1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Virtual Function Driver
4 * Copyright(c) 2013 - 2014 Intel Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 ******************************************************************************/
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
31 #include "i40e_prototype.h"
33 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
36 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
37 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
38 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
39 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
40 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
43 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
46 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
47 * @ring: the ring that owns the buffer
48 * @tx_buffer: the buffer to free
50 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
51 struct i40e_tx_buffer *tx_buffer)
54 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
55 kfree(tx_buffer->raw_buf);
57 dev_kfree_skb_any(tx_buffer->skb);
59 if (dma_unmap_len(tx_buffer, len))
60 dma_unmap_single(ring->dev,
61 dma_unmap_addr(tx_buffer, dma),
62 dma_unmap_len(tx_buffer, len),
64 } else if (dma_unmap_len(tx_buffer, len)) {
65 dma_unmap_page(ring->dev,
66 dma_unmap_addr(tx_buffer, dma),
67 dma_unmap_len(tx_buffer, len),
70 tx_buffer->next_to_watch = NULL;
71 tx_buffer->skb = NULL;
72 dma_unmap_len_set(tx_buffer, len, 0);
73 /* tx_buffer must be completely set up in the transmit path */
77 * i40evf_clean_tx_ring - Free any empty Tx buffers
78 * @tx_ring: ring to be cleaned
80 void i40evf_clean_tx_ring(struct i40e_ring *tx_ring)
82 unsigned long bi_size;
85 /* ring already cleared, nothing to do */
89 /* Free all the Tx ring sk_buffs */
90 for (i = 0; i < tx_ring->count; i++)
91 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
93 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
94 memset(tx_ring->tx_bi, 0, bi_size);
96 /* Zero out the descriptor ring */
97 memset(tx_ring->desc, 0, tx_ring->size);
99 tx_ring->next_to_use = 0;
100 tx_ring->next_to_clean = 0;
102 if (!tx_ring->netdev)
105 /* cleanup Tx queue statistics */
106 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
107 tx_ring->queue_index));
111 * i40evf_free_tx_resources - Free Tx resources per queue
112 * @tx_ring: Tx descriptor ring for a specific queue
114 * Free all transmit software resources
116 void i40evf_free_tx_resources(struct i40e_ring *tx_ring)
118 i40evf_clean_tx_ring(tx_ring);
119 kfree(tx_ring->tx_bi);
120 tx_ring->tx_bi = NULL;
123 dma_free_coherent(tx_ring->dev, tx_ring->size,
124 tx_ring->desc, tx_ring->dma);
125 tx_ring->desc = NULL;
130 * i40e_get_head - Retrieve head from head writeback
131 * @tx_ring: tx ring to fetch head of
133 * Returns value of Tx ring head based on value stored
134 * in head write-back location
136 static inline u32 i40e_get_head(struct i40e_ring *tx_ring)
138 void *head = (struct i40e_tx_desc *)tx_ring->desc + tx_ring->count;
140 return le32_to_cpu(*(volatile __le32 *)head);
143 #define WB_STRIDE 0x3
146 * i40e_clean_tx_irq - Reclaim resources after transmit completes
147 * @tx_ring: tx ring to clean
148 * @budget: how many cleans we're allowed
150 * Returns true if there's any budget left (e.g. the clean is finished)
152 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
154 u16 i = tx_ring->next_to_clean;
155 struct i40e_tx_buffer *tx_buf;
156 struct i40e_tx_desc *tx_head;
157 struct i40e_tx_desc *tx_desc;
158 unsigned int total_packets = 0;
159 unsigned int total_bytes = 0;
161 tx_buf = &tx_ring->tx_bi[i];
162 tx_desc = I40E_TX_DESC(tx_ring, i);
165 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
168 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
170 /* if next_to_watch is not set then there is no work pending */
174 /* prevent any other reads prior to eop_desc */
175 read_barrier_depends();
177 /* we have caught up to head, no work left to do */
178 if (tx_head == tx_desc)
181 /* clear next_to_watch to prevent false hangs */
182 tx_buf->next_to_watch = NULL;
184 /* update the statistics for this packet */
185 total_bytes += tx_buf->bytecount;
186 total_packets += tx_buf->gso_segs;
189 dev_kfree_skb_any(tx_buf->skb);
191 /* unmap skb header data */
192 dma_unmap_single(tx_ring->dev,
193 dma_unmap_addr(tx_buf, dma),
194 dma_unmap_len(tx_buf, len),
197 /* clear tx_buffer data */
199 dma_unmap_len_set(tx_buf, len, 0);
201 /* unmap remaining buffers */
202 while (tx_desc != eop_desc) {
209 tx_buf = tx_ring->tx_bi;
210 tx_desc = I40E_TX_DESC(tx_ring, 0);
213 /* unmap any remaining paged data */
214 if (dma_unmap_len(tx_buf, len)) {
215 dma_unmap_page(tx_ring->dev,
216 dma_unmap_addr(tx_buf, dma),
217 dma_unmap_len(tx_buf, len),
219 dma_unmap_len_set(tx_buf, len, 0);
223 /* move us one more past the eop_desc for start of next pkt */
229 tx_buf = tx_ring->tx_bi;
230 tx_desc = I40E_TX_DESC(tx_ring, 0);
235 /* update budget accounting */
237 } while (likely(budget));
240 tx_ring->next_to_clean = i;
241 u64_stats_update_begin(&tx_ring->syncp);
242 tx_ring->stats.bytes += total_bytes;
243 tx_ring->stats.packets += total_packets;
244 u64_stats_update_end(&tx_ring->syncp);
245 tx_ring->q_vector->tx.total_bytes += total_bytes;
246 tx_ring->q_vector->tx.total_packets += total_packets;
248 /* check to see if there are any non-cache aligned descriptors
249 * waiting to be written back, and kick the hardware to force
250 * them to be written back in case of napi polling
253 !((i & WB_STRIDE) == WB_STRIDE) &&
254 !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
255 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
256 tx_ring->arm_wb = true;
258 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
259 tx_ring->queue_index),
260 total_packets, total_bytes);
262 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
263 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
264 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
265 /* Make sure that anybody stopping the queue after this
266 * sees the new next_to_clean.
