1 /*******************************************************************************
3 * Intel Ethernet Controller XL710 Family Linux Driver
4 * Copyright(c) 2013 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 with
16 * this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 * The full GNU General Public License is included in this distribution in
20 * the file called "COPYING".
22 * Contact Information:
23 * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 ******************************************************************************/
30 static inline __le64 build_ctob(u32 td_cmd, u32 td_offset, unsigned int size,
33 return cpu_to_le64(I40E_TX_DESC_DTYPE_DATA |
34 ((u64)td_cmd << I40E_TXD_QW1_CMD_SHIFT) |
35 ((u64)td_offset << I40E_TXD_QW1_OFFSET_SHIFT) |
36 ((u64)size << I40E_TXD_QW1_TX_BUF_SZ_SHIFT) |
37 ((u64)td_tag << I40E_TXD_QW1_L2TAG1_SHIFT));
40 #define I40E_TXD_CMD (I40E_TX_DESC_CMD_EOP | I40E_TX_DESC_CMD_RS)
42 * i40e_program_fdir_filter - Program a Flow Director filter
43 * @fdir_input: Packet data that will be filter parameters
45 * @add: True for add/update, False for remove
47 int i40e_program_fdir_filter(struct i40e_fdir_data *fdir_data,
48 struct i40e_pf *pf, bool add)
50 struct i40e_filter_program_desc *fdir_desc;
51 struct i40e_tx_buffer *tx_buf;
52 struct i40e_tx_desc *tx_desc;
53 struct i40e_ring *tx_ring;
54 unsigned int fpt, dcc;
61 /* find existing FDIR VSI */
63 for (i = 0; i < pf->hw.func_caps.num_vsis; i++)
64 if (pf->vsi[i] && pf->vsi[i]->type == I40E_VSI_FDIR)
69 tx_ring = vsi->tx_rings[0];
72 dma = dma_map_single(dev, fdir_data->raw_packet,
73 I40E_FDIR_MAX_RAW_PACKET_LOOKUP, DMA_TO_DEVICE);
74 if (dma_mapping_error(dev, dma))
77 /* grab the next descriptor */
78 i = tx_ring->next_to_use;
79 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
80 tx_buf = &tx_ring->tx_bi[i];
82 tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0;
84 fpt = (fdir_data->q_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
85 I40E_TXD_FLTR_QW0_QINDEX_MASK;
87 fpt |= (fdir_data->flex_off << I40E_TXD_FLTR_QW0_FLEXOFF_SHIFT) &
88 I40E_TXD_FLTR_QW0_FLEXOFF_MASK;
90 fpt |= (fdir_data->pctype << I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) &
91 I40E_TXD_FLTR_QW0_PCTYPE_MASK;
93 /* Use LAN VSI Id if not programmed by user */
94 if (fdir_data->dest_vsi == 0)
95 fpt |= (pf->vsi[pf->lan_vsi]->id) <<
96 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
98 fpt |= ((u32)fdir_data->dest_vsi <<
99 I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT) &
100 I40E_TXD_FLTR_QW0_DEST_VSI_MASK;
102 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(fpt);
104 dcc = I40E_TX_DESC_DTYPE_FILTER_PROG;
107 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
108 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
110 dcc |= I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
111 I40E_TXD_FLTR_QW1_PCMD_SHIFT;
113 dcc |= (fdir_data->dest_ctl << I40E_TXD_FLTR_QW1_DEST_SHIFT) &
114 I40E_TXD_FLTR_QW1_DEST_MASK;
116 dcc |= (fdir_data->fd_status << I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT) &
117 I40E_TXD_FLTR_QW1_FD_STATUS_MASK;
119 if (fdir_data->cnt_index != 0) {
120 dcc |= I40E_TXD_FLTR_QW1_CNT_ENA_MASK;
121 dcc |= ((u32)fdir_data->cnt_index <<
122 I40E_TXD_FLTR_QW1_CNTINDEX_SHIFT) &
123 I40E_TXD_FLTR_QW1_CNTINDEX_MASK;
126 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dcc);
127 fdir_desc->fd_id = cpu_to_le32(fdir_data->fd_id);
129 /* Now program a dummy descriptor */
130 i = tx_ring->next_to_use;
131 tx_desc = I40E_TX_DESC(tx_ring, i);
133 tx_ring->next_to_use = (i + 1 < tx_ring->count) ? i + 1 : 0;
135 tx_desc->buffer_addr = cpu_to_le64(dma);
136 td_cmd = I40E_TXD_CMD | I40E_TX_DESC_CMD_DUMMY;
138 tx_desc->cmd_type_offset_bsz =
139 build_ctob(td_cmd, 0, I40E_FDIR_MAX_RAW_PACKET_LOOKUP, 0);
141 /* Force memory writes to complete before letting h/w
142 * know there are new descriptors to fetch. (Only
143 * applicable for weak-ordered memory model archs,
148 /* Mark the data descriptor to be watched */
149 tx_buf->next_to_watch = tx_desc;
151 writel(tx_ring->next_to_use, tx_ring->tail);
159 * i40e_fd_handle_status - check the Programming Status for FD
160 * @rx_ring: the Rx ring for this descriptor
161 * @qw: the descriptor data
162 * @prog_id: the id originally used for programming
164 * This is used to verify if the FD programming or invalidation
165 * requested by SW to the HW is successful or not and take actions accordingly.
167 static void i40e_fd_handle_status(struct i40e_ring *rx_ring, u32 qw, u8 prog_id)
169 struct pci_dev *pdev = rx_ring->vsi->back->pdev;
172 error = (qw & I40E_RX_PROG_STATUS_DESC_QW1_ERROR_MASK) >>
173 I40E_RX_PROG_STATUS_DESC_QW1_ERROR_SHIFT;
175 /* for now just print the Status */
176 dev_info(&pdev->dev, "FD programming id %02x, Status %08x\n",
181 * i40e_unmap_and_free_tx_resource - Release a Tx buffer
182 * @ring: the ring that owns the buffer
183 * @tx_buffer: the buffer to free
185 static void i40e_unmap_and_free_tx_resource(struct i40e_ring *ring,
186 struct i40e_tx_buffer *tx_buffer)
188 if (tx_buffer->skb) {
189 dev_kfree_skb_any(tx_buffer->skb);
190 if (dma_unmap_len(tx_buffer, len))
191 dma_unmap_single(ring->dev,
192 dma_unmap_addr(tx_buffer, dma),
193 dma_unmap_len(tx_buffer, len),
195 } else if (dma_unmap_len(tx_buffer, len)) {
196 dma_unmap_page(ring->dev,
197 dma_unmap_addr(tx_buffer, dma),
198 dma_unmap_len(tx_buffer, len),
201 tx_buffer->next_to_watch = NULL;
202 tx_buffer->skb = NULL;
203 dma_unmap_len_set(tx_buffer, len, 0);
204 /* tx_buffer must be completely set up in the transmit path */
208 * i40e_clean_tx_ring - Free any empty Tx buffers
209 * @tx_ring: ring to be cleaned
211 void i40e_clean_tx_ring(struct i40e_ring *tx_ring)
213 unsigned long bi_size;
216 /* ring already cleared, nothing to do */
220 /* Free all the Tx ring sk_buffs */
221 for (i = 0; i < tx_ring->count; i++)
222 i40e_unmap_and_free_tx_resource(tx_ring, &tx_ring->tx_bi[i]);
224 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
225 memset(tx_ring->tx_bi, 0, bi_size);
227 /* Zero out the descriptor ring */
228 memset(tx_ring->desc, 0, tx_ring->size);
230 tx_ring->next_to_use = 0;
231 tx_ring->next_to_clean = 0;
233 if (!tx_ring->netdev)
236 /* cleanup Tx queue statistics */
237 netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
238 tx_ring->queue_index));
242 * i40e_free_tx_resources - Free Tx resources per queue
243 * @tx_ring: Tx descriptor ring for a specific queue
245 * Free all transmit software resources
247 void i40e_free_tx_resources(struct i40e_ring *tx_ring)
249 i40e_clean_tx_ring(tx_ring);
250 kfree(tx_ring->tx_bi);
251 tx_ring->tx_bi = NULL;
254 dma_free_coherent(tx_ring->dev, tx_ring->size,
255 tx_ring->desc, tx_ring->dma);
256 tx_ring->desc = NULL;
261 * i40e_get_tx_pending - how many tx descriptors not processed
262 * @tx_ring: the ring of descriptors
264 * Since there is no access to the ring head register
265 * in XL710, we need to use our local copies
267 static u32 i40e_get_tx_pending(struct i40e_ring *ring)
269 u32 ntu = ((ring->next_to_clean <= ring->next_to_use)
271 : ring->next_to_use + ring->count);
272 return ntu - ring->next_to_clean;
276 * i40e_check_tx_hang - Is there a hang in the Tx queue
277 * @tx_ring: the ring of descriptors
279 static bool i40e_check_tx_hang(struct i40e_ring *tx_ring)
281 u32 tx_pending = i40e_get_tx_pending(tx_ring);
284 clear_check_for_tx_hang(tx_ring);
286 /* Check for a hung queue, but be thorough. This verifies
287 * that a transmit has been completed since the previous
288 * check AND there is at least one packet pending. The
289 * ARMED bit is set to indicate a potential hang. The
290 * bit is cleared if a pause frame is received to remove
291 * false hang detection due to PFC or 802.3x frames. By
292 * requiring this to fail twice we avoid races with
293 * PFC clearing the ARMED bit and conditions where we
294 * run the check_tx_hang logic with a transmit completion
295 * pending but without time to complete it yet.
