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_head - Retrieve head from head writeback
131 * @tx_ring: tx ring to fetch head of
133 * Returns value of Tx ring head based on value stored
134 * in head write-back location
136 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
138 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
140 return le32_to_cpu(*(volatile __le32 *)head);
144 * i40e_get_tx_pending - how many tx descriptors not processed
145 * @tx_ring: the ring of descriptors
147 * Since there is no access to the ring head register
148 * in XL710, we need to use our local copies
150 static u32 i40e_get_tx_pending(struct i40e_ring *ring)
154 head = i40e_get_head(ring);
155 tail = readl(ring->tail);
158 return (head < tail) ?
159 tail - head : (tail + ring->count - head);
165 * i40e_check_tx_hang - Is there a hang in the Tx queue
166 * @tx_ring: the ring of descriptors
168 static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
170 u32 tx_done = tx_ring->stats.packets;
171 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
172 u32 tx_pending = i40e_get_tx_pending(tx_ring);
175 clear_check_for_tx_hang(tx_ring);
177 /* Check for a hung queue, but be thorough. This verifies
178 * that a transmit has been completed since the previous
179 * check AND there is at least one packet pending. The
180 * ARMED bit is set to indicate a potential hang. The
181 * bit is cleared if a pause frame is received to remove
182 * false hang detection due to PFC or 802.3x frames. By
183 * requiring this to fail twice we avoid races with
184 * PFC clearing the ARMED bit and conditions where we
185 * run the check_tx_hang logic with a transmit completion
186 * pending but without time to complete it yet.
188 if ((tx_done_old == tx_done) && tx_pending) {
189 /* make sure it is true for two checks in a row */
190 ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED,
192 } else if (tx_done_old == tx_done &&
193 (tx_pending < I40E_MIN_DESC_PENDING) && (tx_pending > 0)) {
194 /* update completed stats and disarm the hang check */
195 tx_ring->tx_stats.tx_done_old = tx_done;
196 clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state);
202 #define WB_STRIDE 0x3
205 * i40e_clean_tx_irq - Reclaim resources after transmit completes
206 * @tx_ring: tx ring to clean
207 * @budget: how many cleans we're allowed
209 * Returns true if there's any budget left (e.g. the clean is finished)
211 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
213 u16 i = tx_ring->next_to_clean;
214 struct i40e_tx_buffer *tx_buf;
215 struct i40e_tx_desc *tx_head;
216 struct i40e_tx_desc *tx_desc;
217 unsigned int total_packets = 0;
218 unsigned int total_bytes = 0;
220 tx_buf = &tx_ring->tx_bi[i];
221 tx_desc = I40E_TX_DESC(tx_ring, i);
224 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
227 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
229 /* if next_to_watch is not set then there is no work pending */
233 /* prevent any other reads prior to eop_desc */
234 read_barrier_depends();
236 /* we have caught up to head, no work left to do */
237 if (tx_head == tx_desc)
240 /* clear next_to_watch to prevent false hangs */
241 tx_buf->next_to_watch = NULL;
243 /* update the statistics for this packet */
244 total_bytes += tx_buf->bytecount;
245 total_packets += tx_buf->gso_segs;
248 dev_kfree_skb_any(tx_buf->skb);
250 /* unmap skb header data */
251 dma_unmap_single(tx_ring->dev,
252 dma_unmap_addr(tx_buf, dma),
253 dma_unmap_len(tx_buf, len),
256 /* clear tx_buffer data */
258 dma_unmap_len_set(tx_buf, len, 0);
260 /* unmap remaining buffers */
261 while (tx_desc != eop_desc) {
268 tx_buf = tx_ring->tx_bi;
269 tx_desc = I40E_TX_DESC(tx_ring, 0);
272 /* unmap any remaining paged data */
273 if (dma_unmap_len(tx_buf, len)) {
274 dma_unmap_page(tx_ring->dev,
275 dma_unmap_addr(tx_buf, dma),
276 dma_unmap_len(tx_buf, len),
278 dma_unmap_len_set(tx_buf, len, 0);
282 /* move us one more past the eop_desc for start of next pkt */
288 tx_buf = tx_ring->tx_bi;
289 tx_desc = I40E_TX_DESC(tx_ring, 0);
294 /* update budget accounting */
296 } while (likely(budget));
299 tx_ring->next_to_clean = i;
300 u64_stats_update_begin(&tx_ring->syncp);
301 tx_ring->stats.bytes += total_bytes;
302 tx_ring->stats.packets += total_packets;
303 u64_stats_update_end(&tx_ring->syncp);
304 tx_ring->q_vector->tx.total_bytes += total_bytes;
305 tx_ring->q_vector->tx.total_packets += total_packets;
308 !((i & WB_STRIDE) == WB_STRIDE) &&
309 !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
310 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
311 tx_ring->arm_wb = true;
313 tx_ring->arm_wb = false;
315 if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) {
316 /* schedule immediate reset if we believe we hung */
317 dev_info(tx_ring->dev, "Detected Tx Unit Hang\n"
320 " next_to_use <%x>\n"
321 " next_to_clean <%x>\n",
323 tx_ring->queue_index,
324 tx_ring->next_to_use, i);
326 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
328 dev_info(tx_ring->dev,
329 "tx hang detected on queue %d, resetting adapter\n",
330 tx_ring->queue_index);
332 tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev);
334 /* the adapter is about to reset, no point in enabling stuff */
338 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
339 tx_ring->queue_index),
340 total_packets, total_bytes);
342 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
343 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
344 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
345 /* Make sure that anybody stopping the queue after this
346 * sees the new next_to_clean.