269 if (__netif_subqueue_stopped(tx_ring->netdev,
270 tx_ring->queue_index) &&
271 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
272 netif_wake_subqueue(tx_ring->netdev,
273 tx_ring->queue_index);
274 ++tx_ring->tx_stats.restart_queue;
282 * i40evf_force_wb -Arm hardware to do a wb on noncache aligned descriptors
283 * @vsi: the VSI we care about
284 * @q_vector: the vector on which to force writeback
287 static void i40evf_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
289 u16 flags = q_vector->tx.ring[0].flags;
291 if (flags & I40E_TXR_FLAGS_WB_ON_ITR) {
294 if (q_vector->arm_wb_state)
297 val = I40E_VFINT_DYN_CTLN1_WB_ON_ITR_MASK;
300 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
301 vsi->base_vector - 1),
303 q_vector->arm_wb_state = true;
305 u32 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
306 I40E_VFINT_DYN_CTLN1_ITR_INDX_MASK | /* set noitr */
307 I40E_VFINT_DYN_CTLN1_SWINT_TRIG_MASK |
308 I40E_VFINT_DYN_CTLN1_SW_ITR_INDX_ENA_MASK;
309 /* allow 00 to be written to the index */
312 I40E_VFINT_DYN_CTLN1(q_vector->v_idx +
313 vsi->base_vector - 1), val);
318 * i40e_set_new_dynamic_itr - Find new ITR level
319 * @rc: structure containing ring performance data
321 * Returns true if ITR changed, false if not
323 * Stores a new ITR value based on packets and byte counts during
324 * the last interrupt. The advantage of per interrupt computation
325 * is faster updates and more accurate ITR for the current traffic
326 * pattern. Constants in this function were computed based on
327 * theoretical maximum wire speed and thresholds were set based on
328 * testing data as well as attempting to minimize response time
329 * while increasing bulk throughput.
331 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
333 enum i40e_latency_range new_latency_range = rc->latency_range;
334 u32 new_itr = rc->itr;
338 if (rc->total_packets == 0 || !rc->itr)
341 /* simple throttlerate management
342 * 0-10MB/s lowest (100000 ints/s)
343 * 10-20MB/s low (20000 ints/s)
344 * 20-1249MB/s bulk (8000 ints/s)
346 * The math works out because the divisor is in 10^(-6) which
347 * turns the bytes/us input value into MB/s values, but
348 * make sure to use usecs, as the register values written
349 * are in 2 usec increments in the ITR registers.
351 usecs = (rc->itr << 1);
352 bytes_per_int = rc->total_bytes / usecs;
353 switch (new_latency_range) {
354 case I40E_LOWEST_LATENCY:
355 if (bytes_per_int > 10)
356 new_latency_range = I40E_LOW_LATENCY;
358 case I40E_LOW_LATENCY:
359 if (bytes_per_int > 20)
360 new_latency_range = I40E_BULK_LATENCY;
361 else if (bytes_per_int <= 10)
362 new_latency_range = I40E_LOWEST_LATENCY;
364 case I40E_BULK_LATENCY:
365 if (bytes_per_int <= 20)
366 new_latency_range = I40E_LOW_LATENCY;
369 if (bytes_per_int <= 20)
370 new_latency_range = I40E_LOW_LATENCY;
373 rc->latency_range = new_latency_range;
375 switch (new_latency_range) {
376 case I40E_LOWEST_LATENCY:
377 new_itr = I40E_ITR_100K;
379 case I40E_LOW_LATENCY:
380 new_itr = I40E_ITR_20K;
382 case I40E_BULK_LATENCY:
383 new_itr = I40E_ITR_8K;
390 rc->total_packets = 0;
392 if (new_itr != rc->itr) {
401 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
402 * @tx_ring: the tx ring to set up
404 * Return 0 on success, negative on error
406 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
408 struct device *dev = tx_ring->dev;
414 /* warn if we are about to overwrite the pointer */
415 WARN_ON(tx_ring->tx_bi);
416 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
417 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
421 /* round up to nearest 4K */
422 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
423 /* add u32 for head writeback, align after this takes care of
424 * guaranteeing this is at least one cache line in size
426 tx_ring->size += sizeof(u32);
427 tx_ring->size = ALIGN(tx_ring->size, 4096);
428 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
429 &tx_ring->dma, GFP_KERNEL);
430 if (!tx_ring->desc) {
431 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
436 tx_ring->next_to_use = 0;
437 tx_ring->next_to_clean = 0;
441 kfree(tx_ring->tx_bi);
442 tx_ring->tx_bi = NULL;
447 * i40evf_clean_rx_ring - Free Rx buffers
448 * @rx_ring: ring to be cleaned
450 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
452 struct device *dev = rx_ring->dev;
453 struct i40e_rx_buffer *rx_bi;
454 unsigned long bi_size;
457 /* ring already cleared, nothing to do */
461 if (ring_is_ps_enabled(rx_ring)) {
462 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
464 rx_bi = &rx_ring->rx_bi[0];
465 if (rx_bi->hdr_buf) {
466 dma_free_coherent(dev,
470 for (i = 0; i < rx_ring->count; i++) {
471 rx_bi = &rx_ring->rx_bi[i];
473 rx_bi->hdr_buf = NULL;
477 /* Free all the Rx ring sk_buffs */
478 for (i = 0; i < rx_ring->count; i++) {
479 rx_bi = &rx_ring->rx_bi[i];
481 dma_unmap_single(dev,
488 dev_kfree_skb(rx_bi->skb);
492 if (rx_bi->page_dma) {
499 __free_page(rx_bi->page);
501 rx_bi->page_offset = 0;
505 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
506 memset(rx_ring->rx_bi, 0, bi_size);
508 /* Zero out the descriptor ring */
509 memset(rx_ring->desc, 0, rx_ring->size);
511 rx_ring->next_to_clean = 0;
512 rx_ring->next_to_use = 0;
516 * i40evf_free_rx_resources - Free Rx resources
517 * @rx_ring: ring to clean the resources from
519 * Free all receive software resources
521 void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
523 i40evf_clean_rx_ring(rx_ring);
524 kfree(rx_ring->rx_bi);
525 rx_ring->rx_bi = NULL;
528 dma_free_coherent(rx_ring->dev, rx_ring->size,
529 rx_ring->desc, rx_ring->dma);
530 rx_ring->desc = NULL;
535 * i40evf_alloc_rx_headers - allocate rx header buffers
536 * @rx_ring: ring to alloc buffers
538 * Allocate rx header buffers for the entire ring. As these are static,
539 * this is only called when setting up a new ring.
