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;
337 if (rc->total_packets == 0 || !rc->itr)
340 /* simple throttlerate management
341 * 0-10MB/s lowest (100000 ints/s)
342 * 10-20MB/s low (20000 ints/s)
343 * 20-1249MB/s bulk (8000 ints/s)
345 bytes_per_int = rc->total_bytes / rc->itr;
346 switch (new_latency_range) {
347 case I40E_LOWEST_LATENCY:
348 if (bytes_per_int > 10)
349 new_latency_range = I40E_LOW_LATENCY;
351 case I40E_LOW_LATENCY:
352 if (bytes_per_int > 20)
353 new_latency_range = I40E_BULK_LATENCY;
354 else if (bytes_per_int <= 10)
355 new_latency_range = I40E_LOWEST_LATENCY;
357 case I40E_BULK_LATENCY:
358 if (bytes_per_int <= 20)
359 new_latency_range = I40E_LOW_LATENCY;
362 if (bytes_per_int <= 20)
363 new_latency_range = I40E_LOW_LATENCY;
366 rc->latency_range = new_latency_range;
368 switch (new_latency_range) {
369 case I40E_LOWEST_LATENCY:
370 new_itr = I40E_ITR_100K;
372 case I40E_LOW_LATENCY:
373 new_itr = I40E_ITR_20K;
375 case I40E_BULK_LATENCY:
376 new_itr = I40E_ITR_8K;
383 rc->total_packets = 0;
385 if (new_itr != rc->itr) {
394 * i40evf_setup_tx_descriptors - Allocate the Tx descriptors
395 * @tx_ring: the tx ring to set up
397 * Return 0 on success, negative on error
399 int i40evf_setup_tx_descriptors(struct i40e_ring *tx_ring)
401 struct device *dev = tx_ring->dev;
407 /* warn if we are about to overwrite the pointer */
408 WARN_ON(tx_ring->tx_bi);
409 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
410 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
414 /* round up to nearest 4K */
415 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
416 /* add u32 for head writeback, align after this takes care of
417 * guaranteeing this is at least one cache line in size
419 tx_ring->size += sizeof(u32);
420 tx_ring->size = ALIGN(tx_ring->size, 4096);
421 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
422 &tx_ring->dma, GFP_KERNEL);
423 if (!tx_ring->desc) {
424 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
429 tx_ring->next_to_use = 0;
430 tx_ring->next_to_clean = 0;
434 kfree(tx_ring->tx_bi);
435 tx_ring->tx_bi = NULL;
440 * i40evf_clean_rx_ring - Free Rx buffers
441 * @rx_ring: ring to be cleaned
443 void i40evf_clean_rx_ring(struct i40e_ring *rx_ring)
445 struct device *dev = rx_ring->dev;
446 struct i40e_rx_buffer *rx_bi;
447 unsigned long bi_size;
450 /* ring already cleared, nothing to do */
454 if (ring_is_ps_enabled(rx_ring)) {
455 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
457 rx_bi = &rx_ring->rx_bi[0];
458 if (rx_bi->hdr_buf) {
459 dma_free_coherent(dev,
463 for (i = 0; i < rx_ring->count; i++) {
464 rx_bi = &rx_ring->rx_bi[i];
466 rx_bi->hdr_buf = NULL;
470 /* Free all the Rx ring sk_buffs */
471 for (i = 0; i < rx_ring->count; i++) {
472 rx_bi = &rx_ring->rx_bi[i];
474 dma_unmap_single(dev,
481 dev_kfree_skb(rx_bi->skb);
485 if (rx_bi->page_dma) {
492 __free_page(rx_bi->page);
494 rx_bi->page_offset = 0;
498 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
499 memset(rx_ring->rx_bi, 0, bi_size);
501 /* Zero out the descriptor ring */
502 memset(rx_ring->desc, 0, rx_ring->size);
504 rx_ring->next_to_clean = 0;
505 rx_ring->next_to_use = 0;
509 * i40evf_free_rx_resources - Free Rx resources
510 * @rx_ring: ring to clean the resources from
512 * Free all receive software resources
514 void i40evf_free_rx_resources(struct i40e_ring *rx_ring)
516 i40evf_clean_rx_ring(rx_ring);
517 kfree(rx_ring->rx_bi);
518 rx_ring->rx_bi = NULL;
521 dma_free_coherent(rx_ring->dev, rx_ring->size,
522 rx_ring->desc, rx_ring->dma);
523 rx_ring->desc = NULL;
528 * i40evf_alloc_rx_headers - allocate rx header buffers
529 * @rx_ring: ring to alloc buffers
531 * Allocate rx header buffers for the entire ring. As these are static,
532 * this is only called when setting up a new ring.
