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 /* Change the state of the device to trigger reset
84 * Check that we are not in the middle or a trigger already
87 if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
90 u64_stats_update_begin(&adapter
->syncp
);
91 adapter
->dev_stats
.tx_timeout
++;
92 u64_stats_update_end(&adapter
->syncp
);
94 netif_err(adapter
, tx_err
, dev
, "Transmit time out\n");
97 static void update_rx_ring_mtu(struct ena_adapter
*adapter
, int mtu
)
101 for (i
= 0; i
< adapter
->num_queues
; i
++)
102 adapter
->rx_ring
[i
].mtu
= mtu
;
105 static int ena_change_mtu(struct net_device
*dev
, int new_mtu
)
107 struct ena_adapter
*adapter
= netdev_priv(dev
);
110 ret
= ena_com_set_dev_mtu(adapter
->ena_dev
, new_mtu
);
112 netif_dbg(adapter
, drv
, dev
, "set MTU to %d\n", new_mtu
);
113 update_rx_ring_mtu(adapter
, new_mtu
);
116 netif_err(adapter
, drv
, dev
, "Failed to set MTU to %d\n",
123 static int ena_init_rx_cpu_rmap(struct ena_adapter
*adapter
)
125 #ifdef CONFIG_RFS_ACCEL
129 adapter
->netdev
->rx_cpu_rmap
= alloc_irq_cpu_rmap(adapter
->num_queues
);
130 if (!adapter
->netdev
->rx_cpu_rmap
)
132 for (i
= 0; i
< adapter
->num_queues
; i
++) {
133 int irq_idx
= ENA_IO_IRQ_IDX(i
);
135 rc
= irq_cpu_rmap_add(adapter
->netdev
->rx_cpu_rmap
,
136 adapter
->msix_entries
[irq_idx
].vector
);
138 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
139 adapter
->netdev
->rx_cpu_rmap
= NULL
;
143 #endif /* CONFIG_RFS_ACCEL */
147 static void ena_init_io_rings_common(struct ena_adapter
*adapter
,
148 struct ena_ring
*ring
, u16 qid
)
151 ring
->pdev
= adapter
->pdev
;
152 ring
->dev
= &adapter
->pdev
->dev
;
153 ring
->netdev
= adapter
->netdev
;
154 ring
->napi
= &adapter
->ena_napi
[qid
].napi
;
155 ring
->adapter
= adapter
;
156 ring
->ena_dev
= adapter
->ena_dev
;
157 ring
->per_napi_packets
= 0;
158 ring
->per_napi_bytes
= 0;
160 u64_stats_init(&ring
->syncp
);
163 static void ena_init_io_rings(struct ena_adapter
*adapter
)
165 struct ena_com_dev
*ena_dev
;
166 struct ena_ring
*txr
, *rxr
;
169 ena_dev
= adapter
->ena_dev
;
171 for (i
= 0; i
< adapter
->num_queues
; i
++) {
172 txr
= &adapter
->tx_ring
[i
];
173 rxr
= &adapter
->rx_ring
[i
];
175 /* TX/RX common ring state */
176 ena_init_io_rings_common(adapter
, txr
, i
);
177 ena_init_io_rings_common(adapter
, rxr
, i
);
179 /* TX specific ring state */
180 txr
->ring_size
= adapter
->tx_ring_size
;
181 txr
->tx_max_header_size
= ena_dev
->tx_max_header_size
;
182 txr
->tx_mem_queue_type
= ena_dev
->tx_mem_queue_type
;
183 txr
->sgl_size
= adapter
->max_tx_sgl_size
;
184 txr
->smoothed_interval
=
185 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev
);
187 /* RX specific ring state */
188 rxr
->ring_size
= adapter
->rx_ring_size
;
189 rxr
->rx_copybreak
= adapter
->rx_copybreak
;
190 rxr
->sgl_size
= adapter
->max_rx_sgl_size
;
191 rxr
->smoothed_interval
=
192 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev
);
196 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
197 * @adapter: network interface device structure
200 * Return 0 on success, negative on failure
202 static int ena_setup_tx_resources(struct ena_adapter
*adapter
, int qid
)
204 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
205 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
208 if (tx_ring
->tx_buffer_info
) {
209 netif_err(adapter
, ifup
,
210 adapter
->netdev
, "tx_buffer_info info is not NULL");
214 size
= sizeof(struct ena_tx_buffer
) * tx_ring
->ring_size
;
215 node
= cpu_to_node(ena_irq
->cpu
);
217 tx_ring
->tx_buffer_info
= vzalloc_node(size
, node
);
218 if (!tx_ring
->tx_buffer_info
) {
219 tx_ring
->tx_buffer_info
= vzalloc(size
);
220 if (!tx_ring
->tx_buffer_info
)
224 size
= sizeof(u16
) * tx_ring
->ring_size
;
225 tx_ring
->free_tx_ids
= vzalloc_node(size
, node
);
226 if (!tx_ring
->free_tx_ids
) {
227 tx_ring
->free_tx_ids
= vzalloc(size
);
228 if (!tx_ring
->free_tx_ids
) {
229 vfree(tx_ring
->tx_buffer_info
);
234 /* Req id ring for TX out of order completions */
235 for (i
= 0; i
< tx_ring
->ring_size
; i
++)
236 tx_ring
->free_tx_ids
[i
] = i
;
238 /* Reset tx statistics */
239 memset(&tx_ring
->tx_stats
, 0x0, sizeof(tx_ring
->tx_stats
));
241 tx_ring
->next_to_use
= 0;
242 tx_ring
->next_to_clean
= 0;
243 tx_ring
->cpu
= ena_irq
->cpu
;
247 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
248 * @adapter: network interface device structure
251 * Free all transmit software resources
253 static void ena_free_tx_resources(struct ena_adapter
*adapter
, int qid
)
255 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
257 vfree(tx_ring
->tx_buffer_info
);
258 tx_ring
->tx_buffer_info
= NULL
;
260 vfree(tx_ring
->free_tx_ids
);
261 tx_ring
->free_tx_ids
= NULL
;
264 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
265 * @adapter: private structure
267 * Return 0 on success, negative on failure
269 static int ena_setup_all_tx_resources(struct ena_adapter
*adapter
)
273 for (i
= 0; i
< adapter
->num_queues
; i
++) {
274 rc
= ena_setup_tx_resources(adapter
, i
);
283 netif_err(adapter
, ifup
, adapter
->netdev
,
284 "Tx queue %d: allocation failed\n", i
);
286 /* rewind the index freeing the rings as we go */
288 ena_free_tx_resources(adapter
, i
);
292 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
293 * @adapter: board private structure
295 * Free all transmit software resources
297 static void ena_free_all_io_tx_resources(struct ena_adapter
*adapter
)
301 for (i
= 0; i
< adapter
->num_queues
; i
++)
302 ena_free_tx_resources(adapter
, i
);
305 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
306 * @adapter: network interface device structure
309 * Returns 0 on success, negative on failure
311 static int ena_setup_rx_resources(struct ena_adapter
*adapter
,
314 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
315 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
318 if (rx_ring
->rx_buffer_info
) {
319 netif_err(adapter
, ifup
, adapter
->netdev
,
320 "rx_buffer_info is not NULL");
324 /* alloc extra element so in rx path
325 * we can always prefetch rx_info + 1
327 size
= sizeof(struct ena_rx_buffer
) * (rx_ring
->ring_size
+ 1);
328 node
= cpu_to_node(ena_irq
->cpu
);
330 rx_ring
->rx_buffer_info
= vzalloc_node(size
, node
);
331 if (!rx_ring
->rx_buffer_info
) {
332 rx_ring
->rx_buffer_info
= vzalloc(size
);
333 if (!rx_ring
->rx_buffer_info
)
337 /* Reset rx statistics */
338 memset(&rx_ring
->rx_stats
, 0x0, sizeof(rx_ring
->rx_stats
));
340 rx_ring
->next_to_clean
= 0;
341 rx_ring
->next_to_use
= 0;
342 rx_ring
->cpu
= ena_irq
->cpu
;
347 /* ena_free_rx_resources - Free I/O Rx Resources
348 * @adapter: network interface device structure
351 * Free all receive software resources
353 static void ena_free_rx_resources(struct ena_adapter
*adapter
,
356 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
358 vfree(rx_ring
->rx_buffer_info
);
359 rx_ring
->rx_buffer_info
= NULL
;
362 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
363 * @adapter: board private structure
365 * Return 0 on success, negative on failure
367 static int ena_setup_all_rx_resources(struct ena_adapter
*adapter
)
371 for (i
= 0; i
< adapter
->num_queues
; i
++) {
372 rc
= ena_setup_rx_resources(adapter
, i
);
381 netif_err(adapter
, ifup
, adapter
->netdev
,
382 "Rx queue %d: allocation failed\n", i
);
384 /* rewind the index freeing the rings as we go */
386 ena_free_rx_resources(adapter
, i
);
390 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
391 * @adapter: board private structure
393 * Free all receive software resources
395 static void ena_free_all_io_rx_resources(struct ena_adapter
*adapter
)
399 for (i
= 0; i
< adapter
->num_queues
; i
++)
400 ena_free_rx_resources(adapter
, i
);
403 static inline int ena_alloc_rx_page(struct ena_ring
*rx_ring
,
404 struct ena_rx_buffer
*rx_info
, gfp_t gfp
)
406 struct ena_com_buf
*ena_buf
;
410 /* if previous allocated page is not used */
411 if (unlikely(rx_info
->page
))
414 page
= alloc_page(gfp
);
415 if (unlikely(!page
)) {
416 u64_stats_update_begin(&rx_ring
->syncp
);
417 rx_ring
->rx_stats
.page_alloc_fail
++;
418 u64_stats_update_end(&rx_ring
->syncp
);
422 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
,
424 if (unlikely(dma_mapping_error(rx_ring
->dev
, dma
))) {
425 u64_stats_update_begin(&rx_ring
->syncp
);
426 rx_ring
->rx_stats
.dma_mapping_err
++;
427 u64_stats_update_end(&rx_ring
->syncp
);
432 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
433 "alloc page %p, rx_info %p\n", page
, rx_info
);
435 rx_info
->page
= page
;
436 rx_info
->page_offset
= 0;
437 ena_buf
= &rx_info
->ena_buf
;
438 ena_buf
->paddr
= dma
;
439 ena_buf
->len
= PAGE_SIZE
;
444 static void ena_free_rx_page(struct ena_ring
*rx_ring
,
445 struct ena_rx_buffer
*rx_info
)
447 struct page
*page
= rx_info
->page
;
448 struct ena_com_buf
*ena_buf
= &rx_info
->ena_buf
;
450 if (unlikely(!page
)) {
451 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
452 "Trying to free unallocated buffer\n");
456 dma_unmap_page(rx_ring
->dev
, ena_buf
->paddr
, PAGE_SIZE
,
460 rx_info
->page
= NULL
;
463 static int ena_refill_rx_bufs(struct ena_ring
*rx_ring
, u32 num
)
469 next_to_use
= rx_ring
->next_to_use
;
471 for (i
= 0; i
< num
; i
++) {
472 struct ena_rx_buffer
*rx_info
=
473 &rx_ring
->rx_buffer_info
[next_to_use
];
475 rc
= ena_alloc_rx_page(rx_ring
, rx_info
,
476 __GFP_COLD
| GFP_ATOMIC
| __GFP_COMP
);
477 if (unlikely(rc
< 0)) {
478 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
479 "failed to alloc buffer for rx queue %d\n",
483 rc
= ena_com_add_single_rx_desc(rx_ring
->ena_com_io_sq
,
487 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
488 "failed to add buffer for rx queue %d\n",
492 next_to_use
= ENA_RX_RING_IDX_NEXT(next_to_use
,
496 if (unlikely(i
< num
)) {
497 u64_stats_update_begin(&rx_ring
->syncp
);
498 rx_ring
->rx_stats
.refil_partial
++;
499 u64_stats_update_end(&rx_ring
->syncp
);
500 netdev_warn(rx_ring
->netdev
,
501 "refilled rx qid %d with only %d buffers (from %d)\n",
502 rx_ring
->qid
, i
, num
);
506 /* Add memory barrier to make sure the desc were written before
510 ena_com_write_sq_doorbell(rx_ring
->ena_com_io_sq
);
513 rx_ring
->next_to_use
= next_to_use
;
518 static void ena_free_rx_bufs(struct ena_adapter
*adapter
,
