1 /*******************************************************************************
3 Intel 82599 Virtual Function driver
4 Copyright(c) 1999 - 2015 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *******************************************************************************/
27 /******************************************************************************
28 Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
29 ******************************************************************************/
31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 #include <linux/types.h>
34 #include <linux/bitops.h>
35 #include <linux/module.h>
36 #include <linux/pci.h>
37 #include <linux/netdevice.h>
38 #include <linux/vmalloc.h>
39 #include <linux/string.h>
42 #include <linux/tcp.h>
43 #include <linux/sctp.h>
44 #include <linux/ipv6.h>
45 #include <linux/slab.h>
46 #include <net/checksum.h>
47 #include <net/ip6_checksum.h>
48 #include <linux/ethtool.h>
50 #include <linux/if_vlan.h>
51 #include <linux/prefetch.h>
55 const char ixgbevf_driver_name
[] = "ixgbevf";
56 static const char ixgbevf_driver_string
[] =
57 "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
59 #define DRV_VERSION "3.2.2-k"
60 const char ixgbevf_driver_version
[] = DRV_VERSION
;
61 static char ixgbevf_copyright
[] =
62 "Copyright (c) 2009 - 2015 Intel Corporation.";
64 static const struct ixgbevf_info
*ixgbevf_info_tbl
[] = {
65 [board_82599_vf
] = &ixgbevf_82599_vf_info
,
66 [board_82599_vf_hv
] = &ixgbevf_82599_vf_hv_info
,
67 [board_X540_vf
] = &ixgbevf_X540_vf_info
,
68 [board_X540_vf_hv
] = &ixgbevf_X540_vf_hv_info
,
69 [board_X550_vf
] = &ixgbevf_X550_vf_info
,
70 [board_X550_vf_hv
] = &ixgbevf_X550_vf_hv_info
,
71 [board_X550EM_x_vf
] = &ixgbevf_X550EM_x_vf_info
,
72 [board_X550EM_x_vf_hv
] = &ixgbevf_X550EM_x_vf_hv_info
,
73 [board_x550em_a_vf
] = &ixgbevf_x550em_a_vf_info
,
76 /* ixgbevf_pci_tbl - PCI Device ID Table
78 * Wildcard entries (PCI_ANY_ID) should come last
79 * Last entry must be all 0s
81 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
82 * Class, Class Mask, private data (not used) }
84 static const struct pci_device_id ixgbevf_pci_tbl
[] = {
85 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_VF
), board_82599_vf
},
86 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_VF_HV
), board_82599_vf_hv
},
87 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540_VF
), board_X540_vf
},
88 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X540_VF_HV
), board_X540_vf_hv
},
89 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550_VF
), board_X550_vf
},
90 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550_VF_HV
), board_X550_vf_hv
},
91 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_VF
), board_X550EM_x_vf
},
92 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_X_VF_HV
), board_X550EM_x_vf_hv
},
93 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_X550EM_A_VF
), board_x550em_a_vf
},
94 /* required last entry */
97 MODULE_DEVICE_TABLE(pci
, ixgbevf_pci_tbl
);
99 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
100 MODULE_DESCRIPTION("Intel(R) 10 Gigabit Virtual Function Network Driver");
101 MODULE_LICENSE("GPL");
102 MODULE_VERSION(DRV_VERSION
);
104 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
105 static int debug
= -1;
106 module_param(debug
, int, 0);
107 MODULE_PARM_DESC(debug
, "Debug level (0=none,...,16=all)");
109 static struct workqueue_struct
*ixgbevf_wq
;
111 static void ixgbevf_service_event_schedule(struct ixgbevf_adapter
*adapter
)
113 if (!test_bit(__IXGBEVF_DOWN
, &adapter
->state
) &&
114 !test_bit(__IXGBEVF_REMOVING
, &adapter
->state
) &&
115 !test_and_set_bit(__IXGBEVF_SERVICE_SCHED
, &adapter
->state
))
116 queue_work(ixgbevf_wq
, &adapter
->service_task
);
119 static void ixgbevf_service_event_complete(struct ixgbevf_adapter
*adapter
)
121 BUG_ON(!test_bit(__IXGBEVF_SERVICE_SCHED
, &adapter
->state
));
123 /* flush memory to make sure state is correct before next watchdog */
124 smp_mb__before_atomic();
125 clear_bit(__IXGBEVF_SERVICE_SCHED
, &adapter
->state
);
129 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter
*adapter
);
130 static void ixgbevf_set_itr(struct ixgbevf_q_vector
*q_vector
);
131 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter
*adapter
);
133 static void ixgbevf_remove_adapter(struct ixgbe_hw
*hw
)
135 struct ixgbevf_adapter
*adapter
= hw
->back
;
140 dev_err(&adapter
->pdev
->dev
, "Adapter removed\n");
141 if (test_bit(__IXGBEVF_SERVICE_INITED
, &adapter
->state
))
142 ixgbevf_service_event_schedule(adapter
);
145 static void ixgbevf_check_remove(struct ixgbe_hw
*hw
, u32 reg
)
149 /* The following check not only optimizes a bit by not
150 * performing a read on the status register when the
151 * register just read was a status register read that
152 * returned IXGBE_FAILED_READ_REG. It also blocks any
153 * potential recursion.
155 if (reg
== IXGBE_VFSTATUS
) {
156 ixgbevf_remove_adapter(hw
);
159 value
= ixgbevf_read_reg(hw
, IXGBE_VFSTATUS
);
160 if (value
== IXGBE_FAILED_READ_REG
)
161 ixgbevf_remove_adapter(hw
);
164 u32
ixgbevf_read_reg(struct ixgbe_hw
*hw
, u32 reg
)
166 u8 __iomem
*reg_addr
= ACCESS_ONCE(hw
->hw_addr
);
169 if (IXGBE_REMOVED(reg_addr
))
170 return IXGBE_FAILED_READ_REG
;
171 value
= readl(reg_addr
+ reg
);
172 if (unlikely(value
== IXGBE_FAILED_READ_REG
))
173 ixgbevf_check_remove(hw
, reg
);
178 * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
179 * @adapter: pointer to adapter struct
180 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
181 * @queue: queue to map the corresponding interrupt to
182 * @msix_vector: the vector to map to the corresponding queue
184 static void ixgbevf_set_ivar(struct ixgbevf_adapter
*adapter
, s8 direction
,
185 u8 queue
, u8 msix_vector
)
188 struct ixgbe_hw
*hw
= &adapter
->hw
;
190 if (direction
== -1) {
192 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
193 ivar
= IXGBE_READ_REG(hw
, IXGBE_VTIVAR_MISC
);
196 IXGBE_WRITE_REG(hw
, IXGBE_VTIVAR_MISC
, ivar
);
198 /* Tx or Rx causes */
199 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
200 index
= ((16 * (queue
& 1)) + (8 * direction
));
201 ivar
= IXGBE_READ_REG(hw
, IXGBE_VTIVAR(queue
>> 1));
202 ivar
&= ~(0xFF << index
);
203 ivar
|= (msix_vector
<< index
);
204 IXGBE_WRITE_REG(hw
, IXGBE_VTIVAR(queue
>> 1), ivar
);
208 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring
*tx_ring
,
209 struct ixgbevf_tx_buffer
*tx_buffer
)
211 if (tx_buffer
->skb
) {
212 dev_kfree_skb_any(tx_buffer
->skb
);
213 if (dma_unmap_len(tx_buffer
, len
))
214 dma_unmap_single(tx_ring
->dev
,
215 dma_unmap_addr(tx_buffer
, dma
),
216 dma_unmap_len(tx_buffer
, len
),
218 } else if (dma_unmap_len(tx_buffer
, len
)) {
219 dma_unmap_page(tx_ring
->dev
,
220 dma_unmap_addr(tx_buffer
, dma
),
221 dma_unmap_len(tx_buffer
, len
),
224 tx_buffer
->next_to_watch
= NULL
;
225 tx_buffer
->skb
= NULL
;
226 dma_unmap_len_set(tx_buffer
, len
, 0);
227 /* tx_buffer must be completely set up in the transmit path */
230 static u64
ixgbevf_get_tx_completed(struct ixgbevf_ring
*ring
)
232 return ring
->stats
.packets
;
235 static u32
ixgbevf_get_tx_pending(struct ixgbevf_ring
*ring
)
237 struct ixgbevf_adapter
*adapter
= netdev_priv(ring
->netdev
);
238 struct ixgbe_hw
*hw
= &adapter
->hw
;
240 u32 head
= IXGBE_READ_REG(hw
, IXGBE_VFTDH(ring
->reg_idx
));
241 u32 tail
= IXGBE_READ_REG(hw
, IXGBE_VFTDT(ring
->reg_idx
));
244 return (head
< tail
) ?
245 tail
- head
: (tail
+ ring
->count
- head
);
250 static inline bool ixgbevf_check_tx_hang(struct ixgbevf_ring
*tx_ring
)
252 u32 tx_done
= ixgbevf_get_tx_completed(tx_ring
);
253 u32 tx_done_old
= tx_ring
->tx_stats
.tx_done_old
;
254 u32 tx_pending
= ixgbevf_get_tx_pending(tx_ring
);
256 clear_check_for_tx_hang(tx_ring
);
258 /* Check for a hung queue, but be thorough. This verifies
259 * that a transmit has been completed since the previous
260 * check AND there is at least one packet pending. The
261 * ARMED bit is set to indicate a potential hang.
263 if ((tx_done_old
== tx_done
) && tx_pending
) {
264 /* make sure it is true for two checks in a row */
265 return test_and_set_bit(__IXGBEVF_HANG_CHECK_ARMED
,
268 /* reset the countdown */
269 clear_bit(__IXGBEVF_HANG_CHECK_ARMED
, &tx_ring
->state
);
271 /* update completed stats and continue */
272 tx_ring
->tx_stats
.tx_done_old
= tx_done
;
277 static void ixgbevf_tx_timeout_reset(struct ixgbevf_adapter
*adapter
)
279 /* Do the reset outside of interrupt context */
280 if (!test_bit(__IXGBEVF_DOWN
, &adapter
->state
)) {
281 set_bit(__IXGBEVF_RESET_REQUESTED
, &adapter
->state
);
282 ixgbevf_service_event_schedule(adapter
);
287 * ixgbevf_tx_timeout - Respond to a Tx Hang
288 * @netdev: network interface device structure
290 static void ixgbevf_tx_timeout(struct net_device
*netdev
)
292 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
294 ixgbevf_tx_timeout_reset(adapter
);
298 * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
299 * @q_vector: board private structure
300 * @tx_ring: tx ring to clean
301 * @napi_budget: Used to determine if we are in netpoll
303 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector
*q_vector
,
304 struct ixgbevf_ring
*tx_ring
, int napi_budget
)
306 struct ixgbevf_adapter
*adapter
= q_vector
->adapter
;
307 struct ixgbevf_tx_buffer
*tx_buffer
;
308 union ixgbe_adv_tx_desc
*tx_desc
;
309 unsigned int total_bytes
= 0, total_packets
= 0;
310 unsigned int budget
= tx_ring
->count
/ 2;
311 unsigned int i
= tx_ring
->next_to_clean
;
313 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
))
316 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
317 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, i
);
321 union ixgbe_adv_tx_desc
*eop_desc
= tx_buffer
->next_to_watch
;
323 /* if next_to_watch is not set then there is no work pending */
327 /* prevent any other reads prior to eop_desc */
328 read_barrier_depends();
330 /* if DD is not set pending work has not been completed */
331 if (!(eop_desc
->wb
.status
& cpu_to_le32(IXGBE_TXD_STAT_DD
)))
334 /* clear next_to_watch to prevent false hangs */
335 tx_buffer
->next_to_watch
= NULL
;
337 /* update the statistics for this packet */
338 total_bytes
+= tx_buffer
->bytecount
;
339 total_packets
+= tx_buffer
->gso_segs
;
342 napi_consume_skb(tx_buffer
->skb
, napi_budget
);
344 /* unmap skb header data */
345 dma_unmap_single(tx_ring
->dev
,
346 dma_unmap_addr(tx_buffer
, dma
),
347 dma_unmap_len(tx_buffer
, len
),
350 /* clear tx_buffer data */
351 tx_buffer
->skb
= NULL
;
352 dma_unmap_len_set(tx_buffer
, len
, 0);
354 /* unmap remaining buffers */
355 while (tx_desc
!= eop_desc
) {
361 tx_buffer
= tx_ring
->tx_buffer_info
;
362 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, 0);
365 /* unmap any remaining paged data */
366 if (dma_unmap_len(tx_buffer
, len
)) {
367 dma_unmap_page(tx_ring
->dev
,
368 dma_unmap_addr(tx_buffer
, dma
),
369 dma_unmap_len(tx_buffer
, len
),
371 dma_unmap_len_set(tx_buffer
, len
, 0);
375 /* move us one more past the eop_desc for start of next pkt */
381 tx_buffer
= tx_ring
->tx_buffer_info
;
382 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, 0);
385 /* issue prefetch for next Tx descriptor */
388 /* update budget accounting */
390 } while (likely(budget
));
393 tx_ring
->next_to_clean
= i
;
394 u64_stats_update_begin(&tx_ring
->syncp
);
395 tx_ring
->stats
.bytes
+= total_bytes
;
396 tx_ring
->stats
.packets
+= total_packets
;
397 u64_stats_update_end(&tx_ring
->syncp
);
398 q_vector
->tx
.total_bytes
+= total_bytes
;
399 q_vector
->tx
.total_packets
+= total_packets
;
401 if (check_for_tx_hang(tx_ring
) && ixgbevf_check_tx_hang(tx_ring
)) {
402 struct ixgbe_hw
*hw
= &adapter
->hw
;
403 union ixgbe_adv_tx_desc
*eop_desc
;
405 eop_desc
= tx_ring
->tx_buffer_info
[i
].next_to_watch
;
407 pr_err("Detected Tx Unit Hang\n"
409 " TDH, TDT <%x>, <%x>\n"
410 " next_to_use <%x>\n"
411 " next_to_clean <%x>\n"
412 "tx_buffer_info[next_to_clean]\n"
413 " next_to_watch <%p>\n"
414 " eop_desc->wb.status <%x>\n"
415 " time_stamp <%lx>\n"
417 tx_ring
->queue_index
,
418 IXGBE_READ_REG(hw
, IXGBE_VFTDH(tx_ring
->reg_idx
)),
419 IXGBE_READ_REG(hw
, IXGBE_VFTDT(tx_ring
->reg_idx
)),
420 tx_ring
->next_to_use
, i
,
421 eop_desc
, (eop_desc
? eop_desc
->wb
.status
: 0),
422 tx_ring
->tx_buffer_info
[i
].time_stamp
, jiffies
);
424 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
426 /* schedule immediate reset if we believe we hung */
427 ixgbevf_tx_timeout_reset(adapter
);
432 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
433 if (unlikely(total_packets
&& netif_carrier_ok(tx_ring
->netdev
) &&
434 (ixgbevf_desc_unused(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
435 /* Make sure that anybody stopping the queue after this
436 * sees the new next_to_clean.