297 if ((tx_ring->tx_stats.tx_done_old == tx_ring->stats.packets) &&
299 /* make sure it is true for two checks in a row */
300 ret = test_and_set_bit(__I40E_HANG_CHECK_ARMED,
303 /* update completed stats and disarm the hang check */
304 tx_ring->tx_stats.tx_done_old = tx_ring->stats.packets;
305 clear_bit(__I40E_HANG_CHECK_ARMED, &tx_ring->state);
312 * i40e_clean_tx_irq - Reclaim resources after transmit completes
313 * @tx_ring: tx ring to clean
314 * @budget: how many cleans we're allowed
316 * Returns true if there's any budget left (e.g. the clean is finished)
318 static bool i40e_clean_tx_irq(struct i40e_ring *tx_ring, int budget)
320 u16 i = tx_ring->next_to_clean;
321 struct i40e_tx_buffer *tx_buf;
322 struct i40e_tx_desc *tx_desc;
323 unsigned int total_packets = 0;
324 unsigned int total_bytes = 0;
326 tx_buf = &tx_ring->tx_bi[i];
327 tx_desc = I40E_TX_DESC(tx_ring, i);
331 struct i40e_tx_desc *eop_desc = tx_buf->next_to_watch;
333 /* if next_to_watch is not set then there is no work pending */
337 /* prevent any other reads prior to eop_desc */
338 read_barrier_depends();
340 /* if the descriptor isn't done, no work yet to do */
341 if (!(eop_desc->cmd_type_offset_bsz &
342 cpu_to_le64(I40E_TX_DESC_DTYPE_DESC_DONE)))
345 /* clear next_to_watch to prevent false hangs */
346 tx_buf->next_to_watch = NULL;
348 /* update the statistics for this packet */
349 total_bytes += tx_buf->bytecount;
350 total_packets += tx_buf->gso_segs;
353 dev_kfree_skb_any(tx_buf->skb);
355 /* unmap skb header data */
356 dma_unmap_single(tx_ring->dev,
357 dma_unmap_addr(tx_buf, dma),
358 dma_unmap_len(tx_buf, len),
361 /* clear tx_buffer data */
363 dma_unmap_len_set(tx_buf, len, 0);
365 /* unmap remaining buffers */
366 while (tx_desc != eop_desc) {
373 tx_buf = tx_ring->tx_bi;
374 tx_desc = I40E_TX_DESC(tx_ring, 0);
377 /* unmap any remaining paged data */
378 if (dma_unmap_len(tx_buf, len)) {
379 dma_unmap_page(tx_ring->dev,
380 dma_unmap_addr(tx_buf, dma),
381 dma_unmap_len(tx_buf, len),
383 dma_unmap_len_set(tx_buf, len, 0);
387 /* move us one more past the eop_desc for start of next pkt */
393 tx_buf = tx_ring->tx_bi;
394 tx_desc = I40E_TX_DESC(tx_ring, 0);
397 /* update budget accounting */
399 } while (likely(budget));
402 tx_ring->next_to_clean = i;
403 u64_stats_update_begin(&tx_ring->syncp);
404 tx_ring->stats.bytes += total_bytes;
405 tx_ring->stats.packets += total_packets;
406 u64_stats_update_end(&tx_ring->syncp);
407 tx_ring->q_vector->tx.total_bytes += total_bytes;
408 tx_ring->q_vector->tx.total_packets += total_packets;
410 if (check_for_tx_hang(tx_ring) && i40e_check_tx_hang(tx_ring)) {
411 /* schedule immediate reset if we believe we hung */
412 dev_info(tx_ring->dev, "Detected Tx Unit Hang\n"
415 " next_to_use <%x>\n"
416 " next_to_clean <%x>\n",
418 tx_ring->queue_index,
419 tx_ring->next_to_use, i);
420 dev_info(tx_ring->dev, "tx_bi[next_to_clean]\n"
421 " time_stamp <%lx>\n"
423 tx_ring->tx_bi[i].time_stamp, jiffies);
425 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
427 dev_info(tx_ring->dev,
428 "tx hang detected on queue %d, resetting adapter\n",
429 tx_ring->queue_index);
431 tx_ring->netdev->netdev_ops->ndo_tx_timeout(tx_ring->netdev);
433 /* the adapter is about to reset, no point in enabling stuff */
437 netdev_tx_completed_queue(netdev_get_tx_queue(tx_ring->netdev,
438 tx_ring->queue_index),
439 total_packets, total_bytes);
441 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
442 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
443 (I40E_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
444 /* Make sure that anybody stopping the queue after this
445 * sees the new next_to_clean.
448 if (__netif_subqueue_stopped(tx_ring->netdev,
449 tx_ring->queue_index) &&
450 !test_bit(__I40E_DOWN, &tx_ring->vsi->state)) {
451 netif_wake_subqueue(tx_ring->netdev,
452 tx_ring->queue_index);
453 ++tx_ring->tx_stats.restart_queue;
461 * i40e_set_new_dynamic_itr - Find new ITR level
462 * @rc: structure containing ring performance data
464 * Stores a new ITR value based on packets and byte counts during
465 * the last interrupt. The advantage of per interrupt computation
466 * is faster updates and more accurate ITR for the current traffic
467 * pattern. Constants in this function were computed based on
468 * theoretical maximum wire speed and thresholds were set based on
469 * testing data as well as attempting to minimize response time
470 * while increasing bulk throughput.