349 if (__netif_subqueue_stopped(tx_ring->netdev,
350 tx_ring->queue_index) &&
351 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
352 netif_wake_subqueue(tx_ring->netdev,
353 tx_ring->queue_index);
354 ++tx_ring->tx_stats.restart_queue;
362 * i40e_force_wb -Arm hardware to do a wb on noncache aligned descriptors
363 * @vsi: the VSI we care about
364 * @q_vector: the vector on which to force writeback
367 static void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
369 u16 flags = q_vector->tx.ring[0].flags;
371 if (flags & I40E_TXR_FLAGS_WB_ON_ITR) {
374 if (q_vector->arm_wb_state)
377 val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK;
380 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
381 vsi->base_vector - 1),
383 q_vector->arm_wb_state = true;
385 u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
386 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
387 I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
388 I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK;
389 /* allow 00 to be written to the index */
392 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
393 vsi->base_vector - 1), val);
398 * i40e_set_new_dynamic_itr - Find new ITR level
399 * @rc: structure containing ring performance data
401 * Stores a new ITR value based on packets and byte counts during
402 * the last interrupt. The advantage of per interrupt computation
403 * is faster updates and more accurate ITR for the current traffic
404 * pattern. Constants in this function were computed based on
405 * theoretical maximum wire speed and thresholds were set based on
406 * testing data as well as attempting to minimize response time
407 * while increasing bulk throughput.
409 static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
411 enum i40e_latency_range new_latency_range = rc->latency_range;
412 u32 new_itr = rc->itr;
415 if (rc->total_packets == 0 || !rc->itr)
418 /* simple throttlerate management
419 * 0-10MB/s lowest (100000 ints/s)
420 * 10-20MB/s low (20000 ints/s)
421 * 20-1249MB/s bulk (8000 ints/s)
423 bytes_per_int = rc->total_bytes / rc->itr;
424 switch (new_latency_range) {
425 case I40E_LOWEST_LATENCY:
426 if (bytes_per_int > 10)
427 new_latency_range = I40E_LOW_LATENCY;
429 case I40E_LOW_LATENCY:
430 if (bytes_per_int > 20)
431 new_latency_range = I40E_BULK_LATENCY;
432 else if (bytes_per_int <= 10)
433 new_latency_range = I40E_LOWEST_LATENCY;
435 case I40E_BULK_LATENCY:
436 if (bytes_per_int <= 20)
437 new_latency_range = I40E_LOW_LATENCY;
440 if (bytes_per_int <= 20)
441 new_latency_range = I40E_LOW_LATENCY;
444 rc->latency_range = new_latency_range;
446 switch (new_latency_range) {
447 case I40E_LOWEST_LATENCY:
448 new_itr = I40E_ITR_100K;
450 case I40E_LOW_LATENCY:
451 new_itr = I40E_ITR_20K;
453 case I40E_BULK_LATENCY:
454 new_itr = I40E_ITR_8K;
460 if (new_itr != rc->itr)
464 rc->total_packets = 0;
468 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
469 * @tx_ring: the tx ring to set up
471 * Return 0 on success, negative on error
473 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
475 struct device *dev = tx_ring->dev;
481 /* warn if we are about to overwrite the pointer */
482 WARN_ON(tx_ring->tx_bi);
483 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
484 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
488 /* round up to nearest 4K */
489 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
490 /* add u32 for head writeback, align after this takes care of
491 * guaranteeing this is at least one cache line in size
493 tx_ring->size += sizeof(u32);
494 tx_ring->size = ALIGN(tx_ring->size, 4096);
495 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
496 &tx_ring->dma, GFP_KERNEL);
497 if (!tx_ring->desc) {
498 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
503 tx_ring->next_to_use = 0;
504 tx_ring->next_to_clean = 0;
508 kfree(tx_ring->tx_bi);
509 tx_ring->tx_bi = NULL;
514 * i40evf_clean_rx_ring - Free Rx buffers
515 * @rx_ring: ring to be cleaned
517 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
519 struct device *dev = rx_ring->dev;
520 struct i40e_rx_buffer *rx_bi;
521 unsigned long bi_size;
524 /* ring already cleared, nothing to do */
528 if (ring_is_ps_enabled(rx_ring)) {
529 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
531 rx_bi = &rx_ring->rx_bi[0];
532 if (rx_bi->hdr_buf) {
533 dma_free_coherent(dev,
537 for (i = 0; i < rx_ring->count; i++) {
538 rx_bi = &rx_ring->rx_bi[i];
540 rx_bi->hdr_buf = NULL;
544 /* Free all the Rx ring sk_buffs */
545 for (i = 0; i < rx_ring->count; i++) {
546 rx_bi = &rx_ring->rx_bi[i];
548 dma_unmap_single(dev,
555 dev_kfree_skb(rx_bi->skb);
559 if (rx_bi->page_dma) {
566 __free_page(rx_bi->page);
568 rx_bi->page_offset = 0;
572 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
573 memset(rx_ring->rx_bi, 0, bi_size);
575 /* Zero out the descriptor ring */
576 memset(rx_ring->desc, 0, rx_ring->size);
578 rx_ring->next_to_clean = 0;
579 rx_ring->next_to_use = 0;
583 * i40evf_free_rx_resources - Free Rx resources
584 * @rx_ring: ring to clean the resources from
586 * Free all receive software resources
588 void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
590 i40evf_clean_rx_ring(rx_ring);
591 kfree(rx_ring->rx_bi);
592 rx_ring->rx_bi = NULL;
595 dma_free_coherent(rx_ring->dev, rx_ring->size,
596 rx_ring->desc, rx_ring->dma);
597 rx_ring->desc = NULL;
602 * i40evf_alloc_rx_headers - allocate rx header buffers
603 * @rx_ring: ring to alloc buffers
605 * Allocate rx header buffers for the entire ring. As these are static,
606 * this is only called when setting up a new ring.
