1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
4 * Copyright(c) 2013 - 2016 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_trace.h"
32 #include "i40e_prototype.h"
34 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
37 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
38 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
39 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
40 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
41 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
44 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
47 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
48 * @ring: the ring that owns the buffer
49 * @tx_buffer: the buffer to free
51 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
52 struct i40e_tx_buffer *tx_buffer)
55 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
56 kfree(tx_buffer->raw_buf);
58 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),
71 tx_buffer->next_to_watch = NULL;
72 tx_buffer->skb = NULL;
73 dma_unmap_len_set(tx_buffer, len, 0);
74 /* tx_buffer must be completely set up in the transmit path */
78 * i40evf_clean_tx_ring - Free any empty Tx buffers
79 * @tx_ring: ring to be cleaned
81 void i40evf_clean_tx_ring(struct i40e_ring *tx_ring)
83 unsigned long bi_size;
86 /* ring already cleared, nothing to do */
90 /* Free all the Tx ring sk_buffs */
91 for (i = 0; i < tx_ring->count; i++)
92 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
94 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
95 memset(tx_ring->tx_bi, 0, bi_size);
97 /* Zero out the descriptor ring */
98 memset(tx_ring->desc, 0, tx_ring->size);
100 tx_ring->next_to_use = 0;
101 tx_ring->next_to_clean = 0;
103 if (!tx_ring->netdev)
106 /* cleanup Tx queue statistics */
107 netdev_tx_reset_queue(txring_txq(tx_ring));
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 * i40evf_get_tx_pending - how many Tx descriptors not processed
131 * @tx_ring: the ring of descriptors
132 * @in_sw: is tx_pending being checked in SW or HW
134 * Since there is no access to the ring head register
135 * in XL710, we need to use our local copies
137 u32 i40evf_get_tx_pending(struct i40e_ring *ring, bool in_sw)
141 head = ring->next_to_clean;
142 tail = readl(ring->tail);
145 return (head < tail) ?
146 tail - head : (tail + ring->count - head);
154 * i40e_clean_tx_irq - Reclaim resources after transmit completes
155 * @vsi: the VSI we care about
156 * @tx_ring: Tx ring to clean
157 * @napi_budget: Used to determine if we are in netpoll
159 * Returns true if there's any budget left (e.g. the clean is finished)
161 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
162 struct i40e_ring *tx_ring, int napi_budget)
164 u16 i = tx_ring->next_to_clean;
165 struct i40e_tx_buffer *tx_buf;
166 struct i40e_tx_desc *tx_desc;
167 unsigned int total_bytes = 0, total_packets = 0;
168 unsigned int budget = vsi->work_limit;
170 tx_buf = &tx_ring->tx_bi[i];
171 tx_desc = I40E_TX_DESC(tx_ring, i);
175 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
177 /* if next_to_watch is not set then there is no work pending */
181 /* prevent any other reads prior to eop_desc */
184 i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
185 /* if the descriptor isn't done, no work yet to do */
186 if (!(eop_desc->cmd_type_offset_bsz &
187 cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE)))
190 /* clear next_to_watch to prevent false hangs */
191 tx_buf->next_to_watch = NULL;
193 /* update the statistics for this packet */
194 total_bytes += tx_buf->bytecount;
195 total_packets += tx_buf->gso_segs;
198 napi_consume_skb(tx_buf->skb, napi_budget);
200 /* unmap skb header data */
201 dma_unmap_single(tx_ring->dev,
202 dma_unmap_addr(tx_buf, dma),
203 dma_unmap_len(tx_buf, len),
206 /* clear tx_buffer data */
208 dma_unmap_len_set(tx_buf, len, 0);
210 /* unmap remaining buffers */
211 while (tx_desc != eop_desc) {
212 i40e_trace(clean_tx_irq_unmap,
213 tx_ring, tx_desc, tx_buf);
220 tx_buf = tx_ring->tx_bi;
221 tx_desc = I40E_TX_DESC(tx_ring, 0);
224 /* unmap any remaining paged data */
225 if (dma_unmap_len(tx_buf, len)) {
226 dma_unmap_page(tx_ring->dev,
227 dma_unmap_addr(tx_buf, dma),
228 dma_unmap_len(tx_buf, len),
230 dma_unmap_len_set(tx_buf, len, 0);
234 /* move us one more past the eop_desc for start of next pkt */
240 tx_buf = tx_ring->tx_bi;
241 tx_desc = I40E_TX_DESC(tx_ring, 0);
246 /* update budget accounting */
248 } while (likely(budget));
251 tx_ring->next_to_clean = i;
252 u64_stats_update_begin(&tx_ring->syncp);
253 tx_ring->stats.bytes += total_bytes;
254 tx_ring->stats.packets += total_packets;
255 u64_stats_update_end(&tx_ring->syncp);
256 tx_ring->q_vector->tx.total_bytes += total_bytes;
257 tx_ring->q_vector->tx.total_packets += total_packets;
259 if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
260 /* check to see if there are < 4 descriptors
261 * waiting to be written back, then kick the hardware to force
262 * them to be written back in case we stay in NAPI.
263 * In this mode on X722 we do not enable Interrupt.
265 unsigned int j = i40evf_get_tx_pending(tx_ring, false);
268 ((j / WB_STRIDE) == 0) && (j > 0) &&
269 !test_bit(__I40E_VSI_DOWN, vsi->state) &&
270 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
271 tx_ring->arm_wb = true;
274 /* notify netdev of completed buffers */
275 netdev_tx_completed_queue(txring_txq(tx_ring),
276 total_packets, total_bytes);
278 #define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
279 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
280 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
281 /* Make sure that anybody stopping the queue after this
282 * sees the new next_to_clean.
285 if (__netif_subqueue_stopped(tx_ring->netdev,
286 tx_ring->queue_index) &&
287 !test_bit(__I40E_VSI_DOWN, vsi->state)) {
288 netif_wake_subqueue(tx_ring->netdev,
289 tx_ring->queue_index);
290 ++tx_ring->tx_stats.restart_queue;
298 * i40evf_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
299 * @vsi: the VSI we care about
300 * @q_vector: the vector on which to enable writeback
303 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
304 struct i40e_q_vector *q_vector)
306 u16 flags = q_vector->tx.ring[0].flags;
309 if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
312 if (q_vector->arm_wb_state)
315 val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
316 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK; /* set noitr */
319 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
320 vsi->base_vector - 1), val);
321 q_vector->arm_wb_state = true;
325 * i40evf_force_wb - Issue SW Interrupt so HW does a wb
326 * @vsi: the VSI we care about
327 * @q_vector: the vector on which to force writeback
330 void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
332 u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
333 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
334 I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
335 I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK
336 /* allow 00 to be written to the index */;
339 I40E_VFINT_DYN_CTLN1(q_vector->v_idx + vsi->base_vector - 1),
344 * i40e_set_new_dynamic_itr - Find new ITR level
345 * @rc: structure containing ring performance data
347 * Returns true if ITR changed, false if not
349 * Stores a new ITR value based on packets and byte counts during
350 * the last interrupt. The advantage of per interrupt computation
351 * is faster updates and more accurate ITR for the current traffic
352 * pattern. Constants in this function were computed based on
353 * theoretical maximum wire speed and thresholds were set based on
354 * testing data as well as attempting to minimize response time
355 * while increasing bulk throughput.
