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
);
78 static void check_for_admin_com_state(struct ena_adapter
*adapter
);
79 static void ena_destroy_device(struct ena_adapter
*adapter
);
80 static int ena_restore_device(struct ena_adapter
*adapter
);
82 static void ena_tx_timeout(struct net_device
*dev
)
84 struct ena_adapter
*adapter
= netdev_priv(dev
);
86 /* Change the state of the device to trigger reset
87 * Check that we are not in the middle or a trigger already
90 if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
93 adapter
->reset_reason
= ENA_REGS_RESET_OS_NETDEV_WD
;
94 u64_stats_update_begin(&adapter
->syncp
);
95 adapter
->dev_stats
.tx_timeout
++;
96 u64_stats_update_end(&adapter
->syncp
);
98 netif_err(adapter
, tx_err
, dev
, "Transmit time out\n");
101 static void update_rx_ring_mtu(struct ena_adapter
*adapter
, int mtu
)
105 for (i
= 0; i
< adapter
->num_queues
; i
++)
106 adapter
->rx_ring
[i
].mtu
= mtu
;
109 static int ena_change_mtu(struct net_device
*dev
, int new_mtu
)
111 struct ena_adapter
*adapter
= netdev_priv(dev
);
114 ret
= ena_com_set_dev_mtu(adapter
->ena_dev
, new_mtu
);
116 netif_dbg(adapter
, drv
, dev
, "set MTU to %d\n", new_mtu
);
117 update_rx_ring_mtu(adapter
, new_mtu
);
120 netif_err(adapter
, drv
, dev
, "Failed to set MTU to %d\n",
127 static int ena_init_rx_cpu_rmap(struct ena_adapter
*adapter
)
129 #ifdef CONFIG_RFS_ACCEL
133 adapter
->netdev
->rx_cpu_rmap
= alloc_irq_cpu_rmap(adapter
->num_queues
);
134 if (!adapter
->netdev
->rx_cpu_rmap
)
136 for (i
= 0; i
< adapter
->num_queues
; i
++) {
137 int irq_idx
= ENA_IO_IRQ_IDX(i
);
139 rc
= irq_cpu_rmap_add(adapter
->netdev
->rx_cpu_rmap
,
140 pci_irq_vector(adapter
->pdev
, irq_idx
));
142 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
143 adapter
->netdev
->rx_cpu_rmap
= NULL
;
147 #endif /* CONFIG_RFS_ACCEL */
151 static void ena_init_io_rings_common(struct ena_adapter
*adapter
,
152 struct ena_ring
*ring
, u16 qid
)
155 ring
->pdev
= adapter
->pdev
;
156 ring
->dev
= &adapter
->pdev
->dev
;
157 ring
->netdev
= adapter
->netdev
;
158 ring
->napi
= &adapter
->ena_napi
[qid
].napi
;
159 ring
->adapter
= adapter
;
160 ring
->ena_dev
= adapter
->ena_dev
;
161 ring
->per_napi_packets
= 0;
162 ring
->per_napi_bytes
= 0;
164 u64_stats_init(&ring
->syncp
);
167 static void ena_init_io_rings(struct ena_adapter
*adapter
)
169 struct ena_com_dev
*ena_dev
;
170 struct ena_ring
*txr
, *rxr
;
173 ena_dev
= adapter
->ena_dev
;
175 for (i
= 0; i
< adapter
->num_queues
; i
++) {
176 txr
= &adapter
->tx_ring
[i
];
177 rxr
= &adapter
->rx_ring
[i
];
179 /* TX/RX common ring state */
180 ena_init_io_rings_common(adapter
, txr
, i
);
181 ena_init_io_rings_common(adapter
, rxr
, i
);
183 /* TX specific ring state */
184 txr
->ring_size
= adapter
->tx_ring_size
;
185 txr
->tx_max_header_size
= ena_dev
->tx_max_header_size
;
186 txr
->tx_mem_queue_type
= ena_dev
->tx_mem_queue_type
;
187 txr
->sgl_size
= adapter
->max_tx_sgl_size
;
188 txr
->smoothed_interval
=
189 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev
);
191 /* RX specific ring state */
192 rxr
->ring_size
= adapter
->rx_ring_size
;
193 rxr
->rx_copybreak
= adapter
->rx_copybreak
;
194 rxr
->sgl_size
= adapter
->max_rx_sgl_size
;
195 rxr
->smoothed_interval
=
196 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev
);
197 rxr
->empty_rx_queue
= 0;
201 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
202 * @adapter: network interface device structure
205 * Return 0 on success, negative on failure
207 static int ena_setup_tx_resources(struct ena_adapter
*adapter
, int qid
)
209 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
210 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
213 if (tx_ring
->tx_buffer_info
) {
214 netif_err(adapter
, ifup
,
215 adapter
->netdev
, "tx_buffer_info info is not NULL");
219 size
= sizeof(struct ena_tx_buffer
) * tx_ring
->ring_size
;
220 node
= cpu_to_node(ena_irq
->cpu
);
222 tx_ring
->tx_buffer_info
= vzalloc_node(size
, node
);
223 if (!tx_ring
->tx_buffer_info
) {
224 tx_ring
->tx_buffer_info
= vzalloc(size
);
225 if (!tx_ring
->tx_buffer_info
)
229 size
= sizeof(u16
) * tx_ring
->ring_size
;
230 tx_ring
->free_tx_ids
= vzalloc_node(size
, node
);
231 if (!tx_ring
->free_tx_ids
) {
232 tx_ring
->free_tx_ids
= vzalloc(size
);
233 if (!tx_ring
->free_tx_ids
) {
234 vfree(tx_ring
->tx_buffer_info
);
239 /* Req id ring for TX out of order completions */
240 for (i
= 0; i
< tx_ring
->ring_size
; i
++)
241 tx_ring
->free_tx_ids
[i
] = i
;
243 /* Reset tx statistics */
244 memset(&tx_ring
->tx_stats
, 0x0, sizeof(tx_ring
->tx_stats
));
246 tx_ring
->next_to_use
= 0;
247 tx_ring
->next_to_clean
= 0;
248 tx_ring
->cpu
= ena_irq
->cpu
;
252 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
253 * @adapter: network interface device structure
256 * Free all transmit software resources
258 static void ena_free_tx_resources(struct ena_adapter
*adapter
, int qid
)
260 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
262 vfree(tx_ring
->tx_buffer_info
);
263 tx_ring
->tx_buffer_info
= NULL
;
265 vfree(tx_ring
->free_tx_ids
);
266 tx_ring
->free_tx_ids
= NULL
;
269 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
270 * @adapter: private structure
272 * Return 0 on success, negative on failure
274 static int ena_setup_all_tx_resources(struct ena_adapter
*adapter
)
278 for (i
= 0; i
< adapter
->num_queues
; i
++) {
279 rc
= ena_setup_tx_resources(adapter
, i
);
288 netif_err(adapter
, ifup
, adapter
->netdev
,
289 "Tx queue %d: allocation failed\n", i
);
291 /* rewind the index freeing the rings as we go */
293 ena_free_tx_resources(adapter
, i
);
297 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
298 * @adapter: board private structure
300 * Free all transmit software resources
302 static void ena_free_all_io_tx_resources(struct ena_adapter
*adapter
)
306 for (i
= 0; i
< adapter
->num_queues
; i
++)
307 ena_free_tx_resources(adapter
, i
);
310 static inline int validate_rx_req_id(struct ena_ring
*rx_ring
, u16 req_id
)
312 if (likely(req_id
< rx_ring
->ring_size
))
315 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
316 "Invalid rx req_id: %hu\n", req_id
);
318 u64_stats_update_begin(&rx_ring
->syncp
);
319 rx_ring
->rx_stats
.bad_req_id
++;
320 u64_stats_update_end(&rx_ring
->syncp
);
322 /* Trigger device reset */
323 rx_ring
->adapter
->reset_reason
= ENA_REGS_RESET_INV_RX_REQ_ID
;
324 set_bit(ENA_FLAG_TRIGGER_RESET
, &rx_ring
->adapter
->flags
);
328 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
329 * @adapter: network interface device structure
332 * Returns 0 on success, negative on failure
334 static int ena_setup_rx_resources(struct ena_adapter
*adapter
,
337 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
338 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
341 if (rx_ring
->rx_buffer_info
) {
342 netif_err(adapter
, ifup
, adapter
->netdev
,
343 "rx_buffer_info is not NULL");
347 /* alloc extra element so in rx path
348 * we can always prefetch rx_info + 1
350 size
= sizeof(struct ena_rx_buffer
) * (rx_ring
->ring_size
+ 1);
351 node
= cpu_to_node(ena_irq
->cpu
);
353 rx_ring
->rx_buffer_info
= vzalloc_node(size
, node
);
354 if (!rx_ring
->rx_buffer_info
) {
355 rx_ring
->rx_buffer_info
= vzalloc(size
);
356 if (!rx_ring
->rx_buffer_info
)
360 size
= sizeof(u16
) * rx_ring
->ring_size
;
361 rx_ring
->free_rx_ids
= vzalloc_node(size
, node
);
362 if (!rx_ring
->free_rx_ids
) {
363 rx_ring
->free_rx_ids
= vzalloc(size
);
364 if (!rx_ring
->free_rx_ids
) {
365 vfree(rx_ring
->rx_buffer_info
);
370 /* Req id ring for receiving RX pkts out of order */
371 for (i
= 0; i
< rx_ring
->ring_size
; i
++)
372 rx_ring
->free_rx_ids
[i
] = i
;
374 /* Reset rx statistics */
375 memset(&rx_ring
->rx_stats
, 0x0, sizeof(rx_ring
->rx_stats
));
377 rx_ring
->next_to_clean
= 0;
378 rx_ring
->next_to_use
= 0;
379 rx_ring
->cpu
= ena_irq
->cpu
;
384 /* ena_free_rx_resources - Free I/O Rx Resources
385 * @adapter: network interface device structure
388 * Free all receive software resources
390 static void ena_free_rx_resources(struct ena_adapter
*adapter
,
393 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
395 vfree(rx_ring
->rx_buffer_info
);
396 rx_ring
->rx_buffer_info
= NULL
;
398 vfree(rx_ring
->free_rx_ids
);
399 rx_ring
->free_rx_ids
= NULL
;
402 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
403 * @adapter: board private structure
405 * Return 0 on success, negative on failure
407 static int ena_setup_all_rx_resources(struct ena_adapter
*adapter
)
411 for (i
= 0; i
< adapter
->num_queues
; i
++) {
412 rc
= ena_setup_rx_resources(adapter
, i
);
421 netif_err(adapter
, ifup
, adapter
->netdev
,
422 "Rx queue %d: allocation failed\n", i
);
424 /* rewind the index freeing the rings as we go */
426 ena_free_rx_resources(adapter
, i
);
430 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
431 * @adapter: board private structure
433 * Free all receive software resources
435 static void ena_free_all_io_rx_resources(struct ena_adapter
*adapter
)
439 for (i
= 0; i
< adapter
->num_queues
; i
++)
440 ena_free_rx_resources(adapter
, i
);
443 static inline int ena_alloc_rx_page(struct ena_ring
*rx_ring
,
444 struct ena_rx_buffer
*rx_info
, gfp_t gfp
)
446 struct ena_com_buf
*ena_buf
;
450 /* if previous allocated page is not used */
451 if (unlikely(rx_info
->page
))
454 page
= alloc_page(gfp
);
455 if (unlikely(!page
)) {
456 u64_stats_update_begin(&rx_ring
->syncp
);
457 rx_ring
->rx_stats
.page_alloc_fail
++;
458 u64_stats_update_end(&rx_ring
->syncp
);
462 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
,
464 if (unlikely(dma_mapping_error(rx_ring
->dev
, dma
))) {
465 u64_stats_update_begin(&rx_ring
->syncp
);
466 rx_ring
->rx_stats
.dma_mapping_err
++;
467 u64_stats_update_end(&rx_ring
->syncp
);
472 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
473 "alloc page %p, rx_info %p\n", page
, rx_info
);
475 rx_info
->page
= page
;
476 rx_info
->page_offset
= 0;
477 ena_buf
= &rx_info
->ena_buf
;
478 ena_buf
->paddr
= dma
;
479 ena_buf
->len
= PAGE_SIZE
;
484 static void ena_free_rx_page(struct ena_ring
*rx_ring
,
485 struct ena_rx_buffer
*rx_info
)
487 struct page
*page
= rx_info
->page
;
488 struct ena_com_buf
*ena_buf
= &rx_info
->ena_buf
;
490 if (unlikely(!page
)) {
491 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
492 "Trying to free unallocated buffer\n");
496 dma_unmap_page(rx_ring
->dev
, ena_buf
->paddr
, PAGE_SIZE
,
500 rx_info
->page
= NULL
;
503 static int ena_refill_rx_bufs(struct ena_ring
*rx_ring
, u32 num
)
505 u16 next_to_use
, req_id
;
509 next_to_use
= rx_ring
->next_to_use
;
511 for (i
= 0; i
< num
; i
++) {
512 struct ena_rx_buffer
*rx_info
;
514 req_id
= rx_ring
->free_rx_ids
[next_to_use
];
515 rc
= validate_rx_req_id(rx_ring
, req_id
);
516 if (unlikely(rc
< 0))
519 rx_info
= &rx_ring
->rx_buffer_info
[req_id
];
522 rc
= ena_alloc_rx_page(rx_ring
, rx_info
,
523 GFP_ATOMIC
| __GFP_COMP
);
524 if (unlikely(rc
< 0)) {
525 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
526 "failed to alloc buffer for rx queue %d\n",
530 rc
= ena_com_add_single_rx_desc(rx_ring
->ena_com_io_sq
,
534 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
