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 dev_kfree_skb_any(tx_buffer->skb);
55 if (dma_unmap_len(tx_buffer, len))
56 dma_unmap_single(ring->dev,
57 dma_unmap_addr(tx_buffer, dma),
58 dma_unmap_len(tx_buffer, len),
60 } else if (dma_unmap_len(tx_buffer, len)) {
61 dma_unmap_page(ring->dev,
62 dma_unmap_addr(tx_buffer, dma),
63 dma_unmap_len(tx_buffer, len),
67 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
68 kfree(tx_buffer->raw_buf);
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 * i40evf_get_tx_pending - how many Tx descriptors not processed
131 * @tx_ring: the ring of descriptors
133 * Since there is no access to the ring head register
134 * in XL710, we need to use our local copies
136 u32 i40evf_get_tx_pending(struct i40e_ring *ring)
140 head = i40e_get_head(ring);
141 tail = readl(ring->tail);
144 return (head < tail) ?
145 tail - head : (tail + ring->count - head);
150 #define WB_STRIDE 0x3
153 * i40e_clean_tx_irq - Reclaim resources after transmit completes
154 * @tx_ring: tx ring to clean
155 * @budget: how many cleans we're allowed
157 * Returns true if there's any budget left (e.g. the clean is finished)
159 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
161 u16 i = tx_ring->next_to_clean;
162 struct i40e_tx_buffer *tx_buf;
163 struct i40e_tx_desc *tx_head;
164 struct i40e_tx_desc *tx_desc;
165 unsigned int total_packets = 0;
166 unsigned int total_bytes = 0;
168 tx_buf = &tx_ring->tx_bi[i];
169 tx_desc = I40E_TX_DESC(tx_ring, i);
172 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
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 */
182 read_barrier_depends();
184 /* we have caught up to head, no work left to do */
185 if (tx_head == tx_desc)
188 /* clear next_to_watch to prevent false hangs */
189 tx_buf->next_to_watch = NULL;
191 /* update the statistics for this packet */
192 total_bytes += tx_buf->bytecount;
193 total_packets += tx_buf->gso_segs;
196 dev_kfree_skb_any(tx_buf->skb);
198 /* unmap skb header data */
199 dma_unmap_single(tx_ring->dev,
200 dma_unmap_addr(tx_buf, dma),
201 dma_unmap_len(tx_buf, len),
204 /* clear tx_buffer data */
206 dma_unmap_len_set(tx_buf, len, 0);
208 /* unmap remaining buffers */
209 while (tx_desc != eop_desc) {
216 tx_buf = tx_ring->tx_bi;
217 tx_desc = I40E_TX_DESC(tx_ring, 0);
220 /* unmap any remaining paged data */
221 if (dma_unmap_len(tx_buf, len)) {
222 dma_unmap_page(tx_ring->dev,
223 dma_unmap_addr(tx_buf, dma),
224 dma_unmap_len(tx_buf, len),
226 dma_unmap_len_set(tx_buf, len, 0);
230 /* move us one more past the eop_desc for start of next pkt */
236 tx_buf = tx_ring->tx_bi;
237 tx_desc = I40E_TX_DESC(tx_ring, 0);
242 /* update budget accounting */
244 } while (likely(budget));
247 tx_ring->next_to_clean = i;
248 u64_stats_update_begin(&tx_ring->syncp);
249 tx_ring->stats.bytes += total_bytes;
250 tx_ring->stats.packets += total_packets;
251 u64_stats_update_end(&tx_ring->syncp);
252 tx_ring->q_vector->tx.total_bytes += total_bytes;
253 tx_ring->q_vector->tx.total_packets += total_packets;
255 if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
257 /* check to see if there are < 4 descriptors
258 * waiting to be written back, then kick the hardware to force
259 * them to be written back in case we stay in NAPI.
260 * In this mode on X722 we do not enable Interrupt.
262 j = i40evf_get_tx_pending(tx_ring);
265 ((j / (WB_STRIDE + 1)) == 0) && (j > 0) &&
266 !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
267 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
268 tx_ring->arm_wb = true;
271 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
272 tx_ring->queue_index),
273 total_packets, total_bytes);
275 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
276 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
277 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
278 /* Make sure that anybody stopping the queue after this
279 * sees the new next_to_clean.
282 if (__netif_subqueue_stopped(tx_ring->netdev,
283 tx_ring->queue_index) &&
284 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
285 netif_wake_subqueue(tx_ring->netdev,
286 tx_ring->queue_index);
287 ++tx_ring->tx_stats.restart_queue;
295 * i40evf_force_wb -Arm hardware to do a wb on noncache aligned descriptors
296 * @vsi: the VSI we care about
297 * @q_vector: the vector on which to force writeback
300 static void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
302 u16 flags = q_vector->tx.ring[0].flags;
304 if (flags & I40E_TXR_FLAGS_WB_ON_ITR) {
307 if (q_vector->arm_wb_state)
310 val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
311 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK; /* set noitr */
314 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
315 vsi->base_vector - 1),
317 q_vector->arm_wb_state = true;
319 u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
320 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
321 I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
322 I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK;
323 /* allow 00 to be written to the index */
326 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
327 vsi->base_vector - 1), val);
332 * i40e_set_new_dynamic_itr - Find new ITR level
333 * @rc: structure containing ring performance data
335 * Returns true if ITR changed, false if not
337 * Stores a new ITR value based on packets and byte counts during
338 * the last interrupt. The advantage of per interrupt computation
339 * is faster updates and more accurate ITR for the current traffic
340 * pattern. Constants in this function were computed based on
341 * theoretical maximum wire speed and thresholds were set based on
342 * testing data as well as attempting to minimize response time
343 * while increasing bulk throughput.