608 void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
610 struct device *dev = rx_ring->dev;
611 struct i40e_rx_buffer *rx_bi;
617 if (rx_ring->rx_bi[0].hdr_buf)
619 /* Make sure the buffers don't cross cache line boundaries. */
620 buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
621 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
625 for (i = 0; i < rx_ring->count; i++) {
626 rx_bi = &rx_ring->rx_bi[i];
627 rx_bi->dma = dma + (i * buf_size);
628 rx_bi->hdr_buf = buffer + (i * buf_size);
633 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
634 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
636 * Returns 0 on success, negative on failure
638 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
640 struct device *dev = rx_ring->dev;
643 /* warn if we are about to overwrite the pointer */
644 WARN_ON(rx_ring->rx_bi);
645 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
646 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
650 u64_stats_init(&rx_ring->syncp);
652 /* Round up to nearest 4K */
653 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
654 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
655 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
656 rx_ring->size = ALIGN(rx_ring->size, 4096);
657 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
658 &rx_ring->dma, GFP_KERNEL);
660 if (!rx_ring->desc) {
661 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
666 rx_ring->next_to_clean = 0;
667 rx_ring->next_to_use = 0;
671 kfree(rx_ring->rx_bi);
672 rx_ring->rx_bi = NULL;
677 * i40e_release_rx_desc - Store the new tail and head values
678 * @rx_ring: ring to bump
679 * @val: new head index
681 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
683 rx_ring->next_to_use = val;
684 /* Force memory writes to complete before letting h/w
685 * know there are new descriptors to fetch. (Only
686 * applicable for weak-ordered memory model archs,
690 writel(val, rx_ring->tail);
694 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
695 * @rx_ring: ring to place buffers on
696 * @cleaned_count: number of buffers to replace
698 void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
700 u16 i = rx_ring->next_to_use;
701 union i40e_rx_desc *rx_desc;
702 struct i40e_rx_buffer *bi;
704 /* do nothing if no valid netdev defined */
705 if (!rx_ring->netdev || !cleaned_count)
708 while (cleaned_count--) {
709 rx_desc = I40E_RX_DESC(rx_ring, i);
710 bi = &rx_ring->rx_bi[i];
712 if (bi->skb) /* desc is in use */
715 bi->page = alloc_page(GFP_ATOMIC);
717 rx_ring->rx_stats.alloc_page_failed++;
723 /* use a half page if we're re-using */
724 bi->page_offset ^= PAGE_SIZE / 2;
725 bi->page_dma = dma_map_page(rx_ring->dev,
730 if (dma_mapping_error(rx_ring->dev,
732 rx_ring->rx_stats.alloc_page_failed++;
738 dma_sync_single_range_for_device(rx_ring->dev,
743 /* Refresh the desc even if buffer_addrs didn't change
744 * because each write-back erases this info.
746 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
747 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
749 if (i == rx_ring->count)
754 if (rx_ring->next_to_use != i)
755 i40e_release_rx_desc(rx_ring, i);
759 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
760 * @rx_ring: ring to place buffers on
761 * @cleaned_count: number of buffers to replace
763 void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
765 u16 i = rx_ring->next_to_use;
766 union i40e_rx_desc *rx_desc;
767 struct i40e_rx_buffer *bi;
770 /* do nothing if no valid netdev defined */
771 if (!rx_ring->netdev || !cleaned_count)
774 while (cleaned_count--) {
775 rx_desc = I40E_RX_DESC(rx_ring, i);
776 bi = &rx_ring->rx_bi[i];
780 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
781 rx_ring->rx_buf_len);
783 rx_ring->rx_stats.alloc_buff_failed++;
786 /* initialize queue mapping */
787 skb_record_rx_queue(skb, rx_ring->queue_index);
792 bi->dma = dma_map_single(rx_ring->dev,
796 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
797 rx_ring->rx_stats.alloc_buff_failed++;
803 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
804 rx_desc->read.hdr_addr = 0;
806 if (i == rx_ring->count)
811 if (rx_ring->next_to_use != i)
812 i40e_release_rx_desc(rx_ring, i);
816 * i40e_receive_skb - Send a completed packet up the stack
817 * @rx_ring: rx ring in play
818 * @skb: packet to send up
819 * @vlan_tag: vlan tag for packet
821 static void i40e_receive_skb(struct i40e_ring *rx_ring,
822 struct sk_buff *skb, u16 vlan_tag)
824 struct i40e_q_vector *q_vector = rx_ring->q_vector;
825 struct i40e_vsi *vsi = rx_ring->vsi;
826 u64 flags = vsi->back->flags;
828 if (vlan_tag & VLAN_VID_MASK)
829 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
831 if (flags & I40E_FLAG_IN_NETPOLL)
834 napi_gro_receive(&q_vector->napi, skb);
838 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
839 * @vsi: the VSI we care about
840 * @skb: skb currently being received and modified
841 * @rx_status: status value of last descriptor in packet
842 * @rx_error: error value of last descriptor in packet
843 * @rx_ptype: ptype value of last descriptor in packet
845 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
851 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
852 bool ipv4 = false, ipv6 = false;
853 bool ipv4_tunnel, ipv6_tunnel;
858 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
859 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
860 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
861 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
863 skb->ip_summed = CHECKSUM_NONE;
865 /* Rx csum enabled and ip headers found? */
866 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
869 /* did the hardware decode the packet and checksum? */
870 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
873 /* both known and outer_ip must be set for the below code to work */
874 if (!(decoded.known && decoded.outer_ip))
877 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
878 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
880 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
881 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
885 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
886 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
889 /* likely incorrect csum if alternate IP extension headers found */
891 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
892 /* don't increment checksum err here, non-fatal err */
895 /* there was some L4 error, count error and punt packet to the stack */
896 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
899 /* handle packets that were not able to be checksummed due
900 * to arrival speed, in this case the stack can compute
903 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
906 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
907 * it in the driver, hardware does not do it for us.
