1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Driver
4 * Copyright(c) 2013 - 2016 Intel Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms and conditions of the GNU General Public License,
8 * version 2, as published by the Free Software Foundation.
10 * This program is distributed in the hope it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along
16 * with this program. If not, see <http://www.gnu.org/licenses/>.
18 * The full GNU General Public License is included in this distribution in
19 * the file called "COPYING".
21 * Contact Information:
22 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 ******************************************************************************/
27 #include <linux/prefetch.h>
28 #include <net/busy_poll.h>
30 #include "i40e_trace.h"
31 #include "i40e_prototype.h"
33 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
36 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
37 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
38 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
39 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
40 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
43 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
45 * i40e_fdir - Generate a Flow Director descriptor based on fdata
46 * @tx_ring: Tx ring to send buffer on
47 * @fdata: Flow director filter data
48 * @add: Indicate if we are adding a rule or deleting one
51 static void i40e_fdir(struct i40e_ring *tx_ring,
52 struct i40e_fdir_filter *fdata, bool add)
54 struct i40e_filter_program_desc *fdir_desc;
55 struct i40e_pf *pf = tx_ring->vsi->back;
56 u32 flex_ptype, dtype_cmd;
59 /* grab the next descriptor */
60 i = tx_ring->next_to_use;
61 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
64 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
66 flex_ptype = I40E_TXD_FLTR_QW0_QINDEX_MASK &
67 (fdata->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT);
69 flex_ptype |= I40E_TXD_FLTR_QW0_FLEXOFF_MASK &
70 (fdata->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
72 flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
73 (fdata->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
75 flex_ptype |= I40E_TXD_FLTR_QW0_PCTYPE_MASK &
76 (fdata->flex_offset << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT);
78 /* Use LAN VSI Id if not programmed by user */
79 flex_ptype |= I40E_TXD_FLTR_QW0_DEST_VSI_MASK &
80 ((u32)(fdata->dest_vsi ? : pf->vsi[pf->lan_vsi]->id) <<
81 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT);
83 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
86 I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
87 I40E_TXD_FLTR_QW1_PCMD_SHIFT :
88 I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
89 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
91 dtype_cmd |= I40E_TXD_FLTR_QW1_DEST_MASK &
92 (fdata->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT);
94 dtype_cmd |= I40E_TXD_FLTR_QW1_FD_STATUS_MASK &
95 (fdata->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT);
97 if (fdata->cnt_index) {
98 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
99 dtype_cmd |= I40E_TXD_FLTR_QW1_CNTINDEX_MASK &
100 ((u32)fdata->cnt_index <<
101 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT);
104 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
105 fdir_desc->rsvd = cpu_to_le32(0);
106 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
107 fdir_desc->fd_id = cpu_to_le32(fdata->fd_id);
110 #define I40E_FD_CLEAN_DELAY 10
112 * i40e_program_fdir_filter - Program a Flow Director filter
113 * @fdir_data: Packet data that will be filter parameters
114 * @raw_packet: the pre-allocated packet buffer for FDir
115 * @pf: The PF pointer
116 * @add: True for add/update, False for remove
118 static int i40e_program_fdir_filter(struct i40e_fdir_filter *fdir_data,
119 u8 *raw_packet, struct i40e_pf *pf,
122 struct i40e_tx_buffer *tx_buf, *first;
123 struct i40e_tx_desc *tx_desc;
124 struct i40e_ring *tx_ring;
125 struct i40e_vsi *vsi;
131 /* find existing FDIR VSI */
132 vsi = i40e_find_vsi_by_type(pf, I40E_VSI_FDIR);
136 tx_ring = vsi->tx_rings[0];
139 /* we need two descriptors to add/del a filter and we can wait */
140 for (i = I40E_FD_CLEAN_DELAY; I40E_DESC_UNUSED(tx_ring) < 2; i--) {
143 msleep_interruptible(1);
146 dma = dma_map_single(dev, raw_packet,
147 I40E_FDIR_MAX_RAW_PACKET_SIZE, DMA_TO_DEVICE);
148 if (dma_mapping_error(dev, dma))
151 /* grab the next descriptor */
152 i = tx_ring->next_to_use;
153 first = &tx_ring->tx_bi[i];
154 i40e_fdir(tx_ring, fdir_data, add);
156 /* Now program a dummy descriptor */
157 i = tx_ring->next_to_use;
158 tx_desc = I40E_TX_DESC(tx_ring, i);
159 tx_buf = &tx_ring->tx_bi[i];
161 tx_ring->next_to_use = ((i + 1) < tx_ring->count) ? i + 1 : 0;
163 memset(tx_buf, 0, sizeof(struct i40e_tx_buffer));
165 /* record length, and DMA address */
166 dma_unmap_len_set(tx_buf, len, I40E_FDIR_MAX_RAW_PACKET_SIZE);
167 dma_unmap_addr_set(tx_buf, dma, dma);
169 tx_desc->buffer_addr = cpu_to_le64(dma);
170 td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
172 tx_buf->tx_flags = I40E_TX_FLAGS_FD_SB;
173 tx_buf->raw_buf = (void *)raw_packet;
175 tx_desc->cmd_type_offset_bsz =
176 build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_SIZE, 0);
178 /* Force memory writes to complete before letting h/w
179 * know there are new descriptors to fetch.
183 /* Mark the data descriptor to be watched */
184 first->next_to_watch = tx_desc;
186 writel(tx_ring->next_to_use, tx_ring->tail);
193 #define IP_HEADER_OFFSET 14
194 #define I40E_UDPIP_DUMMY_PACKET_LEN 42
196 * i40e_add_del_fdir_udpv4 - Add/Remove UDPv4 filters
197 * @vsi: pointer to the targeted VSI
198 * @fd_data: the flow director data required for the FDir descriptor
199 * @add: true adds a filter, false removes it
201 * Returns 0 if the filters were successfully added or removed
203 static int i40e_add_del_fdir_udpv4(struct i40e_vsi *vsi,
204 struct i40e_fdir_filter *fd_data,
207 struct i40e_pf *pf = vsi->back;
212 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
213 0x45, 0, 0, 0x1c, 0, 0, 0x40, 0, 0x40, 0x11, 0, 0, 0, 0, 0, 0,
214 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
216 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
219 memcpy(raw_packet, packet, I40E_UDPIP_DUMMY_PACKET_LEN);
221 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
222 udp = (struct udphdr *)(raw_packet + IP_HEADER_OFFSET
223 + sizeof(struct iphdr));
225 ip->daddr = fd_data->dst_ip;
226 udp->dest = fd_data->dst_port;
227 ip->saddr = fd_data->src_ip;
228 udp->source = fd_data->src_port;
230 if (fd_data->flex_filter) {
231 u8 *payload = raw_packet + I40E_UDPIP_DUMMY_PACKET_LEN;
232 __be16 pattern = fd_data->flex_word;
233 u16 off = fd_data->flex_offset;
235 *((__force __be16 *)(payload + off)) = pattern;
238 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_UDP;
239 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
241 dev_info(&pf->pdev->dev,
242 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
243 fd_data->pctype, fd_data->fd_id, ret);
244 /* Free the packet buffer since it wasn't added to the ring */
247 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
249 dev_info(&pf->pdev->dev,
250 "Filter OK for PCTYPE %d loc = %d\n",
251 fd_data->pctype, fd_data->fd_id);
253 dev_info(&pf->pdev->dev,
254 "Filter deleted for PCTYPE %d loc = %d\n",
255 fd_data->pctype, fd_data->fd_id);
259 pf->fd_udp4_filter_cnt++;
261 pf->fd_udp4_filter_cnt--;
266 #define I40E_TCPIP_DUMMY_PACKET_LEN 54
268 * i40e_add_del_fdir_tcpv4 - Add/Remove TCPv4 filters
269 * @vsi: pointer to the targeted VSI
270 * @fd_data: the flow director data required for the FDir descriptor
271 * @add: true adds a filter, false removes it
273 * Returns 0 if the filters were successfully added or removed
275 static int i40e_add_del_fdir_tcpv4(struct i40e_vsi *vsi,
276 struct i40e_fdir_filter *fd_data,
279 struct i40e_pf *pf = vsi->back;
285 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
286 0x45, 0, 0, 0x28, 0, 0, 0x40, 0, 0x40, 0x6, 0, 0, 0, 0, 0, 0,
287 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x80, 0x11,
288 0x0, 0x72, 0, 0, 0, 0};
290 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
293 memcpy(raw_packet, packet, I40E_TCPIP_DUMMY_PACKET_LEN);
295 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
296 tcp = (struct tcphdr *)(raw_packet + IP_HEADER_OFFSET
297 + sizeof(struct iphdr));
299 ip->daddr = fd_data->dst_ip;
300 tcp->dest = fd_data->dst_port;
301 ip->saddr = fd_data->src_ip;
302 tcp->source = fd_data->src_port;
304 if (fd_data->flex_filter) {
305 u8 *payload = raw_packet + I40E_TCPIP_DUMMY_PACKET_LEN;
306 __be16 pattern = fd_data->flex_word;
307 u16 off = fd_data->flex_offset;
309 *((__force __be16 *)(payload + off)) = pattern;
312 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_TCP;
313 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
315 dev_info(&pf->pdev->dev,
316 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
317 fd_data->pctype, fd_data->fd_id, ret);
318 /* Free the packet buffer since it wasn't added to the ring */
321 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
323 dev_info(&pf->pdev->dev, "Filter OK for PCTYPE %d loc = %d)\n",
324 fd_data->pctype, fd_data->fd_id);
326 dev_info(&pf->pdev->dev,
327 "Filter deleted for PCTYPE %d loc = %d\n",
328 fd_data->pctype, fd_data->fd_id);
332 pf->fd_tcp4_filter_cnt++;
333 if ((pf->flags & I40E_FLAG_FD_ATR_ENABLED) &&
334 I40E_DEBUG_FD & pf->hw.debug_mask)
335 dev_info(&pf->pdev->dev, "Forcing ATR off, sideband rules for TCP/IPv4 flow being applied\n");
336 pf->flags |= I40E_FLAG_FD_ATR_AUTO_DISABLED;
338 pf->fd_tcp4_filter_cnt--;
344 #define I40E_SCTPIP_DUMMY_PACKET_LEN 46
346 * i40e_add_del_fdir_sctpv4 - Add/Remove SCTPv4 Flow Director filters for
347 * a specific flow spec
348 * @vsi: pointer to the targeted VSI
349 * @fd_data: the flow director data required for the FDir descriptor
350 * @add: true adds a filter, false removes it
352 * Returns 0 if the filters were successfully added or removed
354 static int i40e_add_del_fdir_sctpv4(struct i40e_vsi *vsi,
355 struct i40e_fdir_filter *fd_data,
358 struct i40e_pf *pf = vsi->back;
359 struct sctphdr *sctp;
364 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
365 0x45, 0, 0, 0x20, 0, 0, 0x40, 0, 0x40, 0x84, 0, 0, 0, 0, 0, 0,
366 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
368 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
371 memcpy(raw_packet, packet, I40E_SCTPIP_DUMMY_PACKET_LEN);
373 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
374 sctp = (struct sctphdr *)(raw_packet + IP_HEADER_OFFSET
375 + sizeof(struct iphdr));
377 ip->daddr = fd_data->dst_ip;
378 sctp->dest = fd_data->dst_port;
379 ip->saddr = fd_data->src_ip;
380 sctp->source = fd_data->src_port;
382 if (fd_data->flex_filter) {
383 u8 *payload = raw_packet + I40E_SCTPIP_DUMMY_PACKET_LEN;
384 __be16 pattern = fd_data->flex_word;
385 u16 off = fd_data->flex_offset;
387 *((__force __be16 *)(payload + off)) = pattern;
390 fd_data->pctype = I40E_FILTER_PCTYPE_NONF_IPV4_SCTP;
391 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
393 dev_info(&pf->pdev->dev,
394 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
395 fd_data->pctype, fd_data->fd_id, ret);
396 /* Free the packet buffer since it wasn't added to the ring */
399 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
401 dev_info(&pf->pdev->dev,
402 "Filter OK for PCTYPE %d loc = %d\n",
403 fd_data->pctype, fd_data->fd_id);
405 dev_info(&pf->pdev->dev,
406 "Filter deleted for PCTYPE %d loc = %d\n",
407 fd_data->pctype, fd_data->fd_id);
411 pf->fd_sctp4_filter_cnt++;
413 pf->fd_sctp4_filter_cnt--;
418 #define I40E_IP_DUMMY_PACKET_LEN 34
420 * i40e_add_del_fdir_ipv4 - Add/Remove IPv4 Flow Director filters for
421 * a specific flow spec
422 * @vsi: pointer to the targeted VSI
423 * @fd_data: the flow director data required for the FDir descriptor
424 * @add: true adds a filter, false removes it
426 * Returns 0 if the filters were successfully added or removed
428 static int i40e_add_del_fdir_ipv4(struct i40e_vsi *vsi,
429 struct i40e_fdir_filter *fd_data,
432 struct i40e_pf *pf = vsi->back;
437 static char packet[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0,
438 0x45, 0, 0, 0x14, 0, 0, 0x40, 0, 0x40, 0x10, 0, 0, 0, 0, 0, 0,
441 for (i = I40E_FILTER_PCTYPE_NONF_IPV4_OTHER;
442 i <= I40E_FILTER_PCTYPE_FRAG_IPV4; i++) {
443 raw_packet = kzalloc(I40E_FDIR_MAX_RAW_PACKET_SIZE, GFP_KERNEL);
446 memcpy(raw_packet, packet, I40E_IP_DUMMY_PACKET_LEN);
447 ip = (struct iphdr *)(raw_packet + IP_HEADER_OFFSET);
449 ip->saddr = fd_data->src_ip;
450 ip->daddr = fd_data->dst_ip;
453 if (fd_data->flex_filter) {
454 u8 *payload = raw_packet + I40E_IP_DUMMY_PACKET_LEN;
455 __be16 pattern = fd_data->flex_word;
456 u16 off = fd_data->flex_offset;
458 *((__force __be16 *)(payload + off)) = pattern;
462 ret = i40e_program_fdir_filter(fd_data, raw_packet, pf, add);
464 dev_info(&pf->pdev->dev,
465 "PCTYPE:%d, Filter command send failed for fd_id:%d (ret = %d)\n",
466 fd_data->pctype, fd_data->fd_id, ret);
467 /* The packet buffer wasn't added to the ring so we
468 * need to free it now.
