);
861 rx_info
->page
= NULL
;
863 ENA_RX_RING_IDX_NEXT(*next_to_clean
,
865 if (likely(--descs
== 0))
867 rx_info
= &rx_ring
->rx_buffer_info
[*next_to_clean
];
868 len
= ena_bufs
[++buf
].len
;
874 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
875 * @adapter: structure containing adapter specific data
876 * @ena_rx_ctx: received packet context/metadata
877 * @skb: skb currently being received and modified
879 static inline void ena_rx_checksum(struct ena_ring
*rx_ring
,
880 struct ena_com_rx_ctx
*ena_rx_ctx
,
883 /* Rx csum disabled */
884 if (unlikely(!(rx_ring
->netdev
->features
& NETIF_F_RXCSUM
))) {
885 skb
->ip_summed
= CHECKSUM_NONE
;
889 /* For fragmented packets the checksum isn't valid */
890 if (ena_rx_ctx
->frag
) {
891 skb
->ip_summed
= CHECKSUM_NONE
;
895 /* if IP and error */
896 if (unlikely((ena_rx_ctx
->l3_proto
== ENA_ETH_IO_L3_PROTO_IPV4
) &&
897 (ena_rx_ctx
->l3_csum_err
))) {
898 /* ipv4 checksum error */
899 skb
->ip_summed
= CHECKSUM_NONE
;
900 u64_stats_update_begin(&rx_ring
->syncp
);
901 rx_ring
->rx_stats
.bad_csum
++;
902 u64_stats_update_end(&rx_ring
->syncp
);
903 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
904 "RX IPv4 header checksum error\n");
909 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
910 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
))) {
911 if (unlikely(ena_rx_ctx
->l4_csum_err
)) {
912 /* TCP/UDP checksum error */
913 u64_stats_update_begin(&rx_ring
->syncp
);
914 rx_ring
->rx_stats
.bad_csum
++;
915 u64_stats_update_end(&rx_ring
->syncp
);
916 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
917 "RX L4 checksum error\n");
918 skb
->ip_summed
= CHECKSUM_NONE
;
922 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
926 static void ena_set_rx_hash(struct ena_ring
*rx_ring
,
927 struct ena_com_rx_ctx
*ena_rx_ctx
,
930 enum pkt_hash_types hash_type
;
932 if (likely(rx_ring
->netdev
->features
& NETIF_F_RXHASH
)) {
933 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
934 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
)))
936 hash_type
= PKT_HASH_TYPE_L4
;
938 hash_type
= PKT_HASH_TYPE_NONE
;
940 /* Override hash type if the packet is fragmented */
941 if (ena_rx_ctx
->frag
)
942 hash_type
= PKT_HASH_TYPE_NONE
;
944 skb_set_hash(skb
, ena_rx_ctx
->hash
, hash_type
);
948 /* ena_clean_rx_irq - Cleanup RX irq
949 * @rx_ring: RX ring to clean
950 * @napi: napi handler
951 * @budget: how many packets driver is allowed to clean
953 * Returns the number of cleaned buffers.
955 static int ena_clean_rx_irq(struct ena_ring
*rx_ring
, struct napi_struct
*napi
,
958 u16 next_to_clean
= rx_ring
->next_to_clean
;
959 u32 res_budget
, work_done
;
961 struct ena_com_rx_ctx ena_rx_ctx
;
962 struct ena_adapter
*adapter
;
965 int refill_threshold
;
968 int rx_copybreak_pkt
= 0;
970 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
971 "%s qid %d\n", __func__
, rx_ring
->qid
);
975 ena_rx_ctx
.ena_bufs
= rx_ring
->ena_bufs
;
976 ena_rx_ctx
.max_bufs
= rx_ring
->sgl_size
;
977 ena_rx_ctx
.descs
= 0;
978 rc
= ena_com_rx_pkt(rx_ring
->ena_com_io_cq
,
979 rx_ring
->ena_com_io_sq
,
984 if (unlikely(ena_rx_ctx
.descs
== 0))
987 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
988 "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
989 rx_ring
->qid
, ena_rx_ctx
.descs
, ena_rx_ctx
.l3_proto
,
990 ena_rx_ctx
.l4_proto
, ena_rx_ctx
.hash
);
992 /* allocate skb and fill it */
993 skb
= ena_rx_skb(rx_ring
, rx_ring
->ena_bufs
, ena_rx_ctx
.descs
,
996 /* exit if we failed to retrieve a buffer */
997 if (unlikely(!skb
)) {
998 next_to_clean
= ENA_RX_RING_IDX_ADD(next_to_clean
,
1000 rx_ring
->ring_size
);
1004 ena_rx_checksum(rx_ring
, &ena_rx_ctx
, skb
);
1006 ena_set_rx_hash(rx_ring
, &ena_rx_ctx
, skb
);
1008 skb_record_rx_queue(skb
, rx_ring
->qid
);
1010 if (rx_ring
->ena_bufs
[0].len
<= rx_ring
->rx_copybreak
) {
1011 total_len
+= rx_ring
->ena_bufs
[0].len
;
1013 napi_gro_receive(napi
, skb
);
1015 total_len
+= skb
->len
;
1016 napi_gro_frags(napi
);
1020 } while (likely(res_budget
));
1022 work_done
= budget
- res_budget
;
1023 rx_ring
->per_napi_bytes
+= total_len
;
1024 rx_ring
->per_napi_packets
+= work_done
;
1025 u64_stats_update_begin(&rx_ring
->syncp
);
1026 rx_ring
->rx_stats
.bytes
+= total_len
;
1027 rx_ring
->rx_stats
.cnt
+= work_done
;
1028 rx_ring
->rx_stats
.rx_copybreak_pkt
+= rx_copybreak_pkt
;
1029 u64_stats_update_end(&rx_ring
->syncp
);
1031 rx_ring
->next_to_clean
= next_to_clean
;
1033 refill_required
= ena_com_sq_empty_space(rx_ring
->ena_com_io_sq
);
1034 refill_threshold
= rx_ring
->ring_size
/ ENA_RX_REFILL_THRESH_DIVIDER
;
1036 /* Optimization, try to batch new rx buffers */
1037 if (refill_required
> refill_threshold
) {
1038 ena_com_update_dev_comp_head(rx_ring
->ena_com_io_cq
);
1039 ena_refill_rx_bufs(rx_ring
, refill_required
);
1045 adapter
= netdev_priv(rx_ring
->netdev
);
1047 u64_stats_update_begin(&rx_ring
->syncp
);
1048 rx_ring
->rx_stats
.bad_desc_num
++;
1049 u64_stats_update_end(&rx_ring
->syncp
);
1051 /* Too many desc from the device. Trigger reset */
1052 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
1057 inline void ena_adjust_intr_moderation(struct ena_ring
*rx_ring
,
1058 struct ena_ring
*tx_ring
)
1060 /* We apply adaptive moderation on Rx path only.
1061 * Tx uses static interrupt moderation.
1063 ena_com_calculate_interrupt_delay(rx_ring
->ena_dev
,
1064 rx_ring
->per_napi_packets
,
1065 rx_ring
->per_napi_bytes
,
1066 &rx_ring
->smoothed_interval
,
1067 &rx_ring
->moder_tbl_idx
);
1069 /* Reset per napi packets/bytes */
1070 tx_ring
->per_napi_packets
= 0;
1071 tx_ring
->per_napi_bytes
= 0;
1072 rx_ring
->per_napi_packets
= 0;
1073 rx_ring
->per_napi_bytes
= 0;
1076 static inline void ena_update_ring_numa_node(struct ena_ring
*tx_ring
,
1077 struct ena_ring
*rx_ring
)
1079 int cpu
= get_cpu();
1082 /* Check only one ring since the 2 rings are running on the same cpu */
1083 if (likely(tx_ring
->cpu
== cpu
))
1086 numa_node
= cpu_to_node(cpu
);
1089 if (numa_node
!= NUMA_NO_NODE
) {
1090 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, numa_node
);
1091 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, numa_node
);
1102 static int ena_io_poll(struct napi_struct
*napi
, int budget
)
1104 struct ena_napi
*ena_napi
= container_of(napi
, struct ena_napi
, napi
);
1105 struct ena_ring
*tx_ring
, *rx_ring
;
1106 struct ena_eth_io_intr_reg intr_reg
;
1111 int napi_comp_call
= 0;
1114 tx_ring
= ena_napi
->tx_ring
;
1115 rx_ring
= ena_napi
->rx_ring
;
1117 tx_budget
= tx_ring
->ring_size
/ ENA_TX_POLL_BUDGET_DIVIDER
;
1119 if (!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
)) {
1120 napi_complete_done(napi
, 0);
1124 tx_work_done
= ena_clean_tx_irq(tx_ring
, tx_budget
);
1125 rx_work_done
= ena_clean_rx_irq(rx_ring
, napi
, budget
);
1127 if ((budget
> rx_work_done
) && (tx_budget
> tx_work_done
)) {
1128 napi_complete_done(napi
, rx_work_done
);
1131 /* Tx and Rx share the same interrupt vector */
1132 if (ena_com_get_adaptive_moderation_enabled(rx_ring
->ena_dev
))
1133 ena_adjust_intr_moderation(rx_ring
, tx_ring
);
1135 /* Update intr register: rx intr delay, tx intr delay and
1138 ena_com_update_intr_reg(&intr_reg
,
1139 rx_ring
->smoothed_interval
,
1140 tx_ring
->smoothed_interval
,
1143 /* It is a shared MSI-X. Tx and Rx CQ have pointer to it.
