Loading...
1/*******************************************************************************
2
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2018 Intel Corporation.
5
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.
9
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
13 more details.
14
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
17
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
20
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
24
25*******************************************************************************/
26
27/******************************************************************************
28 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
29******************************************************************************/
30
31#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
32
33#include <linux/types.h>
34#include <linux/bitops.h>
35#include <linux/module.h>
36#include <linux/pci.h>
37#include <linux/netdevice.h>
38#include <linux/vmalloc.h>
39#include <linux/string.h>
40#include <linux/in.h>
41#include <linux/ip.h>
42#include <linux/tcp.h>
43#include <linux/sctp.h>
44#include <linux/ipv6.h>
45#include <linux/slab.h>
46#include <net/checksum.h>
47#include <net/ip6_checksum.h>
48#include <linux/ethtool.h>
49#include <linux/if.h>
50#include <linux/if_vlan.h>
51#include <linux/prefetch.h>
52#include <net/mpls.h>
53#include <linux/bpf.h>
54#include <linux/bpf_trace.h>
55#include <linux/atomic.h>
56
57#include "ixgbevf.h"
58
59const char ixgbevf_driver_name[] = "ixgbevf";
60static const char ixgbevf_driver_string[] =
61 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
62
63#define DRV_VERSION "4.1.0-k"
64const char ixgbevf_driver_version[] = DRV_VERSION;
65static char ixgbevf_copyright[] =
66 "Copyright (c) 2009 - 2015 Intel Corporation.";
67
68static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
69 [board_82599_vf] = &ixgbevf_82599_vf_info,
70 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info,
71 [board_X540_vf] = &ixgbevf_X540_vf_info,
72 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info,
73 [board_X550_vf] = &ixgbevf_X550_vf_info,
74 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info,
75 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
76 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info,
77 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info,
78};
79
80/* ixgbevf_pci_tbl - PCI Device ID Table
81 *
82 * Wildcard entries (PCI_ANY_ID) should come last
83 * Last entry must be all 0s
84 *
85 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
86 * Class, Class Mask, private data (not used) }
87 */
88static const struct pci_device_id ixgbevf_pci_tbl[] = {
89 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
90 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
91 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
92 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
93 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
94 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
95 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
96 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
97 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
98 /* required last entry */
99 {0, }
100};
101MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
102
103MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
104MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
105MODULE_LICENSE("GPL");
106MODULE_VERSION(DRV_VERSION);
107
108#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
109static int debug = -1;
110module_param(debug, int, 0);
111MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
112
113static struct workqueue_struct *ixgbevf_wq;
114
115static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
116{
117 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
118 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
119 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
120 queue_work(ixgbevf_wq, &adapter->service_task);
121}
122
123static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
124{
125 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
126
127 /* flush memory to make sure state is correct before next watchdog */
128 smp_mb__before_atomic();
129 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
130}
131
132/* forward decls */
133static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
134static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
135static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
136static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
137static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
138 struct ixgbevf_rx_buffer *old_buff);
139
140static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
141{
142 struct ixgbevf_adapter *adapter = hw->back;
143
144 if (!hw->hw_addr)
145 return;
146 hw->hw_addr = NULL;
147 dev_err(&adapter->pdev->dev, "Adapter removed\n");
148 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
149 ixgbevf_service_event_schedule(adapter);
150}
151
152static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
153{
154 u32 value;
155
156 /* The following check not only optimizes a bit by not
157 * performing a read on the status register when the
158 * register just read was a status register read that
159 * returned IXGBE_FAILED_READ_REG. It also blocks any
160 * potential recursion.
161 */
162 if (reg == IXGBE_VFSTATUS) {
163 ixgbevf_remove_adapter(hw);
164 return;
165 }
166 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
167 if (value == IXGBE_FAILED_READ_REG)
168 ixgbevf_remove_adapter(hw);
169}
170
171u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
172{
173 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
174 u32 value;
175
176 if (IXGBE_REMOVED(reg_addr))
177 return IXGBE_FAILED_READ_REG;
178 value = readl(reg_addr + reg);
179 if (unlikely(value == IXGBE_FAILED_READ_REG))
180 ixgbevf_check_remove(hw, reg);
181 return value;
182}
183
184/**
185 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
186 * @adapter: pointer to adapter struct
187 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
188 * @queue: queue to map the corresponding interrupt to
189 * @msix_vector: the vector to map to the corresponding queue
190 **/
191static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
192 u8 queue, u8 msix_vector)
193{
194 u32 ivar, index;
195 struct ixgbe_hw *hw = &adapter->hw;
196
197 if (direction == -1) {
198 /* other causes */
199 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
200 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
201 ivar &= ~0xFF;
202 ivar |= msix_vector;
203 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
204 } else {
205 /* Tx or Rx causes */
206 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
207 index = ((16 * (queue & 1)) + (8 * direction));
208 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
209 ivar &= ~(0xFF << index);
210 ivar |= (msix_vector << index);
211 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
212 }
213}
214
215static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
216{
217 return ring->stats.packets;
218}
219
220static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
221{
222 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
223 struct ixgbe_hw *hw = &adapter->hw;
224
225 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
226 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
227
228 if (head != tail)
229 return (head < tail) ?
230 tail - head : (tail + ring->count - head);
231
232 return 0;
233}
234
235static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
236{
237 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
238 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
239 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
240
241 clear_check_for_tx_hang(tx_ring);
242
243 /* Check for a hung queue, but be thorough. This verifies
244 * that a transmit has been completed since the previous
245 * check AND there is at least one packet pending. The
246 * ARMED bit is set to indicate a potential hang.
247 */
248 if ((tx_done_old == tx_done) && tx_pending) {
249 /* make sure it is true for two checks in a row */
250 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
251 &tx_ring->state);
252 }
253 /* reset the countdown */
254 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
255
256 /* update completed stats and continue */
257 tx_ring->tx_stats.tx_done_old = tx_done;
258
259 return false;
260}
261
262static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
263{
264 /* Do the reset outside of interrupt context */
265 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
266 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
267 ixgbevf_service_event_schedule(adapter);
268 }
269}
270
271/**
272 * ixgbevf_tx_timeout - Respond to a Tx Hang
273 * @netdev: network interface device structure
274 **/
275static void ixgbevf_tx_timeout(struct net_device *netdev)
276{
277 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
278
279 ixgbevf_tx_timeout_reset(adapter);
280}
281
282/**
283 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
284 * @q_vector: board private structure
285 * @tx_ring: tx ring to clean
286 * @napi_budget: Used to determine if we are in netpoll
287 **/
288static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
289 struct ixgbevf_ring *tx_ring, int napi_budget)
290{
291 struct ixgbevf_adapter *adapter = q_vector->adapter;
292 struct ixgbevf_tx_buffer *tx_buffer;
293 union ixgbe_adv_tx_desc *tx_desc;
294 unsigned int total_bytes = 0, total_packets = 0;
295 unsigned int budget = tx_ring->count / 2;
296 unsigned int i = tx_ring->next_to_clean;
297
298 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
299 return true;
300
301 tx_buffer = &tx_ring->tx_buffer_info[i];
302 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
303 i -= tx_ring->count;
304
305 do {
306 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
307
308 /* if next_to_watch is not set then there is no work pending */
309 if (!eop_desc)
310 break;
311
312 /* prevent any other reads prior to eop_desc */
313 smp_rmb();
314
315 /* if DD is not set pending work has not been completed */
316 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
317 break;
318
319 /* clear next_to_watch to prevent false hangs */
320 tx_buffer->next_to_watch = NULL;
321
322 /* update the statistics for this packet */
323 total_bytes += tx_buffer->bytecount;
324 total_packets += tx_buffer->gso_segs;
325
326 /* free the skb */
327 if (ring_is_xdp(tx_ring))
328 page_frag_free(tx_buffer->data);
329 else
330 napi_consume_skb(tx_buffer->skb, napi_budget);
331
332 /* unmap skb header data */
333 dma_unmap_single(tx_ring->dev,
334 dma_unmap_addr(tx_buffer, dma),
335 dma_unmap_len(tx_buffer, len),
336 DMA_TO_DEVICE);
337
338 /* clear tx_buffer data */
339 dma_unmap_len_set(tx_buffer, len, 0);
340
341 /* unmap remaining buffers */
342 while (tx_desc != eop_desc) {
343 tx_buffer++;
344 tx_desc++;
345 i++;
346 if (unlikely(!i)) {
347 i -= tx_ring->count;
348 tx_buffer = tx_ring->tx_buffer_info;
349 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
350 }
351
352 /* unmap any remaining paged data */
353 if (dma_unmap_len(tx_buffer, len)) {
354 dma_unmap_page(tx_ring->dev,
355 dma_unmap_addr(tx_buffer, dma),
356 dma_unmap_len(tx_buffer, len),
357 DMA_TO_DEVICE);
358 dma_unmap_len_set(tx_buffer, len, 0);
359 }
360 }
361
362 /* move us one more past the eop_desc for start of next pkt */
363 tx_buffer++;
364 tx_desc++;
365 i++;
366 if (unlikely(!i)) {
367 i -= tx_ring->count;
368 tx_buffer = tx_ring->tx_buffer_info;
369 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
370 }
371
372 /* issue prefetch for next Tx descriptor */
373 prefetch(tx_desc);
374
375 /* update budget accounting */
376 budget--;
377 } while (likely(budget));
378
379 i += tx_ring->count;
380 tx_ring->next_to_clean = i;
381 u64_stats_update_begin(&tx_ring->syncp);
382 tx_ring->stats.bytes += total_bytes;
383 tx_ring->stats.packets += total_packets;
384 u64_stats_update_end(&tx_ring->syncp);
385 q_vector->tx.total_bytes += total_bytes;
386 q_vector->tx.total_packets += total_packets;
387
388 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
389 struct ixgbe_hw *hw = &adapter->hw;
390 union ixgbe_adv_tx_desc *eop_desc;
391
392 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
393
394 pr_err("Detected Tx Unit Hang%s\n"
395 " Tx Queue <%d>\n"
396 " TDH, TDT <%x>, <%x>\n"
397 " next_to_use <%x>\n"
398 " next_to_clean <%x>\n"
399 "tx_buffer_info[next_to_clean]\n"
400 " next_to_watch <%p>\n"
401 " eop_desc->wb.status <%x>\n"
402 " time_stamp <%lx>\n"
403 " jiffies <%lx>\n",
404 ring_is_xdp(tx_ring) ? " XDP" : "",
405 tx_ring->queue_index,
406 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
407 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
408 tx_ring->next_to_use, i,
409 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
410 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
411
412 if (!ring_is_xdp(tx_ring))
413 netif_stop_subqueue(tx_ring->netdev,
414 tx_ring->queue_index);
415
416 /* schedule immediate reset if we believe we hung */
417 ixgbevf_tx_timeout_reset(adapter);
418
419 return true;
420 }
421
422 if (ring_is_xdp(tx_ring))
423 return !!budget;
424
425#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
426 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
427 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
428 /* Make sure that anybody stopping the queue after this
429 * sees the new next_to_clean.
430 */
431 smp_mb();
432
433 if (__netif_subqueue_stopped(tx_ring->netdev,
434 tx_ring->queue_index) &&
435 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
436 netif_wake_subqueue(tx_ring->netdev,
437 tx_ring->queue_index);
438 ++tx_ring->tx_stats.restart_queue;
439 }
440 }
441
442 return !!budget;
443}
444
445/**
446 * ixgbevf_rx_skb - Helper function to determine proper Rx method
447 * @q_vector: structure containing interrupt and ring information
448 * @skb: packet to send up
449 **/
450static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
451 struct sk_buff *skb)
452{
453 napi_gro_receive(&q_vector->napi, skb);
454}
455
456#define IXGBE_RSS_L4_TYPES_MASK \
457 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
458 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
459 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
460 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
461
462static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
463 union ixgbe_adv_rx_desc *rx_desc,
464 struct sk_buff *skb)
465{
466 u16 rss_type;
467
468 if (!(ring->netdev->features & NETIF_F_RXHASH))
469 return;
470
471 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
472 IXGBE_RXDADV_RSSTYPE_MASK;
473
474 if (!rss_type)
475 return;
476
477 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
478 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
479 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
480}
481
482/**
483 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
484 * @ring: structure containig ring specific data
485 * @rx_desc: current Rx descriptor being processed
486 * @skb: skb currently being received and modified
487 **/
488static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
489 union ixgbe_adv_rx_desc *rx_desc,
490 struct sk_buff *skb)
491{
492 skb_checksum_none_assert(skb);
493
494 /* Rx csum disabled */
495 if (!(ring->netdev->features & NETIF_F_RXCSUM))
496 return;
497
498 /* if IP and error */
499 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
500 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
501 ring->rx_stats.csum_err++;
502 return;
503 }
504
505 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
506 return;
507
508 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
509 ring->rx_stats.csum_err++;
510 return;
511 }
512
513 /* It must be a TCP or UDP packet with a valid checksum */
514 skb->ip_summed = CHECKSUM_UNNECESSARY;
515}
516
517/**
518 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
519 * @rx_ring: rx descriptor ring packet is being transacted on
520 * @rx_desc: pointer to the EOP Rx descriptor
521 * @skb: pointer to current skb being populated
522 *
523 * This function checks the ring, descriptor, and packet information in
524 * order to populate the checksum, VLAN, protocol, and other fields within
525 * the skb.
526 **/
527static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
528 union ixgbe_adv_rx_desc *rx_desc,
529 struct sk_buff *skb)
530{
531 ixgbevf_rx_hash(rx_ring, rx_desc, skb);
532 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
533
534 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
535 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
536 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
537
538 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
539 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
540 }
541
542 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
543}
544
545static
546struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
547 const unsigned int size)
548{
549 struct ixgbevf_rx_buffer *rx_buffer;
550
551 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
552 prefetchw(rx_buffer->page);
553
554 /* we are reusing so sync this buffer for CPU use */
555 dma_sync_single_range_for_cpu(rx_ring->dev,
556 rx_buffer->dma,
557 rx_buffer->page_offset,
558 size,
559 DMA_FROM_DEVICE);
560
561 rx_buffer->pagecnt_bias--;
562
563 return rx_buffer;
564}
565
566static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
567 struct ixgbevf_rx_buffer *rx_buffer,
568 struct sk_buff *skb)
569{
570 if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
571 /* hand second half of page back to the ring */
572 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
573 } else {
574 if (IS_ERR(skb))
575 /* We are not reusing the buffer so unmap it and free
576 * any references we are holding to it
577 */
578 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
579 ixgbevf_rx_pg_size(rx_ring),
580 DMA_FROM_DEVICE,
581 IXGBEVF_RX_DMA_ATTR);
582 __page_frag_cache_drain(rx_buffer->page,
583 rx_buffer->pagecnt_bias);
584 }
585
586 /* clear contents of rx_buffer */
587 rx_buffer->page = NULL;
588}
589
590/**
591 * ixgbevf_is_non_eop - process handling of non-EOP buffers
592 * @rx_ring: Rx ring being processed
593 * @rx_desc: Rx descriptor for current buffer
594 *
595 * This function updates next to clean. If the buffer is an EOP buffer
596 * this function exits returning false, otherwise it will place the
597 * sk_buff in the next buffer to be chained and return true indicating
598 * that this is in fact a non-EOP buffer.
599 **/
600static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
601 union ixgbe_adv_rx_desc *rx_desc)
602{
603 u32 ntc = rx_ring->next_to_clean + 1;
604
605 /* fetch, update, and store next to clean */
606 ntc = (ntc < rx_ring->count) ? ntc : 0;
607 rx_ring->next_to_clean = ntc;
608
609 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
610
611 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
612 return false;
613
614 return true;
615}
616
617static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
618{
619 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
620}
621
622static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
623 struct ixgbevf_rx_buffer *bi)
624{
625 struct page *page = bi->page;
626 dma_addr_t dma;
627
628 /* since we are recycling buffers we should seldom need to alloc */
629 if (likely(page))
630 return true;
631
632 /* alloc new page for storage */
633 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
634 if (unlikely(!page)) {
635 rx_ring->rx_stats.alloc_rx_page_failed++;
636 return false;
637 }
638
639 /* map page for use */
640 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
641 ixgbevf_rx_pg_size(rx_ring),
642 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
643
644 /* if mapping failed free memory back to system since
645 * there isn't much point in holding memory we can't use
646 */
647 if (dma_mapping_error(rx_ring->dev, dma)) {
648 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
649
650 rx_ring->rx_stats.alloc_rx_page_failed++;
651 return false;
652 }
653
654 bi->dma = dma;
655 bi->page = page;
656 bi->page_offset = ixgbevf_rx_offset(rx_ring);
657 bi->pagecnt_bias = 1;
658 rx_ring->rx_stats.alloc_rx_page++;
659
660 return true;
661}
662
663/**
664 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
665 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
666 * @cleaned_count: number of buffers to replace
667 **/
668static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
669 u16 cleaned_count)
670{
671 union ixgbe_adv_rx_desc *rx_desc;
672 struct ixgbevf_rx_buffer *bi;
673 unsigned int i = rx_ring->next_to_use;
674
675 /* nothing to do or no valid netdev defined */
676 if (!cleaned_count || !rx_ring->netdev)
677 return;
678
679 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
680 bi = &rx_ring->rx_buffer_info[i];
681 i -= rx_ring->count;
682
683 do {
684 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
685 break;
686
687 /* sync the buffer for use by the device */
688 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
689 bi->page_offset,
690 ixgbevf_rx_bufsz(rx_ring),
691 DMA_FROM_DEVICE);
692
693 /* Refresh the desc even if pkt_addr didn't change
694 * because each write-back erases this info.
695 */
696 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
697
698 rx_desc++;
699 bi++;
700 i++;
701 if (unlikely(!i)) {
702 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
703 bi = rx_ring->rx_buffer_info;
704 i -= rx_ring->count;
705 }
706
707 /* clear the length for the next_to_use descriptor */
708 rx_desc->wb.upper.length = 0;
709
710 cleaned_count--;
711 } while (cleaned_count);
712
713 i += rx_ring->count;
714
715 if (rx_ring->next_to_use != i) {
716 /* record the next descriptor to use */
717 rx_ring->next_to_use = i;
718
719 /* update next to alloc since we have filled the ring */
720 rx_ring->next_to_alloc = i;
721
722 /* Force memory writes to complete before letting h/w
723 * know there are new descriptors to fetch. (Only
724 * applicable for weak-ordered memory model archs,
725 * such as IA-64).
726 */
727 wmb();
728 ixgbevf_write_tail(rx_ring, i);
729 }
730}
731
732/**
733 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
734 * @rx_ring: rx descriptor ring packet is being transacted on
735 * @rx_desc: pointer to the EOP Rx descriptor
736 * @skb: pointer to current skb being fixed
737 *
738 * Check for corrupted packet headers caused by senders on the local L2
739 * embedded NIC switch not setting up their Tx Descriptors right. These
740 * should be very rare.
741 *
742 * Also address the case where we are pulling data in on pages only
743 * and as such no data is present in the skb header.
744 *
745 * In addition if skb is not at least 60 bytes we need to pad it so that
746 * it is large enough to qualify as a valid Ethernet frame.
747 *
748 * Returns true if an error was encountered and skb was freed.
749 **/
750static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
751 union ixgbe_adv_rx_desc *rx_desc,
752 struct sk_buff *skb)
753{
754 /* XDP packets use error pointer so abort at this point */
755 if (IS_ERR(skb))
756 return true;
757
758 /* verify that the packet does not have any known errors */
759 if (unlikely(ixgbevf_test_staterr(rx_desc,
760 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
761 struct net_device *netdev = rx_ring->netdev;
762
763 if (!(netdev->features & NETIF_F_RXALL)) {
764 dev_kfree_skb_any(skb);
765 return true;
766 }
767 }
768
769 /* if eth_skb_pad returns an error the skb was freed */
770 if (eth_skb_pad(skb))
771 return true;
772
773 return false;
774}
775
776/**
777 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
778 * @rx_ring: rx descriptor ring to store buffers on
779 * @old_buff: donor buffer to have page reused
780 *
781 * Synchronizes page for reuse by the adapter
782 **/
783static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
784 struct ixgbevf_rx_buffer *old_buff)
785{
786 struct ixgbevf_rx_buffer *new_buff;
787 u16 nta = rx_ring->next_to_alloc;
788
789 new_buff = &rx_ring->rx_buffer_info[nta];
790
791 /* update, and store next to alloc */
792 nta++;
793 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
794
795 /* transfer page from old buffer to new buffer */
796 new_buff->page = old_buff->page;
797 new_buff->dma = old_buff->dma;
798 new_buff->page_offset = old_buff->page_offset;
799 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
800}
801
802static inline bool ixgbevf_page_is_reserved(struct page *page)
803{
804 return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
805}
806
807static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
808{
809 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
810 struct page *page = rx_buffer->page;
811
812 /* avoid re-using remote pages */
813 if (unlikely(ixgbevf_page_is_reserved(page)))
814 return false;
815
816#if (PAGE_SIZE < 8192)
817 /* if we are only owner of page we can reuse it */
818 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
819 return false;
820#else
821#define IXGBEVF_LAST_OFFSET \
822 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
823
824 if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
825 return false;
826
827#endif
828
829 /* If we have drained the page fragment pool we need to update
830 * the pagecnt_bias and page count so that we fully restock the
831 * number of references the driver holds.
832 */
833 if (unlikely(!pagecnt_bias)) {
834 page_ref_add(page, USHRT_MAX);
835 rx_buffer->pagecnt_bias = USHRT_MAX;
836 }
837
838 return true;
839}
840
841/**
842 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
843 * @rx_ring: rx descriptor ring to transact packets on
844 * @rx_buffer: buffer containing page to add
845 * @skb: sk_buff to place the data into
846 * @size: size of buffer to be added
847 *
848 * This function will add the data contained in rx_buffer->page to the skb.
849 **/
850static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
851 struct ixgbevf_rx_buffer *rx_buffer,
852 struct sk_buff *skb,
853 unsigned int size)
854{
855#if (PAGE_SIZE < 8192)
856 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
857#else
858 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
859 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
860 SKB_DATA_ALIGN(size);
861#endif
862 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
863 rx_buffer->page_offset, size, truesize);
864#if (PAGE_SIZE < 8192)
865 rx_buffer->page_offset ^= truesize;
866#else
867 rx_buffer->page_offset += truesize;
868#endif
869}
870
871static
872struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
873 struct ixgbevf_rx_buffer *rx_buffer,
874 struct xdp_buff *xdp,
875 union ixgbe_adv_rx_desc *rx_desc)
876{
877 unsigned int size = xdp->data_end - xdp->data;
878#if (PAGE_SIZE < 8192)
879 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
880#else
881 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
882 xdp->data_hard_start);
883#endif
884 unsigned int headlen;
885 struct sk_buff *skb;
886
887 /* prefetch first cache line of first page */
888 prefetch(xdp->data);
889#if L1_CACHE_BYTES < 128
890 prefetch(xdp->data + L1_CACHE_BYTES);
891#endif
892 /* Note, we get here by enabling legacy-rx via:
893 *
894 * ethtool --set-priv-flags <dev> legacy-rx on
895 *
896 * In this mode, we currently get 0 extra XDP headroom as
897 * opposed to having legacy-rx off, where we process XDP
898 * packets going to stack via ixgbevf_build_skb().
899 *
900 * For ixgbevf_construct_skb() mode it means that the
901 * xdp->data_meta will always point to xdp->data, since
902 * the helper cannot expand the head. Should this ever
903 * changed in future for legacy-rx mode on, then lets also
904 * add xdp->data_meta handling here.
905 */
906
907 /* allocate a skb to store the frags */
908 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
909 if (unlikely(!skb))
910 return NULL;
911
912 /* Determine available headroom for copy */
913 headlen = size;
914 if (headlen > IXGBEVF_RX_HDR_SIZE)
915 headlen = eth_get_headlen(xdp->data, IXGBEVF_RX_HDR_SIZE);
916
917 /* align pull length to size of long to optimize memcpy performance */
918 memcpy(__skb_put(skb, headlen), xdp->data,
919 ALIGN(headlen, sizeof(long)));
920
921 /* update all of the pointers */
922 size -= headlen;
923 if (size) {
924 skb_add_rx_frag(skb, 0, rx_buffer->page,
925 (xdp->data + headlen) -
926 page_address(rx_buffer->page),
927 size, truesize);
928#if (PAGE_SIZE < 8192)
929 rx_buffer->page_offset ^= truesize;
930#else
931 rx_buffer->page_offset += truesize;
932#endif
933 } else {
934 rx_buffer->pagecnt_bias++;
935 }
936
937 return skb;
938}
939
940static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
941 u32 qmask)
942{
943 struct ixgbe_hw *hw = &adapter->hw;
944
945 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
946}
947
948static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
949 struct ixgbevf_rx_buffer *rx_buffer,
950 struct xdp_buff *xdp,
951 union ixgbe_adv_rx_desc *rx_desc)
952{
953 unsigned int metasize = xdp->data - xdp->data_meta;
954#if (PAGE_SIZE < 8192)
955 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
956#else
957 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
958 SKB_DATA_ALIGN(xdp->data_end -
959 xdp->data_hard_start);
960#endif
961 struct sk_buff *skb;
962
963 /* Prefetch first cache line of first page. If xdp->data_meta
964 * is unused, this points to xdp->data, otherwise, we likely
965 * have a consumer accessing first few bytes of meta data,
966 * and then actual data.
967 */
968 prefetch(xdp->data_meta);
969#if L1_CACHE_BYTES < 128
970 prefetch(xdp->data_meta + L1_CACHE_BYTES);
971#endif
972
973 /* build an skb around the page buffer */
974 skb = build_skb(xdp->data_hard_start, truesize);
975 if (unlikely(!skb))
976 return NULL;
977
978 /* update pointers within the skb to store the data */
979 skb_reserve(skb, xdp->data - xdp->data_hard_start);
980 __skb_put(skb, xdp->data_end - xdp->data);
981 if (metasize)
982 skb_metadata_set(skb, metasize);
983
984 /* update buffer offset */
985#if (PAGE_SIZE < 8192)
986 rx_buffer->page_offset ^= truesize;
987#else
988 rx_buffer->page_offset += truesize;
989#endif
990
991 return skb;
992}
993
994#define IXGBEVF_XDP_PASS 0
995#define IXGBEVF_XDP_CONSUMED 1
996#define IXGBEVF_XDP_TX 2
997
998static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
999 struct xdp_buff *xdp)
1000{
1001 struct ixgbevf_tx_buffer *tx_buffer;
1002 union ixgbe_adv_tx_desc *tx_desc;
1003 u32 len, cmd_type;
1004 dma_addr_t dma;
1005 u16 i;
1006
1007 len = xdp->data_end - xdp->data;
1008
1009 if (unlikely(!ixgbevf_desc_unused(ring)))
1010 return IXGBEVF_XDP_CONSUMED;
1011
1012 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
1013 if (dma_mapping_error(ring->dev, dma))
1014 return IXGBEVF_XDP_CONSUMED;
1015
1016 /* record the location of the first descriptor for this packet */
1017 tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
1018 tx_buffer->bytecount = len;
1019 tx_buffer->gso_segs = 1;
1020 tx_buffer->protocol = 0;
1021
1022 i = ring->next_to_use;
1023 tx_desc = IXGBEVF_TX_DESC(ring, i);
1024
1025 dma_unmap_len_set(tx_buffer, len, len);
1026 dma_unmap_addr_set(tx_buffer, dma, dma);
1027 tx_buffer->data = xdp->data;
1028 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1029
1030 /* put descriptor type bits */
1031 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1032 IXGBE_ADVTXD_DCMD_DEXT |
1033 IXGBE_ADVTXD_DCMD_IFCS;
1034 cmd_type |= len | IXGBE_TXD_CMD;
1035 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1036 tx_desc->read.olinfo_status =
1037 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1038 IXGBE_ADVTXD_CC);
1039
1040 /* Avoid any potential race with cleanup */
1041 smp_wmb();
1042
1043 /* set next_to_watch value indicating a packet is present */
1044 i++;
1045 if (i == ring->count)
1046 i = 0;
1047
1048 tx_buffer->next_to_watch = tx_desc;
1049 ring->next_to_use = i;
1050
1051 return IXGBEVF_XDP_TX;
1052}
1053
1054static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1055 struct ixgbevf_ring *rx_ring,
1056 struct xdp_buff *xdp)
1057{
1058 int result = IXGBEVF_XDP_PASS;
1059 struct ixgbevf_ring *xdp_ring;
1060 struct bpf_prog *xdp_prog;
1061 u32 act;
1062
1063 rcu_read_lock();
1064 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1065
1066 if (!xdp_prog)
1067 goto xdp_out;
1068
1069 act = bpf_prog_run_xdp(xdp_prog, xdp);
1070 switch (act) {
1071 case XDP_PASS:
1072 break;
1073 case XDP_TX:
1074 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1075 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1076 break;
1077 default:
1078 bpf_warn_invalid_xdp_action(act);
1079 /* fallthrough */
1080 case XDP_ABORTED:
1081 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1082 /* fallthrough -- handle aborts by dropping packet */
1083 case XDP_DROP:
1084 result = IXGBEVF_XDP_CONSUMED;
1085 break;
1086 }
1087xdp_out:
1088 rcu_read_unlock();
1089 return ERR_PTR(-result);
1090}
1091
1092static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1093 struct ixgbevf_rx_buffer *rx_buffer,
1094 unsigned int size)
1095{
1096#if (PAGE_SIZE < 8192)
1097 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
1098
1099 rx_buffer->page_offset ^= truesize;
1100#else
1101 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
1102 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
1103 SKB_DATA_ALIGN(size);
1104
1105 rx_buffer->page_offset += truesize;
1106#endif
1107}
1108
1109static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1110 struct ixgbevf_ring *rx_ring,
1111 int budget)
1112{
1113 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1114 struct ixgbevf_adapter *adapter = q_vector->adapter;
1115 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1116 struct sk_buff *skb = rx_ring->skb;
1117 bool xdp_xmit = false;
1118 struct xdp_buff xdp;
1119
1120 xdp.rxq = &rx_ring->xdp_rxq;
1121
1122 while (likely(total_rx_packets < budget)) {
1123 struct ixgbevf_rx_buffer *rx_buffer;
1124 union ixgbe_adv_rx_desc *rx_desc;
1125 unsigned int size;
1126
1127 /* return some buffers to hardware, one at a time is too slow */
1128 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1129 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1130 cleaned_count = 0;
1131 }
1132
1133 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1134 size = le16_to_cpu(rx_desc->wb.upper.length);
1135 if (!size)
1136 break;
1137
1138 /* This memory barrier is needed to keep us from reading
1139 * any other fields out of the rx_desc until we know the
1140 * RXD_STAT_DD bit is set
1141 */
1142 rmb();
1143
1144 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1145
1146 /* retrieve a buffer from the ring */
1147 if (!skb) {
1148 xdp.data = page_address(rx_buffer->page) +
1149 rx_buffer->page_offset;
1150 xdp.data_meta = xdp.data;
1151 xdp.data_hard_start = xdp.data -
1152 ixgbevf_rx_offset(rx_ring);
1153 xdp.data_end = xdp.data + size;
1154
1155 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1156 }
1157
1158 if (IS_ERR(skb)) {
1159 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1160 xdp_xmit = true;
1161 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1162 size);
1163 } else {
1164 rx_buffer->pagecnt_bias++;
1165 }
1166 total_rx_packets++;
1167 total_rx_bytes += size;
1168 } else if (skb) {
1169 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1170 } else if (ring_uses_build_skb(rx_ring)) {
1171 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1172 &xdp, rx_desc);
1173 } else {
1174 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1175 &xdp, rx_desc);
1176 }
1177
1178 /* exit if we failed to retrieve a buffer */
1179 if (!skb) {
1180 rx_ring->rx_stats.alloc_rx_buff_failed++;
1181 rx_buffer->pagecnt_bias++;
1182 break;
1183 }
1184
1185 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1186 cleaned_count++;
1187
1188 /* fetch next buffer in frame if non-eop */
1189 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1190 continue;
1191
1192 /* verify the packet layout is correct */
1193 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1194 skb = NULL;
1195 continue;
1196 }
1197
1198 /* probably a little skewed due to removing CRC */
1199 total_rx_bytes += skb->len;
1200
1201 /* Workaround hardware that can't do proper VEPA multicast
1202 * source pruning.
1203 */
1204 if ((skb->pkt_type == PACKET_BROADCAST ||
1205 skb->pkt_type == PACKET_MULTICAST) &&
1206 ether_addr_equal(rx_ring->netdev->dev_addr,
1207 eth_hdr(skb)->h_source)) {
1208 dev_kfree_skb_irq(skb);
1209 continue;
1210 }
1211
1212 /* populate checksum, VLAN, and protocol */
1213 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1214
1215 ixgbevf_rx_skb(q_vector, skb);
1216
1217 /* reset skb pointer */
1218 skb = NULL;
1219
1220 /* update budget accounting */
1221 total_rx_packets++;
1222 }
1223
1224 /* place incomplete frames back on ring for completion */
1225 rx_ring->skb = skb;
1226
1227 if (xdp_xmit) {
1228 struct ixgbevf_ring *xdp_ring =
1229 adapter->xdp_ring[rx_ring->queue_index];
1230
1231 /* Force memory writes to complete before letting h/w
1232 * know there are new descriptors to fetch.
1233 */
1234 wmb();
1235 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1236 }
1237
1238 u64_stats_update_begin(&rx_ring->syncp);
1239 rx_ring->stats.packets += total_rx_packets;
1240 rx_ring->stats.bytes += total_rx_bytes;
1241 u64_stats_update_end(&rx_ring->syncp);
1242 q_vector->rx.total_packets += total_rx_packets;
1243 q_vector->rx.total_bytes += total_rx_bytes;
1244
1245 return total_rx_packets;
1246}
1247
1248/**
1249 * ixgbevf_poll - NAPI polling calback
1250 * @napi: napi struct with our devices info in it
1251 * @budget: amount of work driver is allowed to do this pass, in packets
1252 *
1253 * This function will clean more than one or more rings associated with a
1254 * q_vector.
1255 **/
1256static int ixgbevf_poll(struct napi_struct *napi, int budget)
1257{
1258 struct ixgbevf_q_vector *q_vector =
1259 container_of(napi, struct ixgbevf_q_vector, napi);
1260 struct ixgbevf_adapter *adapter = q_vector->adapter;
1261 struct ixgbevf_ring *ring;
1262 int per_ring_budget, work_done = 0;
1263 bool clean_complete = true;
1264
1265 ixgbevf_for_each_ring(ring, q_vector->tx) {
1266 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1267 clean_complete = false;
1268 }
1269
1270 if (budget <= 0)
1271 return budget;
1272
1273 /* attempt to distribute budget to each queue fairly, but don't allow
1274 * the budget to go below 1 because we'll exit polling
1275 */
1276 if (q_vector->rx.count > 1)
1277 per_ring_budget = max(budget/q_vector->rx.count, 1);
1278 else
1279 per_ring_budget = budget;
1280
1281 ixgbevf_for_each_ring(ring, q_vector->rx) {
1282 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1283 per_ring_budget);
1284 work_done += cleaned;
1285 if (cleaned >= per_ring_budget)
1286 clean_complete = false;
1287 }
1288
1289 /* If all work not completed, return budget and keep polling */
1290 if (!clean_complete)
1291 return budget;
1292 /* all work done, exit the polling mode */
1293 napi_complete_done(napi, work_done);
1294 if (adapter->rx_itr_setting == 1)
1295 ixgbevf_set_itr(q_vector);
1296 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1297 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1298 ixgbevf_irq_enable_queues(adapter,
1299 BIT(q_vector->v_idx));
1300
1301 return 0;
1302}
1303
1304/**
1305 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1306 * @q_vector: structure containing interrupt and ring information
1307 **/
1308void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1309{
1310 struct ixgbevf_adapter *adapter = q_vector->adapter;
1311 struct ixgbe_hw *hw = &adapter->hw;
1312 int v_idx = q_vector->v_idx;
1313 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1314
1315 /* set the WDIS bit to not clear the timer bits and cause an
1316 * immediate assertion of the interrupt
1317 */
1318 itr_reg |= IXGBE_EITR_CNT_WDIS;
1319
1320 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1321}
1322
1323/**
1324 * ixgbevf_configure_msix - Configure MSI-X hardware
1325 * @adapter: board private structure
1326 *
1327 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1328 * interrupts.
1329 **/
1330static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1331{
1332 struct ixgbevf_q_vector *q_vector;
1333 int q_vectors, v_idx;
1334
1335 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1336 adapter->eims_enable_mask = 0;
1337
1338 /* Populate the IVAR table and set the ITR values to the
1339 * corresponding register.
