2 * Copyright 2015 Amazon.com, Inc. or its affiliates.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #ifdef CONFIG_RFS_ACCEL
36 #include <linux/cpu_rmap.h>
37 #endif /* CONFIG_RFS_ACCEL */
38 #include <linux/ethtool.h>
39 #include <linux/if_vlan.h>
40 #include <linux/kernel.h>
41 #include <linux/module.h>
42 #include <linux/moduleparam.h>
43 #include <linux/numa.h>
44 #include <linux/pci.h>
45 #include <linux/utsname.h>
46 #include <linux/version.h>
47 #include <linux/vmalloc.h>
50 #include "ena_netdev.h"
51 #include "ena_pci_id_tbl.h"
53 static char version
[] = DEVICE_NAME
" v" DRV_MODULE_VERSION
"\n";
55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
56 MODULE_DESCRIPTION(DEVICE_NAME
);
57 MODULE_LICENSE("GPL");
58 MODULE_VERSION(DRV_MODULE_VERSION
);
60 /* Time in jiffies before concluding the transmitter is hung. */
61 #define TX_TIMEOUT (5 * HZ)
63 #define ENA_NAPI_BUDGET 64
65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
66 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
67 static int debug
= -1;
68 module_param(debug
, int, 0);
69 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
71 static struct ena_aenq_handlers aenq_handlers
;
73 static struct workqueue_struct
*ena_wq
;
75 MODULE_DEVICE_TABLE(pci
, ena_pci_tbl
);
77 static int ena_rss_init_default(struct ena_adapter
*adapter
);
79 static void ena_tx_timeout(struct net_device
*dev
)
81 struct ena_adapter
*adapter
= netdev_priv(dev
);
83 /* Change the state of the device to trigger reset
84 * Check that we are not in the middle or a trigger already
87 if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
90 u64_stats_update_begin(&adapter
->syncp
);
91 adapter
->dev_stats
.tx_timeout
++;
92 u64_stats_update_end(&adapter
->syncp
);
94 netif_err(adapter
, tx_err
, dev
, "Transmit time out\n");
97 static void update_rx_ring_mtu(struct ena_adapter
*adapter
, int mtu
)
101 for (i
= 0; i
< adapter
->num_queues
; i
++)
102 adapter
->rx_ring
[i
].mtu
= mtu
;
105 static int ena_change_mtu(struct net_device
*dev
, int new_mtu
)
107 struct ena_adapter
*adapter
= netdev_priv(dev
);
110 ret
= ena_com_set_dev_mtu(adapter
->ena_dev
, new_mtu
);
112 netif_dbg(adapter
, drv
, dev
, "set MTU to %d\n", new_mtu
);
113 update_rx_ring_mtu(adapter
, new_mtu
);
116 netif_err(adapter
, drv
, dev
, "Failed to set MTU to %d\n",
123 static int ena_init_rx_cpu_rmap(struct ena_adapter
*adapter
)
125 #ifdef CONFIG_RFS_ACCEL
129 adapter
->netdev
->rx_cpu_rmap
= alloc_irq_cpu_rmap(adapter
->num_queues
);
130 if (!adapter
->netdev
->rx_cpu_rmap
)
132 for (i
= 0; i
< adapter
->num_queues
; i
++) {
133 int irq_idx
= ENA_IO_IRQ_IDX(i
);
135 rc
= irq_cpu_rmap_add(adapter
->netdev
->rx_cpu_rmap
,
136 adapter
->msix_entries
[irq_idx
].vector
);
138 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
139 adapter
->netdev
->rx_cpu_rmap
= NULL
;
143 #endif /* CONFIG_RFS_ACCEL */
147 static void ena_init_io_rings_common(struct ena_adapter
*adapter
,
148 struct ena_ring
*ring
, u16 qid
)
151 ring
->pdev
= adapter
->pdev
;
152 ring
->dev
= &adapter
->pdev
->dev
;
153 ring
->netdev
= adapter
->netdev
;
154 ring
->napi
= &adapter
->ena_napi
[qid
].napi
;
155 ring
->adapter
= adapter
;
156 ring
->ena_dev
= adapter
->ena_dev
;
157 ring
->per_napi_packets
= 0;
158 ring
->per_napi_bytes
= 0;
160 u64_stats_init(&ring
->syncp
);
163 static void ena_init_io_rings(struct ena_adapter
*adapter
)
165 struct ena_com_dev
*ena_dev
;
166 struct ena_ring
*txr
, *rxr
;
169 ena_dev
= adapter
->ena_dev
;
171 for (i
= 0; i
< adapter
->num_queues
; i
++) {
172 txr
= &adapter
->tx_ring
[i
];
173 rxr
= &adapter
->rx_ring
[i
];
175 /* TX/RX common ring state */
176 ena_init_io_rings_common(adapter
, txr
, i
);
177 ena_init_io_rings_common(adapter
, rxr
, i
);
179 /* TX specific ring state */
180 txr
->ring_size
= adapter
->tx_ring_size
;
181 txr
->tx_max_header_size
= ena_dev
->tx_max_header_size
;
182 txr
->tx_mem_queue_type
= ena_dev
->tx_mem_queue_type
;
183 txr
->sgl_size
= adapter
->max_tx_sgl_size
;
184 txr
->smoothed_interval
=
185 ena_com_get_nonadaptive_moderation_interval_tx(ena_dev
);
187 /* RX specific ring state */
188 rxr
->ring_size
= adapter
->rx_ring_size
;
189 rxr
->rx_copybreak
= adapter
->rx_copybreak
;
190 rxr
->sgl_size
= adapter
->max_rx_sgl_size
;
191 rxr
->smoothed_interval
=
192 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev
);
196 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
197 * @adapter: network interface device structure
200 * Return 0 on success, negative on failure
202 static int ena_setup_tx_resources(struct ena_adapter
*adapter
, int qid
)
204 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
205 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
208 if (tx_ring
->tx_buffer_info
) {
209 netif_err(adapter
, ifup
,
210 adapter
->netdev
, "tx_buffer_info info is not NULL");
214 size
= sizeof(struct ena_tx_buffer
) * tx_ring
->ring_size
;
215 node
= cpu_to_node(ena_irq
->cpu
);
217 tx_ring
->tx_buffer_info
= vzalloc_node(size
, node
);
218 if (!tx_ring
->tx_buffer_info
) {
219 tx_ring
->tx_buffer_info
= vzalloc(size
);
220 if (!tx_ring
->tx_buffer_info
)
224 size
= sizeof(u16
) * tx_ring
->ring_size
;
225 tx_ring
->free_tx_ids
= vzalloc_node(size
, node
);
226 if (!tx_ring
->free_tx_ids
) {
227 tx_ring
->free_tx_ids
= vzalloc(size
);
228 if (!tx_ring
->free_tx_ids
) {
229 vfree(tx_ring
->tx_buffer_info
);
234 /* Req id ring for TX out of order completions */
235 for (i
= 0; i
< tx_ring
->ring_size
; i
++)
236 tx_ring
->free_tx_ids
[i
] = i
;
238 /* Reset tx statistics */
239 memset(&tx_ring
->tx_stats
, 0x0, sizeof(tx_ring
->tx_stats
));
241 tx_ring
->next_to_use
= 0;
242 tx_ring
->next_to_clean
= 0;
243 tx_ring
->cpu
= ena_irq
->cpu
;
247 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
248 * @adapter: network interface device structure
251 * Free all transmit software resources
253 static void ena_free_tx_resources(struct ena_adapter
*adapter
, int qid
)
255 struct ena_ring
*tx_ring
= &adapter
->tx_ring
[qid
];
257 vfree(tx_ring
->tx_buffer_info
);
258 tx_ring
->tx_buffer_info
= NULL
;
260 vfree(tx_ring
->free_tx_ids
);
261 tx_ring
->free_tx_ids
= NULL
;
264 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
265 * @adapter: private structure
267 * Return 0 on success, negative on failure
269 static int ena_setup_all_tx_resources(struct ena_adapter
*adapter
)
273 for (i
= 0; i
< adapter
->num_queues
; i
++) {
274 rc
= ena_setup_tx_resources(adapter
, i
);
283 netif_err(adapter
, ifup
, adapter
->netdev
,
284 "Tx queue %d: allocation failed\n", i
);
286 /* rewind the index freeing the rings as we go */
288 ena_free_tx_resources(adapter
, i
);
292 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
293 * @adapter: board private structure
295 * Free all transmit software resources
297 static void ena_free_all_io_tx_resources(struct ena_adapter
*adapter
)
301 for (i
= 0; i
< adapter
->num_queues
; i
++)
302 ena_free_tx_resources(adapter
, i
);
305 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
306 * @adapter: network interface device structure
309 * Returns 0 on success, negative on failure
311 static int ena_setup_rx_resources(struct ena_adapter
*adapter
,
314 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
315 struct ena_irq
*ena_irq
= &adapter
->irq_tbl
[ENA_IO_IRQ_IDX(qid
)];
318 if (rx_ring
->rx_buffer_info
) {
319 netif_err(adapter
, ifup
, adapter
->netdev
,
320 "rx_buffer_info is not NULL");
324 /* alloc extra element so in rx path
325 * we can always prefetch rx_info + 1
327 size
= sizeof(struct ena_rx_buffer
) * (rx_ring
->ring_size
+ 1);
328 node
= cpu_to_node(ena_irq
->cpu
);
330 rx_ring
->rx_buffer_info
= vzalloc_node(size
, node
);
331 if (!rx_ring
->rx_buffer_info
) {
332 rx_ring
->rx_buffer_info
= vzalloc(size
);
333 if (!rx_ring
->rx_buffer_info
)
337 /* Reset rx statistics */
338 memset(&rx_ring
->rx_stats
, 0x0, sizeof(rx_ring
->rx_stats
));
340 rx_ring
->next_to_clean
= 0;
341 rx_ring
->next_to_use
= 0;
342 rx_ring
->cpu
= ena_irq
->cpu
;
347 /* ena_free_rx_resources - Free I/O Rx Resources
348 * @adapter: network interface device structure
351 * Free all receive software resources
353 static void ena_free_rx_resources(struct ena_adapter
*adapter
,
356 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
358 vfree(rx_ring
->rx_buffer_info
);
359 rx_ring
->rx_buffer_info
= NULL
;
362 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
363 * @adapter: board private structure
365 * Return 0 on success, negative on failure
367 static int ena_setup_all_rx_resources(struct ena_adapter
*adapter
)
371 for (i
= 0; i
< adapter
->num_queues
; i
++) {
372 rc
= ena_setup_rx_resources(adapter
, i
);
381 netif_err(adapter
, ifup
, adapter
->netdev
,
382 "Rx queue %d: allocation failed\n", i
);
384 /* rewind the index freeing the rings as we go */
386 ena_free_rx_resources(adapter
, i
);
390 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
391 * @adapter: board private structure
393 * Free all receive software resources
395 static void ena_free_all_io_rx_resources(struct ena_adapter
*adapter
)
399 for (i
= 0; i
< adapter
->num_queues
; i
++)
400 ena_free_rx_resources(adapter
, i
);
403 static inline int ena_alloc_rx_page(struct ena_ring
*rx_ring
,
404 struct ena_rx_buffer
*rx_info
, gfp_t gfp
)
406 struct ena_com_buf
*ena_buf
;
410 /* if previous allocated page is not used */
411 if (unlikely(rx_info
->page
))
414 page
= alloc_page(gfp
);
415 if (unlikely(!page
)) {
416 u64_stats_update_begin(&rx_ring
->syncp
);
417 rx_ring
->rx_stats
.page_alloc_fail
++;
418 u64_stats_update_end(&rx_ring
->syncp
);
422 dma
= dma_map_page(rx_ring
->dev
, page
, 0, PAGE_SIZE
,
424 if (unlikely(dma_mapping_error(rx_ring
->dev
, dma
))) {
425 u64_stats_update_begin(&rx_ring
->syncp
);
426 rx_ring
->rx_stats
.dma_mapping_err
++;
427 u64_stats_update_end(&rx_ring
->syncp
);
432 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
433 "alloc page %p, rx_info %p\n", page
, rx_info
);
435 rx_info
->page
= page
;
436 rx_info
->page_offset
= 0;
437 ena_buf
= &rx_info
->ena_buf
;
438 ena_buf
->paddr
= dma
;
439 ena_buf
->len
= PAGE_SIZE
;
444 static void ena_free_rx_page(struct ena_ring
*rx_ring
,
445 struct ena_rx_buffer
*rx_info
)
447 struct page
*page
= rx_info
->page
;
448 struct ena_com_buf
*ena_buf
= &rx_info
->ena_buf
;
450 if (unlikely(!page
)) {
451 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
452 "Trying to free unallocated buffer\n");
456 dma_unmap_page(rx_ring
->dev
, ena_buf
->paddr
, PAGE_SIZE
,
460 rx_info
->page
= NULL
;
463 static int ena_refill_rx_bufs(struct ena_ring
*rx_ring
, u32 num
)
469 next_to_use
= rx_ring
->next_to_use
;
471 for (i
= 0; i
< num
; i
++) {
472 struct ena_rx_buffer
*rx_info
=
473 &rx_ring
->rx_buffer_info
[next_to_use
];
475 rc
= ena_alloc_rx_page(rx_ring
, rx_info
,
476 __GFP_COLD
| GFP_ATOMIC
| __GFP_COMP
);
477 if (unlikely(rc
< 0)) {
478 netif_warn(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
479 "failed to alloc buffer for rx queue %d\n",
483 rc