908 * Since L3L4P bit was set we assume a valid IHL value (>=5)
909 * so the total length of IPv4 header is IHL*4 bytes
910 * The UDP_0 bit *may* bet set if the *inner* header is UDP
913 skb->transport_header = skb->mac_header +
914 sizeof(struct ethhdr) +
915 (ip_hdr(skb)->ihl * 4);
917 /* Add 4 bytes for VLAN tagged packets */
918 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
919 skb->protocol == htons(ETH_P_8021AD))
922 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
923 (udp_hdr(skb)->check != 0)) {
924 rx_udp_csum = udp_csum(skb);
926 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
928 skb_transport_offset(skb)),
929 IPPROTO_UDP, rx_udp_csum);
931 if (udp_hdr(skb)->check != csum)
934 } /* else its GRE and so no outer UDP header */
937 skb->ip_summed = CHECKSUM_UNNECESSARY;
938 skb->csum_level = ipv4_tunnel || ipv6_tunnel;
943 vsi->back->hw_csum_rx_error++;
947 * i40e_rx_hash - returns the hash value from the Rx descriptor
948 * @ring: descriptor ring
949 * @rx_desc: specific descriptor
951 static inline u32 i40e_rx_hash(struct i40e_ring *ring,
952 union i40e_rx_desc *rx_desc)
954 const __le64 rss_mask =
955 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
956 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
958 if ((ring->netdev->features & NETIF_F_RXHASH) &&
959 (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
960 return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
966 * i40e_ptype_to_hash - get a hash type
967 * @ptype: the ptype value from the descriptor
969 * Returns a hash type to be used by skb_set_hash
971 static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
973 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
976 return PKT_HASH_TYPE_NONE;
978 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
979 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
980 return PKT_HASH_TYPE_L4;
981 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
982 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
983 return PKT_HASH_TYPE_L3;
985 return PKT_HASH_TYPE_L2;
989 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
990 * @rx_ring: rx ring to clean
991 * @budget: how many cleans we're allowed
993 * Returns true if there's any budget left (e.g. the clean is finished)
995 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
997 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
998 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
999 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1000 const int current_node = numa_node_id();
1001 struct i40e_vsi *vsi = rx_ring->vsi;
1002 u16 i = rx_ring->next_to_clean;
1003 union i40e_rx_desc *rx_desc;
1004 u32 rx_error, rx_status;
1009 struct i40e_rx_buffer *rx_bi;
1010 struct sk_buff *skb;
1012 /* return some buffers to hardware, one at a time is too slow */
1013 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1014 i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
1018 i = rx_ring->next_to_clean;
1019 rx_desc = I40E_RX_DESC(rx_ring, i);
1020 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1021 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1022 I40E_RXD_QW1_STATUS_SHIFT;
1024 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1027 /* This memory barrier is needed to keep us from reading
1028 * any other fields out of the rx_desc until we know the
1032 rx_bi = &rx_ring->rx_bi[i];
1035 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1036 rx_ring->rx_hdr_len);
1038 rx_ring->rx_stats.alloc_buff_failed++;
1042 /* initialize queue mapping */
1043 skb_record_rx_queue(skb, rx_ring->queue_index);
1044 /* we are reusing so sync this buffer for CPU use */
1045 dma_sync_single_range_for_cpu(rx_ring->dev,
1048 rx_ring->rx_hdr_len,
1051 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1052 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1053 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
1054 I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1055 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1056 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1058 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1059 I40E_RXD_QW1_ERROR_SHIFT;
1060 rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1061 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1063 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1064 I40E_RXD_QW1_PTYPE_SHIFT;
1065 prefetch(rx_bi->page);
1068 if (rx_hbo || rx_sph) {
1071 len = I40E_RX_HDR_SIZE;
1073 len = rx_header_len;
1074 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1075 } else if (skb->len == 0) {
1078 len = (rx_packet_len > skb_headlen(skb) ?