472 static void i40e_set_new_dynamic_itr(struct i40e_ring_container *rc)
474 enum i40e_latency_range new_latency_range = rc->latency_range;
475 u32 new_itr = rc->itr;
478 if (rc->total_packets == 0 || !rc->itr)
481 /* simple throttlerate management
482 * 0-10MB/s lowest (100000 ints/s)
483 * 10-20MB/s low (20000 ints/s)
484 * 20-1249MB/s bulk (8000 ints/s)
486 bytes_per_int = rc->total_bytes / rc->itr;
488 case I40E_LOWEST_LATENCY:
489 if (bytes_per_int > 10)
490 new_latency_range = I40E_LOW_LATENCY;
492 case I40E_LOW_LATENCY:
493 if (bytes_per_int > 20)
494 new_latency_range = I40E_BULK_LATENCY;
495 else if (bytes_per_int <= 10)
496 new_latency_range = I40E_LOWEST_LATENCY;
498 case I40E_BULK_LATENCY:
499 if (bytes_per_int <= 20)
500 rc->latency_range = I40E_LOW_LATENCY;
504 switch (new_latency_range) {
505 case I40E_LOWEST_LATENCY:
506 new_itr = I40E_ITR_100K;
508 case I40E_LOW_LATENCY:
509 new_itr = I40E_ITR_20K;
511 case I40E_BULK_LATENCY:
512 new_itr = I40E_ITR_8K;
518 if (new_itr != rc->itr) {
519 /* do an exponential smoothing */
520 new_itr = (10 * new_itr * rc->itr) /
521 ((9 * new_itr) + rc->itr);
522 rc->itr = new_itr & I40E_MAX_ITR;
526 rc->total_packets = 0;
530 * i40e_update_dynamic_itr - Adjust ITR based on bytes per int
531 * @q_vector: the vector to adjust
533 static void i40e_update_dynamic_itr(struct i40e_q_vector *q_vector)
535 u16 vector = q_vector->vsi->base_vector + q_vector->v_idx;
536 struct i40e_hw *hw = &q_vector->vsi->back->hw;
540 reg_addr = I40E_PFINT_ITRN(I40E_RX_ITR, vector - 1);
541 old_itr = q_vector->rx.itr;
542 i40e_set_new_dynamic_itr(&q_vector->rx);
543 if (old_itr != q_vector->rx.itr)
544 wr32(hw, reg_addr, q_vector->rx.itr);
546 reg_addr = I40E_PFINT_ITRN(I40E_TX_ITR, vector - 1);
547 old_itr = q_vector->tx.itr;
548 i40e_set_new_dynamic_itr(&q_vector->tx);
549 if (old_itr != q_vector->tx.itr)
550 wr32(hw, reg_addr, q_vector->tx.itr);
554 * i40e_clean_programming_status - clean the programming status descriptor
555 * @rx_ring: the rx ring that has this descriptor
556 * @rx_desc: the rx descriptor written back by HW
558 * Flow director should handle FD_FILTER_STATUS to check its filter programming
559 * status being successful or not and take actions accordingly. FCoE should
560 * handle its context/filter programming/invalidation status and take actions.
563 static void i40e_clean_programming_status(struct i40e_ring *rx_ring,
564 union i40e_rx_desc *rx_desc)
569 qw = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
570 id = (qw & I40E_RX_PROG_STATUS_DESC_QW1_PROGID_MASK) >>
571 I40E_RX_PROG_STATUS_DESC_QW1_PROGID_SHIFT;
573 if (id == I40E_RX_PROG_STATUS_DESC_FD_FILTER_STATUS)
574 i40e_fd_handle_status(rx_ring, qw, id);
578 * i40e_setup_tx_descriptors - Allocate the Tx descriptors
579 * @tx_ring: the tx ring to set up
581 * Return 0 on success, negative on error
583 int i40e_setup_tx_descriptors(struct i40e_ring *tx_ring)
585 struct device *dev = tx_ring->dev;
591 bi_size = sizeof(struct i40e_tx_buffer) * tx_ring->count;
592 tx_ring->tx_bi = kzalloc(bi_size, GFP_KERNEL);
596 /* round up to nearest 4K */
597 tx_ring->size = tx_ring->count * sizeof(struct i40e_tx_desc);
598 tx_ring->size = ALIGN(tx_ring->size, 4096);
599 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
600 &tx_ring->dma, GFP_KERNEL);
601 if (!tx_ring->desc) {
602 dev_info(dev, "Unable to allocate memory for the Tx descriptor ring, size=%d\n",
607 tx_ring->next_to_use = 0;
608 tx_ring->next_to_clean = 0;
612 kfree(tx_ring->tx_bi);
613 tx_ring->tx_bi = NULL;
618 * i40e_clean_rx_ring - Free Rx buffers
619 * @rx_ring: ring to be cleaned
621 void i40e_clean_rx_ring(struct i40e_ring *rx_ring)
623 struct device *dev = rx_ring->dev;
624 struct i40e_rx_buffer *rx_bi;
625 unsigned long bi_size;
628 /* ring already cleared, nothing to do */
632 /* Free all the Rx ring sk_buffs */
633 for (i = 0; i < rx_ring->count; i++) {
634 rx_bi = &rx_ring->rx_bi[i];
636 dma_unmap_single(dev,
643 dev_kfree_skb(rx_bi->skb);
647 if (rx_bi->page_dma) {
654 __free_page(rx_bi->page);
656 rx_bi->page_offset = 0;
660 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
661 memset(rx_ring->rx_bi, 0, bi_size);
663 /* Zero out the descriptor ring */
664 memset(rx_ring->desc, 0, rx_ring->size);
666 rx_ring->next_to_clean = 0;
667 rx_ring->next_to_use = 0;
671 * i40e_free_rx_resources - Free Rx resources
672 * @rx_ring: ring to clean the resources from
674 * Free all receive software resources
676 void i40e_free_rx_resources(struct i40e_ring *rx_ring)
678 i40e_clean_rx_ring(rx_ring);
679 kfree(rx_ring->rx_bi);
680 rx_ring->rx_bi = NULL;
683 dma_free_coherent(rx_ring->dev, rx_ring->size,
684 rx_ring->desc, rx_ring->dma);
685 rx_ring->desc = NULL;
690 * i40e_setup_rx_descriptors - Allocate Rx descriptors
691 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
693 * Returns 0 on success, negative on failure
695 int i40e_setup_rx_descriptors(struct i40e_ring *rx_ring)
697 struct device *dev = rx_ring->dev;
700 bi_size = sizeof(struct i40e_rx_buffer) * rx_ring->count;
701 rx_ring->rx_bi = kzalloc(bi_size, GFP_KERNEL);
705 /* Round up to nearest 4K */
706 rx_ring->size = ring_is_16byte_desc_enabled(rx_ring)
707 ? rx_ring->count * sizeof(union i40e_16byte_rx_desc)
708 : rx_ring->count * sizeof(union i40e_32byte_rx_desc);
709 rx_ring->size = ALIGN(rx_ring->size, 4096);
710 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
711 &rx_ring->dma, GFP_KERNEL);
713 if (!rx_ring->desc) {
714 dev_info(dev, "Unable to allocate memory for the Rx descriptor ring, size=%d\n",
719 rx_ring->next_to_clean = 0;
720 rx_ring->next_to_use = 0;
724 kfree(rx_ring->rx_bi);
725 rx_ring->rx_bi = NULL;
730 * i40e_release_rx_desc - Store the new tail and head values
731 * @rx_ring: ring to bump
732 * @val: new head index
734 static inline void i40e_release_rx_desc(struct i40e_ring *rx_ring, u32 val)
736 rx_ring->next_to_use = val;
737 /* Force memory writes to complete before letting h/w
738 * know there are new descriptors to fetch. (Only
739 * applicable for weak-ordered memory model archs,
743 writel(val, rx_ring->tail);
747 * i40e_alloc_rx_buffers - Replace used receive buffers; packet split
748 * @rx_ring: ring to place buffers on
749 * @cleaned_count: number of buffers to replace
751 void i40e_alloc_rx_buffers(struct i40e_ring *rx_ring, u16 cleaned_count)
753 u16 i = rx_ring->next_to_use;
754 union i40e_rx_desc *rx_desc;
755 struct i40e_rx_buffer *bi;
758 /* do nothing if no valid netdev defined */
759 if (!rx_ring->netdev || !cleaned_count)
762 while (cleaned_count--) {
763 rx_desc = I40E_RX_DESC(rx_ring, i);
764 bi = &rx_ring->rx_bi[i];
768 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
769 rx_ring->rx_buf_len);
771 rx_ring->rx_stats.alloc_rx_buff_failed++;
774 /* initialize queue mapping */
775 skb_record_rx_queue(skb, rx_ring->queue_index);
780 bi->dma = dma_map_single(rx_ring->dev,
784 if (dma_mapping_error(rx_ring->dev, bi->dma)) {
785 rx_ring->rx_stats.alloc_rx_buff_failed++;
791 if (ring_is_ps_enabled(rx_ring)) {
793 bi->page = alloc_page(GFP_ATOMIC);
795 rx_ring->rx_stats.alloc_rx_page_failed++;
801 /* use a half page if we're re-using */
802 bi->page_offset ^= PAGE_SIZE / 2;
803 bi->page_dma = dma_map_page(rx_ring->dev,
808 if (dma_mapping_error(rx_ring->dev,
810 rx_ring->rx_stats.alloc_rx_page_failed++;
816 /* Refresh the desc even if buffer_addrs didn't change
817 * because each write-back erases this info.