521 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
524 for (i
= 0; i
< rx_ring
->ring_size
; i
++) {
525 struct ena_rx_buffer
*rx_info
= &rx_ring
->rx_buffer_info
[i
];
528 ena_free_rx_page(rx_ring
, rx_info
);
532 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
533 * @adapter: board private structure
536 static void ena_refill_all_rx_bufs(struct ena_adapter
*adapter
)
538 struct ena_ring
*rx_ring
;
541 for (i
= 0; i
< adapter
->num_queues
; i
++) {
542 rx_ring
= &adapter
->rx_ring
[i
];
543 bufs_num
= rx_ring
->ring_size
- 1;
544 rc
= ena_refill_rx_bufs(rx_ring
, bufs_num
);
546 if (unlikely(rc
!= bufs_num
))
547 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
548 "refilling Queue %d failed. allocated %d buffers from: %d\n",
553 static void ena_free_all_rx_bufs(struct ena_adapter
*adapter
)
557 for (i
= 0; i
< adapter
->num_queues
; i
++)
558 ena_free_rx_bufs(adapter
, i
);
561 /* ena_free_tx_bufs - Free Tx Buffers per Queue
562 * @tx_ring: TX ring for which buffers be freed
564 static void ena_free_tx_bufs(struct ena_ring
*tx_ring
)
566 bool print_once
= true;
569 for (i
= 0; i
< tx_ring
->ring_size
; i
++) {
570 struct ena_tx_buffer
*tx_info
= &tx_ring
->tx_buffer_info
[i
];
571 struct ena_com_buf
*ena_buf
;
579 netdev_notice(tx_ring
->netdev
,
580 "free uncompleted tx skb qid %d idx 0x%x\n",
584 netdev_dbg(tx_ring
->netdev
,
585 "free uncompleted tx skb qid %d idx 0x%x\n",
589 ena_buf
= tx_info
->bufs
;
590 dma_unmap_single(tx_ring
->dev
,
595 /* unmap remaining mapped pages */
596 nr_frags
= tx_info
->num_of_bufs
- 1;
597 for (j
= 0; j
< nr_frags
; j
++) {
599 dma_unmap_page(tx_ring
->dev
,
605 dev_kfree_skb_any(tx_info
->skb
);
607 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring
->netdev
,
611 static void ena_free_all_tx_bufs(struct ena_adapter
*adapter
)
613 struct ena_ring
*tx_ring
;
616 for (i
= 0; i
< adapter
->num_queues
; i
++) {
617 tx_ring
= &adapter
->tx_ring
[i
];
618 ena_free_tx_bufs(tx_ring
);
622 static void ena_destroy_all_tx_queues(struct ena_adapter
*adapter
)
627 for (i
= 0; i
< adapter
->num_queues
; i
++) {
628 ena_qid
= ENA_IO_TXQ_IDX(i
);
629 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
633 static void ena_destroy_all_rx_queues(struct ena_adapter
*adapter
)
638 for (i
= 0; i
< adapter
->num_queues
; i
++) {
639 ena_qid
= ENA_IO_RXQ_IDX(i
);
640 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
644 static void ena_destroy_all_io_queues(struct ena_adapter
*adapter
)
646 ena_destroy_all_tx_queues(adapter
);
647 ena_destroy_all_rx_queues(adapter
);
650 static int validate_tx_req_id(struct ena_ring
*tx_ring
, u16 req_id
)
652 struct ena_tx_buffer
*tx_info
= NULL
;
654 if (likely(req_id
< tx_ring
->ring_size
)) {
655 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
656 if (likely(tx_info
->skb
))
661 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
662 "tx_info doesn't have valid skb\n");
664 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
665 "Invalid req_id: %hu\n", req_id
);
667 u64_stats_update_begin(&tx_ring
->syncp
);
668 tx_ring
->tx_stats
.bad_req_id
++;
669 u64_stats_update_end(&tx_ring
->syncp
);
671 /* Trigger device reset */
672 set_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
);
676 static int ena_clean_tx_irq(struct ena_ring
*tx_ring
, u32 budget
)
678 struct netdev_queue
*txq
;
687 next_to_clean
= tx_ring
->next_to_clean
;
688 txq
= netdev_get_tx_queue(tx_ring
->netdev
, tx_ring
->qid
);
690 while (tx_pkts
< budget
) {
691 struct ena_tx_buffer
*tx_info
;
693 struct ena_com_buf
*ena_buf
;
696 rc
= ena_com_tx_comp_req_id_get(tx_ring
->ena_com_io_cq
,
701 rc
= validate_tx_req_id(tx_ring
, req_id
);
705 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
708 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
712 tx_info
->last_jiffies
= 0;
714 if (likely(tx_info
->num_of_bufs
!= 0)) {
715 ena_buf
= tx_info
->bufs
;
717 dma_unmap_single(tx_ring
->dev
,
718 dma_unmap_addr(ena_buf
, paddr
),
719 dma_unmap_len(ena_buf
, len
),
722 /* unmap remaining mapped pages */
723 nr_frags
= tx_info
->num_of_bufs
- 1;
724 for (i
= 0; i
< nr_frags
; i
++) {
726 dma_unmap_page(tx_ring
->dev
,
727 dma_unmap_addr(ena_buf
, paddr
),
728 dma_unmap_len(ena_buf
, len
),
733 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
734 "tx_poll: q %d skb %p completed\n", tx_ring
->qid
,
737 tx_bytes
+= skb
->len
;
740 total_done
+= tx_info
->tx_descs
;
742 tx_ring
->free_tx_ids
[next_to_clean
] = req_id
;
743 next_to_clean
= ENA_TX_RING_IDX_NEXT(next_to_clean
,
747 tx_ring
->next_to_clean
= next_to_clean
;
748 ena_com_comp_ack(tx_ring
->ena_com_io_sq
, total_done
);
749 ena_com_update_dev_comp_head(tx_ring
->ena_com_io_cq
);
751 netdev_tx_completed_queue(txq
, tx_pkts
, tx_bytes
);
753 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
754 "tx_poll: q %d done. total pkts: %d\n",
755 tx_ring
->qid
, tx_pkts
);
757 /* need to make the rings circular update visible to
758 * ena_start_xmit() before checking for netif_queue_stopped().
762 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
763 ENA_TX_WAKEUP_THRESH
;
764 if (unlikely(netif_tx_queue_stopped(txq
) && above_thresh
)) {
765 __netif_tx_lock(txq
, smp_processor_id());
766 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
767 ENA_TX_WAKEUP_THRESH
;
768 if (netif_tx_queue_stopped(txq
) && above_thresh
) {
769 netif_tx_wake_queue(txq
);
770 u64_stats_update_begin(&tx_ring
->syncp
);
771 tx_ring
->tx_stats
.queue_wakeup
++;
772 u64_stats_update_end(&tx_ring
->syncp
);
774 __netif_tx_unlock(txq
);
777 tx_ring
->per_napi_bytes
+= tx_bytes
;
778 tx_ring
->per_napi_packets
+= tx_pkts
;
783 static struct sk_buff
*ena_rx_skb(struct ena_ring
*rx_ring
,
784 struct ena_com_rx_buf_info
*ena_bufs
,
789 struct ena_rx_buffer
*rx_info
=
790 &rx_ring
->rx_buffer_info
[*next_to_clean
];
795 len
= ena_bufs
[0].len
;
796 if (unlikely(!rx_info
->page
)) {
797 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
802 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
803 "rx_info %p page %p\n",
804 rx_info
, rx_info
->page
);
806 /* save virt address of first buffer */
807 va
= page_address(rx_info
->page
) + rx_info
->page_offset
;
808 prefetch(va
+ NET_IP_ALIGN
);
810 if (len
<= rx_ring
->rx_copybreak
) {
811 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
812 rx_ring
->rx_copybreak
);
813 if (unlikely(!skb
)) {
814 u64_stats_update_begin(&rx_ring
->syncp
);
815 rx_ring
->rx_stats
.skb_alloc_fail
++;
816 u64_stats_update_end(&rx_ring
->syncp
);
817 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
818 "Failed to allocate skb\n");
822 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
823 "rx allocated small packet. len %d. data_len %d\n",
824 skb
->len
, skb
->data_len
);
826 /* sync this buffer for CPU use */
827 dma_sync_single_for_cpu(rx_ring
->dev
,
828 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
831 skb_copy_to_linear_data(skb
, va
, len
);
832 dma_sync_single_for_device(rx_ring
->dev
,
833 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
838 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
839 *next_to_clean
= ENA_RX_RING_IDX_ADD(*next_to_clean
, descs
,
844 skb
= napi_get_frags(rx_ring
->napi
);
845 if (unlikely(!skb
)) {
846 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
847 "Failed allocating skb\n");
848 u64_stats_update_begin(&rx_ring
->syncp
);
849 rx_ring
->rx_stats
.skb_alloc_fail
++;
850 u64_stats_update_end(&rx_ring
->syncp
);
855 dma_unmap_page(rx_ring
->dev
,
856 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
857 PAGE_SIZE
, DMA_FROM_DEVICE
);
859 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, rx_info
->page
,
860 rx_info
->page_offset
, len
, PAGE_SIZE
);
862 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
863 "rx skb updated. len %d. data_len %d\n",
864 skb
->len
, skb
->data_len
);
866 rx_info
->page
= NULL
;
868 ENA_RX_RING_IDX_NEXT(*next_to_clean
,
870 if (likely(--descs
== 0))
872 rx_info
= &rx_ring
->rx_buffer_info
[*next_to_clean
];
873 len
= ena_bufs
[++buf
].len
;
879 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
880 * @adapter: structure containing adapter specific data
881 * @ena_rx_ctx: received packet context/metadata
882 * @skb: skb currently being received and modified
884 static inline void ena_rx_checksum(struct ena_ring
*rx_ring
,
885 struct ena_com_rx_ctx
*ena_rx_ctx
,
888 /* Rx csum disabled */
889 if (unlikely(!(rx_ring
->netdev
->features
& NETIF_F_RXCSUM
))) {
890 skb
->ip_summed
= CHECKSUM_NONE
;
894 /* For fragmented packets the checksum isn't valid */
895 if (ena_rx_ctx
->frag
) {
896 skb
->ip_summed
= CHECKSUM_NONE
;
900 /* if IP and error */
901 if (unlikely((ena_rx_ctx
->l3_proto
== ENA_ETH_IO_L3_PROTO_IPV4
) &&
902 (ena_rx_ctx
->l3_csum_err
))) {
903 /* ipv4 checksum error */
904 skb
->ip_summed
= CHECKSUM_NONE
;
905 u64_stats_update_begin(&rx_ring
->syncp
);
906 rx_ring
->rx_stats
.bad_csum
++;
907 u64_stats_update_end(&rx_ring
->syncp
);
908 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
909 "RX IPv4 header checksum error\n");
914 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
915 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
))) {
916 if (unlikely(ena_rx_ctx
->l4_csum_err
)) {
917 /* TCP/UDP checksum error */
918 u64_stats_update_begin(&rx_ring
->syncp
);
919 rx_ring
->rx_stats
.bad_csum
++;
920 u64_stats_update_end(&rx_ring
->syncp
);
921 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
922 "RX L4 checksum error\n");
923 skb
->ip_summed
= CHECKSUM_NONE
;
927 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
931 static void ena_set_rx_hash(struct ena_ring
*rx_ring
,
932 struct ena_com_rx_ctx
*ena_rx_ctx
,
935 enum pkt_hash_types hash_type
;
937 if (likely(rx_ring
->netdev
->features
& NETIF_F_RXHASH
)) {
938 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
939 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
)))
941 hash_type
= PKT_HASH_TYPE_L4
;
943 hash_type
= PKT_HASH_TYPE_NONE
;
945 /* Override hash type if the packet is fragmented */
946 if (ena_rx_ctx
->frag
)
947 hash_type
= PKT_HASH_TYPE_NONE
;
949 skb_set_hash(skb
, ena_rx_ctx
->hash
, hash_type
);
953 /* ena_clean_rx_irq - Cleanup RX irq
954 * @rx_ring: RX ring to clean
955 * @napi: napi handler
956 * @budget: how many packets driver is allowed to clean
958 * Returns the number of cleaned buffers.