534 void i40evf_alloc_rx_headers(struct i40e_ring *rx_ring)
536 struct device *dev = rx_ring->dev;
537 struct i40e_rx_buffer *rx_bi;
543 if (rx_ring->rx_bi[0].hdr_buf)
545 /* Make sure the buffers don't cross cache line boundaries. */
546 buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
547 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
551 for (i = 0; i < rx_ring->count; i++) {
552 rx_bi = &rx_ring->rx_bi[i];
553 rx_bi->dma = dma + (i * buf_size);
554 rx_bi->hdr_buf = buffer + (i * buf_size);
559 * i40evf_setup_rx_descriptors - Allocate Rx descriptors
560 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
562 * Returns 0 on success, negative on failure
564 int i40evf_setup_rx_descriptors(struct i40e_ring *rx_ring)
566 struct device *dev = rx_ring->dev;
569 /* warn if we are about to overwrite the pointer */
570 WARN_ON(rx_ring->rx_bi);
571 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
572 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
576 u64_stats_init(&rx_ring->syncp);
578 /* Round up to nearest 4K */
579 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
580 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
581 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
582 rx_ring->size = ALIGN(rx_ring->size, 4096);
583 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
584 &rx_ring->dma, GFP_KERNEL);
586 if (!rx_ring->desc) {
587 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
592 rx_ring->next_to_clean = 0;
593 rx_ring->next_to_use = 0;
597 kfree(rx_ring->rx_bi);
598 rx_ring->rx_bi = NULL;
603 * i40e_release_rx_desc - Store the new tail and head values
604 * @rx_ring: ring to bump
605 * @val: new head index
607 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
609 rx_ring->next_to_use = val;
610 /* Force memory writes to complete before letting h/w
611 * know there are new descriptors to fetch. (Only
612 * applicable for weak-ordered memory model archs,
616 writel(val, rx_ring->tail);
620 * i40evf_alloc_rx_buffers_ps - Replace used receive buffers; packet split
621 * @rx_ring: ring to place buffers on
622 * @cleaned_count: number of buffers to replace
624 void i40evf_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
626 u16 i = rx_ring->next_to_use;
627 union i40e_rx_desc *rx_desc;
628 struct i40e_rx_buffer *bi;
630 /* do nothing if no valid netdev defined */
631 if (!rx_ring->netdev || !cleaned_count)
634 while (cleaned_count--) {
635 rx_desc = I40E_RX_DESC(rx_ring, i);
636 bi = &rx_ring->rx_bi[i];
638 if (bi->skb) /* desc is in use */
641 bi->page = alloc_page(GFP_ATOMIC);
643 rx_ring->rx_stats.alloc_page_failed++;
649 /* use a half page if we're re-using */
650 bi->page_offset ^= PAGE_SIZE / 2;
651 bi->page_dma = dma_map_page(rx_ring->dev,
656 if (dma_mapping_error(rx_ring->dev,
658 rx_ring->rx_stats.alloc_page_failed++;
664 dma_sync_single_range_for_device(rx_ring->dev,
669 /* Refresh the desc even if buffer_addrs didn't change
670 * because each write-back erases this info.
672 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
673 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
675 if (i == rx_ring->count)
680 if (rx_ring->next_to_use != i)
681 i40e_release_rx_desc(rx_ring, i);
685 * i40evf_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
686 * @rx_ring: ring to place buffers on
687 * @cleaned_count: number of buffers to replace
689 void i40evf_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
691 u16 i = rx_ring->next_to_use;
692 union i40e_rx_desc *rx_desc;
693 struct i40e_rx_buffer *bi;
696 /* do nothing if no valid netdev defined */
697 if (!rx_ring->netdev || !cleaned_count)
700 while (cleaned_count--) {
701 rx_desc = I40E_RX_DESC(rx_ring, i);
702 bi = &rx_ring->rx_bi[i];
706 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
707 rx_ring->rx_buf_len);
709 rx_ring->rx_stats.alloc_buff_failed++;
712 /* initialize queue mapping */
713 skb_record_rx_queue(skb, rx_ring->queue_index);
718 bi->dma = dma_map_single(rx_ring->dev,
722 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
723 rx_ring->rx_stats.alloc_buff_failed++;
729 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
730 rx_desc->read.hdr_addr = 0;
732 if (i == rx_ring->count)
737 if (rx_ring->next_to_use != i)
738 i40e_release_rx_desc(rx_ring, i);
742 * i40e_receive_skb - Send a completed packet up the stack
743 * @rx_ring: rx ring in play
744 * @skb: packet to send up
745 * @vlan_tag: vlan tag for packet
747 static void i40e_receive_skb(struct i40e_ring *rx_ring,
748 struct sk_buff *skb, u16 vlan_tag)
750 struct i40e_q_vector *q_vector = rx_ring->q_vector;
752 if (vlan_tag & VLAN_VID_MASK)
753 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
755 napi_gro_receive(&q_vector->napi, skb);
759 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
760 * @vsi: the VSI we care about
761 * @skb: skb currently being received and modified
762 * @rx_status: status value of last descriptor in packet
763 * @rx_error: error value of last descriptor in packet
764 * @rx_ptype: ptype value of last descriptor in packet
766 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
772 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
773 bool ipv4 = false, ipv6 = false;
774 bool ipv4_tunnel, ipv6_tunnel;
779 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
780 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
781 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
782 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
784 skb->ip_summed = CHECKSUM_NONE;
786 /* Rx csum enabled and ip headers found? */
787 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
790 /* did the hardware decode the packet and checksum? */
791 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
794 /* both known and outer_ip must be set for the below code to work */
795 if (!(decoded.known && decoded.outer_ip))
798 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
799 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
801 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
802 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
806 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
807 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
810 /* likely incorrect csum if alternate IP extension headers found */
812 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
813 /* don't increment checksum err here, non-fatal err */
816 /* there was some L4 error, count error and punt packet to the stack */
817 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
820 /* handle packets that were not able to be checksummed due
821 * to arrival speed, in this case the stack can compute
824 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
827 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
828 * it in the driver, hardware does not do it for us.
