ixgbevf: fix softirq-safe to unsafe splat on internal mbx_lock
[linux-2.6/libata-dev.git] / drivers / net / ethernet / intel / ixgbevf / ixgbevf_main.c
blob846ac5193584b3f4fd60e1f0f8876e8d4fa01744
1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 more details.
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
22 Contact Information:
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
29 /******************************************************************************
30 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
51 #include <linux/if.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
55 #include "ixgbevf.h"
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
61 #define DRV_VERSION "2.6.0-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64 "Copyright (c) 2009 - 2012 Intel Corporation.";
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67 [board_82599_vf] = &ixgbevf_82599_vf_info,
68 [board_X540_vf] = &ixgbevf_X540_vf_info,
71 /* ixgbevf_pci_tbl - PCI Device ID Table
73 * Wildcard entries (PCI_ANY_ID) should come last
74 * Last entry must be all 0s
76 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77 * Class, Class Mask, private data (not used) }
79 static struct pci_device_id ixgbevf_pci_tbl[] = {
80 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
81 board_82599_vf},
82 {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
83 board_X540_vf},
85 /* required last entry */
86 {0, }
88 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
90 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
91 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
92 MODULE_LICENSE("GPL");
93 MODULE_VERSION(DRV_VERSION);
95 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
96 static int debug = -1;
97 module_param(debug, int, 0);
98 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
100 /* forward decls */
101 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
102 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
104 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
105 struct ixgbevf_ring *rx_ring,
106 u32 val)
109 * Force memory writes to complete before letting h/w
110 * know there are new descriptors to fetch. (Only
111 * applicable for weak-ordered memory model archs,
112 * such as IA-64).
114 wmb();
115 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
119 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
120 * @adapter: pointer to adapter struct
121 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
122 * @queue: queue to map the corresponding interrupt to
123 * @msix_vector: the vector to map to the corresponding queue
126 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
127 u8 queue, u8 msix_vector)
129 u32 ivar, index;
130 struct ixgbe_hw *hw = &adapter->hw;
131 if (direction == -1) {
132 /* other causes */
133 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
134 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
135 ivar &= ~0xFF;
136 ivar |= msix_vector;
137 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
138 } else {
139 /* tx or rx causes */
140 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
141 index = ((16 * (queue & 1)) + (8 * direction));
142 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
143 ivar &= ~(0xFF << index);
144 ivar |= (msix_vector << index);
145 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
149 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
150 struct ixgbevf_tx_buffer
151 *tx_buffer_info)
153 if (tx_buffer_info->dma) {
154 if (tx_buffer_info->mapped_as_page)
155 dma_unmap_page(tx_ring->dev,
156 tx_buffer_info->dma,
157 tx_buffer_info->length,
158 DMA_TO_DEVICE);
159 else
160 dma_unmap_single(tx_ring->dev,
161 tx_buffer_info->dma,
162 tx_buffer_info->length,
163 DMA_TO_DEVICE);
164 tx_buffer_info->dma = 0;
166 if (tx_buffer_info->skb) {
167 dev_kfree_skb_any(tx_buffer_info->skb);
168 tx_buffer_info->skb = NULL;
170 tx_buffer_info->time_stamp = 0;
171 /* tx_buffer_info must be completely set up in the transmit path */
174 #define IXGBE_MAX_TXD_PWR 14
175 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
177 /* Tx Descriptors needed, worst case */
178 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
179 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
181 static void ixgbevf_tx_timeout(struct net_device *netdev);
184 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
185 * @q_vector: board private structure
186 * @tx_ring: tx ring to clean
188 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
189 struct ixgbevf_ring *tx_ring)
191 struct ixgbevf_adapter *adapter = q_vector->adapter;
192 union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
193 struct ixgbevf_tx_buffer *tx_buffer_info;
194 unsigned int i, eop, count = 0;
195 unsigned int total_bytes = 0, total_packets = 0;
197 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
198 return true;
200 i = tx_ring->next_to_clean;
201 eop = tx_ring->tx_buffer_info[i].next_to_watch;
202 eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
204 while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
205 (count < tx_ring->count)) {
206 bool cleaned = false;
207 rmb(); /* read buffer_info after eop_desc */
208 /* eop could change between read and DD-check */
209 if (unlikely(eop != tx_ring->tx_buffer_info[i].next_to_watch))
210 goto cont_loop;
211 for ( ; !cleaned; count++) {
212 struct sk_buff *skb;
213 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
214 tx_buffer_info = &tx_ring->tx_buffer_info[i];
215 cleaned = (i == eop);
216 skb = tx_buffer_info->skb;
218 if (cleaned && skb) {
219 unsigned int segs, bytecount;
221 /* gso_segs is currently only valid for tcp */
222 segs = skb_shinfo(skb)->gso_segs ?: 1;
223 /* multiply data chunks by size of headers */
224 bytecount = ((segs - 1) * skb_headlen(skb)) +
225 skb->len;
226 total_packets += segs;
227 total_bytes += bytecount;
230 ixgbevf_unmap_and_free_tx_resource(tx_ring,
231 tx_buffer_info);
233 tx_desc->wb.status = 0;
235 i++;
236 if (i == tx_ring->count)
237 i = 0;
240 cont_loop:
241 eop = tx_ring->tx_buffer_info[i].next_to_watch;
242 eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
245 tx_ring->next_to_clean = i;
247 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
248 if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
249 (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
250 /* Make sure that anybody stopping the queue after this
251 * sees the new next_to_clean.
253 smp_mb();
254 if (__netif_subqueue_stopped(tx_ring->netdev,
255 tx_ring->queue_index) &&
256 !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
257 netif_wake_subqueue(tx_ring->netdev,
258 tx_ring->queue_index);
259 ++adapter->restart_queue;
263 u64_stats_update_begin(&tx_ring->syncp);
264 tx_ring->total_bytes += total_bytes;
265 tx_ring->total_packets += total_packets;
266 u64_stats_update_end(&tx_ring->syncp);
267 q_vector->tx.total_bytes += total_bytes;
268 q_vector->tx.total_packets += total_packets;
270 return count < tx_ring->count;
274 * ixgbevf_receive_skb - Send a completed packet up the stack
275 * @q_vector: structure containing interrupt and ring information
276 * @skb: packet to send up
277 * @status: hardware indication of status of receive
278 * @rx_desc: rx descriptor
280 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
281 struct sk_buff *skb, u8 status,
282 union ixgbe_adv_rx_desc *rx_desc)
284 struct ixgbevf_adapter *adapter = q_vector->adapter;
285 bool is_vlan = (status & IXGBE_RXD_STAT_VP);
286 u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
288 if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
289 __vlan_hwaccel_put_tag(skb, tag);
291 napi_gro_receive(&q_vector->napi, skb);
295 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
296 * @adapter: address of board private structure
297 * @status_err: hardware indication of status of receive
298 * @skb: skb currently being received and modified
300 static inline void ixgbevf_rx_checksum(struct ixgbevf_adapter *adapter,
301 struct ixgbevf_ring *ring,
302 u32 status_err, struct sk_buff *skb)
304 skb_checksum_none_assert(skb);
306 /* Rx csum disabled */
307 if (!(ring->netdev->features & NETIF_F_RXCSUM))
308 return;
310 /* if IP and error */
311 if ((status_err & IXGBE_RXD_STAT_IPCS) &&
312 (status_err & IXGBE_RXDADV_ERR_IPE)) {
313 adapter->hw_csum_rx_error++;
314 return;
317 if (!(status_err & IXGBE_RXD_STAT_L4CS))
318 return;
320 if (status_err & IXGBE_RXDADV_ERR_TCPE) {
321 adapter->hw_csum_rx_error++;
322 return;
325 /* It must be a TCP or UDP packet with a valid checksum */
326 skb->ip_summed = CHECKSUM_UNNECESSARY;
327 adapter->hw_csum_rx_good++;
331 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
332 * @adapter: address of board private structure
334 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
335 struct ixgbevf_ring *rx_ring,
336 int cleaned_count)
338 struct pci_dev *pdev = adapter->pdev;
339 union ixgbe_adv_rx_desc *rx_desc;
340 struct ixgbevf_rx_buffer *bi;
341 struct sk_buff *skb;
342 unsigned int i = rx_ring->next_to_use;
344 bi = &rx_ring->rx_buffer_info[i];
346 while (cleaned_count--) {
347 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
348 skb = bi->skb;
349 if (!skb) {
350 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
351 rx_ring->rx_buf_len);
352 if (!skb) {
353 adapter->alloc_rx_buff_failed++;
354 goto no_buffers;
356 bi->skb = skb;
358 if (!bi->dma) {
359 bi->dma = dma_map_single(&pdev->dev, skb->data,
360 rx_ring->rx_buf_len,
361 DMA_FROM_DEVICE);
362 if (dma_mapping_error(&pdev->dev, bi->dma)) {
363 dev_kfree_skb(skb);
364 bi->skb = NULL;
365 dev_err(&pdev->dev, "RX DMA map failed\n");
366 break;
369 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
371 i++;
372 if (i == rx_ring->count)
373 i = 0;
374 bi = &rx_ring->rx_buffer_info[i];
377 no_buffers:
378 if (rx_ring->next_to_use != i) {
379 rx_ring->next_to_use = i;
381 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
385 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
386 u32 qmask)
388 struct ixgbe_hw *hw = &adapter->hw;
390 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
393 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
394 struct ixgbevf_ring *rx_ring,
395 int budget)
397 struct ixgbevf_adapter *adapter = q_vector->adapter;
398 struct pci_dev *pdev = adapter->pdev;
399 union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
400 struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
401 struct sk_buff *skb;
402 unsigned int i;
403 u32 len, staterr;
404 int cleaned_count = 0;
405 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
407 i = rx_ring->next_to_clean;
408 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
409 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
410 rx_buffer_info = &rx_ring->rx_buffer_info[i];
412 while (staterr & IXGBE_RXD_STAT_DD) {
413 if (!budget)
414 break;
415 budget--;
417 rmb(); /* read descriptor and rx_buffer_info after status DD */
418 len = le16_to_cpu(rx_desc->wb.upper.length);
419 skb = rx_buffer_info->skb;
420 prefetch(skb->data - NET_IP_ALIGN);
421 rx_buffer_info->skb = NULL;
423 if (rx_buffer_info->dma) {
424 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
425 rx_ring->rx_buf_len,
426 DMA_FROM_DEVICE);
427 rx_buffer_info->dma = 0;
428 skb_put(skb, len);
431 i++;
432 if (i == rx_ring->count)
433 i = 0;
435 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
436 prefetch(next_rxd);
437 cleaned_count++;
439 next_buffer = &rx_ring->rx_buffer_info[i];
441 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
442 skb->next = next_buffer->skb;
443 IXGBE_CB(skb->next)->prev = skb;
444 adapter->non_eop_descs++;
445 goto next_desc;
448 /* we should not be chaining buffers, if we did drop the skb */
449 if (IXGBE_CB(skb)->prev) {
450 do {
451 struct sk_buff *this = skb;
452 skb = IXGBE_CB(skb)->prev;
453 dev_kfree_skb(this);
454 } while (skb);
455 goto next_desc;
458 /* ERR_MASK will only have valid bits if EOP set */
459 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
460 dev_kfree_skb_irq(skb);
461 goto next_desc;
464 ixgbevf_rx_checksum(adapter, rx_ring, staterr, skb);
466 /* probably a little skewed due to removing CRC */
467 total_rx_bytes += skb->len;
468 total_rx_packets++;
471 * Work around issue of some types of VM to VM loop back
472 * packets not getting split correctly
474 if (staterr & IXGBE_RXD_STAT_LB) {
475 u32 header_fixup_len = skb_headlen(skb);
476 if (header_fixup_len < 14)
477 skb_push(skb, header_fixup_len);
479 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
481 ixgbevf_receive_skb(q_vector, skb, staterr, rx_desc);
483 next_desc:
484 rx_desc->wb.upper.status_error = 0;
486 /* return some buffers to hardware, one at a time is too slow */
487 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
488 ixgbevf_alloc_rx_buffers(adapter, rx_ring,
489 cleaned_count);
490 cleaned_count = 0;
493 /* use prefetched values */
494 rx_desc = next_rxd;
495 rx_buffer_info = &rx_ring->rx_buffer_info[i];
497 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
500 rx_ring->next_to_clean = i;
501 cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
503 if (cleaned_count)
504 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
506 u64_stats_update_begin(&rx_ring->syncp);
507 rx_ring->total_packets += total_rx_packets;
508 rx_ring->total_bytes += total_rx_bytes;
509 u64_stats_update_end(&rx_ring->syncp);
510 q_vector->rx.total_packets += total_rx_packets;
511 q_vector->rx.total_bytes += total_rx_bytes;
513 return !!budget;
517 * ixgbevf_poll - NAPI polling calback
518 * @napi: napi struct with our devices info in it
519 * @budget: amount of work driver is allowed to do this pass, in packets
521 * This function will clean more than one or more rings associated with a
522 * q_vector.
