1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Driver
4 * Copyright(c) 2013 - 2016 Intel Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 ******************************************************************************/
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
30 #include "i40e_prototype.h"
32 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
35 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
36 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
37 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
38 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
39 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
42 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
43 #define I40E_FD_CLEAN_DELAY 10
45 * i40e_program_fdir_filter - Program a Flow Director filter
46 * @fdir_data: Packet data that will be filter parameters
47 * @raw_packet: the pre-allocated packet buffer for FDir
49 * @add: True for add/update, False for remove
51 int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data, u8 *raw_packet,
52 struct i40e_pf *pf, bool add)
54 struct i40e_filter_program_desc *fdir_desc;
55 struct i40e_tx_buffer *tx_buf, *first;
56 struct i40e_tx_desc *tx_desc;
57 struct i40e_ring *tx_ring;
58 unsigned int fpt, dcc;
66 /* find existing FDIR VSI */
68 for (i = 0; i < pf->num_alloc_vsi; i++)
69 if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR)
74 tx_ring = vsi->tx_rings[0];
77 /* we need two descriptors to add/del a filter and we can wait */
79 if (I40E_DESC_UNUSED(tx_ring) > 1)
81 msleep_interruptible(1);
83 } while (delay < I40E_FD_CLEAN_DELAY);
85 if (!(I40E_DESC_UNUSED(tx_ring) > 1))
88 dma = dma_map_single(dev, raw_packet,
89 I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
90 if (dma_mapping_error(dev, dma))
93 /* grab the next descriptor */
94 i = tx_ring->next_to_use;
95 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
96 first = &tx_ring->tx_bi[i];
97 memset(first, 0, sizeof(struct i40e_tx_buffer));
99 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
101 fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
102 I40E_TXD_FLTR_QW0_QINDEX_MASK;
104 fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) &
105 I40E_TXD_FLTR_QW0_FLEXOFF_MASK;
107 fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) &
108 I40E_TXD_FLTR_QW0_PCTYPE_MASK;
110 /* Use LAN VSI Id if not programmed by user */
111 if (fdir_data->dest_vsi == 0)
112 fpt |= (pf->vsi[pf->lan_vsi]->id) <<
113 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
115 fpt |= ((u32)fdir_data->dest_vsi <<
116 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) &
117 I40E_TXD_FLTR_QW0_DEST_VSI_MASK;
119 dcc = I40E_TX_DESC_DTYPE_FILTER_PROG;
122 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
123 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
125 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
126 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
128 dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) &
129 I40E_TXD_FLTR_QW1_DEST_MASK;
131 dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) &
132 I40E_TXD_FLTR_QW1_FD_STATUS_MASK;
134 if (fdir_data->cnt_index != 0) {
135 dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
136 dcc |= ((u32)fdir_data->cnt_index <<
137 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
138 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
141 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt);
142 fdir_desc->rsvd = cpu_to_le32(0);
143 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc);
144 fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id);
146 /* Now program a dummy descriptor */
147 i = tx_ring->next_to_use;
148 tx_desc = I40E_TX_DESC(tx_ring, i);
149 tx_buf = &tx_ring->tx_bi[i];
151 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
153 memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
155 /* record length, and DMA address */
156 dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
157 dma_unmap_addr_set(tx_buf, dma, dma);
159 tx_desc->buffer_addr = cpu_to_le64(dma);
160 td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
162 tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
163 tx_buf->raw_buf = (void *)raw_packet;
165 tx_desc->cmd_type_offset_bsz =
166 build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
168 /* Force memory writes to complete before letting h/w
169 * know there are new descriptors to fetch.
173 /* Mark the data descriptor to be watched */
174 first->next_to_watch = tx_desc;
176 writel(tx_ring->next_to_use, tx_ring->tail);
183 #define IP_HEADER_OFFSET 14
184 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
186 * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
187 * @vsi: pointer to the targeted VSI
188 * @fd_data: the flow director data required for the FDir descriptor
189 * @add: true adds a filter, false removes it
191 * Returns 0 if the filters were successfully added or removed
193 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
194 struct i40e_fdir_filter *fd_data,
197 struct i40e_pf *pf = vsi->back;
203 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
204 0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
205 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
207 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
210 memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
212 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
213 udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
214 + sizeof(struct iphdr));
216 ip->daddr = fd_data->dst_ip[0];
217 udp->dest = fd_data->dst_port;
218 ip->saddr = fd_data->src_ip[0];
219 udp->source = fd_data->src_port;
221 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
222 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
224 dev_info(&pf->pdev->dev,
225 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
226 fd_data->pctype, fd_data->fd_id, ret);
228 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
230 dev_info(&pf->pdev->dev,
231 "Filter OK for PCTYPE %d loc = %d\n",
232 fd_data->pctype, fd_data->fd_id);
234 dev_info(&pf->pdev->dev,
235 "Filter deleted for PCTYPE %d loc = %d\n",
236 fd_data->pctype, fd_data->fd_id);
241 return err ? -EOPNOTSUPP : 0;
244 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
246 * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
247 * @vsi: pointer to the targeted VSI
248 * @fd_data: the flow director data required for the FDir descriptor
249 * @add: true adds a filter, false removes it
251 * Returns 0 if the filters were successfully added or removed
253 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
254 struct i40e_fdir_filter *fd_data,
257 struct i40e_pf *pf = vsi->back;
264 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
265 0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
266 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
267 0x0, 0x72, 0, 0, 0, 0};
269 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
272 memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
274 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
275 tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
276 + sizeof(struct iphdr));
278 ip->daddr = fd_data->dst_ip[0];
279 tcp->dest = fd_data->dst_port;
280 ip->saddr = fd_data->src_ip[0];
281 tcp->source = fd_data->src_port;
285 if (pf->flags & I40E_FLAG_FD_ATR_ENABLED) {
286 if (I40E_DEBUG_FD & pf->hw.debug_mask)
287 dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
288 pf->flags &= ~I40E_FLAG_FD_ATR_ENABLED;
291 pf->fd_tcp_rule = (pf->fd_tcp_rule > 0) ?
292 (pf->fd_tcp_rule - 1) : 0;
293 if (pf->fd_tcp_rule == 0) {
294 pf->flags |= I40E_FLAG_FD_ATR_ENABLED;
295 if (I40E_DEBUG_FD & pf->hw.debug_mask)
296 dev_info(&pf->pdev->dev, "ATR re-enabled due to no sideband TCP/IPv4 rules\n");
300 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
301 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
304 dev_info(&pf->pdev->dev,
305 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
306 fd_data->pctype, fd_data->fd_id, ret);
308 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
310 dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
311 fd_data->pctype, fd_data->fd_id);
313 dev_info(&pf->pdev->dev,
314 "Filter deleted for PCTYPE %d loc = %d\n",
315 fd_data->pctype, fd_data->fd_id);
321 return err ? -EOPNOTSUPP : 0;
325 * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
326 * a specific flow spec
327 * @vsi: pointer to the targeted VSI
328 * @fd_data: the flow director data required for the FDir descriptor
329 * @add: true adds a filter, false removes it
331 * Returns 0 if the filters were successfully added or removed
333 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
334 struct i40e_fdir_filter *fd_data,
340 #define I40E_IP_DUMMY_PACKET_LEN 34
342 * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
343 * a specific flow spec
344 * @vsi: pointer to the targeted VSI
345 * @fd_data: the flow director data required for the FDir descriptor
346 * @add: true adds a filter, false removes it
348 * Returns 0 if the filters were successfully added or removed
350 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
351 struct i40e_fdir_filter *fd_data,
354 struct i40e_pf *pf = vsi->back;
360 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
361 0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
364 for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
365 i <= I40E_FILTER_PCTYPE_FRAG_IPV4; i++) {
366 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
369 memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
370 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
372 ip->saddr = fd_data->src_ip[0];
373 ip->daddr = fd_data->dst_ip[0];
377 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
380 dev_info(&pf->pdev->dev,
381 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
382 fd_data->pctype, fd_data->fd_id, ret);
384 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
386 dev_info(&pf->pdev->dev,
387 "Filter OK for PCTYPE %d loc = %d\n",
388 fd_data->pctype, fd_data->fd_id);
390 dev_info(&pf->pdev->dev,
391 "Filter deleted for PCTYPE %d loc = %d\n",
392 fd_data->pctype, fd_data->fd_id);
399 return err ? -EOPNOTSUPP : 0;
403 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
404 * @vsi: pointer to the targeted VSI
405 * @cmd: command to get or set RX flow classification rules
406 * @add: true adds a filter, false removes it
409 int i40e_add_del_fdir(struct i40e_vsi *vsi,
410 struct i40e_fdir_filter *input, bool add)
412 struct i40e_pf *pf = vsi->back;
415 switch (input->flow_type & ~FLOW_EXT) {
417 ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
420 ret = i40e_add_del_fdir_udpv4(vsi, input, add);
423 ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
426 ret = i40e_add_del_fdir_ipv4(vsi, input, add);
429 switch (input->ip4_proto) {
431 ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
434 ret = i40e_add_del_fdir_udpv4(vsi, input, add);
437 ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
440 ret = i40e_add_del_fdir_ipv4(vsi, input, add);
445 dev_info(&pf->pdev->dev, "Could not specify spec type %d\n",
450 /* The buffer allocated here is freed by the i40e_clean_tx_ring() */
455 * i40e_fd_handle_status - check the Programming Status for FD
456 * @rx_ring: the Rx ring for this descriptor
457 * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
458 * @prog_id: the id originally used for programming
460 * This is used to verify if the FD programming or invalidation
461 * requested by SW to the HW is successful or not and take actions accordingly.