535 "failed to add buffer for rx queue %d\n",
539 next_to_use
= ENA_RX_RING_IDX_NEXT(next_to_use
,
543 if (unlikely(i
< num
)) {
544 u64_stats_update_begin(&rx_ring
->syncp
);
545 rx_ring
->rx_stats
.refil_partial
++;
546 u64_stats_update_end(&rx_ring
->syncp
);
547 netdev_warn(rx_ring
->netdev
,
548 "refilled rx qid %d with only %d buffers (from %d)\n",
549 rx_ring
->qid
, i
, num
);
553 /* Add memory barrier to make sure the desc were written before
557 ena_com_write_sq_doorbell(rx_ring
->ena_com_io_sq
);
560 rx_ring
->next_to_use
= next_to_use
;
565 static void ena_free_rx_bufs(struct ena_adapter
*adapter
,
568 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
571 for (i
= 0; i
< rx_ring
->ring_size
; i
++) {
572 struct ena_rx_buffer
*rx_info
= &rx_ring
->rx_buffer_info
[i
];
575 ena_free_rx_page(rx_ring
, rx_info
);
579 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
580 * @adapter: board private structure
583 static void ena_refill_all_rx_bufs(struct ena_adapter
*adapter
)
585 struct ena_ring
*rx_ring
;
588 for (i
= 0; i
< adapter
->num_queues
; i
++) {
589 rx_ring
= &adapter
->rx_ring
[i
];
590 bufs_num
= rx_ring
->ring_size
- 1;
591 rc
= ena_refill_rx_bufs(rx_ring
, bufs_num
);
593 if (unlikely(rc
!= bufs_num
))
594 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
595 "refilling Queue %d failed. allocated %d buffers from: %d\n",
600 static void ena_free_all_rx_bufs(struct ena_adapter
*adapter
)
604 for (i
= 0; i
< adapter
->num_queues
; i
++)
605 ena_free_rx_bufs(adapter
, i
);
608 /* ena_free_tx_bufs - Free Tx Buffers per Queue
609 * @tx_ring: TX ring for which buffers be freed
611 static void ena_free_tx_bufs(struct ena_ring
*tx_ring
)
613 bool print_once
= true;
616 for (i
= 0; i
< tx_ring
->ring_size
; i
++) {
617 struct ena_tx_buffer
*tx_info
= &tx_ring
->tx_buffer_info
[i
];
618 struct ena_com_buf
*ena_buf
;
626 netdev_notice(tx_ring
->netdev
,
627 "free uncompleted tx skb qid %d idx 0x%x\n",
631 netdev_dbg(tx_ring
->netdev
,
632 "free uncompleted tx skb qid %d idx 0x%x\n",
636 ena_buf
= tx_info
->bufs
;
637 dma_unmap_single(tx_ring
->dev
,
642 /* unmap remaining mapped pages */
643 nr_frags
= tx_info
->num_of_bufs
- 1;
644 for (j
= 0; j
< nr_frags
; j
++) {
646 dma_unmap_page(tx_ring
->dev
,
652 dev_kfree_skb_any(tx_info
->skb
);
654 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring
->netdev
,
658 static void ena_free_all_tx_bufs(struct ena_adapter
*adapter
)
660 struct ena_ring
*tx_ring
;
663 for (i
= 0; i
< adapter
->num_queues
; i
++) {
664 tx_ring
= &adapter
->tx_ring
[i
];
665 ena_free_tx_bufs(tx_ring
);
669 static void ena_destroy_all_tx_queues(struct ena_adapter
*adapter
)
674 for (i
= 0; i
< adapter
->num_queues
; i
++) {
675 ena_qid
= ENA_IO_TXQ_IDX(i
);
676 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
680 static void ena_destroy_all_rx_queues(struct ena_adapter
*adapter
)
685 for (i
= 0; i
< adapter
->num_queues
; i
++) {
686 ena_qid
= ENA_IO_RXQ_IDX(i
);
687 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
691 static void ena_destroy_all_io_queues(struct ena_adapter
*adapter
)
693 ena_destroy_all_tx_queues(adapter
);
694 ena_destroy_all_rx_queues(adapter
);
697 static int validate_tx_req_id(struct ena_ring
*tx_ring
, u16 req_id
)
699 struct ena_tx_buffer
*tx_info
= NULL
;
701 if (likely(req_id
< tx_ring
->ring_size
)) {
702 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
703 if (likely(tx_info
->skb
))
708 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
709 "tx_info doesn't have valid skb\n");
711 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
712 "Invalid req_id: %hu\n", req_id
);
714 u64_stats_update_begin(&tx_ring
->syncp
);
715 tx_ring
->tx_stats
.bad_req_id
++;
716 u64_stats_update_end(&tx_ring
->syncp
);
718 /* Trigger device reset */
719 tx_ring
->adapter
->reset_reason
= ENA_REGS_RESET_INV_TX_REQ_ID
;
720 set_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
);
724 static int ena_clean_tx_irq(struct ena_ring
*tx_ring
, u32 budget
)
726 struct netdev_queue
*txq
;
735 next_to_clean
= tx_ring
->next_to_clean
;
736 txq
= netdev_get_tx_queue(tx_ring
->netdev
, tx_ring
->qid
);
738 while (tx_pkts
< budget
) {
739 struct ena_tx_buffer
*tx_info
;
741 struct ena_com_buf
*ena_buf
;
744 rc
= ena_com_tx_comp_req_id_get(tx_ring
->ena_com_io_cq
,
749 rc
= validate_tx_req_id(tx_ring
, req_id
);
753 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
756 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
760 tx_info
->last_jiffies
= 0;
762 if (likely(tx_info
->num_of_bufs
!= 0)) {
763 ena_buf
= tx_info
->bufs
;
765 dma_unmap_single(tx_ring
->dev
,
766 dma_unmap_addr(ena_buf
, paddr
),
767 dma_unmap_len(ena_buf
, len
),
770 /* unmap remaining mapped pages */
771 nr_frags
= tx_info
->num_of_bufs
- 1;
772 for (i
= 0; i
< nr_frags
; i
++) {
774 dma_unmap_page(tx_ring
->dev
,
775 dma_unmap_addr(ena_buf
, paddr
),
776 dma_unmap_len(ena_buf
, len
),
781 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
782 "tx_poll: q %d skb %p completed\n", tx_ring
->qid
,
785 tx_bytes
+= skb
->len
;
788 total_done
+= tx_info
->tx_descs
;
790 tx_ring
->free_tx_ids
[next_to_clean
] = req_id
;
791 next_to_clean
= ENA_TX_RING_IDX_NEXT(next_to_clean
,
795 tx_ring
->next_to_clean
= next_to_clean
;
796 ena_com_comp_ack(tx_ring
->ena_com_io_sq
, total_done
);
797 ena_com_update_dev_comp_head(tx_ring
->ena_com_io_cq
);
799 netdev_tx_completed_queue(txq
, tx_pkts
, tx_bytes
);
801 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
802 "tx_poll: q %d done. total pkts: %d\n",
803 tx_ring
->qid
, tx_pkts
);
805 /* need to make the rings circular update visible to
806 * ena_start_xmit() before checking for netif_queue_stopped().
810 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
811 ENA_TX_WAKEUP_THRESH
;
812 if (unlikely(netif_tx_queue_stopped(txq
) && above_thresh
)) {
813 __netif_tx_lock(txq
, smp_processor_id());
814 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
815 ENA_TX_WAKEUP_THRESH
;
816 if (netif_tx_queue_stopped(txq
) && above_thresh
) {
817 netif_tx_wake_queue(txq
);
818 u64_stats_update_begin(&tx_ring
->syncp
);
819 tx_ring
->tx_stats
.queue_wakeup
++;
820 u64_stats_update_end(&tx_ring
->syncp
);
822 __netif_tx_unlock(txq
);
825 tx_ring
->per_napi_bytes
+= tx_bytes
;
826 tx_ring
->per_napi_packets
+= tx_pkts
;
831 static struct sk_buff
*ena_alloc_skb(struct ena_ring
*rx_ring
, bool frags
)
836 skb
= napi_get_frags(rx_ring
->napi
);
838 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
839 rx_ring
->rx_copybreak
);
841 if (unlikely(!skb
)) {
842 u64_stats_update_begin(&rx_ring
->syncp
);
843 rx_ring
->rx_stats
.skb_alloc_fail
++;
844 u64_stats_update_end(&rx_ring
->syncp
);
845 netif_dbg(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
846 "Failed to allocate skb. frags: %d\n", frags
);
853 static struct sk_buff
*ena_rx_skb(struct ena_ring
*rx_ring
,
854 struct ena_com_rx_buf_info
*ena_bufs
,
859 struct ena_rx_buffer
*rx_info
;
860 u16 len
, req_id
, buf
= 0;
863 len
= ena_bufs
[buf
].len
;
864 req_id
= ena_bufs
[buf
].req_id
;
865 rx_info
= &rx_ring
->rx_buffer_info
[req_id
];
867 if (unlikely(!rx_info
->page
)) {
868 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
873 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
874 "rx_info %p page %p\n",
875 rx_info
, rx_info
->page
);
877 /* save virt address of first buffer */
878 va
= page_address(rx_info
->page
) + rx_info
->page_offset
;
879 prefetch(va
+ NET_IP_ALIGN
);
881 if (len
<= rx_ring
->rx_copybreak
) {
882 skb
= ena_alloc_skb(rx_ring
, false);
886 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
887 "rx allocated small packet. len %d. data_len %d\n",
888 skb
->len
, skb
->data_len
);
890 /* sync this buffer for CPU use */
891 dma_sync_single_for_cpu(rx_ring
->dev
,
892 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
895 skb_copy_to_linear_data(skb
, va
, len
);
896 dma_sync_single_for_device(rx_ring
->dev
,
897 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
902 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
903 rx_ring
->free_rx_ids
[*next_to_clean
] = req_id
;
904 *next_to_clean
= ENA_RX_RING_IDX_ADD(*next_to_clean
, descs
,
909 skb
= ena_alloc_skb(rx_ring
, true);
914 dma_unmap_page(rx_ring
->dev
,
915 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
916 PAGE_SIZE
, DMA_FROM_DEVICE
);
918 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, rx_info
->page
,
919 rx_info
->page_offset
, len
, PAGE_SIZE
);
921 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
922 "rx skb updated. len %d. data_len %d\n",
923 skb
->len
, skb
->data_len
);
925 rx_info
->page
= NULL
;
927 rx_ring
->free_rx_ids
[*next_to_clean
] = req_id
;
929 ENA_RX_RING_IDX_NEXT(*next_to_clean
,
931 if (likely(--descs
== 0))
935 len
= ena_bufs
[buf
].len
;
936 req_id
= ena_bufs
[buf
].req_id
;
937 rx_info
= &rx_ring
->rx_buffer_info
[req_id
];
943 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
944 * @adapter: structure containing adapter specific data
945 * @ena_rx_ctx: received packet context/metadata
946 * @skb: skb currently being received and modified
948 static inline void ena_rx_checksum(struct ena_ring
*rx_ring
,
949 struct ena_com_rx_ctx
*ena_rx_ctx
,
952 /* Rx csum disabled */
953 if (unlikely(!(rx_ring
->netdev
->features
& NETIF_F_RXCSUM
))) {
954 skb
->ip_summed
= CHECKSUM_NONE
;
958 /* For fragmented packets the checksum isn't valid */
959 if (ena_rx_ctx
->frag
) {
960 skb
->ip_summed
= CHECKSUM_NONE
;
964 /* if IP and error */
965 if (unlikely((ena_rx_ctx
->l3_proto
== ENA_ETH_IO_L3_PROTO_IPV4
) &&
966 (ena_rx_ctx
->l3_csum_err
))) {
967 /* ipv4 checksum error */
968 skb
->ip_summed
= CHECKSUM_NONE
;
969 u64_stats_update_begin(&rx_ring
->syncp
);
970 rx_ring
->rx_stats
.bad_csum
++;
971 u64_stats_update_end(&rx_ring
->syncp
);
972 netif_dbg(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
973 "RX IPv4 header checksum error\n");
978 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
979 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
))) {
980 if (unlikely(ena_rx_ctx
->l4_csum_err
)) {
981 /* TCP/UDP checksum error */
982 u64_stats_update_begin(&rx_ring
->syncp
);
983 rx_ring
->rx_stats
.bad_csum
++;
984 u64_stats_update_end(&rx_ring
->syncp
);
985 netif_dbg(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
986 "RX L4 checksum error\n");
987 skb
->ip_summed
= CHECKSUM_NONE
;
991 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
995 static void ena_set_rx_hash(struct ena_ring
*rx_ring
,
996 struct ena_com_rx_ctx
*ena_rx_ctx
,
999 enum pkt_hash_types hash_type
;
1001 if (likely(rx_ring
->netdev
->features
& NETIF_F_RXHASH
)) {
1002 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
1003 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
)))
1005 hash_type
= PKT_HASH_TYPE_L4
;
1007 hash_type
= PKT_HASH_TYPE_NONE
;
1009 /* Override hash type if the packet is fragmented */
1010 if (ena_rx_ctx
->frag
)
1011 hash_type
= PKT_HASH_TYPE_NONE
;
1013 skb_set_hash(skb
, ena_rx_ctx
->hash
, hash_type
);
1017 /* ena_clean_rx_irq - Cleanup RX irq
1018 * @rx_ring: RX ring to clean
1019 * @napi: napi handler
1020 * @budget: how many packets driver is allowed to clean
1022 * Returns the number of cleaned buffers.