440 if (__netif_subqueue_stopped(tx_ring
->netdev
,
441 tx_ring
->queue_index
) &&
442 !test_bit(__IXGBEVF_DOWN
, &adapter
->state
)) {
443 netif_wake_subqueue(tx_ring
->netdev
,
444 tx_ring
->queue_index
);
445 ++tx_ring
->tx_stats
.restart_queue
;
453 * ixgbevf_rx_skb - Helper function to determine proper Rx method
454 * @q_vector: structure containing interrupt and ring information
455 * @skb: packet to send up
457 static void ixgbevf_rx_skb(struct ixgbevf_q_vector
*q_vector
,
460 #ifdef CONFIG_NET_RX_BUSY_POLL
461 skb_mark_napi_id(skb
, &q_vector
->napi
);
463 if (ixgbevf_qv_busy_polling(q_vector
)) {
464 netif_receive_skb(skb
);
465 /* exit early if we busy polled */
468 #endif /* CONFIG_NET_RX_BUSY_POLL */
470 napi_gro_receive(&q_vector
->napi
, skb
);
473 #define IXGBE_RSS_L4_TYPES_MASK \
474 ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
475 (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
476 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
477 (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
479 static inline void ixgbevf_rx_hash(struct ixgbevf_ring
*ring
,
480 union ixgbe_adv_rx_desc
*rx_desc
,
485 if (!(ring
->netdev
->features
& NETIF_F_RXHASH
))
488 rss_type
= le16_to_cpu(rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
) &
489 IXGBE_RXDADV_RSSTYPE_MASK
;
494 skb_set_hash(skb
, le32_to_cpu(rx_desc
->wb
.lower
.hi_dword
.rss
),
495 (IXGBE_RSS_L4_TYPES_MASK
& (1ul << rss_type
)) ?
496 PKT_HASH_TYPE_L4
: PKT_HASH_TYPE_L3
);
500 * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
501 * @ring: structure containig ring specific data
502 * @rx_desc: current Rx descriptor being processed
503 * @skb: skb currently being received and modified
505 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring
*ring
,
506 union ixgbe_adv_rx_desc
*rx_desc
,
509 skb_checksum_none_assert(skb
);
511 /* Rx csum disabled */
512 if (!(ring
->netdev
->features
& NETIF_F_RXCSUM
))
515 /* if IP and error */
516 if (ixgbevf_test_staterr(rx_desc
, IXGBE_RXD_STAT_IPCS
) &&
517 ixgbevf_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_IPE
)) {
518 ring
->rx_stats
.csum_err
++;
522 if (!ixgbevf_test_staterr(rx_desc
, IXGBE_RXD_STAT_L4CS
))
525 if (ixgbevf_test_staterr(rx_desc
, IXGBE_RXDADV_ERR_TCPE
)) {
526 ring
->rx_stats
.csum_err
++;
530 /* It must be a TCP or UDP packet with a valid checksum */
531 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
535 * ixgbevf_process_skb_fields - Populate skb header fields from Rx descriptor
536 * @rx_ring: rx descriptor ring packet is being transacted on
537 * @rx_desc: pointer to the EOP Rx descriptor
538 * @skb: pointer to current skb being populated
540 * This function checks the ring, descriptor, and packet information in
541 * order to populate the checksum, VLAN, protocol, and other fields within
544 static void ixgbevf_process_skb_fields(struct ixgbevf_ring
*rx_ring
,
545 union ixgbe_adv_rx_desc
*rx_desc
,
548 ixgbevf_rx_hash(rx_ring
, rx_desc
, skb
);
549 ixgbevf_rx_checksum(rx_ring
, rx_desc
, skb
);
551 if (ixgbevf_test_staterr(rx_desc
, IXGBE_RXD_STAT_VP
)) {
552 u16 vid
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
553 unsigned long *active_vlans
= netdev_priv(rx_ring
->netdev
);
555 if (test_bit(vid
& VLAN_VID_MASK
, active_vlans
))
556 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), vid
);
559 skb
->protocol
= eth_type_trans(skb
, rx_ring
->netdev
);
563 * ixgbevf_is_non_eop - process handling of non-EOP buffers
564 * @rx_ring: Rx ring being processed
565 * @rx_desc: Rx descriptor for current buffer
566 * @skb: current socket buffer containing buffer in progress
568 * This function updates next to clean. If the buffer is an EOP buffer
569 * this function exits returning false, otherwise it will place the
570 * sk_buff in the next buffer to be chained and return true indicating
571 * that this is in fact a non-EOP buffer.
573 static bool ixgbevf_is_non_eop(struct ixgbevf_ring
*rx_ring
,
574 union ixgbe_adv_rx_desc
*rx_desc
)
576 u32 ntc
= rx_ring
->next_to_clean
+ 1;
578 /* fetch, update, and store next to clean */
579 ntc
= (ntc
< rx_ring
->count
) ? ntc
: 0;
580 rx_ring
->next_to_clean
= ntc
;
582 prefetch(IXGBEVF_RX_DESC(rx_ring
, ntc
));
584 if (likely(ixgbevf_test_staterr(rx_desc
, IXGBE_RXD_STAT_EOP
)))
590 static bool ixgbevf_alloc_mapped_page(struct ixgbevf_ring
*rx_ring
,
591 struct ixgbevf_rx_buffer
*bi
)
593 struct page
*page
= bi
->page
;
594 dma_addr_t dma
= bi
->dma
;
596 /* since we are recycling buffers we should seldom need to alloc */
600 /* alloc new page for storage */
601 page
= dev_alloc_page();
602 if (unlikely(!page
)) {
603 rx_ring
->rx_stats
.alloc_rx_page_failed
++;
607 /* map page for use */
608 dma
= dma_map_page(rx_ring
->dev
, page
, 0,
609 PAGE_SIZE
, DMA_FROM_DEVICE
);
611 /* if mapping failed free memory back to system since
612 * there isn't much point in holding memory we can't use
614 if (dma_mapping_error(rx_ring
->dev
, dma
)) {
617 rx_ring
->rx_stats
.alloc_rx_buff_failed
++;
629 * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
630 * @rx_ring: rx descriptor ring (for a specific queue) to setup buffers on
631 * @cleaned_count: number of buffers to replace
633 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_ring
*rx_ring
,
636 union ixgbe_adv_rx_desc
*rx_desc
;
637 struct ixgbevf_rx_buffer
*bi
;
638 unsigned int i
= rx_ring
->next_to_use
;
640 /* nothing to do or no valid netdev defined */
641 if (!cleaned_count
|| !rx_ring
->netdev
)
644 rx_desc
= IXGBEVF_RX_DESC(rx_ring
, i
);
645 bi
= &rx_ring
->rx_buffer_info
[i
];
649 if (!ixgbevf_alloc_mapped_page(rx_ring
, bi
))
652 /* Refresh the desc even if pkt_addr didn't change
653 * because each write-back erases this info.
655 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
+ bi
->page_offset
);
661 rx_desc
= IXGBEVF_RX_DESC(rx_ring
, 0);
662 bi
= rx_ring
->rx_buffer_info
;
666 /* clear the hdr_addr for the next_to_use descriptor */
667 rx_desc
->read
.hdr_addr
= 0;
670 } while (cleaned_count
);
674 if (rx_ring
->next_to_use
!= i
) {
675 /* record the next descriptor to use */
676 rx_ring
->next_to_use
= i
;
678 /* update next to alloc since we have filled the ring */
679 rx_ring
->next_to_alloc
= i
;
681 /* Force memory writes to complete before letting h/w
682 * know there are new descriptors to fetch. (Only
683 * applicable for weak-ordered memory model archs,
687 ixgbevf_write_tail(rx_ring
, i
);
692 * ixgbevf_cleanup_headers - Correct corrupted or empty headers
693 * @rx_ring: rx descriptor ring packet is being transacted on
694 * @rx_desc: pointer to the EOP Rx descriptor
695 * @skb: pointer to current skb being fixed
697 * Check for corrupted packet headers caused by senders on the local L2
698 * embedded NIC switch not setting up their Tx Descriptors right. These
699 * should be very rare.
701 * Also address the case where we are pulling data in on pages only
702 * and as such no data is present in the skb header.
704 * In addition if skb is not at least 60 bytes we need to pad it so that
705 * it is large enough to qualify as a valid Ethernet frame.
707 * Returns true if an error was encountered and skb was freed.
709 static bool ixgbevf_cleanup_headers(struct ixgbevf_ring
*rx_ring
,
710 union ixgbe_adv_rx_desc
*rx_desc
,
713 /* verify that the packet does not have any known errors */
714 if (unlikely(ixgbevf_test_staterr(rx_desc
,
715 IXGBE_RXDADV_ERR_FRAME_ERR_MASK
))) {
716 struct net_device
*netdev
= rx_ring
->netdev
;
718 if (!(netdev
->features
& NETIF_F_RXALL
)) {
719 dev_kfree_skb_any(skb
);
724 /* if eth_skb_pad returns an error the skb was freed */
725 if (eth_skb_pad(skb
))
732 * ixgbevf_reuse_rx_page - page flip buffer and store it back on the ring
733 * @rx_ring: rx descriptor ring to store buffers on
734 * @old_buff: donor buffer to have page reused
736 * Synchronizes page for reuse by the adapter
738 static void ixgbevf_reuse_rx_page(struct ixgbevf_ring
*rx_ring
,
739 struct ixgbevf_rx_buffer
*old_buff
)
741 struct ixgbevf_rx_buffer
*new_buff
;
742 u16 nta
= rx_ring
->next_to_alloc
;
744 new_buff
= &rx_ring
->rx_buffer_info
[nta
];
746 /* update, and store next to alloc */
748 rx_ring
->next_to_alloc
= (nta
< rx_ring
->count
) ? nta
: 0;
750 /* transfer page from old buffer to new buffer */
751 new_buff
->page
= old_buff
->page
;
752 new_buff
->dma
= old_buff
->dma
;
753 new_buff
->page_offset
= old_buff
->page_offset
;
755 /* sync the buffer for use by the device */
756 dma_sync_single_range_for_device(rx_ring
->dev
, new_buff
->dma
,
757 new_buff
->page_offset
,
762 static inline bool ixgbevf_page_is_reserved(struct page
*page
)
764 return (page_to_nid(page
) != numa_mem_id()) || page_is_pfmemalloc(page
);
768 * ixgbevf_add_rx_frag - Add contents of Rx buffer to sk_buff
769 * @rx_ring: rx descriptor ring to transact packets on
770 * @rx_buffer: buffer containing page to add
771 * @rx_desc: descriptor containing length of buffer written by hardware
772 * @skb: sk_buff to place the data into
774 * This function will add the data contained in rx_buffer->page to the skb.
775 * This is done either through a direct copy if the data in the buffer is
776 * less than the skb header size, otherwise it will just attach the page as
779 * The function will then update the page offset if necessary and return
780 * true if the buffer can be reused by the adapter.
782 static bool ixgbevf_add_rx_frag(struct ixgbevf_ring
*rx_ring
,
783 struct ixgbevf_rx_buffer
*rx_buffer
,
784 union ixgbe_adv_rx_desc
*rx_desc
,
787 struct page
*page
= rx_buffer
->page
;
788 unsigned char *va
= page_address(page
) + rx_buffer
->page_offset
;
789 unsigned int size
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
790 #if (PAGE_SIZE < 8192)
791 unsigned int truesize
= IXGBEVF_RX_BUFSZ
;
793 unsigned int truesize
= ALIGN(size
, L1_CACHE_BYTES
);
795 unsigned int pull_len
;
797 if (unlikely(skb_is_nonlinear(skb
)))
800 if (likely(size
<= IXGBEVF_RX_HDR_SIZE
)) {
801 memcpy(__skb_put(skb
, size
), va
, ALIGN(size
, sizeof(long)));
803 /* page is not reserved, we can reuse buffer as is */
804 if (likely(!ixgbevf_page_is_reserved(page
)))
807 /* this page cannot be reused so discard it */
812 /* we need the header to contain the greater of either ETH_HLEN or
813 * 60 bytes if the skb->len is less than 60 for skb_pad.
815 pull_len
= eth_get_headlen(va
, IXGBEVF_RX_HDR_SIZE
);
817 /* align pull length to size of long to optimize memcpy performance */
818 memcpy(__skb_put(skb
, pull_len
), va
, ALIGN(pull_len
, sizeof(long)));
820 /* update all of the pointers */
825 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
,
826 (unsigned long)va
& ~PAGE_MASK
, size
, truesize
);
828 /* avoid re-using remote pages */
829 if (unlikely(ixgbevf_page_is_reserved(page
)))
832 #if (PAGE_SIZE < 8192)
833 /* if we are only owner of page we can reuse it */
834 if (unlikely(page_count(page
) != 1))
837 /* flip page offset to other buffer */
838 rx_buffer
->page_offset
^= IXGBEVF_RX_BUFSZ
;
841 /* move offset up to the next cache line */
842 rx_buffer
->page_offset
+= truesize
;
844 if (rx_buffer
->page_offset
> (PAGE_SIZE
- IXGBEVF_RX_BUFSZ
))
848 /* Even if we own the page, we are not allowed to use atomic_set()
849 * This would break get_page_unless_zero() users.