541 void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
543 struct device *dev = rx_ring->dev;
544 struct i40e_rx_buffer *rx_bi;
550 if (rx_ring->rx_bi[0].hdr_buf)
552 /* Make sure the buffers don't cross cache line boundaries. */
553 buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
554 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
558 for (i = 0; i < rx_ring->count; i++) {
559 rx_bi = &rx_ring->rx_bi[i];
560 rx_bi->dma = dma + (i * buf_size);
561 rx_bi->hdr_buf = buffer + (i * buf_size);
566 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
567 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
569 * Returns 0 on success, negative on failure
571 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
573 struct device *dev = rx_ring->dev;
576 /* warn if we are about to overwrite the pointer */
577 WARN_ON(rx_ring->rx_bi);
578 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
579 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
583 u64_stats_init(&rx_ring->syncp);
585 /* Round up to nearest 4K */
586 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
587 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
588 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
589 rx_ring->size = ALIGN(rx_ring->size, 4096);
590 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
591 &rx_ring->dma, GFP_KERNEL);
593 if (!rx_ring->desc) {
594 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
599 rx_ring->next_to_clean = 0;
600 rx_ring->next_to_use = 0;
604 kfree(rx_ring->rx_bi);
605 rx_ring->rx_bi = NULL;
610 * i40e_release_rx_desc - Store the new tail and head values
611 * @rx_ring: ring to bump
612 * @val: new head index
614 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
616 rx_ring->next_to_use = val;
617 /* Force memory writes to complete before letting h/w
618 * know there are new descriptors to fetch. (Only
619 * applicable for weak-ordered memory model archs,
623 writel(val, rx_ring->tail);
627 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
628 * @rx_ring: ring to place buffers on
629 * @cleaned_count: number of buffers to replace
631 void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
633 u16 i = rx_ring->next_to_use;
634 union i40e_rx_desc *rx_desc;
635 struct i40e_rx_buffer *bi;
637 /* do nothing if no valid netdev defined */
638 if (!rx_ring->netdev || !cleaned_count)
641 while (cleaned_count--) {
642 rx_desc = I40E_RX_DESC(rx_ring, i);
643 bi = &rx_ring->rx_bi[i];
645 if (bi->skb) /* desc is in use */
648 bi->page = alloc_page(GFP_ATOMIC);
650 rx_ring->rx_stats.alloc_page_failed++;
656 /* use a half page if we're re-using */
657 bi->page_offset ^= PAGE_SIZE / 2;
658 bi->page_dma = dma_map_page(rx_ring->dev,
663 if (dma_mapping_error(rx_ring->dev,
665 rx_ring->rx_stats.alloc_page_failed++;
671 dma_sync_single_range_for_device(rx_ring->dev,
676 /* Refresh the desc even if buffer_addrs didn't change
677 * because each write-back erases this info.