345 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
347 enum i40e_latency_range new_latency_range = rc->latency_range;
348 struct i40e_q_vector *qv = rc->ring->q_vector;
349 u32 new_itr = rc->itr;
353 if (rc->total_packets == 0 || !rc->itr)
356 /* simple throttlerate management
357 * 0-10MB/s lowest (50000 ints/s)
358 * 10-20MB/s low (20000 ints/s)
359 * 20-1249MB/s bulk (18000 ints/s)
360 * > 40000 Rx packets per second (8000 ints/s)
362 * The math works out because the divisor is in 10^(-6) which
363 * turns the bytes/us input value into MB/s values, but
364 * make sure to use usecs, as the register values written
365 * are in 2 usec increments in the ITR registers, and make sure
366 * to use the smoothed values that the countdown timer gives us.
368 usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
369 bytes_per_int = rc->total_bytes / usecs;
371 switch (new_latency_range) {
372 case I40E_LOWEST_LATENCY:
373 if (bytes_per_int > 10)
374 new_latency_range = I40E_LOW_LATENCY;
376 case I40E_LOW_LATENCY:
377 if (bytes_per_int > 20)
378 new_latency_range = I40E_BULK_LATENCY;
379 else if (bytes_per_int <= 10)
380 new_latency_range = I40E_LOWEST_LATENCY;
382 case I40E_BULK_LATENCY:
383 case I40E_ULTRA_LATENCY:
385 if (bytes_per_int <= 20)
386 new_latency_range = I40E_LOW_LATENCY;
390 /* this is to adjust RX more aggressively when streaming small
391 * packets. The value of 40000 was picked as it is just beyond
392 * what the hardware can receive per second if in low latency
395 #define RX_ULTRA_PACKET_RATE 40000
397 if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
399 new_latency_range = I40E_ULTRA_LATENCY;
401 rc->latency_range = new_latency_range;
403 switch (new_latency_range) {
404 case I40E_LOWEST_LATENCY:
405 new_itr = I40E_ITR_50K;
407 case I40E_LOW_LATENCY:
408 new_itr = I40E_ITR_20K;
410 case I40E_BULK_LATENCY:
411 new_itr = I40E_ITR_18K;
413 case I40E_ULTRA_LATENCY:
414 new_itr = I40E_ITR_8K;
421 rc->total_packets = 0;
423 if (new_itr != rc->itr) {
432 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
433 * @tx_ring: the tx ring to set up
435 * Return 0 on success, negative on error
437 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
439 struct device *dev = tx_ring->dev;
445 /* warn if we are about to overwrite the pointer */
446 WARN_ON(tx_ring->tx_bi);
447 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
448 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
452 /* round up to nearest 4K */
453 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
454 /* add u32 for head writeback, align after this takes care of
455 * guaranteeing this is at least one cache line in size
457 tx_ring->size += sizeof(u32);
458 tx_ring->size = ALIGN(tx_ring->size, 4096);
459 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
460 &tx_ring->dma, GFP_KERNEL);
461 if (!tx_ring->desc) {
462 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
467 tx_ring->next_to_use = 0;
468 tx_ring->next_to_clean = 0;
472 kfree(tx_ring->tx_bi);
473 tx_ring->tx_bi = NULL;
478 * i40evf_clean_rx_ring - Free Rx buffers
479 * @rx_ring: ring to be cleaned
481 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
483 struct device *dev = rx_ring->dev;
484 struct i40e_rx_buffer *rx_bi;
485 unsigned long bi_size;
488 /* ring already cleared, nothing to do */
492 if (ring_is_ps_enabled(rx_ring)) {
493 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
495 rx_bi = &rx_ring->rx_bi[0];
496 if (rx_bi->hdr_buf) {
497 dma_free_coherent(dev,
501 for (i = 0; i < rx_ring->count; i++) {
502 rx_bi = &rx_ring->rx_bi[i];
504 rx_bi->hdr_buf = NULL;
508 /* Free all the Rx ring sk_buffs */
509 for (i = 0; i < rx_ring->count; i++) {
510 rx_bi = &rx_ring->rx_bi[i];
512 dma_unmap_single(dev,
519 dev_kfree_skb(rx_bi->skb);
523 if (rx_bi->page_dma) {
530 __free_page(rx_bi->page);
532 rx_bi->page_offset = 0;
536 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
537 memset(rx_ring->rx_bi, 0, bi_size);
539 /* Zero out the descriptor ring */
540 memset(rx_ring->desc, 0, rx_ring->size);
542 rx_ring->next_to_clean = 0;
543 rx_ring->next_to_use = 0;
547 * i40evf_free_rx_resources - Free Rx resources
548 * @rx_ring: ring to clean the resources from
550 * Free all receive software resources
552 void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
554 i40evf_clean_rx_ring(rx_ring);
555 kfree(rx_ring->rx_bi);
556 rx_ring->rx_bi = NULL;
559 dma_free_coherent(rx_ring->dev, rx_ring->size,
560 rx_ring->desc, rx_ring->dma);
561 rx_ring->desc = NULL;
566 * i40evf_alloc_rx_headers - allocate rx header buffers
567 * @rx_ring: ring to alloc buffers
569 * Allocate rx header buffers for the entire ring. As these are static,
570 * this is only called when setting up a new ring.