472 } else if (I40E_DEBUG_FD & pf->hw.debug_mask) {
474 dev_info(&pf->pdev->dev,
475 "Filter OK for PCTYPE %d loc = %d\n",
476 fd_data->pctype, fd_data->fd_id);
478 dev_info(&pf->pdev->dev,
479 "Filter deleted for PCTYPE %d loc = %d\n",
480 fd_data->pctype, fd_data->fd_id);
485 pf->fd_ip4_filter_cnt++;
487 pf->fd_ip4_filter_cnt--;
493 * i40e_add_del_fdir - Build raw packets to add/del fdir filter
494 * @vsi: pointer to the targeted VSI
495 * @cmd: command to get or set RX flow classification rules
496 * @add: true adds a filter, false removes it
499 int i40e_add_del_fdir(struct i40e_vsi *vsi,
500 struct i40e_fdir_filter *input, bool add)
502 struct i40e_pf *pf = vsi->back;
505 switch (input->flow_type & ~FLOW_EXT) {
507 ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
510 ret = i40e_add_del_fdir_udpv4(vsi, input, add);
513 ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
516 switch (input->ip4_proto) {
518 ret = i40e_add_del_fdir_tcpv4(vsi, input, add);
521 ret = i40e_add_del_fdir_udpv4(vsi, input, add);
524 ret = i40e_add_del_fdir_sctpv4(vsi, input, add);
527 ret = i40e_add_del_fdir_ipv4(vsi, input, add);
530 /* We cannot support masking based on protocol */
531 dev_info(&pf->pdev->dev, "Unsupported IPv4 protocol 0x%02x\n",
537 dev_info(&pf->pdev->dev, "Unsupported flow type 0x%02x\n",
542 /* The buffer allocated here will be normally be freed by
543 * i40e_clean_fdir_tx_irq() as it reclaims resources after transmit
544 * completion. In the event of an error adding the buffer to the FDIR
545 * ring, it will immediately be freed. It may also be freed by
546 * i40e_clean_tx_ring() when closing the VSI.
552 * i40e_fd_handle_status - check the Programming Status for FD
553 * @rx_ring: the Rx ring for this descriptor
554 * @rx_desc: the Rx descriptor for programming Status, not a packet descriptor.
555 * @prog_id: the id originally used for programming
557 * This is used to verify if the FD programming or invalidation
558 * requested by SW to the HW is successful or not and take actions accordingly.
560 static void i40e_fd_handle_status(struct i40e_ring *rx_ring,
561 union i40e_rx_desc *rx_desc, u8 prog_id)
563 struct i40e_pf *pf = rx_ring->vsi->back;
564 struct pci_dev *pdev = pf->pdev;
565 u32 fcnt_prog, fcnt_avail;
569 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
570 error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
571 I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
573 if (error == BIT(I40E_RX_PROG_STATUS_DESC_FD_TBL_FULL_SHIFT)) {
574 pf->fd_inv = le32_to_cpu(rx_desc->wb.qword0.hi_dword.fd_id);
575 if ((rx_desc->wb.qword0.hi_dword.fd_id != 0) ||
576 (I40E_DEBUG_FD & pf->hw.debug_mask))
577 dev_warn(&pdev->dev, "ntuple filter loc = %d, could not be added\n",
580 /* Check if the programming error is for ATR.
581 * If so, auto disable ATR and set a state for
582 * flush in progress. Next time we come here if flush is in
583 * progress do nothing, once flush is complete the state will
586 if (test_bit(__I40E_FD_FLUSH_REQUESTED, pf->state))
590 /* store the current atr filter count */
591 pf->fd_atr_cnt = i40e_get_current_atr_cnt(pf);
593 if ((rx_desc->wb.qword0.hi_dword.fd_id == 0) &&
594 pf->flags & I40E_FLAG_FD_SB_AUTO_DISABLED) {
595 pf->flags |= I40E_FLAG_FD_ATR_AUTO_DISABLED;
596 set_bit(__I40E_FD_FLUSH_REQUESTED, pf->state);
599 /* filter programming failed most likely due to table full */
600 fcnt_prog = i40e_get_global_fd_count(pf);
601 fcnt_avail = pf->fdir_pf_filter_count;
602 /* If ATR is running fcnt_prog can quickly change,
603 * if we are very close to full, it makes sense to disable
604 * FD ATR/SB and then re-enable it when there is room.
606 if (fcnt_prog >= (fcnt_avail - I40E_FDIR_BUFFER_FULL_MARGIN)) {
607 if ((pf->flags & I40E_FLAG_FD_SB_ENABLED) &&
608 !(pf->flags & I40E_FLAG_FD_SB_AUTO_DISABLED)) {
609 pf->flags |= I40E_FLAG_FD_SB_AUTO_DISABLED;
610 if (I40E_DEBUG_FD & pf->hw.debug_mask)
611 dev_warn(&pdev->dev, "FD filter space full, new ntuple rules will not be added\n");
614 } else if (error == BIT(I40E_RX_PROG_STATUS_DESC_NO_FD_ENTRY_SHIFT)) {
615 if (I40E_DEBUG_FD & pf->hw.debug_mask)
616 dev_info(&pdev->dev, "ntuple filter fd_id = %d, could not be removed\n",
617 rx_desc->wb.qword0.hi_dword.fd_id);
622 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
623 * @ring: the ring that owns the buffer
624 * @tx_buffer: the buffer to free
626 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
627 struct i40e_tx_buffer *tx_buffer)
629 if (tx_buffer->skb) {
630 if (tx_buffer->tx_flags & I40E_TX_FLAGS_FD_SB)
631 kfree(tx_buffer->raw_buf);
633 dev_kfree_skb_any(tx_buffer->skb);
634 if (dma_unmap_len(tx_buffer, len))
635 dma_unmap_single(ring->dev,
636 dma_unmap_addr(tx_buffer, dma),
637 dma_unmap_len(tx_buffer, len),
639 } else if (dma_unmap_len(tx_buffer, len)) {
640 dma_unmap_page(ring->dev,
641 dma_unmap_addr(tx_buffer, dma),
642 dma_unmap_len(tx_buffer, len),
646 tx_buffer->next_to_watch = NULL;
647 tx_buffer->skb = NULL;
648 dma_unmap_len_set(tx_buffer, len, 0);
649 /* tx_buffer must be completely set up in the transmit path */
653 * i40e_clean_tx_ring - Free any empty Tx buffers
654 * @tx_ring: ring to be cleaned
656 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
658 unsigned long bi_size;
661 /* ring already cleared, nothing to do */
665 /* Free all the Tx ring sk_buffs */
666 for (i = 0; i < tx_ring->count; i++)
667 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
669 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
670 memset(tx_ring->tx_bi, 0, bi_size);
672 /* Zero out the descriptor ring */
673 memset(tx_ring->desc, 0, tx_ring->size);
675 tx_ring->next_to_use = 0;
676 tx_ring->next_to_clean = 0;
678 if (!tx_ring->netdev)
681 /* cleanup Tx queue statistics */
682 netdev_tx_reset_queue(txring_txq(tx_ring));
686 * i40e_free_tx_resources - Free Tx resources per queue
687 * @tx_ring: Tx descriptor ring for a specific queue
689 * Free all transmit software resources
691 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
693 i40e_clean_tx_ring(tx_ring);
694 kfree(tx_ring->tx_bi);
695 tx_ring->tx_bi = NULL;
698 dma_free_coherent(tx_ring->dev, tx_ring->size,
699 tx_ring->desc, tx_ring->dma);
700 tx_ring->desc = NULL;
705 * i40e_get_tx_pending - how many tx descriptors not processed
706 * @tx_ring: the ring of descriptors
708 * Since there is no access to the ring head register
709 * in XL710, we need to use our local copies
711 u32 i40e_get_tx_pending(struct i40e_ring *ring)
715 head = i40e_get_head(ring);
716 tail = readl(ring->tail);
719 return (head < tail) ?
720 tail - head : (tail + ring->count - head);
728 * i40e_clean_tx_irq - Reclaim resources after transmit completes
729 * @vsi: the VSI we care about
730 * @tx_ring: Tx ring to clean
731 * @napi_budget: Used to determine if we are in netpoll
733 * Returns true if there's any budget left (e.g. the clean is finished)
735 static bool i40e_clean_tx_irq(struct i40e_vsi *vsi,
736 struct i40e_ring *tx_ring, int napi_budget)
738 u16 i = tx_ring->next_to_clean;
739 struct i40e_tx_buffer *tx_buf;
740 struct i40e_tx_desc *tx_head;
741 struct i40e_tx_desc *tx_desc;
742 unsigned int total_bytes = 0, total_packets = 0;
743 unsigned int budget = vsi->work_limit;
745 tx_buf = &tx_ring->tx_bi[i];
746 tx_desc = I40E_TX_DESC(tx_ring, i);
749 tx_head = I40E_TX_DESC(tx_ring, i40e_get_head(tx_ring));
752 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
754 /* if next_to_watch is not set then there is no work pending */
758 /* prevent any other reads prior to eop_desc */
759 read_barrier_depends();
761 i40e_trace(clean_tx_irq, tx_ring, tx_desc, tx_buf);
762 /* we have caught up to head, no work left to do */
763 if (tx_head == tx_desc)
766 /* clear next_to_watch to prevent false hangs */
767 tx_buf->next_to_watch = NULL;
769 /* update the statistics for this packet */
770 total_bytes += tx_buf->bytecount;
771 total_packets += tx_buf->gso_segs;
774 napi_consume_skb(tx_buf->skb, napi_budget);
776 /* unmap skb header data */
777 dma_unmap_single(tx_ring->dev,
778 dma_unmap_addr(tx_buf, dma),
779 dma_unmap_len(tx_buf, len),
782 /* clear tx_buffer data */
784 dma_unmap_len_set(tx_buf, len, 0);
786 /* unmap remaining buffers */
787 while (tx_desc != eop_desc) {
788 i40e_trace(clean_tx_irq_unmap,
789 tx_ring, tx_desc, tx_buf);
796 tx_buf = tx_ring->tx_bi;
797 tx_desc = I40E_TX_DESC(tx_ring, 0);
800 /* unmap any remaining paged data */
801 if (dma_unmap_len(tx_buf, len)) {
802 dma_unmap_page(tx_ring->dev,
803 dma_unmap_addr(tx_buf, dma),
804 dma_unmap_len(tx_buf, len),
806 dma_unmap_len_set(tx_buf, len, 0);
810 /* move us one more past the eop_desc for start of next pkt */
816 tx_buf = tx_ring->tx_bi;
817 tx_desc = I40E_TX_DESC(tx_ring, 0);
822 /* update budget accounting */
824 } while (likely(budget));
827 tx_ring->next_to_clean = i;
828 u64_stats_update_begin(&tx_ring->syncp);
829 tx_ring->stats.bytes += total_bytes;
830 tx_ring->stats.packets += total_packets;
831 u64_stats_update_end(&tx_ring->syncp);
832 tx_ring->q_vector->tx.total_bytes += total_bytes;
833 tx_ring->q_vector->tx.total_packets += total_packets;
835 if (tx_ring->flags & I40E_TXR_FLAGS_WB_ON_ITR) {
836 /* check to see if there are < 4 descriptors
837 * waiting to be written back, then kick the hardware to force
838 * them to be written back in case we stay in NAPI.