819 rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
820 rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
822 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
823 rx_desc->read.hdr_addr = 0;
826 if (i == rx_ring->count)
831 if (rx_ring->next_to_use != i)
832 i40e_release_rx_desc(rx_ring, i);
836 * i40e_receive_skb - Send a completed packet up the stack
837 * @rx_ring: rx ring in play
838 * @skb: packet to send up
839 * @vlan_tag: vlan tag for packet
841 static void i40e_receive_skb(struct i40e_ring *rx_ring,
842 struct sk_buff *skb, u16 vlan_tag)
844 struct i40e_q_vector *q_vector = rx_ring->q_vector;
845 struct i40e_vsi *vsi = rx_ring->vsi;
846 u64 flags = vsi->back->flags;
848 if (vlan_tag & VLAN_VID_MASK)
849 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tag);
851 if (flags & I40E_FLAG_IN_NETPOLL)
854 napi_gro_receive(&q_vector->napi, skb);
858 * i40e_rx_checksum - Indicate in skb if hw indicated a good cksum
859 * @vsi: the VSI we care about
860 * @skb: skb currently being received and modified
861 * @rx_status: status value of last descriptor in packet
862 * @rx_error: error value of last descriptor in packet
863 * @rx_ptype: ptype value of last descriptor in packet
865 static inline void i40e_rx_checksum(struct i40e_vsi *vsi,
871 bool ipv4_tunnel, ipv6_tunnel;
876 ipv4_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT4_MAC_PAY3) &&
877 (rx_ptype < I40E_RX_PTYPE_GRENAT4_MACVLAN_IPV6_ICMP_PAY4);
878 ipv6_tunnel = (rx_ptype > I40E_RX_PTYPE_GRENAT6_MAC_PAY3) &&
879 (rx_ptype < I40E_RX_PTYPE_GRENAT6_MACVLAN_IPV6_ICMP_PAY4);
881 skb->encapsulation = ipv4_tunnel || ipv6_tunnel;
882 skb->ip_summed = CHECKSUM_NONE;
884 /* Rx csum enabled and ip headers found? */
885 if (!(vsi->netdev->features & NETIF_F_RXCSUM &&
886 rx_status & (1 << I40E_RX_DESC_STATUS_L3L4P_SHIFT)))
889 /* IP or L4 or outmost IP checksum error */
890 if (rx_error & ((1 << I40E_RX_DESC_ERROR_IPE_SHIFT) |
891 (1 << I40E_RX_DESC_ERROR_L4E_SHIFT) |
892 (1 << I40E_RX_DESC_ERROR_EIPE_SHIFT))) {
893 vsi->back->hw_csum_rx_error++;
898 !(rx_status & (1 << I40E_RX_DESC_STATUS_UDP_0_SHIFT))) {
899 /* If VXLAN traffic has an outer UDPv4 checksum we need to check
900 * it in the driver, hardware does not do it for us.
901 * Since L3L4P bit was set we assume a valid IHL value (>=5)
902 * so the total length of IPv4 header is IHL*4 bytes
904 skb->transport_header = skb->mac_header +
905 sizeof(struct ethhdr) +
906 (ip_hdr(skb)->ihl * 4);
908 /* Add 4 bytes for VLAN tagged packets */
909 skb->transport_header += (skb->protocol == htons(ETH_P_8021Q) ||
910 skb->protocol == htons(ETH_P_8021AD))
913 rx_udp_csum = udp_csum(skb);
915 csum = csum_tcpudp_magic(
916 iph->saddr, iph->daddr,
917 (skb->len - skb_transport_offset(skb)),
918 IPPROTO_UDP, rx_udp_csum);
920 if (udp_hdr(skb)->check != csum) {
921 vsi->back->hw_csum_rx_error++;
926 skb->ip_summed = CHECKSUM_UNNECESSARY;
930 * i40e_rx_hash - returns the hash value from the Rx descriptor
931 * @ring: descriptor ring
932 * @rx_desc: specific descriptor
934 static inline u32 i40e_rx_hash(struct i40e_ring *ring,
935 union i40e_rx_desc *rx_desc)
937 const __le64 rss_mask =
938 cpu_to_le64((u64)I40E_RX_DESC_FLTSTAT_RSS_HASH <<
939 I40E_RX_DESC_STATUS_FLTSTAT_SHIFT);
941 if ((ring->netdev->features & NETIF_F_RXHASH) &&
942 (rx_desc->wb.qword1.status_error_len & rss_mask) == rss_mask)
943 return le32_to_cpu(rx_desc->wb.qword0.hi_dword.rss);
949 * i40e_clean_rx_irq - Reclaim resources after receive completes
950 * @rx_ring: rx ring to clean
951 * @budget: how many cleans we're allowed
953 * Returns true if there's any budget left (e.g. the clean is finished)
955 static int i40e_clean_rx_irq(struct i40e_ring *rx_ring, int budget)
957 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
958 u16 rx_packet_len, rx_header_len, rx_sph, rx_hbo;
959 u16 cleaned_count = I40E_DESC_UNUSED(rx_ring);
960 const int current_node = numa_node_id();
961 struct i40e_vsi *vsi = rx_ring->vsi;
962 u16 i = rx_ring->next_to_clean;
963 union i40e_rx_desc *rx_desc;
964 u32 rx_error, rx_status;
968 rx_desc = I40E_RX_DESC(rx_ring, i);
969 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
970 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK)
971 >> I40E_RXD_QW1_STATUS_SHIFT;
973 while (rx_status & (1 << I40E_RX_DESC_STATUS_DD_SHIFT)) {
974 union i40e_rx_desc *next_rxd;
975 struct i40e_rx_buffer *rx_bi;
978 if (i40e_rx_is_programming_status(qword)) {
979 i40e_clean_programming_status(rx_ring, rx_desc);
980 I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd);
983 rx_bi = &rx_ring->rx_bi[i];
987 rx_packet_len = (qword & I40E_RXD_QW1_LENGTH_PBUF_MASK)
988 >> I40E_RXD_QW1_LENGTH_PBUF_SHIFT;
989 rx_header_len = (qword & I40E_RXD_QW1_LENGTH_HBUF_MASK)
990 >> I40E_RXD_QW1_LENGTH_HBUF_SHIFT;
991 rx_sph = (qword & I40E_RXD_QW1_LENGTH_SPH_MASK)
992 >> I40E_RXD_QW1_LENGTH_SPH_SHIFT;
994 rx_error = (qword & I40E_RXD_QW1_ERROR_MASK)
995 >> I40E_RXD_QW1_ERROR_SHIFT;
996 rx_hbo = rx_error & (1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
997 rx_error &= ~(1 << I40E_RX_DESC_ERROR_HBO_SHIFT);
999 rx_ptype = (qword & I40E_RXD_QW1_PTYPE_MASK) >>
1000 I40E_RXD_QW1_PTYPE_SHIFT;
1003 /* This memory barrier is needed to keep us from reading
1004 * any other fields out of the rx_desc until we know the
1005 * STATUS_DD bit is set
1009 /* Get the header and possibly the whole packet
1010 * If this is an skb from previous receive dma will be 0
1016 len = I40E_RX_HDR_SIZE;
1018 len = rx_header_len;
1019 else if (rx_packet_len)
1020 len = rx_packet_len; /* 1buf/no split found */
1022 len = rx_header_len; /* split always mode */
1025 dma_unmap_single(rx_ring->dev,
1027 rx_ring->rx_buf_len,
1032 /* Get the rest of the data if this was a header split */
1033 if (ring_is_ps_enabled(rx_ring) && rx_packet_len) {
1035 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
1040 skb->len += rx_packet_len;
1041 skb->data_len += rx_packet_len;
1042 skb->truesize += rx_packet_len;