856 static struct sk_buff
*ixgbevf_fetch_rx_buffer(struct ixgbevf_ring
*rx_ring
,
857 union ixgbe_adv_rx_desc
*rx_desc
,
860 struct ixgbevf_rx_buffer
*rx_buffer
;
863 rx_buffer
= &rx_ring
->rx_buffer_info
[rx_ring
->next_to_clean
];
864 page
= rx_buffer
->page
;
868 void *page_addr
= page_address(page
) +
869 rx_buffer
->page_offset
;
871 /* prefetch first cache line of first page */
873 #if L1_CACHE_BYTES < 128
874 prefetch(page_addr
+ L1_CACHE_BYTES
);
877 /* allocate a skb to store the frags */
878 skb
= netdev_alloc_skb_ip_align(rx_ring
->netdev
,
879 IXGBEVF_RX_HDR_SIZE
);
880 if (unlikely(!skb
)) {
881 rx_ring
->rx_stats
.alloc_rx_buff_failed
++;
885 /* we will be copying header into skb->data in
886 * pskb_may_pull so it is in our interest to prefetch
887 * it now to avoid a possible cache miss
889 prefetchw(skb
->data
);
892 /* we are reusing so sync this buffer for CPU use */
893 dma_sync_single_range_for_cpu(rx_ring
->dev
,
895 rx_buffer
->page_offset
,
899 /* pull page into skb */
900 if (ixgbevf_add_rx_frag(rx_ring
, rx_buffer
, rx_desc
, skb
)) {
901 /* hand second half of page back to the ring */
902 ixgbevf_reuse_rx_page(rx_ring
, rx_buffer
);
904 /* we are not reusing the buffer so unmap it */
905 dma_unmap_page(rx_ring
->dev
, rx_buffer
->dma
,
906 PAGE_SIZE
, DMA_FROM_DEVICE
);
909 /* clear contents of buffer_info */
911 rx_buffer
->page
= NULL
;
916 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter
*adapter
,
919 struct ixgbe_hw
*hw
= &adapter
->hw
;
921 IXGBE_WRITE_REG(hw
, IXGBE_VTEIMS
, qmask
);
924 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector
*q_vector
,
925 struct ixgbevf_ring
*rx_ring
,
928 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
929 u16 cleaned_count
= ixgbevf_desc_unused(rx_ring
);
930 struct sk_buff
*skb
= rx_ring
->skb
;
932 while (likely(total_rx_packets
< budget
)) {
933 union ixgbe_adv_rx_desc
*rx_desc
;
935 /* return some buffers to hardware, one at a time is too slow */
936 if (cleaned_count
>= IXGBEVF_RX_BUFFER_WRITE
) {
937 ixgbevf_alloc_rx_buffers(rx_ring
, cleaned_count
);
941 rx_desc
= IXGBEVF_RX_DESC(rx_ring
, rx_ring
->next_to_clean
);
943 if (!ixgbevf_test_staterr(rx_desc
, IXGBE_RXD_STAT_DD
))
946 /* This memory barrier is needed to keep us from reading
947 * any other fields out of the rx_desc until we know the
948 * RXD_STAT_DD bit is set
952 /* retrieve a buffer from the ring */
953 skb
= ixgbevf_fetch_rx_buffer(rx_ring
, rx_desc
, skb
);
955 /* exit if we failed to retrieve a buffer */
961 /* fetch next buffer in frame if non-eop */
962 if (ixgbevf_is_non_eop(rx_ring
, rx_desc
))
965 /* verify the packet layout is correct */
966 if (ixgbevf_cleanup_headers(rx_ring
, rx_desc
, skb
)) {
971 /* probably a little skewed due to removing CRC */
972 total_rx_bytes
+= skb
->len
;
974 /* Workaround hardware that can't do proper VEPA multicast
977 if ((skb
->pkt_type
== PACKET_BROADCAST
||
978 skb
->pkt_type
== PACKET_MULTICAST
) &&
979 ether_addr_equal(rx_ring
->netdev
->dev_addr
,
980 eth_hdr(skb
)->h_source
)) {
981 dev_kfree_skb_irq(skb
);
985 /* populate checksum, VLAN, and protocol */
986 ixgbevf_process_skb_fields(rx_ring
, rx_desc
, skb
);
988 ixgbevf_rx_skb(q_vector
, skb
);
990 /* reset skb pointer */
993 /* update budget accounting */
997 /* place incomplete frames back on ring for completion */
1000 u64_stats_update_begin(&rx_ring
->syncp
);
1001 rx_ring
->stats
.packets
+= total_rx_packets
;
1002 rx_ring
->stats
.bytes
+= total_rx_bytes
;
1003 u64_stats_update_end(&rx_ring
->syncp
);
1004 q_vector
->rx
.total_packets
+= total_rx_packets
;
1005 q_vector
->rx
.total_bytes
+= total_rx_bytes
;
1007 return total_rx_packets
;
1011 * ixgbevf_poll - NAPI polling calback
1012 * @napi: napi struct with our devices info in it
1013 * @budget: amount of work driver is allowed to do this pass, in packets
1015 * This function will clean more than one or more rings associated with a
1018 static int ixgbevf_poll(struct napi_struct
*napi
, int budget
)
1020 struct ixgbevf_q_vector
*q_vector
=
1021 container_of(napi
, struct ixgbevf_q_vector
, napi
);
1022 struct ixgbevf_adapter
*adapter
= q_vector
->adapter
;
1023 struct ixgbevf_ring
*ring
;
1024 int per_ring_budget
, work_done
= 0;
1025 bool clean_complete
= true;
1027 ixgbevf_for_each_ring(ring
, q_vector
->tx
) {
1028 if (!ixgbevf_clean_tx_irq(q_vector
, ring
, budget
))
1029 clean_complete
= false;
1034 #ifdef CONFIG_NET_RX_BUSY_POLL
1035 if (!ixgbevf_qv_lock_napi(q_vector
))
1039 /* attempt to distribute budget to each queue fairly, but don't allow
1040 * the budget to go below 1 because we'll exit polling
1042 if (q_vector
->rx
.count
> 1)
1043 per_ring_budget
= max(budget
/q_vector
->rx
.count
, 1);
1045 per_ring_budget
= budget
;
1047 ixgbevf_for_each_ring(ring
, q_vector
->rx
) {
1048 int cleaned
= ixgbevf_clean_rx_irq(q_vector
, ring
,
1050 work_done
+= cleaned
;
1051 if (cleaned
>= per_ring_budget
)
1052 clean_complete
= false;
1055 #ifdef CONFIG_NET_RX_BUSY_POLL
1056 ixgbevf_qv_unlock_napi(q_vector
);
1059 /* If all work not completed, return budget and keep polling */
1060 if (!clean_complete
)
1062 /* all work done, exit the polling mode */
1063 napi_complete_done(napi
, work_done
);
1064 if (adapter
->rx_itr_setting
== 1)
1065 ixgbevf_set_itr(q_vector
);
1066 if (!test_bit(__IXGBEVF_DOWN
, &adapter
->state
) &&
1067 !test_bit(__IXGBEVF_REMOVING
, &adapter
->state
))
1068 ixgbevf_irq_enable_queues(adapter
,
1069 BIT(q_vector
->v_idx
));
1075 * ixgbevf_write_eitr - write VTEITR register in hardware specific way
1076 * @q_vector: structure containing interrupt and ring information
1078 void ixgbevf_write_eitr(struct ixgbevf_q_vector
*q_vector
)
1080 struct ixgbevf_adapter
*adapter
= q_vector
->adapter
;
1081 struct ixgbe_hw
*hw
= &adapter
->hw
;
1082 int v_idx
= q_vector
->v_idx
;
1083 u32 itr_reg
= q_vector
->itr
& IXGBE_MAX_EITR
;
1085 /* set the WDIS bit to not clear the timer bits and cause an
1086 * immediate assertion of the interrupt
1088 itr_reg
|= IXGBE_EITR_CNT_WDIS
;
1090 IXGBE_WRITE_REG(hw
, IXGBE_VTEITR(v_idx
), itr_reg
);
1093 #ifdef CONFIG_NET_RX_BUSY_POLL
1094 /* must be called with local_bh_disable()d */
1095 static int ixgbevf_busy_poll_recv(struct napi_struct
*napi
)
1097 struct ixgbevf_q_vector
*q_vector
=
1098 container_of(napi
, struct ixgbevf_q_vector
, napi
);
1099 struct ixgbevf_adapter
*adapter
= q_vector
->adapter
;
1100 struct ixgbevf_ring
*ring
;
1103 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
))
1104 return LL_FLUSH_FAILED
;
1106 if (!ixgbevf_qv_lock_poll(q_vector
))
1107 return LL_FLUSH_BUSY
;
1109 ixgbevf_for_each_ring(ring
, q_vector
->rx
) {
1110 found
= ixgbevf_clean_rx_irq(q_vector
, ring
, 4);
1111 #ifdef BP_EXTENDED_STATS
1113 ring
->stats
.cleaned
+= found
;
1115 ring
->stats
.misses
++;
1121 ixgbevf_qv_unlock_poll(q_vector
);
1125 #endif /* CONFIG_NET_RX_BUSY_POLL */
1128 * ixgbevf_configure_msix - Configure MSI-X hardware
1129 * @adapter: board private structure
1131 * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
1134 static void ixgbevf_configure_msix(struct ixgbevf_adapter
*adapter
)
1136 struct ixgbevf_q_vector
*q_vector
;
1137 int q_vectors
, v_idx
;
1139 q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1140 adapter
->eims_enable_mask
= 0;
1142 /* Populate the IVAR table and set the ITR values to the
1143 * corresponding register.
1145 for (v_idx
= 0; v_idx
< q_vectors
; v_idx
++) {
1146 struct ixgbevf_ring
*ring
;
1148 q_vector
= adapter
->q_vector
[v_idx
];
1150 ixgbevf_for_each_ring(ring
, q_vector
->rx
)
1151 ixgbevf_set_ivar(adapter
, 0, ring
->reg_idx
, v_idx
);
1153 ixgbevf_for_each_ring(ring
, q_vector
->tx
)
1154 ixgbevf_set_ivar(adapter
, 1, ring
->reg_idx
, v_idx
);
1156 if (q_vector
->tx
.ring
&& !q_vector
->rx
.ring
) {
1157 /* Tx only vector */
1158 if (adapter
->tx_itr_setting
== 1)
1159 q_vector
->itr
= IXGBE_12K_ITR
;
1161 q_vector
->itr
= adapter
->tx_itr_setting
;
1163 /* Rx or Rx/Tx vector */
1164 if (adapter
->rx_itr_setting
== 1)
1165 q_vector
->itr
= IXGBE_20K_ITR
;
1167 q_vector
->itr
= adapter
->rx_itr_setting
;
1170 /* add q_vector eims value to global eims_enable_mask */
1171 adapter
->eims_enable_mask
|= BIT(v_idx
);
1173 ixgbevf_write_eitr(q_vector
);
1176 ixgbevf_set_ivar(adapter
, -1, 1, v_idx
);
1177 /* setup eims_other and add value to global eims_enable_mask */
1178 adapter
->eims_other
= BIT(v_idx
);
1179 adapter
->eims_enable_mask
|= adapter
->eims_other
;
1182 enum latency_range
{
1186 latency_invalid
= 255
1190 * ixgbevf_update_itr - update the dynamic ITR value based on statistics
1191 * @q_vector: structure containing interrupt and ring information
1192 * @ring_container: structure containing ring performance data
1194 * Stores a new ITR value based on packets and byte
1195 * counts during the last interrupt. The advantage of per interrupt
1196 * computation is faster updates and more accurate ITR for the current
1197 * traffic pattern. Constants in this function were computed
1198 * based on theoretical maximum wire speed and thresholds were set based
1199 * on testing data as well as attempting to minimize response time
1200 * while increasing bulk throughput.
1202 static void ixgbevf_update_itr(struct ixgbevf_q_vector
*q_vector
,
1203 struct ixgbevf_ring_container
*ring_container
)
1205 int bytes
= ring_container
->total_bytes
;
1206 int packets
= ring_container
->total_packets
;
1209 u8 itr_setting
= ring_container
->itr
;
1214 /* simple throttle rate management
1215 * 0-20MB/s lowest (100000 ints/s)
1216 * 20-100MB/s low (20000 ints/s)
1217 * 100-1249MB/s bulk (12000 ints/s)
1219 /* what was last interrupt timeslice? */
1220 timepassed_us
= q_vector
->itr
>> 2;
1221 bytes_perint
= bytes
/ timepassed_us
; /* bytes/usec */
1223 switch (itr_setting
) {
1224 case lowest_latency
:
1225 if (bytes_perint
> 10)
1226 itr_setting
= low_latency
;
1229 if (bytes_perint
> 20)
1230 itr_setting
= bulk_latency
;
1231 else if (bytes_perint
<= 10)
1232 itr_setting
= lowest_latency
;
1235 if (bytes_perint
<= 20)
1236 itr_setting
= low_latency
;
1240 /* clear work counters since we have the values we need */
1241 ring_container
->total_bytes
= 0;
1242 ring_container
->total_packets
= 0;
1244 /* write updated itr to ring container */
1245 ring_container
->itr
= itr_setting
;
1248 static void ixgbevf_set_itr(struct ixgbevf_q_vector
*q_vector
)
1250 u32 new_itr
= q_vector
->itr
;
1253 ixgbevf_update_itr(q_vector
, &q_vector
->tx
);
1254 ixgbevf_update_itr(q_vector
, &q_vector
->rx
);
1256 current_itr
= max(q_vector
->rx
.itr
, q_vector
->tx
.itr
);
1258 switch (current_itr
) {
1259 /* counts and packets in update_itr are dependent on these numbers */
1260 case lowest_latency
:
1261 new_itr
= IXGBE_100K_ITR
;
1264 new_itr
= IXGBE_20K_ITR
;
1267 new_itr
= IXGBE_12K_ITR
;
1273 if (new_itr
!= q_vector
->itr
) {
1274 /* do an exponential smoothing */
1275 new_itr
= (10 * new_itr
* q_vector
->itr
) /
1276 ((9 * new_itr
) + q_vector
->itr
);
1278 /* save the algorithm value here */
1279 q_vector
->itr
= new_itr
;
1281 ixgbevf_write_eitr(q_vector
);
1285 static irqreturn_t
ixgbevf_msix_other(int irq
, void *data
)
1287 struct ixgbevf_adapter
*adapter
= data
;
1288 struct ixgbe_hw
*hw
= &adapter
->hw
;
1290 hw
->mac
.get_link_status
= 1;
1292 ixgbevf_service_event_schedule(adapter
);
1294 IXGBE_WRITE_REG(hw
, IXGBE_VTEIMS
, adapter
->eims_other
);
1300 * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
1302 * @data: pointer to our q_vector struct for this interrupt vector
1304 static irqreturn_t
ixgbevf_msix_clean_rings(int irq
, void *data
)
1306 struct ixgbevf_q_vector
*q_vector
= data
;
1308 /* EIAM disabled interrupts (on this vector) for us */
1309 if (q_vector
->rx
.ring
|| q_vector
->tx
.ring
)
1310 napi_schedule_irqoff(&q_vector
->napi
);
1315 static inline void map_vector_to_rxq(struct ixgbevf_adapter
*a
, int v_idx
,
1318 struct ixgbevf_q_vector
*q_vector
= a
->q_vector
[v_idx
];
1320 a
->rx_ring
[r_idx
]->next
= q_vector
->rx
.ring
;
1321 q_vector
->rx
.ring
= a
->rx_ring
[r_idx
];
1322 q_vector
->rx
.count
++;
1325 static inline void map_vector_to_txq(struct ixgbevf_adapter
*a
, int v_idx
,
1328 struct ixgbevf_q_vector
*q_vector
= a
->q_vector
[v_idx
];
1330 a
->tx_ring
[t_idx
]->next
= q_vector
->tx
.ring
;
1331 q_vector
->tx
.ring
= a
->tx_ring
[t_idx
];
1332 q_vector
->tx
.count
++;
1336 * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
1337 * @adapter: board private structure to initialize
1339 * This function maps descriptor rings to the queue-specific vectors
1340 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1341 * one vector per ring/queue, but on a constrained vector budget, we
1342 * group the rings as "efficiently" as possible. You would add new
1343 * mapping configurations in here.