839 * In this mode on X722 we do not enable Interrupt.
841 unsigned int j = i40e_get_tx_pending(tx_ring);
844 ((j / WB_STRIDE) == 0) && (j > 0) &&
845 !test_bit(__I40E_VSI_DOWN, vsi->state) &&
846 (I40E_DESC_UNUSED(tx_ring) != tx_ring->count))
847 tx_ring->arm_wb = true;
850 /* notify netdev of completed buffers */
851 netdev_tx_completed_queue(txring_txq(tx_ring),
852 total_packets, total_bytes);
854 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
855 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
856 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
857 /* Make sure that anybody stopping the queue after this
858 * sees the new next_to_clean.
861 if (__netif_subqueue_stopped(tx_ring->netdev,
862 tx_ring->queue_index) &&
863 !test_bit(__I40E_VSI_DOWN, vsi->state)) {
864 netif_wake_subqueue(tx_ring->netdev,
865 tx_ring->queue_index);
866 ++tx_ring->tx_stats.restart_queue;
874 * i40e_enable_wb_on_itr - Arm hardware to do a wb, interrupts are not enabled
875 * @vsi: the VSI we care about
876 * @q_vector: the vector on which to enable writeback
879 static void i40e_enable_wb_on_itr(struct i40e_vsi *vsi,
880 struct i40e_q_vector *q_vector)
882 u16 flags = q_vector->tx.ring[0].flags;
885 if (!(flags & I40E_TXR_FLAGS_WB_ON_ITR))
888 if (q_vector->arm_wb_state)
891 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
892 val = I40E_PFINT_DYN_CTLN_WB_ON_ITR_MASK |
893 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK; /* set noitr */
896 I40E_PFINT_DYN_CTLN(q_vector->v_idx + vsi->base_vector - 1),
899 val = I40E_PFINT_DYN_CTL0_WB_ON_ITR_MASK |
900 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK; /* set noitr */
902 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
904 q_vector->arm_wb_state = true;
908 * i40e_force_wb - Issue SW Interrupt so HW does a wb
909 * @vsi: the VSI we care about
910 * @q_vector: the vector on which to force writeback
913 void i40e_force_wb(struct i40e_vsi *vsi, struct i40e_q_vector *q_vector)
915 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
916 u32 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
917 I40E_PFINT_DYN_CTLN_ITR_INDX_MASK | /* set noitr */
918 I40E_PFINT_DYN_CTLN_SWINT_TRIG_MASK |
919 I40E_PFINT_DYN_CTLN_SW_ITR_INDX_ENA_MASK;
920 /* allow 00 to be written to the index */
923 I40E_PFINT_DYN_CTLN(q_vector->v_idx +
924 vsi->base_vector - 1), val);
926 u32 val = I40E_PFINT_DYN_CTL0_INTENA_MASK |
927 I40E_PFINT_DYN_CTL0_ITR_INDX_MASK | /* set noitr */
928 I40E_PFINT_DYN_CTL0_SWINT_TRIG_MASK |
929 I40E_PFINT_DYN_CTL0_SW_ITR_INDX_ENA_MASK;
930 /* allow 00 to be written to the index */
932 wr32(&vsi->back->hw, I40E_PFINT_DYN_CTL0, val);
937 * i40e_set_new_dynamic_itr - Find new ITR level
938 * @rc: structure containing ring performance data
940 * Returns true if ITR changed, false if not
942 * Stores a new ITR value based on packets and byte counts during
943 * the last interrupt. The advantage of per interrupt computation
944 * is faster updates and more accurate ITR for the current traffic
945 * pattern. Constants in this function were computed based on
946 * theoretical maximum wire speed and thresholds were set based on
947 * testing data as well as attempting to minimize response time
948 * while increasing bulk throughput.
950 static bool i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
952 enum i40e_latency_range new_latency_range = rc->latency_range;
953 struct i40e_q_vector *qv = rc->ring->q_vector;
954 u32 new_itr = rc->itr;
958 if (rc->total_packets == 0 || !rc->itr)
961 /* simple throttlerate management
962 * 0-10MB/s lowest (50000 ints/s)
963 * 10-20MB/s low (20000 ints/s)
964 * 20-1249MB/s bulk (18000 ints/s)
965 * > 40000 Rx packets per second (8000 ints/s)
967 * The math works out because the divisor is in 10^(-6) which
968 * turns the bytes/us input value into MB/s values, but
969 * make sure to use usecs, as the register values written
970 * are in 2 usec increments in the ITR registers, and make sure
971 * to use the smoothed values that the countdown timer gives us.
973 usecs = (rc->itr << 1) * ITR_COUNTDOWN_START;
974 bytes_per_int = rc->total_bytes / usecs;
976 switch (new_latency_range) {
977 case I40E_LOWEST_LATENCY:
978 if (bytes_per_int > 10)
979 new_latency_range = I40E_LOW_LATENCY;
981 case I40E_LOW_LATENCY:
982 if (bytes_per_int > 20)
983 new_latency_range = I40E_BULK_LATENCY;
984 else if (bytes_per_int <= 10)
985 new_latency_range = I40E_LOWEST_LATENCY;
987 case I40E_BULK_LATENCY:
988 case I40E_ULTRA_LATENCY:
990 if (bytes_per_int <= 20)
991 new_latency_range = I40E_LOW_LATENCY;
995 /* this is to adjust RX more aggressively when streaming small
996 * packets. The value of 40000 was picked as it is just beyond
997 * what the hardware can receive per second if in low latency
1000 #define RX_ULTRA_PACKET_RATE 40000
1002 if ((((rc->total_packets * 1000000) / usecs) > RX_ULTRA_PACKET_RATE) &&
1004 new_latency_range = I40E_ULTRA_LATENCY;
1006 rc->latency_range = new_latency_range;
1008 switch (new_latency_range) {
1009 case I40E_LOWEST_LATENCY:
1010 new_itr = I40E_ITR_50K;
1012 case I40E_LOW_LATENCY:
1013 new_itr = I40E_ITR_20K;
1015 case I40E_BULK_LATENCY:
1016 new_itr = I40E_ITR_18K;
1018 case I40E_ULTRA_LATENCY:
1019 new_itr = I40E_ITR_8K;
1025 rc->total_bytes = 0;
1026 rc->total_packets = 0;
1028 if (new_itr != rc->itr) {
1037 * i40e_rx_is_programming_status - check for programming status descriptor
1038 * @qw: qword representing status_error_len in CPU ordering
1040 * The value of in the descriptor length field indicate if this
1041 * is a programming status descriptor for flow director or FCoE
1042 * by the value of I40E_RX_PROG_STATUS_DESC_LENGTH, otherwise
1043 * it is a packet descriptor.
1045 static inline bool i40e_rx_is_programming_status(u64 qw)
1047 /* The Rx filter programming status and SPH bit occupy the same
1048 * spot in the descriptor. Since we don't support packet split we
1049 * can just reuse the bit as an indication that this is a
1050 * programming status descriptor.
1052 return qw & I40E_RXD_QW1_LENGTH_SPH_MASK;
1056 * i40e_clean_programming_status - clean the programming status descriptor
1057 * @rx_ring: the rx ring that has this descriptor
1058 * @rx_desc: the rx descriptor written back by HW
1059 * @qw: qword representing status_error_len in CPU ordering
1061 * Flow director should handle FD_FILTER_STATUS to check its filter programming
1062 * status being successful or not and take actions accordingly. FCoE should
1063 * handle its context/filter programming/invalidation status and take actions.
1066 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
1067 union i40e_rx_desc *rx_desc,
1070 u32 ntc = rx_ring->next_to_clean + 1;
1073 /* fetch, update, and store next to clean */
1074 ntc = (ntc < rx_ring->count) ? ntc : 0;
1075 rx_ring->next_to_clean = ntc;
1077 prefetch(I40E_RX_DESC(rx_ring, ntc));
1079 id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
1080 I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
1082 if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
1083 i40e_fd_handle_status(rx_ring, rx_desc, id);
1087 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
1088 * @tx_ring: the tx ring to set up
1090 * Return 0 on success, negative on error
1092 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
1094 struct device *dev = tx_ring->dev;
1100 /* warn if we are about to overwrite the pointer */
1101 WARN_ON(tx_ring->tx_bi);
1102 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
1103 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
1104 if (!tx_ring->tx_bi)
1107 /* round up to nearest 4K */
1108 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
1109 /* add u32 for head writeback, align after this takes care of
1110 * guaranteeing this is at least one cache line in size
1112 tx_ring->size += sizeof(u32);
1113 tx_ring->size = ALIGN(tx_ring->size, 4096);
1114 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
1115 &tx_ring->dma, GFP_KERNEL);
1116 if (!tx_ring->desc) {
1117 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
1122 tx_ring->next_to_use = 0;
1123 tx_ring->next_to_clean = 0;
1127 kfree(tx_ring->tx_bi);
1128 tx_ring->tx_bi = NULL;
1133 * i40e_clean_rx_ring - Free Rx buffers
1134 * @rx_ring: ring to be cleaned
1136 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
1138 unsigned long bi_size;
1141 /* ring already cleared, nothing to do */
1142 if (!rx_ring->rx_bi)
1146 dev_kfree_skb(rx_ring->skb);
1147 rx_ring->skb = NULL;
1150 /* Free all the Rx ring sk_buffs */
1151 for (i = 0; i < rx_ring->count; i++) {
1152 struct i40e_rx_buffer *rx_bi = &rx_ring->rx_bi[i];
1157 /* Invalidate cache lines that may have been written to by
1158 * device so that we avoid corrupting memory.