1024 static int ena_clean_rx_irq(struct ena_ring
*rx_ring
, struct napi_struct
*napi
,
1027 u16 next_to_clean
= rx_ring
->next_to_clean
;
1028 u32 res_budget
, work_done
;
1030 struct ena_com_rx_ctx ena_rx_ctx
;
1031 struct ena_adapter
*adapter
;
1032 struct sk_buff
*skb
;
1033 int refill_required
;
1034 int refill_threshold
;
1037 int rx_copybreak_pkt
= 0;
1040 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
1041 "%s qid %d\n", __func__
, rx_ring
->qid
);
1042 res_budget
= budget
;
1045 ena_rx_ctx
.ena_bufs
= rx_ring
->ena_bufs
;
1046 ena_rx_ctx
.max_bufs
= rx_ring
->sgl_size
;
1047 ena_rx_ctx
.descs
= 0;
1048 rc
= ena_com_rx_pkt(rx_ring
->ena_com_io_cq
,
1049 rx_ring
->ena_com_io_sq
,
1054 if (unlikely(ena_rx_ctx
.descs
== 0))
1057 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
1058 "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1059 rx_ring
->qid
, ena_rx_ctx
.descs
, ena_rx_ctx
.l3_proto
,
1060 ena_rx_ctx
.l4_proto
, ena_rx_ctx
.hash
);
1062 /* allocate skb and fill it */
1063 skb
= ena_rx_skb(rx_ring
, rx_ring
->ena_bufs
, ena_rx_ctx
.descs
,
1066 /* exit if we failed to retrieve a buffer */
1067 if (unlikely(!skb
)) {
1068 for (i
= 0; i
< ena_rx_ctx
.descs
; i
++) {
1069 rx_ring
->free_tx_ids
[next_to_clean
] =
1070 rx_ring
->ena_bufs
[i
].req_id
;
1072 ENA_RX_RING_IDX_NEXT(next_to_clean
,
1073 rx_ring
->ring_size
);
1078 ena_rx_checksum(rx_ring
, &ena_rx_ctx
, skb
);
1080 ena_set_rx_hash(rx_ring
, &ena_rx_ctx
, skb
);
1082 skb_record_rx_queue(skb
, rx_ring
->qid
);
1084 if (rx_ring
->ena_bufs
[0].len
<= rx_ring
->rx_copybreak
) {
1085 total_len
+= rx_ring
->ena_bufs
[0].len
;
1087 napi_gro_receive(napi
, skb
);
1089 total_len
+= skb
->len
;
1090 napi_gro_frags(napi
);
1094 } while (likely(res_budget
));
1096 work_done
= budget
- res_budget
;
1097 rx_ring
->per_napi_bytes
+= total_len
;
1098 rx_ring
->per_napi_packets
+= work_done
;
1099 u64_stats_update_begin(&rx_ring
->syncp
);
1100 rx_ring
->rx_stats
.bytes
+= total_len
;
1101 rx_ring
->rx_stats
.cnt
+= work_done
;
1102 rx_ring
->rx_stats
.rx_copybreak_pkt
+= rx_copybreak_pkt
;
1103 u64_stats_update_end(&rx_ring
->syncp
);
1105 rx_ring
->next_to_clean
= next_to_clean
;
1107 refill_required
= ena_com_sq_empty_space(rx_ring
->ena_com_io_sq
);
1108 refill_threshold
= rx_ring
->ring_size
/ ENA_RX_REFILL_THRESH_DIVIDER
;
1110 /* Optimization, try to batch new rx buffers */
1111 if (refill_required
> refill_threshold
) {
1112 ena_com_update_dev_comp_head(rx_ring
->ena_com_io_cq
);
1113 ena_refill_rx_bufs(rx_ring
, refill_required
);
1119 adapter
= netdev_priv(rx_ring
->netdev
);
1121 u64_stats_update_begin(&rx_ring
->syncp
);
1122 rx_ring
->rx_stats
.bad_desc_num
++;
1123 u64_stats_update_end(&rx_ring
->syncp
);
1125 /* Too many desc from the device. Trigger reset */
1126 adapter
->reset_reason
= ENA_REGS_RESET_TOO_MANY_RX_DESCS
;
1127 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
1132 inline void ena_adjust_intr_moderation(struct ena_ring
*rx_ring
,
1133 struct ena_ring
*tx_ring
)
1135 /* We apply adaptive moderation on Rx path only.
1136 * Tx uses static interrupt moderation.
1138 ena_com_calculate_interrupt_delay(rx_ring
->ena_dev
,
1139 rx_ring
->per_napi_packets
,
1140 rx_ring
->per_napi_bytes
,
1141 &rx_ring
->smoothed_interval
,
1142 &rx_ring
->moder_tbl_idx
);
1144 /* Reset per napi packets/bytes */
1145 tx_ring
->per_napi_packets
= 0;
1146 tx_ring
->per_napi_bytes
= 0;
1147 rx_ring
->per_napi_packets
= 0;
1148 rx_ring
->per_napi_bytes
= 0;
1151 static inline void ena_unmask_interrupt(struct ena_ring
*tx_ring
,
1152 struct ena_ring
*rx_ring
)
1154 struct ena_eth_io_intr_reg intr_reg
;
1156 /* Update intr register: rx intr delay,
1157 * tx intr delay and interrupt unmask
1159 ena_com_update_intr_reg(&intr_reg
,
1160 rx_ring
->smoothed_interval
,
1161 tx_ring
->smoothed_interval
,
1164 /* It is a shared MSI-X.
1165 * Tx and Rx CQ have pointer to it.
1166 * So we use one of them to reach the intr reg
1168 ena_com_unmask_intr(rx_ring
->ena_com_io_cq
, &intr_reg
);
1171 static inline void ena_update_ring_numa_node(struct ena_ring
*tx_ring
,
1172 struct ena_ring
*rx_ring
)
1174 int cpu
= get_cpu();
1177 /* Check only one ring since the 2 rings are running on the same cpu */
1178 if (likely(tx_ring
->cpu
== cpu
))
1181 numa_node
= cpu_to_node(cpu
);
1184 if (numa_node
!= NUMA_NO_NODE
) {
1185 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, numa_node
);
1186 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, numa_node
);
1197 static int ena_io_poll(struct napi_struct
*napi
, int budget
)
1199 struct ena_napi
*ena_napi
= container_of(napi
, struct ena_napi
, napi
);
1200 struct ena_ring
*tx_ring
, *rx_ring
;
1205 int napi_comp_call
= 0;
1208 tx_ring
= ena_napi
->tx_ring
;
1209 rx_ring
= ena_napi
->rx_ring
;
1211 tx_budget
= tx_ring
->ring_size
/ ENA_TX_POLL_BUDGET_DIVIDER
;
1213 if (!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
) ||
1214 test_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
)) {
1215 napi_complete_done(napi
, 0);
1219 tx_work_done
= ena_clean_tx_irq(tx_ring
, tx_budget
);
1220 rx_work_done
= ena_clean_rx_irq(rx_ring
, napi
, budget
);
1222 /* If the device is about to reset or down, avoid unmask
1223 * the interrupt and return 0 so NAPI won't reschedule
1225 if (unlikely(!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
) ||
1226 test_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
))) {
1227 napi_complete_done(napi
, 0);
1230 } else if ((budget
> rx_work_done
) && (tx_budget
> tx_work_done
)) {
1233 /* Update numa and unmask the interrupt only when schedule
1234 * from the interrupt context (vs from sk_busy_loop)
1236 if (napi_complete_done(napi
, rx_work_done
)) {
1237 /* Tx and Rx share the same interrupt vector */
1238 if (ena_com_get_adaptive_moderation_enabled(rx_ring
->ena_dev
))
1239 ena_adjust_intr_moderation(rx_ring
, tx_ring
);
1241 ena_unmask_interrupt(tx_ring
, rx_ring
);
1244 ena_update_ring_numa_node(tx_ring
, rx_ring
);
1251 u64_stats_update_begin(&tx_ring
->syncp
);
1252 tx_ring
->tx_stats
.napi_comp
+= napi_comp_call
;
1253 tx_ring
->tx_stats
.tx_poll
++;
1254 u64_stats_update_end(&tx_ring
->syncp
);
1259 static irqreturn_t
ena_intr_msix_mgmnt(int irq
, void *data
)
1261 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
1263 ena_com_admin_q_comp_intr_handler(adapter
->ena_dev
);
1265 /* Don't call the aenq handler before probe is done */
1266 if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
)))
1267 ena_com_aenq_intr_handler(adapter
->ena_dev
, data
);
1272 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1273 * @irq: interrupt number
1274 * @data: pointer to a network interface private napi device structure
1276 static irqreturn_t
ena_intr_msix_io(int irq
, void *data
)
1278 struct ena_napi
*ena_napi
= data
;
1280 napi_schedule_irqoff(&ena_napi
->napi
);
1285 /* Reserve a single MSI-X vector for management (admin + aenq).
1286 * plus reserve one vector for each potential io queue.
1287 * the number of potential io queues is the minimum of what the device
1288 * supports and the number of vCPUs.
1290 static int ena_enable_msix(struct ena_adapter
*adapter
, int num_queues
)
1292 int msix_vecs
, irq_cnt
;
1294 if (test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1295 netif_err(adapter
, probe
, adapter
->netdev
,
1296 "Error, MSI-X is already enabled\n");
1300 /* Reserved the max msix vectors we might need */
1301 msix_vecs
= ENA_MAX_MSIX_VEC(num_queues
);
1303 netif_dbg(adapter
, probe
, adapter
->netdev
,
1304 "trying to enable MSI-X, vectors %d\n", msix_vecs
);
1306 irq_cnt
= pci_alloc_irq_vectors(adapter
->pdev
, ENA_MIN_MSIX_VEC
,
1307 msix_vecs
, PCI_IRQ_MSIX
);
1310 netif_err(adapter
, probe
, adapter
->netdev
,
1311 "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt
);
1315 if (irq_cnt
!= msix_vecs
) {
1316 netif_notice(adapter
, probe
, adapter
->netdev
,
1317 "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1318 irq_cnt
, msix_vecs
);
1319 adapter
->num_queues
= irq_cnt
- ENA_ADMIN_MSIX_VEC
;
1322 if (ena_init_rx_cpu_rmap(adapter
))
1323 netif_warn(adapter
, probe
, adapter
->netdev
,
1324 "Failed to map IRQs to CPUs\n");
1326 adapter
->msix_vecs
= irq_cnt
;
1327 set_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
);
1332 static void ena_setup_mgmnt_intr(struct ena_adapter
*adapter
)
1336 snprintf(adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].name
,
1337 ENA_IRQNAME_SIZE
, "ena-mgmnt@pci:%s",
1338 pci_name(adapter
->pdev
));
1339 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].handler
=
1340 ena_intr_msix_mgmnt
;
1341 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].data
= adapter
;
1342 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].vector
=
1343 pci_irq_vector(adapter
->pdev
, ENA_MGMNT_IRQ_IDX
);
1344 cpu
= cpumask_first(cpu_online_mask
);
1345 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].cpu
= cpu
;
1346 cpumask_set_cpu(cpu
,
1347 &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].affinity_hint_mask
);
1350 static void ena_setup_io_intr(struct ena_adapter
*adapter
)
1352 struct net_device
*netdev
;
1353 int irq_idx
, i
, cpu
;
1355 netdev
= adapter
->netdev
;
1357 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1358 irq_idx
= ENA_IO_IRQ_IDX(i
);
1359 cpu
= i
% num_online_cpus();
1361 snprintf(adapter
->irq_tbl
[irq_idx
].name
, ENA_IRQNAME_SIZE
,
1362 "%s-Tx-Rx-%d", netdev
->name
, i
);
1363 adapter
->irq_tbl
[irq_idx
].handler
= ena_intr_msix_io
;
1364 adapter
->irq_tbl
[irq_idx
].data
= &adapter
->ena_napi
[i
];
1365 adapter
->irq_tbl
[irq_idx
].vector
=
1366 pci_irq_vector(adapter
->pdev
, irq_idx
);
1367 adapter
->irq_tbl
[irq_idx
].cpu
= cpu
;
1369 cpumask_set_cpu(cpu
,
1370 &adapter
->irq_tbl
[irq_idx
].affinity_hint_mask
);
1374 static int ena_request_mgmnt_irq(struct ena_adapter
*adapter
)
1376 unsigned long flags
= 0;
1377 struct ena_irq
*irq
;
1380 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1381 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1384 netif_err(adapter
, probe
, adapter
->netdev
,
1385 "failed to request admin irq\n");
1389 netif_dbg(adapter
, probe
, adapter
->netdev
,
1390 "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1391 irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1393 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1398 static int ena_request_io_irq(struct ena_adapter
*adapter
)
1400 unsigned long flags
= 0;
1401 struct ena_irq
*irq
;
1404 if (!test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1405 netif_err(adapter
, ifup
, adapter
->netdev
,
1406 "Failed to request I/O IRQ: MSI-X is not enabled\n");
1410 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1411 irq
= &adapter
->irq_tbl
[i
];
1412 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1415 netif_err(adapter
, ifup
, adapter
->netdev
,