463 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
464 union i40e_rx_desc *rx_desc, u8 prog_id)
466 struct i40e_pf *pf = rx_ring->vsi->back;
467 struct pci_dev *pdev = pf->pdev;
468 u32 fcnt_prog, fcnt_avail;
472 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
473 error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
474 I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
476 if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
477 pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
478 if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
479 (I40E_DEBUG_FD & pf->hw.debug_mask))
480 dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
483 /* Check if the programming error is for ATR.
484 * If so, auto disable ATR and set a state for
485 * flush in progress. Next time we come here if flush is in
486 * progress do nothing, once flush is complete the state will
489 if (test_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state))
493 /* store the current atr filter count */
494 pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
496 if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
497 (pf->auto_disable_flags & I40E_FLAG_FD_SB_ENABLED)) {
498 pf->auto_disable_flags |= I40E_FLAG_FD_ATR_ENABLED;
499 set_bit(__I40E_FD_FLUSH_REQUESTED, &pf->state);
502 /* filter programming failed most likely due to table full */
503 fcnt_prog = i40e_get_global_fd_count(pf);
504 fcnt_avail = pf->fdir_pf_filter_count;
505 /* If ATR is running fcnt_prog can quickly change,
506 * if we are very close to full, it makes sense to disable
507 * FD ATR/SB and then re-enable it when there is room.
509 if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
510 if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
511 !(pf->auto_disable_flags &
512 I40E_FLAG_FD_SB_ENABLED)) {
513 if (I40E_DEBUG_FD & pf->hw.debug_mask)
514 dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
515 pf->auto_disable_flags |=
516 I40E_FLAG_FD_SB_ENABLED;
519 } else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
520 if (I40E_DEBUG_FD & pf->hw.debug_mask)
521 dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
522 rx_desc->wb.qword0.hi_dword.fd_id);
527 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
528 * @ring: the ring that owns the buffer
529 * @tx_buffer: the buffer to free
531 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
532 struct i40e_tx_buffer *tx_buffer)
534 if (tx_buffer->skb) {
535 dev_kfree_skb_any(tx_buffer->skb);
536 if (dma_unmap_len(tx_buffer, len))
537 dma_unmap_single(ring->dev,
538 dma_unmap_addr(tx_buffer, dma),
539 dma_unmap_len(tx_buffer, len),
541 } else if (dma_unmap_len(tx_buffer, len)) {
542 dma_unmap_page(ring->dev,
543 dma_unmap_addr(tx_buffer, dma),
544 dma_unmap_len(tx_buffer, len),
548 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
549 kfree(tx_buffer->raw_buf);
551 tx_buffer->next_to_watch = NULL;
552 tx_buffer->skb = NULL;
553 dma_unmap_len_set(tx_buffer, len, 0);
554 /* tx_buffer must be completely set up in the transmit path */
558 * i40e_clean_tx_ring - Free any empty Tx buffers
559 * @tx_ring: ring to be cleaned
561 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
563 unsigned long bi_size;
566 /* ring already cleared, nothing to do */
570 /* Free all the Tx ring sk_buffs */
571 for (i = 0; i < tx_ring->count; i++)
572 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
574 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
575 memset(tx_ring->tx_bi, 0, bi_size);
577 /* Zero out the descriptor ring */
578 memset(tx_ring->desc, 0, tx_ring->size);
580 tx_ring->next_to_use = 0;
581 tx_ring->next_to_clean = 0;
583 if (!tx_ring->netdev)
586 /* cleanup Tx queue statistics */
587 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
588 tx_ring->queue_index));
592 * i40e_free_tx_resources - Free Tx resources per queue
593 * @tx_ring: Tx descriptor ring for a specific queue
595 * Free all transmit software resources
597 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
599 i40e_clean_tx_ring(tx_ring);
600 kfree(tx_ring->tx_bi);
601 tx_ring->tx_bi = NULL;
604 dma_free_coherent(tx_ring->dev, tx_ring->size,
605 tx_ring->desc, tx_ring->dma);
606 tx_ring->desc = NULL;
611 * i40e_get_tx_pending - how many tx descriptors not processed
612 * @tx_ring: the ring of descriptors
614 * Since there is no access to the ring head register
615 * in XL710, we need to use our local copies
617 u32 i40e_get_tx_pending(struct i40e_ring *ring)
621 head = i40e_get_head(ring);
622 tail = readl(ring->tail);
625 return (head < tail) ?
626 tail - head : (tail + ring->count - head);
631 #define WB_STRIDE 0x3
634 * i40e_clean_tx_irq - Reclaim resources after transmit completes
635 * @tx_ring: tx ring to clean
636 * @budget: how many cleans we're allowed
638 * Returns true if there's any budget left (e.g. the clean is finished)
640 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
642 u16 i = tx_ring->next_to_clean;
643 struct i40e_tx_buffer *tx_buf;
644 struct i40e_tx_desc *tx_head;
645 struct i40e_tx_desc *tx_desc;
646 unsigned int total_packets = 0;
647 unsigned int total_bytes = 0;
649 tx_buf = &tx_ring->tx_bi[i];
650 tx_desc = I40E_TX_DESC(tx_ring, i);
653 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
656 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
658 /* if next_to_watch is not set then there is no work pending */
662 /* prevent any other reads prior to eop_desc */
663 read_barrier_depends();
665 /* we have caught up to head, no work left to do */
666 if (tx_head == tx_desc)
669 /* clear next_to_watch to prevent false hangs */
670 tx_buf->next_to_watch = NULL;
672 /* update the statistics for this packet */
673 total_bytes += tx_buf->bytecount;
674 total_packets += tx_buf->gso_segs;
677 dev_consume_skb_any(tx_buf->skb);
679 /* unmap skb header data */
680 dma_unmap_single(tx_ring->dev,
681 dma_unmap_addr(tx_buf, dma),
682 dma_unmap_len(tx_buf, len),
685 /* clear tx_buffer data */
687 dma_unmap_len_set(tx_buf, len, 0);
689 /* unmap remaining buffers */
690 while (tx_desc != eop_desc) {
697 tx_buf = tx_ring->tx_bi;
698 tx_desc = I40E_TX_DESC(tx_ring, 0);
701 /* unmap any remaining paged data */
702 if (dma_unmap_len(tx_buf, len)) {
703 dma_unmap_page(tx_ring->dev,
704 dma_unmap_addr(tx_buf, dma),
705 dma_unmap_len(tx_buf, len),
707 dma_unmap_len_set(tx_buf, len, 0);
711 /* move us one more past the eop_desc for start of next pkt */
717 tx_buf = tx_ring->tx_bi;
718 tx_desc = I40E_TX_DESC(tx_ring, 0);
723 /* update budget accounting */
725 } while (likely(budget));
728 tx_ring->next_to_clean = i;
729 u64_stats_update_begin(&tx_ring->syncp);
730 tx_ring->stats.bytes += total_bytes;
731 tx_ring->stats.packets += total_packets;
732 u64_stats_update_end(&tx_ring->syncp);
733 tx_ring->q_vector->tx.total_bytes += total_bytes;
734 tx_ring->q_vector->tx.total_packets += total_packets;
736 if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
739 /* check to see if there are < 4 descriptors
740 * waiting to be written back, then kick the hardware to force
741 * them to be written back in case we stay in NAPI.
742 * In this mode on X722 we do not enable Interrupt.
744 j = i40e_get_tx_pending(tx_ring);
747 ((j / (WB_STRIDE + 1)) == 0) && (j != 0) &&
748 !test_bit(__I40E_DOWN, &tx_ring->vsi->state) &&
749 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
750 tx_ring->arm_wb = true;
753 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
754 tx_ring->queue_index),
755 total_packets, total_bytes);
757 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
758 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
759 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
760 /* Make sure that anybody stopping the queue after this
761 * sees the new next_to_clean.