1160 dma_sync_single_range_for_cpu(rx_ring->dev,
1163 rx_ring->rx_buf_len,
1166 /* free resources associated with mapping */
1167 dma_unmap_page_attrs(rx_ring->dev, rx_bi->dma,
1168 i40e_rx_pg_size(rx_ring),
1172 __page_frag_cache_drain(rx_bi->page, rx_bi->pagecnt_bias);
1175 rx_bi->page_offset = 0;
1178 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1179 memset(rx_ring->rx_bi, 0, bi_size);
1181 /* Zero out the descriptor ring */
1182 memset(rx_ring->desc, 0, rx_ring->size);
1184 rx_ring->next_to_alloc = 0;
1185 rx_ring->next_to_clean = 0;
1186 rx_ring->next_to_use = 0;
1190 * i40e_free_rx_resources - Free Rx resources
1191 * @rx_ring: ring to clean the resources from
1193 * Free all receive software resources
1195 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
1197 i40e_clean_rx_ring(rx_ring);
1198 kfree(rx_ring->rx_bi);
1199 rx_ring->rx_bi = NULL;
1201 if (rx_ring->desc) {
1202 dma_free_coherent(rx_ring->dev, rx_ring->size,
1203 rx_ring->desc, rx_ring->dma);
1204 rx_ring->desc = NULL;
1209 * i40e_setup_rx_descriptors - Allocate Rx descriptors
1210 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
1212 * Returns 0 on success, negative on failure
1214 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
1216 struct device *dev = rx_ring->dev;
1219 /* warn if we are about to overwrite the pointer */
1220 WARN_ON(rx_ring->rx_bi);
1221 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
1222 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
1223 if (!rx_ring->rx_bi)
1226 u64_stats_init(&rx_ring->syncp);
1228 /* Round up to nearest 4K */
1229 rx_ring->size = rx_ring->count * sizeof(union i40e_32byte_rx_desc);
1230 rx_ring->size = ALIGN(rx_ring->size, 4096);
1231 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
1232 &rx_ring->dma, GFP_KERNEL);
1234 if (!rx_ring->desc) {
1235 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
1240 rx_ring->next_to_alloc = 0;
1241 rx_ring->next_to_clean = 0;
1242 rx_ring->next_to_use = 0;
1246 kfree(rx_ring->rx_bi);
1247 rx_ring->rx_bi = NULL;
1252 * i40e_release_rx_desc - Store the new tail and head values
1253 * @rx_ring: ring to bump
1254 * @val: new head index
1256 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
1258 rx_ring->next_to_use = val;
1260 /* update next to alloc since we have filled the ring */
1261 rx_ring->next_to_alloc = val;
1263 /* Force memory writes to complete before letting h/w
1264 * know there are new descriptors to fetch. (Only
1265 * applicable for weak-ordered memory model archs,
1269 writel(val, rx_ring->tail);
1273 * i40e_rx_offset - Return expected offset into page to access data
1274 * @rx_ring: Ring we are requesting offset of
1276 * Returns the offset value for ring into the data buffer.
1278 static inline unsigned int i40e_rx_offset(struct i40e_ring *rx_ring)
1280 return ring_uses_build_skb(rx_ring) ? I40E_SKB_PAD : 0;
1284 * i40e_alloc_mapped_page - recycle or make a new page
1285 * @rx_ring: ring to use
1286 * @bi: rx_buffer struct to modify
1288 * Returns true if the page was successfully allocated or
1291 static bool i40e_alloc_mapped_page(struct i40e_ring *rx_ring,
1292 struct i40e_rx_buffer *bi)
1294 struct page *page = bi->page;
1297 /* since we are recycling buffers we should seldom need to alloc */
1299 rx_ring->rx_stats.page_reuse_count++;
1303 /* alloc new page for storage */
1304 page = dev_alloc_pages(i40e_rx_pg_order(rx_ring));
1305 if (unlikely(!page)) {
1306 rx_ring->rx_stats.alloc_page_failed++;
1310 /* map page for use */
1311 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1312 i40e_rx_pg_size(rx_ring),
1316 /* if mapping failed free memory back to system since
1317 * there isn't much point in holding memory we can't use
1319 if (dma_mapping_error(rx_ring->dev, dma)) {
1320 __free_pages(page, i40e_rx_pg_order(rx_ring));
1321 rx_ring->rx_stats.alloc_page_failed++;
1327 bi->page_offset = i40e_rx_offset(rx_ring);
1329 /* initialize pagecnt_bias to 1 representing we fully own page */
1330 bi->pagecnt_bias = 1;
1336 * i40e_receive_skb - Send a completed packet up the stack
1337 * @rx_ring: rx ring in play
1338 * @skb: packet to send up
1339 * @vlan_tag: vlan tag for packet
1341 static void i40e_receive_skb(struct i40e_ring *rx_ring,
1342 struct sk_buff *skb, u16 vlan_tag)
1344 struct i40e_q_vector *q_vector = rx_ring->q_vector;
1346 if ((rx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1347 (vlan_tag & VLAN_VID_MASK))
1348 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
1350 napi_gro_receive(&q_vector->napi, skb);
1354 * i40e_alloc_rx_buffers - Replace used receive buffers
1355 * @rx_ring: ring to place buffers on
1356 * @cleaned_count: number of buffers to replace
1358 * Returns false if all allocations were successful, true if any fail
1360 bool i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
1362 u16 ntu = rx_ring->next_to_use;
1363 union i40e_rx_desc *rx_desc;
1364 struct i40e_rx_buffer *bi;
1366 /* do nothing if no valid netdev defined */
1367 if (!rx_ring->netdev || !cleaned_count)
1370 rx_desc = I40E_RX_DESC(rx_ring, ntu);
1371 bi = &rx_ring->rx_bi[ntu];
1374 if (!i40e_alloc_mapped_page(rx_ring, bi))
1377 /* sync the buffer for use by the device */
1378 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1380 rx_ring->rx_buf_len,
1383 /* Refresh the desc even if buffer_addrs didn't change
1384 * because each write-back erases this info.
1386 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1391 if (unlikely(ntu == rx_ring->count)) {
1392 rx_desc = I40E_RX_DESC(rx_ring, 0);
1393 bi = rx_ring->rx_bi;
1397 /* clear the status bits for the next_to_use descriptor */
1398 rx_desc->wb.qword1.status_error_len = 0;
1401 } while (cleaned_count);
1403 if (rx_ring->next_to_use != ntu)
1404 i40e_release_rx_desc(rx_ring, ntu);
1409 if (rx_ring->next_to_use != ntu)
1410 i40e_release_rx_desc(rx_ring, ntu);
1412 /* make sure to come back via polling to try again after
1413 * allocation failure
1419 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
1420 * @vsi: the VSI we care about
1421 * @skb: skb currently being received and modified
1422 * @rx_desc: the receive descriptor
1424 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
1425 struct sk_buff *skb,
1426 union i40e_rx_desc *rx_desc)
1428 struct i40e_rx_ptype_decoded decoded;
1429 u32 rx_error, rx_status;
1434 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1435 ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >> I40E_RXD_QW1_PTYPE_SHIFT;
1436 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK) >>
1437 I40E_RXD_QW1_ERROR_SHIFT;
1438 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1439 I40E_RXD_QW1_STATUS_SHIFT;
1440 decoded = decode_rx_desc_ptype(ptype);
1442 skb->ip_summed = CHECKSUM_NONE;
1444 skb_checksum_none_assert(skb);
1446 /* Rx csum enabled and ip headers found? */
1447 if (!(vsi->netdev->features & NETIF_F_RXCSUM))
1450 /* did the hardware decode the packet and checksum? */
1451 if (!(rx_status & BIT(I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
1454 /* both known and outer_ip must be set for the below code to work */
1455 if (!(decoded.known && decoded.outer_ip))
1458 ipv4 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1459 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV4);
1460 ipv6 = (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP) &&
1461 (decoded.outer_ip_ver == I40E_RX_PTYPE_OUTER_IPV6);
1464 (rx_error & (BIT(I40E_RX_DESC_ERROR_IPE_SHIFT) |
1465 BIT(I40E_RX_DESC_ERROR_EIPE_SHIFT))))
1468 /* likely incorrect csum if alternate IP extension headers found */
1470 rx_status & BIT(I40E_RX_DESC_STATUS_IPV6EXADD_SHIFT))
1471 /* don't increment checksum err here, non-fatal err */
1474 /* there was some L4 error, count error and punt packet to the stack */
1475 if (rx_error & BIT(I40E_RX_DESC_ERROR_L4E_SHIFT))
1478 /* handle packets that were not able to be checksummed due
1479 * to arrival speed, in this case the stack can compute
1482 if (rx_error & BIT(I40E_RX_DESC_ERROR_PPRS_SHIFT))
1485 /* If there is an outer header present that might contain a checksum
1486 * we need to bump the checksum level by 1 to reflect the fact that
1487 * we are indicating we validated the inner checksum.
1489 if (decoded.tunnel_type >= I40E_RX_PTYPE_TUNNEL_IP_GRENAT)
1490 skb->csum_level = 1;
1492 /* Only report checksum unnecessary for TCP, UDP, or SCTP */
1493 switch (decoded.inner_prot) {
1494 case I40E_RX_PTYPE_INNER_PROT_TCP:
1495 case I40E_RX_PTYPE_INNER_PROT_UDP:
1496 case I40E_RX_PTYPE_INNER_PROT_SCTP:
1497 skb->ip_summed = CHECKSUM_UNNECESSARY;
1506 vsi->back->hw_csum_rx_error++;
1510 * i40e_ptype_to_htype - get a hash type
1511 * @ptype: the ptype value from the descriptor
1513 * Returns a hash type to be used by skb_set_hash
1515 static inline int i40e_ptype_to_htype(u8 ptype)
1517 struct i40e_rx_ptype_decoded decoded = decode_rx_desc_ptype(ptype);
1520 return PKT_HASH_TYPE_NONE;
1522 if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1523 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY4)
1524 return PKT_HASH_TYPE_L4;
1525 else if (decoded.outer_ip == I40E_RX_PTYPE_OUTER_IP &&
1526 decoded.payload_layer == I40E_RX_PTYPE_PAYLOAD_LAYER_PAY3)
1527 return PKT_HASH_TYPE_L3;
1529 return PKT_HASH_TYPE_L2;
1533 * i40e_rx_hash - set the hash value in the skb
1534 * @ring: descriptor ring
1535 * @rx_desc: specific descriptor
1537 static inline void i40e_rx_hash(struct i40e_ring *ring,
1538 union i40e_rx_desc *rx_desc,
1539 struct sk_buff *skb,
1543 const __le64 rss_mask =
1544 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
1545 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
1547 if (!(ring->netdev->features & NETIF_F_RXHASH))
1550 if ((rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask) {
1551 hash = le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
1552 skb_set_hash(skb, hash, i40e_ptype_to_htype(rx_ptype));
1557 * i40e_process_skb_fields - Populate skb header fields from Rx descriptor
1558 * @rx_ring: rx descriptor ring packet is being transacted on
1559 * @rx_desc: pointer to the EOP Rx descriptor
1560 * @skb: pointer to current skb being populated
1561 * @rx_ptype: the packet type decoded by hardware
1563 * This function checks the ring, descriptor, and packet information in
1564 * order to populate the hash, checksum, VLAN, protocol, and
1565 * other fields within the skb.
1568 void i40e_process_skb_fields(struct i40e_ring *rx_ring,
1569 union i40e_rx_desc *rx_desc, struct sk_buff *skb,
1572 u64 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1573 u32 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK) >>
1574 I40E_RXD_QW1_STATUS_SHIFT;
1575 u32 tsynvalid = rx_status & I40E_RXD_QW1_STATUS_TSYNVALID_MASK;
1576 u32 tsyn = (rx_status & I40E_RXD_QW1_STATUS_TSYNINDX_MASK) >>
1577 I40E_RXD_QW1_STATUS_TSYNINDX_SHIFT;
1579 if (unlikely(tsynvalid))
1580 i40e_ptp_rx_hwtstamp(rx_ring->vsi->back, skb, tsyn);
1582 i40e_rx_hash(rx_ring, rx_desc, skb, rx_ptype);
1584 i40e_rx_checksum(rx_ring->vsi, skb, rx_desc);
1586 skb_record_rx_queue(skb, rx_ring->queue_index);
1588 /* modifies the skb - consumes the enet header */
1589 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1593 * i40e_cleanup_headers - Correct empty headers
1594 * @rx_ring: rx descriptor ring packet is being transacted on
1595 * @skb: pointer to current skb being fixed
1597 * Also address the case where we are pulling data in on pages only
1598 * and as such no data is present in the skb header.
1600 * In addition if skb is not at least 60 bytes we need to pad it so that
1601 * it is large enough to qualify as a valid Ethernet frame.
1603 * Returns true if an error was encountered and skb was freed.