679 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
680 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
682 if (i == rx_ring->count)
687 if (rx_ring->next_to_use != i)
688 i40e_release_rx_desc(rx_ring, i);
692 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
693 * @rx_ring: ring to place buffers on
694 * @cleaned_count: number of buffers to replace
696 void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
698 u16 i = rx_ring->next_to_use;
699 union i40e_rx_desc *rx_desc;
700 struct i40e_rx_buffer *bi;
703 /* do nothing if no valid netdev defined */
704 if (!rx_ring->netdev || !cleaned_count)
707 while (cleaned_count--) {
708 rx_desc = I40E_RX_DESC(rx_ring, i);
709 bi = &rx_ring->rx_bi[i];
713 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
714 rx_ring->rx_buf_len);
716 rx_ring->rx_stats.alloc_buff_failed++;
719 /* initialize queue mapping */
720 skb_record_rx_queue(skb, rx_ring->queue_index);
725 bi->dma = dma_map_single(rx_ring->dev,
729 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
730 rx_ring->rx_stats.alloc_buff_failed++;
736 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
737 rx_desc->read.hdr_addr = 0;
739 if (i == rx_ring->count)
744 if (rx_ring->next_to_use != i)
745 i40e_release_rx_desc(rx_ring, i);
749 * i40e_receive_skb - Send a completed packet up the stack
750 * @rx_ring: rx ring in play
751 * @skb: packet to send up
752 * @vlan_tag: vlan tag for packet
754 static void i40e_receive_skb(struct i40e_ring *rx_ring,
755 struct sk_buff *skb, u16 vlan_tag)
757 struct i40e_q_vector *q_vector = rx_ring->q_vector;
759 if (vlan_tag & VLAN_VID_MASK)
760 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
762 napi_gro_receive(&q_vector->napi, skb);
766 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
767 * @vsi: the VSI we care about
768 * @skb: skb currently being received and modified
769 * @rx_status: status value of last descriptor in packet
770 * @rx_error: error value of last descriptor in packet
771 * @rx_ptype: ptype value of last descriptor in packet
773 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
779 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
780 bool ipv4 = false, ipv6 = false;
781 bool ipv4_tunnel, ipv6_tunnel;
786 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
787 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
788 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
789 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
791 skb->ip_summed = CHECKSUM_NONE;
793 /* Rx csum enabled and ip headers found? */
794 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
797 /* did the hardware decode the packet and checksum? */
798 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
801 /* both known and outer_ip must be set for the below code to work */
802 if (!(decoded.known && decoded.outer_ip))
805 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
806 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
808 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
809 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
813 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
814 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
817 /* likely incorrect csum if alternate IP extension headers found */
819 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
820 /* don't increment checksum err here, non-fatal err */
823 /* there was some L4 error, count error and punt packet to the stack */
824 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
827 /* handle packets that were not able to be checksummed due
828 * to arrival speed, in this case the stack can compute
831 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
834 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
835 * it in the driver, hardware does not do it for us.
836 * Since L3L4P bit was set we assume a valid IHL value (>=5)
837 * so the total length of IPv4 header is IHL*4 bytes
838 * The UDP_0 bit *may* bet set if the *inner* header is UDP
841 skb->transport_header = skb->mac_header +
842 sizeof(struct ethhdr) +
843 (ip_hdr(skb)->ihl * 4);
845 /* Add 4 bytes for VLAN tagged packets */
846 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
847 skb->protocol == htons(ETH_P_8021AD))
850 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
851 (udp_hdr(skb)->check != 0)) {
852 rx_udp_csum = udp_csum(skb);
854 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
856 skb_transport_offset(skb)),
857 IPPROTO_UDP, rx_udp_csum);
859 if (udp_hdr(skb)->check != csum)
862 } /* else its GRE and so no outer UDP header */
865 skb->ip_summed = CHECKSUM_UNNECESSARY;
866 skb->csum_level = ipv4_tunnel || ipv6_tunnel;
871 vsi->back->hw_csum_rx_error++;
875 * i40e_rx_hash - returns the hash value from the Rx descriptor
876 * @ring: descriptor ring
877 * @rx_desc: specific descriptor
879 static inline u32 i40e_rx_hash(struct i40e_ring *ring,
880 union i40e_rx_desc *rx_desc)
882 const __le64 rss_mask =
883 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
884 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
886 if ((ring->netdev->features & NETIF_F_RXHASH) &&
887 (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
888 return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
894 * i40e_ptype_to_hash - get a hash type
895 * @ptype: the ptype value from the descriptor
897 * Returns a hash type to be used by skb_set_hash
899 static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
901 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
904 return PKT_HASH_TYPE_NONE;
906 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
907 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
908 return PKT_HASH_TYPE_L4;
909 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
910 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
911 return PKT_HASH_TYPE_L3;
913 return PKT_HASH_TYPE_L2;
917 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
918 * @rx_ring: rx ring to clean
919 * @budget: how many cleans we're allowed
921 * Returns true if there's any budget left (e.g. the clean is finished)
923 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
925 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
926 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
927 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
928 const int current_node = numa_mem_id();
929 struct i40e_vsi *vsi = rx_ring->vsi;
930 u16 i = rx_ring->next_to_clean;
931 union i40e_rx_desc *rx_desc;
932 u32 rx_error, rx_status;
937 struct i40e_rx_buffer *rx_bi;
940 /* return some buffers to hardware, one at a time is too slow */
941 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
942 i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
946 i = rx_ring->next_to_clean;
947 rx_desc = I40E_RX_DESC(rx_ring, i);
948 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
949 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
950 I40E_RXD_QW1_STATUS_SHIFT;
952 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
955 /* This memory barrier is needed to keep us from reading
956 * any other fields out of the rx_desc until we know the
960 rx_bi = &rx_ring->rx_bi[i];
963 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
964 rx_ring->rx_hdr_len);
966 rx_ring->rx_stats.alloc_buff_failed++;
970 /* initialize queue mapping */
971 skb_record_rx_queue(skb, rx_ring->queue_index);
972 /* we are reusing so sync this buffer for CPU use */
973 dma_sync_single_range_for_cpu(rx_ring->dev,
979 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
980 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
981 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
982 I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
983 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
984 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
986 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
987 I40E_RXD_QW1_ERROR_SHIFT;
988 rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
989 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
991 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
992 I40E_RXD_QW1_PTYPE_SHIFT;
993 prefetch(rx_bi->page);
996 if (rx_hbo || rx_sph) {
1000 len = I40E_RX_HDR_SIZE;
1002 len = rx_header_len;
1003 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1004 } else if (skb->len == 0) {
1007 len = (rx_packet_len > skb_headlen(skb) ?
