1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2013 - 2018 Intel Corporation. */
4 #include <linux/prefetch.h>
5 #include <linux/bpf_trace.h>
8 #include "i40e_trace.h"
9 #include "i40e_prototype.h"
10 #include "i40e_txrx_common.h"
13 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
15 * i40e_fdir - Generate a Flow Director descriptor based on fdata
16 * @tx_ring: Tx ring to send buffer on
17 * @fdata: Flow director filter data
18 * @add: Indicate if we are adding a rule or deleting one
21 static void i40e_fdir(struct i40e_ring *tx_ring,
22 struct i40e_fdir_filter *fdata, bool add)
24 struct i40e_filter_program_desc *fdir_desc;
25 struct i40e_pf *pf = tx_ring->vsi->back;
26 u32 flex_ptype, dtype_cmd;
29 /* grab the next descriptor */
30 i = tx_ring->next_to_use;
31 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
34 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
36 flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
37 (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
39 flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
40 (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
42 flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
43 (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
45 /* Use LAN VSI Id if not programmed by user */
46 flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
47 ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
48 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
50 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
53 I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
54 I40E_TXD_FLTR_QW1_PCMD_SHIFT :
55 I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
56 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
58 dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
59 (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
61 dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
62 (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
64 if (fdata->cnt_index) {
65 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
66 dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
67 ((u32)fdata->cnt_index <<
68 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
71 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
72 fdir_desc->rsvd = cpu_to_le32(0);
73 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
74 fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
77 #define I40E_FD_CLEAN_DELAY 10
79 * i40e_program_fdir_filter - Program a Flow Director filter
80 * @fdir_data: Packet data that will be filter parameters
81 * @raw_packet: the pre-allocated packet buffer for FDir
83 * @add: True for add/update, False for remove
85 static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
86 u8 *raw_packet, struct i40e_pf *pf,
89 struct i40e_tx_buffer *tx_buf, *first;
90 struct i40e_tx_desc *tx_desc;
91 struct i40e_ring *tx_ring;
98 /* find existing FDIR VSI */
99 vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
103 tx_ring = vsi->tx_rings[0];
106 /* we need two descriptors to add/del a filter and we can wait */
107 for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
110 msleep_interruptible(1);
113 dma = dma_map_single(dev, raw_packet,
114 I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
115 if (dma_mapping_error(dev, dma))
118 /* grab the next descriptor */
119 i = tx_ring->next_to_use;
120 first = &tx_ring->tx_bi[i];
121 i40e_fdir(tx_ring, fdir_data, add);
123 /* Now program a dummy descriptor */
124 i = tx_ring->next_to_use;
125 tx_desc = I40E_TX_DESC(tx_ring, i);
126 tx_buf = &tx_ring->tx_bi[i];
128 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
130 memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
132 /* record length, and DMA address */
133 dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
134 dma_unmap_addr_set(tx_buf, dma, dma);
136 tx_desc->buffer_addr = cpu_to_le64(dma);
137 td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
139 tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
140 tx_buf->raw_buf = (void *)raw_packet;
142 tx_desc->cmd_type_offset_bsz =
143 build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
145 /* Force memory writes to complete before letting h/w
146 * know there are new descriptors to fetch.
150 /* Mark the data descriptor to be watched */
151 first->next_to_watch = tx_desc;
153 writel(tx_ring->next_to_use, tx_ring->tail);
161 * i40e_create_dummy_packet - Constructs dummy packet for HW
162 * @dummy_packet: preallocated space for dummy packet
163 * @ipv4: is layer 3 packet of version 4 or 6
164 * @l4proto: next level protocol used in data portion of l3
167 * Returns address of layer 4 protocol dummy packet.
169 static char *i40e_create_dummy_packet(u8 *dummy_packet, bool ipv4, u8 l4proto,
170 struct i40e_fdir_filter *data)
172 bool is_vlan = !!data->vlan_tag;
173 struct vlan_hdr vlan;
180 eth.h_proto = cpu_to_be16(ETH_P_IP);
181 ip.protocol = l4proto;
185 ip.daddr = data->dst_ip;
186 ip.saddr = data->src_ip;
188 eth.h_proto = cpu_to_be16(ETH_P_IPV6);
189 ipv6.nexthdr = l4proto;
192 memcpy(&ipv6.saddr.in6_u.u6_addr32, data->src_ip6,
194 memcpy(&ipv6.daddr.in6_u.u6_addr32, data->dst_ip6,
199 vlan.h_vlan_TCI = data->vlan_tag;
200 vlan.h_vlan_encapsulated_proto = eth.h_proto;
201 eth.h_proto = data->vlan_etype;
205 memcpy(tmp, ð, sizeof(eth));
209 memcpy(tmp, &vlan, sizeof(vlan));
214 memcpy(tmp, &ip, sizeof(ip));
217 memcpy(tmp, &ipv6, sizeof(ipv6));
225 * i40e_create_dummy_udp_packet - helper function to create UDP packet
226 * @raw_packet: preallocated space for dummy packet
227 * @ipv4: is layer 3 packet of version 4 or 6
228 * @l4proto: next level protocol used in data portion of l3
231 * Helper function to populate udp fields.
233 static void i40e_create_dummy_udp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
234 struct i40e_fdir_filter *data)
239 tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_UDP, data);
240 udp = (struct udphdr *)(tmp);
241 udp->dest = data->dst_port;
242 udp->source = data->src_port;
246 * i40e_create_dummy_tcp_packet - helper function to create TCP packet
247 * @raw_packet: preallocated space for dummy packet
248 * @ipv4: is layer 3 packet of version 4 or 6
249 * @l4proto: next level protocol used in data portion of l3
252 * Helper function to populate tcp fields.
254 static void i40e_create_dummy_tcp_packet(u8 *raw_packet, bool ipv4, u8 l4proto,
255 struct i40e_fdir_filter *data)
259 /* Dummy tcp packet */
260 static const char tcp_packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
261 0x50, 0x11, 0x0, 0x72, 0, 0, 0, 0};
263 tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_TCP, data);
265 tcp = (struct tcphdr *)tmp;
266 memcpy(tcp, tcp_packet, sizeof(tcp_packet));
267 tcp->dest = data->dst_port;
268 tcp->source = data->src_port;
272 * i40e_create_dummy_sctp_packet - helper function to create SCTP packet
273 * @raw_packet: preallocated space for dummy packet
274 * @ipv4: is layer 3 packet of version 4 or 6
275 * @l4proto: next level protocol used in data portion of l3
278 * Helper function to populate sctp fields.
280 static void i40e_create_dummy_sctp_packet(u8 *raw_packet, bool ipv4,
282 struct i40e_fdir_filter *data)
284 struct sctphdr *sctp;
287 tmp = i40e_create_dummy_packet(raw_packet, ipv4, IPPROTO_SCTP, data);
289 sctp = (struct sctphdr *)tmp;
290 sctp->dest = data->dst_port;
291 sctp->source = data->src_port;
295 * i40e_prepare_fdir_filter - Prepare and program fdir filter
296 * @pf: physical function to attach filter to
297 * @fd_data: filter data
298 * @add: add or delete filter
299 * @packet_addr: address of dummy packet, used in filtering
300 * @payload_offset: offset from dummy packet address to user defined data
301 * @pctype: Packet type for which filter is used
303 * Helper function to offset data of dummy packet, program it and
306 static int i40e_prepare_fdir_filter(struct i40e_pf *pf,
307 struct i40e_fdir_filter *fd_data,
308 bool add, char *packet_addr,
309 int payload_offset, u8 pctype)
313 if (fd_data->flex_filter) {
315 __be16 pattern = fd_data->flex_word;
316 u16 off = fd_data->flex_offset;
318 payload = packet_addr + payload_offset;
320 /* If user provided vlan, offset payload by vlan header length */
321 if (!!fd_data->vlan_tag)
322 payload += VLAN_HLEN;
324 *((__force __be16 *)(payload + off)) = pattern;
327 fd_data->pctype = pctype;
328 ret = i40e_program_fdir_filter(fd_data, packet_addr, pf, add);
330 dev_info(&pf->pdev->dev,
331 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
332 fd_data->pctype, fd_data->fd_id, ret);
333 /* Free the packet buffer since it wasn't added to the ring */
335 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
337 dev_info(&pf->pdev->dev,
338 "Filter OK for PCTYPE %d loc = %d\n",
339 fd_data->pctype, fd_data->fd_id);
341 dev_info(&pf->pdev->dev,
342 "Filter deleted for PCTYPE %d loc = %d\n",
343 fd_data->pctype, fd_data->fd_id);
350 * i40e_change_filter_num - Prepare and program fdir filter
351 * @ipv4: is layer 3 packet of version 4 or 6
352 * @add: add or delete filter
353 * @ipv4_filter_num: field to update
354 * @ipv6_filter_num: field to update
356 * Update filter number field for pf.
358 static void i40e_change_filter_num(bool ipv4, bool add, u16 *ipv4_filter_num,
359 u16 *ipv6_filter_num)
363 (*ipv4_filter_num)++;
365 (*ipv6_filter_num)++;
368 (*ipv4_filter_num)--;
370 (*ipv6_filter_num)--;
374 #define IP_HEADER_OFFSET 14
375 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
376 #define I40E_UDPIP6_DUMMY_PACKET_LEN 62
378 * i40e_add_del_fdir_udp - Add/Remove UDP filters
379 * @vsi: pointer to the targeted VSI
380 * @fd_data: the flow director data required for the FDir descriptor
381 * @add: true adds a filter, false removes it
382 * @ipv4: true is v4, false is v6
384 * Returns 0 if the filters were successfully added or removed
386 static int i40e_add_del_fdir_udp(struct i40e_vsi *vsi,
387 struct i40e_fdir_filter *fd_data,
391 struct i40e_pf *pf = vsi->back;
395 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
399 i40e_create_dummy_udp_packet(raw_packet, ipv4, IPPROTO_UDP, fd_data);
402 ret = i40e_prepare_fdir_filter
403 (pf, fd_data, add, raw_packet,
404 I40E_UDPIP_DUMMY_PACKET_LEN,
405 I40E_FILTER_PCTYPE_NONF_IPV4_UDP);
407 ret = i40e_prepare_fdir_filter
408 (pf, fd_data, add, raw_packet,
409 I40E_UDPIP6_DUMMY_PACKET_LEN,
410 I40E_FILTER_PCTYPE_NONF_IPV6_UDP);
417 i40e_change_filter_num(ipv4, add, &pf->fd_udp4_filter_cnt,
418 &pf->fd_udp6_filter_cnt);
423 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
424 #define I40E_TCPIP6_DUMMY_PACKET_LEN 74
426 * i40e_add_del_fdir_tcp - Add/Remove TCPv4 filters
427 * @vsi: pointer to the targeted VSI
428 * @fd_data: the flow director data required for the FDir descriptor
429 * @add: true adds a filter, false removes it
430 * @ipv4: true is v4, false is v6
432 * Returns 0 if the filters were successfully added or removed
434 static int i40e_add_del_fdir_tcp(struct i40e_vsi *vsi,
435 struct i40e_fdir_filter *fd_data,
439 struct i40e_pf *pf = vsi->back;
443 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
447 i40e_create_dummy_tcp_packet(raw_packet, ipv4, IPPROTO_TCP, fd_data);
449 ret = i40e_prepare_fdir_filter
450 (pf, fd_data, add, raw_packet,
451 I40E_TCPIP_DUMMY_PACKET_LEN,
452 I40E_FILTER_PCTYPE_NONF_IPV4_TCP);
454 ret = i40e_prepare_fdir_filter
455 (pf, fd_data, add, raw_packet,
456 I40E_TCPIP6_DUMMY_PACKET_LEN,
457 I40E_FILTER_PCTYPE_NONF_IPV6_TCP);
464 i40e_change_filter_num(ipv4, add, &pf->fd_tcp4_filter_cnt,
465 &pf->fd_tcp6_filter_cnt);
468 if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
469 I40E_DEBUG_FD & pf->hw.debug_mask)
470 dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
471 set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
476 #define I40E_SCTPIP_DUMMY_PACKET_LEN 46
477 #define I40E_SCTPIP6_DUMMY_PACKET_LEN 66
479 * i40e_add_del_fdir_sctp - Add/Remove SCTPv4 Flow Director filters for
480 * a specific flow spec
481 * @vsi: pointer to the targeted VSI
482 * @fd_data: the flow director data required for the FDir descriptor
483 * @add: true adds a filter, false removes it
484 * @ipv4: true is v4, false is v6
486 * Returns 0 if the filters were successfully added or removed
488 static int i40e_add_del_fdir_sctp(struct i40e_vsi *vsi,
489 struct i40e_fdir_filter *fd_data,
493 struct i40e_pf *pf = vsi->back;
497 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
501 i40e_create_dummy_sctp_packet(raw_packet, ipv4, IPPROTO_SCTP, fd_data);
504 ret = i40e_prepare_fdir_filter
505 (pf, fd_data, add, raw_packet,
506 I40E_SCTPIP_DUMMY_PACKET_LEN,
507 I40E_FILTER_PCTYPE_NONF_IPV4_SCTP);
509 ret = i40e_prepare_fdir_filter
510 (pf, fd_data, add, raw_packet,