1605 static bool i40e_cleanup_headers(struct i40e_ring *rx_ring, struct sk_buff *skb)
1607 /* if eth_skb_pad returns an error the skb was freed */
1608 if (eth_skb_pad(skb))
1615 * i40e_reuse_rx_page - page flip buffer and store it back on the ring
1616 * @rx_ring: rx descriptor ring to store buffers on
1617 * @old_buff: donor buffer to have page reused
1619 * Synchronizes page for reuse by the adapter
1621 static void i40e_reuse_rx_page(struct i40e_ring *rx_ring,
1622 struct i40e_rx_buffer *old_buff)
1624 struct i40e_rx_buffer *new_buff;
1625 u16 nta = rx_ring->next_to_alloc;
1627 new_buff = &rx_ring->rx_bi[nta];
1629 /* update, and store next to alloc */
1631 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1633 /* transfer page from old buffer to new buffer */
1634 new_buff->dma = old_buff->dma;
1635 new_buff->page = old_buff->page;
1636 new_buff->page_offset = old_buff->page_offset;
1637 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
1641 * i40e_page_is_reusable - check if any reuse is possible
1642 * @page: page struct to check
1644 * A page is not reusable if it was allocated under low memory
1645 * conditions, or it's not in the same NUMA node as this CPU.
1647 static inline bool i40e_page_is_reusable(struct page *page)
1649 return (page_to_nid(page) == numa_mem_id()) &&
1650 !page_is_pfmemalloc(page);
1654 * i40e_can_reuse_rx_page - Determine if this page can be reused by
1655 * the adapter for another receive
1657 * @rx_buffer: buffer containing the page
1659 * If page is reusable, rx_buffer->page_offset is adjusted to point to
1660 * an unused region in the page.
1662 * For small pages, @truesize will be a constant value, half the size
1663 * of the memory at page. We'll attempt to alternate between high and
1664 * low halves of the page, with one half ready for use by the hardware
1665 * and the other half being consumed by the stack. We use the page
1666 * ref count to determine whether the stack has finished consuming the
1667 * portion of this page that was passed up with a previous packet. If
1668 * the page ref count is >1, we'll assume the "other" half page is
1669 * still busy, and this page cannot be reused.
1671 * For larger pages, @truesize will be the actual space used by the
1672 * received packet (adjusted upward to an even multiple of the cache
1673 * line size). This will advance through the page by the amount
1674 * actually consumed by the received packets while there is still
1675 * space for a buffer. Each region of larger pages will be used at
1676 * most once, after which the page will not be reused.
1678 * In either case, if the page is reusable its refcount is increased.
1680 static bool i40e_can_reuse_rx_page(struct i40e_rx_buffer *rx_buffer)
1682 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1683 struct page *page = rx_buffer->page;
1685 /* Is any reuse possible? */
1686 if (unlikely(!i40e_page_is_reusable(page)))
1689 #if (PAGE_SIZE < 8192)
1690 /* if we are only owner of page we can reuse it */
1691 if (unlikely((page_count(page) - pagecnt_bias) > 1))
1694 #define I40E_LAST_OFFSET \
1695 (SKB_WITH_OVERHEAD(PAGE_SIZE) - I40E_RXBUFFER_2048)
1696 if (rx_buffer->page_offset > I40E_LAST_OFFSET)
1700 /* If we have drained the page fragment pool we need to update
1701 * the pagecnt_bias and page count so that we fully restock the
1702 * number of references the driver holds.
1704 if (unlikely(!pagecnt_bias)) {
1705 page_ref_add(page, USHRT_MAX);
1706 rx_buffer->pagecnt_bias = USHRT_MAX;
1713 * i40e_add_rx_frag - Add contents of Rx buffer to sk_buff
1714 * @rx_ring: rx descriptor ring to transact packets on
1715 * @rx_buffer: buffer containing page to add
1716 * @skb: sk_buff to place the data into
1717 * @size: packet length from rx_desc
1719 * This function will add the data contained in rx_buffer->page to the skb.
1720 * It will just attach the page as a frag to the skb.
1722 * The function will then update the page offset.
1724 static void i40e_add_rx_frag(struct i40e_ring *rx_ring,
1725 struct i40e_rx_buffer *rx_buffer,
1726 struct sk_buff *skb,
1729 #if (PAGE_SIZE < 8192)
1730 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1732 unsigned int truesize = SKB_DATA_ALIGN(size + i40e_rx_offset(rx_ring));
1735 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1736 rx_buffer->page_offset, size, truesize);
1738 /* page is being used so we must update the page offset */
1739 #if (PAGE_SIZE < 8192)
1740 rx_buffer->page_offset ^= truesize;
1742 rx_buffer->page_offset += truesize;
1747 * i40e_get_rx_buffer - Fetch Rx buffer and synchronize data for use
1748 * @rx_ring: rx descriptor ring to transact packets on
1749 * @size: size of buffer to add to skb
1751 * This function will pull an Rx buffer from the ring and synchronize it
1752 * for use by the CPU.
1754 static struct i40e_rx_buffer *i40e_get_rx_buffer(struct i40e_ring *rx_ring,
1755 const unsigned int size)
1757 struct i40e_rx_buffer *rx_buffer;
1759 rx_buffer = &rx_ring->rx_bi[rx_ring->next_to_clean];
1760 prefetchw(rx_buffer->page);
1762 /* we are reusing so sync this buffer for CPU use */
1763 dma_sync_single_range_for_cpu(rx_ring->dev,
1765 rx_buffer->page_offset,
1769 /* We have pulled a buffer for use, so decrement pagecnt_bias */
1770 rx_buffer->pagecnt_bias--;
1776 * i40e_construct_skb - Allocate skb and populate it
1777 * @rx_ring: rx descriptor ring to transact packets on
1778 * @rx_buffer: rx buffer to pull data from
1779 * @size: size of buffer to add to skb
1781 * This function allocates an skb. It then populates it with the page
1782 * data from the current receive descriptor, taking care to set up the
1785 static struct sk_buff *i40e_construct_skb(struct i40e_ring *rx_ring,
1786 struct i40e_rx_buffer *rx_buffer,
1789 void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1790 #if (PAGE_SIZE < 8192)
1791 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1793 unsigned int truesize = SKB_DATA_ALIGN(size);
1795 unsigned int headlen;
1796 struct sk_buff *skb;
1798 /* prefetch first cache line of first page */
1800 #if L1_CACHE_BYTES < 128
1801 prefetch(va + L1_CACHE_BYTES);
1804 /* allocate a skb to store the frags */
1805 skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
1807 GFP_ATOMIC | __GFP_NOWARN);
1811 /* Determine available headroom for copy */
1813 if (headlen > I40E_RX_HDR_SIZE)
1814 headlen = eth_get_headlen(va, I40E_RX_HDR_SIZE);
1816 /* align pull length to size of long to optimize memcpy performance */
1817 memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1819 /* update all of the pointers */
1822 skb_add_rx_frag(skb, 0, rx_buffer->page,
1823 rx_buffer->page_offset + headlen,
1826 /* buffer is used by skb, update page_offset */
1827 #if (PAGE_SIZE < 8192)
1828 rx_buffer->page_offset ^= truesize;
1830 rx_buffer->page_offset += truesize;
1833 /* buffer is unused, reset bias back to rx_buffer */
1834 rx_buffer->pagecnt_bias++;
1841 * i40e_build_skb - Build skb around an existing buffer
1842 * @rx_ring: Rx descriptor ring to transact packets on
1843 * @rx_buffer: Rx buffer to pull data from
1844 * @size: size of buffer to add to skb
1846 * This function builds an skb around an existing Rx buffer, taking care
1847 * to set up the skb correctly and avoid any memcpy overhead.
1849 static struct sk_buff *i40e_build_skb(struct i40e_ring *rx_ring,
1850 struct i40e_rx_buffer *rx_buffer,
1853 void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1854 #if (PAGE_SIZE < 8192)
1855 unsigned int truesize = i40e_rx_pg_size(rx_ring) / 2;
1857 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1858 SKB_DATA_ALIGN(I40E_SKB_PAD + size);
1860 struct sk_buff *skb;
1862 /* prefetch first cache line of first page */
1864 #if L1_CACHE_BYTES < 128
1865 prefetch(va + L1_CACHE_BYTES);
1867 /* build an skb around the page buffer */
1868 skb = build_skb(va - I40E_SKB_PAD, truesize);
1872 /* update pointers within the skb to store the data */
1873 skb_reserve(skb, I40E_SKB_PAD);
1874 __skb_put(skb, size);
1876 /* buffer is used by skb, update page_offset */
1877 #if (PAGE_SIZE < 8192)
1878 rx_buffer->page_offset ^= truesize;
1880 rx_buffer->page_offset += truesize;
1887 * i40e_put_rx_buffer - Clean up used buffer and either recycle or free
1888 * @rx_ring: rx descriptor ring to transact packets on
1889 * @rx_buffer: rx buffer to pull data from
1891 * This function will clean up the contents of the rx_buffer. It will
1892 * either recycle the bufer or unmap it and free the associated resources.
1894 static void i40e_put_rx_buffer(struct i40e_ring *rx_ring,
1895 struct i40e_rx_buffer *rx_buffer)
1897 if (i40e_can_reuse_rx_page(rx_buffer)) {
1898 /* hand second half of page back to the ring */
1899 i40e_reuse_rx_page(rx_ring, rx_buffer);
1900 rx_ring->rx_stats.page_reuse_count++;
1902 /* we are not reusing the buffer so unmap it */
1903 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
1904 i40e_rx_pg_size(rx_ring),
1905 DMA_FROM_DEVICE, I40E_RX_DMA_ATTR);
1906 __page_frag_cache_drain(rx_buffer->page,
1907 rx_buffer->pagecnt_bias);
1910 /* clear contents of buffer_info */
1911 rx_buffer->page = NULL;
1915 * i40e_is_non_eop - process handling of non-EOP buffers
1916 * @rx_ring: Rx ring being processed
1917 * @rx_desc: Rx descriptor for current buffer
1918 * @skb: Current socket buffer containing buffer in progress
1920 * This function updates next to clean. If the buffer is an EOP buffer
1921 * this function exits returning false, otherwise it will place the
1922 * sk_buff in the next buffer to be chained and return true indicating
1923 * that this is in fact a non-EOP buffer.
1925 static bool i40e_is_non_eop(struct i40e_ring *rx_ring,
1926 union i40e_rx_desc *rx_desc,
1927 struct sk_buff *skb)
1929 u32 ntc = rx_ring->next_to_clean + 1;
1931 /* fetch, update, and store next to clean */
1932 ntc = (ntc < rx_ring->count) ? ntc : 0;
1933 rx_ring->next_to_clean = ntc;
1935 prefetch(I40E_RX_DESC(rx_ring, ntc));
1937 /* if we are the last buffer then there is nothing else to do */
1938 #define I40E_RXD_EOF BIT(I40E_RX_DESC_STATUS_EOF_SHIFT)
1939 if (likely(i40e_test_staterr(rx_desc, I40E_RXD_EOF)))
1942 rx_ring->rx_stats.non_eop_descs++;
1948 * i40e_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
1949 * @rx_ring: rx descriptor ring to transact packets on
1950 * @budget: Total limit on number of packets to process
1952 * This function provides a "bounce buffer" approach to Rx interrupt
1953 * processing. The advantage to this is that on systems that have
1954 * expensive overhead for IOMMU access this provides a means of avoiding
1955 * it by maintaining the mapping of the page to the system.
1957 * Returns amount of work completed
1959 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
1961 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1962 struct sk_buff *skb = rx_ring->skb;
1963 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
1964 bool failure = false;
1966 while (likely(total_rx_packets < budget)) {
1967 struct i40e_rx_buffer *rx_buffer;
1968 union i40e_rx_desc *rx_desc;
1974 /* return some buffers to hardware, one at a time is too slow */
1975 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1976 failure = failure ||
1977 i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1981 rx_desc = I40E_RX_DESC(rx_ring, rx_ring->next_to_clean);
1983 /* status_error_len will always be zero for unused descriptors
1984 * because it's cleared in cleanup, and overlaps with hdr_addr
1985 * which is always zero because packet split isn't used, if the
1986 * hardware wrote DD then the length will be non-zero
1988 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1990 /* This memory barrier is needed to keep us from reading
1991 * any other fields out of the rx_desc until we have
1992 * verified the descriptor has been written back.