572 void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
574 struct device *dev = rx_ring->dev;
575 struct i40e_rx_buffer *rx_bi;
581 if (rx_ring->rx_bi[0].hdr_buf)
583 /* Make sure the buffers don't cross cache line boundaries. */
584 buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
585 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
589 for (i = 0; i < rx_ring->count; i++) {
590 rx_bi = &rx_ring->rx_bi[i];
591 rx_bi->dma = dma + (i * buf_size);
592 rx_bi->hdr_buf = buffer + (i * buf_size);
597 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
598 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
600 * Returns 0 on success, negative on failure
602 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
604 struct device *dev = rx_ring->dev;
607 /* warn if we are about to overwrite the pointer */
608 WARN_ON(rx_ring->rx_bi);
609 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
610 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
614 u64_stats_init(&rx_ring->syncp);
616 /* Round up to nearest 4K */
617 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
618 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
619 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
620 rx_ring->size = ALIGN(rx_ring->size, 4096);
621 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
622 &rx_ring->dma, GFP_KERNEL);
624 if (!rx_ring->desc) {
625 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
630 rx_ring->next_to_clean = 0;
631 rx_ring->next_to_use = 0;
635 kfree(rx_ring->rx_bi);
636 rx_ring->rx_bi = NULL;
641 * i40e_release_rx_desc - Store the new tail and head values
642 * @rx_ring: ring to bump
643 * @val: new head index
645 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
647 rx_ring->next_to_use = val;
648 /* Force memory writes to complete before letting h/w
649 * know there are new descriptors to fetch. (Only
650 * applicable for weak-ordered memory model archs,
654 writel(val, rx_ring->tail);
658 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
659 * @rx_ring: ring to place buffers on
660 * @cleaned_count: number of buffers to replace
662 void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
664 u16 i = rx_ring->next_to_use;
665 union i40e_rx_desc *rx_desc;
666 struct i40e_rx_buffer *bi;
668 /* do nothing if no valid netdev defined */
669 if (!rx_ring->netdev || !cleaned_count)
672 while (cleaned_count--) {
673 rx_desc = I40E_RX_DESC(rx_ring, i);
674 bi = &rx_ring->rx_bi[i];
676 if (bi->skb) /* desc is in use */
679 bi->page = alloc_page(GFP_ATOMIC);
681 rx_ring->rx_stats.alloc_page_failed++;
687 /* use a half page if we're re-using */
688 bi->page_offset ^= PAGE_SIZE / 2;
689 bi->page_dma = dma_map_page(rx_ring->dev,
694 if (dma_mapping_error(rx_ring->dev,
696 rx_ring->rx_stats.alloc_page_failed++;
702 dma_sync_single_range_for_device(rx_ring->dev,
703 rx_ring->rx_bi[0].dma,
704 i * rx_ring->rx_hdr_len,
707 /* Refresh the desc even if buffer_addrs didn't change
708 * because each write-back erases this info.
710 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
711 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
713 if (i == rx_ring->count)
718 if (rx_ring->next_to_use != i)
719 i40e_release_rx_desc(rx_ring, i);
723 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
724 * @rx_ring: ring to place buffers on
725 * @cleaned_count: number of buffers to replace
727 void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
729 u16 i = rx_ring->next_to_use;
730 union i40e_rx_desc *rx_desc;
731 struct i40e_rx_buffer *bi;
734 /* do nothing if no valid netdev defined */
735 if (!rx_ring->netdev || !cleaned_count)
738 while (cleaned_count--) {
739 rx_desc = I40E_RX_DESC(rx_ring, i);
740 bi = &rx_ring->rx_bi[i];
744 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
745 rx_ring->rx_buf_len);
747 rx_ring->rx_stats.alloc_buff_failed++;
750 /* initialize queue mapping */
751 skb_record_rx_queue(skb, rx_ring->queue_index);
756 bi->dma = dma_map_single(rx_ring->dev,
760 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
761 rx_ring->rx_stats.alloc_buff_failed++;
767 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
768 rx_desc->read.hdr_addr = 0;
770 if (i == rx_ring->count)
775 if (rx_ring->next_to_use != i)
776 i40e_release_rx_desc(rx_ring, i);
780 * i40e_receive_skb - Send a completed packet up the stack
781 * @rx_ring: rx ring in play
782 * @skb: packet to send up
783 * @vlan_tag: vlan tag for packet
785 static void i40e_receive_skb(struct i40e_ring *rx_ring,
786 struct sk_buff *skb, u16 vlan_tag)
788 struct i40e_q_vector *q_vector = rx_ring->q_vector;
790 if (vlan_tag & VLAN_VID_MASK)
791 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
793 napi_gro_receive(&q_vector->napi, skb);
797 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
798 * @vsi: the VSI we care about
799 * @skb: skb currently being received and modified
800 * @rx_status: status value of last descriptor in packet
801 * @rx_error: error value of last descriptor in packet
802 * @rx_ptype: ptype value of last descriptor in packet
804 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
810 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
811 bool ipv4 = false, ipv6 = false;
812 bool ipv4_tunnel, ipv6_tunnel;
817 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
818 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
819 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
820 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
822 skb->ip_summed = CHECKSUM_NONE;
824 /* Rx csum enabled and ip headers found? */
825 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
828 /* did the hardware decode the packet and checksum? */
829 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
832 /* both known and outer_ip must be set for the below code to work */
833 if (!(decoded.known && decoded.outer_ip))
836 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
837 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
839 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
840 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
844 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
845 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
848 /* likely incorrect csum if alternate IP extension headers found */
850 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
851 /* don't increment checksum err here, non-fatal err */
854 /* there was some L4 error, count error and punt packet to the stack */
855 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
858 /* handle packets that were not able to be checksummed due
859 * to arrival speed, in this case the stack can compute
862 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
865 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
866 * it in the driver, hardware does not do it for us.