511 I40E_SCTPIP6_DUMMY_PACKET_LEN,
512 I40E_FILTER_PCTYPE_NONF_IPV6_SCTP);
519 i40e_change_filter_num(ipv4, add, &pf->fd_sctp4_filter_cnt,
520 &pf->fd_sctp6_filter_cnt);
525 #define I40E_IP_DUMMY_PACKET_LEN 34
526 #define I40E_IP6_DUMMY_PACKET_LEN 54
528 * i40e_add_del_fdir_ip - Add/Remove IPv4 Flow Director filters for
529 * a specific flow spec
530 * @vsi: pointer to the targeted VSI
531 * @fd_data: the flow director data required for the FDir descriptor
532 * @add: true adds a filter, false removes it
533 * @ipv4: true is v4, false is v6
535 * Returns 0 if the filters were successfully added or removed
537 static int i40e_add_del_fdir_ip(struct i40e_vsi *vsi,
538 struct i40e_fdir_filter *fd_data,
542 struct i40e_pf *pf = vsi->back;
551 iter_start = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
552 iter_end = I40E_FILTER_PCTYPE_FRAG_IPV4;
554 iter_start = I40E_FILTER_PCTYPE_NONF_IPV6_OTHER;
555 iter_end = I40E_FILTER_PCTYPE_FRAG_IPV6;
558 for (i = iter_start; i <= iter_end; i++) {
559 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
563 /* IPv6 no header option differs from IPv4 */
564 (void)i40e_create_dummy_packet
565 (raw_packet, ipv4, (ipv4) ? IPPROTO_IP : IPPROTO_NONE,
568 payload_offset = (ipv4) ? I40E_IP_DUMMY_PACKET_LEN :
569 I40E_IP6_DUMMY_PACKET_LEN;
570 ret = i40e_prepare_fdir_filter(pf, fd_data, add, raw_packet,
576 i40e_change_filter_num(ipv4, add, &pf->fd_ip4_filter_cnt,
577 &pf->fd_ip6_filter_cnt);
586 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
587 * @vsi: pointer to the targeted VSI
588 * @input: filter to add or delete
589 * @add: true adds a filter, false removes it
592 int i40e_add_del_fdir(struct i40e_vsi *vsi,
593 struct i40e_fdir_filter *input, bool add)
595 enum ip_ver { ipv6 = 0, ipv4 = 1 };
596 struct i40e_pf *pf = vsi->back;
599 switch (input->flow_type & ~FLOW_EXT) {
601 ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
604 ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
607 ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
610 ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
613 ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
616 ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
619 switch (input->ipl4_proto) {
621 ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv4);
624 ret = i40e_add_del_fdir_udp(vsi, input, add, ipv4);
627 ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv4);
630 ret = i40e_add_del_fdir_ip(vsi, input, add, ipv4);
633 /* We cannot support masking based on protocol */
634 dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
640 switch (input->ipl4_proto) {
642 ret = i40e_add_del_fdir_tcp(vsi, input, add, ipv6);
645 ret = i40e_add_del_fdir_udp(vsi, input, add, ipv6);
648 ret = i40e_add_del_fdir_sctp(vsi, input, add, ipv6);
651 ret = i40e_add_del_fdir_ip(vsi, input, add, ipv6);
654 /* We cannot support masking based on protocol */
655 dev_info(&pf->pdev->dev, "Unsupported IPv6 protocol 0x%02x\n",
661 dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
666 /* The buffer allocated here will be normally be freed by
667 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
668 * completion. In the event of an error adding the buffer to the FDIR
669 * ring, it will immediately be freed. It may also be freed by
670 * i40e_clean_tx_ring() when closing the VSI.
676 * i40e_fd_handle_status - check the Programming Status for FD
677 * @rx_ring: the Rx ring for this descriptor
678 * @qword0_raw: qword0
679 * @qword1: qword1 after le_to_cpu
680 * @prog_id: the id originally used for programming
682 * This is used to verify if the FD programming or invalidation
683 * requested by SW to the HW is successful or not and take actions accordingly.
685 static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u64 qword0_raw,
686 u64 qword1, u8 prog_id)
688 struct i40e_pf *pf = rx_ring->vsi->back;
689 struct pci_dev *pdev = pf->pdev;
690 struct i40e_16b_rx_wb_qw0 *qw0;
691 u32 fcnt_prog, fcnt_avail;
694 qw0 = (struct i40e_16b_rx_wb_qw0 *)&qword0_raw;
695 error = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
696 I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
698 if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
699 pf->fd_inv = le32_to_cpu(qw0->hi_dword.fd_id);
700 if (qw0->hi_dword.fd_id != 0 ||
701 (I40E_DEBUG_FD & pf->hw.debug_mask))
702 dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
705 /* Check if the programming error is for ATR.
706 * If so, auto disable ATR and set a state for
707 * flush in progress. Next time we come here if flush is in
708 * progress do nothing, once flush is complete the state will
711 if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
715 /* store the current atr filter count */
716 pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
718 if (qw0->hi_dword.fd_id == 0 &&
719 test_bit(__I40E_FD_SB_AUTO_DISABLED, pf->state)) {
720 /* These set_bit() calls aren't atomic with the
721 * test_bit() here, but worse case we potentially
722 * disable ATR and queue a flush right after SB
723 * support is re-enabled. That shouldn't cause an
726 set_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state);
727 set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
730 /* filter programming failed most likely due to table full */
731 fcnt_prog = i40e_get_global_fd_count(pf);
732 fcnt_avail = pf->fdir_pf_filter_count;
733 /* If ATR is running fcnt_prog can quickly change,
734 * if we are very close to full, it makes sense to disable
735 * FD ATR/SB and then re-enable it when there is room.
737 if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
738 if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
739 !test_and_set_bit(__I40E_FD_SB_AUTO_DISABLED,
741 if (I40E_DEBUG_FD & pf->hw.debug_mask)
742 dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
744 } else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
745 if (I40E_DEBUG_FD & pf->hw.debug_mask)
746 dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
747 qw0->hi_dword.fd_id);
752 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
753 * @ring: the ring that owns the buffer
754 * @tx_buffer: the buffer to free
756 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
757 struct i40e_tx_buffer *tx_buffer)
759 if (tx_buffer->skb) {
760 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
761 kfree(tx_buffer->raw_buf);
762 else if (ring_is_xdp(ring))
763 xdp_return_frame(tx_buffer->xdpf);
765 dev_kfree_skb_any(tx_buffer->skb);
766 if (dma_unmap_len(tx_buffer, len))
767 dma_unmap_single(ring->dev,
768 dma_unmap_addr(tx_buffer, dma),
769 dma_unmap_len(tx_buffer, len),
771 } else if (dma_unmap_len(tx_buffer, len)) {
772 dma_unmap_page(ring->dev,
773 dma_unmap_addr(tx_buffer, dma),
774 dma_unmap_len(tx_buffer, len),
778 tx_buffer->next_to_watch = NULL;
779 tx_buffer->skb = NULL;
780 dma_unmap_len_set(tx_buffer, len, 0);
781 /* tx_buffer must be completely set up in the transmit path */
785 * i40e_clean_tx_ring - Free any empty Tx buffers
786 * @tx_ring: ring to be cleaned
788 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
790 unsigned long bi_size;
793 if (ring_is_xdp(tx_ring) && tx_ring->xsk_pool) {
794 i40e_xsk_clean_tx_ring(tx_ring);
796 /* ring already cleared, nothing to do */
800 /* Free all the Tx ring sk_buffs */
801 for (i = 0; i < tx_ring->count; i++)
802 i40e_unmap_and_free_tx_resource(tx_ring,
806 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
807 memset(tx_ring->tx_bi, 0, bi_size);
809 /* Zero out the descriptor ring */
810 memset(tx_ring->desc, 0, tx_ring->size);
812 tx_ring->next_to_use = 0;
813 tx_ring->next_to_clean = 0;
815 if (!tx_ring->netdev)
818 /* cleanup Tx queue statistics */
819 netdev_tx_reset_queue(txring_txq(tx_ring));
823 * i40e_free_tx_resources - Free Tx resources per queue
824 * @tx_ring: Tx descriptor ring for a specific queue
826 * Free all transmit software resources
828 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
830 i40e_clean_tx_ring(tx_ring);
831 kfree(tx_ring->tx_bi);
832 tx_ring->tx_bi = NULL;
833 kfree(tx_ring->xsk_descs);
834 tx_ring->xsk_descs = NULL;
837 dma_free_coherent(tx_ring->dev, tx_ring->size,
838 tx_ring->desc, tx_ring->dma);
839 tx_ring->desc = NULL;
844 * i40e_get_tx_pending - how many tx descriptors not processed
845 * @ring: the ring of descriptors
846 * @in_sw: use SW variables
848 * Since there is no access to the ring head register
849 * in XL710, we need to use our local copies
851 u32 i40e_get_tx_pending(struct i40e_ring *ring, bool in_sw)
856 head = i40e_get_head(ring);
857 tail = readl(ring->tail);
859 head = ring->next_to_clean;
860 tail = ring->next_to_use;
864 return (head < tail) ?
865 tail - head : (tail + ring->count - head);
871 * i40e_detect_recover_hung - Function to detect and recover hung_queues
872 * @vsi: pointer to vsi struct with tx queues
874 * VSI has netdev and netdev has TX queues. This function is to check each of
875 * those TX queues if they are hung, trigger recovery by issuing SW interrupt.
877 void i40e_detect_recover_hung(struct i40e_vsi *vsi)
879 struct i40e_ring *tx_ring = NULL;
880 struct net_device *netdev;
887 if (test_bit(__I40E_VSI_DOWN, vsi->state))
890 netdev = vsi->netdev;
894 if (!netif_carrier_ok(netdev))
897 for (i = 0; i < vsi->num_queue_pairs; i++) {
898 tx_ring = vsi->tx_rings[i];
899 if (tx_ring && tx_ring->desc) {
900 /* If packet counter has not changed the queue is
901 * likely stalled, so force an interrupt for this
904 * prev_pkt_ctr would be negative if there was no
907 packets = tx_ring->stats.packets & INT_MAX;
908 if (tx_ring->tx_stats.prev_pkt_ctr == packets) {
909 i40e_force_wb(vsi, tx_ring->q_vector);
913 /* Memory barrier between read of packet count and call
914 * to i40e_get_tx_pending()
917 tx_ring->tx_stats.prev_pkt_ctr =
918 i40e_get_tx_pending(tx_ring, true) ? packets : -1;
924 * i40e_clean_tx_irq - Reclaim resources after transmit completes
925 * @vsi: the VSI we care about
926 * @tx_ring: Tx ring to clean
927 * @napi_budget: Used to determine if we are in netpoll
929 * Returns true if there's any budget left (e.g. the clean is finished)
931 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
932 struct i40e_ring *tx_ring, int napi_budget)
934 int i = tx_ring->next_to_clean;
935 struct i40e_tx_buffer *tx_buf;
936 struct i40e_tx_desc *tx_head;
937 struct i40e_tx_desc *tx_desc;
938 unsigned int total_bytes = 0, total_packets = 0;
939 unsigned int budget = vsi->work_limit;
941 tx_buf = &tx_ring->tx_bi[i];
942 tx_desc = I40E_TX_DESC(tx_ring, i);
945 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
948 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
950 /* if next_to_watch is not set then there is no work pending */
954 /* prevent any other reads prior to eop_desc */
957 i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
958 /* we have caught up to head, no work left to do */
959 if (tx_head == tx_desc)
962 /* clear next_to_watch to prevent false hangs */
963 tx_buf->next_to_watch = NULL;
965 /* update the statistics for this packet */
966 total_bytes += tx_buf->bytecount;
967 total_packets += tx_buf->gso_segs;
969 /* free the skb/XDP data */
970 if (ring_is_xdp(tx_ring))
971 xdp_return_frame(tx_buf->xdpf);
973 napi_consume_skb(tx_buf->skb, napi_budget);
975 /* unmap skb header data */
976 dma_unmap_single(tx_ring->dev,
977 dma_unmap_addr(tx_buf, dma),
978 dma_unmap_len(tx_buf, len),
981 /* clear tx_buffer data */
983 dma_unmap_len_set(tx_buf, len, 0);
985 /* unmap remaining buffers */
986 while (tx_desc != eop_desc) {
987 i40e_trace(clean_tx_irq_unmap,
988 tx_ring, tx_desc, tx_buf);
995 tx_buf = tx_ring->tx_bi;
996 tx_desc = I40E_TX_DESC(tx_ring, 0);
999 /* unmap any remaining paged data */
1000 if (dma_unmap_len(tx_buf, len)) {
1001 dma_unmap_page(tx_ring->dev,
1002 dma_unmap_addr(tx_buf, dma),
1003 dma_unmap_len(tx_buf, len),
1005 dma_unmap_len_set(tx_buf, len, 0);
1009 /* move us one more past the eop_desc for start of next pkt */
1014 i -= tx_ring->count;
1015 tx_buf = tx_ring->tx_bi;
1016 tx_desc = I40E_TX_DESC(tx_ring, 0);
1021 /* update budget accounting */
1023 } while (likely(budget));
1025 i += tx_ring->count;
1026 tx_ring->next_to_clean = i;
1027 i40e_update_tx_stats(tx_ring, total_packets, total_bytes);
1028 i40e_arm_wb(tx_ring, vsi, budget);
1030 if (ring_is_xdp(tx_ring))
1033 /* notify netdev of completed buffers */
1034 netdev_tx_completed_queue(txring_txq(tx_ring),
1035 total_packets, total_bytes);
1037 #define TX_WAKE_THRESHOLD ((s16)(DESC_NEEDED * 2))
1038 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1039 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
1040 /* Make sure that anybody stopping the queue after this
1041 * sees the new next_to_clean.