1996 if (unlikely(i40e_rx_is_programming_status(qword))) {
1997 i40e_clean_programming_status(rx_ring, rx_desc, qword);
2000 size = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK) >>
2001 I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
2005 i40e_trace(clean_rx_irq, rx_ring, rx_desc, skb);
2006 rx_buffer = i40e_get_rx_buffer(rx_ring, size);
2008 /* retrieve a buffer from the ring */
2010 i40e_add_rx_frag(rx_ring, rx_buffer, skb, size);
2011 else if (ring_uses_build_skb(rx_ring))
2012 skb = i40e_build_skb(rx_ring, rx_buffer, size);
2014 skb = i40e_construct_skb(rx_ring, rx_buffer, size);
2016 /* exit if we failed to retrieve a buffer */
2018 rx_ring->rx_stats.alloc_buff_failed++;
2019 rx_buffer->pagecnt_bias++;
2023 i40e_put_rx_buffer(rx_ring, rx_buffer);
2026 if (i40e_is_non_eop(rx_ring, rx_desc, skb))
2029 /* ERR_MASK will only have valid bits if EOP set, and
2030 * what we are doing here is actually checking
2031 * I40E_RX_DESC_ERROR_RXE_SHIFT, since it is the zeroth bit in
2034 if (unlikely(i40e_test_staterr(rx_desc, BIT(I40E_RXD_QW1_ERROR_SHIFT)))) {
2035 dev_kfree_skb_any(skb);
2040 if (i40e_cleanup_headers(rx_ring, skb)) {
2045 /* probably a little skewed due to removing CRC */
2046 total_rx_bytes += skb->len;
2048 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
2049 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
2050 I40E_RXD_QW1_PTYPE_SHIFT;
2052 /* populate checksum, VLAN, and protocol */
2053 i40e_process_skb_fields(rx_ring, rx_desc, skb, rx_ptype);
2055 vlan_tag = (qword & BIT(I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)) ?
2056 le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1) : 0;
2058 i40e_trace(clean_rx_irq_rx, rx_ring, rx_desc, skb);
2059 i40e_receive_skb(rx_ring, skb, vlan_tag);
2062 /* update budget accounting */
2068 u64_stats_update_begin(&rx_ring->syncp);
2069 rx_ring->stats.packets += total_rx_packets;
2070 rx_ring->stats.bytes += total_rx_bytes;
2071 u64_stats_update_end(&rx_ring->syncp);
2072 rx_ring->q_vector->rx.total_packets += total_rx_packets;
2073 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
2075 /* guarantee a trip back through this routine if there was a failure */
2076 return failure ? budget : total_rx_packets;
2079 static u32 i40e_buildreg_itr(const int type, const u16 itr)
2083 val = I40E_PFINT_DYN_CTLN_INTENA_MASK |
2084 /* Don't clear PBA because that can cause lost interrupts that
2085 * came in while we were cleaning/polling
2087 (type << I40E_PFINT_DYN_CTLN_ITR_INDX_SHIFT) |
2088 (itr << I40E_PFINT_DYN_CTLN_INTERVAL_SHIFT);
2093 /* a small macro to shorten up some long lines */
2094 #define INTREG I40E_PFINT_DYN_CTLN
2095 static inline int get_rx_itr(struct i40e_vsi *vsi, int idx)
2097 return vsi->rx_rings[idx]->rx_itr_setting;
2100 static inline int get_tx_itr(struct i40e_vsi *vsi, int idx)
2102 return vsi->tx_rings[idx]->tx_itr_setting;
2106 * i40e_update_enable_itr - Update itr and re-enable MSIX interrupt
2107 * @vsi: the VSI we care about
2108 * @q_vector: q_vector for which itr is being updated and interrupt enabled
2111 static inline void i40e_update_enable_itr(struct i40e_vsi *vsi,
2112 struct i40e_q_vector *q_vector)
2114 struct i40e_hw *hw = &vsi->back->hw;
2115 bool rx = false, tx = false;
2118 int idx = q_vector->v_idx;
2119 int rx_itr_setting, tx_itr_setting;
2121 vector = (q_vector->v_idx + vsi->base_vector);
2123 /* avoid dynamic calculation if in countdown mode OR if
2124 * all dynamic is disabled
2126 rxval = txval = i40e_buildreg_itr(I40E_ITR_NONE, 0);
2128 rx_itr_setting = get_rx_itr(vsi, idx);
2129 tx_itr_setting = get_tx_itr(vsi, idx);
2131 if (q_vector->itr_countdown > 0 ||
2132 (!ITR_IS_DYNAMIC(rx_itr_setting) &&
2133 !ITR_IS_DYNAMIC(tx_itr_setting))) {
2137 if (ITR_IS_DYNAMIC(tx_itr_setting)) {
2138 rx = i40e_set_new_dynamic_itr(&q_vector->rx);
2139 rxval = i40e_buildreg_itr(I40E_RX_ITR, q_vector->rx.itr);
2142 if (ITR_IS_DYNAMIC(tx_itr_setting)) {
2143 tx = i40e_set_new_dynamic_itr(&q_vector->tx);
2144 txval = i40e_buildreg_itr(I40E_TX_ITR, q_vector->tx.itr);
2148 /* get the higher of the two ITR adjustments and
2149 * use the same value for both ITR registers
2150 * when in adaptive mode (Rx and/or Tx)
2152 u16 itr = max(q_vector->tx.itr, q_vector->rx.itr);
2154 q_vector->tx.itr = q_vector->rx.itr = itr;
2155 txval = i40e_buildreg_itr(I40E_TX_ITR, itr);
2157 rxval = i40e_buildreg_itr(I40E_RX_ITR, itr);
2161 /* only need to enable the interrupt once, but need
2162 * to possibly update both ITR values
2165 /* set the INTENA_MSK_MASK so that this first write
2166 * won't actually enable the interrupt, instead just
2167 * updating the ITR (it's bit 31 PF and VF)
2170 /* don't check _DOWN because interrupt isn't being enabled */
2171 wr32(hw, INTREG(vector - 1), rxval);
2175 if (!test_bit(__I40E_VSI_DOWN, vsi->state))
2176 wr32(hw, INTREG(vector - 1), txval);
2178 if (q_vector->itr_countdown)
2179 q_vector->itr_countdown--;
2181 q_vector->itr_countdown = ITR_COUNTDOWN_START;
2185 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
2186 * @napi: napi struct with our devices info in it
2187 * @budget: amount of work driver is allowed to do this pass, in packets
2189 * This function will clean all queues associated with a q_vector.
2191 * Returns the amount of work done
2193 int i40e_napi_poll(struct napi_struct *napi, int budget)
2195 struct i40e_q_vector *q_vector =
2196 container_of(napi, struct i40e_q_vector, napi);
2197 struct i40e_vsi *vsi = q_vector->vsi;
2198 struct i40e_ring *ring;
2199 bool clean_complete = true;
2200 bool arm_wb = false;
2201 int budget_per_ring;
2204 if (test_bit(__I40E_VSI_DOWN, vsi->state)) {
2205 napi_complete(napi);
2209 /* Since the actual Tx work is minimal, we can give the Tx a larger
2210 * budget and be more aggressive about cleaning up the Tx descriptors.
2212 i40e_for_each_ring(ring, q_vector->tx) {
2213 if (!i40e_clean_tx_irq(vsi, ring, budget)) {
2214 clean_complete = false;
2217 arm_wb |= ring->arm_wb;
2218 ring->arm_wb = false;
2221 /* Handle case where we are called by netpoll with a budget of 0 */
2225 /* We attempt to distribute budget to each Rx queue fairly, but don't
2226 * allow the budget to go below 1 because that would exit polling early.
2228 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
2230 i40e_for_each_ring(ring, q_vector->rx) {
2231 int cleaned = i40e_clean_rx_irq(ring, budget_per_ring);
2233 work_done += cleaned;
2234 /* if we clean as many as budgeted, we must not be done */
2235 if (cleaned >= budget_per_ring)
2236 clean_complete = false;
2239 /* If work not completed, return budget and polling will return */
2240 if (!clean_complete) {
2241 const cpumask_t *aff_mask = &q_vector->affinity_mask;
2242 int cpu_id = smp_processor_id();
2244 /* It is possible that the interrupt affinity has changed but,
2245 * if the cpu is pegged at 100%, polling will never exit while
2246 * traffic continues and the interrupt will be stuck on this
2247 * cpu. We check to make sure affinity is correct before we
2248 * continue to poll, otherwise we must stop polling so the
2249 * interrupt can move to the correct cpu.
2251 if (likely(cpumask_test_cpu(cpu_id, aff_mask) ||
2252 !(vsi->back->flags & I40E_FLAG_MSIX_ENABLED))) {
2255 q_vector->tx.ring[0].tx_stats.tx_force_wb++;
2256 i40e_enable_wb_on_itr(vsi, q_vector);
2262 if (vsi->back->flags & I40E_TXR_FLAGS_WB_ON_ITR)
2263 q_vector->arm_wb_state = false;
2265 /* Work is done so exit the polling mode and re-enable the interrupt */
2266 napi_complete_done(napi, work_done);
2268 /* If we're prematurely stopping polling to fix the interrupt
2269 * affinity we want to make sure polling starts back up so we
2270 * issue a call to i40e_force_wb which triggers a SW interrupt.
2272 if (!clean_complete)
2273 i40e_force_wb(vsi, q_vector);
2274 else if (!(vsi->back->flags & I40E_FLAG_MSIX_ENABLED))
2275 i40e_irq_dynamic_enable_icr0(vsi->back, false);
2277 i40e_update_enable_itr(vsi, q_vector);
2279 return min(work_done, budget - 1);
2283 * i40e_atr - Add a Flow Director ATR filter
2284 * @tx_ring: ring to add programming descriptor to
2286 * @tx_flags: send tx flags
2288 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
2291 struct i40e_filter_program_desc *fdir_desc;
2292 struct i40e_pf *pf = tx_ring->vsi->back;
2294 unsigned char *network;
2296 struct ipv6hdr *ipv6;
2300 u32 flex_ptype, dtype_cmd;
2304 /* make sure ATR is enabled */
2305 if (!(pf->flags & I40E_FLAG_FD_ATR_ENABLED))
2308 if (pf->flags & I40E_FLAG_FD_ATR_AUTO_DISABLED)
2311 /* if sampling is disabled do nothing */
2312 if (!tx_ring->atr_sample_rate)
2315 /* Currently only IPv4/IPv6 with TCP is supported */
2316 if (!(tx_flags & (I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6)))
2319 /* snag network header to get L4 type and address */
2320 hdr.network = (tx_flags & I40E_TX_FLAGS_UDP_TUNNEL) ?
2321 skb_inner_network_header(skb) : skb_network_header(skb);
2323 /* Note: tx_flags gets modified to reflect inner protocols in
2324 * tx_enable_csum function if encap is enabled.
2326 if (tx_flags & I40E_TX_FLAGS_IPV4) {
2327 /* access ihl as u8 to avoid unaligned access on ia64 */
2328 hlen = (hdr.network[0] & 0x0F) << 2;
2329 l4_proto = hdr.ipv4->protocol;
2331 hlen = hdr.network - skb->data;
2332 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
2333 hlen -= hdr.network - skb->data;
2336 if (l4_proto != IPPROTO_TCP)
2339 th = (struct tcphdr *)(hdr.network + hlen);
2341 /* Due to lack of space, no more new filters can be programmed */
2342 if (th->syn && (pf->flags & I40E_FLAG_FD_ATR_AUTO_DISABLED))
2344 if (pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE) {
2345 /* HW ATR eviction will take care of removing filters on FIN
2348 if (th->fin || th->rst)
2352 tx_ring->atr_count++;
2354 /* sample on all syn/fin/rst packets or once every atr sample rate */
2358 (tx_ring->atr_count < tx_ring->atr_sample_rate))
2361 tx_ring->atr_count = 0;
2363 /* grab the next descriptor */
2364 i = tx_ring->next_to_use;
2365 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
2368 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2370 flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
2371 I40E_TXD_FLTR_QW0_QINDEX_MASK;
2372 flex_ptype |= (tx_flags & I40E_TX_FLAGS_IPV4) ?