524 static int ixgbevf_poll(struct napi_struct *napi, int budget)
526 struct ixgbevf_q_vector *q_vector =
527 container_of(napi, struct ixgbevf_q_vector, napi);
528 struct ixgbevf_adapter *adapter = q_vector->adapter;
529 struct ixgbevf_ring *ring;
530 int per_ring_budget;
531 bool clean_complete = true;
533 ixgbevf_for_each_ring(ring, q_vector->tx)
534 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
536 /* attempt to distribute budget to each queue fairly, but don't allow
537 * the budget to go below 1 because we'll exit polling */
538 if (q_vector->rx.count > 1)
539 per_ring_budget = max(budget/q_vector->rx.count, 1);
540 else
541 per_ring_budget = budget;
543 ixgbevf_for_each_ring(ring, q_vector->rx)
544 clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
545 per_ring_budget);
547 /* If all work not completed, return budget and keep polling */
548 if (!clean_complete)
549 return budget;
550 /* all work done, exit the polling mode */
551 napi_complete(napi);
552 if (adapter->rx_itr_setting & 1)
553 ixgbevf_set_itr(q_vector);
554 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
555 ixgbevf_irq_enable_queues(adapter,
556 1 << q_vector->v_idx);
558 return 0;
562 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
563 * @q_vector: structure containing interrupt and ring information
565 static void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
567 struct ixgbevf_adapter *adapter = q_vector->adapter;
568 struct ixgbe_hw *hw = &adapter->hw;
569 int v_idx = q_vector->v_idx;
570 u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
573 * set the WDIS bit to not clear the timer bits and cause an
574 * immediate assertion of the interrupt
576 itr_reg |= IXGBE_EITR_CNT_WDIS;
578 IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
582 * ixgbevf_configure_msix - Configure MSI-X hardware
583 * @adapter: board private structure
585 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
586 * interrupts.
588 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
590 struct ixgbevf_q_vector *q_vector;
591 int q_vectors, v_idx;
593 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
594 adapter->eims_enable_mask = 0;
597 * Populate the IVAR table and set the ITR values to the
598 * corresponding register.
600 for (v_idx = 0; v_idx < q_vectors; v_idx++) {
601 struct ixgbevf_ring *ring;
602 q_vector = adapter->q_vector[v_idx];
604 ixgbevf_for_each_ring(ring, q_vector->rx)
605 ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
607 ixgbevf_for_each_ring(ring, q_vector->tx)
608 ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
610 if (q_vector->tx.ring && !q_vector->rx.ring) {
611 /* tx only vector */
612 if (adapter->tx_itr_setting == 1)
613 q_vector->itr = IXGBE_10K_ITR;
614 else
615 q_vector->itr = adapter->tx_itr_setting;
616 } else {
617 /* rx or rx/tx vector */
618 if (adapter->rx_itr_setting == 1)
619 q_vector->itr = IXGBE_20K_ITR;
620 else
621 q_vector->itr = adapter->rx_itr_setting;
624 /* add q_vector eims value to global eims_enable_mask */
625 adapter->eims_enable_mask |= 1 << v_idx;
627 ixgbevf_write_eitr(q_vector);
630 ixgbevf_set_ivar(adapter, -1, 1, v_idx);
631 /* setup eims_other and add value to global eims_enable_mask */
632 adapter->eims_other = 1 << v_idx;
633 adapter->eims_enable_mask |= adapter->eims_other;
636 enum latency_range {
637 lowest_latency = 0,
638 low_latency = 1,
639 bulk_latency = 2,
640 latency_invalid = 255
644 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
645 * @q_vector: structure containing interrupt and ring information
646 * @ring_container: structure containing ring performance data
648 * Stores a new ITR value based on packets and byte
649 * counts during the last interrupt. The advantage of per interrupt
650 * computation is faster updates and more accurate ITR for the current
651 * traffic pattern. Constants in this function were computed
652 * based on theoretical maximum wire speed and thresholds were set based
653 * on testing data as well as attempting to minimize response time
654 * while increasing bulk throughput.
656 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
657 struct ixgbevf_ring_container *ring_container)
659 int bytes = ring_container->total_bytes;
660 int packets = ring_container->total_packets;
661 u32 timepassed_us;
662 u64 bytes_perint;
663 u8 itr_setting = ring_container->itr;
665 if (packets == 0)
666 return;
668 /* simple throttlerate management
669 * 0-20MB/s lowest (100000 ints/s)
670 * 20-100MB/s low (20000 ints/s)
671 * 100-1249MB/s bulk (8000 ints/s)
673 /* what was last interrupt timeslice? */
674 timepassed_us = q_vector->itr >> 2;
675 bytes_perint = bytes / timepassed_us; /* bytes/usec */
677 switch (itr_setting) {
678 case lowest_latency:
679 if (bytes_perint > 10)
680 itr_setting = low_latency;
681 break;
682 case low_latency:
683 if (bytes_perint > 20)
684 itr_setting = bulk_latency;
685 else if (bytes_perint <= 10)
686 itr_setting = lowest_latency;
687 break;
688 case bulk_latency:
689 if (bytes_perint <= 20)
690 itr_setting = low_latency;
691 break;
694 /* clear work counters since we have the values we need */
695 ring_container->total_bytes = 0;
696 ring_container->total_packets = 0;
698 /* write updated itr to ring container */
699 ring_container->itr = itr_setting;
702 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
704 u32 new_itr = q_vector->itr;
705 u8 current_itr;
707 ixgbevf_update_itr(q_vector, &q_vector->tx);
708 ixgbevf_update_itr(q_vector, &q_vector->rx);
710 current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
712 switch (current_itr) {
713 /* counts and packets in update_itr are dependent on these numbers */
714 case lowest_latency:
715 new_itr = IXGBE_100K_ITR;
716 break;
717 case low_latency:
718 new_itr = IXGBE_20K_ITR;
719 break;
720 case bulk_latency:
721 default:
722 new_itr = IXGBE_8K_ITR;
723 break;
726 if (new_itr != q_vector->itr) {
727 /* do an exponential smoothing */
728 new_itr = (10 * new_itr * q_vector->itr) /
729 ((9 * new_itr) + q_vector->itr);
731 /* save the algorithm value here */
732 q_vector->itr = new_itr;
734 ixgbevf_write_eitr(q_vector);
738 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
740 struct ixgbevf_adapter *adapter = data;
741 struct ixgbe_hw *hw = &adapter->hw;
743 hw->mac.get_link_status = 1;
745 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
746 mod_timer(&adapter->watchdog_timer, jiffies);
748 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
750 return IRQ_HANDLED;
755 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
756 * @irq: unused
757 * @data: pointer to our q_vector struct for this interrupt vector
759 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
761 struct ixgbevf_q_vector *q_vector = data;
763 /* EIAM disabled interrupts (on this vector) for us */
764 if (q_vector->rx.ring || q_vector->tx.ring)
765 napi_schedule(&q_vector->napi);
767 return IRQ_HANDLED;
770 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
771 int r_idx)
773 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
775 a->rx_ring[r_idx].next = q_vector->rx.ring;
776 q_vector->rx.ring = &a->rx_ring[r_idx];
777 q_vector->rx.count++;
780 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
781 int t_idx)
783 struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
785 a->tx_ring[t_idx].next = q_vector->tx.ring;
786 q_vector->tx.ring = &a->tx_ring[t_idx];
787 q_vector->tx.count++;
791 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
792 * @adapter: board private structure to initialize
794 * This function maps descriptor rings to the queue-specific vectors
795 * we were allotted through the MSI-X enabling code. Ideally, we'd have
796 * one vector per ring/queue, but on a constrained vector budget, we
797 * group the rings as "efficiently" as possible. You would add new
798 * mapping configurations in here.
800 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
802 int q_vectors;
803 int v_start = 0;
804 int rxr_idx = 0, txr_idx = 0;
805 int rxr_remaining = adapter->num_rx_queues;
806 int txr_remaining = adapter->num_tx_queues;
807 int i, j;
808 int rqpv, tqpv;
809 int err = 0;
811 q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
814 * The ideal configuration...
815 * We have enough vectors to map one per queue.
817 if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
818 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
819 map_vector_to_rxq(adapter, v_start, rxr_idx);
821 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
822 map_vector_to_txq(adapter, v_start, txr_idx);
823 goto out;
827 * If we don't have enough vectors for a 1-to-1
828 * mapping, we'll have to group them so there are
829 * multiple queues per vector.
831 /* Re-adjusting *qpv takes care of the remainder. */
832 for (i = v_start; i < q_vectors; i++) {
833 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
834 for (j = 0; j < rqpv; j++) {
835 map_vector_to_rxq(adapter, i, rxr_idx);
836 rxr_idx++;
837 rxr_remaining--;
840 for (i = v_start; i < q_vectors; i++) {
841 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
842 for (j = 0; j < tqpv; j++) {
843 map_vector_to_txq(adapter, i, txr_idx);
844 txr_idx++;
845 txr_remaining--;
849 out:
850 return err;
854 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
855 * @adapter: board private structure
857 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
858 * interrupts from the kernel.
860 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
862 struct net_device *netdev = adapter->netdev;
863 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
864 int vector, err;
865 int ri = 0, ti = 0;
867 for (vector = 0; vector < q_vectors; vector++) {
868 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
869 struct msix_entry *entry = &adapter->msix_entries[vector];
871 if (q_vector->tx.ring && q_vector->rx.ring) {
872 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
873 "%s-%s-%d", netdev->name, "TxRx", ri++);
874 ti++;
875 } else if (q_vector->rx.ring) {
876 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
877 "%s-%s-%d", netdev->name, "rx", ri++);
878 } else if (q_vector->tx.ring) {
879 snprintf(q_vector->name, sizeof(q_vector->name) - 1,
880 "%s-%s-%d", netdev->name, "tx", ti++);
881 } else {
882 /* skip this unused q_vector */
883 continue;
885 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
886 q_vector->name, q_vector);
887 if (err) {
888 hw_dbg(&adapter->hw,
889 "request_irq failed for MSIX interrupt "
890 "Error: %d\n", err);
891 goto free_queue_irqs;
895 err = request_irq(adapter->msix_entries[vector].vector,
896 &ixgbevf_msix_other, 0, netdev->name, adapter);
897 if (err) {
898 hw_dbg(&adapter->hw,
899 "request_irq for msix_other failed: %d\n", err);
900 goto free_queue_irqs;
903 return 0;
905 free_queue_irqs:
906 while (vector) {
907 vector--;
908 free_irq(adapter->msix_entries[vector].vector,
909 adapter->q_vector[vector]);
911 pci_disable_msix(adapter->pdev);
912 kfree(adapter->msix_entries);
913 adapter->msix_entries = NULL;
914 return err;
917 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
919 int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
921 for (i = 0; i < q_vectors; i++) {
922 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
923 q_vector->rx.ring = NULL;
924 q_vector->tx.ring = NULL;
925 q_vector->rx.count = 0;
926 q_vector->tx.count = 0;
931 * ixgbevf_request_irq - initialize interrupts
932 * @adapter: board private structure
934 * Attempts to configure interrupts using the best available
935 * capabilities of the hardware and kernel.