960 static int ena_clean_rx_irq(struct ena_ring
*rx_ring
, struct napi_struct
*napi
,
963 u16 next_to_clean
= rx_ring
->next_to_clean
;
964 u32 res_budget
, work_done
;
966 struct ena_com_rx_ctx ena_rx_ctx
;
967 struct ena_adapter
*adapter
;
970 int refill_threshold
;
973 int rx_copybreak_pkt
= 0;
975 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
976 "%s qid %d\n", __func__
, rx_ring
->qid
);
980 ena_rx_ctx
.ena_bufs
= rx_ring
->ena_bufs
;
981 ena_rx_ctx
.max_bufs
= rx_ring
->sgl_size
;
982 ena_rx_ctx
.descs
= 0;
983 rc
= ena_com_rx_pkt(rx_ring
->ena_com_io_cq
,
984 rx_ring
->ena_com_io_sq
,
989 if (unlikely(ena_rx_ctx
.descs
== 0))
992 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
993 "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
994 rx_ring
->qid
, ena_rx_ctx
.descs
, ena_rx_ctx
.l3_proto
,
995 ena_rx_ctx
.l4_proto
, ena_rx_ctx
.hash
);
997 /* allocate skb and fill it */
998 skb
= ena_rx_skb(rx_ring
, rx_ring
->ena_bufs
, ena_rx_ctx
.descs
,
1001 /* exit if we failed to retrieve a buffer */
1002 if (unlikely(!skb
)) {
1003 next_to_clean
= ENA_RX_RING_IDX_ADD(next_to_clean
,
1005 rx_ring
->ring_size
);
1009 ena_rx_checksum(rx_ring
, &ena_rx_ctx
, skb
);
1011 ena_set_rx_hash(rx_ring
, &ena_rx_ctx
, skb
);
1013 skb_record_rx_queue(skb
, rx_ring
->qid
);
1015 if (rx_ring
->ena_bufs
[0].len
<= rx_ring
->rx_copybreak
) {
1016 total_len
+= rx_ring
->ena_bufs
[0].len
;
1018 napi_gro_receive(napi
, skb
);
1020 total_len
+= skb
->len
;
1021 napi_gro_frags(napi
);
1025 } while (likely(res_budget
));
1027 work_done
= budget
- res_budget
;
1028 rx_ring
->per_napi_bytes
+= total_len
;
1029 rx_ring
->per_napi_packets
+= work_done
;
1030 u64_stats_update_begin(&rx_ring
->syncp
);
1031 rx_ring
->rx_stats
.bytes
+= total_len
;
1032 rx_ring
->rx_stats
.cnt
+= work_done
;
1033 rx_ring
->rx_stats
.rx_copybreak_pkt
+= rx_copybreak_pkt
;
1034 u64_stats_update_end(&rx_ring
->syncp
);
1036 rx_ring
->next_to_clean
= next_to_clean
;
1038 refill_required
= ena_com_sq_empty_space(rx_ring
->ena_com_io_sq
);
1039 refill_threshold
= rx_ring
->ring_size
/ ENA_RX_REFILL_THRESH_DIVIDER
;
1041 /* Optimization, try to batch new rx buffers */
1042 if (refill_required
> refill_threshold
) {
1043 ena_com_update_dev_comp_head(rx_ring
->ena_com_io_cq
);
1044 ena_refill_rx_bufs(rx_ring
, refill_required
);
1050 adapter
= netdev_priv(rx_ring
->netdev
);
1052 u64_stats_update_begin(&rx_ring
->syncp
);
1053 rx_ring
->rx_stats
.bad_desc_num
++;
1054 u64_stats_update_end(&rx_ring
->syncp
);
1056 /* Too many desc from the device. Trigger reset */
1057 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
1062 inline void ena_adjust_intr_moderation(struct ena_ring
*rx_ring
,
1063 struct ena_ring
*tx_ring
)
1065 /* We apply adaptive moderation on Rx path only.
1066 * Tx uses static interrupt moderation.
1068 ena_com_calculate_interrupt_delay(rx_ring
->ena_dev
,
1069 rx_ring
->per_napi_packets
,
1070 rx_ring
->per_napi_bytes
,
1071 &rx_ring
->smoothed_interval
,
1072 &rx_ring
->moder_tbl_idx
);
1074 /* Reset per napi packets/bytes */
1075 tx_ring
->per_napi_packets
= 0;
1076 tx_ring
->per_napi_bytes
= 0;
1077 rx_ring
->per_napi_packets
= 0;
1078 rx_ring
->per_napi_bytes
= 0;
1081 static inline void ena_update_ring_numa_node(struct ena_ring
*tx_ring
,
1082 struct ena_ring
*rx_ring
)
1084 int cpu
= get_cpu();
1087 /* Check only one ring since the 2 rings are running on the same cpu */
1088 if (likely(tx_ring
->cpu
== cpu
))
1091 numa_node
= cpu_to_node(cpu
);
1094 if (numa_node
!= NUMA_NO_NODE
) {
1095 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, numa_node
);
1096 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, numa_node
);
1107 static int ena_io_poll(struct napi_struct
*napi
, int budget
)
1109 struct ena_napi
*ena_napi
= container_of(napi
, struct ena_napi
, napi
);
1110 struct ena_ring
*tx_ring
, *rx_ring
;
1111 struct ena_eth_io_intr_reg intr_reg
;
1116 int napi_comp_call
= 0;
1119 tx_ring
= ena_napi
->tx_ring
;
1120 rx_ring
= ena_napi
->rx_ring
;
1122 tx_budget
= tx_ring
->ring_size
/ ENA_TX_POLL_BUDGET_DIVIDER
;
1124 if (!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
) ||
1125 test_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
)) {
1126 napi_complete_done(napi
, 0);
1130 tx_work_done
= ena_clean_tx_irq(tx_ring
, tx_budget
);
1131 rx_work_done
= ena_clean_rx_irq(rx_ring
, napi
, budget
);
1133 /* If the device is about to reset or down, avoid unmask
1134 * the interrupt and return 0 so NAPI won't reschedule
1136 if (unlikely(!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
) ||
1137 test_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
))) {
1138 napi_complete_done(napi
, 0);
1141 } else if ((budget
> rx_work_done
) && (tx_budget
> tx_work_done
)) {
1144 /* Update numa and unmask the interrupt only when schedule
1145 * from the interrupt context (vs from sk_busy_loop)
1147 if (napi_complete_done(napi
, rx_work_done
)) {
1148 /* Tx and Rx share the same interrupt vector */
1149 if (ena_com_get_adaptive_moderation_enabled(rx_ring
->ena_dev
))
1150 ena_adjust_intr_moderation(rx_ring
, tx_ring
);
1152 /* Update intr register: rx intr delay,
1153 * tx intr delay and interrupt unmask
1155 ena_com_update_intr_reg(&intr_reg
,
1156 rx_ring
->smoothed_interval
,
1157 tx_ring
->smoothed_interval
,
1160 /* It is a shared MSI-X.
1161 * Tx and Rx CQ have pointer to it.
1162 * So we use one of them to reach the intr reg
1164 ena_com_unmask_intr(rx_ring
->ena_com_io_cq
, &intr_reg
);
1168 ena_update_ring_numa_node(tx_ring
, rx_ring
);
1175 u64_stats_update_begin(&tx_ring
->syncp
);
1176 tx_ring
->tx_stats
.napi_comp
+= napi_comp_call
;
1177 tx_ring
->tx_stats
.tx_poll
++;
1178 u64_stats_update_end(&tx_ring
->syncp
);
1183 static irqreturn_t
ena_intr_msix_mgmnt(int irq
, void *data
)
1185 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
1187 ena_com_admin_q_comp_intr_handler(adapter
->ena_dev
);
1189 /* Don't call the aenq handler before probe is done */
1190 if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
)))
1191 ena_com_aenq_intr_handler(adapter
->ena_dev
, data
);
1196 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1197 * @irq: interrupt number
1198 * @data: pointer to a network interface private napi device structure
1200 static irqreturn_t
ena_intr_msix_io(int irq
, void *data
)
1202 struct ena_napi
*ena_napi
= data
;
1204 napi_schedule(&ena_napi
->napi
);
1209 static int ena_enable_msix(struct ena_adapter
*adapter
, int num_queues
)
1211 int i
, msix_vecs
, rc
;
1213 if (test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1214 netif_err(adapter
, probe
, adapter
->netdev
,
1215 "Error, MSI-X is already enabled\n");
1219 /* Reserved the max msix vectors we might need */
1220 msix_vecs
= ENA_MAX_MSIX_VEC(num_queues
);
1222 netif_dbg(adapter
, probe
, adapter
->netdev
,
1223 "trying to enable MSI-X, vectors %d\n", msix_vecs
);
1225 adapter
->msix_entries
= vzalloc(msix_vecs
* sizeof(struct msix_entry
));
1227 if (!adapter
->msix_entries
)
1230 for (i
= 0; i
< msix_vecs
; i
++)
1231 adapter
->msix_entries
[i
].entry
= i
;
1233 rc
= pci_enable_msix(adapter
->pdev
, adapter
->msix_entries
, msix_vecs
);
1235 netif_err(adapter
, probe
, adapter
->netdev
,
1236 "Failed to enable MSI-X, vectors %d rc %d\n",
1241 netif_dbg(adapter
, probe
, adapter
->netdev
, "enable MSI-X, vectors %d\n",
1244 if (msix_vecs
>= 1) {
1245 if (ena_init_rx_cpu_rmap(adapter
))
1246 netif_warn(adapter
, probe
, adapter
->netdev
,
1247 "Failed to map IRQs to CPUs\n");
1250 adapter
->msix_vecs
= msix_vecs
;
1251 set_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
);
1256 static void ena_setup_mgmnt_intr(struct ena_adapter
*adapter
)
1260 snprintf(adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].name
,
1261 ENA_IRQNAME_SIZE
, "ena-mgmnt@pci:%s",
1262 pci_name(adapter
->pdev
));
1263 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].handler
=
1264 ena_intr_msix_mgmnt
;
1265 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].data
= adapter
;
1266 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].vector
=
1267 adapter
->msix_entries
[ENA_MGMNT_IRQ_IDX
].vector
;
1268 cpu
= cpumask_first(cpu_online_mask
);
1269 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].cpu
= cpu
;
1270 cpumask_set_cpu(cpu
,
1271 &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].affinity_hint_mask
);
1274 static void ena_setup_io_intr(struct ena_adapter
*adapter
)
1276 struct net_device
*netdev
;
1277 int irq_idx
, i
, cpu
;
1279 netdev
= adapter
->netdev
;
1281 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1282 irq_idx
= ENA_IO_IRQ_IDX(i
);
1283 cpu
= i
% num_online_cpus();
1285 snprintf(adapter
->irq_tbl
[irq_idx
].name
, ENA_IRQNAME_SIZE
,
1286 "%s-Tx-Rx-%d", netdev
->name
, i
);
1287 adapter
->irq_tbl
[irq_idx
].handler
= ena_intr_msix_io
;
1288 adapter
->irq_tbl
[irq_idx
].data
= &adapter
->ena_napi
[i
];
1289 adapter
->irq_tbl
[irq_idx
].vector
=
1290 adapter
->msix_entries
[irq_idx
].vector
;
1291 adapter
->irq_tbl
[irq_idx
].cpu
= cpu
;
1293 cpumask_set_cpu(cpu
,
1294 &adapter
->irq_tbl
[irq_idx
].affinity_hint_mask
);
1298 static int ena_request_mgmnt_irq(struct ena_adapter
*adapter
)
1300 unsigned long flags
= 0;
1301 struct ena_irq
*irq
;
1304 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1305 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1308 netif_err(adapter
, probe
, adapter
->netdev
,
1309 "failed to request admin irq\n");
1313 netif_dbg(adapter
, probe
, adapter
->netdev
,
1314 "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1315 irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1317 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1322 static int ena_request_io_irq(struct ena_adapter
*adapter
)
1324 unsigned long flags
= 0;
1325 struct ena_irq
*irq
;
1328 if (!test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1329 netif_err(adapter
, ifup
, adapter
->netdev
,
1330 "Failed to request I/O IRQ: MSI-X is not enabled\n");
1334 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1335 irq
= &adapter
->irq_tbl
[i
];
1336 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1339 netif_err(adapter
, ifup
, adapter
->netdev
,
1340 "Failed to request I/O IRQ. index %d rc %d\n",
1345 netif_dbg(adapter
, ifup
, adapter
->netdev
,
1346 "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1347 i
, irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1349 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1355 for (k
