1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
4 * Copyright(c) 2013 - 2014 Intel Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 ******************************************************************************/
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
31 #include "i40e_prototype.h"
33 static inline __le64
build_ctob(u32 td_cmd
, u32 td_offset
, unsigned int size
,
36 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA
|
37 ((u64
)td_cmd
<< I40E_TXD_QW1_CMD_SHIFT
) |
38 ((u64
)td_offset
<< I40E_TXD_QW1_OFFSET_SHIFT
) |
39 ((u64
)size
<< I40E_TXD_QW1_TX_BUF_SZ_SHIFT
) |
40 ((u64
)td_tag
<< I40E_TXD_QW1_L2TAG1_SHIFT
));
43 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
46 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
47 * @ring: the ring that owns the buffer
48 * @tx_buffer: the buffer to free
50 static void i40e_unmap_and_free_tx_resource(struct i40e_ring
*ring
,
51 struct i40e_tx_buffer
*tx_buffer
)
54 if (tx_buffer
->tx_flags
& I40E_TX_FLAGS_FD_SB
)
55 kfree(tx_buffer
->raw_buf
);
57 dev_kfree_skb_any(tx_buffer
->skb
);
59 if (dma_unmap_len(tx_buffer
, len
))
60 dma_unmap_single(ring
->dev
,
61 dma_unmap_addr(tx_buffer
, dma
),
62 dma_unmap_len(tx_buffer
, len
),
64 } else if (dma_unmap_len(tx_buffer
, len
)) {
65 dma_unmap_page(ring
->dev
,
66 dma_unmap_addr(tx_buffer
, dma
),
67 dma_unmap_len(tx_buffer
, len
),
70 tx_buffer
->next_to_watch
= NULL
;
71 tx_buffer
->skb
= NULL
;
72 dma_unmap_len_set(tx_buffer
, len
, 0);
73 /* tx_buffer must be completely set up in the transmit path */
77 * i40evf_clean_tx_ring - Free any empty Tx buffers
78 * @tx_ring: ring to be cleaned
80 void i40evf_clean_tx_ring(struct i40e_ring
*tx_ring
)
82 unsigned long bi_size
;
85 /* ring already cleared, nothing to do */
89 /* Free all the Tx ring sk_buffs */
90 for (i
= 0; i
< tx_ring
->count
; i
++)
91 i40e_unmap_and_free_tx_resource(tx_ring
, &tx_ring
->tx_bi
[i
]);
93 bi_size
= sizeof(struct i40e_tx_buffer
) * tx_ring
->count
;
94 memset(tx_ring
->tx_bi
, 0, bi_size
);
96 /* Zero out the descriptor ring */
97 memset(tx_ring
->desc
, 0, tx_ring
->size
);
99 tx_ring
->next_to_use
= 0;
100 tx_ring
->next_to_clean
= 0;
102 if (!tx_ring
->netdev
)
105 /* cleanup Tx queue statistics */
106 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring
->netdev
,
107 tx_ring
->queue_index
));
111 * i40evf_free_tx_resources - Free Tx resources per queue
112 * @tx_ring: Tx descriptor ring for a specific queue
114 * Free all transmit software resources
116 void i40evf_free_tx_resources(struct i40e_ring
*tx_ring
)
118 i40evf_clean_tx_ring(tx_ring
);
119 kfree(tx_ring
->tx_bi
);
120 tx_ring
->tx_bi
= NULL
;
123 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
,
124 tx_ring
->desc
, tx_ring
->dma
);
125 tx_ring
->desc
= NULL
;
130 * i40e_get_tx_pending - how many tx descriptors not processed
131 * @tx_ring: the ring of descriptors
133 * Since there is no access to the ring head register
134 * in XL710, we need to use our local copies
136 static u32
i40e_get_tx_pending(struct i40e_ring
*ring
)
138 u32 ntu
= ((ring
->next_to_clean
<= ring
->next_to_use
)
140 : ring
->next_to_use
+ ring
->count
);
141 return ntu
- ring
->next_to_clean
;
145 * i40e_check_tx_hang - Is there a hang in the Tx queue
146 * @tx_ring: the ring of descriptors
148 static bool i40e_check_tx_hang(struct i40e_ring
*tx_ring
)
150 u32 tx_pending
= i40e_get_tx_pending(tx_ring
);
153 clear_check_for_tx_hang(tx_ring
);
155 /* Check for a hung queue, but be thorough. This verifies
156 * that a transmit has been completed since the previous
157 * check AND there is at least one packet pending. The
158 * ARMED bit is set to indicate a potential hang. The
159 * bit is cleared if a pause frame is received to remove
160 * false hang detection due to PFC or 802.3x frames. By
161 * requiring this to fail twice we avoid races with
162 * PFC clearing the ARMED bit and conditions where we
163 * run the check_tx_hang logic with a transmit completion
164 * pending but without time to complete it yet.
166 if ((tx_ring
->tx_stats
.tx_done_old
== tx_ring
->stats
.packets
) &&
167 (tx_pending
>= I40E_MIN_DESC_PENDING
)) {
168 /* make sure it is true for two checks in a row */
169 ret
= test_and_set_bit(__I40E_HANG_CHECK_ARMED
,
171 } else if (!(tx_ring
->tx_stats
.tx_done_old
== tx_ring
->stats
.packets
) ||
172 !(tx_pending
< I40E_MIN_DESC_PENDING
) ||
174 /* update completed stats and disarm the hang check */
175 tx_ring
->tx_stats
.tx_done_old
= tx_ring
->stats
.packets
;
176 clear_bit(__I40E_HANG_CHECK_ARMED
, &tx_ring
->state
);
183 * i40e_get_head - Retrieve head from head writeback
184 * @tx_ring: tx ring to fetch head of
186 * Returns value of Tx ring head based on value stored
187 * in head write-back location
189 static inline u32
i40e_get_head(struct i40e_ring
*tx_ring
)
191 void *head
= (struct i40e_tx_desc
*)tx_ring
->desc
+ tx_ring
->count
;
193 return le32_to_cpu(*(volatile __le32
*)head
);
196 #define WB_STRIDE 0x3
199 * i40e_clean_tx_irq - Reclaim resources after transmit completes
200 * @tx_ring: tx ring to clean
201 * @budget: how many cleans we're allowed
203 * Returns true if there's any budget left (e.g. the clean is finished)
205 static bool i40e_clean_tx_irq(struct i40e_ring
*tx_ring
, int budget
)
207 u16 i
= tx_ring
->next_to_clean
;
208 struct i40e_tx_buffer
*tx_buf
;
209 struct i40e_tx_desc
*tx_head
;
210 struct i40e_tx_desc
*tx_desc
;
211 unsigned int total_packets
= 0;
212 unsigned int total_bytes
= 0;
214 tx_buf
= &tx_ring
->tx_bi
[i
];
215 tx_desc
= I40E_TX_DESC(tx_ring
, i
);
218 tx_head
= I40E_TX_DESC(tx_ring
, i40e_get_head(tx_ring
));
221 struct i40e_tx_desc
*eop_desc
= tx_buf
->next_to_watch
;
223 /* if next_to_watch is not set then there is no work pending */
227 /* prevent any other reads prior to eop_desc */
228 read_barrier_depends();
230 /* we have caught up to head, no work left to do */
231 if (tx_head
== tx_desc
)