357 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
359 enum i40e_latency_range new_latency_range = rc->latency_range;
360 u32 new_itr = rc->itr;
362 unsigned int usecs, estimated_usecs;
364 if (rc->total_packets == 0 || !rc->itr)
367 usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
368 bytes_per_int = rc->total_bytes / usecs;
370 /* The calculations in this algorithm depend on interrupts actually
371 * firing at the ITR rate. This may not happen if the packet rate is
372 * really low, or if we've been napi polling. Check to make sure
373 * that's not the case before we continue.
375 estimated_usecs = jiffies_to_usecs(jiffies - rc->last_itr_update);
376 if (estimated_usecs > usecs) {
377 new_latency_range = I40E_LOW_LATENCY;
381 /* simple throttlerate management
382 * 0-10MB/s lowest (50000 ints/s)
383 * 10-20MB/s low (20000 ints/s)
384 * 20-1249MB/s bulk (18000 ints/s)
386 * The math works out because the divisor is in 10^(-6) which
387 * turns the bytes/us input value into MB/s values, but
388 * make sure to use usecs, as the register values written
389 * are in 2 usec increments in the ITR registers, and make sure
390 * to use the smoothed values that the countdown timer gives us.
392 switch (new_latency_range) {
393 case I40E_LOWEST_LATENCY:
394 if (bytes_per_int > 10)
395 new_latency_range = I40E_LOW_LATENCY;
397 case I40E_LOW_LATENCY:
398 if (bytes_per_int > 20)
399 new_latency_range = I40E_BULK_LATENCY;
400 else if (bytes_per_int <= 10)
401 new_latency_range = I40E_LOWEST_LATENCY;
403 case I40E_BULK_LATENCY:
405 if (bytes_per_int <= 20)
406 new_latency_range = I40E_LOW_LATENCY;
411 rc->latency_range = new_latency_range;
413 switch (new_latency_range) {
414 case I40E_LOWEST_LATENCY:
415 new_itr = I40E_ITR_50K;
417 case I40E_LOW_LATENCY:
418 new_itr = I40E_ITR_20K;
420 case I40E_BULK_LATENCY:
421 new_itr = I40E_ITR_18K;
428 rc->total_packets = 0;
429 rc->last_itr_update = jiffies;
431 if (new_itr != rc->itr) {
439 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
440 * @tx_ring: the tx ring to set up
442 * Return 0 on success, negative on error
444 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
446 struct device *dev = tx_ring->dev;
452 /* warn if we are about to overwrite the pointer */
453 WARN_ON(tx_ring->tx_bi);
454 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
455 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
459 /* round up to nearest 4K */
460 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
461 tx_ring->size = ALIGN(tx_ring->size, 4096);
462 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
463 &tx_ring->dma, GFP_KERNEL);
464 if (!tx_ring->desc) {
465 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
470 tx_ring->next_to_use = 0;
471 tx_ring->next_to_clean = 0;
475 kfree(tx_ring->tx_bi);
476 tx_ring->tx_bi = NULL;
481 * i40evf_clean_rx_ring - Free Rx buffers
482 * @rx_ring: ring to be cleaned
484 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
486 unsigned long bi_size;
489 /* ring already cleared, nothing to do */
494 dev_kfree_skb(rx_ring->skb);
498 /* Free all the Rx ring sk_buffs */
499 for (i = 0; i < rx_ring->count; i++) {
500 struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
505 /* Invalidate cache lines that may have been written to by
506 * device so that we avoid corrupting memory.
508 dma_sync_single_range_for_cpu(rx_ring->dev,
514 /* free resources associated with mapping */
515 dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
516 i40e_rx_pg_size(rx_ring),
520 __page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
523 rx_bi->page_offset = 0;
526 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
527 memset(rx_ring->rx_bi, 0, bi_size);
529 /* Zero out the descriptor ring */
530 memset(rx_ring->desc, 0, rx_ring->size);
532 rx_ring->next_to_alloc = 0;
533 rx_ring->next_to_clean = 0;
534 rx_ring->next_to_use = 0;
538 * i40evf_free_rx_resources - Free Rx resources
539 * @rx_ring: ring to clean the resources from
541 * Free all receive software resources
543 void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
545 i40evf_clean_rx_ring(rx_ring);
546 kfree(rx_ring->rx_bi);
547 rx_ring->rx_bi = NULL;
550 dma_free_coherent(rx_ring->dev, rx_ring->size,
551 rx_ring->desc, rx_ring->dma);
552 rx_ring->desc = NULL;
557 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
558 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
560 * Returns 0 on success, negative on failure
562 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
564 struct device *dev = rx_ring->dev;
567 /* warn if we are about to overwrite the pointer */
568 WARN_ON(rx_ring->rx_bi);
569 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
570 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
574 u64_stats_init(&rx_ring->syncp);
576 /* Round up to nearest 4K */
577 rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
578 rx_ring->size = ALIGN(rx_ring->size, 4096);
579 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
580 &rx_ring->dma, GFP_KERNEL);
582 if (!rx_ring->desc) {
583 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
588 rx_ring->next_to_alloc = 0;
589 rx_ring->next_to_clean = 0;
590 rx_ring->next_to_use = 0;
594 kfree(rx_ring->rx_bi);
595 rx_ring->rx_bi = NULL;
600 * i40e_release_rx_desc - Store the new tail and head values
601 * @rx_ring: ring to bump
602 * @val: new head index
604 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
606 rx_ring->next_to_use = val;
608 /* update next to alloc since we have filled the ring */
609 rx_ring->next_to_alloc = val;
611 /* Force memory writes to complete before letting h/w
612 * know there are new descriptors to fetch. (Only
613 * applicable for weak-ordered memory model archs,
617 writel(val, rx_ring->tail);
621 * i40e_rx_offset - Return expected offset into page to access data
622 * @rx_ring: Ring we are requesting offset of
624 * Returns the offset value for ring into the data buffer.