= ENA_IO_IRQ_FIRST_IDX
; k
< i
; k
++) {
1356 irq
= &adapter
->irq_tbl
[k
];
1357 free_irq(irq
->vector
, irq
->data
);
1363 static void ena_free_mgmnt_irq(struct ena_adapter
*adapter
)
1365 struct ena_irq
*irq
;
1367 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1368 synchronize_irq(irq
->vector
);
1369 irq_set_affinity_hint(irq
->vector
, NULL
);
1370 free_irq(irq
->vector
, irq
->data
);
1373 static void ena_free_io_irq(struct ena_adapter
*adapter
)
1375 struct ena_irq
*irq
;
1378 #ifdef CONFIG_RFS_ACCEL
1379 if (adapter
->msix_vecs
>= 1) {
1380 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
1381 adapter
->netdev
->rx_cpu_rmap
= NULL
;
1383 #endif /* CONFIG_RFS_ACCEL */
1385 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1386 irq
= &adapter
->irq_tbl
[i
];
1387 irq_set_affinity_hint(irq
->vector
, NULL
);
1388 free_irq(irq
->vector
, irq
->data
);
1392 static void ena_disable_msix(struct ena_adapter
*adapter
)
1394 if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
))
1395 pci_disable_msix(adapter
->pdev
);
1397 if (adapter
->msix_entries
)
1398 vfree(adapter
->msix_entries
);
1399 adapter
->msix_entries
= NULL
;
1402 static void ena_disable_io_intr_sync(struct ena_adapter
*adapter
)
1406 if (!netif_running(adapter
->netdev
))
1409 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++)
1410 synchronize_irq(adapter
->irq_tbl
[i
].vector
);
1413 static void ena_del_napi(struct ena_adapter
*adapter
)
1417 for (i
= 0; i
< adapter
->num_queues
; i
++)
1418 netif_napi_del(&adapter
->ena_napi
[i
].napi
);
1421 static void ena_init_napi(struct ena_adapter
*adapter
)
1423 struct ena_napi
*napi
;
1426 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1427 napi
= &adapter
->ena_napi
[i
];
1429 netif_napi_add(adapter
->netdev
,
1430 &adapter
->ena_napi
[i
].napi
,
1433 napi
->rx_ring
= &adapter
->rx_ring
[i
];
1434 napi
->tx_ring
= &adapter
->tx_ring
[i
];
1439 static void ena_napi_disable_all(struct ena_adapter
*adapter
)
1443 for (i
= 0; i
< adapter
->num_queues
; i
++)
1444 napi_disable(&adapter
->ena_napi
[i
].napi
);
1447 static void ena_napi_enable_all(struct ena_adapter
*adapter
)
1451 for (i
= 0; i
< adapter
->num_queues
; i
++)
1452 napi_enable(&adapter
->ena_napi
[i
].napi
);
1455 static void ena_restore_ethtool_params(struct ena_adapter
*adapter
)
1457 adapter
->tx_usecs
= 0;
1458 adapter
->rx_usecs
= 0;
1459 adapter
->tx_frames
= 1;
1460 adapter
->rx_frames
= 1;
1463 /* Configure the Rx forwarding */
1464 static int ena_rss_configure(struct ena_adapter
*adapter
)
1466 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1469 /* In case the RSS table wasn't initialized by probe */
1470 if (!ena_dev
->rss
.tbl_log_size
) {
1471 rc
= ena_rss_init_default(adapter
);
1472 if (rc
&& (rc
!= -EPERM
)) {
1473 netif_err(adapter
, ifup
, adapter
->netdev
,
1474 "Failed to init RSS rc: %d\n", rc
);
1479 /* Set indirect table */
1480 rc
= ena_com_indirect_table_set(ena_dev
);
1481 if (unlikely(rc
&& rc
!= -EPERM
))
1484 /* Configure hash function (if supported) */
1485 rc
= ena_com_set_hash_function(ena_dev
);
1486 if (unlikely(rc
&& (rc
!= -EPERM
)))
1489 /* Configure hash inputs (if supported) */
1490 rc
= ena_com_set_hash_ctrl(ena_dev
);
1491 if (unlikely(rc
&& (rc
!= -EPERM
)))
1497 static int ena_up_complete(struct ena_adapter
*adapter
)
1501 rc
= ena_rss_configure(adapter
);
1505 ena_init_napi(adapter
);
1507 ena_change_mtu(adapter
->netdev
, adapter
->netdev
->mtu
);
1509 ena_refill_all_rx_bufs(adapter
);
1511 /* enable transmits */
1512 netif_tx_start_all_queues(adapter
->netdev
);
1514 ena_restore_ethtool_params(adapter
);
1516 ena_napi_enable_all(adapter
);
1518 /* schedule napi in case we had pending packets
1519 * from the last time we disable napi
1521 for (i
= 0; i
< adapter
->num_queues
; i
++)
1522 napi_schedule(&adapter
->ena_napi
[i
].napi
);
1527 static int ena_create_io_tx_queue(struct ena_adapter
*adapter
, int qid
)
1529 struct ena_com_create_io_ctx ctx
= { 0 };
1530 struct ena_com_dev
*ena_dev
;
1531 struct ena_ring
*tx_ring
;
1536 ena_dev
= adapter
->ena_dev
;
1538 tx_ring
= &adapter
->tx_ring
[qid
];
1539 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1540 ena_qid
= ENA_IO_TXQ_IDX(qid
);
1542 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_TX
;
1544 ctx
.mem_queue_type
= ena_dev
->tx_mem_queue_type
;
1545 ctx
.msix_vector
= msix_vector
;
1546 ctx
.queue_size
= adapter
->tx_ring_size
;
1547 ctx
.numa_node
= cpu_to_node(tx_ring
->cpu
);
1549 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1551 netif_err(adapter
, ifup
, adapter
->netdev
,
1552 "Failed to create I/O TX queue num %d rc: %d\n",
1557 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1558 &tx_ring
->ena_com_io_sq
,
1559 &tx_ring
->ena_com_io_cq
);
1561 netif_err(adapter
, ifup
, adapter
->netdev
,
1562 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1564 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1567 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1571 static int ena_create_all_io_tx_queues(struct ena_adapter
*adapter
)
1573 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1576 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1577 rc
= ena_create_io_tx_queue(adapter
, i
);
1586 ena_com_destroy_io_queue(ena_dev
, ENA_IO_TXQ_IDX(i
));
1591 static int ena_create_io_rx_queue(struct ena_adapter
*adapter
, int qid
)
1593 struct ena_com_dev
*ena_dev
;
1594 struct ena_com_create_io_ctx ctx
= { 0 };
1595 struct ena_ring
*rx_ring
;
1600 ena_dev
= adapter
->ena_dev
;
1602 rx_ring
= &adapter
->rx_ring
[qid
];
1603 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1604 ena_qid
= ENA_IO_RXQ_IDX(qid
);
1607 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_RX
;
1608 ctx
.mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
1609 ctx
.msix_vector
= msix_vector
;
1610 ctx
.queue_size
= adapter
->rx_ring_size
;
1611 ctx
.numa_node
= cpu_to_node(rx_ring
->cpu
);
1613 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1615 netif_err(adapter
, ifup
, adapter
->netdev
,
1616 "Failed to create I/O RX queue num %d rc: %d\n",
1621 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1622 &rx_ring
->ena_com_io_sq
,
1623 &rx_ring
->ena_com_io_cq
);
1625 netif_err(adapter
, ifup
, adapter
->netdev
,
1626 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1628 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1631 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1636 static int ena_create_all_io_rx_queues(struct ena_adapter
*adapter
)
1638 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1641 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1642 rc
= ena_create_io_rx_queue(adapter
, i
);
1651 ena_com_destroy_io_queue(ena_dev
, ENA_IO_RXQ_IDX(i
));
1656 static int ena_up(struct ena_adapter
*adapter
)
1660 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
1662 ena_setup_io_intr(adapter
);
1664 rc
= ena_request_io_irq(adapter
);
1668 /* allocate transmit descriptors */
1669 rc
= ena_setup_all_tx_resources(adapter
);
1673 /* allocate receive descriptors */
1674 rc
= ena_setup_all_rx_resources(adapter
);
1678 /* Create TX queues */
1679 rc
= ena_create_all_io_tx_queues(adapter
);
1681 goto err_create_tx_queues
;
1683 /* Create RX queues */
1684 rc
= ena_create_all_io_rx_queues(adapter
);
1686 goto err_create_rx_queues
;
1688 rc
= ena_up_complete(adapter
);
1692 if (test_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
))
1693 netif_carrier_on(adapter
->netdev
);
1695 u64_stats_update_begin(&adapter
->syncp
);
1696 adapter
->dev_stats
.interface_up
++;
1697 u64_stats_update_end(&adapter
->syncp
);
1699 set_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1704 ena_destroy_all_rx_queues(adapter
);
1705 err_create_rx_queues
:
1706 ena_destroy_all_tx_queues(adapter
);
1707 err_create_tx_queues
:
1708 ena_free_all_io_rx_resources(adapter
);
1710 ena_free_all_io_tx_resources(adapter
);
1712 ena_free_io_irq(adapter
);
1718 static void ena_down(struct ena_adapter
*adapter
)
1720 netif_info(adapter
, ifdown
, adapter
->netdev
, "%s\n", __func__
);
1722 clear_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1724 u64_stats_update_begin(&adapter
->syncp
);
1725 adapter
->dev_stats
.interface_down
++;
1726 u64_stats_update_end(&adapter
->syncp
);
1728 netif_carrier_off(adapter
->netdev
);
1729 netif_tx_disable(adapter
->netdev
);
1731 /* After this point the napi handler won't enable the tx queue */
1732 ena_napi_disable_all(adapter
);
1734 /* After destroy the queue there won't be any new interrupts */
1736 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
)) {
1739 rc
= ena_com_dev_reset(adapter
->ena_dev
);
1741 dev_err(&adapter
->pdev
->dev
, "Device reset failed\n");
1744 ena_destroy_all_io_queues(adapter
);
1746 ena_disable_io_intr_sync(adapter
);
1747 ena_free_io_irq(adapter
);
1748 ena_del_napi(adapter
);
1750 ena_free_all_tx_bufs(adapter
);
1751 ena_free_all_rx_bufs(adapter
);
1752 ena_free_all_io_tx_resources(adapter
);
1753 ena_free_all_io_rx_resources(adapter
);
1756 /* ena_open - Called when a network interface is made active
1757 * @netdev: network interface device structure
1759 * Returns 0 on success, negative value on failure
1761 * The open entry point is called when a network interface is made
1762 * active by the system (IFF_UP). At this point all resources needed
1763 * for transmit and receive operations are allocated, the interrupt
1764 * handler is registered with the OS, the watchdog timer is started,
1765 * and the stack is notified that the interface is ready.
1767 static int ena_open(struct net_device
*netdev
)
1769 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1772 /* Notify the stack of the actual queue counts. */
1773 rc
= netif_set_real_num_tx_queues(netdev
, adapter
->num_queues
);
1775 netif_err(adapter
, ifup
, netdev
, "Can't set num tx queues\n");
1779 rc
= netif_set_real_num_rx_queues(netdev
, adapter
->num_queues
);
1781 netif_err(adapter
, ifup
, netdev
, "Can't set num rx queues\n");
1785 rc
= ena_up(adapter
);
1792 /* ena_close - Disables a network interface
1793 * @netdev: network interface device structure
1795 * Returns 0, this is not allowed to fail
1797 * The close entry point is called when an interface is de-activated
1798 * by the OS. The hardware is still under the drivers control, but
1799 * needs to be disabled. A global MAC reset is issued to stop the
1800 * hardware, and all transmit and receive resources are freed.