= ena_com_add_single_rx_desc(rx_ring
->ena_com_io_sq
,
487 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
488 "failed to add buffer for rx queue %d\n",
492 next_to_use
= ENA_RX_RING_IDX_NEXT(next_to_use
,
496 if (unlikely(i
< num
)) {
497 u64_stats_update_begin(&rx_ring
->syncp
);
498 rx_ring
->rx_stats
.refil_partial
++;
499 u64_stats_update_end(&rx_ring
->syncp
);
500 netdev_warn(rx_ring
->netdev
,
501 "refilled rx qid %d with only %d buffers (from %d)\n",
502 rx_ring
->qid
, i
, num
);
506 /* Add memory barrier to make sure the desc were written before
510 ena_com_write_sq_doorbell(rx_ring
->ena_com_io_sq
);
513 rx_ring
->next_to_use
= next_to_use
;
518 static void ena_free_rx_bufs(struct ena_adapter
*adapter
,
521 struct ena_ring
*rx_ring
= &adapter
->rx_ring
[qid
];
524 for (i
= 0; i
< rx_ring
->ring_size
; i
++) {
525 struct ena_rx_buffer
*rx_info
= &rx_ring
->rx_buffer_info
[i
];
528 ena_free_rx_page(rx_ring
, rx_info
);
532 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
533 * @adapter: board private structure
536 static void ena_refill_all_rx_bufs(struct ena_adapter
*adapter
)
538 struct ena_ring
*rx_ring
;
541 for (i
= 0; i
< adapter
->num_queues
; i
++) {
542 rx_ring
= &adapter
->rx_ring
[i
];
543 bufs_num
= rx_ring
->ring_size
- 1;
544 rc
= ena_refill_rx_bufs(rx_ring
, bufs_num
);
546 if (unlikely(rc
!= bufs_num
))
547 netif_warn(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
548 "refilling Queue %d failed. allocated %d buffers from: %d\n",
553 static void ena_free_all_rx_bufs(struct ena_adapter
*adapter
)
557 for (i
= 0; i
< adapter
->num_queues
; i
++)
558 ena_free_rx_bufs(adapter
, i
);
561 /* ena_free_tx_bufs - Free Tx Buffers per Queue
562 * @tx_ring: TX ring for which buffers be freed
564 static void ena_free_tx_bufs(struct ena_ring
*tx_ring
)
568 for (i
= 0; i
< tx_ring
->ring_size
; i
++) {
569 struct ena_tx_buffer
*tx_info
= &tx_ring
->tx_buffer_info
[i
];
570 struct ena_com_buf
*ena_buf
;
577 netdev_notice(tx_ring
->netdev
,
578 "free uncompleted tx skb qid %d idx 0x%x\n",
581 ena_buf
= tx_info
->bufs
;
582 dma_unmap_single(tx_ring
->dev
,
587 /* unmap remaining mapped pages */
588 nr_frags
= tx_info
->num_of_bufs
- 1;
589 for (j
= 0; j
< nr_frags
; j
++) {
591 dma_unmap_page(tx_ring
->dev
,
597 dev_kfree_skb_any(tx_info
->skb
);
599 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring
->netdev
,
603 static void ena_free_all_tx_bufs(struct ena_adapter
*adapter
)
605 struct ena_ring
*tx_ring
;
608 for (i
= 0; i
< adapter
->num_queues
; i
++) {
609 tx_ring
= &adapter
->tx_ring
[i
];
610 ena_free_tx_bufs(tx_ring
);
614 static void ena_destroy_all_tx_queues(struct ena_adapter
*adapter
)
619 for (i
= 0; i
< adapter
->num_queues
; i
++) {
620 ena_qid
= ENA_IO_TXQ_IDX(i
);
621 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
625 static void ena_destroy_all_rx_queues(struct ena_adapter
*adapter
)
630 for (i
= 0; i
< adapter
->num_queues
; i
++) {
631 ena_qid
= ENA_IO_RXQ_IDX(i
);
632 ena_com_destroy_io_queue(adapter
->ena_dev
, ena_qid
);
636 static void ena_destroy_all_io_queues(struct ena_adapter
*adapter
)
638 ena_destroy_all_tx_queues(adapter
);
639 ena_destroy_all_rx_queues(adapter
);
642 static int validate_tx_req_id(struct ena_ring
*tx_ring
, u16 req_id
)
644 struct ena_tx_buffer
*tx_info
= NULL
;
646 if (likely(req_id
< tx_ring
->ring_size
)) {
647 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
648 if (likely(tx_info
->skb
))
653 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
654 "tx_info doesn't have valid skb\n");
656 netif_err(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
657 "Invalid req_id: %hu\n", req_id
);
659 u64_stats_update_begin(&tx_ring
->syncp
);
660 tx_ring
->tx_stats
.bad_req_id
++;
661 u64_stats_update_end(&tx_ring
->syncp
);
663 /* Trigger device reset */
664 set_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
);
668 static int ena_clean_tx_irq(struct ena_ring
*tx_ring
, u32 budget
)
670 struct netdev_queue
*txq
;
679 next_to_clean
= tx_ring
->next_to_clean
;
680 txq
= netdev_get_tx_queue(tx_ring
->netdev
, tx_ring
->qid
);
682 while (tx_pkts
< budget
) {
683 struct ena_tx_buffer
*tx_info
;
685 struct ena_com_buf
*ena_buf
;
688 rc
= ena_com_tx_comp_req_id_get(tx_ring
->ena_com_io_cq
,
693 rc
= validate_tx_req_id(tx_ring
, req_id
);
697 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
700 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
704 tx_info
->last_jiffies
= 0;
706 if (likely(tx_info
->num_of_bufs
!= 0)) {
707 ena_buf
= tx_info
->bufs
;
709 dma_unmap_single(tx_ring
->dev
,
710 dma_unmap_addr(ena_buf
, paddr
),
711 dma_unmap_len(ena_buf
, len
),
714 /* unmap remaining mapped pages */
715 nr_frags
= tx_info
->num_of_bufs
- 1;
716 for (i
= 0; i
< nr_frags
; i
++) {
718 dma_unmap_page(tx_ring
->dev
,
719 dma_unmap_addr(ena_buf
, paddr
),
720 dma_unmap_len(ena_buf
, len
),
725 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
726 "tx_poll: q %d skb %p completed\n", tx_ring
->qid
,
729 tx_bytes
+= skb
->len
;
732 total_done
+= tx_info
->tx_descs
;
734 tx_ring
->free_tx_ids
[next_to_clean
] = req_id
;
735 next_to_clean
= ENA_TX_RING_IDX_NEXT(next_to_clean
,
739 tx_ring
->next_to_clean
= next_to_clean
;
740 ena_com_comp_ack(tx_ring
->ena_com_io_sq
, total_done
);
741 ena_com_update_dev_comp_head(tx_ring
->ena_com_io_cq
);
743 netdev_tx_completed_queue(txq
, tx_pkts
, tx_bytes
);
745 netif_dbg(tx_ring
->adapter
, tx_done
, tx_ring
->netdev
,
746 "tx_poll: q %d done. total pkts: %d\n",
747 tx_ring
->qid
, tx_pkts
);
749 /* need to make the rings circular update visible to
750 * ena_start_xmit() before checking for netif_queue_stopped().
754 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
755 ENA_TX_WAKEUP_THRESH
;
756 if (unlikely(netif_tx_queue_stopped(txq
) && above_thresh
)) {
757 __netif_tx_lock(txq
, smp_processor_id());
758 above_thresh
= ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) >
759 ENA_TX_WAKEUP_THRESH
;
760 if (netif_tx_queue_stopped(txq
) && above_thresh
) {
761 netif_tx_wake_queue(txq
);
762 u64_stats_update_begin(&tx_ring
->syncp
);
763 tx_ring
->tx_stats
.queue_wakeup
++;
764 u64_stats_update_end(&tx_ring
->syncp
);
766 __netif_tx_unlock(txq
);
769 tx_ring
->per_napi_bytes
+= tx_bytes
;
770 tx_ring
->per_napi_packets
+= tx_pkts
;
775 static struct sk_buff
*ena_rx_skb(struct ena_ring
*rx_ring
,
776 struct ena_com_rx_buf_info
*ena_bufs
,
781 struct ena_rx_buffer
*rx_info
=
782 &rx_ring
->rx_buffer_info
[*next_to_clean
];
787 len
= ena_bufs
[0].len
;
788 if (unlikely(!rx_info
->page
)) {
789 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
794 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
795 "rx_info %p page %p\n",
796 rx_info
, rx_info
->page
);
798 /* save virt address of first buffer */
799 va
= page_address(rx_info
->page
) + rx_info
->page_offset
;
800 prefetch(va
+ NET_IP_ALIGN
);
802 if (len
<= rx_ring
->rx_copybreak
) {
803 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
804 rx_ring
->rx_copybreak
);
805 if (unlikely(!skb
)) {
806 u64_stats_update_begin(&rx_ring
->syncp
);
807 rx_ring
->rx_stats
.skb_alloc_fail
++;
808 u64_stats_update_end(&rx_ring
->syncp
);
809 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
810 "Failed to allocate skb\n");
814 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
815 "rx allocated small packet. len %d. data_len %d\n",
816 skb
->len
, skb
->data_len
);
818 /* sync this buffer for CPU use */
819 dma_sync_single_for_cpu(rx_ring
->dev
,
820 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
823 skb_copy_to_linear_data(skb
, va
, len
);
824 dma_sync_single_for_device(rx_ring
->dev
,
825 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
830 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
831 *next_to_clean
= ENA_RX_RING_IDX_ADD(*next_to_clean
, descs
,
836 skb
= napi_get_frags(rx_ring
->napi
);
837 if (unlikely(!skb
)) {
838 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
839 "Failed allocating skb\n");
840 u64_stats_update_begin(&rx_ring
->syncp
);
841 rx_ring
->rx_stats
.skb_alloc_fail
++;
842 u64_stats_update_end(&rx_ring
->syncp
);
847 dma_unmap_page(rx_ring
->dev
,
848 dma_unmap_addr(&rx_info
->ena_buf
, paddr
),
849 PAGE_SIZE
, DMA_FROM_DEVICE
);
851 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, rx_info
->page
,
852 rx_info
->page_offset
, len
, PAGE_SIZE
);
854 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
855 "rx skb updated. len %d. data_len %d\n",
856 skb
->len
, skb
->data_len
);
858 rx_info
->page
= NULL
;
860 ENA_RX_RING_IDX_NEXT(*next_to_clean
,
862 if (likely(--descs
== 0))
864 rx_info
= &rx_ring
->rx_buffer_info
[*next_to_clean
];
865 len
= ena_bufs
[++buf
].len
;
871 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
872 * @adapter: structure containing adapter specific data
873 * @ena_rx_ctx: received packet context/metadata
874 * @skb: skb currently being received and modified
876 static inline void ena_rx_checksum(struct ena_ring
*rx_ring
,
877 struct ena_com_rx_ctx
*ena_rx_ctx
,
880 /* Rx csum disabled */
881 if (unlikely(!(rx_ring
->netdev
->features
& NETIF_F_RXCSUM
))) {
882 skb
->ip_summed
= CHECKSUM_NONE
;
886 /* For fragmented packets the checksum isn't valid */
887 if (ena_rx_ctx
->frag
) {
888 skb
->ip_summed
= CHECKSUM_NONE
;
892 /* if IP and error */
893 if (unlikely((ena_rx_ctx
->l3_proto
== ENA_ETH_IO_L3_PROTO_IPV4
) &&
894 (ena_rx_ctx
->l3_csum_err
))) {
895 /* ipv4 checksum error */
896 skb
->ip_summed
= CHECKSUM_NONE
;
897 u64_stats_update_begin(&rx_ring
->syncp
);
898 rx_ring
->rx_stats
.bad_csum
++;
899 u64_stats_update_end(&rx_ring
->syncp
);
900 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
901 "RX IPv4 header checksum error\n");
906 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
907 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
))) {
908 if (unlikely(ena_rx_ctx
->l4_csum_err
)) {
909 /* TCP/UDP checksum error */
910 u64_stats_update_begin(&rx_ring
->syncp
);
911 rx_ring
->rx_stats
.bad_csum
++;
912 u64_stats_update_end(&rx_ring
->syncp
);
913 netif_err(rx_ring
->adapter
, rx_err
, rx_ring
->netdev
,
914 "RX L4 checksum error\n");
915 skb
->ip_summed
= CHECKSUM_NONE
;
919 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
923 static void ena_set_rx_hash(struct ena_ring
*rx_ring
,
924 struct ena_com_rx_ctx
*ena_rx_ctx
,
927 enum pkt_hash_types hash_type
;
929 if (likely(rx_ring
->netdev
->features
& NETIF_F_RXHASH
)) {
930 if (likely((ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_TCP
) ||
931 (ena_rx_ctx
->l4_proto
== ENA_ETH_IO_L4_PROTO_UDP
)))
933 hash_type
= PKT_HASH_TYPE_L4
;
935 hash_type
= PKT_HASH_TYPE_NONE
;
937 /* Override hash type if the packet is fragmented */
938 if (ena_rx_ctx
->frag
)
939 hash_type
= PKT_HASH_TYPE_NONE
;
941 skb_set_hash(skb
, ena_rx_ctx
->hash
, hash_type
);
945 /* ena_clean_rx_irq - Cleanup RX irq
946 * @rx_ring: RX ring to clean
947 * @napi: napi handler
948 * @budget: how many packets driver is allowed to clean
950 * Returns the number of cleaned buffers.