1008 skb_headlen(skb) : rx_packet_len);
1009 memcpy(__skb_put(skb, len),
1010 rx_bi->page + rx_bi->page_offset,
1012 rx_bi->page_offset += len;
1013 rx_packet_len -= len;
1016 /* Get the rest of the data if this was a header split */
1017 if (rx_packet_len) {
1018 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1023 skb->len += rx_packet_len;
1024 skb->data_len += rx_packet_len;
1025 skb->truesize += rx_packet_len;
1027 if ((page_count(rx_bi->page) == 1) &&
1028 (page_to_nid(rx_bi->page) == current_node))
1029 get_page(rx_bi->page);
1033 dma_unmap_page(rx_ring->dev,
1037 rx_bi->page_dma = 0;
1039 I40E_RX_INCREMENT(rx_ring, i);
1042 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1043 struct i40e_rx_buffer *next_buffer;
1045 next_buffer = &rx_ring->rx_bi[i];
1046 next_buffer->skb = skb;
1047 rx_ring->rx_stats.non_eop_descs++;
1051 /* ERR_MASK will only have valid bits if EOP set */
1052 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1053 dev_kfree_skb_any(skb);
1057 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1058 i40e_ptype_to_hash(rx_ptype));
1059 /* probably a little skewed due to removing CRC */
1060 total_rx_bytes += skb->len;
1063 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1065 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1067 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1068 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1071 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1072 dev_kfree_skb_any(skb);
1076 skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
1077 i40e_receive_skb(rx_ring, skb, vlan_tag);
1079 rx_desc->wb.qword1.status_error_len = 0;
1081 } while (likely(total_rx_packets < budget));
1083 u64_stats_update_begin(&rx_ring->syncp);
1084 rx_ring->stats.packets += total_rx_packets;
1085 rx_ring->stats.bytes += total_rx_bytes;
1086 u64_stats_update_end(&rx_ring->syncp);
1087 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1088 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1090 return total_rx_packets;
1094 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1095 * @rx_ring: rx ring to clean
1096 * @budget: how many cleans we're allowed
1098 * Returns number of packets cleaned
1100 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1102 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1103 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1104 struct i40e_vsi *vsi = rx_ring->vsi;
1105 union i40e_rx_desc *rx_desc;
1106 u32 rx_error, rx_status;
1113 struct i40e_rx_buffer *rx_bi;
1114 struct sk_buff *skb;
1116 /* return some buffers to hardware, one at a time is too slow */
1117 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1118 i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1122 i = rx_ring->next_to_clean;
1123 rx_desc = I40E_RX_DESC(rx_ring, i);
1124 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1125 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1126 I40E_RXD_QW1_STATUS_SHIFT;
1128 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1131 /* This memory barrier is needed to keep us from reading
1132 * any other fields out of the rx_desc until we know the
1137 rx_bi = &rx_ring->rx_bi[i];
1139 prefetch(skb->data);
1141 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1142 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1144 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1145 I40E_RXD_QW1_ERROR_SHIFT;
1146 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1148 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1149 I40E_RXD_QW1_PTYPE_SHIFT;
1153 /* Get the header and possibly the whole packet
1154 * If this is an skb from previous receive dma will be 0
1156 skb_put(skb, rx_packet_len);
1157 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1161 I40E_RX_INCREMENT(rx_ring, i);
1164 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1165 rx_ring->rx_stats.non_eop_descs++;
1169 /* ERR_MASK will only have valid bits if EOP set */
1170 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1171 dev_kfree_skb_any(skb);
1175 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1176 i40e_ptype_to_hash(rx_ptype));
1177 /* probably a little skewed due to removing CRC */
1178 total_rx_bytes += skb->len;
1181 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1183 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1185 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1186 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1188 i40e_receive_skb(rx_ring, skb, vlan_tag);
1190 rx_desc->wb.qword1.status_error_len = 0;
1191 } while (likely(total_rx_packets < budget));
1193 u64_stats_update_begin(&rx_ring->syncp);
1194 rx_ring->stats.packets += total_rx_packets;
1195 rx_ring->stats.bytes += total_rx_bytes;
1196 u64_stats_update_end(&rx_ring->syncp);
1197 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1198 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1200 return total_rx_packets;
1203 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1207 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1208 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1209 (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1210 (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1215 /* a small macro to shorten up some long lines */
1216 #define INTREG I40E_VFINT_DYN_CTLN1
1219 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1220 * @vsi: the VSI we care about
1221 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1224 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1225 struct i40e_q_vector *q_vector)
1227 struct i40e_hw *hw = &vsi->back->hw;
1228 bool rx = false, tx = false;
1232 vector = (q_vector->v_idx + vsi->base_vector);
1233 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1235 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1236 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1237 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1239 if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1240 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1241 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1244 /* get the higher of the two ITR adjustments and
1245 * use the same value for both ITR registers
1246 * when in adaptive mode (Rx and/or Tx)
1248 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1250 q_vector->tx.itr = q_vector->rx.itr = itr;
1251 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1253 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1257 /* only need to enable the interrupt once, but need
1258 * to possibly update both ITR values
1261 /* set the INTENA_MSK_MASK so that this first write
1262 * won't actually enable the interrupt, instead just
1263 * updating the ITR (it's bit 31 PF and VF)
1266 /* don't check _DOWN because interrupt isn't being enabled */
1267 wr32(hw, INTREG(vector - 1), rxval);
1270 if (!test_bit(__I40E_DOWN, &vsi->state))
1271 wr32(hw, INTREG(vector - 1), txval);
1275 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1276 * @napi: napi struct with our devices info in it
1277 * @budget: amount of work driver is allowed to do this pass, in packets
1279 * This function will clean all queues associated with a q_vector.
1281 * Returns the amount of work done
1283 int i40evf_napi_poll(struct napi_struct *napi, int budget)
1285 struct i40e_q_vector *q_vector =
1286 container_of(napi, struct i40e_q_vector, napi);
1287 struct i40e_vsi *vsi = q_vector->vsi;
1288 struct i40e_ring *ring;
1289 bool clean_complete = true;
1290 bool arm_wb = false;
1291 int budget_per_ring;
1294 if (test_bit(__I40E_DOWN, &vsi->state)) {
1295 napi_complete(napi);
1299 /* Since the actual Tx work is minimal, we can give the Tx a larger
1300 * budget and be more aggressive about cleaning up the Tx descriptors.