1340 */
1341 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1342 struct ixgbevf_ring *ring;
1343
1344 q_vector = adapter->q_vector[v_idx];
1345
1346 ixgbevf_for_each_ring(ring, q_vector->rx)
1347 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1348
1349 ixgbevf_for_each_ring(ring, q_vector->tx)
1350 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1351
1352 if (q_vector->tx.ring && !q_vector->rx.ring) {
1353 /* Tx only vector */
1354 if (adapter->tx_itr_setting == 1)
1355 q_vector->itr = IXGBE_12K_ITR;
1356 else
1357 q_vector->itr = adapter->tx_itr_setting;
1358 } else {
1359 /* Rx or Rx/Tx vector */
1360 if (adapter->rx_itr_setting == 1)
1361 q_vector->itr = IXGBE_20K_ITR;
1362 else
1363 q_vector->itr = adapter->rx_itr_setting;
1364 }
1365
1366 /* add q_vector eims value to global eims_enable_mask */
1367 adapter->eims_enable_mask |= BIT(v_idx);
1368
1369 ixgbevf_write_eitr(q_vector);
1370 }
1371
1372 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1373 /* setup eims_other and add value to global eims_enable_mask */
1374 adapter->eims_other = BIT(v_idx);
1375 adapter->eims_enable_mask |= adapter->eims_other;
1376}
1377
1378enum latency_range {
1379 lowest_latency = 0,
1380 low_latency = 1,
1381 bulk_latency = 2,
1382 latency_invalid = 255
1383};
1384
1385/**
1386 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1387 * @q_vector: structure containing interrupt and ring information
1388 * @ring_container: structure containing ring performance data
1389 *
1390 * Stores a new ITR value based on packets and byte
1391 * counts during the last interrupt. The advantage of per interrupt
1392 * computation is faster updates and more accurate ITR for the current
1393 * traffic pattern. Constants in this function were computed
1394 * based on theoretical maximum wire speed and thresholds were set based
1395 * on testing data as well as attempting to minimize response time
1396 * while increasing bulk throughput.
1397 **/
1398static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1399 struct ixgbevf_ring_container *ring_container)
1400{
1401 int bytes = ring_container->total_bytes;
1402 int packets = ring_container->total_packets;
1403 u32 timepassed_us;
1404 u64 bytes_perint;
1405 u8 itr_setting = ring_container->itr;
1406
1407 if (packets == 0)
1408 return;
1409
1410 /* simple throttle rate management
1411 * 0-20MB/s lowest (100000 ints/s)
1412 * 20-100MB/s low (20000 ints/s)
1413 * 100-1249MB/s bulk (12000 ints/s)
1414 */
1415 /* what was last interrupt timeslice? */
1416 timepassed_us = q_vector->itr >> 2;
1417 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1418
1419 switch (itr_setting) {
1420 case lowest_latency:
1421 if (bytes_perint > 10)
1422 itr_setting = low_latency;
1423 break;
1424 case low_latency:
1425 if (bytes_perint > 20)
1426 itr_setting = bulk_latency;
1427 else if (bytes_perint <= 10)
1428 itr_setting = lowest_latency;
1429 break;
1430 case bulk_latency:
1431 if (bytes_perint <= 20)
1432 itr_setting = low_latency;
1433 break;
1434 }
1435
1436 /* clear work counters since we have the values we need */
1437 ring_container->total_bytes = 0;
1438 ring_container->total_packets = 0;
1439
1440 /* write updated itr to ring container */
1441 ring_container->itr = itr_setting;
1442}
1443
1444static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1445{
1446 u32 new_itr = q_vector->itr;
1447 u8 current_itr;
1448
1449 ixgbevf_update_itr(q_vector, &q_vector->tx);
1450 ixgbevf_update_itr(q_vector, &q_vector->rx);
1451
1452 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1453
1454 switch (current_itr) {
1455 /* counts and packets in update_itr are dependent on these numbers */
1456 case lowest_latency:
1457 new_itr = IXGBE_100K_ITR;
1458 break;
1459 case low_latency:
1460 new_itr = IXGBE_20K_ITR;
1461 break;
1462 case bulk_latency:
1463 new_itr = IXGBE_12K_ITR;
1464 break;
1465 default:
1466 break;
1467 }
1468
1469 if (new_itr != q_vector->itr) {
1470 /* do an exponential smoothing */
1471 new_itr = (10 * new_itr * q_vector->itr) /
1472 ((9 * new_itr) + q_vector->itr);
1473
1474 /* save the algorithm value here */
1475 q_vector->itr = new_itr;
1476
1477 ixgbevf_write_eitr(q_vector);
1478 }
1479}
1480
1481static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1482{
1483 struct ixgbevf_adapter *adapter = data;
1484 struct ixgbe_hw *hw = &adapter->hw;
1485
1486 hw->mac.get_link_status = 1;
1487
1488 ixgbevf_service_event_schedule(adapter);
1489
1490 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1491
1492 return IRQ_HANDLED;
1493}
1494
1495/**
1496 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1497 * @irq: unused
1498 * @data: pointer to our q_vector struct for this interrupt vector
1499 **/
1500static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1501{
1502 struct ixgbevf_q_vector *q_vector = data;
1503
1504 /* EIAM disabled interrupts (on this vector) for us */
1505 if (q_vector->rx.ring || q_vector->tx.ring)
1506 napi_schedule_irqoff(&q_vector->napi);
1507
1508 return IRQ_HANDLED;
1509}
1510
1511/**
1512 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1513 * @adapter: board private structure
1514 *
1515 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1516 * interrupts from the kernel.
1517 **/
1518static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1519{
1520 struct net_device *netdev = adapter->netdev;
1521 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1522 unsigned int ri = 0, ti = 0;
1523 int vector, err;
1524
1525 for (vector = 0; vector < q_vectors; vector++) {
1526 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1527 struct msix_entry *entry = &adapter->msix_entries[vector];
1528
1529 if (q_vector->tx.ring && q_vector->rx.ring) {
1530 snprintf(q_vector->name, sizeof(q_vector->name),
1531 "%s-TxRx-%u", netdev->name, ri++);
1532 ti++;
1533 } else if (q_vector->rx.ring) {
1534 snprintf(q_vector->name, sizeof(q_vector->name),
1535 "%s-rx-%u", netdev->name, ri++);
1536 } else if (q_vector->tx.ring) {
1537 snprintf(q_vector->name, sizeof(q_vector->name),
1538 "%s-tx-%u", netdev->name, ti++);
1539 } else {
1540 /* skip this unused q_vector */
1541 continue;
1542 }
1543 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1544 q_vector->name, q_vector);
1545 if (err) {
1546 hw_dbg(&adapter->hw,
1547 "request_irq failed for MSIX interrupt Error: %d\n",
1548 err);
1549 goto free_queue_irqs;
1550 }
1551 }
1552
1553 err = request_irq(adapter->msix_entries[vector].vector,
1554 &ixgbevf_msix_other, 0, netdev->name, adapter);
1555 if (err) {
1556 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1557 err);
1558 goto free_queue_irqs;
1559 }
1560
1561 return 0;
1562
1563free_queue_irqs:
1564 while (vector) {
1565 vector--;
1566 free_irq(adapter->msix_entries[vector].vector,
1567 adapter->q_vector[vector]);
1568 }
1569 /* This failure is non-recoverable - it indicates the system is
1570 * out of MSIX vector resources and the VF driver cannot run
1571 * without them. Set the number of msix vectors to zero
1572 * indicating that not enough can be allocated. The error
1573 * will be returned to the user indicating device open failed.
1574 * Any further attempts to force the driver to open will also
1575 * fail. The only way to recover is to unload the driver and
1576 * reload it again. If the system has recovered some MSIX
1577 * vectors then it may succeed.
1578 */
1579 adapter->num_msix_vectors = 0;
1580 return err;
1581}
1582
1583/**
1584 * ixgbevf_request_irq - initialize interrupts
1585 * @adapter: board private structure
1586 *
1587 * Attempts to configure interrupts using the best available
1588 * capabilities of the hardware and kernel.
1589 **/
1590static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1591{
1592 int err = ixgbevf_request_msix_irqs(adapter);
1593
1594 if (err)
1595 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1596
1597 return err;
1598}
1599
1600static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1601{
1602 int i, q_vectors;
1603
1604 if (!adapter->msix_entries)
1605 return;
1606
1607 q_vectors = adapter->num_msix_vectors;
1608 i = q_vectors - 1;
1609
1610 free_irq(adapter->msix_entries[i].vector, adapter);
1611 i--;
1612
1613 for (; i >= 0; i--) {
1614 /* free only the irqs that were actually requested */
1615 if (!adapter->q_vector[i]->rx.ring &&
1616 !adapter->q_vector[i]->tx.ring)
1617 continue;
1618
1619 free_irq(adapter->msix_entries[i].vector,
1620 adapter->q_vector[i]);
1621 }
1622}
1623
1624/**
1625 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1626 * @adapter: board private structure
1627 **/
1628static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1629{
1630 struct ixgbe_hw *hw = &adapter->hw;
1631 int i;
1632
1633 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1634 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1635 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1636
1637 IXGBE_WRITE_FLUSH(hw);
1638
1639 for (i = 0; i < adapter->num_msix_vectors; i++)
1640 synchronize_irq(adapter->msix_entries[i].vector);
1641}
1642
1643/**
1644 * ixgbevf_irq_enable - Enable default interrupt generation settings
1645 * @adapter: board private structure
1646 **/
1647static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1648{
1649 struct ixgbe_hw *hw = &adapter->hw;
1650
1651 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1652 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1653 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1654}
1655
1656/**
1657 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1658 * @adapter: board private structure
1659 * @ring: structure containing ring specific data
1660 *
1661 * Configure the Tx descriptor ring after a reset.
1662 **/
1663static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1664 struct ixgbevf_ring *ring)
1665{
1666 struct ixgbe_hw *hw = &adapter->hw;
1667 u64 tdba = ring->dma;
1668 int wait_loop = 10;
1669 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1670 u8 reg_idx = ring->reg_idx;
1671
1672 /* disable queue to avoid issues while updating state */
1673 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1674 IXGBE_WRITE_FLUSH(hw);
1675
1676 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1677 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1678 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1679 ring->count * sizeof(union ixgbe_adv_tx_desc));
1680
1681 /* disable head writeback */
1682 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1683 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1684
1685 /* enable relaxed ordering */
1686 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1687 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1688 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1689
1690 /* reset head and tail pointers */
1691 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1692 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1693 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1694
1695 /* reset ntu and ntc to place SW in sync with hardwdare */
1696 ring->next_to_clean = 0;
1697 ring->next_to_use = 0;
1698
1699 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1700 * to or less than the number of on chip descriptors, which is
1701 * currently 40.
1702 */
1703 txdctl |= (8 << 16); /* WTHRESH = 8 */
1704
1705 /* Setting PTHRESH to 32 both improves performance */
1706 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1707 32; /* PTHRESH = 32 */
1708
1709 /* reinitialize tx_buffer_info */
1710 memset(ring->tx_buffer_info, 0,
1711 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1712
1713 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1714
1715 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1716
1717 /* poll to verify queue is enabled */
1718 do {
1719 usleep_range(1000, 2000);
1720 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1721 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1722 if (!wait_loop)
1723 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1724}
1725
1726/**
1727 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1728 * @adapter: board private structure
1729 *
1730 * Configure the Tx unit of the MAC after a reset.
1731 **/
1732static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1733{
1734 u32 i;
1735
1736 /* Setup the HW Tx Head and Tail descriptor pointers */
1737 for (i = 0; i < adapter->num_tx_queues; i++)
1738 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1739 for (i = 0; i < adapter->num_xdp_queues; i++)
1740 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1741}
1742
1743#define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1744
1745static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1746 struct ixgbevf_ring *ring, int index)
1747{
1748 struct ixgbe_hw *hw = &adapter->hw;
1749 u32 srrctl;
1750
1751 srrctl = IXGBE_SRRCTL_DROP_EN;
1752
1753 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1754 if (ring_uses_large_buffer(ring))
1755 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1756 else
1757 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1758 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1759
1760 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1761}
1762
1763static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1764{
1765 struct ixgbe_hw *hw = &adapter->hw;
1766
1767 /* PSRTYPE must be initialized in 82599 */
1768 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1769 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1770 IXGBE_PSRTYPE_L2HDR;
1771
1772 if (adapter->num_rx_queues > 1)
1773 psrtype |= BIT(29);
1774
1775 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1776}
1777
1778#define IXGBEVF_MAX_RX_DESC_POLL 10
1779static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1780 struct ixgbevf_ring *ring)
1781{
1782 struct ixgbe_hw *hw = &adapter->hw;
1783 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1784 u32 rxdctl;
1785 u8 reg_idx = ring->reg_idx;
1786
1787 if (IXGBE_REMOVED(hw->hw_addr))
1788 return;
1789 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1790 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1791
1792 /* write value back with RXDCTL.ENABLE bit cleared */
1793 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1794
1795 /* the hardware may take up to 100us to really disable the Rx queue */
1796 do {
1797 udelay(10);
1798 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1799 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1800
1801 if (!wait_loop)
1802 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1803 reg_idx);
1804}
1805
1806static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1807 struct ixgbevf_ring *ring)
1808{
1809 struct ixgbe_hw *hw = &adapter->hw;
1810 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1811 u32 rxdctl;
1812 u8 reg_idx = ring->reg_idx;
1813
1814 if (IXGBE_REMOVED(hw->hw_addr))
1815 return;
1816 do {
1817 usleep_range(1000, 2000);
1818 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1819 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1820
1821 if (!wait_loop)
1822 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1823 reg_idx);
1824}
1825
1826/**
1827 * ixgbevf_init_rss_key - Initialize adapter RSS key
1828 * @adapter: device handle
1829 *
1830 * Allocates and initializes the RSS key if it is not allocated.
1831 **/
1832static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1833{
1834 u32 *rss_key;
1835
1836 if (!adapter->rss_key) {
1837 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1838 if (unlikely(!rss_key))
1839 return -ENOMEM;
1840
1841 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1842 adapter->rss_key = rss_key;
1843 }
1844
1845 return 0;
1846}
1847
1848static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1849{
1850 struct ixgbe_hw *hw = &adapter->hw;
1851 u32 vfmrqc = 0, vfreta = 0;
1852 u16 rss_i = adapter->num_rx_queues;
1853 u8 i, j;
1854
1855 /* Fill out hash function seeds */
1856 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1857 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1858
1859 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1860 if (j == rss_i)
1861 j = 0;
1862
1863 adapter->rss_indir_tbl[i] = j;
1864
1865 vfreta |= j << (i & 0x3) * 8;
1866 if ((i & 3) == 3) {
1867 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1868 vfreta = 0;
1869 }
1870 }
1871
1872 /* Perform hash on these packet types */
1873 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1874 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1875 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1876 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1877
1878 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1879
1880 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1881}
1882
1883static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1884 struct ixgbevf_ring *ring)
1885{
1886 struct ixgbe_hw *hw = &adapter->hw;
1887 union ixgbe_adv_rx_desc *rx_desc;
1888 u64 rdba = ring->dma;
1889 u32 rxdctl;
1890 u8 reg_idx = ring->reg_idx;
1891
1892 /* disable queue to avoid issues while updating state */
1893 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1894 ixgbevf_disable_rx_queue(adapter, ring);
1895
1896 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1897 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1898 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1899 ring->count * sizeof(union ixgbe_adv_rx_desc));
1900
1901#ifndef CONFIG_SPARC
1902 /* enable relaxed ordering */
1903 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1904 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1905#else
1906 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1907 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1908 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1909#endif
1910
1911 /* reset head and tail pointers */
1912 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1913 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1914 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1915
1916 /* initialize rx_buffer_info */
1917 memset(ring->rx_buffer_info, 0,
1918 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1919
1920 /* initialize Rx descriptor 0 */
1921 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1922 rx_desc->wb.upper.length = 0;
1923
1924 /* reset ntu and ntc to place SW in sync with hardwdare */
1925 ring->next_to_clean = 0;
1926 ring->next_to_use = 0;
1927 ring->next_to_alloc = 0;
1928
1929 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1930
1931 /* RXDCTL.RLPML does not work on 82599 */
1932 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1933 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1934 IXGBE_RXDCTL_RLPML_EN);
1935
1936#if (PAGE_SIZE < 8192)
1937 /* Limit the maximum frame size so we don't overrun the skb */
1938 if (ring_uses_build_skb(ring) &&
1939 !ring_uses_large_buffer(ring))
1940 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1941 IXGBE_RXDCTL_RLPML_EN;
1942#endif
1943 }
1944
1945 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1946 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1947
1948 ixgbevf_rx_desc_queue_enable(adapter, ring);
1949 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1950}
1951
1952static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1953 struct ixgbevf_ring *rx_ring)
1954{
1955 struct net_device *netdev = adapter->netdev;
1956 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1957
1958 /* set build_skb and buffer size flags */
1959 clear_ring_build_skb_enabled(rx_ring);
1960 clear_ring_uses_large_buffer(rx_ring);
1961
1962 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1963 return;
1964
1965 set_ring_build_skb_enabled(rx_ring);
1966
1967 if (PAGE_SIZE < 8192) {
1968 if (max_frame <= IXGBEVF_MAX_FRAME_BUILD_SKB)
1969 return;
1970
1971 set_ring_uses_large_buffer(rx_ring);
1972 }
1973}
1974
1975/**
1976 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1977 * @adapter: board private structure
1978 *
1979 * Configure the Rx unit of the MAC after a reset.
1980 **/
1981static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1982{
1983 struct ixgbe_hw *hw = &adapter->hw;
1984 struct net_device *netdev = adapter->netdev;
1985 int i, ret;
1986
1987 ixgbevf_setup_psrtype(adapter);
1988 if (hw->mac.type >= ixgbe_mac_X550_vf)
1989 ixgbevf_setup_vfmrqc(adapter);
1990
1991 spin_lock_bh(&adapter->mbx_lock);
1992 /* notify the PF of our intent to use this size of frame */
1993 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
1994 spin_unlock_bh(&adapter->mbx_lock);
1995 if (ret)
1996 dev_err(&adapter->pdev->dev,
1997 "Failed to set MTU at %d\n", netdev->mtu);
1998
1999 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2000 * the Base and Length of the Rx Descriptor Ring
2001 */
2002 for (i = 0; i < adapter->num_rx_queues; i++) {
2003 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2004
2005 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2006 ixgbevf_configure_rx_ring(adapter, rx_ring);
2007 }
2008}
2009
2010static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2011 __be16 proto, u16 vid)
2012{
2013 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2014 struct ixgbe_hw *hw = &adapter->hw;
2015 int err;
2016
2017 spin_lock_bh(&adapter->mbx_lock);
2018
2019 /* add VID to filter table */
2020 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2021
2022 spin_unlock_bh(&adapter->mbx_lock);
2023
2024 /* translate error return types so error makes sense */
2025 if (err == IXGBE_ERR_MBX)
2026 return -EIO;
2027
2028 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2029 return -EACCES;
2030
2031 set_bit(vid, adapter->active_vlans);
2032
2033 return err;
2034}
2035
2036static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2037 __be16 proto, u16 vid)
2038{
2039 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2040 struct ixgbe_hw *hw = &adapter->hw;
2041 int err;
2042
2043 spin_lock_bh(&adapter->mbx_lock);
2044
2045 /* remove VID from filter table */
2046 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2047
2048 spin_unlock_bh(&adapter->mbx_lock);
2049
2050 clear_bit(vid, adapter->active_vlans);
2051
2052 return err;
2053}
2054
2055static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2056{
2057 u16 vid;
2058
2059 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2060 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2061 htons(ETH_P_8021Q), vid);
2062}
2063
2064static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2065{
2066 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2067 struct ixgbe_hw *hw = &adapter->hw;
2068 int count = 0;
2069
2070 if ((netdev_uc_count(netdev)) > 10) {
2071 pr_err("Too many unicast filters - No Space\n");
2072 return -ENOSPC;
2073 }
2074
2075 if (!netdev_uc_empty(netdev)) {
2076 struct netdev_hw_addr *ha;
2077
2078 netdev_for_each_uc_addr(ha, netdev) {
2079 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2080 udelay(200);
2081 }
2082 } else {
2083 /* If the list is empty then send message to PF driver to
2084 * clear all MAC VLANs on this VF.
2085 */
2086 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2087 }
2088
2089 return count;
2090}
2091
2092/**
2093 * ixgbevf_set_rx_mode - Multicast and unicast set
2094 * @netdev: network interface device structure
2095 *
2096 * The set_rx_method entry point is called whenever the multicast address
2097 * list, unicast address list or the network interface flags are updated.
2098 * This routine is responsible for configuring the hardware for proper
2099 * multicast mode and configuring requested unicast filters.
2100 **/
2101static void ixgbevf_set_rx_mode(struct net_device *netdev)
2102{
2103 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2104 struct ixgbe_hw *hw = &adapter->hw;
2105 unsigned int flags = netdev->flags;
2106 int xcast_mode;
2107
2108 /* request the most inclusive mode we need */
2109 if (flags & IFF_PROMISC)
2110 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2111 else if (flags & IFF_ALLMULTI)
2112 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2113 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2114 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2115 else
2116 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2117
2118 spin_lock_bh(&adapter->mbx_lock);
2119
2120 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2121
2122 /* reprogram multicast list */
2123 hw->mac.ops.update_mc_addr_list(hw, netdev);
2124
2125 ixgbevf_write_uc_addr_list(netdev);
2126
2127 spin_unlock_bh(&adapter->mbx_lock);
2128}
2129
2130static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2131{
2132 int q_idx;
2133 struct ixgbevf_q_vector *q_vector;
2134 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2135
2136 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2137 q_vector = adapter->q_vector[q_idx];
2138 napi_enable(&q_vector->napi);
2139 }
2140}
2141
2142static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2143{
2144 int q_idx;
2145 struct ixgbevf_q_vector *q_vector;
2146 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2147
2148 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2149 q_vector = adapter->q_vector[q_idx];
2150 napi_disable(&q_vector->napi);
2151 }
2152}
2153
2154static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2155{
2156 struct ixgbe_hw *hw = &adapter->hw;
2157 unsigned int def_q = 0;
2158 unsigned int num_tcs = 0;
2159 unsigned int num_rx_queues = adapter->num_rx_queues;
2160 unsigned int num_tx_queues = adapter->num_tx_queues;
2161 int err;
2162
2163 spin_lock_bh(&adapter->mbx_lock);
2164
2165 /* fetch queue configuration from the PF */
2166 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2167
2168 spin_unlock_bh(&adapter->mbx_lock);
2169
2170 if (err)
2171 return err;
2172
2173 if (num_tcs > 1) {
2174 /* we need only one Tx queue */
2175 num_tx_queues = 1;
2176
2177 /* update default Tx ring register index */
2178 adapter->tx_ring[0]->reg_idx = def_q;
2179
2180 /* we need as many queues as traffic classes */
2181 num_rx_queues = num_tcs;
2182 }
2183
2184 /* if we have a bad config abort request queue reset */
2185 if ((adapter->num_rx_queues != num_rx_queues) ||
2186 (adapter->num_tx_queues != num_tx_queues)) {
2187 /* force mailbox timeout to prevent further messages */
2188 hw->mbx.timeout = 0;
2189
2190 /* wait for watchdog to come around and bail us out */
2191 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2192 }
2193
2194 return 0;
2195}
2196
2197static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2198{
2199 ixgbevf_configure_dcb(adapter);
2200
2201 ixgbevf_set_rx_mode(adapter->netdev);
2202
2203 ixgbevf_restore_vlan(adapter);
2204
2205 ixgbevf_configure_tx(adapter);
2206 ixgbevf_configure_rx(adapter);
2207}
2208
2209static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2210{
2211 /* Only save pre-reset stats if there are some */
2212 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2213 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2214 adapter->stats.base_vfgprc;
2215 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2216 adapter->stats.base_vfgptc;
2217 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2218 adapter->stats.base_vfgorc;
2219 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2220 adapter->stats.base_vfgotc;
2221 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2222 adapter->stats.base_vfmprc;
2223 }
2224}
2225
2226static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2227{
2228 struct ixgbe_hw *hw = &adapter->hw;
2229
2230 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2231 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2232 adapter->stats.last_vfgorc |=
2233 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2234 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2235 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2236 adapter->stats.last_vfgotc |=
2237 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2238 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2239
2240 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2241 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2242 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2243 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2244 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2245}
2246
2247static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2248{
2249 struct ixgbe_hw *hw = &adapter->hw;
2250 int api[] = { ixgbe_mbox_api_13,
2251 ixgbe_mbox_api_12,
2252 ixgbe_mbox_api_11,
2253 ixgbe_mbox_api_10,
2254 ixgbe_mbox_api_unknown };
2255 int err, idx = 0;
2256
2257 spin_lock_bh(&adapter->mbx_lock);
2258
2259 while (api[idx] != ixgbe_mbox_api_unknown) {
2260 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2261 if (!err)
2262 break;
2263 idx++;
2264 }
2265
2266 spin_unlock_bh(&adapter->mbx_lock);
2267}
2268
2269static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2270{
2271 struct net_device *netdev = adapter->netdev;
2272 struct ixgbe_hw *hw = &adapter->hw;
2273
2274 ixgbevf_configure_msix(adapter);
2275
2276 spin_lock_bh(&adapter->mbx_lock);
2277
2278 if (is_valid_ether_addr(hw->mac.addr))
2279 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2280 else
2281 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2282
2283 spin_unlock_bh(&adapter->mbx_lock);
2284
2285 smp_mb__before_atomic();
2286 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2287 ixgbevf_napi_enable_all(adapter);
2288
2289 /* clear any pending interrupts, may auto mask */
2290 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2291 ixgbevf_irq_enable(adapter);
2292
2293 /* enable transmits */
2294 netif_tx_start_all_queues(netdev);
2295
2296 ixgbevf_save_reset_stats(adapter);
2297 ixgbevf_init_last_counter_stats(adapter);
2298
2299 hw->mac.get_link_status = 1;
2300 mod_timer(&adapter->service_timer, jiffies);
2301}
2302
2303void ixgbevf_up(struct ixgbevf_adapter *adapter)
2304{
2305 ixgbevf_configure(adapter);
2306
2307 ixgbevf_up_complete(adapter);
2308}
2309
2310/**
2311 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2312 * @rx_ring: ring to free buffers from
2313 **/
2314static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2315{
2316 u16 i = rx_ring->next_to_clean;
2317
2318 /* Free Rx ring sk_buff */
2319 if (rx_ring->skb) {
2320 dev_kfree_skb(rx_ring->skb);
2321 rx_ring->skb = NULL;
2322 }
2323
2324 /* Free all the Rx ring pages */
2325 while (i != rx_ring->next_to_alloc) {
2326 struct ixgbevf_rx_buffer *rx_buffer;
2327
2328 rx_buffer = &rx_ring->rx_buffer_info[i];
2329
2330 /* Invalidate cache lines that may have been written to by
2331 * device so that we avoid corrupting memory.
2332 */
2333 dma_sync_single_range_for_cpu(rx_ring->dev,
2334 rx_buffer->dma,
2335 rx_buffer->page_offset,
2336 ixgbevf_rx_bufsz(rx_ring),
2337 DMA_FROM_DEVICE);
2338
2339 /* free resources associated with mapping */
2340 dma_unmap_page_attrs(rx_ring->dev,
2341 rx_buffer->dma,
2342 ixgbevf_rx_pg_size(rx_ring),
2343 DMA_FROM_DEVICE,
2344 IXGBEVF_RX_DMA_ATTR);
2345
2346 __page_frag_cache_drain(rx_buffer->page,
2347 rx_buffer->pagecnt_bias);
2348
2349 i++;
2350 if (i == rx_ring->count)
2351 i = 0;
2352 }
2353
2354 rx_ring->next_to_alloc = 0;
2355 rx_ring->next_to_clean = 0;
2356 rx_ring->next_to_use = 0;
2357}
2358
2359/**
2360 * ixgbevf_clean_tx_ring - Free Tx Buffers
2361 * @tx_ring: ring to be cleaned
2362 **/
2363static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2364{
2365 u16 i = tx_ring->next_to_clean;
2366 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2367
2368 while (i != tx_ring->next_to_use) {
2369 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2370
2371 /* Free all the Tx ring sk_buffs */
2372 if (ring_is_xdp(tx_ring))
2373 page_frag_free(tx_buffer->data);
2374 else
2375 dev_kfree_skb_any(tx_buffer->skb);
2376
2377 /* unmap skb header data */
2378 dma_unmap_single(tx_ring->dev,
2379 dma_unmap_addr(tx_buffer, dma),
2380 dma_unmap_len(tx_buffer, len),
2381 DMA_TO_DEVICE);
2382
2383 /* check for eop_desc to determine the end of the packet */
2384 eop_desc = tx_buffer->next_to_watch;
2385 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2386
2387 /* unmap remaining buffers */
2388 while (tx_desc != eop_desc) {
2389 tx_buffer++;
2390 tx_desc++;
2391 i++;
2392 if (unlikely(i == tx_ring->count)) {
2393 i = 0;
2394 tx_buffer = tx_ring->tx_buffer_info;
2395 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2396 }
2397
2398 /* unmap any remaining paged data */
2399 if (dma_unmap_len(tx_buffer, len))
2400 dma_unmap_page(tx_ring->dev,
2401 dma_unmap_addr(tx_buffer, dma),
2402 dma_unmap_len(tx_buffer, len),
2403 DMA_TO_DEVICE);
2404 }
2405
2406 /* move us one more past the eop_desc for start of next pkt */
2407 tx_buffer++;
2408 i++;
2409 if (unlikely(i == tx_ring->count)) {
2410 i = 0;
2411 tx_buffer = tx_ring->tx_buffer_info;
2412 }
2413 }
2414
2415 /* reset next_to_use and next_to_clean */
2416 tx_ring->next_to_use = 0;
2417 tx_ring->next_to_clean = 0;
2418
2419}
2420
2421/**
2422 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2423 * @adapter: board private structure
2424 **/
2425static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2426{
2427 int i;
2428
2429 for (i = 0; i < adapter->num_rx_queues; i++)
2430 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2431}
2432
2433/**
2434 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2435 * @adapter: board private structure
2436 **/
2437static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2438{
2439 int i;
2440
2441 for (i = 0; i < adapter->num_tx_queues; i++)
2442 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2443 for (i = 0; i < adapter->num_xdp_queues; i++)
2444 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2445}
2446
2447void ixgbevf_down(struct ixgbevf_adapter *adapter)
2448{
2449 struct net_device *netdev = adapter->netdev;
2450 struct ixgbe_hw *hw = &adapter->hw;
2451 int i;
2452
2453 /* signal that we are down to the interrupt handler */
2454 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2455 return; /* do nothing if already down */
2456
2457 /* disable all enabled Rx queues */
2458 for (i = 0; i < adapter->num_rx_queues; i++)
2459 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2460
2461 usleep_range(10000, 20000);
2462
2463 netif_tx_stop_all_queues(netdev);
2464
2465 /* call carrier off first to avoid false dev_watchdog timeouts */
2466 netif_carrier_off(netdev);
2467 netif_tx_disable(netdev);
2468
2469 ixgbevf_irq_disable(adapter);
2470
2471 ixgbevf_napi_disable_all(adapter);
2472
2473 del_timer_sync(&adapter->service_timer);
2474
2475 /* disable transmits in the hardware now that interrupts are off */
2476 for (i = 0; i < adapter->num_tx_queues; i++) {
2477 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2478
2479 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2480 IXGBE_TXDCTL_SWFLSH);
2481 }
2482
2483 for (i = 0; i < adapter->num_xdp_queues; i++) {
2484 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2485
2486 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2487 IXGBE_TXDCTL_SWFLSH);
2488 }
2489
2490 if (!pci_channel_offline(adapter->pdev))
2491 ixgbevf_reset(adapter);
2492
2493 ixgbevf_clean_all_tx_rings(adapter);
2494 ixgbevf_clean_all_rx_rings(adapter);
2495}
2496
2497void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2498{
2499 WARN_ON(in_interrupt());
2500
2501 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2502 msleep(1);
2503
2504 ixgbevf_down(adapter);
2505 ixgbevf_up(adapter);
2506
2507 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2508}
2509
2510void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2511{
2512 struct ixgbe_hw *hw = &adapter->hw;
2513 struct net_device *netdev = adapter->netdev;
2514
2515 if (hw->mac.ops.reset_hw(hw)) {
2516 hw_dbg(hw, "PF still resetting\n");
2517 } else {
2518 hw->mac.ops.init_hw(hw);
2519 ixgbevf_negotiate_api(adapter);
2520 }
2521
2522 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2523 ether_addr_copy(netdev->dev_addr, adapter->hw.mac.addr);
2524 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2525 }
2526
2527 adapter->last_reset = jiffies;
2528}
2529
2530static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2531 int vectors)
2532{
2533 int vector_threshold;
2534
2535 /* We'll want at least 2 (vector_threshold):
2536 * 1) TxQ[0] + RxQ[0] handler
2537 * 2) Other (Link Status Change, etc.)
2538 */
2539 vector_threshold = MIN_MSIX_COUNT;
2540
2541 /* The more we get, the more we will assign to Tx/Rx Cleanup
2542 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2543 * Right now, we simply care about how many we'll get; we'll
2544 * set them up later while requesting irq's.
2545 */
2546 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2547 vector_threshold, vectors);
2548
2549 if (vectors < 0) {
2550 dev_err(&adapter->pdev->dev,
2551 "Unable to allocate MSI-X interrupts\n");
2552 kfree(adapter->msix_entries);
2553 adapter->msix_entries = NULL;
2554 return vectors;
2555 }
2556
2557 /* Adjust for only the vectors we'll use, which is minimum
2558 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2559 * vectors we were allocated.
2560 */
2561 adapter->num_msix_vectors = vectors;
2562
2563 return 0;
2564}
2565
2566/**
2567 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2568 * @adapter: board private structure to initialize
2569 *
2570 * This is the top level queue allocation routine. The order here is very
2571 * important, starting with the "most" number of features turned on at once,
2572 * and ending with the smallest set of features. This way large combinations
2573 * can be allocated if they're turned on, and smaller combinations are the
2574 * fallthrough conditions.
2575 *
2576 **/
2577static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2578{
2579 struct ixgbe_hw *hw = &adapter->hw;
2580 unsigned int def_q = 0;
2581 unsigned int num_tcs = 0;
2582 int err;
2583
2584 /* Start with base case */
2585 adapter->num_rx_queues = 1;
2586 adapter->num_tx_queues = 1;
2587 adapter->num_xdp_queues = 0;
2588
2589 spin_lock_bh(&adapter->mbx_lock);
2590
2591 /* fetch queue configuration from the PF */
2592 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2593
2594 spin_unlock_bh(&adapter->mbx_lock);
2595
2596 if (err)
2597 return;
2598
2599 /* we need as many queues as traffic classes */
2600 if (num_tcs > 1) {
2601 adapter->num_rx_queues = num_tcs;
2602 } else {
2603 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2604
2605 switch (hw->api_version) {
2606 case ixgbe_mbox_api_11:
2607 case ixgbe_mbox_api_12:
2608 case ixgbe_mbox_api_13:
2609 if (adapter->xdp_prog &&
2610 hw->mac.max_tx_queues == rss)
2611 rss = rss > 3 ? 2 : 1;
2612
2613 adapter->num_rx_queues = rss;
2614 adapter->num_tx_queues = rss;
2615 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2616 default:
2617 break;
2618 }
2619 }
2620}
2621
2622/**
2623 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2624 * @adapter: board private structure to initialize
2625 *
2626 * Attempt to configure the interrupts using the best available
2627 * capabilities of the hardware and the kernel.
2628 **/
2629static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2630{
2631 int vector, v_budget;
2632
2633 /* It's easy to be greedy for MSI-X vectors, but it really
2634 * doesn't do us much good if we have a lot more vectors
2635 * than CPU's. So let's be conservative and only ask for
2636 * (roughly) the same number of vectors as there are CPU's.
2637 * The default is to use pairs of vectors.
2638 */
2639 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2640 v_budget = min_t(int, v_budget, num_online_cpus());
2641 v_budget += NON_Q_VECTORS;
2642
2643 adapter->msix_entries = kcalloc(v_budget,
2644 sizeof(struct msix_entry), GFP_KERNEL);
2645 if (!adapter->msix_entries)
2646 return -ENOMEM;
2647
2648 for (vector = 0; vector < v_budget; vector++)
2649 adapter->msix_entries[vector].entry = vector;
2650
2651 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2652 * does not support any other modes, so we will simply fail here. Note
2653 * that we clean up the msix_entries pointer else-where.