1144 * So we use one of them to reach the intr reg
1146 ena_com_unmask_intr(rx_ring
->ena_com_io_cq
, &intr_reg
);
1148 ena_update_ring_numa_node(tx_ring
, rx_ring
);
1155 u64_stats_update_begin(&tx_ring
->syncp
);
1156 tx_ring
->tx_stats
.napi_comp
+= napi_comp_call
;
1157 tx_ring
->tx_stats
.tx_poll
++;
1158 u64_stats_update_end(&tx_ring
->syncp
);
1163 static irqreturn_t
ena_intr_msix_mgmnt(int irq
, void *data
)
1165 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
1167 ena_com_admin_q_comp_intr_handler(adapter
->ena_dev
);
1169 /* Don't call the aenq handler before probe is done */
1170 if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
)))
1171 ena_com_aenq_intr_handler(adapter
->ena_dev
, data
);
1176 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1177 * @irq: interrupt number
1178 * @data: pointer to a network interface private napi device structure
1180 static irqreturn_t
ena_intr_msix_io(int irq
, void *data
)
1182 struct ena_napi
*ena_napi
= data
;
1184 napi_schedule(&ena_napi
->napi
);
1189 static int ena_enable_msix(struct ena_adapter
*adapter
, int num_queues
)
1191 int i
, msix_vecs
, rc
;
1193 if (test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1194 netif_err(adapter
, probe
, adapter
->netdev
,
1195 "Error, MSI-X is already enabled\n");
1199 /* Reserved the max msix vectors we might need */
1200 msix_vecs
= ENA_MAX_MSIX_VEC(num_queues
);
1202 netif_dbg(adapter
, probe
, adapter
->netdev
,
1203 "trying to enable MSI-X, vectors %d\n", msix_vecs
);
1205 adapter
->msix_entries
= vzalloc(msix_vecs
* sizeof(struct msix_entry
));
1207 if (!adapter
->msix_entries
)
1210 for (i
= 0; i
< msix_vecs
; i
++)
1211 adapter
->msix_entries
[i
].entry
= i
;
1213 rc
= pci_enable_msix(adapter
->pdev
, adapter
->msix_entries
, msix_vecs
);
1215 netif_err(adapter
, probe
, adapter
->netdev
,
1216 "Failed to enable MSI-X, vectors %d rc %d\n",
1221 netif_dbg(adapter
, probe
, adapter
->netdev
, "enable MSI-X, vectors %d\n",
1224 if (msix_vecs
>= 1) {
1225 if (ena_init_rx_cpu_rmap(adapter
))
1226 netif_warn(adapter
, probe
, adapter
->netdev
,
1227 "Failed to map IRQs to CPUs\n");
1230 adapter
->msix_vecs
= msix_vecs
;
1231 set_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
);
1236 static void ena_setup_mgmnt_intr(struct ena_adapter
*adapter
)
1240 snprintf(adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].name
,
1241 ENA_IRQNAME_SIZE
, "ena-mgmnt@pci:%s",
1242 pci_name(adapter
->pdev
));
1243 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].handler
=
1244 ena_intr_msix_mgmnt
;
1245 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].data
= adapter
;
1246 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].vector
=
1247 adapter
->msix_entries
[ENA_MGMNT_IRQ_IDX
].vector
;
1248 cpu
= cpumask_first(cpu_online_mask
);
1249 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].cpu
= cpu
;
1250 cpumask_set_cpu(cpu
,
1251 &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].affinity_hint_mask
);
1254 static void ena_setup_io_intr(struct ena_adapter
*adapter
)
1256 struct net_device
*netdev
;
1257 int irq_idx
, i
, cpu
;
1259 netdev
= adapter
->netdev
;
1261 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1262 irq_idx
= ENA_IO_IRQ_IDX(i
);
1263 cpu
= i
% num_online_cpus();
1265 snprintf(adapter
->irq_tbl
[irq_idx
].name
, ENA_IRQNAME_SIZE
,
1266 "%s-Tx-Rx-%d", netdev
->name
, i
);
1267 adapter
->irq_tbl
[irq_idx
].handler
= ena_intr_msix_io
;
1268 adapter
->irq_tbl
[irq_idx
].data
= &adapter
->ena_napi
[i
];
1269 adapter
->irq_tbl
[irq_idx
].vector
=
1270 adapter
->msix_entries
[irq_idx
].vector
;
1271 adapter
->irq_tbl
[irq_idx
].cpu
= cpu
;
1273 cpumask_set_cpu(cpu
,
1274 &adapter
->irq_tbl
[irq_idx
].affinity_hint_mask
);
1278 static int ena_request_mgmnt_irq(struct ena_adapter
*adapter
)
1280 unsigned long flags
= 0;
1281 struct ena_irq
*irq
;
1284 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1285 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1288 netif_err(adapter
, probe
, adapter
->netdev
,
1289 "failed to request admin irq\n");
1293 netif_dbg(adapter
, probe
, adapter
->netdev
,
1294 "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1295 irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1297 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1302 static int ena_request_io_irq(struct ena_adapter
*adapter
)
1304 unsigned long flags
= 0;
1305 struct ena_irq
*irq
;
1308 if (!test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1309 netif_err(adapter
, ifup
, adapter
->netdev
,
1310 "Failed to request I/O IRQ: MSI-X is not enabled\n");
1314 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1315 irq
= &adapter
->irq_tbl
[i
];
1316 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1319 netif_err(adapter
, ifup
, adapter
->netdev
,
1320 "Failed to request I/O IRQ. index %d rc %d\n",
1325 netif_dbg(adapter
, ifup
, adapter
->netdev
,
1326 "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1327 i
, irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1329 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1335 for (k
= ENA_IO_IRQ_FIRST_IDX
; k
< i
; k
++) {
1336 irq
= &adapter
->irq_tbl
[k
];
1337 free_irq(irq
->vector
, irq
->data
);
1343 static void ena_free_mgmnt_irq(struct ena_adapter
*adapter
)
1345 struct ena_irq
*irq
;
1347 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1348 synchronize_irq(irq
->vector
);
1349 irq_set_affinity_hint(irq
->vector
, NULL
);
1350 free_irq(irq
->vector
, irq
->data
);
1353 static void ena_free_io_irq(struct ena_adapter
*adapter
)
1355 struct ena_irq
*irq
;
1358 #ifdef CONFIG_RFS_ACCEL
1359 if (adapter
->msix_vecs
>= 1) {
1360 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
1361 adapter
->netdev
->rx_cpu_rmap
= NULL
;
1363 #endif /* CONFIG_RFS_ACCEL */
1365 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1366 irq
= &adapter
->irq_tbl
[i
];
1367 irq_set_affinity_hint(irq
->vector
, NULL
);
1368 free_irq(irq
->vector
, irq
->data
);
1372 static void ena_disable_msix(struct ena_adapter
*adapter
)
1374 if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
))
1375 pci_disable_msix(adapter
->pdev
);
1377 if (adapter
->msix_entries
)
1378 vfree(adapter
->msix_entries
);
1379 adapter
->msix_entries
= NULL
;
1382 static void ena_disable_io_intr_sync(struct ena_adapter
*adapter
)
1386 if (!netif_running(adapter
->netdev
))
1389 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++)
1390 synchronize_irq(adapter
->irq_tbl
[i
].vector
);
1393 static void ena_del_napi(struct ena_adapter
*adapter
)
1397 for (i
= 0; i
< adapter
->num_queues
; i
++)
1398 netif_napi_del(&adapter
->ena_napi
[i
].napi
);
1401 static void ena_init_napi(struct ena_adapter
*adapter
)
1403 struct ena_napi
*napi
;
1406 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1407 napi
= &adapter
->ena_napi
[i
];
1409 netif_napi_add(adapter
->netdev
,
1410 &adapter
->ena_napi
[i
].napi
,
1413 napi
->rx_ring
= &adapter
->rx_ring
[i
];
1414 napi
->tx_ring
= &adapter
->tx_ring
[i
];
1419 static void ena_napi_disable_all(struct ena_adapter
*adapter
)
1423 for (i
= 0; i
< adapter
->num_queues
; i
++)
1424 napi_disable(&adapter
->ena_napi
[i
].napi
);
1427 static void ena_napi_enable_all(struct ena_adapter
*adapter
)
1431 for (i
= 0; i
< adapter
->num_queues
; i
++)
1432 napi_enable(&adapter
->ena_napi
[i
].napi
);
1435 static void ena_restore_ethtool_params(struct ena_adapter
*adapter
)
1437 adapter
->tx_usecs
= 0;
1438 adapter
->rx_usecs
= 0;
1439 adapter
->tx_frames
= 1;
1440 adapter
->rx_frames
= 1;
1443 /* Configure the Rx forwarding */
1444 static int ena_rss_configure(struct ena_adapter
*adapter
)