1044 if (__netif_subqueue_stopped(tx_ring->netdev,
1045 tx_ring->queue_index) &&
1046 !test_bit(__I40E_VSI_DOWN, vsi->state)) {
1047 netif_wake_subqueue(tx_ring->netdev,
1048 tx_ring->queue_index);
1049 ++tx_ring->tx_stats.restart_queue;
1057 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
1058 * @vsi: the VSI we care about
1059 * @q_vector: the vector on which to enable writeback
1062 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
1063 struct i40e_q_vector *q_vector)
1065 u16 flags = q_vector->tx.ring[0].flags;
1068 if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
1071 if (q_vector->arm_wb_state)
1074 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1075 val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
1076 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
1078 wr32(&vsi->back->hw,
1079 I40E_PFINT_DYN_CTLN(q_vector->reg_idx),
1082 val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
1083 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
1085 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1087 q_vector->arm_wb_state = true;
1091 * i40e_force_wb - Issue SW Interrupt so HW does a wb
1092 * @vsi: the VSI we care about
1093 * @q_vector: the vector on which to force writeback
1096 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
1098 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1099 u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
1100 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
1101 I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
1102 I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
1103 /* allow 00 to be written to the index */
1105 wr32(&vsi->back->hw,
1106 I40E_PFINT_DYN_CTLN(q_vector->reg_idx), val);
1108 u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
1109 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
1110 I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
1111 I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
1112 /* allow 00 to be written to the index */
1114 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
1118 static inline bool i40e_container_is_rx(struct i40e_q_vector *q_vector,
1119 struct i40e_ring_container *rc)
1121 return &q_vector->rx == rc;
1124 static inline unsigned int i40e_itr_divisor(struct i40e_q_vector *q_vector)
1126 unsigned int divisor;
1128 switch (q_vector->vsi->back->hw.phy.link_info.link_speed) {
1129 case I40E_LINK_SPEED_40GB:
1130 divisor = I40E_ITR_ADAPTIVE_MIN_INC * 1024;
1132 case I40E_LINK_SPEED_25GB:
1133 case I40E_LINK_SPEED_20GB:
1134 divisor = I40E_ITR_ADAPTIVE_MIN_INC * 512;
1137 case I40E_LINK_SPEED_10GB:
1138 divisor = I40E_ITR_ADAPTIVE_MIN_INC * 256;
1140 case I40E_LINK_SPEED_1GB:
1141 case I40E_LINK_SPEED_100MB:
1142 divisor = I40E_ITR_ADAPTIVE_MIN_INC * 32;
1150 * i40e_update_itr - update the dynamic ITR value based on statistics
1151 * @q_vector: structure containing interrupt and ring information
1152 * @rc: structure containing ring performance data
1154 * Stores a new ITR value based on packets and byte
1155 * counts during the last interrupt. The advantage of per interrupt
1156 * computation is faster updates and more accurate ITR for the current
1157 * traffic pattern. Constants in this function were computed
1158 * based on theoretical maximum wire speed and thresholds were set based
1159 * on testing data as well as attempting to minimize response time
1160 * while increasing bulk throughput.
1162 static void i40e_update_itr(struct i40e_q_vector *q_vector,
1163 struct i40e_ring_container *rc)
1165 unsigned int avg_wire_size, packets, bytes, itr;
1166 unsigned long next_update = jiffies;
1168 /* If we don't have any rings just leave ourselves set for maximum
1169 * possible latency so we take ourselves out of the equation.
1171 if (!rc->ring || !ITR_IS_DYNAMIC(rc->ring->itr_setting))
1174 /* For Rx we want to push the delay up and default to low latency.
1175 * for Tx we want to pull the delay down and default to high latency.
1177 itr = i40e_container_is_rx(q_vector, rc) ?
1178 I40E_ITR_ADAPTIVE_MIN_USECS | I40E_ITR_ADAPTIVE_LATENCY :
1179 I40E_ITR_ADAPTIVE_MAX_USECS | I40E_ITR_ADAPTIVE_LATENCY;
1181 /* If we didn't update within up to 1 - 2 jiffies we can assume
1182 * that either packets are coming in so slow there hasn't been
1183 * any work, or that there is so much work that NAPI is dealing
1184 * with interrupt moderation and we don't need to do anything.
1186 if (time_after(next_update, rc->next_update))
1189 /* If itr_countdown is set it means we programmed an ITR within
1190 * the last 4 interrupt cycles. This has a side effect of us
1191 * potentially firing an early interrupt. In order to work around
1192 * this we need to throw out any data received for a few
1193 * interrupts following the update.
1195 if (q_vector->itr_countdown) {
1196 itr = rc->target_itr;
1200 packets = rc->total_packets;
1201 bytes = rc->total_bytes;
1203 if (i40e_container_is_rx(q_vector, rc)) {
1204 /* If Rx there are 1 to 4 packets and bytes are less than
1205 * 9000 assume insufficient data to use bulk rate limiting
1206 * approach unless Tx is already in bulk rate limiting. We
1207 * are likely latency driven.
1209 if (packets && packets < 4 && bytes < 9000 &&
1210 (q_vector->tx.target_itr & I40E_ITR_ADAPTIVE_LATENCY)) {
1211 itr = I40E_ITR_ADAPTIVE_LATENCY;
1212 goto adjust_by_size;
1214 } else if (packets < 4) {
1215 /* If we have Tx and Rx ITR maxed and Tx ITR is running in
1216 * bulk mode and we are receiving 4 or fewer packets just
1217 * reset the ITR_ADAPTIVE_LATENCY bit for latency mode so
1218 * that the Rx can relax.
1220 if (rc->target_itr == I40E_ITR_ADAPTIVE_MAX_USECS &&
1221 (q_vector->rx.target_itr & I40E_ITR_MASK) ==
1222 I40E_ITR_ADAPTIVE_MAX_USECS)
1224 } else if (packets > 32) {
1225 /* If we have processed over 32 packets in a single interrupt
1226 * for Tx assume we need to switch over to "bulk" mode.
1228 rc->target_itr &= ~I40E_ITR_ADAPTIVE_LATENCY;
1231 /* We have no packets to actually measure against. This means
1232 * either one of the other queues on this vector is active or
1233 * we are a Tx queue doing TSO with too high of an interrupt rate.
1235 * Between 4 and 56 we can assume that our current interrupt delay
1236 * is only slightly too low. As such we should increase it by a small
1240 itr = rc->target_itr + I40E_ITR_ADAPTIVE_MIN_INC;
1241 if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1242 itr &= I40E_ITR_ADAPTIVE_LATENCY;
1243 itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1248 if (packets <= 256) {
1249 itr = min(q_vector->tx.current_itr, q_vector->rx.current_itr);
1250 itr &= I40E_ITR_MASK;
1252 /* Between 56 and 112 is our "goldilocks" zone where we are
1253 * working out "just right". Just report that our current
1254 * ITR is good for us.
1259 /* If packet count is 128 or greater we are likely looking
1260 * at a slight overrun of the delay we want. Try halving
1261 * our delay to see if that will cut the number of packets
1262 * in half per interrupt.
1265 itr &= I40E_ITR_MASK;
1266 if (itr < I40E_ITR_ADAPTIVE_MIN_USECS)
1267 itr = I40E_ITR_ADAPTIVE_MIN_USECS;
1272 /* The paths below assume we are dealing with a bulk ITR since
1273 * number of packets is greater than 256. We are just going to have
1274 * to compute a value and try to bring the count under control,
1275 * though for smaller packet sizes there isn't much we can do as
1276 * NAPI polling will likely be kicking in sooner rather than later.
1278 itr = I40E_ITR_ADAPTIVE_BULK;
1281 /* If packet counts are 256 or greater we can assume we have a gross
1282 * overestimation of what the rate should be. Instead of trying to fine
1283 * tune it just use the formula below to try and dial in an exact value
1284 * give the current packet size of the frame.
1286 avg_wire_size = bytes / packets;
1288 /* The following is a crude approximation of:
1289 * wmem_default / (size + overhead) = desired_pkts_per_int
1290 * rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1291 * (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1293 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1294 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1297 * (170 * (size + 24)) / (size + 640) = ITR
1299 * We first do some math on the packet size and then finally bitshift
1300 * by 8 after rounding up. We also have to account for PCIe link speed
1301 * difference as ITR scales based on this.
1303 if (avg_wire_size <= 60) {
1304 /* Start at 250k ints/sec */
1305 avg_wire_size = 4096;
1306 } else if (avg_wire_size <= 380) {
1307 /* 250K ints/sec to 60K ints/sec */
1308 avg_wire_size *= 40;
1309 avg_wire_size += 1696;
1310 } else if (avg_wire_size <= 1084) {
1311 /* 60K ints/sec to 36K ints/sec */
1312 avg_wire_size *= 15;
1313 avg_wire_size += 11452;
1314 } else if (avg_wire_size <= 1980) {
1315 /* 36K ints/sec to 30K ints/sec */
1317 avg_wire_size += 22420;
1319 /* plateau at a limit of 30K ints/sec */
1320 avg_wire_size = 32256;
1323 /* If we are in low latency mode halve our delay which doubles the
1324 * rate to somewhere between 100K to 16K ints/sec
1326 if (itr & I40E_ITR_ADAPTIVE_LATENCY)
1329 /* Resultant value is 256 times larger than it needs to be. This
1330 * gives us room to adjust the value as needed to either increase
1331 * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
1333 * Use addition as we have already recorded the new latency flag
1334 * for the ITR value.
1336 itr += DIV_ROUND_UP(avg_wire_size, i40e_itr_divisor(q_vector)) *
1337 I40E_ITR_ADAPTIVE_MIN_INC;
1339 if ((itr & I40E_ITR_MASK) > I40E_ITR_ADAPTIVE_MAX_USECS) {
1340 itr &= I40E_ITR_ADAPTIVE_LATENCY;
1341 itr += I40E_ITR_ADAPTIVE_MAX_USECS;
1345 /* write back value */
1346 rc->target_itr = itr;
1348 /* next update should occur within next jiffy */
1349 rc->next_update = next_update + 1;
1351 rc->total_bytes = 0;
1352 rc->total_packets = 0;
1355 static struct i40e_rx_buffer *i40e_rx_bi(struct i40e_ring *rx_ring, u32 idx)
1357 return &rx_ring->rx_bi[idx];
1361 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1362 * @rx_ring: rx descriptor ring to store buffers on
1363 * @old_buff: donor buffer to have page reused
1365 * Synchronizes page for reuse by the adapter
1367 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1368 struct i40e_rx_buffer *old_buff)
1370 struct i40e_rx_buffer *new_buff;
1371 u16 nta = rx_ring->next_to_alloc;
1373 new_buff = i40e_rx_bi(rx_ring, nta);
1375 /* update, and store next to alloc */
1377 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1379 /* transfer page from old buffer to new buffer */
1380 new_buff->dma = old_buff->dma;
1381 new_buff->page = old_buff->page;
1382 new_buff->page_offset = old_buff->page_offset;
1383 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1385 rx_ring->rx_stats.page_reuse_count++;
1387 /* clear contents of buffer_info */
1388 old_buff->page = NULL;
1392 * i40e_clean_programming_status - clean the programming status descriptor
1393 * @rx_ring: the rx ring that has this descriptor
1394 * @qword0_raw: qword0
1395 * @qword1: qword1 representing status_error_len in CPU ordering
1397 * Flow director should handle FD_FILTER_STATUS to check its filter programming
1398 * status being successful or not and take actions accordingly. FCoE should
1399 * handle its context/filter programming/invalidation status and take actions.
1401 * Returns an i40e_rx_buffer to reuse if the cleanup occurred, otherwise NULL.