2654 */
2655 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2656}
2657
2658static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2659 struct ixgbevf_ring_container *head)
2660{
2661 ring->next = head->ring;
2662 head->ring = ring;
2663 head->count++;
2664}
2665
2666/**
2667 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2668 * @adapter: board private structure to initialize
2669 * @v_idx: index of vector in adapter struct
2670 * @txr_count: number of Tx rings for q vector
2671 * @txr_idx: index of first Tx ring to assign
2672 * @xdp_count: total number of XDP rings to allocate
2673 * @xdp_idx: index of first XDP ring to allocate
2674 * @rxr_count: number of Rx rings for q vector
2675 * @rxr_idx: index of first Rx ring to assign
2676 *
2677 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2678 **/
2679static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2680 int txr_count, int txr_idx,
2681 int xdp_count, int xdp_idx,
2682 int rxr_count, int rxr_idx)
2683{
2684 struct ixgbevf_q_vector *q_vector;
2685 int reg_idx = txr_idx + xdp_idx;
2686 struct ixgbevf_ring *ring;
2687 int ring_count, size;
2688
2689 ring_count = txr_count + xdp_count + rxr_count;
2690 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2691
2692 /* allocate q_vector and rings */
2693 q_vector = kzalloc(size, GFP_KERNEL);
2694 if (!q_vector)
2695 return -ENOMEM;
2696
2697 /* initialize NAPI */
2698 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll, 64);
2699
2700 /* tie q_vector and adapter together */
2701 adapter->q_vector[v_idx] = q_vector;
2702 q_vector->adapter = adapter;
2703 q_vector->v_idx = v_idx;
2704
2705 /* initialize pointer to rings */
2706 ring = q_vector->ring;
2707
2708 while (txr_count) {
2709 /* assign generic ring traits */
2710 ring->dev = &adapter->pdev->dev;
2711 ring->netdev = adapter->netdev;
2712
2713 /* configure backlink on ring */
2714 ring->q_vector = q_vector;
2715
2716 /* update q_vector Tx values */
2717 ixgbevf_add_ring(ring, &q_vector->tx);
2718
2719 /* apply Tx specific ring traits */
2720 ring->count = adapter->tx_ring_count;
2721 ring->queue_index = txr_idx;
2722 ring->reg_idx = reg_idx;
2723
2724 /* assign ring to adapter */
2725 adapter->tx_ring[txr_idx] = ring;
2726
2727 /* update count and index */
2728 txr_count--;
2729 txr_idx++;
2730 reg_idx++;
2731
2732 /* push pointer to next ring */
2733 ring++;
2734 }
2735
2736 while (xdp_count) {
2737 /* assign generic ring traits */
2738 ring->dev = &adapter->pdev->dev;
2739 ring->netdev = adapter->netdev;
2740
2741 /* configure backlink on ring */
2742 ring->q_vector = q_vector;
2743
2744 /* update q_vector Tx values */
2745 ixgbevf_add_ring(ring, &q_vector->tx);
2746
2747 /* apply Tx specific ring traits */
2748 ring->count = adapter->tx_ring_count;
2749 ring->queue_index = xdp_idx;
2750 ring->reg_idx = reg_idx;
2751 set_ring_xdp(ring);
2752
2753 /* assign ring to adapter */
2754 adapter->xdp_ring[xdp_idx] = ring;
2755
2756 /* update count and index */
2757 xdp_count--;
2758 xdp_idx++;
2759 reg_idx++;
2760
2761 /* push pointer to next ring */
2762 ring++;
2763 }
2764
2765 while (rxr_count) {
2766 /* assign generic ring traits */
2767 ring->dev = &adapter->pdev->dev;
2768 ring->netdev = adapter->netdev;
2769
2770 /* configure backlink on ring */
2771 ring->q_vector = q_vector;
2772
2773 /* update q_vector Rx values */
2774 ixgbevf_add_ring(ring, &q_vector->rx);
2775
2776 /* apply Rx specific ring traits */
2777 ring->count = adapter->rx_ring_count;
2778 ring->queue_index = rxr_idx;
2779 ring->reg_idx = rxr_idx;
2780
2781 /* assign ring to adapter */
2782 adapter->rx_ring[rxr_idx] = ring;
2783
2784 /* update count and index */
2785 rxr_count--;
2786 rxr_idx++;
2787
2788 /* push pointer to next ring */
2789 ring++;
2790 }
2791
2792 return 0;
2793}
2794
2795/**
2796 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2797 * @adapter: board private structure to initialize
2798 * @v_idx: index of vector in adapter struct
2799 *
2800 * This function frees the memory allocated to the q_vector. In addition if
2801 * NAPI is enabled it will delete any references to the NAPI struct prior
2802 * to freeing the q_vector.
2803 **/
2804static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2805{
2806 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2807 struct ixgbevf_ring *ring;
2808
2809 ixgbevf_for_each_ring(ring, q_vector->tx) {
2810 if (ring_is_xdp(ring))
2811 adapter->xdp_ring[ring->queue_index] = NULL;
2812 else
2813 adapter->tx_ring[ring->queue_index] = NULL;
2814 }
2815
2816 ixgbevf_for_each_ring(ring, q_vector->rx)
2817 adapter->rx_ring[ring->queue_index] = NULL;
2818
2819 adapter->q_vector[v_idx] = NULL;
2820 netif_napi_del(&q_vector->napi);
2821
2822 /* ixgbevf_get_stats() might access the rings on this vector,
2823 * we must wait a grace period before freeing it.
2824 */
2825 kfree_rcu(q_vector, rcu);
2826}
2827
2828/**
2829 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2830 * @adapter: board private structure to initialize
2831 *
2832 * We allocate one q_vector per queue interrupt. If allocation fails we
2833 * return -ENOMEM.
2834 **/
2835static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2836{
2837 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2838 int rxr_remaining = adapter->num_rx_queues;
2839 int txr_remaining = adapter->num_tx_queues;
2840 int xdp_remaining = adapter->num_xdp_queues;
2841 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2842 int err;
2843
2844 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2845 for (; rxr_remaining; v_idx++, q_vectors--) {
2846 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2847
2848 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2849 0, 0, 0, 0, rqpv, rxr_idx);
2850 if (err)
2851 goto err_out;
2852
2853 /* update counts and index */
2854 rxr_remaining -= rqpv;
2855 rxr_idx += rqpv;
2856 }
2857 }
2858
2859 for (; q_vectors; v_idx++, q_vectors--) {
2860 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2861 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2862 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2863
2864 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2865 tqpv, txr_idx,
2866 xqpv, xdp_idx,
2867 rqpv, rxr_idx);
2868
2869 if (err)
2870 goto err_out;
2871
2872 /* update counts and index */
2873 rxr_remaining -= rqpv;
2874 rxr_idx += rqpv;
2875 txr_remaining -= tqpv;
2876 txr_idx += tqpv;
2877 xdp_remaining -= xqpv;
2878 xdp_idx += xqpv;
2879 }
2880
2881 return 0;
2882
2883err_out:
2884 while (v_idx) {
2885 v_idx--;
2886 ixgbevf_free_q_vector(adapter, v_idx);
2887 }
2888
2889 return -ENOMEM;
2890}
2891
2892/**
2893 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2894 * @adapter: board private structure to initialize
2895 *
2896 * This function frees the memory allocated to the q_vectors. In addition if
2897 * NAPI is enabled it will delete any references to the NAPI struct prior
2898 * to freeing the q_vector.
2899 **/
2900static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2901{
2902 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2903
2904 while (q_vectors) {
2905 q_vectors--;
2906 ixgbevf_free_q_vector(adapter, q_vectors);
2907 }
2908}
2909
2910/**
2911 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2912 * @adapter: board private structure
2913 *
2914 **/
2915static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2916{
2917 if (!adapter->msix_entries)
2918 return;
2919
2920 pci_disable_msix(adapter->pdev);
2921 kfree(adapter->msix_entries);
2922 adapter->msix_entries = NULL;
2923}
2924
2925/**
2926 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2927 * @adapter: board private structure to initialize
2928 *
2929 **/
2930static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2931{
2932 int err;
2933
2934 /* Number of supported queues */
2935 ixgbevf_set_num_queues(adapter);
2936
2937 err = ixgbevf_set_interrupt_capability(adapter);
2938 if (err) {
2939 hw_dbg(&adapter->hw,
2940 "Unable to setup interrupt capabilities\n");
2941 goto err_set_interrupt;
2942 }
2943
2944 err = ixgbevf_alloc_q_vectors(adapter);
2945 if (err) {
2946 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2947 goto err_alloc_q_vectors;
2948 }
2949
2950 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2951 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2952 adapter->num_rx_queues, adapter->num_tx_queues,
2953 adapter->num_xdp_queues);
2954
2955 set_bit(__IXGBEVF_DOWN, &adapter->state);
2956
2957 return 0;
2958err_alloc_q_vectors:
2959 ixgbevf_reset_interrupt_capability(adapter);
2960err_set_interrupt:
2961 return err;
2962}
2963
2964/**
2965 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2966 * @adapter: board private structure to clear interrupt scheme on
2967 *
2968 * We go through and clear interrupt specific resources and reset the structure
2969 * to pre-load conditions
2970 **/
2971static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2972{
2973 adapter->num_tx_queues = 0;
2974 adapter->num_xdp_queues = 0;
2975 adapter->num_rx_queues = 0;
2976
2977 ixgbevf_free_q_vectors(adapter);
2978 ixgbevf_reset_interrupt_capability(adapter);
2979}
2980
2981/**
2982 * ixgbevf_sw_init - Initialize general software structures
2983 * @adapter: board private structure to initialize
2984 *
2985 * ixgbevf_sw_init initializes the Adapter private data structure.
2986 * Fields are initialized based on PCI device information and
2987 * OS network device settings (MTU size).
2988 **/
2989static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2990{
2991 struct ixgbe_hw *hw = &adapter->hw;
2992 struct pci_dev *pdev = adapter->pdev;
2993 struct net_device *netdev = adapter->netdev;
2994 int err;
2995
2996 /* PCI config space info */
2997 hw->vendor_id = pdev->vendor;
2998 hw->device_id = pdev->device;
2999 hw->revision_id = pdev->revision;
3000 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3001 hw->subsystem_device_id = pdev->subsystem_device;
3002
3003 hw->mbx.ops.init_params(hw);
3004
3005 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3006 err = ixgbevf_init_rss_key(adapter);
3007 if (err)
3008 goto out;
3009 }
3010
3011 /* assume legacy case in which PF would only give VF 2 queues */
3012 hw->mac.max_tx_queues = 2;
3013 hw->mac.max_rx_queues = 2;
3014
3015 /* lock to protect mailbox accesses */
3016 spin_lock_init(&adapter->mbx_lock);
3017
3018 err = hw->mac.ops.reset_hw(hw);
3019 if (err) {
3020 dev_info(&pdev->dev,
3021 "PF still in reset state. Is the PF interface up?\n");
3022 } else {
3023 err = hw->mac.ops.init_hw(hw);
3024 if (err) {
3025 pr_err("init_shared_code failed: %d\n", err);
3026 goto out;
3027 }
3028 ixgbevf_negotiate_api(adapter);
3029 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3030 if (err)
3031 dev_info(&pdev->dev, "Error reading MAC address\n");
3032 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3033 dev_info(&pdev->dev,
3034 "MAC address not assigned by administrator.\n");
3035 ether_addr_copy(netdev->dev_addr, hw->mac.addr);
3036 }
3037
3038 if (!is_valid_ether_addr(netdev->dev_addr)) {
3039 dev_info(&pdev->dev, "Assigning random MAC address\n");
3040 eth_hw_addr_random(netdev);
3041 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3042 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3043 }
3044
3045 /* Enable dynamic interrupt throttling rates */
3046 adapter->rx_itr_setting = 1;
3047 adapter->tx_itr_setting = 1;
3048
3049 /* set default ring sizes */
3050 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3051 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3052
3053 set_bit(__IXGBEVF_DOWN, &adapter->state);
3054 return 0;
3055
3056out:
3057 return err;
3058}
3059
3060#define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3061 { \
3062 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3063 if (current_counter < last_counter) \
3064 counter += 0x100000000LL; \
3065 last_counter = current_counter; \
3066 counter &= 0xFFFFFFFF00000000LL; \
3067 counter |= current_counter; \
3068 }
3069
3070#define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3071 { \
3072 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3073 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3074 u64 current_counter = (current_counter_msb << 32) | \
3075 current_counter_lsb; \
3076 if (current_counter < last_counter) \
3077 counter += 0x1000000000LL; \
3078 last_counter = current_counter; \
3079 counter &= 0xFFFFFFF000000000LL; \
3080 counter |= current_counter; \
3081 }
3082/**
3083 * ixgbevf_update_stats - Update the board statistics counters.
3084 * @adapter: board private structure
3085 **/
3086void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3087{
3088 struct ixgbe_hw *hw = &adapter->hw;
3089 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3090 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3091 int i;
3092
3093 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3094 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3095 return;
3096
3097 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3098 adapter->stats.vfgprc);
3099 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3100 adapter->stats.vfgptc);
3101 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3102 adapter->stats.last_vfgorc,
3103 adapter->stats.vfgorc);
3104 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3105 adapter->stats.last_vfgotc,
3106 adapter->stats.vfgotc);
3107 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3108 adapter->stats.vfmprc);
3109
3110 for (i = 0; i < adapter->num_rx_queues; i++) {
3111 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3112
3113 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3114 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3115 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3116 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3117 }
3118
3119 adapter->hw_csum_rx_error = hw_csum_rx_error;
3120 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3121 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3122 adapter->alloc_rx_page = alloc_rx_page;
3123}
3124
3125/**
3126 * ixgbevf_service_timer - Timer Call-back
3127 * @t: pointer to timer_list struct
3128 **/
3129static void ixgbevf_service_timer(struct timer_list *t)
3130{
3131 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3132 service_timer);
3133
3134 /* Reset the timer */
3135 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3136
3137 ixgbevf_service_event_schedule(adapter);
3138}
3139
3140static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3141{
3142 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3143 return;
3144
3145 /* If we're already down or resetting, just bail */
3146 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3147 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3148 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3149 return;
3150
3151 adapter->tx_timeout_count++;
3152
3153 rtnl_lock();
3154 ixgbevf_reinit_locked(adapter);
3155 rtnl_unlock();
3156}
3157
3158/**
3159 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3160 * @adapter: pointer to the device adapter structure
3161 *
3162 * This function serves two purposes. First it strobes the interrupt lines
3163 * in order to make certain interrupts are occurring. Secondly it sets the
3164 * bits needed to check for TX hangs. As a result we should immediately
3165 * determine if a hang has occurred.
3166 **/
3167static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3168{
3169 struct ixgbe_hw *hw = &adapter->hw;
3170 u32 eics = 0;
3171 int i;
3172
3173 /* If we're down or resetting, just bail */
3174 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3175 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3176 return;
3177
3178 /* Force detection of hung controller */
3179 if (netif_carrier_ok(adapter->netdev)) {
3180 for (i = 0; i < adapter->num_tx_queues; i++)
3181 set_check_for_tx_hang(adapter->tx_ring[i]);
3182 for (i = 0; i < adapter->num_xdp_queues; i++)
3183 set_check_for_tx_hang(adapter->xdp_ring[i]);
3184 }
3185
3186 /* get one bit for every active Tx/Rx interrupt vector */
3187 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3188 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3189
3190 if (qv->rx.ring || qv->tx.ring)
3191 eics |= BIT(i);
3192 }
3193
3194 /* Cause software interrupt to ensure rings are cleaned */
3195 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3196}
3197
3198/**
3199 * ixgbevf_watchdog_update_link - update the link status
3200 * @adapter: pointer to the device adapter structure
3201 **/
3202static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3203{
3204 struct ixgbe_hw *hw = &adapter->hw;
3205 u32 link_speed = adapter->link_speed;
3206 bool link_up = adapter->link_up;
3207 s32 err;
3208
3209 spin_lock_bh(&adapter->mbx_lock);
3210
3211 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3212
3213 spin_unlock_bh(&adapter->mbx_lock);
3214
3215 /* if check for link returns error we will need to reset */
3216 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3217 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3218 link_up = false;
3219 }
3220
3221 adapter->link_up = link_up;
3222 adapter->link_speed = link_speed;
3223}
3224
3225/**
3226 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3227 * print link up message
3228 * @adapter: pointer to the device adapter structure
3229 **/
3230static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3231{
3232 struct net_device *netdev = adapter->netdev;
3233
3234 /* only continue if link was previously down */
3235 if (netif_carrier_ok(netdev))
3236 return;
3237
3238 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3239 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3240 "10 Gbps" :
3241 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3242 "1 Gbps" :
3243 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3244 "100 Mbps" :
3245 "unknown speed");
3246
3247 netif_carrier_on(netdev);
3248}
3249
3250/**
3251 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3252 * print link down message
3253 * @adapter: pointer to the adapter structure
3254 **/
3255static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3256{
3257 struct net_device *netdev = adapter->netdev;
3258
3259 adapter->link_speed = 0;
3260
3261 /* only continue if link was up previously */
3262 if (!netif_carrier_ok(netdev))
3263 return;
3264
3265 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3266
3267 netif_carrier_off(netdev);
3268}
3269
3270/**
3271 * ixgbevf_watchdog_subtask - worker thread to bring link up
3272 * @adapter: board private structure
3273 **/
3274static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3275{
3276 /* if interface is down do nothing */
3277 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3278 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3279 return;
3280
3281 ixgbevf_watchdog_update_link(adapter);
3282
3283 if (adapter->link_up)
3284 ixgbevf_watchdog_link_is_up(adapter);
3285 else
3286 ixgbevf_watchdog_link_is_down(adapter);
3287
3288 ixgbevf_update_stats(adapter);
3289}
3290
3291/**
3292 * ixgbevf_service_task - manages and runs subtasks
3293 * @work: pointer to work_struct containing our data
3294 **/
3295static void ixgbevf_service_task(struct work_struct *work)
3296{
3297 struct ixgbevf_adapter *adapter = container_of(work,
3298 struct ixgbevf_adapter,
3299 service_task);
3300 struct ixgbe_hw *hw = &adapter->hw;
3301
3302 if (IXGBE_REMOVED(hw->hw_addr)) {
3303 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3304 rtnl_lock();
3305 ixgbevf_down(adapter);
3306 rtnl_unlock();
3307 }
3308 return;
3309 }
3310
3311 ixgbevf_queue_reset_subtask(adapter);
3312 ixgbevf_reset_subtask(adapter);
3313 ixgbevf_watchdog_subtask(adapter);
3314 ixgbevf_check_hang_subtask(adapter);
3315
3316 ixgbevf_service_event_complete(adapter);
3317}
3318
3319/**
3320 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3321 * @tx_ring: Tx descriptor ring for a specific queue
3322 *
3323 * Free all transmit software resources
3324 **/
3325void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3326{
3327 ixgbevf_clean_tx_ring(tx_ring);
3328
3329 vfree(tx_ring->tx_buffer_info);
3330 tx_ring->tx_buffer_info = NULL;
3331
3332 /* if not set, then don't free */
3333 if (!tx_ring->desc)
3334 return;
3335
3336 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3337 tx_ring->dma);
3338
3339 tx_ring->desc = NULL;
3340}
3341
3342/**
3343 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3344 * @adapter: board private structure
3345 *
3346 * Free all transmit software resources
3347 **/
3348static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3349{
3350 int i;
3351
3352 for (i = 0; i < adapter->num_tx_queues; i++)
3353 if (adapter->tx_ring[i]->desc)
3354 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3355 for (i = 0; i < adapter->num_xdp_queues; i++)
3356 if (adapter->xdp_ring[i]->desc)
3357 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3358}
3359
3360/**
3361 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3362 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3363 *
3364 * Return 0 on success, negative on failure
3365 **/
3366int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3367{
3368 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3369 int size;
3370
3371 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3372 tx_ring->tx_buffer_info = vmalloc(size);
3373 if (!tx_ring->tx_buffer_info)
3374 goto err;
3375
3376 u64_stats_init(&tx_ring->syncp);
3377
3378 /* round up to nearest 4K */
3379 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3380 tx_ring->size = ALIGN(tx_ring->size, 4096);
3381
3382 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3383 &tx_ring->dma, GFP_KERNEL);
3384 if (!tx_ring->desc)
3385 goto err;
3386
3387 return 0;
3388
3389err:
3390 vfree(tx_ring->tx_buffer_info);
3391 tx_ring->tx_buffer_info = NULL;
3392 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3393 return -ENOMEM;
3394}
3395
3396/**
3397 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3398 * @adapter: board private structure
3399 *
3400 * If this function returns with an error, then it's possible one or
3401 * more of the rings is populated (while the rest are not). It is the
3402 * callers duty to clean those orphaned rings.
3403 *
3404 * Return 0 on success, negative on failure
3405 **/
3406static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3407{
3408 int i, j = 0, err = 0;
3409
3410 for (i = 0; i < adapter->num_tx_queues; i++) {
3411 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3412 if (!err)
3413 continue;
3414 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3415 goto err_setup_tx;
3416 }
3417
3418 for (j = 0; j < adapter->num_xdp_queues; j++) {
3419 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3420 if (!err)
3421 continue;
3422 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3423 goto err_setup_tx;
3424 }
3425
3426 return 0;
3427err_setup_tx:
3428 /* rewind the index freeing the rings as we go */
3429 while (j--)
3430 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3431 while (i--)
3432 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3433
3434 return err;
3435}
3436
3437/**
3438 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3439 * @adapter: board private structure
3440 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3441 *
3442 * Returns 0 on success, negative on failure
3443 **/
3444int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3445 struct ixgbevf_ring *rx_ring)
3446{
3447 int size;
3448
3449 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3450 rx_ring->rx_buffer_info = vmalloc(size);
3451 if (!rx_ring->rx_buffer_info)
3452 goto err;
3453
3454 u64_stats_init(&rx_ring->syncp);
3455
3456 /* Round up to nearest 4K */
3457 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3458 rx_ring->size = ALIGN(rx_ring->size, 4096);
3459
3460 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3461 &rx_ring->dma, GFP_KERNEL);
3462
3463 if (!rx_ring->desc)
3464 goto err;
3465
3466 /* XDP RX-queue info */
3467 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3468 rx_ring->queue_index) < 0)
3469 goto err;
3470
3471 rx_ring->xdp_prog = adapter->xdp_prog;
3472
3473 return 0;
3474err:
3475 vfree(rx_ring->rx_buffer_info);
3476 rx_ring->rx_buffer_info = NULL;
3477 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3478 return -ENOMEM;
3479}
3480
3481/**
3482 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3483 * @adapter: board private structure
3484 *
3485 * If this function returns with an error, then it's possible one or
3486 * more of the rings is populated (while the rest are not). It is the
3487 * callers duty to clean those orphaned rings.
3488 *
3489 * Return 0 on success, negative on failure
3490 **/
3491static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3492{
3493 int i, err = 0;
3494
3495 for (i = 0; i < adapter->num_rx_queues; i++) {
3496 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3497 if (!err)
3498 continue;
3499 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3500 goto err_setup_rx;
3501 }
3502
3503 return 0;
3504err_setup_rx:
3505 /* rewind the index freeing the rings as we go */
3506 while (i--)
3507 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3508 return err;
3509}
3510
3511/**
3512 * ixgbevf_free_rx_resources - Free Rx Resources
3513 * @rx_ring: ring to clean the resources from
3514 *
3515 * Free all receive software resources
3516 **/
3517void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3518{
3519 ixgbevf_clean_rx_ring(rx_ring);
3520
3521 rx_ring->xdp_prog = NULL;
3522 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3523 vfree(rx_ring->rx_buffer_info);
3524 rx_ring->rx_buffer_info = NULL;
3525
3526 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3527 rx_ring->dma);
3528
3529 rx_ring->desc = NULL;
3530}
3531
3532/**
3533 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3534 * @adapter: board private structure
3535 *
3536 * Free all receive software resources
3537 **/
3538static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3539{
3540 int i;
3541
3542 for (i = 0; i < adapter->num_rx_queues; i++)
3543 if (adapter->rx_ring[i]->desc)
3544 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3545}
3546
3547/**
3548 * ixgbevf_open - Called when a network interface is made active
3549 * @netdev: network interface device structure
3550 *
3551 * Returns 0 on success, negative value on failure
3552 *
3553 * The open entry point is called when a network interface is made
3554 * active by the system (IFF_UP). At this point all resources needed
3555 * for transmit and receive operations are allocated, the interrupt
3556 * handler is registered with the OS, the watchdog timer is started,
3557 * and the stack is notified that the interface is ready.
3558 **/
3559int ixgbevf_open(struct net_device *netdev)
3560{
3561 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3562 struct ixgbe_hw *hw = &adapter->hw;
3563 int err;
3564
3565 /* A previous failure to open the device because of a lack of
3566 * available MSIX vector resources may have reset the number
3567 * of msix vectors variable to zero. The only way to recover
3568 * is to unload/reload the driver and hope that the system has
3569 * been able to recover some MSIX vector resources.
3570 */
3571 if (!adapter->num_msix_vectors)
3572 return -ENOMEM;
3573
3574 if (hw->adapter_stopped) {
3575 ixgbevf_reset(adapter);
3576 /* if adapter is still stopped then PF isn't up and
3577 * the VF can't start.
3578 */
3579 if (hw->adapter_stopped) {
3580 err = IXGBE_ERR_MBX;
3581 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3582 goto err_setup_reset;
3583 }
3584 }
3585
3586 /* disallow open during test */
3587 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3588 return -EBUSY;
3589
3590 netif_carrier_off(netdev);
3591
3592 /* allocate transmit descriptors */
3593 err = ixgbevf_setup_all_tx_resources(adapter);
3594 if (err)
3595 goto err_setup_tx;
3596
3597 /* allocate receive descriptors */
3598 err = ixgbevf_setup_all_rx_resources(adapter);
3599 if (err)
3600 goto err_setup_rx;
3601
3602 ixgbevf_configure(adapter);
3603
3604 err = ixgbevf_request_irq(adapter);
3605 if (err)
3606 goto err_req_irq;
3607
3608 /* Notify the stack of the actual queue counts. */
3609 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3610 if (err)
3611 goto err_set_queues;
3612
3613 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3614 if (err)
3615 goto err_set_queues;
3616
3617 ixgbevf_up_complete(adapter);
3618
3619 return 0;
3620
3621err_set_queues:
3622 ixgbevf_free_irq(adapter);
3623err_req_irq:
3624 ixgbevf_free_all_rx_resources(adapter);
3625err_setup_rx:
3626 ixgbevf_free_all_tx_resources(adapter);
3627err_setup_tx:
3628 ixgbevf_reset(adapter);
3629err_setup_reset:
3630
3631 return err;
3632}
3633
3634/**
3635 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3636 * @adapter: the private adapter struct
3637 *
3638 * This function should contain the necessary work common to both suspending
3639 * and closing of the device.
3640 */
3641static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3642{
3643 ixgbevf_down(adapter);
3644 ixgbevf_free_irq(adapter);
3645 ixgbevf_free_all_tx_resources(adapter);
3646 ixgbevf_free_all_rx_resources(adapter);
3647}
3648
3649/**
3650 * ixgbevf_close - Disables a network interface
3651 * @netdev: network interface device structure
3652 *
3653 * Returns 0, this is not allowed to fail
3654 *
3655 * The close entry point is called when an interface is de-activated
3656 * by the OS. The hardware is still under the drivers control, but
3657 * needs to be disabled. A global MAC reset is issued to stop the
3658 * hardware, and all transmit and receive resources are freed.
3659 **/
3660int ixgbevf_close(struct net_device *netdev)
3661{
3662 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3663
3664 if (netif_device_present(netdev))
3665 ixgbevf_close_suspend(adapter);
3666
3667 return 0;
3668}
3669
3670static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3671{
3672 struct net_device *dev = adapter->netdev;
3673
3674 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3675 &adapter->state))
3676 return;
3677
3678 /* if interface is down do nothing */
3679 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3680 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3681 return;
3682
3683 /* Hardware has to reinitialize queues and interrupts to
3684 * match packet buffer alignment. Unfortunately, the
3685 * hardware is not flexible enough to do this dynamically.
3686 */
3687 rtnl_lock();
3688
3689 if (netif_running(dev))
3690 ixgbevf_close(dev);
3691
3692 ixgbevf_clear_interrupt_scheme(adapter);
3693 ixgbevf_init_interrupt_scheme(adapter);
3694
3695 if (netif_running(dev))
3696 ixgbevf_open(dev);
3697
3698 rtnl_unlock();
3699}
3700
3701static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3702 u32 vlan_macip_lens, u32 type_tucmd,
3703 u32 mss_l4len_idx)
3704{
3705 struct ixgbe_adv_tx_context_desc *context_desc;
3706 u16 i = tx_ring->next_to_use;
3707
3708 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3709
3710 i++;
3711 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3712
3713 /* set bits to identify this as an advanced context descriptor */
3714 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3715
3716 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3717 context_desc->seqnum_seed = 0;
3718 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3719 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3720}
3721
3722static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3723 struct ixgbevf_tx_buffer *first,
3724 u8 *hdr_len)
3725{
3726 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3727 struct sk_buff *skb = first->skb;
3728 union {
3729 struct iphdr *v4;
3730 struct ipv6hdr *v6;
3731 unsigned char *hdr;
3732 } ip;
3733 union {
3734 struct tcphdr *tcp;
3735 unsigned char *hdr;
3736 } l4;
3737 u32 paylen, l4_offset;
3738 int err;
3739
3740 if (skb->ip_summed != CHECKSUM_PARTIAL)
3741 return 0;
3742
3743 if (!skb_is_gso(skb))
3744 return 0;
3745
3746 err = skb_cow_head(skb, 0);
3747 if (err < 0)
3748 return err;
3749
3750 if (eth_p_mpls(first->protocol))
3751 ip.hdr = skb_inner_network_header(skb);
3752 else
3753 ip.hdr = skb_network_header(skb);
3754 l4.hdr = skb_checksum_start(skb);
3755
3756 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3757 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3758
3759 /* initialize outer IP header fields */
3760 if (ip.v4->version == 4) {
3761 unsigned char *csum_start = skb_checksum_start(skb);
3762 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3763
3764 /* IP header will have to cancel out any data that
3765 * is not a part of the outer IP header
3766 */
3767 ip.v4->check = csum_fold(csum_partial(trans_start,
3768 csum_start - trans_start,
3769 0));
3770 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3771
3772 ip.v4->tot_len = 0;
3773 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3774 IXGBE_TX_FLAGS_CSUM |
3775 IXGBE_TX_FLAGS_IPV4;
3776 } else {
3777 ip.v6->payload_len = 0;
3778 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3779 IXGBE_TX_FLAGS_CSUM;
3780 }
3781
3782 /* determine offset of inner transport header */
3783 l4_offset = l4.hdr - skb->data;
3784
3785 /* compute length of segmentation header */
3786 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3787
3788 /* remove payload length from inner checksum */
3789 paylen = skb->len - l4_offset;
3790 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
3791
3792 /* update gso size and bytecount with header size */
3793 first->gso_segs = skb_shinfo(skb)->gso_segs;
3794 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3795
3796 /* mss_l4len_id: use 1 as index for TSO */
3797 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3798 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3799 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3800
3801 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3802 vlan_macip_lens = l4.hdr - ip.hdr;
3803 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3804 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3805
3806 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3807 type_tucmd, mss_l4len_idx);
3808
3809 return 1;
3810}
3811
3812static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff *skb)
3813{
3814 unsigned int offset = 0;
3815
3816 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
3817
3818 return offset == skb_checksum_start_offset(skb);
3819}
3820
3821static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3822 struct ixgbevf_tx_buffer *first)
3823{
3824 struct sk_buff *skb = first->skb;
3825 u32 vlan_macip_lens = 0;
3826 u32 type_tucmd = 0;
3827
3828 if (skb->ip_summed != CHECKSUM_PARTIAL)
3829 goto no_csum;
3830
3831 switch (skb->csum_offset) {
3832 case offsetof(struct tcphdr, check):
3833 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3834 /* fall through */
3835 case offsetof(struct udphdr, check):
3836 break;
3837 case offsetof(struct sctphdr, checksum):
3838 /* validate that this is actually an SCTP request */
3839 if (((first->protocol == htons(ETH_P_IP)) &&
3840 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
3841 ((first->protocol == htons(ETH_P_IPV6)) &&
3842 ixgbevf_ipv6_csum_is_sctp(skb))) {
3843 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3844 break;
3845 }
3846 /* fall through */
3847 default:
3848 skb_checksum_help(skb);
3849 goto no_csum;
3850 }
3851 /* update TX checksum flag */
3852 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3853 vlan_macip_lens = skb_checksum_start_offset(skb) -
3854 skb_network_offset(skb);
3855no_csum:
3856 /* vlan_macip_lens: MACLEN, VLAN tag */
3857 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3858 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3859
3860 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
3861}
3862
3863static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3864{
3865 /* set type for advanced descriptor with frame checksum insertion */
3866 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3867 IXGBE_ADVTXD_DCMD_IFCS |
3868 IXGBE_ADVTXD_DCMD_DEXT);
3869
3870 /* set HW VLAN bit if VLAN is present */
3871 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3872 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3873
3874 /* set segmentation enable bits for TSO/FSO */
3875 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3876 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3877
3878 return cmd_type;
3879}
3880
3881static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3882 u32 tx_flags, unsigned int paylen)
3883{
3884 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3885
3886 /* enable L4 checksum for TSO and TX checksum offload */
3887 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3888 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3889
3890 /* enble IPv4 checksum for TSO */
3891 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3892 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3893
3894 /* use index 1 context for TSO/FSO/FCOE */
3895 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3896 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3897
3898 /* Check Context must be set if Tx switch is enabled, which it
3899 * always is for case where virtual functions are running
3900 */
3901 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3902
3903 tx_desc->read.olinfo_status = olinfo_status;
3904}
3905
3906static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3907 struct ixgbevf_tx_buffer *first,
3908 const u8 hdr_len)
3909{
3910 struct sk_buff *skb = first->skb;
3911 struct ixgbevf_tx_buffer *tx_buffer;
3912 union ixgbe_adv_tx_desc *tx_desc;
3913 struct skb_frag_struct *frag;
3914 dma_addr_t dma;
3915 unsigned int data_len, size;
3916 u32 tx_flags = first->tx_flags;
3917 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3918 u16 i = tx_ring->next_to_use;
3919
3920 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3921
3922 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3923
3924 size = skb_headlen(skb);
3925 data_len = skb->data_len;
3926
3927 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3928
3929 tx_buffer = first;
3930
3931 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3932 if (dma_mapping_error(tx_ring->dev, dma))
3933 goto dma_error;
3934
3935 /* record length, and DMA address */
3936 dma_unmap_len_set(tx_buffer, len, size);
3937 dma_unmap_addr_set(tx_buffer, dma, dma);
3938
3939 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3940
3941 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
3942 tx_desc->read.cmd_type_len =
3943 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
3944
3945 i++;
3946 tx_desc++;
3947 if (i == tx_ring->count) {
3948 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3949 i = 0;
3950 }
3951 tx_desc->read.olinfo_status = 0;
3952
3953 dma += IXGBE_MAX_DATA_PER_TXD;
3954 size -= IXGBE_MAX_DATA_PER_TXD;
3955
3956 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3957 }
3958
3959 if (likely(!data_len))
3960 break;
3961
3962 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
3963
3964 i++;
3965 tx_desc++;
3966 if (i == tx_ring->count) {
3967 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
3968 i = 0;
3969 }
3970 tx_desc->read.olinfo_status = 0;
3971
3972 size = skb_frag_size(frag);
3973 data_len -= size;
3974
3975 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
3976 DMA_TO_DEVICE);
3977
3978 tx_buffer = &tx_ring->tx_buffer_info[i];
3979 }
3980
3981 /* write last descriptor with RS and EOP bits */
3982 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
3983 tx_desc->read.cmd_type_len = cmd_type;
3984
3985 /* set the timestamp */
3986 first->time_stamp = jiffies;
3987
3988 /* Force memory writes to complete before letting h/w know there
3989 * are new descriptors to fetch. (Only applicable for weak-ordered
3990 * memory model archs, such as IA-64).
3991 *
3992 * We also need this memory barrier (wmb) to make certain all of the
3993 * status bits have been updated before next_to_watch is written.
3994 */
3995 wmb();
3996
3997 /* set next_to_watch value indicating a packet is present */
3998 first->next_to_watch = tx_desc;
3999
4000 i++;
4001 if (i == tx_ring->count)
4002 i = 0;
4003
4004 tx_ring->next_to_use = i;
4005
4006 /* notify HW of packet */
4007 ixgbevf_write_tail(tx_ring, i);
4008
4009 return;
4010dma_error:
4011 dev_err(tx_ring->dev, "TX DMA map failed\n");
4012 tx_buffer = &tx_ring->tx_buffer_info[i];
4013
4014 /* clear dma mappings for failed tx_buffer_info map */
4015 while (tx_buffer != first) {
4016 if (dma_unmap_len(tx_buffer, len))
4017 dma_unmap_page(tx_ring->dev,
4018 dma_unmap_addr(tx_buffer, dma),
4019 dma_unmap_len(tx_buffer, len),
4020 DMA_TO_DEVICE);
4021 dma_unmap_len_set(tx_buffer, len, 0);
4022
4023 if (i-- == 0)
4024 i += tx_ring->count;
4025 tx_buffer = &tx_ring->tx_buffer_info[i];
4026 }
4027
4028 if (dma_unmap_len(tx_buffer, len))
4029 dma_unmap_single(tx_ring->dev,
4030 dma_unmap_addr(tx_buffer, dma),
4031 dma_unmap_len(tx_buffer, len),
4032 DMA_TO_DEVICE);
4033 dma_unmap_len_set(tx_buffer, len, 0);
4034
4035 dev_kfree_skb_any(tx_buffer->skb);
4036 tx_buffer->skb = NULL;
4037
4038 tx_ring->next_to_use = i;
4039}
4040
4041static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4042{
4043 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4044 /* Herbert's original patch had:
4045 * smp_mb__after_netif_stop_queue();
4046 * but since that doesn't exist yet, just open code it.
4047 */
4048 smp_mb();
4049
4050 /* We need to check again in a case another CPU has just
4051 * made room available.