1079 skb_headlen(skb) : rx_packet_len);
1080 memcpy(__skb_put(skb, len),
1081 rx_bi->page + rx_bi->page_offset,
1083 rx_bi->page_offset += len;
1084 rx_packet_len -= len;
1087 /* Get the rest of the data if this was a header split */
1088 if (rx_packet_len) {
1089 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1094 skb->len += rx_packet_len;
1095 skb->data_len += rx_packet_len;
1096 skb->truesize += rx_packet_len;
1098 if ((page_count(rx_bi->page) == 1) &&
1099 (page_to_nid(rx_bi->page) == current_node))
1100 get_page(rx_bi->page);
1104 dma_unmap_page(rx_ring->dev,
1108 rx_bi->page_dma = 0;
1110 I40E_RX_INCREMENT(rx_ring, i);
1113 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1114 struct i40e_rx_buffer *next_buffer;
1116 next_buffer = &rx_ring->rx_bi[i];
1117 next_buffer->skb = skb;
1118 rx_ring->rx_stats.non_eop_descs++;
1122 /* ERR_MASK will only have valid bits if EOP set */
1123 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1124 dev_kfree_skb_any(skb);
1128 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1129 i40e_ptype_to_hash(rx_ptype));
1130 /* probably a little skewed due to removing CRC */
1131 total_rx_bytes += skb->len;
1134 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1136 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1138 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1139 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1142 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1143 dev_kfree_skb_any(skb);
1147 skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
1148 i40e_receive_skb(rx_ring, skb, vlan_tag);
1150 rx_desc->wb.qword1.status_error_len = 0;
1152 } while (likely(total_rx_packets < budget));
1154 u64_stats_update_begin(&rx_ring->syncp);
1155 rx_ring->stats.packets += total_rx_packets;
1156 rx_ring->stats.bytes += total_rx_bytes;
1157 u64_stats_update_end(&rx_ring->syncp);
1158 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1159 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1161 return total_rx_packets;
1165 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1166 * @rx_ring: rx ring to clean
1167 * @budget: how many cleans we're allowed
1169 * Returns number of packets cleaned
1171 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1173 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1174 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1175 struct i40e_vsi *vsi = rx_ring->vsi;
1176 union i40e_rx_desc *rx_desc;
1177 u32 rx_error, rx_status;
1184 struct i40e_rx_buffer *rx_bi;
1185 struct sk_buff *skb;
1187 /* return some buffers to hardware, one at a time is too slow */
1188 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1189 i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1193 i = rx_ring->next_to_clean;
1194 rx_desc = I40E_RX_DESC(rx_ring, i);
1195 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1196 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1197 I40E_RXD_QW1_STATUS_SHIFT;
1199 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1202 /* This memory barrier is needed to keep us from reading
1203 * any other fields out of the rx_desc until we know the
1208 rx_bi = &rx_ring->rx_bi[i];
1210 prefetch(skb->data);
1212 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1213 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1215 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1216 I40E_RXD_QW1_ERROR_SHIFT;
1217 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1219 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1220 I40E_RXD_QW1_PTYPE_SHIFT;
1224 /* Get the header and possibly the whole packet
1225 * If this is an skb from previous receive dma will be 0
1227 skb_put(skb, rx_packet_len);
1228 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1232 I40E_RX_INCREMENT(rx_ring, i);
1235 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1236 rx_ring->rx_stats.non_eop_descs++;
1240 /* ERR_MASK will only have valid bits if EOP set */
1241 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1242 dev_kfree_skb_any(skb);
1243 /* TODO: shouldn't we increment a counter indicating the
1249 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1250 i40e_ptype_to_hash(rx_ptype));
1251 /* probably a little skewed due to removing CRC */
1252 total_rx_bytes += skb->len;
1255 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1257 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1259 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1260 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1262 i40e_receive_skb(rx_ring, skb, vlan_tag);
1264 rx_desc->wb.qword1.status_error_len = 0;
1265 } while (likely(total_rx_packets < budget));
1267 u64_stats_update_begin(&rx_ring->syncp);
1268 rx_ring->stats.packets += total_rx_packets;
1269 rx_ring->stats.bytes += total_rx_bytes;
1270 u64_stats_update_end(&rx_ring->syncp);
1271 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1272 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1274 return total_rx_packets;
1278 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1279 * @vsi: the VSI we care about
1280 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1283 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1284 struct i40e_q_vector *q_vector)
1286 struct i40e_hw *hw = &vsi->back->hw;
1291 vector = (q_vector->v_idx + vsi->base_vector);
1292 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1293 old_itr = q_vector->rx.itr;
1294 i40e_set_new_dynamic_itr(&q_vector->rx);
1295 if (old_itr != q_vector->rx.itr) {
1296 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1297 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1299 I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1300 (q_vector->rx.itr <<
1301 I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1303 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1304 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1306 I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT);
1308 if (!test_bit(__I40E_DOWN, &vsi->state))
1309 wr32(hw, I40E_VFINT_DYN_CTLN1(vector - 1), val);
1311 i40evf_irq_enable_queues(vsi->back, 1
1312 << q_vector->v_idx);
1314 if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1315 old_itr = q_vector->tx.itr;
1316 i40e_set_new_dynamic_itr(&q_vector->tx);
1317 if (old_itr != q_vector->tx.itr) {
1318 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1319 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1321 I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1322 (q_vector->tx.itr <<
1323 I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1326 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1327 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1329 I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT);
1331 if (!test_bit(__I40E_DOWN, &vsi->state))
1332 wr32(hw, I40E_VFINT_DYN_CTLN1(vector - 1), val);
1334 i40evf_irq_enable_queues(vsi->back, BIT(q_vector->v_idx));
1339 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1340 * @napi: napi struct with our devices info in it
1341 * @budget: amount of work driver is allowed to do this pass, in packets
1343 * This function will clean all queues associated with a q_vector.
1345 * Returns the amount of work done
1347 int i40evf_napi_poll(struct napi_struct *napi, int budget)
1349 struct i40e_q_vector *q_vector =
1350 container_of(napi, struct i40e_q_vector, napi);
1351 struct i40e_vsi *vsi = q_vector->vsi;
1352 struct i40e_ring *ring;
1353 bool clean_complete = true;
1354 bool arm_wb = false;
1355 int budget_per_ring;
1358 if (test_bit(__I40E_DOWN, &vsi->state)) {
1359 napi_complete(napi);
1363 /* Since the actual Tx work is minimal, we can give the Tx a larger
1364 * budget and be more aggressive about cleaning up the Tx descriptors.