937 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
939 int err = 0;
941 err = ixgbevf_request_msix_irqs(adapter);
943 if (err)
944 hw_dbg(&adapter->hw,
945 "request_irq failed, Error %d\n", err);
947 return err;
950 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
952 int i, q_vectors;
954 q_vectors = adapter->num_msix_vectors;
955 i = q_vectors - 1;
957 free_irq(adapter->msix_entries[i].vector, adapter);
958 i--;
960 for (; i >= 0; i--) {
961 /* free only the irqs that were actually requested */
962 if (!adapter->q_vector[i]->rx.ring &&
963 !adapter->q_vector[i]->tx.ring)
964 continue;
966 free_irq(adapter->msix_entries[i].vector,
967 adapter->q_vector[i]);
970 ixgbevf_reset_q_vectors(adapter);
974 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
975 * @adapter: board private structure
977 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
979 struct ixgbe_hw *hw = &adapter->hw;
980 int i;
982 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
983 IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
984 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
986 IXGBE_WRITE_FLUSH(hw);
988 for (i = 0; i < adapter->num_msix_vectors; i++)
989 synchronize_irq(adapter->msix_entries[i].vector);
993 * ixgbevf_irq_enable - Enable default interrupt generation settings
994 * @adapter: board private structure
996 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
998 struct ixgbe_hw *hw = &adapter->hw;
1000 IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1001 IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1002 IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1006 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1007 * @adapter: board private structure
1009 * Configure the Tx unit of the MAC after a reset.
1011 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1013 u64 tdba;
1014 struct ixgbe_hw *hw = &adapter->hw;
1015 u32 i, j, tdlen, txctrl;
1017 /* Setup the HW Tx Head and Tail descriptor pointers */
1018 for (i = 0; i < adapter->num_tx_queues; i++) {
1019 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1020 j = ring->reg_idx;
1021 tdba = ring->dma;
1022 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1023 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1024 (tdba & DMA_BIT_MASK(32)));
1025 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1026 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1027 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1028 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1029 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1030 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1031 /* Disable Tx Head Writeback RO bit, since this hoses
1032 * bookkeeping if things aren't delivered in order.
1034 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1035 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1036 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1040 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1042 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1044 struct ixgbevf_ring *rx_ring;
1045 struct ixgbe_hw *hw = &adapter->hw;
1046 u32 srrctl;
1048 rx_ring = &adapter->rx_ring[index];
1050 srrctl = IXGBE_SRRCTL_DROP_EN;
1052 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1054 srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1055 IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1057 IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1060 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1062 struct ixgbe_hw *hw = &adapter->hw;
1063 struct net_device *netdev = adapter->netdev;
1064 int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1065 int i;
1066 u16 rx_buf_len;
1068 /* notify the PF of our intent to use this size of frame */
1069 ixgbevf_rlpml_set_vf(hw, max_frame);
1071 /* PF will allow an extra 4 bytes past for vlan tagged frames */
1072 max_frame += VLAN_HLEN;
1075 * Make best use of allocation by using all but 1K of a
1076 * power of 2 allocation that will be used for skb->head.
1078 if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1079 (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1080 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1081 else if (max_frame <= IXGBEVF_RXBUFFER_3K)
1082 rx_buf_len = IXGBEVF_RXBUFFER_3K;
1083 else if (max_frame <= IXGBEVF_RXBUFFER_7K)
1084 rx_buf_len = IXGBEVF_RXBUFFER_7K;
1085 else if (max_frame <= IXGBEVF_RXBUFFER_15K)
1086 rx_buf_len = IXGBEVF_RXBUFFER_15K;
1087 else
1088 rx_buf_len = IXGBEVF_MAX_RXBUFFER;
1090 for (i = 0; i < adapter->num_rx_queues; i++)
1091 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1095 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1096 * @adapter: board private structure
1098 * Configure the Rx unit of the MAC after a reset.
1100 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1102 u64 rdba;
1103 struct ixgbe_hw *hw = &adapter->hw;
1104 int i, j;
1105 u32 rdlen;
1107 /* PSRTYPE must be initialized in 82599 */
1108 IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1110 /* set_rx_buffer_len must be called before ring initialization */
1111 ixgbevf_set_rx_buffer_len(adapter);
1113 rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1114 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1115 * the Base and Length of the Rx Descriptor Ring */
1116 for (i = 0; i < adapter->num_rx_queues; i++) {
1117 rdba = adapter->rx_ring[i].dma;
1118 j = adapter->rx_ring[i].reg_idx;
1119 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1120 (rdba & DMA_BIT_MASK(32)));
1121 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1122 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1123 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1124 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1125 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1126 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1128 ixgbevf_configure_srrctl(adapter, j);
1132 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1134 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1135 struct ixgbe_hw *hw = &adapter->hw;
1136 int err;
1138 if (!hw->mac.ops.set_vfta)
1139 return -EOPNOTSUPP;
1141 spin_lock_bh(&adapter->mbx_lock);
1143 /* add VID to filter table */
1144 err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1146 spin_unlock_bh(&adapter->mbx_lock);
1148 /* translate error return types so error makes sense */
1149 if (err == IXGBE_ERR_MBX)
1150 return -EIO;
1152 if (err == IXGBE_ERR_INVALID_ARGUMENT)
1153 return -EACCES;
1155 set_bit(vid, adapter->active_vlans);
1157 return err;
1160 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1162 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1163 struct ixgbe_hw *hw = &adapter->hw;
1164 int err = -EOPNOTSUPP;
1166 spin_lock_bh(&adapter->mbx_lock);
1168 /* remove VID from filter table */
1169 if (hw->mac.ops.set_vfta)
1170 err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1172 spin_unlock_bh(&adapter->mbx_lock);
1174 clear_bit(vid, adapter->active_vlans);
1176 return err;
1179 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1181 u16 vid;
1183 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1184 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1187 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1189 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1190 struct ixgbe_hw *hw = &adapter->hw;
1191 int count = 0;
1193 if ((netdev_uc_count(netdev)) > 10) {
1194 pr_err("Too many unicast filters - No Space\n");
1195 return -ENOSPC;
1198 if (!netdev_uc_empty(netdev)) {
1199 struct netdev_hw_addr *ha;
1200 netdev_for_each_uc_addr(ha, netdev) {
1201 hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1202 udelay(200);
1204 } else {
1206 * If the list is empty then send message to PF driver to
1207 * clear all macvlans on this VF.
1209 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1212 return count;
1216 * ixgbevf_set_rx_mode - Multicast set
1217 * @netdev: network interface device structure
1219 * The set_rx_method entry point is called whenever the multicast address
1220 * list or the network interface flags are updated. This routine is
1221 * responsible for configuring the hardware for proper multicast mode.
1223 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1225 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1226 struct ixgbe_hw *hw = &adapter->hw;
1228 spin_lock_bh(&adapter->mbx_lock);
1230 /* reprogram multicast list */
1231 if (hw->mac.ops.update_mc_addr_list)
1232 hw->mac.ops.update_mc_addr_list(hw, netdev);
1234 ixgbevf_write_uc_addr_list(netdev);
1236 spin_unlock_bh(&adapter->mbx_lock);
1239 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1241 int q_idx;
1242 struct ixgbevf_q_vector *q_vector;
1243 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1245 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1246 q_vector = adapter->q_vector[q_idx];
1247 napi_enable(&q_vector->napi);
1251 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1253 int q_idx;
1254 struct ixgbevf_q_vector *q_vector;
1255 int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1257 for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1258 q_vector = adapter->q_vector[q_idx];
1259 napi_disable(&q_vector->napi);
1263 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1265 struct net_device *netdev = adapter->netdev;
1266 int i;
1268 ixgbevf_set_rx_mode(netdev);
1270 ixgbevf_restore_vlan(adapter);
1272 ixgbevf_configure_tx(adapter);
1273 ixgbevf_configure_rx(adapter);
1274 for (i = 0; i < adapter->num_rx_queues; i++) {
1275 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1276 ixgbevf_alloc_rx_buffers(adapter, ring,
1277 IXGBE_DESC_UNUSED(ring));
1281 #define IXGBE_MAX_RX_DESC_POLL 10
1282 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1283 int rxr)
1285 struct ixgbe_hw *hw = &adapter->hw;
1286 int j = adapter->rx_ring[rxr].reg_idx;
1287 int k;
1289 for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1290 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1291 break;
1292 else
1293 msleep(1);
1295 if (k >= IXGBE_MAX_RX_DESC_POLL) {
1296 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1297 "not set within the polling period\n", rxr);
1300 ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1301 (adapter->rx_ring[rxr].count - 1));
1304 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1306 /* Only save pre-reset stats if there are some */
1307 if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1308 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1309 adapter->stats.base_vfgprc;
1310 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1311 adapter->stats.base_vfgptc;
1312 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1313 adapter->stats.base_vfgorc;
1314 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1315 adapter->stats.base_vfgotc;
1316 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1317 adapter->stats.base_vfmprc;
1321 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1323 struct ixgbe_hw *hw = &adapter->hw;
1325 adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1326 adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1327 adapter->stats.last_vfgorc |=
1328 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1329 adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1330 adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1331 adapter->stats.last_vfgotc |=
1332 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1333 adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1335 adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1336 adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1337 adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1338 adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1339 adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1342 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1344 struct ixgbe_hw *hw = &adapter->hw;
1345 int api[] = { ixgbe_mbox_api_11,
1346 ixgbe_mbox_api_10,
1347 ixgbe_mbox_api_unknown };
1348 int err = 0, idx = 0;
1350 spin_lock_bh(&adapter->mbx_lock);
1352 while (api[idx] != ixgbe_mbox_api_unknown) {
1353 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1354 if (!err)
1355 break;
1356 idx++;
1359 spin_unlock_bh(&adapter->mbx_lock);
1362 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1364 struct net_device *netdev = adapter->netdev;
1365 struct ixgbe_hw *hw = &adapter->hw;
1366 int i, j = 0;
1367 int num_rx_rings = adapter->num_rx_queues;
1368 u32 txdctl, rxdctl;
1370 for (i = 0; i < adapter->num_tx_queues; i++) {
1371 j = adapter->tx_ring[i].reg_idx;
1372 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1373 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1374 txdctl |= (8 << 16);