952 static int ena_clean_rx_irq(struct ena_ring
*rx_ring
, struct napi_struct
*napi
,
955 u16 next_to_clean
= rx_ring
->next_to_clean
;
956 u32 res_budget
, work_done
;
958 struct ena_com_rx_ctx ena_rx_ctx
;
959 struct ena_adapter
*adapter
;
962 int refill_threshold
;
965 int rx_copybreak_pkt
= 0;
967 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
968 "%s qid %d\n", __func__
, rx_ring
->qid
);
972 ena_rx_ctx
.ena_bufs
= rx_ring
->ena_bufs
;
973 ena_rx_ctx
.max_bufs
= rx_ring
->sgl_size
;
974 ena_rx_ctx
.descs
= 0;
975 rc
= ena_com_rx_pkt(rx_ring
->ena_com_io_cq
,
976 rx_ring
->ena_com_io_sq
,
981 if (unlikely(ena_rx_ctx
.descs
== 0))
984 netif_dbg(rx_ring
->adapter
, rx_status
, rx_ring
->netdev
,
985 "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
986 rx_ring
->qid
, ena_rx_ctx
.descs
, ena_rx_ctx
.l3_proto
,
987 ena_rx_ctx
.l4_proto
, ena_rx_ctx
.hash
);
989 /* allocate skb and fill it */
990 skb
= ena_rx_skb(rx_ring
, rx_ring
->ena_bufs
, ena_rx_ctx
.descs
,
993 /* exit if we failed to retrieve a buffer */
994 if (unlikely(!skb
)) {
995 next_to_clean
= ENA_RX_RING_IDX_ADD(next_to_clean
,
1001 ena_rx_checksum(rx_ring
, &ena_rx_ctx
, skb
);
1003 ena_set_rx_hash(rx_ring
, &ena_rx_ctx
, skb
);
1005 skb_record_rx_queue(skb
, rx_ring
->qid
);
1007 if (rx_ring
->ena_bufs
[0].len
<= rx_ring
->rx_copybreak
) {
1008 total_len
+= rx_ring
->ena_bufs
[0].len
;
1010 napi_gro_receive(napi
, skb
);
1012 total_len
+= skb
->len
;
1013 napi_gro_frags(napi
);
1017 } while (likely(res_budget
));
1019 work_done
= budget
- res_budget
;
1020 rx_ring
->per_napi_bytes
+= total_len
;
1021 rx_ring
->per_napi_packets
+= work_done
;
1022 u64_stats_update_begin(&rx_ring
->syncp
);
1023 rx_ring
->rx_stats
.bytes
+= total_len
;
1024 rx_ring
->rx_stats
.cnt
+= work_done
;
1025 rx_ring
->rx_stats
.rx_copybreak_pkt
+= rx_copybreak_pkt
;
1026 u64_stats_update_end(&rx_ring
->syncp
);
1028 rx_ring
->next_to_clean
= next_to_clean
;
1030 refill_required
= ena_com_sq_empty_space(rx_ring
->ena_com_io_sq
);
1031 refill_threshold
= rx_ring
->ring_size
/ ENA_RX_REFILL_THRESH_DIVIDER
;
1033 /* Optimization, try to batch new rx buffers */
1034 if (refill_required
> refill_threshold
) {
1035 ena_com_update_dev_comp_head(rx_ring
->ena_com_io_cq
);
1036 ena_refill_rx_bufs(rx_ring
, refill_required
);
1042 adapter
= netdev_priv(rx_ring
->netdev
);
1044 u64_stats_update_begin(&rx_ring
->syncp
);
1045 rx_ring
->rx_stats
.bad_desc_num
++;
1046 u64_stats_update_end(&rx_ring
->syncp
);
1048 /* Too many desc from the device. Trigger reset */
1049 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
1054 inline void ena_adjust_intr_moderation(struct ena_ring
*rx_ring
,
1055 struct ena_ring
*tx_ring
)
1057 /* We apply adaptive moderation on Rx path only.
1058 * Tx uses static interrupt moderation.
1060 ena_com_calculate_interrupt_delay(rx_ring
->ena_dev
,
1061 rx_ring
->per_napi_packets
,
1062 rx_ring
->per_napi_bytes
,
1063 &rx_ring
->smoothed_interval
,
1064 &rx_ring
->moder_tbl_idx
);
1066 /* Reset per napi packets/bytes */
1067 tx_ring
->per_napi_packets
= 0;
1068 tx_ring
->per_napi_bytes
= 0;
1069 rx_ring
->per_napi_packets
= 0;
1070 rx_ring
->per_napi_bytes
= 0;
1073 static inline void ena_update_ring_numa_node(struct ena_ring
*tx_ring
,
1074 struct ena_ring
*rx_ring
)
1076 int cpu
= get_cpu();
1079 /* Check only one ring since the 2 rings are running on the same cpu */
1080 if (likely(tx_ring
->cpu
== cpu
))
1083 numa_node
= cpu_to_node(cpu
);
1086 if (numa_node
!= NUMA_NO_NODE
) {
1087 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, numa_node
);
1088 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, numa_node
);
1099 static int ena_io_poll(struct napi_struct
*napi
, int budget
)
1101 struct ena_napi
*ena_napi
= container_of(napi
, struct ena_napi
, napi
);
1102 struct ena_ring
*tx_ring
, *rx_ring
;
1103 struct ena_eth_io_intr_reg intr_reg
;
1108 int napi_comp_call
= 0;
1111 tx_ring
= ena_napi
->tx_ring
;
1112 rx_ring
= ena_napi
->rx_ring
;
1114 tx_budget
= tx_ring
->ring_size
/ ENA_TX_POLL_BUDGET_DIVIDER
;
1116 if (!test_bit(ENA_FLAG_DEV_UP
, &tx_ring
->adapter
->flags
) ||
1117 test_bit(ENA_FLAG_TRIGGER_RESET
, &tx_ring
->adapter
->flags
)) {
1118 napi_complete_done(napi
, 0);
1122 tx_work_done
= ena_clean_tx_irq(tx_ring
, tx_budget
);
1123 rx_work_done
= ena_clean_rx_irq(rx_ring
, napi
, budget
);
1125 if ((budget
> rx_work_done
) && (tx_budget
> tx_work_done
)) {
1126 napi_complete_done(napi
, rx_work_done
);
1129 /* Tx and Rx share the same interrupt vector */
1130 if (ena_com_get_adaptive_moderation_enabled(rx_ring
->ena_dev
))
1131 ena_adjust_intr_moderation(rx_ring
, tx_ring
);
1133 /* Update intr register: rx intr delay, tx intr delay and
1136 ena_com_update_intr_reg(&intr_reg
,
1137 rx_ring
->smoothed_interval
,
1138 tx_ring
->smoothed_interval
,
1141 /* It is a shared MSI-X. Tx and Rx CQ have pointer to it.
1142 * So we use one of them to reach the intr reg
1144 ena_com_unmask_intr(rx_ring
->ena_com_io_cq
, &intr_reg
);
1146 ena_update_ring_numa_node(tx_ring
, rx_ring
);
1153 u64_stats_update_begin(&tx_ring
->syncp
);
1154 tx_ring
->tx_stats
.napi_comp
+= napi_comp_call
;
1155 tx_ring
->tx_stats
.tx_poll
++;
1156 u64_stats_update_end(&tx_ring
->syncp
);
1161 static irqreturn_t
ena_intr_msix_mgmnt(int irq
, void *data
)
1163 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
1165 ena_com_admin_q_comp_intr_handler(adapter
->ena_dev
);
1167 /* Don't call the aenq handler before probe is done */
1168 if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
)))
1169 ena_com_aenq_intr_handler(adapter
->ena_dev
, data
);
1174 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1175 * @irq: interrupt number
1176 * @data: pointer to a network interface private napi device structure
1178 static irqreturn_t
ena_intr_msix_io(int irq
, void *data
)
1180 struct ena_napi
*ena_napi
= data
;
1182 napi_schedule(&ena_napi
->napi
);
1187 static int ena_enable_msix(struct ena_adapter
*adapter
, int num_queues
)
1189 int i
, msix_vecs
, rc
;
1191 if (test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1192 netif_err(adapter
, probe
, adapter
->netdev
,
1193 "Error, MSI-X is already enabled\n");
1197 /* Reserved the max msix vectors we might need */
1198 msix_vecs
= ENA_MAX_MSIX_VEC(num_queues
);
1200 netif_dbg(adapter
, probe
, adapter
->netdev
,
1201 "trying to enable MSI-X, vectors %d\n", msix_vecs
);
1203 adapter
->msix_entries
= vzalloc(msix_vecs
* sizeof(struct msix_entry
));
1205 if (!adapter
->msix_entries
)
1208 for (i
= 0; i
< msix_vecs
; i
++)
1209 adapter
->msix_entries
[i
].entry
= i
;
1211 rc
= pci_enable_msix(adapter
->pdev
, adapter
->msix_entries
, msix_vecs
);
1213 netif_err(adapter
, probe
, adapter
->netdev
,
1214 "Failed to enable MSI-X, vectors %d rc %d\n",
1219 netif_dbg(adapter
, probe
, adapter
->netdev
, "enable MSI-X, vectors %d\n",
1222 if (msix_vecs
>= 1) {
1223 if (ena_init_rx_cpu_rmap(adapter
))
1224 netif_warn(adapter
, probe
, adapter
->netdev
,
1225 "Failed to map IRQs to CPUs\n");
1228 adapter
->msix_vecs
= msix_vecs
;
1229 set_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
);
1234 static void ena_setup_mgmnt_intr(struct ena_adapter
*adapter
)
1238 snprintf(adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].name
,
1239 ENA_IRQNAME_SIZE
, "ena-mgmnt@pci:%s",
1240 pci_name(adapter
->pdev
));
1241 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].handler
=
1242 ena_intr_msix_mgmnt
;
1243 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].data
= adapter
;
1244 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].vector
=
1245 adapter
->msix_entries
[ENA_MGMNT_IRQ_IDX
].vector
;
1246 cpu
= cpumask_first(cpu_online_mask
);
1247 adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].cpu
= cpu
;
1248 cpumask_set_cpu(cpu
,
1249 &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
].affinity_hint_mask
);
1252 static void ena_setup_io_intr(struct ena_adapter
*adapter
)
1254 struct net_device
*netdev
;
1255 int irq_idx
, i
, cpu
;
1257 netdev
= adapter
->netdev
;
1259 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1260 irq_idx
= ENA_IO_IRQ_IDX(i
);
1261 cpu
= i
% num_online_cpus();
1263 snprintf(adapter
->irq_tbl
[irq_idx
].name
, ENA_IRQNAME_SIZE
,
1264 "%s-Tx-Rx-%d", netdev
->name
, i
);
1265 adapter
->irq_tbl
[irq_idx
].handler
= ena_intr_msix_io
;
1266 adapter
->irq_tbl
[irq_idx
].data
= &adapter
->ena_napi
[i
];
1267 adapter
->irq_tbl
[irq_idx
].vector
=
1268 adapter
->msix_entries
[irq_idx
].vector
;
1269 adapter
->irq_tbl
[irq_idx
].cpu
= cpu
;
1271 cpumask_set_cpu(cpu
,
1272 &adapter
->irq_tbl
[irq_idx
].affinity_hint_mask
);
1276 static int ena_request_mgmnt_irq(struct ena_adapter
*adapter
)
1278 unsigned long flags
= 0;
1279 struct ena_irq
*irq
;
1282 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1283 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1286 netif_err(adapter
, probe
, adapter
->netdev
,
1287 "failed to request admin irq\n");
1291 netif_dbg(adapter
, probe
, adapter
->netdev
,
1292 "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1293 irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1295 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1300 static int ena_request_io_irq(struct ena_adapter
*adapter
)
1302 unsigned long flags
= 0;
1303 struct ena_irq
*irq
;
1306 if (!test_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
)) {
1307 netif_err(adapter
, ifup
, adapter
->netdev
,
1308 "Failed to request I/O IRQ: MSI-X is not enabled\n");
1312 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1313 irq
= &adapter
->irq_tbl
[i
];
1314 rc
= request_irq(irq
->vector
, irq
->handler
, flags
, irq
->name
,
1317 netif_err(adapter
, ifup
, adapter
->netdev
,
1318 "Failed to request I/O IRQ. index %d rc %d\n",
1323 netif_dbg(adapter