4052 */
4053 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4054 return -EBUSY;
4055
4056 /* A reprieve! - use start_queue because it doesn't call schedule */
4057 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4058 ++tx_ring->tx_stats.restart_queue;
4059
4060 return 0;
4061}
4062
4063static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4064{
4065 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4066 return 0;
4067 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4068}
4069
4070static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4071 struct ixgbevf_ring *tx_ring)
4072{
4073 struct ixgbevf_tx_buffer *first;
4074 int tso;
4075 u32 tx_flags = 0;
4076 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4077#if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4078 unsigned short f;
4079#endif
4080 u8 hdr_len = 0;
4081 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4082
4083 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4084 dev_kfree_skb_any(skb);
4085 return NETDEV_TX_OK;
4086 }
4087
4088 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4089 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4090 * + 2 desc gap to keep tail from touching head,
4091 * + 1 desc for context descriptor,
4092 * otherwise try next time
4093 */
4094#if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4095 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4096 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4097#else
4098 count += skb_shinfo(skb)->nr_frags;
4099#endif
4100 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4101 tx_ring->tx_stats.tx_busy++;
4102 return NETDEV_TX_BUSY;
4103 }
4104
4105 /* record the location of the first descriptor for this packet */
4106 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4107 first->skb = skb;
4108 first->bytecount = skb->len;
4109 first->gso_segs = 1;
4110
4111 if (skb_vlan_tag_present(skb)) {
4112 tx_flags |= skb_vlan_tag_get(skb);
4113 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4114 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4115 }
4116
4117 /* record initial flags and protocol */
4118 first->tx_flags = tx_flags;
4119 first->protocol = vlan_get_protocol(skb);
4120
4121 tso = ixgbevf_tso(tx_ring, first, &hdr_len);
4122 if (tso < 0)
4123 goto out_drop;
4124 else if (!tso)
4125 ixgbevf_tx_csum(tx_ring, first);
4126
4127 ixgbevf_tx_map(tx_ring, first, hdr_len);
4128
4129 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4130
4131 return NETDEV_TX_OK;
4132
4133out_drop:
4134 dev_kfree_skb_any(first->skb);
4135 first->skb = NULL;
4136
4137 return NETDEV_TX_OK;
4138}
4139
4140static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4141{
4142 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4143 struct ixgbevf_ring *tx_ring;
4144
4145 if (skb->len <= 0) {
4146 dev_kfree_skb_any(skb);
4147 return NETDEV_TX_OK;
4148 }
4149
4150 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4151 * in order to meet this minimum size requirement.
4152 */
4153 if (skb->len < 17) {
4154 if (skb_padto(skb, 17))
4155 return NETDEV_TX_OK;
4156 skb->len = 17;
4157 }
4158
4159 tx_ring = adapter->tx_ring[skb->queue_mapping];
4160 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4161}
4162
4163/**
4164 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4165 * @netdev: network interface device structure
4166 * @p: pointer to an address structure
4167 *
4168 * Returns 0 on success, negative on failure
4169 **/
4170static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4171{
4172 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4173 struct ixgbe_hw *hw = &adapter->hw;
4174 struct sockaddr *addr = p;
4175 int err;
4176
4177 if (!is_valid_ether_addr(addr->sa_data))
4178 return -EADDRNOTAVAIL;
4179
4180 spin_lock_bh(&adapter->mbx_lock);
4181
4182 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4183
4184 spin_unlock_bh(&adapter->mbx_lock);
4185
4186 if (err)
4187 return -EPERM;
4188
4189 ether_addr_copy(hw->mac.addr, addr->sa_data);
4190 ether_addr_copy(netdev->dev_addr, addr->sa_data);
4191
4192 return 0;
4193}
4194
4195/**
4196 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4197 * @netdev: network interface device structure
4198 * @new_mtu: new value for maximum frame size
4199 *
4200 * Returns 0 on success, negative on failure
4201 **/
4202static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4203{
4204 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4205 struct ixgbe_hw *hw = &adapter->hw;
4206 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4207 int ret;
4208
4209 /* prevent MTU being changed to a size unsupported by XDP */
4210 if (adapter->xdp_prog) {
4211 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4212 return -EPERM;
4213 }
4214
4215 spin_lock_bh(&adapter->mbx_lock);
4216 /* notify the PF of our intent to use this size of frame */
4217 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4218 spin_unlock_bh(&adapter->mbx_lock);
4219 if (ret)
4220 return -EINVAL;
4221
4222 hw_dbg(hw, "changing MTU from %d to %d\n",
4223 netdev->mtu, new_mtu);
4224
4225 /* must set new MTU before calling down or up */
4226 netdev->mtu = new_mtu;
4227
4228 if (netif_running(netdev))
4229 ixgbevf_reinit_locked(adapter);
4230
4231 return 0;
4232}
4233
4234#ifdef CONFIG_NET_POLL_CONTROLLER
4235/* Polling 'interrupt' - used by things like netconsole to send skbs
4236 * without having to re-enable interrupts. It's not called while
4237 * the interrupt routine is executing.
4238 */
4239static void ixgbevf_netpoll(struct net_device *netdev)
4240{
4241 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4242 int i;
4243
4244 /* if interface is down do nothing */
4245 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
4246 return;
4247 for (i = 0; i < adapter->num_rx_queues; i++)
4248 ixgbevf_msix_clean_rings(0, adapter->q_vector[i]);
4249}
4250#endif /* CONFIG_NET_POLL_CONTROLLER */
4251
4252static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
4253{
4254 struct net_device *netdev = pci_get_drvdata(pdev);
4255 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4256#ifdef CONFIG_PM
4257 int retval = 0;
4258#endif
4259
4260 rtnl_lock();
4261 netif_device_detach(netdev);
4262
4263 if (netif_running(netdev))
4264 ixgbevf_close_suspend(adapter);
4265
4266 ixgbevf_clear_interrupt_scheme(adapter);
4267 rtnl_unlock();
4268
4269#ifdef CONFIG_PM
4270 retval = pci_save_state(pdev);
4271 if (retval)
4272 return retval;
4273
4274#endif
4275 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4276 pci_disable_device(pdev);
4277
4278 return 0;
4279}
4280
4281#ifdef CONFIG_PM
4282static int ixgbevf_resume(struct pci_dev *pdev)
4283{
4284 struct net_device *netdev = pci_get_drvdata(pdev);
4285 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4286 u32 err;
4287
4288 pci_restore_state(pdev);
4289 /* pci_restore_state clears dev->state_saved so call
4290 * pci_save_state to restore it.
4291 */
4292 pci_save_state(pdev);
4293
4294 err = pci_enable_device_mem(pdev);
4295 if (err) {
4296 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
4297 return err;
4298 }
4299
4300 adapter->hw.hw_addr = adapter->io_addr;
4301 smp_mb__before_atomic();
4302 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4303 pci_set_master(pdev);
4304
4305 ixgbevf_reset(adapter);
4306
4307 rtnl_lock();
4308 err = ixgbevf_init_interrupt_scheme(adapter);
4309 if (!err && netif_running(netdev))
4310 err = ixgbevf_open(netdev);
4311 rtnl_unlock();
4312 if (err)
4313 return err;
4314
4315 netif_device_attach(netdev);
4316
4317 return err;
4318}
4319
4320#endif /* CONFIG_PM */
4321static void ixgbevf_shutdown(struct pci_dev *pdev)
4322{
4323 ixgbevf_suspend(pdev, PMSG_SUSPEND);
4324}
4325
4326static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4327 const struct ixgbevf_ring *ring)
4328{
4329 u64 bytes, packets;
4330 unsigned int start;
4331
4332 if (ring) {
4333 do {
4334 start = u64_stats_fetch_begin_irq(&ring->syncp);
4335 bytes = ring->stats.bytes;
4336 packets = ring->stats.packets;
4337 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4338 stats->tx_bytes += bytes;
4339 stats->tx_packets += packets;
4340 }
4341}
4342
4343static void ixgbevf_get_stats(struct net_device *netdev,
4344 struct rtnl_link_stats64 *stats)
4345{
4346 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4347 unsigned int start;
4348 u64 bytes, packets;
4349 const struct ixgbevf_ring *ring;
4350 int i;
4351
4352 ixgbevf_update_stats(adapter);
4353
4354 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4355
4356 rcu_read_lock();
4357 for (i = 0; i < adapter->num_rx_queues; i++) {
4358 ring = adapter->rx_ring[i];
4359 do {
4360 start = u64_stats_fetch_begin_irq(&ring->syncp);
4361 bytes = ring->stats.bytes;
4362 packets = ring->stats.packets;
4363 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
4364 stats->rx_bytes += bytes;
4365 stats->rx_packets += packets;
4366 }
4367
4368 for (i = 0; i < adapter->num_tx_queues; i++) {
4369 ring = adapter->tx_ring[i];
4370 ixgbevf_get_tx_ring_stats(stats, ring);
4371 }
4372
4373 for (i = 0; i < adapter->num_xdp_queues; i++) {
4374 ring = adapter->xdp_ring[i];
4375 ixgbevf_get_tx_ring_stats(stats, ring);
4376 }
4377 rcu_read_unlock();
4378}
4379
4380#define IXGBEVF_MAX_MAC_HDR_LEN 127
4381#define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4382
4383static netdev_features_t
4384ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4385 netdev_features_t features)
4386{
4387 unsigned int network_hdr_len, mac_hdr_len;
4388
4389 /* Make certain the headers can be described by a context descriptor */
4390 mac_hdr_len = skb_network_header(skb) - skb->data;
4391 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4392 return features & ~(NETIF_F_HW_CSUM |
4393 NETIF_F_SCTP_CRC |
4394 NETIF_F_HW_VLAN_CTAG_TX |
4395 NETIF_F_TSO |
4396 NETIF_F_TSO6);
4397
4398 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4399 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4400 return features & ~(NETIF_F_HW_CSUM |
4401 NETIF_F_SCTP_CRC |
4402 NETIF_F_TSO |
4403 NETIF_F_TSO6);
4404
4405 /* We can only support IPV4 TSO in tunnels if we can mangle the
4406 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4407 */
4408 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4409 features &= ~NETIF_F_TSO;
4410
4411 return features;
4412}
4413
4414static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4415{
4416 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4417 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4418 struct bpf_prog *old_prog;
4419
4420 /* verify ixgbevf ring attributes are sufficient for XDP */
4421 for (i = 0; i < adapter->num_rx_queues; i++) {
4422 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4423
4424 if (frame_size > ixgbevf_rx_bufsz(ring))
4425 return -EINVAL;
4426 }
4427
4428 old_prog = xchg(&adapter->xdp_prog, prog);
4429
4430 /* If transitioning XDP modes reconfigure rings */
4431 if (!!prog != !!old_prog) {
4432 /* Hardware has to reinitialize queues and interrupts to
4433 * match packet buffer alignment. Unfortunately, the
4434 * hardware is not flexible enough to do this dynamically.
4435 */
4436 if (netif_running(dev))
4437 ixgbevf_close(dev);
4438
4439 ixgbevf_clear_interrupt_scheme(adapter);
4440 ixgbevf_init_interrupt_scheme(adapter);
4441
4442 if (netif_running(dev))
4443 ixgbevf_open(dev);
4444 } else {
4445 for (i = 0; i < adapter->num_rx_queues; i++)
4446 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4447 }
4448
4449 if (old_prog)
4450 bpf_prog_put(old_prog);
4451
4452 return 0;
4453}
4454
4455static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4456{
4457 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4458
4459 switch (xdp->command) {
4460 case XDP_SETUP_PROG:
4461 return ixgbevf_xdp_setup(dev, xdp->prog);
4462 case XDP_QUERY_PROG:
4463 xdp->prog_attached = !!(adapter->xdp_prog);
4464 xdp->prog_id = adapter->xdp_prog ?
4465 adapter->xdp_prog->aux->id : 0;
4466 return 0;
4467 default:
4468 return -EINVAL;
4469 }
4470}
4471
4472static const struct net_device_ops ixgbevf_netdev_ops = {
4473 .ndo_open = ixgbevf_open,
4474 .ndo_stop = ixgbevf_close,
4475 .ndo_start_xmit = ixgbevf_xmit_frame,
4476 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4477 .ndo_get_stats64 = ixgbevf_get_stats,
4478 .ndo_validate_addr = eth_validate_addr,
4479 .ndo_set_mac_address = ixgbevf_set_mac,
4480 .ndo_change_mtu = ixgbevf_change_mtu,
4481 .ndo_tx_timeout = ixgbevf_tx_timeout,
4482 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4483 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4484#ifdef CONFIG_NET_POLL_CONTROLLER
4485 .ndo_poll_controller = ixgbevf_netpoll,
4486#endif
4487 .ndo_features_check = ixgbevf_features_check,
4488 .ndo_bpf = ixgbevf_xdp,
4489};
4490
4491static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4492{
4493 dev->netdev_ops = &ixgbevf_netdev_ops;
4494 ixgbevf_set_ethtool_ops(dev);
4495 dev->watchdog_timeo = 5 * HZ;
4496}
4497
4498/**
4499 * ixgbevf_probe - Device Initialization Routine
4500 * @pdev: PCI device information struct
4501 * @ent: entry in ixgbevf_pci_tbl
4502 *
4503 * Returns 0 on success, negative on failure
4504 *
4505 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4506 * The OS initialization, configuring of the adapter private structure,
4507 * and a hardware reset occur.
4508 **/
4509static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4510{
4511 struct net_device *netdev;
4512 struct ixgbevf_adapter *adapter = NULL;
4513 struct ixgbe_hw *hw = NULL;
4514 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4515 int err, pci_using_dac;
4516 bool disable_dev = false;
4517
4518 err = pci_enable_device(pdev);
4519 if (err)
4520 return err;
4521
4522 if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
4523 pci_using_dac = 1;
4524 } else {
4525 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
4526 if (err) {
4527 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4528 goto err_dma;
4529 }
4530 pci_using_dac = 0;
4531 }
4532
4533 err = pci_request_regions(pdev, ixgbevf_driver_name);
4534 if (err) {
4535 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4536 goto err_pci_reg;
4537 }
4538
4539 pci_set_master(pdev);
4540
4541 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4542 MAX_TX_QUEUES);
4543 if (!netdev) {
4544 err = -ENOMEM;
4545 goto err_alloc_etherdev;
4546 }
4547
4548 SET_NETDEV_DEV(netdev, &pdev->dev);
4549
4550 adapter = netdev_priv(netdev);
4551
4552 adapter->netdev = netdev;
4553 adapter->pdev = pdev;
4554 hw = &adapter->hw;
4555 hw->back = adapter;
4556 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4557
4558 /* call save state here in standalone driver because it relies on
4559 * adapter struct to exist, and needs to call netdev_priv
4560 */
4561 pci_save_state(pdev);
4562
4563 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4564 pci_resource_len(pdev, 0));
4565 adapter->io_addr = hw->hw_addr;
4566 if (!hw->hw_addr) {
4567 err = -EIO;
4568 goto err_ioremap;
4569 }
4570
4571 ixgbevf_assign_netdev_ops(netdev);
4572
4573 /* Setup HW API */
4574 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4575 hw->mac.type = ii->mac;
4576
4577 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
4578 sizeof(struct ixgbe_mbx_operations));
4579
4580 /* setup the private structure */
4581 err = ixgbevf_sw_init(adapter);
4582 if (err)
4583 goto err_sw_init;
4584
4585 /* The HW MAC address was set and/or determined in sw_init */
4586 if (!is_valid_ether_addr(netdev->dev_addr)) {
4587 pr_err("invalid MAC address\n");
4588 err = -EIO;
4589 goto err_sw_init;
4590 }
4591
4592 netdev->hw_features = NETIF_F_SG |
4593 NETIF_F_TSO |
4594 NETIF_F_TSO6 |
4595 NETIF_F_RXCSUM |
4596 NETIF_F_HW_CSUM |
4597 NETIF_F_SCTP_CRC;
4598
4599#define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4600 NETIF_F_GSO_GRE_CSUM | \
4601 NETIF_F_GSO_IPXIP4 | \
4602 NETIF_F_GSO_IPXIP6 | \
4603 NETIF_F_GSO_UDP_TUNNEL | \
4604 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4605
4606 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4607 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4608 IXGBEVF_GSO_PARTIAL_FEATURES;
4609
4610 netdev->features = netdev->hw_features;
4611
4612 if (pci_using_dac)
4613 netdev->features |= NETIF_F_HIGHDMA;
4614
4615 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4616 netdev->mpls_features |= NETIF_F_SG |
4617 NETIF_F_TSO |
4618 NETIF_F_TSO6 |
4619 NETIF_F_HW_CSUM;
4620 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4621 netdev->hw_enc_features |= netdev->vlan_features;
4622
4623 /* set this bit last since it cannot be part of vlan_features */
4624 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4625 NETIF_F_HW_VLAN_CTAG_RX |
4626 NETIF_F_HW_VLAN_CTAG_TX;
4627
4628 netdev->priv_flags |= IFF_UNICAST_FLT;
4629
4630 /* MTU range: 68 - 1504 or 9710 */
4631 netdev->min_mtu = ETH_MIN_MTU;
4632 switch (adapter->hw.api_version) {
4633 case ixgbe_mbox_api_11:
4634 case ixgbe_mbox_api_12:
4635 case ixgbe_mbox_api_13:
4636 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4637 (ETH_HLEN + ETH_FCS_LEN);
4638 break;
4639 default:
4640 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4641 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4642 (ETH_HLEN + ETH_FCS_LEN);
4643 else
4644 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4645 break;
4646 }
4647
4648 if (IXGBE_REMOVED(hw->hw_addr)) {
4649 err = -EIO;
4650 goto err_sw_init;
4651 }
4652
4653 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4654
4655 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4656 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4657 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4658
4659 err = ixgbevf_init_interrupt_scheme(adapter);
4660 if (err)
4661 goto err_sw_init;
4662
4663 strcpy(netdev->name, "eth%d");
4664
4665 err = register_netdev(netdev);
4666 if (err)
4667 goto err_register;
4668
4669 pci_set_drvdata(pdev, netdev);
4670 netif_carrier_off(netdev);
4671
4672 ixgbevf_init_last_counter_stats(adapter);
4673
4674 /* print the VF info */
4675 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4676 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4677
4678 switch (hw->mac.type) {
4679 case ixgbe_mac_X550_vf:
4680 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4681 break;
4682 case ixgbe_mac_X540_vf:
4683 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4684 break;
4685 case ixgbe_mac_82599_vf:
4686 default:
4687 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4688 break;
4689 }
4690
4691 return 0;
4692
4693err_register:
4694 ixgbevf_clear_interrupt_scheme(adapter);
4695err_sw_init:
4696 ixgbevf_reset_interrupt_capability(adapter);
4697 iounmap(adapter->io_addr);
4698 kfree(adapter->rss_key);
4699err_ioremap:
4700 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4701 free_netdev(netdev);
4702err_alloc_etherdev:
4703 pci_release_regions(pdev);
4704err_pci_reg:
4705err_dma:
4706 if (!adapter || disable_dev)
4707 pci_disable_device(pdev);
4708 return err;
4709}
4710
4711/**
4712 * ixgbevf_remove - Device Removal Routine
4713 * @pdev: PCI device information struct
4714 *
4715 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4716 * that it should release a PCI device. The could be caused by a
4717 * Hot-Plug event, or because the driver is going to be removed from
4718 * memory.
4719 **/
4720static void ixgbevf_remove(struct pci_dev *pdev)
4721{
4722 struct net_device *netdev = pci_get_drvdata(pdev);
4723 struct ixgbevf_adapter *adapter;
4724 bool disable_dev;
4725
4726 if (!netdev)
4727 return;
4728
4729 adapter = netdev_priv(netdev);
4730
4731 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4732 cancel_work_sync(&adapter->service_task);
4733
4734 if (netdev->reg_state == NETREG_REGISTERED)
4735 unregister_netdev(netdev);
4736
4737 ixgbevf_clear_interrupt_scheme(adapter);
4738 ixgbevf_reset_interrupt_capability(adapter);
4739
4740 iounmap(adapter->io_addr);
4741 pci_release_regions(pdev);
4742
4743 hw_dbg(&adapter->hw, "Remove complete\n");
4744
4745 kfree(adapter->rss_key);
4746 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4747 free_netdev(netdev);
4748
4749 if (disable_dev)
4750 pci_disable_device(pdev);
4751}
4752
4753/**
4754 * ixgbevf_io_error_detected - called when PCI error is detected
4755 * @pdev: Pointer to PCI device
4756 * @state: The current pci connection state
4757 *
4758 * This function is called after a PCI bus error affecting
4759 * this device has been detected.
4760 **/
4761static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4762 pci_channel_state_t state)
4763{
4764 struct net_device *netdev = pci_get_drvdata(pdev);
4765 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4766
4767 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4768 return PCI_ERS_RESULT_DISCONNECT;
4769
4770 rtnl_lock();
4771 netif_device_detach(netdev);
4772
4773 if (state == pci_channel_io_perm_failure) {
4774 rtnl_unlock();
4775 return PCI_ERS_RESULT_DISCONNECT;
4776 }
4777
4778 if (netif_running(netdev))
4779 ixgbevf_close_suspend(adapter);
4780
4781 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4782 pci_disable_device(pdev);
4783 rtnl_unlock();
4784
4785 /* Request a slot slot reset. */
4786 return PCI_ERS_RESULT_NEED_RESET;
4787}
4788
4789/**
4790 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4791 * @pdev: Pointer to PCI device
4792 *
4793 * Restart the card from scratch, as if from a cold-boot. Implementation
4794 * resembles the first-half of the ixgbevf_resume routine.
4795 **/
4796static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4797{
4798 struct net_device *netdev = pci_get_drvdata(pdev);
4799 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4800
4801 if (pci_enable_device_mem(pdev)) {
4802 dev_err(&pdev->dev,
4803 "Cannot re-enable PCI device after reset.\n");
4804 return PCI_ERS_RESULT_DISCONNECT;
4805 }
4806
4807 adapter->hw.hw_addr = adapter->io_addr;
4808 smp_mb__before_atomic();
4809 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4810 pci_set_master(pdev);
4811
4812 ixgbevf_reset(adapter);
4813
4814 return PCI_ERS_RESULT_RECOVERED;
4815}
4816
4817/**
4818 * ixgbevf_io_resume - called when traffic can start flowing again.
4819 * @pdev: Pointer to PCI device
4820 *
4821 * This callback is called when the error recovery driver tells us that
4822 * its OK to resume normal operation. Implementation resembles the
4823 * second-half of the ixgbevf_resume routine.
4824 **/
4825static void ixgbevf_io_resume(struct pci_dev *pdev)
4826{
4827 struct net_device *netdev = pci_get_drvdata(pdev);
4828
4829 rtnl_lock();
4830 if (netif_running(netdev))
4831 ixgbevf_open(netdev);
4832
4833 netif_device_attach(netdev);
4834 rtnl_unlock();
4835}
4836
4837/* PCI Error Recovery (ERS) */
4838static const struct pci_error_handlers ixgbevf_err_handler = {
4839 .error_detected = ixgbevf_io_error_detected,
4840 .slot_reset = ixgbevf_io_slot_reset,
4841 .resume = ixgbevf_io_resume,
4842};
4843
4844static struct pci_driver ixgbevf_driver = {
4845 .name = ixgbevf_driver_name,
4846 .id_table = ixgbevf_pci_tbl,
4847 .probe = ixgbevf_probe,
4848 .remove = ixgbevf_remove,
4849#ifdef CONFIG_PM
4850 /* Power Management Hooks */
4851 .suspend = ixgbevf_suspend,
4852 .resume = ixgbevf_resume,
4853#endif
4854 .shutdown = ixgbevf_shutdown,
4855 .err_handler = &ixgbevf_err_handler
4856};
4857
4858/**
4859 * ixgbevf_init_module - Driver Registration Routine
4860 *
4861 * ixgbevf_init_module is the first routine called when the driver is
4862 * loaded. All it does is register with the PCI subsystem.
4863 **/
4864static int __init ixgbevf_init_module(void)
4865{
4866 pr_info("%s - version %s\n", ixgbevf_driver_string,
4867 ixgbevf_driver_version);
4868
4869 pr_info("%s\n", ixgbevf_copyright);
4870 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4871 if (!ixgbevf_wq) {
4872 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4873 return -ENOMEM;
4874 }
4875
4876 return pci_register_driver(&ixgbevf_driver);
4877}
4878
4879module_init(ixgbevf_init_module);
4880
4881/**
4882 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4883 *
4884 * ixgbevf_exit_module is called just before the driver is removed
4885 * from memory.
4886 **/
4887static void __exit ixgbevf_exit_module(void)
4888{
4889 pci_unregister_driver(&ixgbevf_driver);
4890 if (ixgbevf_wq) {
4891 destroy_workqueue(ixgbevf_wq);
4892 ixgbevf_wq = NULL;
4893 }
4894}
4895
4896#ifdef DEBUG
4897/**
4898 * ixgbevf_get_hw_dev_name - return device name string
4899 * used by hardware layer to print debugging information
4900 * @hw: pointer to private hardware struct
4901 **/
4902char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4903{
4904 struct ixgbevf_adapter *adapter = hw->back;
4905
4906 return adapter->netdev->name;
4907}
4908
4909#endif
4910module_exit(ixgbevf_exit_module);
4911
4912/* ixgbevf_main.c */
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright(c) 1999 - 2018 Intel Corporation. */
3
4/******************************************************************************
5 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
6******************************************************************************/
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10#include <linux/types.h>
11#include <linux/bitops.h>
12#include <linux/module.h>
13#include <linux/pci.h>
14#include <linux/netdevice.h>
15#include <linux/vmalloc.h>
16#include <linux/string.h>
17#include <linux/in.h>
18#include <linux/ip.h>
19#include <linux/tcp.h>
20#include <linux/sctp.h>
21#include <linux/ipv6.h>
22#include <linux/slab.h>
23#include <net/checksum.h>
24#include <net/ip6_checksum.h>
25#include <linux/ethtool.h>
26#include <linux/if.h>
27#include <linux/if_vlan.h>
28#include <linux/prefetch.h>
29#include <net/mpls.h>
30#include <linux/bpf.h>
31#include <linux/bpf_trace.h>
32#include <linux/atomic.h>
33#include <net/xfrm.h>
34
35#include "ixgbevf.h"
36
37const char ixgbevf_driver_name[] = "ixgbevf";
38static const char ixgbevf_driver_string[] =
39 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
40
41static char ixgbevf_copyright[] =
42 "Copyright (c) 2009 - 2018 Intel Corporation.";
43
44static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
45 [board_82599_vf] = &ixgbevf_82599_vf_info,
46 [board_82599_vf_hv] = &ixgbevf_82599_vf_hv_info,
47 [board_X540_vf] = &ixgbevf_X540_vf_info,
48 [board_X540_vf_hv] = &ixgbevf_X540_vf_hv_info,
49 [board_X550_vf] = &ixgbevf_X550_vf_info,
50 [board_X550_vf_hv] = &ixgbevf_X550_vf_hv_info,
51 [board_X550EM_x_vf] = &ixgbevf_X550EM_x_vf_info,
52 [board_X550EM_x_vf_hv] = &ixgbevf_X550EM_x_vf_hv_info,
53 [board_x550em_a_vf] = &ixgbevf_x550em_a_vf_info,
54};
55
56/* ixgbevf_pci_tbl - PCI Device ID Table
57 *
58 * Wildcard entries (PCI_ANY_ID) should come last
59 * Last entry must be all 0s
60 *
61 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
62 * Class, Class Mask, private data (not used) }
63 */
64static const struct pci_device_id ixgbevf_pci_tbl[] = {
65 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
66 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF_HV), board_82599_vf_hv },
67 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
68 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF_HV), board_X540_vf_hv },
69 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF), board_X550_vf },
70 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550_VF_HV), board_X550_vf_hv },
71 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF), board_X550EM_x_vf },
72 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_VF_HV), board_X550EM_x_vf_hv},
73 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_A_VF), board_x550em_a_vf },
74 /* required last entry */
75 {0, }
76};
77MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
78
79MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
80MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
81MODULE_LICENSE("GPL v2");
82
83#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
84static int debug = -1;
85module_param(debug, int, 0);
86MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
87
88static struct workqueue_struct *ixgbevf_wq;
89
90static void ixgbevf_service_event_schedule(struct ixgbevf_adapter *adapter)
91{
92 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
93 !test_bit(__IXGBEVF_REMOVING, &adapter->state) &&
94 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state))
95 queue_work(ixgbevf_wq, &adapter->service_task);
96}
97
98static void ixgbevf_service_event_complete(struct ixgbevf_adapter *adapter)
99{
100 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state));
101
102 /* flush memory to make sure state is correct before next watchdog */
103 smp_mb__before_atomic();
104 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
105}
106
107/* forward decls */
108static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter);
109static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
110static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
111static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer);
112static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
113 struct ixgbevf_rx_buffer *old_buff);
114
115static void ixgbevf_remove_adapter(struct ixgbe_hw *hw)
116{
117 struct ixgbevf_adapter *adapter = hw->back;
118
119 if (!hw->hw_addr)
120 return;
121 hw->hw_addr = NULL;
122 dev_err(&adapter->pdev->dev, "Adapter removed\n");
123 if (test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
124 ixgbevf_service_event_schedule(adapter);
125}
126
127static void ixgbevf_check_remove(struct ixgbe_hw *hw, u32 reg)
128{
129 u32 value;
130
131 /* The following check not only optimizes a bit by not
132 * performing a read on the status register when the
133 * register just read was a status register read that
134 * returned IXGBE_FAILED_READ_REG. It also blocks any
135 * potential recursion.
136 */
137 if (reg == IXGBE_VFSTATUS) {
138 ixgbevf_remove_adapter(hw);
139 return;
140 }
141 value = ixgbevf_read_reg(hw, IXGBE_VFSTATUS);
142 if (value == IXGBE_FAILED_READ_REG)
143 ixgbevf_remove_adapter(hw);
144}
145
146u32 ixgbevf_read_reg(struct ixgbe_hw *hw, u32 reg)
147{
148 u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
149 u32 value;
150
151 if (IXGBE_REMOVED(reg_addr))
152 return IXGBE_FAILED_READ_REG;
153 value = readl(reg_addr + reg);
154 if (unlikely(value == IXGBE_FAILED_READ_REG))
155 ixgbevf_check_remove(hw, reg);
156 return value;
157}
158
159/**
160 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
161 * @adapter: pointer to adapter struct
162 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
163 * @queue: queue to map the corresponding interrupt to
164 * @msix_vector: the vector to map to the corresponding queue
165 **/
166static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
167 u8 queue, u8 msix_vector)
168{
169 u32 ivar, index;
170 struct ixgbe_hw *hw = &adapter->hw;
171
172 if (direction == -1) {
173 /* other causes */
174 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
175 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
176 ivar &= ~0xFF;
177 ivar |= msix_vector;
178 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
179 } else {
180 /* Tx or Rx causes */
181 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
182 index = ((16 * (queue & 1)) + (8 * direction));
183 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
184 ivar &= ~(0xFF << index);
185 ivar |= (msix_vector << index);
186 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
187 }
188}
189
190static u64 ixgbevf_get_tx_completed(struct ixgbevf_ring *ring)
191{
192 return ring->stats.packets;
193}
194
195static u32 ixgbevf_get_tx_pending(struct ixgbevf_ring *ring)
196{
197 struct ixgbevf_adapter *adapter = netdev_priv(ring->netdev);
198 struct ixgbe_hw *hw = &adapter->hw;
199
200 u32 head = IXGBE_READ_REG(hw, IXGBE_VFTDH(ring->reg_idx));
201 u32 tail = IXGBE_READ_REG(hw, IXGBE_VFTDT(ring->reg_idx));
202
203 if (head != tail)
204 return (head < tail) ?
205 tail - head : (tail + ring->count - head);
206
207 return 0;
208}
209
210static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring *tx_ring)
211{
212 u32 tx_done = ixgbevf_get_tx_completed(tx_ring);
213 u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
214 u32 tx_pending = ixgbevf_get_tx_pending(tx_ring);
215
216 clear_check_for_tx_hang(tx_ring);
217
218 /* Check for a hung queue, but be thorough. This verifies
219 * that a transmit has been completed since the previous
220 * check AND there is at least one packet pending. The
221 * ARMED bit is set to indicate a potential hang.
222 */
223 if ((tx_done_old == tx_done) && tx_pending) {
224 /* make sure it is true for two checks in a row */
225 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED,
226 &tx_ring->state);
227 }
228 /* reset the countdown */
229 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &tx_ring->state);
230
231 /* update completed stats and continue */
232 tx_ring->tx_stats.tx_done_old = tx_done;
233
234 return false;
235}
236
237static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter *adapter)
238{
239 /* Do the reset outside of interrupt context */
240 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
241 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
242 ixgbevf_service_event_schedule(adapter);
243 }
244}
245
246/**
247 * ixgbevf_tx_timeout - Respond to a Tx Hang
248 * @netdev: network interface device structure
249 * @txqueue: transmit queue hanging (unused)
250 **/
251static void ixgbevf_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
252{
253 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
254
255 ixgbevf_tx_timeout_reset(adapter);
256}
257
258/**
259 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
260 * @q_vector: board private structure
261 * @tx_ring: tx ring to clean
262 * @napi_budget: Used to determine if we are in netpoll
263 **/
264static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
265 struct ixgbevf_ring *tx_ring, int napi_budget)
266{
267 struct ixgbevf_adapter *adapter = q_vector->adapter;
268 struct ixgbevf_tx_buffer *tx_buffer;
269 union ixgbe_adv_tx_desc *tx_desc;
270 unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
271 unsigned int budget = tx_ring->count / 2;
272 unsigned int i = tx_ring->next_to_clean;
273
274 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
275 return true;
276
277 tx_buffer = &tx_ring->tx_buffer_info[i];
278 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
279 i -= tx_ring->count;
280
281 do {
282 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
283
284 /* if next_to_watch is not set then there is no work pending */
285 if (!eop_desc)
286 break;
287
288 /* prevent any other reads prior to eop_desc */
289 smp_rmb();
290
291 /* if DD is not set pending work has not been completed */
292 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
293 break;
294
295 /* clear next_to_watch to prevent false hangs */
296 tx_buffer->next_to_watch = NULL;
297
298 /* update the statistics for this packet */
299 total_bytes += tx_buffer->bytecount;
300 total_packets += tx_buffer->gso_segs;
301 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
302 total_ipsec++;
303
304 /* free the skb */
305 if (ring_is_xdp(tx_ring))
306 page_frag_free(tx_buffer->data);
307 else
308 napi_consume_skb(tx_buffer->skb, napi_budget);
309
310 /* unmap skb header data */
311 dma_unmap_single(tx_ring->dev,
312 dma_unmap_addr(tx_buffer, dma),
313 dma_unmap_len(tx_buffer, len),
314 DMA_TO_DEVICE);
315
316 /* clear tx_buffer data */
317 dma_unmap_len_set(tx_buffer, len, 0);
318
319 /* unmap remaining buffers */
320 while (tx_desc != eop_desc) {
321 tx_buffer++;
322 tx_desc++;
323 i++;
324 if (unlikely(!i)) {
325 i -= tx_ring->count;
326 tx_buffer = tx_ring->tx_buffer_info;
327 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
328 }
329
330 /* unmap any remaining paged data */
331 if (dma_unmap_len(tx_buffer, len)) {
332 dma_unmap_page(tx_ring->dev,
333 dma_unmap_addr(tx_buffer, dma),
334 dma_unmap_len(tx_buffer, len),
335 DMA_TO_DEVICE);
336 dma_unmap_len_set(tx_buffer, len, 0);
337 }
338 }
339
340 /* move us one more past the eop_desc for start of next pkt */
341 tx_buffer++;
342 tx_desc++;
343 i++;
344 if (unlikely(!i)) {
345 i -= tx_ring->count;
346 tx_buffer = tx_ring->tx_buffer_info;
347 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
348 }
349
350 /* issue prefetch for next Tx descriptor */
351 prefetch(tx_desc);
352
353 /* update budget accounting */
354 budget--;
355 } while (likely(budget));
356
357 i += tx_ring->count;
358 tx_ring->next_to_clean = i;
359 u64_stats_update_begin(&tx_ring->syncp);
360 tx_ring->stats.bytes += total_bytes;
361 tx_ring->stats.packets += total_packets;
362 u64_stats_update_end(&tx_ring->syncp);
363 q_vector->tx.total_bytes += total_bytes;
364 q_vector->tx.total_packets += total_packets;
365 adapter->tx_ipsec += total_ipsec;
366
367 if (check_for_tx_hang(tx_ring) && ixgbevf_check_tx_hang(tx_ring)) {
368 struct ixgbe_hw *hw = &adapter->hw;
369 union ixgbe_adv_tx_desc *eop_desc;
370
371 eop_desc = tx_ring->tx_buffer_info[i].next_to_watch;
372
373 pr_err("Detected Tx Unit Hang%s\n"
374 " Tx Queue <%d>\n"
375 " TDH, TDT <%x>, <%x>\n"
376 " next_to_use <%x>\n"
377 " next_to_clean <%x>\n"
378 "tx_buffer_info[next_to_clean]\n"
379 " next_to_watch <%p>\n"
380 " eop_desc->wb.status <%x>\n"
381 " time_stamp <%lx>\n"
382 " jiffies <%lx>\n",
383 ring_is_xdp(tx_ring) ? " XDP" : "",
384 tx_ring->queue_index,
385 IXGBE_READ_REG(hw, IXGBE_VFTDH(tx_ring->reg_idx)),
386 IXGBE_READ_REG(hw, IXGBE_VFTDT(tx_ring->reg_idx)),
387 tx_ring->next_to_use, i,
388 eop_desc, (eop_desc ? eop_desc->wb.status : 0),
389 tx_ring->tx_buffer_info[i].time_stamp, jiffies);
390
391 if (!ring_is_xdp(tx_ring))
392 netif_stop_subqueue(tx_ring->netdev,
393 tx_ring->queue_index);
394
395 /* schedule immediate reset if we believe we hung */
396 ixgbevf_tx_timeout_reset(adapter);
397
398 return true;
399 }
400
401 if (ring_is_xdp(tx_ring))
402 return !!budget;
403
404#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
405 if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
406 (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
407 /* Make sure that anybody stopping the queue after this
408 * sees the new next_to_clean.