1345 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter
*adapter
)
1349 int rxr_idx
= 0, txr_idx
= 0;
1350 int rxr_remaining
= adapter
->num_rx_queues
;
1351 int txr_remaining
= adapter
->num_tx_queues
;
1355 q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1357 /* The ideal configuration...
1358 * We have enough vectors to map one per queue.
1360 if (q_vectors
== adapter
->num_rx_queues
+ adapter
->num_tx_queues
) {
1361 for (; rxr_idx
< rxr_remaining
; v_start
++, rxr_idx
++)
1362 map_vector_to_rxq(adapter
, v_start
, rxr_idx
);
1364 for (; txr_idx
< txr_remaining
; v_start
++, txr_idx
++)
1365 map_vector_to_txq(adapter
, v_start
, txr_idx
);
1369 /* If we don't have enough vectors for a 1-to-1
1370 * mapping, we'll have to group them so there are
1371 * multiple queues per vector.
1373 /* Re-adjusting *qpv takes care of the remainder. */
1374 for (i
= v_start
; i
< q_vectors
; i
++) {
1375 rqpv
= DIV_ROUND_UP(rxr_remaining
, q_vectors
- i
);
1376 for (j
= 0; j
< rqpv
; j
++) {
1377 map_vector_to_rxq(adapter
, i
, rxr_idx
);
1382 for (i
= v_start
; i
< q_vectors
; i
++) {
1383 tqpv
= DIV_ROUND_UP(txr_remaining
, q_vectors
- i
);
1384 for (j
= 0; j
< tqpv
; j
++) {
1385 map_vector_to_txq(adapter
, i
, txr_idx
);
1395 * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1396 * @adapter: board private structure
1398 * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1399 * interrupts from the kernel.
1401 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter
*adapter
)
1403 struct net_device
*netdev
= adapter
->netdev
;
1404 int q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1408 for (vector
= 0; vector
< q_vectors
; vector
++) {
1409 struct ixgbevf_q_vector
*q_vector
= adapter
->q_vector
[vector
];
1410 struct msix_entry
*entry
= &adapter
->msix_entries
[vector
];
1412 if (q_vector
->tx
.ring
&& q_vector
->rx
.ring
) {
1413 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
1414 "%s-%s-%d", netdev
->name
, "TxRx", ri
++);
1416 } else if (q_vector
->rx
.ring
) {
1417 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
1418 "%s-%s-%d", netdev
->name
, "rx", ri
++);
1419 } else if (q_vector
->tx
.ring
) {
1420 snprintf(q_vector
->name
, sizeof(q_vector
->name
) - 1,
1421 "%s-%s-%d", netdev
->name
, "tx", ti
++);
1423 /* skip this unused q_vector */
1426 err
= request_irq(entry
->vector
, &ixgbevf_msix_clean_rings
, 0,
1427 q_vector
->name
, q_vector
);
1429 hw_dbg(&adapter
->hw
,
1430 "request_irq failed for MSIX interrupt Error: %d\n",
1432 goto free_queue_irqs
;
1436 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1437 &ixgbevf_msix_other
, 0, netdev
->name
, adapter
);
1439 hw_dbg(&adapter
->hw
, "request_irq for msix_other failed: %d\n",
1441 goto free_queue_irqs
;
1449 free_irq(adapter
->msix_entries
[vector
].vector
,
1450 adapter
->q_vector
[vector
]);
1452 /* This failure is non-recoverable - it indicates the system is
1453 * out of MSIX vector resources and the VF driver cannot run
1454 * without them. Set the number of msix vectors to zero
1455 * indicating that not enough can be allocated. The error
1456 * will be returned to the user indicating device open failed.
1457 * Any further attempts to force the driver to open will also
1458 * fail. The only way to recover is to unload the driver and
1459 * reload it again. If the system has recovered some MSIX
1460 * vectors then it may succeed.
1462 adapter
->num_msix_vectors
= 0;
1466 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter
*adapter
)
1468 int i
, q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1470 for (i
= 0; i
< q_vectors
; i
++) {
1471 struct ixgbevf_q_vector
*q_vector
= adapter
->q_vector
[i
];
1473 q_vector
->rx
.ring
= NULL
;
1474 q_vector
->tx
.ring
= NULL
;
1475 q_vector
->rx
.count
= 0;
1476 q_vector
->tx
.count
= 0;
1481 * ixgbevf_request_irq - initialize interrupts
1482 * @adapter: board private structure
1484 * Attempts to configure interrupts using the best available
1485 * capabilities of the hardware and kernel.
1487 static int ixgbevf_request_irq(struct ixgbevf_adapter
*adapter
)
1489 int err
= ixgbevf_request_msix_irqs(adapter
);
1492 hw_dbg(&adapter
->hw
, "request_irq failed, Error %d\n", err
);
1497 static void ixgbevf_free_irq(struct ixgbevf_adapter
*adapter
)
1501 q_vectors
= adapter
->num_msix_vectors
;
1504 free_irq(adapter
->msix_entries
[i
].vector
, adapter
);
1507 for (; i
>= 0; i
--) {
1508 /* free only the irqs that were actually requested */
1509 if (!adapter
->q_vector
[i
]->rx
.ring
&&
1510 !adapter
->q_vector
[i
]->tx
.ring
)
1513 free_irq(adapter
->msix_entries
[i
].vector
,
1514 adapter
->q_vector
[i
]);
1517 ixgbevf_reset_q_vectors(adapter
);
1521 * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1522 * @adapter: board private structure
1524 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter
*adapter
)
1526 struct ixgbe_hw
*hw
= &adapter
->hw
;
1529 IXGBE_WRITE_REG(hw
, IXGBE_VTEIAM
, 0);
1530 IXGBE_WRITE_REG(hw
, IXGBE_VTEIMC
, ~0);
1531 IXGBE_WRITE_REG(hw
, IXGBE_VTEIAC
, 0);
1533 IXGBE_WRITE_FLUSH(hw
);
1535 for (i
= 0; i
< adapter
->num_msix_vectors
; i
++)
1536 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1540 * ixgbevf_irq_enable - Enable default interrupt generation settings
1541 * @adapter: board private structure
1543 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter
*adapter
)
1545 struct ixgbe_hw
*hw
= &adapter
->hw
;
1547 IXGBE_WRITE_REG(hw
, IXGBE_VTEIAM
, adapter
->eims_enable_mask
);
1548 IXGBE_WRITE_REG(hw
, IXGBE_VTEIAC
, adapter
->eims_enable_mask
);
1549 IXGBE_WRITE_REG(hw
, IXGBE_VTEIMS
, adapter
->eims_enable_mask
);
1553 * ixgbevf_configure_tx_ring - Configure 82599 VF Tx ring after Reset
1554 * @adapter: board private structure
1555 * @ring: structure containing ring specific data
1557 * Configure the Tx descriptor ring after a reset.
1559 static void ixgbevf_configure_tx_ring(struct ixgbevf_adapter
*adapter
,
1560 struct ixgbevf_ring
*ring
)
1562 struct ixgbe_hw
*hw
= &adapter
->hw
;
1563 u64 tdba
= ring
->dma
;
1565 u32 txdctl
= IXGBE_TXDCTL_ENABLE
;
1566 u8 reg_idx
= ring
->reg_idx
;
1568 /* disable queue to avoid issues while updating state */
1569 IXGBE_WRITE_REG(hw
, IXGBE_VFTXDCTL(reg_idx
), IXGBE_TXDCTL_SWFLSH
);
1570 IXGBE_WRITE_FLUSH(hw
);
1572 IXGBE_WRITE_REG(hw
, IXGBE_VFTDBAL(reg_idx
), tdba
& DMA_BIT_MASK(32));
1573 IXGBE_WRITE_REG(hw
, IXGBE_VFTDBAH(reg_idx
), tdba
>> 32);
1574 IXGBE_WRITE_REG(hw
, IXGBE_VFTDLEN(reg_idx
),
1575 ring
->count
* sizeof(union ixgbe_adv_tx_desc
));
1577 /* disable head writeback */
1578 IXGBE_WRITE_REG(hw
, IXGBE_VFTDWBAH(reg_idx
), 0);
1579 IXGBE_WRITE_REG(hw
, IXGBE_VFTDWBAL(reg_idx
), 0);
1581 /* enable relaxed ordering */
1582 IXGBE_WRITE_REG(hw
, IXGBE_VFDCA_TXCTRL(reg_idx
),
1583 (IXGBE_DCA_TXCTRL_DESC_RRO_EN
|
1584 IXGBE_DCA_TXCTRL_DATA_RRO_EN
));
1586 /* reset head and tail pointers */
1587 IXGBE_WRITE_REG(hw
, IXGBE_VFTDH(reg_idx
), 0);
1588 IXGBE_WRITE_REG(hw
, IXGBE_VFTDT(reg_idx
), 0);
1589 ring
->tail
= adapter
->io_addr
+ IXGBE_VFTDT(reg_idx
);
1591 /* reset ntu and ntc to place SW in sync with hardwdare */
1592 ring
->next_to_clean
= 0;
1593 ring
->next_to_use
= 0;
1595 /* In order to avoid issues WTHRESH + PTHRESH should always be equal
1596 * to or less than the number of on chip descriptors, which is
1599 txdctl
|= (8 << 16); /* WTHRESH = 8 */
1601 /* Setting PTHRESH to 32 both improves performance */
1602 txdctl
|= (1u << 8) | /* HTHRESH = 1 */
1603 32; /* PTHRESH = 32 */
1605 clear_bit(__IXGBEVF_HANG_CHECK_ARMED
, &ring
->state
);
1607 IXGBE_WRITE_REG(hw
, IXGBE_VFTXDCTL(reg_idx
), txdctl
);
1609 /* poll to verify queue is enabled */
1611 usleep_range(1000, 2000);
1612 txdctl
= IXGBE_READ_REG(hw
, IXGBE_VFTXDCTL(reg_idx
));
1613 } while (--wait_loop
&& !(txdctl
& IXGBE_TXDCTL_ENABLE
));
1615 pr_err("Could not enable Tx Queue %d\n", reg_idx
);
1619 * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1620 * @adapter: board private structure
1622 * Configure the Tx unit of the MAC after a reset.