626 static inline unsigned int i40e_rx_offset(struct i40e_ring *rx_ring)
628 return ring_uses_build_skb(rx_ring) ? I40E_SKB_PAD : 0;
632 * i40e_alloc_mapped_page - recycle or make a new page
633 * @rx_ring: ring to use
634 * @bi: rx_buffer struct to modify
636 * Returns true if the page was successfully allocated or
639 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
640 struct i40e_rx_buffer *bi)
642 struct page *page = bi->page;
645 /* since we are recycling buffers we should seldom need to alloc */
647 rx_ring->rx_stats.page_reuse_count++;
651 /* alloc new page for storage */
652 page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
653 if (unlikely(!page)) {
654 rx_ring->rx_stats.alloc_page_failed++;
658 /* map page for use */
659 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
660 i40e_rx_pg_size(rx_ring),
664 /* if mapping failed free memory back to system since
665 * there isn't much point in holding memory we can't use
667 if (dma_mapping_error(rx_ring->dev, dma)) {
668 __free_pages(page, i40e_rx_pg_order(rx_ring));
669 rx_ring->rx_stats.alloc_page_failed++;
675 bi->page_offset = i40e_rx_offset(rx_ring);
677 /* initialize pagecnt_bias to 1 representing we fully own page */
678 bi->pagecnt_bias = 1;
684 * i40e_receive_skb - Send a completed packet up the stack
685 * @rx_ring: rx ring in play
686 * @skb: packet to send up
687 * @vlan_tag: vlan tag for packet
689 static void i40e_receive_skb(struct i40e_ring *rx_ring,
690 struct sk_buff *skb, u16 vlan_tag)
692 struct i40e_q_vector *q_vector = rx_ring->q_vector;
694 if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
695 (vlan_tag & VLAN_VID_MASK))
696 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
698 napi_gro_receive(&q_vector->napi, skb);
702 * i40evf_alloc_rx_buffers - Replace used receive buffers
703 * @rx_ring: ring to place buffers on
704 * @cleaned_count: number of buffers to replace
706 * Returns false if all allocations were successful, true if any fail
708 bool i40evf_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
710 u16 ntu = rx_ring->next_to_use;
711 union i40e_rx_desc *rx_desc;
712 struct i40e_rx_buffer *bi;
714 /* do nothing if no valid netdev defined */
715 if (!rx_ring->netdev || !cleaned_count)
718 rx_desc = I40E_RX_DESC(rx_ring, ntu);
719 bi = &rx_ring->rx_bi[ntu];
722 if (!i40e_alloc_mapped_page(rx_ring, bi))
725 /* sync the buffer for use by the device */
726 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
731 /* Refresh the desc even if buffer_addrs didn't change
732 * because each write-back erases this info.
734 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
739 if (unlikely(ntu == rx_ring->count)) {
740 rx_desc = I40E_RX_DESC(rx_ring, 0);
745 /* clear the status bits for the next_to_use descriptor */
746 rx_desc->wb.qword1.status_error_len = 0;
749 } while (cleaned_count);
751 if (rx_ring->next_to_use != ntu)
752 i40e_release_rx_desc(rx_ring, ntu);
757 if (rx_ring->next_to_use != ntu)
758 i40e_release_rx_desc(rx_ring, ntu);
760 /* make sure to come back via polling to try again after
767 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
768 * @vsi: the VSI we care about
769 * @skb: skb currently being received and modified
770 * @rx_desc: the receive descriptor
772 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
774 union i40e_rx_desc *rx_desc)
776 struct i40e_rx_ptype_decoded decoded;
777 u32 rx_error, rx_status;
782 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
783 ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
784 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
785 I40E_RXD_QW1_ERROR_SHIFT;
786 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
787 I40E_RXD_QW1_STATUS_SHIFT;
788 decoded = decode_rx_desc_ptype(ptype);
790 skb->ip_summed = CHECKSUM_NONE;
792 skb_checksum_none_assert(skb);
794 /* Rx csum enabled and ip headers found? */
795 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
798 /* did the hardware decode the packet and checksum? */
799 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
802 /* both known and outer_ip must be set for the below code to work */
803 if (!(decoded.known && decoded.outer_ip))
806 ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
807 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
808 ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
809 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
812 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
813 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
816 /* likely incorrect csum if alternate IP extension headers found */
818 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
819 /* don't increment checksum err here, non-fatal err */
822 /* there was some L4 error, count error and punt packet to the stack */
823 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
826 /* handle packets that were not able to be checksummed due
827 * to arrival speed, in this case the stack can compute
830 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
833 /* Only report checksum unnecessary for TCP, UDP, or SCTP */
834 switch (decoded.inner_prot) {
835 case I40E_RX_PTYPE_INNER_PROT_TCP:
836 case I40E_RX_PTYPE_INNER_PROT_UDP:
837 case I40E_RX_PTYPE_INNER_PROT_SCTP:
838 skb->ip_summed = CHECKSUM_UNNECESSARY;
847 vsi->back->hw_csum_rx_error++;
851 * i40e_ptype_to_htype - get a hash type
852 * @ptype: the ptype value from the descriptor
854 * Returns a hash type to be used by skb_set_hash
856 static inline int i40e_ptype_to_htype(u8 ptype)
858 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
861 return PKT_HASH_TYPE_NONE;
863 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
864 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
865 return PKT_HASH_TYPE_L4;
866 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
867 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
868 return PKT_HASH_TYPE_L3;
870 return PKT_HASH_TYPE_L2;
874 * i40e_rx_hash - set the hash value in the skb
875 * @ring: descriptor ring
876 * @rx_desc: specific descriptor
878 static inline void i40e_rx_hash(struct i40e_ring *ring,
879 union i40e_rx_desc *rx_desc,
884 const __le64 rss_mask =
885 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
886 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
888 if (ring->netdev->features & NETIF_F_RXHASH)
891 if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
892 hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
893 skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
898 * i40evf_process_skb_fields - Populate skb header fields from Rx descriptor
899 * @rx_ring: rx descriptor ring packet is being transacted on
900 * @rx_desc: pointer to the EOP Rx descriptor
901 * @skb: pointer to current skb being populated
902 * @rx_ptype: the packet type decoded by hardware
904 * This function checks the ring, descriptor, and packet information in
905 * order to populate the hash, checksum, VLAN, protocol, and
906 * other fields within the skb.
909 void i40evf_process_skb_fields(struct i40e_ring *rx_ring,
910 union i40e_rx_desc *rx_desc, struct sk_buff *skb,
913 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
915 i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
917 skb_record_rx_queue(skb, rx_ring->queue_index);
919 /* modifies the skb - consumes the enet header */
920 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
924 * i40e_cleanup_headers - Correct empty headers
925 * @rx_ring: rx descriptor ring packet is being transacted on
926 * @skb: pointer to current skb being fixed
928 * Also address the case where we are pulling data in on pages only
929 * and as such no data is present in the skb header.
931 * In addition if skb is not at least 60 bytes we need to pad it so that
932 * it is large enough to qualify as a valid Ethernet frame.
934 * Returns true if an error was encountered and skb was freed.
936 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
938 /* if eth_skb_pad returns an error the skb was freed */
939 if (eth_skb_pad(skb))
946 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
947 * @rx_ring: rx descriptor ring to store buffers on
948 * @old_buff: donor buffer to have page reused
950 * Synchronizes page for reuse by the adapter
952 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
953 struct i40e_rx_buffer *old_buff)
955 struct i40e_rx_buffer *new_buff;
956 u16 nta = rx_ring->next_to_alloc;
958 new_buff = &rx_ring->rx_bi[nta];
960 /* update, and store next to alloc */
962 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
964 /* transfer page from old buffer to new buffer */
965 new_buff->dma = old_buff->dma;
966 new_buff->page = old_buff->page;
967 new_buff->page_offset = old_buff->page_offset;
968 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
972 * i40e_page_is_reusable - check if any reuse is possible
973 * @page: page struct to check
975 * A page is not reusable if it was allocated under low memory
976 * conditions, or it's not in the same NUMA node as this CPU.