1416 "Failed to request I/O IRQ. index %d rc %d\n",
1421 netif_dbg(adapter
, ifup
, adapter
->netdev
,
1422 "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1423 i
, irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1425 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1431 for (k
= ENA_IO_IRQ_FIRST_IDX
; k
< i
; k
++) {
1432 irq
= &adapter
->irq_tbl
[k
];
1433 free_irq(irq
->vector
, irq
->data
);
1439 static void ena_free_mgmnt_irq(struct ena_adapter
*adapter
)
1441 struct ena_irq
*irq
;
1443 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1444 synchronize_irq(irq
->vector
);
1445 irq_set_affinity_hint(irq
->vector
, NULL
);
1446 free_irq(irq
->vector
, irq
->data
);
1449 static void ena_free_io_irq(struct ena_adapter
*adapter
)
1451 struct ena_irq
*irq
;
1454 #ifdef CONFIG_RFS_ACCEL
1455 if (adapter
->msix_vecs
>= 1) {
1456 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
1457 adapter
->netdev
->rx_cpu_rmap
= NULL
;
1459 #endif /* CONFIG_RFS_ACCEL */
1461 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1462 irq
= &adapter
->irq_tbl
[i
];
1463 irq_set_affinity_hint(irq
->vector
, NULL
);
1464 free_irq(irq
->vector
, irq
->data
);
1468 static void ena_disable_msix(struct ena_adapter
*adapter
)
1470 if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
))
1471 pci_free_irq_vectors(adapter
->pdev
);
1474 static void ena_disable_io_intr_sync(struct ena_adapter
*adapter
)
1478 if (!netif_running(adapter
->netdev
))
1481 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++)
1482 synchronize_irq(adapter
->irq_tbl
[i
].vector
);
1485 static void ena_del_napi(struct ena_adapter
*adapter
)
1489 for (i
= 0; i
< adapter
->num_queues
; i
++)
1490 netif_napi_del(&adapter
->ena_napi
[i
].napi
);
1493 static void ena_init_napi(struct ena_adapter
*adapter
)
1495 struct ena_napi
*napi
;
1498 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1499 napi
= &adapter
->ena_napi
[i
];
1501 netif_napi_add(adapter
->netdev
,
1502 &adapter
->ena_napi
[i
].napi
,
1505 napi
->rx_ring
= &adapter
->rx_ring
[i
];
1506 napi
->tx_ring
= &adapter
->tx_ring
[i
];
1511 static void ena_napi_disable_all(struct ena_adapter
*adapter
)
1515 for (i
= 0; i
< adapter
->num_queues
; i
++)
1516 napi_disable(&adapter
->ena_napi
[i
].napi
);
1519 static void ena_napi_enable_all(struct ena_adapter
*adapter
)
1523 for (i
= 0; i
< adapter
->num_queues
; i
++)
1524 napi_enable(&adapter
->ena_napi
[i
].napi
);
1527 static void ena_restore_ethtool_params(struct ena_adapter
*adapter
)
1529 adapter
->tx_usecs
= 0;
1530 adapter
->rx_usecs
= 0;
1531 adapter
->tx_frames
= 1;
1532 adapter
->rx_frames
= 1;
1535 /* Configure the Rx forwarding */
1536 static int ena_rss_configure(struct ena_adapter
*adapter
)
1538 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1541 /* In case the RSS table wasn't initialized by probe */
1542 if (!ena_dev
->rss
.tbl_log_size
) {
1543 rc
= ena_rss_init_default(adapter
);
1544 if (rc
&& (rc
!= -EOPNOTSUPP
)) {
1545 netif_err(adapter
, ifup
, adapter
->netdev
,
1546 "Failed to init RSS rc: %d\n", rc
);
1551 /* Set indirect table */
1552 rc
= ena_com_indirect_table_set(ena_dev
);
1553 if (unlikely(rc
&& rc
!= -EOPNOTSUPP
))
1556 /* Configure hash function (if supported) */
1557 rc
= ena_com_set_hash_function(ena_dev
);
1558 if (unlikely(rc
&& (rc
!= -EOPNOTSUPP
)))
1561 /* Configure hash inputs (if supported) */
1562 rc
= ena_com_set_hash_ctrl(ena_dev
);
1563 if (unlikely(rc
&& (rc
!= -EOPNOTSUPP
)))
1569 static int ena_up_complete(struct ena_adapter
*adapter
)
1573 rc
= ena_rss_configure(adapter
);
1577 ena_init_napi(adapter
);
1579 ena_change_mtu(adapter
->netdev
, adapter
->netdev
->mtu
);
1581 ena_refill_all_rx_bufs(adapter
);
1583 /* enable transmits */
1584 netif_tx_start_all_queues(adapter
->netdev
);
1586 ena_restore_ethtool_params(adapter
);
1588 ena_napi_enable_all(adapter
);
1593 static int ena_create_io_tx_queue(struct ena_adapter
*adapter
, int qid
)
1595 struct ena_com_create_io_ctx ctx
= { 0 };
1596 struct ena_com_dev
*ena_dev
;
1597 struct ena_ring
*tx_ring
;
1602 ena_dev
= adapter
->ena_dev
;
1604 tx_ring
= &adapter
->tx_ring
[qid
];
1605 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1606 ena_qid
= ENA_IO_TXQ_IDX(qid
);
1608 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_TX
;
1610 ctx
.mem_queue_type
= ena_dev
->tx_mem_queue_type
;
1611 ctx
.msix_vector
= msix_vector
;
1612 ctx
.queue_size
= adapter
->tx_ring_size
;
1613 ctx
.numa_node
= cpu_to_node(tx_ring
->cpu
);
1615 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1617 netif_err(adapter
, ifup
, adapter
->netdev
,
1618 "Failed to create I/O TX queue num %d rc: %d\n",
1623 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1624 &tx_ring
->ena_com_io_sq
,
1625 &tx_ring
->ena_com_io_cq
);
1627 netif_err(adapter
, ifup
, adapter
->netdev
,
1628 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1630 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1634 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1638 static int ena_create_all_io_tx_queues(struct ena_adapter
*adapter
)
1640 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1643 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1644 rc
= ena_create_io_tx_queue(adapter
, i
);
1653 ena_com_destroy_io_queue(ena_dev
, ENA_IO_TXQ_IDX(i
));
1658 static int ena_create_io_rx_queue(struct ena_adapter
*adapter
, int qid
)
1660 struct ena_com_dev
*ena_dev
;
1661 struct ena_com_create_io_ctx ctx
= { 0 };
1662 struct ena_ring
*rx_ring
;
1667 ena_dev
= adapter
->ena_dev
;
1669 rx_ring
= &adapter
->rx_ring
[qid
];
1670 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1671 ena_qid
= ENA_IO_RXQ_IDX(qid
);
1674 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_RX
;
1675 ctx
.mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
1676 ctx
.msix_vector
= msix_vector
;
1677 ctx
.queue_size
= adapter
->rx_ring_size
;
1678 ctx
.numa_node
= cpu_to_node(rx_ring
->cpu
);
1680 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1682 netif_err(adapter
, ifup
, adapter
->netdev
,
1683 "Failed to create I/O RX queue num %d rc: %d\n",
1688 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1689 &rx_ring
->ena_com_io_sq
,
1690 &rx_ring
->ena_com_io_cq
);
1692 netif_err(adapter
, ifup
, adapter
->netdev
,
1693 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1695 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1699 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1704 static int ena_create_all_io_rx_queues(struct ena_adapter
*adapter
)
1706 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1709 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1710 rc
= ena_create_io_rx_queue(adapter
, i
);
1719 ena_com_destroy_io_queue(ena_dev
, ENA_IO_RXQ_IDX(i
));
1724 static int ena_up(struct ena_adapter
*adapter
)
1728 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
1730 ena_setup_io_intr(adapter
);
1732 rc
= ena_request_io_irq(adapter
);
1736 /* allocate transmit descriptors */
1737 rc
= ena_setup_all_tx_resources(adapter
);
1741 /* allocate receive descriptors */
1742 rc
= ena_setup_all_rx_resources(adapter
);
1746 /* Create TX queues */
1747 rc
= ena_create_all_io_tx_queues(adapter
);
1749 goto err_create_tx_queues
;
1751 /* Create RX queues */
1752 rc
= ena_create_all_io_rx_queues(adapter
);
1754 goto err_create_rx_queues
;
1756 rc
= ena_up_complete(adapter
);
1760 if (test_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
))
1761 netif_carrier_on(adapter
->netdev
);
1763 u64_stats_update_begin(&adapter
->syncp
);
1764 adapter
->dev_stats
.interface_up
++;
1765 u64_stats_update_end(&adapter
->syncp
);
1767 set_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1769 /* Enable completion queues interrupt */
1770 for (i
= 0; i
< adapter
->num_queues
; i
++)
1771 ena_unmask_interrupt(&adapter
->tx_ring
[i
],
1772 &adapter
->rx_ring
[i
]);
1774 /* schedule napi in case we had pending packets
1775 * from the last time we disable napi
1777 for (i
= 0; i
< adapter
->num_queues
; i
++)
1778 napi_schedule(&adapter
->ena_napi
[i
].napi
);
1783 ena_destroy_all_rx_queues(adapter
);
1784 err_create_rx_queues
:
1785 ena_destroy_all_tx_queues(adapter
);
1786 err_create_tx_queues
:
1787 ena_free_all_io_rx_resources(adapter
);
1789 ena_free_all_io_tx_resources(adapter
);
1791 ena_free_io_irq(adapter
);
1797 static void ena_down(struct ena_adapter
*adapter
)
1799 netif_info(adapter
, ifdown
, adapter
->netdev
, "%s\n", __func__
);
1801 clear_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1803 u64_stats_update_begin(&adapter
->syncp
);
1804 adapter
->dev_stats
.interface_down
++;
1805 u64_stats_update_end(&adapter
->syncp
);
1807 netif_carrier_off(adapter
->netdev
);
1808 netif_tx_disable(adapter
->netdev
);
1810 /* After this point the napi handler won't enable the tx queue */
1811 ena_napi_disable_all(adapter
);
1813 /* After destroy the queue there won't be any new interrupts */
1815 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
)) {
1818 rc
= ena_com_dev_reset(adapter
->ena_dev
, adapter
->reset_reason
);
1820 dev_err(&adapter
->pdev
->dev
, "Device reset failed\n");
1823 ena_destroy_all_io_queues(adapter
);
1825 ena_disable_io_intr_sync(adapter
);
1826 ena_free_io_irq(adapter
);
1827 ena_del_napi(adapter
);
1829 ena_free_all_tx_bufs(adapter
);
1830 ena_free_all_rx_bufs(adapter
);
1831 ena_free_all_io_tx_resources(adapter
);
1832 ena_free_all_io_rx_resources(adapter
);
1835 /* ena_open - Called when a network interface is made active
1836 * @netdev: network interface device structure
1838 * Returns 0 on success, negative value on failure
1840 * The open entry point is called when a network interface is made
1841 * active by the system (IFF_UP). At this point all resources needed
1842 * for transmit and receive operations are allocated, the interrupt
1843 * handler is registered with the OS, the watchdog timer is started,
1844 * and the stack is notified that the interface is ready.
1846 static int ena_open(struct net_device
*netdev
)
1848 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1851 /* Notify the stack of the actual queue counts. */
1852 rc
= netif_set_real_num_tx_queues(netdev
, adapter
->num_queues
);
1854 netif_err(adapter
, ifup
, netdev
, "Can't set num tx queues\n");
1858 rc
= netif_set_real_num_rx_queues(netdev
, adapter
->num_queues
);
1860 netif_err(adapter
, ifup
, netdev
, "Can't set num rx queues\n");
1864 rc
= ena_up(adapter
);
1871 /* ena_close - Disables a network interface
1872 * @netdev: network interface device structure
1874 * Returns 0, this is not allowed to fail
1876 * The close entry point is called when an interface is de-activated
1877 * by the OS. The hardware is still under the drivers control, but
1878 * needs to be disabled. A global MAC reset is issued to stop the
1879 * hardware, and all transmit and receive resources are freed.