1044 if ((page_count(rx_bi->page) == 1) &&
1045 (page_to_nid(rx_bi->page) == current_node))
1046 get_page(rx_bi->page);
1050 dma_unmap_page(rx_ring->dev,
1054 rx_bi->page_dma = 0;
1056 I40E_RX_NEXT_DESC_PREFETCH(rx_ring, i, next_rxd);
1059 !(rx_status & (1 << I40E_RX_DESC_STATUS_EOF_SHIFT)))) {
1060 struct i40e_rx_buffer *next_buffer;
1062 next_buffer = &rx_ring->rx_bi[i];
1064 if (ring_is_ps_enabled(rx_ring)) {
1065 rx_bi->skb = next_buffer->skb;
1066 rx_bi->dma = next_buffer->dma;
1067 next_buffer->skb = skb;
1068 next_buffer->dma = 0;
1070 rx_ring->rx_stats.non_eop_descs++;
1074 /* ERR_MASK will only have valid bits if EOP set */
1075 if (unlikely(rx_error & (1 << I40E_RX_DESC_ERROR_RXE_SHIFT))) {
1076 dev_kfree_skb_any(skb);
1080 skb->rxhash = i40e_rx_hash(rx_ring, rx_desc);
1081 /* probably a little skewed due to removing CRC */
1082 total_rx_bytes += skb->len;
1085 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1087 i40e_rx_checksum(vsi, skb, rx_status, rx_error, rx_ptype);
1089 vlan_tag = rx_status & (1 << I40E_RX_DESC_STATUS_L2TAG1P_SHIFT)
1090 ? le16_to_cpu(rx_desc->wb.qword0.lo_dword.l2tag1)
1092 i40e_receive_skb(rx_ring, skb, vlan_tag);
1094 rx_ring->netdev->last_rx = jiffies;
1097 rx_desc->wb.qword1.status_error_len = 0;
1102 /* return some buffers to hardware, one at a time is too slow */
1103 if (cleaned_count >= I40E_RX_BUFFER_WRITE) {
1104 i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1108 /* use prefetched values */
1110 qword = le64_to_cpu(rx_desc->wb.qword1.status_error_len);
1111 rx_status = (qword & I40E_RXD_QW1_STATUS_MASK)
1112 >> I40E_RXD_QW1_STATUS_SHIFT;
1115 rx_ring->next_to_clean = i;
1116 u64_stats_update_begin(&rx_ring->syncp);
1117 rx_ring->stats.packets += total_rx_packets;
1118 rx_ring->stats.bytes += total_rx_bytes;
1119 u64_stats_update_end(&rx_ring->syncp);
1120 rx_ring->q_vector->rx.total_packets += total_rx_packets;
1121 rx_ring->q_vector->rx.total_bytes += total_rx_bytes;
1124 i40e_alloc_rx_buffers(rx_ring, cleaned_count);
1130 * i40e_napi_poll - NAPI polling Rx/Tx cleanup routine
1131 * @napi: napi struct with our devices info in it
1132 * @budget: amount of work driver is allowed to do this pass, in packets
1134 * This function will clean all queues associated with a q_vector.
1136 * Returns the amount of work done
1138 int i40e_napi_poll(struct napi_struct *napi, int budget)
1140 struct i40e_q_vector *q_vector =
1141 container_of(napi, struct i40e_q_vector, napi);
1142 struct i40e_vsi *vsi = q_vector->vsi;
1143 struct i40e_ring *ring;
1144 bool clean_complete = true;
1145 int budget_per_ring;
1147 if (test_bit(__I40E_DOWN, &vsi->state)) {
1148 napi_complete(napi);
1152 /* Since the actual Tx work is minimal, we can give the Tx a larger
1153 * budget and be more aggressive about cleaning up the Tx descriptors.
1155 i40e_for_each_ring(ring, q_vector->tx)
1156 clean_complete &= i40e_clean_tx_irq(ring, vsi->work_limit);
1158 /* We attempt to distribute budget to each Rx queue fairly, but don't
1159 * allow the budget to go below 1 because that would exit polling early.
1161 budget_per_ring = max(budget/q_vector->num_ringpairs, 1);
1163 i40e_for_each_ring(ring, q_vector->rx)
1164 clean_complete &= i40e_clean_rx_irq(ring, budget_per_ring);
1166 /* If work not completed, return budget and polling will return */
1167 if (!clean_complete)
1170 /* Work is done so exit the polling mode and re-enable the interrupt */
1171 napi_complete(napi);
1172 if (ITR_IS_DYNAMIC(vsi->rx_itr_setting) ||
1173 ITR_IS_DYNAMIC(vsi->tx_itr_setting))
1174 i40e_update_dynamic_itr(q_vector);
1176 if (!test_bit(__I40E_DOWN, &vsi->state)) {
1177 if (vsi->back->flags & I40E_FLAG_MSIX_ENABLED) {
1178 i40e_irq_dynamic_enable(vsi,
1179 q_vector->v_idx + vsi->base_vector);
1181 struct i40e_hw *hw = &vsi->back->hw;
1182 /* We re-enable the queue 0 cause, but
1183 * don't worry about dynamic_enable
1184 * because we left it on for the other
1185 * possible interrupts during napi
1187 u32 qval = rd32(hw, I40E_QINT_RQCTL(0));
1188 qval |= I40E_QINT_RQCTL_CAUSE_ENA_MASK;
1189 wr32(hw, I40E_QINT_RQCTL(0), qval);
1191 qval = rd32(hw, I40E_QINT_TQCTL(0));
1192 qval |= I40E_QINT_TQCTL_CAUSE_ENA_MASK;
1193 wr32(hw, I40E_QINT_TQCTL(0), qval);
1195 i40e_irq_dynamic_enable_icr0(vsi->back);
1203 * i40e_atr - Add a Flow Director ATR filter
1204 * @tx_ring: ring to add programming descriptor to
1206 * @flags: send flags
1207 * @protocol: wire protocol
1209 static void i40e_atr(struct i40e_ring *tx_ring, struct sk_buff *skb,
1210 u32 flags, __be16 protocol)
1212 struct i40e_filter_program_desc *fdir_desc;
1213 struct i40e_pf *pf = tx_ring->vsi->back;
1215 unsigned char *network;
1217 struct ipv6hdr *ipv6;
1221 u32 flex_ptype, dtype_cmd;
1224 /* make sure ATR is enabled */
1225 if (!(pf->flags & I40E_FLAG_FDIR_ATR_ENABLED))
1228 /* if sampling is disabled do nothing */
1229 if (!tx_ring->atr_sample_rate)
1232 tx_ring->atr_count++;
1234 /* snag network header to get L4 type and address */
1235 hdr.network = skb_network_header(skb);
1237 /* Currently only IPv4/IPv6 with TCP is supported */
1238 if (protocol == htons(ETH_P_IP)) {
1239 if (hdr.ipv4->protocol != IPPROTO_TCP)
1242 /* access ihl as a u8 to avoid unaligned access on ia64 */
1243 hlen = (hdr.network[0] & 0x0F) << 2;
1244 } else if (protocol == htons(ETH_P_IPV6)) {
1245 if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1248 hlen = sizeof(struct ipv6hdr);
1253 th = (struct tcphdr *)(hdr.network + hlen);
1255 /* sample on all syn/fin packets or once every atr sample rate */
1256 if (!th->fin && !th->syn && (tx_ring->atr_count < tx_ring->atr_sample_rate))
1259 tx_ring->atr_count = 0;
1261 /* grab the next descriptor */
1262 i = tx_ring->next_to_use;
1263 fdir_desc = I40E_TX_FDIRDESC(tx_ring, i);
1266 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1268 flex_ptype = (tx_ring->queue_index << I40E_TXD_FLTR_QW0_QINDEX_SHIFT) &
1269 I40E_TXD_FLTR_QW0_QINDEX_MASK;
1270 flex_ptype |= (protocol == htons(ETH_P_IP)) ?