829 * Since L3L4P bit was set we assume a valid IHL value (>=5)
830 * so the total length of IPv4 header is IHL*4 bytes
831 * The UDP_0 bit *may* bet set if the *inner* header is UDP
834 skb->transport_header = skb->mac_header +
835 sizeof(struct ethhdr) +
836 (ip_hdr(skb)->ihl * 4);
838 /* Add 4 bytes for VLAN tagged packets */
839 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
840 skb->protocol == htons(ETH_P_8021AD))
843 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
844 (udp_hdr(skb)->check != 0)) {
845 rx_udp_csum = udp_csum(skb);
847 csum = csum_tcpudp_magic(iph->saddr, iph->daddr,
849 skb_transport_offset(skb)),
850 IPPROTO_UDP, rx_udp_csum);
852 if (udp_hdr(skb)->check != csum)
855 } /* else its GRE and so no outer UDP header */
858 skb->ip_summed = CHECKSUM_UNNECESSARY;
859 skb->csum_level = ipv4_tunnel || ipv6_tunnel;
864 vsi->back->hw_csum_rx_error++;
868 * i40e_rx_hash - returns the hash value from the Rx descriptor
869 * @ring: descriptor ring
870 * @rx_desc: specific descriptor
872 static inline u32 i40e_rx_hash(struct i40e_ring *ring,
873 union i40e_rx_desc *rx_desc)
875 const __le64 rss_mask =
876 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
877 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
879 if ((ring->netdev->features & NETIF_F_RXHASH) &&
880 (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
881 return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
887 * i40e_ptype_to_hash - get a hash type
888 * @ptype: the ptype value from the descriptor
890 * Returns a hash type to be used by skb_set_hash
892 static inline enum pkt_hash_types i40e_ptype_to_hash(u8 ptype)
894 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
897 return PKT_HASH_TYPE_NONE;
899 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
900 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
901 return PKT_HASH_TYPE_L4;
902 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
903 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
904 return PKT_HASH_TYPE_L3;
906 return PKT_HASH_TYPE_L2;
910 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
911 * @rx_ring: rx ring to clean
912 * @budget: how many cleans we're allowed
914 * Returns true if there's any budget left (e.g. the clean is finished)
916 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, int budget)
918 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
919 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
920 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
921 const int current_node = numa_mem_id();
922 struct i40e_vsi *vsi = rx_ring->vsi;
923 u16 i = rx_ring->next_to_clean;
924 union i40e_rx_desc *rx_desc;
925 u32 rx_error, rx_status;
930 struct i40e_rx_buffer *rx_bi;
933 /* return some buffers to hardware, one at a time is too slow */
934 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
935 i40evf_alloc_rx_buffers_ps(rx_ring, cleaned_count);
939 i = rx_ring->next_to_clean;
940 rx_desc = I40E_RX_DESC(rx_ring, i);
941 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
942 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
943 I40E_RXD_QW1_STATUS_SHIFT;
945 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
948 /* This memory barrier is needed to keep us from reading
949 * any other fields out of the rx_desc until we know the
953 rx_bi = &rx_ring->rx_bi[i];
956 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
957 rx_ring->rx_hdr_len);
959 rx_ring->rx_stats.alloc_buff_failed++;
963 /* initialize queue mapping */
964 skb_record_rx_queue(skb, rx_ring->queue_index);
965 /* we are reusing so sync this buffer for CPU use */
966 dma_sync_single_range_for_cpu(rx_ring->dev,
972 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
973 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
974 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
975 I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
976 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
977 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
979 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
980 I40E_RXD_QW1_ERROR_SHIFT;
981 rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
982 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
984 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
985 I40E_RXD_QW1_PTYPE_SHIFT;
986 prefetch(rx_bi->page);
989 if (rx_hbo || rx_sph) {
993 len = I40E_RX_HDR_SIZE;
996 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
997 } else if (skb->len == 0) {
1000 len = (rx_packet_len > skb_headlen(skb) ?