1802 static int ena_close(struct net_device
*netdev
)
1804 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1806 netif_dbg(adapter
, ifdown
, netdev
, "%s\n", __func__
);
1808 if (test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
1814 static void ena_tx_csum(struct ena_com_tx_ctx
*ena_tx_ctx
, struct sk_buff
*skb
)
1816 u32 mss
= skb_shinfo(skb
)->gso_size
;
1817 struct ena_com_tx_meta
*ena_meta
= &ena_tx_ctx
->ena_meta
;
1820 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) || mss
) {
1821 ena_tx_ctx
->l4_csum_enable
= 1;
1823 ena_tx_ctx
->tso_enable
= 1;
1824 ena_meta
->l4_hdr_len
= tcp_hdr(skb
)->doff
;
1825 ena_tx_ctx
->l4_csum_partial
= 0;
1827 ena_tx_ctx
->tso_enable
= 0;
1828 ena_meta
->l4_hdr_len
= 0;
1829 ena_tx_ctx
->l4_csum_partial
= 1;
1832 switch (ip_hdr(skb
)->version
) {
1834 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV4
;
1835 if (ip_hdr(skb
)->frag_off
& htons(IP_DF
))
1838 ena_tx_ctx
->l3_csum_enable
= 1;
1839 l4_protocol
= ip_hdr(skb
)->protocol
;
1842 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV6
;
1843 l4_protocol
= ipv6_hdr(skb
)->nexthdr
;
1849 if (l4_protocol
== IPPROTO_TCP
)
1850 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_TCP
;
1852 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_UDP
;
1854 ena_meta
->mss
= mss
;
1855 ena_meta
->l3_hdr_len
= skb_network_header_len(skb
);
1856 ena_meta
->l3_hdr_offset
= skb_network_offset(skb
);
1857 ena_tx_ctx
->meta_valid
= 1;
1860 ena_tx_ctx
->meta_valid
= 0;
1864 static int ena_check_and_linearize_skb(struct ena_ring
*tx_ring
,
1865 struct sk_buff
*skb
)
1867 int num_frags
, header_len
, rc
;
1869 num_frags
= skb_shinfo(skb
)->nr_frags
;
1870 header_len
= skb_headlen(skb
);
1872 if (num_frags
< tx_ring
->sgl_size
)
1875 if ((num_frags
== tx_ring
->sgl_size
) &&
1876 (header_len
< tx_ring
->tx_max_header_size
))
1879 u64_stats_update_begin(&tx_ring
->syncp
);
1880 tx_ring
->tx_stats
.linearize
++;
1881 u64_stats_update_end(&tx_ring
->syncp
);
1883 rc
= skb_linearize(skb
);
1885 u64_stats_update_begin(&tx_ring
->syncp
);
1886 tx_ring
->tx_stats
.linearize_failed
++;
1887 u64_stats_update_end(&tx_ring
->syncp
);
1893 /* Called with netif_tx_lock. */
1894 static netdev_tx_t
ena_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1896 struct ena_adapter
*adapter
= netdev_priv(dev
);
1897 struct ena_tx_buffer
*tx_info
;
1898 struct ena_com_tx_ctx ena_tx_ctx
;
1899 struct ena_ring
*tx_ring
;
1900 struct netdev_queue
*txq
;
1901 struct ena_com_buf
*ena_buf
;
1909 int qid
, rc
, nb_hw_desc
;
1912 netif_dbg(adapter
, tx_queued
, dev
, "%s skb %p\n", __func__
, skb
);
1913 /* Determine which tx ring we will be placed on */
1914 qid
= skb_get_queue_mapping(skb
);
1915 tx_ring
= &adapter
->tx_ring
[qid
];
1916 txq
= netdev_get_tx_queue(dev
, qid
);
1918 rc
= ena_check_and_linearize_skb(tx_ring
, skb
);
1920 goto error_drop_packet
;
1922 skb_tx_timestamp(skb
);
1923 len
= skb_headlen(skb
);
1925 next_to_use
= tx_ring
->next_to_use
;
1926 req_id
= tx_ring
->free_tx_ids
[next_to_use
];
1927 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
1928 tx_info
->num_of_bufs
= 0;
1930 WARN(tx_info
->skb
, "SKB isn't NULL req_id %d\n", req_id
);
1931 ena_buf
= tx_info
->bufs
;
1934 if (tx_ring
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
1935 /* prepared the push buffer */
1936 push_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1937 header_len
= push_len
;
1938 push_hdr
= skb
->data
;
1941 header_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1945 netif_dbg(adapter
, tx_queued
, dev
,
1946 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb
,
1947 push_hdr
, push_len
);
1949 if (len
> push_len
) {
1950 dma
= dma_map_single(tx_ring
->dev
, skb
->data
+ push_len
,
1951 len
- push_len
, DMA_TO_DEVICE
);
1952 if (dma_mapping_error(tx_ring
->dev
, dma
))
1953 goto error_report_dma_error
;
1955 ena_buf
->paddr
= dma
;
1956 ena_buf
->len
= len
- push_len
;
1959 tx_info
->num_of_bufs
++;
1962 last_frag
= skb_shinfo(skb
)->nr_frags
;
1964 for (i
= 0; i
< last_frag
; i
++) {
1965 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1967 len
= skb_frag_size(frag
);
1968 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, len
,
1970 if (dma_mapping_error(tx_ring
->dev
, dma
))
1971 goto error_report_dma_error
;
1973 ena_buf
->paddr
= dma
;
1978 tx_info
->num_of_bufs
+= last_frag
;
1980 memset(&ena_tx_ctx
, 0x0, sizeof(struct ena_com_tx_ctx
));
1981 ena_tx_ctx
.ena_bufs
= tx_info
->bufs
;
1982 ena_tx_ctx
.push_header
= push_hdr
;
1983 ena_tx_ctx
.num_bufs
= tx_info
->num_of_bufs
;
1984 ena_tx_ctx
.req_id
= req_id
;
1985 ena_tx_ctx
.header_len
= header_len
;
1987 /* set flags and meta data */
1988 ena_tx_csum(&ena_tx_ctx
, skb
);
1990 /* prepare the packet's descriptors to dma engine */
1991 rc
= ena_com_prepare_tx(tx_ring
->ena_com_io_sq
, &ena_tx_ctx
,
1995 netif_err(adapter
, tx_queued
, dev
,
1996 "failed to prepare tx bufs\n");
1997 u64_stats_update_begin(&tx_ring
->syncp
);
1998 tx_ring
->tx_stats
.queue_stop
++;
1999 tx_ring
->tx_stats
.prepare_ctx_err
++;
2000 u64_stats_update_end(&tx_ring
->syncp
);
2001 netif_tx_stop_queue(txq
);
2002 goto error_unmap_dma
;
2005 netdev_tx_sent_queue(txq
, skb
->len
);
2007 u64_stats_update_begin(&tx_ring
->syncp
);
2008 tx_ring
->tx_stats
.cnt
++;
2009 tx_ring
->tx_stats
.bytes
+= skb
->len
;
2010 u64_stats_update_end(&tx_ring
->syncp
);
2012 tx_info
->tx_descs
= nb_hw_desc
;
2013 tx_info
->last_jiffies
= jiffies
;
2015 tx_ring
->next_to_use
= ENA_TX_RING_IDX_NEXT(next_to_use
,
2016 tx_ring
->ring_size
);
2018 /* This WMB is aimed to:
2019 * 1 - perform smp barrier before reading next_to_completion
2020 * 2 - make sure the desc were written before trigger DB
2024 /* stop the queue when no more space available, the packet can have up
2025 * to sgl_size + 2. one for the meta descriptor and one for header
2026 * (if the header is larger than tx_max_header_size).
2028 if (unlikely(ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) <
2029 (tx_ring
->sgl_size
+ 2))) {
2030 netif_dbg(adapter
, tx_queued
, dev
, "%s stop queue %d\n",
2033 netif_tx_stop_queue(txq
);
2034 u64_stats_update_begin(&tx_ring
->syncp
);
2035 tx_ring
->tx_stats
.queue_stop
++;
2036 u64_stats_update_end(&tx_ring
->syncp
);
2038 /* There is a rare condition where this function decide to
2039 * stop the queue but meanwhile clean_tx_irq updates
2040 * next_to_completion and terminates.
2041 * The queue will remain stopped forever.
2042 * To solve this issue this function perform rmb, check
2043 * the wakeup condition and wake up the queue if needed.
2047 if (ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
)
2048 > ENA_TX_WAKEUP_THRESH
) {
2049 netif_tx_wake_queue(txq
);
2050 u64_stats_update_begin(&tx_ring
->syncp
);
2051 tx_ring
->tx_stats
.queue_wakeup
++;
2052 u64_stats_update_end(&tx_ring
->syncp
);
2056 if (netif_xmit_stopped(txq
) || !skb
->xmit_more
) {
2057 /* trigger the dma engine */
2058 ena_com_write_sq_doorbell(tx_ring
->ena_com_io_sq
);
2059 u64_stats_update_begin(&tx_ring
->syncp
);
2060 tx_ring
->tx_stats
.doorbells
++;
2061 u64_stats_update_end(&tx_ring
->syncp
);
2064 return NETDEV_TX_OK
;
2066 error_report_dma_error
:
2067 u64_stats_update_begin(&tx_ring
->syncp
);
2068 tx_ring
->tx_stats
.dma_mapping_err
++;
2069 u64_stats_update_end(&tx_ring
->syncp
);
2070 netdev_warn(adapter
->netdev
, "failed to map skb\n");
2072 tx_info
->skb
= NULL
;
2076 /* save value of frag that failed */
2079 /* start back at beginning and unmap skb */
2080 tx_info
->skb
= NULL
;
2081 ena_buf
= tx_info
->bufs
;
2082 dma_unmap_single(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2083 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2085 /* unmap remaining mapped pages */
2086 for (i
= 0; i
< last_frag
; i
++) {
2088 dma_unmap_page(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2089 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2096 return NETDEV_TX_OK
;
2099 #ifdef CONFIG_NET_POLL_CONTROLLER
2100 static void ena_netpoll(struct net_device
*netdev
)
2102 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2105 /* Dont schedule NAPI if the driver is in the middle of reset
2106 * or netdev is down.