1375 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1378 for (i = 0; i < adapter->num_tx_queues; i++) {
1379 j = adapter->tx_ring[i].reg_idx;
1380 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1381 txdctl |= IXGBE_TXDCTL_ENABLE;
1382 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1385 for (i = 0; i < num_rx_rings; i++) {
1386 j = adapter->rx_ring[i].reg_idx;
1387 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1388 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1389 if (hw->mac.type == ixgbe_mac_X540_vf) {
1390 rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1391 rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1392 IXGBE_RXDCTL_RLPML_EN);
1394 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1395 ixgbevf_rx_desc_queue_enable(adapter, i);
1398 ixgbevf_configure_msix(adapter);
1400 spin_lock_bh(&adapter->mbx_lock);
1402 if (hw->mac.ops.set_rar) {
1403 if (is_valid_ether_addr(hw->mac.addr))
1404 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1405 else
1406 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1409 spin_unlock_bh(&adapter->mbx_lock);
1411 clear_bit(__IXGBEVF_DOWN, &adapter->state);
1412 ixgbevf_napi_enable_all(adapter);
1414 /* enable transmits */
1415 netif_tx_start_all_queues(netdev);
1417 ixgbevf_save_reset_stats(adapter);
1418 ixgbevf_init_last_counter_stats(adapter);
1420 hw->mac.get_link_status = 1;
1421 mod_timer(&adapter->watchdog_timer, jiffies);
1424 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1426 struct ixgbe_hw *hw = &adapter->hw;
1427 struct ixgbevf_ring *rx_ring;
1428 unsigned int def_q = 0;
1429 unsigned int num_tcs = 0;
1430 unsigned int num_rx_queues = 1;
1431 int err, i;
1433 spin_lock_bh(&adapter->mbx_lock);
1435 /* fetch queue configuration from the PF */
1436 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1438 spin_unlock_bh(&adapter->mbx_lock);
1440 if (err)
1441 return err;
1443 if (num_tcs > 1) {
1444 /* update default Tx ring register index */
1445 adapter->tx_ring[0].reg_idx = def_q;
1447 /* we need as many queues as traffic classes */
1448 num_rx_queues = num_tcs;
1451 /* nothing to do if we have the correct number of queues */
1452 if (adapter->num_rx_queues == num_rx_queues)
1453 return 0;
1455 /* allocate new rings */
1456 rx_ring = kcalloc(num_rx_queues,
1457 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1458 if (!rx_ring)
1459 return -ENOMEM;
1461 /* setup ring fields */
1462 for (i = 0; i < num_rx_queues; i++) {
1463 rx_ring[i].count = adapter->rx_ring_count;
1464 rx_ring[i].queue_index = i;
1465 rx_ring[i].reg_idx = i;
1466 rx_ring[i].dev = &adapter->pdev->dev;
1467 rx_ring[i].netdev = adapter->netdev;
1469 /* allocate resources on the ring */
1470 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1471 if (err) {
1472 while (i) {
1473 i--;
1474 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1476 kfree(rx_ring);
1477 return err;
1481 /* free the existing rings and queues */
1482 ixgbevf_free_all_rx_resources(adapter);
1483 adapter->num_rx_queues = 0;
1484 kfree(adapter->rx_ring);
1486 /* move new rings into position on the adapter struct */
1487 adapter->rx_ring = rx_ring;
1488 adapter->num_rx_queues = num_rx_queues;
1490 /* reset ring to vector mapping */
1491 ixgbevf_reset_q_vectors(adapter);
1492 ixgbevf_map_rings_to_vectors(adapter);
1494 return 0;
1497 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1499 struct ixgbe_hw *hw = &adapter->hw;
1501 ixgbevf_negotiate_api(adapter);
1503 ixgbevf_reset_queues(adapter);
1505 ixgbevf_configure(adapter);
1507 ixgbevf_up_complete(adapter);
1509 /* clear any pending interrupts, may auto mask */
1510 IXGBE_READ_REG(hw, IXGBE_VTEICR);
1512 ixgbevf_irq_enable(adapter);
1516 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1517 * @adapter: board private structure
1518 * @rx_ring: ring to free buffers from
1520 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1521 struct ixgbevf_ring *rx_ring)
1523 struct pci_dev *pdev = adapter->pdev;
1524 unsigned long size;
1525 unsigned int i;
1527 if (!rx_ring->rx_buffer_info)
1528 return;
1530 /* Free all the Rx ring sk_buffs */
1531 for (i = 0; i < rx_ring->count; i++) {
1532 struct ixgbevf_rx_buffer *rx_buffer_info;
1534 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1535 if (rx_buffer_info->dma) {
1536 dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1537 rx_ring->rx_buf_len,
1538 DMA_FROM_DEVICE);
1539 rx_buffer_info->dma = 0;
1541 if (rx_buffer_info->skb) {
1542 struct sk_buff *skb = rx_buffer_info->skb;
1543 rx_buffer_info->skb = NULL;
1544 do {
1545 struct sk_buff *this = skb;
1546 skb = IXGBE_CB(skb)->prev;
1547 dev_kfree_skb(this);
1548 } while (skb);
1552 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1553 memset(rx_ring->rx_buffer_info, 0, size);
1555 /* Zero out the descriptor ring */
1556 memset(rx_ring->desc, 0, rx_ring->size);
1558 rx_ring->next_to_clean = 0;
1559 rx_ring->next_to_use = 0;
1561 if (rx_ring->head)
1562 writel(0, adapter->hw.hw_addr + rx_ring->head);
1563 if (rx_ring->tail)
1564 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1568 * ixgbevf_clean_tx_ring - Free Tx Buffers
1569 * @adapter: board private structure
1570 * @tx_ring: ring to be cleaned
1572 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1573 struct ixgbevf_ring *tx_ring)
1575 struct ixgbevf_tx_buffer *tx_buffer_info;
1576 unsigned long size;
1577 unsigned int i;
1579 if (!tx_ring->tx_buffer_info)
1580 return;
1582 /* Free all the Tx ring sk_buffs */
1584 for (i = 0; i < tx_ring->count; i++) {
1585 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1586 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1589 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1590 memset(tx_ring->tx_buffer_info, 0, size);
1592 memset(tx_ring->desc, 0, tx_ring->size);
1594 tx_ring->next_to_use = 0;
1595 tx_ring->next_to_clean = 0;
1597 if (tx_ring->head)
1598 writel(0, adapter->hw.hw_addr + tx_ring->head);
1599 if (tx_ring->tail)
1600 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1604 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1605 * @adapter: board private structure
1607 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1609 int i;
1611 for (i = 0; i < adapter->num_rx_queues; i++)
1612 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1616 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1617 * @adapter: board private structure
1619 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1621 int i;
1623 for (i = 0; i < adapter->num_tx_queues; i++)
1624 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1627 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1629 struct net_device *netdev = adapter->netdev;
1630 struct ixgbe_hw *hw = &adapter->hw;
1631 u32 txdctl;
1632 int i, j;
1634 /* signal that we are down to the interrupt handler */
1635 set_bit(__IXGBEVF_DOWN, &adapter->state);
1636 /* disable receives */
1638 netif_tx_disable(netdev);
1640 msleep(10);
1642 netif_tx_stop_all_queues(netdev);
1644 ixgbevf_irq_disable(adapter);
1646 ixgbevf_napi_disable_all(adapter);
1648 del_timer_sync(&adapter->watchdog_timer);
1649 /* can't call flush scheduled work here because it can deadlock
1650 * if linkwatch_event tries to acquire the rtnl_lock which we are
1651 * holding */
1652 while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1653 msleep(1);
1655 /* disable transmits in the hardware now that interrupts are off */
1656 for (i = 0; i < adapter->num_tx_queues; i++) {
1657 j = adapter->tx_ring[i].reg_idx;
1658 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1659 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1660 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1663 netif_carrier_off(netdev);
1665 if (!pci_channel_offline(adapter->pdev))
1666 ixgbevf_reset(adapter);
1668 ixgbevf_clean_all_tx_rings(adapter);
1669 ixgbevf_clean_all_rx_rings(adapter);
1672 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1674 WARN_ON(in_interrupt());
1676 while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1677 msleep(1);
1680 * Check if PF is up before re-init. If not then skip until
1681 * later when the PF is up and ready to service requests from
1682 * the VF via mailbox. If the VF is up and running then the
1683 * watchdog task will continue to schedule reset tasks until
1684 * the PF is up and running.
1686 ixgbevf_down(adapter);
1687 ixgbevf_up(adapter);
1689 clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1692 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1694 struct ixgbe_hw *hw = &adapter->hw;
1695 struct net_device *netdev = adapter->netdev;
1697 spin_lock_bh(&adapter->mbx_lock);
1699 if (hw->mac.ops.reset_hw(hw))
1700 hw_dbg(hw, "PF still resetting\n");
1701 else
1702 hw->mac.ops.init_hw(hw);
1704 spin_unlock_bh(&adapter->mbx_lock);
1706 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1707 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1708 netdev->addr_len);
1709 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1710 netdev->addr_len);
1714 static void ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1715 int vectors)
1717 int err, vector_threshold;
1719 /* We'll want at least 2 (vector_threshold):
1720 * 1) TxQ[0] + RxQ[0] handler
1721 * 2) Other (Link Status Change, etc.)
1723 vector_threshold = MIN_MSIX_COUNT;
1725 /* The more we get, the more we will assign to Tx/Rx Cleanup
1726 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1727 * Right now, we simply care about how many we'll get; we'll
1728 * set them up later while requesting irq's.
1730 while (vectors >= vector_threshold) {
1731 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1732 vectors);
1733 if (!err) /* Success in acquiring all requested vectors. */
1734 break;
1735 else if (err < 0)
1736 vectors = 0; /* Nasty failure, quit now */
1737 else /* err == number of vectors we should try again with */
1738 vectors = err;
1741 if (vectors < vector_threshold) {
1742 /* Can't allocate enough MSI-X interrupts? Oh well.
1743 * This just means we'll go with either a single MSI
1744 * vector or fall back to legacy interrupts.
1746 hw_dbg(&adapter->hw,
1747 "Unable to allocate MSI-X interrupts\n");
1748 kfree(adapter->msix_entries);
1749 adapter->msix_entries = NULL;
1750 } else {
1752 * Adjust for only the vectors we'll use, which is minimum
1753 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1754 * vectors we were allocated.
1756 adapter->num_msix_vectors = vectors;
1761 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1762 * @adapter: board private structure to initialize
1764 * This is the top level queue allocation routine. The order here is very
1765 * important, starting with the "most" number of features turned on at once,
1766 * and ending with the smallest set of features. This way large combinations
1767 * can be allocated if they're turned on, and smaller combinations are the
1768 * fallthrough conditions.
1771 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1773 /* Start with base case */
1774 adapter->num_rx_queues = 1;
1775 adapter->num_tx_queues = 1;
1779 * ixgbevf_alloc_queues - Allocate memory for all rings
1780 * @adapter: board private structure to initialize
1782 * We allocate one ring per queue at run-time since we don't know the
1783 * number of queues at compile-time. The polling_netdev array is
1784 * intended for Multiqueue, but should work fine with a single queue.
1786 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1788 int i;
1790 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1791 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1792 if (!adapter->tx_ring)
1793 goto err_tx_ring_allocation;
1795 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1796 sizeof(struct ixgbevf_ring), GFP_KERNEL);
1797 if (!adapter->rx_ring)
1798 goto err_rx_ring_allocation;
1800 for (i = 0; i < adapter->num_tx_queues; i++) {
1801 adapter->tx_ring[i].count = adapter->tx_ring_count;
1802 adapter->tx_ring[i].queue_index = i;
1803 /* reg_idx may be remapped later by DCB config */
1804 adapter->tx_ring[i].reg_idx = i;
1805 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1806 adapter->tx_ring[i].netdev = adapter->netdev;
1809 for (i = 0; i < adapter->num_rx_queues; i++) {
1810 adapter->rx_ring[i].count = adapter->rx_ring_count;
1811 adapter->rx_ring[i].queue_index = i;
1812 adapter->rx_ring[i].reg_idx = i;
1813 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1814 adapter->rx_ring[i].netdev = adapter->netdev;
1817 return 0;
1819 err_rx_ring_allocation:
1820 kfree(adapter->tx_ring);
1821 err_tx_ring_allocation:
1822 return -ENOMEM;
1826 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1827 * @adapter: board private structure to initialize
1829 * Attempt to configure the interrupts using the best available
1830 * capabilities of the hardware and the kernel.