, ifup
, adapter
->netdev
,
1324 "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1325 i
, irq
->affinity_hint_mask
.bits
[0], irq
->vector
);
1327 irq_set_affinity_hint(irq
->vector
, &irq
->affinity_hint_mask
);
1333 for (k
= ENA_IO_IRQ_FIRST_IDX
; k
< i
; k
++) {
1334 irq
= &adapter
->irq_tbl
[k
];
1335 free_irq(irq
->vector
, irq
->data
);
1341 static void ena_free_mgmnt_irq(struct ena_adapter
*adapter
)
1343 struct ena_irq
*irq
;
1345 irq
= &adapter
->irq_tbl
[ENA_MGMNT_IRQ_IDX
];
1346 synchronize_irq(irq
->vector
);
1347 irq_set_affinity_hint(irq
->vector
, NULL
);
1348 free_irq(irq
->vector
, irq
->data
);
1351 static void ena_free_io_irq(struct ena_adapter
*adapter
)
1353 struct ena_irq
*irq
;
1356 #ifdef CONFIG_RFS_ACCEL
1357 if (adapter
->msix_vecs
>= 1) {
1358 free_irq_cpu_rmap(adapter
->netdev
->rx_cpu_rmap
);
1359 adapter
->netdev
->rx_cpu_rmap
= NULL
;
1361 #endif /* CONFIG_RFS_ACCEL */
1363 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++) {
1364 irq
= &adapter
->irq_tbl
[i
];
1365 irq_set_affinity_hint(irq
->vector
, NULL
);
1366 free_irq(irq
->vector
, irq
->data
);
1370 static void ena_disable_msix(struct ena_adapter
*adapter
)
1372 if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED
, &adapter
->flags
))
1373 pci_disable_msix(adapter
->pdev
);
1375 if (adapter
->msix_entries
)
1376 vfree(adapter
->msix_entries
);
1377 adapter
->msix_entries
= NULL
;
1380 static void ena_disable_io_intr_sync(struct ena_adapter
*adapter
)
1384 if (!netif_running(adapter
->netdev
))
1387 for (i
= ENA_IO_IRQ_FIRST_IDX
; i
< adapter
->msix_vecs
; i
++)
1388 synchronize_irq(adapter
->irq_tbl
[i
].vector
);
1391 static void ena_del_napi(struct ena_adapter
*adapter
)
1395 for (i
= 0; i
< adapter
->num_queues
; i
++)
1396 netif_napi_del(&adapter
->ena_napi
[i
].napi
);
1399 static void ena_init_napi(struct ena_adapter
*adapter
)
1401 struct ena_napi
*napi
;
1404 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1405 napi
= &adapter
->ena_napi
[i
];
1407 netif_napi_add(adapter
->netdev
,
1408 &adapter
->ena_napi
[i
].napi
,
1411 napi
->rx_ring
= &adapter
->rx_ring
[i
];
1412 napi
->tx_ring
= &adapter
->tx_ring
[i
];
1417 static void ena_napi_disable_all(struct ena_adapter
*adapter
)
1421 for (i
= 0; i
< adapter
->num_queues
; i
++)
1422 napi_disable(&adapter
->ena_napi
[i
].napi
);
1425 static void ena_napi_enable_all(struct ena_adapter
*adapter
)
1429 for (i
= 0; i
< adapter
->num_queues
; i
++)
1430 napi_enable(&adapter
->ena_napi
[i
].napi
);
1433 static void ena_restore_ethtool_params(struct ena_adapter
*adapter
)
1435 adapter
->tx_usecs
= 0;
1436 adapter
->rx_usecs
= 0;
1437 adapter
->tx_frames
= 1;
1438 adapter
->rx_frames
= 1;
1441 /* Configure the Rx forwarding */
1442 static int ena_rss_configure(struct ena_adapter
*adapter
)
1444 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1447 /* In case the RSS table wasn't initialized by probe */
1448 if (!ena_dev
->rss
.tbl_log_size
) {
1449 rc
= ena_rss_init_default(adapter
);
1450 if (rc
&& (rc
!= -EPERM
)) {
1451 netif_err(adapter
, ifup
, adapter
->netdev
,
1452 "Failed to init RSS rc: %d\n", rc
);
1457 /* Set indirect table */
1458 rc
= ena_com_indirect_table_set(ena_dev
);
1459 if (unlikely(rc
&& rc
!= -EPERM
))
1462 /* Configure hash function (if supported) */
1463 rc
= ena_com_set_hash_function(ena_dev
);
1464 if (unlikely(rc
&& (rc
!= -EPERM
)))
1467 /* Configure hash inputs (if supported) */
1468 rc
= ena_com_set_hash_ctrl(ena_dev
);
1469 if (unlikely(rc
&& (rc
!= -EPERM
)))
1475 static int ena_up_complete(struct ena_adapter
*adapter
)
1479 rc
= ena_rss_configure(adapter
);
1483 ena_init_napi(adapter
);
1485 ena_change_mtu(adapter
->netdev
, adapter
->netdev
->mtu
);
1487 ena_refill_all_rx_bufs(adapter
);
1489 /* enable transmits */
1490 netif_tx_start_all_queues(adapter
->netdev
);
1492 ena_restore_ethtool_params(adapter
);
1494 ena_napi_enable_all(adapter
);
1496 /* schedule napi in case we had pending packets
1497 * from the last time we disable napi
1499 for (i
= 0; i
< adapter
->num_queues
; i
++)
1500 napi_schedule(&adapter
->ena_napi
[i
].napi
);
1505 static int ena_create_io_tx_queue(struct ena_adapter
*adapter
, int qid
)
1507 struct ena_com_create_io_ctx ctx
= { 0 };
1508 struct ena_com_dev
*ena_dev
;
1509 struct ena_ring
*tx_ring
;
1514 ena_dev
= adapter
->ena_dev
;
1516 tx_ring
= &adapter
->tx_ring
[qid
];
1517 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1518 ena_qid
= ENA_IO_TXQ_IDX(qid
);
1520 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_TX
;
1522 ctx
.mem_queue_type
= ena_dev
->tx_mem_queue_type
;
1523 ctx
.msix_vector
= msix_vector
;
1524 ctx
.queue_size
= adapter
->tx_ring_size
;
1525 ctx
.numa_node
= cpu_to_node(tx_ring
->cpu
);
1527 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1529 netif_err(adapter
, ifup
, adapter
->netdev
,
1530 "Failed to create I/O TX queue num %d rc: %d\n",
1535 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1536 &tx_ring
->ena_com_io_sq
,
1537 &tx_ring
->ena_com_io_cq
);
1539 netif_err(adapter
, ifup
, adapter
->netdev
,
1540 "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1542 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1545 ena_com_update_numa_node(tx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1549 static int ena_create_all_io_tx_queues(struct ena_adapter
*adapter
)
1551 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1554 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1555 rc
= ena_create_io_tx_queue(adapter
, i
);
1564 ena_com_destroy_io_queue(ena_dev
, ENA_IO_TXQ_IDX(i
));
1569 static int ena_create_io_rx_queue(struct ena_adapter
*adapter
, int qid
)
1571 struct ena_com_dev
*ena_dev
;
1572 struct ena_com_create_io_ctx ctx
= { 0 };
1573 struct ena_ring
*rx_ring
;
1578 ena_dev
= adapter
->ena_dev
;
1580 rx_ring
= &adapter
->rx_ring
[qid
];
1581 msix_vector
= ENA_IO_IRQ_IDX(qid
);
1582 ena_qid
= ENA_IO_RXQ_IDX(qid
);
1585 ctx
.direction
= ENA_COM_IO_QUEUE_DIRECTION_RX
;
1586 ctx
.mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
1587 ctx
.msix_vector
= msix_vector
;
1588 ctx
.queue_size
= adapter
->rx_ring_size
;
1589 ctx
.numa_node
= cpu_to_node(rx_ring
->cpu
);
1591 rc
= ena_com_create_io_queue(ena_dev
, &ctx
);
1593 netif_err(adapter
, ifup
, adapter
->netdev
,
1594 "Failed to create I/O RX queue num %d rc: %d\n",
1599 rc
= ena_com_get_io_handlers(ena_dev
, ena_qid
,
1600 &rx_ring
->ena_com_io_sq
,
1601 &rx_ring
->ena_com_io_cq
);
1603 netif_err(adapter
, ifup
, adapter
->netdev
,
1604 "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1606 ena_com_destroy_io_queue(ena_dev
, ena_qid
);
1609 ena_com_update_numa_node(rx_ring
->ena_com_io_cq
, ctx
.numa_node
);
1614 static int ena_create_all_io_rx_queues(struct ena_adapter
*adapter
)
1616 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
1619 for (i
= 0; i
< adapter
->num_queues
; i
++) {
1620 rc
= ena_create_io_rx_queue(adapter
, i
);
1629 ena_com_destroy_io_queue(ena_dev
, ENA_IO_RXQ_IDX(i
));
1634 static int ena_up(struct ena_adapter
*adapter
)
1638 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
1640 ena_setup_io_intr(adapter
);
1642 rc
= ena_request_io_irq(adapter
);
1646 /* allocate transmit descriptors */
1647 rc
= ena_setup_all_tx_resources(adapter
);
1651 /* allocate receive descriptors */
1652 rc
= ena_setup_all_rx_resources(adapter
);
1656 /* Create TX queues */
1657 rc
= ena_create_all_io_tx_queues(adapter
);
1659 goto err_create_tx_queues
;
1661 /* Create RX queues */
1662 rc
= ena_create_all_io_rx_queues(adapter
);
1664 goto err_create_rx_queues
;
1666 rc
= ena_up_complete(adapter
);
1670 if (test_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
))
1671 netif_carrier_on(adapter
->netdev
);
1673 u64_stats_update_begin(&adapter
->syncp
);
1674 adapter
->dev_stats
.interface_up
++;
1675 u64_stats_update_end(&adapter
->syncp
);
1677 set_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1682 ena_destroy_all_rx_queues(adapter
);
1683 err_create_rx_queues
:
1684 ena_destroy_all_tx_queues(adapter
);
1685 err_create_tx_queues
:
1686 ena_free_all_io_rx_resources(adapter
);
1688 ena_free_all_io_tx_resources(adapter
);
1690 ena_free_io_irq(adapter
);
1696 static void ena_down(struct ena_adapter
*adapter
)
1698 netif_info(adapter
, ifdown
, adapter
->netdev
, "%s\n", __func__
);
1700 clear_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
1702 u64_stats_update_begin(&adapter
->syncp
);
1703 adapter
->dev_stats
.interface_down
++;
1704 u64_stats_update_end(&adapter
->syncp
);
1706 netif_carrier_off(adapter
->netdev
);
1707 netif_tx_disable(adapter
->netdev
);
1709 /* After this point the napi handler won't enable the tx queue */
1710 ena_napi_disable_all(adapter
);
1712 /* After destroy the queue there won't be any new interrupts */
1714 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
)) {
1717 rc
= ena_com_dev_reset(adapter
->ena_dev
);
1719 dev_err(&adapter
->pdev
->dev
, "Device reset failed\n");
1722 ena_destroy_all_io_queues(adapter
);
1724 ena_disable_io_intr_sync(adapter
);
1725 ena_free_io_irq(adapter
);
1726 ena_del_napi(adapter
);
1728 ena_free_all_tx_bufs(adapter
);
1729 ena_free_all_rx_bufs(adapter
);
1730 ena_free_all_io_tx_resources(adapter
);
1731 ena_free_all_io_rx_resources(adapter
);
1734 /* ena_open - Called when a network interface is made active
1735 * @netdev: network interface device structure
1737 * Returns 0 on success, negative value on failure
1739 * The open entry point is called when a network interface is made
1740 * active by the system (IFF_UP). At this point all resources needed
1741 * for transmit and receive operations are allocated, the interrupt
1742 * handler is registered with the OS, the watchdog timer is started,
1743 * and the stack is notified that the interface is ready.