978 static inline bool i40e_page_is_reusable(struct page *page)
980 return (page_to_nid(page) == numa_mem_id()) &&
981 !page_is_pfmemalloc(page);
985 * i40e_can_reuse_rx_page - Determine if this page can be reused by
986 * the adapter for another receive
988 * @rx_buffer: buffer containing the page
990 * If page is reusable, rx_buffer->page_offset is adjusted to point to
991 * an unused region in the page.
993 * For small pages, @truesize will be a constant value, half the size
994 * of the memory at page. We'll attempt to alternate between high and
995 * low halves of the page, with one half ready for use by the hardware
996 * and the other half being consumed by the stack. We use the page
997 * ref count to determine whether the stack has finished consuming the
998 * portion of this page that was passed up with a previous packet. If
999 * the page ref count is >1, we'll assume the "other" half page is
1000 * still busy, and this page cannot be reused.
1002 * For larger pages, @truesize will be the actual space used by the
1003 * received packet (adjusted upward to an even multiple of the cache
1004 * line size). This will advance through the page by the amount
1005 * actually consumed by the received packets while there is still
1006 * space for a buffer. Each region of larger pages will be used at
1007 * most once, after which the page will not be reused.
1009 * In either case, if the page is reusable its refcount is increased.
1011 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer)
1013 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1014 struct page *page = rx_buffer->page;
1016 /* Is any reuse possible? */
1017 if (unlikely(!i40e_page_is_reusable(page)))
1020 #if (PAGE_SIZE < 8192)
1021 /* if we are only owner of page we can reuse it */
1022 if (unlikely((page_count(page) - pagecnt_bias) > 1))
1025 #define I40E_LAST_OFFSET \
1026 (SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
1027 if (rx_buffer->page_offset > I40E_LAST_OFFSET)
1031 /* If we have drained the page fragment pool we need to update
1032 * the pagecnt_bias and page count so that we fully restock the
1033 * number of references the driver holds.
1035 if (unlikely(!pagecnt_bias)) {
1036 page_ref_add(page, USHRT_MAX);
1037 rx_buffer->pagecnt_bias = USHRT_MAX;
1044 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1045 * @rx_ring: rx descriptor ring to transact packets on
1046 * @rx_buffer: buffer containing page to add
1047 * @skb: sk_buff to place the data into
1048 * @size: packet length from rx_desc
1050 * This function will add the data contained in rx_buffer->page to the skb.
1051 * It will just attach the page as a frag to the skb.
1053 * The function will then update the page offset.
1055 static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
1056 struct i40e_rx_buffer *rx_buffer,
1057 struct sk_buff *skb,
1060 #if (PAGE_SIZE < 8192)
1061 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1063 unsigned int truesize = SKB_DATA_ALIGN(size + i40e_rx_offset(rx_ring));
1066 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1067 rx_buffer->page_offset, size, truesize);
1069 /* page is being used so we must update the page offset */
1070 #if (PAGE_SIZE < 8192)
1071 rx_buffer->page_offset ^= truesize;
1073 rx_buffer->page_offset += truesize;
1078 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
1079 * @rx_ring: rx descriptor ring to transact packets on
1080 * @size: size of buffer to add to skb
1082 * This function will pull an Rx buffer from the ring and synchronize it
1083 * for use by the CPU.
1085 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
1086 const unsigned int size)
1088 struct i40e_rx_buffer *rx_buffer;
1090 rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1091 prefetchw(rx_buffer->page);
1093 /* we are reusing so sync this buffer for CPU use */
1094 dma_sync_single_range_for_cpu(rx_ring->dev,
1096 rx_buffer->page_offset,
1100 /* We have pulled a buffer for use, so decrement pagecnt_bias */
1101 rx_buffer->pagecnt_bias--;
1107 * i40e_construct_skb - Allocate skb and populate it
1108 * @rx_ring: rx descriptor ring to transact packets on
1109 * @rx_buffer: rx buffer to pull data from
1110 * @size: size of buffer to add to skb
1112 * This function allocates an skb. It then populates it with the page
1113 * data from the current receive descriptor, taking care to set up the
1116 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
1117 struct i40e_rx_buffer *rx_buffer,
1121 #if (PAGE_SIZE < 8192)
1122 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1124 unsigned int truesize = SKB_DATA_ALIGN(size);
1126 unsigned int headlen;
1127 struct sk_buff *skb;
1129 /* prefetch first cache line of first page */
1130 va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1132 #if L1_CACHE_BYTES < 128
1133 prefetch(va + L1_CACHE_BYTES);
1136 /* allocate a skb to store the frags */
1137 skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1139 GFP_ATOMIC | __GFP_NOWARN);
1143 /* Determine available headroom for copy */
1145 if (headlen > I40E_RX_HDR_SIZE)
1146 headlen = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1148 /* align pull length to size of long to optimize memcpy performance */
1149 memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1151 /* update all of the pointers */
1154 skb_add_rx_frag(skb, 0, rx_buffer->page,
1155 rx_buffer->page_offset + headlen,
1158 /* buffer is used by skb, update page_offset */
1159 #if (PAGE_SIZE < 8192)
1160 rx_buffer->page_offset ^= truesize;
1162 rx_buffer->page_offset += truesize;
1165 /* buffer is unused, reset bias back to rx_buffer */
1166 rx_buffer->pagecnt_bias++;
1173 * i40e_build_skb - Build skb around an existing buffer
1174 * @rx_ring: Rx descriptor ring to transact packets on
1175 * @rx_buffer: Rx buffer to pull data from
1176 * @size: size of buffer to add to skb
1178 * This function builds an skb around an existing Rx buffer, taking care
1179 * to set up the skb correctly and avoid any memcpy overhead.
1181 static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
1182 struct i40e_rx_buffer *rx_buffer,
1186 #if (PAGE_SIZE < 8192)
1187 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1189 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1190 SKB_DATA_ALIGN(I40E_SKB_PAD + size);
1192 struct sk_buff *skb;
1194 /* prefetch first cache line of first page */
1195 va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1197 #if L1_CACHE_BYTES < 128
1198 prefetch(va + L1_CACHE_BYTES);
1200 /* build an skb around the page buffer */
1201 skb = build_skb(va - I40E_SKB_PAD, truesize);
1205 /* update pointers within the skb to store the data */
1206 skb_reserve(skb, I40E_SKB_PAD);
1207 __skb_put(skb, size);
1209 /* buffer is used by skb, update page_offset */
1210 #if (PAGE_SIZE < 8192)
1211 rx_buffer->page_offset ^= truesize;
1213 rx_buffer->page_offset += truesize;
1220 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
1221 * @rx_ring: rx descriptor ring to transact packets on
1222 * @rx_buffer: rx buffer to pull data from
1224 * This function will clean up the contents of the rx_buffer. It will
1225 * either recycle the bufer or unmap it and free the associated resources.