867 * Since L3L4P bit was set we assume a valid IHL value (>=5)
868 * so the total length of IPv4 header is IHL*4 bytes
869 * The UDP_0 bit *may* bet set if the *inner* header is UDP
872 skb->transport_header = skb->mac_header +
873 sizeof(struct ethhdr) +
874 (ip_hdr(skb)->ihl * 4);
876 /* Add 4 bytes for VLAN tagged packets */
877 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
878 skb->protocol == htons(ETH_P_8021AD))
881 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
882 (udp_hdr(skb)->check != 0)) {
883 rx_udp_csum = udp_csum(skb);
885 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
887 skb_transport_offset(skb)),
888 IPPROTO_UDP, rx_udp_csum);
890 if (udp_hdr(skb)->check != csum)
893 } /* else its GRE and so no outer UDP header */
896 skb->ip_summed = CHECKSUM_UNNECESSARY;
897 skb->csum_level = ipv4_tunnel || ipv6_tunnel;
902 vsi->back->hw_csum_rx_error++;
906 * i40e_ptype_to_htype - get a hash type
907 * @ptype: the ptype value from the descriptor
909 * Returns a hash type to be used by skb_set_hash
911 static inline enum pkt_hash_types i40e_ptype_to_htype(u8 ptype)
913 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
916 return PKT_HASH_TYPE_NONE;
918 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
919 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
920 return PKT_HASH_TYPE_L4;
921 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
922 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
923 return PKT_HASH_TYPE_L3;
925 return PKT_HASH_TYPE_L2;
929 * i40e_rx_hash - set the hash value in the skb
930 * @ring: descriptor ring
931 * @rx_desc: specific descriptor
933 static inline void i40e_rx_hash(struct i40e_ring *ring,
934 union i40e_rx_desc *rx_desc,
939 const __le64 rss_mask =
940 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
941 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
943 if (ring->netdev->features & NETIF_F_RXHASH)
946 if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
947 hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
948 skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
953 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
954 * @rx_ring: rx ring to clean
955 * @budget: how many cleans we're allowed
957 * Returns true if there's any budget left (e.g. the clean is finished)
959 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
961 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
962 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
963 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
964 const int current_node = numa_mem_id();
965 struct i40e_vsi *vsi = rx_ring->vsi;
966 u16 i = rx_ring->next_to_clean;
967 union i40e_rx_desc *rx_desc;
968 u32 rx_error, rx_status;
973 struct i40e_rx_buffer *rx_bi;
976 /* return some buffers to hardware, one at a time is too slow */
977 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
978 i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
982 i = rx_ring->next_to_clean;
983 rx_desc = I40E_RX_DESC(rx_ring, i);
984 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
985 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
986 I40E_RXD_QW1_STATUS_SHIFT;
988 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
991 /* This memory barrier is needed to keep us from reading
992 * any other fields out of the rx_desc until we know the
996 rx_bi = &rx_ring->rx_bi[i];
999 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1000 rx_ring->rx_hdr_len);
1002 rx_ring->rx_stats.alloc_buff_failed++;
1006 /* initialize queue mapping */
1007 skb_record_rx_queue(skb, rx_ring->queue_index);
1008 /* we are reusing so sync this buffer for CPU use */
1009 dma_sync_single_range_for_cpu(rx_ring->dev,
1010 rx_ring->rx_bi[0].dma,
1011 i * rx_ring->rx_hdr_len,
1012 rx_ring->rx_hdr_len,
1015 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1016 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1017 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
1018 I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1019 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1020 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1022 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1023 I40E_RXD_QW1_ERROR_SHIFT;
1024 rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1025 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1027 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1028 I40E_RXD_QW1_PTYPE_SHIFT;
1029 prefetch(rx_bi->page);
1032 if (rx_hbo || rx_sph) {
1036 len = I40E_RX_HDR_SIZE;
1038 len = rx_header_len;
1039 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1040 } else if (skb->len == 0) {
1043 len = (rx_packet_len > skb_headlen(skb) ?