1302 i40e_for_each_ring(ring, q_vector->tx) {
1303 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1304 arm_wb |= ring->arm_wb;
1305 ring->arm_wb = false;
1308 /* Handle case where we are called by netpoll with a budget of 0 */
1312 /* We attempt to distribute budget to each Rx queue fairly, but don't
1313 * allow the budget to go below 1 because that would exit polling early.
1315 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1317 i40e_for_each_ring(ring, q_vector->rx) {
1320 if (ring_is_ps_enabled(ring))
1321 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1323 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1325 work_done += cleaned;
1326 /* if we didn't clean as many as budgeted, we must be done */
1327 clean_complete &= (budget_per_ring != cleaned);
1330 /* If work not completed, return budget and polling will return */
1331 if (!clean_complete) {
1334 i40evf_force_wb(vsi, q_vector);
1338 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1339 q_vector->arm_wb_state = false;
1341 /* Work is done so exit the polling mode and re-enable the interrupt */
1342 napi_complete_done(napi, work_done);
1343 i40e_update_enable_itr(vsi, q_vector);
1348 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1350 * @tx_ring: ring to send buffer on
1351 * @flags: the tx flags to be set
1353 * Checks the skb and set up correspondingly several generic transmit flags
1354 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1356 * Returns error code indicate the frame should be dropped upon error and the
1357 * otherwise returns 0 to indicate the flags has been set properly.
1359 static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1360 struct i40e_ring *tx_ring,
1363 __be16 protocol = skb->protocol;
1366 if (protocol == htons(ETH_P_8021Q) &&
1367 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1368 /* When HW VLAN acceleration is turned off by the user the
1369 * stack sets the protocol to 8021q so that the driver
1370 * can take any steps required to support the SW only
1371 * VLAN handling. In our case the driver doesn't need
1372 * to take any further steps so just set the protocol
1373 * to the encapsulated ethertype.
1375 skb->protocol = vlan_get_protocol(skb);
1379 /* if we have a HW VLAN tag being added, default to the HW one */
1380 if (skb_vlan_tag_present(skb)) {
1381 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1382 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1383 /* else if it is a SW VLAN, check the next protocol and store the tag */
1384 } else if (protocol == htons(ETH_P_8021Q)) {
1385 struct vlan_hdr *vhdr, _vhdr;
1387 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1391 protocol = vhdr->h_vlan_encapsulated_proto;
1392 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1393 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1402 * i40e_tso - set up the tso context descriptor
1403 * @tx_ring: ptr to the ring to send
1404 * @skb: ptr to the skb we're sending
1405 * @hdr_len: ptr to the size of the packet header
1406 * @cd_tunneling: ptr to context descriptor bits
1408 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1410 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1411 u8 *hdr_len, u64 *cd_type_cmd_tso_mss,
1414 u32 cd_cmd, cd_tso_len, cd_mss;
1415 struct ipv6hdr *ipv6h;
1416 struct tcphdr *tcph;
1421 if (!skb_is_gso(skb))
1424 err = skb_cow_head(skb, 0);
1428 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1429 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1431 if (iph->version == 4) {
1432 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1435 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1437 } else if (ipv6h->version == 6) {
1438 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1439 ipv6h->payload_len = 0;
1440 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1444 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1445 *hdr_len = (skb->encapsulation
1446 ? (skb_inner_transport_header(skb) - skb->data)
1447 : skb_transport_offset(skb)) + l4len;
1449 /* find the field values */
1450 cd_cmd = I40E_TX_CTX_DESC_TSO;
1451 cd_tso_len = skb->len - *hdr_len;
1452 cd_mss = skb_shinfo(skb)->gso_size;
1453 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1455 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1456 ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1461 * i40e_tx_enable_csum - Enable Tx checksum offloads
1463 * @tx_flags: pointer to Tx flags currently set
1464 * @td_cmd: Tx descriptor command bits to set
1465 * @td_offset: Tx descriptor header offsets to set
1466 * @cd_tunneling: ptr to context desc bits
1468 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1469 u32 *td_cmd, u32 *td_offset,
1470 struct i40e_ring *tx_ring,
1473 struct ipv6hdr *this_ipv6_hdr;
1474 unsigned int this_tcp_hdrlen;
1475 struct iphdr *this_ip_hdr;
1476 u32 network_hdr_len;
1478 struct udphdr *oudph;
1482 if (skb->encapsulation) {
1483 switch (ip_hdr(skb)->protocol) {
1485 oudph = udp_hdr(skb);
1487 l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
1488 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1493 network_hdr_len = skb_inner_network_header_len(skb);
1494 this_ip_hdr = inner_ip_hdr(skb);
1495 this_ipv6_hdr = inner_ipv6_hdr(skb);
1496 this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1498 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1499 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1500 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1501 ip_hdr(skb)->check = 0;
1504 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1506 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1507 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1508 if (*tx_flags & I40E_TX_FLAGS_TSO)
1509 ip_hdr(skb)->check = 0;
1512 /* Now set the ctx descriptor fields */
1513 *cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1514 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1516 ((skb_inner_network_offset(skb) -
1517 skb_transport_offset(skb)) >> 1) <<
1518 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1519 if (this_ip_hdr->version == 6) {
1520 *tx_flags &= ~I40E_TX_FLAGS_IPV4;
1521 *tx_flags |= I40E_TX_FLAGS_IPV6;
1525 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
1526 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
1527 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
1528 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
1530 (skb->len - skb_transport_offset(skb)),
1532 *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1535 network_hdr_len = skb_network_header_len(skb);
1536 this_ip_hdr = ip_hdr(skb);
1537 this_ipv6_hdr = ipv6_hdr(skb);
1538 this_tcp_hdrlen = tcp_hdrlen(skb);
1541 /* Enable IP checksum offloads */
1542 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1543 l4_hdr = this_ip_hdr->protocol;
1544 /* the stack computes the IP header already, the only time we
1545 * need the hardware to recompute it is in the case of TSO.