1227 static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
1228 struct i40e_rx_buffer *rx_buffer)
1230 if (i40e_can_reuse_rx_page(rx_buffer)) {
1231 /* hand second half of page back to the ring */
1232 i40e_reuse_rx_page(rx_ring, rx_buffer);
1233 rx_ring->rx_stats.page_reuse_count++;
1235 /* we are not reusing the buffer so unmap it */
1236 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
1237 i40e_rx_pg_size(rx_ring),
1238 DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
1239 __page_frag_cache_drain(rx_buffer->page,
1240 rx_buffer->pagecnt_bias);
1243 /* clear contents of buffer_info */
1244 rx_buffer->page = NULL;
1248 * i40e_is_non_eop - process handling of non-EOP buffers
1249 * @rx_ring: Rx ring being processed
1250 * @rx_desc: Rx descriptor for current buffer
1251 * @skb: Current socket buffer containing buffer in progress
1253 * This function updates next to clean. If the buffer is an EOP buffer
1254 * this function exits returning false, otherwise it will place the
1255 * sk_buff in the next buffer to be chained and return true indicating
1256 * that this is in fact a non-EOP buffer.
1258 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1259 union i40e_rx_desc *rx_desc,
1260 struct sk_buff *skb)
1262 u32 ntc = rx_ring->next_to_clean + 1;
1264 /* fetch, update, and store next to clean */
1265 ntc = (ntc < rx_ring->count) ? ntc : 0;
1266 rx_ring->next_to_clean = ntc;
1268 prefetch(I40E_RX_DESC(rx_ring, ntc));
1270 /* if we are the last buffer then there is nothing else to do */
1271 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1272 if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1275 rx_ring->rx_stats.non_eop_descs++;
1281 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1282 * @rx_ring: rx descriptor ring to transact packets on
1283 * @budget: Total limit on number of packets to process
1285 * This function provides a "bounce buffer" approach to Rx interrupt
1286 * processing. The advantage to this is that on systems that have
1287 * expensive overhead for IOMMU access this provides a means of avoiding
1288 * it by maintaining the mapping of the page to the system.
1290 * Returns amount of work completed
1292 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1294 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1295 struct sk_buff *skb = rx_ring->skb;
1296 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1297 bool failure = false;
1299 while (likely(total_rx_packets < (unsigned int)budget)) {
1300 struct i40e_rx_buffer *rx_buffer;
1301 union i40e_rx_desc *rx_desc;
1307 /* return some buffers to hardware, one at a time is too slow */
1308 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1309 failure = failure ||
1310 i40evf_alloc_rx_buffers(rx_ring, cleaned_count);
1314 rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1316 /* status_error_len will always be zero for unused descriptors
1317 * because it's cleared in cleanup, and overlaps with hdr_addr
1318 * which is always zero because packet split isn't used, if the
1319 * hardware wrote DD then the length will be non-zero
1321 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1323 /* This memory barrier is needed to keep us from reading
1324 * any other fields out of the rx_desc until we have
1325 * verified the descriptor has been written back.
1329 size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1330 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1334 i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
1335 rx_buffer = i40e_get_rx_buffer(rx_ring, size);
1337 /* retrieve a buffer from the ring */
1339 i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
1340 else if (ring_uses_build_skb(rx_ring))
1341 skb = i40e_build_skb(rx_ring, rx_buffer, size);
1343 skb = i40e_construct_skb(rx_ring, rx_buffer, size);
1345 /* exit if we failed to retrieve a buffer */
1347 rx_ring->rx_stats.alloc_buff_failed++;
1348 rx_buffer->pagecnt_bias++;
1352 i40e_put_rx_buffer(rx_ring, rx_buffer);
1355 if (i40e_is_non_eop(rx_ring, rx_desc, skb))
1358 /* ERR_MASK will only have valid bits if EOP set, and
1359 * what we are doing here is actually checking
1360 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1363 if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1364 dev_kfree_skb_any(skb);
1369 if (i40e_cleanup_headers(rx_ring, skb)) {
1374 /* probably a little skewed due to removing CRC */
1375 total_rx_bytes += skb->len;
1377 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1378 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1379 I40E_RXD_QW1_PTYPE_SHIFT;
1381 /* populate checksum, VLAN, and protocol */
1382 i40evf_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
1385 vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
1386 le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
1388 i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
1389 i40e_receive_skb(rx_ring, skb, vlan_tag);
1392 /* update budget accounting */
1398 u64_stats_update_begin(&rx_ring->syncp);
1399 rx_ring->stats.packets += total_rx_packets;
1400 rx_ring->stats.bytes += total_rx_bytes;
1401 u64_stats_update_end(&rx_ring->syncp);
1402 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1403 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1405 /* guarantee a trip back through this routine if there was a failure */
1406 return failure ? budget : (int)total_rx_packets;
1409 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1413 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1414 /* Don't clear PBA because that can cause lost interrupts that
1415 * came in while we were cleaning/polling
1417 (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1418 (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1423 /* a small macro to shorten up some long lines */
1424 #define INTREG I40E_VFINT_DYN_CTLN1
1425 static inline int get_rx_itr(struct i40e_vsi *vsi, int idx)
1427 struct i40evf_adapter *adapter = vsi->back;
1429 return adapter->rx_rings[idx].rx_itr_setting;
1432 static inline int get_tx_itr(struct i40e_vsi *vsi, int idx)
1434 struct i40evf_adapter *adapter = vsi->back;
1436 return adapter->tx_rings[idx].tx_itr_setting;
1440 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1441 * @vsi: the VSI we care about
1442 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1445 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1446 struct i40e_q_vector *q_vector)
1448 struct i40e_hw *hw = &vsi->back->hw;
1449 bool rx = false, tx = false;
1452 int idx = q_vector->v_idx;
1453 int rx_itr_setting, tx_itr_setting;
1455 vector = (q_vector->v_idx + vsi->base_vector);
1457 /* avoid dynamic calculation if in countdown mode OR if
1458 * all dynamic is disabled
1460 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1462 rx_itr_setting = get_rx_itr(vsi, idx);
1463 tx_itr_setting = get_tx_itr(vsi, idx);
1465 if (q_vector->itr_countdown > 0 ||
1466 (!ITR_IS_DYNAMIC(rx_itr_setting) &&
1467 !ITR_IS_DYNAMIC(tx_itr_setting))) {
1471 if (ITR_IS_DYNAMIC(rx_itr_setting)) {
1472 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1473 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1476 if (ITR_IS_DYNAMIC(tx_itr_setting)) {
1477 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1478 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1482 /* get the higher of the two ITR adjustments and
1483 * use the same value for both ITR registers
1484 * when in adaptive mode (Rx and/or Tx)
1486 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1488 q_vector->tx.itr = q_vector->rx.itr = itr;
1489 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1491 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1495 /* only need to enable the interrupt once, but need
1496 * to possibly update both ITR values
1499 /* set the INTENA_MSK_MASK so that this first write
1500 * won't actually enable the interrupt, instead just
1501 * updating the ITR (it's bit 31 PF and VF)
1504 /* don't check _DOWN because interrupt isn't being enabled */
1505 wr32(hw, INTREG(vector - 1), rxval);
1509 if (!test_bit(__I40E_VSI_DOWN, vsi->state))
1510 wr32(hw, INTREG(vector - 1), txval);
1512 if (q_vector->itr_countdown)
1513 q_vector->itr_countdown--;
1515 q_vector->itr_countdown = ITR_COUNTDOWN_START;
1519 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1520 * @napi: napi struct with our devices info in it
1521 * @budget: amount of work driver is allowed to do this pass, in packets
1523 * This function will clean all queues associated with a q_vector.