1446 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1449 /* In case the RSS table wasn't initialized by probe */
1450 if (!ena_dev
->rss
.tbl_log_size
) {
1451 rc
= ena_rss_init_default(adapter
);
1452 if (rc
&& (rc
!= -EPERM
)) {
1453 netif_err(adapter
, ifup
, adapter
->netdev
,
1454 "Failed to init RSS rc: %d\n", rc
);
1459 /* Set indirect table */
1460 rc
= ena_com_indirect_table_set(ena_dev
);
1461 if (unlikely(rc
&& rc
!= -EPERM
))
1464 /* Configure hash function (if supported) */
1465 rc
= ena_com_set_hash_function(ena_dev
);
1466 if (unlikely(rc
&& (rc
!= -EPERM
)))
1469 /* Configure hash inputs (if supported) */
1470 rc
= ena_com_set_hash_ctrl(ena_dev
);
1471 if (unlikely(rc
&& (rc
!= -EPERM
)))
1477 static int ena_up_complete(struct ena_adapter
*adapter
)
1481 rc
= ena_rss_configure(adapter
);
1485 ena_init_napi(adapter
);
1487 ena_change_mtu(adapter
->netdev
, adapter
->netdev
->mtu
);
1489 ena_refill_all_rx_bufs(adapter
);
1491 /* enable transmits */
1492 netif_tx_start_all_queues(adapter
->netdev
);
1494 ena_restore_ethtool_params(adapter
);
1496 ena_napi_enable_all(adapter
);
1498 /* schedule napi in case we had pending packets
1499 * from the last time we disable napi
1501 for (i
= 0; i
< adapter
->num_queues
; i
++)
1502 napi_schedule(&adapter
->ena_napi
[i
].napi
);
1507 static int ena_create_io_tx_queue(struct ena_adapter
*adapter
, int qid
)
1509 struct ena_com_create_io_ctx ctx
= { 0 };
1510 struct ena_com_dev
*ena_dev
;
1511 struct ena_ring
*tx_ring
;
1516 ena_dev
= adapter
->ena_dev
;
1518 tx_ring
= &adapter
->tx_ring
[qid
];
1519 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1520 ena_qid
= ENA_IO_TXQ_IDX(qid
);
1522 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_TX
;
1524 ctx
.mem_queue_type
= ena_dev
->tx_mem_queue_type
;
1525 ctx
.msix_vector
= msix_vector
;
1526 ctx
.queue_size
= adapter
->tx_ring_size
;
1527 ctx
.numa_node
= cpu_to_node(tx_ring
->cpu
);
1529 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1531 netif_err(adapter
, ifup
, adapter
->netdev
,
1532 "Failed to create I/O TX queue num %d rc: %d\n",
1537 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1538 &tx_ring
->ena_com_io_sq
,
1539 &tx_ring
->ena_com_io_cq
);
1541 netif_err(adapter
, ifup
, adapter
->netdev
,
1542 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1544 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1547 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1551 static int ena_create_all_io_tx_queues(struct ena_adapter
*adapter
)
1553 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1556 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1557 rc
= ena_create_io_tx_queue(adapter
, i
);
1566 ena_com_destroy_io_queue(ena_dev
, ENA_IO_TXQ_IDX(i
));
1571 static int ena_create_io_rx_queue(struct ena_adapter
*adapter
, int qid
)
1573 struct ena_com_dev
*ena_dev
;
1574 struct ena_com_create_io_ctx ctx
= { 0 };
1575 struct ena_ring
*rx_ring
;
1580 ena_dev
= adapter
->ena_dev
;
1582 rx_ring
= &adapter
->rx_ring
[qid
];
1583 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1584 ena_qid
= ENA_IO_RXQ_IDX(qid
);
1587 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_RX
;
1588 ctx
.mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
1589 ctx
.msix_vector
= msix_vector
;
1590 ctx
.queue_size
= adapter
->rx_ring_size
;
1591 ctx
.numa_node
= cpu_to_node(rx_ring
->cpu
);
1593 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1595 netif_err(adapter
, ifup
, adapter
->netdev
,
1596 "Failed to create I/O RX queue num %d rc: %d\n",
1601 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1602 &rx_ring
->ena_com_io_sq
,
1603 &rx_ring
->ena_com_io_cq
);
1605 netif_err(adapter
, ifup
, adapter
->netdev
,
1606 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1608 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1611 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1616 static int ena_create_all_io_rx_queues(struct ena_adapter
*adapter
)
1618 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1621 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1622 rc
= ena_create_io_rx_queue(adapter
, i
);
1631 ena_com_destroy_io_queue(ena_dev
, ENA_IO_RXQ_IDX(i
));
1636 static int ena_up(struct ena_adapter
*adapter
)
1640 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
1642 ena_setup_io_intr(adapter
);
1644 rc
= ena_request_io_irq(adapter
);
1648 /* allocate transmit descriptors */
1649 rc
= ena_setup_all_tx_resources(adapter
);
1653 /* allocate receive descriptors */
1654 rc
= ena_setup_all_rx_resources(adapter
);
1658 /* Create TX queues */
1659 rc
= ena_create_all_io_tx_queues(adapter
);
1661 goto err_create_tx_queues
;
1663 /* Create RX queues */
1664 rc
= ena_create_all_io_rx_queues(adapter
);
1666 goto err_create_rx_queues
;
1668 rc
= ena_up_complete(adapter
);
1672 if (test_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
))
1673 netif_carrier_on(adapter
->netdev
);
1675 u64_stats_update_begin(&adapter
->syncp
);
1676 adapter
->dev_stats
.interface_up
++;
1677 u64_stats_update_end(&adapter
->syncp
);
1679 set_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1684 ena_destroy_all_rx_queues(adapter
);
1685 err_create_rx_queues
:
1686 ena_destroy_all_tx_queues(adapter
);
1687 err_create_tx_queues
:
1688 ena_free_all_io_rx_resources(adapter
);
1690 ena_free_all_io_tx_resources(adapter
);
1692 ena_free_io_irq(adapter
);
1698 static void ena_down(struct ena_adapter
*adapter
)
1700 netif_info(adapter
, ifdown
, adapter
->netdev
, "%s\n", __func__
);
1702 clear_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1704 u64_stats_update_begin(&adapter
->syncp
);
1705 adapter
->dev_stats
.interface_down
++;
1706 u64_stats_update_end(&adapter
->syncp
);
1708 /* After this point the napi handler won't enable the tx queue */
1709 ena_napi_disable_all(adapter
);
1710 netif_carrier_off(adapter
->netdev
);
1711 netif_tx_disable(adapter
->netdev
);
1713 /* After destroy the queue there won't be any new interrupts */
1714 ena_destroy_all_io_queues(adapter
);
1716 ena_disable_io_intr_sync(adapter
);
1717 ena_free_io_irq(adapter
);
1718 ena_del_napi(adapter
);
1720 ena_free_all_tx_bufs(adapter
);
1721 ena_free_all_rx_bufs(adapter
);
1722 ena_free_all_io_tx_resources(adapter
);
1723 ena_free_all_io_rx_resources(adapter
);
1726 /* ena_open - Called when a network interface is made active
1727 * @netdev: network interface device structure
1729 * Returns 0 on success, negative value on failure
1731 * The open entry point is called when a network interface is made
1732 * active by the system (IFF_UP). At this point all resources needed
1733 * for transmit and receive operations are allocated, the interrupt
1734 * handler is registered with the OS, the watchdog timer is started,
1735 * and the stack is notified that the interface is ready.
1737 static int ena_open(struct net_device
*netdev
)
1739 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1742 /* Notify the stack of the actual queue counts. */
1743 rc
= netif_set_real_num_tx_queues(netdev
, adapter
->num_queues
);
1745 netif_err(adapter
, ifup
, netdev
, "Can't set num tx queues\n");
1749 rc
= netif_set_real_num_rx_queues(netdev
, adapter
->num_queues
);
1751 netif_err(adapter
, ifup
, netdev
, "Can't set num rx queues\n");
1755 rc
= ena_up(adapter
);
1762 /* ena_close - Disables a network interface
1763 * @netdev: network interface device structure
1765 * Returns 0, this is not allowed to fail
1767 * The close entry point is called when an interface is de-activated
1768 * by the OS. The hardware is still under the drivers control, but
1769 * needs to be disabled. A global MAC reset is issued to stop the
1770 * hardware, and all transmit and receive resources are freed.