1044 skb_headlen(skb) : rx_packet_len);
1045 memcpy(__skb_put(skb, len),
1046 rx_bi->page + rx_bi->page_offset,
1048 rx_bi->page_offset += len;
1049 rx_packet_len -= len;
1052 /* Get the rest of the data if this was a header split */
1053 if (rx_packet_len) {
1054 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1059 skb->len += rx_packet_len;
1060 skb->data_len += rx_packet_len;
1061 skb->truesize += rx_packet_len;
1063 if ((page_count(rx_bi->page) == 1) &&
1064 (page_to_nid(rx_bi->page) == current_node))
1065 get_page(rx_bi->page);
1069 dma_unmap_page(rx_ring->dev,
1073 rx_bi->page_dma = 0;
1075 I40E_RX_INCREMENT(rx_ring, i);
1078 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1079 struct i40e_rx_buffer *next_buffer;
1081 next_buffer = &rx_ring->rx_bi[i];
1082 next_buffer->skb = skb;
1083 rx_ring->rx_stats.non_eop_descs++;
1087 /* ERR_MASK will only have valid bits if EOP set */
1088 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1089 dev_kfree_skb_any(skb);
1093 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1095 /* probably a little skewed due to removing CRC */
1096 total_rx_bytes += skb->len;
1099 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1101 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1103 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1104 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1107 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1108 dev_kfree_skb_any(skb);
1112 i40e_receive_skb(rx_ring, skb, vlan_tag);
1114 rx_desc->wb.qword1.status_error_len = 0;
1116 } while (likely(total_rx_packets < budget));
1118 u64_stats_update_begin(&rx_ring->syncp);
1119 rx_ring->stats.packets += total_rx_packets;
1120 rx_ring->stats.bytes += total_rx_bytes;
1121 u64_stats_update_end(&rx_ring->syncp);
1122 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1123 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1125 return total_rx_packets;
1129 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1130 * @rx_ring: rx ring to clean
1131 * @budget: how many cleans we're allowed
1133 * Returns number of packets cleaned
1135 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1137 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1138 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1139 struct i40e_vsi *vsi = rx_ring->vsi;
1140 union i40e_rx_desc *rx_desc;
1141 u32 rx_error, rx_status;
1148 struct i40e_rx_buffer *rx_bi;
1149 struct sk_buff *skb;
1151 /* return some buffers to hardware, one at a time is too slow */
1152 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1153 i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1157 i = rx_ring->next_to_clean;
1158 rx_desc = I40E_RX_DESC(rx_ring, i);
1159 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1160 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1161 I40E_RXD_QW1_STATUS_SHIFT;
1163 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1166 /* This memory barrier is needed to keep us from reading
1167 * any other fields out of the rx_desc until we know the
1172 rx_bi = &rx_ring->rx_bi[i];
1174 prefetch(skb->data);
1176 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1177 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1179 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1180 I40E_RXD_QW1_ERROR_SHIFT;
1181 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1183 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1184 I40E_RXD_QW1_PTYPE_SHIFT;
1188 /* Get the header and possibly the whole packet
1189 * If this is an skb from previous receive dma will be 0
1191 skb_put(skb, rx_packet_len);
1192 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1196 I40E_RX_INCREMENT(rx_ring, i);
1199 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1200 rx_ring->rx_stats.non_eop_descs++;
1204 /* ERR_MASK will only have valid bits if EOP set */
1205 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1206 dev_kfree_skb_any(skb);
1210 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1211 /* probably a little skewed due to removing CRC */
1212 total_rx_bytes += skb->len;
1215 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1217 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1219 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1220 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1222 i40e_receive_skb(rx_ring, skb, vlan_tag);
1224 rx_desc->wb.qword1.status_error_len = 0;
1225 } while (likely(total_rx_packets < budget));
1227 u64_stats_update_begin(&rx_ring->syncp);
1228 rx_ring->stats.packets += total_rx_packets;
1229 rx_ring->stats.bytes += total_rx_bytes;
1230 u64_stats_update_end(&rx_ring->syncp);
1231 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1232 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1234 return total_rx_packets;
1237 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1241 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1242 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1243 (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1244 (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1249 /* a small macro to shorten up some long lines */
1250 #define INTREG I40E_VFINT_DYN_CTLN1
1253 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1254 * @vsi: the VSI we care about
1255 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1258 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1259 struct i40e_q_vector *q_vector)
1261 struct i40e_hw *hw = &vsi->back->hw;
1262 bool rx = false, tx = false;
1266 vector = (q_vector->v_idx + vsi->base_vector);
1268 /* avoid dynamic calculation if in countdown mode OR if
1269 * all dynamic is disabled
1271 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1273 if (q_vector->itr_countdown > 0 ||
1274 (!ITR_IS_DYNAMIC(vsi->rx_itr_setting) &&
1275 !ITR_IS_DYNAMIC(vsi->tx_itr_setting))) {
1279 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1280 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1281 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1284 if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1285 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1286 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1290 /* get the higher of the two ITR adjustments and
1291 * use the same value for both ITR registers
1292 * when in adaptive mode (Rx and/or Tx)
1294 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1296 q_vector->tx.itr = q_vector->rx.itr = itr;
1297 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1299 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1303 /* only need to enable the interrupt once, but need
1304 * to possibly update both ITR values
1307 /* set the INTENA_MSK_MASK so that this first write
1308 * won't actually enable the interrupt, instead just
1309 * updating the ITR (it's bit 31 PF and VF)
1312 /* don't check _DOWN because interrupt isn't being enabled */
1313 wr32(hw, INTREG(vector - 1), rxval);
1317 if (!test_bit(__I40E_DOWN, &vsi->state))
1318 wr32(hw, INTREG(vector - 1), txval);
1320 if (q_vector->itr_countdown)
1321 q_vector->itr_countdown--;
1323 q_vector->itr_countdown = ITR_COUNTDOWN_START;
1327 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1328 * @napi: napi struct with our devices info in it
1329 * @budget: amount of work driver is allowed to do this pass, in packets
1331 * This function will clean all queues associated with a q_vector.