1832 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1834 struct net_device *netdev = adapter->netdev;
1835 int err = 0;
1836 int vector, v_budget;
1839 * It's easy to be greedy for MSI-X vectors, but it really
1840 * doesn't do us much good if we have a lot more vectors
1841 * than CPU's. So let's be conservative and only ask for
1842 * (roughly) the same number of vectors as there are CPU's.
1843 * The default is to use pairs of vectors.
1845 v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1846 v_budget = min_t(int, v_budget, num_online_cpus());
1847 v_budget += NON_Q_VECTORS;
1849 /* A failure in MSI-X entry allocation isn't fatal, but it does
1850 * mean we disable MSI-X capabilities of the adapter. */
1851 adapter->msix_entries = kcalloc(v_budget,
1852 sizeof(struct msix_entry), GFP_KERNEL);
1853 if (!adapter->msix_entries) {
1854 err = -ENOMEM;
1855 goto out;
1858 for (vector = 0; vector < v_budget; vector++)
1859 adapter->msix_entries[vector].entry = vector;
1861 ixgbevf_acquire_msix_vectors(adapter, v_budget);
1863 err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1864 if (err)
1865 goto out;
1867 err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1869 out:
1870 return err;
1874 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1875 * @adapter: board private structure to initialize
1877 * We allocate one q_vector per queue interrupt. If allocation fails we
1878 * return -ENOMEM.
1880 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1882 int q_idx, num_q_vectors;
1883 struct ixgbevf_q_vector *q_vector;
1885 num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1887 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1888 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1889 if (!q_vector)
1890 goto err_out;
1891 q_vector->adapter = adapter;
1892 q_vector->v_idx = q_idx;
1893 netif_napi_add(adapter->netdev, &q_vector->napi,
1894 ixgbevf_poll, 64);
1895 adapter->q_vector[q_idx] = q_vector;
1898 return 0;
1900 err_out:
1901 while (q_idx) {
1902 q_idx--;
1903 q_vector = adapter->q_vector[q_idx];
1904 netif_napi_del(&q_vector->napi);
1905 kfree(q_vector);
1906 adapter->q_vector[q_idx] = NULL;
1908 return -ENOMEM;
1912 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1913 * @adapter: board private structure to initialize
1915 * This function frees the memory allocated to the q_vectors. In addition if
1916 * NAPI is enabled it will delete any references to the NAPI struct prior
1917 * to freeing the q_vector.
1919 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1921 int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1923 for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1924 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1926 adapter->q_vector[q_idx] = NULL;
1927 netif_napi_del(&q_vector->napi);
1928 kfree(q_vector);
1933 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1934 * @adapter: board private structure
1937 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1939 pci_disable_msix(adapter->pdev);
1940 kfree(adapter->msix_entries);
1941 adapter->msix_entries = NULL;
1945 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1946 * @adapter: board private structure to initialize
1949 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1951 int err;
1953 /* Number of supported queues */
1954 ixgbevf_set_num_queues(adapter);
1956 err = ixgbevf_set_interrupt_capability(adapter);
1957 if (err) {
1958 hw_dbg(&adapter->hw,
1959 "Unable to setup interrupt capabilities\n");
1960 goto err_set_interrupt;
1963 err = ixgbevf_alloc_q_vectors(adapter);
1964 if (err) {
1965 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
1966 "vectors\n");
1967 goto err_alloc_q_vectors;
1970 err = ixgbevf_alloc_queues(adapter);
1971 if (err) {
1972 pr_err("Unable to allocate memory for queues\n");
1973 goto err_alloc_queues;
1976 hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
1977 "Tx Queue count = %u\n",
1978 (adapter->num_rx_queues > 1) ? "Enabled" :
1979 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
1981 set_bit(__IXGBEVF_DOWN, &adapter->state);
1983 return 0;
1984 err_alloc_queues:
1985 ixgbevf_free_q_vectors(adapter);
1986 err_alloc_q_vectors:
1987 ixgbevf_reset_interrupt_capability(adapter);
1988 err_set_interrupt:
1989 return err;
1993 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
1994 * @adapter: board private structure to clear interrupt scheme on
1996 * We go through and clear interrupt specific resources and reset the structure
1997 * to pre-load conditions
1999 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2001 adapter->num_tx_queues = 0;
2002 adapter->num_rx_queues = 0;
2004 ixgbevf_free_q_vectors(adapter);
2005 ixgbevf_reset_interrupt_capability(adapter);
2009 * ixgbevf_sw_init - Initialize general software structures
2010 * (struct ixgbevf_adapter)
2011 * @adapter: board private structure to initialize
2013 * ixgbevf_sw_init initializes the Adapter private data structure.
2014 * Fields are initialized based on PCI device information and
2015 * OS network device settings (MTU size).
2017 static int __devinit ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2019 struct ixgbe_hw *hw = &adapter->hw;
2020 struct pci_dev *pdev = adapter->pdev;
2021 int err;
2023 /* PCI config space info */
2025 hw->vendor_id = pdev->vendor;
2026 hw->device_id = pdev->device;
2027 hw->revision_id = pdev->revision;
2028 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2029 hw->subsystem_device_id = pdev->subsystem_device;
2031 hw->mbx.ops.init_params(hw);
2033 /* assume legacy case in which PF would only give VF 2 queues */
2034 hw->mac.max_tx_queues = 2;
2035 hw->mac.max_rx_queues = 2;
2037 err = hw->mac.ops.reset_hw(hw);
2038 if (err) {
2039 dev_info(&pdev->dev,
2040 "PF still in reset state, assigning new address\n");
2041 eth_hw_addr_random(adapter->netdev);
2042 memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
2043 adapter->netdev->addr_len);
2044 } else {
2045 err = hw->mac.ops.init_hw(hw);
2046 if (err) {
2047 pr_err("init_shared_code failed: %d\n", err);
2048 goto out;
2050 memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
2051 adapter->netdev->addr_len);
2054 /* lock to protect mailbox accesses */
2055 spin_lock_init(&adapter->mbx_lock);
2057 /* Enable dynamic interrupt throttling rates */
2058 adapter->rx_itr_setting = 1;
2059 adapter->tx_itr_setting = 1;
2061 /* set default ring sizes */
2062 adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2063 adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2065 set_bit(__IXGBEVF_DOWN, &adapter->state);
2066 return 0;
2068 out:
2069 return err;
2072 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2074 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2075 if (current_counter < last_counter) \
2076 counter += 0x100000000LL; \
2077 last_counter = current_counter; \
2078 counter &= 0xFFFFFFFF00000000LL; \
2079 counter |= current_counter; \
2082 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2084 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2085 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2086 u64 current_counter = (current_counter_msb << 32) | \
2087 current_counter_lsb; \
2088 if (current_counter < last_counter) \
2089 counter += 0x1000000000LL; \
2090 last_counter = current_counter; \
2091 counter &= 0xFFFFFFF000000000LL; \
2092 counter |= current_counter; \
2095 * ixgbevf_update_stats - Update the board statistics counters.
2096 * @adapter: board private structure
2098 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2100 struct ixgbe_hw *hw = &adapter->hw;
2102 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2103 adapter->stats.vfgprc);
2104 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2105 adapter->stats.vfgptc);
2106 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2107 adapter->stats.last_vfgorc,
2108 adapter->stats.vfgorc);
2109 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2110 adapter->stats.last_vfgotc,
2111 adapter->stats.vfgotc);
2112 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2113 adapter->stats.vfmprc);
2117 * ixgbevf_watchdog - Timer Call-back
2118 * @data: pointer to adapter cast into an unsigned long
2120 static void ixgbevf_watchdog(unsigned long data)
2122 struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2123 struct ixgbe_hw *hw = &adapter->hw;
2124 u32 eics = 0;
2125 int i;
2128 * Do the watchdog outside of interrupt context due to the lovely
2129 * delays that some of the newer hardware requires
2132 if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2133 goto watchdog_short_circuit;
2135 /* get one bit for every active tx/rx interrupt vector */
2136 for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2137 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2138 if (qv->rx.ring || qv->tx.ring)
2139 eics |= 1 << i;
2142 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2144 watchdog_short_circuit:
2145 schedule_work(&adapter->watchdog_task);
2149 * ixgbevf_tx_timeout - Respond to a Tx Hang
2150 * @netdev: network interface device structure
2152 static void ixgbevf_tx_timeout(struct net_device *netdev)
2154 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2156 /* Do the reset outside of interrupt context */
2157 schedule_work(&adapter->reset_task);
2160 static void ixgbevf_reset_task(struct work_struct *work)
2162 struct ixgbevf_adapter *adapter;
2163 adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2165 /* If we're already down or resetting, just bail */
2166 if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2167 test_bit(__IXGBEVF_RESETTING, &adapter->state))
2168 return;
2170 adapter->tx_timeout_count++;
2172 ixgbevf_reinit_locked(adapter);
2176 * ixgbevf_watchdog_task - worker thread to bring link up
2177 * @work: pointer to work_struct containing our data
2179 static void ixgbevf_watchdog_task(struct work_struct *work)
2181 struct ixgbevf_adapter *adapter = container_of(work,
2182 struct ixgbevf_adapter,
2183 watchdog_task);
2184 struct net_device *netdev = adapter->netdev;
2185 struct ixgbe_hw *hw = &adapter->hw;
2186 u32 link_speed = adapter->link_speed;
2187 bool link_up = adapter->link_up;
2189 adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2192 * Always check the link on the watchdog because we have
2193 * no LSC interrupt
2195 if (hw->mac.ops.check_link) {
2196 s32 need_reset;
2198 spin_lock_bh(&adapter->mbx_lock);
2200 need_reset = hw->mac.ops.check_link(hw, &link_speed,
2201 &link_up, false);
2203 spin_unlock_bh(&adapter->mbx_lock);
2205 if (need_reset) {
2206 adapter->link_up = link_up;
2207 adapter->link_speed = link_speed;
2208 netif_carrier_off(netdev);
2209 netif_tx_stop_all_queues(netdev);
2210 schedule_work(&adapter->reset_task);
2211 goto pf_has_reset;
2213 } else {
2214 /* always assume link is up, if no check link
2215 * function */
2216 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
2217 link_up = true;
2219 adapter->link_up = link_up;
2220 adapter->link_speed = link_speed;
2222 if (link_up) {
2223 if (!netif_carrier_ok(netdev)) {
2224 hw_dbg(&adapter->hw, "NIC Link is Up, %u Gbps\n",
2225 (link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2226 10 : 1);
2227 netif_carrier_on(netdev);
2228 netif_tx_wake_all_queues(netdev);
2230 } else {
2231 adapter->link_up = false;
2232 adapter->link_speed = 0;
2233 if (netif_carrier_ok(netdev)) {
2234 hw_dbg(&adapter->hw, "NIC Link is Down\n");
2235 netif_carrier_off(netdev);
2236 netif_tx_stop_all_queues(netdev);
2240 ixgbevf_update_stats(adapter);
2242 pf_has_reset:
2243 /* Reset the timer */
2244 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2245 mod_timer(&adapter->watchdog_timer,
2246 round_jiffies(jiffies + (2 * HZ)));
2248 adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2252 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2253 * @adapter: board private structure
2254 * @tx_ring: Tx descriptor ring for a specific queue
2256 * Free all transmit software resources
2258 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2259 struct ixgbevf_ring *tx_ring)
2261 struct pci_dev *pdev = adapter->pdev;
2263 ixgbevf_clean_tx_ring(adapter, tx_ring);
2265 vfree(tx_ring->tx_buffer_info);
2266 tx_ring->tx_buffer_info = NULL;
2268 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2269 tx_ring->dma);
2271 tx_ring->desc = NULL;
2275 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2276 * @adapter: board private structure
2278 * Free all transmit software resources
2280 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2282 int i;
2284 for (i = 0; i < adapter->num_tx_queues; i++)
2285 if (adapter->tx_ring[i].desc)
2286 ixgbevf_free_tx_resources(adapter,
2287 &adapter->tx_ring[i]);
2292 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2293 * @adapter: board private structure
2294 * @tx_ring: tx descriptor ring (for a specific queue) to setup
2296 * Return 0 on success, negative on failure
2298 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2299 struct ixgbevf_ring *tx_ring)
2301 struct pci_dev *pdev = adapter->pdev;
2302 int size;
2304 size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2305 tx_ring->tx_buffer_info = vzalloc(size);
2306 if (!tx_ring->tx_buffer_info)
2307 goto err;
2309 /* round up to nearest 4K */
2310 tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2311 tx_ring->size = ALIGN(tx_ring->size, 4096);
2313 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2314 &tx_ring->dma, GFP_KERNEL);
2315 if (!tx_ring->desc)
2316 goto err;
2318 tx_ring->next_to_use = 0;
2319 tx_ring->next_to_clean = 0;
2320 return 0;
2322 err:
2323 vfree(tx_ring->tx_buffer_info);
2324 tx_ring->tx_buffer_info = NULL;
2325 hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2326 "descriptor ring\n");
2327 return -ENOMEM;
2331 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2332 * @adapter: board private structure
2334 * If this function returns with an error, then it's possible one or
2335 * more of the rings is populated (while the rest are not). It is the
2336 * callers duty to clean those orphaned rings.