1881 static int ena_close(struct net_device
*netdev
)
1883 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1885 netif_dbg(adapter
, ifdown
, netdev
, "%s\n", __func__
);
1887 if (test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
1890 /* Check for device status and issue reset if needed*/
1891 check_for_admin_com_state(adapter
);
1892 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
1893 netif_err(adapter
, ifdown
, adapter
->netdev
,
1894 "Destroy failure, restarting device\n");
1895 ena_dump_stats_to_dmesg(adapter
);
1896 /* rtnl lock already obtained in dev_ioctl() layer */
1897 ena_destroy_device(adapter
);
1898 ena_restore_device(adapter
);
1904 static void ena_tx_csum(struct ena_com_tx_ctx
*ena_tx_ctx
, struct sk_buff
*skb
)
1906 u32 mss
= skb_shinfo(skb
)->gso_size
;
1907 struct ena_com_tx_meta
*ena_meta
= &ena_tx_ctx
->ena_meta
;
1910 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) || mss
) {
1911 ena_tx_ctx
->l4_csum_enable
= 1;
1913 ena_tx_ctx
->tso_enable
= 1;
1914 ena_meta
->l4_hdr_len
= tcp_hdr(skb
)->doff
;
1915 ena_tx_ctx
->l4_csum_partial
= 0;
1917 ena_tx_ctx
->tso_enable
= 0;
1918 ena_meta
->l4_hdr_len
= 0;
1919 ena_tx_ctx
->l4_csum_partial
= 1;
1922 switch (ip_hdr(skb
)->version
) {
1924 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV4
;
1925 if (ip_hdr(skb
)->frag_off
& htons(IP_DF
))
1928 ena_tx_ctx
->l3_csum_enable
= 1;
1929 l4_protocol
= ip_hdr(skb
)->protocol
;
1932 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV6
;
1933 l4_protocol
= ipv6_hdr(skb
)->nexthdr
;
1939 if (l4_protocol
== IPPROTO_TCP
)
1940 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_TCP
;
1942 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_UDP
;
1944 ena_meta
->mss
= mss
;
1945 ena_meta
->l3_hdr_len
= skb_network_header_len(skb
);
1946 ena_meta
->l3_hdr_offset
= skb_network_offset(skb
);
1947 ena_tx_ctx
->meta_valid
= 1;
1950 ena_tx_ctx
->meta_valid
= 0;
1954 static int ena_check_and_linearize_skb(struct ena_ring
*tx_ring
,
1955 struct sk_buff
*skb
)
1957 int num_frags
, header_len
, rc
;
1959 num_frags
= skb_shinfo(skb
)->nr_frags
;
1960 header_len
= skb_headlen(skb
);
1962 if (num_frags
< tx_ring
->sgl_size
)
1965 if ((num_frags
== tx_ring
->sgl_size
) &&
1966 (header_len
< tx_ring
->tx_max_header_size
))
1969 u64_stats_update_begin(&tx_ring
->syncp
);
1970 tx_ring
->tx_stats
.linearize
++;
1971 u64_stats_update_end(&tx_ring
->syncp
);
1973 rc
= skb_linearize(skb
);
1975 u64_stats_update_begin(&tx_ring
->syncp
);
1976 tx_ring
->tx_stats
.linearize_failed
++;
1977 u64_stats_update_end(&tx_ring
->syncp
);
1983 /* Called with netif_tx_lock. */
1984 static netdev_tx_t
ena_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1986 struct ena_adapter
*adapter
= netdev_priv(dev
);
1987 struct ena_tx_buffer
*tx_info
;
1988 struct ena_com_tx_ctx ena_tx_ctx
;
1989 struct ena_ring
*tx_ring
;
1990 struct netdev_queue
*txq
;
1991 struct ena_com_buf
*ena_buf
;
1999 int qid
, rc
, nb_hw_desc
;
2002 netif_dbg(adapter
, tx_queued
, dev
, "%s skb %p\n", __func__
, skb
);
2003 /* Determine which tx ring we will be placed on */
2004 qid
= skb_get_queue_mapping(skb
);
2005 tx_ring
= &adapter
->tx_ring
[qid
];
2006 txq
= netdev_get_tx_queue(dev
, qid
);
2008 rc
= ena_check_and_linearize_skb(tx_ring
, skb
);
2010 goto error_drop_packet
;
2012 skb_tx_timestamp(skb
);
2013 len
= skb_headlen(skb
);
2015 next_to_use
= tx_ring
->next_to_use
;
2016 req_id
= tx_ring
->free_tx_ids
[next_to_use
];
2017 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
2018 tx_info
->num_of_bufs
= 0;
2020 WARN(tx_info
->skb
, "SKB isn't NULL req_id %d\n", req_id
);
2021 ena_buf
= tx_info
->bufs
;
2024 if (tx_ring
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2025 /* prepared the push buffer */
2026 push_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
2027 header_len
= push_len
;
2028 push_hdr
= skb
->data
;
2031 header_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
2035 netif_dbg(adapter
, tx_queued
, dev
,
2036 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb
,
2037 push_hdr
, push_len
);
2039 if (len
> push_len
) {
2040 dma
= dma_map_single(tx_ring
->dev
, skb
->data
+ push_len
,
2041 len
- push_len
, DMA_TO_DEVICE
);
2042 if (dma_mapping_error(tx_ring
->dev
, dma
))
2043 goto error_report_dma_error
;
2045 ena_buf
->paddr
= dma
;
2046 ena_buf
->len
= len
- push_len
;
2049 tx_info
->num_of_bufs
++;
2052 last_frag
= skb_shinfo(skb
)->nr_frags
;
2054 for (i
= 0; i
< last_frag
; i
++) {
2055 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2057 len
= skb_frag_size(frag
);
2058 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, len
,
2060 if (dma_mapping_error(tx_ring
->dev
, dma
))
2061 goto error_report_dma_error
;
2063 ena_buf
->paddr
= dma
;
2068 tx_info
->num_of_bufs
+= last_frag
;
2070 memset(&ena_tx_ctx
, 0x0, sizeof(struct ena_com_tx_ctx
));
2071 ena_tx_ctx
.ena_bufs
= tx_info
->bufs
;
2072 ena_tx_ctx
.push_header
= push_hdr
;
2073 ena_tx_ctx
.num_bufs
= tx_info
->num_of_bufs
;
2074 ena_tx_ctx
.req_id
= req_id
;
2075 ena_tx_ctx
.header_len
= header_len
;
2077 /* set flags and meta data */
2078 ena_tx_csum(&ena_tx_ctx
, skb
);
2080 /* prepare the packet's descriptors to dma engine */
2081 rc
= ena_com_prepare_tx(tx_ring
->ena_com_io_sq
, &ena_tx_ctx
,
2085 netif_err(adapter
, tx_queued
, dev
,
2086 "failed to prepare tx bufs\n");
2087 u64_stats_update_begin(&tx_ring
->syncp
);
2088 tx_ring
->tx_stats
.queue_stop
++;
2089 tx_ring
->tx_stats
.prepare_ctx_err
++;
2090 u64_stats_update_end(&tx_ring
->syncp
);
2091 netif_tx_stop_queue(txq
);
2092 goto error_unmap_dma
;
2095 netdev_tx_sent_queue(txq
, skb
->len
);
2097 u64_stats_update_begin(&tx_ring
->syncp
);
2098 tx_ring
->tx_stats
.cnt
++;
2099 tx_ring
->tx_stats
.bytes
+= skb
->len
;
2100 u64_stats_update_end(&tx_ring
->syncp
);
2102 tx_info
->tx_descs
= nb_hw_desc
;
2103 tx_info
->last_jiffies
= jiffies
;
2104 tx_info
->print_once
= 0;
2106 tx_ring
->next_to_use
= ENA_TX_RING_IDX_NEXT(next_to_use
,
2107 tx_ring
->ring_size
);
2109 /* This WMB is aimed to:
2110 * 1 - perform smp barrier before reading next_to_completion
2111 * 2 - make sure the desc were written before trigger DB
2115 /* stop the queue when no more space available, the packet can have up
2116 * to sgl_size + 2. one for the meta descriptor and one for header
2117 * (if the header is larger than tx_max_header_size).
2119 if (unlikely(ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) <
2120 (tx_ring
->sgl_size
+ 2))) {
2121 netif_dbg(adapter
, tx_queued
, dev
, "%s stop queue %d\n",
2124 netif_tx_stop_queue(txq
);
2125 u64_stats_update_begin(&tx_ring
->syncp
);
2126 tx_ring
->tx_stats
.queue_stop
++;
2127 u64_stats_update_end(&tx_ring
->syncp
);
2129 /* There is a rare condition where this function decide to
2130 * stop the queue but meanwhile clean_tx_irq updates
2131 * next_to_completion and terminates.
2132 * The queue will remain stopped forever.
2133 * To solve this issue this function perform rmb, check
2134 * the wakeup condition and wake up the queue if needed.
2138 if (ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
)
2139 > ENA_TX_WAKEUP_THRESH
) {
2140 netif_tx_wake_queue(txq
);
2141 u64_stats_update_begin(&tx_ring
->syncp
);
2142 tx_ring
->tx_stats
.queue_wakeup
++;
2143 u64_stats_update_end(&tx_ring
->syncp
);
2147 if (netif_xmit_stopped(txq
) || !skb
->xmit_more
) {
2148 /* trigger the dma engine */
2149 ena_com_write_sq_doorbell(tx_ring
->ena_com_io_sq
);
2150 u64_stats_update_begin(&tx_ring
->syncp
);
2151 tx_ring
->tx_stats
.doorbells
++;
2152 u64_stats_update_end(&tx_ring
->syncp
);
2155 return NETDEV_TX_OK
;
2157 error_report_dma_error
:
2158 u64_stats_update_begin(&tx_ring
->syncp
);
2159 tx_ring
->tx_stats
.dma_mapping_err
++;
2160 u64_stats_update_end(&tx_ring
->syncp
);
2161 netdev_warn(adapter
->netdev
, "failed to map skb\n");
2163 tx_info
->skb
= NULL
;
2167 /* save value of frag that failed */
2170 /* start back at beginning and unmap skb */
2171 tx_info
->skb
= NULL
;
2172 ena_buf
= tx_info
->bufs
;
2173 dma_unmap_single(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2174 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2176 /* unmap remaining mapped pages */
2177 for (i
= 0; i
< last_frag
; i
++) {
2179 dma_unmap_page(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2180 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2187 return NETDEV_TX_OK
;
2190 #ifdef CONFIG_NET_POLL_CONTROLLER
2191 static void ena_netpoll(struct net_device
*netdev
)
2193 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2196 /* Dont schedule NAPI if the driver is in the middle of reset
2197 * or netdev is down.
2200 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
) ||
2201 test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2204 for (i
= 0; i
< adapter
->num_queues
; i
++)
2205 napi_schedule(&adapter
->ena_napi
[i
].napi
);
2207 #endif /* CONFIG_NET_POLL_CONTROLLER */
2209 static u16
ena_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
2210 void *accel_priv
, select_queue_fallback_t fallback
)
2213 /* we suspect that this is good for in--kernel network services that
2214 * want to loop incoming skb rx to tx in normal user generated traffic,
2215 * most probably we will not get to this
2217 if (skb_rx_queue_recorded(skb
))
2218 qid
= skb_get_rx_queue(skb
);
2220 qid
= fallback(dev
, skb
);
2225 static void ena_config_host_info(struct ena_com_dev
*ena_dev
)
2227 struct ena_admin_host_info
*host_info
;
2230 /* Allocate only the host info */
2231 rc
= ena_com_allocate_host_info(ena_dev
);
2233 pr_err("Cannot allocate host info\n");
2237 host_info
= ena_dev
->host_attr
.host_info
;
2239 host_info
->os_type
= ENA_ADMIN_OS_LINUX
;
2240 host_info
->kernel_ver
= LINUX_VERSION_CODE
;
2241 strncpy(host_info
->kernel_ver_str
, utsname()->version
,
2242 sizeof(host_info
->kernel_ver_str
) - 1);
2243 host_info
->os_dist
= 0;
2244 strncpy(host_info
->os_dist_str
, utsname()->release
,
2245 sizeof(host_info
->os_dist_str
) - 1);
2246 host_info
->driver_version
=
2247 (DRV_MODULE_VER_MAJOR
) |
2248 (DRV_MODULE_VER_MINOR
<< ENA_ADMIN_HOST_INFO_MINOR_SHIFT
) |
2249 (DRV_MODULE_VER_SUBMINOR
<< ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT
);
2251 rc
= ena_com_set_host_attributes(ena_dev
);
2253 if (rc
== -EOPNOTSUPP
)
2254 pr_warn("Cannot set host attributes\n");
2256 pr_err("Cannot set host attributes\n");
2264 ena_com_delete_host_info(ena_dev
);
2267 static void ena_config_debug_area(struct ena_adapter
*adapter
)
2269 u32 debug_area_size
;
2272 ss_count
= ena_get_sset_count(adapter
->netdev
, ETH_SS_STATS
);
2273 if (ss_count
<= 0) {
2274 netif_err(adapter
, drv
, adapter
->netdev
,
2275 "SS count is negative\n");
2279 /* allocate 32 bytes for each string and 64bit for the value */
2280 debug_area_size
= ss_count
* ETH_GSTRING_LEN
+ sizeof(u64
) * ss_count
;
2282 rc
= ena_com_allocate_debug_area(adapter
->ena_dev
, debug_area_size
);
2284 pr_err("Cannot allocate debug area\n");
2288 rc
= ena_com_set_host_attributes(adapter
->ena_dev
);
2290 if (rc
== -EOPNOTSUPP
)
2291 netif_warn(adapter
, drv
, adapter
->netdev
,
2292 "Cannot set host attributes\n");
2294 netif_err(adapter
, drv
, adapter
->netdev
,
2295 "Cannot set host attributes\n");
2301 ena_com_delete_debug_area(adapter
->ena_dev
);
2304 static void ena_get_stats64(struct net_device
*netdev
,
2305 struct rtnl_link_stats64
*stats
)
2307 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2308 struct ena_ring
*rx_ring