234 /* clear next_to_watch to prevent false hangs */
235 tx_buf
->next_to_watch
= NULL
;
237 /* update the statistics for this packet */
238 total_bytes
+= tx_buf
->bytecount
;
239 total_packets
+= tx_buf
->gso_segs
;
242 dev_kfree_skb_any(tx_buf
->skb
);
244 /* unmap skb header data */
245 dma_unmap_single(tx_ring
->dev
,
246 dma_unmap_addr(tx_buf
, dma
),
247 dma_unmap_len(tx_buf
, len
),
250 /* clear tx_buffer data */
252 dma_unmap_len_set(tx_buf
, len
, 0);
254 /* unmap remaining buffers */
255 while (tx_desc
!= eop_desc
) {
262 tx_buf
= tx_ring
->tx_bi
;
263 tx_desc
= I40E_TX_DESC(tx_ring
, 0);
266 /* unmap any remaining paged data */
267 if (dma_unmap_len(tx_buf
, len
)) {
268 dma_unmap_page(tx_ring
->dev
,
269 dma_unmap_addr(tx_buf
, dma
),
270 dma_unmap_len(tx_buf
, len
),
272 dma_unmap_len_set(tx_buf
, len
, 0);
276 /* move us one more past the eop_desc for start of next pkt */
282 tx_buf
= tx_ring
->tx_bi
;
283 tx_desc
= I40E_TX_DESC(tx_ring
, 0);
286 /* update budget accounting */
288 } while (likely(budget
));
291 tx_ring
->next_to_clean
= i
;
292 u64_stats_update_begin(&tx_ring
->syncp
);
293 tx_ring
->stats
.bytes
+= total_bytes
;
294 tx_ring
->stats
.packets
+= total_packets
;
295 u64_stats_update_end(&tx_ring
->syncp
);
296 tx_ring
->q_vector
->tx
.total_bytes
+= total_bytes
;
297 tx_ring
->q_vector
->tx
.total_packets
+= total_packets
;
300 !((i
& WB_STRIDE
) == WB_STRIDE
) &&
301 !test_bit(__I40E_DOWN
, &tx_ring
->vsi
->state
) &&
302 (I40E_DESC_UNUSED(tx_ring
) != tx_ring
->count
))
303 tx_ring
->arm_wb
= true;
305 tx_ring
->arm_wb
= false;
307 if (check_for_tx_hang(tx_ring
) && i40e_check_tx_hang(tx_ring
)) {
308 /* schedule immediate reset if we believe we hung */
309 dev_info(tx_ring
->dev
, "Detected Tx Unit Hang\n"
312 " next_to_use <%x>\n"
313 " next_to_clean <%x>\n",
315 tx_ring
->queue_index
,
316 tx_ring
->next_to_use
, i
);
317 dev_info(tx_ring
->dev
, "tx_bi[next_to_clean]\n"
318 " time_stamp <%lx>\n"
320 tx_ring
->tx_bi
[i
].time_stamp
, jiffies
);
322 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
324 dev_info(tx_ring
->dev
,
325 "tx hang detected on queue %d, resetting adapter\n",
326 tx_ring
->queue_index
);
328 tx_ring
->netdev
->netdev_ops
->ndo_tx_timeout(tx_ring
->netdev
);
330 /* the adapter is about to reset, no point in enabling stuff */
334 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring
->netdev
,
335 tx_ring
->queue_index
),
336 total_packets
, total_bytes
);
338 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
339 if (unlikely(total_packets
&& netif_carrier_ok(tx_ring
->netdev
) &&
340 (I40E_DESC_UNUSED(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
341 /* Make sure that anybody stopping the queue after this
342 * sees the new next_to_clean.
345 if (__netif_subqueue_stopped(tx_ring
->netdev
,
346 tx_ring
->queue_index
) &&
347 !test_bit(__I40E_DOWN
, &tx_ring
->vsi
->state
)) {
348 netif_wake_subqueue(tx_ring
->netdev
,
349 tx_ring
->queue_index
);
350 ++tx_ring
->tx_stats
.restart_queue
;
358 * i40e_force_wb -Arm hardware to do a wb on noncache aligned descriptors
359 * @vsi: the VSI we care about
360 * @q_vector: the vector on which to force writeback
363 static void i40e_force_wb(struct i40e_vsi
*vsi
, struct i40e_q_vector
*q_vector
)
365 u32 val
= I40E_VFINT_DYN_CTLN_INTENA_MASK
|
366 I40E_VFINT_DYN_CTLN_SWINT_TRIG_MASK
|
367 I40E_VFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK
;
368 /* allow 00 to be written to the index */
371 I40E_VFINT_DYN_CTLN1(q_vector
->v_idx
+ vsi
->base_vector
- 1),
376 * i40e_set_new_dynamic_itr - Find new ITR level
377 * @rc: structure containing ring performance data
379 * Stores a new ITR value based on packets and byte counts during
380 * the last interrupt. The advantage of per interrupt computation
381 * is faster updates and more accurate ITR for the current traffic
382 * pattern. Constants in this function were computed based on
383 * theoretical maximum wire speed and thresholds were set based on
384 * testing data as well as attempting to minimize response time
385 * while increasing bulk throughput.
387 static void i40e_set_new_dynamic_itr(struct i40e_ring_container
*rc
)
389 enum i40e_latency_range new_latency_range
= rc
->latency_range
;
390 u32 new_itr
= rc
->itr
;
393 if (rc
->total_packets
== 0 || !rc
->itr
)
396 /* simple throttlerate management
397 * 0-10MB/s lowest (100000 ints/s)
398 * 10-20MB/s low (20000 ints/s)
399 * 20-1249MB/s bulk (8000 ints/s)
401 bytes_per_int
= rc
->total_bytes
/ rc
->itr
;
403 case I40E_LOWEST_LATENCY
:
404 if (bytes_per_int
> 10)
405 new_latency_range
= I40E_LOW_LATENCY
;
407 case I40E_LOW_LATENCY
:
408 if (bytes_per_int
> 20)
409 new_latency_range
= I40E_BULK_LATENCY
;
410 else if (bytes_per_int
<= 10)
411 new_latency_range
= I40E_LOWEST_LATENCY
;
413 case I40E_BULK_LATENCY
:
414 if (bytes_per_int
<= 20)
415 rc
->latency_range
= I40E_LOW_LATENCY
;
419 switch (new_latency_range
) {
420 case I40E_LOWEST_LATENCY
:
421 new_itr
= I40E_ITR_100K
;
423 case I40E_LOW_LATENCY
:
424 new_itr
= I40E_ITR_20K
;
426 case I40E_BULK_LATENCY
:
427 new_itr
= I40E_ITR_8K
;
433 if (new_itr
!= rc
->itr
) {
434 /* do an exponential smoothing */
435 new_itr
= (10 * new_itr
* rc
->itr
) /
436 ((9 * new_itr
) + rc
->itr
);
437 rc
->itr
= new_itr
& I40E_MAX_ITR
;
441 rc
->total_packets
= 0;
445 * i40e_update_dynamic_itr - Adjust ITR based on bytes per int
446 * @q_vector: the vector to adjust
448 static void i40e_update_dynamic_itr(struct i40e_q_vector
*q_vector
)
450 u16 vector
= q_vector
->vsi
->base_vector
+ q_vector
->v_idx
;
451 struct i40e_hw
*hw
= &q_vector
->vsi
->back
->hw
;
455 reg_addr
= I40E_VFINT_ITRN1(I40E_RX_ITR
, vector
- 1);
456 old_itr
= q_vector
->rx
.itr
;
457 i40e_set_new_dynamic_itr(&q_vector
->rx
);
458 if (old_itr
!= q_vector
->rx
.itr
)
459 wr32(hw
, reg_addr
, q_vector
->rx
.itr
);
461 reg_addr
= I40E_VFINT_ITRN1(I40E_TX_ITR
, vector
- 1);
462 old_itr
= q_vector