1333 * Returns the amount of work done
1335 int i40evf_napi_poll(struct napi_struct *napi, int budget)
1337 struct i40e_q_vector *q_vector =
1338 container_of(napi, struct i40e_q_vector, napi);
1339 struct i40e_vsi *vsi = q_vector->vsi;
1340 struct i40e_ring *ring;
1341 bool clean_complete = true;
1342 bool arm_wb = false;
1343 int budget_per_ring;
1346 if (test_bit(__I40E_DOWN, &vsi->state)) {
1347 napi_complete(napi);
1351 /* Since the actual Tx work is minimal, we can give the Tx a larger
1352 * budget and be more aggressive about cleaning up the Tx descriptors.
1354 i40e_for_each_ring(ring, q_vector->tx) {
1355 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1356 arm_wb = arm_wb || ring->arm_wb;
1357 ring->arm_wb = false;
1360 /* Handle case where we are called by netpoll with a budget of 0 */
1364 /* We attempt to distribute budget to each Rx queue fairly, but don't
1365 * allow the budget to go below 1 because that would exit polling early.
1367 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1369 i40e_for_each_ring(ring, q_vector->rx) {
1372 if (ring_is_ps_enabled(ring))
1373 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1375 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1377 work_done += cleaned;
1378 /* if we didn't clean as many as budgeted, we must be done */
1379 clean_complete &= (budget_per_ring != cleaned);
1382 /* If work not completed, return budget and polling will return */
1383 if (!clean_complete) {
1386 q_vector->tx.ring[0].tx_stats.tx_force_wb++;
1387 i40evf_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_done(napi, work_done);
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;
1441 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1445 protocol = vhdr->h_vlan_encapsulated_proto;
1446 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1447 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1456 * i40e_tso - set up the tso context descriptor
1457 * @tx_ring: ptr to the ring to send
1458 * @skb: ptr to the skb we're sending
1459 * @hdr_len: ptr to the size of the packet header
1460 * @cd_type_cmd_tso_mss: Quad Word 1
1462 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1464 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1465 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;
1577 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
1578 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
1579 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
1580 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
1582 (skb->len - skb_transport_offset(skb)),
1584 *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1587 network_hdr_len = skb_network_header_len(skb);
1588 this_ip_hdr = ip_hdr(skb);
1589 this_ipv6_hdr = ipv6_hdr(skb);
1590 this_tcp_hdrlen = tcp_hdrlen(skb);
1593 /* Enable IP checksum offloads */
1594 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1595 l4_hdr = this_ip_hdr->protocol;
1596 /* the stack computes the IP header already, the only time we
1597 * need the hardware to recompute it is in the case of TSO.
1599 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1600 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1601 this_ip_hdr->check = 0;
1603 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1605 /* Now set the td_offset for IP header length */
1606 *td_offset = (network_hdr_len >> 2) <<
1607 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1608 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1609 l4_hdr = this_ipv6_hdr->nexthdr;
1610 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1611 /* Now set the td_offset for IP header length */
1612 *td_offset = (network_hdr_len >> 2) <<
1613 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1615 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1616 *td_offset |= (skb_network_offset(skb) >> 1) <<
1617 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1619 /* Enable L4 checksum offloads */
1622 /* enable checksum offloads */
1623 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1624 *td_offset |= (this_tcp_hdrlen >> 2) <<
1625 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1628 /* enable SCTP checksum offload */
1629 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1630 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
1631 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1634 /* enable UDP checksum offload */
1635 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1636 *td_offset |= (sizeof(struct udphdr) >> 2) <<
1637 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1645 * i40e_create_tx_ctx Build the Tx context descriptor
1646 * @tx_ring: ring to create the descriptor on
1647 * @cd_type_cmd_tso_mss: Quad Word 1
1648 * @cd_tunneling: Quad Word 0 - bits 0-31
1649 * @cd_l2tag2: Quad Word 0 - bits 32-63
1651 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1652 const u64 cd_type_cmd_tso_mss,
1653 const u32 cd_tunneling, const u32 cd_l2tag2)
1655 struct i40e_tx_context_desc *context_desc;
1656 int i = tx_ring->next_to_use;
1658 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1659 !cd_tunneling && !cd_l2tag2)
1662 /* grab the next descriptor */
1663 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1666 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1668 /* cpu_to_le32 and assign to struct fields */
1669 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1670 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1671 context_desc->rsvd = cpu_to_le16(0);
1672 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1676 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1678 * @tx_flags: collected send information
1680 * Note: Our HW can't scatter-gather more than 8 fragments to build
1681 * a packet on the wire and so we need to figure out the cases where we
1682 * need to linearize the skb.