1366 i40e_for_each_ring(ring, q_vector->tx) {
1367 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1368 arm_wb |= ring->arm_wb;
1371 /* We attempt to distribute budget to each Rx queue fairly, but don't
1372 * allow the budget to go below 1 because that would exit polling early.
1374 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1376 i40e_for_each_ring(ring, q_vector->rx) {
1377 if (ring_is_ps_enabled(ring))
1378 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1380 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1381 /* if we didn't clean as many as budgeted, we must be done */
1382 clean_complete &= (budget_per_ring != cleaned);
1385 /* If work not completed, return budget and polling will return */
1386 if (!clean_complete) {
1388 i40e_force_wb(vsi, q_vector);
1392 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1393 q_vector->arm_wb_state = false;
1395 /* Work is done so exit the polling mode and re-enable the interrupt */
1396 napi_complete(napi);
1397 i40e_update_enable_itr(vsi, q_vector);
1402 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1404 * @tx_ring: ring to send buffer on
1405 * @flags: the tx flags to be set
1407 * Checks the skb and set up correspondingly several generic transmit flags
1408 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1410 * Returns error code indicate the frame should be dropped upon error and the
1411 * otherwise returns 0 to indicate the flags has been set properly.
1413 static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1414 struct i40e_ring *tx_ring,
1417 __be16 protocol = skb->protocol;
1420 if (protocol == htons(ETH_P_8021Q) &&
1421 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1422 /* When HW VLAN acceleration is turned off by the user the
1423 * stack sets the protocol to 8021q so that the driver
1424 * can take any steps required to support the SW only
1425 * VLAN handling. In our case the driver doesn't need
1426 * to take any further steps so just set the protocol
1427 * to the encapsulated ethertype.
1429 skb->protocol = vlan_get_protocol(skb);
1433 /* if we have a HW VLAN tag being added, default to the HW one */
1434 if (skb_vlan_tag_present(skb)) {
1435 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1436 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1437 /* else if it is a SW VLAN, check the next protocol and store the tag */
1438 } else if (protocol == htons(ETH_P_8021Q)) {
1439 struct vlan_hdr *vhdr, _vhdr;
1440 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1444 protocol = vhdr->h_vlan_encapsulated_proto;
1445 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1446 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1455 * i40e_tso - set up the tso context descriptor
1456 * @tx_ring: ptr to the ring to send
1457 * @skb: ptr to the skb we're sending
1458 * @hdr_len: ptr to the size of the packet header
1459 * @cd_tunneling: ptr to context descriptor bits
1461 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1463 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1464 u8 *hdr_len, u64 *cd_type_cmd_tso_mss,
1467 u32 cd_cmd, cd_tso_len, cd_mss;
1468 struct ipv6hdr *ipv6h;
1469 struct tcphdr *tcph;
1474 if (!skb_is_gso(skb))
1477 err = skb_cow_head(skb, 0);
1481 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1482 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1484 if (iph->version == 4) {
1485 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1488 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1490 } else if (ipv6h->version == 6) {
1491 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1492 ipv6h->payload_len = 0;
1493 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1497 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1498 *hdr_len = (skb->encapsulation
1499 ? (skb_inner_transport_header(skb) - skb->data)
1500 : skb_transport_offset(skb)) + l4len;
1502 /* find the field values */
1503 cd_cmd = I40E_TX_CTX_DESC_TSO;
1504 cd_tso_len = skb->len - *hdr_len;
1505 cd_mss = skb_shinfo(skb)->gso_size;
1506 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1508 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1509 ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1514 * i40e_tx_enable_csum - Enable Tx checksum offloads
1516 * @tx_flags: pointer to Tx flags currently set
1517 * @td_cmd: Tx descriptor command bits to set
1518 * @td_offset: Tx descriptor header offsets to set
1519 * @cd_tunneling: ptr to context desc bits
1521 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1522 u32 *td_cmd, u32 *td_offset,
1523 struct i40e_ring *tx_ring,
1526 struct ipv6hdr *this_ipv6_hdr;
1527 unsigned int this_tcp_hdrlen;
1528 struct iphdr *this_ip_hdr;
1529 u32 network_hdr_len;
1531 struct udphdr *oudph;
1535 if (skb->encapsulation) {
1536 switch (ip_hdr(skb)->protocol) {
1538 oudph = udp_hdr(skb);
1540 l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
1541 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1546 network_hdr_len = skb_inner_network_header_len(skb);
1547 this_ip_hdr = inner_ip_hdr(skb);
1548 this_ipv6_hdr = inner_ipv6_hdr(skb);
1549 this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1551 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1552 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1553 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1554 ip_hdr(skb)->check = 0;
1557 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1559 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1560 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1561 if (*tx_flags & I40E_TX_FLAGS_TSO)
1562 ip_hdr(skb)->check = 0;
1565 /* Now set the ctx descriptor fields */
1566 *cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1567 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1569 ((skb_inner_network_offset(skb) -
1570 skb_transport_offset(skb)) >> 1) <<
1571 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1572 if (this_ip_hdr->version == 6) {
1573 *tx_flags &= ~I40E_TX_FLAGS_IPV4;
1574 *tx_flags |= I40E_TX_FLAGS_IPV6;
1578 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
1579 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
1580 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
1581 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
1583 (skb->len - skb_transport_offset(skb)),
1585 *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1588 network_hdr_len = skb_network_header_len(skb);
1589 this_ip_hdr = ip_hdr(skb);
1590 this_ipv6_hdr = ipv6_hdr(skb);
1591 this_tcp_hdrlen = tcp_hdrlen(skb);
1594 /* Enable IP checksum offloads */
1595 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1596 l4_hdr = this_ip_hdr->protocol;
1597 /* the stack computes the IP header already, the only time we
1598 * need the hardware to recompute it is in the case of TSO.