->tx
.itr
;
463 i40e_set_new_dynamic_itr(&q_vector
->tx
);
464 if (old_itr
!= q_vector
->tx
.itr
)
465 wr32(hw
, reg_addr
, q_vector
->tx
.itr
);
469 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
470 * @tx_ring: the tx ring to set up
472 * Return 0 on success, negative on error
474 int i40evf_setup_tx_descriptors(struct i40e_ring
*tx_ring
)
476 struct device
*dev
= tx_ring
->dev
;
482 bi_size
= sizeof(struct i40e_tx_buffer
) * tx_ring
->count
;
483 tx_ring
->tx_bi
= kzalloc(bi_size
, GFP_KERNEL
);
487 /* round up to nearest 4K */
488 tx_ring
->size
= tx_ring
->count
* sizeof(struct i40e_tx_desc
);
489 /* add u32 for head writeback, align after this takes care of
490 * guaranteeing this is at least one cache line in size
492 tx_ring
->size
+= sizeof(u32
);
493 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
494 tx_ring
->desc
= dma_alloc_coherent(dev
, tx_ring
->size
,
495 &tx_ring
->dma
, GFP_KERNEL
);
496 if (!tx_ring
->desc
) {
497 dev_info(dev
, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
502 tx_ring
->next_to_use
= 0;
503 tx_ring
->next_to_clean
= 0;
507 kfree(tx_ring
->tx_bi
);
508 tx_ring
->tx_bi
= NULL
;
513 * i40evf_clean_rx_ring - Free Rx buffers
514 * @rx_ring: ring to be cleaned
516 void i40evf_clean_rx_ring(struct i40e_ring
*rx_ring
)
518 struct device
*dev
= rx_ring
->dev
;
519 struct i40e_rx_buffer
*rx_bi
;
520 unsigned long bi_size
;
523 /* ring already cleared, nothing to do */
527 if (ring_is_ps_enabled(rx_ring
)) {
528 int bufsz
= ALIGN(rx_ring
->rx_hdr_len
, 256) * rx_ring
->count
;
530 rx_bi
= &rx_ring
->rx_bi
[0];
531 if (rx_bi
->hdr_buf
) {
532 dma_free_coherent(dev
,
536 for (i
= 0; i
< rx_ring
->count
; i
++) {
537 rx_bi
= &rx_ring
->rx_bi
[i
];
543 /* Free all the Rx ring sk_buffs */
544 for (i
= 0; i
< rx_ring
->count
; i
++) {
545 rx_bi
= &rx_ring
->rx_bi
[i
];
547 dma_unmap_single(dev
,
554 dev_kfree_skb(rx_bi
->skb
);
558 if (rx_bi
->page_dma
) {
565 __free_page(rx_bi
->page
);
567 rx_bi
->page_offset
= 0;
571 bi_size
= sizeof(struct i40e_rx_buffer
) * rx_ring
->count
;
572 memset(rx_ring
->rx_bi
, 0, bi_size
);
574 /* Zero out the descriptor ring */
575 memset(rx_ring
->desc
, 0, rx_ring
->size
);
577 rx_ring
->next_to_clean
= 0;
578 rx_ring
->next_to_use
= 0;
582 * i40evf_free_rx_resources - Free Rx resources
583 * @rx_ring: ring to clean the resources from
585 * Free all receive software resources
587 void i40evf_free_rx_resources(struct i40e_ring
*rx_ring
)
589 i40evf_clean_rx_ring(rx_ring
);
590 kfree(rx_ring
->rx_bi
);
591 rx_ring
->rx_bi
= NULL
;
594 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
,
595 rx_ring
->desc
, rx_ring
->dma
);
596 rx_ring
->desc
= NULL
;
601 * i40evf_alloc_rx_headers - allocate rx header buffers
602 * @rx_ring: ring to alloc buffers
604 * Allocate rx header buffers for the entire ring. As these are static,
605 * this is only called when setting up a new ring.
607 void i40evf_alloc_rx_headers(struct i40e_ring
*rx_ring
)
609 struct device
*dev
= rx_ring
->dev
;
610 struct i40e_rx_buffer
*rx_bi
;
616 if (rx_ring
->rx_bi
[0].hdr_buf
)
618 /* Make sure the buffers don't cross cache line boundaries. */
619 buf_size
= ALIGN(rx_ring
->rx_hdr_len
, 256);
620 buffer
= dma_alloc_coherent(dev
, buf_size
* rx_ring
->count
,
624 for (i
= 0; i
< rx_ring
->count
; i
++) {
625 rx_bi
= &rx_ring
->rx_bi
[i
];
626 rx_bi
->dma
= dma
+ (i
* buf_size
);
627 rx_bi
->hdr_buf
= buffer
+ (i
* buf_size
);
632 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
633 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
635 * Returns 0 on success, negative on failure
637 int i40evf_setup_rx_descriptors(struct i40e_ring
*rx_ring
)
639 struct device
*dev
= rx_ring
->dev
;
642 bi_size
= sizeof(struct i40e_rx_buffer
) * rx_ring
->count
;
643 rx_ring
->rx_bi
= kzalloc(bi_size
, GFP_KERNEL
);
647 u64_stats_init(&rx_ring
->syncp
);
649 /* Round up to nearest 4K */
650 rx_ring
->size
= ring_is_16byte_desc_enabled(rx_ring
)
651 ? rx_ring
->count
* sizeof(union i40e_16byte_rx_desc
)
652 : rx_ring
->count
* sizeof(union i40e_32byte_rx_desc
);
653 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
654 rx_ring
->desc
= dma_alloc_coherent(dev
, rx_ring
->size
,
655 &rx_ring
->dma
, GFP_KERNEL
);
657 if (!rx_ring
->desc
) {
658 dev_info(dev
, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
663 rx_ring
->next_to_clean
= 0;
664 rx_ring
->next_to_use
= 0;
668 kfree(rx_ring
->rx_bi
);
669 rx_ring
->rx_bi
= NULL
;
674 * i40e_release_rx_desc - Store the new tail and head values
675 * @rx_ring: ring to bump
676 * @val: new head index
678 static inline void i40e_release_rx_desc(struct i40e_ring
*rx_ring
, u32 val
)
680 rx_ring
->next_to_use
= val
;
681 /* Force memory writes to complete before letting h/w
682 * know there are new descriptors to fetch. (Only
683 * applicable for weak-ordered memory model archs,
687 writel(val
, rx_ring
->tail
);
691 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
692 * @rx_ring: ring to place buffers on
693 * @cleaned_count: number of buffers to replace
695 void i40evf_alloc_rx_buffers_ps(struct i40e_ring
*rx_ring
, u16 cleaned_count
)
697 u16 i
= rx_ring
->next_to_use
;
698 union i40e_rx_desc
*rx_desc
;
699 struct i40e_rx_buffer
*bi
;
701 /* do nothing if no valid netdev defined */
702 if (!rx_ring
->netdev
|| !cleaned_count
)
705 while (cleaned_count
--) {
706 rx_desc
= I40E_RX_DESC(rx_ring
, i
);
707 bi
= &rx_ring
->rx_bi
[i
];
709 if (bi
->skb
) /* desc is in use */
712 bi
->page
= alloc_page(GFP_ATOMIC
);
714 rx_ring
->rx_stats
.alloc_page_failed
++;
720 /* use a half page if we're re-using */
721 bi
->page_offset
^= PAGE_SIZE
/ 2;
722 bi
->page_dma
= dma_map_page(rx_ring
->dev
,
727 if (dma_mapping_error(rx_ring
->dev
,
729 rx_ring
->rx_stats
.alloc_page_failed
++;
735 dma_sync_single_range_for_device(rx_ring
->dev
,
740 /* Refresh the desc even if buffer_addrs didn't change
741 * because each write-back erases this info.