1525 * Returns the amount of work done
1527 int i40evf_napi_poll(struct napi_struct *napi, int budget)
1529 struct i40e_q_vector *q_vector =
1530 container_of(napi, struct i40e_q_vector, napi);
1531 struct i40e_vsi *vsi = q_vector->vsi;
1532 struct i40e_ring *ring;
1533 bool clean_complete = true;
1534 bool arm_wb = false;
1535 int budget_per_ring;
1538 if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
1539 napi_complete(napi);
1543 /* Since the actual Tx work is minimal, we can give the Tx a larger
1544 * budget and be more aggressive about cleaning up the Tx descriptors.
1546 i40e_for_each_ring(ring, q_vector->tx) {
1547 if (!i40e_clean_tx_irq(vsi, ring, budget)) {
1548 clean_complete = false;
1551 arm_wb |= ring->arm_wb;
1552 ring->arm_wb = false;
1555 /* Handle case where we are called by netpoll with a budget of 0 */
1559 /* We attempt to distribute budget to each Rx queue fairly, but don't
1560 * allow the budget to go below 1 because that would exit polling early.
1562 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1564 i40e_for_each_ring(ring, q_vector->rx) {
1565 int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
1567 work_done += cleaned;
1568 /* if we clean as many as budgeted, we must not be done */
1569 if (cleaned >= budget_per_ring)
1570 clean_complete = false;
1573 /* If work not completed, return budget and polling will return */
1574 if (!clean_complete) {
1575 int cpu_id = smp_processor_id();
1577 /* It is possible that the interrupt affinity has changed but,
1578 * if the cpu is pegged at 100%, polling will never exit while
1579 * traffic continues and the interrupt will be stuck on this
1580 * cpu. We check to make sure affinity is correct before we
1581 * continue to poll, otherwise we must stop polling so the
1582 * interrupt can move to the correct cpu.
1584 if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
1585 /* Tell napi that we are done polling */
1586 napi_complete_done(napi, work_done);
1588 /* Force an interrupt */
1589 i40evf_force_wb(vsi, q_vector);
1591 /* Return budget-1 so that polling stops */
1596 q_vector->tx.ring[0].tx_stats.tx_force_wb++;
1597 i40e_enable_wb_on_itr(vsi, q_vector);
1602 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1603 q_vector->arm_wb_state = false;
1605 /* Work is done so exit the polling mode and re-enable the interrupt */
1606 napi_complete_done(napi, work_done);
1608 i40e_update_enable_itr(vsi, q_vector);
1610 return min(work_done, budget - 1);
1614 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1616 * @tx_ring: ring to send buffer on
1617 * @flags: the tx flags to be set
1619 * Checks the skb and set up correspondingly several generic transmit flags
1620 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1622 * Returns error code indicate the frame should be dropped upon error and the
1623 * otherwise returns 0 to indicate the flags has been set properly.
1625 static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1626 struct i40e_ring *tx_ring,
1629 __be16 protocol = skb->protocol;
1632 if (protocol == htons(ETH_P_8021Q) &&
1633 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1634 /* When HW VLAN acceleration is turned off by the user the
1635 * stack sets the protocol to 8021q so that the driver
1636 * can take any steps required to support the SW only
1637 * VLAN handling. In our case the driver doesn't need
1638 * to take any further steps so just set the protocol
1639 * to the encapsulated ethertype.
1641 skb->protocol = vlan_get_protocol(skb);
1645 /* if we have a HW VLAN tag being added, default to the HW one */
1646 if (skb_vlan_tag_present(skb)) {
1647 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1648 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1649 /* else if it is a SW VLAN, check the next protocol and store the tag */
1650 } else if (protocol == htons(ETH_P_8021Q)) {
1651 struct vlan_hdr *vhdr, _vhdr;
1653 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1657 protocol = vhdr->h_vlan_encapsulated_proto;
1658 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1659 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1668 * i40e_tso - set up the tso context descriptor
1669 * @first: pointer to first Tx buffer for xmit
1670 * @hdr_len: ptr to the size of the packet header
1671 * @cd_type_cmd_tso_mss: Quad Word 1
1673 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1675 static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
1676 u64 *cd_type_cmd_tso_mss)
1678 struct sk_buff *skb = first->skb;
1679 u64 cd_cmd, cd_tso_len, cd_mss;
1690 u32 paylen, l4_offset;
1691 u16 gso_segs, gso_size;
1694 if (skb->ip_summed != CHECKSUM_PARTIAL)
1697 if (!skb_is_gso(skb))
1700 err = skb_cow_head(skb, 0);
1704 ip.hdr = skb_network_header(skb);
1705 l4.hdr = skb_transport_header(skb);
1707 /* initialize outer IP header fields */
1708 if (ip.v4->version == 4) {
1712 ip.v6->payload_len = 0;
1715 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
1719 SKB_GSO_UDP_TUNNEL |
1720 SKB_GSO_UDP_TUNNEL_CSUM)) {
1721 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
1722 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
1725 /* determine offset of outer transport header */
1726 l4_offset = l4.hdr - skb->data;
1728 /* remove payload length from outer checksum */
1729 paylen = skb->len - l4_offset;
1730 csum_replace_by_diff(&l4.udp->check,
1731 (__force __wsum)htonl(paylen));
1734 /* reset pointers to inner headers */
1735 ip.hdr = skb_inner_network_header(skb);
1736 l4.hdr = skb_inner_transport_header(skb);
1738 /* initialize inner IP header fields */
1739 if (ip.v4->version == 4) {
1743 ip.v6->payload_len = 0;
1747 /* determine offset of inner transport header */
1748 l4_offset = l4.hdr - skb->data;
1750 /* remove payload length from inner checksum */
1751 paylen = skb->len - l4_offset;
1752 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
1754 /* compute length of segmentation header */
1755 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
1757 /* pull values out of skb_shinfo */
1758 gso_size = skb_shinfo(skb)->gso_size;
1759 gso_segs = skb_shinfo(skb)->gso_segs;
1761 /* update GSO size and bytecount with header size */
1762 first->gso_segs = gso_segs;
1763 first->bytecount += (first->gso_segs - 1) * *hdr_len;
1765 /* find the field values */
1766 cd_cmd = I40E_TX_CTX_DESC_TSO;
1767 cd_tso_len = skb->len - *hdr_len;
1769 *cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1770 (cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1771 (cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1776 * i40e_tx_enable_csum - Enable Tx checksum offloads
1778 * @tx_flags: pointer to Tx flags currently set
1779 * @td_cmd: Tx descriptor command bits to set
1780 * @td_offset: Tx descriptor header offsets to set
1781 * @tx_ring: Tx descriptor ring
1782 * @cd_tunneling: ptr to context desc bits
1784 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1785 u32 *td_cmd, u32 *td_offset,
1786 struct i40e_ring *tx_ring,
1799 unsigned char *exthdr;
1800 u32 offset, cmd = 0;