1624 static void ixgbevf_configure_tx(struct ixgbevf_adapter
*adapter
)
1628 /* Setup the HW Tx Head and Tail descriptor pointers */
1629 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
1630 ixgbevf_configure_tx_ring(adapter
, adapter
->tx_ring
[i
]);
1633 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1635 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter
*adapter
, int index
)
1637 struct ixgbe_hw
*hw
= &adapter
->hw
;
1640 srrctl
= IXGBE_SRRCTL_DROP_EN
;
1642 srrctl
|= IXGBEVF_RX_HDR_SIZE
<< IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT
;
1643 srrctl
|= IXGBEVF_RX_BUFSZ
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
1644 srrctl
|= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF
;
1646 IXGBE_WRITE_REG(hw
, IXGBE_VFSRRCTL(index
), srrctl
);
1649 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter
*adapter
)
1651 struct ixgbe_hw
*hw
= &adapter
->hw
;
1653 /* PSRTYPE must be initialized in 82599 */
1654 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
| IXGBE_PSRTYPE_UDPHDR
|
1655 IXGBE_PSRTYPE_IPV4HDR
| IXGBE_PSRTYPE_IPV6HDR
|
1656 IXGBE_PSRTYPE_L2HDR
;
1658 if (adapter
->num_rx_queues
> 1)
1661 IXGBE_WRITE_REG(hw
, IXGBE_VFPSRTYPE
, psrtype
);
1664 #define IXGBEVF_MAX_RX_DESC_POLL 10
1665 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter
*adapter
,
1666 struct ixgbevf_ring
*ring
)
1668 struct ixgbe_hw
*hw
= &adapter
->hw
;
1669 int wait_loop
= IXGBEVF_MAX_RX_DESC_POLL
;
1671 u8 reg_idx
= ring
->reg_idx
;
1673 if (IXGBE_REMOVED(hw
->hw_addr
))
1675 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_VFRXDCTL(reg_idx
));
1676 rxdctl
&= ~IXGBE_RXDCTL_ENABLE
;
1678 /* write value back with RXDCTL.ENABLE bit cleared */
1679 IXGBE_WRITE_REG(hw
, IXGBE_VFRXDCTL(reg_idx
), rxdctl
);
1681 /* the hardware may take up to 100us to really disable the Rx queue */
1684 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_VFRXDCTL(reg_idx
));
1685 } while (--wait_loop
&& (rxdctl
& IXGBE_RXDCTL_ENABLE
));
1688 pr_err("RXDCTL.ENABLE queue %d not cleared while polling\n",
1692 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter
*adapter
,
1693 struct ixgbevf_ring
*ring
)
1695 struct ixgbe_hw
*hw
= &adapter
->hw
;
1696 int wait_loop
= IXGBEVF_MAX_RX_DESC_POLL
;
1698 u8 reg_idx
= ring
->reg_idx
;
1700 if (IXGBE_REMOVED(hw
->hw_addr
))
1703 usleep_range(1000, 2000);
1704 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_VFRXDCTL(reg_idx
));
1705 } while (--wait_loop
&& !(rxdctl
& IXGBE_RXDCTL_ENABLE
));
1708 pr_err("RXDCTL.ENABLE queue %d not set while polling\n",
1712 static void ixgbevf_setup_vfmrqc(struct ixgbevf_adapter
*adapter
)
1714 struct ixgbe_hw
*hw
= &adapter
->hw
;
1715 u32 vfmrqc
= 0, vfreta
= 0;
1716 u16 rss_i
= adapter
->num_rx_queues
;
1719 /* Fill out hash function seeds */
1720 netdev_rss_key_fill(adapter
->rss_key
, sizeof(adapter
->rss_key
));
1721 for (i
= 0; i
< IXGBEVF_VFRSSRK_REGS
; i
++)
1722 IXGBE_WRITE_REG(hw
, IXGBE_VFRSSRK(i
), adapter
->rss_key
[i
]);
1724 for (i
= 0, j
= 0; i
< IXGBEVF_X550_VFRETA_SIZE
; i
++, j
++) {
1728 adapter
->rss_indir_tbl
[i
] = j
;
1730 vfreta
|= j
<< (i
& 0x3) * 8;
1732 IXGBE_WRITE_REG(hw
, IXGBE_VFRETA(i
>> 2), vfreta
);
1737 /* Perform hash on these packet types */
1738 vfmrqc
|= IXGBE_VFMRQC_RSS_FIELD_IPV4
|
1739 IXGBE_VFMRQC_RSS_FIELD_IPV4_TCP
|
1740 IXGBE_VFMRQC_RSS_FIELD_IPV6
|
1741 IXGBE_VFMRQC_RSS_FIELD_IPV6_TCP
;
1743 vfmrqc
|= IXGBE_VFMRQC_RSSEN
;
1745 IXGBE_WRITE_REG(hw
, IXGBE_VFMRQC
, vfmrqc
);
1748 static void ixgbevf_configure_rx_ring(struct ixgbevf_adapter
*adapter
,
1749 struct ixgbevf_ring
*ring
)
1751 struct ixgbe_hw
*hw
= &adapter
->hw
;
1752 u64 rdba
= ring
->dma
;
1754 u8 reg_idx
= ring
->reg_idx
;
1756 /* disable queue to avoid issues while updating state */
1757 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_VFRXDCTL(reg_idx
));
1758 ixgbevf_disable_rx_queue(adapter
, ring
);
1760 IXGBE_WRITE_REG(hw
, IXGBE_VFRDBAL(reg_idx
), rdba
& DMA_BIT_MASK(32));
1761 IXGBE_WRITE_REG(hw
, IXGBE_VFRDBAH(reg_idx
), rdba
>> 32);
1762 IXGBE_WRITE_REG(hw
, IXGBE_VFRDLEN(reg_idx
),
1763 ring
->count
* sizeof(union ixgbe_adv_rx_desc
));
1765 #ifndef CONFIG_SPARC
1766 /* enable relaxed ordering */
1767 IXGBE_WRITE_REG(hw
, IXGBE_VFDCA_RXCTRL(reg_idx
),
1768 IXGBE_DCA_RXCTRL_DESC_RRO_EN
);
1770 IXGBE_WRITE_REG(hw
, IXGBE_VFDCA_RXCTRL(reg_idx
),
1771 IXGBE_DCA_RXCTRL_DESC_RRO_EN
|
1772 IXGBE_DCA_RXCTRL_DATA_WRO_EN
);
1775 /* reset head and tail pointers */
1776 IXGBE_WRITE_REG(hw
, IXGBE_VFRDH(reg_idx
), 0);
1777 IXGBE_WRITE_REG(hw
, IXGBE_VFRDT(reg_idx
), 0);
1778 ring
->tail
= adapter
->io_addr
+ IXGBE_VFRDT(reg_idx
);
1780 /* reset ntu and ntc to place SW in sync with hardwdare */
1781 ring
->next_to_clean
= 0;
1782 ring
->next_to_use
= 0;
1783 ring
->next_to_alloc
= 0;
1785 ixgbevf_configure_srrctl(adapter
, reg_idx
);
1787 /* allow any size packet since we can handle overflow */
1788 rxdctl
&= ~IXGBE_RXDCTL_RLPML_EN
;
1790 rxdctl
|= IXGBE_RXDCTL_ENABLE
| IXGBE_RXDCTL_VME
;
1791 IXGBE_WRITE_REG(hw
, IXGBE_VFRXDCTL(reg_idx
), rxdctl
);
1793 ixgbevf_rx_desc_queue_enable(adapter
, ring
);
1794 ixgbevf_alloc_rx_buffers(ring
, ixgbevf_desc_unused(ring
));
1798 * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1799 * @adapter: board private structure
1801 * Configure the Rx unit of the MAC after a reset.
1803 static void ixgbevf_configure_rx(struct ixgbevf_adapter
*adapter
)
1806 struct ixgbe_hw
*hw
= &adapter
->hw
;
1807 struct net_device
*netdev
= adapter
->netdev
;
1809 ixgbevf_setup_psrtype(adapter
);
1810 if (hw
->mac
.type
>= ixgbe_mac_X550_vf
)
1811 ixgbevf_setup_vfmrqc(adapter
);
1813 /* notify the PF of our intent to use this size of frame */
1814 hw
->mac
.ops
.set_rlpml(hw
, netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
);
1816 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1817 * the Base and Length of the Rx Descriptor Ring
1819 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
1820 ixgbevf_configure_rx_ring(adapter
, adapter
->rx_ring
[i
]);
1823 static int ixgbevf_vlan_rx_add_vid(struct net_device
*netdev
,
1824 __be16 proto
, u16 vid
)
1826 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
1827 struct ixgbe_hw
*hw
= &adapter
->hw
;
1830 spin_lock_bh(&adapter
->mbx_lock
);
1832 /* add VID to filter table */
1833 err
= hw
->mac
.ops
.set_vfta(hw
, vid
, 0, true);
1835 spin_unlock_bh(&adapter
->mbx_lock
);
1837 /* translate error return types so error makes sense */
1838 if (err
== IXGBE_ERR_MBX
)
1841 if (err
== IXGBE_ERR_INVALID_ARGUMENT
)
1844 set_bit(vid
, adapter
->active_vlans
);
1849 static int ixgbevf_vlan_rx_kill_vid(struct net_device
*netdev
,
1850 __be16 proto
, u16 vid
)
1852 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
1853 struct ixgbe_hw
*hw
= &adapter
->hw
;
1856 spin_lock_bh(&adapter
->mbx_lock
);
1858 /* remove VID from filter table */
1859 err
= hw
->mac
.ops
.set_vfta(hw
, vid
, 0, false);
1861 spin_unlock_bh(&adapter
->mbx_lock
);
1863 clear_bit(vid
, adapter
->active_vlans
);
1868 static void ixgbevf_restore_vlan(struct ixgbevf_adapter
*adapter
)
1872 for_each_set_bit(vid
, adapter
->active_vlans
, VLAN_N_VID
)
1873 ixgbevf_vlan_rx_add_vid(adapter
->netdev
,
1874 htons(ETH_P_8021Q
), vid
);
1877 static int ixgbevf_write_uc_addr_list(struct net_device
*netdev
)
1879 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
1880 struct ixgbe_hw
*hw
= &adapter
->hw
;
1883 if ((netdev_uc_count(netdev
)) > 10) {
1884 pr_err("Too many unicast filters - No Space\n");
1888 if (!netdev_uc_empty(netdev
)) {
1889 struct netdev_hw_addr
*ha
;
1891 netdev_for_each_uc_addr(ha
, netdev
) {
1892 hw
->mac
.ops
.set_uc_addr(hw
, ++count
, ha
->addr
);
1896 /* If the list is empty then send message to PF driver to
1897 * clear all MAC VLANs on this VF.
1899 hw
->mac
.ops
.set_uc_addr(hw
, 0, NULL
);
1906 * ixgbevf_set_rx_mode - Multicast and unicast set
1907 * @netdev: network interface device structure
1909 * The set_rx_method entry point is called whenever the multicast address
1910 * list, unicast address list or the network interface flags are updated.
1911 * This routine is responsible for configuring the hardware for proper
1912 * multicast mode and configuring requested unicast filters.
1914 static void ixgbevf_set_rx_mode(struct net_device
*netdev
)
1916 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
1917 struct ixgbe_hw
*hw
= &adapter
->hw
;
1918 unsigned int flags
= netdev
->flags
;
1921 xcast_mode
= (flags
& IFF_ALLMULTI
) ? IXGBEVF_XCAST_MODE_ALLMULTI
:
1922 (flags
& (IFF_BROADCAST
| IFF_MULTICAST
)) ?
1923 IXGBEVF_XCAST_MODE_MULTI
: IXGBEVF_XCAST_MODE_NONE
;
1925 spin_lock_bh(&adapter
->mbx_lock
);
1927 hw
->mac
.ops
.update_xcast_mode(hw
, xcast_mode
);
1929 /* reprogram multicast list */
1930 hw
->mac
.ops
.update_mc_addr_list(hw
, netdev
);
1932 ixgbevf_write_uc_addr_list(netdev
);
1934 spin_unlock_bh(&adapter
->mbx_lock
);
1937 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter
*adapter
)
1940 struct ixgbevf_q_vector
*q_vector
;
1941 int q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1943 for (q_idx
= 0; q_idx
< q_vectors
; q_idx
++) {
1944 q_vector
= adapter
->q_vector
[q_idx
];
1945 #ifdef CONFIG_NET_RX_BUSY_POLL
1946 ixgbevf_qv_init_lock(adapter
->q_vector
[q_idx
]);
1948 napi_enable(&q_vector
->napi
);
1952 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter
*adapter
)
1955 struct ixgbevf_q_vector
*q_vector
;
1956 int q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1958 for (q_idx
= 0; q_idx
< q_vectors
; q_idx
++) {
1959 q_vector
= adapter
->q_vector
[q_idx
];
1960 napi_disable(&q_vector
->napi
);
1961 #ifdef CONFIG_NET_RX_BUSY_POLL
1962 while (!ixgbevf_qv_disable(adapter
->q_vector
[q_idx
])) {
1963 pr_info("QV %d locked\n", q_idx
);
1964 usleep_range(1000, 20000);
1966 #endif /* CONFIG_NET_RX_BUSY_POLL */
1970 static int ixgbevf_configure_dcb(struct ixgbevf_adapter
*adapter
)
1972 struct ixgbe_hw
*hw
= &adapter
->hw
;
1973 unsigned int def_q
= 0;
1974 unsigned int num_tcs
= 0;
1975 unsigned int num_rx_queues
= adapter
->num_rx_queues
;
1976 unsigned int num_tx_queues
= adapter
->num_tx_queues
;
1979 spin_lock_bh(&adapter
->mbx_lock
);
1981 /* fetch queue configuration from the PF */
1982 err
= ixgbevf_get_queues(hw
, &num_tcs
, &def_q
);
1984 spin_unlock_bh(&adapter
->mbx_lock
);
1990 /* we need only one Tx queue */
1993 /* update default Tx ring register index */
1994 adapter
->tx_ring
[0]->reg_idx
= def_q
;
1996 /* we need as many queues as traffic classes */
1997 num_rx_queues
= num_tcs
;
2000 /* if we have a bad config abort request queue reset */
2001 if ((adapter
->num_rx_queues
!= num_rx_queues
) ||
2002 (adapter
->num_tx_queues
!= num_tx_queues
)) {
2003 /* force mailbox timeout to prevent further messages */
2004 hw
->mbx
.timeout
= 0;
2006 /* wait for watchdog to come around and bail us out */
2007 set_bit(__IXGBEVF_QUEUE_RESET_REQUESTED
, &adapter
->state
);
2013 static void ixgbevf_configure(struct ixgbevf_adapter
*adapter
)
2015 ixgbevf_configure_dcb(adapter
);
2017 ixgbevf_set_rx_mode(adapter
->netdev
);
2019 ixgbevf_restore_vlan(adapter
);
2021 ixgbevf_configure_tx(adapter
);
2022 ixgbevf_configure_rx(adapter
);
2025 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter
*adapter
)
2027 /* Only save pre-reset stats if there are some */
2028 if (adapter
->stats
.vfgprc
|| adapter
->stats
.vfgptc
) {
2029 adapter
->stats
.saved_reset_vfgprc
+= adapter
->stats
.vfgprc
-
2030 adapter
->stats
.base_vfgprc
;
2031 adapter
->stats
.saved_reset_vfgptc
+= adapter
->stats
.vfgptc
-
2032 adapter