409 */
410 smp_mb();
411
412 if (__netif_subqueue_stopped(tx_ring->netdev,
413 tx_ring->queue_index) &&
414 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
415 netif_wake_subqueue(tx_ring->netdev,
416 tx_ring->queue_index);
417 ++tx_ring->tx_stats.restart_queue;
418 }
419 }
420
421 return !!budget;
422}
423
424/**
425 * ixgbevf_rx_skb - Helper function to determine proper Rx method
426 * @q_vector: structure containing interrupt and ring information
427 * @skb: packet to send up
428 **/
429static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
430 struct sk_buff *skb)
431{
432 napi_gro_receive(&q_vector->napi, skb);
433}
434
435#define IXGBE_RSS_L4_TYPES_MASK \
436 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
437 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
438 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
439 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
440
441static inline void ixgbevf_rx_hash(struct ixgbevf_ring *ring,
442 union ixgbe_adv_rx_desc *rx_desc,
443 struct sk_buff *skb)
444{
445 u16 rss_type;
446
447 if (!(ring->netdev->features & NETIF_F_RXHASH))
448 return;
449
450 rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
451 IXGBE_RXDADV_RSSTYPE_MASK;
452
453 if (!rss_type)
454 return;
455
456 skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
457 (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
458 PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
459}
460
461/**
462 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
463 * @ring: structure containig ring specific data
464 * @rx_desc: current Rx descriptor being processed
465 * @skb: skb currently being received and modified
466 **/
467static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
468 union ixgbe_adv_rx_desc *rx_desc,
469 struct sk_buff *skb)
470{
471 skb_checksum_none_assert(skb);
472
473 /* Rx csum disabled */
474 if (!(ring->netdev->features & NETIF_F_RXCSUM))
475 return;
476
477 /* if IP and error */
478 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
479 ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
480 ring->rx_stats.csum_err++;
481 return;
482 }
483
484 if (!ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
485 return;
486
487 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
488 ring->rx_stats.csum_err++;
489 return;
490 }
491
492 /* It must be a TCP or UDP packet with a valid checksum */
493 skb->ip_summed = CHECKSUM_UNNECESSARY;
494}
495
496/**
497 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
498 * @rx_ring: rx descriptor ring packet is being transacted on
499 * @rx_desc: pointer to the EOP Rx descriptor
500 * @skb: pointer to current skb being populated
501 *
502 * This function checks the ring, descriptor, and packet information in
503 * order to populate the checksum, VLAN, protocol, and other fields within
504 * the skb.
505 **/
506static void ixgbevf_process_skb_fields(struct ixgbevf_ring *rx_ring,
507 union ixgbe_adv_rx_desc *rx_desc,
508 struct sk_buff *skb)
509{
510 ixgbevf_rx_hash(rx_ring, rx_desc, skb);
511 ixgbevf_rx_checksum(rx_ring, rx_desc, skb);
512
513 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
514 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
515 unsigned long *active_vlans = netdev_priv(rx_ring->netdev);
516
517 if (test_bit(vid & VLAN_VID_MASK, active_vlans))
518 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
519 }
520
521 if (ixgbevf_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
522 ixgbevf_ipsec_rx(rx_ring, rx_desc, skb);
523
524 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
525}
526
527static
528struct ixgbevf_rx_buffer *ixgbevf_get_rx_buffer(struct ixgbevf_ring *rx_ring,
529 const unsigned int size)
530{
531 struct ixgbevf_rx_buffer *rx_buffer;
532
533 rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
534 prefetchw(rx_buffer->page);
535
536 /* we are reusing so sync this buffer for CPU use */
537 dma_sync_single_range_for_cpu(rx_ring->dev,
538 rx_buffer->dma,
539 rx_buffer->page_offset,
540 size,
541 DMA_FROM_DEVICE);
542
543 rx_buffer->pagecnt_bias--;
544
545 return rx_buffer;
546}
547
548static void ixgbevf_put_rx_buffer(struct ixgbevf_ring *rx_ring,
549 struct ixgbevf_rx_buffer *rx_buffer,
550 struct sk_buff *skb)
551{
552 if (ixgbevf_can_reuse_rx_page(rx_buffer)) {
553 /* hand second half of page back to the ring */
554 ixgbevf_reuse_rx_page(rx_ring, rx_buffer);
555 } else {
556 if (IS_ERR(skb))
557 /* We are not reusing the buffer so unmap it and free
558 * any references we are holding to it
559 */
560 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
561 ixgbevf_rx_pg_size(rx_ring),
562 DMA_FROM_DEVICE,
563 IXGBEVF_RX_DMA_ATTR);
564 __page_frag_cache_drain(rx_buffer->page,
565 rx_buffer->pagecnt_bias);
566 }
567
568 /* clear contents of rx_buffer */
569 rx_buffer->page = NULL;
570}
571
572/**
573 * ixgbevf_is_non_eop - process handling of non-EOP buffers
574 * @rx_ring: Rx ring being processed
575 * @rx_desc: Rx descriptor for current buffer
576 *
577 * This function updates next to clean. If the buffer is an EOP buffer
578 * this function exits returning false, otherwise it will place the
579 * sk_buff in the next buffer to be chained and return true indicating
580 * that this is in fact a non-EOP buffer.
581 **/
582static bool ixgbevf_is_non_eop(struct ixgbevf_ring *rx_ring,
583 union ixgbe_adv_rx_desc *rx_desc)
584{
585 u32 ntc = rx_ring->next_to_clean + 1;
586
587 /* fetch, update, and store next to clean */
588 ntc = (ntc < rx_ring->count) ? ntc : 0;
589 rx_ring->next_to_clean = ntc;
590
591 prefetch(IXGBEVF_RX_DESC(rx_ring, ntc));
592
593 if (likely(ixgbevf_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
594 return false;
595
596 return true;
597}
598
599static inline unsigned int ixgbevf_rx_offset(struct ixgbevf_ring *rx_ring)
600{
601 return ring_uses_build_skb(rx_ring) ? IXGBEVF_SKB_PAD : 0;
602}
603
604static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring *rx_ring,
605 struct ixgbevf_rx_buffer *bi)
606{
607 struct page *page = bi->page;
608 dma_addr_t dma;
609
610 /* since we are recycling buffers we should seldom need to alloc */
611 if (likely(page))
612 return true;
613
614 /* alloc new page for storage */
615 page = dev_alloc_pages(ixgbevf_rx_pg_order(rx_ring));
616 if (unlikely(!page)) {
617 rx_ring->rx_stats.alloc_rx_page_failed++;
618 return false;
619 }
620
621 /* map page for use */
622 dma = dma_map_page_attrs(rx_ring->dev, page, 0,
623 ixgbevf_rx_pg_size(rx_ring),
624 DMA_FROM_DEVICE, IXGBEVF_RX_DMA_ATTR);
625
626 /* if mapping failed free memory back to system since
627 * there isn't much point in holding memory we can't use
628 */
629 if (dma_mapping_error(rx_ring->dev, dma)) {
630 __free_pages(page, ixgbevf_rx_pg_order(rx_ring));
631
632 rx_ring->rx_stats.alloc_rx_page_failed++;
633 return false;
634 }
635
636 bi->dma = dma;
637 bi->page = page;
638 bi->page_offset = ixgbevf_rx_offset(rx_ring);
639 bi->pagecnt_bias = 1;
640 rx_ring->rx_stats.alloc_rx_page++;
641
642 return true;
643}
644
645/**
646 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
647 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
648 * @cleaned_count: number of buffers to replace
649 **/
650static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring *rx_ring,
651 u16 cleaned_count)
652{
653 union ixgbe_adv_rx_desc *rx_desc;
654 struct ixgbevf_rx_buffer *bi;
655 unsigned int i = rx_ring->next_to_use;
656
657 /* nothing to do or no valid netdev defined */
658 if (!cleaned_count || !rx_ring->netdev)
659 return;
660
661 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
662 bi = &rx_ring->rx_buffer_info[i];
663 i -= rx_ring->count;
664
665 do {
666 if (!ixgbevf_alloc_mapped_page(rx_ring, bi))
667 break;
668
669 /* sync the buffer for use by the device */
670 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
671 bi->page_offset,
672 ixgbevf_rx_bufsz(rx_ring),
673 DMA_FROM_DEVICE);
674
675 /* Refresh the desc even if pkt_addr didn't change
676 * because each write-back erases this info.
677 */
678 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
679
680 rx_desc++;
681 bi++;
682 i++;
683 if (unlikely(!i)) {
684 rx_desc = IXGBEVF_RX_DESC(rx_ring, 0);
685 bi = rx_ring->rx_buffer_info;
686 i -= rx_ring->count;
687 }
688
689 /* clear the length for the next_to_use descriptor */
690 rx_desc->wb.upper.length = 0;
691
692 cleaned_count--;
693 } while (cleaned_count);
694
695 i += rx_ring->count;
696
697 if (rx_ring->next_to_use != i) {
698 /* record the next descriptor to use */
699 rx_ring->next_to_use = i;
700
701 /* update next to alloc since we have filled the ring */
702 rx_ring->next_to_alloc = i;
703
704 /* Force memory writes to complete before letting h/w
705 * know there are new descriptors to fetch. (Only
706 * applicable for weak-ordered memory model archs,
707 * such as IA-64).
708 */
709 wmb();
710 ixgbevf_write_tail(rx_ring, i);
711 }
712}
713
714/**
715 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
716 * @rx_ring: rx descriptor ring packet is being transacted on
717 * @rx_desc: pointer to the EOP Rx descriptor
718 * @skb: pointer to current skb being fixed
719 *
720 * Check for corrupted packet headers caused by senders on the local L2
721 * embedded NIC switch not setting up their Tx Descriptors right. These
722 * should be very rare.
723 *
724 * Also address the case where we are pulling data in on pages only
725 * and as such no data is present in the skb header.
726 *
727 * In addition if skb is not at least 60 bytes we need to pad it so that
728 * it is large enough to qualify as a valid Ethernet frame.
729 *
730 * Returns true if an error was encountered and skb was freed.
731 **/
732static bool ixgbevf_cleanup_headers(struct ixgbevf_ring *rx_ring,
733 union ixgbe_adv_rx_desc *rx_desc,
734 struct sk_buff *skb)
735{
736 /* XDP packets use error pointer so abort at this point */
737 if (IS_ERR(skb))
738 return true;
739
740 /* verify that the packet does not have any known errors */
741 if (unlikely(ixgbevf_test_staterr(rx_desc,
742 IXGBE_RXDADV_ERR_FRAME_ERR_MASK))) {
743 struct net_device *netdev = rx_ring->netdev;
744
745 if (!(netdev->features & NETIF_F_RXALL)) {
746 dev_kfree_skb_any(skb);
747 return true;
748 }
749 }
750
751 /* if eth_skb_pad returns an error the skb was freed */
752 if (eth_skb_pad(skb))
753 return true;
754
755 return false;
756}
757
758/**
759 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
760 * @rx_ring: rx descriptor ring to store buffers on
761 * @old_buff: donor buffer to have page reused
762 *
763 * Synchronizes page for reuse by the adapter
764 **/
765static void ixgbevf_reuse_rx_page(struct ixgbevf_ring *rx_ring,
766 struct ixgbevf_rx_buffer *old_buff)
767{
768 struct ixgbevf_rx_buffer *new_buff;
769 u16 nta = rx_ring->next_to_alloc;
770
771 new_buff = &rx_ring->rx_buffer_info[nta];
772
773 /* update, and store next to alloc */
774 nta++;
775 rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
776
777 /* transfer page from old buffer to new buffer */
778 new_buff->page = old_buff->page;
779 new_buff->dma = old_buff->dma;
780 new_buff->page_offset = old_buff->page_offset;
781 new_buff->pagecnt_bias = old_buff->pagecnt_bias;
782}
783
784static bool ixgbevf_can_reuse_rx_page(struct ixgbevf_rx_buffer *rx_buffer)
785{
786 unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
787 struct page *page = rx_buffer->page;
788
789 /* avoid re-using remote and pfmemalloc pages */
790 if (!dev_page_is_reusable(page))
791 return false;
792
793#if (PAGE_SIZE < 8192)
794 /* if we are only owner of page we can reuse it */
795 if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
796 return false;
797#else
798#define IXGBEVF_LAST_OFFSET \
799 (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBEVF_RXBUFFER_2048)
800
801 if (rx_buffer->page_offset > IXGBEVF_LAST_OFFSET)
802 return false;
803
804#endif
805
806 /* If we have drained the page fragment pool we need to update
807 * the pagecnt_bias and page count so that we fully restock the
808 * number of references the driver holds.
809 */
810 if (unlikely(!pagecnt_bias)) {
811 page_ref_add(page, USHRT_MAX);
812 rx_buffer->pagecnt_bias = USHRT_MAX;
813 }
814
815 return true;
816}
817
818/**
819 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
820 * @rx_ring: rx descriptor ring to transact packets on
821 * @rx_buffer: buffer containing page to add
822 * @skb: sk_buff to place the data into
823 * @size: size of buffer to be added
824 *
825 * This function will add the data contained in rx_buffer->page to the skb.
826 **/
827static void ixgbevf_add_rx_frag(struct ixgbevf_ring *rx_ring,
828 struct ixgbevf_rx_buffer *rx_buffer,
829 struct sk_buff *skb,
830 unsigned int size)
831{
832#if (PAGE_SIZE < 8192)
833 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
834#else
835 unsigned int truesize = ring_uses_build_skb(rx_ring) ?
836 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) :
837 SKB_DATA_ALIGN(size);
838#endif
839 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
840 rx_buffer->page_offset, size, truesize);
841#if (PAGE_SIZE < 8192)
842 rx_buffer->page_offset ^= truesize;
843#else
844 rx_buffer->page_offset += truesize;
845#endif
846}
847
848static
849struct sk_buff *ixgbevf_construct_skb(struct ixgbevf_ring *rx_ring,
850 struct ixgbevf_rx_buffer *rx_buffer,
851 struct xdp_buff *xdp,
852 union ixgbe_adv_rx_desc *rx_desc)
853{
854 unsigned int size = xdp->data_end - xdp->data;
855#if (PAGE_SIZE < 8192)
856 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
857#else
858 unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
859 xdp->data_hard_start);
860#endif
861 unsigned int headlen;
862 struct sk_buff *skb;
863
864 /* prefetch first cache line of first page */
865 net_prefetch(xdp->data);
866
867 /* Note, we get here by enabling legacy-rx via:
868 *
869 * ethtool --set-priv-flags <dev> legacy-rx on
870 *
871 * In this mode, we currently get 0 extra XDP headroom as
872 * opposed to having legacy-rx off, where we process XDP
873 * packets going to stack via ixgbevf_build_skb().
874 *
875 * For ixgbevf_construct_skb() mode it means that the
876 * xdp->data_meta will always point to xdp->data, since
877 * the helper cannot expand the head. Should this ever
878 * changed in future for legacy-rx mode on, then lets also
879 * add xdp->data_meta handling here.
880 */
881
882 /* allocate a skb to store the frags */
883 skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBEVF_RX_HDR_SIZE);
884 if (unlikely(!skb))
885 return NULL;
886
887 /* Determine available headroom for copy */
888 headlen = size;
889 if (headlen > IXGBEVF_RX_HDR_SIZE)
890 headlen = eth_get_headlen(skb->dev, xdp->data,
891 IXGBEVF_RX_HDR_SIZE);
892
893 /* align pull length to size of long to optimize memcpy performance */
894 memcpy(__skb_put(skb, headlen), xdp->data,
895 ALIGN(headlen, sizeof(long)));
896
897 /* update all of the pointers */
898 size -= headlen;
899 if (size) {
900 skb_add_rx_frag(skb, 0, rx_buffer->page,
901 (xdp->data + headlen) -
902 page_address(rx_buffer->page),
903 size, truesize);
904#if (PAGE_SIZE < 8192)
905 rx_buffer->page_offset ^= truesize;
906#else
907 rx_buffer->page_offset += truesize;
908#endif
909 } else {
910 rx_buffer->pagecnt_bias++;
911 }
912
913 return skb;
914}
915
916static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
917 u32 qmask)
918{
919 struct ixgbe_hw *hw = &adapter->hw;
920
921 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
922}
923
924static struct sk_buff *ixgbevf_build_skb(struct ixgbevf_ring *rx_ring,
925 struct ixgbevf_rx_buffer *rx_buffer,
926 struct xdp_buff *xdp,
927 union ixgbe_adv_rx_desc *rx_desc)
928{
929 unsigned int metasize = xdp->data - xdp->data_meta;
930#if (PAGE_SIZE < 8192)
931 unsigned int truesize = ixgbevf_rx_pg_size(rx_ring) / 2;
932#else
933 unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
934 SKB_DATA_ALIGN(xdp->data_end -
935 xdp->data_hard_start);
936#endif
937 struct sk_buff *skb;
938
939 /* Prefetch first cache line of first page. If xdp->data_meta
940 * is unused, this points to xdp->data, otherwise, we likely
941 * have a consumer accessing first few bytes of meta data,
942 * and then actual data.
943 */
944 net_prefetch(xdp->data_meta);
945
946 /* build an skb around the page buffer */
947 skb = napi_build_skb(xdp->data_hard_start, truesize);
948 if (unlikely(!skb))
949 return NULL;
950
951 /* update pointers within the skb to store the data */
952 skb_reserve(skb, xdp->data - xdp->data_hard_start);
953 __skb_put(skb, xdp->data_end - xdp->data);
954 if (metasize)
955 skb_metadata_set(skb, metasize);
956
957 /* update buffer offset */
958#if (PAGE_SIZE < 8192)
959 rx_buffer->page_offset ^= truesize;
960#else
961 rx_buffer->page_offset += truesize;
962#endif
963
964 return skb;
965}
966
967#define IXGBEVF_XDP_PASS 0
968#define IXGBEVF_XDP_CONSUMED 1
969#define IXGBEVF_XDP_TX 2
970
971static int ixgbevf_xmit_xdp_ring(struct ixgbevf_ring *ring,
972 struct xdp_buff *xdp)
973{
974 struct ixgbevf_tx_buffer *tx_buffer;
975 union ixgbe_adv_tx_desc *tx_desc;
976 u32 len, cmd_type;
977 dma_addr_t dma;
978 u16 i;
979
980 len = xdp->data_end - xdp->data;
981
982 if (unlikely(!ixgbevf_desc_unused(ring)))
983 return IXGBEVF_XDP_CONSUMED;
984
985 dma = dma_map_single(ring->dev, xdp->data, len, DMA_TO_DEVICE);
986 if (dma_mapping_error(ring->dev, dma))
987 return IXGBEVF_XDP_CONSUMED;
988
989 /* record the location of the first descriptor for this packet */
990 i = ring->next_to_use;
991 tx_buffer = &ring->tx_buffer_info[i];
992
993 dma_unmap_len_set(tx_buffer, len, len);
994 dma_unmap_addr_set(tx_buffer, dma, dma);
995 tx_buffer->data = xdp->data;
996 tx_buffer->bytecount = len;
997 tx_buffer->gso_segs = 1;
998 tx_buffer->protocol = 0;
999
1000 /* Populate minimal context descriptor that will provide for the
1001 * fact that we are expected to process Ethernet frames.
1002 */
1003 if (!test_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state)) {
1004 struct ixgbe_adv_tx_context_desc *context_desc;
1005
1006 set_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1007
1008 context_desc = IXGBEVF_TX_CTXTDESC(ring, 0);
1009 context_desc->vlan_macip_lens =
1010 cpu_to_le32(ETH_HLEN << IXGBE_ADVTXD_MACLEN_SHIFT);
1011 context_desc->fceof_saidx = 0;
1012 context_desc->type_tucmd_mlhl =
1013 cpu_to_le32(IXGBE_TXD_CMD_DEXT |
1014 IXGBE_ADVTXD_DTYP_CTXT);
1015 context_desc->mss_l4len_idx = 0;
1016
1017 i = 1;
1018 }
1019
1020 /* put descriptor type bits */
1021 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
1022 IXGBE_ADVTXD_DCMD_DEXT |
1023 IXGBE_ADVTXD_DCMD_IFCS;
1024 cmd_type |= len | IXGBE_TXD_CMD;
1025
1026 tx_desc = IXGBEVF_TX_DESC(ring, i);
1027 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1028
1029 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1030 tx_desc->read.olinfo_status =
1031 cpu_to_le32((len << IXGBE_ADVTXD_PAYLEN_SHIFT) |
1032 IXGBE_ADVTXD_CC);
1033
1034 /* Avoid any potential race with cleanup */
1035 smp_wmb();
1036
1037 /* set next_to_watch value indicating a packet is present */
1038 i++;
1039 if (i == ring->count)
1040 i = 0;
1041
1042 tx_buffer->next_to_watch = tx_desc;
1043 ring->next_to_use = i;
1044
1045 return IXGBEVF_XDP_TX;
1046}
1047
1048static struct sk_buff *ixgbevf_run_xdp(struct ixgbevf_adapter *adapter,
1049 struct ixgbevf_ring *rx_ring,
1050 struct xdp_buff *xdp)
1051{
1052 int result = IXGBEVF_XDP_PASS;
1053 struct ixgbevf_ring *xdp_ring;
1054 struct bpf_prog *xdp_prog;
1055 u32 act;
1056
1057 xdp_prog = READ_ONCE(rx_ring->xdp_prog);
1058
1059 if (!xdp_prog)
1060 goto xdp_out;
1061
1062 act = bpf_prog_run_xdp(xdp_prog, xdp);
1063 switch (act) {
1064 case XDP_PASS:
1065 break;
1066 case XDP_TX:
1067 xdp_ring = adapter->xdp_ring[rx_ring->queue_index];
1068 result = ixgbevf_xmit_xdp_ring(xdp_ring, xdp);
1069 if (result == IXGBEVF_XDP_CONSUMED)
1070 goto out_failure;
1071 break;
1072 default:
1073 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, act);
1074 fallthrough;
1075 case XDP_ABORTED:
1076out_failure:
1077 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
1078 fallthrough; /* handle aborts by dropping packet */
1079 case XDP_DROP:
1080 result = IXGBEVF_XDP_CONSUMED;
1081 break;
1082 }
1083xdp_out:
1084 return ERR_PTR(-result);
1085}
1086
1087static unsigned int ixgbevf_rx_frame_truesize(struct ixgbevf_ring *rx_ring,
1088 unsigned int size)
1089{
1090 unsigned int truesize;
1091
1092#if (PAGE_SIZE < 8192)
1093 truesize = ixgbevf_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
1094#else
1095 truesize = ring_uses_build_skb(rx_ring) ?
1096 SKB_DATA_ALIGN(IXGBEVF_SKB_PAD + size) +
1097 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
1098 SKB_DATA_ALIGN(size);
1099#endif
1100 return truesize;
1101}
1102
1103static void ixgbevf_rx_buffer_flip(struct ixgbevf_ring *rx_ring,
1104 struct ixgbevf_rx_buffer *rx_buffer,
1105 unsigned int size)
1106{
1107 unsigned int truesize = ixgbevf_rx_frame_truesize(rx_ring, size);
1108
1109#if (PAGE_SIZE < 8192)
1110 rx_buffer->page_offset ^= truesize;
1111#else
1112 rx_buffer->page_offset += truesize;
1113#endif
1114}
1115
1116static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
1117 struct ixgbevf_ring *rx_ring,
1118 int budget)
1119{
1120 unsigned int total_rx_bytes = 0, total_rx_packets = 0, frame_sz = 0;
1121 struct ixgbevf_adapter *adapter = q_vector->adapter;
1122 u16 cleaned_count = ixgbevf_desc_unused(rx_ring);
1123 struct sk_buff *skb = rx_ring->skb;
1124 bool xdp_xmit = false;
1125 struct xdp_buff xdp;
1126
1127 /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
1128#if (PAGE_SIZE < 8192)
1129 frame_sz = ixgbevf_rx_frame_truesize(rx_ring, 0);
1130#endif
1131 xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
1132
1133 while (likely(total_rx_packets < budget)) {
1134 struct ixgbevf_rx_buffer *rx_buffer;
1135 union ixgbe_adv_rx_desc *rx_desc;
1136 unsigned int size;
1137
1138 /* return some buffers to hardware, one at a time is too slow */
1139 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
1140 ixgbevf_alloc_rx_buffers(rx_ring, cleaned_count);
1141 cleaned_count = 0;
1142 }
1143
1144 rx_desc = IXGBEVF_RX_DESC(rx_ring, rx_ring->next_to_clean);
1145 size = le16_to_cpu(rx_desc->wb.upper.length);
1146 if (!size)
1147 break;
1148
1149 /* This memory barrier is needed to keep us from reading
1150 * any other fields out of the rx_desc until we know the
1151 * RXD_STAT_DD bit is set
1152 */
1153 rmb();
1154
1155 rx_buffer = ixgbevf_get_rx_buffer(rx_ring, size);
1156
1157 /* retrieve a buffer from the ring */
1158 if (!skb) {
1159 unsigned int offset = ixgbevf_rx_offset(rx_ring);
1160 unsigned char *hard_start;
1161
1162 hard_start = page_address(rx_buffer->page) +
1163 rx_buffer->page_offset - offset;
1164 xdp_prepare_buff(&xdp, hard_start, offset, size, true);
1165#if (PAGE_SIZE > 4096)
1166 /* At larger PAGE_SIZE, frame_sz depend on len size */
1167 xdp.frame_sz = ixgbevf_rx_frame_truesize(rx_ring, size);
1168#endif
1169 skb = ixgbevf_run_xdp(adapter, rx_ring, &xdp);
1170 }
1171
1172 if (IS_ERR(skb)) {
1173 if (PTR_ERR(skb) == -IXGBEVF_XDP_TX) {
1174 xdp_xmit = true;
1175 ixgbevf_rx_buffer_flip(rx_ring, rx_buffer,
1176 size);
1177 } else {
1178 rx_buffer->pagecnt_bias++;
1179 }
1180 total_rx_packets++;
1181 total_rx_bytes += size;
1182 } else if (skb) {
1183 ixgbevf_add_rx_frag(rx_ring, rx_buffer, skb, size);
1184 } else if (ring_uses_build_skb(rx_ring)) {
1185 skb = ixgbevf_build_skb(rx_ring, rx_buffer,
1186 &xdp, rx_desc);
1187 } else {
1188 skb = ixgbevf_construct_skb(rx_ring, rx_buffer,
1189 &xdp, rx_desc);
1190 }
1191
1192 /* exit if we failed to retrieve a buffer */
1193 if (!skb) {
1194 rx_ring->rx_stats.alloc_rx_buff_failed++;
1195 rx_buffer->pagecnt_bias++;
1196 break;
1197 }
1198
1199 ixgbevf_put_rx_buffer(rx_ring, rx_buffer, skb);
1200 cleaned_count++;
1201
1202 /* fetch next buffer in frame if non-eop */
1203 if (ixgbevf_is_non_eop(rx_ring, rx_desc))
1204 continue;
1205
1206 /* verify the packet layout is correct */
1207 if (ixgbevf_cleanup_headers(rx_ring, rx_desc, skb)) {
1208 skb = NULL;
1209 continue;
1210 }
1211
1212 /* probably a little skewed due to removing CRC */
1213 total_rx_bytes += skb->len;
1214
1215 /* Workaround hardware that can't do proper VEPA multicast
1216 * source pruning.
1217 */
1218 if ((skb->pkt_type == PACKET_BROADCAST ||
1219 skb->pkt_type == PACKET_MULTICAST) &&
1220 ether_addr_equal(rx_ring->netdev->dev_addr,
1221 eth_hdr(skb)->h_source)) {
1222 dev_kfree_skb_irq(skb);
1223 continue;
1224 }
1225
1226 /* populate checksum, VLAN, and protocol */
1227 ixgbevf_process_skb_fields(rx_ring, rx_desc, skb);
1228
1229 ixgbevf_rx_skb(q_vector, skb);
1230
1231 /* reset skb pointer */
1232 skb = NULL;
1233
1234 /* update budget accounting */
1235 total_rx_packets++;
1236 }
1237
1238 /* place incomplete frames back on ring for completion */
1239 rx_ring->skb = skb;
1240
1241 if (xdp_xmit) {
1242 struct ixgbevf_ring *xdp_ring =
1243 adapter->xdp_ring[rx_ring->queue_index];
1244
1245 /* Force memory writes to complete before letting h/w
1246 * know there are new descriptors to fetch.
1247 */
1248 wmb();
1249 ixgbevf_write_tail(xdp_ring, xdp_ring->next_to_use);
1250 }
1251
1252 u64_stats_update_begin(&rx_ring->syncp);
1253 rx_ring->stats.packets += total_rx_packets;
1254 rx_ring->stats.bytes += total_rx_bytes;
1255 u64_stats_update_end(&rx_ring->syncp);
1256 q_vector->rx.total_packets += total_rx_packets;
1257 q_vector->rx.total_bytes += total_rx_bytes;
1258
1259 return total_rx_packets;
1260}
1261
1262/**
1263 * ixgbevf_poll - NAPI polling calback
1264 * @napi: napi struct with our devices info in it
1265 * @budget: amount of work driver is allowed to do this pass, in packets
1266 *
1267 * This function will clean more than one or more rings associated with a
1268 * q_vector.
1269 **/
1270static int ixgbevf_poll(struct napi_struct *napi, int budget)
1271{
1272 struct ixgbevf_q_vector *q_vector =
1273 container_of(napi, struct ixgbevf_q_vector, napi);
1274 struct ixgbevf_adapter *adapter = q_vector->adapter;
1275 struct ixgbevf_ring *ring;
1276 int per_ring_budget, work_done = 0;
1277 bool clean_complete = true;
1278
1279 ixgbevf_for_each_ring(ring, q_vector->tx) {
1280 if (!ixgbevf_clean_tx_irq(q_vector, ring, budget))
1281 clean_complete = false;
1282 }
1283
1284 if (budget <= 0)
1285 return budget;
1286
1287 /* attempt to distribute budget to each queue fairly, but don't allow
1288 * the budget to go below 1 because we'll exit polling
1289 */
1290 if (q_vector->rx.count > 1)
1291 per_ring_budget = max(budget/q_vector->rx.count, 1);
1292 else
1293 per_ring_budget = budget;
1294
1295 ixgbevf_for_each_ring(ring, q_vector->rx) {
1296 int cleaned = ixgbevf_clean_rx_irq(q_vector, ring,
1297 per_ring_budget);
1298 work_done += cleaned;
1299 if (cleaned >= per_ring_budget)
1300 clean_complete = false;
1301 }
1302
1303 /* If all work not completed, return budget and keep polling */
1304 if (!clean_complete)
1305 return budget;
1306
1307 /* Exit the polling mode, but don't re-enable interrupts if stack might
1308 * poll us due to busy-polling
1309 */
1310 if (likely(napi_complete_done(napi, work_done))) {
1311 if (adapter->rx_itr_setting == 1)
1312 ixgbevf_set_itr(q_vector);
1313 if (!test_bit(__IXGBEVF_DOWN, &adapter->state) &&
1314 !test_bit(__IXGBEVF_REMOVING, &adapter->state))
1315 ixgbevf_irq_enable_queues(adapter,
1316 BIT(q_vector->v_idx));
1317 }
1318
1319 return min(work_done, budget - 1);
1320}
1321
1322/**
1323 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1324 * @q_vector: structure containing interrupt and ring information
1325 **/
1326void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
1327{
1328 struct ixgbevf_adapter *adapter = q_vector->adapter;
1329 struct ixgbe_hw *hw = &adapter->hw;
1330 int v_idx = q_vector->v_idx;
1331 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
1332
1333 /* set the WDIS bit to not clear the timer bits and cause an
1334 * immediate assertion of the interrupt
1335 */
1336 itr_reg |= IXGBE_EITR_CNT_WDIS;
1337
1338 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
1339}
1340
1341/**
1342 * ixgbevf_configure_msix - Configure MSI-X hardware
1343 * @adapter: board private structure
1344 *
1345 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1346 * interrupts.
1347 **/
1348static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
1349{
1350 struct ixgbevf_q_vector *q_vector;
1351 int q_vectors, v_idx;
1352
1353 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1354 adapter->eims_enable_mask = 0;
1355
1356 /* Populate the IVAR table and set the ITR values to the
1357 * corresponding register.
1358 */
1359 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
1360 struct ixgbevf_ring *ring;
1361
1362 q_vector = adapter->q_vector[v_idx];
1363
1364 ixgbevf_for_each_ring(ring, q_vector->rx)
1365 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
1366
1367 ixgbevf_for_each_ring(ring, q_vector->tx)
1368 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
1369
1370 if (q_vector->tx.ring && !q_vector->rx.ring) {
1371 /* Tx only vector */
1372 if (adapter->tx_itr_setting == 1)
1373 q_vector->itr = IXGBE_12K_ITR;
1374 else
1375 q_vector->itr = adapter->tx_itr_setting;
1376 } else {
1377 /* Rx or Rx/Tx vector */
1378 if (adapter->rx_itr_setting == 1)
1379 q_vector->itr = IXGBE_20K_ITR;
1380 else
1381 q_vector->itr = adapter->rx_itr_setting;
1382 }
1383
1384 /* add q_vector eims value to global eims_enable_mask */
1385 adapter->eims_enable_mask |= BIT(v_idx);
1386
1387 ixgbevf_write_eitr(q_vector);
1388 }
1389
1390 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
1391 /* setup eims_other and add value to global eims_enable_mask */
1392 adapter->eims_other = BIT(v_idx);
1393 adapter->eims_enable_mask |= adapter->eims_other;
1394}
1395
1396enum latency_range {
1397 lowest_latency = 0,
1398 low_latency = 1,
1399 bulk_latency = 2,
1400 latency_invalid = 255
1401};
1402
1403/**
1404 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1405 * @q_vector: structure containing interrupt and ring information
1406 * @ring_container: structure containing ring performance data
1407 *
1408 * Stores a new ITR value based on packets and byte
1409 * counts during the last interrupt. The advantage of per interrupt
1410 * computation is faster updates and more accurate ITR for the current
1411 * traffic pattern. Constants in this function were computed
1412 * based on theoretical maximum wire speed and thresholds were set based
1413 * on testing data as well as attempting to minimize response time
1414 * while increasing bulk throughput.
1415 **/
1416static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
1417 struct ixgbevf_ring_container *ring_container)
1418{
1419 int bytes = ring_container->total_bytes;
1420 int packets = ring_container->total_packets;
1421 u32 timepassed_us;
1422 u64 bytes_perint;
1423 u8 itr_setting = ring_container->itr;
1424
1425 if (packets == 0)
1426 return;
1427
1428 /* simple throttle rate management
1429 * 0-20MB/s lowest (100000 ints/s)
1430 * 20-100MB/s low (20000 ints/s)
1431 * 100-1249MB/s bulk (12000 ints/s)
1432 */
1433 /* what was last interrupt timeslice? */
1434 timepassed_us = q_vector->itr >> 2;
1435 if (timepassed_us == 0)
1436 return;
1437
1438 bytes_perint = bytes / timepassed_us; /* bytes/usec */
1439
1440 switch (itr_setting) {
1441 case lowest_latency:
1442 if (bytes_perint > 10)
1443 itr_setting = low_latency;
1444 break;
1445 case low_latency:
1446 if (bytes_perint > 20)
1447 itr_setting = bulk_latency;
1448 else if (bytes_perint <= 10)
1449 itr_setting = lowest_latency;
1450 break;
1451 case bulk_latency:
1452 if (bytes_perint <= 20)
1453 itr_setting = low_latency;
1454 break;
1455 }
1456
1457 /* clear work counters since we have the values we need */
1458 ring_container->total_bytes = 0;
1459 ring_container->total_packets = 0;
1460
1461 /* write updated itr to ring container */
1462 ring_container->itr = itr_setting;
1463}
1464
1465static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
1466{
1467 u32 new_itr = q_vector->itr;
1468 u8 current_itr;
1469
1470 ixgbevf_update_itr(q_vector, &q_vector->tx);
1471 ixgbevf_update_itr(q_vector, &q_vector->rx);
1472
1473 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
1474
1475 switch (current_itr) {
1476 /* counts and packets in update_itr are dependent on these numbers */
1477 case lowest_latency:
1478 new_itr = IXGBE_100K_ITR;
1479 break;
1480 case low_latency:
1481 new_itr = IXGBE_20K_ITR;
1482 break;
1483 case bulk_latency:
1484 new_itr = IXGBE_12K_ITR;
1485 break;
1486 default:
1487 break;
1488 }
1489
1490 if (new_itr != q_vector->itr) {
1491 /* do an exponential smoothing */
1492 new_itr = (10 * new_itr * q_vector->itr) /
1493 ((9 * new_itr) + q_vector->itr);
1494
1495 /* save the algorithm value here */
1496 q_vector->itr = new_itr;
1497
1498 ixgbevf_write_eitr(q_vector);
1499 }
1500}
1501
1502static irqreturn_t ixgbevf_msix_other(int irq, void *data)
1503{
1504 struct ixgbevf_adapter *adapter = data;
1505 struct ixgbe_hw *hw = &adapter->hw;
1506
1507 hw->mac.get_link_status = 1;
1508
1509 ixgbevf_service_event_schedule(adapter);
1510
1511 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
1512
1513 return IRQ_HANDLED;
1514}
1515
1516/**
1517 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1518 * @irq: unused
1519 * @data: pointer to our q_vector struct for this interrupt vector
1520 **/
1521static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
1522{
1523 struct ixgbevf_q_vector *q_vector = data;
1524
1525 /* EIAM disabled interrupts (on this vector) for us */
1526 if (q_vector->rx.ring || q_vector->tx.ring)
1527 napi_schedule_irqoff(&q_vector->napi);
1528
1529 return IRQ_HANDLED;
1530}
1531
1532/**
1533 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1534 * @adapter: board private structure
1535 *
1536 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1537 * interrupts from the kernel.