1745 static int ena_open(struct net_device
*netdev
)
1747 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1750 /* Notify the stack of the actual queue counts. */
1751 rc
= netif_set_real_num_tx_queues(netdev
, adapter
->num_queues
);
1753 netif_err(adapter
, ifup
, netdev
, "Can't set num tx queues\n");
1757 rc
= netif_set_real_num_rx_queues(netdev
, adapter
->num_queues
);
1759 netif_err(adapter
, ifup
, netdev
, "Can't set num rx queues\n");
1763 rc
= ena_up(adapter
);
1770 /* ena_close - Disables a network interface
1771 * @netdev: network interface device structure
1773 * Returns 0, this is not allowed to fail
1775 * The close entry point is called when an interface is de-activated
1776 * by the OS. The hardware is still under the drivers control, but
1777 * needs to be disabled. A global MAC reset is issued to stop the
1778 * hardware, and all transmit and receive resources are freed.
1780 static int ena_close(struct net_device
*netdev
)
1782 struct ena_adapter
*adapter
= netdev_priv(netdev
);
1784 netif_dbg(adapter
, ifdown
, netdev
, "%s\n", __func__
);
1786 if (test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
1792 static void ena_tx_csum(struct ena_com_tx_ctx
*ena_tx_ctx
, struct sk_buff
*skb
)
1794 u32 mss
= skb_shinfo(skb
)->gso_size
;
1795 struct ena_com_tx_meta
*ena_meta
= &ena_tx_ctx
->ena_meta
;
1798 if ((skb
->ip_summed
== CHECKSUM_PARTIAL
) || mss
) {
1799 ena_tx_ctx
->l4_csum_enable
= 1;
1801 ena_tx_ctx
->tso_enable
= 1;
1802 ena_meta
->l4_hdr_len
= tcp_hdr(skb
)->doff
;
1803 ena_tx_ctx
->l4_csum_partial
= 0;
1805 ena_tx_ctx
->tso_enable
= 0;
1806 ena_meta
->l4_hdr_len
= 0;
1807 ena_tx_ctx
->l4_csum_partial
= 1;
1810 switch (ip_hdr(skb
)->version
) {
1812 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV4
;
1813 if (ip_hdr(skb
)->frag_off
& htons(IP_DF
))
1816 ena_tx_ctx
->l3_csum_enable
= 1;
1817 l4_protocol
= ip_hdr(skb
)->protocol
;
1820 ena_tx_ctx
->l3_proto
= ENA_ETH_IO_L3_PROTO_IPV6
;
1821 l4_protocol
= ipv6_hdr(skb
)->nexthdr
;
1827 if (l4_protocol
== IPPROTO_TCP
)
1828 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_TCP
;
1830 ena_tx_ctx
->l4_proto
= ENA_ETH_IO_L4_PROTO_UDP
;
1832 ena_meta
->mss
= mss
;
1833 ena_meta
->l3_hdr_len
= skb_network_header_len(skb
);
1834 ena_meta
->l3_hdr_offset
= skb_network_offset(skb
);
1835 ena_tx_ctx
->meta_valid
= 1;
1838 ena_tx_ctx
->meta_valid
= 0;
1842 static int ena_check_and_linearize_skb(struct ena_ring
*tx_ring
,
1843 struct sk_buff
*skb
)
1845 int num_frags
, header_len
, rc
;
1847 num_frags
= skb_shinfo(skb
)->nr_frags
;
1848 header_len
= skb_headlen(skb
);
1850 if (num_frags
< tx_ring
->sgl_size
)
1853 if ((num_frags
== tx_ring
->sgl_size
) &&
1854 (header_len
< tx_ring
->tx_max_header_size
))
1857 u64_stats_update_begin(&tx_ring
->syncp
);
1858 tx_ring
->tx_stats
.linearize
++;
1859 u64_stats_update_end(&tx_ring
->syncp
);
1861 rc
= skb_linearize(skb
);
1863 u64_stats_update_begin(&tx_ring
->syncp
);
1864 tx_ring
->tx_stats
.linearize_failed
++;
1865 u64_stats_update_end(&tx_ring
->syncp
);
1871 /* Called with netif_tx_lock. */
1872 static netdev_tx_t
ena_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1874 struct ena_adapter
*adapter
= netdev_priv(dev
);
1875 struct ena_tx_buffer
*tx_info
;
1876 struct ena_com_tx_ctx ena_tx_ctx
;
1877 struct ena_ring
*tx_ring
;
1878 struct netdev_queue
*txq
;
1879 struct ena_com_buf
*ena_buf
;
1887 int qid
, rc
, nb_hw_desc
;
1890 netif_dbg(adapter
, tx_queued
, dev
, "%s skb %p\n", __func__
, skb
);
1891 /* Determine which tx ring we will be placed on */
1892 qid
= skb_get_queue_mapping(skb
);
1893 tx_ring
= &adapter
->tx_ring
[qid
];
1894 txq
= netdev_get_tx_queue(dev
, qid
);
1896 rc
= ena_check_and_linearize_skb(tx_ring
, skb
);
1898 goto error_drop_packet
;
1900 skb_tx_timestamp(skb
);
1901 len
= skb_headlen(skb
);
1903 next_to_use
= tx_ring
->next_to_use
;
1904 req_id
= tx_ring
->free_tx_ids
[next_to_use
];
1905 tx_info
= &tx_ring
->tx_buffer_info
[req_id
];
1906 tx_info
->num_of_bufs
= 0;
1908 WARN(tx_info
->skb
, "SKB isn't NULL req_id %d\n", req_id
);
1909 ena_buf
= tx_info
->bufs
;
1912 if (tx_ring
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
1913 /* prepared the push buffer */
1914 push_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1915 header_len
= push_len
;
1916 push_hdr
= skb
->data
;
1919 header_len
= min_t(u32
, len
, tx_ring
->tx_max_header_size
);
1923 netif_dbg(adapter
, tx_queued
, dev
,
1924 "skb: %p header_buf->vaddr: %p push_len: %d\n", skb
,
1925 push_hdr
, push_len
);
1927 if (len
> push_len
) {
1928 dma
= dma_map_single(tx_ring
->dev
, skb
->data
+ push_len
,
1929 len
- push_len
, DMA_TO_DEVICE
);
1930 if (dma_mapping_error(tx_ring
->dev
, dma
))
1931 goto error_report_dma_error
;
1933 ena_buf
->paddr
= dma
;
1934 ena_buf
->len
= len
- push_len
;
1937 tx_info
->num_of_bufs
++;
1940 last_frag
= skb_shinfo(skb
)->nr_frags
;
1942 for (i
= 0; i
< last_frag
; i
++) {
1943 const skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
1945 len
= skb_frag_size(frag
);
1946 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, len
,
1948 if (dma_mapping_error(tx_ring
->dev
, dma
))
1949 goto error_report_dma_error
;
1951 ena_buf
->paddr
= dma
;
1956 tx_info
->num_of_bufs
+= last_frag
;
1958 memset(&ena_tx_ctx
, 0x0, sizeof(struct ena_com_tx_ctx
));
1959 ena_tx_ctx
.ena_bufs
= tx_info
->bufs
;
1960 ena_tx_ctx
.push_header
= push_hdr
;
1961 ena_tx_ctx
.num_bufs
= tx_info
->num_of_bufs
;
1962 ena_tx_ctx
.req_id
= req_id
;
1963 ena_tx_ctx
.header_len
= header_len
;
1965 /* set flags and meta data */
1966 ena_tx_csum(&ena_tx_ctx
, skb
);
1968 /* prepare the packet's descriptors to dma engine */
1969 rc
= ena_com_prepare_tx(tx_ring
->ena_com_io_sq
, &ena_tx_ctx
,
1973 netif_err(adapter
, tx_queued
, dev
,
1974 "failed to prepare tx bufs\n");
1975 u64_stats_update_begin(&tx_ring
->syncp
);
1976 tx_ring
->tx_stats
.queue_stop
++;
1977 tx_ring
->tx_stats
.prepare_ctx_err
++;
1978 u64_stats_update_end(&tx_ring
->syncp
);
1979 netif_tx_stop_queue(txq
);
1980 goto error_unmap_dma
;
1983 netdev_tx_sent_queue(txq
, skb
->len
);
1985 u64_stats_update_begin(&tx_ring
->syncp
);
1986 tx_ring
->tx_stats
.cnt
++;
1987 tx_ring
->tx_stats
.bytes
+= skb
->len
;
1988 u64_stats_update_end(&tx_ring
->syncp
);
1990 tx_info
->tx_descs
= nb_hw_desc
;
1991 tx_info
->last_jiffies
= jiffies
;
1993 tx_ring
->next_to_use
= ENA_TX_RING_IDX_NEXT(next_to_use
,
1994 tx_ring
->ring_size
);
1996 /* This WMB is aimed to:
1997 * 1 - perform smp barrier before reading next_to_completion
1998 * 2 - make sure the desc were written before trigger DB
2002 /* stop the queue when no more space available, the packet can have up
2003 * to sgl_size + 2. one for the meta descriptor and one for header
2004 * (if the header is larger than tx_max_header_size).
2006 if (unlikely(ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
) <
2007 (tx_ring
->sgl_size
+ 2))) {
2008 netif_dbg(adapter
, tx_queued
, dev
, "%s stop queue %d\n",
2011 netif_tx_stop_queue(txq
);
2012 u64_stats_update_begin(&tx_ring
->syncp
);
2013 tx_ring
->tx_stats
.queue_stop
++;
2014 u64_stats_update_end(&tx_ring
->syncp
);
2016 /* There is a rare condition where this function decide to
2017 * stop the queue but meanwhile clean_tx_irq updates
2018 * next_to_completion and terminates.
2019 * The queue will remain stopped forever.
2020 * To solve this issue this function perform rmb, check
2021 * the wakeup condition and wake up the queue if needed.
2025 if (ena_com_sq_empty_space(tx_ring
->ena_com_io_sq
)
2026 > ENA_TX_WAKEUP_THRESH
) {
2027 netif_tx_wake_queue(txq
);
2028 u64_stats_update_begin(&tx_ring
->syncp
);
2029 tx_ring
->tx_stats
.queue_wakeup
++;
2030 u64_stats_update_end(&tx_ring
->syncp
);
2034 if (netif_xmit_stopped(txq
) || !skb
->xmit_more
) {
2035 /* trigger the dma engine */
2036 ena_com_write_sq_doorbell(tx_ring
->ena_com_io_sq
);
2037 u64_stats_update_begin(&tx_ring
->syncp
);
2038 tx_ring
->tx_stats
.doorbells
++;
2039 u64_stats_update_end(&tx_ring
->syncp
);
2042 return NETDEV_TX_OK
;
2044 error_report_dma_error
:
2045 u64_stats_update_begin(&tx_ring
->syncp
);
2046 tx_ring
->tx_stats
.dma_mapping_err
++;
2047 u64_stats_update_end(&tx_ring
->syncp
);
2048 netdev_warn(adapter
->netdev
, "failed to map skb\n");
2050 tx_info
->skb
= NULL
;
2054 /* save value of frag that failed */
2057 /* start back at beginning and unmap skb */
2058 tx_info
->skb
= NULL
;
2059 ena_buf
= tx_info
->bufs
;
2060 dma_unmap_single(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2061 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2063 /* unmap remaining mapped pages */
2064 for (i
= 0; i
< last_frag
; i
++) {
2066 dma_unmap_page(tx_ring
->dev
, dma_unmap_addr(ena_buf
, paddr
),
2067 dma_unmap_len(ena_buf
, len
), DMA_TO_DEVICE
);
2074 return NETDEV_TX_OK
;
2077 #ifdef CONFIG_NET_POLL_CONTROLLER
2078 static void ena_netpoll(struct net_device
*netdev
)
2080 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2083 /* Dont schedule NAPI if the driver is in the middle of reset
2084 * or netdev is down.