1772 static int ena_close(struct net_device
*netdev
)
1774 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1776 netif_dbg(adapter
, ifdown
, netdev
, "%s\n", __func__
);
1778 if (test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
1784 static void ena_tx_csum(struct ena_com_tx_ctx
*ena_tx_ctx
, struct sk_buff
*skb
)
1786 u32 mss
= skb_shinfo(skb
)->gso_size
;
1787 struct ena_com_tx_meta
*ena_meta
= &ena_tx_ctx
->ena_meta
;
1790 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) || mss
) {
1791 ena_tx_ctx
->l4_csum_enable
= 1;
1793 ena_tx_ctx
->tso_enable
= 1;
1794 ena_meta
->l4_hdr_len
= tcp_hdr(skb
)->doff
;
1795 ena_tx_ctx
->l4_csum_partial
= 0;
1797 ena_tx_ctx
->tso_enable
= 0;
1798 ena_meta
->l4_hdr_len
= 0;
1799 ena_tx_ctx
->l4_csum_partial
= 1;
1802 switch (ip_hdr(skb
)->version
) {
1804 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV4
;
1805 if (ip_hdr(skb
)->frag_off
& htons(IP_DF
))
1808 ena_tx_ctx
->l3_csum_enable
= 1;
1809 l4_protocol
= ip_hdr(skb
)->protocol
;
1812 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV6
;
1813 l4_protocol
= ipv6_hdr(skb
)->nexthdr
;
1819 if (l4_protocol
== IPPROTO_TCP
)
1820 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_TCP
;
1822 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_UDP
;
1824 ena_meta
->mss
= mss
;
1825 ena_meta
->l3_hdr_len
= skb_network_header_len(skb
);
1826 ena_meta
->l3_hdr_offset
= skb_network_offset(skb
);
1827 ena_tx_ctx
->meta_valid
= 1;
1830 ena_tx_ctx
->meta_valid
= 0;
1834 static int ena_check_and_linearize_skb(struct ena_ring
*tx_ring
,
1835 struct sk_buff
*skb
)
1837 int num_frags
, header_len
, rc
;
1839 num_frags
= skb_shinfo(skb
)->nr_frags
;
1840 header_len
= skb_headlen(skb
);
1842 if (num_frags
< tx_ring
->sgl_size
)
1845 if ((num_frags
== tx_ring
->sgl_size
) &&
1846 (header_len
< tx_ring
->tx_max_header_size
))
1849 u64_stats_update_begin(&tx_ring
->syncp
);
1850 tx_ring
->tx_stats
.linearize
++;
1851 u64_stats_update_end(&tx_ring
->syncp
);
1853 rc
= skb_linearize(skb
);
1855 u64_stats_update_begin(&tx_ring
->syncp
);
1856 tx_ring
->tx_stats
.linearize_failed
++;
1857 u64_stats_update_end(&tx_ring
->syncp
);
1863 /* Called with netif_tx_lock. */
1864 static netdev_tx_t
ena_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1866 struct ena_adapter
*adapter
= netdev_priv(dev
);
1867 struct ena_tx_buffer
*tx_info
;
1868 struct ena_com_tx_ctx ena_tx_ctx
;
1869 struct ena_ring
*tx_ring
;
1870 struct netdev_queue
*txq
;
1871 struct ena_com_buf
*ena_buf
;
1879 int qid
, rc
, nb_hw_desc
;
1882 netif_dbg(adapter
, tx_queued
, dev
, "%s skb %p\n", __func__
, skb
);
1883 /* Determine which tx ring we will be placed on */
1884 qid
= skb_get_queue_mapping(skb
);
1885 tx_ring
= &adapter
->tx_ring
[qid
];
1886 txq
= netdev_get_tx_queue(dev
, qid
);
1888 rc
= ena_check_and_linearize_skb(tx_ring
, skb
);
1890 goto error_drop_packet
;
1892 skb_tx_timestamp(skb
);
1893 len
= skb_headlen(skb
);
1895 next_to_use
= tx_ring
->next_to_use
;
1896 req_id
= tx_ring
->free_tx_ids
[next_to_use
];
1897 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
1898 tx_info
->num_of_bufs
= 0;
1900 WARN(tx_info
->skb
, "SKB isn't NULL req_id %d\n", req_id
);
1901 ena_buf
= tx_info
->bufs
;
1904 if (tx_ring
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
1905 /* prepared the push buffer */
1906 push_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1907 header_len
= push_len
;
1908 push_hdr
= skb
->data
;
1911 header_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1915 netif_dbg(adapter
, tx_queued
, dev
,
1916 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb
,
1917 push_hdr
, push_len
);
1919 if (len
> push_len
) {
1920 dma
= dma_map_single(tx_ring
->dev
, skb
->data
+ push_len
,
1921 len
- push_len
, DMA_TO_DEVICE
);
1922 if (dma_mapping_error(tx_ring
->dev
, dma
))
1923 goto error_report_dma_error
;
1925 ena_buf
->paddr
= dma
;
1926 ena_buf
->len
= len
- push_len
;
1929 tx_info
->num_of_bufs
++;
1932 last_frag
= skb_shinfo(skb
)->nr_frags
;
1934 for (i
= 0; i
< last_frag
; i
++) {
1935 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1937 len
= skb_frag_size(frag
);
1938 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, len
,
1940 if (dma_mapping_error(tx_ring
->dev
, dma
))
1941 goto error_report_dma_error
;
1943 ena_buf
->paddr
= dma
;
1948 tx_info
->num_of_bufs
+= last_frag
;
1950 memset(&ena_tx_ctx
, 0x0, sizeof(struct ena_com_tx_ctx
));
1951 ena_tx_ctx
.ena_bufs
= tx_info
->bufs
;
1952 ena_tx_ctx
.push_header
= push_hdr
;
1953 ena_tx_ctx
.num_bufs
= tx_info
->num_of_bufs
;
1954 ena_tx_ctx
.req_id
= req_id
;
1955 ena_tx_ctx
.header_len
= header_len
;
1957 /* set flags and meta data */
1958 ena_tx_csum(&ena_tx_ctx
, skb
);
1960 /* prepare the packet's descriptors to dma engine */
1961 rc
= ena_com_prepare_tx(tx_ring
->ena_com_io_sq
, &ena_tx_ctx
,
1965 netif_err(adapter
, tx_queued
, dev
,
1966 "failed to prepare tx bufs\n");
1967 u64_stats_update_begin(&tx_ring
->syncp
);
1968 tx_ring
->tx_stats
.queue_stop
++;
1969 tx_ring
->tx_stats
.prepare_ctx_err
++;
1970 u64_stats_update_end(&tx_ring
->syncp
);
1971 netif_tx_stop_queue(txq
);
1972 goto error_unmap_dma
;
1975 netdev_tx_sent_queue(txq
, skb
->len
);
1977 u64_stats_update_begin(&tx_ring
->syncp
);
1978 tx_ring
->tx_stats
.cnt
++;
1979 tx_ring
->tx_stats
.bytes
+= skb
->len
;
1980 u64_stats_update_end(&tx_ring
->syncp
);
1982 tx_info
->tx_descs
= nb_hw_desc
;
1983 tx_info
->last_jiffies
= jiffies
;
1985 tx_ring
->next_to_use
= ENA_TX_RING_IDX_NEXT(next_to_use
,
1986 tx_ring
->ring_size
);
1988 /* This WMB is aimed to:
1989 * 1 - perform smp barrier before reading next_to_completion
1990 * 2 - make sure the desc were written before trigger DB
1994 /* stop the queue when no more space available, the packet can have up
1995 * to sgl_size + 2. one for the meta descriptor and one for header
1996 * (if the header is larger than tx_max_header_size).
1998 if (unlikely(ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) <
1999 (tx_ring
->sgl_size
+ 2))) {
2000 netif_dbg(adapter
, tx_queued
, dev
, "%s stop queue %d\n",
2003 netif_tx_stop_queue(txq
);
2004 u64_stats_update_begin(&tx_ring
->syncp
);
2005 tx_ring
->tx_stats
.queue_stop
++;
2006 u64_stats_update_end(&tx_ring
->syncp
);
2008 /* There is a rare condition where this function decide to
2009 * stop the queue but meanwhile clean_tx_irq updates
2010 * next_to_completion and terminates.
2011 * The queue will remain stopped forever.
2012 * To solve this issue this function perform rmb, check
2013 * the wakeup condition and wake up the queue if needed.