->stats
.base_vfgptc
;
2033 adapter
->stats
.saved_reset_vfgorc
+= adapter
->stats
.vfgorc
-
2034 adapter
->stats
.base_vfgorc
;
2035 adapter
->stats
.saved_reset_vfgotc
+= adapter
->stats
.vfgotc
-
2036 adapter
->stats
.base_vfgotc
;
2037 adapter
->stats
.saved_reset_vfmprc
+= adapter
->stats
.vfmprc
-
2038 adapter
->stats
.base_vfmprc
;
2042 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter
*adapter
)
2044 struct ixgbe_hw
*hw
= &adapter
->hw
;
2046 adapter
->stats
.last_vfgprc
= IXGBE_READ_REG(hw
, IXGBE_VFGPRC
);
2047 adapter
->stats
.last_vfgorc
= IXGBE_READ_REG(hw
, IXGBE_VFGORC_LSB
);
2048 adapter
->stats
.last_vfgorc
|=
2049 (((u64
)(IXGBE_READ_REG(hw
, IXGBE_VFGORC_MSB
))) << 32);
2050 adapter
->stats
.last_vfgptc
= IXGBE_READ_REG(hw
, IXGBE_VFGPTC
);
2051 adapter
->stats
.last_vfgotc
= IXGBE_READ_REG(hw
, IXGBE_VFGOTC_LSB
);
2052 adapter
->stats
.last_vfgotc
|=
2053 (((u64
)(IXGBE_READ_REG(hw
, IXGBE_VFGOTC_MSB
))) << 32);
2054 adapter
->stats
.last_vfmprc
= IXGBE_READ_REG(hw
, IXGBE_VFMPRC
);
2056 adapter
->stats
.base_vfgprc
= adapter
->stats
.last_vfgprc
;
2057 adapter
->stats
.base_vfgorc
= adapter
->stats
.last_vfgorc
;
2058 adapter
->stats
.base_vfgptc
= adapter
->stats
.last_vfgptc
;
2059 adapter
->stats
.base_vfgotc
= adapter
->stats
.last_vfgotc
;
2060 adapter
->stats
.base_vfmprc
= adapter
->stats
.last_vfmprc
;
2063 static void ixgbevf_negotiate_api(struct ixgbevf_adapter
*adapter
)
2065 struct ixgbe_hw
*hw
= &adapter
->hw
;
2066 int api
[] = { ixgbe_mbox_api_12
,
2069 ixgbe_mbox_api_unknown
};
2072 spin_lock_bh(&adapter
->mbx_lock
);
2074 while (api
[idx
] != ixgbe_mbox_api_unknown
) {
2075 err
= hw
->mac
.ops
.negotiate_api_version(hw
, api
[idx
]);
2081 spin_unlock_bh(&adapter
->mbx_lock
);
2084 static void ixgbevf_up_complete(struct ixgbevf_adapter
*adapter
)
2086 struct net_device
*netdev
= adapter
->netdev
;
2087 struct ixgbe_hw
*hw
= &adapter
->hw
;
2089 ixgbevf_configure_msix(adapter
);
2091 spin_lock_bh(&adapter
->mbx_lock
);
2093 if (is_valid_ether_addr(hw
->mac
.addr
))
2094 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.addr
, 0);
2096 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.perm_addr
, 0);
2098 spin_unlock_bh(&adapter
->mbx_lock
);
2100 smp_mb__before_atomic();
2101 clear_bit(__IXGBEVF_DOWN
, &adapter
->state
);
2102 ixgbevf_napi_enable_all(adapter
);
2104 /* clear any pending interrupts, may auto mask */
2105 IXGBE_READ_REG(hw
, IXGBE_VTEICR
);
2106 ixgbevf_irq_enable(adapter
);
2108 /* enable transmits */
2109 netif_tx_start_all_queues(netdev
);
2111 ixgbevf_save_reset_stats(adapter
);
2112 ixgbevf_init_last_counter_stats(adapter
);
2114 hw
->mac
.get_link_status
= 1;
2115 mod_timer(&adapter
->service_timer
, jiffies
);
2118 void ixgbevf_up(struct ixgbevf_adapter
*adapter
)
2120 ixgbevf_configure(adapter
);
2122 ixgbevf_up_complete(adapter
);
2126 * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
2127 * @rx_ring: ring to free buffers from
2129 static void ixgbevf_clean_rx_ring(struct ixgbevf_ring
*rx_ring
)
2131 struct device
*dev
= rx_ring
->dev
;
2135 /* Free Rx ring sk_buff */
2137 dev_kfree_skb(rx_ring
->skb
);
2138 rx_ring
->skb
= NULL
;
2141 /* ring already cleared, nothing to do */
2142 if (!rx_ring
->rx_buffer_info
)
2145 /* Free all the Rx ring pages */
2146 for (i
= 0; i
< rx_ring
->count
; i
++) {
2147 struct ixgbevf_rx_buffer
*rx_buffer
;
2149 rx_buffer
= &rx_ring
->rx_buffer_info
[i
];
2151 dma_unmap_page(dev
, rx_buffer
->dma
,
2152 PAGE_SIZE
, DMA_FROM_DEVICE
);
2154 if (rx_buffer
->page
)
2155 __free_page(rx_buffer
->page
);
2156 rx_buffer
->page
= NULL
;
2159 size
= sizeof(struct ixgbevf_rx_buffer
) * rx_ring
->count
;
2160 memset(rx_ring
->rx_buffer_info
, 0, size
);
2162 /* Zero out the descriptor ring */
2163 memset(rx_ring
->desc
, 0, rx_ring
->size
);
2167 * ixgbevf_clean_tx_ring - Free Tx Buffers
2168 * @tx_ring: ring to be cleaned
2170 static void ixgbevf_clean_tx_ring(struct ixgbevf_ring
*tx_ring
)
2172 struct ixgbevf_tx_buffer
*tx_buffer_info
;
2176 if (!tx_ring
->tx_buffer_info
)
2179 /* Free all the Tx ring sk_buffs */
2180 for (i
= 0; i
< tx_ring
->count
; i
++) {
2181 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
2182 ixgbevf_unmap_and_free_tx_resource(tx_ring
, tx_buffer_info
);
2185 size
= sizeof(struct ixgbevf_tx_buffer
) * tx_ring
->count
;
2186 memset(tx_ring
->tx_buffer_info
, 0, size
);
2188 memset(tx_ring
->desc
, 0, tx_ring
->size
);
2192 * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
2193 * @adapter: board private structure
2195 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter
*adapter
)
2199 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2200 ixgbevf_clean_rx_ring(adapter
->rx_ring
[i
]);
2204 * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
2205 * @adapter: board private structure
2207 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter
*adapter
)
2211 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2212 ixgbevf_clean_tx_ring(adapter
->tx_ring
[i
]);
2215 void ixgbevf_down(struct ixgbevf_adapter
*adapter
)
2217 struct net_device
*netdev
= adapter
->netdev
;
2218 struct ixgbe_hw
*hw
= &adapter
->hw
;
2221 /* signal that we are down to the interrupt handler */
2222 if (test_and_set_bit(__IXGBEVF_DOWN
, &adapter
->state
))
2223 return; /* do nothing if already down */
2225 /* disable all enabled Rx queues */
2226 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2227 ixgbevf_disable_rx_queue(adapter
, adapter
->rx_ring
[i
]);
2229 usleep_range(10000, 20000);
2231 netif_tx_stop_all_queues(netdev
);
2233 /* call carrier off first to avoid false dev_watchdog timeouts */
2234 netif_carrier_off(netdev
);
2235 netif_tx_disable(netdev
);
2237 ixgbevf_irq_disable(adapter
);
2239 ixgbevf_napi_disable_all(adapter
);
2241 del_timer_sync(&adapter
->service_timer
);
2243 /* disable transmits in the hardware now that interrupts are off */
2244 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2245 u8 reg_idx
= adapter
->tx_ring
[i
]->reg_idx
;
2247 IXGBE_WRITE_REG(hw
, IXGBE_VFTXDCTL(reg_idx
),
2248 IXGBE_TXDCTL_SWFLSH
);
2251 if (!pci_channel_offline(adapter
->pdev
))
2252 ixgbevf_reset(adapter
);
2254 ixgbevf_clean_all_tx_rings(adapter
);
2255 ixgbevf_clean_all_rx_rings(adapter
);
2258 void ixgbevf_reinit_locked(struct ixgbevf_adapter
*adapter
)
2260 WARN_ON(in_interrupt());
2262 while (test_and_set_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
2265 ixgbevf_down(adapter
);
2266 ixgbevf_up(adapter
);
2268 clear_bit(__IXGBEVF_RESETTING
, &adapter
->state
);
2271 void ixgbevf_reset(struct ixgbevf_adapter
*adapter
)
2273 struct ixgbe_hw
*hw
= &adapter
->hw
;
2274 struct net_device
*netdev
= adapter
->netdev
;
2276 if (hw
->mac
.ops
.reset_hw(hw
)) {
2277 hw_dbg(hw
, "PF still resetting\n");
2279 hw
->mac
.ops
.init_hw(hw
);
2280 ixgbevf_negotiate_api(adapter
);
2283 if (is_valid_ether_addr(adapter
->hw
.mac
.addr
)) {
2284 ether_addr_copy(netdev
->dev_addr
, adapter
->hw
.mac
.addr
);
2285 ether_addr_copy(netdev
->perm_addr
, adapter
->hw
.mac
.addr
);
2288 adapter
->last_reset
= jiffies
;
2291 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter
*adapter
,
2294 int vector_threshold
;
2296 /* We'll want at least 2 (vector_threshold):
2297 * 1) TxQ[0] + RxQ[0] handler
2298 * 2) Other (Link Status Change, etc.)
2300 vector_threshold
= MIN_MSIX_COUNT
;
2302 /* The more we get, the more we will assign to Tx/Rx Cleanup
2303 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2304 * Right now, we simply care about how many we'll get; we'll
2305 * set them up later while requesting irq's.
2307 vectors
= pci_enable_msix_range(adapter
->pdev
, adapter
->msix_entries
,
2308 vector_threshold
, vectors
);
2311 dev_err(&adapter
->pdev
->dev
,
2312 "Unable to allocate MSI-X interrupts\n");
2313 kfree(adapter
->msix_entries
);
2314 adapter
->msix_entries
= NULL
;
2318 /* Adjust for only the vectors we'll use, which is minimum
2319 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
2320 * vectors we were allocated.
2322 adapter
->num_msix_vectors
= vectors
;
2328 * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
2329 * @adapter: board private structure to initialize
2331 * This is the top level queue allocation routine. The order here is very
2332 * important, starting with the "most" number of features turned on at once,
2333 * and ending with the smallest set of features. This way large combinations
2334 * can be allocated if they're turned on, and smaller combinations are the
2335 * fallthrough conditions.
2338 static void ixgbevf_set_num_queues(struct ixgbevf_adapter
*adapter
)
2340 struct ixgbe_hw
*hw
= &adapter
->hw
;
2341 unsigned int def_q
= 0;
2342 unsigned int num_tcs
= 0;
2345 /* Start with base case */
2346 adapter
->num_rx_queues
= 1;
2347 adapter
->num_tx_queues
= 1;
2349 spin_lock_bh(&adapter
->mbx_lock
);
2351 /* fetch queue configuration from the PF */
2352 err
= ixgbevf_get_queues(hw
, &num_tcs
, &def_q
);
2354 spin_unlock_bh(&adapter
->mbx_lock
);
2359 /* we need as many queues as traffic classes */
2361 adapter
->num_rx_queues
= num_tcs
;
2363 u16 rss
= min_t(u16
, num_online_cpus(), IXGBEVF_MAX_RSS_QUEUES
);
2365 switch (hw
->api_version
) {
2366 case ixgbe_mbox_api_11
:
2367 case ixgbe_mbox_api_12
:
2368 adapter
->num_rx_queues
= rss
;
2369 adapter
->num_tx_queues
= rss
;
2377 * ixgbevf_alloc_queues - Allocate memory for all rings
2378 * @adapter: board private structure to initialize
2380 * We allocate one ring per queue at run-time since we don't know the
2381 * number of queues at compile-time. The polling_netdev array is
2382 * intended for Multiqueue, but should work fine with a single queue.
2384 static int ixgbevf_alloc_queues(struct ixgbevf_adapter
*adapter
)
2386 struct ixgbevf_ring
*ring
;
2389 for (; tx
< adapter
->num_tx_queues
; tx
++) {
2390 ring
= kzalloc(sizeof(*ring
), GFP_KERNEL
);
2392 goto err_allocation
;
2394 ring
->dev
= &adapter
->pdev
->dev
;
2395 ring
->netdev
= adapter
->netdev
;
2396 ring
->count
= adapter
->tx_ring_count
;
2397 ring
->queue_index
= tx
;
2400 adapter
->tx_ring
[tx
] = ring
;
2403 for (; rx
< adapter
->num_rx_queues
; rx
++) {
2404 ring
= kzalloc(sizeof(*ring
), GFP_KERNEL
);
2406 goto err_allocation
;
2408 ring
->dev
= &adapter
->pdev
->dev
;
2409 ring
->netdev
= adapter
->netdev
;
2411 ring
->count
= adapter
->rx_ring_count
;
2412 ring
->queue_index
= rx
;
2415 adapter
->rx_ring
[rx
] = ring
;
2422 kfree(adapter
->tx_ring
[--tx
]);
2423 adapter
->tx_ring
[tx
] = NULL
;
2427 kfree(adapter
->rx_ring
[--rx
]);
2428 adapter
->rx_ring
[rx
] = NULL
;
2434 * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
2435 * @adapter: board private structure to initialize
2437 * Attempt to configure the interrupts using the best available
2438 * capabilities of the hardware and the kernel.
2440 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter
*adapter
)
2442 struct net_device
*netdev
= adapter
->netdev
;
2444 int vector
, v_budget
;
2446 /* It's easy to be greedy for MSI-X vectors, but it really
2447 * doesn't do us much good if we have a lot more vectors
2448 * than CPU's. So let's be conservative and only ask for
2449 * (roughly) the same number of vectors as there are CPU's.
2450 * The default is to use pairs of vectors.
2452 v_budget
= max(adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2453 v_budget
= min_t(int, v_budget
, num_online_cpus());
2454 v_budget
+= NON_Q_VECTORS
;
2456 /* A failure in MSI-X entry allocation isn't fatal, but it does
2457 * mean we disable MSI-X capabilities of the adapter.