2087 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
) ||
2088 test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2091 for (i
= 0; i
< adapter
->num_queues
; i
++)
2092 napi_schedule(&adapter
->ena_napi
[i
].napi
);
2094 #endif /* CONFIG_NET_POLL_CONTROLLER */
2096 static u16
ena_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
2097 void *accel_priv
, select_queue_fallback_t fallback
)
2100 /* we suspect that this is good for in--kernel network services that
2101 * want to loop incoming skb rx to tx in normal user generated traffic,
2102 * most probably we will not get to this
2104 if (skb_rx_queue_recorded(skb
))
2105 qid
= skb_get_rx_queue(skb
);
2107 qid
= fallback(dev
, skb
);
2112 static void ena_config_host_info(struct ena_com_dev
*ena_dev
)
2114 struct ena_admin_host_info
*host_info
;
2117 /* Allocate only the host info */
2118 rc
= ena_com_allocate_host_info(ena_dev
);
2120 pr_err("Cannot allocate host info\n");
2124 host_info
= ena_dev
->host_attr
.host_info
;
2126 host_info
->os_type
= ENA_ADMIN_OS_LINUX
;
2127 host_info
->kernel_ver
= LINUX_VERSION_CODE
;
2128 strncpy(host_info
->kernel_ver_str
, utsname()->version
,
2129 sizeof(host_info
->kernel_ver_str
) - 1);
2130 host_info
->os_dist
= 0;
2131 strncpy(host_info
->os_dist_str
, utsname()->release
,
2132 sizeof(host_info
->os_dist_str
) - 1);
2133 host_info
->driver_version
=
2134 (DRV_MODULE_VER_MAJOR
) |
2135 (DRV_MODULE_VER_MINOR
<< ENA_ADMIN_HOST_INFO_MINOR_SHIFT
) |
2136 (DRV_MODULE_VER_SUBMINOR
<< ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT
);
2138 rc
= ena_com_set_host_attributes(ena_dev
);
2141 pr_warn("Cannot set host attributes\n");
2143 pr_err("Cannot set host attributes\n");
2151 ena_com_delete_host_info(ena_dev
);
2154 static void ena_config_debug_area(struct ena_adapter
*adapter
)
2156 u32 debug_area_size
;
2159 ss_count
= ena_get_sset_count(adapter
->netdev
, ETH_SS_STATS
);
2160 if (ss_count
<= 0) {
2161 netif_err(adapter
, drv
, adapter
->netdev
,
2162 "SS count is negative\n");
2166 /* allocate 32 bytes for each string and 64bit for the value */
2167 debug_area_size
= ss_count
* ETH_GSTRING_LEN
+ sizeof(u64
) * ss_count
;
2169 rc
= ena_com_allocate_debug_area(adapter
->ena_dev
, debug_area_size
);
2171 pr_err("Cannot allocate debug area\n");
2175 rc
= ena_com_set_host_attributes(adapter
->ena_dev
);
2178 netif_warn(adapter
, drv
, adapter
->netdev
,
2179 "Cannot set host attributes\n");
2181 netif_err(adapter
, drv
, adapter
->netdev
,
2182 "Cannot set host attributes\n");
2188 ena_com_delete_debug_area(adapter
->ena_dev
);
2191 static void ena_get_stats64(struct net_device
*netdev
,
2192 struct rtnl_link_stats64
*stats
)
2194 struct ena_adapter
*adapter
= netdev_priv(netdev
);
2195 struct ena_ring
*rx_ring
, *tx_ring
;
2200 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2203 for (i
= 0; i
< adapter
->num_queues
; i
++) {
2206 tx_ring
= &adapter
->tx_ring
[i
];
2209 start
= u64_stats_fetch_begin_irq(&tx_ring
->syncp
);
2210 packets
= tx_ring
->tx_stats
.cnt
;
2211 bytes
= tx_ring
->tx_stats
.bytes
;
2212 } while (u64_stats_fetch_retry_irq(&tx_ring
->syncp
, start
));
2214 stats
->tx_packets
+= packets
;
2215 stats
->tx_bytes
+= bytes
;
2217 rx_ring
= &adapter
->rx_ring
[i
];
2220 start
= u64_stats_fetch_begin_irq(&rx_ring
->syncp
);
2221 packets
= rx_ring
->rx_stats
.cnt
;
2222 bytes
= rx_ring
->rx_stats
.bytes
;
2223 } while (u64_stats_fetch_retry_irq(&rx_ring
->syncp
, start
));
2225 stats
->rx_packets
+= packets
;
2226 stats
->rx_bytes
+= bytes
;
2230 start
= u64_stats_fetch_begin_irq(&adapter
->syncp
);
2231 rx_drops
= adapter
->dev_stats
.rx_drops
;
2232 } while (u64_stats_fetch_retry_irq(&adapter
->syncp
, start
));
2234 stats
->rx_dropped
= rx_drops
;
2236 stats
->multicast
= 0;
2237 stats
->collisions
= 0;
2239 stats
->rx_length_errors
= 0;
2240 stats
->rx_crc_errors
= 0;
2241 stats
->rx_frame_errors
= 0;
2242 stats
->rx_fifo_errors
= 0;
2243 stats
->rx_missed_errors
= 0;
2244 stats
->tx_window_errors
= 0;
2246 stats
->rx_errors
= 0;
2247 stats
->tx_errors
= 0;
2250 static const struct net_device_ops ena_netdev_ops
= {
2251 .ndo_open
= ena_open
,
2252 .ndo_stop
= ena_close
,
2253 .ndo_start_xmit
= ena_start_xmit
,
2254 .ndo_select_queue
= ena_select_queue
,
2255 .ndo_get_stats64
= ena_get_stats64
,
2256 .ndo_tx_timeout
= ena_tx_timeout
,
2257 .ndo_change_mtu
= ena_change_mtu
,
2258 .ndo_set_mac_address
= NULL
,
2259 .ndo_validate_addr
= eth_validate_addr
,
2260 #ifdef CONFIG_NET_POLL_CONTROLLER
2261 .ndo_poll_controller
= ena_netpoll
,
2262 #endif /* CONFIG_NET_POLL_CONTROLLER */
2265 static void ena_device_io_suspend(struct work_struct
*work
)
2267 struct ena_adapter
*adapter
=
2268 container_of(work
, struct ena_adapter
, suspend_io_task
);
2269 struct net_device
*netdev
= adapter
->netdev
;
2271 /* ena_napi_disable_all disables only the IO handling.
2272 * We are still subject to AENQ keep alive watchdog.
2274 u64_stats_update_begin(&adapter
->syncp
);
2275 adapter
->dev_stats
.io_suspend
++;
2276 u64_stats_update_begin(&adapter
->syncp
);
2277 ena_napi_disable_all(adapter
);
2278 netif_tx_lock(netdev
);
2279 netif_device_detach(netdev
);
2280 netif_tx_unlock(netdev
);
2283 static void ena_device_io_resume(struct work_struct
*work
)
2285 struct ena_adapter
*adapter
=
2286 container_of(work
, struct ena_adapter
, resume_io_task
);
2287 struct net_device
*netdev
= adapter
->netdev
;
2289 u64_stats_update_begin(&adapter
->syncp
);
2290 adapter
->dev_stats
.io_resume
++;
2291 u64_stats_update_end(&adapter
->syncp
);
2293 netif_device_attach(netdev
);
2294 ena_napi_enable_all(adapter
);
2297 static int ena_device_validate_params(struct ena_adapter
*adapter
,
2298 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2300 struct net_device
*netdev
= adapter
->netdev
;
2303 rc
= ether_addr_equal(get_feat_ctx
->dev_attr
.mac_addr
,
2306 netif_err(adapter
, drv
, netdev
,
2307 "Error, mac address are different\n");
2311 if ((get_feat_ctx
->max_queues
.max_cq_num
< adapter
->num_queues
) ||
2312 (get_feat_ctx
->max_queues
.max_sq_num
< adapter
->num_queues
)) {
2313 netif_err(adapter
, drv
, netdev
,
2314 "Error, device doesn't support enough queues\n");
2318 if (get_feat_ctx
->dev_attr
.max_mtu
< netdev
->mtu
) {
2319 netif_err(adapter
, drv
, netdev
,
2320 "Error, device max mtu is smaller than netdev MTU\n");
2327 static int ena_device_init(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
,
2328 struct ena_com_dev_get_features_ctx
*get_feat_ctx
,
2331 struct device
*dev
= &pdev
->dev
;
2332 bool readless_supported
;
2337 rc
= ena_com_mmio_reg_read_request_init(ena_dev
);
2339 dev_err(dev
, "failed to init mmio read less\n");
2343 /* The PCIe configuration space revision id indicate if mmio reg
2346 readless_supported
= !(pdev
->revision
& ENA_MMIO_DISABLE_REG_READ
);
2347 ena_com_set_mmio_read_mode(ena_dev
, readless_supported
);
2349 rc
= ena_com_dev_reset(ena_dev
);
2351 dev_err(dev
, "Can not reset device\n");
2352 goto err_mmio_read_less
;
2355 rc
= ena_com_validate_version(ena_dev
);
2357 dev_err(dev
, "device version is too low\n");
2358 goto err_mmio_read_less
;
2361 dma_width
= ena_com_get_dma_width(ena_dev
);
2362 if (dma_width
< 0) {
2363 dev_err(dev
, "Invalid dma width value %d", dma_width
);
2365 goto err_mmio_read_less
;
2368 rc
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2370 dev_err(dev
, "pci_set_dma_mask failed 0x%x\n", rc
);
2371 goto err_mmio_read_less
;
2374 rc
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(dma_width
));
2376 dev_err(dev
, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2378 goto err_mmio_read_less
;
2381 /* ENA admin level init */
2382 rc
= ena_com_admin_init(ena_dev
, &aenq_handlers
, true);
2385 "Can not initialize ena admin queue with device\n");
2386 goto err_mmio_read_less
;
2389 /* To enable the msix interrupts the driver needs to know the number
2390 * of queues. So the driver uses polling mode to retrieve this
2393 ena_com_set_admin_polling_mode(ena_dev
, true);
2395 /* Get Device Attributes*/
2396 rc
= ena_com_get_dev_attr_feat(ena_dev
, get_feat_ctx
);
2398 dev_err(dev
, "Cannot get attribute for ena device rc=%d\n", rc
);
2399 goto err_admin_init
;
2402 /* Try to turn all the available aenq groups */
2403 aenq_groups
= BIT(ENA_ADMIN_LINK_CHANGE
) |
2404 BIT(ENA_ADMIN_FATAL_ERROR
) |
2405 BIT(ENA_ADMIN_WARNING
) |
2406 BIT(ENA_ADMIN_NOTIFICATION
) |
2407 BIT(ENA_ADMIN_KEEP_ALIVE
);
2409 aenq_groups
&= get_feat_ctx
->aenq
.supported_groups
;
2411 rc
= ena_com_set_aenq_config(ena_dev
, aenq_groups
);
2413 dev_err(dev
, "Cannot configure aenq groups rc= %d\n", rc
);
2414 goto err_admin_init
;
2417 *wd_state
= !!(aenq_groups
& BIT(ENA_ADMIN_KEEP_ALIVE
));
2419 ena_config_host_info(ena_dev
);
2424 ena_com_admin_destroy(ena_dev
);
2426 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2431 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter
*adapter
,
2434 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2435 struct device
*dev
= &adapter
->pdev
->dev
;
2438 rc
= ena_enable_msix(adapter
, io_vectors
);
2440 dev_err(dev
, "Can not reserve msix vectors\n");
2444 ena_setup_mgmnt_intr(adapter
);
2446 rc
= ena_request_mgmnt_irq(adapter
);
2448 dev_err(dev
, "Can not setup management interrupts\n");
2449 goto err_disable_msix
;
2452 ena_com_set_admin_polling_mode(ena_dev
, false);
2454 ena_com_admin_aenq_enable(ena_dev
);
2459 ena_disable_msix(adapter
);
2464 static void ena_fw_reset_device(struct work_struct
*work
)
2466 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2467 struct ena_adapter
*adapter
=
2468 container_of(work
, struct ena_adapter
, reset_task
);
2469 struct net_device
*netdev
= adapter
->netdev
;
2470 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2471 struct pci_dev
*pdev
= adapter
->pdev
;
2472 bool dev_up
, wd_state
;
2475 if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2477 "device reset schedule while reset bit is off\n");
2481 netif_carrier_off(netdev
);
2483 del_timer_sync(&adapter
->timer_service
);
2487 dev_up
= test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
);
2488 ena_com_set_admin_running_state(ena_dev
, false);
2490 /* After calling ena_close the tx queues and the napi
2491 * are disabled so no one can interfere or touch the
2496 ena_free_mgmnt_irq(adapter
);
2498 ena_disable_msix(adapter
);
2500 ena_com_abort_admin_commands(ena_dev
);
2502 ena_com_wait_for_abort_completion(ena_dev
);
2504 ena_com_admin_destroy(ena_dev
);
2506 ena_com_mmio_reg_read_request_destroy(ena_dev
);
2508 clear_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2510 /* Finish with the destroy part. Start the init part */
2512 rc
= ena_device_init(ena_dev
, adapter
->pdev
, &get_feat_ctx
, &wd_state
);
2514 dev_err(&pdev
->dev
, "Can not initialize device\n");
2517 adapter