1271 (I40E_FILTER_PCTYPE_NONF_IPV4_TCP <<
1272 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT) :
1273 (I40E_FILTER_PCTYPE_NONF_IPV6_TCP <<
1274 I40E_TXD_FLTR_QW0_PCTYPE_SHIFT);
1276 flex_ptype |= tx_ring->vsi->id << I40E_TXD_FLTR_QW0_DEST_VSI_SHIFT;
1278 dtype_cmd = I40E_TX_DESC_DTYPE_FILTER_PROG;
1280 dtype_cmd |= th->fin ?
1281 (I40E_FILTER_PROGRAM_DESC_PCMD_REMOVE <<
1282 I40E_TXD_FLTR_QW1_PCMD_SHIFT) :
1283 (I40E_FILTER_PROGRAM_DESC_PCMD_ADD_UPDATE <<
1284 I40E_TXD_FLTR_QW1_PCMD_SHIFT);
1286 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_DEST_DIRECT_PACKET_QINDEX <<
1287 I40E_TXD_FLTR_QW1_DEST_SHIFT;
1289 dtype_cmd |= I40E_FILTER_PROGRAM_DESC_FD_STATUS_FD_ID <<
1290 I40E_TXD_FLTR_QW1_FD_STATUS_SHIFT;
1292 fdir_desc->qindex_flex_ptype_vsi = cpu_to_le32(flex_ptype);
1293 fdir_desc->dtype_cmd_cntindex = cpu_to_le32(dtype_cmd);
1297 * i40e_tx_prepare_vlan_flags - prepare generic TX VLAN tagging flags for HW
1299 * @tx_ring: ring to send buffer on
1300 * @flags: the tx flags to be set
1302 * Checks the skb and set up correspondingly several generic transmit flags
1303 * related to VLAN tagging for the HW, such as VLAN, DCB, etc.
1305 * Returns error code indicate the frame should be dropped upon error and the
1306 * otherwise returns 0 to indicate the flags has been set properly.
1308 static int i40e_tx_prepare_vlan_flags(struct sk_buff *skb,
1309 struct i40e_ring *tx_ring,
1312 __be16 protocol = skb->protocol;
1315 /* if we have a HW VLAN tag being added, default to the HW one */
1316 if (vlan_tx_tag_present(skb)) {
1317 tx_flags |= vlan_tx_tag_get(skb) << I40E_TX_FLAGS_VLAN_SHIFT;
1318 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1319 /* else if it is a SW VLAN, check the next protocol and store the tag */
1320 } else if (protocol == htons(ETH_P_8021Q)) {
1321 struct vlan_hdr *vhdr, _vhdr;
1322 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
1326 protocol = vhdr->h_vlan_encapsulated_proto;
1327 tx_flags |= ntohs(vhdr->h_vlan_TCI) << I40E_TX_FLAGS_VLAN_SHIFT;
1328 tx_flags |= I40E_TX_FLAGS_SW_VLAN;
1331 /* Insert 802.1p priority into VLAN header */
1332 if ((tx_ring->vsi->back->flags & I40E_FLAG_DCB_ENABLED) &&
1333 ((tx_flags & (I40E_TX_FLAGS_HW_VLAN | I40E_TX_FLAGS_SW_VLAN)) ||
1334 (skb->priority != TC_PRIO_CONTROL))) {
1335 tx_flags &= ~I40E_TX_FLAGS_VLAN_PRIO_MASK;
1336 tx_flags |= (skb->priority & 0x7) <<
1337 I40E_TX_FLAGS_VLAN_PRIO_SHIFT;
1338 if (tx_flags & I40E_TX_FLAGS_SW_VLAN) {
1339 struct vlan_ethhdr *vhdr;
1340 if (skb_header_cloned(skb) &&
1341 pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
1343 vhdr = (struct vlan_ethhdr *)skb->data;
1344 vhdr->h_vlan_TCI = htons(tx_flags >>
1345 I40E_TX_FLAGS_VLAN_SHIFT);
1347 tx_flags |= I40E_TX_FLAGS_HW_VLAN;
1355 * i40e_tso - set up the tso context descriptor
1356 * @tx_ring: ptr to the ring to send
1357 * @skb: ptr to the skb we're sending
1358 * @tx_flags: the collected send information
1359 * @protocol: the send protocol
1360 * @hdr_len: ptr to the size of the packet header
1361 * @cd_tunneling: ptr to context descriptor bits
1363 * Returns 0 if no TSO can happen, 1 if tso is going, or error
1365 static int i40e_tso(struct i40e_ring *tx_ring, struct sk_buff *skb,
1366 u32 tx_flags, __be16 protocol, u8 *hdr_len,
1367 u64 *cd_type_cmd_tso_mss, u32 *cd_tunneling)
1369 u32 cd_cmd, cd_tso_len, cd_mss;
1370 struct tcphdr *tcph;
1374 struct ipv6hdr *ipv6h;
1376 if (!skb_is_gso(skb))
1379 if (skb_header_cloned(skb)) {
1380 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1385 if (protocol == htons(ETH_P_IP)) {
1386 iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
1387 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1390 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1392 } else if (skb_is_gso_v6(skb)) {
1394 ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb)
1396 tcph = skb->encapsulation ? inner_tcp_hdr(skb) : tcp_hdr(skb);
1397 ipv6h->payload_len = 0;
1398 tcph->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
1402 l4len = skb->encapsulation ? inner_tcp_hdrlen(skb) : tcp_hdrlen(skb);
1403 *hdr_len = (skb->encapsulation
1404 ? (skb_inner_transport_header(skb) - skb->data)
1405 : skb_transport_offset(skb)) + l4len;
1407 /* find the field values */
1408 cd_cmd = I40E_TX_CTX_DESC_TSO;
1409 cd_tso_len = skb->len - *hdr_len;
1410 cd_mss = skb_shinfo(skb)->gso_size;
1411 *cd_type_cmd_tso_mss |= ((u64)cd_cmd << I40E_TXD_CTX_QW1_CMD_SHIFT)
1413 << I40E_TXD_CTX_QW1_TSO_LEN_SHIFT)
1414 | ((u64)cd_mss << I40E_TXD_CTX_QW1_MSS_SHIFT);
1419 * i40e_tx_enable_csum - Enable Tx checksum offloads
1421 * @tx_flags: Tx flags currently set
1422 * @td_cmd: Tx descriptor command bits to set
1423 * @td_offset: Tx descriptor header offsets to set
1424 * @cd_tunneling: ptr to context desc bits
1426 static void i40e_tx_enable_csum(struct sk_buff *skb, u32 tx_flags,
1427 u32 *td_cmd, u32 *td_offset,
1428 struct i40e_ring *tx_ring,
1431 struct ipv6hdr *this_ipv6_hdr;
1432 unsigned int this_tcp_hdrlen;
1433 struct iphdr *this_ip_hdr;
1434 u32 network_hdr_len;
1437 if (skb->encapsulation) {
1438 network_hdr_len = skb_inner_network_header_len(skb);
1439 this_ip_hdr = inner_ip_hdr(skb);
1440 this_ipv6_hdr = inner_ipv6_hdr(skb);
1441 this_tcp_hdrlen = inner_tcp_hdrlen(skb);
1443 if (tx_flags & I40E_TX_FLAGS_IPV4) {
1445 if (tx_flags & I40E_TX_FLAGS_TSO) {
1446 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV4;
1447 ip_hdr(skb)->check = 0;
1450 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1452 } else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1453 if (tx_flags & I40E_TX_FLAGS_TSO) {