1538 **/
1539static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1540{
1541 struct net_device *netdev = adapter->netdev;
1542 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1543 unsigned int ri = 0, ti = 0;
1544 int vector, err;
1545
1546 for (vector = 0; vector < q_vectors; vector++) {
1547 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
1548 struct msix_entry *entry = &adapter->msix_entries[vector];
1549
1550 if (q_vector->tx.ring && q_vector->rx.ring) {
1551 snprintf(q_vector->name, sizeof(q_vector->name),
1552 "%s-TxRx-%u", netdev->name, ri++);
1553 ti++;
1554 } else if (q_vector->rx.ring) {
1555 snprintf(q_vector->name, sizeof(q_vector->name),
1556 "%s-rx-%u", netdev->name, ri++);
1557 } else if (q_vector->tx.ring) {
1558 snprintf(q_vector->name, sizeof(q_vector->name),
1559 "%s-tx-%u", netdev->name, ti++);
1560 } else {
1561 /* skip this unused q_vector */
1562 continue;
1563 }
1564 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
1565 q_vector->name, q_vector);
1566 if (err) {
1567 hw_dbg(&adapter->hw,
1568 "request_irq failed for MSIX interrupt Error: %d\n",
1569 err);
1570 goto free_queue_irqs;
1571 }
1572 }
1573
1574 err = request_irq(adapter->msix_entries[vector].vector,
1575 &ixgbevf_msix_other, 0, netdev->name, adapter);
1576 if (err) {
1577 hw_dbg(&adapter->hw, "request_irq for msix_other failed: %d\n",
1578 err);
1579 goto free_queue_irqs;
1580 }
1581
1582 return 0;
1583
1584free_queue_irqs:
1585 while (vector) {
1586 vector--;
1587 free_irq(adapter->msix_entries[vector].vector,
1588 adapter->q_vector[vector]);
1589 }
1590 /* This failure is non-recoverable - it indicates the system is
1591 * out of MSIX vector resources and the VF driver cannot run
1592 * without them. Set the number of msix vectors to zero
1593 * indicating that not enough can be allocated. The error
1594 * will be returned to the user indicating device open failed.
1595 * Any further attempts to force the driver to open will also
1596 * fail. The only way to recover is to unload the driver and
1597 * reload it again. If the system has recovered some MSIX
1598 * vectors then it may succeed.
1599 */
1600 adapter->num_msix_vectors = 0;
1601 return err;
1602}
1603
1604/**
1605 * ixgbevf_request_irq - initialize interrupts
1606 * @adapter: board private structure
1607 *
1608 * Attempts to configure interrupts using the best available
1609 * capabilities of the hardware and kernel.
1610 **/
1611static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1612{
1613 int err = ixgbevf_request_msix_irqs(adapter);
1614
1615 if (err)
1616 hw_dbg(&adapter->hw, "request_irq failed, Error %d\n", err);
1617
1618 return err;
1619}
1620
1621static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1622{
1623 int i, q_vectors;
1624
1625 if (!adapter->msix_entries)
1626 return;
1627
1628 q_vectors = adapter->num_msix_vectors;
1629 i = q_vectors - 1;
1630
1631 free_irq(adapter->msix_entries[i].vector, adapter);
1632 i--;
1633
1634 for (; i >= 0; i--) {
1635 /* free only the irqs that were actually requested */
1636 if (!adapter->q_vector[i]->rx.ring &&
1637 !adapter->q_vector[i]->tx.ring)
1638 continue;
1639
1640 free_irq(adapter->msix_entries[i].vector,
1641 adapter->q_vector[i]);
1642 }
1643}
1644
1645/**
1646 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1647 * @adapter: board private structure
1648 **/
1649static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1650{
1651 struct ixgbe_hw *hw = &adapter->hw;
1652 int i;
1653
1654 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1655 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1656 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1657
1658 IXGBE_WRITE_FLUSH(hw);
1659
1660 for (i = 0; i < adapter->num_msix_vectors; i++)
1661 synchronize_irq(adapter->msix_entries[i].vector);
1662}
1663
1664/**
1665 * ixgbevf_irq_enable - Enable default interrupt generation settings
1666 * @adapter: board private structure
1667 **/
1668static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1669{
1670 struct ixgbe_hw *hw = &adapter->hw;
1671
1672 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1673 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1674 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1675}
1676
1677/**
1678 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1679 * @adapter: board private structure
1680 * @ring: structure containing ring specific data
1681 *
1682 * Configure the Tx descriptor ring after a reset.
1683 **/
1684static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter *adapter,
1685 struct ixgbevf_ring *ring)
1686{
1687 struct ixgbe_hw *hw = &adapter->hw;
1688 u64 tdba = ring->dma;
1689 int wait_loop = 10;
1690 u32 txdctl = IXGBE_TXDCTL_ENABLE;
1691 u8 reg_idx = ring->reg_idx;
1692
1693 /* disable queue to avoid issues while updating state */
1694 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
1695 IXGBE_WRITE_FLUSH(hw);
1696
1697 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(reg_idx), tdba & DMA_BIT_MASK(32));
1698 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(reg_idx), tdba >> 32);
1699 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(reg_idx),
1700 ring->count * sizeof(union ixgbe_adv_tx_desc));
1701
1702 /* disable head writeback */
1703 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAH(reg_idx), 0);
1704 IXGBE_WRITE_REG(hw, IXGBE_VFTDWBAL(reg_idx), 0);
1705
1706 /* enable relaxed ordering */
1707 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(reg_idx),
1708 (IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1709 IXGBE_DCA_TXCTRL_DATA_RRO_EN));
1710
1711 /* reset head and tail pointers */
1712 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(reg_idx), 0);
1713 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(reg_idx), 0);
1714 ring->tail = adapter->io_addr + IXGBE_VFTDT(reg_idx);
1715
1716 /* reset ntu and ntc to place SW in sync with hardwdare */
1717 ring->next_to_clean = 0;
1718 ring->next_to_use = 0;
1719
1720 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1721 * to or less than the number of on chip descriptors, which is
1722 * currently 40.
1723 */
1724 txdctl |= (8 << 16); /* WTHRESH = 8 */
1725
1726 /* Setting PTHRESH to 32 both improves performance */
1727 txdctl |= (1u << 8) | /* HTHRESH = 1 */
1728 32; /* PTHRESH = 32 */
1729
1730 /* reinitialize tx_buffer_info */
1731 memset(ring->tx_buffer_info, 0,
1732 sizeof(struct ixgbevf_tx_buffer) * ring->count);
1733
1734 clear_bit(__IXGBEVF_HANG_CHECK_ARMED, &ring->state);
1735 clear_bit(__IXGBEVF_TX_XDP_RING_PRIMED, &ring->state);
1736
1737 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx), txdctl);
1738
1739 /* poll to verify queue is enabled */
1740 do {
1741 usleep_range(1000, 2000);
1742 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(reg_idx));
1743 } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
1744 if (!wait_loop)
1745 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
1746}
1747
1748/**
1749 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1750 * @adapter: board private structure
1751 *
1752 * Configure the Tx unit of the MAC after a reset.
1753 **/
1754static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1755{
1756 u32 i;
1757
1758 /* Setup the HW Tx Head and Tail descriptor pointers */
1759 for (i = 0; i < adapter->num_tx_queues; i++)
1760 ixgbevf_configure_tx_ring(adapter, adapter->tx_ring[i]);
1761 for (i = 0; i < adapter->num_xdp_queues; i++)
1762 ixgbevf_configure_tx_ring(adapter, adapter->xdp_ring[i]);
1763}
1764
1765#define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1766
1767static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter,
1768 struct ixgbevf_ring *ring, int index)
1769{
1770 struct ixgbe_hw *hw = &adapter->hw;
1771 u32 srrctl;
1772
1773 srrctl = IXGBE_SRRCTL_DROP_EN;
1774
1775 srrctl |= IXGBEVF_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
1776 if (ring_uses_large_buffer(ring))
1777 srrctl |= IXGBEVF_RXBUFFER_3072 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1778 else
1779 srrctl |= IXGBEVF_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1780 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1781
1782 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1783}
1784
1785static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1786{
1787 struct ixgbe_hw *hw = &adapter->hw;
1788
1789 /* PSRTYPE must be initialized in 82599 */
1790 u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1791 IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1792 IXGBE_PSRTYPE_L2HDR;
1793
1794 if (adapter->num_rx_queues > 1)
1795 psrtype |= BIT(29);
1796
1797 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1798}
1799
1800#define IXGBEVF_MAX_RX_DESC_POLL 10
1801static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1802 struct ixgbevf_ring *ring)
1803{
1804 struct ixgbe_hw *hw = &adapter->hw;
1805 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1806 u32 rxdctl;
1807 u8 reg_idx = ring->reg_idx;
1808
1809 if (IXGBE_REMOVED(hw->hw_addr))
1810 return;
1811 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1812 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1813
1814 /* write value back with RXDCTL.ENABLE bit cleared */
1815 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1816
1817 /* the hardware may take up to 100us to really disable the Rx queue */
1818 do {
1819 udelay(10);
1820 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1821 } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1822
1823 if (!wait_loop)
1824 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1825 reg_idx);
1826}
1827
1828static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1829 struct ixgbevf_ring *ring)
1830{
1831 struct ixgbe_hw *hw = &adapter->hw;
1832 int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1833 u32 rxdctl;
1834 u8 reg_idx = ring->reg_idx;
1835
1836 if (IXGBE_REMOVED(hw->hw_addr))
1837 return;
1838 do {
1839 usleep_range(1000, 2000);
1840 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1841 } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1842
1843 if (!wait_loop)
1844 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1845 reg_idx);
1846}
1847
1848/**
1849 * ixgbevf_init_rss_key - Initialize adapter RSS key
1850 * @adapter: device handle
1851 *
1852 * Allocates and initializes the RSS key if it is not allocated.
1853 **/
1854static inline int ixgbevf_init_rss_key(struct ixgbevf_adapter *adapter)
1855{
1856 u32 *rss_key;
1857
1858 if (!adapter->rss_key) {
1859 rss_key = kzalloc(IXGBEVF_RSS_HASH_KEY_SIZE, GFP_KERNEL);
1860 if (unlikely(!rss_key))
1861 return -ENOMEM;
1862
1863 netdev_rss_key_fill(rss_key, IXGBEVF_RSS_HASH_KEY_SIZE);
1864 adapter->rss_key = rss_key;
1865 }
1866
1867 return 0;
1868}
1869
1870static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter *adapter)
1871{
1872 struct ixgbe_hw *hw = &adapter->hw;
1873 u32 vfmrqc = 0, vfreta = 0;
1874 u16 rss_i = adapter->num_rx_queues;
1875 u8 i, j;
1876
1877 /* Fill out hash function seeds */
1878 for (i = 0; i < IXGBEVF_VFRSSRK_REGS; i++)
1879 IXGBE_WRITE_REG(hw, IXGBE_VFRSSRK(i), *(adapter->rss_key + i));
1880
1881 for (i = 0, j = 0; i < IXGBEVF_X550_VFRETA_SIZE; i++, j++) {
1882 if (j == rss_i)
1883 j = 0;
1884
1885 adapter->rss_indir_tbl[i] = j;
1886
1887 vfreta |= j << (i & 0x3) * 8;
1888 if ((i & 3) == 3) {
1889 IXGBE_WRITE_REG(hw, IXGBE_VFRETA(i >> 2), vfreta);
1890 vfreta = 0;
1891 }
1892 }
1893
1894 /* Perform hash on these packet types */
1895 vfmrqc |= IXGBE_VFMRQC_RSS_FIELD_IPV4 |
1896 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP |
1897 IXGBE_VFMRQC_RSS_FIELD_IPV6 |
1898 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP;
1899
1900 vfmrqc |= IXGBE_VFMRQC_RSSEN;
1901
1902 IXGBE_WRITE_REG(hw, IXGBE_VFMRQC, vfmrqc);
1903}
1904
1905static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter *adapter,
1906 struct ixgbevf_ring *ring)
1907{
1908 struct ixgbe_hw *hw = &adapter->hw;
1909 union ixgbe_adv_rx_desc *rx_desc;
1910 u64 rdba = ring->dma;
1911 u32 rxdctl;
1912 u8 reg_idx = ring->reg_idx;
1913
1914 /* disable queue to avoid issues while updating state */
1915 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1916 ixgbevf_disable_rx_queue(adapter, ring);
1917
1918 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(reg_idx), rdba & DMA_BIT_MASK(32));
1919 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(reg_idx), rdba >> 32);
1920 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(reg_idx),
1921 ring->count * sizeof(union ixgbe_adv_rx_desc));
1922
1923#ifndef CONFIG_SPARC
1924 /* enable relaxed ordering */
1925 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1926 IXGBE_DCA_RXCTRL_DESC_RRO_EN);
1927#else
1928 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_RXCTRL(reg_idx),
1929 IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1930 IXGBE_DCA_RXCTRL_DATA_WRO_EN);
1931#endif
1932
1933 /* reset head and tail pointers */
1934 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(reg_idx), 0);
1935 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(reg_idx), 0);
1936 ring->tail = adapter->io_addr + IXGBE_VFRDT(reg_idx);
1937
1938 /* initialize rx_buffer_info */
1939 memset(ring->rx_buffer_info, 0,
1940 sizeof(struct ixgbevf_rx_buffer) * ring->count);
1941
1942 /* initialize Rx descriptor 0 */
1943 rx_desc = IXGBEVF_RX_DESC(ring, 0);
1944 rx_desc->wb.upper.length = 0;
1945
1946 /* reset ntu and ntc to place SW in sync with hardwdare */
1947 ring->next_to_clean = 0;
1948 ring->next_to_use = 0;
1949 ring->next_to_alloc = 0;
1950
1951 ixgbevf_configure_srrctl(adapter, ring, reg_idx);
1952
1953 /* RXDCTL.RLPML does not work on 82599 */
1954 if (adapter->hw.mac.type != ixgbe_mac_82599_vf) {
1955 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
1956 IXGBE_RXDCTL_RLPML_EN);
1957
1958#if (PAGE_SIZE < 8192)
1959 /* Limit the maximum frame size so we don't overrun the skb */
1960 if (ring_uses_build_skb(ring) &&
1961 !ring_uses_large_buffer(ring))
1962 rxdctl |= IXGBEVF_MAX_FRAME_BUILD_SKB |
1963 IXGBE_RXDCTL_RLPML_EN;
1964#endif
1965 }
1966
1967 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1968 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1969
1970 ixgbevf_rx_desc_queue_enable(adapter, ring);
1971 ixgbevf_alloc_rx_buffers(ring, ixgbevf_desc_unused(ring));
1972}
1973
1974static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter,
1975 struct ixgbevf_ring *rx_ring)
1976{
1977 struct net_device *netdev = adapter->netdev;
1978 unsigned int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1979
1980 /* set build_skb and buffer size flags */
1981 clear_ring_build_skb_enabled(rx_ring);
1982 clear_ring_uses_large_buffer(rx_ring);
1983
1984 if (adapter->flags & IXGBEVF_FLAGS_LEGACY_RX)
1985 return;
1986
1987 if (PAGE_SIZE < 8192)
1988 if (max_frame > IXGBEVF_MAX_FRAME_BUILD_SKB)
1989 set_ring_uses_large_buffer(rx_ring);
1990
1991 /* 82599 can't rely on RXDCTL.RLPML to restrict the size of the frame */
1992 if (adapter->hw.mac.type == ixgbe_mac_82599_vf && !ring_uses_large_buffer(rx_ring))
1993 return;
1994
1995 set_ring_build_skb_enabled(rx_ring);
1996}
1997
1998/**
1999 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
2000 * @adapter: board private structure
2001 *
2002 * Configure the Rx unit of the MAC after a reset.
2003 **/
2004static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
2005{
2006 struct ixgbe_hw *hw = &adapter->hw;
2007 struct net_device *netdev = adapter->netdev;
2008 int i, ret;
2009
2010 ixgbevf_setup_psrtype(adapter);
2011 if (hw->mac.type >= ixgbe_mac_X550_vf)
2012 ixgbevf_setup_vfmrqc(adapter);
2013
2014 spin_lock_bh(&adapter->mbx_lock);
2015 /* notify the PF of our intent to use this size of frame */
2016 ret = hw->mac.ops.set_rlpml(hw, netdev->mtu + ETH_HLEN + ETH_FCS_LEN);
2017 spin_unlock_bh(&adapter->mbx_lock);
2018 if (ret)
2019 dev_err(&adapter->pdev->dev,
2020 "Failed to set MTU at %d\n", netdev->mtu);
2021
2022 /* Setup the HW Rx Head and Tail Descriptor Pointers and
2023 * the Base and Length of the Rx Descriptor Ring
2024 */
2025 for (i = 0; i < adapter->num_rx_queues; i++) {
2026 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
2027
2028 ixgbevf_set_rx_buffer_len(adapter, rx_ring);
2029 ixgbevf_configure_rx_ring(adapter, rx_ring);
2030 }
2031}
2032
2033static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
2034 __be16 proto, u16 vid)
2035{
2036 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2037 struct ixgbe_hw *hw = &adapter->hw;
2038 int err;
2039
2040 spin_lock_bh(&adapter->mbx_lock);
2041
2042 /* add VID to filter table */
2043 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
2044
2045 spin_unlock_bh(&adapter->mbx_lock);
2046
2047 if (err) {
2048 netdev_err(netdev, "VF could not set VLAN %d\n", vid);
2049
2050 /* translate error return types so error makes sense */
2051 if (err == IXGBE_ERR_MBX)
2052 return -EIO;
2053
2054 if (err == IXGBE_ERR_INVALID_ARGUMENT)
2055 return -EACCES;
2056 }
2057
2058 set_bit(vid, adapter->active_vlans);
2059
2060 return err;
2061}
2062
2063static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
2064 __be16 proto, u16 vid)
2065{
2066 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2067 struct ixgbe_hw *hw = &adapter->hw;
2068 int err;
2069
2070 spin_lock_bh(&adapter->mbx_lock);
2071
2072 /* remove VID from filter table */
2073 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
2074
2075 spin_unlock_bh(&adapter->mbx_lock);
2076
2077 if (err)
2078 netdev_err(netdev, "Could not remove VLAN %d\n", vid);
2079
2080 clear_bit(vid, adapter->active_vlans);
2081
2082 return err;
2083}
2084
2085static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
2086{
2087 u16 vid;
2088
2089 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2090 ixgbevf_vlan_rx_add_vid(adapter->netdev,
2091 htons(ETH_P_8021Q), vid);
2092}
2093
2094static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
2095{
2096 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2097 struct ixgbe_hw *hw = &adapter->hw;
2098 int count = 0;
2099
2100 if (!netdev_uc_empty(netdev)) {
2101 struct netdev_hw_addr *ha;
2102
2103 netdev_for_each_uc_addr(ha, netdev) {
2104 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
2105 udelay(200);
2106 }
2107 } else {
2108 /* If the list is empty then send message to PF driver to
2109 * clear all MAC VLANs on this VF.
2110 */
2111 hw->mac.ops.set_uc_addr(hw, 0, NULL);
2112 }
2113
2114 return count;
2115}
2116
2117/**
2118 * ixgbevf_set_rx_mode - Multicast and unicast set
2119 * @netdev: network interface device structure
2120 *
2121 * The set_rx_method entry point is called whenever the multicast address
2122 * list, unicast address list or the network interface flags are updated.
2123 * This routine is responsible for configuring the hardware for proper
2124 * multicast mode and configuring requested unicast filters.
2125 **/
2126static void ixgbevf_set_rx_mode(struct net_device *netdev)
2127{
2128 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2129 struct ixgbe_hw *hw = &adapter->hw;
2130 unsigned int flags = netdev->flags;
2131 int xcast_mode;
2132
2133 /* request the most inclusive mode we need */
2134 if (flags & IFF_PROMISC)
2135 xcast_mode = IXGBEVF_XCAST_MODE_PROMISC;
2136 else if (flags & IFF_ALLMULTI)
2137 xcast_mode = IXGBEVF_XCAST_MODE_ALLMULTI;
2138 else if (flags & (IFF_BROADCAST | IFF_MULTICAST))
2139 xcast_mode = IXGBEVF_XCAST_MODE_MULTI;
2140 else
2141 xcast_mode = IXGBEVF_XCAST_MODE_NONE;
2142
2143 spin_lock_bh(&adapter->mbx_lock);
2144
2145 hw->mac.ops.update_xcast_mode(hw, xcast_mode);
2146
2147 /* reprogram multicast list */
2148 hw->mac.ops.update_mc_addr_list(hw, netdev);
2149
2150 ixgbevf_write_uc_addr_list(netdev);
2151
2152 spin_unlock_bh(&adapter->mbx_lock);
2153}
2154
2155static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
2156{
2157 int q_idx;
2158 struct ixgbevf_q_vector *q_vector;
2159 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2160
2161 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2162 q_vector = adapter->q_vector[q_idx];
2163 napi_enable(&q_vector->napi);
2164 }
2165}
2166
2167static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
2168{
2169 int q_idx;
2170 struct ixgbevf_q_vector *q_vector;
2171 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2172
2173 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2174 q_vector = adapter->q_vector[q_idx];
2175 napi_disable(&q_vector->napi);
2176 }
2177}
2178
2179static int ixgbevf_configure_dcb(struct ixgbevf_adapter *adapter)
2180{
2181 struct ixgbe_hw *hw = &adapter->hw;
2182 unsigned int def_q = 0;
2183 unsigned int num_tcs = 0;
2184 unsigned int num_rx_queues = adapter->num_rx_queues;
2185 unsigned int num_tx_queues = adapter->num_tx_queues;
2186 int err;
2187
2188 spin_lock_bh(&adapter->mbx_lock);
2189
2190 /* fetch queue configuration from the PF */
2191 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2192
2193 spin_unlock_bh(&adapter->mbx_lock);
2194
2195 if (err)
2196 return err;
2197
2198 if (num_tcs > 1) {
2199 /* we need only one Tx queue */
2200 num_tx_queues = 1;
2201
2202 /* update default Tx ring register index */
2203 adapter->tx_ring[0]->reg_idx = def_q;
2204
2205 /* we need as many queues as traffic classes */
2206 num_rx_queues = num_tcs;
2207 }
2208
2209 /* if we have a bad config abort request queue reset */
2210 if ((adapter->num_rx_queues != num_rx_queues) ||
2211 (adapter->num_tx_queues != num_tx_queues)) {
2212 /* force mailbox timeout to prevent further messages */
2213 hw->mbx.timeout = 0;
2214
2215 /* wait for watchdog to come around and bail us out */
2216 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED, &adapter->state);
2217 }
2218
2219 return 0;
2220}
2221
2222static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
2223{
2224 ixgbevf_configure_dcb(adapter);
2225
2226 ixgbevf_set_rx_mode(adapter->netdev);
2227
2228 ixgbevf_restore_vlan(adapter);
2229 ixgbevf_ipsec_restore(adapter);
2230
2231 ixgbevf_configure_tx(adapter);
2232 ixgbevf_configure_rx(adapter);
2233}
2234
2235static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
2236{
2237 /* Only save pre-reset stats if there are some */
2238 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
2239 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
2240 adapter->stats.base_vfgprc;
2241 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
2242 adapter->stats.base_vfgptc;
2243 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
2244 adapter->stats.base_vfgorc;
2245 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
2246 adapter->stats.base_vfgotc;
2247 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
2248 adapter->stats.base_vfmprc;
2249 }
2250}
2251
2252static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
2253{
2254 struct ixgbe_hw *hw = &adapter->hw;
2255
2256 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
2257 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
2258 adapter->stats.last_vfgorc |=
2259 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
2260 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
2261 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
2262 adapter->stats.last_vfgotc |=
2263 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
2264 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
2265
2266 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
2267 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
2268 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
2269 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
2270 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
2271}
2272
2273static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
2274{
2275 struct ixgbe_hw *hw = &adapter->hw;
2276 static const int api[] = {
2277 ixgbe_mbox_api_15,
2278 ixgbe_mbox_api_14,
2279 ixgbe_mbox_api_13,
2280 ixgbe_mbox_api_12,
2281 ixgbe_mbox_api_11,
2282 ixgbe_mbox_api_10,
2283 ixgbe_mbox_api_unknown
2284 };
2285 int err, idx = 0;
2286
2287 spin_lock_bh(&adapter->mbx_lock);
2288
2289 while (api[idx] != ixgbe_mbox_api_unknown) {
2290 err = hw->mac.ops.negotiate_api_version(hw, api[idx]);
2291 if (!err)
2292 break;
2293 idx++;
2294 }
2295
2296 if (hw->api_version >= ixgbe_mbox_api_15) {
2297 hw->mbx.ops.init_params(hw);
2298 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
2299 sizeof(struct ixgbe_mbx_operations));
2300 }
2301
2302 spin_unlock_bh(&adapter->mbx_lock);
2303}
2304
2305static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
2306{
2307 struct net_device *netdev = adapter->netdev;
2308 struct pci_dev *pdev = adapter->pdev;
2309 struct ixgbe_hw *hw = &adapter->hw;
2310 bool state;
2311
2312 ixgbevf_configure_msix(adapter);
2313
2314 spin_lock_bh(&adapter->mbx_lock);
2315
2316 if (is_valid_ether_addr(hw->mac.addr))
2317 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
2318 else
2319 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
2320
2321 spin_unlock_bh(&adapter->mbx_lock);
2322
2323 state = adapter->link_state;
2324 hw->mac.ops.get_link_state(hw, &adapter->link_state);
2325 if (state && state != adapter->link_state)
2326 dev_info(&pdev->dev, "VF is administratively disabled\n");
2327
2328 smp_mb__before_atomic();
2329 clear_bit(__IXGBEVF_DOWN, &adapter->state);
2330 ixgbevf_napi_enable_all(adapter);
2331
2332 /* clear any pending interrupts, may auto mask */
2333 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2334 ixgbevf_irq_enable(adapter);
2335
2336 /* enable transmits */
2337 netif_tx_start_all_queues(netdev);
2338
2339 ixgbevf_save_reset_stats(adapter);
2340 ixgbevf_init_last_counter_stats(adapter);
2341
2342 hw->mac.get_link_status = 1;
2343 mod_timer(&adapter->service_timer, jiffies);
2344}
2345
2346void ixgbevf_up(struct ixgbevf_adapter *adapter)
2347{
2348 ixgbevf_configure(adapter);
2349
2350 ixgbevf_up_complete(adapter);
2351}
2352
2353/**
2354 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2355 * @rx_ring: ring to free buffers from
2356 **/
2357static void ixgbevf_clean_rx_ring(struct ixgbevf_ring *rx_ring)
2358{
2359 u16 i = rx_ring->next_to_clean;
2360
2361 /* Free Rx ring sk_buff */
2362 if (rx_ring->skb) {
2363 dev_kfree_skb(rx_ring->skb);
2364 rx_ring->skb = NULL;
2365 }
2366
2367 /* Free all the Rx ring pages */
2368 while (i != rx_ring->next_to_alloc) {
2369 struct ixgbevf_rx_buffer *rx_buffer;
2370
2371 rx_buffer = &rx_ring->rx_buffer_info[i];
2372
2373 /* Invalidate cache lines that may have been written to by
2374 * device so that we avoid corrupting memory.
2375 */
2376 dma_sync_single_range_for_cpu(rx_ring->dev,
2377 rx_buffer->dma,
2378 rx_buffer->page_offset,
2379 ixgbevf_rx_bufsz(rx_ring),
2380 DMA_FROM_DEVICE);
2381
2382 /* free resources associated with mapping */
2383 dma_unmap_page_attrs(rx_ring->dev,
2384 rx_buffer->dma,
2385 ixgbevf_rx_pg_size(rx_ring),
2386 DMA_FROM_DEVICE,
2387 IXGBEVF_RX_DMA_ATTR);
2388
2389 __page_frag_cache_drain(rx_buffer->page,
2390 rx_buffer->pagecnt_bias);
2391
2392 i++;
2393 if (i == rx_ring->count)
2394 i = 0;
2395 }
2396
2397 rx_ring->next_to_alloc = 0;
2398 rx_ring->next_to_clean = 0;
2399 rx_ring->next_to_use = 0;
2400}
2401
2402/**
2403 * ixgbevf_clean_tx_ring - Free Tx Buffers
2404 * @tx_ring: ring to be cleaned
2405 **/
2406static void ixgbevf_clean_tx_ring(struct ixgbevf_ring *tx_ring)
2407{
2408 u16 i = tx_ring->next_to_clean;
2409 struct ixgbevf_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
2410
2411 while (i != tx_ring->next_to_use) {
2412 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
2413
2414 /* Free all the Tx ring sk_buffs */
2415 if (ring_is_xdp(tx_ring))
2416 page_frag_free(tx_buffer->data);
2417 else
2418 dev_kfree_skb_any(tx_buffer->skb);
2419
2420 /* unmap skb header data */
2421 dma_unmap_single(tx_ring->dev,
2422 dma_unmap_addr(tx_buffer, dma),
2423 dma_unmap_len(tx_buffer, len),
2424 DMA_TO_DEVICE);
2425
2426 /* check for eop_desc to determine the end of the packet */
2427 eop_desc = tx_buffer->next_to_watch;
2428 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2429
2430 /* unmap remaining buffers */
2431 while (tx_desc != eop_desc) {
2432 tx_buffer++;
2433 tx_desc++;
2434 i++;
2435 if (unlikely(i == tx_ring->count)) {
2436 i = 0;
2437 tx_buffer = tx_ring->tx_buffer_info;
2438 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
2439 }
2440
2441 /* unmap any remaining paged data */
2442 if (dma_unmap_len(tx_buffer, len))
2443 dma_unmap_page(tx_ring->dev,
2444 dma_unmap_addr(tx_buffer, dma),
2445 dma_unmap_len(tx_buffer, len),
2446 DMA_TO_DEVICE);
2447 }
2448
2449 /* move us one more past the eop_desc for start of next pkt */
2450 tx_buffer++;
2451 i++;
2452 if (unlikely(i == tx_ring->count)) {
2453 i = 0;
2454 tx_buffer = tx_ring->tx_buffer_info;
2455 }
2456 }
2457
2458 /* reset next_to_use and next_to_clean */
2459 tx_ring->next_to_use = 0;
2460 tx_ring->next_to_clean = 0;
2461
2462}
2463
2464/**
2465 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2466 * @adapter: board private structure
2467 **/
2468static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
2469{
2470 int i;
2471
2472 for (i = 0; i < adapter->num_rx_queues; i++)
2473 ixgbevf_clean_rx_ring(adapter->rx_ring[i]);
2474}
2475
2476/**
2477 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2478 * @adapter: board private structure
2479 **/
2480static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
2481{
2482 int i;
2483
2484 for (i = 0; i < adapter->num_tx_queues; i++)
2485 ixgbevf_clean_tx_ring(adapter->tx_ring[i]);
2486 for (i = 0; i < adapter->num_xdp_queues; i++)
2487 ixgbevf_clean_tx_ring(adapter->xdp_ring[i]);
2488}
2489
2490void ixgbevf_down(struct ixgbevf_adapter *adapter)
2491{
2492 struct net_device *netdev = adapter->netdev;
2493 struct ixgbe_hw *hw = &adapter->hw;
2494 int i;
2495
2496 /* signal that we are down to the interrupt handler */
2497 if (test_and_set_bit(__IXGBEVF_DOWN, &adapter->state))
2498 return; /* do nothing if already down */
2499
2500 /* disable all enabled Rx queues */
2501 for (i = 0; i < adapter->num_rx_queues; i++)
2502 ixgbevf_disable_rx_queue(adapter, adapter->rx_ring[i]);
2503
2504 usleep_range(10000, 20000);
2505
2506 netif_tx_stop_all_queues(netdev);
2507
2508 /* call carrier off first to avoid false dev_watchdog timeouts */
2509 netif_carrier_off(netdev);
2510 netif_tx_disable(netdev);
2511
2512 ixgbevf_irq_disable(adapter);
2513
2514 ixgbevf_napi_disable_all(adapter);
2515
2516 del_timer_sync(&adapter->service_timer);
2517
2518 /* disable transmits in the hardware now that interrupts are off */
2519 for (i = 0; i < adapter->num_tx_queues; i++) {
2520 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
2521
2522 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2523 IXGBE_TXDCTL_SWFLSH);
2524 }
2525
2526 for (i = 0; i < adapter->num_xdp_queues; i++) {
2527 u8 reg_idx = adapter->xdp_ring[i]->reg_idx;
2528
2529 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(reg_idx),
2530 IXGBE_TXDCTL_SWFLSH);
2531 }
2532
2533 if (!pci_channel_offline(adapter->pdev))
2534 ixgbevf_reset(adapter);
2535
2536 ixgbevf_clean_all_tx_rings(adapter);
2537 ixgbevf_clean_all_rx_rings(adapter);
2538}
2539
2540void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
2541{
2542 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
2543 msleep(1);
2544
2545 ixgbevf_down(adapter);
2546 pci_set_master(adapter->pdev);
2547 ixgbevf_up(adapter);
2548
2549 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
2550}
2551
2552void ixgbevf_reset(struct ixgbevf_adapter *adapter)
2553{
2554 struct ixgbe_hw *hw = &adapter->hw;
2555 struct net_device *netdev = adapter->netdev;
2556
2557 if (hw->mac.ops.reset_hw(hw)) {
2558 hw_dbg(hw, "PF still resetting\n");
2559 } else {
2560 hw->mac.ops.init_hw(hw);
2561 ixgbevf_negotiate_api(adapter);
2562 }
2563
2564 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
2565 eth_hw_addr_set(netdev, adapter->hw.mac.addr);
2566 ether_addr_copy(netdev->perm_addr, adapter->hw.mac.addr);
2567 }
2568
2569 adapter->last_reset = jiffies;
2570}
2571
2572static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
2573 int vectors)
2574{
2575 int vector_threshold;
2576
2577 /* We'll want at least 2 (vector_threshold):
2578 * 1) TxQ[0] + RxQ[0] handler
2579 * 2) Other (Link Status Change, etc.)
2580 */
2581 vector_threshold = MIN_MSIX_COUNT;
2582
2583 /* The more we get, the more we will assign to Tx/Rx Cleanup
2584 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2585 * Right now, we simply care about how many we'll get; we'll
2586 * set them up later while requesting irq's.
2587 */
2588 vectors = pci_enable_msix_range(adapter->pdev, adapter->msix_entries,
2589 vector_threshold, vectors);
2590
2591 if (vectors < 0) {
2592 dev_err(&adapter->pdev->dev,
2593 "Unable to allocate MSI-X interrupts\n");
2594 kfree(adapter->msix_entries);
2595 adapter->msix_entries = NULL;
2596 return vectors;
2597 }
2598
2599 /* Adjust for only the vectors we'll use, which is minimum
2600 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2601 * vectors we were allocated.
2602 */
2603 adapter->num_msix_vectors = vectors;
2604
2605 return 0;
2606}
2607
2608/**
2609 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2610 * @adapter: board private structure to initialize
2611 *
2612 * This is the top level queue allocation routine. The order here is very
2613 * important, starting with the "most" number of features turned on at once,
2614 * and ending with the smallest set of features. This way large combinations
2615 * can be allocated if they're turned on, and smaller combinations are the
2616 * fall through conditions.