764 if (__netif_subqueue_stopped(tx_ring->netdev,
765 tx_ring->queue_index) &&
766 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
767 netif_wake_subqueue(tx_ring->netdev,
768 tx_ring->queue_index);
769 ++tx_ring->tx_stats.restart_queue;
777 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
778 * @vsi: the VSI we care about
779 * @q_vector: the vector on which to enable writeback
782 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
783 struct i40e_q_vector *q_vector)
785 u16 flags = q_vector->tx.ring[0].flags;
788 if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
791 if (q_vector->arm_wb_state)
794 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
795 val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
796 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
799 I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
802 val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
803 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
805 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
807 q_vector->arm_wb_state = true;
811 * i40e_force_wb - Issue SW Interrupt so HW does a wb
812 * @vsi: the VSI we care about
813 * @q_vector: the vector on which to force writeback
816 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
818 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
819 u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
820 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
821 I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
822 I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
823 /* allow 00 to be written to the index */
826 I40E_PFINT_DYN_CTLN(q_vector->v_idx +
827 vsi->base_vector - 1), val);
829 u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
830 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
831 I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
832 I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
833 /* allow 00 to be written to the index */
835 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
840 * i40e_set_new_dynamic_itr - Find new ITR level
841 * @rc: structure containing ring performance data
843 * Returns true if ITR changed, false if not
845 * Stores a new ITR value based on packets and byte counts during
846 * the last interrupt. The advantage of per interrupt computation
847 * is faster updates and more accurate ITR for the current traffic
848 * pattern. Constants in this function were computed based on
849 * theoretical maximum wire speed and thresholds were set based on
850 * testing data as well as attempting to minimize response time
851 * while increasing bulk throughput.
853 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
855 enum i40e_latency_range new_latency_range = rc->latency_range;
856 struct i40e_q_vector *qv = rc->ring->q_vector;
857 u32 new_itr = rc->itr;
861 if (rc->total_packets == 0 || !rc->itr)
864 /* simple throttlerate management
865 * 0-10MB/s lowest (50000 ints/s)
866 * 10-20MB/s low (20000 ints/s)
867 * 20-1249MB/s bulk (18000 ints/s)
868 * > 40000 Rx packets per second (8000 ints/s)
870 * The math works out because the divisor is in 10^(-6) which
871 * turns the bytes/us input value into MB/s values, but
872 * make sure to use usecs, as the register values written
873 * are in 2 usec increments in the ITR registers, and make sure
874 * to use the smoothed values that the countdown timer gives us.
876 usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
877 bytes_per_int = rc->total_bytes / usecs;
879 switch (new_latency_range) {
880 case I40E_LOWEST_LATENCY:
881 if (bytes_per_int > 10)
882 new_latency_range = I40E_LOW_LATENCY;
884 case I40E_LOW_LATENCY:
885 if (bytes_per_int > 20)
886 new_latency_range = I40E_BULK_LATENCY;
887 else if (bytes_per_int <= 10)
888 new_latency_range = I40E_LOWEST_LATENCY;
890 case I40E_BULK_LATENCY:
891 case I40E_ULTRA_LATENCY:
893 if (bytes_per_int <= 20)
894 new_latency_range = I40E_LOW_LATENCY;
898 /* this is to adjust RX more aggressively when streaming small
899 * packets. The value of 40000 was picked as it is just beyond
900 * what the hardware can receive per second if in low latency
903 #define RX_ULTRA_PACKET_RATE 40000
905 if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
907 new_latency_range = I40E_ULTRA_LATENCY;
909 rc->latency_range = new_latency_range;
911 switch (new_latency_range) {
912 case I40E_LOWEST_LATENCY:
913 new_itr = I40E_ITR_50K;
915 case I40E_LOW_LATENCY:
916 new_itr = I40E_ITR_20K;
918 case I40E_BULK_LATENCY:
919 new_itr = I40E_ITR_18K;
921 case I40E_ULTRA_LATENCY:
922 new_itr = I40E_ITR_8K;
929 rc->total_packets = 0;
931 if (new_itr != rc->itr) {
940 * i40e_clean_programming_status - clean the programming status descriptor
941 * @rx_ring: the rx ring that has this descriptor
942 * @rx_desc: the rx descriptor written back by HW
944 * Flow director should handle FD_FILTER_STATUS to check its filter programming
945 * status being successful or not and take actions accordingly. FCoE should
946 * handle its context/filter programming/invalidation status and take actions.
949 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
950 union i40e_rx_desc *rx_desc)
955 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
956 id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
957 I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
959 if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
960 i40e_fd_handle_status(rx_ring, rx_desc, id);
962 else if ((id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_PROG_STATUS) ||
963 (id == I40E_RX_PROG_STATUS_DESC_FCOE_CTXT_INVL_STATUS))
964 i40e_fcoe_handle_status(rx_ring, rx_desc, id);
969 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
970 * @tx_ring: the tx ring to set up
972 * Return 0 on success, negative on error
974 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
976 struct device *dev = tx_ring->dev;
982 /* warn if we are about to overwrite the pointer */
983 WARN_ON(tx_ring->tx_bi);
984 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
985 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
989 /* round up to nearest 4K */
990 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
991 /* add u32 for head writeback, align after this takes care of
992 * guaranteeing this is at least one cache line in size
994 tx_ring->size += sizeof(u32);
995 tx_ring->size = ALIGN(tx_ring->size, 4096);
996 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
997 &tx_ring->dma, GFP_KERNEL);
998 if (!tx_ring->desc) {
999 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1004 tx_ring->next_to_use = 0;
1005 tx_ring->next_to_clean = 0;
1009 kfree(tx_ring->tx_bi);
1010 tx_ring->tx_bi = NULL;
1015 * i40e_clean_rx_ring - Free Rx buffers
1016 * @rx_ring: ring to be cleaned
1018 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1020 struct device *dev = rx_ring->dev;
1021 struct i40e_rx_buffer *rx_bi;
1022 unsigned long bi_size;
1025 /* ring already cleared, nothing to do */
1026 if (!rx_ring->rx_bi)
1029 if (ring_is_ps_enabled(rx_ring)) {
1030 int bufsz = ALIGN(rx_ring->rx_hdr_len, 256) * rx_ring->count;
1032 rx_bi = &rx_ring->rx_bi[0];
1033 if (rx_bi->hdr_buf) {
1034 dma_free_coherent(dev,
1038 for (i = 0; i < rx_ring->count; i++) {
1039 rx_bi = &rx_ring->rx_bi[i];
1041 rx_bi->hdr_buf = NULL;
1045 /* Free all the Rx ring sk_buffs */
1046 for (i = 0; i < rx_ring->count; i++) {
1047 rx_bi = &rx_ring->rx_bi[i];
1049 dma_unmap_single(dev,
1051 rx_ring->rx_buf_len,
1056 dev_kfree_skb(rx_bi->skb);
1060 if (rx_bi->page_dma) {
1065 rx_bi->page_dma = 0;
1067 __free_page(rx_bi->page);
1069 rx_bi->page_offset = 0;
1073 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1074 memset(rx_ring->rx_bi, 0, bi_size);
1076 /* Zero out the descriptor ring */
1077 memset(rx_ring->desc, 0, rx_ring->size);
1079 rx_ring->next_to_clean = 0;
1080 rx_ring->next_to_use = 0;
1084 * i40e_free_rx_resources - Free Rx resources
1085 * @rx_ring: ring to clean the resources from
1087 * Free all receive software resources
1089 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1091 i40e_clean_rx_ring(rx_ring);
1092 kfree(rx_ring->rx_bi);
1093 rx_ring->rx_bi = NULL;
1095 if (rx_ring->desc) {
1096 dma_free_coherent(rx_ring->dev, rx_ring->size,
1097 rx_ring->desc, rx_ring->dma);
1098 rx_ring->desc = NULL;
1103 * i40e_alloc_rx_headers - allocate rx header buffers
1104 * @rx_ring: ring to alloc buffers
1106 * Allocate rx header buffers for the entire ring. As these are static,
1107 * this is only called when setting up a new ring.