1454 *cd_tunneling |= I40E_TX_CTX_EXT_IP_IPV6;
1455 ip_hdr(skb)->check = 0;
1458 I40E_TX_CTX_EXT_IP_IPV4_NO_CSUM;
1462 /* Now set the ctx descriptor fields */
1463 *cd_tunneling |= (skb_network_header_len(skb) >> 2) <<
1464 I40E_TXD_CTX_QW0_EXT_IPLEN_SHIFT |
1465 I40E_TXD_CTX_UDP_TUNNELING |
1466 ((skb_inner_network_offset(skb) -
1467 skb_transport_offset(skb)) >> 1) <<
1468 I40E_TXD_CTX_QW0_NATLEN_SHIFT;
1471 network_hdr_len = skb_network_header_len(skb);
1472 this_ip_hdr = ip_hdr(skb);
1473 this_ipv6_hdr = ipv6_hdr(skb);
1474 this_tcp_hdrlen = tcp_hdrlen(skb);
1477 /* Enable IP checksum offloads */
1478 if (tx_flags & I40E_TX_FLAGS_IPV4) {
1479 l4_hdr = this_ip_hdr->protocol;
1480 /* the stack computes the IP header already, the only time we
1481 * need the hardware to recompute it is in the case of TSO.
1483 if (tx_flags & I40E_TX_FLAGS_TSO) {
1484 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4_CSUM;
1485 this_ip_hdr->check = 0;
1487 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV4;
1489 /* Now set the td_offset for IP header length */
1490 *td_offset = (network_hdr_len >> 2) <<
1491 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1492 } else if (tx_flags & I40E_TX_FLAGS_IPV6) {
1493 l4_hdr = this_ipv6_hdr->nexthdr;
1494 *td_cmd |= I40E_TX_DESC_CMD_IIPT_IPV6;
1495 /* Now set the td_offset for IP header length */
1496 *td_offset = (network_hdr_len >> 2) <<
1497 I40E_TX_DESC_LENGTH_IPLEN_SHIFT;
1499 /* words in MACLEN + dwords in IPLEN + dwords in L4Len */
1500 *td_offset |= (skb_network_offset(skb) >> 1) <<
1501 I40E_TX_DESC_LENGTH_MACLEN_SHIFT;
1503 /* Enable L4 checksum offloads */
1506 /* enable checksum offloads */
1507 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_TCP;
1508 *td_offset |= (this_tcp_hdrlen >> 2) <<
1509 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1512 /* enable SCTP checksum offload */
1513 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_SCTP;
1514 *td_offset |= (sizeof(struct sctphdr) >> 2) <<
1515 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1518 /* enable UDP checksum offload */
1519 *td_cmd |= I40E_TX_DESC_CMD_L4T_EOFT_UDP;
1520 *td_offset |= (sizeof(struct udphdr) >> 2) <<
1521 I40E_TX_DESC_LENGTH_L4_FC_LEN_SHIFT;
1529 * i40e_create_tx_ctx Build the Tx context descriptor
1530 * @tx_ring: ring to create the descriptor on
1531 * @cd_type_cmd_tso_mss: Quad Word 1
1532 * @cd_tunneling: Quad Word 0 - bits 0-31
1533 * @cd_l2tag2: Quad Word 0 - bits 32-63
1535 static void i40e_create_tx_ctx(struct i40e_ring *tx_ring,
1536 const u64 cd_type_cmd_tso_mss,
1537 const u32 cd_tunneling, const u32 cd_l2tag2)
1539 struct i40e_tx_context_desc *context_desc;
1540 int i = tx_ring->next_to_use;
1542 if (!cd_type_cmd_tso_mss && !cd_tunneling && !cd_l2tag2)
1545 /* grab the next descriptor */
1546 context_desc = I40E_TX_CTXTDESC(tx_ring, i);
1549 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1551 /* cpu_to_le32 and assign to struct fields */
1552 context_desc->tunneling_params = cpu_to_le32(cd_tunneling);
1553 context_desc->l2tag2 = cpu_to_le16(cd_l2tag2);
1554 context_desc->type_cmd_tso_mss = cpu_to_le64(cd_type_cmd_tso_mss);
1558 * i40e_tx_map - Build the Tx descriptor
1559 * @tx_ring: ring to send buffer on
1561 * @first: first buffer info buffer to use
1562 * @tx_flags: collected send information
1563 * @hdr_len: size of the packet header
1564 * @td_cmd: the command field in the descriptor
1565 * @td_offset: offset for checksum or crc
1567 static void i40e_tx_map(struct i40e_ring *tx_ring, struct sk_buff *skb,
1568 struct i40e_tx_buffer *first, u32 tx_flags,
1569 const u8 hdr_len, u32 td_cmd, u32 td_offset)
1571 unsigned int data_len = skb->data_len;
1572 unsigned int size = skb_headlen(skb);
1573 struct skb_frag_struct *frag;
1574 struct i40e_tx_buffer *tx_bi;
1575 struct i40e_tx_desc *tx_desc;
1576 u16 i = tx_ring->next_to_use;
1581 if (tx_flags & I40E_TX_FLAGS_HW_VLAN) {
1582 td_cmd |= I40E_TX_DESC_CMD_IL2TAG1;
1583 td_tag = (tx_flags & I40E_TX_FLAGS_VLAN_MASK) >>
1584 I40E_TX_FLAGS_VLAN_SHIFT;
1587 if (tx_flags & (I40E_TX_FLAGS_TSO | I40E_TX_FLAGS_FSO))
1588 gso_segs = skb_shinfo(skb)->gso_segs;
1592 /* multiply data chunks by size of headers */
1593 first->bytecount = skb->len - hdr_len + (gso_segs * hdr_len);
1594 first->gso_segs = gso_segs;
1596 first->tx_flags = tx_flags;
1598 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1600 tx_desc = I40E_TX_DESC(tx_ring, i);
1603 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1604 if (dma_mapping_error(tx_ring->dev, dma))
1607 /* record length, and DMA address */
1608 dma_unmap_len_set(tx_bi, len, size);
1609 dma_unmap_addr_set(tx_bi, dma, dma);
1611 tx_desc->buffer_addr = cpu_to_le64(dma);
1613 while (unlikely(size > I40E_MAX_DATA_PER_TXD)) {
1614 tx_desc->cmd_type_offset_bsz =
1615 build_ctob(td_cmd, td_offset,
1616 I40E_MAX_DATA_PER_TXD, td_tag);
1620 if (i == tx_ring->count) {
1621 tx_desc = I40E_TX_DESC(tx_ring, 0);
1625 dma += I40E_MAX_DATA_PER_TXD;
1626 size -= I40E_MAX_DATA_PER_TXD;
1628 tx_desc->buffer_addr = cpu_to_le64(dma);
1631 if (likely(!data_len))
1634 tx_desc->cmd_type_offset_bsz = build_ctob(td_cmd, td_offset,
1639 if (i == tx_ring->count) {
1640 tx_desc = I40E_TX_DESC(tx_ring, 0);
1644 size = skb_frag_size(frag);
1647 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1650 tx_bi = &tx_ring->tx_bi[i];
1653 tx_desc->cmd_type_offset_bsz =
1654 build_ctob(td_cmd, td_offset, size, td_tag) |