2338 * Return 0 on success, negative on failure
2340 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2342 int i, err = 0;
2344 for (i = 0; i < adapter->num_tx_queues; i++) {
2345 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2346 if (!err)
2347 continue;
2348 hw_dbg(&adapter->hw,
2349 "Allocation for Tx Queue %u failed\n", i);
2350 break;
2353 return err;
2357 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2358 * @adapter: board private structure
2359 * @rx_ring: rx descriptor ring (for a specific queue) to setup
2361 * Returns 0 on success, negative on failure
2363 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2364 struct ixgbevf_ring *rx_ring)
2366 struct pci_dev *pdev = adapter->pdev;
2367 int size;
2369 size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2370 rx_ring->rx_buffer_info = vzalloc(size);
2371 if (!rx_ring->rx_buffer_info)
2372 goto alloc_failed;
2374 /* Round up to nearest 4K */
2375 rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2376 rx_ring->size = ALIGN(rx_ring->size, 4096);
2378 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2379 &rx_ring->dma, GFP_KERNEL);
2381 if (!rx_ring->desc) {
2382 hw_dbg(&adapter->hw,
2383 "Unable to allocate memory for "
2384 "the receive descriptor ring\n");
2385 vfree(rx_ring->rx_buffer_info);
2386 rx_ring->rx_buffer_info = NULL;
2387 goto alloc_failed;
2390 rx_ring->next_to_clean = 0;
2391 rx_ring->next_to_use = 0;
2393 return 0;
2394 alloc_failed:
2395 return -ENOMEM;
2399 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2400 * @adapter: board private structure
2402 * If this function returns with an error, then it's possible one or
2403 * more of the rings is populated (while the rest are not). It is the
2404 * callers duty to clean those orphaned rings.
2406 * Return 0 on success, negative on failure
2408 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2410 int i, err = 0;
2412 for (i = 0; i < adapter->num_rx_queues; i++) {
2413 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2414 if (!err)
2415 continue;
2416 hw_dbg(&adapter->hw,
2417 "Allocation for Rx Queue %u failed\n", i);
2418 break;
2420 return err;
2424 * ixgbevf_free_rx_resources - Free Rx Resources
2425 * @adapter: board private structure
2426 * @rx_ring: ring to clean the resources from
2428 * Free all receive software resources
2430 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2431 struct ixgbevf_ring *rx_ring)
2433 struct pci_dev *pdev = adapter->pdev;
2435 ixgbevf_clean_rx_ring(adapter, rx_ring);
2437 vfree(rx_ring->rx_buffer_info);
2438 rx_ring->rx_buffer_info = NULL;
2440 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2441 rx_ring->dma);
2443 rx_ring->desc = NULL;
2447 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2448 * @adapter: board private structure
2450 * Free all receive software resources
2452 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2454 int i;
2456 for (i = 0; i < adapter->num_rx_queues; i++)
2457 if (adapter->rx_ring[i].desc)
2458 ixgbevf_free_rx_resources(adapter,
2459 &adapter->rx_ring[i]);
2462 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2464 struct ixgbe_hw *hw = &adapter->hw;
2465 struct ixgbevf_ring *rx_ring;
2466 unsigned int def_q = 0;
2467 unsigned int num_tcs = 0;
2468 unsigned int num_rx_queues = 1;
2469 int err, i;
2471 spin_lock_bh(&adapter->mbx_lock);
2473 /* fetch queue configuration from the PF */
2474 err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2476 spin_unlock_bh(&adapter->mbx_lock);
2478 if (err)
2479 return err;
2481 if (num_tcs > 1) {
2482 /* update default Tx ring register index */
2483 adapter->tx_ring[0].reg_idx = def_q;
2485 /* we need as many queues as traffic classes */
2486 num_rx_queues = num_tcs;
2489 /* nothing to do if we have the correct number of queues */
2490 if (adapter->num_rx_queues == num_rx_queues)
2491 return 0;
2493 /* allocate new rings */
2494 rx_ring = kcalloc(num_rx_queues,
2495 sizeof(struct ixgbevf_ring), GFP_KERNEL);
2496 if (!rx_ring)
2497 return -ENOMEM;
2499 /* setup ring fields */
2500 for (i = 0; i < num_rx_queues; i++) {
2501 rx_ring[i].count = adapter->rx_ring_count;
2502 rx_ring[i].queue_index = i;
2503 rx_ring[i].reg_idx = i;
2504 rx_ring[i].dev = &adapter->pdev->dev;
2505 rx_ring[i].netdev = adapter->netdev;
2508 /* free the existing ring and queues */
2509 adapter->num_rx_queues = 0;
2510 kfree(adapter->rx_ring);
2512 /* move new rings into position on the adapter struct */
2513 adapter->rx_ring = rx_ring;
2514 adapter->num_rx_queues = num_rx_queues;
2516 return 0;
2520 * ixgbevf_open - Called when a network interface is made active
2521 * @netdev: network interface device structure
2523 * Returns 0 on success, negative value on failure
2525 * The open entry point is called when a network interface is made
2526 * active by the system (IFF_UP). At this point all resources needed
2527 * for transmit and receive operations are allocated, the interrupt
2528 * handler is registered with the OS, the watchdog timer is started,
2529 * and the stack is notified that the interface is ready.
2531 static int ixgbevf_open(struct net_device *netdev)
2533 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2534 struct ixgbe_hw *hw = &adapter->hw;
2535 int err;
2537 /* disallow open during test */
2538 if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2539 return -EBUSY;
2541 if (hw->adapter_stopped) {
2542 ixgbevf_reset(adapter);
2543 /* if adapter is still stopped then PF isn't up and
2544 * the vf can't start. */
2545 if (hw->adapter_stopped) {
2546 err = IXGBE_ERR_MBX;
2547 pr_err("Unable to start - perhaps the PF Driver isn't "
2548 "up yet\n");
2549 goto err_setup_reset;
2553 ixgbevf_negotiate_api(adapter);
2555 /* setup queue reg_idx and Rx queue count */
2556 err = ixgbevf_setup_queues(adapter);
2557 if (err)
2558 goto err_setup_queues;
2560 /* allocate transmit descriptors */
2561 err = ixgbevf_setup_all_tx_resources(adapter);
2562 if (err)
2563 goto err_setup_tx;
2565 /* allocate receive descriptors */
2566 err = ixgbevf_setup_all_rx_resources(adapter);
2567 if (err)
2568 goto err_setup_rx;
2570 ixgbevf_configure(adapter);
2573 * Map the Tx/Rx rings to the vectors we were allotted.
2574 * if request_irq will be called in this function map_rings
2575 * must be called *before* up_complete
2577 ixgbevf_map_rings_to_vectors(adapter);
2579 ixgbevf_up_complete(adapter);
2581 /* clear any pending interrupts, may auto mask */
2582 IXGBE_READ_REG(hw, IXGBE_VTEICR);
2583 err = ixgbevf_request_irq(adapter);
2584 if (err)
2585 goto err_req_irq;
2587 ixgbevf_irq_enable(adapter);
2589 return 0;
2591 err_req_irq:
2592 ixgbevf_down(adapter);
2593 ixgbevf_free_irq(adapter);
2594 err_setup_rx:
2595 ixgbevf_free_all_rx_resources(adapter);
2596 err_setup_tx:
2597 ixgbevf_free_all_tx_resources(adapter);
2598 err_setup_queues:
2599 ixgbevf_reset(adapter);
2601 err_setup_reset:
2603 return err;
2607 * ixgbevf_close - Disables a network interface
2608 * @netdev: network interface device structure
2610 * Returns 0, this is not allowed to fail
2612 * The close entry point is called when an interface is de-activated
2613 * by the OS. The hardware is still under the drivers control, but
2614 * needs to be disabled. A global MAC reset is issued to stop the
2615 * hardware, and all transmit and receive resources are freed.