2017 if (ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
)
2018 > ENA_TX_WAKEUP_THRESH
) {
2019 netif_tx_wake_queue(txq
);
2020 u64_stats_update_begin(&tx_ring
->syncp
);
2021 tx_ring
->tx_stats
.queue_wakeup
++;
2022 u64_stats_update_end(&tx_ring
->syncp
);
2026 if (netif_xmit_stopped(txq
) || !skb
->xmit_more
) {
2027 /* trigger the dma engine */
2028 ena_com_write_sq_doorbell(tx_ring
->ena_com_io_sq
);
2029 u64_stats_update_begin(&tx_ring
->syncp
);
2030 tx_ring
->tx_stats
.doorbells
++;
2031 u64_stats_update_end(&tx_ring
->syncp
);
2034 return NETDEV_TX_OK
;
2036 error_report_dma_error
:
2037 u64_stats_update_begin(&tx_ring
->syncp
);
2038 tx_ring
->tx_stats
.dma_mapping_err
++;
2039 u64_stats_update_end(&tx_ring
->syncp
);
2040 netdev_warn(adapter
->netdev
, "failed to map skb\n");
2042 tx_info
->skb
= NULL
;
2046 /* save value of frag that failed */
2049 /* start back at beginning and unmap skb */
2050 tx_info
->skb
= NULL
;
2051 ena_buf
= tx_info
->bufs
;
2052 dma_unmap_single(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2053 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2055 /* unmap remaining mapped pages */
2056 for (i
= 0; i
< last_frag
; i
++) {
2058 dma_unmap_page(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2059 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2066 return NETDEV_TX_OK
;
2069 #ifdef CONFIG_NET_POLL_CONTROLLER
2070 static void ena_netpoll(struct net_device
*netdev
)
2072 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2075 for (i
= 0; i
< adapter
->num_queues
; i
++)
2076 napi_schedule(&adapter
->ena_napi
[i
].napi
);
2078 #endif /* CONFIG_NET_POLL_CONTROLLER */
2080 static u16
ena_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
2081 void *accel_priv
, select_queue_fallback_t fallback
)
2084 /* we suspect that this is good for in--kernel network services that
2085 * want to loop incoming skb rx to tx in normal user generated traffic,
2086 * most probably we will not get to this
2088 if (skb_rx_queue_recorded(skb
))
2089 qid
= skb_get_rx_queue(skb
);
2091 qid
= fallback(dev
, skb
);
2096 static void ena_config_host_info(struct ena_com_dev
*ena_dev
)
2098 struct ena_admin_host_info
*host_info
;
2101 /* Allocate only the host info */
2102 rc
= ena_com_allocate_host_info(ena_dev
);
2104 pr_err("Cannot allocate host info\n");
2108 host_info
= ena_dev
->host_attr
.host_info
;
2110 host_info
->os_type
= ENA_ADMIN_OS_LINUX
;
2111 host_info
->kernel_ver
= LINUX_VERSION_CODE
;
2112 strncpy(host_info
->kernel_ver_str
, utsname()->version
,
2113 sizeof(host_info
->kernel_ver_str
) - 1);
2114 host_info
->os_dist
= 0;
2115 strncpy(host_info
->os_dist_str
, utsname()->release
,
2116 sizeof(host_info
->os_dist_str
) - 1);
2117 host_info
->driver_version
=
2118 (DRV_MODULE_VER_MAJOR
) |
2119 (DRV_MODULE_VER_MINOR
<< ENA_ADMIN_HOST_INFO_MINOR_SHIFT
) |
2120 (DRV_MODULE_VER_SUBMINOR
<< ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT
);
2122 rc
= ena_com_set_host_attributes(ena_dev
);
2125 pr_warn("Cannot set host attributes\n");
2127 pr_err("Cannot set host attributes\n");
2135 ena_com_delete_host_info(ena_dev
);
2138 static void ena_config_debug_area(struct ena_adapter
*adapter
)
2140 u32 debug_area_size
;
2143 ss_count
= ena_get_sset_count(adapter
->netdev
, ETH_SS_STATS
);
2144 if (ss_count
<= 0) {
2145 netif_err(adapter
, drv
, adapter
->netdev
,
2146 "SS count is negative\n");
2150 /* allocate 32 bytes for each string and 64bit for the value */
2151 debug_area_size
= ss_count
* ETH_GSTRING_LEN
+ sizeof(u64
) * ss_count
;
2153 rc
= ena_com_allocate_debug_area(adapter
->ena_dev
, debug_area_size
);
2155 pr_err("Cannot allocate debug area\n");
2159 rc
= ena_com_set_host_attributes(adapter
->ena_dev
);
2162 netif_warn(adapter
, drv
, adapter
->netdev
,
2163 "Cannot set host attributes\n");
2165 netif_err(adapter
, drv
, adapter
->netdev
,
2166 "Cannot set host attributes\n");
2172 ena_com_delete_debug_area(adapter
->ena_dev
);
2175 static struct rtnl_link_stats64
*ena_get_stats64(struct net_device
*netdev
,
2176 struct rtnl_link_stats64
*stats
)
2178 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2179 struct ena_admin_basic_stats ena_stats
;
2182 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2185 rc
= ena_com_get_dev_basic_stats(adapter
->ena_dev
, &ena_stats
);
2189 stats
->tx_bytes
= ((u64
)ena_stats
.tx_bytes_high
<< 32) |
2190 ena_stats
.tx_bytes_low
;
2191 stats
->rx_bytes
= ((u64
)ena_stats
.rx_bytes_high
<< 32) |
2192 ena_stats
.rx_bytes_low
;
2194 stats
->rx_packets
= ((u64
)ena_stats
.rx_pkts_high
<< 32) |
2195 ena_stats
.rx_pkts_low
;
2196 stats
->tx_packets
= ((u64
)ena_stats
.tx_pkts_high
<< 32) |
2197 ena_stats
.tx_pkts_low
;
2199 stats
->rx_dropped
= ((u64
)ena_stats
.rx_drops_high
<< 32) |
2200 ena_stats
.rx_drops_low
;
2202 stats
->multicast
= 0;
2203 stats
->collisions
= 0;
2205 stats
->rx_length_errors
= 0;
2206 stats
->rx_crc_errors
= 0;
2207 stats
->rx_frame_errors
= 0;
2208 stats
->rx_fifo_errors
= 0;
2209 stats
->rx_missed_errors
= 0;
2210 stats
->tx_window_errors
= 0;
2212 stats
->rx_errors
= 0;
2213 stats
->tx_errors
= 0;
2218 static const struct net_device_ops ena_netdev_ops
= {
2219 .ndo_open
= ena_open
,
2220 .ndo_stop
= ena_close
,
2221 .ndo_start_xmit
= ena_start_xmit
,
2222 .ndo_select_queue
= ena_select_queue
,
2223 .ndo_get_stats64
= ena_get_stats64
,
2224 .ndo_tx_timeout
= ena_tx_timeout
,
2225 .ndo_change_mtu
= ena_change_mtu
,
2226 .ndo_set_mac_address
= NULL
,
2227 .ndo_validate_addr
= eth_validate_addr
,
2228 #ifdef CONFIG_NET_POLL_CONTROLLER
2229 .ndo_poll_controller
= ena_netpoll
,
2230 #endif /* CONFIG_NET_POLL_CONTROLLER */
2233 static void ena_device_io_suspend(struct work_struct
*work
)
2235 struct ena_adapter
*adapter
=
2236 container_of(work
, struct ena_adapter
, suspend_io_task
);
2237 struct net_device
*netdev
= adapter
->netdev
;
2239 /* ena_napi_disable_all disables only the IO handling.
2240 * We are still subject to AENQ keep alive watchdog.
2242 u64_stats_update_begin(&adapter
->syncp
);
2243 adapter
->dev_stats
.io_suspend
++;
2244 u64_stats_update_begin(&adapter
->syncp
);
2245 ena_napi_disable_all(adapter
);
2246 netif_tx_lock(netdev
);
2247 netif_device_detach(netdev
);
2248 netif_tx_unlock(netdev
);
2251 static void ena_device_io_resume(struct work_struct
*work
)
2253 struct ena_adapter
*adapter
=
2254 container_of(work
, struct ena_adapter
, resume_io_task
);
2255 struct net_device
*netdev
= adapter
->netdev
;
2257 u64_stats_update_begin(&adapter
->syncp
);
2258 adapter
->dev_stats
.io_resume
++;
2259 u64_stats_update_end(&adapter
->syncp
);
2261 netif_device_attach(netdev
);
2262 ena_napi_enable_all(adapter
);
2265 static int ena_device_validate_params(struct ena_adapter
*adapter
,
2266 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2268 struct net_device
*netdev
= adapter
->netdev
;
2271 rc
= ether_addr_equal(get_feat_ctx
->dev_attr
.mac_addr
,
2274 netif_err(adapter
, drv
, netdev
,
2275 "Error, mac address are different\n");
2279 if ((get_feat_ctx
->max_queues
.max_cq_num
< adapter
->num_queues
) ||
2280 (get_feat_ctx
->max_queues
.max_sq_num
< adapter
->num_queues
)) {
2281 netif_err(adapter
, drv
, netdev
,
2282 "Error, device doesn't support enough queues\n");
2286 if (get_feat_ctx
->dev_attr
.max_mtu
< netdev
->mtu
) {
2287 netif_err(adapter
, drv
, netdev
,
2288 "Error, device max mtu is smaller than netdev MTU\n");
2295 static int ena_device_init(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
,
2296 struct ena_com_dev_get_features_ctx
*get_feat_ctx
,
2299 struct device
*dev
= &pdev
->dev
;
2300 bool readless_supported
;
2305 rc
= ena_com_mmio_reg_read_request_init(ena_dev
);
2307 dev_err(dev
, "failed to init mmio read less\n");
2311 /* The PCIe configuration space revision id indicate if mmio reg
2314 readless_supported
= !(pdev
->revision
& ENA_MMIO_DISABLE_REG_READ
);
2315 ena_com_set_mmio_read_mode(ena_dev
, readless_supported
);
2317 rc
= ena_com_dev_reset(ena_dev
);
2319 dev_err(dev
, "Can not reset device\n");