2109 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
) ||
2110 test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2113 for (i
= 0; i
< adapter
->num_queues
; i
++)
2114 napi_schedule(&adapter
->ena_napi
[i
].napi
);
2116 #endif /* CONFIG_NET_POLL_CONTROLLER */
2118 static u16
ena_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
2119 void *accel_priv
, select_queue_fallback_t fallback
)
2122 /* we suspect that this is good for in--kernel network services that
2123 * want to loop incoming skb rx to tx in normal user generated traffic,
2124 * most probably we will not get to this
2126 if (skb_rx_queue_recorded(skb
))
2127 qid
= skb_get_rx_queue(skb
);
2129 qid
= fallback(dev
, skb
);
2134 static void ena_config_host_info(struct ena_com_dev
*ena_dev
)
2136 struct ena_admin_host_info
*host_info
;
2139 /* Allocate only the host info */
2140 rc
= ena_com_allocate_host_info(ena_dev
);
2142 pr_err("Cannot allocate host info\n");
2146 host_info
= ena_dev
->host_attr
.host_info
;
2148 host_info
->os_type
= ENA_ADMIN_OS_LINUX
;
2149 host_info
->kernel_ver
= LINUX_VERSION_CODE
;
2150 strncpy(host_info
->kernel_ver_str
, utsname()->version
,
2151 sizeof(host_info
->kernel_ver_str
) - 1);
2152 host_info
->os_dist
= 0;
2153 strncpy(host_info
->os_dist_str
, utsname()->release
,
2154 sizeof(host_info
->os_dist_str
) - 1);
2155 host_info
->driver_version
=
2156 (DRV_MODULE_VER_MAJOR
) |
2157 (DRV_MODULE_VER_MINOR
<< ENA_ADMIN_HOST_INFO_MINOR_SHIFT
) |
2158 (DRV_MODULE_VER_SUBMINOR
<< ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT
);
2160 rc
= ena_com_set_host_attributes(ena_dev
);
2163 pr_warn("Cannot set host attributes\n");
2165 pr_err("Cannot set host attributes\n");
2173 ena_com_delete_host_info(ena_dev
);
2176 static void ena_config_debug_area(struct ena_adapter
*adapter
)
2178 u32 debug_area_size
;
2181 ss_count
= ena_get_sset_count(adapter
->netdev
, ETH_SS_STATS
);
2182 if (ss_count
<= 0) {
2183 netif_err(adapter
, drv
, adapter
->netdev
,
2184 "SS count is negative\n");
2188 /* allocate 32 bytes for each string and 64bit for the value */
2189 debug_area_size
= ss_count
* ETH_GSTRING_LEN
+ sizeof(u64
) * ss_count
;
2191 rc
= ena_com_allocate_debug_area(adapter
->ena_dev
, debug_area_size
);
2193 pr_err("Cannot allocate debug area\n");
2197 rc
= ena_com_set_host_attributes(adapter
->ena_dev
);
2200 netif_warn(adapter
, drv
, adapter
->netdev
,
2201 "Cannot set host attributes\n");
2203 netif_err(adapter
, drv
, adapter
->netdev
,
2204 "Cannot set host attributes\n");
2210 ena_com_delete_debug_area(adapter
->ena_dev
);
2213 static struct rtnl_link_stats64
*ena_get_stats64(struct net_device
*netdev
,
2214 struct rtnl_link_stats64
*stats
)
2216 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2217 struct ena_ring
*rx_ring
, *tx_ring
;
2222 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2225 for (i
= 0; i
< adapter
->num_queues
; i
++) {
2228 tx_ring
= &adapter
->tx_ring
[i
];
2231 start
= u64_stats_fetch_begin_irq(&tx_ring
->syncp
);
2232 packets
= tx_ring
->tx_stats
.cnt
;
2233 bytes
= tx_ring
->tx_stats
.bytes
;
2234 } while (u64_stats_fetch_retry_irq(&tx_ring
->syncp
, start
));
2236 stats
->tx_packets
+= packets
;
2237 stats
->tx_bytes
+= bytes
;
2239 rx_ring
= &adapter
->rx_ring
[i
];
2242 start
= u64_stats_fetch_begin_irq(&rx_ring
->syncp
);
2243 packets
= rx_ring
->rx_stats
.cnt
;
2244 bytes
= rx_ring
->rx_stats
.bytes
;
2245 } while (u64_stats_fetch_retry_irq(&rx_ring
->syncp
, start
));
2247 stats
->rx_packets
+= packets
;
2248 stats
->rx_bytes
+= bytes
;
2252 start
= u64_stats_fetch_begin_irq(&adapter
->syncp
);
2253 rx_drops
= adapter
->dev_stats
.rx_drops
;
2254 } while (u64_stats_fetch_retry_irq(&adapter
->syncp
, start
));
2256 stats
->rx_dropped
= rx_drops
;
2258 stats
->multicast
= 0;
2259 stats
->collisions
= 0;
2261 stats
->rx_length_errors
= 0;
2262 stats
->rx_crc_errors
= 0;
2263 stats
->rx_frame_errors
= 0;
2264 stats
->rx_fifo_errors
= 0;
2265 stats
->rx_missed_errors
= 0;
2266 stats
->tx_window_errors
= 0;
2268 stats
->rx_errors
= 0;
2269 stats
->tx_errors
= 0;
2274 static const struct net_device_ops ena_netdev_ops
= {
2275 .ndo_open
= ena_open
,
2276 .ndo_stop
= ena_close
,
2277 .ndo_start_xmit
= ena_start_xmit
,
2278 .ndo_select_queue
= ena_select_queue
,
2279 .ndo_get_stats64
= ena_get_stats64
,
2280 .ndo_tx_timeout
= ena_tx_timeout
,
2281 .ndo_change_mtu
= ena_change_mtu
,
2282 .ndo_set_mac_address
= NULL
,
2283 .ndo_validate_addr
= eth_validate_addr
,
2284 #ifdef CONFIG_NET_POLL_CONTROLLER
2285 .ndo_poll_controller
= ena_netpoll
,
2286 #endif /* CONFIG_NET_POLL_CONTROLLER */
2289 static void ena_device_io_suspend(struct work_struct
*work
)
2291 struct ena_adapter
*adapter
=
2292 container_of(work
, struct ena_adapter
, suspend_io_task
);
2293 struct net_device
*netdev
= adapter
->netdev
;
2295 /* ena_napi_disable_all disables only the IO handling.
2296 * We are still subject to AENQ keep alive watchdog.
2298 u64_stats_update_begin(&adapter
->syncp
);
2299 adapter
->dev_stats
.io_suspend
++;
2300 u64_stats_update_begin(&adapter
->syncp
);
2301 ena_napi_disable_all(adapter
);
2302 netif_tx_lock(netdev
);
2303 netif_device_detach(netdev
);
2304 netif_tx_unlock(netdev
);
2307 static void ena_device_io_resume(struct work_struct
*work
)
2309 struct ena_adapter
*adapter
=
2310 container_of(work
, struct ena_adapter
, resume_io_task
);
2311 struct net_device
*netdev
= adapter
->netdev
;
2313 u64_stats_update_begin(&adapter
->syncp
);
2314 adapter
->dev_stats
.io_resume
++;
2315 u64_stats_update_end(&adapter
->syncp
);
2317 netif_device_attach(netdev
);
2318 ena_napi_enable_all(adapter
);
2321 static int ena_device_validate_params(struct ena_adapter
*adapter
,
2322 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2324 struct net_device
*netdev
= adapter
->netdev
;
2327 rc
= ether_addr_equal(get_feat_ctx
->dev_attr
.mac_addr
,
2330 netif_err(adapter
, drv
, netdev
,
2331 "Error, mac address are different\n");
2335 if ((get_feat_ctx
->max_queues
.max_cq_num
< adapter
->num_queues
) ||
2336 (get_feat_ctx
->max_queues
.max_sq_num
< adapter
->num_queues
)) {
2337 netif_err(adapter
, drv
, netdev
,
2338 "Error, device doesn't support enough queues\n");
2342 if (get_feat_ctx
->dev_attr
.max_mtu
< netdev
->mtu
) {
2343 netif_err(adapter
, drv
, netdev
,
2344 "Error, device max mtu is smaller than netdev MTU\n");
2351 static int ena_device_init(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
,
2352 struct ena_com_dev_get_features_ctx
*get_feat_ctx
,
2355 struct device
*dev
= &pdev
->dev
;
2356 bool readless_supported
;
2361 rc
= ena_com_mmio_reg_read_request_init(ena_dev
);
2363 dev_err(dev
, "failed to init mmio read less\n");
2367 /* The PCIe configuration space revision id indicate if mmio reg
2370 readless_supported
= !(pdev
->revision
& ENA_MMIO_DISABLE_REG_READ
);
2371 ena_com_set_mmio_read_mode(ena_dev
, readless_supported
);
2373 rc
= ena_com_dev_reset(ena_dev
);
2375 dev_err(dev
, "Can not reset device\n");
2376 goto err_mmio_read_less
;
2379 rc
= ena_com_validate_version(ena_dev
);
2381 dev_err(dev
, "device version is too low\n");
2382 goto err_mmio_read_less
;
2385 dma_width
= ena_com_get_dma_width(ena_dev
);
2386 if (dma_width
< 0) {
2387 dev_err(dev
, "Invalid dma width value %d", dma_width
);
2389 goto err_mmio_read_less
;
2392 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2394 dev_err(dev
, "pci_set_dma_mask failed 0x%x\n", rc
);
2395 goto err_mmio_read_less
;
2398 rc
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2400 dev_err(dev
, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2402 goto err_mmio_read_less
;
2405 /* ENA admin level init */
2406 rc
= ena_com_admin_init(ena_dev
, &aenq_handlers
, true);
2409 "Can not initialize ena admin queue with device\n");
2410 goto err_mmio_read_less
;
2413 /* To enable the msix interrupts the driver needs to know the number
2414 * of queues. So the driver uses polling mode to retrieve this
2417 ena_com_set_admin_polling_mode(ena_dev
, true);
2419 ena_config_host_info(ena_dev
);
2421 /* Get Device Attributes*/
2422 rc
= ena_com_get_dev_attr_feat(ena_dev
, get_feat_ctx
);
2424 dev_err(dev
, "Cannot get attribute for ena device rc=%d\n", rc
);
2425 goto err_admin_init
;
2428 /* Try to turn all the available aenq groups */
2429 aenq_groups
= BIT(ENA_ADMIN_LINK_CHANGE
) |
2430 BIT(ENA_ADMIN_FATAL_ERROR
) |
2431 BIT(ENA_ADMIN_WARNING
) |
2432 BIT(ENA_ADMIN_NOTIFICATION
) |
2433 BIT(ENA_ADMIN_KEEP_ALIVE
);
2435 aenq_groups
&= get_feat_ctx
->aenq
.supported_groups
;
2437 rc
= ena_com_set_aenq_config(ena_dev
, aenq_groups
);
2439 dev_err(dev
, "Cannot configure aenq groups rc= %d\n", rc
);
2440 goto err_admin_init
;
2443 *wd_state
= !!(aenq_groups
& BIT(ENA_ADMIN_KEEP_ALIVE
));
2448 ena_com_delete_host_info(ena_dev
);
2449 ena_com_admin_destroy(ena_dev
);
2451 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2456 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter
*adapter
,
2459 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2460 struct device
*dev
= &adapter
->pdev
->dev
;
2463 rc
= ena_enable_msix(adapter
, io_vectors
);
2465 dev_err(dev
, "Can not reserve msix vectors\n");
2469 ena_setup_mgmnt_intr(adapter
);
2471 rc
= ena_request_mgmnt_irq(adapter
);
2473 dev_err(dev
, "Can not setup management interrupts\n");
2474 goto err_disable_msix
;
2477 ena_com_set_admin_polling_mode(ena_dev
, false);
2479 ena_com_admin_aenq_enable(ena_dev
);
2484 ena_disable_msix(adapter
);
2489 static void ena_fw_reset_device(struct work_struct
*work
)
2491 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2492 struct ena_adapter
*adapter
=
2493 container_of(work
, struct ena_adapter
, reset_task
);
2494 struct net_device
*netdev
= adapter
->netdev
;
2495 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2496 struct pci_dev
*pdev
= adapter
->pdev
;
2497 bool dev_up
, wd_state
;
2500 if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2502 "device reset schedule while reset bit is off\n");
2506 netif_carrier_off(netdev
);
2508 del_timer_sync(&adapter
->timer_service
);
2512 dev_up
= test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
2513 ena_com_set_admin_running_state(ena_dev
, false);
2515 /* After calling ena_close the tx queues and the napi
2516 * are disabled so no one can interfere or touch the
2521 ena_free_mgmnt_irq(adapter
);
2523 ena_disable_msix(adapter
);
2525 ena_com_abort_admin_commands(ena_dev
);
2527 ena_com_wait_for_abort_completion(ena_dev
);
2529 ena_com_admin_destroy(ena_dev
);
2531 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2533 clear_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2535 /* Finish with the destroy part. Start the init part */
2537 rc
= ena_device_init(ena_dev
, adapter
->pdev
, &get_feat_ctx
, &wd_state
);
2539 dev_err(&pdev
->dev