1109 void i40e_alloc_rx_headers(struct i40e_ring *rx_ring)
1111 struct device *dev = rx_ring->dev;
1112 struct i40e_rx_buffer *rx_bi;
1118 if (rx_ring->rx_bi[0].hdr_buf)
1120 /* Make sure the buffers don't cross cache line boundaries. */
1121 buf_size = ALIGN(rx_ring->rx_hdr_len, 256);
1122 buffer = dma_alloc_coherent(dev, buf_size * rx_ring->count,
1126 for (i = 0; i < rx_ring->count; i++) {
1127 rx_bi = &rx_ring->rx_bi[i];
1128 rx_bi->dma = dma + (i * buf_size);
1129 rx_bi->hdr_buf = buffer + (i * buf_size);
1134 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1135 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1137 * Returns 0 on success, negative on failure
1139 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1141 struct device *dev = rx_ring->dev;
1144 /* warn if we are about to overwrite the pointer */
1145 WARN_ON(rx_ring->rx_bi);
1146 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1147 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1148 if (!rx_ring->rx_bi)
1151 u64_stats_init(&rx_ring->syncp);
1153 /* Round up to nearest 4K */
1154 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
1155 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
1156 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1157 rx_ring->size = ALIGN(rx_ring->size, 4096);
1158 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1159 &rx_ring->dma, GFP_KERNEL);
1161 if (!rx_ring->desc) {
1162 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1167 rx_ring->next_to_clean = 0;
1168 rx_ring->next_to_use = 0;
1172 kfree(rx_ring->rx_bi);
1173 rx_ring->rx_bi = NULL;
1178 * i40e_release_rx_desc - Store the new tail and head values
1179 * @rx_ring: ring to bump
1180 * @val: new head index
1182 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1184 rx_ring->next_to_use = val;
1185 /* Force memory writes to complete before letting h/w
1186 * know there are new descriptors to fetch. (Only
1187 * applicable for weak-ordered memory model archs,
1191 writel(val, rx_ring->tail);
1195 * i40e_alloc_rx_buffers_ps - Replace used receive buffers; packet split
1196 * @rx_ring: ring to place buffers on
1197 * @cleaned_count: number of buffers to replace
1199 * Returns true if any errors on allocation
1201 bool i40e_alloc_rx_buffers_ps(struct i40e_ring *rx_ring, u16 cleaned_count)
1203 u16 i = rx_ring->next_to_use;
1204 union i40e_rx_desc *rx_desc;
1205 struct i40e_rx_buffer *bi;
1207 /* do nothing if no valid netdev defined */
1208 if (!rx_ring->netdev || !cleaned_count)
1211 while (cleaned_count--) {
1212 rx_desc = I40E_RX_DESC(rx_ring, i);
1213 bi = &rx_ring->rx_bi[i];
1215 if (bi->skb) /* desc is in use */
1218 bi->page = alloc_page(GFP_ATOMIC);
1220 rx_ring->rx_stats.alloc_page_failed++;
1225 if (!bi->page_dma) {
1226 /* use a half page if we're re-using */
1227 bi->page_offset ^= PAGE_SIZE / 2;
1228 bi->page_dma = dma_map_page(rx_ring->dev,
1233 if (dma_mapping_error(rx_ring->dev,
1235 rx_ring->rx_stats.alloc_page_failed++;
1241 dma_sync_single_range_for_device(rx_ring->dev,
1242 rx_ring->rx_bi[0].dma,
1243 i * rx_ring->rx_hdr_len,
1244 rx_ring->rx_hdr_len,
1246 /* Refresh the desc even if buffer_addrs didn't change
1247 * because each write-back erases this info.
1249 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
1250 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
1252 if (i == rx_ring->count)
1256 if (rx_ring->next_to_use != i)
1257 i40e_release_rx_desc(rx_ring, i);
1262 if (rx_ring->next_to_use != i)
1263 i40e_release_rx_desc(rx_ring, i);
1265 /* make sure to come back via polling to try again after
1266 * allocation failure
1272 * i40e_alloc_rx_buffers_1buf - Replace used receive buffers; single buffer
1273 * @rx_ring: ring to place buffers on
1274 * @cleaned_count: number of buffers to replace
1276 * Returns true if any errors on allocation
1278 bool i40e_alloc_rx_buffers_1buf(struct i40e_ring *rx_ring, u16 cleaned_count)
1280 u16 i = rx_ring->next_to_use;
1281 union i40e_rx_desc *rx_desc;
1282 struct i40e_rx_buffer *bi;
1283 struct sk_buff *skb;
1285 /* do nothing if no valid netdev defined */
1286 if (!rx_ring->netdev || !cleaned_count)
1289 while (cleaned_count--) {
1290 rx_desc = I40E_RX_DESC(rx_ring, i);
1291 bi = &rx_ring->rx_bi[i];
1295 skb = __netdev_alloc_skb_ip_align(rx_ring->netdev,
1296 rx_ring->rx_buf_len,
1300 rx_ring->rx_stats.alloc_buff_failed++;
1303 /* initialize queue mapping */
1304 skb_record_rx_queue(skb, rx_ring->queue_index);
1309 bi->dma = dma_map_single(rx_ring->dev,
1311 rx_ring->rx_buf_len,
1313 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
1314 rx_ring->rx_stats.alloc_buff_failed++;
1316 dev_kfree_skb(bi->skb);
1322 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
1323 rx_desc->read.hdr_addr = 0;
1325 if (i == rx_ring->count)
1329 if (rx_ring->next_to_use != i)
1330 i40e_release_rx_desc(rx_ring, i);
1335 if (rx_ring->next_to_use != i)
1336 i40e_release_rx_desc(rx_ring, i);
1338 /* make sure to come back via polling to try again after
1339 * allocation failure
1345 * i40e_receive_skb - Send a completed packet up the stack
1346 * @rx_ring: rx ring in play
1347 * @skb: packet to send up
1348 * @vlan_tag: vlan tag for packet
1350 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1351 struct sk_buff *skb, u16 vlan_tag)
1353 struct i40e_q_vector *q_vector = rx_ring->q_vector;
1355 if (vlan_tag & VLAN_VID_MASK)
1356 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1358 napi_gro_receive(&q_vector->napi, skb);
1362 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1363 * @vsi: the VSI we care about
1364 * @skb: skb currently being received and modified
1365 * @rx_status: status value of last descriptor in packet
1366 * @rx_error: error value of last descriptor in packet
1367 * @rx_ptype: ptype value of last descriptor in packet
1369 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1370 struct sk_buff *skb,
1375 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(rx_ptype);
1376 bool ipv4 = false, ipv6 = false;
1377 bool ipv4_tunnel, ipv6_tunnel;
1382 ipv4_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
1383 (rx_ptype <= I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
1384 ipv6_tunnel = (rx_ptype >= I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
1385 (rx_ptype <= I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
1387 skb->ip_summed = CHECKSUM_NONE;
1389 /* Rx csum enabled and ip headers found? */
1390 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1393 /* did the hardware decode the packet and checksum? */
1394 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1397 /* both known and outer_ip must be set for the below code to work */
1398 if (!(decoded.known && decoded.outer_ip))
1401 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1402 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4)
1404 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1405 decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6)
1409 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1410 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1413 /* likely incorrect csum if alternate IP extension headers found */
1415 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1416 /* don't increment checksum err here, non-fatal err */
1419 /* there was some L4 error, count error and punt packet to the stack */
1420 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1423 /* handle packets that were not able to be checksummed due
1424 * to arrival speed, in this case the stack can compute
1427 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1430 /* If VXLAN/GENEVE traffic has an outer UDPv4 checksum we need to check
1431 * it in the driver, hardware does not do it for us.
1432 * Since L3L4P bit was set we assume a valid IHL value (>=5)
1433 * so the total length of IPv4 header is IHL*4 bytes
1434 * The UDP_0 bit *may* bet set if the *inner* header is UDP
1436 if (!(vsi->back->flags & I40E_FLAG_OUTER_UDP_CSUM_CAPABLE) &&
1438 skb->transport_header = skb->mac_header +
1439 sizeof(struct ethhdr) +
1440 (ip_hdr(skb)->ihl * 4);
1442 /* Add 4 bytes for VLAN tagged packets */
1443 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
1444 skb->protocol == htons(ETH_P_8021AD))
1447 if ((ip_hdr(skb)->protocol == IPPROTO_UDP) &&
1448 (udp_hdr(skb)->check != 0)) {
1449 rx_udp_csum = udp_csum(skb);
1451 csum = csum_tcpudp_magic(
1452 iph->saddr, iph->daddr,
1453 (skb->len - skb_transport_offset(skb)),
1454 IPPROTO_UDP, rx_udp_csum);
1456 if (udp_hdr(skb)->check != csum)
1459 } /* else its GRE and so no outer UDP header */
1462 skb->ip_summed = CHECKSUM_UNNECESSARY;
1463 skb->csum_level = ipv4_tunnel || ipv6_tunnel;
1468 vsi->back->hw_csum_rx_error++;
1472 * i40e_ptype_to_htype - get a hash type
1473 * @ptype: the ptype value from the descriptor
1475 * Returns a hash type to be used by skb_set_hash
1477 static inline enum pkt_hash_types i40e_ptype_to_htype(u8 ptype)
1479 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1482 return PKT_HASH_TYPE_NONE;
1484 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1485 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1486 return PKT_HASH_TYPE_L4;
1487 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1488 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1489 return PKT_HASH_TYPE_L3;
1491 return PKT_HASH_TYPE_L2;
1495 * i40e_rx_hash - set the hash value in the skb
1496 * @ring: descriptor ring
1497 * @rx_desc: specific descriptor
1499 static inline void i40e_rx_hash(struct i40e_ring *ring,
1500 union i40e_rx_desc *rx_desc,
1501 struct sk_buff *skb,
1505 const __le64 rss_mask =
1506 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1507 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1509 if (ring->netdev->features & NETIF_F_RXHASH)
1512 if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1513 hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1514 skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1519 * i40e_clean_rx_irq_ps - Reclaim resources after receive; packet split
1520 * @rx_ring: rx ring to clean
1521 * @budget: how many cleans we're allowed
1523 * Returns true if there's any budget left (e.g. the clean is finished)
1525 static int i40e_clean_rx_irq_ps(struct i40e_ring *rx_ring, const int budget)
1527 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1528 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
1529 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1530 const int current_node = numa_mem_id();
1531 struct i40e_vsi *vsi = rx_ring->vsi;
1532 u16 i = rx_ring->next_to_clean;
1533 union i40e_rx_desc *rx_desc;
1534 u32 rx_error, rx_status;
1535 bool failure = false;
1543 struct i40e_rx_buffer *rx_bi;
1544 struct sk_buff *skb;
1546 /* return some buffers to hardware, one at a time is too slow */
1547 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1548 failure = failure ||
1549 i40e_alloc_rx_buffers_ps(rx_ring,
1554 i = rx_ring->next_to_clean;
1555 rx_desc = I40E_RX_DESC(rx_ring, i);
1556 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1557 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1558 I40E_RXD_QW1_STATUS_SHIFT;
1560 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1563 /* This memory barrier is needed to keep us from reading
1564 * any other fields out of the rx_desc until we know the
1568 if (i40e_rx_is_programming_status(qword)) {
1569 i40e_clean_programming_status(rx_ring, rx_desc);
1570 I40E_RX_INCREMENT(rx_ring, i);
1573 rx_bi = &rx_ring->rx_bi[i];
1576 skb = __netdev_alloc_skb_ip_align(rx_ring->netdev,
1577 rx_ring->rx_hdr_len,
1581 rx_ring->rx_stats.alloc_buff_failed++;
1586 /* initialize queue mapping */
1587 skb_record_rx_queue(skb, rx_ring->queue_index);
1588 /* we are reusing so sync this buffer for CPU use */
1589 dma_sync_single_range_for_cpu(rx_ring->dev,
1590 rx_ring->rx_bi[0].dma,
1591 i * rx_ring->rx_hdr_len,
1592 rx_ring->rx_hdr_len,
1595 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1596 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1597 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK) >>
1598 I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
1599 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK) >>
1600 I40E_RXD_QW1_LENGTH_SPH_SHIFT;
1602 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1603 I40E_RXD_QW1_ERROR_SHIFT;
1604 rx_hbo = rx_error & BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1605 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1607 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1608 I40E_RXD_QW1_PTYPE_SHIFT;
1609 prefetch(rx_bi->page);
1612 if (rx_hbo || rx_sph) {
1616 len = I40E_RX_HDR_SIZE;
1618 len = rx_header_len;
1619 memcpy(__skb_put(skb, len), rx_bi->hdr_buf, len);
1620 } else if (skb->len == 0) {
1623 len = (rx_packet_len > skb_headlen(skb) ?