2320 goto err_mmio_read_less
;
2323 rc
= ena_com_validate_version(ena_dev
);
2325 dev_err(dev
, "device version is too low\n");
2326 goto err_mmio_read_less
;
2329 dma_width
= ena_com_get_dma_width(ena_dev
);
2330 if (dma_width
< 0) {
2331 dev_err(dev
, "Invalid dma width value %d", dma_width
);
2333 goto err_mmio_read_less
;
2336 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2338 dev_err(dev
, "pci_set_dma_mask failed 0x%x\n", rc
);
2339 goto err_mmio_read_less
;
2342 rc
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2344 dev_err(dev
, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2346 goto err_mmio_read_less
;
2349 /* ENA admin level init */
2350 rc
= ena_com_admin_init(ena_dev
, &aenq_handlers
, true);
2353 "Can not initialize ena admin queue with device\n");
2354 goto err_mmio_read_less
;
2357 /* To enable the msix interrupts the driver needs to know the number
2358 * of queues. So the driver uses polling mode to retrieve this
2361 ena_com_set_admin_polling_mode(ena_dev
, true);
2363 /* Get Device Attributes*/
2364 rc
= ena_com_get_dev_attr_feat(ena_dev
, get_feat_ctx
);
2366 dev_err(dev
, "Cannot get attribute for ena device rc=%d\n", rc
);
2367 goto err_admin_init
;
2370 /* Try to turn all the available aenq groups */
2371 aenq_groups
= BIT(ENA_ADMIN_LINK_CHANGE
) |
2372 BIT(ENA_ADMIN_FATAL_ERROR
) |
2373 BIT(ENA_ADMIN_WARNING
) |
2374 BIT(ENA_ADMIN_NOTIFICATION
) |
2375 BIT(ENA_ADMIN_KEEP_ALIVE
);
2377 aenq_groups
&= get_feat_ctx
->aenq
.supported_groups
;
2379 rc
= ena_com_set_aenq_config(ena_dev
, aenq_groups
);
2381 dev_err(dev
, "Cannot configure aenq groups rc= %d\n", rc
);
2382 goto err_admin_init
;
2385 *wd_state
= !!(aenq_groups
& BIT(ENA_ADMIN_KEEP_ALIVE
));
2387 ena_config_host_info(ena_dev
);
2392 ena_com_admin_destroy(ena_dev
);
2394 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2399 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter
*adapter
,
2402 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2403 struct device
*dev
= &adapter
->pdev
->dev
;
2406 rc
= ena_enable_msix(adapter
, io_vectors
);
2408 dev_err(dev
, "Can not reserve msix vectors\n");
2412 ena_setup_mgmnt_intr(adapter
);
2414 rc
= ena_request_mgmnt_irq(adapter
);
2416 dev_err(dev
, "Can not setup management interrupts\n");
2417 goto err_disable_msix
;
2420 ena_com_set_admin_polling_mode(ena_dev
, false);
2422 ena_com_admin_aenq_enable(ena_dev
);
2427 ena_disable_msix(adapter
);
2432 static void ena_fw_reset_device(struct work_struct
*work
)
2434 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2435 struct ena_adapter
*adapter
=
2436 container_of(work
, struct ena_adapter
, reset_task
);
2437 struct net_device
*netdev
= adapter
->netdev
;
2438 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2439 struct pci_dev
*pdev
= adapter
->pdev
;
2440 bool dev_up
, wd_state
;
2443 del_timer_sync(&adapter
->timer_service
);
2447 dev_up
= test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
2448 ena_com_set_admin_running_state(ena_dev
, false);
2450 /* After calling ena_close the tx queues and the napi
2451 * are disabled so no one can interfere or touch the
2456 rc
= ena_com_dev_reset(ena_dev
);
2458 dev_err(&pdev
->dev
, "Device reset failed\n");
2462 ena_free_mgmnt_irq(adapter
);
2464 ena_disable_msix(adapter
);
2466 ena_com_abort_admin_commands(ena_dev
);
2468 ena_com_wait_for_abort_completion(ena_dev
);
2470 ena_com_admin_destroy(ena_dev
);
2472 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2474 /* Finish with the destroy part. Start the init part */
2476 rc
= ena_device_init(ena_dev
, adapter
->pdev
, &get_feat_ctx
, &wd_state
);
2478 dev_err(&pdev
->dev
, "Can not initialize device\n");
2481 adapter
->wd_state
= wd_state
;
2483 rc
= ena_device_validate_params(adapter
, &get_feat_ctx
);
2485 dev_err(&pdev
->dev
, "Validation of device parameters failed\n");
2486 goto err_device_destroy
;
2489 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
,
2490 adapter
->num_queues
);
2492 dev_err(&pdev
->dev
, "Enable MSI-X failed\n");
2493 goto err_device_destroy
;
2495 /* If the interface was up before the reset bring it up */
2497 rc
= ena_up(adapter
);
2499 dev_err(&pdev
->dev
, "Failed to create I/O queues\n");
2500 goto err_disable_msix
;
2504 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
2508 dev_err(&pdev
->dev
, "Device reset completed successfully\n");
2512 ena_free_mgmnt_irq(adapter
);
2513 ena_disable_msix(adapter
);
2515 ena_com_admin_destroy(ena_dev
);
2520 "Reset attempt failed. Can not reset the device\n");
2523 static void check_for_missing_tx_completions(struct ena_adapter
*adapter
)
2525 struct ena_tx_buffer
*tx_buf
;
2526 unsigned long last_jiffies
;
2527 struct ena_ring
*tx_ring
;
2531 /* Make sure the driver doesn't turn the device in other process */
2534 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2537 budget
= ENA_MONITORED_TX_QUEUES
;
2539 for (i
= adapter
->last_monitored_tx_qid
; i
< adapter
->num_queues
; i
++) {
2540 tx_ring
= &adapter
->tx_ring
[i
];
2542 for (j
= 0; j
< tx_ring
->ring_size
; j
++) {
2543 tx_buf
= &tx_ring
->tx_buffer_info
[j
];
2544 last_jiffies
= tx_buf
->last_jiffies
;
2545 if (unlikely(last_jiffies
&& time_is_before_jiffies(last_jiffies
+ TX_TIMEOUT
))) {
2546 netif_notice(adapter
, tx_err
, adapter
->netdev
,
2547 "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2550 u64_stats_update_begin(&tx_ring
->syncp
);
2551 missed_tx
= tx_ring
->tx_stats
.missing_tx_comp
++;
2552 u64_stats_update_end(&tx_ring
->syncp
);
2554 /* Clear last jiffies so the lost buffer won't
2557 tx_buf
->last_jiffies
= 0;
2559 if (unlikely(missed_tx
> MAX_NUM_OF_TIMEOUTED_PACKETS
)) {
2560 netif_err(adapter
, tx_err
, adapter
->netdev
,
2561 "The number of lost tx completion is above the threshold (%d > %d). Reset the device\n",
2562 missed_tx
, MAX_NUM_OF_TIMEOUTED_PACKETS
);
2563 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2573 adapter
->last_monitored_tx_qid
= i
% adapter
->num_queues
;
2576 /* Check for keep alive expiration */
2577 static void check_for_missing_keep_alive(struct ena_adapter
*adapter
)
2579 unsigned long keep_alive_expired
;
2581 if (!adapter
->wd_state
)
2584 keep_alive_expired
= round_jiffies(adapter
->last_keep_alive_jiffies
2585 + ENA_DEVICE_KALIVE_TIMEOUT
);
2586 if (unlikely(time_is_before_jiffies(keep_alive_expired
))) {
2587 netif_err(adapter
, drv
, adapter
->netdev
,
2588 "Keep alive watchdog timeout.\n");
2589 u64_stats_update_begin(&adapter
->syncp
);
2590 adapter
->dev_stats
.wd_expired
++;
2591 u64_stats_update_end(&adapter
->syncp
);
2592 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2596 static void check_for_admin_com_state(struct ena_adapter
*adapter
)
2598 if (unlikely(!ena_com_get_admin_running_state(adapter
->ena_dev
))) {
2599 netif_err(adapter
, drv
, adapter
->netdev
,
2600 "ENA admin queue is not in running state!\n");
2601 u64_stats_update_begin(&adapter
->syncp
);
2602 adapter
->dev_stats
.admin_q_pause
++;
2603 u64_stats_update_end(&adapter
->syncp
);
2604 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2608 static void ena_update_host_info(struct ena_admin_host_info
*host_info
,
2609 struct net_device
*netdev
)
2611 host_info
->supported_network_features
[0] =
2612 netdev
->features
& GENMASK_ULL(31, 0);
2613 host_info
->supported_network_features
[1] =
2614 (netdev
->features
& GENMASK_ULL(63, 32)) >> 32;
2617 static void ena_timer_service(unsigned long data
)
2619 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
2620 u8
*debug_area
= adapter
->ena_dev
->host_attr
.debug_area_virt_addr
;
2621 struct ena_admin_host_info
*host_info
=
2622 adapter
->ena_dev
->host_attr
.host_info
;
2624 check_for_missing_keep_alive(adapter
);
2626 check_for_admin_com_state(adapter
);
2628 check_for_missing_tx_completions(adapter
);
2631 ena_dump_stats_to_buf(adapter
, debug_area
);
2634 ena_update_host_info(host_info
, adapter
->netdev
);
2636 if (unlikely(test_and_clear_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2637 netif_err(adapter