, *tx_ring
;
2313 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2316 for (i
= 0; i
< adapter
->num_queues
; i
++) {
2319 tx_ring
= &adapter
->tx_ring
[i
];
2322 start
= u64_stats_fetch_begin_irq(&tx_ring
->syncp
);
2323 packets
= tx_ring
->tx_stats
.cnt
;
2324 bytes
= tx_ring
->tx_stats
.bytes
;
2325 } while (u64_stats_fetch_retry_irq(&tx_ring
->syncp
, start
));
2327 stats
->tx_packets
+= packets
;
2328 stats
->tx_bytes
+= bytes
;
2330 rx_ring
= &adapter
->rx_ring
[i
];
2333 start
= u64_stats_fetch_begin_irq(&rx_ring
->syncp
);
2334 packets
= rx_ring
->rx_stats
.cnt
;
2335 bytes
= rx_ring
->rx_stats
.bytes
;
2336 } while (u64_stats_fetch_retry_irq(&rx_ring
->syncp
, start
));
2338 stats
->rx_packets
+= packets
;
2339 stats
->rx_bytes
+= bytes
;
2343 start
= u64_stats_fetch_begin_irq(&adapter
->syncp
);
2344 rx_drops
= adapter
->dev_stats
.rx_drops
;
2345 } while (u64_stats_fetch_retry_irq(&adapter
->syncp
, start
));
2347 stats
->rx_dropped
= rx_drops
;
2349 stats
->multicast
= 0;
2350 stats
->collisions
= 0;
2352 stats
->rx_length_errors
= 0;
2353 stats
->rx_crc_errors
= 0;
2354 stats
->rx_frame_errors
= 0;
2355 stats
->rx_fifo_errors
= 0;
2356 stats
->rx_missed_errors
= 0;
2357 stats
->tx_window_errors
= 0;
2359 stats
->rx_errors
= 0;
2360 stats
->tx_errors
= 0;
2363 static const struct net_device_ops ena_netdev_ops
= {
2364 .ndo_open
= ena_open
,
2365 .ndo_stop
= ena_close
,
2366 .ndo_start_xmit
= ena_start_xmit
,
2367 .ndo_select_queue
= ena_select_queue
,
2368 .ndo_get_stats64
= ena_get_stats64
,
2369 .ndo_tx_timeout
= ena_tx_timeout
,
2370 .ndo_change_mtu
= ena_change_mtu
,
2371 .ndo_set_mac_address
= NULL
,
2372 .ndo_validate_addr
= eth_validate_addr
,
2373 #ifdef CONFIG_NET_POLL_CONTROLLER
2374 .ndo_poll_controller
= ena_netpoll
,
2375 #endif /* CONFIG_NET_POLL_CONTROLLER */
2378 static int ena_device_validate_params(struct ena_adapter
*adapter
,
2379 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2381 struct net_device
*netdev
= adapter
->netdev
;
2384 rc
= ether_addr_equal(get_feat_ctx
->dev_attr
.mac_addr
,
2387 netif_err(adapter
, drv
, netdev
,
2388 "Error, mac address are different\n");
2392 if ((get_feat_ctx
->max_queues
.max_cq_num
< adapter
->num_queues
) ||
2393 (get_feat_ctx
->max_queues
.max_sq_num
< adapter
->num_queues
)) {
2394 netif_err(adapter
, drv
, netdev
,
2395 "Error, device doesn't support enough queues\n");
2399 if (get_feat_ctx
->dev_attr
.max_mtu
< netdev
->mtu
) {
2400 netif_err(adapter
, drv
, netdev
,
2401 "Error, device max mtu is smaller than netdev MTU\n");
2408 static int ena_device_init(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
,
2409 struct ena_com_dev_get_features_ctx
*get_feat_ctx
,
2412 struct device
*dev
= &pdev
->dev
;
2413 bool readless_supported
;
2418 rc
= ena_com_mmio_reg_read_request_init(ena_dev
);
2420 dev_err(dev
, "failed to init mmio read less\n");
2424 /* The PCIe configuration space revision id indicate if mmio reg
2427 readless_supported
= !(pdev
->revision
& ENA_MMIO_DISABLE_REG_READ
);
2428 ena_com_set_mmio_read_mode(ena_dev
, readless_supported
);
2430 rc
= ena_com_dev_reset(ena_dev
, ENA_REGS_RESET_NORMAL
);
2432 dev_err(dev
, "Can not reset device\n");
2433 goto err_mmio_read_less
;
2436 rc
= ena_com_validate_version(ena_dev
);
2438 dev_err(dev
, "device version is too low\n");
2439 goto err_mmio_read_less
;
2442 dma_width
= ena_com_get_dma_width(ena_dev
);
2443 if (dma_width
< 0) {
2444 dev_err(dev
, "Invalid dma width value %d", dma_width
);
2446 goto err_mmio_read_less
;
2449 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2451 dev_err(dev
, "pci_set_dma_mask failed 0x%x\n", rc
);
2452 goto err_mmio_read_less
;
2455 rc
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2457 dev_err(dev
, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2459 goto err_mmio_read_less
;
2462 /* ENA admin level init */
2463 rc
= ena_com_admin_init(ena_dev
, &aenq_handlers
, true);
2466 "Can not initialize ena admin queue with device\n");
2467 goto err_mmio_read_less
;
2470 /* To enable the msix interrupts the driver needs to know the number
2471 * of queues. So the driver uses polling mode to retrieve this
2474 ena_com_set_admin_polling_mode(ena_dev
, true);
2476 ena_config_host_info(ena_dev
);
2478 /* Get Device Attributes*/
2479 rc
= ena_com_get_dev_attr_feat(ena_dev
, get_feat_ctx
);
2481 dev_err(dev
, "Cannot get attribute for ena device rc=%d\n", rc
);
2482 goto err_admin_init
;
2485 /* Try to turn all the available aenq groups */
2486 aenq_groups
= BIT(ENA_ADMIN_LINK_CHANGE
) |
2487 BIT(ENA_ADMIN_FATAL_ERROR
) |
2488 BIT(ENA_ADMIN_WARNING
) |
2489 BIT(ENA_ADMIN_NOTIFICATION
) |
2490 BIT(ENA_ADMIN_KEEP_ALIVE
);
2492 aenq_groups
&= get_feat_ctx
->aenq
.supported_groups
;
2494 rc
= ena_com_set_aenq_config(ena_dev
, aenq_groups
);
2496 dev_err(dev
, "Cannot configure aenq groups rc= %d\n", rc
);
2497 goto err_admin_init
;
2500 *wd_state
= !!(aenq_groups
& BIT(ENA_ADMIN_KEEP_ALIVE
));
2505 ena_com_delete_host_info(ena_dev
);
2506 ena_com_admin_destroy(ena_dev
);
2508 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2513 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter
*adapter
,
2516 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2517 struct device
*dev
= &adapter
->pdev
->dev
;
2520 rc
= ena_enable_msix(adapter
, io_vectors
);
2522 dev_err(dev
, "Can not reserve msix vectors\n");
2526 ena_setup_mgmnt_intr(adapter
);
2528 rc
= ena_request_mgmnt_irq(adapter
);
2530 dev_err(dev
, "Can not setup management interrupts\n");
2531 goto err_disable_msix
;
2534 ena_com_set_admin_polling_mode(ena_dev
, false);
2536 ena_com_admin_aenq_enable(ena_dev
);
2541 ena_disable_msix(adapter
);
2546 static void ena_destroy_device(struct ena_adapter
*adapter
)
2548 struct net_device
*netdev
= adapter
->netdev
;
2549 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2552 netif_carrier_off(netdev
);
2554 del_timer_sync(&adapter
->timer_service
);
2556 dev_up
= test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
2557 adapter
->dev_up_before_reset
= dev_up
;
2559 ena_com_set_admin_running_state(ena_dev
, false);
2561 if (test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2564 /* Before releasing the ENA resources, a device reset is required.
2565 * (to prevent the device from accessing them).
2566 * In case the reset flag is set and the device is up, ena_down()
2567 * already perform the reset, so it can be skipped.
2569 if (!(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
) && dev_up
))
2570 ena_com_dev_reset(adapter
->ena_dev
, adapter
->reset_reason
);
2572 ena_free_mgmnt_irq(adapter
);
2574 ena_disable_msix(adapter
);
2576 ena_com_abort_admin_commands(ena_dev
);
2578 ena_com_wait_for_abort_completion(ena_dev
);
2580 ena_com_admin_destroy(ena_dev
);
2582 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2584 adapter
->reset_reason
= ENA_REGS_RESET_NORMAL
;
2586 clear_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2589 static int ena_restore_device(struct ena_adapter
*adapter
)
2591 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2592 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2593 struct pci_dev
*pdev
= adapter
->pdev
;
2597 set_bit(ENA_FLAG_ONGOING_RESET
, &adapter
->flags
);
2598 rc
= ena_device_init(ena_dev
, adapter
->pdev
, &get_feat_ctx
, &wd_state
);
2600 dev_err(&pdev
->dev
, "Can not initialize device\n");
2603 adapter
->wd_state
= wd_state
;
2605 rc
= ena_device_validate_params(adapter
, &get_feat_ctx
);
2607 dev_err(&pdev
->dev
, "Validation of device parameters failed\n");
2608 goto err_device_destroy
;
2611 clear_bit(ENA_FLAG_ONGOING_RESET
, &adapter
->flags
);
2612 /* Make sure we don't have a race with AENQ Links state handler */
2613 if (test_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
))
2614 netif_carrier_on(adapter
->netdev
);
2616 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
,
2617 adapter
->num_queues
);
2619 dev_err(&pdev
->dev
, "Enable MSI-X failed\n");
2620 goto err_device_destroy
;
2622 /* If the interface was up before the reset bring it up */
2623 if (adapter
->dev_up_before_reset
) {
2624 rc
= ena_up(adapter
);
2626 dev_err(&pdev
->dev
, "Failed to create I/O queues\n");
2627 goto err_disable_msix
;
2631 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
2632 dev_err(&pdev
->dev
, "Device reset completed successfully\n");
2636 ena_free_mgmnt_irq(adapter
);
2637 ena_disable_msix(adapter
);
2639 ena_com_admin_destroy(ena_dev
);
2641 clear_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
2642 clear_bit(ENA_FLAG_ONGOING_RESET
, &adapter
->flags
);
2644 "Reset attempt failed. Can not reset the device\n");
2649 static void ena_fw_reset_device(struct work_struct
*work
)
2651 struct ena_adapter
*adapter
=
2652 container_of(work
, struct ena_adapter
, reset_task
);
2653 struct pci_dev
*pdev
= adapter
->pdev
;
2655 if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2657 "device reset schedule while reset bit is off\n");
2661 ena_destroy_device(adapter
);
2662 ena_restore_device(adapter
);
2666 static int check_missing_comp_in_queue(struct ena_adapter
*adapter
,
2667 struct ena_ring
*tx_ring
)
2669 struct ena_tx_buffer
*tx_buf
;
2670 unsigned long last_jiffies
;
2674 for (i
= 0; i
< tx_ring
->ring_size
; i
++) {
2675 tx_buf
= &tx_ring
->tx_buffer_info
[i
];
2676 last_jiffies
= tx_buf
->last_jiffies
;
2677 if (unlikely(last_jiffies
&&
2678 time_is_before_jiffies(last_jiffies
+ adapter
->missing_tx_completion_to
))) {
2679 if (!tx_buf
->print_once
)
2680 netif_notice(adapter
, tx_err
, adapter
->netdev
,
2681 "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2684 tx_buf
->print_once
= 1;
2689 if (unlikely(missed_tx
> adapter
->missing_tx_completion_threshold
)) {
2690 netif_err(adapter
, tx_err
, adapter
->netdev
,
2691 "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2693 adapter
->missing_tx_completion_threshold
);
2694 adapter
->reset_reason
=
2695 ENA_REGS_RESET_MISS_TX_CMPL
;
2696 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2700 u64_stats_update_begin(&tx_ring
->syncp
);
2701 tx_ring
->tx_stats
.missed_tx
= missed_tx
;
2702 u64_stats_update_end(&tx_ring
->syncp
);
2707 static void check_for_missing_tx_completions(struct ena_adapter
*adapter
)
2709 struct ena_ring
*tx_ring
;
2712 /* Make sure the driver doesn't turn the device in other process */
2715 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2718 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2721 if (adapter
->missing_tx_completion_to
== ENA_HW_HINTS_NO_TIMEOUT
)
2724 budget
= ENA_MONITORED_TX_QUEUES
;
2726 for (i
= adapter
->last_monitored_tx_qid
; i
< adapter
->num_queues
; i
++) {
2727 tx_ring
= &adapter
->tx_ring
[i
];
2729 rc
= check_missing_comp_in_queue(adapter
, tx_ring
);
2738 adapter
->last_monitored_tx_qid
= i
% adapter
->num_queues
;
2741 /* trigger napi schedule after 2 consecutive detections */
2742 #define EMPTY_RX_REFILL 2
2743 /* For the rare case where the device runs out of Rx descriptors and the
2744 * napi handler failed to refill new Rx descriptors (due to a lack of memory
2746 * This case will lead to a deadlock:
2747 * The device won't send interrupts since all the new Rx packets will be dropped
2748 * The napi handler won't allocate new Rx descriptors so the device will be
2749 * able to send new packets.
2751 * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2752 * It is recommended to have at least 512MB, with a minimum of 128MB for
2753 * constrained environment).