1547 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1548 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1549 this_ip_hdr->check = 0;
1551 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1553 /* Now set the td_offset for IP header length */
1554 *td_offset = (network_hdr_len >> 2) <<
1555 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1556 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1557 l4_hdr = this_ipv6_hdr->nexthdr;
1558 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1559 /* Now set the td_offset for IP header length */
1560 *td_offset = (network_hdr_len >> 2) <<
1561 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1563 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1564 *td_offset |= (skb_network_offset(skb) >> 1) <<
1565 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1567 /* Enable L4 checksum offloads */
1570 /* enable checksum offloads */
1571 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1572 *td_offset |= (this_tcp_hdrlen >> 2) <<
1573 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1576 /* enable SCTP checksum offload */
1577 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1578 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
1579 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1582 /* enable UDP checksum offload */
1583 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1584 *td_offset |= (sizeof(struct udphdr) >> 2) <<
1585 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1593 * i40e_create_tx_ctx Build the Tx context descriptor
1594 * @tx_ring: ring to create the descriptor on
1595 * @cd_type_cmd_tso_mss: Quad Word 1
1596 * @cd_tunneling: Quad Word 0 - bits 0-31
1597 * @cd_l2tag2: Quad Word 0 - bits 32-63
1599 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1600 const u64 cd_type_cmd_tso_mss,
1601 const u32 cd_tunneling, const u32 cd_l2tag2)
1603 struct i40e_tx_context_desc *context_desc;
1604 int i = tx_ring->next_to_use;
1606 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1607 !cd_tunneling && !cd_l2tag2)
1610 /* grab the next descriptor */
1611 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1614 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1616 /* cpu_to_le32 and assign to struct fields */
1617 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1618 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1619 context_desc->rsvd = cpu_to_le16(0);
1620 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1624 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1626 * @tx_flags: collected send information
1628 * Note: Our HW can't scatter-gather more than 8 fragments to build
1629 * a packet on the wire and so we need to figure out the cases where we
1630 * need to linearize the skb.
1632 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1634 struct skb_frag_struct *frag;
1635 bool linearize = false;
1636 unsigned int size = 0;
1640 num_frags = skb_shinfo(skb)->nr_frags;
1641 gso_segs = skb_shinfo(skb)->gso_segs;
1643 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1646 if (num_frags < (I40E_MAX_BUFFER_TXD))
1647 goto linearize_chk_done;
1648 /* try the simple math, if we have too many frags per segment */
1649 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1650 I40E_MAX_BUFFER_TXD) {
1652 goto linearize_chk_done;
1654 frag = &skb_shinfo(skb)->frags[0];
1655 /* we might still have more fragments per segment */
1657 size += skb_frag_size(frag);
1659 if ((size >= skb_shinfo(skb)->gso_size) &&
1660 (j < I40E_MAX_BUFFER_TXD)) {
1661 size = (size % skb_shinfo(skb)->gso_size);
1664 if (j == I40E_MAX_BUFFER_TXD) {
1669 } while (num_frags);
1671 if (num_frags >= I40E_MAX_BUFFER_TXD)
1680 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1681 * @tx_ring: the ring to be checked
1682 * @size: the size buffer we want to assure is available
1684 * Returns -EBUSY if a stop is needed, else 0
1686 static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1688 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1689 /* Memory barrier before checking head and tail */
1692 /* Check again in a case another CPU has just made room available. */
1693 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1696 /* A reprieve! - use start_queue because it doesn't call schedule */
1697 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1698 ++tx_ring->tx_stats.restart_queue;
1703 * i40evf_maybe_stop_tx - 1st level check for tx stop conditions
1704 * @tx_ring: the ring to be checked
1705 * @size: the size buffer we want to assure is available
1707 * Returns 0 if stop is not needed
1709 static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1711 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1713 return __i40evf_maybe_stop_tx(tx_ring, size);
1717 * i40evf_tx_map - Build the Tx descriptor
1718 * @tx_ring: ring to send buffer on
1720 * @first: first buffer info buffer to use
1721 * @tx_flags: collected send information
1722 * @hdr_len: size of the packet header
1723 * @td_cmd: the command field in the descriptor
1724 * @td_offset: offset for checksum or crc
1726 static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1727 struct i40e_tx_buffer *first, u32 tx_flags,
1728 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1730 unsigned int data_len = skb->data_len;
1731 unsigned int size = skb_headlen(skb);
1732 struct skb_frag_struct *frag;
1733 struct i40e_tx_buffer *tx_bi;
1734 struct i40e_tx_desc *tx_desc;
1735 u16 i = tx_ring->next_to_use;
1740 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1741 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1742 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1743 I40E_TX_FLAGS_VLAN_SHIFT;
1746 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1747 gso_segs = skb_shinfo(skb)->gso_segs;
1751 /* multiply data chunks by size of headers */
1752 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1753 first->gso_segs = gso_segs;