2459 adapter
->msix_entries
= kcalloc(v_budget
,
2460 sizeof(struct msix_entry
), GFP_KERNEL
);
2461 if (!adapter
->msix_entries
)
2464 for (vector
= 0; vector
< v_budget
; vector
++)
2465 adapter
->msix_entries
[vector
].entry
= vector
;
2467 err
= ixgbevf_acquire_msix_vectors(adapter
, v_budget
);
2471 err
= netif_set_real_num_tx_queues(netdev
, adapter
->num_tx_queues
);
2475 return netif_set_real_num_rx_queues(netdev
, adapter
->num_rx_queues
);
2479 * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
2480 * @adapter: board private structure to initialize
2482 * We allocate one q_vector per queue interrupt. If allocation fails we
2485 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter
*adapter
)
2487 int q_idx
, num_q_vectors
;
2488 struct ixgbevf_q_vector
*q_vector
;
2490 num_q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
2492 for (q_idx
= 0; q_idx
< num_q_vectors
; q_idx
++) {
2493 q_vector
= kzalloc(sizeof(struct ixgbevf_q_vector
), GFP_KERNEL
);
2496 q_vector
->adapter
= adapter
;
2497 q_vector
->v_idx
= q_idx
;
2498 netif_napi_add(adapter
->netdev
, &q_vector
->napi
,
2500 adapter
->q_vector
[q_idx
] = q_vector
;
2508 q_vector
= adapter
->q_vector
[q_idx
];
2509 #ifdef CONFIG_NET_RX_BUSY_POLL
2510 napi_hash_del(&q_vector
->napi
);
2512 netif_napi_del(&q_vector
->napi
);
2514 adapter
->q_vector
[q_idx
] = NULL
;
2520 * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2521 * @adapter: board private structure to initialize
2523 * This function frees the memory allocated to the q_vectors. In addition if
2524 * NAPI is enabled it will delete any references to the NAPI struct prior
2525 * to freeing the q_vector.
2527 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter
*adapter
)
2529 int q_idx
, num_q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
2531 for (q_idx
= 0; q_idx
< num_q_vectors
; q_idx
++) {
2532 struct ixgbevf_q_vector
*q_vector
= adapter
->q_vector
[q_idx
];
2534 adapter
->q_vector
[q_idx
] = NULL
;
2535 #ifdef CONFIG_NET_RX_BUSY_POLL
2536 napi_hash_del(&q_vector
->napi
);
2538 netif_napi_del(&q_vector
->napi
);
2544 * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2545 * @adapter: board private structure
2548 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter
*adapter
)
2550 pci_disable_msix(adapter
->pdev
);
2551 kfree(adapter
->msix_entries
);
2552 adapter
->msix_entries
= NULL
;
2556 * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2557 * @adapter: board private structure to initialize
2560 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter
*adapter
)
2564 /* Number of supported queues */
2565 ixgbevf_set_num_queues(adapter
);
2567 err
= ixgbevf_set_interrupt_capability(adapter
);
2569 hw_dbg(&adapter
->hw
,
2570 "Unable to setup interrupt capabilities\n");
2571 goto err_set_interrupt
;
2574 err
= ixgbevf_alloc_q_vectors(adapter
);
2576 hw_dbg(&adapter
->hw
, "Unable to allocate memory for queue vectors\n");
2577 goto err_alloc_q_vectors
;
2580 err
= ixgbevf_alloc_queues(adapter
);
2582 pr_err("Unable to allocate memory for queues\n");
2583 goto err_alloc_queues
;
2586 hw_dbg(&adapter
->hw
, "Multiqueue %s: Rx Queue count = %u, Tx Queue count = %u\n",
2587 (adapter
->num_rx_queues
> 1) ? "Enabled" :
2588 "Disabled", adapter
->num_rx_queues
, adapter
->num_tx_queues
);
2590 set_bit(__IXGBEVF_DOWN
, &adapter
->state
);
2594 ixgbevf_free_q_vectors(adapter
);
2595 err_alloc_q_vectors
:
2596 ixgbevf_reset_interrupt_capability(adapter
);
2602 * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2603 * @adapter: board private structure to clear interrupt scheme on
2605 * We go through and clear interrupt specific resources and reset the structure
2606 * to pre-load conditions
2608 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter
*adapter
)
2612 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2613 kfree(adapter
->tx_ring
[i
]);
2614 adapter
->tx_ring
[i
] = NULL
;
2616 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2617 kfree(adapter
->rx_ring
[i
]);
2618 adapter
->rx_ring
[i
] = NULL
;
2621 adapter
->num_tx_queues
= 0;
2622 adapter
->num_rx_queues
= 0;
2624 ixgbevf_free_q_vectors(adapter
);
2625 ixgbevf_reset_interrupt_capability(adapter
);
2629 * ixgbevf_sw_init - Initialize general software structures
2630 * @adapter: board private structure to initialize
2632 * ixgbevf_sw_init initializes the Adapter private data structure.
2633 * Fields are initialized based on PCI device information and
2634 * OS network device settings (MTU size).
2636 static int ixgbevf_sw_init(struct ixgbevf_adapter
*adapter
)
2638 struct ixgbe_hw
*hw
= &adapter
->hw
;
2639 struct pci_dev
*pdev
= adapter
->pdev
;
2640 struct net_device
*netdev
= adapter
->netdev
;
2643 /* PCI config space info */
2644 hw
->vendor_id
= pdev
->vendor
;
2645 hw
->device_id
= pdev
->device
;
2646 hw
->revision_id
= pdev
->revision
;
2647 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
2648 hw
->subsystem_device_id
= pdev
->subsystem_device
;
2650 hw
->mbx
.ops
.init_params(hw
);
2652 /* assume legacy case in which PF would only give VF 2 queues */
2653 hw
->mac
.max_tx_queues
= 2;
2654 hw
->mac
.max_rx_queues
= 2;
2656 /* lock to protect mailbox accesses */
2657 spin_lock_init(&adapter
->mbx_lock
);
2659 err
= hw
->mac
.ops
.reset_hw(hw
);
2661 dev_info(&pdev
->dev
,
2662 "PF still in reset state. Is the PF interface up?\n");
2664 err
= hw
->mac
.ops
.init_hw(hw
);
2666 pr_err("init_shared_code failed: %d\n", err
);
2669 ixgbevf_negotiate_api(adapter
);
2670 err
= hw
->mac
.ops
.get_mac_addr(hw
, hw
->mac
.addr
);
2672 dev_info(&pdev
->dev
, "Error reading MAC address\n");
2673 else if (is_zero_ether_addr(adapter
->hw
.mac
.addr
))
2674 dev_info(&pdev
->dev
,
2675 "MAC address not assigned by administrator.\n");
2676 ether_addr_copy(netdev
->dev_addr
, hw
->mac
.addr
);
2679 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
2680 dev_info(&pdev
->dev
, "Assigning random MAC address\n");
2681 eth_hw_addr_random(netdev
);
2682 ether_addr_copy(hw
->mac
.addr
, netdev
->dev_addr
);
2683 ether_addr_copy(hw
->mac
.perm_addr
, netdev
->dev_addr
);
2686 /* Enable dynamic interrupt throttling rates */
2687 adapter
->rx_itr_setting
= 1;
2688 adapter
->tx_itr_setting
= 1;
2690 /* set default ring sizes */
2691 adapter
->tx_ring_count
= IXGBEVF_DEFAULT_TXD
;
2692 adapter
->rx_ring_count
= IXGBEVF_DEFAULT_RXD
;
2694 set_bit(__IXGBEVF_DOWN
, &adapter
->state
);
2701 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter) \
2703 u32 current_counter = IXGBE_READ_REG(hw, reg); \
2704 if (current_counter < last_counter) \
2705 counter += 0x100000000LL; \
2706 last_counter = current_counter; \
2707 counter &= 0xFFFFFFFF00000000LL; \
2708 counter |= current_counter; \
2711 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2713 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb); \
2714 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb); \
2715 u64 current_counter = (current_counter_msb << 32) | \
2716 current_counter_lsb; \
2717 if (current_counter < last_counter) \
2718 counter += 0x1000000000LL; \
2719 last_counter = current_counter; \
2720 counter &= 0xFFFFFFF000000000LL; \
2721 counter |= current_counter; \
2724 * ixgbevf_update_stats - Update the board statistics counters.
2725 * @adapter: board private structure
2727 void ixgbevf_update_stats(struct ixgbevf_adapter
*adapter
)
2729 struct ixgbe_hw
*hw
= &adapter
->hw
;
2732 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
) ||
2733 test_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
2736 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC
, adapter
->stats
.last_vfgprc
,
2737 adapter
->stats
.vfgprc
);
2738 UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC
, adapter
->stats
.last_vfgptc
,
2739 adapter
->stats
.vfgptc
);
2740 UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB
, IXGBE_VFGORC_MSB
,
2741 adapter
->stats
.last_vfgorc
,
2742 adapter
->stats
.vfgorc
);
2743 UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB
, IXGBE_VFGOTC_MSB
,
2744 adapter
->stats
.last_vfgotc
,
2745 adapter
->stats
.vfgotc
);
2746 UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC
, adapter
->stats
.last_vfmprc
,
2747 adapter
->stats
.vfmprc
);
2749 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2750 adapter
->hw_csum_rx_error
+=
2751 adapter
->rx_ring
[i
]->hw_csum_rx_error
;
2752 adapter
->rx_ring
[i
]->hw_csum_rx_error
= 0;
2757 * ixgbevf_service_timer - Timer Call-back
2758 * @data: pointer to adapter cast into an unsigned long
2760 static void ixgbevf_service_timer(unsigned long data
)
2762 struct ixgbevf_adapter
*adapter
= (struct ixgbevf_adapter
*)data
;
2764 /* Reset the timer */
2765 mod_timer(&adapter
->service_timer
, (HZ
* 2) + jiffies
);
2767 ixgbevf_service_event_schedule(adapter
);
2770 static void ixgbevf_reset_subtask(struct ixgbevf_adapter
*adapter
)
2772 if (!test_and_clear_bit(__IXGBEVF_RESET_REQUESTED
, &adapter
->state
))
2775 /* If we're already down or resetting, just bail */
2776 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
) ||
2777 test_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
2780 adapter
->tx_timeout_count
++;
2782 ixgbevf_reinit_locked(adapter
);
2786 * ixgbevf_check_hang_subtask - check for hung queues and dropped interrupts
2787 * @adapter: pointer to the device adapter structure
2789 * This function serves two purposes. First it strobes the interrupt lines
2790 * in order to make certain interrupts are occurring. Secondly it sets the
2791 * bits needed to check for TX hangs. As a result we should immediately
2792 * determine if a hang has occurred.
2794 static void ixgbevf_check_hang_subtask(struct ixgbevf_adapter
*adapter
)
2796 struct ixgbe_hw
*hw
= &adapter
->hw
;
2800 /* If we're down or resetting, just bail */
2801 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
) ||
2802 test_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
2805 /* Force detection of hung controller */
2806 if (netif_carrier_ok(adapter
->netdev
)) {
2807 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2808 set_check_for_tx_hang(adapter
->tx_ring
[i
]);
2811 /* get one bit for every active Tx/Rx interrupt vector */
2812 for (i
= 0; i
< adapter
->num_msix_vectors
- NON_Q_VECTORS
; i
++) {
2813 struct ixgbevf_q_vector
*qv
= adapter
->q_vector
[i
];
2815 if (qv
->rx
.ring
|| qv
->tx
.ring
)
2819 /* Cause software interrupt to ensure rings are cleaned */
2820 IXGBE_WRITE_REG(hw
, IXGBE_VTEICS
, eics
);
2824 * ixgbevf_watchdog_update_link - update the link status
2825 * @adapter: pointer to the device adapter structure
2827 static void ixgbevf_watchdog_update_link(struct ixgbevf_adapter
*adapter
)
2829 struct ixgbe_hw
*hw
= &adapter
->hw
;
2830 u32 link_speed
= adapter
->link_speed
;
2831 bool link_up
= adapter
->link_up
;
2834 spin_lock_bh(&adapter
->mbx_lock
);
2836 err
= hw
->mac
.ops
.check_link(hw
, &link_speed
, &link_up
, false);
2838 spin_unlock_bh(&adapter
->mbx_lock
);
2840 /* if check for link returns error we will need to reset */
2841 if (err
&& time_after(jiffies
, adapter
->last_reset
+ (10 * HZ
))) {
2842 set_bit(__IXGBEVF_RESET_REQUESTED
, &adapter
->state
);
2846 adapter
->link_up
= link_up
;
2847 adapter
->link_speed
= link_speed
;
2851 * ixgbevf_watchdog_link_is_up - update netif_carrier status and
2852 * print link up message
2853 * @adapter: pointer to the device adapter structure
2855 static void ixgbevf_watchdog_link_is_up(struct ixgbevf_adapter
*adapter
)
2857 struct net_device
*netdev
= adapter
->netdev
;
2859 /* only continue if link was previously down */
2860 if (netif_carrier_ok(netdev
))
2863 dev_info(&adapter
->pdev
->dev
, "NIC Link is Up %s\n",
2864 (adapter
->link_speed
== IXGBE_LINK_SPEED_10GB_FULL
) ?
2866 (adapter
->link_speed
== IXGBE_LINK_SPEED_1GB_FULL
) ?
2868 (adapter
->link_speed
== IXGBE_LINK_SPEED_100_FULL
) ?