2617 *
2618 **/
2619static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
2620{
2621 struct ixgbe_hw *hw = &adapter->hw;
2622 unsigned int def_q = 0;
2623 unsigned int num_tcs = 0;
2624 int err;
2625
2626 /* Start with base case */
2627 adapter->num_rx_queues = 1;
2628 adapter->num_tx_queues = 1;
2629 adapter->num_xdp_queues = 0;
2630
2631 spin_lock_bh(&adapter->mbx_lock);
2632
2633 /* fetch queue configuration from the PF */
2634 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2635
2636 spin_unlock_bh(&adapter->mbx_lock);
2637
2638 if (err)
2639 return;
2640
2641 /* we need as many queues as traffic classes */
2642 if (num_tcs > 1) {
2643 adapter->num_rx_queues = num_tcs;
2644 } else {
2645 u16 rss = min_t(u16, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES);
2646
2647 switch (hw->api_version) {
2648 case ixgbe_mbox_api_11:
2649 case ixgbe_mbox_api_12:
2650 case ixgbe_mbox_api_13:
2651 case ixgbe_mbox_api_14:
2652 case ixgbe_mbox_api_15:
2653 if (adapter->xdp_prog &&
2654 hw->mac.max_tx_queues == rss)
2655 rss = rss > 3 ? 2 : 1;
2656
2657 adapter->num_rx_queues = rss;
2658 adapter->num_tx_queues = rss;
2659 adapter->num_xdp_queues = adapter->xdp_prog ? rss : 0;
2660 break;
2661 default:
2662 break;
2663 }
2664 }
2665}
2666
2667/**
2668 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2669 * @adapter: board private structure to initialize
2670 *
2671 * Attempt to configure the interrupts using the best available
2672 * capabilities of the hardware and the kernel.
2673 **/
2674static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
2675{
2676 int vector, v_budget;
2677
2678 /* It's easy to be greedy for MSI-X vectors, but it really
2679 * doesn't do us much good if we have a lot more vectors
2680 * than CPU's. So let's be conservative and only ask for
2681 * (roughly) the same number of vectors as there are CPU's.
2682 * The default is to use pairs of vectors.
2683 */
2684 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
2685 v_budget = min_t(int, v_budget, num_online_cpus());
2686 v_budget += NON_Q_VECTORS;
2687
2688 adapter->msix_entries = kcalloc(v_budget,
2689 sizeof(struct msix_entry), GFP_KERNEL);
2690 if (!adapter->msix_entries)
2691 return -ENOMEM;
2692
2693 for (vector = 0; vector < v_budget; vector++)
2694 adapter->msix_entries[vector].entry = vector;
2695
2696 /* A failure in MSI-X entry allocation isn't fatal, but the VF driver
2697 * does not support any other modes, so we will simply fail here. Note
2698 * that we clean up the msix_entries pointer else-where.
2699 */
2700 return ixgbevf_acquire_msix_vectors(adapter, v_budget);
2701}
2702
2703static void ixgbevf_add_ring(struct ixgbevf_ring *ring,
2704 struct ixgbevf_ring_container *head)
2705{
2706 ring->next = head->ring;
2707 head->ring = ring;
2708 head->count++;
2709}
2710
2711/**
2712 * ixgbevf_alloc_q_vector - Allocate memory for a single interrupt vector
2713 * @adapter: board private structure to initialize
2714 * @v_idx: index of vector in adapter struct
2715 * @txr_count: number of Tx rings for q vector
2716 * @txr_idx: index of first Tx ring to assign
2717 * @xdp_count: total number of XDP rings to allocate
2718 * @xdp_idx: index of first XDP ring to allocate
2719 * @rxr_count: number of Rx rings for q vector
2720 * @rxr_idx: index of first Rx ring to assign
2721 *
2722 * We allocate one q_vector. If allocation fails we return -ENOMEM.
2723 **/
2724static int ixgbevf_alloc_q_vector(struct ixgbevf_adapter *adapter, int v_idx,
2725 int txr_count, int txr_idx,
2726 int xdp_count, int xdp_idx,
2727 int rxr_count, int rxr_idx)
2728{
2729 struct ixgbevf_q_vector *q_vector;
2730 int reg_idx = txr_idx + xdp_idx;
2731 struct ixgbevf_ring *ring;
2732 int ring_count, size;
2733
2734 ring_count = txr_count + xdp_count + rxr_count;
2735 size = sizeof(*q_vector) + (sizeof(*ring) * ring_count);
2736
2737 /* allocate q_vector and rings */
2738 q_vector = kzalloc(size, GFP_KERNEL);
2739 if (!q_vector)
2740 return -ENOMEM;
2741
2742 /* initialize NAPI */
2743 netif_napi_add(adapter->netdev, &q_vector->napi, ixgbevf_poll);
2744
2745 /* tie q_vector and adapter together */
2746 adapter->q_vector[v_idx] = q_vector;
2747 q_vector->adapter = adapter;
2748 q_vector->v_idx = v_idx;
2749
2750 /* initialize pointer to rings */
2751 ring = q_vector->ring;
2752
2753 while (txr_count) {
2754 /* assign generic ring traits */
2755 ring->dev = &adapter->pdev->dev;
2756 ring->netdev = adapter->netdev;
2757
2758 /* configure backlink on ring */
2759 ring->q_vector = q_vector;
2760
2761 /* update q_vector Tx values */
2762 ixgbevf_add_ring(ring, &q_vector->tx);
2763
2764 /* apply Tx specific ring traits */
2765 ring->count = adapter->tx_ring_count;
2766 ring->queue_index = txr_idx;
2767 ring->reg_idx = reg_idx;
2768
2769 /* assign ring to adapter */
2770 adapter->tx_ring[txr_idx] = ring;
2771
2772 /* update count and index */
2773 txr_count--;
2774 txr_idx++;
2775 reg_idx++;
2776
2777 /* push pointer to next ring */
2778 ring++;
2779 }
2780
2781 while (xdp_count) {
2782 /* assign generic ring traits */
2783 ring->dev = &adapter->pdev->dev;
2784 ring->netdev = adapter->netdev;
2785
2786 /* configure backlink on ring */
2787 ring->q_vector = q_vector;
2788
2789 /* update q_vector Tx values */
2790 ixgbevf_add_ring(ring, &q_vector->tx);
2791
2792 /* apply Tx specific ring traits */
2793 ring->count = adapter->tx_ring_count;
2794 ring->queue_index = xdp_idx;
2795 ring->reg_idx = reg_idx;
2796 set_ring_xdp(ring);
2797
2798 /* assign ring to adapter */
2799 adapter->xdp_ring[xdp_idx] = ring;
2800
2801 /* update count and index */
2802 xdp_count--;
2803 xdp_idx++;
2804 reg_idx++;
2805
2806 /* push pointer to next ring */
2807 ring++;
2808 }
2809
2810 while (rxr_count) {
2811 /* assign generic ring traits */
2812 ring->dev = &adapter->pdev->dev;
2813 ring->netdev = adapter->netdev;
2814
2815 /* configure backlink on ring */
2816 ring->q_vector = q_vector;
2817
2818 /* update q_vector Rx values */
2819 ixgbevf_add_ring(ring, &q_vector->rx);
2820
2821 /* apply Rx specific ring traits */
2822 ring->count = adapter->rx_ring_count;
2823 ring->queue_index = rxr_idx;
2824 ring->reg_idx = rxr_idx;
2825
2826 /* assign ring to adapter */
2827 adapter->rx_ring[rxr_idx] = ring;
2828
2829 /* update count and index */
2830 rxr_count--;
2831 rxr_idx++;
2832
2833 /* push pointer to next ring */
2834 ring++;
2835 }
2836
2837 return 0;
2838}
2839
2840/**
2841 * ixgbevf_free_q_vector - Free memory allocated for specific interrupt vector
2842 * @adapter: board private structure to initialize
2843 * @v_idx: index of vector in adapter struct
2844 *
2845 * This function frees the memory allocated to the q_vector. In addition if
2846 * NAPI is enabled it will delete any references to the NAPI struct prior
2847 * to freeing the q_vector.
2848 **/
2849static void ixgbevf_free_q_vector(struct ixgbevf_adapter *adapter, int v_idx)
2850{
2851 struct ixgbevf_q_vector *q_vector = adapter->q_vector[v_idx];
2852 struct ixgbevf_ring *ring;
2853
2854 ixgbevf_for_each_ring(ring, q_vector->tx) {
2855 if (ring_is_xdp(ring))
2856 adapter->xdp_ring[ring->queue_index] = NULL;
2857 else
2858 adapter->tx_ring[ring->queue_index] = NULL;
2859 }
2860
2861 ixgbevf_for_each_ring(ring, q_vector->rx)
2862 adapter->rx_ring[ring->queue_index] = NULL;
2863
2864 adapter->q_vector[v_idx] = NULL;
2865 netif_napi_del(&q_vector->napi);
2866
2867 /* ixgbevf_get_stats() might access the rings on this vector,
2868 * we must wait a grace period before freeing it.
2869 */
2870 kfree_rcu(q_vector, rcu);
2871}
2872
2873/**
2874 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2875 * @adapter: board private structure to initialize
2876 *
2877 * We allocate one q_vector per queue interrupt. If allocation fails we
2878 * return -ENOMEM.
2879 **/
2880static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2881{
2882 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2883 int rxr_remaining = adapter->num_rx_queues;
2884 int txr_remaining = adapter->num_tx_queues;
2885 int xdp_remaining = adapter->num_xdp_queues;
2886 int rxr_idx = 0, txr_idx = 0, xdp_idx = 0, v_idx = 0;
2887 int err;
2888
2889 if (q_vectors >= (rxr_remaining + txr_remaining + xdp_remaining)) {
2890 for (; rxr_remaining; v_idx++, q_vectors--) {
2891 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2892
2893 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2894 0, 0, 0, 0, rqpv, rxr_idx);
2895 if (err)
2896 goto err_out;
2897
2898 /* update counts and index */
2899 rxr_remaining -= rqpv;
2900 rxr_idx += rqpv;
2901 }
2902 }
2903
2904 for (; q_vectors; v_idx++, q_vectors--) {
2905 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors);
2906 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors);
2907 int xqpv = DIV_ROUND_UP(xdp_remaining, q_vectors);
2908
2909 err = ixgbevf_alloc_q_vector(adapter, v_idx,
2910 tqpv, txr_idx,
2911 xqpv, xdp_idx,
2912 rqpv, rxr_idx);
2913
2914 if (err)
2915 goto err_out;
2916
2917 /* update counts and index */
2918 rxr_remaining -= rqpv;
2919 rxr_idx += rqpv;
2920 txr_remaining -= tqpv;
2921 txr_idx += tqpv;
2922 xdp_remaining -= xqpv;
2923 xdp_idx += xqpv;
2924 }
2925
2926 return 0;
2927
2928err_out:
2929 while (v_idx) {
2930 v_idx--;
2931 ixgbevf_free_q_vector(adapter, v_idx);
2932 }
2933
2934 return -ENOMEM;
2935}
2936
2937/**
2938 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2939 * @adapter: board private structure to initialize
2940 *
2941 * This function frees the memory allocated to the q_vectors. In addition if
2942 * NAPI is enabled it will delete any references to the NAPI struct prior
2943 * to freeing the q_vector.
2944 **/
2945static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2946{
2947 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2948
2949 while (q_vectors) {
2950 q_vectors--;
2951 ixgbevf_free_q_vector(adapter, q_vectors);
2952 }
2953}
2954
2955/**
2956 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2957 * @adapter: board private structure
2958 *
2959 **/
2960static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2961{
2962 if (!adapter->msix_entries)
2963 return;
2964
2965 pci_disable_msix(adapter->pdev);
2966 kfree(adapter->msix_entries);
2967 adapter->msix_entries = NULL;
2968}
2969
2970/**
2971 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2972 * @adapter: board private structure to initialize
2973 *
2974 **/
2975static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2976{
2977 int err;
2978
2979 /* Number of supported queues */
2980 ixgbevf_set_num_queues(adapter);
2981
2982 err = ixgbevf_set_interrupt_capability(adapter);
2983 if (err) {
2984 hw_dbg(&adapter->hw,
2985 "Unable to setup interrupt capabilities\n");
2986 goto err_set_interrupt;
2987 }
2988
2989 err = ixgbevf_alloc_q_vectors(adapter);
2990 if (err) {
2991 hw_dbg(&adapter->hw, "Unable to allocate memory for queue vectors\n");
2992 goto err_alloc_q_vectors;
2993 }
2994
2995 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u XDP Queue count %u\n",
2996 (adapter->num_rx_queues > 1) ? "Enabled" : "Disabled",
2997 adapter->num_rx_queues, adapter->num_tx_queues,
2998 adapter->num_xdp_queues);
2999
3000 set_bit(__IXGBEVF_DOWN, &adapter->state);
3001
3002 return 0;
3003err_alloc_q_vectors:
3004 ixgbevf_reset_interrupt_capability(adapter);
3005err_set_interrupt:
3006 return err;
3007}
3008
3009/**
3010 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
3011 * @adapter: board private structure to clear interrupt scheme on
3012 *
3013 * We go through and clear interrupt specific resources and reset the structure
3014 * to pre-load conditions
3015 **/
3016static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
3017{
3018 adapter->num_tx_queues = 0;
3019 adapter->num_xdp_queues = 0;
3020 adapter->num_rx_queues = 0;
3021
3022 ixgbevf_free_q_vectors(adapter);
3023 ixgbevf_reset_interrupt_capability(adapter);
3024}
3025
3026/**
3027 * ixgbevf_sw_init - Initialize general software structures
3028 * @adapter: board private structure to initialize
3029 *
3030 * ixgbevf_sw_init initializes the Adapter private data structure.
3031 * Fields are initialized based on PCI device information and
3032 * OS network device settings (MTU size).
3033 **/
3034static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
3035{
3036 struct ixgbe_hw *hw = &adapter->hw;
3037 struct pci_dev *pdev = adapter->pdev;
3038 struct net_device *netdev = adapter->netdev;
3039 int err;
3040
3041 /* PCI config space info */
3042 hw->vendor_id = pdev->vendor;
3043 hw->device_id = pdev->device;
3044 hw->revision_id = pdev->revision;
3045 hw->subsystem_vendor_id = pdev->subsystem_vendor;
3046 hw->subsystem_device_id = pdev->subsystem_device;
3047
3048 hw->mbx.ops.init_params(hw);
3049
3050 if (hw->mac.type >= ixgbe_mac_X550_vf) {
3051 err = ixgbevf_init_rss_key(adapter);
3052 if (err)
3053 goto out;
3054 }
3055
3056 /* assume legacy case in which PF would only give VF 2 queues */
3057 hw->mac.max_tx_queues = 2;
3058 hw->mac.max_rx_queues = 2;
3059
3060 /* lock to protect mailbox accesses */
3061 spin_lock_init(&adapter->mbx_lock);
3062
3063 err = hw->mac.ops.reset_hw(hw);
3064 if (err) {
3065 dev_info(&pdev->dev,
3066 "PF still in reset state. Is the PF interface up?\n");
3067 } else {
3068 err = hw->mac.ops.init_hw(hw);
3069 if (err) {
3070 pr_err("init_shared_code failed: %d\n", err);
3071 goto out;
3072 }
3073 ixgbevf_negotiate_api(adapter);
3074 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
3075 if (err)
3076 dev_info(&pdev->dev, "Error reading MAC address\n");
3077 else if (is_zero_ether_addr(adapter->hw.mac.addr))
3078 dev_info(&pdev->dev,
3079 "MAC address not assigned by administrator.\n");
3080 eth_hw_addr_set(netdev, hw->mac.addr);
3081 }
3082
3083 if (!is_valid_ether_addr(netdev->dev_addr)) {
3084 dev_info(&pdev->dev, "Assigning random MAC address\n");
3085 eth_hw_addr_random(netdev);
3086 ether_addr_copy(hw->mac.addr, netdev->dev_addr);
3087 ether_addr_copy(hw->mac.perm_addr, netdev->dev_addr);
3088 }
3089
3090 /* Enable dynamic interrupt throttling rates */
3091 adapter->rx_itr_setting = 1;
3092 adapter->tx_itr_setting = 1;
3093
3094 /* set default ring sizes */
3095 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
3096 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
3097
3098 adapter->link_state = true;
3099
3100 set_bit(__IXGBEVF_DOWN, &adapter->state);
3101 return 0;
3102
3103out:
3104 return err;
3105}
3106
3107#define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
3108 { \
3109 u32 current_counter = IXGBE_READ_REG(hw, reg); \
3110 if (current_counter < last_counter) \
3111 counter += 0x100000000LL; \
3112 last_counter = current_counter; \
3113 counter &= 0xFFFFFFFF00000000LL; \
3114 counter |= current_counter; \
3115 }
3116
3117#define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
3118 { \
3119 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
3120 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
3121 u64 current_counter = (current_counter_msb << 32) | \
3122 current_counter_lsb; \
3123 if (current_counter < last_counter) \
3124 counter += 0x1000000000LL; \
3125 last_counter = current_counter; \
3126 counter &= 0xFFFFFFF000000000LL; \
3127 counter |= current_counter; \
3128 }
3129/**
3130 * ixgbevf_update_stats - Update the board statistics counters.
3131 * @adapter: board private structure
3132 **/
3133void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
3134{
3135 struct ixgbe_hw *hw = &adapter->hw;
3136 u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
3137 u64 alloc_rx_page = 0, hw_csum_rx_error = 0;
3138 int i;
3139
3140 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3141 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3142 return;
3143
3144 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
3145 adapter->stats.vfgprc);
3146 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
3147 adapter->stats.vfgptc);
3148 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
3149 adapter->stats.last_vfgorc,
3150 adapter->stats.vfgorc);
3151 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
3152 adapter->stats.last_vfgotc,
3153 adapter->stats.vfgotc);
3154 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
3155 adapter->stats.vfmprc);
3156
3157 for (i = 0; i < adapter->num_rx_queues; i++) {
3158 struct ixgbevf_ring *rx_ring = adapter->rx_ring[i];
3159
3160 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
3161 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
3162 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
3163 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
3164 }
3165
3166 adapter->hw_csum_rx_error = hw_csum_rx_error;
3167 adapter->alloc_rx_page_failed = alloc_rx_page_failed;
3168 adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
3169 adapter->alloc_rx_page = alloc_rx_page;
3170}
3171
3172/**
3173 * ixgbevf_service_timer - Timer Call-back
3174 * @t: pointer to timer_list struct
3175 **/
3176static void ixgbevf_service_timer(struct timer_list *t)
3177{
3178 struct ixgbevf_adapter *adapter = from_timer(adapter, t,
3179 service_timer);
3180
3181 /* Reset the timer */
3182 mod_timer(&adapter->service_timer, (HZ * 2) + jiffies);
3183
3184 ixgbevf_service_event_schedule(adapter);
3185}
3186
3187static void ixgbevf_reset_subtask(struct ixgbevf_adapter *adapter)
3188{
3189 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state))
3190 return;
3191
3192 rtnl_lock();
3193 /* If we're already down or resetting, just bail */
3194 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3195 test_bit(__IXGBEVF_REMOVING, &adapter->state) ||
3196 test_bit(__IXGBEVF_RESETTING, &adapter->state)) {
3197 rtnl_unlock();
3198 return;
3199 }
3200
3201 adapter->tx_timeout_count++;
3202
3203 ixgbevf_reinit_locked(adapter);
3204 rtnl_unlock();
3205}
3206
3207/**
3208 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
3209 * @adapter: pointer to the device adapter structure
3210 *
3211 * This function serves two purposes. First it strobes the interrupt lines
3212 * in order to make certain interrupts are occurring. Secondly it sets the
3213 * bits needed to check for TX hangs. As a result we should immediately
3214 * determine if a hang has occurred.
3215 **/
3216static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter *adapter)
3217{
3218 struct ixgbe_hw *hw = &adapter->hw;
3219 u32 eics = 0;
3220 int i;
3221
3222 /* If we're down or resetting, just bail */
3223 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3224 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3225 return;
3226
3227 /* Force detection of hung controller */
3228 if (netif_carrier_ok(adapter->netdev)) {
3229 for (i = 0; i < adapter->num_tx_queues; i++)
3230 set_check_for_tx_hang(adapter->tx_ring[i]);
3231 for (i = 0; i < adapter->num_xdp_queues; i++)
3232 set_check_for_tx_hang(adapter->xdp_ring[i]);
3233 }
3234
3235 /* get one bit for every active Tx/Rx interrupt vector */
3236 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
3237 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
3238
3239 if (qv->rx.ring || qv->tx.ring)
3240 eics |= BIT(i);
3241 }
3242
3243 /* Cause software interrupt to ensure rings are cleaned */
3244 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
3245}
3246
3247/**
3248 * ixgbevf_watchdog_update_link - update the link status
3249 * @adapter: pointer to the device adapter structure
3250 **/
3251static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter *adapter)
3252{
3253 struct ixgbe_hw *hw = &adapter->hw;
3254 u32 link_speed = adapter->link_speed;
3255 bool link_up = adapter->link_up;
3256 s32 err;
3257
3258 spin_lock_bh(&adapter->mbx_lock);
3259
3260 err = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3261
3262 spin_unlock_bh(&adapter->mbx_lock);
3263
3264 /* if check for link returns error we will need to reset */
3265 if (err && time_after(jiffies, adapter->last_reset + (10 * HZ))) {
3266 set_bit(__IXGBEVF_RESET_REQUESTED, &adapter->state);
3267 link_up = false;
3268 }
3269
3270 adapter->link_up = link_up;
3271 adapter->link_speed = link_speed;
3272}
3273
3274/**
3275 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
3276 * print link up message
3277 * @adapter: pointer to the device adapter structure
3278 **/
3279static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter *adapter)
3280{
3281 struct net_device *netdev = adapter->netdev;
3282
3283 /* only continue if link was previously down */
3284 if (netif_carrier_ok(netdev))
3285 return;
3286
3287 dev_info(&adapter->pdev->dev, "NIC Link is Up %s\n",
3288 (adapter->link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
3289 "10 Gbps" :
3290 (adapter->link_speed == IXGBE_LINK_SPEED_1GB_FULL) ?
3291 "1 Gbps" :
3292 (adapter->link_speed == IXGBE_LINK_SPEED_100_FULL) ?
3293 "100 Mbps" :
3294 "unknown speed");
3295
3296 netif_carrier_on(netdev);
3297}
3298
3299/**
3300 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
3301 * print link down message
3302 * @adapter: pointer to the adapter structure
3303 **/
3304static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter *adapter)
3305{
3306 struct net_device *netdev = adapter->netdev;
3307
3308 adapter->link_speed = 0;
3309
3310 /* only continue if link was up previously */
3311 if (!netif_carrier_ok(netdev))
3312 return;
3313
3314 dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
3315
3316 netif_carrier_off(netdev);
3317}
3318
3319/**
3320 * ixgbevf_watchdog_subtask - worker thread to bring link up
3321 * @adapter: board private structure
3322 **/
3323static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter *adapter)
3324{
3325 /* if interface is down do nothing */
3326 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3327 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3328 return;
3329
3330 ixgbevf_watchdog_update_link(adapter);
3331
3332 if (adapter->link_up && adapter->link_state)
3333 ixgbevf_watchdog_link_is_up(adapter);
3334 else
3335 ixgbevf_watchdog_link_is_down(adapter);
3336
3337 ixgbevf_update_stats(adapter);
3338}
3339
3340/**
3341 * ixgbevf_service_task - manages and runs subtasks
3342 * @work: pointer to work_struct containing our data
3343 **/
3344static void ixgbevf_service_task(struct work_struct *work)
3345{
3346 struct ixgbevf_adapter *adapter = container_of(work,
3347 struct ixgbevf_adapter,
3348 service_task);
3349 struct ixgbe_hw *hw = &adapter->hw;
3350
3351 if (IXGBE_REMOVED(hw->hw_addr)) {
3352 if (!test_bit(__IXGBEVF_DOWN, &adapter->state)) {
3353 rtnl_lock();
3354 ixgbevf_down(adapter);
3355 rtnl_unlock();
3356 }
3357 return;
3358 }
3359
3360 ixgbevf_queue_reset_subtask(adapter);
3361 ixgbevf_reset_subtask(adapter);
3362 ixgbevf_watchdog_subtask(adapter);
3363 ixgbevf_check_hang_subtask(adapter);
3364
3365 ixgbevf_service_event_complete(adapter);
3366}
3367
3368/**
3369 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
3370 * @tx_ring: Tx descriptor ring for a specific queue
3371 *
3372 * Free all transmit software resources
3373 **/
3374void ixgbevf_free_tx_resources(struct ixgbevf_ring *tx_ring)
3375{
3376 ixgbevf_clean_tx_ring(tx_ring);
3377
3378 vfree(tx_ring->tx_buffer_info);
3379 tx_ring->tx_buffer_info = NULL;
3380
3381 /* if not set, then don't free */
3382 if (!tx_ring->desc)
3383 return;
3384
3385 dma_free_coherent(tx_ring->dev, tx_ring->size, tx_ring->desc,
3386 tx_ring->dma);
3387
3388 tx_ring->desc = NULL;
3389}
3390
3391/**
3392 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
3393 * @adapter: board private structure
3394 *
3395 * Free all transmit software resources
3396 **/
3397static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
3398{
3399 int i;
3400
3401 for (i = 0; i < adapter->num_tx_queues; i++)
3402 if (adapter->tx_ring[i]->desc)
3403 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3404 for (i = 0; i < adapter->num_xdp_queues; i++)
3405 if (adapter->xdp_ring[i]->desc)
3406 ixgbevf_free_tx_resources(adapter->xdp_ring[i]);
3407}
3408
3409/**
3410 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
3411 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
3412 *
3413 * Return 0 on success, negative on failure
3414 **/
3415int ixgbevf_setup_tx_resources(struct ixgbevf_ring *tx_ring)
3416{
3417 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3418 int size;
3419
3420 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
3421 tx_ring->tx_buffer_info = vmalloc(size);
3422 if (!tx_ring->tx_buffer_info)
3423 goto err;
3424
3425 u64_stats_init(&tx_ring->syncp);
3426
3427 /* round up to nearest 4K */
3428 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
3429 tx_ring->size = ALIGN(tx_ring->size, 4096);
3430
3431 tx_ring->desc = dma_alloc_coherent(tx_ring->dev, tx_ring->size,
3432 &tx_ring->dma, GFP_KERNEL);
3433 if (!tx_ring->desc)
3434 goto err;
3435
3436 return 0;
3437
3438err:
3439 vfree(tx_ring->tx_buffer_info);
3440 tx_ring->tx_buffer_info = NULL;
3441 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit descriptor ring\n");
3442 return -ENOMEM;
3443}
3444
3445/**
3446 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3447 * @adapter: board private structure
3448 *
3449 * If this function returns with an error, then it's possible one or
3450 * more of the rings is populated (while the rest are not). It is the
3451 * callers duty to clean those orphaned rings.
3452 *
3453 * Return 0 on success, negative on failure
3454 **/
3455static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
3456{
3457 int i, j = 0, err = 0;
3458
3459 for (i = 0; i < adapter->num_tx_queues; i++) {
3460 err = ixgbevf_setup_tx_resources(adapter->tx_ring[i]);
3461 if (!err)
3462 continue;
3463 hw_dbg(&adapter->hw, "Allocation for Tx Queue %u failed\n", i);
3464 goto err_setup_tx;
3465 }
3466
3467 for (j = 0; j < adapter->num_xdp_queues; j++) {
3468 err = ixgbevf_setup_tx_resources(adapter->xdp_ring[j]);
3469 if (!err)
3470 continue;
3471 hw_dbg(&adapter->hw, "Allocation for XDP Queue %u failed\n", j);
3472 goto err_setup_tx;
3473 }
3474
3475 return 0;
3476err_setup_tx:
3477 /* rewind the index freeing the rings as we go */
3478 while (j--)
3479 ixgbevf_free_tx_resources(adapter->xdp_ring[j]);
3480 while (i--)
3481 ixgbevf_free_tx_resources(adapter->tx_ring[i]);
3482
3483 return err;
3484}
3485
3486/**
3487 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3488 * @adapter: board private structure
3489 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3490 *
3491 * Returns 0 on success, negative on failure
3492 **/
3493int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
3494 struct ixgbevf_ring *rx_ring)
3495{
3496 int size;
3497
3498 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
3499 rx_ring->rx_buffer_info = vmalloc(size);
3500 if (!rx_ring->rx_buffer_info)
3501 goto err;
3502
3503 u64_stats_init(&rx_ring->syncp);
3504
3505 /* Round up to nearest 4K */
3506 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
3507 rx_ring->size = ALIGN(rx_ring->size, 4096);
3508
3509 rx_ring->desc = dma_alloc_coherent(rx_ring->dev, rx_ring->size,
3510 &rx_ring->dma, GFP_KERNEL);
3511
3512 if (!rx_ring->desc)
3513 goto err;
3514
3515 /* XDP RX-queue info */
3516 if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
3517 rx_ring->queue_index, 0) < 0)
3518 goto err;
3519
3520 rx_ring->xdp_prog = adapter->xdp_prog;
3521
3522 return 0;
3523err:
3524 vfree(rx_ring->rx_buffer_info);
3525 rx_ring->rx_buffer_info = NULL;
3526 dev_err(rx_ring->dev, "Unable to allocate memory for the Rx descriptor ring\n");
3527 return -ENOMEM;
3528}
3529
3530/**
3531 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3532 * @adapter: board private structure
3533 *
3534 * If this function returns with an error, then it's possible one or
3535 * more of the rings is populated (while the rest are not). It is the
3536 * callers duty to clean those orphaned rings.
3537 *
3538 * Return 0 on success, negative on failure
3539 **/
3540static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
3541{
3542 int i, err = 0;
3543
3544 for (i = 0; i < adapter->num_rx_queues; i++) {
3545 err = ixgbevf_setup_rx_resources(adapter, adapter->rx_ring[i]);
3546 if (!err)
3547 continue;
3548 hw_dbg(&adapter->hw, "Allocation for Rx Queue %u failed\n", i);
3549 goto err_setup_rx;
3550 }
3551
3552 return 0;
3553err_setup_rx:
3554 /* rewind the index freeing the rings as we go */
3555 while (i--)
3556 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3557 return err;
3558}
3559
3560/**
3561 * ixgbevf_free_rx_resources - Free Rx Resources
3562 * @rx_ring: ring to clean the resources from
3563 *
3564 * Free all receive software resources
3565 **/
3566void ixgbevf_free_rx_resources(struct ixgbevf_ring *rx_ring)
3567{
3568 ixgbevf_clean_rx_ring(rx_ring);
3569
3570 rx_ring->xdp_prog = NULL;
3571 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
3572 vfree(rx_ring->rx_buffer_info);
3573 rx_ring->rx_buffer_info = NULL;
3574
3575 dma_free_coherent(rx_ring->dev, rx_ring->size, rx_ring->desc,
3576 rx_ring->dma);
3577
3578 rx_ring->desc = NULL;
3579}
3580
3581/**
3582 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3583 * @adapter: board private structure
3584 *
3585 * Free all receive software resources
3586 **/
3587static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
3588{
3589 int i;
3590
3591 for (i = 0; i < adapter->num_rx_queues; i++)
3592 if (adapter->rx_ring[i]->desc)
3593 ixgbevf_free_rx_resources(adapter->rx_ring[i]);
3594}
3595
3596/**
3597 * ixgbevf_open - Called when a network interface is made active
3598 * @netdev: network interface device structure
3599 *
3600 * Returns 0 on success, negative value on failure
3601 *
3602 * The open entry point is called when a network interface is made
3603 * active by the system (IFF_UP). At this point all resources needed
3604 * for transmit and receive operations are allocated, the interrupt
3605 * handler is registered with the OS, the watchdog timer is started,
3606 * and the stack is notified that the interface is ready.
3607 **/
3608int ixgbevf_open(struct net_device *netdev)
3609{
3610 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3611 struct ixgbe_hw *hw = &adapter->hw;
3612 int err;
3613
3614 /* A previous failure to open the device because of a lack of
3615 * available MSIX vector resources may have reset the number
3616 * of msix vectors variable to zero. The only way to recover
3617 * is to unload/reload the driver and hope that the system has
3618 * been able to recover some MSIX vector resources.
3619 */
3620 if (!adapter->num_msix_vectors)
3621 return -ENOMEM;
3622
3623 if (hw->adapter_stopped) {
3624 ixgbevf_reset(adapter);
3625 /* if adapter is still stopped then PF isn't up and
3626 * the VF can't start.
3627 */
3628 if (hw->adapter_stopped) {
3629 err = IXGBE_ERR_MBX;
3630 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3631 goto err_setup_reset;
3632 }
3633 }
3634
3635 /* disallow open during test */
3636 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
3637 return -EBUSY;
3638
3639 netif_carrier_off(netdev);
3640
3641 /* allocate transmit descriptors */
3642 err = ixgbevf_setup_all_tx_resources(adapter);
3643 if (err)
3644 goto err_setup_tx;
3645
3646 /* allocate receive descriptors */
3647 err = ixgbevf_setup_all_rx_resources(adapter);
3648 if (err)
3649 goto err_setup_rx;
3650
3651 ixgbevf_configure(adapter);
3652
3653 err = ixgbevf_request_irq(adapter);
3654 if (err)
3655 goto err_req_irq;
3656
3657 /* Notify the stack of the actual queue counts. */
3658 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
3659 if (err)
3660 goto err_set_queues;
3661
3662 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
3663 if (err)
3664 goto err_set_queues;
3665
3666 ixgbevf_up_complete(adapter);
3667
3668 return 0;
3669
3670err_set_queues:
3671 ixgbevf_free_irq(adapter);
3672err_req_irq:
3673 ixgbevf_free_all_rx_resources(adapter);
3674err_setup_rx:
3675 ixgbevf_free_all_tx_resources(adapter);
3676err_setup_tx:
3677 ixgbevf_reset(adapter);
3678err_setup_reset:
3679
3680 return err;
3681}
3682
3683/**
3684 * ixgbevf_close_suspend - actions necessary to both suspend and close flows
3685 * @adapter: the private adapter struct
3686 *
3687 * This function should contain the necessary work common to both suspending
3688 * and closing of the device.
3689 */
3690static void ixgbevf_close_suspend(struct ixgbevf_adapter *adapter)
3691{
3692 ixgbevf_down(adapter);
3693 ixgbevf_free_irq(adapter);
3694 ixgbevf_free_all_tx_resources(adapter);
3695 ixgbevf_free_all_rx_resources(adapter);
3696}
3697
3698/**
3699 * ixgbevf_close - Disables a network interface
3700 * @netdev: network interface device structure
3701 *
3702 * Returns 0, this is not allowed to fail
3703 *
3704 * The close entry point is called when an interface is de-activated
3705 * by the OS. The hardware is still under the drivers control, but
3706 * needs to be disabled. A global MAC reset is issued to stop the
3707 * hardware, and all transmit and receive resources are freed.
3708 **/
3709int ixgbevf_close(struct net_device *netdev)
3710{
3711 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3712
3713 if (netif_device_present(netdev))
3714 ixgbevf_close_suspend(adapter);
3715
3716 return 0;
3717}
3718
3719static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter *adapter)
3720{
3721 struct net_device *dev = adapter->netdev;
3722
3723 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED,
3724 &adapter->state))
3725 return;
3726
3727 /* if interface is down do nothing */
3728 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
3729 test_bit(__IXGBEVF_RESETTING, &adapter->state))
3730 return;
3731
3732 /* Hardware has to reinitialize queues and interrupts to
3733 * match packet buffer alignment. Unfortunately, the
3734 * hardware is not flexible enough to do this dynamically.
3735 */
3736 rtnl_lock();
3737
3738 if (netif_running(dev))
3739 ixgbevf_close(dev);
3740
3741 ixgbevf_clear_interrupt_scheme(adapter);
3742 ixgbevf_init_interrupt_scheme(adapter);
3743
3744 if (netif_running(dev))
3745 ixgbevf_open(dev);
3746
3747 rtnl_unlock();
3748}
3749
3750static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
3751 u32 vlan_macip_lens, u32 fceof_saidx,
3752 u32 type_tucmd, u32 mss_l4len_idx)
3753{
3754 struct ixgbe_adv_tx_context_desc *context_desc;
3755 u16 i = tx_ring->next_to_use;
3756
3757 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
3758
3759 i++;
3760 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
3761
3762 /* set bits to identify this as an advanced context descriptor */
3763 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
3764
3765 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3766 context_desc->fceof_saidx = cpu_to_le32(fceof_saidx);
3767 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
3768 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3769}
3770
3771static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
3772 struct ixgbevf_tx_buffer *first,
3773 u8 *hdr_len,
3774 struct ixgbevf_ipsec_tx_data *itd)
3775{
3776 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
3777 struct sk_buff *skb = first->skb;
3778 union {
3779 struct iphdr *v4;
3780 struct ipv6hdr *v6;
3781 unsigned char *hdr;
3782 } ip;
3783 union {
3784 struct tcphdr *tcp;
3785 unsigned char *hdr;
3786 } l4;
3787 u32 paylen, l4_offset;
3788 u32 fceof_saidx = 0;
3789 int err;
3790
3791 if (skb->ip_summed != CHECKSUM_PARTIAL)
3792 return 0;
3793
3794 if (!skb_is_gso(skb))
3795 return 0;
3796
3797 err = skb_cow_head(skb, 0);
3798 if (err < 0)
3799 return err;
3800
3801 if (eth_p_mpls(first->protocol))
3802 ip.hdr = skb_inner_network_header(skb);
3803 else
3804 ip.hdr = skb_network_header(skb);
3805 l4.hdr = skb_checksum_start(skb);
3806
3807 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3808 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3809
3810 /* initialize outer IP header fields */
3811 if (ip.v4->version == 4) {
3812 unsigned char *csum_start = skb_checksum_start(skb);
3813 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
3814 int len = csum_start - trans_start;
3815
3816 /* IP header will have to cancel out any data that
3817 * is not a part of the outer IP header, so set to
3818 * a reverse csum if needed, else init check to 0.