2617 static int ixgbevf_close(struct net_device *netdev)
2619 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2621 ixgbevf_down(adapter);
2622 ixgbevf_free_irq(adapter);
2624 ixgbevf_free_all_tx_resources(adapter);
2625 ixgbevf_free_all_rx_resources(adapter);
2627 return 0;
2630 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2631 u32 vlan_macip_lens, u32 type_tucmd,
2632 u32 mss_l4len_idx)
2634 struct ixgbe_adv_tx_context_desc *context_desc;
2635 u16 i = tx_ring->next_to_use;
2637 context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2639 i++;
2640 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2642 /* set bits to identify this as an advanced context descriptor */
2643 type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2645 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2646 context_desc->seqnum_seed = 0;
2647 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
2648 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2651 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2652 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2654 u32 vlan_macip_lens, type_tucmd;
2655 u32 mss_l4len_idx, l4len;
2657 if (!skb_is_gso(skb))
2658 return 0;
2660 if (skb_header_cloned(skb)) {
2661 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2662 if (err)
2663 return err;
2666 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2667 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2669 if (skb->protocol == htons(ETH_P_IP)) {
2670 struct iphdr *iph = ip_hdr(skb);
2671 iph->tot_len = 0;
2672 iph->check = 0;
2673 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2674 iph->daddr, 0,
2675 IPPROTO_TCP,
2677 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2678 } else if (skb_is_gso_v6(skb)) {
2679 ipv6_hdr(skb)->payload_len = 0;
2680 tcp_hdr(skb)->check =
2681 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2682 &ipv6_hdr(skb)->daddr,
2683 0, IPPROTO_TCP, 0);
2686 /* compute header lengths */
2687 l4len = tcp_hdrlen(skb);
2688 *hdr_len += l4len;
2689 *hdr_len = skb_transport_offset(skb) + l4len;
2691 /* mss_l4len_id: use 1 as index for TSO */
2692 mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2693 mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2694 mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2696 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2697 vlan_macip_lens = skb_network_header_len(skb);
2698 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2699 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2701 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2702 type_tucmd, mss_l4len_idx);
2704 return 1;
2707 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2708 struct sk_buff *skb, u32 tx_flags)
2713 u32 vlan_macip_lens = 0;
2714 u32 mss_l4len_idx = 0;
2715 u32 type_tucmd = 0;
2717 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2718 u8 l4_hdr = 0;
2719 switch (skb->protocol) {
2720 case __constant_htons(ETH_P_IP):
2721 vlan_macip_lens |= skb_network_header_len(skb);
2722 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2723 l4_hdr = ip_hdr(skb)->protocol;
2724 break;
2725 case __constant_htons(ETH_P_IPV6):
2726 vlan_macip_lens |= skb_network_header_len(skb);
2727 l4_hdr = ipv6_hdr(skb)->nexthdr;
2728 break;
2729 default:
2730 if (unlikely(net_ratelimit())) {
2731 dev_warn(tx_ring->dev,
2732 "partial checksum but proto=%x!\n",
2733 skb->protocol);
2735 break;
2738 switch (l4_hdr) {
2739 case IPPROTO_TCP:
2740 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2741 mss_l4len_idx = tcp_hdrlen(skb) <<
2742 IXGBE_ADVTXD_L4LEN_SHIFT;
2743 break;
2744 case IPPROTO_SCTP:
2745 type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2746 mss_l4len_idx = sizeof(struct sctphdr) <<
2747 IXGBE_ADVTXD_L4LEN_SHIFT;
2748 break;
2749 case IPPROTO_UDP:
2750 mss_l4len_idx = sizeof(struct udphdr) <<
2751 IXGBE_ADVTXD_L4LEN_SHIFT;
2752 break;
2753 default:
2754 if (unlikely(net_ratelimit())) {
2755 dev_warn(tx_ring->dev,
2756 "partial checksum but l4 proto=%x!\n",
2757 l4_hdr);
2759 break;
2763 /* vlan_macip_lens: MACLEN, VLAN tag */
2764 vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2765 vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2767 ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2768 type_tucmd, mss_l4len_idx);
2770 return (skb->ip_summed == CHECKSUM_PARTIAL);
2773 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2774 struct sk_buff *skb, u32 tx_flags,
2775 unsigned int first)
2777 struct ixgbevf_tx_buffer *tx_buffer_info;
2778 unsigned int len;
2779 unsigned int total = skb->len;
2780 unsigned int offset = 0, size;
2781 int count = 0;
2782 unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2783 unsigned int f;
2784 int i;
2786 i = tx_ring->next_to_use;
2788 len = min(skb_headlen(skb), total);
2789 while (len) {
2790 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2791 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2793 tx_buffer_info->length = size;
2794 tx_buffer_info->mapped_as_page = false;
2795 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2796 skb->data + offset,
2797 size, DMA_TO_DEVICE);
2798 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2799 goto dma_error;
2800 tx_buffer_info->next_to_watch = i;
2802 len -= size;
2803 total -= size;
2804 offset += size;
2805 count++;
2806 i++;
2807 if (i == tx_ring->count)
2808 i = 0;
2811 for (f = 0; f < nr_frags; f++) {
2812 const struct skb_frag_struct *frag;
2814 frag = &skb_shinfo(skb)->frags[f];
2815 len = min((unsigned int)skb_frag_size(frag), total);
2816 offset = 0;
2818 while (len) {
2819 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2820 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2822 tx_buffer_info->length = size;
2823 tx_buffer_info->dma =
2824 skb_frag_dma_map(tx_ring->dev, frag,
2825 offset, size, DMA_TO_DEVICE);
2826 if (dma_mapping_error(tx_ring->dev,
2827 tx_buffer_info->dma))
2828 goto dma_error;
2829 tx_buffer_info->mapped_as_page = true;
2830 tx_buffer_info->next_to_watch = i;
2832 len -= size;
2833 total -= size;
2834 offset += size;
2835 count++;
2836 i++;
2837 if (i == tx_ring->count)
2838 i = 0;
2840 if (total == 0)
2841 break;
2844 if (i == 0)
2845 i = tx_ring->count - 1;
2846 else
2847 i = i - 1;
2848 tx_ring->tx_buffer_info[i].skb = skb;
2849 tx_ring->tx_buffer_info[first].next_to_watch = i;
2850 tx_ring->tx_buffer_info[first].time_stamp = jiffies;
2852 return count;
2854 dma_error:
2855 dev_err(tx_ring->dev, "TX DMA map failed\n");
2857 /* clear timestamp and dma mappings for failed tx_buffer_info map */
2858 tx_buffer_info->dma = 0;
2859 tx_buffer_info->next_to_watch = 0;
2860 count--;
2862 /* clear timestamp and dma mappings for remaining portion of packet */
2863 while (count >= 0) {
2864 count--;
2865 i--;
2866 if (i < 0)
2867 i += tx_ring->count;
2868 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2869 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2872 return count;
2875 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2876 int count, u32 paylen, u8 hdr_len)
2878 union ixgbe_adv_tx_desc *tx_desc = NULL;
2879 struct ixgbevf_tx_buffer *tx_buffer_info;
2880 u32 olinfo_status = 0, cmd_type_len = 0;
2881 unsigned int i;
2883 u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2885 cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2887 cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2889 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2890 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2892 if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2893 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2895 if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2896 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2898 /* use index 1 context for tso */
2899 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2900 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2901 olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2906 * Check Context must be set if Tx switch is enabled, which it
2907 * always is for case where virtual functions are running
2909 olinfo_status |= IXGBE_ADVTXD_CC;
2911 olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2913 i = tx_ring->next_to_use;
2914 while (count--) {
2915 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2916 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2917 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2918 tx_desc->read.cmd_type_len =
2919 cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2920 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2921 i++;
2922 if (i == tx_ring->count)
2923 i = 0;
2926 tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2928 tx_ring->next_to_use = i;
2931 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2933 struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2935 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2936 /* Herbert's original patch had:
2937 * smp_mb__after_netif_stop_queue();
2938 * but since that doesn't exist yet, just open code it. */
2939 smp_mb();
2941 /* We need to check again in a case another CPU has just
2942 * made room available. */
2943 if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
2944 return -EBUSY;
2946 /* A reprieve! - use start_queue because it doesn't call schedule */
2947 netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2948 ++adapter->restart_queue;
2949 return 0;
2952 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2954 if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
2955 return 0;
2956 return __ixgbevf_maybe_stop_tx(tx_ring, size);
2959 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2961 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2962 struct ixgbevf_ring *tx_ring;
2963 unsigned int first;
2964 unsigned int tx_flags = 0;
2965 u8 hdr_len = 0;
2966 int r_idx = 0, tso;
2967 u16 count = TXD_USE_COUNT(skb_headlen(skb));
2968 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
2969 unsigned short f;
2970 #endif
2972 tx_ring = &adapter->tx_ring[r_idx];
2975 * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
2976 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
2977 * + 2 desc gap to keep tail from touching head,
2978 * + 1 desc for context descriptor,
2979 * otherwise try next time
2981 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
2982 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
2983 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
2984 #else
2985 count += skb_shinfo(skb)->nr_frags;
2986 #endif
2987 if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
2988 adapter->tx_busy++;
2989 return NETDEV_TX_BUSY;
2992 if (vlan_tx_tag_present(skb)) {
2993 tx_flags |= vlan_tx_tag_get(skb);
2994 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
2995 tx_flags |= IXGBE_TX_FLAGS_VLAN;
2998 first = tx_ring->next_to_use;
3000 if (skb->protocol == htons(ETH_P_IP))
3001 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3002 tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3003 if (tso < 0) {
3004 dev_kfree_skb_any(skb);
3005 return NETDEV_TX_OK;
3008 if (tso)
3009 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3010 else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3011 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3013 ixgbevf_tx_queue(tx_ring, tx_flags,
3014 ixgbevf_tx_map(tx_ring, skb, tx_flags, first),
3015 skb->len, hdr_len);
3017 * Force memory writes to complete before letting h/w
3018 * know there are new descriptors to fetch. (Only
3019 * applicable for weak-ordered memory model archs,
3020 * such as IA-64).
3022 wmb();
3024 writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3026 ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3028 return NETDEV_TX_OK;
3032 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3033 * @netdev: network interface device structure
3034 * @p: pointer to an address structure
3036 * Returns 0 on success, negative on failure
3038 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3040 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3041 struct ixgbe_hw *hw = &adapter->hw;
3042 struct sockaddr *addr = p;
3044 if (!is_valid_ether_addr(addr->sa_data))
3045 return -EADDRNOTAVAIL;
3047 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3048 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3050 spin_lock_bh(&adapter->mbx_lock);
3052 if (hw->mac.ops.set_rar)
3053 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3055 spin_unlock_bh(&adapter->mbx_lock);
3057 return 0;
3061 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3062 * @netdev: network interface device structure
3063 * @new_mtu: new value for maximum frame size
3065 * Returns 0 on success, negative on failure
3067 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3069 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3070 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3071 int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3073 switch (adapter->hw.api_version) {
3074 case ixgbe_mbox_api_11:
3075 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3076 break;
3077 default:
3078 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3079 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3080 break;
3083 /* MTU < 68 is an error and causes problems on some kernels */
3084 if ((new_mtu < 68) || (max_frame > max_possible_frame))
3085 return -EINVAL;
3087 hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3088 netdev->mtu, new_mtu);
3089 /* must set new MTU before calling down or up */
3090 netdev->mtu = new_mtu;
3092 if (netif_running(netdev))
3093 ixgbevf_reinit_locked(adapter);
3095 return 0;
3098 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3100 struct net_device *netdev = pci_get_drvdata(pdev);
3101 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3102 #ifdef CONFIG_PM
3103 int retval = 0;
3104 #endif
3106 netif_device_detach(netdev);
3108 if (netif_running(netdev)) {
3109 rtnl_lock();
3110 ixgbevf_down(adapter);
3111 ixgbevf_free_irq(adapter);
3112 ixgbevf_free_all_tx_resources(adapter);
3113 ixgbevf_free_all_rx_resources(adapter);
3114 rtnl_unlock();
3117 ixgbevf_clear_interrupt_scheme(adapter);
3119 #ifdef CONFIG_PM
3120 retval = pci_save_state(pdev);
3121 if (retval)
3122 return retval;
3124 #endif
3125 pci_disable_device(pdev);
3127 return 0;
3130 #ifdef CONFIG_PM
3131 static int ixgbevf_resume(struct pci_dev *pdev)
3133 struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3134 struct net_device *netdev = adapter->netdev;
3135 u32 err;
3137 pci_set_power_state(pdev, PCI_D0);
3138 pci_restore_state(pdev);
3140 * pci_restore_state clears dev->state_saved so call
3141 * pci_save_state to restore it.