1684 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1686 struct skb_frag_struct *frag;
1687 bool linearize = false;
1688 unsigned int size = 0;
1692 num_frags = skb_shinfo(skb)->nr_frags;
1693 gso_segs = skb_shinfo(skb)->gso_segs;
1695 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1698 if (num_frags < (I40E_MAX_BUFFER_TXD))
1699 goto linearize_chk_done;
1700 /* try the simple math, if we have too many frags per segment */
1701 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1702 I40E_MAX_BUFFER_TXD) {
1704 goto linearize_chk_done;
1706 frag = &skb_shinfo(skb)->frags[0];
1707 /* we might still have more fragments per segment */
1709 size += skb_frag_size(frag);
1711 if ((size >= skb_shinfo(skb)->gso_size) &&
1712 (j < I40E_MAX_BUFFER_TXD)) {
1713 size = (size % skb_shinfo(skb)->gso_size);
1716 if (j == I40E_MAX_BUFFER_TXD) {
1721 } while (num_frags);
1723 if (num_frags >= I40E_MAX_BUFFER_TXD)
1732 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1733 * @tx_ring: the ring to be checked
1734 * @size: the size buffer we want to assure is available
1736 * Returns -EBUSY if a stop is needed, else 0
1738 static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1740 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1741 /* Memory barrier before checking head and tail */
1744 /* Check again in a case another CPU has just made room available. */
1745 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1748 /* A reprieve! - use start_queue because it doesn't call schedule */
1749 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1750 ++tx_ring->tx_stats.restart_queue;
1755 * i40evf_maybe_stop_tx - 1st level check for tx stop conditions
1756 * @tx_ring: the ring to be checked
1757 * @size: the size buffer we want to assure is available
1759 * Returns 0 if stop is not needed
1761 static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1763 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1765 return __i40evf_maybe_stop_tx(tx_ring, size);
1769 * i40evf_tx_map - Build the Tx descriptor
1770 * @tx_ring: ring to send buffer on
1772 * @first: first buffer info buffer to use
1773 * @tx_flags: collected send information
1774 * @hdr_len: size of the packet header
1775 * @td_cmd: the command field in the descriptor
1776 * @td_offset: offset for checksum or crc
1778 static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1779 struct i40e_tx_buffer *first, u32 tx_flags,
1780 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1782 unsigned int data_len = skb->data_len;
1783 unsigned int size = skb_headlen(skb);
1784 struct skb_frag_struct *frag;
1785 struct i40e_tx_buffer *tx_bi;
1786 struct i40e_tx_desc *tx_desc;
1787 u16 i = tx_ring->next_to_use;
1792 bool tail_bump = true;
1795 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1796 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1797 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1798 I40E_TX_FLAGS_VLAN_SHIFT;
1801 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1802 gso_segs = skb_shinfo(skb)->gso_segs;
1806 /* multiply data chunks by size of headers */
1807 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1808 first->gso_segs = gso_segs;
1810 first->tx_flags = tx_flags;
1812 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1814 tx_desc = I40E_TX_DESC(tx_ring, i);
1817 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1818 if (dma_mapping_error(tx_ring->dev, dma))
1821 /* record length, and DMA address */
1822 dma_unmap_len_set(tx_bi, len, size);
1823 dma_unmap_addr_set(tx_bi, dma, dma);
1825 tx_desc->buffer_addr = cpu_to_le64(dma);
1827 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1828 tx_desc->cmd_type_offset_bsz =
1829 build_ctob(td_cmd, td_offset,
1830 I40E_MAX_DATA_PER_TXD, td_tag);
1836 if (i == tx_ring->count) {
1837 tx_desc = I40E_TX_DESC(tx_ring, 0);
1841 dma += I40E_MAX_DATA_PER_TXD;
1842 size -= I40E_MAX_DATA_PER_TXD;
1844 tx_desc->buffer_addr = cpu_to_le64(dma);
1847 if (likely(!data_len))
1850 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1857 if (i == tx_ring->count) {
1858 tx_desc = I40E_TX_DESC(tx_ring, 0);
1862 size = skb_frag_size(frag);
1865 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1868 tx_bi = &tx_ring->tx_bi[i];
1871 /* set next_to_watch value indicating a packet is present */
1872 first->next_to_watch = tx_desc;
1875 if (i == tx_ring->count)
1878 tx_ring->next_to_use = i;
1880 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1881 tx_ring->queue_index),
1883 i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED);
1885 /* Algorithm to optimize tail and RS bit setting:
1886 * if xmit_more is supported
1887 * if xmit_more is true
1888 * do not update tail and do not mark RS bit.