1624 skb_headlen(skb) : rx_packet_len);
1625 memcpy(__skb_put(skb, len),
1626 rx_bi->page + rx_bi->page_offset,
1628 rx_bi->page_offset += len;
1629 rx_packet_len -= len;
1632 /* Get the rest of the data if this was a header split */
1633 if (rx_packet_len) {
1634 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1639 skb->len += rx_packet_len;
1640 skb->data_len += rx_packet_len;
1641 skb->truesize += rx_packet_len;
1643 if ((page_count(rx_bi->page) == 1) &&
1644 (page_to_nid(rx_bi->page) == current_node))
1645 get_page(rx_bi->page);
1649 dma_unmap_page(rx_ring->dev,
1653 rx_bi->page_dma = 0;
1655 I40E_RX_INCREMENT(rx_ring, i);
1658 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1659 struct i40e_rx_buffer *next_buffer;
1661 next_buffer = &rx_ring->rx_bi[i];
1662 next_buffer->skb = skb;
1663 rx_ring->rx_stats.non_eop_descs++;
1667 /* ERR_MASK will only have valid bits if EOP set */
1668 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1669 dev_kfree_skb_any(skb);
1673 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1675 if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) {
1676 i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status &
1677 I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1678 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT);
1679 rx_ring->last_rx_timestamp = jiffies;
1682 /* probably a little skewed due to removing CRC */
1683 total_rx_bytes += skb->len;
1686 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1688 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1690 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1691 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1694 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1695 dev_kfree_skb_any(skb);
1699 i40e_receive_skb(rx_ring, skb, vlan_tag);
1701 rx_desc->wb.qword1.status_error_len = 0;
1703 } while (likely(total_rx_packets < budget));
1705 u64_stats_update_begin(&rx_ring->syncp);
1706 rx_ring->stats.packets += total_rx_packets;
1707 rx_ring->stats.bytes += total_rx_bytes;
1708 u64_stats_update_end(&rx_ring->syncp);
1709 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1710 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1712 return failure ? budget : total_rx_packets;
1716 * i40e_clean_rx_irq_1buf - Reclaim resources after receive; single buffer
1717 * @rx_ring: rx ring to clean
1718 * @budget: how many cleans we're allowed
1720 * Returns number of packets cleaned
1722 static int i40e_clean_rx_irq_1buf(struct i40e_ring *rx_ring, int budget)
1724 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1725 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1726 struct i40e_vsi *vsi = rx_ring->vsi;
1727 union i40e_rx_desc *rx_desc;
1728 u32 rx_error, rx_status;
1730 bool failure = false;
1736 struct i40e_rx_buffer *rx_bi;
1737 struct sk_buff *skb;
1739 /* return some buffers to hardware, one at a time is too slow */
1740 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1741 failure = failure ||
1742 i40e_alloc_rx_buffers_1buf(rx_ring,
1747 i = rx_ring->next_to_clean;
1748 rx_desc = I40E_RX_DESC(rx_ring, i);
1749 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1750 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1751 I40E_RXD_QW1_STATUS_SHIFT;
1753 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_DD_SHIFT)))
1756 /* This memory barrier is needed to keep us from reading
1757 * any other fields out of the rx_desc until we know the
1762 if (i40e_rx_is_programming_status(qword)) {
1763 i40e_clean_programming_status(rx_ring, rx_desc);
1764 I40E_RX_INCREMENT(rx_ring, i);
1767 rx_bi = &rx_ring->rx_bi[i];
1769 prefetch(skb->data);
1771 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
1772 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
1774 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1775 I40E_RXD_QW1_ERROR_SHIFT;
1776 rx_error &= ~BIT(I40E_RX_DESC_ERROR_HBO_SHIFT);
1778 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1779 I40E_RXD_QW1_PTYPE_SHIFT;
1783 /* Get the header and possibly the whole packet
1784 * If this is an skb from previous receive dma will be 0
1786 skb_put(skb, rx_packet_len);
1787 dma_unmap_single(rx_ring->dev, rx_bi->dma, rx_ring->rx_buf_len,
1791 I40E_RX_INCREMENT(rx_ring, i);
1794 !(rx_status & BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1795 rx_ring->rx_stats.non_eop_descs++;
1799 /* ERR_MASK will only have valid bits if EOP set */
1800 if (unlikely(rx_error & BIT(I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1801 dev_kfree_skb_any(skb);
1805 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1806 if (unlikely(rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK)) {
1807 i40e_ptp_rx_hwtstamp(vsi->back, skb, (rx_status &
1808 I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1809 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT);
1810 rx_ring->last_rx_timestamp = jiffies;
1813 /* probably a little skewed due to removing CRC */
1814 total_rx_bytes += skb->len;
1817 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1819 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1821 vlan_tag = rx_status & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1822 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1825 if (!i40e_fcoe_handle_offload(rx_ring, rx_desc, skb)) {
1826 dev_kfree_skb_any(skb);
1830 i40e_receive_skb(rx_ring, skb, vlan_tag);
1832 rx_desc->wb.qword1.status_error_len = 0;
1833 } while (likely(total_rx_packets < budget));
1835 u64_stats_update_begin(&rx_ring->syncp);
1836 rx_ring->stats.packets += total_rx_packets;
1837 rx_ring->stats.bytes += total_rx_bytes;
1838 u64_stats_update_end(&rx_ring->syncp);
1839 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1840 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1842 return failure ? budget : total_rx_packets;
1845 static u32 i40e_buildreg_itr(const int type, const u16 itr)
1849 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1850 /* Don't clear PBA because that can cause lost interrupts that
1851 * came in while we were cleaning/polling
1853 (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
1854 (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
1859 /* a small macro to shorten up some long lines */
1860 #define INTREG I40E_PFINT_DYN_CTLN
1863 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
1864 * @vsi: the VSI we care about
1865 * @q_vector: q_vector for which itr is being updated and interrupt enabled
1868 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
1869 struct i40e_q_vector *q_vector)
1871 struct i40e_hw *hw = &vsi->back->hw;
1872 bool rx = false, tx = false;
1876 vector = (q_vector->v_idx + vsi->base_vector);
1878 /* avoid dynamic calculation if in countdown mode OR if
1879 * all dynamic is disabled
1881 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
1883 if (q_vector->itr_countdown > 0 ||
1884 (!ITR_IS_DYNAMIC(vsi->rx_itr_setting) &&
1885 !ITR_IS_DYNAMIC(vsi->tx_itr_setting))) {
1889 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting)) {
1890 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
1891 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
1894 if (ITR_IS_DYNAMIC(vsi->tx_itr_setting)) {
1895 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
1896 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
1900 /* get the higher of the two ITR adjustments and
1901 * use the same value for both ITR registers
1902 * when in adaptive mode (Rx and/or Tx)
1904 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
1906 q_vector->tx.itr = q_vector->rx.itr = itr;
1907 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
1909 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
1913 /* only need to enable the interrupt once, but need
1914 * to possibly update both ITR values
1917 /* set the INTENA_MSK_MASK so that this first write
1918 * won't actually enable the interrupt, instead just
1919 * updating the ITR (it's bit 31 PF and VF)
1922 /* don't check _DOWN because interrupt isn't being enabled */
1923 wr32(hw, INTREG(vector - 1), rxval);
1927 if (!test_bit(__I40E_DOWN, &vsi->state))
1928 wr32(hw, INTREG(vector - 1), txval);
1930 if (q_vector->itr_countdown)
1931 q_vector->itr_countdown--;
1933 q_vector->itr_countdown = ITR_COUNTDOWN_START;
1937 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1938 * @napi: napi struct with our devices info in it
1939 * @budget: amount of work driver is allowed to do this pass, in packets
1941 * This function will clean all queues associated with a q_vector.