, drv
, adapter
->netdev
,
2638 "Trigger reset is on\n");
2639 ena_dump_stats_to_dmesg(adapter
);
2640 queue_work(ena_wq
, &adapter
->reset_task
);
2644 /* Reset the timer */
2645 mod_timer(&adapter
->timer_service
, jiffies
+ HZ
);
2648 static int ena_calc_io_queue_num(struct pci_dev
*pdev
,
2649 struct ena_com_dev
*ena_dev
,
2650 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2652 int io_sq_num
, io_queue_num
;
2654 /* In case of LLQ use the llq number in the get feature cmd */
2655 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2656 io_sq_num
= get_feat_ctx
->max_queues
.max_llq_num
;
2658 if (io_sq_num
== 0) {
2660 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2662 ena_dev
->tx_mem_queue_type
=
2663 ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2664 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2667 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2670 io_queue_num
= min_t(int, num_possible_cpus(), ENA_MAX_NUM_IO_QUEUES
);
2671 io_queue_num
= min_t(int, io_queue_num
, io_sq_num
);
2672 io_queue_num
= min_t(int, io_queue_num
,
2673 get_feat_ctx
->max_queues
.max_cq_num
);
2674 /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2675 io_queue_num
= min_t(int, io_queue_num
, pci_msix_vec_count(pdev
) - 1);
2676 if (unlikely(!io_queue_num
)) {
2677 dev_err(&pdev
->dev
, "The device doesn't have io queues\n");
2681 return io_queue_num
;
2684 static void ena_set_push_mode(struct pci_dev
*pdev
, struct ena_com_dev
*ena_dev
,
2685 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2689 has_mem_bar
= pci_select_bars(pdev
, IORESOURCE_MEM
) & BIT(ENA_MEM_BAR
);
2691 /* Enable push mode if device supports LLQ */
2692 if (has_mem_bar
&& (get_feat_ctx
->max_queues
.max_llq_num
> 0))
2693 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_DEV
;
2695 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2698 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx
*feat
,
2699 struct net_device
*netdev
)
2701 netdev_features_t dev_features
= 0;
2703 /* Set offload features */
2704 if (feat
->offload
.tx
&
2705 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK
)
2706 dev_features
|= NETIF_F_IP_CSUM
;
2708 if (feat
->offload
.tx
&
2709 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK
)
2710 dev_features
|= NETIF_F_IPV6_CSUM
;
2712 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK
)
2713 dev_features
|= NETIF_F_TSO
;
2715 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK
)
2716 dev_features
|= NETIF_F_TSO6
;
2718 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK
)
2719 dev_features
|= NETIF_F_TSO_ECN
;
2721 if (feat
->offload
.rx_supported
&
2722 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK
)
2723 dev_features
|= NETIF_F_RXCSUM
;
2725 if (feat
->offload
.rx_supported
&
2726 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK
)
2727 dev_features
|= NETIF_F_RXCSUM
;
2736 netdev
->hw_features
|= netdev
->features
;
2737 netdev
->vlan_features
|= netdev
->features
;
2740 static void ena_set_conf_feat_params(struct ena_adapter
*adapter
,
2741 struct ena_com_dev_get_features_ctx
*feat
)
2743 struct net_device
*netdev
= adapter
->netdev
;
2745 /* Copy mac address */
2746 if (!is_valid_ether_addr(feat
->dev_attr
.mac_addr
)) {
2747 eth_hw_addr_random(netdev
);
2748 ether_addr_copy(adapter
->mac_addr
, netdev
->dev_addr
);
2750 ether_addr_copy(adapter
->mac_addr
, feat
->dev_attr
.mac_addr
);
2751 ether_addr_copy(netdev
->dev_addr
, adapter
->mac_addr
);
2754 /* Set offload features */
2755 ena_set_dev_offloads(feat
, netdev
);
2757 adapter
->max_mtu
= feat
->dev_attr
.max_mtu
;
2760 static int ena_rss_init_default(struct ena_adapter
*adapter
)
2762 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2763 struct device
*dev
= &adapter
->pdev
->dev
;
2767 rc
= ena_com_rss_init(ena_dev
, ENA_RX_RSS_TABLE_LOG_SIZE
);
2769 dev_err(dev
, "Cannot init indirect table\n");
2773 for (i
= 0; i
< ENA_RX_RSS_TABLE_SIZE
; i
++) {
2774 val
= ethtool_rxfh_indir_default(i
, adapter
->num_queues
);
2775 rc
= ena_com_indirect_table_fill_entry(ena_dev
, i
,
2776 ENA_IO_RXQ_IDX(val
));
2777 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2778 dev_err(dev
, "Cannot fill indirect table\n");
2779 goto err_fill_indir
;
2783 rc
= ena_com_fill_hash_function(ena_dev
, ENA_ADMIN_CRC32
, NULL
,
2784 ENA_HASH_KEY_SIZE
, 0xFFFFFFFF);
2785 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2786 dev_err(dev
, "Cannot fill hash function\n");
2787 goto err_fill_indir
;
2790 rc
= ena_com_set_default_hash_ctrl(ena_dev
);
2791 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2792 dev_err(dev
, "Cannot fill hash control\n");
2793 goto err_fill_indir
;
2799 ena_com_rss_destroy(ena_dev
);
2805 static void ena_release_bars(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
)
2809 release_bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2810 pci_release_selected_regions(pdev
, release_bars
);
2813 static int ena_calc_queue_size(struct pci_dev
*pdev
,
2814 struct ena_com_dev
*ena_dev
,
2815 u16
*max_tx_sgl_size
,
2816 u16
*max_rx_sgl_size
,
2817 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2819 u32 queue_size
= ENA_DEFAULT_RING_SIZE
;
2821 queue_size
= min_t(u32
, queue_size
,
2822 get_feat_ctx
->max_queues
.max_cq_depth
);
2823 queue_size
= min_t(u32
, queue_size
,
2824 get_feat_ctx
->max_queues
.max_sq_depth
);
2826 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
)
2827 queue_size
= min_t(u32
, queue_size
,
2828 get_feat_ctx
->max_queues
.max_llq_depth
);
2830 queue_size
= rounddown_pow_of_two(queue_size
);
2832 if (unlikely(!queue_size
)) {
2833 dev_err(&pdev
->dev
, "Invalid queue size\n");
2837 *max_tx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2838 get_feat_ctx
->max_queues
.max_packet_tx_descs
);
2839 *max_rx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2840 get_feat_ctx
->max_queues
.max_packet_rx_descs
);
2845 /* ena_probe - Device Initialization Routine
2846 * @pdev: PCI device information struct
2847 * @ent: entry in ena_pci_tbl
2849 * Returns 0 on success, negative on failure
2851 * ena_probe initializes an adapter identified by a pci_dev structure.
2852 * The OS initialization, configuring of the adapter private structure,
2853 * and a hardware reset occur.
2855 static int ena_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2857 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2858 static int version_printed
;
2859 struct net_device
*netdev
;
2860 struct ena_adapter
*adapter
;
2861 struct ena_com_dev
*ena_dev
= NULL
;
2862 static int adapters_found
;
2863 int io_queue_num
, bars
, rc
;
2865 u16 tx_sgl_size
= 0;
2866 u16 rx_sgl_size
= 0;
2869 dev_dbg(&pdev
->dev
, "%s\n", __func__
);
2871 if (version_printed
++ == 0)
2872 dev_info(&pdev
->dev
, "%s", version
);
2874 rc
= pci_enable_device_mem(pdev
);
2876 dev_err(&pdev
->dev
, "pci_enable_device_mem() failed!\n");
2880 pci_set_master(pdev
);
2882 ena_dev
= vzalloc(sizeof(*ena_dev
));
2885 goto err_disable_device
;
2888 bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2889 rc
= pci_request_selected_regions(pdev
, bars
, DRV_MODULE_NAME
);
2891 dev_err(&pdev
->dev
, "pci_request_selected_regions failed %d\n",
2893 goto err_free_ena_dev
;
2896 ena_dev
->reg_bar
= ioremap(pci_resource_start(pdev
, ENA_REG_BAR
),
2897 pci_resource_len(pdev
, ENA_REG_BAR
));
2898 if (!ena_dev
->reg_bar
) {
2899 dev_err(&pdev
->dev
, "failed to remap regs bar\n");
2901 goto err_free_region
;
2904 ena_dev
->dmadev
= &pdev
->dev
;
2906 rc
= ena_device_init(ena_dev
, pdev
, &get_feat_ctx
, &wd_state
);
2908 dev_err(&pdev
->dev
, "ena device init failed\n");
2911 goto err_free_region
;
2914 ena_set_push_mode(pdev
, ena_dev
, &get_feat_ctx
);
2916 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2917 ena_dev
->mem_bar
= ioremap_wc(pci_resource_start(pdev
, ENA_MEM_BAR
),
2918 pci_resource_len(pdev
, ENA_MEM_BAR
));
2919 if (!ena_dev
->mem_bar
) {
2921 goto err_device_destroy
;
2925 /* initial Tx interrupt delay, Assumes 1 usec granularity.