1001 skb_headlen(skb) : rx_packet_len);
1002 memcpy(__skb_put(skb, len),
1003 rx_bi->page + rx_bi->page_offset,
1005 rx_bi->page_offset += len;
1006 rx_packet_len -= len;
1009 /* Get the rest of the data if this was a header split */
1010 if (rx_packet_len) {
1011 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1016 skb->len += rx_packet_len;
1017 skb->data_len += rx_packet_len;
1018 skb->truesize += rx_packet_len;
1020 if ((page_count(rx_bi->page) == 1) &&
1021 (page_to_nid(rx_bi->page) == current_node))
1022 get_page(rx_bi->page);
1026 dma_unmap_page(rx_ring->dev,
1030 rx_bi->page_dma = 0;
1032 I40E_RX_INCREMENT(rx_ring, i);
1035 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1036 struct i40e_rx_buffer *next_buffer;
1038 next_buffer = &rx_ring->rx_bi[i];
1039 next_buffer->skb = skb;
1040 rx_ring->rx_stats.non_eop_descs++;
1044 /* ERR_MASK will only have valid bits if EOP set */
1045 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1046 dev_kfree_skb_any(skb);
1050 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1051 i40e_ptype_to_hash(rx_ptype));
1052 /* probably a little skewed due to removing CRC */
1053 total_rx_bytes += skb->len;
1056 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1058 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1060 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1061 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1064 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1065 dev_kfree_skb_any(skb);
1069 skb_mark_napi_id(skb, &rx_ring->q_vector->napi);
1070 i40e_receive_skb(rx_ring, skb, vlan_tag);
1072 rx_desc->wb.qword1.status_error_len = 0;
1074 } while (likely(total_rx_packets < budget));
1076 u64_stats_update_begin(&rx_ring->syncp);
1077 rx_ring->stats.packets += total_rx_packets;
1078 rx_ring->stats.bytes += total_rx_bytes;
1079 u64_stats_update_end(&rx_ring->syncp);
1080 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1081 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1083 return total_rx_packets;
1087 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1088 * @rx_ring: rx ring to clean
1089 * @budget: how many cleans we're allowed
1091 * Returns number of packets cleaned
1093 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1095 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1096 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1097 struct i40e_vsi *vsi = rx_ring->vsi;
1098 union i40e_rx_desc *rx_desc;
1099 u32 rx_error, rx_status;
1106 struct i40e_rx_buffer *rx_bi;
1107 struct sk_buff *skb;
1109 /* return some buffers to hardware, one at a time is too slow */
1110 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1111 i40evf_alloc_rx_buffers_1buf(rx_ring, cleaned_count);
1115 i = rx_ring->next_to_clean;
1116 rx_desc = I40E_RX_DESC(rx_ring, i);
1117 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1118 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1119 I40E_RXD_QW1_STATUS_SHIFT;
1121 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1124 /* This memory barrier is needed to keep us from reading
1125 * any other fields out of the rx_desc until we know the
1130 rx_bi = &rx_ring->rx_bi[i];
1132 prefetch(skb->data);
1134 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1135 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1137 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1138 I40E_RXD_QW1_ERROR_SHIFT;
1139 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1141 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1142 I40E_RXD_QW1_PTYPE_SHIFT;
1146 /* Get the header and possibly the whole packet
1147 * If this is an skb from previous receive dma will be 0
1149 skb_put(skb, rx_packet_len);
1150 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1154 I40E_RX_INCREMENT(rx_ring, i);
1157 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1158 rx_ring->rx_stats.non_eop_descs++;
1162 /* ERR_MASK will only have valid bits if EOP set */
1163 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1164 dev_kfree_skb_any(skb);
1168 skb_set_hash(skb, i40e_rx_hash(rx_ring, rx_desc),
1169 i40e_ptype_to_hash(rx_ptype));
1170 /* probably a little skewed due to removing CRC */
1171 total_rx_bytes += skb->len;
1174 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1176 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1178 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1179 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1181 i40e_receive_skb(rx_ring, skb, vlan_tag);
1183 rx_desc->wb.qword1.status_error_len = 0;
1184 } while (likely(total_rx_packets < budget));
1186 u64_stats_update_begin(&rx_ring->syncp);
1187 rx_ring->stats.packets += total_rx_packets;
1188 rx_ring->stats.bytes += total_rx_bytes;
1189 u64_stats_update_end(&rx_ring->syncp);
1190 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1191 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1193 return total_rx_packets;
1196 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1200 val = I40E_VFINT_DYN_CTLN1_INTENA_MASK |
1201 I40E_VFINT_DYN_CTLN1_CLEARPBA_MASK |
1202 (type << I40E_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
1203 (itr << I40E_VFINT_DYN_CTLN1_INTERVAL_SHIFT);
1208 /* a small macro to shorten up some long lines */
1209 #define INTREG I40E_VFINT_DYN_CTLN1
1212 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1213 * @vsi: the VSI we care about
1214 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1217 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1218 struct i40e_q_vector *q_vector)
1220 struct i40e_hw *hw = &vsi->back->hw;
1221 bool rx = false, tx = false;
1225 vector = (q_vector->v_idx + vsi->base_vector);
1226 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1228 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1229 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1230 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1232 if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1233 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1234 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1237 /* get the higher of the two ITR adjustments and
1238 * use the same value for both ITR registers
1239 * when in adaptive mode (Rx and/or Tx)
1241 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1243 q_vector->tx.itr = q_vector->rx.itr = itr;
1244 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1246 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1250 /* only need to enable the interrupt once, but need
1251 * to possibly update both ITR values
1254 /* set the INTENA_MSK_MASK so that this first write
1255 * won't actually enable the interrupt, instead just
1256 * updating the ITR (it's bit 31 PF and VF)
1259 /* don't check _DOWN because interrupt isn't being enabled */
1260 wr32(hw, INTREG(vector - 1), rxval);
1263 if (!test_bit(__I40E_DOWN, &vsi->state))
1264 wr32(hw, INTREG(vector - 1), txval);
1268 * i40evf_napi_poll - NAPI polling Rx/Tx cleanup routine
1269 * @napi: napi struct with our devices info in it
1270 * @budget: amount of work driver is allowed to do this pass, in packets
1272 * This function will clean all queues associated with a q_vector.