1943 * Returns the amount of work done
1945 int i40e_napi_poll(struct napi_struct *napi, int budget)
1947 struct i40e_q_vector *q_vector =
1948 container_of(napi, struct i40e_q_vector, napi);
1949 struct i40e_vsi *vsi = q_vector->vsi;
1950 struct i40e_ring *ring;
1951 bool clean_complete = true;
1952 bool arm_wb = false;
1953 int budget_per_ring;
1956 if (test_bit(__I40E_DOWN, &vsi->state)) {
1957 napi_complete(napi);
1961 /* Clear hung_detected bit */
1962 clear_bit(I40E_Q_VECTOR_HUNG_DETECT, &q_vector->hung_detected);
1963 /* Since the actual Tx work is minimal, we can give the Tx a larger
1964 * budget and be more aggressive about cleaning up the Tx descriptors.
1966 i40e_for_each_ring(ring, q_vector->tx) {
1967 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1968 arm_wb = arm_wb || ring->arm_wb;
1969 ring->arm_wb = false;
1972 /* Handle case where we are called by netpoll with a budget of 0 */
1976 /* We attempt to distribute budget to each Rx queue fairly, but don't
1977 * allow the budget to go below 1 because that would exit polling early.
1979 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1981 i40e_for_each_ring(ring, q_vector->rx) {
1984 if (ring_is_ps_enabled(ring))
1985 cleaned = i40e_clean_rx_irq_ps(ring, budget_per_ring);
1987 cleaned = i40e_clean_rx_irq_1buf(ring, budget_per_ring);
1989 work_done += cleaned;
1990 /* if we didn't clean as many as budgeted, we must be done */
1991 clean_complete &= (budget_per_ring != cleaned);
1994 /* If work not completed, return budget and polling will return */
1995 if (!clean_complete) {
1998 q_vector->tx.ring[0].tx_stats.tx_force_wb++;
1999 i40e_enable_wb_on_itr(vsi, q_vector);
2004 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2005 q_vector->arm_wb_state = false;
2007 /* Work is done so exit the polling mode and re-enable the interrupt */
2008 napi_complete_done(napi, work_done);
2009 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
2010 i40e_update_enable_itr(vsi, q_vector);
2011 } else { /* Legacy mode */
2012 struct i40e_hw *hw = &vsi->back->hw;
2013 /* We re-enable the queue 0 cause, but
2014 * don't worry about dynamic_enable
2015 * because we left it on for the other
2016 * possible interrupts during napi
2018 u32 qval = rd32(hw, I40E_QINT_RQCTL(0)) |
2019 I40E_QINT_RQCTL_CAUSE_ENA_MASK;
2021 wr32(hw, I40E_QINT_RQCTL(0), qval);
2022 qval = rd32(hw, I40E_QINT_TQCTL(0)) |
2023 I40E_QINT_TQCTL_CAUSE_ENA_MASK;
2024 wr32(hw, I40E_QINT_TQCTL(0), qval);
2025 i40e_irq_dynamic_enable_icr0(vsi->back, false);
2031 * i40e_atr - Add a Flow Director ATR filter
2032 * @tx_ring: ring to add programming descriptor to
2034 * @tx_flags: send tx flags
2035 * @protocol: wire protocol
2037 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2038 u32 tx_flags, __be16 protocol)
2040 struct i40e_filter_program_desc *fdir_desc;
2041 struct i40e_pf *pf = tx_ring->vsi->back;
2043 unsigned char *network;
2045 struct ipv6hdr *ipv6;
2049 u32 flex_ptype, dtype_cmd;
2052 /* make sure ATR is enabled */
2053 if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2056 if ((pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2059 /* if sampling is disabled do nothing */
2060 if (!tx_ring->atr_sample_rate)
2063 if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2066 if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL)) {
2067 /* snag network header to get L4 type and address */
2068 hdr.network = skb_network_header(skb);
2070 /* Currently only IPv4/IPv6 with TCP is supported
2071 * access ihl as u8 to avoid unaligned access on ia64
2073 if (tx_flags & I40E_TX_FLAGS_IPV4)
2074 hlen = (hdr.network[0] & 0x0F) << 2;
2075 else if (protocol == htons(ETH_P_IPV6))
2076 hlen = sizeof(struct ipv6hdr);
2080 hdr.network = skb_inner_network_header(skb);
2081 hlen = skb_inner_network_header_len(skb);
2084 /* Currently only IPv4/IPv6 with TCP is supported
2085 * Note: tx_flags gets modified to reflect inner protocols in
2086 * tx_enable_csum function if encap is enabled.
2088 if ((tx_flags & I40E_TX_FLAGS_IPV4) &&
2089 (hdr.ipv4->protocol != IPPROTO_TCP))
2091 else if ((tx_flags & I40E_TX_FLAGS_IPV6) &&
2092 (hdr.ipv6->nexthdr != IPPROTO_TCP))
2095 th = (struct tcphdr *)(hdr.network + hlen);
2097 /* Due to lack of space, no more new filters can be programmed */
2098 if (th->syn && (pf->auto_disable_flags & I40E_FLAG_FD_ATR_ENABLED))
2100 if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2101 (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE))) {
2102 /* HW ATR eviction will take care of removing filters on FIN
2105 if (th->fin || th->rst)
2109 tx_ring->atr_count++;
2111 /* sample on all syn/fin/rst packets or once every atr sample rate */
2115 (tx_ring->atr_count < tx_ring->atr_sample_rate))
2118 tx_ring->atr_count = 0;
2120 /* grab the next descriptor */
2121 i = tx_ring->next_to_use;
2122 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2125 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2127 flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2128 I40E_TXD_FLTR_QW0_QINDEX_MASK;
2129 flex_ptype |= (protocol == htons(ETH_P_IP)) ?
2130 (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2131 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2132 (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2133 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2135 flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2137 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2139 dtype_cmd |= (th->fin || th->rst) ?
2140 (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2141 I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2142 (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2143 I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2145 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2146 I40E_TXD_FLTR_QW1_DEST_SHIFT;
2148 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2149 I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2151 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2152 if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2154 ((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2155 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2156 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2159 ((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2160 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2161 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2163 if ((pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) &&
2164 (!(pf->auto_disable_flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)))
2165 dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2167 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2168 fdir_desc->rsvd = cpu_to_le32(0);
2169 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2170 fdir_desc->fd_id = cpu_to_le32(0);
2174 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2176 * @tx_ring: ring to send buffer on
2177 * @flags: the tx flags to be set
2179 * Checks the skb and set up correspondingly several generic transmit flags
2180 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2182 * Returns error code indicate the frame should be dropped upon error and the
2183 * otherwise returns 0 to indicate the flags has been set properly.
2186 inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2187 struct i40e_ring *tx_ring,
2190 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2191 struct i40e_ring *tx_ring,
2195 __be16 protocol = skb->protocol;
2198 if (protocol == htons(ETH_P_8021Q) &&
2199 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2200 /* When HW VLAN acceleration is turned off by the user the
2201 * stack sets the protocol to 8021q so that the driver
2202 * can take any steps required to support the SW only
2203 * VLAN handling. In our case the driver doesn't need
2204 * to take any further steps so just set the protocol
2205 * to the encapsulated ethertype.