2755 * When such a situation is detected - Reschedule napi
2757 static void check_for_empty_rx_ring(struct ena_adapter
*adapter
)
2759 struct ena_ring
*rx_ring
;
2760 int i
, refill_required
;
2762 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2765 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2768 for (i
= 0; i
< adapter
->num_queues
; i
++) {
2769 rx_ring
= &adapter
->rx_ring
[i
];
2772 ena_com_sq_empty_space(rx_ring
->ena_com_io_sq
);
2773 if (unlikely(refill_required
== (rx_ring
->ring_size
- 1))) {
2774 rx_ring
->empty_rx_queue
++;
2776 if (rx_ring
->empty_rx_queue
>= EMPTY_RX_REFILL
) {
2777 u64_stats_update_begin(&rx_ring
->syncp
);
2778 rx_ring
->rx_stats
.empty_rx_ring
++;
2779 u64_stats_update_end(&rx_ring
->syncp
);
2781 netif_err(adapter
, drv
, adapter
->netdev
,
2782 "trigger refill for ring %d\n", i
);
2784 napi_schedule(rx_ring
->napi
);
2785 rx_ring
->empty_rx_queue
= 0;
2788 rx_ring
->empty_rx_queue
= 0;
2793 /* Check for keep alive expiration */
2794 static void check_for_missing_keep_alive(struct ena_adapter
*adapter
)
2796 unsigned long keep_alive_expired
;
2798 if (!adapter
->wd_state
)
2801 if (adapter
->keep_alive_timeout
== ENA_HW_HINTS_NO_TIMEOUT
)
2804 keep_alive_expired
= round_jiffies(adapter
->last_keep_alive_jiffies
+
2805 adapter
->keep_alive_timeout
);
2806 if (unlikely(time_is_before_jiffies(keep_alive_expired
))) {
2807 netif_err(adapter
, drv
, adapter
->netdev
,
2808 "Keep alive watchdog timeout.\n");
2809 u64_stats_update_begin(&adapter
->syncp
);
2810 adapter
->dev_stats
.wd_expired
++;
2811 u64_stats_update_end(&adapter
->syncp
);
2812 adapter
->reset_reason
= ENA_REGS_RESET_KEEP_ALIVE_TO
;
2813 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2817 static void check_for_admin_com_state(struct ena_adapter
*adapter
)
2819 if (unlikely(!ena_com_get_admin_running_state(adapter
->ena_dev
))) {
2820 netif_err(adapter
, drv
, adapter
->netdev
,
2821 "ENA admin queue is not in running state!\n");
2822 u64_stats_update_begin(&adapter
->syncp
);
2823 adapter
->dev_stats
.admin_q_pause
++;
2824 u64_stats_update_end(&adapter
->syncp
);
2825 adapter
->reset_reason
= ENA_REGS_RESET_ADMIN_TO
;
2826 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2830 static void ena_update_hints(struct ena_adapter
*adapter
,
2831 struct ena_admin_ena_hw_hints
*hints
)
2833 struct net_device
*netdev
= adapter
->netdev
;
2835 if (hints
->admin_completion_tx_timeout
)
2836 adapter
->ena_dev
->admin_queue
.completion_timeout
=
2837 hints
->admin_completion_tx_timeout
* 1000;
2839 if (hints
->mmio_read_timeout
)
2840 /* convert to usec */
2841 adapter
->ena_dev
->mmio_read
.reg_read_to
=
2842 hints
->mmio_read_timeout
* 1000;
2844 if (hints
->missed_tx_completion_count_threshold_to_reset
)
2845 adapter
->missing_tx_completion_threshold
=
2846 hints
->missed_tx_completion_count_threshold_to_reset
;
2848 if (hints
->missing_tx_completion_timeout
) {
2849 if (hints
->missing_tx_completion_timeout
== ENA_HW_HINTS_NO_TIMEOUT
)
2850 adapter
->missing_tx_completion_to
= ENA_HW_HINTS_NO_TIMEOUT
;
2852 adapter
->missing_tx_completion_to
=
2853 msecs_to_jiffies(hints
->missing_tx_completion_timeout
);
2856 if (hints
->netdev_wd_timeout
)
2857 netdev
->watchdog_timeo
= msecs_to_jiffies(hints
->netdev_wd_timeout
);
2859 if (hints
->driver_watchdog_timeout
) {
2860 if (hints
->driver_watchdog_timeout
== ENA_HW_HINTS_NO_TIMEOUT
)
2861 adapter
->keep_alive_timeout
= ENA_HW_HINTS_NO_TIMEOUT
;
2863 adapter
->keep_alive_timeout
=
2864 msecs_to_jiffies(hints
->driver_watchdog_timeout
);
2868 static void ena_update_host_info(struct ena_admin_host_info
*host_info
,
2869 struct net_device
*netdev
)
2871 host_info
->supported_network_features
[0] =
2872 netdev
->features
& GENMASK_ULL(31, 0);
2873 host_info
->supported_network_features
[1] =
2874 (netdev
->features
& GENMASK_ULL(63, 32)) >> 32;
2877 static void ena_timer_service(struct timer_list
*t
)
2879 struct ena_adapter
*adapter
= from_timer(adapter
, t
, timer_service
);
2880 u8
*debug_area
= adapter
->ena_dev
->host_attr
.debug_area_virt_addr
;
2881 struct ena_admin_host_info
*host_info
=
2882 adapter
->ena_dev
->host_attr
.host_info
;
2884 check_for_missing_keep_alive(adapter
);
2886 check_for_admin_com_state(adapter
);
2888 check_for_missing_tx_completions(adapter
);
2890 check_for_empty_rx_ring(adapter
);
2893 ena_dump_stats_to_buf(adapter
, debug_area
);
2896 ena_update_host_info(host_info
, adapter
->netdev
);
2898 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2899 netif_err(adapter
, drv
, adapter
->netdev
,
2900 "Trigger reset is on\n");
2901 ena_dump_stats_to_dmesg(adapter
);
2902 queue_work(ena_wq
, &adapter
->reset_task
);
2906 /* Reset the timer */
2907 mod_timer(&adapter
->timer_service
, jiffies
+ HZ
);
2910 static int ena_calc_io_queue_num(struct pci_dev
*pdev
,
2911 struct ena_com_dev
*ena_dev
,
2912 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2914 int io_sq_num
, io_queue_num
;
2916 /* In case of LLQ use the llq number in the get feature cmd */
2917 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2918 io_sq_num
= get_feat_ctx
->max_queues
.max_llq_num
;
2920 if (io_sq_num
== 0) {
2922 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2924 ena_dev
->tx_mem_queue_type
=
2925 ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2926 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2929 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2932 io_queue_num
= min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES
);
2933 io_queue_num
= min_t(int, io_queue_num
, io_sq_num
);
2934 io_queue_num
= min_t(int, io_queue_num
,
2935 get_feat_ctx
->max_queues
.max_cq_num
);
2936 /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2937 io_queue_num
= min_t(int, io_queue_num
, pci_msix_vec_count(pdev
) - 1);
2938 if (unlikely(!io_queue_num
)) {
2939 dev_err(&pdev
->dev
, "The device doesn't have io queues\n");
2943 return io_queue_num
;
2946 static void ena_set_push_mode(struct pci_dev
*pdev
, struct ena_com_dev
*ena_dev
,
2947 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2951 has_mem_bar
= pci_select_bars(pdev
, IORESOURCE_MEM
) & BIT(ENA_MEM_BAR
);
2953 /* Enable push mode if device supports LLQ */
2954 if (has_mem_bar
&& (get_feat_ctx
->max_queues
.max_llq_num
> 0))
2955 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_DEV
;
2957 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2960 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx
*feat
,
2961 struct net_device
*netdev
)
2963 netdev_features_t dev_features
= 0;
2965 /* Set offload features */
2966 if (feat
->offload
.tx
&
2967 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK
)
2968 dev_features
|= NETIF_F_IP_CSUM
;
2970 if (feat
->offload
.tx
&
2971 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK
)
2972 dev_features
|= NETIF_F_IPV6_CSUM
;
2974 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK
)
2975 dev_features
|= NETIF_F_TSO
;
2977 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK
)
2978 dev_features
|= NETIF_F_TSO6
;
2980 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK
)
2981 dev_features
|= NETIF_F_TSO_ECN
;
2983 if (feat
->offload
.rx_supported
&
2984 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK
)
2985 dev_features
|= NETIF_F_RXCSUM
;
2987 if (feat
->offload
.rx_supported
&
2988 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK
)
2989 dev_features
|= NETIF_F_RXCSUM
;
2997 netdev
->hw_features
|= netdev
->features
;
2998 netdev
->vlan_features
|= netdev
->features
;
3001 static void ena_set_conf_feat_params(struct ena_adapter
*adapter
,
3002 struct ena_com_dev_get_features_ctx
*feat
)
3004 struct net_device
*netdev
= adapter
->netdev
;
3006 /* Copy mac address */
3007 if (!is_valid_ether_addr(feat
->dev_attr
.mac_addr
)) {
3008 eth_hw_addr_random(netdev
);
3009 ether_addr_copy(adapter
->mac_addr
, netdev
->dev_addr
);
3011 ether_addr_copy(adapter
->mac_addr
, feat
->dev_attr
.mac_addr
);
3012 ether_addr_copy(netdev
->dev_addr
, adapter
->mac_addr
);
3015 /* Set offload features */
3016 ena_set_dev_offloads(feat
, netdev
);
3018 adapter
->max_mtu
= feat
->dev_attr
.max_mtu
;
3019 netdev
->max_mtu
= adapter
->max_mtu
;
3020 netdev
->min_mtu
= ENA_MIN_MTU
;
3023 static int ena_rss_init_default(struct ena_adapter
*adapter
)
3025 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
3026 struct device
*dev
= &adapter
->pdev
->dev
;
3030 rc
= ena_com_rss_init(ena_dev
, ENA_RX_RSS_TABLE_LOG_SIZE
);
3032 dev_err(dev
, "Cannot init indirect table\n");
3036 for (i
= 0; i
< ENA_RX_RSS_TABLE_SIZE
; i
++) {
3037 val
= ethtool_rxfh_indir_default(i
, adapter
->num_queues
);
3038 rc
= ena_com_indirect_table_fill_entry(ena_dev
, i
,
3039 ENA_IO_RXQ_IDX(val
));
3040 if (unlikely(rc
&& (rc
!= -EOPNOTSUPP
))) {
3041 dev_err(dev
, "Cannot fill indirect table\n");
3042 goto err_fill_indir
;
3046 rc
= ena_com_fill_hash_function(ena_dev
, ENA_ADMIN_CRC32
, NULL
,
3047 ENA_HASH_KEY_SIZE
, 0xFFFFFFFF);
3048 if (unlikely(rc
&& (rc
!= -EOPNOTSUPP
))) {
3049 dev_err(dev
, "Cannot fill hash function\n");
3050 goto err_fill_indir
;
3053 rc
= ena_com_set_default_hash_ctrl(ena_dev
);
3054 if (unlikely(rc
&& (rc
!= -EOPNOTSUPP
))) {
3055 dev_err(dev
, "Cannot fill hash control\n");
3056 goto err_fill_indir
;
3062 ena_com_rss_destroy(ena_dev
);
3068 static void ena_release_bars(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
)
3072 if (ena_dev
->mem_bar
)
3073 devm_iounmap(&pdev
->dev
, ena_dev
->mem_bar
);
3075 if (ena_dev
->reg_bar
)
3076 devm_iounmap(&pdev
->dev
, ena_dev
->reg_bar
);
3078 release_bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
3079 pci_release_selected_regions(pdev
, release_bars
);
3082 static int ena_calc_queue_size(struct pci_dev
*pdev
,
3083 struct ena_com_dev
*ena_dev
,
3084 u16
*max_tx_sgl_size
,
3085 u16
*max_rx_sgl_size
,
3086 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
3088 u32 queue_size
= ENA_DEFAULT_RING_SIZE
;
3090 queue_size
= min_t(u32
, queue_size
,
3091 get_feat_ctx
->max_queues
.max_cq_depth
);
3092 queue_size
= min_t(u32
, queue_size
,
3093 get_feat_ctx
->max_queues
.max_sq_depth
);
3095 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
)
3096 queue_size
= min_t(u32
, queue_size
,
3097 get_feat_ctx
->max_queues
.max_llq_depth
);
3099 queue_size
= rounddown_pow_of_two(queue_size
);
3101 if (unlikely(!queue_size
)) {
3102 dev_err(&pdev
->dev
, "Invalid queue size\n");
3106 *max_tx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
3107 get_feat_ctx
->max_queues
.max_packet_tx_descs
);
3108 *max_rx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
3109 get_feat_ctx
->max_queues
.max_packet_rx_descs
);
3114 /* ena_probe - Device Initialization Routine
3115 * @pdev: PCI device information struct
3116 * @ent: entry in ena_pci_tbl
3118 * Returns 0 on success, negative on failure
3120 * ena_probe initializes an adapter identified by a pci_dev structure.
3121 * The OS initialization, configuring of the adapter private structure,
3122 * and a hardware reset occur.