->wd_state
= wd_state
;
2519 rc
= ena_device_validate_params(adapter
, &get_feat_ctx
);
2521 dev_err(&pdev
->dev
, "Validation of device parameters failed\n");
2522 goto err_device_destroy
;
2525 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
,
2526 adapter
->num_queues
);
2528 dev_err(&pdev
->dev
, "Enable MSI-X failed\n");
2529 goto err_device_destroy
;
2531 /* If the interface was up before the reset bring it up */
2533 rc
= ena_up(adapter
);
2535 dev_err(&pdev
->dev
, "Failed to create I/O queues\n");
2536 goto err_disable_msix
;
2540 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
2544 dev_err(&pdev
->dev
, "Device reset completed successfully\n");
2548 ena_free_mgmnt_irq(adapter
);
2549 ena_disable_msix(adapter
);
2551 ena_com_admin_destroy(ena_dev
);
2555 clear_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
2558 "Reset attempt failed. Can not reset the device\n");
2561 static void check_for_missing_tx_completions(struct ena_adapter
*adapter
)
2563 struct ena_tx_buffer
*tx_buf
;
2564 unsigned long last_jiffies
;
2565 struct ena_ring
*tx_ring
;
2569 /* Make sure the driver doesn't turn the device in other process */
2572 if (!test_bit(ENA_FLAG_DEV_UP
, &adapter
->flags
))
2575 if (test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))
2578 budget
= ENA_MONITORED_TX_QUEUES
;
2580 for (i
= adapter
->last_monitored_tx_qid
; i
< adapter
->num_queues
; i
++) {
2581 tx_ring
= &adapter
->tx_ring
[i
];
2583 for (j
= 0; j
< tx_ring
->ring_size
; j
++) {
2584 tx_buf
= &tx_ring
->tx_buffer_info
[j
];
2585 last_jiffies
= tx_buf
->last_jiffies
;
2586 if (unlikely(last_jiffies
&& time_is_before_jiffies(last_jiffies
+ TX_TIMEOUT
))) {
2587 netif_notice(adapter
, tx_err
, adapter
->netdev
,
2588 "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2591 u64_stats_update_begin(&tx_ring
->syncp
);
2592 missed_tx
= tx_ring
->tx_stats
.missing_tx_comp
++;
2593 u64_stats_update_end(&tx_ring
->syncp
);
2595 /* Clear last jiffies so the lost buffer won't
2598 tx_buf
->last_jiffies
= 0;
2600 if (unlikely(missed_tx
> MAX_NUM_OF_TIMEOUTED_PACKETS
)) {
2601 netif_err(adapter
, tx_err
, adapter
->netdev
,
2602 "The number of lost tx completion is above the threshold (%d > %d). Reset the device\n",
2603 missed_tx
, MAX_NUM_OF_TIMEOUTED_PACKETS
);
2604 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2614 adapter
->last_monitored_tx_qid
= i
% adapter
->num_queues
;
2617 /* Check for keep alive expiration */
2618 static void check_for_missing_keep_alive(struct ena_adapter
*adapter
)
2620 unsigned long keep_alive_expired
;
2622 if (!adapter
->wd_state
)
2625 keep_alive_expired
= round_jiffies(adapter
->last_keep_alive_jiffies
2626 + ENA_DEVICE_KALIVE_TIMEOUT
);
2627 if (unlikely(time_is_before_jiffies(keep_alive_expired
))) {
2628 netif_err(adapter
, drv
, adapter
->netdev
,
2629 "Keep alive watchdog timeout.\n");
2630 u64_stats_update_begin(&adapter
->syncp
);
2631 adapter
->dev_stats
.wd_expired
++;
2632 u64_stats_update_end(&adapter
->syncp
);
2633 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2637 static void check_for_admin_com_state(struct ena_adapter
*adapter
)
2639 if (unlikely(!ena_com_get_admin_running_state(adapter
->ena_dev
))) {
2640 netif_err(adapter
, drv
, adapter
->netdev
,
2641 "ENA admin queue is not in running state!\n");
2642 u64_stats_update_begin(&adapter
->syncp
);
2643 adapter
->dev_stats
.admin_q_pause
++;
2644 u64_stats_update_end(&adapter
->syncp
);
2645 set_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
);
2649 static void ena_update_host_info(struct ena_admin_host_info
*host_info
,
2650 struct net_device
*netdev
)
2652 host_info
->supported_network_features
[0] =
2653 netdev
->features
& GENMASK_ULL(31, 0);
2654 host_info
->supported_network_features
[1] =
2655 (netdev
->features
& GENMASK_ULL(63, 32)) >> 32;
2658 static void ena_timer_service(unsigned long data
)
2660 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
2661 u8
*debug_area
= adapter
->ena_dev
->host_attr
.debug_area_virt_addr
;
2662 struct ena_admin_host_info
*host_info
=
2663 adapter
->ena_dev
->host_attr
.host_info
;
2665 check_for_missing_keep_alive(adapter
);
2667 check_for_admin_com_state(adapter
);
2669 check_for_missing_tx_completions(adapter
);
2672 ena_dump_stats_to_buf(adapter
, debug_area
);
2675 ena_update_host_info(host_info
, adapter
->netdev
);
2677 if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET
, &adapter
->flags
))) {
2678 netif_err(adapter
, drv
, adapter
->netdev
,
2679 "Trigger reset is on\n");
2680 ena_dump_stats_to_dmesg(adapter
);
2681 queue_work(ena_wq
, &adapter
->reset_task
);
2685 /* Reset the timer */
2686 mod_timer(&adapter
->timer_service
, jiffies
+ HZ
);
2689 static int ena_calc_io_queue_num(struct pci_dev
*pdev
,
2690 struct ena_com_dev
*ena_dev
,
2691 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2693 int io_sq_num
, io_queue_num
;
2695 /* In case of LLQ use the llq number in the get feature cmd */
2696 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2697 io_sq_num
= get_feat_ctx
->max_queues
.max_llq_num
;
2699 if (io_sq_num
== 0) {
2701 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2703 ena_dev
->tx_mem_queue_type
=
2704 ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2705 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2708 io_sq_num
= get_feat_ctx
->max_queues
.max_sq_num
;
2711 io_queue_num
= min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES
);
2712 io_queue_num
= min_t(int, io_queue_num
, io_sq_num
);
2713 io_queue_num
= min_t(int, io_queue_num
,
2714 get_feat_ctx
->max_queues
.max_cq_num
);
2715 /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2716 io_queue_num
= min_t(int, io_queue_num
, pci_msix_vec_count(pdev
) - 1);
2717 if (unlikely(!io_queue_num
)) {
2718 dev_err(&pdev
->dev
, "The device doesn't have io queues\n");
2722 return io_queue_num
;
2725 static void ena_set_push_mode(struct pci_dev
*pdev
, struct ena_com_dev
*ena_dev
,
2726 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2730 has_mem_bar
= pci_select_bars(pdev
, IORESOURCE_MEM
) & BIT(ENA_MEM_BAR
);
2732 /* Enable push mode if device supports LLQ */
2733 if (has_mem_bar
&& (get_feat_ctx
->max_queues
.max_llq_num
> 0))
2734 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_DEV
;
2736 ena_dev
->tx_mem_queue_type
= ENA_ADMIN_PLACEMENT_POLICY_HOST
;
2739 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx
*feat
,
2740 struct net_device
*netdev
)
2742 netdev_features_t dev_features
= 0;
2744 /* Set offload features */
2745 if (feat
->offload
.tx
&
2746 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK
)
2747 dev_features
|= NETIF_F_IP_CSUM
;
2749 if (feat
->offload
.tx
&
2750 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK
)
2751 dev_features
|= NETIF_F_IPV6_CSUM
;
2753 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK
)
2754 dev_features
|= NETIF_F_TSO
;
2756 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK
)
2757 dev_features
|= NETIF_F_TSO6
;
2759 if (feat
->offload
.tx
& ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK
)
2760 dev_features
|= NETIF_F_TSO_ECN
;
2762 if (feat
->offload
.rx_supported
&
2763 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK
)
2764 dev_features
|= NETIF_F_RXCSUM
;
2766 if (feat
->offload
.rx_supported
&
2767 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK
)
2768 dev_features
|= NETIF_F_RXCSUM
;
2776 netdev
->hw_features
|= netdev
->features
;
2777 netdev
->vlan_features
|= netdev
->features
;
2780 static void ena_set_conf_feat_params(struct ena_adapter
*adapter
,
2781 struct ena_com_dev_get_features_ctx
*feat
)
2783 struct net_device
*netdev
= adapter
->netdev
;
2785 /* Copy mac address */
2786 if (!is_valid_ether_addr(feat
->dev_attr
.mac_addr
)) {
2787 eth_hw_addr_random(netdev
);
2788 ether_addr_copy(adapter
->mac_addr
, netdev
->dev_addr
);
2790 ether_addr_copy(adapter
->mac_addr
, feat
->dev_attr
.mac_addr
);
2791 ether_addr_copy(netdev
->dev_addr
, adapter
->mac_addr
);
2794 /* Set offload features */
2795 ena_set_dev_offloads(feat
, netdev
);
2797 adapter
->max_mtu
= feat
->dev_attr
.max_mtu
;
2798 netdev
->max_mtu
= adapter
->max_mtu
;
2799 netdev
->min_mtu
= ENA_MIN_MTU
;
2802 static int ena_rss_init_default(struct ena_adapter
*adapter
)
2804 struct ena_com_dev
*ena_dev
= adapter
->ena_dev
;
2805 struct device
*dev
= &adapter
->pdev
->dev
;
2809 rc
= ena_com_rss_init(ena_dev
, ENA_RX_RSS_TABLE_LOG_SIZE
);
2811 dev_err(dev
, "Cannot init indirect table\n");
2815 for (i
= 0; i
< ENA_RX_RSS_TABLE_SIZE
; i
++) {
2816 val
= ethtool_rxfh_indir_default(i
, adapter
->num_queues
);
2817 rc
= ena_com_indirect_table_fill_entry(ena_dev
, i
,
2818 ENA_IO_RXQ_IDX(val
));
2819 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2820 dev_err(dev
, "Cannot fill indirect table\n");
2821 goto err_fill_indir
;
2825 rc
= ena_com_fill_hash_function(ena_dev
, ENA_ADMIN_CRC32
, NULL
,
2826 ENA_HASH_KEY_SIZE
, 0xFFFFFFFF);
2827 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2828 dev_err(dev
, "Cannot fill hash function\n");
2829 goto err_fill_indir
;
2832 rc
= ena_com_set_default_hash_ctrl(ena_dev
);
2833 if (unlikely(rc
&& (rc
!= -EPERM
))) {
2834 dev_err(dev
, "Cannot fill hash control\n");
2835 goto err_fill_indir
;
2841 ena_com_rss_destroy(ena_dev
);
2847 static void ena_release_bars(struct ena_com_dev
*ena_dev
, struct pci_dev
*pdev
)
2851 release_bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2852 pci_release_selected_regions(pdev
, release_bars
);
2855 static int ena_calc_queue_size(struct pci_dev
*pdev
,
2856 struct ena_com_dev
*ena_dev
,
2857 u16
*max_tx_sgl_size
,
2858 u16
*max_rx_sgl_size
,
2859 struct ena_com_dev_get_features_ctx
*get_feat_ctx
)
2861 u32 queue_size
= ENA_DEFAULT_RING_SIZE
;
2863 queue_size
= min_t(u32
, queue_size
,
2864 get_feat_ctx
->max_queues
.max_cq_depth
);
2865 queue_size
= min_t(u32
, queue_size
,
2866 get_feat_ctx
->max_queues
.max_sq_depth
);
2868 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
)
2869 queue_size
= min_t(u32
, queue_size
,
2870 get_feat_ctx
->max_queues
.max_llq_depth
);
2872 queue_size
= rounddown_pow_of_two(queue_size
);
2874 if (unlikely(!queue_size
)) {
2875 dev_err(&pdev
->dev
, "Invalid queue size\n");
2879 *max_tx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2880 get_feat_ctx
->max_queues
.max_packet_tx_descs
);
2881 *max_rx_sgl_size
= min_t(u16
, ENA_PKT_MAX_BUFS
,
2882 get_feat_ctx
->max_queues
.max_packet_rx_descs
);
2887 /* ena_probe - Device Initialization Routine
2888 * @pdev: PCI device information struct
2889 * @ent: entry in ena_pci_tbl
2891 * Returns 0 on success, negative on failure
2893 * ena_probe initializes an adapter identified by a pci_dev structure.
2894 * The OS initialization, configuring of the adapter private structure,
2895 * and a hardware reset occur.