2207 skb->protocol = vlan_get_protocol(skb);
2211 /* if we have a HW VLAN tag being added, default to the HW one */
2212 if (skb_vlan_tag_present(skb)) {
2213 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2214 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2215 /* else if it is a SW VLAN, check the next protocol and store the tag */
2216 } else if (protocol == htons(ETH_P_8021Q)) {
2217 struct vlan_hdr *vhdr, _vhdr;
2219 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2223 protocol = vhdr->h_vlan_encapsulated_proto;
2224 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2225 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2228 if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2231 /* Insert 802.1p priority into VLAN header */
2232 if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2233 (skb->priority != TC_PRIO_CONTROL)) {
2234 tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2235 tx_flags |= (skb->priority & 0x7) <<
2236 I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2237 if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2238 struct vlan_ethhdr *vhdr;
2241 rc = skb_cow_head(skb, 0);
2244 vhdr = (struct vlan_ethhdr *)skb->data;
2245 vhdr->h_vlan_TCI = htons(tx_flags >>
2246 I40E_TX_FLAGS_VLAN_SHIFT);
2248 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2258 * i40e_tso - set up the tso context descriptor
2259 * @tx_ring: ptr to the ring to send
2260 * @skb: ptr to the skb we're sending
2261 * @hdr_len: ptr to the size of the packet header
2262 * @cd_type_cmd_tso_mss: Quad Word 1
2264 * Returns 0 if no TSO can happen, 1 if tso is going, or error
2266 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
2267 u8 *hdr_len, u64 *cd_type_cmd_tso_mss)
2269 u32 cd_cmd, cd_tso_len, cd_mss;
2270 struct ipv6hdr *ipv6h;
2271 struct tcphdr *tcph;
2276 if (skb->ip_summed != CHECKSUM_PARTIAL)
2279 if (!skb_is_gso(skb))
2282 err = skb_cow_head(skb, 0);
2286 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
2287 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
2289 if (iph->version == 4) {
2290 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
2293 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2295 } else if (ipv6h->version == 6) {
2296 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
2297 ipv6h->payload_len = 0;
2298 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2302 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
2303 *hdr_len = (skb->encapsulation
2304 ? (skb_inner_transport_header(skb) - skb->data)
2305 : skb_transport_offset(skb)) + l4len;
2307 /* find the field values */
2308 cd_cmd = I40E_TX_CTX_DESC_TSO;
2309 cd_tso_len = skb->len - *hdr_len;
2310 cd_mss = skb_shinfo(skb)->gso_size;
2311 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2313 I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2314 ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2319 * i40e_tsyn - set up the tsyn context descriptor
2320 * @tx_ring: ptr to the ring to send
2321 * @skb: ptr to the skb we're sending
2322 * @tx_flags: the collected send information
2323 * @cd_type_cmd_tso_mss: Quad Word 1
2325 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2327 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2328 u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2332 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2335 /* Tx timestamps cannot be sampled when doing TSO */
2336 if (tx_flags & I40E_TX_FLAGS_TSO)
2339 /* only timestamp the outbound packet if the user has requested it and
2340 * we are not already transmitting a packet to be timestamped
2342 pf = i40e_netdev_to_pf(tx_ring->netdev);
2343 if (!(pf->flags & I40E_FLAG_PTP))
2347 !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, &pf->state)) {
2348 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2349 pf->ptp_tx_skb = skb_get(skb);
2354 *cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2355 I40E_TXD_CTX_QW1_CMD_SHIFT;
2361 * i40e_tx_enable_csum - Enable Tx checksum offloads
2363 * @tx_flags: pointer to Tx flags currently set
2364 * @td_cmd: Tx descriptor command bits to set
2365 * @td_offset: Tx descriptor header offsets to set
2366 * @tx_ring: Tx descriptor ring
2367 * @cd_tunneling: ptr to context desc bits
2369 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2370 u32 *td_cmd, u32 *td_offset,
2371 struct i40e_ring *tx_ring,
2374 struct ipv6hdr *this_ipv6_hdr;
2375 unsigned int this_tcp_hdrlen;
2376 struct iphdr *this_ip_hdr;
2377 u32 network_hdr_len;
2379 struct udphdr *oudph = NULL;
2380 struct iphdr *oiph = NULL;
2383 if (skb->encapsulation) {
2384 switch (ip_hdr(skb)->protocol) {
2386 oudph = udp_hdr(skb);
2388 l4_tunnel = I40E_TXD_CTX_UDP_TUNNELING;
2389 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2392 l4_tunnel = I40E_TXD_CTX_GRE_TUNNELING;
2397 network_hdr_len = skb_inner_network_header_len(skb);
2398 this_ip_hdr = inner_ip_hdr(skb);
2399 this_ipv6_hdr = inner_ipv6_hdr(skb);
2400 this_tcp_hdrlen = inner_tcp_hdrlen(skb);
2402 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2403 if (*tx_flags & I40E_TX_FLAGS_TSO) {
2404 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
2405 ip_hdr(skb)->check = 0;
2408 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2410 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2411 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
2412 if (*tx_flags & I40E_TX_FLAGS_TSO)
2413 ip_hdr(skb)->check = 0;
2416 /* Now set the ctx descriptor fields */
2417 *cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
2418 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
2420 ((skb_inner_network_offset(skb) -
2421 skb_transport_offset(skb)) >> 1) <<
2422 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2423 if (this_ip_hdr->version == 6) {
2424 *tx_flags &= ~I40E_TX_FLAGS_IPV4;
2425 *tx_flags |= I40E_TX_FLAGS_IPV6;
2427 if ((tx_ring->flags & I40E_TXR_FLAGS_OUTER_UDP_CSUM) &&
2428 (l4_tunnel == I40E_TXD_CTX_UDP_TUNNELING) &&
2429 (*cd_tunneling & I40E_TXD_CTX_QW0_EXT_IP_MASK)) {
2430 oudph->check = ~csum_tcpudp_magic(oiph->saddr,
2432 (skb->len - skb_transport_offset(skb)),
2434 *cd_tunneling |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2437 network_hdr_len = skb_network_header_len(skb);
2438 this_ip_hdr = ip_hdr(skb);
2439 this_ipv6_hdr = ipv6_hdr(skb);
2440 this_tcp_hdrlen = tcp_hdrlen(skb);
2443 /* Enable IP checksum offloads */
2444 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2445 l4_hdr = this_ip_hdr->protocol;
2446 /* the stack computes the IP header already, the only time we
2447 * need the hardware to recompute it is in the case of TSO.
2449 if (*tx_flags & I40E_TX_FLAGS_TSO) {
2450 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
2451 this_ip_hdr->check = 0;
2453 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
2455 /* Now set the td_offset for IP header length */
2456 *td_offset = (network_hdr_len >> 2) <<
2457 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2458 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2459 l4_hdr = this_ipv6_hdr->nexthdr;
2460 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2461 /* Now set the td_offset for IP header length */
2462 *td_offset = (network_hdr_len >> 2) <<
2463 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2465 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
2466 *td_offset |= (skb_network_offset(skb) >> 1) <<
2467 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2469 /* Enable L4 checksum offloads */
2472 /* enable checksum offloads */
2473 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2474 *td_offset |= (this_tcp_hdrlen >> 2) <<
2475 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2478 /* enable SCTP checksum offload */
2479 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2480 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
2481 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2484 /* enable UDP checksum offload */
2485 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2486 *td_offset |= (sizeof(struct udphdr) >> 2) <<
2487 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2495 * i40e_create_tx_ctx Build the Tx context descriptor
2496 * @tx_ring: ring to create the descriptor on
2497 * @cd_type_cmd_tso_mss: Quad Word 1
2498 * @cd_tunneling: Quad Word 0 - bits 0-31
2499 * @cd_l2tag2: Quad Word 0 - bits 32-63
2501 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2502 const u64 cd_type_cmd_tso_mss,
2503 const u32 cd_tunneling, const u32 cd_l2tag2)
2505 struct i40e_tx_context_desc *context_desc;
2506 int i = tx_ring->next_to_use;
2508 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2509 !cd_tunneling && !cd_l2tag2)
2512 /* grab the next descriptor */
2513 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2516 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2518 /* cpu_to_le32 and assign to struct fields */
2519 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2520 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2521 context_desc->rsvd = cpu_to_le16(0);
2522 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2526 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2527 * @tx_ring: the ring to be checked
2528 * @size: the size buffer we want to assure is available
2530 * Returns -EBUSY if a stop is needed, else 0
2532 static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2534 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2535 /* Memory barrier before checking head and tail */
2538 /* Check again in a case another CPU has just made room available. */
2539 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2542 /* A reprieve! - use start_queue because it doesn't call schedule */
2543 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2544 ++tx_ring->tx_stats.restart_queue;
2549 * i40e_maybe_stop_tx - 1st level check for tx stop conditions
2550 * @tx_ring: the ring to be checked
2551 * @size: the size buffer we want to assure is available
2553 * Returns 0 if stop is not needed
2556 inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2558 static inline int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2561 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
2563 return __i40e_maybe_stop_tx(tx_ring, size);
2567 * i40e_chk_linearize - Check if there are more than 8 fragments per packet
2569 * @tx_flags: collected send information
2571 * Note: Our HW can't scatter-gather more than 8 fragments to build
2572 * a packet on the wire and so we need to figure out the cases where we
2573 * need to linearize the skb.