743 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->page_dma
);
744 rx_desc
->read
.hdr_addr
= cpu_to_le64(bi
->dma
);
746 if (i
== rx_ring
->count
)
751 if (rx_ring
->next_to_use
!= i
)
752 i40e_release_rx_desc(rx_ring
, i
);
756 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
757 * @rx_ring: ring to place buffers on
758 * @cleaned_count: number of buffers to replace
760 void i40evf_alloc_rx_buffers_1buf(struct i40e_ring
*rx_ring
, u16 cleaned_count
)
762 u16 i
= rx_ring
->next_to_use
;
763 union i40e_rx_desc
*rx_desc
;
764 struct i40e_rx_buffer
*bi
;
767 /* do nothing if no valid netdev defined */
768 if (!rx_ring
->netdev
|| !cleaned_count
)
771 while (cleaned_count
--) {
772 rx_desc
= I40E_RX_DESC(rx_ring
, i
);
773 bi
= &rx_ring
->rx_bi
[i
];
777 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
778 rx_ring
->rx_buf_len
);
780 rx_ring
->rx_stats
.alloc_buff_failed
++;
783 /* initialize queue mapping */
784 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
789 bi
->dma
= dma_map_single(rx_ring
->dev
,
793 if (dma_mapping_error(rx_ring
->dev
, bi
->dma
)) {
794 rx_ring
->rx_stats
.alloc_buff_failed
++;
800 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
);
801 rx_desc
->read
.hdr_addr
= 0;
803 if (i
== rx_ring
->count
)
808 if (rx_ring
->next_to_use
!= i
)
809 i40e_release_rx_desc(rx_ring
, i
);
813 * i40e_receive_skb - Send a completed packet up the stack
814 * @rx_ring: rx ring in play
815 * @skb: packet to send up
816 * @vlan_tag: vlan tag for packet
818 static void i40e_receive_skb(struct i40e_ring
*rx_ring
,
819 struct sk_buff
*skb
, u16 vlan_tag
)
821 struct i40e_q_vector
*q_vector
= rx_ring
->q_vector
;
822 struct i40e_vsi
*vsi
= rx_ring
->vsi
;
823 u64 flags
= vsi
->back
->flags
;
825 if (vlan_tag
& VLAN_VID_MASK
)
826 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vlan_tag
);
828 if (flags
& I40E_FLAG_IN_NETPOLL
)
831 napi_gro_receive(&q_vector
->napi
, skb
);
835 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
836 * @vsi: the VSI we care about
837 * @skb: skb currently being received and modified
838 * @rx_status: status value of last descriptor in packet
839 * @rx_error: error value of last descriptor in packet
840 * @rx_ptype: ptype value of last descriptor in packet
842 static inline void i40e_rx_checksum(struct i40e_vsi
*vsi
,
848 struct i40e_rx_ptype_decoded decoded
= decode_rx_desc_ptype(rx_ptype
);
849 bool ipv4
= false, ipv6
= false;
850 bool ipv4_tunnel
, ipv6_tunnel
;
855 ipv4_tunnel
= (rx_ptype
>= I40E_RX_PTYPE_GRENAT4_MAC_PAY3
) &&
856 (rx_ptype
<= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4
);
857 ipv6_tunnel
= (rx_ptype
>= I40E_RX_PTYPE_GRENAT6_MAC_PAY3
) &&
858 (rx_ptype
<= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4
);
860 skb
->ip_summed
= CHECKSUM_NONE
;
862 /* Rx csum enabled and ip headers found? */
863 if (!(vsi
->netdev
->features
& NETIF_F_RXCSUM
))
866 /* did the hardware decode the packet and checksum? */
867 if (!(rx_status
& (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT
)))
870 /* both known and outer_ip must be set for the below code to work */
871 if (!(decoded
.known
&& decoded
.outer_ip
))
874 if (decoded
.outer_ip
== I40E_RX_PTYPE_OUTER_IP
&&
875 decoded
.outer_ip_ver
== I40E_RX_PTYPE_OUTER_IPV4
)
877 else if (decoded
.outer_ip
== I40E_RX_PTYPE_OUTER_IP
&&
878 decoded
.outer_ip_ver
== I40E_RX_PTYPE_OUTER_IPV6
)
882 (rx_error
& ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT
) |
883 (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT
))))
886 /* likely incorrect csum if alternate IP extension headers found */
888 rx_status
& (1 << I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT
))
889 /* don't increment checksum err here, non-fatal err */
892 /* there was some L4 error, count error and punt packet to the stack */
893 if (rx_error
& (1 << I40E_RX_DESC_ERROR_L4E_SHIFT
))
896 /* handle packets that were not able to be checksummed due
897 * to arrival speed, in this case the stack can compute
900 if (rx_error
& (1 << I40E_RX_DESC_ERROR_PPRS_SHIFT
))
903 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
904 * it in the driver, hardware does not do it for us.
905 * Since L3L4P bit was set we assume a valid IHL value (>=5)
906 * so the total length of IPv4 header is IHL*4 bytes
907 * The UDP_0 bit *may* bet set if the *inner* header is UDP
910 (decoded
.inner_prot
!= I40E_RX_PTYPE_INNER_PROT_UDP
) &&
911 !(rx_status
& (1 << I40E_RX_DESC_STATUS_UDP_0_SHIFT
))) {
912 skb
->transport_header
= skb
->mac_header
+
913 sizeof(struct ethhdr
) +
914 (ip_hdr(skb
)->ihl
* 4);
916 /* Add 4 bytes for VLAN tagged packets */
917 skb
->transport_header
+= (skb
->protocol
== htons(ETH_P_8021Q
) ||
918 skb
->protocol
== htons(ETH_P_8021AD
))
921 rx_udp_csum
= udp_csum(skb
);
923 csum
= csum_tcpudp_magic(
924 iph
->saddr
, iph
->daddr
,
925 (skb
->len
- skb_transport_offset(skb
)),
926 IPPROTO_UDP
, rx_udp_csum
);
928 if (udp_hdr(skb
)->check
!= csum
)
932 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
933 skb
->csum_level
= ipv4_tunnel
|| ipv6_tunnel
;
938 vsi
->back
->hw_csum_rx_error
++;
942 * i40e_rx_hash - returns the hash value from the Rx descriptor
943 * @ring: descriptor ring
944 * @rx_desc: specific descriptor
946 static inline u32
i40e_rx_hash(struct i40e_ring
*ring
,
947 union i40e_rx_desc
*rx_desc
)
949 const __le64 rss_mask
=
950 cpu_to_le64((u64
)I40E_RX_DESC_FLTSTAT_RSS_HASH
<<
951 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT
);
953 if ((ring
->netdev
->features
& NETIF_F_RXHASH
) &&
954 (rx_desc
->wb
.qword1
.status_error_len
& rss_mask
) == rss_mask
)
955 return le32_to_cpu(rx_desc
->wb
.qword0
.hi_dword
.rss
);
961 * i40e_ptype_to_hash - get a hash type
962 * @ptype: the ptype value from the descriptor
964 * Returns a hash type to be used by skb_set_hash
966 static inline enum pkt_hash_types
i40e_ptype_to_hash(u8 ptype
)
968 struct i40e_rx_ptype_decoded decoded
= decode_rx_desc_ptype(ptype
);
971 return PKT_HASH_TYPE_NONE
;
973 if (decoded
.outer_ip
== I40E_RX_PTYPE_OUTER_IP
&&
974 decoded
.payload_layer
== I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4
)
975 return PKT_HASH_TYPE_L4
;
976 else if (decoded
.outer_ip
== I40E_RX_PTYPE_OUTER_IP
&&
977 decoded
.payload_layer
== I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3
)
978 return PKT_HASH_TYPE_L3
;
980 return PKT_HASH_TYPE_L2
;
984 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
985 * @rx_ring: rx ring to clean
986 * @budget: how many cleans we're allowed
988 * Returns true if there's any budget left (e.g. the clean is finished)
990 static int i40e_clean_rx_irq_ps(struct i40e_ring
*rx_ring
, int budget
)
992 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
993 u16 rx_packet_len
, rx_header_len
, rx_sph
, rx_hbo
;
994 u16 cleaned_count
= I40E_DESC_UNUSED(rx_ring
);
995 const int current_node
= numa_node_id();
996 struct i40e_vsi
*vsi
= rx_ring
->vsi
;
997 u16 i
= rx_ring
->next_to_clean
;
998 union i40e_rx_desc
*rx_desc
;
999 u32 rx_error
, rx_status
;
1004 struct i40e_rx_buffer
*rx_bi
;
1005 struct sk_buff
*skb
;
1007 /* return some buffers to hardware, one at a time is too slow */
1008 if (cleaned_count
>= I40E_RX_BUFFER_WRITE
) {
1009 i40evf_alloc_rx_buffers_ps(rx_ring
, cleaned_count
);
1013 i
= rx_ring
->next_to_clean
;
1014 rx_desc
= I40E_RX_DESC(rx_ring
, i
);
1015 qword
= le64_to_cpu(rx_desc
->wb
.qword1
.status_error_len
);
1016 rx_status
= (qword
& I40E_RXD_QW1_STATUS_MASK
) >>
1017 I40E_RXD_QW1_STATUS_SHIFT
;
1019 if (!(rx_status
& (1 << I40E_RX_DESC_STATUS_DD_SHIFT
)))
1022 /* This memory barrier is needed to keep us from reading
1023 * any other fields out of the rx_desc until we know the
1027 rx_bi
= &rx_ring
->rx_bi
[i
];
1030 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
1031 rx_ring
->rx_hdr_len
);
1033 rx_ring
->rx_stats
.alloc_buff_failed
++;
1034 /* initialize queue mapping */
1035 skb_record_rx_queue(skb
, rx_ring
->queue_index
);
1036 /* we are reusing so sync this buffer for CPU use */
1037 dma_sync_single_range_for_cpu(rx_ring
->dev
,
1040 rx_ring
->rx_hdr_len
,
1043 rx_packet_len
= (qword
& I40E_RXD_QW1_LENGTH_PBUF_MASK
) >>
1044 I40E_RXD_QW1_LENGTH_PBUF_SHIFT
;
1045 rx_header_len
= (qword
& I40E_RXD_QW1_LENGTH_HBUF_MASK
) >>
1046 I40E_RXD_QW1_LENGTH_HBUF_SHIFT
;
1047 rx_sph
= (qword
& I40E_RXD_QW1_LENGTH_SPH_MASK
) >>
1048 I40E_RXD_QW1_LENGTH_SPH_SHIFT
;
1050 rx_error
= (qword
& I40E_RXD_QW1_ERROR_MASK
) >>
1051 I40E_RXD_QW1_ERROR_SHIFT
;
1052 rx_hbo
= rx_error
& (1 << I40E_RX_DESC_ERROR_HBO_SHIFT
);
1053 rx_error
&= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT
);
1055 rx_ptype
= (qword
& I40E_RXD_QW1_PTYPE_MASK
) >>
1056 I40E_RXD_QW1_PTYPE_SHIFT
;
1057 prefetch(rx_bi
->page
);
1060 if (rx_hbo
|| rx_sph
) {
1063 len
= I40E_RX_HDR_SIZE
;
1065 len
= rx_header_len
;
1066 memcpy(__skb_put(skb
, len
), rx_bi
->hdr_buf
, len
);
1067 } else if (skb
->len
== 0) {
1070 len
= (rx_packet_len
> skb_headlen(skb
) ?