1403 void i40e_clean_programming_status(struct i40e_ring *rx_ring, u64 qword0_raw,
1408 id = (qword1 & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1409 I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1411 if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1412 i40e_fd_handle_status(rx_ring, qword0_raw, qword1, id);
1416 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1417 * @tx_ring: the tx ring to set up
1419 * Return 0 on success, negative on error
1421 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1423 struct device *dev = tx_ring->dev;
1429 /* warn if we are about to overwrite the pointer */
1430 WARN_ON(tx_ring->tx_bi);
1431 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1432 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1433 if (!tx_ring->tx_bi)
1436 if (ring_is_xdp(tx_ring)) {
1437 tx_ring->xsk_descs = kcalloc(I40E_MAX_NUM_DESCRIPTORS, sizeof(*tx_ring->xsk_descs),
1439 if (!tx_ring->xsk_descs)
1443 u64_stats_init(&tx_ring->syncp);
1445 /* round up to nearest 4K */
1446 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1447 /* add u32 for head writeback, align after this takes care of
1448 * guaranteeing this is at least one cache line in size
1450 tx_ring->size += sizeof(u32);
1451 tx_ring->size = ALIGN(tx_ring->size, 4096);
1452 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1453 &tx_ring->dma, GFP_KERNEL);
1454 if (!tx_ring->desc) {
1455 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1460 tx_ring->next_to_use = 0;
1461 tx_ring->next_to_clean = 0;
1462 tx_ring->tx_stats.prev_pkt_ctr = -1;
1466 kfree(tx_ring->xsk_descs);
1467 tx_ring->xsk_descs = NULL;
1468 kfree(tx_ring->tx_bi);
1469 tx_ring->tx_bi = NULL;
1473 int i40e_alloc_rx_bi(struct i40e_ring *rx_ring)
1475 unsigned long sz = sizeof(*rx_ring->rx_bi) * rx_ring->count;
1477 rx_ring->rx_bi = kzalloc(sz, GFP_KERNEL);
1478 return rx_ring->rx_bi ? 0 : -ENOMEM;
1481 static void i40e_clear_rx_bi(struct i40e_ring *rx_ring)
1483 memset(rx_ring->rx_bi, 0, sizeof(*rx_ring->rx_bi) * rx_ring->count);
1487 * i40e_clean_rx_ring - Free Rx buffers
1488 * @rx_ring: ring to be cleaned
1490 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1494 /* ring already cleared, nothing to do */
1495 if (!rx_ring->rx_bi)
1499 dev_kfree_skb(rx_ring->skb);
1500 rx_ring->skb = NULL;
1503 if (rx_ring->xsk_pool) {
1504 i40e_xsk_clean_rx_ring(rx_ring);
1508 /* Free all the Rx ring sk_buffs */
1509 for (i = 0; i < rx_ring->count; i++) {
1510 struct i40e_rx_buffer *rx_bi = i40e_rx_bi(rx_ring, i);
1515 /* Invalidate cache lines that may have been written to by
1516 * device so that we avoid corrupting memory.
1518 dma_sync_single_range_for_cpu(rx_ring->dev,
1521 rx_ring->rx_buf_len,
1524 /* free resources associated with mapping */
1525 dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1526 i40e_rx_pg_size(rx_ring),
1530 __page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1533 rx_bi->page_offset = 0;
1537 if (rx_ring->xsk_pool)
1538 i40e_clear_rx_bi_zc(rx_ring);
1540 i40e_clear_rx_bi(rx_ring);
1542 /* Zero out the descriptor ring */
1543 memset(rx_ring->desc, 0, rx_ring->size);
1545 rx_ring->next_to_alloc = 0;
1546 rx_ring->next_to_clean = 0;
1547 rx_ring->next_to_use = 0;
1551 * i40e_free_rx_resources - Free Rx resources
1552 * @rx_ring: ring to clean the resources from
1554 * Free all receive software resources
1556 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1558 i40e_clean_rx_ring(rx_ring);
1559 if (rx_ring->vsi->type == I40E_VSI_MAIN)
1560 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
1561 rx_ring->xdp_prog = NULL;
1562 kfree(rx_ring->rx_bi);
1563 rx_ring->rx_bi = NULL;
1565 if (rx_ring->desc) {
1566 dma_free_coherent(rx_ring->dev, rx_ring->size,
1567 rx_ring->desc, rx_ring->dma);
1568 rx_ring->desc = NULL;
1573 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1574 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1576 * Returns 0 on success, negative on failure
1578 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1580 struct device *dev = rx_ring->dev;
1583 u64_stats_init(&rx_ring->syncp);
1585 /* Round up to nearest 4K */
1586 rx_ring->size = rx_ring->count * sizeof(union i40e_rx_desc);
1587 rx_ring->size = ALIGN(rx_ring->size, 4096);
1588 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1589 &rx_ring->dma, GFP_KERNEL);
1591 if (!rx_ring->desc) {
1592 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1597 rx_ring->next_to_alloc = 0;
1598 rx_ring->next_to_clean = 0;
1599 rx_ring->next_to_use = 0;
1601 /* XDP RX-queue info only needed for RX rings exposed to XDP */
1602 if (rx_ring->vsi->type == I40E_VSI_MAIN) {
1603 err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
1604 rx_ring->queue_index, rx_ring->q_vector->napi.napi_id);
1609 rx_ring->xdp_prog = rx_ring->vsi->xdp_prog;
1615 * i40e_release_rx_desc - Store the new tail and head values
1616 * @rx_ring: ring to bump
1617 * @val: new head index
1619 void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1621 rx_ring->next_to_use = val;
1623 /* update next to alloc since we have filled the ring */
1624 rx_ring->next_to_alloc = val;
1626 /* Force memory writes to complete before letting h/w
1627 * know there are new descriptors to fetch. (Only
1628 * applicable for weak-ordered memory model archs,
1632 writel(val, rx_ring->tail);
1635 static unsigned int i40e_rx_frame_truesize(struct i40e_ring *rx_ring,
1638 unsigned int truesize;
1640 #if (PAGE_SIZE < 8192)
1641 truesize = i40e_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1643 truesize = rx_ring->rx_offset ?
1644 SKB_DATA_ALIGN(size + rx_ring->rx_offset) +
1645 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1646 SKB_DATA_ALIGN(size);
1652 * i40e_alloc_mapped_page - recycle or make a new page
1653 * @rx_ring: ring to use
1654 * @bi: rx_buffer struct to modify
1656 * Returns true if the page was successfully allocated or
1659 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1660 struct i40e_rx_buffer *bi)
1662 struct page *page = bi->page;
1665 /* since we are recycling buffers we should seldom need to alloc */
1667 rx_ring->rx_stats.page_reuse_count++;
1671 /* alloc new page for storage */
1672 page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1673 if (unlikely(!page)) {
1674 rx_ring->rx_stats.alloc_page_failed++;
1678 /* map page for use */
1679 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1680 i40e_rx_pg_size(rx_ring),
1684 /* if mapping failed free memory back to system since
1685 * there isn't much point in holding memory we can't use
1687 if (dma_mapping_error(rx_ring->dev, dma)) {
1688 __free_pages(page, i40e_rx_pg_order(rx_ring));
1689 rx_ring->rx_stats.alloc_page_failed++;
1695 bi->page_offset = rx_ring->rx_offset;
1696 page_ref_add(page, USHRT_MAX - 1);
1697 bi->pagecnt_bias = USHRT_MAX;
1703 * i40e_alloc_rx_buffers - Replace used receive buffers
1704 * @rx_ring: ring to place buffers on
1705 * @cleaned_count: number of buffers to replace
1707 * Returns false if all allocations were successful, true if any fail
1709 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1711 u16 ntu = rx_ring->next_to_use;
1712 union i40e_rx_desc *rx_desc;
1713 struct i40e_rx_buffer *bi;
1715 /* do nothing if no valid netdev defined */
1716 if (!rx_ring->netdev || !cleaned_count)
1719 rx_desc = I40E_RX_DESC(rx_ring, ntu);
1720 bi = i40e_rx_bi(rx_ring, ntu);
1723 if (!i40e_alloc_mapped_page(rx_ring, bi))
1726 /* sync the buffer for use by the device */
1727 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1729 rx_ring->rx_buf_len,
1732 /* Refresh the desc even if buffer_addrs didn't change
1733 * because each write-back erases this info.
1735 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1740 if (unlikely(ntu == rx_ring->count)) {
1741 rx_desc = I40E_RX_DESC(rx_ring, 0);
1742 bi = i40e_rx_bi(rx_ring, 0);
1746 /* clear the status bits for the next_to_use descriptor */
1747 rx_desc->wb.qword1.status_error_len = 0;
1750 } while (cleaned_count);
1752 if (rx_ring->next_to_use != ntu)
1753 i40e_release_rx_desc(rx_ring, ntu);
1758 if (rx_ring->next_to_use != ntu)
1759 i40e_release_rx_desc(rx_ring, ntu);
1761 /* make sure to come back via polling to try again after
1762 * allocation failure
1768 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1769 * @vsi: the VSI we care about
1770 * @skb: skb currently being received and modified
1771 * @rx_desc: the receive descriptor
1773 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1774 struct sk_buff *skb,
1775 union i40e_rx_desc *rx_desc)
1777 struct i40e_rx_ptype_decoded decoded;
1778 u32 rx_error, rx_status;
1783 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1784 ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1785 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1786 I40E_RXD_QW1_ERROR_SHIFT;
1787 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1788 I40E_RXD_QW1_STATUS_SHIFT;
1789 decoded = decode_rx_desc_ptype(ptype);
1791 skb->ip_summed = CHECKSUM_NONE;
1793 skb_checksum_none_assert(skb);
1795 /* Rx csum enabled and ip headers found? */
1796 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1799 /* did the hardware decode the packet and checksum? */
1800 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1803 /* both known and outer_ip must be set for the below code to work */
1804 if (!(decoded.known && decoded.outer_ip))
1807 ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1808 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1809 ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1810 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1813 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1814 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1817 /* likely incorrect csum if alternate IP extension headers found */
1819 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1820 /* don't increment checksum err here, non-fatal err */
1823 /* there was some L4 error, count error and punt packet to the stack */
1824 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1827 /* handle packets that were not able to be checksummed due
1828 * to arrival speed, in this case the stack can compute
1831 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1834 /* If there is an outer header present that might contain a checksum
1835 * we need to bump the checksum level by 1 to reflect the fact that
1836 * we are indicating we validated the inner checksum.
1838 if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1839 skb->csum_level = 1;
1841 /* Only report checksum unnecessary for TCP, UDP, or SCTP */
1842 switch (decoded.inner_prot) {
1843 case I40E_RX_PTYPE_INNER_PROT_TCP:
1844 case I40E_RX_PTYPE_INNER_PROT_UDP:
1845 case I40E_RX_PTYPE_INNER_PROT_SCTP:
1846 skb->ip_summed = CHECKSUM_UNNECESSARY;
1855 vsi->back->hw_csum_rx_error++;
1859 * i40e_ptype_to_htype - get a hash type
1860 * @ptype: the ptype value from the descriptor
1862 * Returns a hash type to be used by skb_set_hash
1864 static inline int i40e_ptype_to_htype(u8 ptype)
1866 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1869 return PKT_HASH_TYPE_NONE;
1871 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1872 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1873 return PKT_HASH_TYPE_L4;
1874 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1875 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1876 return PKT_HASH_TYPE_L3;
1878 return PKT_HASH_TYPE_L2;
1882 * i40e_rx_hash - set the hash value in the skb
1883 * @ring: descriptor ring
1884 * @rx_desc: specific descriptor
1885 * @skb: skb currently being received and modified
1886 * @rx_ptype: Rx packet type
1888 static inline void i40e_rx_hash(struct i40e_ring *ring,
1889 union i40e_rx_desc *rx_desc,
1890 struct sk_buff *skb,
1894 const __le64 rss_mask =
1895 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1896 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1898 if (!(ring->netdev->features & NETIF_F_RXHASH))
1901 if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1902 hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1903 skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1908 * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1909 * @rx_ring: rx descriptor ring packet is being transacted on
1910 * @rx_desc: pointer to the EOP Rx descriptor
1911 * @skb: pointer to current skb being populated
1913 * This function checks the ring, descriptor, and packet information in
1914 * order to populate the hash, checksum, VLAN, protocol, and
1915 * other fields within the skb.
1917 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1918 union i40e_rx_desc *rx_desc, struct sk_buff *skb)
1920 u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1921 u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1922 I40E_RXD_QW1_STATUS_SHIFT;
1923 u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1924 u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1925 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1926 u8 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1927 I40E_RXD_QW1_PTYPE_SHIFT;
1929 if (unlikely(tsynvalid))
1930 i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1932 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1934 i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1936 skb_record_rx_queue(skb, rx_ring->queue_index);
1938 if (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) {
1939 __le16 vlan_tag = rx_desc->wb.qword0.lo_dword.l2tag1;
1941 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1942 le16_to_cpu(vlan_tag));
1945 /* modifies the skb - consumes the enet header */
1946 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1950 * i40e_cleanup_headers - Correct empty headers
1951 * @rx_ring: rx descriptor ring packet is being transacted on
1952 * @skb: pointer to current skb being fixed
1953 * @rx_desc: pointer to the EOP Rx descriptor
1955 * In addition if skb is not at least 60 bytes we need to pad it so that
1956 * it is large enough to qualify as a valid Ethernet frame.