1274 * Returns the amount of work done
1276 int i40evf_napi_poll(struct napi_struct *napi, int budget)
1278 struct i40e_q_vector *q_vector =
1279 container_of(napi, struct i40e_q_vector, napi);
1280 struct i40e_vsi *vsi = q_vector->vsi;
1281 struct i40e_ring *ring;
1282 bool clean_complete = true;
1283 bool arm_wb = false;
1284 int budget_per_ring;
1287 if (test_bit(__I40E_DOWN, &vsi->state)) {
1288 napi_complete(napi);
1292 /* Since the actual Tx work is minimal, we can give the Tx a larger
1293 * budget and be more aggressive about cleaning up the Tx descriptors.
1295 i40e_for_each_ring(ring, q_vector->tx) {
1296 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1297 arm_wb |= ring->arm_wb;
1298 ring->arm_wb = false;
1301 /* Handle case where we are called by netpoll with a budget of 0 */
1305 /* We attempt to distribute budget to each Rx queue fairly, but don't
1306 * allow the budget to go below 1 because that would exit polling early.
1308 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1310 i40e_for_each_ring(ring, q_vector->rx) {
1313 if (ring_is_ps_enabled(ring))
1314 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1316 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1318 work_done += cleaned;
1319 /* if we didn't clean as many as budgeted, we must be done */
1320 clean_complete &= (budget_per_ring != cleaned);
1323 /* If work not completed, return budget and polling will return */
1324 if (!clean_complete) {
1327 i40evf_force_wb(vsi, q_vector);
1331 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
1332 q_vector->arm_wb_state = false;
1334 /* Work is done so exit the polling mode and re-enable the interrupt */
1335 napi_complete_done(napi, work_done);
1336 i40e_update_enable_itr(vsi, q_vector);
1341 * i40evf_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1343 * @tx_ring: ring to send buffer on
1344 * @flags: the tx flags to be set
1346 * Checks the skb and set up correspondingly several generic transmit flags
1347 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1349 * Returns error code indicate the frame should be dropped upon error and the
1350 * otherwise returns 0 to indicate the flags has been set properly.
1352 static inline int i40evf_tx_prepare_vlan_flags(struct sk_buff *skb,
1353 struct i40e_ring *tx_ring,
1356 __be16 protocol = skb->protocol;
1359 if (protocol == htons(ETH_P_8021Q) &&
1360 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
1361 /* When HW VLAN acceleration is turned off by the user the
1362 * stack sets the protocol to 8021q so that the driver
1363 * can take any steps required to support the SW only
1364 * VLAN handling. In our case the driver doesn't need
1365 * to take any further steps so just set the protocol
1366 * to the encapsulated ethertype.