, "Can not initialize device\n");
2542 adapter
->wd_state
= wd_state
;
2544 rc
= ena_device_validate_params(adapter
, &get_feat_ctx
);
2546 dev_err(&pdev
->dev
, "Validation of device parameters failed\n");
2547 goto err_device_destroy
;
2550 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
,
2551 adapter
->num_queues
);
2553 dev_err(&pdev
->dev
, "Enable MSI-X failed\n");
2554 goto err_device_destroy
;
2556 /* If the interface was up before the reset bring it up */
2558 rc
= ena_up(adapter
);
2560 dev_err(&pdev
->dev
, "Failed to create I/O queues\n");
2561 goto err_disable_msix
;
2565 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
2569 dev_err(&pdev
->dev
, "Device reset completed successfully\n");
2573 ena_free_mgmnt_irq(adapter
);
2574 ena_disable_msix(adapter
);
2576 ena_com_admin_destroy(ena_dev
);
2580 clear_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
2583 "Reset attempt failed. Can not reset the device\n");
2586 static void check_for_missing_tx_completions(struct ena_adapter
*adapter
)
2588 struct ena_tx_buffer
*tx_buf
;
2589 unsigned long last_jiffies
;
2590 struct ena_ring
*tx_ring
;
2594 /* Make sure the driver doesn't turn the device in other process */
2597 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2600 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2603 budget
= ENA_MONITORED_TX_QUEUES
;
2605 for (i
= adapter
->last_monitored_tx_qid
; i
< adapter
->num_queues
; i
++) {
2606 tx_ring
= &adapter
->tx_ring
[i
];
2608 for (j
= 0; j
< tx_ring
->ring_size
; j
++) {
2609 tx_buf
= &tx_ring
->tx_buffer_info
[j
];
2610 last_jiffies
= tx_buf
->last_jiffies
;
2611 if (unlikely(last_jiffies
&& time_is_before_jiffies(last_jiffies
+ TX_TIMEOUT
))) {
2612 netif_notice(adapter
, tx_err
, adapter
->netdev
,
2613 "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2616 u64_stats_update_begin(&tx_ring
->syncp
);
2617 missed_tx
= tx_ring
->tx_stats
.missing_tx_comp
++;
2618 u64_stats_update_end(&tx_ring
->syncp
);
2620 /* Clear last jiffies so the lost buffer won't
2623 tx_buf
->last_jiffies
= 0;
2625 if (unlikely(missed_tx
> MAX_NUM_OF_TIMEOUTED_PACKETS
)) {
2626 netif_err(adapter
, tx_err
, adapter
->netdev
,
2627 "The number of lost tx completion is above the threshold (%d > %d). Reset the device\n",
2628 missed_tx
, MAX_NUM_OF_TIMEOUTED_PACKETS
);
2629 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2639 adapter
->last_monitored_tx_qid
= i
% adapter
->num_queues
;
2642 /* Check for keep alive expiration */
2643 static void check_for_missing_keep_alive(struct ena_adapter
*adapter
)
2645 unsigned long keep_alive_expired
;
2647 if (!adapter
->wd_state
)
2650 keep_alive_expired
= round_jiffies(adapter
->last_keep_alive_jiffies
2651 + ENA_DEVICE_KALIVE_TIMEOUT
);
2652 if (unlikely(time_is_before_jiffies(keep_alive_expired
))) {
2653 netif_err(adapter
, drv
, adapter
->netdev
,
2654 "Keep alive watchdog timeout.\n");
2655 u64_stats_update_begin(&adapter
->syncp
);
2656 adapter
->dev_stats
.wd_expired
++;
2657 u64_stats_update_end(&adapter
->syncp
);
2658 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2662 static void check_for_admin_com_state(struct ena_adapter
*adapter
)
2664 if (unlikely(!ena_com_get_admin_running_state(adapter
->ena_dev
))) {
2665 netif_err(adapter
, drv
, adapter
->netdev
,
2666 "ENA admin queue is not in running state!\n");
2667 u64_stats_update_begin(&adapter
->syncp
);
2668 adapter
->dev_stats
.admin_q_pause
++;
2669 u64_stats_update_end(&adapter
->syncp
);
2670 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2674 static void ena_update_host_info(struct ena_admin_host_info
*host_info
,
2675 struct net_device
*netdev
)
2677 host_info
->supported_network_features
[0] =
2678 netdev
->features
& GENMASK_ULL(31, 0);
2679 host_info
->supported_network_features
[1] =
2680 (netdev
->features
& GENMASK_ULL(63, 32)) >> 32;
2683 static void ena_timer_service(unsigned long data
)
2685 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
2686 u8
*debug_area
= adapter
->ena_dev
->host_attr
.debug_area_virt_addr
;
2687 struct ena_admin_host_info
*host_info
=
2688 adapter
->ena_dev
->host_attr
.host_info
;
2690 check_for_missing_keep_alive(adapter
);
2692 check_for_admin_com_state(adapter
);
2694 check_for_missing_tx_completions(adapter
);
2697 ena_dump_stats_to_buf(adapter
, debug_area
);
2700 ena_update_host_info(host_info
, adapter
->netdev
);
2702 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2703 netif_err(adapter
, drv
, adapter
->netdev
,
2704 "Trigger reset is on\n");
2705 ena_dump_stats_to_dmesg(adapter
);
2706 queue_work(ena_wq
, &adapter
->reset_task
);
2710 /* Reset the timer */
2711 mod_timer(&adapter
->timer_service
, jiffies
+ HZ
);
2714 static int ena_calc_io_queue_num(struct pci_dev
*pdev
,
2715 struct ena_com_dev
*ena_dev
,
2716 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2718 int io_sq_num
, io_queue_num
;
2720 /* In case of LLQ use the llq number in the get feature cmd */
2721 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2722 io_sq_num
= get_feat_ctx
->max_queues
.max_llq_num
;
2724 if (io_sq_num
== 0) {
2726 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2728 ena_dev
->tx_mem_queue_type
=
2729 ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2730 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2733 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2736 io_queue_num
= min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES
);
2737 io_queue_num
= min_t(int, io_queue_num
, io_sq_num
);
2738 io_queue_num
= min_t(int, io_queue_num
,
2739 get_feat_ctx
->max_queues
.max_cq_num
);
2740 /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2741 io_queue_num
= min_t(int, io_queue_num
, pci_msix_vec_count(pdev
) - 1);
2742 if (unlikely(!io_queue_num
)) {
2743 dev_err(&pdev
->dev
, "The device doesn't have io queues\n");
2747 return io_queue_num
;
2750 static void ena_set_push_mode(struct pci_dev
*pdev
, struct ena_com_dev
*ena_dev
,
2751 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2755 has_mem_bar
= pci_select_bars(pdev
, IORESOURCE_MEM
) & BIT(ENA_MEM_BAR
);
2757 /* Enable push mode if device supports LLQ */
2758 if (has_mem_bar
&& (get_feat_ctx
->max_queues
.max_llq_num
> 0))
2759 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_DEV
;
2761 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2764 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx
*feat
,
2765 struct net_device
*netdev
)
2767 netdev_features_t dev_features
= 0;
2769 /* Set offload features */
2770 if (feat
->offload
.tx
&
2771 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK
)
2772 dev_features
|= NETIF_F_IP_CSUM
;
2774 if (feat
->offload
.tx
&
2775 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK
)
2776 dev_features
|= NETIF_F_IPV6_CSUM
;
2778 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK
)
2779 dev_features
|= NETIF_F_TSO
;
2781 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK
)
2782 dev_features
|= NETIF_F_TSO6
;
2784 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK
)
2785 dev_features
|= NETIF_F_TSO_ECN
;
2787 if (feat
->offload
.rx_supported
&
2788 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK
)
2789 dev_features
|= NETIF_F_RXCSUM
;
2791 if (feat
->offload
.rx_supported
&
2792 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK
)
2793 dev_features
|= NETIF_F_RXCSUM
;
2801 netdev
->hw_features
|= netdev
->features
;
2802 netdev
->vlan_features
|= netdev
->features
;
2805 static void ena_set_conf_feat_params(struct ena_adapter
*adapter
,
2806 struct ena_com_dev_get_features_ctx
*feat
)
2808 struct net_device
*netdev
= adapter
->netdev
;
2810 /* Copy mac address */
2811 if (!is_valid_ether_addr(feat
->dev_attr
.mac_addr
)) {
2812 eth_hw_addr_random(netdev
);
2813 ether_addr_copy(adapter
->mac_addr
, netdev
->dev_addr
);
2815 ether_addr_copy(adapter
->mac_addr
, feat
->dev_attr
.mac_addr
);
2816 ether_addr_copy(netdev
->dev_addr
, adapter
->mac_addr
);
2819 /* Set offload features */
2820 ena_set_dev_offloads(feat
, netdev
);
2822 adapter
->max_mtu
= feat
->dev_attr
.max_mtu
;
2823 netdev
->max_mtu
= adapter
->max_mtu
;
2824 netdev
->min_mtu
= ENA_MIN_MTU
;
2827 static int ena_rss_init_default(struct ena_adapter
*adapter
)
2829 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2830 struct device
*dev
= &adapter
->pdev
->dev
;
2834 rc
= ena_com_rss_init(ena_dev
, ENA_RX_RSS_TABLE_LOG_SIZE
);
2836 dev_err(dev
, "Cannot init indirect table\n");
2840 for (i
= 0; i
< ENA_RX_RSS_TABLE_SIZE
; i
++) {
2841 val
= ethtool_rxfh_indir_default(i
, adapter
->num_queues
);
2842 rc
= ena_com_indirect_table_fill_entry(ena_dev
, i
,
2843 ENA_IO_RXQ_IDX(val
));
2844 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2845 dev_err(dev
, "Cannot fill indirect table\n");
2846 goto err_fill_indir
;
2850 rc
= ena_com_fill_hash_function(ena_dev
, ENA_ADMIN_CRC32
, NULL
,
2851 ENA_HASH_KEY_SIZE
, 0xFFFFFFFF);
2852 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2853 dev_err(dev
, "Cannot fill hash function\n");
2854 goto err_fill_indir
;
2857 rc
= ena_com_set_default_hash_ctrl(ena_dev
);
2858 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2859 dev_err(dev
, "Cannot fill hash control\n");
2860 goto err_fill_indir
;
2866 ena_com_rss_destroy(ena_dev
);
2872 static void ena_release_bars(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
)
2876 release_bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2877 pci_release_selected_regions(pdev
, release_bars
);
2880 static int ena_calc_queue_size(struct pci_dev
*pdev
,
2881 struct ena_com_dev
*ena_dev
,
2882 u16
*max_tx_sgl_size
,
2883 u16
*max_rx_sgl_size
,
2884 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2886 u32 queue_size
= ENA_DEFAULT_RING_SIZE
;
2888 queue_size
= min_t(u32
, queue_size
,
2889 get_feat_ctx
->max_queues
.max_cq_depth
);
2890 queue_size
= min_t(u32
, queue_size
,
2891 get_feat_ctx
->max_queues
.max_sq_depth
);
2893 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
)
2894 queue_size
= min_t(u32
, queue_size
,
2895 get_feat_ctx
->max_queues
.max_llq_depth
);
2897 queue_size
= rounddown_pow_of_two(queue_size
);
2899 if (unlikely(!queue_size
)) {
2900 dev_err(&pdev
->dev
, "Invalid queue size\n");
2904 *max_tx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2905 get_feat_ctx
->max_queues
.max_packet_tx_descs
);
2906 *max_rx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2907 get_feat_ctx
->max_queues
.max_packet_rx_descs
);
2912 /* ena_probe - Device Initialization Routine
2913 * @pdev: PCI device information struct
2914 * @ent: entry in ena_pci_tbl
2916 * Returns 0 on success, negative on failure
2918 * ena_probe initializes an adapter identified by a pci_dev structure.
2919 * The OS initialization, configuring of the adapter private structure,
2920 * and a hardware reset occur.