2373 (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
2374 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
2375 (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
2376 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
2378 flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
2380 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
2382 dtype_cmd |= (th->fin || th->rst) ?
2383 (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
2384 I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
2385 (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
2386 I40E_TXD_FLTR_QW1_PCMD_SHIFT);
2388 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
2389 I40E_TXD_FLTR_QW1_DEST_SHIFT;
2391 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
2392 I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
2394 dtype_cmd |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
2395 if (!(tx_flags & I40E_TX_FLAGS_UDP_TUNNEL))
2397 ((u32)I40E_FD_ATR_STAT_IDX(pf->hw.pf_id) <<
2398 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2399 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2402 ((u32)I40E_FD_ATR_TUNNEL_STAT_IDX(pf->hw.pf_id) <<
2403 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
2404 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
2406 if (pf->flags & I40E_FLAG_HW_ATR_EVICT_CAPABLE)
2407 dtype_cmd |= I40E_TXD_FLTR_QW1_ATR_MASK;
2409 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
2410 fdir_desc->rsvd = cpu_to_le32(0);
2411 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
2412 fdir_desc->fd_id = cpu_to_le32(0);
2416 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
2418 * @tx_ring: ring to send buffer on
2419 * @flags: the tx flags to be set
2421 * Checks the skb and set up correspondingly several generic transmit flags
2422 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
2424 * Returns error code indicate the frame should be dropped upon error and the
2425 * otherwise returns 0 to indicate the flags has been set properly.
2427 static inline int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
2428 struct i40e_ring *tx_ring,
2431 __be16 protocol = skb->protocol;
2434 if (protocol == htons(ETH_P_8021Q) &&
2435 !(tx_ring->netdev->features & NETIF_F_HW_VLAN_CTAG_TX)) {
2436 /* When HW VLAN acceleration is turned off by the user the
2437 * stack sets the protocol to 8021q so that the driver
2438 * can take any steps required to support the SW only
2439 * VLAN handling. In our case the driver doesn't need
2440 * to take any further steps so just set the protocol
2441 * to the encapsulated ethertype.
2443 skb->protocol = vlan_get_protocol(skb);
2447 /* if we have a HW VLAN tag being added, default to the HW one */
2448 if (skb_vlan_tag_present(skb)) {
2449 tx_flags |= skb_vlan_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
2450 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2451 /* else if it is a SW VLAN, check the next protocol and store the tag */
2452 } else if (protocol == htons(ETH_P_8021Q)) {
2453 struct vlan_hdr *vhdr, _vhdr;
2455 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
2459 protocol = vhdr->h_vlan_encapsulated_proto;
2460 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
2461 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
2464 if (!(tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED))
2467 /* Insert 802.1p priority into VLAN header */
2468 if ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
2469 (skb->priority != TC_PRIO_CONTROL)) {
2470 tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
2471 tx_flags |= (skb->priority & 0x7) <<
2472 I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
2473 if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
2474 struct vlan_ethhdr *vhdr;
2477 rc = skb_cow_head(skb, 0);
2480 vhdr = (struct vlan_ethhdr *)skb->data;
2481 vhdr->h_vlan_TCI = htons(tx_flags >>
2482 I40E_TX_FLAGS_VLAN_SHIFT);
2484 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
2494 * i40e_tso - set up the tso context descriptor
2495 * @first: pointer to first Tx buffer for xmit
2496 * @hdr_len: ptr to the size of the packet header
2497 * @cd_type_cmd_tso_mss: Quad Word 1
2499 * Returns 0 if no TSO can happen, 1 if tso is going, or error
2501 static int i40e_tso(struct i40e_tx_buffer *first, u8 *hdr_len,
2502 u64 *cd_type_cmd_tso_mss)
2504 struct sk_buff *skb = first->skb;
2505 u64 cd_cmd, cd_tso_len, cd_mss;
2516 u32 paylen, l4_offset;
2517 u16 gso_segs, gso_size;
2520 if (skb->ip_summed != CHECKSUM_PARTIAL)
2523 if (!skb_is_gso(skb))
2526 err = skb_cow_head(skb, 0);
2530 ip.hdr = skb_network_header(skb);
2531 l4.hdr = skb_transport_header(skb);
2533 /* initialize outer IP header fields */
2534 if (ip.v4->version == 4) {
2538 ip.v6->payload_len = 0;
2541 if (skb_shinfo(skb)->gso_type & (SKB_GSO_GRE |
2545 SKB_GSO_UDP_TUNNEL |
2546 SKB_GSO_UDP_TUNNEL_CSUM)) {
2547 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2548 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM)) {
2551 /* determine offset of outer transport header */
2552 l4_offset = l4.hdr - skb->data;
2554 /* remove payload length from outer checksum */
2555 paylen = skb->len - l4_offset;
2556 csum_replace_by_diff(&l4.udp->check,
2557 (__force __wsum)htonl(paylen));
2560 /* reset pointers to inner headers */
2561 ip.hdr = skb_inner_network_header(skb);
2562 l4.hdr = skb_inner_transport_header(skb);
2564 /* initialize inner IP header fields */
2565 if (ip.v4->version == 4) {
2569 ip.v6->payload_len = 0;
2573 /* determine offset of inner transport header */
2574 l4_offset = l4.hdr - skb->data;
2576 /* remove payload length from inner checksum */
2577 paylen = skb->len - l4_offset;
2578 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
2580 /* compute length of segmentation header */
2581 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
2583 /* pull values out of skb_shinfo */
2584 gso_size = skb_shinfo(skb)->gso_size;
2585 gso_segs = skb_shinfo(skb)->gso_segs;
2587 /* update GSO size and bytecount with header size */
2588 first->gso_segs = gso_segs;
2589 first->bytecount += (first->gso_segs - 1) * *hdr_len;
2591 /* find the field values */
2592 cd_cmd = I40E_TX_CTX_DESC_TSO;
2593 cd_tso_len = skb->len - *hdr_len;
2595 *cd_type_cmd_tso_mss |= (cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT) |
2596 (cd_tso_len << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT) |
2597 (cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
2602 * i40e_tsyn - set up the tsyn context descriptor
2603 * @tx_ring: ptr to the ring to send
2604 * @skb: ptr to the skb we're sending
2605 * @tx_flags: the collected send information
2606 * @cd_type_cmd_tso_mss: Quad Word 1
2608 * Returns 0 if no Tx timestamp can happen and 1 if the timestamp will happen
2610 static int i40e_tsyn(struct i40e_ring *tx_ring, struct sk_buff *skb,
2611 u32 tx_flags, u64 *cd_type_cmd_tso_mss)
2615 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)))
2618 /* Tx timestamps cannot be sampled when doing TSO */
2619 if (tx_flags & I40E_TX_FLAGS_TSO)
2622 /* only timestamp the outbound packet if the user has requested it and
2623 * we are not already transmitting a packet to be timestamped
2625 pf = i40e_netdev_to_pf(tx_ring->netdev);
2626 if (!(pf->flags & I40E_FLAG_PTP))
2630 !test_and_set_bit_lock(__I40E_PTP_TX_IN_PROGRESS, pf->state)) {
2631 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2632 pf->ptp_tx_skb = skb_get(skb);
2637 *cd_type_cmd_tso_mss |= (u64)I40E_TX_CTX_DESC_TSYN <<
2638 I40E_TXD_CTX_QW1_CMD_SHIFT;
2644 * i40e_tx_enable_csum - Enable Tx checksum offloads
2646 * @tx_flags: pointer to Tx flags currently set
2647 * @td_cmd: Tx descriptor command bits to set
2648 * @td_offset: Tx descriptor header offsets to set
2649 * @tx_ring: Tx descriptor ring
2650 * @cd_tunneling: ptr to context desc bits
2652 static int i40e_tx_enable_csum(struct sk_buff *skb, u32 *tx_flags,
2653 u32 *td_cmd, u32 *td_offset,
2654 struct i40e_ring *tx_ring,
2667 unsigned char *exthdr;
2668 u32 offset, cmd = 0;
2672 if (skb->ip_summed != CHECKSUM_PARTIAL)
2675 ip.hdr = skb_network_header(skb);
2676 l4.hdr = skb_transport_header(skb);
2678 /* compute outer L2 header size */
2679 offset = ((ip.hdr - skb->data) / 2) << I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
2681 if (skb->encapsulation) {
2683 /* define outer network header type */
2684 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2685 tunnel |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2686 I40E_TX_CTX_EXT_IP_IPV4 :
2687 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
2689 l4_proto = ip.v4->protocol;
2690 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2691 tunnel |= I40E_TX_CTX_EXT_IP_IPV6;
2693 exthdr = ip.hdr + sizeof(*ip.v6);
2694 l4_proto = ip.v6->nexthdr;
2695 if (l4.hdr != exthdr)
2696 ipv6_skip_exthdr(skb, exthdr - skb->data,
2697 &l4_proto, &frag_off);
2700 /* define outer transport */
2703 tunnel |= I40E_TXD_CTX_UDP_TUNNELING;
2704 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2707 tunnel |= I40E_TXD_CTX_GRE_TUNNELING;
2708 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2712 *tx_flags |= I40E_TX_FLAGS_UDP_TUNNEL;
2713 l4.hdr = skb_inner_network_header(skb);
2716 if (*tx_flags & I40E_TX_FLAGS_TSO)
2719 skb_checksum_help(skb);
2723 /* compute outer L3 header size */
2724 tunnel |= ((l4.hdr - ip.hdr) / 4) <<
2725 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT;
2727 /* switch IP header pointer from outer to inner header */
2728 ip.hdr = skb_inner_network_header(skb);
2730 /* compute tunnel header size */
2731 tunnel |= ((ip.hdr - l4.hdr) / 2) <<
2732 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
2734 /* indicate if we need to offload outer UDP header */
2735 if ((*tx_flags & I40E_TX_FLAGS_TSO) &&
2736 !(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) &&
2737 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM))
2738 tunnel |= I40E_TXD_CTX_QW0_L4T_CS_MASK;
2740 /* record tunnel offload values */
2741 *cd_tunneling |= tunnel;
2743 /* switch L4 header pointer from outer to inner */
2744 l4.hdr = skb_inner_transport_header(skb);
2747 /* reset type as we transition from outer to inner headers */
2748 *tx_flags &= ~(I40E_TX_FLAGS_IPV4 | I40E_TX_FLAGS_IPV6);
2749 if (ip.v4->version == 4)
2750 *tx_flags |= I40E_TX_FLAGS_IPV4;
2751 if (ip.v6->version == 6)
2752 *tx_flags |= I40E_TX_FLAGS_IPV6;
2755 /* Enable IP checksum offloads */
2756 if (*tx_flags & I40E_TX_FLAGS_IPV4) {
2757 l4_proto = ip.v4->protocol;
2758 /* the stack computes the IP header already, the only time we
2759 * need the hardware to recompute it is in the case of TSO.
2761 cmd |= (*tx_flags & I40E_TX_FLAGS_TSO) ?