3819 */
3820 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
3821 csum_fold(csum_partial(trans_start,
3822 len, 0)) : 0;
3823 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3824
3825 ip.v4->tot_len = 0;
3826 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3827 IXGBE_TX_FLAGS_CSUM |
3828 IXGBE_TX_FLAGS_IPV4;
3829 } else {
3830 ip.v6->payload_len = 0;
3831 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
3832 IXGBE_TX_FLAGS_CSUM;
3833 }
3834
3835 /* determine offset of inner transport header */
3836 l4_offset = l4.hdr - skb->data;
3837
3838 /* compute length of segmentation header */
3839 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
3840
3841 /* remove payload length from inner checksum */
3842 paylen = skb->len - l4_offset;
3843 csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
3844
3845 /* update gso size and bytecount with header size */
3846 first->gso_segs = skb_shinfo(skb)->gso_segs;
3847 first->bytecount += (first->gso_segs - 1) * *hdr_len;
3848
3849 /* mss_l4len_id: use 1 as index for TSO */
3850 mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
3851 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
3852 mss_l4len_idx |= (1u << IXGBE_ADVTXD_IDX_SHIFT);
3853
3854 fceof_saidx |= itd->pfsa;
3855 type_tucmd |= itd->flags | itd->trailer_len;
3856
3857 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3858 vlan_macip_lens = l4.hdr - ip.hdr;
3859 vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
3860 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3861
3862 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
3863 mss_l4len_idx);
3864
3865 return 1;
3866}
3867
3868static void ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
3869 struct ixgbevf_tx_buffer *first,
3870 struct ixgbevf_ipsec_tx_data *itd)
3871{
3872 struct sk_buff *skb = first->skb;
3873 u32 vlan_macip_lens = 0;
3874 u32 fceof_saidx = 0;
3875 u32 type_tucmd = 0;
3876
3877 if (skb->ip_summed != CHECKSUM_PARTIAL)
3878 goto no_csum;
3879
3880 switch (skb->csum_offset) {
3881 case offsetof(struct tcphdr, check):
3882 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
3883 fallthrough;
3884 case offsetof(struct udphdr, check):
3885 break;
3886 case offsetof(struct sctphdr, checksum):
3887 /* validate that this is actually an SCTP request */
3888 if (skb_csum_is_sctp(skb)) {
3889 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
3890 break;
3891 }
3892 fallthrough;
3893 default:
3894 skb_checksum_help(skb);
3895 goto no_csum;
3896 }
3897
3898 if (first->protocol == htons(ETH_P_IP))
3899 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
3900
3901 /* update TX checksum flag */
3902 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
3903 vlan_macip_lens = skb_checksum_start_offset(skb) -
3904 skb_network_offset(skb);
3905no_csum:
3906 /* vlan_macip_lens: MACLEN, VLAN tag */
3907 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
3908 vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
3909
3910 fceof_saidx |= itd->pfsa;
3911 type_tucmd |= itd->flags | itd->trailer_len;
3912
3913 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
3914 fceof_saidx, type_tucmd, 0);
3915}
3916
3917static __le32 ixgbevf_tx_cmd_type(u32 tx_flags)
3918{
3919 /* set type for advanced descriptor with frame checksum insertion */
3920 __le32 cmd_type = cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA |
3921 IXGBE_ADVTXD_DCMD_IFCS |
3922 IXGBE_ADVTXD_DCMD_DEXT);
3923
3924 /* set HW VLAN bit if VLAN is present */
3925 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3926 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE);
3927
3928 /* set segmentation enable bits for TSO/FSO */
3929 if (tx_flags & IXGBE_TX_FLAGS_TSO)
3930 cmd_type |= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE);
3931
3932 return cmd_type;
3933}
3934
3935static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
3936 u32 tx_flags, unsigned int paylen)
3937{
3938 __le32 olinfo_status = cpu_to_le32(paylen << IXGBE_ADVTXD_PAYLEN_SHIFT);
3939
3940 /* enable L4 checksum for TSO and TX checksum offload */
3941 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3942 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM);
3943
3944 /* enble IPv4 checksum for TSO */
3945 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3946 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM);
3947
3948 /* enable IPsec */
3949 if (tx_flags & IXGBE_TX_FLAGS_IPSEC)
3950 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_POPTS_IPSEC);
3951
3952 /* use index 1 context for TSO/FSO/FCOE/IPSEC */
3953 if (tx_flags & (IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_IPSEC))
3954 olinfo_status |= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT);
3955
3956 /* Check Context must be set if Tx switch is enabled, which it
3957 * always is for case where virtual functions are running
3958 */
3959 olinfo_status |= cpu_to_le32(IXGBE_ADVTXD_CC);
3960
3961 tx_desc->read.olinfo_status = olinfo_status;
3962}
3963
3964static void ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
3965 struct ixgbevf_tx_buffer *first,
3966 const u8 hdr_len)
3967{
3968 struct sk_buff *skb = first->skb;
3969 struct ixgbevf_tx_buffer *tx_buffer;
3970 union ixgbe_adv_tx_desc *tx_desc;
3971 skb_frag_t *frag;
3972 dma_addr_t dma;
3973 unsigned int data_len, size;
3974 u32 tx_flags = first->tx_flags;
3975 __le32 cmd_type = ixgbevf_tx_cmd_type(tx_flags);
3976 u16 i = tx_ring->next_to_use;
3977
3978 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3979
3980 ixgbevf_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
3981
3982 size = skb_headlen(skb);
3983 data_len = skb->data_len;
3984
3985 dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
3986
3987 tx_buffer = first;
3988
3989 for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
3990 if (dma_mapping_error(tx_ring->dev, dma))
3991 goto dma_error;
3992
3993 /* record length, and DMA address */
3994 dma_unmap_len_set(tx_buffer, len, size);
3995 dma_unmap_addr_set(tx_buffer, dma, dma);
3996
3997 tx_desc->read.buffer_addr = cpu_to_le64(dma);
3998
3999 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
4000 tx_desc->read.cmd_type_len =
4001 cmd_type | cpu_to_le32(IXGBE_MAX_DATA_PER_TXD);
4002
4003 i++;
4004 tx_desc++;
4005 if (i == tx_ring->count) {
4006 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4007 i = 0;
4008 }
4009 tx_desc->read.olinfo_status = 0;
4010
4011 dma += IXGBE_MAX_DATA_PER_TXD;
4012 size -= IXGBE_MAX_DATA_PER_TXD;
4013
4014 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4015 }
4016
4017 if (likely(!data_len))
4018 break;
4019
4020 tx_desc->read.cmd_type_len = cmd_type | cpu_to_le32(size);
4021
4022 i++;
4023 tx_desc++;
4024 if (i == tx_ring->count) {
4025 tx_desc = IXGBEVF_TX_DESC(tx_ring, 0);
4026 i = 0;
4027 }
4028 tx_desc->read.olinfo_status = 0;
4029
4030 size = skb_frag_size(frag);
4031 data_len -= size;
4032
4033 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
4034 DMA_TO_DEVICE);
4035
4036 tx_buffer = &tx_ring->tx_buffer_info[i];
4037 }
4038
4039 /* write last descriptor with RS and EOP bits */
4040 cmd_type |= cpu_to_le32(size) | cpu_to_le32(IXGBE_TXD_CMD);
4041 tx_desc->read.cmd_type_len = cmd_type;
4042
4043 /* set the timestamp */
4044 first->time_stamp = jiffies;
4045
4046 skb_tx_timestamp(skb);
4047
4048 /* Force memory writes to complete before letting h/w know there
4049 * are new descriptors to fetch. (Only applicable for weak-ordered
4050 * memory model archs, such as IA-64).
4051 *
4052 * We also need this memory barrier (wmb) to make certain all of the
4053 * status bits have been updated before next_to_watch is written.
4054 */
4055 wmb();
4056
4057 /* set next_to_watch value indicating a packet is present */
4058 first->next_to_watch = tx_desc;
4059
4060 i++;
4061 if (i == tx_ring->count)
4062 i = 0;
4063
4064 tx_ring->next_to_use = i;
4065
4066 /* notify HW of packet */
4067 ixgbevf_write_tail(tx_ring, i);
4068
4069 return;
4070dma_error:
4071 dev_err(tx_ring->dev, "TX DMA map failed\n");
4072 tx_buffer = &tx_ring->tx_buffer_info[i];
4073
4074 /* clear dma mappings for failed tx_buffer_info map */
4075 while (tx_buffer != first) {
4076 if (dma_unmap_len(tx_buffer, len))
4077 dma_unmap_page(tx_ring->dev,
4078 dma_unmap_addr(tx_buffer, dma),
4079 dma_unmap_len(tx_buffer, len),
4080 DMA_TO_DEVICE);
4081 dma_unmap_len_set(tx_buffer, len, 0);
4082
4083 if (i-- == 0)
4084 i += tx_ring->count;
4085 tx_buffer = &tx_ring->tx_buffer_info[i];
4086 }
4087
4088 if (dma_unmap_len(tx_buffer, len))
4089 dma_unmap_single(tx_ring->dev,
4090 dma_unmap_addr(tx_buffer, dma),
4091 dma_unmap_len(tx_buffer, len),
4092 DMA_TO_DEVICE);
4093 dma_unmap_len_set(tx_buffer, len, 0);
4094
4095 dev_kfree_skb_any(tx_buffer->skb);
4096 tx_buffer->skb = NULL;
4097
4098 tx_ring->next_to_use = i;
4099}
4100
4101static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4102{
4103 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
4104 /* Herbert's original patch had:
4105 * smp_mb__after_netif_stop_queue();
4106 * but since that doesn't exist yet, just open code it.
4107 */
4108 smp_mb();
4109
4110 /* We need to check again in a case another CPU has just
4111 * made room available.
4112 */
4113 if (likely(ixgbevf_desc_unused(tx_ring) < size))
4114 return -EBUSY;
4115
4116 /* A reprieve! - use start_queue because it doesn't call schedule */
4117 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
4118 ++tx_ring->tx_stats.restart_queue;
4119
4120 return 0;
4121}
4122
4123static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
4124{
4125 if (likely(ixgbevf_desc_unused(tx_ring) >= size))
4126 return 0;
4127 return __ixgbevf_maybe_stop_tx(tx_ring, size);
4128}
4129
4130static int ixgbevf_xmit_frame_ring(struct sk_buff *skb,
4131 struct ixgbevf_ring *tx_ring)
4132{
4133 struct ixgbevf_tx_buffer *first;
4134 int tso;
4135 u32 tx_flags = 0;
4136 u16 count = TXD_USE_COUNT(skb_headlen(skb));
4137 struct ixgbevf_ipsec_tx_data ipsec_tx = { 0 };
4138#if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4139 unsigned short f;
4140#endif
4141 u8 hdr_len = 0;
4142 u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
4143
4144 if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
4145 dev_kfree_skb_any(skb);
4146 return NETDEV_TX_OK;
4147 }
4148
4149 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
4150 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
4151 * + 2 desc gap to keep tail from touching head,
4152 * + 1 desc for context descriptor,
4153 * otherwise try next time
4154 */
4155#if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
4156 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
4157 skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
4158
4159 count += TXD_USE_COUNT(skb_frag_size(frag));
4160 }
4161#else
4162 count += skb_shinfo(skb)->nr_frags;
4163#endif
4164 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
4165 tx_ring->tx_stats.tx_busy++;
4166 return NETDEV_TX_BUSY;
4167 }
4168
4169 /* record the location of the first descriptor for this packet */
4170 first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4171 first->skb = skb;
4172 first->bytecount = skb->len;
4173 first->gso_segs = 1;
4174
4175 if (skb_vlan_tag_present(skb)) {
4176 tx_flags |= skb_vlan_tag_get(skb);
4177 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
4178 tx_flags |= IXGBE_TX_FLAGS_VLAN;
4179 }
4180
4181 /* record initial flags and protocol */
4182 first->tx_flags = tx_flags;
4183 first->protocol = vlan_get_protocol(skb);
4184
4185#ifdef CONFIG_IXGBEVF_IPSEC
4186 if (xfrm_offload(skb) && !ixgbevf_ipsec_tx(tx_ring, first, &ipsec_tx))
4187 goto out_drop;
4188#endif
4189 tso = ixgbevf_tso(tx_ring, first, &hdr_len, &ipsec_tx);
4190 if (tso < 0)
4191 goto out_drop;
4192 else if (!tso)
4193 ixgbevf_tx_csum(tx_ring, first, &ipsec_tx);
4194
4195 ixgbevf_tx_map(tx_ring, first, hdr_len);
4196
4197 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
4198
4199 return NETDEV_TX_OK;
4200
4201out_drop:
4202 dev_kfree_skb_any(first->skb);
4203 first->skb = NULL;
4204
4205 return NETDEV_TX_OK;
4206}
4207
4208static netdev_tx_t ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
4209{
4210 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4211 struct ixgbevf_ring *tx_ring;
4212
4213 if (skb->len <= 0) {
4214 dev_kfree_skb_any(skb);
4215 return NETDEV_TX_OK;
4216 }
4217
4218 /* The minimum packet size for olinfo paylen is 17 so pad the skb
4219 * in order to meet this minimum size requirement.
4220 */
4221 if (skb->len < 17) {
4222 if (skb_padto(skb, 17))
4223 return NETDEV_TX_OK;
4224 skb->len = 17;
4225 }
4226
4227 tx_ring = adapter->tx_ring[skb->queue_mapping];
4228 return ixgbevf_xmit_frame_ring(skb, tx_ring);
4229}
4230
4231/**
4232 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
4233 * @netdev: network interface device structure
4234 * @p: pointer to an address structure
4235 *
4236 * Returns 0 on success, negative on failure
4237 **/
4238static int ixgbevf_set_mac(struct net_device *netdev, void *p)
4239{
4240 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4241 struct ixgbe_hw *hw = &adapter->hw;
4242 struct sockaddr *addr = p;
4243 int err;
4244
4245 if (!is_valid_ether_addr(addr->sa_data))
4246 return -EADDRNOTAVAIL;
4247
4248 spin_lock_bh(&adapter->mbx_lock);
4249
4250 err = hw->mac.ops.set_rar(hw, 0, addr->sa_data, 0);
4251
4252 spin_unlock_bh(&adapter->mbx_lock);
4253
4254 if (err)
4255 return -EPERM;
4256
4257 ether_addr_copy(hw->mac.addr, addr->sa_data);
4258 ether_addr_copy(hw->mac.perm_addr, addr->sa_data);
4259 eth_hw_addr_set(netdev, addr->sa_data);
4260
4261 return 0;
4262}
4263
4264/**
4265 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
4266 * @netdev: network interface device structure
4267 * @new_mtu: new value for maximum frame size
4268 *
4269 * Returns 0 on success, negative on failure
4270 **/
4271static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
4272{
4273 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4274 struct ixgbe_hw *hw = &adapter->hw;
4275 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4276 int ret;
4277
4278 /* prevent MTU being changed to a size unsupported by XDP */
4279 if (adapter->xdp_prog) {
4280 dev_warn(&adapter->pdev->dev, "MTU cannot be changed while XDP program is loaded\n");
4281 return -EPERM;
4282 }
4283
4284 spin_lock_bh(&adapter->mbx_lock);
4285 /* notify the PF of our intent to use this size of frame */
4286 ret = hw->mac.ops.set_rlpml(hw, max_frame);
4287 spin_unlock_bh(&adapter->mbx_lock);
4288 if (ret)
4289 return -EINVAL;
4290
4291 hw_dbg(hw, "changing MTU from %d to %d\n",
4292 netdev->mtu, new_mtu);
4293
4294 /* must set new MTU before calling down or up */
4295 netdev->mtu = new_mtu;
4296
4297 if (netif_running(netdev))
4298 ixgbevf_reinit_locked(adapter);
4299
4300 return 0;
4301}
4302
4303static int __maybe_unused ixgbevf_suspend(struct device *dev_d)
4304{
4305 struct net_device *netdev = dev_get_drvdata(dev_d);
4306 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4307
4308 rtnl_lock();
4309 netif_device_detach(netdev);
4310
4311 if (netif_running(netdev))
4312 ixgbevf_close_suspend(adapter);
4313
4314 ixgbevf_clear_interrupt_scheme(adapter);
4315 rtnl_unlock();
4316
4317 return 0;
4318}
4319
4320static int __maybe_unused ixgbevf_resume(struct device *dev_d)
4321{
4322 struct pci_dev *pdev = to_pci_dev(dev_d);
4323 struct net_device *netdev = pci_get_drvdata(pdev);
4324 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4325 u32 err;
4326
4327 adapter->hw.hw_addr = adapter->io_addr;
4328 smp_mb__before_atomic();
4329 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4330 pci_set_master(pdev);
4331
4332 ixgbevf_reset(adapter);
4333
4334 rtnl_lock();
4335 err = ixgbevf_init_interrupt_scheme(adapter);
4336 if (!err && netif_running(netdev))
4337 err = ixgbevf_open(netdev);
4338 rtnl_unlock();
4339 if (err)
4340 return err;
4341
4342 netif_device_attach(netdev);
4343
4344 return err;
4345}
4346
4347static void ixgbevf_shutdown(struct pci_dev *pdev)
4348{
4349 ixgbevf_suspend(&pdev->dev);
4350}
4351
4352static void ixgbevf_get_tx_ring_stats(struct rtnl_link_stats64 *stats,
4353 const struct ixgbevf_ring *ring)
4354{
4355 u64 bytes, packets;
4356 unsigned int start;
4357
4358 if (ring) {
4359 do {
4360 start = u64_stats_fetch_begin(&ring->syncp);
4361 bytes = ring->stats.bytes;
4362 packets = ring->stats.packets;
4363 } while (u64_stats_fetch_retry(&ring->syncp, start));
4364 stats->tx_bytes += bytes;
4365 stats->tx_packets += packets;
4366 }
4367}
4368
4369static void ixgbevf_get_stats(struct net_device *netdev,
4370 struct rtnl_link_stats64 *stats)
4371{
4372 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4373 unsigned int start;
4374 u64 bytes, packets;
4375 const struct ixgbevf_ring *ring;
4376 int i;
4377
4378 ixgbevf_update_stats(adapter);
4379
4380 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
4381
4382 rcu_read_lock();
4383 for (i = 0; i < adapter->num_rx_queues; i++) {
4384 ring = adapter->rx_ring[i];
4385 do {
4386 start = u64_stats_fetch_begin(&ring->syncp);
4387 bytes = ring->stats.bytes;
4388 packets = ring->stats.packets;
4389 } while (u64_stats_fetch_retry(&ring->syncp, start));
4390 stats->rx_bytes += bytes;
4391 stats->rx_packets += packets;
4392 }
4393
4394 for (i = 0; i < adapter->num_tx_queues; i++) {
4395 ring = adapter->tx_ring[i];
4396 ixgbevf_get_tx_ring_stats(stats, ring);
4397 }
4398
4399 for (i = 0; i < adapter->num_xdp_queues; i++) {
4400 ring = adapter->xdp_ring[i];
4401 ixgbevf_get_tx_ring_stats(stats, ring);
4402 }
4403 rcu_read_unlock();
4404}
4405
4406#define IXGBEVF_MAX_MAC_HDR_LEN 127
4407#define IXGBEVF_MAX_NETWORK_HDR_LEN 511
4408
4409static netdev_features_t
4410ixgbevf_features_check(struct sk_buff *skb, struct net_device *dev,
4411 netdev_features_t features)
4412{
4413 unsigned int network_hdr_len, mac_hdr_len;
4414
4415 /* Make certain the headers can be described by a context descriptor */
4416 mac_hdr_len = skb_network_header(skb) - skb->data;
4417 if (unlikely(mac_hdr_len > IXGBEVF_MAX_MAC_HDR_LEN))
4418 return features & ~(NETIF_F_HW_CSUM |
4419 NETIF_F_SCTP_CRC |
4420 NETIF_F_HW_VLAN_CTAG_TX |
4421 NETIF_F_TSO |
4422 NETIF_F_TSO6);
4423
4424 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4425 if (unlikely(network_hdr_len > IXGBEVF_MAX_NETWORK_HDR_LEN))
4426 return features & ~(NETIF_F_HW_CSUM |
4427 NETIF_F_SCTP_CRC |
4428 NETIF_F_TSO |
4429 NETIF_F_TSO6);
4430
4431 /* We can only support IPV4 TSO in tunnels if we can mangle the
4432 * inner IP ID field, so strip TSO if MANGLEID is not supported.
4433 */
4434 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4435 features &= ~NETIF_F_TSO;
4436
4437 return features;
4438}
4439
4440static int ixgbevf_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
4441{
4442 int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4443 struct ixgbevf_adapter *adapter = netdev_priv(dev);
4444 struct bpf_prog *old_prog;
4445
4446 /* verify ixgbevf ring attributes are sufficient for XDP */
4447 for (i = 0; i < adapter->num_rx_queues; i++) {
4448 struct ixgbevf_ring *ring = adapter->rx_ring[i];
4449
4450 if (frame_size > ixgbevf_rx_bufsz(ring))
4451 return -EINVAL;
4452 }
4453
4454 old_prog = xchg(&adapter->xdp_prog, prog);
4455
4456 /* If transitioning XDP modes reconfigure rings */
4457 if (!!prog != !!old_prog) {
4458 /* Hardware has to reinitialize queues and interrupts to
4459 * match packet buffer alignment. Unfortunately, the
4460 * hardware is not flexible enough to do this dynamically.
4461 */
4462 if (netif_running(dev))
4463 ixgbevf_close(dev);
4464
4465 ixgbevf_clear_interrupt_scheme(adapter);
4466 ixgbevf_init_interrupt_scheme(adapter);
4467
4468 if (netif_running(dev))
4469 ixgbevf_open(dev);
4470 } else {
4471 for (i = 0; i < adapter->num_rx_queues; i++)
4472 xchg(&adapter->rx_ring[i]->xdp_prog, adapter->xdp_prog);
4473 }
4474
4475 if (old_prog)
4476 bpf_prog_put(old_prog);
4477
4478 return 0;
4479}
4480
4481static int ixgbevf_xdp(struct net_device *dev, struct netdev_bpf *xdp)
4482{
4483 switch (xdp->command) {
4484 case XDP_SETUP_PROG:
4485 return ixgbevf_xdp_setup(dev, xdp->prog);
4486 default:
4487 return -EINVAL;
4488 }
4489}
4490
4491static const struct net_device_ops ixgbevf_netdev_ops = {
4492 .ndo_open = ixgbevf_open,
4493 .ndo_stop = ixgbevf_close,
4494 .ndo_start_xmit = ixgbevf_xmit_frame,
4495 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
4496 .ndo_get_stats64 = ixgbevf_get_stats,
4497 .ndo_validate_addr = eth_validate_addr,
4498 .ndo_set_mac_address = ixgbevf_set_mac,
4499 .ndo_change_mtu = ixgbevf_change_mtu,
4500 .ndo_tx_timeout = ixgbevf_tx_timeout,
4501 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
4502 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
4503 .ndo_features_check = ixgbevf_features_check,
4504 .ndo_bpf = ixgbevf_xdp,
4505};
4506
4507static void ixgbevf_assign_netdev_ops(struct net_device *dev)
4508{
4509 dev->netdev_ops = &ixgbevf_netdev_ops;
4510 ixgbevf_set_ethtool_ops(dev);
4511 dev->watchdog_timeo = 5 * HZ;
4512}
4513
4514/**
4515 * ixgbevf_probe - Device Initialization Routine
4516 * @pdev: PCI device information struct
4517 * @ent: entry in ixgbevf_pci_tbl
4518 *
4519 * Returns 0 on success, negative on failure
4520 *
4521 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
4522 * The OS initialization, configuring of the adapter private structure,
4523 * and a hardware reset occur.
4524 **/
4525static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4526{
4527 struct net_device *netdev;
4528 struct ixgbevf_adapter *adapter = NULL;
4529 struct ixgbe_hw *hw = NULL;
4530 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
4531 bool disable_dev = false;
4532 int err;
4533
4534 err = pci_enable_device(pdev);
4535 if (err)
4536 return err;
4537
4538 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
4539 if (err) {
4540 dev_err(&pdev->dev, "No usable DMA configuration, aborting\n");
4541 goto err_dma;
4542 }
4543
4544 err = pci_request_regions(pdev, ixgbevf_driver_name);
4545 if (err) {
4546 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
4547 goto err_pci_reg;
4548 }
4549
4550 pci_set_master(pdev);
4551
4552 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
4553 MAX_TX_QUEUES);
4554 if (!netdev) {
4555 err = -ENOMEM;
4556 goto err_alloc_etherdev;
4557 }
4558
4559 SET_NETDEV_DEV(netdev, &pdev->dev);
4560
4561 adapter = netdev_priv(netdev);
4562
4563 adapter->netdev = netdev;
4564 adapter->pdev = pdev;
4565 hw = &adapter->hw;
4566 hw->back = adapter;
4567 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4568
4569 /* call save state here in standalone driver because it relies on
4570 * adapter struct to exist, and needs to call netdev_priv
4571 */
4572 pci_save_state(pdev);
4573
4574 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
4575 pci_resource_len(pdev, 0));
4576 adapter->io_addr = hw->hw_addr;
4577 if (!hw->hw_addr) {
4578 err = -EIO;
4579 goto err_ioremap;
4580 }
4581
4582 ixgbevf_assign_netdev_ops(netdev);
4583
4584 /* Setup HW API */
4585 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
4586 hw->mac.type = ii->mac;
4587
4588 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops_legacy,
4589 sizeof(struct ixgbe_mbx_operations));
4590
4591 /* setup the private structure */
4592 err = ixgbevf_sw_init(adapter);
4593 if (err)
4594 goto err_sw_init;
4595
4596 /* The HW MAC address was set and/or determined in sw_init */
4597 if (!is_valid_ether_addr(netdev->dev_addr)) {
4598 pr_err("invalid MAC address\n");
4599 err = -EIO;
4600 goto err_sw_init;
4601 }
4602
4603 netdev->hw_features = NETIF_F_SG |
4604 NETIF_F_TSO |
4605 NETIF_F_TSO6 |
4606 NETIF_F_RXCSUM |
4607 NETIF_F_HW_CSUM |
4608 NETIF_F_SCTP_CRC;
4609
4610#define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4611 NETIF_F_GSO_GRE_CSUM | \
4612 NETIF_F_GSO_IPXIP4 | \
4613 NETIF_F_GSO_IPXIP6 | \
4614 NETIF_F_GSO_UDP_TUNNEL | \
4615 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4616
4617 netdev->gso_partial_features = IXGBEVF_GSO_PARTIAL_FEATURES;
4618 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
4619 IXGBEVF_GSO_PARTIAL_FEATURES;
4620
4621 netdev->features = netdev->hw_features | NETIF_F_HIGHDMA;
4622
4623 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
4624 netdev->mpls_features |= NETIF_F_SG |
4625 NETIF_F_TSO |
4626 NETIF_F_TSO6 |
4627 NETIF_F_HW_CSUM;
4628 netdev->mpls_features |= IXGBEVF_GSO_PARTIAL_FEATURES;
4629 netdev->hw_enc_features |= netdev->vlan_features;
4630
4631 /* set this bit last since it cannot be part of vlan_features */
4632 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
4633 NETIF_F_HW_VLAN_CTAG_RX |
4634 NETIF_F_HW_VLAN_CTAG_TX;
4635
4636 netdev->priv_flags |= IFF_UNICAST_FLT;
4637
4638 /* MTU range: 68 - 1504 or 9710 */
4639 netdev->min_mtu = ETH_MIN_MTU;
4640 switch (adapter->hw.api_version) {
4641 case ixgbe_mbox_api_11:
4642 case ixgbe_mbox_api_12:
4643 case ixgbe_mbox_api_13:
4644 case ixgbe_mbox_api_14:
4645 case ixgbe_mbox_api_15:
4646 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4647 (ETH_HLEN + ETH_FCS_LEN);
4648 break;
4649 default:
4650 if (adapter->hw.mac.type != ixgbe_mac_82599_vf)
4651 netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE -
4652 (ETH_HLEN + ETH_FCS_LEN);
4653 else
4654 netdev->max_mtu = ETH_DATA_LEN + ETH_FCS_LEN;
4655 break;
4656 }
4657
4658 if (IXGBE_REMOVED(hw->hw_addr)) {
4659 err = -EIO;
4660 goto err_sw_init;
4661 }
4662
4663 timer_setup(&adapter->service_timer, ixgbevf_service_timer, 0);
4664
4665 INIT_WORK(&adapter->service_task, ixgbevf_service_task);
4666 set_bit(__IXGBEVF_SERVICE_INITED, &adapter->state);
4667 clear_bit(__IXGBEVF_SERVICE_SCHED, &adapter->state);
4668
4669 err = ixgbevf_init_interrupt_scheme(adapter);
4670 if (err)
4671 goto err_sw_init;
4672
4673 strcpy(netdev->name, "eth%d");
4674
4675 err = register_netdev(netdev);
4676 if (err)
4677 goto err_register;
4678
4679 pci_set_drvdata(pdev, netdev);
4680 netif_carrier_off(netdev);
4681 ixgbevf_init_ipsec_offload(adapter);
4682
4683 ixgbevf_init_last_counter_stats(adapter);
4684
4685 /* print the VF info */
4686 dev_info(&pdev->dev, "%pM\n", netdev->dev_addr);
4687 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
4688
4689 switch (hw->mac.type) {
4690 case ixgbe_mac_X550_vf:
4691 dev_info(&pdev->dev, "Intel(R) X550 Virtual Function\n");
4692 break;
4693 case ixgbe_mac_X540_vf:
4694 dev_info(&pdev->dev, "Intel(R) X540 Virtual Function\n");
4695 break;
4696 case ixgbe_mac_82599_vf:
4697 default:
4698 dev_info(&pdev->dev, "Intel(R) 82599 Virtual Function\n");
4699 break;
4700 }
4701
4702 return 0;
4703
4704err_register:
4705 ixgbevf_clear_interrupt_scheme(adapter);
4706err_sw_init:
4707 ixgbevf_reset_interrupt_capability(adapter);
4708 iounmap(adapter->io_addr);
4709 kfree(adapter->rss_key);
4710err_ioremap:
4711 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4712 free_netdev(netdev);
4713err_alloc_etherdev:
4714 pci_release_regions(pdev);
4715err_pci_reg:
4716err_dma:
4717 if (!adapter || disable_dev)
4718 pci_disable_device(pdev);
4719 return err;
4720}
4721
4722/**
4723 * ixgbevf_remove - Device Removal Routine
4724 * @pdev: PCI device information struct
4725 *
4726 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4727 * that it should release a PCI device. The could be caused by a
4728 * Hot-Plug event, or because the driver is going to be removed from
4729 * memory.
4730 **/
4731static void ixgbevf_remove(struct pci_dev *pdev)
4732{
4733 struct net_device *netdev = pci_get_drvdata(pdev);
4734 struct ixgbevf_adapter *adapter;
4735 bool disable_dev;
4736
4737 if (!netdev)
4738 return;
4739
4740 adapter = netdev_priv(netdev);
4741
4742 set_bit(__IXGBEVF_REMOVING, &adapter->state);
4743 cancel_work_sync(&adapter->service_task);
4744
4745 if (netdev->reg_state == NETREG_REGISTERED)
4746 unregister_netdev(netdev);
4747
4748 ixgbevf_stop_ipsec_offload(adapter);
4749 ixgbevf_clear_interrupt_scheme(adapter);
4750 ixgbevf_reset_interrupt_capability(adapter);
4751
4752 iounmap(adapter->io_addr);
4753 pci_release_regions(pdev);
4754
4755 hw_dbg(&adapter->hw, "Remove complete\n");
4756
4757 kfree(adapter->rss_key);
4758 disable_dev = !test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state);
4759 free_netdev(netdev);
4760
4761 if (disable_dev)
4762 pci_disable_device(pdev);
4763}
4764
4765/**
4766 * ixgbevf_io_error_detected - called when PCI error is detected
4767 * @pdev: Pointer to PCI device
4768 * @state: The current pci connection state
4769 *
4770 * This function is called after a PCI bus error affecting
4771 * this device has been detected.
4772 **/
4773static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
4774 pci_channel_state_t state)
4775{
4776 struct net_device *netdev = pci_get_drvdata(pdev);
4777 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4778
4779 if (!test_bit(__IXGBEVF_SERVICE_INITED, &adapter->state))
4780 return PCI_ERS_RESULT_DISCONNECT;
4781
4782 rtnl_lock();
4783 netif_device_detach(netdev);
4784
4785 if (netif_running(netdev))
4786 ixgbevf_close_suspend(adapter);
4787
4788 if (state == pci_channel_io_perm_failure) {
4789 rtnl_unlock();
4790 return PCI_ERS_RESULT_DISCONNECT;
4791 }
4792
4793 if (!test_and_set_bit(__IXGBEVF_DISABLED, &adapter->state))
4794 pci_disable_device(pdev);
4795 rtnl_unlock();
4796
4797 /* Request a slot reset. */
4798 return PCI_ERS_RESULT_NEED_RESET;
4799}
4800
4801/**
4802 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4803 * @pdev: Pointer to PCI device
4804 *
4805 * Restart the card from scratch, as if from a cold-boot. Implementation
4806 * resembles the first-half of the ixgbevf_resume routine.
4807 **/
4808static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
4809{
4810 struct net_device *netdev = pci_get_drvdata(pdev);
4811 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
4812
4813 if (pci_enable_device_mem(pdev)) {
4814 dev_err(&pdev->dev,
4815 "Cannot re-enable PCI device after reset.\n");
4816 return PCI_ERS_RESULT_DISCONNECT;
4817 }
4818
4819 adapter->hw.hw_addr = adapter->io_addr;
4820 smp_mb__before_atomic();
4821 clear_bit(__IXGBEVF_DISABLED, &adapter->state);
4822 pci_set_master(pdev);
4823
4824 ixgbevf_reset(adapter);
4825
4826 return PCI_ERS_RESULT_RECOVERED;
4827}
4828
4829/**
4830 * ixgbevf_io_resume - called when traffic can start flowing again.
4831 * @pdev: Pointer to PCI device
4832 *
4833 * This callback is called when the error recovery driver tells us that
4834 * its OK to resume normal operation. Implementation resembles the
4835 * second-half of the ixgbevf_resume routine.
4836 **/
4837static void ixgbevf_io_resume(struct pci_dev *pdev)
4838{
4839 struct net_device *netdev = pci_get_drvdata(pdev);
4840
4841 rtnl_lock();
4842 if (netif_running(netdev))
4843 ixgbevf_open(netdev);
4844
4845 netif_device_attach(netdev);
4846 rtnl_unlock();
4847}
4848
4849/* PCI Error Recovery (ERS) */
4850static const struct pci_error_handlers ixgbevf_err_handler = {
4851 .error_detected = ixgbevf_io_error_detected,
4852 .slot_reset = ixgbevf_io_slot_reset,
4853 .resume = ixgbevf_io_resume,
4854};
4855
4856static SIMPLE_DEV_PM_OPS(ixgbevf_pm_ops, ixgbevf_suspend, ixgbevf_resume);
4857
4858static struct pci_driver ixgbevf_driver = {
4859 .name = ixgbevf_driver_name,
4860 .id_table = ixgbevf_pci_tbl,
4861 .probe = ixgbevf_probe,
4862 .remove = ixgbevf_remove,
4863
4864 /* Power Management Hooks */
4865 .driver.pm = &ixgbevf_pm_ops,
4866
4867 .shutdown = ixgbevf_shutdown,
4868 .err_handler = &ixgbevf_err_handler
4869};
4870
4871/**
4872 * ixgbevf_init_module - Driver Registration Routine
4873 *
4874 * ixgbevf_init_module is the first routine called when the driver is
4875 * loaded. All it does is register with the PCI subsystem.
4876 **/
4877static int __init ixgbevf_init_module(void)
4878{
4879 int err;
4880
4881 pr_info("%s\n", ixgbevf_driver_string);
4882 pr_info("%s\n", ixgbevf_copyright);
4883 ixgbevf_wq = create_singlethread_workqueue(ixgbevf_driver_name);
4884 if (!ixgbevf_wq) {
4885 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name);
4886 return -ENOMEM;
4887 }
4888
4889 err = pci_register_driver(&ixgbevf_driver);
4890 if (err) {
4891 destroy_workqueue(ixgbevf_wq);
4892 return err;
4893 }
4894
4895 return 0;
4896}
4897
4898module_init(ixgbevf_init_module);
4899
4900/**
4901 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4902 *
4903 * ixgbevf_exit_module is called just before the driver is removed
4904 * from memory.
4905 **/
4906static void __exit ixgbevf_exit_module(void)
4907{
4908 pci_unregister_driver(&ixgbevf_driver);
4909 if (ixgbevf_wq) {
4910 destroy_workqueue(ixgbevf_wq);
4911 ixgbevf_wq = NULL;
4912 }
4913}
4914
4915#ifdef DEBUG
4916/**
4917 * ixgbevf_get_hw_dev_name - return device name string
4918 * used by hardware layer to print debugging information
4919 * @hw: pointer to private hardware struct
4920 **/
4921char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
4922{
4923 struct ixgbevf_adapter *adapter = hw->back;
4924
4925 return adapter->netdev->name;
4926}
4927
4928#endif
4929module_exit(ixgbevf_exit_module);
4930
4931/* ixgbevf_main.c */