1368 skb->protocol = vlan_get_protocol(skb);
1372 /* if we have a HW VLAN tag being added, default to the HW one */
1373 if (skb_vlan_tag_present(skb)) {
1374 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1375 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1376 /* else if it is a SW VLAN, check the next protocol and store the tag */
1377 } else if (protocol == htons(ETH_P_8021Q)) {
1378 struct vlan_hdr *vhdr, _vhdr;
1380 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1384 protocol = vhdr->h_vlan_encapsulated_proto;
1385 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1386 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1395 * i40e_tso - set up the tso context descriptor
1396 * @tx_ring: ptr to the ring to send
1397 * @skb: ptr to the skb we're sending
1398 * @hdr_len: ptr to the size of the packet header
1399 * @cd_tunneling: ptr to context descriptor bits
1401 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1403 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1404 u8 *hdr_len, u64 *cd_type_cmd_tso_mss,
1407 u32 cd_cmd, cd_tso_len, cd_mss;
1408 struct ipv6hdr *ipv6h;
1409 struct tcphdr *tcph;
1414 if (!skb_is_gso(skb))
1417 err = skb_cow_head(skb, 0);
1421 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1422 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
1424 if (iph->version == 4) {
1425 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1428 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1430 } else if (ipv6h->version == 6) {
1431 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1432 ipv6h->payload_len = 0;
1433 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1437 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1438 *hdr_len = (skb->encapsulation
1439 ? (skb_inner_transport_header(skb) - skb->data)
1440 : skb_transport_offset(skb)) + l4len;
1442 /* find the field values */
1443 cd_cmd = I40E_TX_CTX_DESC_TSO;
1444 cd_tso_len = skb->len - *hdr_len;
1445 cd_mss = skb_shinfo(skb)->gso_size;
1446 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
1448 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
1449 ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1454 * i40e_tx_enable_csum - Enable Tx checksum offloads
1456 * @tx_flags: pointer to Tx flags currently set
1457 * @td_cmd: Tx descriptor command bits to set
1458 * @td_offset: Tx descriptor header offsets to set
1459 * @cd_tunneling: ptr to context desc bits
1461 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
1462 u32 *td_cmd, u32 *td_offset,
1463 struct i40e_ring *tx_ring,
1466 struct ipv6hdr *this_ipv6_hdr;
1467 unsigned int this_tcp_hdrlen;
1468 struct iphdr *this_ip_hdr;
1469 u32 network_hdr_len;
1471 struct udphdr *oudph;
1475 if (skb->encapsulation) {
1476 switch (ip_hdr(skb)->protocol) {
1478 oudph = udp_hdr(skb);
1480 l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
1481 *tx_flags |= I40E_TX_FLAGS_VXLAN_TUNNEL;
1486 network_hdr_len = skb_inner_network_header_len(skb);
1487 this_ip_hdr = inner_ip_hdr(skb);
1488 this_ipv6_hdr = inner_ipv6_hdr(skb);
1489 this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1491 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1492 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1493 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1494 ip_hdr(skb)->check = 0;
1497 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1499 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1500 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1501 if (*tx_flags & I40E_TX_FLAGS_TSO)
1502 ip_hdr(skb)->check = 0;
1505 /* Now set the ctx descriptor fields */
1506 *cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1507 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1509 ((skb_inner_network_offset(skb) -
1510 skb_transport_offset(skb)) >> 1) <<
1511 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1512 if (this_ip_hdr->version == 6) {
1513 *tx_flags &= ~I40E_TX_FLAGS_IPV4;
1514 *tx_flags |= I40E_TX_FLAGS_IPV6;
1518 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
1519 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
1520 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
1521 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
1523 (skb->len - skb_transport_offset(skb)),
1525 *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
1528 network_hdr_len = skb_network_header_len(skb);
1529 this_ip_hdr = ip_hdr(skb);
1530 this_ipv6_hdr = ipv6_hdr(skb);
1531 this_tcp_hdrlen = tcp_hdrlen(skb);
1534 /* Enable IP checksum offloads */
1535 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
1536 l4_hdr = this_ip_hdr->protocol;
1537 /* the stack computes the IP header already, the only time we
1538 * need the hardware to recompute it is in the case of TSO.
1540 if (*tx_flags & I40E_TX_FLAGS_TSO) {
1541 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1542 this_ip_hdr->check = 0;
1544 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1546 /* Now set the td_offset for IP header length */
1547 *td_offset = (network_hdr_len >> 2) <<
1548 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1549 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
1550 l4_hdr = this_ipv6_hdr->nexthdr;
1551 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1552 /* Now set the td_offset for IP header length */
1553 *td_offset = (network_hdr_len >> 2) <<
1554 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1556 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1557 *td_offset |= (skb_network_offset(skb) >> 1) <<
1558 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1560 /* Enable L4 checksum offloads */
1563 /* enable checksum offloads */
1564 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1565 *td_offset |= (this_tcp_hdrlen >> 2) <<
1566 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1569 /* enable SCTP checksum offload */
1570 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1571 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
1572 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1575 /* enable UDP checksum offload */
1576 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1577 *td_offset |= (sizeof(struct udphdr) >> 2) <<
1578 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1586 * i40e_create_tx_ctx Build the Tx context descriptor
1587 * @tx_ring: ring to create the descriptor on
1588 * @cd_type_cmd_tso_mss: Quad Word 1
1589 * @cd_tunneling: Quad Word 0 - bits 0-31
1590 * @cd_l2tag2: Quad Word 0 - bits 32-63
1592 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1593 const u64 cd_type_cmd_tso_mss,
1594 const u32 cd_tunneling, const u32 cd_l2tag2)
1596 struct i40e_tx_context_desc *context_desc;
1597 int i = tx_ring->next_to_use;
1599 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
1600 !cd_tunneling && !cd_l2tag2)
1603 /* grab the next descriptor */
1604 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1607 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1609 /* cpu_to_le32 and assign to struct fields */
1610 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1611 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1612 context_desc->rsvd = cpu_to_le16(0);
1613 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1617 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
1619 * @tx_flags: collected send information
1621 * Note: Our HW can't scatter-gather more than 8 fragments to build
1622 * a packet on the wire and so we need to figure out the cases where we
1623 * need to linearize the skb.