1889 * if xmit_more is false and last xmit_more was false
1890 * if every packet spanned less than 4 desc
1891 * then set RS bit on 4th packet and update tail
1894 * update tail and set RS bit on every packet.
1895 * if xmit_more is false and last_xmit_more was true
1896 * update tail and set RS bit.
1898 * Optimization: wmb to be issued only in case of tail update.
1899 * Also optimize the Descriptor WB path for RS bit with the same
1902 * Note: If there are less than 4 packets
1903 * pending and interrupts were disabled the service task will
1904 * trigger a force WB.
1906 if (skb->xmit_more &&
1907 !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1908 tx_ring->queue_index))) {
1909 tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
1911 } else if (!skb->xmit_more &&
1912 !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1913 tx_ring->queue_index)) &&
1914 (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) &&
1915 (tx_ring->packet_stride < WB_STRIDE) &&
1916 (desc_count < WB_STRIDE)) {
1917 tx_ring->packet_stride++;
1919 tx_ring->packet_stride = 0;
1920 tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
1924 tx_ring->packet_stride = 0;
1926 tx_desc->cmd_type_offset_bsz =
1927 build_ctob(td_cmd, td_offset, size, td_tag) |
1928 cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD :
1929 I40E_TX_DESC_CMD_EOP) <<
1930 I40E_TXD_QW1_CMD_SHIFT);
1932 /* notify HW of packet */
1934 prefetchw(tx_desc + 1);
1937 /* Force memory writes to complete before letting h/w
1938 * know there are new descriptors to fetch. (Only
1939 * applicable for weak-ordered memory model archs,
1943 writel(i, tx_ring->tail);
1949 dev_info(tx_ring->dev, "TX DMA map failed\n");
1951 /* clear dma mappings for failed tx_bi map */
1953 tx_bi = &tx_ring->tx_bi[i];
1954 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1962 tx_ring->next_to_use = i;
1966 * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed
1968 * @tx_ring: ring to send buffer on
1970 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1971 * there is not enough descriptors available in this ring since we need at least
1974 static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb,
1975 struct i40e_ring *tx_ring)
1980 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1981 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1982 * + 4 desc gap to avoid the cache line where head is,
1983 * + 1 desc for context descriptor,
1984 * otherwise try next time
1986 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1987 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1989 count += TXD_USE_COUNT(skb_headlen(skb));
1990 if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1991 tx_ring->tx_stats.tx_busy++;
1998 * i40e_xmit_frame_ring - Sends buffer on Tx ring
2000 * @tx_ring: ring to send buffer on
2002 * Returns NETDEV_TX_OK if sent, else an error code
2004 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2005 struct i40e_ring *tx_ring)
2007 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2008 u32 cd_tunneling = 0, cd_l2tag2 = 0;
2009 struct i40e_tx_buffer *first;
2017 /* prefetch the data, we'll need it later */
2018 prefetch(skb->data);
2020 if (0 == i40evf_xmit_descriptor_count(skb, tx_ring))
2021 return NETDEV_TX_BUSY;
2023 /* prepare the xmit flags */
2024 if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2027 /* obtain protocol of skb */
2028 protocol = vlan_get_protocol(skb);
2030 /* record the location of the first descriptor for this packet */
2031 first = &tx_ring->tx_bi[tx_ring->next_to_use];
2033 /* setup IPv4/IPv6 offloads */
2034 if (protocol == htons(ETH_P_IP))
2035 tx_flags |= I40E_TX_FLAGS_IPV4;
2036 else if (protocol == htons(ETH_P_IPV6))
2037 tx_flags |= I40E_TX_FLAGS_IPV6;
2039 tso = i40e_tso(tx_ring, skb, &hdr_len, &cd_type_cmd_tso_mss);
2044 tx_flags |= I40E_TX_FLAGS_TSO;
2046 if (i40e_chk_linearize(skb, tx_flags)) {
2047 if (skb_linearize(skb))
2049 tx_ring->tx_stats.tx_linearize++;
2051 skb_tx_timestamp(skb);
2053 /* always enable CRC insertion offload */
2054 td_cmd |= I40E_TX_DESC_CMD_ICRC;
2056 /* Always offload the checksum, since it's in the data descriptor */
2057 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2058 tx_flags |= I40E_TX_FLAGS_CSUM;
2060 i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2061 tx_ring, &cd_tunneling);
2064 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2065 cd_tunneling, cd_l2tag2);
2067 i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2070 return NETDEV_TX_OK;
2073 dev_kfree_skb_any(skb);
2074 return NETDEV_TX_OK;
2078 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2080 * @netdev: network interface device structure
2082 * Returns NETDEV_TX_OK if sent, else an error code
2084 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2086 struct i40evf_adapter *adapter = netdev_priv(netdev);
2087 struct i40e_ring *tx_ring = &adapter->tx_rings[skb->queue_mapping];
2089 /* hardware can't handle really short frames, hardware padding works
2092 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2093 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2094 return NETDEV_TX_OK;
2095 skb->len = I40E_MIN_TX_LEN;
2096 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2099 return i40e_xmit_frame_ring(skb, tx_ring);