1600 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1601 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1602 this_ip_hdr->check = 0;
1604 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1606 /* Now set the td_offset for IP header length */
1607 *td_offset = (network_hdr_len >> 2) <<
1608 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1609 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1610 l4_hdr = this_ipv6_hdr->nexthdr;
1611 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1612 /* Now set the td_offset for IP header length */
1613 *td_offset = (network_hdr_len >> 2) <<
1614 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1616 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1617 *td_offset |= (skb_network_offset(skb) >> 1) <<
1618 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1620 /* Enable L4 checksum offloads */
1623 /* enable checksum offloads */
1624 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1625 *td_offset |= (this_tcp_hdrlen >> 2) <<
1626 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1629 /* enable SCTP checksum offload */
1630 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1631 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
1632 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1635 /* enable UDP checksum offload */
1636 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1637 *td_offset |= (sizeof(struct udphdr) >> 2) <<
1638 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1646 * i40e_create_tx_ctx Build the Tx context descriptor
1647 * @tx_ring: ring to create the descriptor on
1648 * @cd_type_cmd_tso_mss: Quad Word 1
1649 * @cd_tunneling: Quad Word 0 - bits 0-31
1650 * @cd_l2tag2: Quad Word 0 - bits 32-63
1652 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1653 const u64 cd_type_cmd_tso_mss,
1654 const u32 cd_tunneling, const u32 cd_l2tag2)
1656 struct i40e_tx_context_desc *context_desc;
1657 int i = tx_ring->next_to_use;
1659 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1660 !cd_tunneling && !cd_l2tag2)
1663 /* grab the next descriptor */
1664 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1667 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1669 /* cpu_to_le32 and assign to struct fields */
1670 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1671 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1672 context_desc->rsvd = cpu_to_le16(0);
1673 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1677 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1679 * @tx_flags: collected send information
1681 * Note: Our HW can't scatter-gather more than 8 fragments to build
1682 * a packet on the wire and so we need to figure out the cases where we
1683 * need to linearize the skb.
1685 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1687 struct skb_frag_struct *frag;
1688 bool linearize = false;
1689 unsigned int size = 0;
1693 num_frags = skb_shinfo(skb)->nr_frags;
1694 gso_segs = skb_shinfo(skb)->gso_segs;
1696 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1699 if (num_frags < (I40E_MAX_BUFFER_TXD))
1700 goto linearize_chk_done;
1701 /* try the simple math, if we have too many frags per segment */
1702 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1703 I40E_MAX_BUFFER_TXD) {
1705 goto linearize_chk_done;
1707 frag = &skb_shinfo(skb)->frags[0];
1708 /* we might still have more fragments per segment */
1710 size += skb_frag_size(frag);
1712 if ((size >= skb_shinfo(skb)->gso_size) &&
1713 (j < I40E_MAX_BUFFER_TXD)) {
1714 size = (size % skb_shinfo(skb)->gso_size);
1717 if (j == I40E_MAX_BUFFER_TXD) {
1722 } while (num_frags);
1724 if (num_frags >= I40E_MAX_BUFFER_TXD)
1733 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1734 * @tx_ring: the ring to be checked
1735 * @size: the size buffer we want to assure is available
1737 * Returns -EBUSY if a stop is needed, else 0
1739 static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1741 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1742 /* Memory barrier before checking head and tail */
1745 /* Check again in a case another CPU has just made room available. */
1746 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1749 /* A reprieve! - use start_queue because it doesn't call schedule */
1750 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1751 ++tx_ring->tx_stats.restart_queue;
1756 * i40evf_maybe_stop_tx - 1st level check for tx stop conditions
1757 * @tx_ring: the ring to be checked
1758 * @size: the size buffer we want to assure is available
1760 * Returns 0 if stop is not needed
1762 static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1764 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1766 return __i40evf_maybe_stop_tx(tx_ring, size);
1770 * i40evf_tx_map - Build the Tx descriptor
1771 * @tx_ring: ring to send buffer on
1773 * @first: first buffer info buffer to use
1774 * @tx_flags: collected send information
1775 * @hdr_len: size of the packet header
1776 * @td_cmd: the command field in the descriptor
1777 * @td_offset: offset for checksum or crc
1779 static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1780 struct i40e_tx_buffer *first, u32 tx_flags,
1781 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1783 unsigned int data_len = skb->data_len;
1784 unsigned int size = skb_headlen(skb);
1785 struct skb_frag_struct *frag;
1786 struct i40e_tx_buffer *tx_bi;
1787 struct i40e_tx_desc *tx_desc;
1788 u16 i = tx_ring->next_to_use;
1793 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1794 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1795 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1796 I40E_TX_FLAGS_VLAN_SHIFT;
1799 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1800 gso_segs = skb_shinfo(skb)->gso_segs;
1804 /* multiply data chunks by size of headers */
1805 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1806 first->gso_segs = gso_segs;