1804 if (skb->ip_summed != CHECKSUM_PARTIAL)
1807 ip.hdr = skb_network_header(skb);
1808 l4.hdr = skb_transport_header(skb);
1810 /* compute outer L2 header size */
1811 offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1813 if (skb->encapsulation) {
1815 /* define outer network header type */
1816 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1817 tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
1818 I40E_TX_CTX_EXT_IP_IPV4 :
1819 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1821 l4_proto = ip.v4->protocol;
1822 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1823 tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
1825 exthdr = ip.hdr + sizeof(*ip.v6);
1826 l4_proto = ip.v6->nexthdr;
1827 if (l4.hdr != exthdr)
1828 ipv6_skip_exthdr(skb, exthdr - skb->data,
1829 &l4_proto, &frag_off);
1832 /* define outer transport */
1835 tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
1836 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1839 tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
1840 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1844 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1845 l4.hdr = skb_inner_network_header(skb);
1848 if (*tx_flags & I40E_TX_FLAGS_TSO)
1851 skb_checksum_help(skb);
1855 /* compute outer L3 header size */
1856 tunnel |= ((l4.hdr - ip.hdr) / 4) <<
1857 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
1859 /* switch IP header pointer from outer to inner header */
1860 ip.hdr = skb_inner_network_header(skb);
1862 /* compute tunnel header size */
1863 tunnel |= ((ip.hdr - l4.hdr) / 2) <<
1864 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1866 /* indicate if we need to offload outer UDP header */
1867 if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
1868 !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
1869 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
1870 tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1872 /* record tunnel offload values */
1873 *cd_tunneling |= tunnel;
1875 /* switch L4 header pointer from outer to inner */
1876 l4.hdr = skb_inner_transport_header(skb);
1879 /* reset type as we transition from outer to inner headers */
1880 *tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
1881 if (ip.v4->version == 4)
1882 *tx_flags |= I40E_TX_FLAGS_IPV4;
1883 if (ip.v6->version == 6)
1884 *tx_flags |= I40E_TX_FLAGS_IPV6;
1887 /* Enable IP checksum offloads */
1888 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1889 l4_proto = ip.v4->protocol;
1890 /* the stack computes the IP header already, the only time we
1891 * need the hardware to recompute it is in the case of TSO.
1893 cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
1894 I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
1895 I40E_TX_DESC_CMD_IIPT_IPV4;
1896 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1897 cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1899 exthdr = ip.hdr + sizeof(*ip.v6);
1900 l4_proto = ip.v6->nexthdr;
1901 if (l4.hdr != exthdr)
1902 ipv6_skip_exthdr(skb, exthdr - skb->data,
1903 &l4_proto, &frag_off);
1906 /* compute inner L3 header size */
1907 offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1909 /* Enable L4 checksum offloads */
1912 /* enable checksum offloads */
1913 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1914 offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1917 /* enable SCTP checksum offload */
1918 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1919 offset |= (sizeof(struct sctphdr) >> 2) <<
1920 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1923 /* enable UDP checksum offload */
1924 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1925 offset |= (sizeof(struct udphdr) >> 2) <<
1926 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1929 if (*tx_flags & I40E_TX_FLAGS_TSO)
1931 skb_checksum_help(skb);
1936 *td_offset |= offset;
1942 * i40e_create_tx_ctx Build the Tx context descriptor
1943 * @tx_ring: ring to create the descriptor on
1944 * @cd_type_cmd_tso_mss: Quad Word 1
1945 * @cd_tunneling: Quad Word 0 - bits 0-31
1946 * @cd_l2tag2: Quad Word 0 - bits 32-63
1948 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1949 const u64 cd_type_cmd_tso_mss,
1950 const u32 cd_tunneling, const u32 cd_l2tag2)
1952 struct i40e_tx_context_desc *context_desc;
1953 int i = tx_ring->next_to_use;
1955 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1956 !cd_tunneling && !cd_l2tag2)
1959 /* grab the next descriptor */
1960 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1963 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1965 /* cpu_to_le32 and assign to struct fields */
1966 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1967 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1968 context_desc->rsvd = cpu_to_le16(0);
1969 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1973 * __i40evf_chk_linearize - Check if there are more than 8 buffers per packet
1976 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
1977 * and so we need to figure out the cases where we need to linearize the skb.
1979 * For TSO we need to count the TSO header and segment payload separately.
1980 * As such we need to check cases where we have 7 fragments or more as we
1981 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
1982 * the segment payload in the first descriptor, and another 7 for the
1985 bool __i40evf_chk_linearize(struct sk_buff *skb)
1987 const struct skb_frag_struct *frag, *stale;
1990 /* no need to check if number of frags is less than 7 */
1991 nr_frags = skb_shinfo(skb)->nr_frags;
1992 if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
1995 /* We need to walk through the list and validate that each group
1996 * of 6 fragments totals at least gso_size.
1998 nr_frags -= I40E_MAX_BUFFER_TXD - 2;
1999 frag = &skb_shinfo(skb)->frags[0];
2001 /* Initialize size to the negative value of gso_size minus 1. We
2002 * use this as the worst case scenerio in which the frag ahead
2003 * of us only provides one byte which is why we are limited to 6
2004 * descriptors for a single transmit as the header and previous
2005 * fragment are already consuming 2 descriptors.
2007 sum = 1 - skb_shinfo(skb)->gso_size;
2009 /* Add size of frags 0 through 4 to create our initial sum */
2010 sum += skb_frag_size(frag++);
2011 sum += skb_frag_size(frag++);
2012 sum += skb_frag_size(frag++);
2013 sum += skb_frag_size(frag++);
2014 sum += skb_frag_size(frag++);
2016 /* Walk through fragments adding latest fragment, testing it, and
2017 * then removing stale fragments from the sum.
2019 for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
2020 int stale_size = skb_frag_size(stale);
2022 sum += skb_frag_size(frag++);
2024 /* The stale fragment may present us with a smaller
2025 * descriptor than the actual fragment size. To account
2026 * for that we need to remove all the data on the front and
2027 * figure out what the remainder would be in the last
2028 * descriptor associated with the fragment.