1958 * Returns true if an error was encountered and skb was freed.
1960 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb,
1961 union i40e_rx_desc *rx_desc)
1964 /* ERR_MASK will only have valid bits if EOP set, and
1965 * what we are doing here is actually checking
1966 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
1969 if (unlikely(i40e_test_staterr(rx_desc,
1970 BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
1971 dev_kfree_skb_any(skb);
1975 /* if eth_skb_pad returns an error the skb was freed */
1976 if (eth_skb_pad(skb))
1983 * i40e_can_reuse_rx_page - Determine if page can be reused for another Rx
1984 * @rx_buffer: buffer containing the page
1985 * @rx_buffer_pgcnt: buffer page refcount pre xdp_do_redirect() call
1987 * If page is reusable, we have a green light for calling i40e_reuse_rx_page,
1988 * which will assign the current buffer to the buffer that next_to_alloc is
1989 * pointing to; otherwise, the DMA mapping needs to be destroyed and
1992 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer,
1993 int rx_buffer_pgcnt)
1995 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1996 struct page *page = rx_buffer->page;
1998 /* Is any reuse possible? */
1999 if (!dev_page_is_reusable(page))
2002 #if (PAGE_SIZE < 8192)
2003 /* if we are only owner of page we can reuse it */
2004 if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
2007 #define I40E_LAST_OFFSET \
2008 (SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
2009 if (rx_buffer->page_offset > I40E_LAST_OFFSET)
2013 /* If we have drained the page fragment pool we need to update
2014 * the pagecnt_bias and page count so that we fully restock the
2015 * number of references the driver holds.
2017 if (unlikely(pagecnt_bias == 1)) {
2018 page_ref_add(page, USHRT_MAX - 1);
2019 rx_buffer->pagecnt_bias = USHRT_MAX;
2026 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
2027 * @rx_ring: rx descriptor ring to transact packets on
2028 * @rx_buffer: buffer containing page to add
2029 * @skb: sk_buff to place the data into
2030 * @size: packet length from rx_desc
2032 * This function will add the data contained in rx_buffer->page to the skb.
2033 * It will just attach the page as a frag to the skb.
2035 * The function will then update the page offset.
2037 static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
2038 struct i40e_rx_buffer *rx_buffer,
2039 struct sk_buff *skb,
2042 #if (PAGE_SIZE < 8192)
2043 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2045 unsigned int truesize = SKB_DATA_ALIGN(size + rx_ring->rx_offset);
2048 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2049 rx_buffer->page_offset, size, truesize);
2051 /* page is being used so we must update the page offset */
2052 #if (PAGE_SIZE < 8192)
2053 rx_buffer->page_offset ^= truesize;
2055 rx_buffer->page_offset += truesize;
2060 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
2061 * @rx_ring: rx descriptor ring to transact packets on
2062 * @size: size of buffer to add to skb
2063 * @rx_buffer_pgcnt: buffer page refcount
2065 * This function will pull an Rx buffer from the ring and synchronize it
2066 * for use by the CPU.
2068 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
2069 const unsigned int size,
2070 int *rx_buffer_pgcnt)
2072 struct i40e_rx_buffer *rx_buffer;
2074 rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2076 #if (PAGE_SIZE < 8192)
2077 page_count(rx_buffer->page);
2081 prefetch_page_address(rx_buffer->page);
2083 /* we are reusing so sync this buffer for CPU use */
2084 dma_sync_single_range_for_cpu(rx_ring->dev,
2086 rx_buffer->page_offset,
2090 /* We have pulled a buffer for use, so decrement pagecnt_bias */
2091 rx_buffer->pagecnt_bias--;
2097 * i40e_construct_skb - Allocate skb and populate it
2098 * @rx_ring: rx descriptor ring to transact packets on
2099 * @rx_buffer: rx buffer to pull data from
2100 * @xdp: xdp_buff pointing to the data
2102 * This function allocates an skb. It then populates it with the page
2103 * data from the current receive descriptor, taking care to set up the
2106 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
2107 struct i40e_rx_buffer *rx_buffer,
2108 struct xdp_buff *xdp)
2110 unsigned int size = xdp->data_end - xdp->data;
2111 #if (PAGE_SIZE < 8192)
2112 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2114 unsigned int truesize = SKB_DATA_ALIGN(size);
2116 unsigned int headlen;
2117 struct sk_buff *skb;
2119 /* prefetch first cache line of first page */
2120 net_prefetch(xdp->data);
2122 /* Note, we get here by enabling legacy-rx via:
2124 * ethtool --set-priv-flags <dev> legacy-rx on
2126 * In this mode, we currently get 0 extra XDP headroom as
2127 * opposed to having legacy-rx off, where we process XDP
2128 * packets going to stack via i40e_build_skb(). The latter
2129 * provides us currently with 192 bytes of headroom.
2131 * For i40e_construct_skb() mode it means that the
2132 * xdp->data_meta will always point to xdp->data, since
2133 * the helper cannot expand the head. Should this ever
2134 * change in future for legacy-rx mode on, then lets also
2135 * add xdp->data_meta handling here.
2138 /* allocate a skb to store the frags */
2139 skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
2141 GFP_ATOMIC | __GFP_NOWARN);
2145 /* Determine available headroom for copy */
2147 if (headlen > I40E_RX_HDR_SIZE)
2148 headlen = eth_get_headlen(skb->dev, xdp->data,
2151 /* align pull length to size of long to optimize memcpy performance */
2152 memcpy(__skb_put(skb, headlen), xdp->data,
2153 ALIGN(headlen, sizeof(long)));
2155 /* update all of the pointers */
2158 skb_add_rx_frag(skb, 0, rx_buffer->page,
2159 rx_buffer->page_offset + headlen,
2162 /* buffer is used by skb, update page_offset */
2163 #if (PAGE_SIZE < 8192)
2164 rx_buffer->page_offset ^= truesize;
2166 rx_buffer->page_offset += truesize;
2169 /* buffer is unused, reset bias back to rx_buffer */
2170 rx_buffer->pagecnt_bias++;
2177 * i40e_build_skb - Build skb around an existing buffer
2178 * @rx_ring: Rx descriptor ring to transact packets on
2179 * @rx_buffer: Rx buffer to pull data from
2180 * @xdp: xdp_buff pointing to the data
2182 * This function builds an skb around an existing Rx buffer, taking care
2183 * to set up the skb correctly and avoid any memcpy overhead.
2185 static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
2186 struct i40e_rx_buffer *rx_buffer,
2187 struct xdp_buff *xdp)
2189 unsigned int metasize = xdp->data - xdp->data_meta;
2190 #if (PAGE_SIZE < 8192)
2191 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
2193 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2194 SKB_DATA_ALIGN(xdp->data_end -
2195 xdp->data_hard_start);
2197 struct sk_buff *skb;
2199 /* Prefetch first cache line of first page. If xdp->data_meta
2200 * is unused, this points exactly as xdp->data, otherwise we
2201 * likely have a consumer accessing first few bytes of meta
2202 * data, and then actual data.
2204 net_prefetch(xdp->data_meta);
2206 /* build an skb around the page buffer */
2207 skb = build_skb(xdp->data_hard_start, truesize);
2211 /* update pointers within the skb to store the data */
2212 skb_reserve(skb, xdp->data - xdp->data_hard_start);
2213 __skb_put(skb, xdp->data_end - xdp->data);
2215 skb_metadata_set(skb, metasize);
2217 /* buffer is used by skb, update page_offset */
2218 #if (PAGE_SIZE < 8192)
2219 rx_buffer->page_offset ^= truesize;
2221 rx_buffer->page_offset += truesize;
2228 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
2229 * @rx_ring: rx descriptor ring to transact packets on
2230 * @rx_buffer: rx buffer to pull data from
2231 * @rx_buffer_pgcnt: rx buffer page refcount pre xdp_do_redirect() call
2233 * This function will clean up the contents of the rx_buffer. It will
2234 * either recycle the buffer or unmap it and free the associated resources.
2236 static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
2237 struct i40e_rx_buffer *rx_buffer,
2238 int rx_buffer_pgcnt)
2240 if (i40e_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2241 /* hand second half of page back to the ring */
2242 i40e_reuse_rx_page(rx_ring, rx_buffer);
2244 /* we are not reusing the buffer so unmap it */
2245 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2246 i40e_rx_pg_size(rx_ring),
2247 DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
2248 __page_frag_cache_drain(rx_buffer->page,
2249 rx_buffer->pagecnt_bias);
2250 /* clear contents of buffer_info */
2251 rx_buffer->page = NULL;
2256 * i40e_is_non_eop - process handling of non-EOP buffers
2257 * @rx_ring: Rx ring being processed
2258 * @rx_desc: Rx descriptor for current buffer
2260 * If the buffer is an EOP buffer, this function exits returning false,
2261 * otherwise return true indicating that this is in fact a non-EOP buffer.
2263 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
2264 union i40e_rx_desc *rx_desc)
2266 /* if we are the last buffer then there is nothing else to do */
2267 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
2268 if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
2271 rx_ring->rx_stats.non_eop_descs++;
2276 static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
2277 struct i40e_ring *xdp_ring);
2279 int i40e_xmit_xdp_tx_ring(struct xdp_buff *xdp, struct i40e_ring *xdp_ring)
2281 struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2283 if (unlikely(!xdpf))
2284 return I40E_XDP_CONSUMED;
2286 return i40e_xmit_xdp_ring(xdpf, xdp_ring);
2290 * i40e_run_xdp - run an XDP program
2291 * @rx_ring: Rx ring being processed
2292 * @xdp: XDP buffer containing the frame
2294 static int i40e_run_xdp(struct i40e_ring *rx_ring, struct xdp_buff *xdp)
2296 int err, result = I40E_XDP_PASS;
2297 struct i40e_ring *xdp_ring;
2298 struct bpf_prog *xdp_prog;
2302 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
2307 prefetchw(xdp->data_hard_start); /* xdp_frame write */
2309 act = bpf_prog_run_xdp(xdp_prog, xdp);
2314 xdp_ring = rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2315 result = i40e_xmit_xdp_tx_ring(xdp, xdp_ring);
2316 if (result == I40E_XDP_CONSUMED)
2320 err = xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog);
2323 result = I40E_XDP_REDIR;
2326 bpf_warn_invalid_xdp_action(act);
2330 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
2331 fallthrough; /* handle aborts by dropping packet */
2333 result = I40E_XDP_CONSUMED;
2342 * i40e_rx_buffer_flip - adjusted rx_buffer to point to an unused region
2344 * @rx_buffer: Rx buffer to adjust
2345 * @size: Size of adjustment
2347 static void i40e_rx_buffer_flip(struct i40e_ring *rx_ring,
2348 struct i40e_rx_buffer *rx_buffer,
2351 unsigned int truesize = i40e_rx_frame_truesize(rx_ring, size);
2353 #if (PAGE_SIZE < 8192)
2354 rx_buffer->page_offset ^= truesize;
2356 rx_buffer->page_offset += truesize;
2361 * i40e_xdp_ring_update_tail - Updates the XDP Tx ring tail register
2362 * @xdp_ring: XDP Tx ring
2364 * This function updates the XDP Tx ring tail register.
2366 void i40e_xdp_ring_update_tail(struct i40e_ring *xdp_ring)
2368 /* Force memory writes to complete before letting h/w
2369 * know there are new descriptors to fetch.
2372 writel_relaxed(xdp_ring->next_to_use, xdp_ring->tail);
2376 * i40e_update_rx_stats - Update Rx ring statistics
2377 * @rx_ring: rx descriptor ring
2378 * @total_rx_bytes: number of bytes received
2379 * @total_rx_packets: number of packets received
2381 * This function updates the Rx ring statistics.