2926 * Updated during device initialization with the real granularity
2928 ena_dev
->intr_moder_tx_interval
= ENA_INTR_INITIAL_TX_INTERVAL_USECS
;
2929 io_queue_num
= ena_calc_io_queue_num(pdev
, ena_dev
, &get_feat_ctx
);
2930 queue_size
= ena_calc_queue_size(pdev
, ena_dev
, &tx_sgl_size
,
2931 &rx_sgl_size
, &get_feat_ctx
);
2932 if ((queue_size
<= 0) || (io_queue_num
<= 0)) {
2934 goto err_device_destroy
;
2937 dev_info(&pdev
->dev
, "creating %d io queues. queue size: %d\n",
2938 io_queue_num
, queue_size
);
2940 /* dev zeroed in init_etherdev */
2941 netdev
= alloc_etherdev_mq(sizeof(struct ena_adapter
), io_queue_num
);
2943 dev_err(&pdev
->dev
, "alloc_etherdev_mq failed\n");
2945 goto err_device_destroy
;
2948 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2950 adapter
= netdev_priv(netdev
);
2951 pci_set_drvdata(pdev
, adapter
);
2953 adapter
->ena_dev
= ena_dev
;
2954 adapter
->netdev
= netdev
;
2955 adapter
->pdev
= pdev
;
2957 ena_set_conf_feat_params(adapter
, &get_feat_ctx
);
2959 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
2961 adapter
->tx_ring_size
= queue_size
;
2962 adapter
->rx_ring_size
= queue_size
;
2964 adapter
->max_tx_sgl_size
= tx_sgl_size
;
2965 adapter
->max_rx_sgl_size
= rx_sgl_size
;
2967 adapter
->num_queues
= io_queue_num
;
2968 adapter
->last_monitored_tx_qid
= 0;
2970 adapter
->rx_copybreak
= ENA_DEFAULT_RX_COPYBREAK
;
2971 adapter
->wd_state
= wd_state
;
2973 snprintf(adapter
->name
, ENA_NAME_MAX_LEN
, "ena_%d", adapters_found
);
2975 rc
= ena_com_init_interrupt_moderation(adapter
->ena_dev
);
2978 "Failed to query interrupt moderation feature\n");
2979 goto err_netdev_destroy
;
2981 ena_init_io_rings(adapter
);
2983 netdev
->netdev_ops
= &ena_netdev_ops
;
2984 netdev
->watchdog_timeo
= TX_TIMEOUT
;
2985 ena_set_ethtool_ops(netdev
);
2987 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
2989 u64_stats_init(&adapter
->syncp
);
2991 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
, io_queue_num
);
2994 "Failed to enable and set the admin interrupts\n");
2995 goto err_worker_destroy
;
2997 rc
= ena_rss_init_default(adapter
);
2998 if (rc
&& (rc
!= -EPERM
)) {
2999 dev_err(&pdev
->dev
, "Cannot init RSS rc: %d\n", rc
);
3003 ena_config_debug_area(adapter
);
3005 memcpy(adapter
->netdev
->perm_addr
, adapter
->mac_addr
, netdev
->addr_len
);
3007 netif_carrier_off(netdev
);
3009 rc
= register_netdev(netdev
);
3011 dev_err(&pdev
->dev
, "Cannot register net device\n");
3015 INIT_WORK(&adapter
->suspend_io_task
, ena_device_io_suspend
);
3016 INIT_WORK(&adapter
->resume_io_task
, ena_device_io_resume
);
3017 INIT_WORK(&adapter
->reset_task
, ena_fw_reset_device
);
3019 adapter
->last_keep_alive_jiffies
= jiffies
;
3021 init_timer(&adapter
->timer_service
);
3022 adapter
->timer_service
.expires
= round_jiffies(jiffies
+ HZ
);
3023 adapter
->timer_service
.function
= ena_timer_service
;
3024 adapter
->timer_service
.data
= (unsigned long)adapter
;
3026 add_timer(&adapter
->timer_service
);
3028 dev_info(&pdev
->dev
, "%s found at mem %lx, mac addr %pM Queues %d\n",
3029 DEVICE_NAME
, (long)pci_resource_start(pdev
, 0),
3030 netdev
->dev_addr
, io_queue_num
);
3032 set_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
3039 ena_com_delete_debug_area(ena_dev
);
3040 ena_com_rss_destroy(ena_dev
);
3042 ena_com_dev_reset(ena_dev
);
3043 ena_free_mgmnt_irq(adapter
);
3044 ena_disable_msix(adapter
);
3046 ena_com_destroy_interrupt_moderation(ena_dev
);
3047 del_timer(&adapter
->timer_service
);
3048 cancel_work_sync(&adapter
->suspend_io_task
);
3049 cancel_work_sync(&adapter
->resume_io_task
);
3051 free_netdev(netdev
);
3053 ena_com_delete_host_info(ena_dev
);
3054 ena_com_admin_destroy(ena_dev
);
3056 ena_release_bars(ena_dev
, pdev
);
3060 pci_disable_device(pdev
);
3064 /*****************************************************************************/
3065 static int ena_sriov_configure(struct pci_dev
*dev
, int numvfs
)
3070 rc
= pci_enable_sriov(dev
, numvfs
);
3073 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3082 pci_disable_sriov(dev
);
3089 /*****************************************************************************/
3090 /*****************************************************************************/
3092 /* ena_remove - Device Removal Routine
3093 * @pdev: PCI device information struct
3095 * ena_remove is called by the PCI subsystem to alert the driver
3096 * that it should release a PCI device.
3098 static void ena_remove(struct pci_dev
*pdev
)
3100 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3101 struct ena_com_dev
*ena_dev
;
3102 struct net_device
*netdev
;
3105 /* This device didn't load properly and it's resources
3106 * already released, nothing to do
3110 ena_dev
= adapter
->ena_dev
;
3111 netdev
= adapter
->netdev
;
3113 #ifdef CONFIG_RFS_ACCEL
3114 if ((adapter
->msix_vecs
>= 1) && (netdev
->rx_cpu_rmap
)) {
3115 free_irq_cpu_rmap(netdev
->rx_cpu_rmap
);
3116 netdev
->rx_cpu_rmap
= NULL
;
3118 #endif /* CONFIG_RFS_ACCEL */
3120 unregister_netdev(netdev
);
3121 del_timer_sync(&adapter
->timer_service
);
3123 cancel_work_sync(&adapter
->reset_task
);
3125 cancel_work_sync(&adapter
->suspend_io_task
);
3127 cancel_work_sync(&adapter
->resume_io_task
);
3129 ena_com_dev_reset(ena_dev
);
3131 ena_free_mgmnt_irq(adapter
);
3133 ena_disable_msix(adapter
);
3135 free_netdev(netdev
);
3137 ena_com_mmio_reg_read_request_destroy(ena_dev
);
3139 ena_com_abort_admin_commands(ena_dev
);
3141 ena_com_wait_for_abort_completion(ena_dev
);
3143 ena_com_admin_destroy(ena_dev
);
3145 ena_com_rss_destroy(ena_dev
);
3147 ena_com_delete_debug_area(ena_dev
);
3149 ena_com_delete_host_info(ena_dev
);
3151 ena_release_bars(ena_dev
, pdev
);
3153 pci_disable_device(pdev
);
3155 ena_com_destroy_interrupt_moderation(ena_dev
);
3160 static struct pci_driver ena_pci_driver
= {
3161 .name
= DRV_MODULE_NAME
,
3162 .id_table
= ena_pci_tbl
,
3164 .remove
= ena_remove
,
3165 .sriov_configure
= ena_sriov_configure
,
3168 static int __init
ena_init(void)
3170 pr_info("%s", version
);
3172 ena_wq
= create_singlethread_workqueue(DRV_MODULE_NAME
);
3174 pr_err("Failed to create workqueue\n");
3178 return pci_register_driver(&ena_pci_driver
);
3181 static void __exit
ena_cleanup(void)
3183 pci_unregister_driver(&ena_pci_driver
);
3186 destroy_workqueue(ena_wq
);
3191 /******************************************************************************
3192 ******************************** AENQ Handlers *******************************
3193 *****************************************************************************/
3194 /* ena_update_on_link_change:
3195 * Notify the network interface about the change in link status
3197 static void ena_update_on_link_change(void *adapter_data
,
3198 struct ena_admin_aenq_entry
*aenq_e
)
3200 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3201 struct ena_admin_aenq_link_change_desc
*aenq_desc
=
3202 (struct ena_admin_aenq_link_change_desc
*)aenq_e
;
3203 int status
= aenq_desc
->flags
&
3204 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK
;
3207 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
3208 set_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3209 netif_carrier_on(adapter
->netdev
);
3211 clear_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3212 netif_carrier_off(adapter
->netdev
);
3216 static void ena_keep_alive_wd(void *adapter_data
,
3217 struct ena_admin_aenq_entry
*aenq_e
)
3219 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3221 adapter
->last_keep_alive_jiffies
= jiffies
;
3224 static void ena_notification(void *adapter_data
,
3225 struct ena_admin_aenq_entry
*aenq_e
)
3227 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3229 WARN(aenq_e
->aenq_common_desc
.group
!= ENA_ADMIN_NOTIFICATION
,
3230 "Invalid group(%x) expected %x\n",
3231 aenq_e
->aenq_common_desc
.group
,
3232 ENA_ADMIN_NOTIFICATION
);
3234 switch (aenq_e
->aenq_common_desc
.syndrom
) {
3235 case ENA_ADMIN_SUSPEND
:
3236 /* Suspend just the IO queues.
3237 * We deliberately don't suspend admin so the timer and
3238 * the keep_alive events should remain.
3240 queue_work(ena_wq
, &adapter
->suspend_io_task
);
3242 case ENA_ADMIN_RESUME
:
3243 queue_work(ena_wq
, &adapter
->resume_io_task
);
3246 netif_err(adapter
, drv
, adapter
->netdev
,
3247 "Invalid aenq notification link state %d\n",
3248 aenq_e
->aenq_common_desc
.syndrom
);
3252 /* This handler will called for unknown event group or unimplemented handlers*/
3253 static void unimplemented_aenq_handler(void *data
,
3254 struct ena_admin_aenq_entry
*aenq_e
)
3256 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
3258 netif_err(adapter
, drv
, adapter
->netdev
,
3259 "Unknown event was received or event with unimplemented handler\n");
3262 static struct ena_aenq_handlers aenq_handlers
= {
3264 [ENA_ADMIN_LINK_CHANGE
] = ena_update_on_link_change
,
3265 [ENA_ADMIN_NOTIFICATION
] = ena_notification
,
3266 [ENA_ADMIN_KEEP_ALIVE
] = ena_keep_alive_wd
,
3268 .unimplemented_handler
= unimplemented_aenq_handler
3271 module_init(ena_init
);
3272 module_exit(ena_cleanup
);