3143 pci_save_state(pdev);
3145 err = pci_enable_device_mem(pdev);
3146 if (err) {
3147 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3148 return err;
3150 pci_set_master(pdev);
3152 rtnl_lock();
3153 err = ixgbevf_init_interrupt_scheme(adapter);
3154 rtnl_unlock();
3155 if (err) {
3156 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3157 return err;
3160 ixgbevf_reset(adapter);
3162 if (netif_running(netdev)) {
3163 err = ixgbevf_open(netdev);
3164 if (err)
3165 return err;
3168 netif_device_attach(netdev);
3170 return err;
3173 #endif /* CONFIG_PM */
3174 static void ixgbevf_shutdown(struct pci_dev *pdev)
3176 ixgbevf_suspend(pdev, PMSG_SUSPEND);
3179 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3180 struct rtnl_link_stats64 *stats)
3182 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3183 unsigned int start;
3184 u64 bytes, packets;
3185 const struct ixgbevf_ring *ring;
3186 int i;
3188 ixgbevf_update_stats(adapter);
3190 stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3192 for (i = 0; i < adapter->num_rx_queues; i++) {
3193 ring = &adapter->rx_ring[i];
3194 do {
3195 start = u64_stats_fetch_begin_bh(&ring->syncp);
3196 bytes = ring->total_bytes;
3197 packets = ring->total_packets;
3198 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3199 stats->rx_bytes += bytes;
3200 stats->rx_packets += packets;
3203 for (i = 0; i < adapter->num_tx_queues; i++) {
3204 ring = &adapter->tx_ring[i];
3205 do {
3206 start = u64_stats_fetch_begin_bh(&ring->syncp);
3207 bytes = ring->total_bytes;
3208 packets = ring->total_packets;
3209 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3210 stats->tx_bytes += bytes;
3211 stats->tx_packets += packets;
3214 return stats;
3217 static const struct net_device_ops ixgbevf_netdev_ops = {
3218 .ndo_open = ixgbevf_open,
3219 .ndo_stop = ixgbevf_close,
3220 .ndo_start_xmit = ixgbevf_xmit_frame,
3221 .ndo_set_rx_mode = ixgbevf_set_rx_mode,
3222 .ndo_get_stats64 = ixgbevf_get_stats,
3223 .ndo_validate_addr = eth_validate_addr,
3224 .ndo_set_mac_address = ixgbevf_set_mac,
3225 .ndo_change_mtu = ixgbevf_change_mtu,
3226 .ndo_tx_timeout = ixgbevf_tx_timeout,
3227 .ndo_vlan_rx_add_vid = ixgbevf_vlan_rx_add_vid,
3228 .ndo_vlan_rx_kill_vid = ixgbevf_vlan_rx_kill_vid,
3231 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3233 dev->netdev_ops = &ixgbevf_netdev_ops;
3234 ixgbevf_set_ethtool_ops(dev);
3235 dev->watchdog_timeo = 5 * HZ;
3239 * ixgbevf_probe - Device Initialization Routine
3240 * @pdev: PCI device information struct
3241 * @ent: entry in ixgbevf_pci_tbl
3243 * Returns 0 on success, negative on failure
3245 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3246 * The OS initialization, configuring of the adapter private structure,
3247 * and a hardware reset occur.
3249 static int __devinit ixgbevf_probe(struct pci_dev *pdev,
3250 const struct pci_device_id *ent)
3252 struct net_device *netdev;
3253 struct ixgbevf_adapter *adapter = NULL;
3254 struct ixgbe_hw *hw = NULL;
3255 const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3256 static int cards_found;
3257 int err, pci_using_dac;
3259 err = pci_enable_device(pdev);
3260 if (err)
3261 return err;
3263 if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3264 !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3265 pci_using_dac = 1;
3266 } else {
3267 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3268 if (err) {
3269 err = dma_set_coherent_mask(&pdev->dev,
3270 DMA_BIT_MASK(32));
3271 if (err) {
3272 dev_err(&pdev->dev, "No usable DMA "
3273 "configuration, aborting\n");
3274 goto err_dma;
3277 pci_using_dac = 0;
3280 err = pci_request_regions(pdev, ixgbevf_driver_name);
3281 if (err) {
3282 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3283 goto err_pci_reg;
3286 pci_set_master(pdev);
3288 netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3289 MAX_TX_QUEUES);
3290 if (!netdev) {
3291 err = -ENOMEM;
3292 goto err_alloc_etherdev;
3295 SET_NETDEV_DEV(netdev, &pdev->dev);
3297 pci_set_drvdata(pdev, netdev);
3298 adapter = netdev_priv(netdev);
3300 adapter->netdev = netdev;
3301 adapter->pdev = pdev;
3302 hw = &adapter->hw;
3303 hw->back = adapter;
3304 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3307 * call save state here in standalone driver because it relies on
3308 * adapter struct to exist, and needs to call netdev_priv
3310 pci_save_state(pdev);
3312 hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3313 pci_resource_len(pdev, 0));
3314 if (!hw->hw_addr) {
3315 err = -EIO;
3316 goto err_ioremap;
3319 ixgbevf_assign_netdev_ops(netdev);
3321 adapter->bd_number = cards_found;
3323 /* Setup hw api */
3324 memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3325 hw->mac.type = ii->mac;
3327 memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3328 sizeof(struct ixgbe_mbx_operations));
3330 /* setup the private structure */
3331 err = ixgbevf_sw_init(adapter);
3332 if (err)
3333 goto err_sw_init;
3335 /* The HW MAC address was set and/or determined in sw_init */
3336 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3338 if (!is_valid_ether_addr(netdev->dev_addr)) {
3339 pr_err("invalid MAC address\n");
3340 err = -EIO;
3341 goto err_sw_init;
3344 netdev->hw_features = NETIF_F_SG |
3345 NETIF_F_IP_CSUM |
3346 NETIF_F_IPV6_CSUM |
3347 NETIF_F_TSO |
3348 NETIF_F_TSO6 |
3349 NETIF_F_RXCSUM;
3351 netdev->features = netdev->hw_features |
3352 NETIF_F_HW_VLAN_TX |
3353 NETIF_F_HW_VLAN_RX |
3354 NETIF_F_HW_VLAN_FILTER;
3356 netdev->vlan_features |= NETIF_F_TSO;
3357 netdev->vlan_features |= NETIF_F_TSO6;
3358 netdev->vlan_features |= NETIF_F_IP_CSUM;
3359 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3360 netdev->vlan_features |= NETIF_F_SG;
3362 if (pci_using_dac)
3363 netdev->features |= NETIF_F_HIGHDMA;
3365 netdev->priv_flags |= IFF_UNICAST_FLT;
3367 init_timer(&adapter->watchdog_timer);
3368 adapter->watchdog_timer.function = ixgbevf_watchdog;
3369 adapter->watchdog_timer.data = (unsigned long)adapter;
3371 INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3372 INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3374 err = ixgbevf_init_interrupt_scheme(adapter);
3375 if (err)
3376 goto err_sw_init;
3378 /* pick up the PCI bus settings for reporting later */
3379 if (hw->mac.ops.get_bus_info)
3380 hw->mac.ops.get_bus_info(hw);
3382 strcpy(netdev->name, "eth%d");
3384 err = register_netdev(netdev);
3385 if (err)
3386 goto err_register;
3388 netif_carrier_off(netdev);
3390 ixgbevf_init_last_counter_stats(adapter);
3392 /* print the MAC address */
3393 hw_dbg(hw, "%pM\n", netdev->dev_addr);
3395 hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3397 hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3398 cards_found++;
3399 return 0;
3401 err_register:
3402 ixgbevf_clear_interrupt_scheme(adapter);
3403 err_sw_init:
3404 ixgbevf_reset_interrupt_capability(adapter);
3405 iounmap(hw->hw_addr);
3406 err_ioremap:
3407 free_netdev(netdev);
3408 err_alloc_etherdev:
3409 pci_release_regions(pdev);
3410 err_pci_reg:
3411 err_dma:
3412 pci_disable_device(pdev);
3413 return err;
3417 * ixgbevf_remove - Device Removal Routine
3418 * @pdev: PCI device information struct
3420 * ixgbevf_remove is called by the PCI subsystem to alert the driver
3421 * that it should release a PCI device. The could be caused by a
3422 * Hot-Plug event, or because the driver is going to be removed from
3423 * memory.
3425 static void __devexit ixgbevf_remove(struct pci_dev *pdev)
3427 struct net_device *netdev = pci_get_drvdata(pdev);
3428 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3430 set_bit(__IXGBEVF_DOWN, &adapter->state);
3432 del_timer_sync(&adapter->watchdog_timer);
3434 cancel_work_sync(&adapter->reset_task);
3435 cancel_work_sync(&adapter->watchdog_task);
3437 if (netdev->reg_state == NETREG_REGISTERED)
3438 unregister_netdev(netdev);
3440 ixgbevf_clear_interrupt_scheme(adapter);
3441 ixgbevf_reset_interrupt_capability(adapter);
3443 iounmap(adapter->hw.hw_addr);
3444 pci_release_regions(pdev);
3446 hw_dbg(&adapter->hw, "Remove complete\n");
3448 kfree(adapter->tx_ring);
3449 kfree(adapter->rx_ring);
3451 free_netdev(netdev);
3453 pci_disable_device(pdev);
3457 * ixgbevf_io_error_detected - called when PCI error is detected
3458 * @pdev: Pointer to PCI device
3459 * @state: The current pci connection state
3461 * This function is called after a PCI bus error affecting
3462 * this device has been detected.
3464 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3465 pci_channel_state_t state)
3467 struct net_device *netdev = pci_get_drvdata(pdev);
3468 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3470 netif_device_detach(netdev);
3472 if (state == pci_channel_io_perm_failure)
3473 return PCI_ERS_RESULT_DISCONNECT;
3475 if (netif_running(netdev))
3476 ixgbevf_down(adapter);
3478 pci_disable_device(pdev);
3480 /* Request a slot slot reset. */
3481 return PCI_ERS_RESULT_NEED_RESET;
3485 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3486 * @pdev: Pointer to PCI device
3488 * Restart the card from scratch, as if from a cold-boot. Implementation
3489 * resembles the first-half of the ixgbevf_resume routine.
3491 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3493 struct net_device *netdev = pci_get_drvdata(pdev);
3494 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3496 if (pci_enable_device_mem(pdev)) {
3497 dev_err(&pdev->dev,
3498 "Cannot re-enable PCI device after reset.\n");
3499 return PCI_ERS_RESULT_DISCONNECT;
3502 pci_set_master(pdev);
3504 ixgbevf_reset(adapter);
3506 return PCI_ERS_RESULT_RECOVERED;
3510 * ixgbevf_io_resume - called when traffic can start flowing again.
3511 * @pdev: Pointer to PCI device
3513 * This callback is called when the error recovery driver tells us that
3514 * its OK to resume normal operation. Implementation resembles the
3515 * second-half of the ixgbevf_resume routine.
3517 static void ixgbevf_io_resume(struct pci_dev *pdev)
3519 struct net_device *netdev = pci_get_drvdata(pdev);
3520 struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3522 if (netif_running(netdev))
3523 ixgbevf_up(adapter);
3525 netif_device_attach(netdev);
3528 /* PCI Error Recovery (ERS) */
3529 static const struct pci_error_handlers ixgbevf_err_handler = {
3530 .error_detected = ixgbevf_io_error_detected,
3531 .slot_reset = ixgbevf_io_slot_reset,
3532 .resume = ixgbevf_io_resume,
3535 static struct pci_driver ixgbevf_driver = {
3536 .name = ixgbevf_driver_name,
3537 .id_table = ixgbevf_pci_tbl,
3538 .probe = ixgbevf_probe,
3539 .remove = __devexit_p(ixgbevf_remove),
3540 #ifdef CONFIG_PM
3541 /* Power Management Hooks */
3542 .suspend = ixgbevf_suspend,
3543 .resume = ixgbevf_resume,
3544 #endif
3545 .shutdown = ixgbevf_shutdown,
3546 .err_handler = &ixgbevf_err_handler
3550 * ixgbevf_init_module - Driver Registration Routine
3552 * ixgbevf_init_module is the first routine called when the driver is
3553 * loaded. All it does is register with the PCI subsystem.
3555 static int __init ixgbevf_init_module(void)
3557 int ret;
3558 pr_info("%s - version %s\n", ixgbevf_driver_string,
3559 ixgbevf_driver_version);
3561 pr_info("%s\n", ixgbevf_copyright);
3563 ret = pci_register_driver(&ixgbevf_driver);
3564 return ret;
3567 module_init(ixgbevf_init_module);
3570 * ixgbevf_exit_module - Driver Exit Cleanup Routine
3572 * ixgbevf_exit_module is called just before the driver is removed
3573 * from memory.
3575 static void __exit ixgbevf_exit_module(void)
3577 pci_unregister_driver(&ixgbevf_driver);
3580 #ifdef DEBUG
3582 * ixgbevf_get_hw_dev_name - return device name string
3583 * used by hardware layer to print debugging information
3585 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3587 struct ixgbevf_adapter *adapter = hw->back;
3588 return adapter->netdev->name;
3591 #endif
3592 module_exit(ixgbevf_exit_module);
3594 /* ixgbevf_main.c */