2897 static int ena_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
2899 struct ena_com_dev_get_features_ctx get_feat_ctx
;
2900 static int version_printed
;
2901 struct net_device
*netdev
;
2902 struct ena_adapter
*adapter
;
2903 struct ena_com_dev
*ena_dev
= NULL
;
2904 static int adapters_found
;
2905 int io_queue_num
, bars
, rc
;
2907 u16 tx_sgl_size
= 0;
2908 u16 rx_sgl_size
= 0;
2911 dev_dbg(&pdev
->dev
, "%s\n", __func__
);
2913 if (version_printed
++ == 0)
2914 dev_info(&pdev
->dev
, "%s", version
);
2916 rc
= pci_enable_device_mem(pdev
);
2918 dev_err(&pdev
->dev
, "pci_enable_device_mem() failed!\n");
2922 pci_set_master(pdev
);
2924 ena_dev
= vzalloc(sizeof(*ena_dev
));
2927 goto err_disable_device
;
2930 bars
= pci_select_bars(pdev
, IORESOURCE_MEM
) & ENA_BAR_MASK
;
2931 rc
= pci_request_selected_regions(pdev
, bars
, DRV_MODULE_NAME
);
2933 dev_err(&pdev
->dev
, "pci_request_selected_regions failed %d\n",
2935 goto err_free_ena_dev
;
2938 ena_dev
->reg_bar
= ioremap(pci_resource_start(pdev
, ENA_REG_BAR
),
2939 pci_resource_len(pdev
, ENA_REG_BAR
));
2940 if (!ena_dev
->reg_bar
) {
2941 dev_err(&pdev
->dev
, "failed to remap regs bar\n");
2943 goto err_free_region
;
2946 ena_dev
->dmadev
= &pdev
->dev
;
2948 rc
= ena_device_init(ena_dev
, pdev
, &get_feat_ctx
, &wd_state
);
2950 dev_err(&pdev
->dev
, "ena device init failed\n");
2953 goto err_free_region
;
2956 ena_set_push_mode(pdev
, ena_dev
, &get_feat_ctx
);
2958 if (ena_dev
->tx_mem_queue_type
== ENA_ADMIN_PLACEMENT_POLICY_DEV
) {
2959 ena_dev
->mem_bar
= ioremap_wc(pci_resource_start(pdev
, ENA_MEM_BAR
),
2960 pci_resource_len(pdev
, ENA_MEM_BAR
));
2961 if (!ena_dev
->mem_bar
) {
2963 goto err_device_destroy
;
2967 /* initial Tx interrupt delay, Assumes 1 usec granularity.
2968 * Updated during device initialization with the real granularity
2970 ena_dev
->intr_moder_tx_interval
= ENA_INTR_INITIAL_TX_INTERVAL_USECS
;
2971 io_queue_num
= ena_calc_io_queue_num(pdev
, ena_dev
, &get_feat_ctx
);
2972 queue_size
= ena_calc_queue_size(pdev
, ena_dev
, &tx_sgl_size
,
2973 &rx_sgl_size
, &get_feat_ctx
);
2974 if ((queue_size
<= 0) || (io_queue_num
<= 0)) {
2976 goto err_device_destroy
;
2979 dev_info(&pdev
->dev
, "creating %d io queues. queue size: %d\n",
2980 io_queue_num
, queue_size
);
2982 /* dev zeroed in init_etherdev */
2983 netdev
= alloc_etherdev_mq(sizeof(struct ena_adapter
), io_queue_num
);
2985 dev_err(&pdev
->dev
, "alloc_etherdev_mq failed\n");
2987 goto err_device_destroy
;
2990 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
2992 adapter
= netdev_priv(netdev
);
2993 pci_set_drvdata(pdev
, adapter
);
2995 adapter
->ena_dev
= ena_dev
;
2996 adapter
->netdev
= netdev
;
2997 adapter
->pdev
= pdev
;
2999 ena_set_conf_feat_params(adapter
, &get_feat_ctx
);
3001 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
3003 adapter
->tx_ring_size
= queue_size
;
3004 adapter
->rx_ring_size
= queue_size
;
3006 adapter
->max_tx_sgl_size
= tx_sgl_size
;
3007 adapter
->max_rx_sgl_size
= rx_sgl_size
;
3009 adapter
->num_queues
= io_queue_num
;
3010 adapter
->last_monitored_tx_qid
= 0;
3012 adapter
->rx_copybreak
= ENA_DEFAULT_RX_COPYBREAK
;
3013 adapter
->wd_state
= wd_state
;
3015 snprintf(adapter
->name
, ENA_NAME_MAX_LEN
, "ena_%d", adapters_found
);
3017 rc
= ena_com_init_interrupt_moderation(adapter
->ena_dev
);
3020 "Failed to query interrupt moderation feature\n");
3021 goto err_netdev_destroy
;
3023 ena_init_io_rings(adapter
);
3025 netdev
->netdev_ops
= &ena_netdev_ops
;
3026 netdev
->watchdog_timeo
= TX_TIMEOUT
;
3027 ena_set_ethtool_ops(netdev
);
3029 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
3031 u64_stats_init(&adapter
->syncp
);
3033 rc
= ena_enable_msix_and_set_admin_interrupts(adapter
, io_queue_num
);
3036 "Failed to enable and set the admin interrupts\n");
3037 goto err_worker_destroy
;
3039 rc
= ena_rss_init_default(adapter
);
3040 if (rc
&& (rc
!= -EPERM
)) {
3041 dev_err(&pdev
->dev
, "Cannot init RSS rc: %d\n", rc
);
3045 ena_config_debug_area(adapter
);
3047 memcpy(adapter
->netdev
->perm_addr
, adapter
->mac_addr
, netdev
->addr_len
);
3049 netif_carrier_off(netdev
);
3051 rc
= register_netdev(netdev
);
3053 dev_err(&pdev
->dev
, "Cannot register net device\n");
3057 INIT_WORK(&adapter
->suspend_io_task
, ena_device_io_suspend
);
3058 INIT_WORK(&adapter
->resume_io_task
, ena_device_io_resume
);
3059 INIT_WORK(&adapter
->reset_task
, ena_fw_reset_device
);
3061 adapter
->last_keep_alive_jiffies
= jiffies
;
3063 setup_timer(&adapter
->timer_service
, ena_timer_service
,
3064 (unsigned long)adapter
);
3065 mod_timer(&adapter
->timer_service
, round_jiffies(jiffies
+ HZ
));
3067 dev_info(&pdev
->dev
, "%s found at mem %lx, mac addr %pM Queues %d\n",
3068 DEVICE_NAME
, (long)pci_resource_start(pdev
, 0),
3069 netdev
->dev_addr
, io_queue_num
);
3071 set_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
);
3078 ena_com_delete_debug_area(ena_dev
);
3079 ena_com_rss_destroy(ena_dev
);
3081 ena_com_dev_reset(ena_dev
);
3082 ena_free_mgmnt_irq(adapter
);
3083 ena_disable_msix(adapter
);
3085 ena_com_destroy_interrupt_moderation(ena_dev
);
3086 del_timer(&adapter
->timer_service
);
3087 cancel_work_sync(&adapter
->suspend_io_task
);
3088 cancel_work_sync(&adapter
->resume_io_task
);
3090 free_netdev(netdev
);
3092 ena_com_delete_host_info(ena_dev
);
3093 ena_com_admin_destroy(ena_dev
);
3095 ena_release_bars(ena_dev
, pdev
);
3099 pci_disable_device(pdev
);
3103 /*****************************************************************************/
3104 static int ena_sriov_configure(struct pci_dev
*dev
, int numvfs
)
3109 rc
= pci_enable_sriov(dev
, numvfs
);
3112 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3121 pci_disable_sriov(dev
);
3128 /*****************************************************************************/
3129 /*****************************************************************************/
3131 /* ena_remove - Device Removal Routine
3132 * @pdev: PCI device information struct
3134 * ena_remove is called by the PCI subsystem to alert the driver
3135 * that it should release a PCI device.
3137 static void ena_remove(struct pci_dev
*pdev
)
3139 struct ena_adapter
*adapter
= pci_get_drvdata(pdev
);
3140 struct ena_com_dev
*ena_dev
;
3141 struct net_device
*netdev
;
3144 /* This device didn't load properly and it's resources
3145 * already released, nothing to do
3149 ena_dev
= adapter
->ena_dev
;
3150 netdev
= adapter
->netdev
;
3152 #ifdef CONFIG_RFS_ACCEL
3153 if ((adapter
->msix_vecs
>= 1) && (netdev
->rx_cpu_rmap
)) {
3154 free_irq_cpu_rmap(netdev
->rx_cpu_rmap
);
3155 netdev
->rx_cpu_rmap
= NULL
;
3157 #endif /* CONFIG_RFS_ACCEL */
3159 unregister_netdev(netdev
);
3160 del_timer_sync(&adapter
->timer_service
);
3162 cancel_work_sync(&adapter
->reset_task
);
3164 cancel_work_sync(&adapter
->suspend_io_task
);
3166 cancel_work_sync(&adapter
->resume_io_task
);
3168 /* Reset the device only if the device is running. */
3169 if (test_bit(ENA_FLAG_DEVICE_RUNNING
, &adapter
->flags
))
3170 ena_com_dev_reset(ena_dev
);
3172 ena_free_mgmnt_irq(adapter
);
3174 ena_disable_msix(adapter
);
3176 free_netdev(netdev
);
3178 ena_com_mmio_reg_read_request_destroy(ena_dev
);
3180 ena_com_abort_admin_commands(ena_dev
);
3182 ena_com_wait_for_abort_completion(ena_dev
);
3184 ena_com_admin_destroy(ena_dev
);
3186 ena_com_rss_destroy(ena_dev
);
3188 ena_com_delete_debug_area(ena_dev
);
3190 ena_com_delete_host_info(ena_dev
);
3192 ena_release_bars(ena_dev
, pdev
);
3194 pci_disable_device(pdev
);
3196 ena_com_destroy_interrupt_moderation(ena_dev
);
3201 static struct pci_driver ena_pci_driver
= {
3202 .name
= DRV_MODULE_NAME
,
3203 .id_table
= ena_pci_tbl
,
3205 .remove
= ena_remove
,
3206 .sriov_configure
= ena_sriov_configure
,
3209 static int __init
ena_init(void)
3211 pr_info("%s", version
);
3213 ena_wq
= create_singlethread_workqueue(DRV_MODULE_NAME
);
3215 pr_err("Failed to create workqueue\n");
3219 return pci_register_driver(&ena_pci_driver
);
3222 static void __exit
ena_cleanup(void)
3224 pci_unregister_driver(&ena_pci_driver
);
3227 destroy_workqueue(ena_wq
);
3232 /******************************************************************************
3233 ******************************** AENQ Handlers *******************************
3234 *****************************************************************************/
3235 /* ena_update_on_link_change:
3236 * Notify the network interface about the change in link status
3238 static void ena_update_on_link_change(void *adapter_data
,
3239 struct ena_admin_aenq_entry
*aenq_e
)
3241 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3242 struct ena_admin_aenq_link_change_desc
*aenq_desc
=
3243 (struct ena_admin_aenq_link_change_desc
*)aenq_e
;
3244 int status
= aenq_desc
->flags
&
3245 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK
;
3248 netdev_dbg(adapter
->netdev
, "%s\n", __func__
);
3249 set_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3250 netif_carrier_on(adapter
->netdev
);
3252 clear_bit(ENA_FLAG_LINK_UP
, &adapter
->flags
);
3253 netif_carrier_off(adapter
->netdev
);
3257 static void ena_keep_alive_wd(void *adapter_data
,
3258 struct ena_admin_aenq_entry
*aenq_e
)
3260 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3262 adapter
->last_keep_alive_jiffies
= jiffies
;
3265 static void ena_notification(void *adapter_data
,
3266 struct ena_admin_aenq_entry
*aenq_e
)
3268 struct ena_adapter
*adapter
= (struct ena_adapter
*)adapter_data
;
3270 WARN(aenq_e
->aenq_common_desc
.group
!= ENA_ADMIN_NOTIFICATION
,
3271 "Invalid group(%x) expected %x\n",
3272 aenq_e
->aenq_common_desc
.group
,
3273 ENA_ADMIN_NOTIFICATION
);
3275 switch (aenq_e
->aenq_common_desc
.syndrom
) {
3276 case ENA_ADMIN_SUSPEND
:
3277 /* Suspend just the IO queues.
3278 * We deliberately don't suspend admin so the timer and
3279 * the keep_alive events should remain.
3281 queue_work(ena_wq
, &adapter
->suspend_io_task
);
3283 case ENA_ADMIN_RESUME
:
3284 queue_work(ena_wq
, &adapter
->resume_io_task
);
3287 netif_err(adapter
, drv
, adapter
->netdev
,
3288 "Invalid aenq notification link state %d\n",
3289 aenq_e
->aenq_common_desc
.syndrom
);
3293 /* This handler will called for unknown event group or unimplemented handlers*/
3294 static void unimplemented_aenq_handler(void *data
,
3295 struct ena_admin_aenq_entry
*aenq_e
)
3297 struct ena_adapter
*adapter
= (struct ena_adapter
*)data
;
3299 netif_err(adapter
, drv
, adapter
->netdev
,
3300 "Unknown event was received or event with unimplemented handler\n");
3303 static struct ena_aenq_handlers aenq_handlers
= {
3305 [ENA_ADMIN_LINK_CHANGE
] = ena_update_on_link_change
,
3306 [ENA_ADMIN_NOTIFICATION
] = ena_notification
,
3307 [ENA_ADMIN_KEEP_ALIVE
] = ena_keep_alive_wd
,
3309 .unimplemented_handler
= unimplemented_aenq_handler
3312 module_init(ena_init
);
3313 module_exit(ena_cleanup
);