2762 I40E_TX_DESC_CMD_IIPT_IPV4_CSUM :
2763 I40E_TX_DESC_CMD_IIPT_IPV4;
2764 } else if (*tx_flags & I40E_TX_FLAGS_IPV6) {
2765 cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
2767 exthdr = ip.hdr + sizeof(*ip.v6);
2768 l4_proto = ip.v6->nexthdr;
2769 if (l4.hdr != exthdr)
2770 ipv6_skip_exthdr(skb, exthdr - skb->data,
2771 &l4_proto, &frag_off);
2774 /* compute inner L3 header size */
2775 offset |= ((l4.hdr - ip.hdr) / 4) << I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
2777 /* Enable L4 checksum offloads */
2780 /* enable checksum offloads */
2781 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
2782 offset |= l4.tcp->doff << I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2785 /* enable SCTP checksum offload */
2786 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
2787 offset |= (sizeof(struct sctphdr) >> 2) <<
2788 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2791 /* enable UDP checksum offload */
2792 cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
2793 offset |= (sizeof(struct udphdr) >> 2) <<
2794 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
2797 if (*tx_flags & I40E_TX_FLAGS_TSO)
2799 skb_checksum_help(skb);
2804 *td_offset |= offset;
2810 * i40e_create_tx_ctx Build the Tx context descriptor
2811 * @tx_ring: ring to create the descriptor on
2812 * @cd_type_cmd_tso_mss: Quad Word 1
2813 * @cd_tunneling: Quad Word 0 - bits 0-31
2814 * @cd_l2tag2: Quad Word 0 - bits 32-63
2816 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
2817 const u64 cd_type_cmd_tso_mss,
2818 const u32 cd_tunneling, const u32 cd_l2tag2)
2820 struct i40e_tx_context_desc *context_desc;
2821 int i = tx_ring->next_to_use;
2823 if ((cd_type_cmd_tso_mss == I40E_TX_DESC_DTYPE_CONTEXT) &&
2824 !cd_tunneling && !cd_l2tag2)
2827 /* grab the next descriptor */
2828 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
2831 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2833 /* cpu_to_le32 and assign to struct fields */
2834 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
2835 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
2836 context_desc->rsvd = cpu_to_le16(0);
2837 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
2841 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
2842 * @tx_ring: the ring to be checked
2843 * @size: the size buffer we want to assure is available
2845 * Returns -EBUSY if a stop is needed, else 0
2847 int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
2849 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2850 /* Memory barrier before checking head and tail */
2853 /* Check again in a case another CPU has just made room available. */
2854 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
2857 /* A reprieve! - use start_queue because it doesn't call schedule */
2858 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2859 ++tx_ring->tx_stats.restart_queue;
2864 * __i40e_chk_linearize - Check if there are more than 8 buffers per packet
2867 * Note: Our HW can't DMA more than 8 buffers to build a packet on the wire
2868 * and so we need to figure out the cases where we need to linearize the skb.
2870 * For TSO we need to count the TSO header and segment payload separately.
2871 * As such we need to check cases where we have 7 fragments or more as we
2872 * can potentially require 9 DMA transactions, 1 for the TSO header, 1 for
2873 * the segment payload in the first descriptor, and another 7 for the
2876 bool __i40e_chk_linearize(struct sk_buff *skb)
2878 const struct skb_frag_struct *frag, *stale;
2881 /* no need to check if number of frags is less than 7 */
2882 nr_frags = skb_shinfo(skb)->nr_frags;
2883 if (nr_frags < (I40E_MAX_BUFFER_TXD - 1))
2886 /* We need to walk through the list and validate that each group
2887 * of 6 fragments totals at least gso_size.
2889 nr_frags -= I40E_MAX_BUFFER_TXD - 2;
2890 frag = &skb_shinfo(skb)->frags[0];
2892 /* Initialize size to the negative value of gso_size minus 1. We
2893 * use this as the worst case scenerio in which the frag ahead
2894 * of us only provides one byte which is why we are limited to 6
2895 * descriptors for a single transmit as the header and previous
2896 * fragment are already consuming 2 descriptors.
2898 sum = 1 - skb_shinfo(skb)->gso_size;
2900 /* Add size of frags 0 through 4 to create our initial sum */
2901 sum += skb_frag_size(frag++);
2902 sum += skb_frag_size(frag++);
2903 sum += skb_frag_size(frag++);
2904 sum += skb_frag_size(frag++);
2905 sum += skb_frag_size(frag++);
2907 /* Walk through fragments adding latest fragment, testing it, and
2908 * then removing stale fragments from the sum.
2910 stale = &skb_shinfo(skb)->frags[0];
2912 sum += skb_frag_size(frag++);
2914 /* if sum is negative we failed to make sufficient progress */
2921 sum -= skb_frag_size(stale++);
2928 * i40e_tx_map - Build the Tx descriptor
2929 * @tx_ring: ring to send buffer on
2931 * @first: first buffer info buffer to use
2932 * @tx_flags: collected send information
2933 * @hdr_len: size of the packet header
2934 * @td_cmd: the command field in the descriptor
2935 * @td_offset: offset for checksum or crc
2937 static inline void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
2938 struct i40e_tx_buffer *first, u32 tx_flags,
2939 const u8 hdr_len, u32 td_cmd, u32 td_offset)
2941 unsigned int data_len = skb->data_len;
2942 unsigned int size = skb_headlen(skb);
2943 struct skb_frag_struct *frag;
2944 struct i40e_tx_buffer *tx_bi;
2945 struct i40e_tx_desc *tx_desc;
2946 u16 i = tx_ring->next_to_use;
2951 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
2952 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
2953 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
2954 I40E_TX_FLAGS_VLAN_SHIFT;
2957 first->tx_flags = tx_flags;
2959 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
2961 tx_desc = I40E_TX_DESC(tx_ring, i);
2964 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
2965 unsigned int max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2967 if (dma_mapping_error(tx_ring->dev, dma))
2970 /* record length, and DMA address */
2971 dma_unmap_len_set(tx_bi, len, size);
2972 dma_unmap_addr_set(tx_bi, dma, dma);
2974 /* align size to end of page */
2975 max_data += -dma & (I40E_MAX_READ_REQ_SIZE - 1);
2976 tx_desc->buffer_addr = cpu_to_le64(dma);
2978 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
2979 tx_desc->cmd_type_offset_bsz =
2980 build_ctob(td_cmd, td_offset,
2987 if (i == tx_ring->count) {
2988 tx_desc = I40E_TX_DESC(tx_ring, 0);
2995 max_data = I40E_MAX_DATA_PER_TXD_ALIGNED;
2996 tx_desc->buffer_addr = cpu_to_le64(dma);
2999 if (likely(!data_len))
3002 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
3009 if (i == tx_ring->count) {
3010 tx_desc = I40E_TX_DESC(tx_ring, 0);
3014 size = skb_frag_size(frag);
3017 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3020 tx_bi = &tx_ring->tx_bi[i];
3023 netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
3026 if (i == tx_ring->count)
3029 tx_ring->next_to_use = i;
3031 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
3033 /* write last descriptor with EOP bit */
3034 td_cmd |= I40E_TX_DESC_CMD_EOP;
3036 /* We can OR these values together as they both are checked against
3037 * 4 below and at this point desc_count will be used as a boolean value
3038 * after this if/else block.
3040 desc_count |= ++tx_ring->packet_stride;
3042 /* Algorithm to optimize tail and RS bit setting:
3043 * if queue is stopped
3045 * reset packet counter
3046 * else if xmit_more is supported and is true
3047 * advance packet counter to 4
3048 * reset desc_count to 0
3050 * if desc_count >= 4
3052 * reset packet counter
3056 * Note: If there are less than 4 descriptors
3057 * pending and interrupts were disabled the service task will
3058 * trigger a force WB.
3060 if (netif_xmit_stopped(txring_txq(tx_ring))) {
3062 } else if (skb->xmit_more) {
3063 /* set stride to arm on next packet and reset desc_count */
3064 tx_ring->packet_stride = WB_STRIDE;
3066 } else if (desc_count >= WB_STRIDE) {
3068 /* write last descriptor with RS bit set */
3069 td_cmd |= I40E_TX_DESC_CMD_RS;
3070 tx_ring->packet_stride = 0;
3073 tx_desc->cmd_type_offset_bsz =
3074 build_ctob(td_cmd, td_offset, size, td_tag);
3076 /* Force memory writes to complete before letting h/w know there
3077 * are new descriptors to fetch.
3079 * We also use this memory barrier to make certain all of the
3080 * status bits have been updated before next_to_watch is written.
3084 /* set next_to_watch value indicating a packet is present */
3085 first->next_to_watch = tx_desc;
3087 /* notify HW of packet */
3089 writel(i, tx_ring->tail);
3091 /* we need this if more than one processor can write to our tail
3092 * at a time, it synchronizes IO on IA64/Altix systems
3100 dev_info(tx_ring->dev, "TX DMA map failed\n");
3102 /* clear dma mappings for failed tx_bi map */
3104 tx_bi = &tx_ring->tx_bi[i];
3105 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
3113 tx_ring->next_to_use = i;
3117 * i40e_xmit_frame_ring - Sends buffer on Tx ring
3119 * @tx_ring: ring to send buffer on
3121 * Returns NETDEV_TX_OK if sent, else an error code
3123 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
3124 struct i40e_ring *tx_ring)
3126 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
3127 u32 cd_tunneling = 0, cd_l2tag2 = 0;
3128 struct i40e_tx_buffer *first;
3137 /* prefetch the data, we'll need it later */
3138 prefetch(skb->data);
3140 i40e_trace(xmit_frame_ring, skb, tx_ring);
3142 count = i40e_xmit_descriptor_count(skb);
3143 if (i40e_chk_linearize(skb, count)) {
3144 if (__skb_linearize(skb)) {
3145 dev_kfree_skb_any(skb);
3146 return NETDEV_TX_OK;
3148 count = i40e_txd_use_count(skb->len);
3149 tx_ring->tx_stats.tx_linearize++;
3152 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
3153 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
3154 * + 4 desc gap to avoid the cache line where head is,
3155 * + 1 desc for context descriptor,
3156 * otherwise try next time
3158 if (i40e_maybe_stop_tx(tx_ring, count + 4 + 1)) {
3159 tx_ring->tx_stats.tx_busy++;
3160 return NETDEV_TX_BUSY;
3163 /* record the location of the first descriptor for this packet */
3164 first = &tx_ring->tx_bi[tx_ring->next_to_use];
3166 first->bytecount = skb->len;
3167 first->gso_segs = 1;
3169 /* prepare the xmit flags */
3170 if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
3173 /* obtain protocol of skb */
3174 protocol = vlan_get_protocol(skb);
3176 /* setup IPv4/IPv6 offloads */
3177 if (protocol == htons(ETH_P_IP))
3178 tx_flags |= I40E_TX_FLAGS_IPV4;
3179 else if (protocol == htons(ETH_P_IPV6))
3180 tx_flags |= I40E_TX_FLAGS_IPV6;
3182 tso = i40e_tso(first, &hdr_len, &cd_type_cmd_tso_mss);
3187 tx_flags |= I40E_TX_FLAGS_TSO;
3189 /* Always offload the checksum, since it's in the data descriptor */
3190 tso = i40e_tx_enable_csum(skb, &tx_flags, &td_cmd, &td_offset,
3191 tx_ring, &cd_tunneling);
3195 tsyn = i40e_tsyn(tx_ring, skb, tx_flags, &cd_type_cmd_tso_mss);
3198 tx_flags |= I40E_TX_FLAGS_TSYN;
3200 skb_tx_timestamp(skb);
3202 /* always enable CRC insertion offload */
3203 td_cmd |= I40E_TX_DESC_CMD_ICRC;
3205 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
3206 cd_tunneling, cd_l2tag2);
3208 /* Add Flow Director ATR if it's enabled.
3210 * NOTE: this must always be directly before the data descriptor.
3212 i40e_atr(tx_ring, skb, tx_flags);
3214 i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
3217 return NETDEV_TX_OK;
3220 i40e_trace(xmit_frame_ring_drop, first->skb, tx_ring);
3221 dev_kfree_skb_any(first->skb);
3223 return NETDEV_TX_OK;
3227 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
3229 * @netdev: network interface device structure
3231 * Returns NETDEV_TX_OK if sent, else an error code
3233 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3235 struct i40e_netdev_priv *np = netdev_priv(netdev);
3236 struct i40e_vsi *vsi = np->vsi;
3237 struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
3239 /* hardware can't handle really short frames, hardware padding works
3242 if (skb_put_padto(skb, I40E_MIN_TX_LEN))
3243 return NETDEV_TX_OK;
3245 return i40e_xmit_frame_ring(skb, tx_ring);