1625 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
1627 struct skb_frag_struct *frag;
1628 bool linearize = false;
1629 unsigned int size = 0;
1633 num_frags = skb_shinfo(skb)->nr_frags;
1634 gso_segs = skb_shinfo(skb)->gso_segs;
1636 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
1639 if (num_frags < (I40E_MAX_BUFFER_TXD))
1640 goto linearize_chk_done;
1641 /* try the simple math, if we have too many frags per segment */
1642 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
1643 I40E_MAX_BUFFER_TXD) {
1645 goto linearize_chk_done;
1647 frag = &skb_shinfo(skb)->frags[0];
1648 /* we might still have more fragments per segment */
1650 size += skb_frag_size(frag);
1652 if ((size >= skb_shinfo(skb)->gso_size) &&
1653 (j < I40E_MAX_BUFFER_TXD)) {
1654 size = (size % skb_shinfo(skb)->gso_size);
1657 if (j == I40E_MAX_BUFFER_TXD) {
1662 } while (num_frags);
1664 if (num_frags >= I40E_MAX_BUFFER_TXD)
1673 * __i40evf_maybe_stop_tx - 2nd level check for tx stop conditions
1674 * @tx_ring: the ring to be checked
1675 * @size: the size buffer we want to assure is available
1677 * Returns -EBUSY if a stop is needed, else 0
1679 static inline int __i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1681 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1682 /* Memory barrier before checking head and tail */
1685 /* Check again in a case another CPU has just made room available. */
1686 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1689 /* A reprieve! - use start_queue because it doesn't call schedule */
1690 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1691 ++tx_ring->tx_stats.restart_queue;
1696 * i40evf_maybe_stop_tx - 1st level check for tx stop conditions
1697 * @tx_ring: the ring to be checked
1698 * @size: the size buffer we want to assure is available
1700 * Returns 0 if stop is not needed
1702 static inline int i40evf_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1704 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1706 return __i40evf_maybe_stop_tx(tx_ring, size);
1710 * i40evf_tx_map - Build the Tx descriptor
1711 * @tx_ring: ring to send buffer on
1713 * @first: first buffer info buffer to use
1714 * @tx_flags: collected send information
1715 * @hdr_len: size of the packet header
1716 * @td_cmd: the command field in the descriptor
1717 * @td_offset: offset for checksum or crc
1719 static inline void i40evf_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1720 struct i40e_tx_buffer *first, u32 tx_flags,
1721 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1723 unsigned int data_len = skb->data_len;
1724 unsigned int size = skb_headlen(skb);
1725 struct skb_frag_struct *frag;
1726 struct i40e_tx_buffer *tx_bi;
1727 struct i40e_tx_desc *tx_desc;
1728 u16 i = tx_ring->next_to_use;
1733 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1734 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1735 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1736 I40E_TX_FLAGS_VLAN_SHIFT;
1739 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1740 gso_segs = skb_shinfo(skb)->gso_segs;
1744 /* multiply data chunks by size of headers */
1745 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1746 first->gso_segs = gso_segs;
1748 first->tx_flags = tx_flags;
1750 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1752 tx_desc = I40E_TX_DESC(tx_ring, i);
1755 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1756 if (dma_mapping_error(tx_ring->dev, dma))
1759 /* record length, and DMA address */
1760 dma_unmap_len_set(tx_bi, len, size);
1761 dma_unmap_addr_set(tx_bi, dma, dma);
1763 tx_desc->buffer_addr = cpu_to_le64(dma);
1765 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1766 tx_desc->cmd_type_offset_bsz =
1767 build_ctob(td_cmd, td_offset,
1768 I40E_MAX_DATA_PER_TXD, td_tag);
1772 if (i == tx_ring->count) {
1773 tx_desc = I40E_TX_DESC(tx_ring, 0);
1777 dma += I40E_MAX_DATA_PER_TXD;
1778 size -= I40E_MAX_DATA_PER_TXD;
1780 tx_desc->buffer_addr = cpu_to_le64(dma);
1783 if (likely(!data_len))
1786 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1791 if (i == tx_ring->count) {
1792 tx_desc = I40E_TX_DESC(tx_ring, 0);
1796 size = skb_frag_size(frag);
1799 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1802 tx_bi = &tx_ring->tx_bi[i];
1805 /* Place RS bit on last descriptor of any packet that spans across the
1806 * 4th descriptor (WB_STRIDE aka 0x3) in a 64B cacheline.