1655 cpu_to_le64((u64)I40E_TXD_CMD << I40E_TXD_QW1_CMD_SHIFT);
1657 netdev_tx_sent_queue(netdev_get_tx_queue(tx_ring->netdev,
1658 tx_ring->queue_index),
1661 /* set the timestamp */
1662 first->time_stamp = jiffies;
1664 /* Force memory writes to complete before letting h/w
1665 * know there are new descriptors to fetch. (Only
1666 * applicable for weak-ordered memory model archs,
1671 /* set next_to_watch value indicating a packet is present */
1672 first->next_to_watch = tx_desc;
1675 if (i == tx_ring->count)
1678 tx_ring->next_to_use = i;
1680 /* notify HW of packet */
1681 writel(i, tx_ring->tail);
1686 dev_info(tx_ring->dev, "TX DMA map failed\n");
1688 /* clear dma mappings for failed tx_bi map */
1690 tx_bi = &tx_ring->tx_bi[i];
1691 i40e_unmap_and_free_tx_resource(tx_ring, tx_bi);
1699 tx_ring->next_to_use = i;
1703 * __i40e_maybe_stop_tx - 2nd level check for tx stop conditions
1704 * @tx_ring: the ring to be checked
1705 * @size: the size buffer we want to assure is available
1707 * Returns -EBUSY if a stop is needed, else 0
1709 static inline int __i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1711 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1714 /* Check again in a case another CPU has just made room available. */
1715 if (likely(I40E_DESC_UNUSED(tx_ring) < size))
1718 /* A reprieve! - use start_queue because it doesn't call schedule */
1719 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
1720 ++tx_ring->tx_stats.restart_queue;
1725 * i40e_maybe_stop_tx - 1st level check for tx stop conditions
1726 * @tx_ring: the ring to be checked
1727 * @size: the size buffer we want to assure is available
1729 * Returns 0 if stop is not needed
1731 static int i40e_maybe_stop_tx(struct i40e_ring *tx_ring, int size)
1733 if (likely(I40E_DESC_UNUSED(tx_ring) >= size))
1735 return __i40e_maybe_stop_tx(tx_ring, size);
1739 * i40e_xmit_descriptor_count - calculate number of tx descriptors needed
1741 * @tx_ring: ring to send buffer on
1743 * Returns number of data descriptors needed for this skb. Returns 0 to indicate
1744 * there is not enough descriptors available in this ring since we need at least
1747 static int i40e_xmit_descriptor_count(struct sk_buff *skb,
1748 struct i40e_ring *tx_ring)
1750 #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD
1755 /* need: 1 descriptor per page * PAGE_SIZE/I40E_MAX_DATA_PER_TXD,
1756 * + 1 desc for skb_head_len/I40E_MAX_DATA_PER_TXD,
1757 * + 2 desc gap to keep tail from touching head,
1758 * + 1 desc for context descriptor,
1759 * otherwise try next time
1761 #if PAGE_SIZE > I40E_MAX_DATA_PER_TXD
1762 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1763 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
1765 count += skb_shinfo(skb)->nr_frags;
1767 count += TXD_USE_COUNT(skb_headlen(skb));
1768 if (i40e_maybe_stop_tx(tx_ring, count + 3)) {
1769 tx_ring->tx_stats.tx_busy++;
1776 * i40e_xmit_frame_ring - Sends buffer on Tx ring
1778 * @tx_ring: ring to send buffer on
1780 * Returns NETDEV_TX_OK if sent, else an error code
1782 static netdev_tx_t i40e_xmit_frame_ring(struct sk_buff *skb,
1783 struct i40e_ring *tx_ring)
1785 u64 cd_type_cmd_tso_mss = I40E_TX_DESC_DTYPE_CONTEXT;
1786 u32 cd_tunneling = 0, cd_l2tag2 = 0;
1787 struct i40e_tx_buffer *first;
1794 if (0 == i40e_xmit_descriptor_count(skb, tx_ring))
1795 return NETDEV_TX_BUSY;
1797 /* prepare the xmit flags */
1798 if (i40e_tx_prepare_vlan_flags(skb, tx_ring, &tx_flags))
1801 /* obtain protocol of skb */
1802 protocol = skb->protocol;
1804 /* record the location of the first descriptor for this packet */
1805 first = &tx_ring->tx_bi[tx_ring->next_to_use];
1807 /* setup IPv4/IPv6 offloads */
1808 if (protocol == htons(ETH_P_IP))
1809 tx_flags |= I40E_TX_FLAGS_IPV4;
1810 else if (protocol == htons(ETH_P_IPV6))
1811 tx_flags |= I40E_TX_FLAGS_IPV6;
1813 tso = i40e_tso(tx_ring, skb, tx_flags, protocol, &hdr_len,
1814 &cd_type_cmd_tso_mss, &cd_tunneling);
1819 tx_flags |= I40E_TX_FLAGS_TSO;
1821 skb_tx_timestamp(skb);
1823 /* always enable CRC insertion offload */
1824 td_cmd |= I40E_TX_DESC_CMD_ICRC;
1826 /* Always offload the checksum, since it's in the data descriptor */
1827 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1828 tx_flags |= I40E_TX_FLAGS_CSUM;
1830 i40e_tx_enable_csum(skb, tx_flags, &td_cmd, &td_offset,
1831 tx_ring, &cd_tunneling);
1834 i40e_create_tx_ctx(tx_ring, cd_type_cmd_tso_mss,
1835 cd_tunneling, cd_l2tag2);
1837 /* Add Flow Director ATR if it's enabled.
1839 * NOTE: this must always be directly before the data descriptor.
1841 i40e_atr(tx_ring, skb, tx_flags, protocol);
1843 i40e_tx_map(tx_ring, skb, first, tx_flags, hdr_len,
1846 i40e_maybe_stop_tx(tx_ring, DESC_NEEDED);
1848 return NETDEV_TX_OK;
1851 dev_kfree_skb_any(skb);
1852 return NETDEV_TX_OK;
1856 * i40e_lan_xmit_frame - Selects the correct VSI and Tx queue to send buffer
1858 * @netdev: network interface device structure
1860 * Returns NETDEV_TX_OK if sent, else an error code
1862 netdev_tx_t i40e_lan_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1864 struct i40e_netdev_priv *np = netdev_priv(netdev);
1865 struct i40e_vsi *vsi = np->vsi;
1866 struct i40e_ring *tx_ring = vsi->tx_rings[skb->queue_mapping];
1868 /* hardware can't handle really short frames, hardware padding works
1871 if (unlikely(skb->len < I40E_MIN_TX_LEN)) {
1872 if (skb_pad(skb, I40E_MIN_TX_LEN - skb->len))
1873 return NETDEV_TX_OK;
1874 skb->len = I40E_MIN_TX_LEN;
1875 skb_set_tail_pointer(skb, I40E_MIN_TX_LEN);
1878 return i40e_xmit_frame_ring(skb, tx_ring);