2383 void i40e_update_rx_stats(struct i40e_ring *rx_ring,
2384 unsigned int total_rx_bytes,
2385 unsigned int total_rx_packets)
2387 u64_stats_update_begin(&rx_ring->syncp);
2388 rx_ring->stats.packets += total_rx_packets;
2389 rx_ring->stats.bytes += total_rx_bytes;
2390 u64_stats_update_end(&rx_ring->syncp);
2391 rx_ring->q_vector->rx.total_packets += total_rx_packets;
2392 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2396 * i40e_finalize_xdp_rx - Bump XDP Tx tail and/or flush redirect map
2398 * @xdp_res: Result of the receive batch
2400 * This function bumps XDP Tx tail and/or flush redirect map, and
2401 * should be called when a batch of packets has been processed in the
2404 void i40e_finalize_xdp_rx(struct i40e_ring *rx_ring, unsigned int xdp_res)
2406 if (xdp_res & I40E_XDP_REDIR)
2409 if (xdp_res & I40E_XDP_TX) {
2410 struct i40e_ring *xdp_ring =
2411 rx_ring->vsi->xdp_rings[rx_ring->queue_index];
2413 i40e_xdp_ring_update_tail(xdp_ring);
2418 * i40e_inc_ntc: Advance the next_to_clean index
2421 static void i40e_inc_ntc(struct i40e_ring *rx_ring)
2423 u32 ntc = rx_ring->next_to_clean + 1;
2425 ntc = (ntc < rx_ring->count) ? ntc : 0;
2426 rx_ring->next_to_clean = ntc;
2427 prefetch(I40E_RX_DESC(rx_ring, ntc));
2431 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2432 * @rx_ring: rx descriptor ring to transact packets on
2433 * @budget: Total limit on number of packets to process
2435 * This function provides a "bounce buffer" approach to Rx interrupt
2436 * processing. The advantage to this is that on systems that have
2437 * expensive overhead for IOMMU access this provides a means of avoiding
2438 * it by maintaining the mapping of the page to the system.
2440 * Returns amount of work completed
2442 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
2444 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
2445 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
2446 unsigned int offset = rx_ring->rx_offset;
2447 struct sk_buff *skb = rx_ring->skb;
2448 unsigned int xdp_xmit = 0;
2449 bool failure = false;
2450 struct xdp_buff xdp;
2453 #if (PAGE_SIZE < 8192)
2454 frame_sz = i40e_rx_frame_truesize(rx_ring, 0);
2456 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
2458 while (likely(total_rx_packets < (unsigned int)budget)) {
2459 struct i40e_rx_buffer *rx_buffer;
2460 union i40e_rx_desc *rx_desc;
2461 int rx_buffer_pgcnt;
2465 /* return some buffers to hardware, one at a time is too slow */
2466 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
2467 failure = failure ||
2468 i40e_alloc_rx_buffers(rx_ring, cleaned_count);
2472 rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
2474 /* status_error_len will always be zero for unused descriptors
2475 * because it's cleared in cleanup, and overlaps with hdr_addr
2476 * which is always zero because packet split isn't used, if the
2477 * hardware wrote DD then the length will be non-zero
2479 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2481 /* This memory barrier is needed to keep us from reading
2482 * any other fields out of the rx_desc until we have
2483 * verified the descriptor has been written back.
2487 if (i40e_rx_is_programming_status(qword)) {
2488 i40e_clean_programming_status(rx_ring,
2489 rx_desc->raw.qword[0],
2491 rx_buffer = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
2492 i40e_inc_ntc(rx_ring);
2493 i40e_reuse_rx_page(rx_ring, rx_buffer);
2498 size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
2499 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
2503 i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
2504 rx_buffer = i40e_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
2506 /* retrieve a buffer from the ring */
2508 unsigned char *hard_start;
2510 hard_start = page_address(rx_buffer->page) +
2511 rx_buffer->page_offset - offset;
2512 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
2513 #if (PAGE_SIZE > 4096)
2514 /* At larger PAGE_SIZE, frame_sz depend on len size */
2515 xdp.frame_sz = i40e_rx_frame_truesize(rx_ring, size);
2517 xdp_res = i40e_run_xdp(rx_ring, &xdp);
2521 if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
2522 xdp_xmit |= xdp_res;
2523 i40e_rx_buffer_flip(rx_ring, rx_buffer, size);
2525 rx_buffer->pagecnt_bias++;
2527 total_rx_bytes += size;
2530 i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
2531 } else if (ring_uses_build_skb(rx_ring)) {
2532 skb = i40e_build_skb(rx_ring, rx_buffer, &xdp);
2534 skb = i40e_construct_skb(rx_ring, rx_buffer, &xdp);
2537 /* exit if we failed to retrieve a buffer */
2538 if (!xdp_res && !skb) {
2539 rx_ring->rx_stats.alloc_buff_failed++;
2540 rx_buffer->pagecnt_bias++;
2544 i40e_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
2547 i40e_inc_ntc(rx_ring);
2548 if (i40e_is_non_eop(rx_ring, rx_desc))
2551 if (xdp_res || i40e_cleanup_headers(rx_ring, skb, rx_desc)) {
2556 /* probably a little skewed due to removing CRC */
2557 total_rx_bytes += skb->len;
2559 /* populate checksum, VLAN, and protocol */
2560 i40e_process_skb_fields(rx_ring, rx_desc, skb);
2562 i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
2563 napi_gro_receive(&rx_ring->q_vector->napi, skb);
2566 /* update budget accounting */
2570 i40e_finalize_xdp_rx(rx_ring, xdp_xmit);
2573 i40e_update_rx_stats(rx_ring, total_rx_bytes, total_rx_packets);
2575 /* guarantee a trip back through this routine if there was a failure */
2576 return failure ? budget : (int)total_rx_packets;
2579 static inline u32 i40e_buildreg_itr(const int type, u16 itr)
2583 /* We don't bother with setting the CLEARPBA bit as the data sheet
2584 * points out doing so is "meaningless since it was already
2585 * auto-cleared". The auto-clearing happens when the interrupt is
2588 * Hardware errata 28 for also indicates that writing to a
2589 * xxINT_DYN_CTLx CSR with INTENA_MSK (bit 31) set to 0 will clear
2590 * an event in the PBA anyway so we need to rely on the automask
2591 * to hold pending events for us until the interrupt is re-enabled
2593 * The itr value is reported in microseconds, and the register
2594 * value is recorded in 2 microsecond units. For this reason we
2595 * only need to shift by the interval shift - 1 instead of the
2598 itr &= I40E_ITR_MASK;
2600 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2601 (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2602 (itr << (I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT - 1));
2607 /* a small macro to shorten up some long lines */
2608 #define INTREG I40E_PFINT_DYN_CTLN
2610 /* The act of updating the ITR will cause it to immediately trigger. In order
2611 * to prevent this from throwing off adaptive update statistics we defer the
2612 * update so that it can only happen so often. So after either Tx or Rx are
2613 * updated we make the adaptive scheme wait until either the ITR completely
2614 * expires via the next_update expiration or we have been through at least
2617 #define ITR_COUNTDOWN_START 3
2620 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2621 * @vsi: the VSI we care about
2622 * @q_vector: q_vector for which itr is being updated and interrupt enabled
2625 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2626 struct i40e_q_vector *q_vector)
2628 struct i40e_hw *hw = &vsi->back->hw;
2631 /* If we don't have MSIX, then we only need to re-enable icr0 */
2632 if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED)) {
2633 i40e_irq_dynamic_enable_icr0(vsi->back);
2637 /* These will do nothing if dynamic updates are not enabled */
2638 i40e_update_itr(q_vector, &q_vector->tx);
2639 i40e_update_itr(q_vector, &q_vector->rx);
2641 /* This block of logic allows us to get away with only updating
2642 * one ITR value with each interrupt. The idea is to perform a
2643 * pseudo-lazy update with the following criteria.
2645 * 1. Rx is given higher priority than Tx if both are in same state
2646 * 2. If we must reduce an ITR that is given highest priority.
2647 * 3. We then give priority to increasing ITR based on amount.
2649 if (q_vector->rx.target_itr < q_vector->rx.current_itr) {
2650 /* Rx ITR needs to be reduced, this is highest priority */
2651 intval = i40e_buildreg_itr(I40E_RX_ITR,
2652 q_vector->rx.target_itr);
2653 q_vector->rx.current_itr = q_vector->rx.target_itr;
2654 q_vector->itr_countdown = ITR_COUNTDOWN_START;
2655 } else if ((q_vector->tx.target_itr < q_vector->tx.current_itr) ||
2656 ((q_vector->rx.target_itr - q_vector->rx.current_itr) <
2657 (q_vector->tx.target_itr - q_vector->tx.current_itr))) {
2658 /* Tx ITR needs to be reduced, this is second priority
2659 * Tx ITR needs to be increased more than Rx, fourth priority
2661 intval = i40e_buildreg_itr(I40E_TX_ITR,
2662 q_vector->tx.target_itr);
2663 q_vector->tx.current_itr = q_vector->tx.target_itr;
2664 q_vector->itr_countdown = ITR_COUNTDOWN_START;
2665 } else if (q_vector->rx.current_itr != q_vector->rx.target_itr) {
2666 /* Rx ITR needs to be increased, third priority */
2667 intval = i40e_buildreg_itr(I40E_RX_ITR,
2668 q_vector->rx.target_itr);
2669 q_vector->rx.current_itr = q_vector->rx.target_itr;
2670 q_vector->itr_countdown = ITR_COUNTDOWN_START;
2672 /* No ITR update, lowest priority */
2673 intval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2674 if (q_vector->itr_countdown)
2675 q_vector->itr_countdown--;
2678 if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2679 wr32(hw, INTREG(q_vector->reg_idx), intval);
2683 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2684 * @napi: napi struct with our devices info in it
2685 * @budget: amount of work driver is allowed to do this pass, in packets
2687 * This function will clean all queues associated with a q_vector.
2689 * Returns the amount of work done
2691 int i40e_napi_poll(struct napi_struct *napi, int budget)
2693 struct i40e_q_vector *q_vector =
2694 container_of(napi, struct i40e_q_vector, napi);
2695 struct i40e_vsi *vsi = q_vector->vsi;
2696 struct i40e_ring *ring;
2697 bool clean_complete = true;
2698 bool arm_wb = false;
2699 int budget_per_ring;
2702 if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2703 napi_complete(napi);
2707 /* Since the actual Tx work is minimal, we can give the Tx a larger
2708 * budget and be more aggressive about cleaning up the Tx descriptors.
2710 i40e_for_each_ring(ring, q_vector->tx) {
2711 bool wd = ring->xsk_pool ?
2712 i40e_clean_xdp_tx_irq(vsi, ring) :
2713 i40e_clean_tx_irq(vsi, ring, budget);
2716 clean_complete = false;
2719 arm_wb |= ring->arm_wb;
2720 ring->arm_wb = false;
2723 /* Handle case where we are called by netpoll with a budget of 0 */
2727 /* normally we have 1 Rx ring per q_vector */
2728 if (unlikely(q_vector->num_ringpairs > 1))
2729 /* We attempt to distribute budget to each Rx queue fairly, but
2730 * don't allow the budget to go below 1 because that would exit
2733 budget_per_ring = max_t(int, budget / q_vector->num_ringpairs, 1);
2735 /* Max of 1 Rx ring in this q_vector so give it the budget */
2736 budget_per_ring = budget;
2738 i40e_for_each_ring(ring, q_vector->rx) {
2739 int cleaned = ring->xsk_pool ?
2740 i40e_clean_rx_irq_zc(ring, budget_per_ring) :
2741 i40e_clean_rx_irq(ring, budget_per_ring);
2743 work_done += cleaned;
2744 /* if we clean as many as budgeted, we must not be done */
2745 if (cleaned >= budget_per_ring)
2746 clean_complete = false;
2749 /* If work not completed, return budget and polling will return */
2750 if (!clean_complete) {
2751 int cpu_id = smp_processor_id();
2753 /* It is possible that the interrupt affinity has changed but,
2754 * if the cpu is pegged at 100%, polling will never exit while
2755 * traffic continues and the interrupt will be stuck on this
2756 * cpu. We check to make sure affinity is correct before we
2757 * continue to poll, otherwise we must stop polling so the
2758 * interrupt can move to the correct cpu.
2760 if (!cpumask_test_cpu(cpu_id, &q_vector->affinity_mask)) {
2761 /* Tell napi that we are done polling */
2762 napi_complete_done(napi, work_done);
2764 /* Force an interrupt */
2765 i40e_force_wb(vsi, q_vector);
2767 /* Return budget-1 so that polling stops */
2772 q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2773 i40e_enable_wb_on_itr(vsi, q_vector);
2778 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2779 q_vector->arm_wb_state = false;
2781 /* Exit the polling mode, but don't re-enable interrupts if stack might
2782 * poll us due to busy-polling
2784 if (likely(napi_complete_done(napi, work_done)))
2785 i40e_update_enable_itr(vsi, q_vector);
2787 return min(work_done, budget - 1);
2791 * i40e_atr - Add a Flow Director ATR filter
2792 * @tx_ring: ring to add programming descriptor to
2794 * @tx_flags: send tx flags
2796 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2799 struct i40e_filter_program_desc *fdir_desc;
2800 struct i40e_pf *pf = tx_ring->vsi->back;
2802 unsigned char *network;
2804 struct ipv6hdr *ipv6;
2808 u32 flex_ptype, dtype_cmd;
2812 /* make sure ATR is enabled */
2813 if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2816 if (test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2819 /* if sampling is disabled do nothing */
2820 if (!tx_ring->atr_sample_rate)
2823 /* Currently only IPv4/IPv6 with TCP is supported */
2824 if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2827 /* snag network header to get L4 type and address */
2828 hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2829 skb_inner_network_header(skb) : skb_network_header(skb);
2831 /* Note: tx_flags gets modified to reflect inner protocols in
2832 * tx_enable_csum function if encap is enabled.