1071 skb_headlen(skb
) : rx_packet_len
);
1072 memcpy(__skb_put(skb
, len
),
1073 rx_bi
->page
+ rx_bi
->page_offset
,
1075 rx_bi
->page_offset
+= len
;
1076 rx_packet_len
-= len
;
1079 /* Get the rest of the data if this was a header split */
1080 if (rx_packet_len
) {
1081 skb_fill_page_desc(skb
, skb_shinfo(skb
)->nr_frags
,
1086 skb
->len
+= rx_packet_len
;
1087 skb
->data_len
+= rx_packet_len
;
1088 skb
->truesize
+= rx_packet_len
;
1090 if ((page_count(rx_bi
->page
) == 1) &&
1091 (page_to_nid(rx_bi
->page
) == current_node
))
1092 get_page(rx_bi
->page
);
1096 dma_unmap_page(rx_ring
->dev
,
1100 rx_bi
->page_dma
= 0;
1102 I40E_RX_INCREMENT(rx_ring
, i
);
1105 !(rx_status
& (1 << I40E_RX_DESC_STATUS_EOF_SHIFT
)))) {
1106 struct i40e_rx_buffer
*next_buffer
;
1108 next_buffer
= &rx_ring
->rx_bi
[i
];
1109 next_buffer
->skb
= skb
;
1110 rx_ring
->rx_stats
.non_eop_descs
++;
1114 /* ERR_MASK will only have valid bits if EOP set */
1115 if (unlikely(rx_error
& (1 << I40E_RX_DESC_ERROR_RXE_SHIFT
))) {
1116 dev_kfree_skb_any(skb
);
1117 /* TODO: shouldn't we increment a counter indicating the
1123 skb_set_hash(skb
, i40e_rx_hash(rx_ring
, rx_desc
),
1124 i40e_ptype_to_hash(rx_ptype
));
1125 /* probably a little skewed due to removing CRC */
1126 total_rx_bytes
+= skb
->len
;
1129 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
1131 i40e_rx_checksum(vsi
, skb
, rx_status
, rx_error
, rx_ptype
);
1133 vlan_tag
= rx_status
& (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT
)
1134 ? le16_to_cpu(rx_desc
->wb
.qword0
.lo_dword
.l2tag1
)
1137 if (!i40e_fcoe_handle_offload(rx_ring
, rx_desc
, skb
)) {
1138 dev_kfree_skb_any(skb
);
1142 skb_mark_napi_id(skb
, &rx_ring
->q_vector
->napi
);
1143 i40e_receive_skb(rx_ring
, skb
, vlan_tag
);
1145 rx_ring
->netdev
->last_rx
= jiffies
;
1146 rx_desc
->wb
.qword1
.status_error_len
= 0;
1148 } while (likely(total_rx_packets
< budget
));
1150 u64_stats_update_begin(&rx_ring
->syncp
);
1151 rx_ring
->stats
.packets
+= total_rx_packets
;
1152 rx_ring
->stats
.bytes
+= total_rx_bytes
;
1153 u64_stats_update_end(&rx_ring
->syncp
);
1154 rx_ring
->q_vector
->rx
.total_packets
+= total_rx_packets
;
1155 rx_ring
->q_vector
->rx
.total_bytes
+= total_rx_bytes
;
1157 return total_rx_packets
;
1161 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1162 * @rx_ring: rx ring to clean
1163 * @budget: how many cleans we're allowed
1165 * Returns number of packets cleaned
1167 static int i40e_clean_rx_irq_1buf(struct i40e_ring
*rx_ring
, int budget
)
1169 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
1170 u16 cleaned_count
= I40E_DESC_UNUSED(rx_ring
);
1171 struct i40e_vsi
*vsi
= rx_ring
->vsi
;
1172 union i40e_rx_desc
*rx_desc
;
1173 u32 rx_error
, rx_status
;
1180 struct i40e_rx_buffer
*rx_bi
;
1181 struct sk_buff
*skb
;
1183 /* return some buffers to hardware, one at a time is too slow */
1184 if (cleaned_count
>= I40E_RX_BUFFER_WRITE
) {
1185 i40evf_alloc_rx_buffers_1buf(rx_ring
, cleaned_count
);
1189 i
= rx_ring
->next_to_clean
;
1190 rx_desc
= I40E_RX_DESC(rx_ring
, i
);
1191 qword
= le64_to_cpu(rx_desc
->wb
.qword1
.status_error_len
);
1192 rx_status
= (qword
& I40E_RXD_QW1_STATUS_MASK
) >>
1193 I40E_RXD_QW1_STATUS_SHIFT
;
1195 if (!(rx_status
& (1 << I40E_RX_DESC_STATUS_DD_SHIFT
)))
1198 /* This memory barrier is needed to keep us from reading
1199 * any other fields out of the rx_desc until we know the
1204 rx_bi
= &rx_ring
->rx_bi
[i
];
1206 prefetch(skb
->data
);
1208 rx_packet_len
= (qword
& I40E_RXD_QW1_LENGTH_PBUF_MASK
) >>
1209 I40E_RXD_QW1_LENGTH_PBUF_SHIFT
;
1211 rx_error
= (qword
& I40E_RXD_QW1_ERROR_MASK
) >>
1212 I40E_RXD_QW1_ERROR_SHIFT
;
1213 rx_error
&= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT
);
1215 rx_ptype
= (qword
& I40E_RXD_QW1_PTYPE_MASK
) >>
1216 I40E_RXD_QW1_PTYPE_SHIFT
;
1220 /* Get the header and possibly the whole packet
1221 * If this is an skb from previous receive dma will be 0
1223 skb_put(skb
, rx_packet_len
);
1224 dma_unmap_single(rx_ring
->dev
, rx_bi
->dma
, rx_ring
->rx_buf_len
,
1228 I40E_RX_INCREMENT(rx_ring
, i
);
1231 !(rx_status
& (1 << I40E_RX_DESC_STATUS_EOF_SHIFT
)))) {
1232 rx_ring
->rx_stats
.non_eop_descs
++;
1236 /* ERR_MASK will only have valid bits if EOP set */
1237 if (unlikely(rx_error
& (1 << I40E_RX_DESC_ERROR_RXE_SHIFT
))) {
1238 dev_kfree_skb_any(skb
);
1239 /* TODO: shouldn't we increment a counter indicating the
1245 skb_set_hash(skb
, i40e_rx_hash(rx_ring
, rx_desc
),
1246 i40e_ptype_to_hash(rx_ptype
));
1247 /* probably a little skewed due to removing CRC */
1248 total_rx_bytes
+= skb
->len
;
1251 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
1253 i40e_rx_checksum(vsi
, skb
, rx_status
, rx_error
, rx_ptype
);
1255 vlan_tag
= rx_status
& (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT
)
1256 ? le16_to_cpu(rx_desc
->wb
.qword0
.lo_dword
.l2tag1
)
1258 i40e_receive_skb(rx_ring
, skb
, vlan_tag
);
1260 rx_ring
->netdev
->last_rx
= jiffies
;
1261 rx_desc
->wb
.qword1
.status_error_len
= 0;
1262 } while (likely(total_rx_packets
< budget
));
1264 u64_stats_update_begin(&rx_ring
->syncp
);
1265 rx_ring
->stats
.packets
+= total_rx_packets
;
1266 rx_ring
->stats
.bytes
+= total_rx_bytes
;
1267 u64_stats_update_end(&rx_ring
->syncp
);
1268 rx_ring
->q_vector
->rx
.total_packets
+= total_rx_packets
;
1269 rx_ring
->q_vector
->rx
.total_bytes
+= total_rx_bytes
;
1271 return total_rx_packets
;
1275 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1276 * @napi: napi struct with our devices info in it
1277 * @budget: amount of work driver is allowed to do this pass, in packets
1279 * This function will clean all queues associated with a q_vector.