2575 static bool i40e_chk_linearize(struct sk_buff *skb, u32 tx_flags)
2577 struct skb_frag_struct *frag;
2578 bool linearize = false;
2579 unsigned int size = 0;
2583 num_frags = skb_shinfo(skb)->nr_frags;
2584 gso_segs = skb_shinfo(skb)->gso_segs;
2586 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO)) {
2589 if (num_frags < (I40E_MAX_BUFFER_TXD))
2590 goto linearize_chk_done;
2591 /* try the simple math, if we have too many frags per segment */
2592 if (DIV_ROUND_UP((num_frags + gso_segs), gso_segs) >
2593 I40E_MAX_BUFFER_TXD) {
2595 goto linearize_chk_done;
2597 frag = &skb_shinfo(skb)->frags[0];
2598 /* we might still have more fragments per segment */
2600 size += skb_frag_size(frag);
2602 if ((size >= skb_shinfo(skb)->gso_size) &&
2603 (j < I40E_MAX_BUFFER_TXD)) {
2604 size = (size % skb_shinfo(skb)->gso_size);
2607 if (j == I40E_MAX_BUFFER_TXD) {
2612 } while (num_frags);
2614 if (num_frags >= I40E_MAX_BUFFER_TXD)
2623 * i40e_tx_map - Build the Tx descriptor
2624 * @tx_ring: ring to send buffer on
2626 * @first: first buffer info buffer to use
2627 * @tx_flags: collected send information
2628 * @hdr_len: size of the packet header
2629 * @td_cmd: the command field in the descriptor
2630 * @td_offset: offset for checksum or crc
2633 inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2634 struct i40e_tx_buffer *first, u32 tx_flags,
2635 const u8 hdr_len, u32 td_cmd, u32 td_offset)
2637 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2638 struct i40e_tx_buffer *first, u32 tx_flags,
2639 const u8 hdr_len, u32 td_cmd, u32 td_offset)
2642 unsigned int data_len = skb->data_len;
2643 unsigned int size = skb_headlen(skb);
2644 struct skb_frag_struct *frag;
2645 struct i40e_tx_buffer *tx_bi;
2646 struct i40e_tx_desc *tx_desc;
2647 u16 i = tx_ring->next_to_use;
2652 bool tail_bump = true;
2655 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2656 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2657 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2658 I40E_TX_FLAGS_VLAN_SHIFT;
2661 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
2662 gso_segs = skb_shinfo(skb)->gso_segs;
2666 /* multiply data chunks by size of headers */
2667 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
2668 first->gso_segs = gso_segs;
2670 first->tx_flags = tx_flags;
2672 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2674 tx_desc = I40E_TX_DESC(tx_ring, i);
2677 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2678 if (dma_mapping_error(tx_ring->dev, dma))
2681 /* record length, and DMA address */
2682 dma_unmap_len_set(tx_bi, len, size);
2683 dma_unmap_addr_set(tx_bi, dma, dma);
2685 tx_desc->buffer_addr = cpu_to_le64(dma);
2687 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2688 tx_desc->cmd_type_offset_bsz =
2689 build_ctob(td_cmd, td_offset,
2690 I40E_MAX_DATA_PER_TXD, td_tag);
2696 if (i == tx_ring->count) {
2697 tx_desc = I40E_TX_DESC(tx_ring, 0);
2701 dma += I40E_MAX_DATA_PER_TXD;
2702 size -= I40E_MAX_DATA_PER_TXD;
2704 tx_desc->buffer_addr = cpu_to_le64(dma);
2707 if (likely(!data_len))
2710 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
2717 if (i == tx_ring->count) {
2718 tx_desc = I40E_TX_DESC(tx_ring, 0);
2722 size = skb_frag_size(frag);
2725 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
2728 tx_bi = &tx_ring->tx_bi[i];
2731 /* set next_to_watch value indicating a packet is present */
2732 first->next_to_watch = tx_desc;
2735 if (i == tx_ring->count)
2738 tx_ring->next_to_use = i;
2740 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
2741 tx_ring->queue_index),
2743 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
2745 /* Algorithm to optimize tail and RS bit setting:
2746 * if xmit_more is supported
2747 * if xmit_more is true
2748 * do not update tail and do not mark RS bit.
2749 * if xmit_more is false and last xmit_more was false
2750 * if every packet spanned less than 4 desc
2751 * then set RS bit on 4th packet and update tail
2754 * update tail and set RS bit on every packet.
2755 * if xmit_more is false and last_xmit_more was true
2756 * update tail and set RS bit.
2758 * Optimization: wmb to be issued only in case of tail update.
2759 * Also optimize the Descriptor WB path for RS bit with the same
2762 * Note: If there are less than 4 packets
2763 * pending and interrupts were disabled the service task will
2764 * trigger a force WB.
2766 if (skb->xmit_more &&
2767 !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2768 tx_ring->queue_index))) {
2769 tx_ring->flags |= I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2771 } else if (!skb->xmit_more &&
2772 !netif_xmit_stopped(netdev_get_tx_queue(tx_ring->netdev,
2773 tx_ring->queue_index)) &&
2774 (!(tx_ring->flags & I40E_TXR_FLAGS_LAST_XMIT_MORE_SET)) &&
2775 (tx_ring->packet_stride < WB_STRIDE) &&
2776 (desc_count < WB_STRIDE)) {
2777 tx_ring->packet_stride++;
2779 tx_ring->packet_stride = 0;
2780 tx_ring->flags &= ~I40E_TXR_FLAGS_LAST_XMIT_MORE_SET;
2784 tx_ring->packet_stride = 0;
2786 tx_desc->cmd_type_offset_bsz =
2787 build_ctob(td_cmd, td_offset, size, td_tag) |
2788 cpu_to_le64((u64)(do_rs ? I40E_TXD_CMD :
2789 I40E_TX_DESC_CMD_EOP) <<
2790 I40E_TXD_QW1_CMD_SHIFT);
2792 /* notify HW of packet */
2794 prefetchw(tx_desc + 1);
2797 /* Force memory writes to complete before letting h/w
2798 * know there are new descriptors to fetch. (Only
2799 * applicable for weak-ordered memory model archs,
2803 writel(i, tx_ring->tail);
2809 dev_info(tx_ring->dev, "TX DMA map failed\n");
2811 /* clear dma mappings for failed tx_bi map */
2813 tx_bi = &tx_ring->tx_bi[i];
2814 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
2822 tx_ring->next_to_use = i;
2826 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
2828 * @tx_ring: ring to send buffer on
2830 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
2831 * there is not enough descriptors available in this ring since we need at least
2835 inline int i40e_xmit_descriptor_count(struct sk_buff *skb,
2836 struct i40e_ring *tx_ring)
2838 static inline int i40e_xmit_descriptor_count(struct sk_buff *skb,
2839 struct i40e_ring *tx_ring)
2845 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
2846 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
2847 * + 4 desc gap to avoid the cache line where head is,
2848 * + 1 desc for context descriptor,
2849 * otherwise try next time
2851 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
2852 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
2854 count += TXD_USE_COUNT(skb_headlen(skb));
2855 if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
2856 tx_ring->tx_stats.tx_busy++;
2863 * i40e_xmit_frame_ring - Sends buffer on Tx ring
2865 * @tx_ring: ring to send buffer on
2867 * Returns NETDEV_TX_OK if sent, else an error code
2869 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
2870 struct i40e_ring *tx_ring)
2872 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
2873 u32 cd_tunneling = 0, cd_l2tag2 = 0;
2874 struct i40e_tx_buffer *first;
2883 /* prefetch the data, we'll need it later */
2884 prefetch(skb->data);
2886 if (0 == i40e_xmit_descriptor_count(skb, tx_ring))
2887 return NETDEV_TX_BUSY;
2889 /* prepare the xmit flags */
2890 if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
2893 /* obtain protocol of skb */
2894 protocol = vlan_get_protocol(skb);
2896 /* record the location of the first descriptor for this packet */
2897 first = &tx_ring->tx_bi[tx_ring->next_to_use];
2899 /* setup IPv4/IPv6 offloads */
2900 if (protocol == htons(ETH_P_IP))
2901 tx_flags |= I40E_TX_FLAGS_IPV4;
2902 else if (protocol == htons(ETH_P_IPV6))
2903 tx_flags |= I40E_TX_FLAGS_IPV6;
2905 tso = i40e_tso(tx_ring, skb, &hdr_len, &cd_type_cmd_tso_mss);
2910 tx_flags |= I40E_TX_FLAGS_TSO;
2912 tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
2915 tx_flags |= I40E_TX_FLAGS_TSYN;
2917 if (i40e_chk_linearize(skb, tx_flags)) {
2918 if (skb_linearize(skb))
2920 tx_ring->tx_stats.tx_linearize++;
2922 skb_tx_timestamp(skb);
2924 /* always enable CRC insertion offload */
2925 td_cmd |= I40E_TX_DESC_CMD_ICRC;
2927 /* Always offload the checksum, since it's in the data descriptor */
2928 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2929 tx_flags |= I40E_TX_FLAGS_CSUM;
2931 i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
2932 tx_ring, &cd_tunneling);
2935 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
2936 cd_tunneling, cd_l2tag2);
2938 /* Add Flow Director ATR if it's enabled.
2940 * NOTE: this must always be directly before the data descriptor.
2942 i40e_atr(tx_ring, skb, tx_flags, protocol);
2944 i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
2947 return NETDEV_TX_OK;
2950 dev_kfree_skb_any(skb);
2951 return NETDEV_TX_OK;
2955 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
2957 * @netdev: network interface device structure
2959 * Returns NETDEV_TX_OK if sent, else an error code
2961 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2963 struct i40e_netdev_priv *np = netdev_priv(netdev);
2964 struct i40e_vsi *vsi = np->vsi;
2965 struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
2967 /* hardware can't handle really short frames, hardware padding works
2970 if (skb_put_padto(skb, I40E_MIN_TX_LEN))
2971 return NETDEV_TX_OK;
2973 return i40e_xmit_frame_ring(skb, tx_ring);