1808 #define WB_STRIDE 0x3
1809 if (((i & WB_STRIDE) != WB_STRIDE) &&
1810 (first <= &tx_ring->tx_bi[i]) &&
1811 (first >= &tx_ring->tx_bi[i & ~WB_STRIDE])) {
1812 tx_desc->cmd_type_offset_bsz =
1813 build_ctob(td_cmd, td_offset, size, td_tag) |
1814 cpu_to_le64((u64)I40E_TX_DESC_CMD_EOP <<
1815 I40E_TXD_QW1_CMD_SHIFT);
1817 tx_desc->cmd_type_offset_bsz =
1818 build_ctob(td_cmd, td_offset, size, td_tag) |
1819 cpu_to_le64((u64)I40E_TXD_CMD <<
1820 I40E_TXD_QW1_CMD_SHIFT);
1823 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1824 tx_ring->queue_index),
1827 /* Force memory writes to complete before letting h/w
1828 * know there are new descriptors to fetch. (Only
1829 * applicable for weak-ordered memory model archs,
1834 /* set next_to_watch value indicating a packet is present */
1835 first->next_to_watch = tx_desc;
1838 if (i == tx_ring->count)
1841 tx_ring->next_to_use = i;
1843 i40evf_maybe_stop_tx(tx_ring, DESC_NEEDED);
1844 /* notify HW of packet */
1845 if (!skb->xmit_more ||
1846 netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
1847 tx_ring->queue_index)))
1848 writel(i, tx_ring->tail);
1850 prefetchw(tx_desc + 1);
1855 dev_info(tx_ring->dev, "TX DMA map failed\n");
1857 /* clear dma mappings for failed tx_bi map */
1859 tx_bi = &tx_ring->tx_bi[i];
1860 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1868 tx_ring->next_to_use = i;
1872 * i40evf_xmit_descriptor_count - calculate number of tx descriptors needed
1874 * @tx_ring: ring to send buffer on
1876 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1877 * there is not enough descriptors available in this ring since we need at least
1880 static inline int i40evf_xmit_descriptor_count(struct sk_buff *skb,
1881 struct i40e_ring *tx_ring)
1886 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1887 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1888 * + 4 desc gap to avoid the cache line where head is,
1889 * + 1 desc for context descriptor,
1890 * otherwise try next time
1892 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1893 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1895 count += TXD_USE_COUNT(skb_headlen(skb));
1896 if (i40evf_maybe_stop_tx(tx_ring, count + 4 + 1)) {
1897 tx_ring->tx_stats.tx_busy++;
1904 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1906 * @tx_ring: ring to send buffer on
1908 * Returns NETDEV_TX_OK if sent, else an error code
1910 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1911 struct i40e_ring *tx_ring)
1913 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1914 u32 cd_tunneling = 0, cd_l2tag2 = 0;
1915 struct i40e_tx_buffer *first;
1923 if (0 == i40evf_xmit_descriptor_count(skb, tx_ring))
1924 return NETDEV_TX_BUSY;
1926 /* prepare the xmit flags */
1927 if (i40evf_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1930 /* obtain protocol of skb */
1931 protocol = vlan_get_protocol(skb);
1933 /* record the location of the first descriptor for this packet */
1934 first = &tx_ring->tx_bi[tx_ring->next_to_use];
1936 /* setup IPv4/IPv6 offloads */
1937 if (protocol == htons(ETH_P_IP))
1938 tx_flags |= I40E_TX_FLAGS_IPV4;
1939 else if (protocol == htons(ETH_P_IPV6))
1940 tx_flags |= I40E_TX_FLAGS_IPV6;
1942 tso = i40e_tso(tx_ring, skb, &hdr_len,
1943 &cd_type_cmd_tso_mss, &cd_tunneling);
1948 tx_flags |= I40E_TX_FLAGS_TSO;
1950 if (i40e_chk_linearize(skb, tx_flags)) {
1951 if (skb_linearize(skb))
1953 tx_ring->tx_stats.tx_linearize++;
1955 skb_tx_timestamp(skb);
1957 /* always enable CRC insertion offload */
1958 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1960 /* Always offload the checksum, since it's in the data descriptor */
1961 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1962 tx_flags |= I40E_TX_FLAGS_CSUM;
1964 i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
1965 tx_ring, &cd_tunneling);
1968 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
1969 cd_tunneling, cd_l2tag2);
1971 i40evf_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
1974 return NETDEV_TX_OK;
1977 dev_kfree_skb_any(skb);
1978 return NETDEV_TX_OK;
1982 * i40evf_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1984 * @netdev: network interface device structure
1986 * Returns NETDEV_TX_OK if sent, else an error code
1988 netdev_tx_t i40evf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1990 struct i40evf_adapter *adapter = netdev_priv(netdev);
1991 struct i40e_ring *tx_ring = adapter->tx_rings[skb->queue_mapping];
1993 /* hardware can't handle really short frames, hardware padding works
1996 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
1997 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
1998 return NETDEV_TX_OK;
1999 skb->len = I40E_MIN_TX_LEN;
2000 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
2003 return i40e_xmit_frame_ring(skb, tx_ring);