1281 * Returns the amount of work done
1283 int i40evf_napi_poll(struct napi_struct
*napi
, int budget
)
1285 struct i40e_q_vector
*q_vector
=
1286 container_of(napi
, struct i40e_q_vector
, napi
);
1287 struct i40e_vsi
*vsi
= q_vector
->vsi
;
1288 struct i40e_ring
*ring
;
1289 bool clean_complete
= true;
1290 bool arm_wb
= false;
1291 int budget_per_ring
;
1294 if (test_bit(__I40E_DOWN
, &vsi
->state
)) {
1295 napi_complete(napi
);
1299 /* Since the actual Tx work is minimal, we can give the Tx a larger
1300 * budget and be more aggressive about cleaning up the Tx descriptors.
1302 i40e_for_each_ring(ring
, q_vector
->tx
) {
1303 clean_complete
&= i40e_clean_tx_irq(ring
, vsi
->work_limit
);
1304 arm_wb
|= ring
->arm_wb
;
1307 /* We attempt to distribute budget to each Rx queue fairly, but don't
1308 * allow the budget to go below 1 because that would exit polling early.
1310 budget_per_ring
= max(budget
/q_vector
->num_ringpairs
, 1);
1312 i40e_for_each_ring(ring
, q_vector
->rx
) {
1313 if (ring_is_ps_enabled(ring
))
1314 cleaned
= i40e_clean_rx_irq_ps(ring
, budget_per_ring
);
1316 cleaned
= i40e_clean_rx_irq_1buf(ring
, budget_per_ring
);
1317 /* if we didn't clean as many as budgeted, we must be done */
1318 clean_complete
&= (budget_per_ring
!= cleaned
);
1321 /* If work not completed, return budget and polling will return */
1322 if (!clean_complete
) {
1324 i40e_force_wb(vsi
, q_vector
);
1328 /* Work is done so exit the polling mode and re-enable the interrupt */
1329 napi_complete(napi
);
1330 if (ITR_IS_DYNAMIC(vsi
->rx_itr_setting
) ||
1331 ITR_IS_DYNAMIC(vsi
->tx_itr_setting
))
1332 i40e_update_dynamic_itr(q_vector
);
1334 if (!test_bit(__I40E_DOWN
, &vsi
->state
))
1335 i40evf_irq_enable_queues(vsi
->back
, 1 << q_vector
->v_idx
);
1341 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1343 * @tx_ring: ring to send buffer on
1344 * @flags: the tx flags to be set
1346 * Checks the skb and set up correspondingly several generic transmit flags
1347 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1349 * Returns error code indicate the frame should be dropped upon error and the
1350 * otherwise returns 0 to indicate the flags has been set properly.
1352 static int i40e_tx_prepare_vlan_flags(struct sk_buff
*skb
,
1353 struct i40e_ring
*tx_ring
,
1356 __be16 protocol
= skb
->protocol
;
1359 /* if we have a HW VLAN tag being added, default to the HW one */
1360 if (skb_vlan_tag_present(skb
)) {
1361 tx_flags
|= skb_vlan_tag_get(skb
) << I40E_TX_FLAGS_VLAN_SHIFT
;
1362 tx_flags
|= I40E_TX_FLAGS_HW_VLAN
;
1363 /* else if it is a SW VLAN, check the next protocol and store the tag */
1364 } else if (protocol
== htons(ETH_P_8021Q
)) {
1365 struct vlan_hdr
*vhdr
, _vhdr
;
1366 vhdr
= skb_header_pointer(skb
, ETH_HLEN
, sizeof(_vhdr
), &_vhdr
);
1370 protocol
= vhdr
->h_vlan_encapsulated_proto
;
1371 tx_flags
|= ntohs(vhdr
->h_vlan_TCI
) << I40E_TX_FLAGS_VLAN_SHIFT
;
1372 tx_flags
|= I40E_TX_FLAGS_SW_VLAN
;
1380 * i40e_tso - set up the tso context descriptor
1381 * @tx_ring: ptr to the ring to send
1382 * @skb: ptr to the skb we're sending
1383 * @tx_flags: the collected send information
1384 * @protocol: the send protocol
1385 * @hdr_len: ptr to the size of the packet header
1386 * @cd_tunneling: ptr to context descriptor bits
1388 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1390 static int i40e_tso(struct i40e_ring
*tx_ring
, struct sk_buff
*skb
,
1391 u32 tx_flags
, __be16 protocol
, u8
*hdr_len
,
1392 u64
*cd_type_cmd_tso_mss
, u32
*cd_tunneling
)
1394 u32 cd_cmd
, cd_tso_len
, cd_mss
;
1395 struct ipv6hdr
*ipv6h
;
1396 struct tcphdr
*tcph
;
1401 if (!skb_is_gso(skb
))
1404 err
= skb_cow_head(skb
, 0);
1408 if (protocol
== htons(ETH_P_IP
)) {
1409 iph
= skb
->encapsulation
? inner_ip_hdr(skb
) : ip_hdr(skb
);
1410 tcph
= skb
->encapsulation
? inner_tcp_hdr(skb
) : tcp_hdr(skb
);
1413 tcph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
1415 } else if (skb_is_gso_v6(skb
)) {
1417 ipv6h
= skb
->encapsulation
? inner_ipv6_hdr(skb
)
1419 tcph
= skb
->encapsulation
? inner_tcp_hdr(skb
) : tcp_hdr(skb
);
1420 ipv6h
->payload_len
= 0;
1421 tcph
->check
= ~csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
,
1425 l4len
= skb
->encapsulation
? inner_tcp_hdrlen(skb
) : tcp_hdrlen(skb
);
1426 *hdr_len
= (skb
->encapsulation
1427 ? (skb_inner_transport_header(skb
) - skb
->data
)
1428 : skb_transport_offset(skb
)) + l4len
;
1430 /* find the field values */
1431 cd_cmd
= I40E_TX_CTX_DESC_TSO
;
1432 cd_tso_len
= skb
->len
- *hdr_len
;
1433 cd_mss
= skb_shinfo(skb
)->gso_size
;
1434 *cd_type_cmd_tso_mss
|= ((u64
)cd_cmd
<< I40E_TXD_CTX_QW1_CMD_SHIFT
) |
1436 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT
) |
1437 ((u64
)cd_mss
<< I40E_TXD_CTX_QW1_MSS_SHIFT
);
1442 * i40e_tx_enable_csum - Enable Tx checksum offloads
1444 * @tx_flags: Tx flags currently set
1445 * @td_cmd: Tx descriptor command bits to set
1446 * @td_offset: Tx descriptor header offsets to set
1447 * @cd_tunneling: ptr to context desc bits
1449 static void i40e_tx_enable_csum(struct sk_buff
*skb
, u32 tx_flags
,
1450 u32
*td_cmd
, u32
*td_offset
,
1451 struct i40e_ring
*tx_ring
,
1454 struct ipv6hdr
*this_ipv6_hdr
;
1455 unsigned int this_tcp_hdrlen
;
1456 struct iphdr
*this_ip_hdr
;
1457 u32 network_hdr_len
;
1460 if (skb
->encapsulation
) {
1461 network_hdr_len
= skb_inner_network_header_len(skb
);
1462 this_ip_hdr
= inner_ip_hdr(skb
);
1463 this_ipv6_hdr
= inner_ipv6_hdr(skb
);
1464 this_tcp_hdrlen
= inner_tcp_hdrlen(skb
);
1466 if (tx_flags
& I40E_TX_FLAGS_IPV4
) {
1468 if (tx_flags
& I40E_TX_FLAGS_TSO
) {
1469 *cd_tunneling
|= I40E_TX_CTX_EXT_IP_IPV4
;
1470 ip_hdr(skb
)->check
= 0;
1473 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM
;
1475 } else if (tx_flags
& I40E_TX_FLAGS_IPV6
) {
1476 if (tx_flags
& I40E_TX_FLAGS_TSO
) {
1477 *cd_tunneling
|= I40E_TX_CTX_EXT_IP_IPV6
;
1478 ip_hdr(skb
)->check
= 0;
1481 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM
;
1485 /* Now set the ctx descriptor fields */
1486 *cd_tunneling
|= (skb_network_header_len(skb
) >> 2) <<
1487 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT
|
1488 I40E_TXD_CTX_UDP_TUNNELING
|
1489 ((skb_inner_network_offset(skb
) -
1490 skb_transport_offset(skb
)) >> 1) <<
1491 I40E_TXD_CTX_QW0_NATLEN_SHIFT
;
1494 network_hdr_len
= skb_network_header_len(skb
);
1495 this_ip_hdr
= ip_hdr(skb
);
1496 this_ipv6_hdr
= ipv6_hdr(skb
);
1497 this_tcp_hdrlen
= tcp_hdrlen(skb
);
1500 /* Enable IP checksum offloads */
1501 if (tx_flags
& I40E_TX_FLAGS_IPV4
) {
1502 l4_hdr
= this_ip_hdr
->protocol
;
1503 /* the stack computes the IP header already, the only time we
1504 * need the hardware to recompute it is in the case of TSO.