1755 first->tx_flags = tx_flags;
1757 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1759 tx_desc = I40E_TX_DESC(tx_ring, i);
1762 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1763 if (dma_mapping_error(tx_ring->dev, dma))
1766 /* record length, and DMA address */
1767 dma_unmap_len_set(tx_bi, len, size);
1768 dma_unmap_addr_set(tx_bi, dma, dma);
1770 tx_desc->buffer_addr = cpu_to_le64(dma);
1772 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1773 tx_desc->cmd_type_offset_bsz =
1774 build_ctob(td_cmd, td_offset,
1775 I40E_MAX_DATA_PER_TXD, td_tag);
1779 if (i == tx_ring->count) {
1780 tx_desc = I40E_TX_DESC(tx_ring, 0);
1784 dma += I40E_MAX_DATA_PER_TXD;
1785 size -= I40E_MAX_DATA_PER_TXD;
1787 tx_desc->buffer_addr = cpu_to_le64(dma);
1790 if (likely(!data_len))
1793 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1798 if (i == tx_ring->count) {
1799 tx_desc = I40E_TX_DESC(tx_ring, 0);
1803 size = skb_frag_size(frag);
1806 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1809 tx_bi = &tx_ring->tx_bi[i];
1812 /* Place RS bit on last descriptor of any packet that spans across the
1813 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1815 #define WB_STRIDE 0x3
1816 if (((i & WB_STRIDE) != WB_STRIDE) &&
1817 (first <= &tx_ring->tx_bi[i]) &&
1818 (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1819 tx_desc->cmd_type_offset_bsz =
1820 build_ctob(td_cmd, td_offset, size, td_tag) |
1821 cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1822 I40E_TXD_QW1_CMD_SHIFT);
1824 tx_desc->cmd_type_offset_bsz =
1825 build_ctob(td_cmd, td_offset, size, td_tag) |
1826 cpu_to_le64((u64)I40E_TXD_CMD <<
1827 I40E_TXD_QW1_CMD_SHIFT);
1830 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1831 tx_ring->queue_index),
1834 /* Force memory writes to complete before letting h/w
1835 * know there are new descriptors to fetch. (Only
1836 * applicable for weak-ordered memory model archs,
1841 /* set next_to_watch value indicating a packet is present */
1842 first->next_to_watch = tx_desc;
1845 if (i == tx_ring->count)
1848 tx_ring->next_to_use = i;
1850 i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED);
1851 /* notify HW of packet */
1852 if (!skb->xmit_more ||
1853 netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1854 tx_ring->queue_index)))
1855 writel(i, tx_ring->tail);
1857 prefetchw(tx_desc + 1);
1862 dev_info(tx_ring->dev, "TX DMA map failed\n");
1864 /* clear dma mappings for failed tx_bi map */
1866 tx_bi = &tx_ring->tx_bi[i];
1867 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1875 tx_ring->next_to_use = i;
1879 * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed
1881 * @tx_ring: ring to send buffer on
1883 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1884 * there is not enough descriptors available in this ring since we need at least
1887 static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb,
1888 struct i40e_ring *tx_ring)
1893 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1894 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1895 * + 4 desc gap to avoid the cache line where head is,
1896 * + 1 desc for context descriptor,
1897 * otherwise try next time
1899 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1900 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1902 count += TXD_USE_COUNT(skb_headlen(skb));
1903 if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1904 tx_ring->tx_stats.tx_busy++;
1911 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1913 * @tx_ring: ring to send buffer on
1915 * Returns NETDEV_TX_OK if sent, else an error code
1917 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1918 struct i40e_ring *tx_ring)
1920 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1921 u32 cd_tunneling = 0, cd_l2tag2 = 0;
1922 struct i40e_tx_buffer *first;
1930 if (0 == i40evf_xmit_descriptor_count(skb, tx_ring))
1931 return NETDEV_TX_BUSY;
1933 /* prepare the xmit flags */
1934 if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1937 /* obtain protocol of skb */
1938 protocol = vlan_get_protocol(skb);
1940 /* record the location of the first descriptor for this packet */
1941 first = &tx_ring->tx_bi[tx_ring->next_to_use];
1943 /* setup IPv4/IPv6 offloads */
1944 if (protocol == htons(ETH_P_IP))
1945 tx_flags |= I40E_TX_FLAGS_IPV4;
1946 else if (protocol == htons(ETH_P_IPV6))
1947 tx_flags |= I40E_TX_FLAGS_IPV6;
1949 tso = i40e_tso(tx_ring, skb, &hdr_len,
1950 &cd_type_cmd_tso_mss, &cd_tunneling);
1955 tx_flags |= I40E_TX_FLAGS_TSO;
1957 if (i40e_chk_linearize(skb, tx_flags)) {
1958 if (skb_linearize(skb))
1960 tx_ring->tx_stats.tx_linearize++;
1962 skb_tx_timestamp(skb);
1964 /* always enable CRC insertion offload */
1965 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1967 /* Always offload the checksum, since it's in the data descriptor */
1968 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1969 tx_flags |= I40E_TX_FLAGS_CSUM;
1971 i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
1972 tx_ring, &cd_tunneling);
1975 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
1976 cd_tunneling, cd_l2tag2);
1978 i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
1981 return NETDEV_TX_OK;
1984 dev_kfree_skb_any(skb);
1985 return NETDEV_TX_OK;
1989 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1991 * @netdev: network interface device structure
1993 * Returns NETDEV_TX_OK if sent, else an error code
1995 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1997 struct i40evf_adapter *adapter = netdev_priv(netdev);
1998 struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
2000 /* hardware can't handle really short frames, hardware padding works
2003 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
2004 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
2005 return NETDEV_TX_OK;
2006 skb->len = I40E_MIN_TX_LEN;
2007 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2010 return i40e_xmit_frame_ring(skb, tx_ring);