1808 first->tx_flags = tx_flags;
1810 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1812 tx_desc = I40E_TX_DESC(tx_ring, i);
1815 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1816 if (dma_mapping_error(tx_ring->dev, dma))
1819 /* record length, and DMA address */
1820 dma_unmap_len_set(tx_bi, len, size);
1821 dma_unmap_addr_set(tx_bi, dma, dma);
1823 tx_desc->buffer_addr = cpu_to_le64(dma);
1825 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1826 tx_desc->cmd_type_offset_bsz =
1827 build_ctob(td_cmd, td_offset,
1828 I40E_MAX_DATA_PER_TXD, td_tag);
1832 if (i == tx_ring->count) {
1833 tx_desc = I40E_TX_DESC(tx_ring, 0);
1837 dma += I40E_MAX_DATA_PER_TXD;
1838 size -= I40E_MAX_DATA_PER_TXD;
1840 tx_desc->buffer_addr = cpu_to_le64(dma);
1843 if (likely(!data_len))
1846 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1851 if (i == tx_ring->count) {
1852 tx_desc = I40E_TX_DESC(tx_ring, 0);
1856 size = skb_frag_size(frag);
1859 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1862 tx_bi = &tx_ring->tx_bi[i];
1865 /* Place RS bit on last descriptor of any packet that spans across the
1866 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1868 #define WB_STRIDE 0x3
1869 if (((i & WB_STRIDE) != WB_STRIDE) &&
1870 (first <= &tx_ring->tx_bi[i]) &&
1871 (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1872 tx_desc->cmd_type_offset_bsz =
1873 build_ctob(td_cmd, td_offset, size, td_tag) |
1874 cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1875 I40E_TXD_QW1_CMD_SHIFT);
1877 tx_desc->cmd_type_offset_bsz =
1878 build_ctob(td_cmd, td_offset, size, td_tag) |
1879 cpu_to_le64((u64)I40E_TXD_CMD <<
1880 I40E_TXD_QW1_CMD_SHIFT);
1883 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1884 tx_ring->queue_index),
1887 /* Force memory writes to complete before letting h/w
1888 * know there are new descriptors to fetch. (Only
1889 * applicable for weak-ordered memory model archs,
1894 /* set next_to_watch value indicating a packet is present */
1895 first->next_to_watch = tx_desc;
1898 if (i == tx_ring->count)
1901 tx_ring->next_to_use = i;
1903 i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED);
1904 /* notify HW of packet */
1905 if (!skb->xmit_more ||
1906 netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1907 tx_ring->queue_index)))
1908 writel(i, tx_ring->tail);
1910 prefetchw(tx_desc + 1);
1915 dev_info(tx_ring->dev, "TX DMA map failed\n");
1917 /* clear dma mappings for failed tx_bi map */
1919 tx_bi = &tx_ring->tx_bi[i];
1920 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1928 tx_ring->next_to_use = i;
1932 * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed
1934 * @tx_ring: ring to send buffer on
1936 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1937 * there is not enough descriptors available in this ring since we need at least
1940 static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb,
1941 struct i40e_ring *tx_ring)
1946 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1947 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1948 * + 4 desc gap to avoid the cache line where head is,
1949 * + 1 desc for context descriptor,
1950 * otherwise try next time
1952 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1953 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1955 count += TXD_USE_COUNT(skb_headlen(skb));
1956 if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1957 tx_ring->tx_stats.tx_busy++;
1964 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1966 * @tx_ring: ring to send buffer on
1968 * Returns NETDEV_TX_OK if sent, else an error code
1970 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1971 struct i40e_ring *tx_ring)
1973 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1974 u32 cd_tunneling = 0, cd_l2tag2 = 0;
1975 struct i40e_tx_buffer *first;
1982 if (0 == i40evf_xmit_descriptor_count(skb, tx_ring))
1983 return NETDEV_TX_BUSY;
1985 /* prepare the xmit flags */
1986 if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1989 /* obtain protocol of skb */
1990 protocol = vlan_get_protocol(skb);
1992 /* record the location of the first descriptor for this packet */
1993 first = &tx_ring->tx_bi[tx_ring->next_to_use];
1995 /* setup IPv4/IPv6 offloads */
1996 if (protocol == htons(ETH_P_IP))
1997 tx_flags |= I40E_TX_FLAGS_IPV4;
1998 else if (protocol == htons(ETH_P_IPV6))
1999 tx_flags |= I40E_TX_FLAGS_IPV6;
2001 tso = i40e_tso(tx_ring, skb, &hdr_len,
2002 &cd_type_cmd_tso_mss, &cd_tunneling);
2007 tx_flags |= I40E_TX_FLAGS_TSO;
2009 if (i40e_chk_linearize(skb, tx_flags))
2010 if (skb_linearize(skb))
2013 skb_tx_timestamp(skb);
2015 /* always enable CRC insertion offload */
2016 td_cmd |= I40E_TX_DESC_CMD_ICRC;
2018 /* Always offload the checksum, since it's in the data descriptor */
2019 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2020 tx_flags |= I40E_TX_FLAGS_CSUM;
2022 i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2023 tx_ring, &cd_tunneling);
2026 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2027 cd_tunneling, cd_l2tag2);
2029 i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2032 return NETDEV_TX_OK;
2035 dev_kfree_skb_any(skb);
2036 return NETDEV_TX_OK;
2040 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2042 * @netdev: network interface device structure
2044 * Returns NETDEV_TX_OK if sent, else an error code
2046 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2048 struct i40evf_adapter *adapter = netdev_priv(netdev);
2049 struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
2051 /* hardware can't handle really short frames, hardware padding works
2054 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2055 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2056 return NETDEV_TX_OK;
2057 skb->len = I40E_MIN_TX_LEN;
2058 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2061 return i40e_xmit_frame_ring(skb, tx_ring);