1506 if (tx_flags
& I40E_TX_FLAGS_TSO
) {
1507 *td_cmd
|= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM
;
1508 this_ip_hdr
->check
= 0;
1510 *td_cmd
|= I40E_TX_DESC_CMD_IIPT_IPV4
;
1512 /* Now set the td_offset for IP header length */
1513 *td_offset
= (network_hdr_len
>> 2) <<
1514 I40E_TX_DESC_LENGTH_IPLEN_SHIFT
;
1515 } else if (tx_flags
& I40E_TX_FLAGS_IPV6
) {
1516 l4_hdr
= this_ipv6_hdr
->nexthdr
;
1517 *td_cmd
|= I40E_TX_DESC_CMD_IIPT_IPV6
;
1518 /* Now set the td_offset for IP header length */
1519 *td_offset
= (network_hdr_len
>> 2) <<
1520 I40E_TX_DESC_LENGTH_IPLEN_SHIFT
;
1522 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1523 *td_offset
|= (skb_network_offset(skb
) >> 1) <<
1524 I40E_TX_DESC_LENGTH_MACLEN_SHIFT
;
1526 /* Enable L4 checksum offloads */
1529 /* enable checksum offloads */
1530 *td_cmd
|= I40E_TX_DESC_CMD_L4T_EOFT_TCP
;
1531 *td_offset
|= (this_tcp_hdrlen
>> 2) <<
1532 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT
;
1535 /* enable SCTP checksum offload */
1536 *td_cmd
|= I40E_TX_DESC_CMD_L4T_EOFT_SCTP
;
1537 *td_offset
|= (sizeof(struct sctphdr
) >> 2) <<
1538 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT
;
1541 /* enable UDP checksum offload */
1542 *td_cmd
|= I40E_TX_DESC_CMD_L4T_EOFT_UDP
;
1543 *td_offset
|= (sizeof(struct udphdr
) >> 2) <<
1544 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT
;
1552 * i40e_create_tx_ctx Build the Tx context descriptor
1553 * @tx_ring: ring to create the descriptor on
1554 * @cd_type_cmd_tso_mss: Quad Word 1
1555 * @cd_tunneling: Quad Word 0 - bits 0-31
1556 * @cd_l2tag2: Quad Word 0 - bits 32-63
1558 static void i40e_create_tx_ctx(struct i40e_ring
*tx_ring
,
1559 const u64 cd_type_cmd_tso_mss
,
1560 const u32 cd_tunneling
, const u32 cd_l2tag2
)
1562 struct i40e_tx_context_desc
*context_desc
;
1563 int i
= tx_ring
->next_to_use
;
1565 if ((cd_type_cmd_tso_mss
== I40E_TX_DESC_DTYPE_CONTEXT
) &&
1566 !cd_tunneling
&& !cd_l2tag2
)
1569 /* grab the next descriptor */
1570 context_desc
= I40E_TX_CTXTDESC(tx_ring
, i
);
1573 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
1575 /* cpu_to_le32 and assign to struct fields */
1576 context_desc
->tunneling_params
= cpu_to_le32(cd_tunneling
);
1577 context_desc
->l2tag2
= cpu_to_le16(cd_l2tag2
);
1578 context_desc
->rsvd
= cpu_to_le16(0);
1579 context_desc
->type_cmd_tso_mss
= cpu_to_le64(cd_type_cmd_tso_mss
);
1583 * i40e_tx_map - Build the Tx descriptor
1584 * @tx_ring: ring to send buffer on
1586 * @first: first buffer info buffer to use
1587 * @tx_flags: collected send information
1588 * @hdr_len: size of the packet header
1589 * @td_cmd: the command field in the descriptor
1590 * @td_offset: offset for checksum or crc
1592 static void i40e_tx_map(struct i40e_ring
*tx_ring
, struct sk_buff
*skb
,
1593 struct i40e_tx_buffer
*first
, u32 tx_flags
,
1594 const u8 hdr_len
, u32 td_cmd
, u32 td_offset
)
1596 unsigned int data_len
= skb
->data_len
;
1597 unsigned int size
= skb_headlen(skb
);
1598 struct skb_frag_struct
*frag
;
1599 struct i40e_tx_buffer
*tx_bi
;
1600 struct i40e_tx_desc
*tx_desc
;
1601 u16 i
= tx_ring
->next_to_use
;
1606 if (tx_flags
& I40E_TX_FLAGS_HW_VLAN
) {
1607 td_cmd
|= I40E_TX_DESC_CMD_IL2TAG1
;
1608 td_tag
= (tx_flags
& I40E_TX_FLAGS_VLAN_MASK
) >>
1609 I40E_TX_FLAGS_VLAN_SHIFT
;
1612 if (tx_flags
& (I40E_TX_FLAGS_TSO
| I40E_TX_FLAGS_FSO
))
1613 gso_segs
= skb_shinfo(skb
)->gso_segs
;
1617 /* multiply data chunks by size of headers */
1618 first
->bytecount
= skb
->len
- hdr_len
+ (gso_segs
* hdr_len
);
1619 first
->gso_segs
= gso_segs
;
1621 first
->tx_flags
= tx_flags
;
1623 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
1625 tx_desc
= I40E_TX_DESC(tx_ring
, i
);
1628 for (frag
= &skb_shinfo(skb
)->frags
[0];; frag
++) {
1629 if (dma_mapping_error(tx_ring
->dev
, dma
))
1632 /* record length, and DMA address */
1633 dma_unmap_len_set(tx_bi
, len
, size
);
1634 dma_unmap_addr_set(tx_bi
, dma
, dma
);
1636 tx_desc
->buffer_addr
= cpu_to_le64(dma
);
1638 while (unlikely(size
> I40E_MAX_DATA_PER_TXD
)) {
1639 tx_desc
->cmd_type_offset_bsz
=
1640 build_ctob(td_cmd
, td_offset
,
1641 I40E_MAX_DATA_PER_TXD
, td_tag
);
1645 if (i
== tx_ring
->count
) {
1646 tx_desc
= I40E_TX_DESC(tx_ring
, 0);
1650 dma
+= I40E_MAX_DATA_PER_TXD
;
1651 size
-= I40E_MAX_DATA_PER_TXD
;
1653 tx_desc
->buffer_addr
= cpu_to_le64(dma
);
1656 if (likely(!data_len
))
1659 tx_desc
->cmd_type_offset_bsz
= build_ctob(td_cmd
, td_offset
,
1664 if (i
== tx_ring
->count
) {
1665 tx_desc
= I40E_TX_DESC(tx_ring
, 0);
1669 size
= skb_frag_size(frag
);
1672 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, size
,
1675 tx_bi
= &tx_ring
->tx_bi
[i
];
1678 /* Place RS bit on last descriptor of any packet that spans across the
1679 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1681 #define WB_STRIDE 0x3
1682 if (((i
& WB_STRIDE
) != WB_STRIDE
) &&
1683 (first
<= &tx_ring
->tx_bi
[i
]) &&
1684 (first
>= &tx_ring
->tx_bi
[i
& ~WB_STRIDE
])) {
1685 tx_desc
->cmd_type_offset_bsz
=
1686 build_ctob(td_cmd
, td_offset
, size
, td_tag
) |
1687 cpu_to_le64((u64
)I40E_TX_DESC_CMD_EOP
<<
1688 I40E_TXD_QW1_CMD_SHIFT
);
1690 tx_desc
->cmd_type_offset_bsz
=
1691 build_ctob(td_cmd
, td_offset
, size
, td_tag
) |
1692 cpu_to_le64((u64
)I40E_TXD_CMD
<<
1693 I40E_TXD_QW1_CMD_SHIFT