2922 static int ena_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2924 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2925 static int version_printed
;
2926 struct net_device
*netdev
;
2927 struct ena_adapter
*adapter
;
2928 struct ena_com_dev
*ena_dev
= NULL
;
2929 static int adapters_found
;
2930 int io_queue_num
, bars
, rc
;
2932 u16 tx_sgl_size
= 0;
2933 u16 rx_sgl_size
= 0;
2936 dev_dbg(&pdev
->dev
, "%s\n", __func__
);
2938 if (version_printed
++ == 0)
2939 dev_info(&pdev
->dev
, "%s", version
);
2941 rc
= pci_enable_device_mem(pdev
);
2943 dev_err(&pdev
->dev
, "pci_enable_device_mem() failed!\n");
2947 pci_set_master(pdev
);
2949 ena_dev
= vzalloc(sizeof(*ena_dev
));
2952 goto err_disable_device
;
2955 bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2956 rc
= pci_request_selected_regions(pdev
, bars
, DRV_MODULE_NAME
);
2958 dev_err(&pdev
->dev
, "pci_request_selected_regions failed %d\n",
2960 goto err_free_ena_dev
;
2963 ena_dev
->reg_bar
= ioremap(pci_resource_start(pdev
, ENA_REG_BAR
),
2964 pci_resource_len(pdev
, ENA_REG_BAR
));
2965 if (!ena_dev
->reg_bar
) {
2966 dev_err(&pdev
->dev
, "failed to remap regs bar\n");
2968 goto err_free_region
;
2971 ena_dev
->dmadev
= &pdev
->dev
;
2973 rc
= ena_device_init(ena_dev
, pdev
, &get_feat_ctx
, &wd_state
);
2975 dev_err(&pdev
->dev
, "ena device init failed\n");
2978 goto err_free_region
;
2981 ena_set_push_mode(pdev
, ena_dev
, &get_feat_ctx
);
2983 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2984 ena_dev
->mem_bar
= ioremap_wc(pci_resource_start(pdev
, ENA_MEM_BAR
),
2985 pci_resource_len(pdev
, ENA_MEM_BAR
));
2986 if (!ena_dev
->mem_bar
) {
2988 goto err_device_destroy
;
2992 /* initial Tx interrupt delay, Assumes 1 usec granularity.
2993 * Updated during device initialization with the real granularity
2995 ena_dev
->intr_moder_tx_interval
= ENA_INTR_INITIAL_TX_INTERVAL_USECS
;
2996 io_queue_num
= ena_calc_io_queue_num(pdev
, ena_dev
, &get_feat_ctx
);
2997 queue_size
= ena_calc_queue_size(pdev
, ena_dev
, &tx_sgl_size
,
2998 &rx_sgl_size
, &get_feat_ctx
);
2999 if ((queue_size
<= 0) || (io_queue_num
<= 0)) {
3001 goto err_device_destroy
;
3004 dev_info(&pdev
->dev
, "creating %d io queues. queue size: %d\n",
3005 io_queue_num
, queue_size
);
3007 /* dev zeroed in init_etherdev */
3008 netdev
= alloc_etherdev_mq(sizeof(struct ena_adapter
), io_queue_num
);
3010 dev_err(&pdev
->dev
, "alloc_etherdev_mq failed\n");
3012 goto err_device_destroy
;
3015 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
3017 adapter
= netdev_priv(netdev
);
3018 pci_set_drvdata(pdev
, adapter
);
3020 adapter
->ena_dev
= ena_dev
;
3021 adapter
->netdev
= netdev
;
3022 adapter
->pdev
= pdev
;
3024 ena_set_conf_feat_params(adapter
, &get_feat_ctx
);
3026 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
3028 adapter
->tx_ring_size
= queue_size
;
3029 adapter
->rx_ring_size
= queue_size
;
3031 adapter
->max_tx_sgl_size
= tx_sgl_size
;
3032 adapter
->max_rx_sgl_size
= rx_sgl_size
;
3034 adapter
->num_queues
= io_queue_num
;
3035 adapter
->last_monitored_tx_qid
= 0;
3037 adapter
->rx_copybreak
= ENA_DEFAULT_RX_COPYBREAK
;
3038 adapter
->wd_state
= wd_state
;
3040 snprintf(adapter
->name
, ENA_NAME_MAX_LEN
, "ena_%d", adapters_found
);
3042 rc
= ena_com_init_interrupt_moderation(adapter
->ena_dev
);
3045 "Failed to query interrupt moderation feature\n");
3046 goto err_netdev_destroy
;
3048 ena_init_io_rings(adapter
);
3050 netdev
->netdev_ops
= &ena_netdev_ops
;
3051 netdev
->watchdog_timeo
= TX_TIMEOUT
;
3052 ena_set_ethtool_ops(netdev
);
3054 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
3056 u64_stats_init(&adapter
->syncp
);
3058 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
, io_queue_num
);
3061 "Failed to enable and set the admin interrupts\n");
3062 goto err_worker_destroy
;
3064 rc
= ena_rss_init_default(adapter
);
3065 if (rc
&& (rc
!= -EPERM
)) {
3066 dev_err(&pdev
->dev
, "Cannot init RSS rc: %d\n", rc
);
3070 ena_config_debug_area(adapter
);
3072 memcpy(adapter
->netdev
->perm_addr
, adapter
->mac_addr
, netdev
->addr_len
);
3074 netif_carrier_off(netdev
);
3076 rc
= register_netdev(netdev
);
3078 dev_err(&pdev
->dev
, "Cannot register net device\n");
3082 INIT_WORK(&adapter
->suspend_io_task
, ena_device_io_suspend
);
3083 INIT_WORK(&adapter
->resume_io_task
, ena_device_io_resume
);
3084 INIT_WORK(&adapter
->reset_task
, ena_fw_reset_device
);
3086 adapter
->last_keep_alive_jiffies
= jiffies
;
3088 setup_timer(&adapter
->timer_service
, ena_timer_service
,
3089 (unsigned long)adapter
);
3090 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
3092 dev_info(&pdev
->dev
, "%s found at mem %lx, mac addr %pM Queues %d\n",
3093 DEVICE_NAME
, (long)pci_resource_start(pdev
, 0),
3094 netdev
->dev_addr
, io_queue_num
);
3096 set_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
3103 ena_com_delete_debug_area(ena_dev
);
3104 ena_com_rss_destroy(ena_dev
);
3106 ena_com_dev_reset(ena_dev
);
3107 ena_free_mgmnt_irq(adapter
);
3108 ena_disable_msix(adapter
);
3110 ena_com_destroy_interrupt_moderation(ena_dev
);
3111 del_timer(&adapter
->timer_service
);
3112 cancel_work_sync(&adapter
->suspend_io_task
);
3113 cancel_work_sync(&adapter
->resume_io_task
);
3115 free_netdev(netdev
);
3117 ena_com_delete_host_info(ena_dev
);
3118 ena_com_admin_destroy(ena_dev
);
3120 ena_release_bars(ena_dev
, pdev
);
3124 pci_disable_device(pdev
);
3128 /*****************************************************************************/
3129 static int ena_sriov_configure(struct pci_dev
*dev
, int numvfs
)
3134 rc
= pci_enable_sriov(dev
, numvfs
);
3137 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3146 pci_disable_sriov(dev
);
3153 /*****************************************************************************/
3154 /*****************************************************************************/
3156 /* ena_remove - Device Removal Routine
3157 * @pdev: PCI device information struct
3159 * ena_remove is called by the PCI subsystem to alert the driver
3160 * that it should release a PCI device.
3162 static void ena_remove(struct pci_dev
*pdev
)
3164 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3165 struct ena_com_dev
*ena_dev
;
3166 struct net_device
*netdev
;
3169 /* This device didn't load properly and it's resources
3170 * already released, nothing to do
3174 ena_dev
= adapter
->ena_dev
;
3175 netdev
= adapter
->netdev
;
3177 #ifdef CONFIG_RFS_ACCEL
3178 if ((adapter
->msix_vecs
>= 1) && (netdev
->rx_cpu_rmap
)) {
3179 free_irq_cpu_rmap(netdev
->rx_cpu_rmap
);
3180 netdev
->rx_cpu_rmap
= NULL
;
3182 #endif /* CONFIG_RFS_ACCEL */
3184 unregister_netdev(netdev
);
3185 del_timer_sync(&adapter
->timer_service
);
3187 cancel_work_sync(&adapter
->reset_task
);
3189 cancel_work_sync(&adapter
->suspend_io_task
);
3191 cancel_work_sync(&adapter
->resume_io_task
);
3193 /* Reset the device only if the device is running. */
3194 if (test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
))
3195 ena_com_dev_reset(ena_dev
);
3197 ena_free_mgmnt_irq(adapter
);
3199 ena_disable_msix(adapter
);
3201 free_netdev(netdev
);
3203 ena_com_mmio_reg_read_request_destroy(ena_dev
);
3205 ena_com_abort_admin_commands(ena_dev
);
3207 ena_com_wait_for_abort_completion(ena_dev
);
3209 ena_com_admin_destroy(ena_dev
);
3211 ena_com_rss_destroy(ena_dev
);
3213 ena_com_delete_debug_area(ena_dev
);
3215 ena_com_delete_host_info(ena_dev
);
3217 ena_release_bars(ena_dev
, pdev
);
3219 pci_disable_device(pdev
);
3221 ena_com_destroy_interrupt_moderation(ena_dev
);
3226 static struct pci_driver ena_pci_driver
= {
3227 .name
= DRV_MODULE_NAME
,
3228 .id_table
= ena_pci_tbl
,
3230 .remove
= ena_remove
,
3231 .sriov_configure
= ena_sriov_configure
,
3234 static int __init
ena_init(void)
3236 pr_info("%s", version
);
3238 ena_wq
= create_singlethread_workqueue(DRV_MODULE_NAME
);
3240 pr_err("Failed to create workqueue\n");
3244 return pci_register_driver(&ena_pci_driver
);
3247 static void __exit
ena_cleanup(void)
3249 pci_unregister_driver(&ena_pci_driver
);
3252 destroy_workqueue(ena_wq
);
3257 /******************************************************************************
3258 ******************************** AENQ Handlers *******************************
3259 *****************************************************************************/
3260 /* ena_update_on_link_change:
3261 * Notify the network interface about the change in link status
3263 static void ena_update_on_link_change(void *adapter_data
,
3264 struct ena_admin_aenq_entry
*aenq_e
)
3266 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3267 struct ena_admin_aenq_link_change_desc
*aenq_desc
=
3268 (struct ena_admin_aenq_link_change_desc
*)aenq_e
;
3269 int status
= aenq_desc
->flags
&
3270 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK
;
3273 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
3274 set_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3275 netif_carrier_on(adapter
->netdev
);
3277 clear_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3278 netif_carrier_off(adapter
->netdev
);
3282 static void ena_keep_alive_wd(void *adapter_data
,
3283 struct ena_admin_aenq_entry
*aenq_e
)
3285 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3287 adapter
->last_keep_alive_jiffies
= jiffies
;
3290 static void ena_notification(void *adapter_data
,
3291 struct ena_admin_aenq_entry
*aenq_e
)
3293 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3295 WARN(aenq_e
->aenq_common_desc
.group
!= ENA_ADMIN_NOTIFICATION
,
3296 "Invalid group(%x) expected %x\n",
3297 aenq_e
->aenq_common_desc
.group
,
3298 ENA_ADMIN_NOTIFICATION
);
3300 switch (aenq_e
->aenq_common_desc
.syndrom
) {
3301 case ENA_ADMIN_SUSPEND
:
3302 /* Suspend just the IO queues.
3303 * We deliberately don't suspend admin so the timer and
3304 * the keep_alive events should remain.
3306 queue_work(ena_wq
, &adapter
->suspend_io_task
);
3308 case ENA_ADMIN_RESUME
:
3309 queue_work(ena_wq
, &adapter
->resume_io_task
);
3312 netif_err(adapter
, drv
, adapter
->netdev
,
3313 "Invalid aenq notification link state %d\n",
3314 aenq_e
->aenq_common_desc
.syndrom
);
3318 /* This handler will called for unknown event group or unimplemented handlers*/
3319 static void unimplemented_aenq_handler(void *data
,
3320 struct ena_admin_aenq_entry
*aenq_e
)
3322 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
3324 netif_err(adapter
, drv
, adapter
->netdev
,
3325 "Unknown event was received or event with unimplemented handler\n");
3328 static struct ena_aenq_handlers aenq_handlers
= {
3330 [ENA_ADMIN_LINK_CHANGE
] = ena_update_on_link_change
,
3331 [ENA_ADMIN_NOTIFICATION
] = ena_notification
,
3332 [ENA_ADMIN_KEEP_ALIVE
] = ena_keep_alive_wd
,
3334 .unimplemented_handler
= unimplemented_aenq_handler
3337 module_init(ena_init
);
3338 module_exit(ena_cleanup
);