3124 static int ena_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
3126 struct ena_com_dev_get_features_ctx get_feat_ctx
;
3127 static int version_printed
;
3128 struct net_device
*netdev
;
3129 struct ena_adapter
*adapter
;
3130 struct ena_com_dev
*ena_dev
= NULL
;
3131 static int adapters_found
;
3132 int io_queue_num
, bars
, rc
;
3134 u16 tx_sgl_size
= 0;
3135 u16 rx_sgl_size
= 0;
3138 dev_dbg(&pdev
->dev
, "%s\n", __func__
);
3140 if (version_printed
++ == 0)
3141 dev_info(&pdev
->dev
, "%s", version
);
3143 rc
= pci_enable_device_mem(pdev
);
3145 dev_err(&pdev
->dev
, "pci_enable_device_mem() failed!\n");
3149 pci_set_master(pdev
);
3151 ena_dev
= vzalloc(sizeof(*ena_dev
));
3154 goto err_disable_device
;
3157 bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
3158 rc
= pci_request_selected_regions(pdev
, bars
, DRV_MODULE_NAME
);
3160 dev_err(&pdev
->dev
, "pci_request_selected_regions failed %d\n",
3162 goto err_free_ena_dev
;
3165 ena_dev
->reg_bar
= devm_ioremap(&pdev
->dev
,
3166 pci_resource_start(pdev
, ENA_REG_BAR
),
3167 pci_resource_len(pdev
, ENA_REG_BAR
));
3168 if (!ena_dev
->reg_bar
) {
3169 dev_err(&pdev
->dev
, "failed to remap regs bar\n");
3171 goto err_free_region
;
3174 ena_dev
->dmadev
= &pdev
->dev
;
3176 rc
= ena_device_init(ena_dev
, pdev
, &get_feat_ctx
, &wd_state
);
3178 dev_err(&pdev
->dev
, "ena device init failed\n");
3181 goto err_free_region
;
3184 ena_set_push_mode(pdev
, ena_dev
, &get_feat_ctx
);
3186 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
3187 ena_dev
->mem_bar
= devm_ioremap_wc(&pdev
->dev
,
3188 pci_resource_start(pdev
, ENA_MEM_BAR
),
3189 pci_resource_len(pdev
, ENA_MEM_BAR
));
3190 if (!ena_dev
->mem_bar
) {
3192 goto err_device_destroy
;
3196 /* initial Tx interrupt delay, Assumes 1 usec granularity.
3197 * Updated during device initialization with the real granularity
3199 ena_dev
->intr_moder_tx_interval
= ENA_INTR_INITIAL_TX_INTERVAL_USECS
;
3200 io_queue_num
= ena_calc_io_queue_num(pdev
, ena_dev
, &get_feat_ctx
);
3201 queue_size
= ena_calc_queue_size(pdev
, ena_dev
, &tx_sgl_size
,
3202 &rx_sgl_size
, &get_feat_ctx
);
3203 if ((queue_size
<= 0) || (io_queue_num
<= 0)) {
3205 goto err_device_destroy
;
3208 dev_info(&pdev
->dev
, "creating %d io queues. queue size: %d\n",
3209 io_queue_num
, queue_size
);
3211 /* dev zeroed in init_etherdev */
3212 netdev
= alloc_etherdev_mq(sizeof(struct ena_adapter
), io_queue_num
);
3214 dev_err(&pdev
->dev
, "alloc_etherdev_mq failed\n");
3216 goto err_device_destroy
;
3219 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
3221 adapter
= netdev_priv(netdev
);
3222 pci_set_drvdata(pdev
, adapter
);
3224 adapter
->ena_dev
= ena_dev
;
3225 adapter
->netdev
= netdev
;
3226 adapter
->pdev
= pdev
;
3228 ena_set_conf_feat_params(adapter
, &get_feat_ctx
);
3230 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
3231 adapter
->reset_reason
= ENA_REGS_RESET_NORMAL
;
3233 adapter
->tx_ring_size
= queue_size
;
3234 adapter
->rx_ring_size
= queue_size
;
3236 adapter
->max_tx_sgl_size
= tx_sgl_size
;
3237 adapter
->max_rx_sgl_size
= rx_sgl_size
;
3239 adapter
->num_queues
= io_queue_num
;
3240 adapter
->last_monitored_tx_qid
= 0;
3242 adapter
->rx_copybreak
= ENA_DEFAULT_RX_COPYBREAK
;
3243 adapter
->wd_state
= wd_state
;
3245 snprintf(adapter
->name
, ENA_NAME_MAX_LEN
, "ena_%d", adapters_found
);
3247 rc
= ena_com_init_interrupt_moderation(adapter
->ena_dev
);
3250 "Failed to query interrupt moderation feature\n");
3251 goto err_netdev_destroy
;
3253 ena_init_io_rings(adapter
);
3255 netdev
->netdev_ops
= &ena_netdev_ops
;
3256 netdev
->watchdog_timeo
= TX_TIMEOUT
;
3257 ena_set_ethtool_ops(netdev
);
3259 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
3261 u64_stats_init(&adapter
->syncp
);
3263 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
, io_queue_num
);
3266 "Failed to enable and set the admin interrupts\n");
3267 goto err_worker_destroy
;
3269 rc
= ena_rss_init_default(adapter
);
3270 if (rc
&& (rc
!= -EOPNOTSUPP
)) {
3271 dev_err(&pdev
->dev
, "Cannot init RSS rc: %d\n", rc
);
3275 ena_config_debug_area(adapter
);
3277 memcpy(adapter
->netdev
->perm_addr
, adapter
->mac_addr
, netdev
->addr_len
);
3279 netif_carrier_off(netdev
);
3281 rc
= register_netdev(netdev
);
3283 dev_err(&pdev
->dev
, "Cannot register net device\n");
3287 INIT_WORK(&adapter
->reset_task
, ena_fw_reset_device
);
3289 adapter
->last_keep_alive_jiffies
= jiffies
;
3290 adapter
->keep_alive_timeout
= ENA_DEVICE_KALIVE_TIMEOUT
;
3291 adapter
->missing_tx_completion_to
= TX_TIMEOUT
;
3292 adapter
->missing_tx_completion_threshold
= MAX_NUM_OF_TIMEOUTED_PACKETS
;
3294 ena_update_hints(adapter
, &get_feat_ctx
.hw_hints
);
3296 timer_setup(&adapter
->timer_service
, ena_timer_service
, 0);
3297 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
3299 dev_info(&pdev
->dev
, "%s found at mem %lx, mac addr %pM Queues %d\n",
3300 DEVICE_NAME
, (long)pci_resource_start(pdev
, 0),
3301 netdev
->dev_addr
, io_queue_num
);
3303 set_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
3310 ena_com_delete_debug_area(ena_dev
);
3311 ena_com_rss_destroy(ena_dev
);
3313 ena_com_dev_reset(ena_dev
, ENA_REGS_RESET_INIT_ERR
);
3314 ena_free_mgmnt_irq(adapter
);
3315 ena_disable_msix(adapter
);
3317 ena_com_destroy_interrupt_moderation(ena_dev
);
3318 del_timer(&adapter
->timer_service
);
3320 free_netdev(netdev
);
3322 ena_com_delete_host_info(ena_dev
);
3323 ena_com_admin_destroy(ena_dev
);
3325 ena_release_bars(ena_dev
, pdev
);
3329 pci_disable_device(pdev
);
3333 /*****************************************************************************/
3334 static int ena_sriov_configure(struct pci_dev
*dev
, int numvfs
)
3339 rc
= pci_enable_sriov(dev
, numvfs
);
3342 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3351 pci_disable_sriov(dev
);
3358 /*****************************************************************************/
3359 /*****************************************************************************/
3361 /* ena_remove - Device Removal Routine
3362 * @pdev: PCI device information struct
3364 * ena_remove is called by the PCI subsystem to alert the driver
3365 * that it should release a PCI device.
3367 static void ena_remove(struct pci_dev
*pdev
)
3369 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3370 struct ena_com_dev
*ena_dev
;
3371 struct net_device
*netdev
;
3373 ena_dev
= adapter
->ena_dev
;
3374 netdev
= adapter
->netdev
;
3376 #ifdef CONFIG_RFS_ACCEL
3377 if ((adapter
->msix_vecs
>= 1) && (netdev
->rx_cpu_rmap
)) {
3378 free_irq_cpu_rmap(netdev
->rx_cpu_rmap
);
3379 netdev
->rx_cpu_rmap
= NULL
;
3381 #endif /* CONFIG_RFS_ACCEL */
3383 unregister_netdev(netdev
);
3384 del_timer_sync(&adapter
->timer_service
);
3386 cancel_work_sync(&adapter
->reset_task
);
3388 /* Reset the device only if the device is running. */
3389 if (test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
))
3390 ena_com_dev_reset(ena_dev
, adapter
->reset_reason
);
3392 ena_free_mgmnt_irq(adapter
);
3394 ena_disable_msix(adapter
);
3396 free_netdev(netdev
);
3398 ena_com_mmio_reg_read_request_destroy(ena_dev
);
3400 ena_com_abort_admin_commands(ena_dev
);
3402 ena_com_wait_for_abort_completion(ena_dev
);
3404 ena_com_admin_destroy(ena_dev
);
3406 ena_com_rss_destroy(ena_dev
);
3408 ena_com_delete_debug_area(ena_dev
);
3410 ena_com_delete_host_info(ena_dev
);
3412 ena_release_bars(ena_dev
, pdev
);
3414 pci_disable_device(pdev
);
3416 ena_com_destroy_interrupt_moderation(ena_dev
);
3422 /* ena_suspend - PM suspend callback
3423 * @pdev: PCI device information struct
3424 * @state:power state
3426 static int ena_suspend(struct pci_dev
*pdev
, pm_message_t state
)
3428 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3430 u64_stats_update_begin(&adapter
->syncp
);
3431 adapter
->dev_stats
.suspend
++;
3432 u64_stats_update_end(&adapter
->syncp
);
3435 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
3437 "ignoring device reset request as the device is being suspended\n");
3438 clear_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
3440 ena_destroy_device(adapter
);
3445 /* ena_resume - PM resume callback
3446 * @pdev: PCI device information struct
3449 static int ena_resume(struct pci_dev
*pdev
)
3451 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3454 u64_stats_update_begin(&adapter
->syncp
);
3455 adapter
->dev_stats
.resume
++;
3456 u64_stats_update_end(&adapter
->syncp
);
3459 rc
= ena_restore_device(adapter
);
3465 static struct pci_driver ena_pci_driver
= {
3466 .name
= DRV_MODULE_NAME
,
3467 .id_table
= ena_pci_tbl
,
3469 .remove
= ena_remove
,
3471 .suspend
= ena_suspend
,
3472 .resume
= ena_resume
,
3474 .sriov_configure
= ena_sriov_configure
,
3477 static int __init
ena_init(void)
3479 pr_info("%s", version
);
3481 ena_wq
= create_singlethread_workqueue(DRV_MODULE_NAME
);
3483 pr_err("Failed to create workqueue\n");
3487 return pci_register_driver(&ena_pci_driver
);
3490 static void __exit
ena_cleanup(void)
3492 pci_unregister_driver(&ena_pci_driver
);
3495 destroy_workqueue(ena_wq
);
3500 /******************************************************************************
3501 ******************************** AENQ Handlers *******************************
3502 *****************************************************************************/
3503 /* ena_update_on_link_change:
3504 * Notify the network interface about the change in link status
3506 static void ena_update_on_link_change(void *adapter_data
,
3507 struct ena_admin_aenq_entry
*aenq_e
)
3509 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3510 struct ena_admin_aenq_link_change_desc
*aenq_desc
=
3511 (struct ena_admin_aenq_link_change_desc
*)aenq_e
;
3512 int status
= aenq_desc
->flags
&
3513 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK
;
3516 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
3517 set_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3518 if (!test_bit(ENA_FLAG_ONGOING_RESET
, &adapter
->flags
))
3519 netif_carrier_on(adapter
->netdev
);
3521 clear_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3522 netif_carrier_off(adapter
->netdev
);
3526 static void ena_keep_alive_wd(void *adapter_data
,
3527 struct ena_admin_aenq_entry
*aenq_e
)
3529 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3530 struct ena_admin_aenq_keep_alive_desc
*desc
;
3533 desc
= (struct ena_admin_aenq_keep_alive_desc
*)aenq_e
;
3534 adapter
->last_keep_alive_jiffies
= jiffies
;
3536 rx_drops
= ((u64
)desc
->rx_drops_high
<< 32) | desc
->rx_drops_low
;
3538 u64_stats_update_begin(&adapter
->syncp
);
3539 adapter
->dev_stats
.rx_drops
= rx_drops
;
3540 u64_stats_update_end(&adapter
->syncp
);
3543 static void ena_notification(void *adapter_data
,
3544 struct ena_admin_aenq_entry
*aenq_e
)
3546 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3547 struct ena_admin_ena_hw_hints
*hints
;
3549 WARN(aenq_e
->aenq_common_desc
.group
!= ENA_ADMIN_NOTIFICATION
,
3550 "Invalid group(%x) expected %x\n",
3551 aenq_e
->aenq_common_desc
.group
,
3552 ENA_ADMIN_NOTIFICATION
);
3554 switch (aenq_e
->aenq_common_desc
.syndrom
) {
3555 case ENA_ADMIN_UPDATE_HINTS
:
3556 hints
= (struct ena_admin_ena_hw_hints
*)
3557 (&aenq_e
->inline_data_w4
);
3558 ena_update_hints(adapter
, hints
);
3561 netif_err(adapter
, drv
, adapter
->netdev
,
3562 "Invalid aenq notification link state %d\n",
3563 aenq_e
->aenq_common_desc
.syndrom
);
3567 /* This handler will called for unknown event group or unimplemented handlers*/
3568 static void unimplemented_aenq_handler(void *data
,
3569 struct ena_admin_aenq_entry
*aenq_e
)
3571 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
3573 netif_err(adapter
, drv
, adapter
->netdev
,
3574 "Unknown event was received or event with unimplemented handler\n");
3577 static struct ena_aenq_handlers aenq_handlers
= {
3579 [ENA_ADMIN_LINK_CHANGE
] = ena_update_on_link_change
,
3580 [ENA_ADMIN_NOTIFICATION
] = ena_notification
,
3581 [ENA_ADMIN_KEEP_ALIVE
] = ena_keep_alive_wd
,
3583 .unimplemented_handler
= unimplemented_aenq_handler
3586 module_init(ena_init
);
3587 module_exit(ena_cleanup
);