2030 if (stale_size > I40E_MAX_DATA_PER_TXD) {
2031 int align_pad = -(stale->page_offset) &
2032 (I40E_MAX_READ_REQ_SIZE - 1);
2035 stale_size -= align_pad;
2038 sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
2039 stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
2040 } while (stale_size > I40E_MAX_DATA_PER_TXD);
2043 /* if sum is negative we failed to make sufficient progress */
2057 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
2058 * @tx_ring: the ring to be checked
2059 * @size: the size buffer we want to assure is available
2061 * Returns -EBUSY if a stop is needed, else 0
2063 int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2065 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2066 /* Memory barrier before checking head and tail */
2069 /* Check again in a case another CPU has just made room available. */
2070 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2073 /* A reprieve! - use start_queue because it doesn't call schedule */
2074 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2075 ++tx_ring->tx_stats.restart_queue;
2080 * i40evf_tx_map - Build the Tx descriptor
2081 * @tx_ring: ring to send buffer on
2083 * @first: first buffer info buffer to use
2084 * @tx_flags: collected send information
2085 * @hdr_len: size of the packet header
2086 * @td_cmd: the command field in the descriptor
2087 * @td_offset: offset for checksum or crc
2089 static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2090 struct i40e_tx_buffer *first, u32 tx_flags,
2091 const u8 hdr_len, u32 td_cmd, u32 td_offset)
2093 unsigned int data_len = skb->data_len;
2094 unsigned int size = skb_headlen(skb);
2095 struct skb_frag_struct *frag;
2096 struct i40e_tx_buffer *tx_bi;
2097 struct i40e_tx_desc *tx_desc;
2098 u16 i = tx_ring->next_to_use;
2102 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2103 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2104 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2105 I40E_TX_FLAGS_VLAN_SHIFT;
2108 first->tx_flags = tx_flags;
2110 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2112 tx_desc = I40E_TX_DESC(tx_ring, i);
2115 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2116 unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2118 if (dma_mapping_error(tx_ring->dev, dma))
2121 /* record length, and DMA address */
2122 dma_unmap_len_set(tx_bi, len, size);
2123 dma_unmap_addr_set(tx_bi, dma, dma);
2125 /* align size to end of page */
2126 max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2127 tx_desc->buffer_addr = cpu_to_le64(dma);
2129 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2130 tx_desc->cmd_type_offset_bsz =
2131 build_ctob(td_cmd, td_offset,
2137 if (i == tx_ring->count) {
2138 tx_desc = I40E_TX_DESC(tx_ring, 0);
2145 max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2146 tx_desc->buffer_addr = cpu_to_le64(dma);
2149 if (likely(!data_len))
2152 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2158 if (i == tx_ring->count) {
2159 tx_desc = I40E_TX_DESC(tx_ring, 0);
2163 size = skb_frag_size(frag);
2166 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2169 tx_bi = &tx_ring->tx_bi[i];
2172 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
2175 if (i == tx_ring->count)
2178 tx_ring->next_to_use = i;
2180 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2182 /* write last descriptor with RS and EOP bits */
2183 td_cmd |= I40E_TXD_CMD;
2184 tx_desc->cmd_type_offset_bsz =
2185 build_ctob(td_cmd, td_offset, size, td_tag);
2187 /* Force memory writes to complete before letting h/w know there
2188 * are new descriptors to fetch.
2190 * We also use this memory barrier to make certain all of the
2191 * status bits have been updated before next_to_watch is written.
2195 /* set next_to_watch value indicating a packet is present */
2196 first->next_to_watch = tx_desc;
2198 /* notify HW of packet */
2199 if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
2200 writel(i, tx_ring->tail);
2202 /* we need this if more than one processor can write to our tail
2203 * at a time, it synchronizes IO on IA64/Altix systems
2211 dev_info(tx_ring->dev, "TX DMA map failed\n");
2213 /* clear dma mappings for failed tx_bi map */
2215 tx_bi = &tx_ring->tx_bi[i];
2216 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2224 tx_ring->next_to_use = i;
2228 * i40e_xmit_frame_ring - Sends buffer on Tx ring
2230 * @tx_ring: ring to send buffer on
2232 * Returns NETDEV_TX_OK if sent, else an error code
2234 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2235 struct i40e_ring *tx_ring)
2237 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2238 u32 cd_tunneling = 0, cd_l2tag2 = 0;
2239 struct i40e_tx_buffer *first;
2247 /* prefetch the data, we'll need it later */
2248 prefetch(skb->data);
2250 i40e_trace(xmit_frame_ring, skb, tx_ring);
2252 count = i40e_xmit_descriptor_count(skb);
2253 if (i40e_chk_linearize(skb, count)) {
2254 if (__skb_linearize(skb)) {
2255 dev_kfree_skb_any(skb);
2256 return NETDEV_TX_OK;
2258 count = i40e_txd_use_count(skb->len);
2259 tx_ring->tx_stats.tx_linearize++;
2262 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2263 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2264 * + 4 desc gap to avoid the cache line where head is,
2265 * + 1 desc for context descriptor,
2266 * otherwise try next time
2268 if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2269 tx_ring->tx_stats.tx_busy++;
2270 return NETDEV_TX_BUSY;
2273 /* record the location of the first descriptor for this packet */
2274 first = &tx_ring->tx_bi[tx_ring->next_to_use];
2276 first->bytecount = skb->len;
2277 first->gso_segs = 1;
2279 /* prepare the xmit flags */
2280 if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2283 /* obtain protocol of skb */
2284 protocol = vlan_get_protocol(skb);
2286 /* setup IPv4/IPv6 offloads */
2287 if (protocol == htons(ETH_P_IP))
2288 tx_flags |= I40E_TX_FLAGS_IPV4;
2289 else if (protocol == htons(ETH_P_IPV6))
2290 tx_flags |= I40E_TX_FLAGS_IPV6;
2292 tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
2297 tx_flags |= I40E_TX_FLAGS_TSO;
2299 /* Always offload the checksum, since it's in the data descriptor */
2300 tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2301 tx_ring, &cd_tunneling);
2305 skb_tx_timestamp(skb);
2307 /* always enable CRC insertion offload */
2308 td_cmd |= I40E_TX_DESC_CMD_ICRC;
2310 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2311 cd_tunneling, cd_l2tag2);
2313 i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2316 return NETDEV_TX_OK;
2319 i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
2320 dev_kfree_skb_any(first->skb);
2322 return NETDEV_TX_OK;
2326 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2328 * @netdev: network interface device structure
2330 * Returns NETDEV_TX_OK if sent, else an error code
2332 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2334 struct i40evf_adapter *adapter = netdev_priv(netdev);
2335 struct i40e_ring *tx_ring = &adapter->tx_rings[skb->queue_mapping];
2337 /* hardware can't handle really short frames, hardware padding works
2340 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2341 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2342 return NETDEV_TX_OK;
2343 skb->len = I40E_MIN_TX_LEN;
2344 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2347 return i40e_xmit_frame_ring(skb, tx_ring);