2834 if (tx_flags & I40E_TX_FLAGS_IPV4) {
2835 /* access ihl as u8 to avoid unaligned access on ia64 */
2836 hlen = (hdr.network[0] & 0x0F) << 2;
2837 l4_proto = hdr.ipv4->protocol;
2839 /* find the start of the innermost ipv6 header */
2840 unsigned int inner_hlen = hdr.network - skb->data;
2841 unsigned int h_offset = inner_hlen;
2843 /* this function updates h_offset to the end of the header */
2845 ipv6_find_hdr(skb, &h_offset, IPPROTO_TCP, NULL, NULL);
2846 /* hlen will contain our best estimate of the tcp header */
2847 hlen = h_offset - inner_hlen;
2850 if (l4_proto != IPPROTO_TCP)
2853 th = (struct tcphdr *)(hdr.network + hlen);
2855 /* Due to lack of space, no more new filters can be programmed */
2856 if (th->syn && test_bit(__I40E_FD_ATR_AUTO_DISABLED, pf->state))
2858 if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED) {
2859 /* HW ATR eviction will take care of removing filters on FIN
2862 if (th->fin || th->rst)
2866 tx_ring->atr_count++;
2868 /* sample on all syn/fin/rst packets or once every atr sample rate */
2872 (tx_ring->atr_count < tx_ring->atr_sample_rate))
2875 tx_ring->atr_count = 0;
2877 /* grab the next descriptor */
2878 i = tx_ring->next_to_use;
2879 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2882 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2884 flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2885 I40E_TXD_FLTR_QW0_QINDEX_MASK;
2886 flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2887 (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2888 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2889 (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2890 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2892 flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2894 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2896 dtype_cmd |= (th->fin || th->rst) ?
2897 (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2898 I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2899 (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2900 I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2902 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2903 I40E_TXD_FLTR_QW1_DEST_SHIFT;
2905 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2906 I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2908 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2909 if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2911 ((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2912 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2913 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2916 ((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2917 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2918 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2920 if (pf->flags & I40E_FLAG_HW_ATR_EVICT_ENABLED)
2921 dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2923 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2924 fdir_desc->rsvd = cpu_to_le32(0);
2925 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2926 fdir_desc->fd_id = cpu_to_le32(0);
2930 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2932 * @tx_ring: ring to send buffer on
2933 * @flags: the tx flags to be set
2935 * Checks the skb and set up correspondingly several generic transmit flags
2936 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2938 * Returns error code indicate the frame should be dropped upon error and the
2939 * otherwise returns 0 to indicate the flags has been set properly.
2941 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2942 struct i40e_ring *tx_ring,
2945 __be16 protocol = skb->protocol;
2948 if (protocol == htons(ETH_P_8021Q) &&
2949 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2950 /* When HW VLAN acceleration is turned off by the user the
2951 * stack sets the protocol to 8021q so that the driver
2952 * can take any steps required to support the SW only
2953 * VLAN handling. In our case the driver doesn't need
2954 * to take any further steps so just set the protocol
2955 * to the encapsulated ethertype.
2957 skb->protocol = vlan_get_protocol(skb);
2961 /* if we have a HW VLAN tag being added, default to the HW one */
2962 if (skb_vlan_tag_present(skb)) {
2963 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2964 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2965 /* else if it is a SW VLAN, check the next protocol and store the tag */
2966 } else if (protocol == htons(ETH_P_8021Q)) {
2967 struct vlan_hdr *vhdr, _vhdr;
2969 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2973 protocol = vhdr->h_vlan_encapsulated_proto;
2974 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2975 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2978 if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2981 /* Insert 802.1p priority into VLAN header */
2982 if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2983 (skb->priority != TC_PRIO_CONTROL)) {
2984 tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2985 tx_flags |= (skb->priority & 0x7) <<
2986 I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2987 if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2988 struct vlan_ethhdr *vhdr;
2991 rc = skb_cow_head(skb, 0);
2994 vhdr = (struct vlan_ethhdr *)skb->data;
2995 vhdr->h_vlan_TCI = htons(tx_flags >>
2996 I40E_TX_FLAGS_VLAN_SHIFT);
2998 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
3008 * i40e_tso - set up the tso context descriptor
3009 * @first: pointer to first Tx buffer for xmit
3010 * @hdr_len: ptr to the size of the packet header
3011 * @cd_type_cmd_tso_mss: Quad Word 1
3013 * Returns 0 if no TSO can happen, 1 if tso is going, or error
3015 static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
3016 u64 *cd_type_cmd_tso_mss)
3018 struct sk_buff *skb = first->skb;
3019 u64 cd_cmd, cd_tso_len, cd_mss;
3030 u32 paylen, l4_offset;
3031 u16 gso_segs, gso_size;
3034 if (skb->ip_summed != CHECKSUM_PARTIAL)
3037 if (!skb_is_gso(skb))
3040 err = skb_cow_head(skb, 0);
3044 ip.hdr = skb_network_header(skb);
3045 l4.hdr = skb_transport_header(skb);
3047 /* initialize outer IP header fields */
3048 if (ip.v4->version == 4) {
3052 ip.v6->payload_len = 0;
3055 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
3059 SKB_GSO_UDP_TUNNEL |
3060 SKB_GSO_UDP_TUNNEL_CSUM)) {
3061 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3062 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
3065 /* determine offset of outer transport header */
3066 l4_offset = l4.hdr - skb->data;
3068 /* remove payload length from outer checksum */
3069 paylen = skb->len - l4_offset;
3070 csum_replace_by_diff(&l4.udp->check,
3071 (__force __wsum)htonl(paylen));
3074 /* reset pointers to inner headers */
3075 ip.hdr = skb_inner_network_header(skb);
3076 l4.hdr = skb_inner_transport_header(skb);
3078 /* initialize inner IP header fields */
3079 if (ip.v4->version == 4) {
3083 ip.v6->payload_len = 0;
3087 /* determine offset of inner transport header */
3088 l4_offset = l4.hdr - skb->data;
3090 /* remove payload length from inner checksum */
3091 paylen = skb->len - l4_offset;
3093 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
3094 csum_replace_by_diff(&l4.udp->check, (__force __wsum)htonl(paylen));
3095 /* compute length of segmentation header */
3096 *hdr_len = sizeof(*l4.udp) + l4_offset;
3098 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3099 /* compute length of segmentation header */
3100 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3103 /* pull values out of skb_shinfo */
3104 gso_size = skb_shinfo(skb)->gso_size;
3105 gso_segs = skb_shinfo(skb)->gso_segs;
3107 /* update GSO size and bytecount with header size */
3108 first->gso_segs = gso_segs;
3109 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3111 /* find the field values */
3112 cd_cmd = I40E_TX_CTX_DESC_TSO;
3113 cd_tso_len = skb->len - *hdr_len;
3115 *cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
3116 (cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
3117 (cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
3122 * i40e_tsyn - set up the tsyn context descriptor
3123 * @tx_ring: ptr to the ring to send
3124 * @skb: ptr to the skb we're sending
3125 * @tx_flags: the collected send information
3126 * @cd_type_cmd_tso_mss: Quad Word 1
3128 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
3130 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
3131 u32 tx_flags, u64 *cd_type_cmd_tso_mss)
3135 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
3138 /* Tx timestamps cannot be sampled when doing TSO */
3139 if (tx_flags & I40E_TX_FLAGS_TSO)
3142 /* only timestamp the outbound packet if the user has requested it and
3143 * we are not already transmitting a packet to be timestamped
3145 pf = i40e_netdev_to_pf(tx_ring->netdev);
3146 if (!(pf->flags & I40E_FLAG_PTP))
3150 !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
3151 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
3152 pf->ptp_tx_start = jiffies;
3153 pf->ptp_tx_skb = skb_get(skb);
3155 pf->tx_hwtstamp_skipped++;
3159 *cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
3160 I40E_TXD_CTX_QW1_CMD_SHIFT;
3166 * i40e_tx_enable_csum - Enable Tx checksum offloads
3168 * @tx_flags: pointer to Tx flags currently set
3169 * @td_cmd: Tx descriptor command bits to set
3170 * @td_offset: Tx descriptor header offsets to set
3171 * @tx_ring: Tx descriptor ring
3172 * @cd_tunneling: ptr to context desc bits
3174 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
3175 u32 *td_cmd, u32 *td_offset,
3176 struct i40e_ring *tx_ring,
3189 unsigned char *exthdr;
3190 u32 offset, cmd = 0;
3194 if (skb->ip_summed != CHECKSUM_PARTIAL)
3197 ip.hdr = skb_network_header(skb);
3198 l4.hdr = skb_transport_header(skb);
3200 /* compute outer L2 header size */
3201 offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
3203 if (skb->encapsulation) {
3205 /* define outer network header type */
3206 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3207 tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3208 I40E_TX_CTX_EXT_IP_IPV4 :
3209 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
3211 l4_proto = ip.v4->protocol;
3212 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3215 tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
3217 exthdr = ip.hdr + sizeof(*ip.v6);
3218 l4_proto = ip.v6->nexthdr;
3219 ret = ipv6_skip_exthdr(skb, exthdr - skb->data,
3220 &l4_proto, &frag_off);
3225 /* define outer transport */
3228 tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
3229 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3232 tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
3233 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3237 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
3238 l4.hdr = skb_inner_network_header(skb);
3241 if (*tx_flags & I40E_TX_FLAGS_TSO)
3244 skb_checksum_help(skb);
3248 /* compute outer L3 header size */
3249 tunnel |= ((l4.hdr - ip.hdr) / 4) <<
3250 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
3252 /* switch IP header pointer from outer to inner header */
3253 ip.hdr = skb_inner_network_header(skb);
3255 /* compute tunnel header size */
3256 tunnel |= ((ip.hdr - l4.hdr) / 2) <<
3257 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
3259 /* indicate if we need to offload outer UDP header */
3260 if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
3261 !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
3262 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
3263 tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
3265 /* record tunnel offload values */
3266 *cd_tunneling |= tunnel;
3268 /* switch L4 header pointer from outer to inner */
3269 l4.hdr = skb_inner_transport_header(skb);
3272 /* reset type as we transition from outer to inner headers */
3273 *tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
3274 if (ip.v4->version == 4)
3275 *tx_flags |= I40E_TX_FLAGS_IPV4;
3276 if (ip.v6->version == 6)
3277 *tx_flags |= I40E_TX_FLAGS_IPV6;
3280 /* Enable IP checksum offloads */
3281 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
3282 l4_proto = ip.v4->protocol;
3283 /* the stack computes the IP header already, the only time we
3284 * need the hardware to recompute it is in the case of TSO.
3286 cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
3287 I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
3288 I40E_TX_DESC_CMD_IIPT_IPV4;
3289 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
3290 cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
3292 exthdr = ip.hdr + sizeof(*ip.v6);
3293 l4_proto = ip.v6->nexthdr;
3294 if (l4.hdr != exthdr)
3295 ipv6_skip_exthdr(skb, exthdr - skb->data,
3296 &l4_proto, &frag_off);
3299 /* compute inner L3 header size */
3300 offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
3302 /* Enable L4 checksum offloads */
3305 /* enable checksum offloads */
3306 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
3307 offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3310 /* enable SCTP checksum offload */
3311 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
3312 offset |= (sizeof(struct sctphdr) >> 2) <<
3313 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3316 /* enable UDP checksum offload */
3317 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
3318 offset |= (sizeof(struct udphdr) >> 2) <<
3319 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
3322 if (*tx_flags & I40E_TX_FLAGS_TSO)
3324 skb_checksum_help(skb);
3329 *td_offset |= offset;
3335 * i40e_create_tx_ctx - Build the Tx context descriptor
3336 * @tx_ring: ring to create the descriptor on
3337 * @cd_type_cmd_tso_mss: Quad Word 1
3338 * @cd_tunneling: Quad Word 0 - bits 0-31
3339 * @cd_l2tag2: Quad Word 0 - bits 32-63
3341 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
3342 const u64 cd_type_cmd_tso_mss,
3343 const u32 cd_tunneling, const u32 cd_l2tag2)
3345 struct i40e_tx_context_desc *context_desc;
3346 int i = tx_ring->next_to_use;
3348 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
3349 !cd_tunneling && !cd_l2tag2)
3352 /* grab the next descriptor */
3353 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
3356 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3358 /* cpu_to_le32 and assign to struct fields */
3359 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
3360 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
3361 context_desc->rsvd = cpu_to_le16(0);
3362 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
3366 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
3367 * @tx_ring: the ring to be checked
3368 * @size: the size buffer we want to assure is available
3370 * Returns -EBUSY if a stop is needed, else 0
3372 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
3374 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3375 /* Memory barrier before checking head and tail */
3378 /* Check again in a case another CPU has just made room available. */
3379 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
3382 /* A reprieve! - use start_queue because it doesn't call schedule */
3383 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3384 ++tx_ring->tx_stats.restart_queue;
3389 * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
3392 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
3393 * and so we need to figure out the cases where we need to linearize the skb.