);
1696 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring
->netdev
,
1697 tx_ring
->queue_index
),
1700 /* set the timestamp */
1701 first
->time_stamp
= jiffies
;
1703 /* Force memory writes to complete before letting h/w
1704 * know there are new descriptors to fetch. (Only
1705 * applicable for weak-ordered memory model archs,
1710 /* set next_to_watch value indicating a packet is present */
1711 first
->next_to_watch
= tx_desc
;
1714 if (i
== tx_ring
->count
)
1717 tx_ring
->next_to_use
= i
;
1719 /* notify HW of packet */
1720 writel(i
, tx_ring
->tail
);
1725 dev_info(tx_ring
->dev
, "TX DMA map failed\n");
1727 /* clear dma mappings for failed tx_bi map */
1729 tx_bi
= &tx_ring
->tx_bi
[i
];
1730 i40e_unmap_and_free_tx_resource(tx_ring
, tx_bi
);
1738 tx_ring
->next_to_use
= i
;
1742 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
1743 * @tx_ring: the ring to be checked
1744 * @size: the size buffer we want to assure is available
1746 * Returns -EBUSY if a stop is needed, else 0
1748 static inline int __i40e_maybe_stop_tx(struct i40e_ring
*tx_ring
, int size
)
1750 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
1751 /* Memory barrier before checking head and tail */
1754 /* Check again in a case another CPU has just made room available. */
1755 if (likely(I40E_DESC_UNUSED(tx_ring
) < size
))
1758 /* A reprieve! - use start_queue because it doesn't call schedule */
1759 netif_start_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
1760 ++tx_ring
->tx_stats
.restart_queue
;
1765 * i40e_maybe_stop_tx - 1st level check for tx stop conditions
1766 * @tx_ring: the ring to be checked
1767 * @size: the size buffer we want to assure is available
1769 * Returns 0 if stop is not needed
1771 static int i40e_maybe_stop_tx(struct i40e_ring
*tx_ring
, int size
)
1773 if (likely(I40E_DESC_UNUSED(tx_ring
) >= size
))
1775 return __i40e_maybe_stop_tx(tx_ring
, size
);
1779 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
1781 * @tx_ring: ring to send buffer on
1783 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1784 * there is not enough descriptors available in this ring since we need at least
1787 static int i40e_xmit_descriptor_count(struct sk_buff
*skb
,
1788 struct i40e_ring
*tx_ring
)
1793 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1794 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1795 * + 4 desc gap to avoid the cache line where head is,
1796 * + 1 desc for context descriptor,
1797 * otherwise try next time
1799 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
1800 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
1802 count
+= TXD_USE_COUNT(skb_headlen(skb
));
1803 if (i40e_maybe_stop_tx(tx_ring
, count
+ 4 + 1)) {
1804 tx_ring
->tx_stats
.tx_busy
++;
1811 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1813 * @tx_ring: ring to send buffer on
1815 * Returns NETDEV_TX_OK if sent, else an error code
1817 static netdev_tx_t
i40e_xmit_frame_ring(struct sk_buff
*skb
,
1818 struct i40e_ring
*tx_ring
)
1820 u64 cd_type_cmd_tso_mss
= I40E_TX_DESC_DTYPE_CONTEXT
;
1821 u32 cd_tunneling
= 0, cd_l2tag2
= 0;
1822 struct i40e_tx_buffer
*first
;
1829 if (0 == i40e_xmit_descriptor_count(skb
, tx_ring
))
1830 return NETDEV_TX_BUSY
;
1832 /* prepare the xmit flags */
1833 if (i40e_tx_prepare_vlan_flags(skb
, tx_ring
, &tx_flags
))
1836 /* obtain protocol of skb */
1837 protocol
= vlan_get_protocol(skb
);
1839 /* record the location of the first descriptor for this packet */
1840 first
= &tx_ring
->tx_bi
[tx_ring
->next_to_use
];
1842 /* setup IPv4/IPv6 offloads */
1843 if (protocol
== htons(ETH_P_IP
))
1844 tx_flags
|= I40E_TX_FLAGS_IPV4
;
1845 else if (protocol
== htons(ETH_P_IPV6
))
1846 tx_flags
|= I40E_TX_FLAGS_IPV6
;
1848 tso
= i40e_tso(tx_ring
, skb
, tx_flags
, protocol
, &hdr_len
,
1849 &cd_type_cmd_tso_mss
, &cd_tunneling
);
1854 tx_flags
|= I40E_TX_FLAGS_TSO
;
1856 skb_tx_timestamp(skb
);
1858 /* always enable CRC insertion offload */
1859 td_cmd
|= I40E_TX_DESC_CMD_ICRC
;
1861 /* Always offload the checksum, since it's in the data descriptor */
1862 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1863 tx_flags
|= I40E_TX_FLAGS_CSUM
;
1865 i40e_tx_enable_csum(skb
, tx_flags
, &td_cmd
, &td_offset
,
1866 tx_ring
, &cd_tunneling
);
1869 i40e_create_tx_ctx(tx_ring
, cd_type_cmd_tso_mss
,
1870 cd_tunneling
, cd_l2tag2
);
1872 i40e_tx_map(tx_ring
, skb
, first
, tx_flags
, hdr_len
,
1875 i40e_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
1877 return NETDEV_TX_OK
;
1880 dev_kfree_skb_any(skb
);
1881 return NETDEV_TX_OK
;
1885 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1887 * @netdev: network interface device structure
1889 * Returns NETDEV_TX_OK if sent, else an error code
1891 netdev_tx_t
i40evf_xmit_frame(struct sk_buff
*skb
, struct net_device
*netdev
)
1893 struct i40evf_adapter
*adapter
= netdev_priv(netdev
);
1894 struct i40e_ring
*tx_ring
= adapter
->tx_rings
[skb
->queue_mapping
];
1896 /* hardware can't handle really short frames, hardware padding works
1899 if (unlikely(skb
->len
< I40E_MIN_TX_LEN
)) {
1900 if (skb_pad(skb
, I40E_MIN_TX_LEN
- skb
->len
))
1901 return NETDEV_TX_OK
;
1902 skb
->len
= I40E_MIN_TX_LEN
;
1903 skb_set_tail_pointer(skb
, I40E_MIN_TX_LEN
);
1906 return i40e_xmit_frame_ring(skb
, tx_ring
);