3395 * For TSO we need to count the TSO header and segment payload separately.
3396 * As such we need to check cases where we have 7 fragments or more as we
3397 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
3398 * the segment payload in the first descriptor, and another 7 for the
3401 bool __i40e_chk_linearize(struct sk_buff *skb)
3403 const skb_frag_t *frag, *stale;
3406 /* no need to check if number of frags is less than 7 */
3407 nr_frags = skb_shinfo(skb)->nr_frags;
3408 if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
3411 /* We need to walk through the list and validate that each group
3412 * of 6 fragments totals at least gso_size.
3414 nr_frags -= I40E_MAX_BUFFER_TXD - 2;
3415 frag = &skb_shinfo(skb)->frags[0];
3417 /* Initialize size to the negative value of gso_size minus 1. We
3418 * use this as the worst case scenerio in which the frag ahead
3419 * of us only provides one byte which is why we are limited to 6
3420 * descriptors for a single transmit as the header and previous
3421 * fragment are already consuming 2 descriptors.
3423 sum = 1 - skb_shinfo(skb)->gso_size;
3425 /* Add size of frags 0 through 4 to create our initial sum */
3426 sum += skb_frag_size(frag++);
3427 sum += skb_frag_size(frag++);
3428 sum += skb_frag_size(frag++);
3429 sum += skb_frag_size(frag++);
3430 sum += skb_frag_size(frag++);
3432 /* Walk through fragments adding latest fragment, testing it, and
3433 * then removing stale fragments from the sum.
3435 for (stale = &skb_shinfo(skb)->frags[0];; stale++) {
3436 int stale_size = skb_frag_size(stale);
3438 sum += skb_frag_size(frag++);
3440 /* The stale fragment may present us with a smaller
3441 * descriptor than the actual fragment size. To account
3442 * for that we need to remove all the data on the front and
3443 * figure out what the remainder would be in the last
3444 * descriptor associated with the fragment.
3446 if (stale_size > I40E_MAX_DATA_PER_TXD) {
3447 int align_pad = -(skb_frag_off(stale)) &
3448 (I40E_MAX_READ_REQ_SIZE - 1);
3451 stale_size -= align_pad;
3454 sum -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3455 stale_size -= I40E_MAX_DATA_PER_TXD_ALIGNED;
3456 } while (stale_size > I40E_MAX_DATA_PER_TXD);
3459 /* if sum is negative we failed to make sufficient progress */
3473 * i40e_tx_map - Build the Tx descriptor
3474 * @tx_ring: ring to send buffer on
3476 * @first: first buffer info buffer to use
3477 * @tx_flags: collected send information
3478 * @hdr_len: size of the packet header
3479 * @td_cmd: the command field in the descriptor
3480 * @td_offset: offset for checksum or crc
3482 * Returns 0 on success, -1 on failure to DMA
3484 static inline int i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
3485 struct i40e_tx_buffer *first, u32 tx_flags,
3486 const u8 hdr_len, u32 td_cmd, u32 td_offset)
3488 unsigned int data_len = skb->data_len;
3489 unsigned int size = skb_headlen(skb);
3491 struct i40e_tx_buffer *tx_bi;
3492 struct i40e_tx_desc *tx_desc;
3493 u16 i = tx_ring->next_to_use;
3498 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
3499 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
3500 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
3501 I40E_TX_FLAGS_VLAN_SHIFT;
3504 first->tx_flags = tx_flags;
3506 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3508 tx_desc = I40E_TX_DESC(tx_ring, i);
3511 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3512 unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3514 if (dma_mapping_error(tx_ring->dev, dma))
3517 /* record length, and DMA address */
3518 dma_unmap_len_set(tx_bi, len, size);
3519 dma_unmap_addr_set(tx_bi, dma, dma);
3521 /* align size to end of page */
3522 max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
3523 tx_desc->buffer_addr = cpu_to_le64(dma);
3525 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
3526 tx_desc->cmd_type_offset_bsz =
3527 build_ctob(td_cmd, td_offset,
3534 if (i == tx_ring->count) {
3535 tx_desc = I40E_TX_DESC(tx_ring, 0);
3542 max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
3543 tx_desc->buffer_addr = cpu_to_le64(dma);
3546 if (likely(!data_len))
3549 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3556 if (i == tx_ring->count) {
3557 tx_desc = I40E_TX_DESC(tx_ring, 0);
3561 size = skb_frag_size(frag);
3564 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3567 tx_bi = &tx_ring->tx_bi[i];
3570 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3573 if (i == tx_ring->count)
3576 tx_ring->next_to_use = i;
3578 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3580 /* write last descriptor with EOP bit */
3581 td_cmd |= I40E_TX_DESC_CMD_EOP;
3583 /* We OR these values together to check both against 4 (WB_STRIDE)
3584 * below. This is safe since we don't re-use desc_count afterwards.
3586 desc_count |= ++tx_ring->packet_stride;
3588 if (desc_count >= WB_STRIDE) {
3589 /* write last descriptor with RS bit set */
3590 td_cmd |= I40E_TX_DESC_CMD_RS;
3591 tx_ring->packet_stride = 0;
3594 tx_desc->cmd_type_offset_bsz =
3595 build_ctob(td_cmd, td_offset, size, td_tag);
3597 skb_tx_timestamp(skb);
3599 /* Force memory writes to complete before letting h/w know there
3600 * are new descriptors to fetch.
3602 * We also use this memory barrier to make certain all of the
3603 * status bits have been updated before next_to_watch is written.
3607 /* set next_to_watch value indicating a packet is present */
3608 first->next_to_watch = tx_desc;
3610 /* notify HW of packet */
3611 if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
3612 writel(i, tx_ring->tail);
3618 dev_info(tx_ring->dev, "TX DMA map failed\n");
3620 /* clear dma mappings for failed tx_bi map */
3622 tx_bi = &tx_ring->tx_bi[i];
3623 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3631 tx_ring->next_to_use = i;
3637 * i40e_xmit_xdp_ring - transmits an XDP buffer to an XDP Tx ring
3638 * @xdpf: data to transmit
3639 * @xdp_ring: XDP Tx ring
3641 static int i40e_xmit_xdp_ring(struct xdp_frame *xdpf,
3642 struct i40e_ring *xdp_ring)
3644 u16 i = xdp_ring->next_to_use;
3645 struct i40e_tx_buffer *tx_bi;
3646 struct i40e_tx_desc *tx_desc;
3647 void *data = xdpf->data;
3648 u32 size = xdpf->len;
3651 if (!unlikely(I40E_DESC_UNUSED(xdp_ring))) {
3652 xdp_ring->tx_stats.tx_busy++;
3653 return I40E_XDP_CONSUMED;
3655 dma = dma_map_single(xdp_ring->dev, data, size, DMA_TO_DEVICE);
3656 if (dma_mapping_error(xdp_ring->dev, dma))
3657 return I40E_XDP_CONSUMED;
3659 tx_bi = &xdp_ring->tx_bi[i];
3660 tx_bi->bytecount = size;
3661 tx_bi->gso_segs = 1;
3664 /* record length, and DMA address */
3665 dma_unmap_len_set(tx_bi, len, size);
3666 dma_unmap_addr_set(tx_bi, dma, dma);
3668 tx_desc = I40E_TX_DESC(xdp_ring, i);
3669 tx_desc->buffer_addr = cpu_to_le64(dma);
3670 tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC
3674 /* Make certain all of the status bits have been updated
3675 * before next_to_watch is written.
3679 xdp_ring->xdp_tx_active++;
3681 if (i == xdp_ring->count)
3684 tx_bi->next_to_watch = tx_desc;
3685 xdp_ring->next_to_use = i;
3691 * i40e_xmit_frame_ring - Sends buffer on Tx ring
3693 * @tx_ring: ring to send buffer on
3695 * Returns NETDEV_TX_OK if sent, else an error code
3697 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3698 struct i40e_ring *tx_ring)
3700 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3701 u32 cd_tunneling = 0, cd_l2tag2 = 0;
3702 struct i40e_tx_buffer *first;
3711 /* prefetch the data, we'll need it later */
3712 prefetch(skb->data);
3714 i40e_trace(xmit_frame_ring, skb, tx_ring);
3716 count = i40e_xmit_descriptor_count(skb);
3717 if (i40e_chk_linearize(skb, count)) {
3718 if (__skb_linearize(skb)) {
3719 dev_kfree_skb_any(skb);
3720 return NETDEV_TX_OK;
3722 count = i40e_txd_use_count(skb->len);
3723 tx_ring->tx_stats.tx_linearize++;
3726 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3727 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3728 * + 4 desc gap to avoid the cache line where head is,
3729 * + 1 desc for context descriptor,
3730 * otherwise try next time
3732 if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3733 tx_ring->tx_stats.tx_busy++;
3734 return NETDEV_TX_BUSY;
3737 /* record the location of the first descriptor for this packet */
3738 first = &tx_ring->tx_bi[tx_ring->next_to_use];
3740 first->bytecount = skb->len;
3741 first->gso_segs = 1;
3743 /* prepare the xmit flags */
3744 if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3747 /* obtain protocol of skb */
3748 protocol = vlan_get_protocol(skb);
3750 /* setup IPv4/IPv6 offloads */
3751 if (protocol == htons(ETH_P_IP))
3752 tx_flags |= I40E_TX_FLAGS_IPV4;
3753 else if (protocol == htons(ETH_P_IPV6))
3754 tx_flags |= I40E_TX_FLAGS_IPV6;
3756 tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3761 tx_flags |= I40E_TX_FLAGS_TSO;
3763 /* Always offload the checksum, since it's in the data descriptor */
3764 tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3765 tx_ring, &cd_tunneling);
3769 tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3772 tx_flags |= I40E_TX_FLAGS_TSYN;
3774 /* always enable CRC insertion offload */
3775 td_cmd |= I40E_TX_DESC_CMD_ICRC;
3777 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3778 cd_tunneling, cd_l2tag2);
3780 /* Add Flow Director ATR if it's enabled.
3782 * NOTE: this must always be directly before the data descriptor.
3784 i40e_atr(tx_ring, skb, tx_flags);
3786 if (i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3788 goto cleanup_tx_tstamp;
3790 return NETDEV_TX_OK;
3793 i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3794 dev_kfree_skb_any(first->skb);
3797 if (unlikely(tx_flags & I40E_TX_FLAGS_TSYN)) {
3798 struct i40e_pf *pf = i40e_netdev_to_pf(tx_ring->netdev);
3800 dev_kfree_skb_any(pf->ptp_tx_skb);
3801 pf->ptp_tx_skb = NULL;
3802 clear_bit_unlock(__I40E_PTP_TX_IN_PROGRESS, pf->state);
3805 return NETDEV_TX_OK;
3809 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3811 * @netdev: network interface device structure
3813 * Returns NETDEV_TX_OK if sent, else an error code
3815 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3817 struct i40e_netdev_priv *np = netdev_priv(netdev);
3818 struct i40e_vsi *vsi = np->vsi;
3819 struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3821 /* hardware can't handle really short frames, hardware padding works
3824 if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3825 return NETDEV_TX_OK;
3827 return i40e_xmit_frame_ring(skb, tx_ring);
3831 * i40e_xdp_xmit - Implements ndo_xdp_xmit
3833 * @n: number of frames
3834 * @frames: array of XDP buffer pointers
3835 * @flags: XDP extra info
3837 * Returns number of frames successfully sent. Failed frames
3838 * will be free'ed by XDP core.
3840 * For error cases, a negative errno code is returned and no-frames
3841 * are transmitted (caller must handle freeing frames).
3843 int i40e_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames,
3846 struct i40e_netdev_priv *np = netdev_priv(dev);
3847 unsigned int queue_index = smp_processor_id();
3848 struct i40e_vsi *vsi = np->vsi;
3849 struct i40e_pf *pf = vsi->back;
3850 struct i40e_ring *xdp_ring;
3854 if (test_bit(__I40E_VSI_DOWN, vsi->state))
3857 if (!i40e_enabled_xdp_vsi(vsi) || queue_index >= vsi->num_queue_pairs ||
3858 test_bit(__I40E_CONFIG_BUSY, pf->state))
3861 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
3864 xdp_ring = vsi->xdp_rings[queue_index];
3866 for (i = 0; i < n; i++) {
3867 struct xdp_frame *xdpf = frames[i];
3870 err = i40e_xmit_xdp_ring(xdpf, xdp_ring);
3871 if (err != I40E_XDP_TX)
3876 if (unlikely(flags & XDP_XMIT_FLUSH))
3877 i40e_xdp_ring_update_tail(xdp_ring);