2872 netif_carrier_on(netdev
);
2876 * ixgbevf_watchdog_link_is_down - update netif_carrier status and
2877 * print link down message
2878 * @adapter: pointer to the adapter structure
2880 static void ixgbevf_watchdog_link_is_down(struct ixgbevf_adapter
*adapter
)
2882 struct net_device
*netdev
= adapter
->netdev
;
2884 adapter
->link_speed
= 0;
2886 /* only continue if link was up previously */
2887 if (!netif_carrier_ok(netdev
))
2890 dev_info(&adapter
->pdev
->dev
, "NIC Link is Down\n");
2892 netif_carrier_off(netdev
);
2896 * ixgbevf_watchdog_subtask - worker thread to bring link up
2897 * @work: pointer to work_struct containing our data
2899 static void ixgbevf_watchdog_subtask(struct ixgbevf_adapter
*adapter
)
2901 /* if interface is down do nothing */
2902 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
) ||
2903 test_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
2906 ixgbevf_watchdog_update_link(adapter
);
2908 if (adapter
->link_up
)
2909 ixgbevf_watchdog_link_is_up(adapter
);
2911 ixgbevf_watchdog_link_is_down(adapter
);
2913 ixgbevf_update_stats(adapter
);
2917 * ixgbevf_service_task - manages and runs subtasks
2918 * @work: pointer to work_struct containing our data
2920 static void ixgbevf_service_task(struct work_struct
*work
)
2922 struct ixgbevf_adapter
*adapter
= container_of(work
,
2923 struct ixgbevf_adapter
,
2925 struct ixgbe_hw
*hw
= &adapter
->hw
;
2927 if (IXGBE_REMOVED(hw
->hw_addr
)) {
2928 if (!test_bit(__IXGBEVF_DOWN
, &adapter
->state
)) {
2930 ixgbevf_down(adapter
);
2936 ixgbevf_queue_reset_subtask(adapter
);
2937 ixgbevf_reset_subtask(adapter
);
2938 ixgbevf_watchdog_subtask(adapter
);
2939 ixgbevf_check_hang_subtask(adapter
);
2941 ixgbevf_service_event_complete(adapter
);
2945 * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2946 * @tx_ring: Tx descriptor ring for a specific queue
2948 * Free all transmit software resources
2950 void ixgbevf_free_tx_resources(struct ixgbevf_ring
*tx_ring
)
2952 ixgbevf_clean_tx_ring(tx_ring
);
2954 vfree(tx_ring
->tx_buffer_info
);
2955 tx_ring
->tx_buffer_info
= NULL
;
2957 /* if not set, then don't free */
2961 dma_free_coherent(tx_ring
->dev
, tx_ring
->size
, tx_ring
->desc
,
2964 tx_ring
->desc
= NULL
;
2968 * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2969 * @adapter: board private structure
2971 * Free all transmit software resources
2973 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter
*adapter
)
2977 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2978 if (adapter
->tx_ring
[i
]->desc
)
2979 ixgbevf_free_tx_resources(adapter
->tx_ring
[i
]);
2983 * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2984 * @tx_ring: Tx descriptor ring (for a specific queue) to setup
2986 * Return 0 on success, negative on failure
2988 int ixgbevf_setup_tx_resources(struct ixgbevf_ring
*tx_ring
)
2992 size
= sizeof(struct ixgbevf_tx_buffer
) * tx_ring
->count
;
2993 tx_ring
->tx_buffer_info
= vzalloc(size
);
2994 if (!tx_ring
->tx_buffer_info
)
2997 /* round up to nearest 4K */
2998 tx_ring
->size
= tx_ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
2999 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
3001 tx_ring
->desc
= dma_alloc_coherent(tx_ring
->dev
, tx_ring
->size
,
3002 &tx_ring
->dma
, GFP_KERNEL
);
3009 vfree(tx_ring
->tx_buffer_info
);
3010 tx_ring
->tx_buffer_info
= NULL
;
3011 hw_dbg(&adapter
->hw
, "Unable to allocate memory for the transmit descriptor ring\n");
3016 * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
3017 * @adapter: board private structure
3019 * If this function returns with an error, then it's possible one or
3020 * more of the rings is populated (while the rest are not). It is the
3021 * callers duty to clean those orphaned rings.
3023 * Return 0 on success, negative on failure
3025 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter
*adapter
)
3029 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3030 err
= ixgbevf_setup_tx_resources(adapter
->tx_ring
[i
]);
3033 hw_dbg(&adapter
->hw
, "Allocation for Tx Queue %u failed\n", i
);
3041 * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
3042 * @rx_ring: Rx descriptor ring (for a specific queue) to setup
3044 * Returns 0 on success, negative on failure
3046 int ixgbevf_setup_rx_resources(struct ixgbevf_ring
*rx_ring
)
3050 size
= sizeof(struct ixgbevf_rx_buffer
) * rx_ring
->count
;
3051 rx_ring
->rx_buffer_info
= vzalloc(size
);
3052 if (!rx_ring
->rx_buffer_info
)
3055 /* Round up to nearest 4K */
3056 rx_ring
->size
= rx_ring
->count
* sizeof(union ixgbe_adv_rx_desc
);
3057 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
3059 rx_ring
->desc
= dma_alloc_coherent(rx_ring
->dev
, rx_ring
->size
,
3060 &rx_ring
->dma
, GFP_KERNEL
);
3067 vfree(rx_ring
->rx_buffer_info
);
3068 rx_ring
->rx_buffer_info
= NULL
;
3069 dev_err(rx_ring
->dev
, "Unable to allocate memory for the Rx descriptor ring\n");
3074 * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
3075 * @adapter: board private structure
3077 * If this function returns with an error, then it's possible one or
3078 * more of the rings is populated (while the rest are not). It is the
3079 * callers duty to clean those orphaned rings.
3081 * Return 0 on success, negative on failure
3083 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter
*adapter
)
3087 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3088 err
= ixgbevf_setup_rx_resources(adapter
->rx_ring
[i
]);
3091 hw_dbg(&adapter
->hw
, "Allocation for Rx Queue %u failed\n", i
);
3098 * ixgbevf_free_rx_resources - Free Rx Resources
3099 * @rx_ring: ring to clean the resources from
3101 * Free all receive software resources
3103 void ixgbevf_free_rx_resources(struct ixgbevf_ring
*rx_ring
)
3105 ixgbevf_clean_rx_ring(rx_ring
);
3107 vfree(rx_ring
->rx_buffer_info
);
3108 rx_ring
->rx_buffer_info
= NULL
;
3110 dma_free_coherent(rx_ring
->dev
, rx_ring
->size
, rx_ring
->desc
,
3113 rx_ring
->desc
= NULL
;
3117 * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
3118 * @adapter: board private structure
3120 * Free all receive software resources
3122 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter
*adapter
)
3126 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3127 if (adapter
->rx_ring
[i
]->desc
)
3128 ixgbevf_free_rx_resources(adapter
->rx_ring
[i
]);
3132 * ixgbevf_open - Called when a network interface is made active
3133 * @netdev: network interface device structure
3135 * Returns 0 on success, negative value on failure
3137 * The open entry point is called when a network interface is made
3138 * active by the system (IFF_UP). At this point all resources needed
3139 * for transmit and receive operations are allocated, the interrupt
3140 * handler is registered with the OS, the watchdog timer is started,
3141 * and the stack is notified that the interface is ready.
3143 int ixgbevf_open(struct net_device
*netdev
)
3145 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3146 struct ixgbe_hw
*hw
= &adapter
->hw
;
3149 /* A previous failure to open the device because of a lack of
3150 * available MSIX vector resources may have reset the number
3151 * of msix vectors variable to zero. The only way to recover
3152 * is to unload/reload the driver and hope that the system has
3153 * been able to recover some MSIX vector resources.
3155 if (!adapter
->num_msix_vectors
)
3158 if (hw
->adapter_stopped
) {
3159 ixgbevf_reset(adapter
);
3160 /* if adapter is still stopped then PF isn't up and
3161 * the VF can't start.
3163 if (hw
->adapter_stopped
) {
3164 err
= IXGBE_ERR_MBX
;
3165 pr_err("Unable to start - perhaps the PF Driver isn't up yet\n");
3166 goto err_setup_reset
;
3170 /* disallow open during test */
3171 if (test_bit(__IXGBEVF_TESTING
, &adapter
->state
))
3174 netif_carrier_off(netdev
);
3176 /* allocate transmit descriptors */
3177 err
= ixgbevf_setup_all_tx_resources(adapter
);
3181 /* allocate receive descriptors */
3182 err
= ixgbevf_setup_all_rx_resources(adapter
);
3186 ixgbevf_configure(adapter
);
3188 /* Map the Tx/Rx rings to the vectors we were allotted.
3189 * if request_irq will be called in this function map_rings
3190 * must be called *before* up_complete
3192 ixgbevf_map_rings_to_vectors(adapter
);
3194 err
= ixgbevf_request_irq(adapter
);
3198 ixgbevf_up_complete(adapter
);
3203 ixgbevf_down(adapter
);
3205 ixgbevf_free_all_rx_resources(adapter
);
3207 ixgbevf_free_all_tx_resources(adapter
);
3208 ixgbevf_reset(adapter
);
3216 * ixgbevf_close - Disables a network interface
3217 * @netdev: network interface device structure
3219 * Returns 0, this is not allowed to fail
3221 * The close entry point is called when an interface is de-activated
3222 * by the OS. The hardware is still under the drivers control, but
3223 * needs to be disabled. A global MAC reset is issued to stop the
3224 * hardware, and all transmit and receive resources are freed.
3226 int ixgbevf_close(struct net_device
*netdev
)
3228 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3230 ixgbevf_down(adapter
);
3231 ixgbevf_free_irq(adapter
);
3233 ixgbevf_free_all_tx_resources(adapter
);
3234 ixgbevf_free_all_rx_resources(adapter
);
3239 static void ixgbevf_queue_reset_subtask(struct ixgbevf_adapter
*adapter
)
3241 struct net_device
*dev
= adapter
->netdev
;
3243 if (!test_and_clear_bit(__IXGBEVF_QUEUE_RESET_REQUESTED
,
3247 /* if interface is down do nothing */
3248 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
) ||
3249 test_bit(__IXGBEVF_RESETTING
, &adapter
->state
))
3252 /* Hardware has to reinitialize queues and interrupts to
3253 * match packet buffer alignment. Unfortunately, the
3254 * hardware is not flexible enough to do this dynamically.
3256 if (netif_running(dev
))
3259 ixgbevf_clear_interrupt_scheme(adapter
);
3260 ixgbevf_init_interrupt_scheme(adapter
);
3262 if (netif_running(dev
))
3266 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring
*tx_ring
,
3267 u32 vlan_macip_lens
, u32 type_tucmd
,
3270 struct ixgbe_adv_tx_context_desc
*context_desc
;
3271 u16 i
= tx_ring
->next_to_use
;
3273 context_desc
= IXGBEVF_TX_CTXTDESC(tx_ring
, i
);
3276 tx_ring
->next_to_use
= (i
< tx_ring
->count
) ? i
: 0;
3278 /* set bits to identify this as an advanced context descriptor */
3279 type_tucmd
|= IXGBE_TXD_CMD_DEXT
| IXGBE_ADVTXD_DTYP_CTXT
;
3281 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
3282 context_desc
->seqnum_seed
= 0;
3283 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd
);
3284 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
3287 static int ixgbevf_tso(struct ixgbevf_ring
*tx_ring
,
3288 struct ixgbevf_tx_buffer
*first
,
3291 u32 vlan_macip_lens
, type_tucmd
, mss_l4len_idx
;
3292 struct sk_buff
*skb
= first
->skb
;
3302 u32 paylen
, l4_offset
;
3305 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
3308 if (!skb_is_gso(skb
))
3311 err
= skb_cow_head(skb
, 0);
3315 ip
.hdr
= skb_network_header(skb
);
3316 l4
.hdr
= skb_checksum_start(skb
);
3318 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3319 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
3321 /* initialize outer IP header fields */
3322 if (ip
.v4
->version
== 4) {
3323 /* IP header will have to cancel out any data that
3324 * is not a part of the outer IP header
3326 ip
.v4
->check
= csum_fold(csum_add(lco_csum(skb
),
3327 csum_unfold(l4
.tcp
->check
)));
3328 type_tucmd
|= IXGBE_ADVTXD_TUCMD_IPV4
;
3331 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
3332 IXGBE_TX_FLAGS_CSUM
|
3333 IXGBE_TX_FLAGS_IPV4
;
3335 ip
.v6
->payload_len
= 0;
3336 first
->tx_flags
|= IXGBE_TX_FLAGS_TSO
|
3337 IXGBE_TX_FLAGS_CSUM
;
3340 /* determine offset of inner transport header */
3341 l4_offset
= l4
.hdr
- skb
->data
;
3343 /* compute length of segmentation header */
3344 *hdr_len
= (l4
.tcp
->doff
* 4) + l4_offset
;
3346 /* remove payload length from inner checksum */
3347 paylen
= skb
->len
- l4_offset
;
3348 csum_replace_by_diff(&l4
.tcp
->check
, htonl(paylen
));
3350 /* update gso size and bytecount with header size */
3351 first
->gso_segs
= skb_shinfo(skb
)->gso_segs
;
3352 first
->bytecount
+= (first
->gso_segs
- 1) * *hdr_len
;
3354 /* mss_l4len_id: use 1 as index for TSO */
3355 mss_l4len_idx
= (*hdr_len
- l4_offset
) << IXGBE_ADVTXD_L4LEN_SHIFT
;
3356 mss_l4len_idx
|= skb_shinfo(skb
)->gso_size
<< IXGBE_ADVTXD_MSS_SHIFT
;
3357 mss_l4len_idx
|= (1u << IXGBE_ADVTXD_IDX_SHIFT
);
3359 /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
3360 vlan_macip_lens
= l4
.hdr
- ip
.hdr
;
3361 vlan_macip_lens
|= (ip
.hdr
- skb
->data
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
3362 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
3364 ixgbevf_tx_ctxtdesc(tx_ring
, vlan_macip_lens
,
3365 type_tucmd
, mss_l4len_idx
);
3370 static inline bool ixgbevf_ipv6_csum_is_sctp(struct sk_buff
*skb
)
3372 unsigned int offset
= 0;
3374 ipv6_find_hdr(skb
, &offset
, IPPROTO_SCTP
, NULL
, NULL
);
3376 return offset
== skb_checksum_start_offset(skb
);
3379 static void ixgbevf_tx_csum(struct ixgbevf_ring
*tx_ring
,
3380 struct ixgbevf_tx_buffer
*first
)
3382 struct sk_buff
*skb
= first
->skb
;
3383 u32 vlan_macip_lens
= 0;
3386 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
3389 switch (skb
->csum_offset
) {
3390 case offsetof(struct tcphdr
, check
):
3391 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
3393 case offsetof(struct udphdr
, check
):
3395 case offsetof(struct sctphdr
, checksum
):
3396 /* validate that this is actually an SCTP request */
3397 if (((first
->protocol
== htons(ETH_P_IP
)) &&
3398 (ip_hdr(skb
)->protocol
== IPPROTO_SCTP
)) ||
3399 ((first
->protocol
== htons(ETH_P_IPV6
)) &&
3400 ixgbevf_ipv6_csum_is_sctp(skb
))) {
3401 type_tucmd
= IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
3406 skb_checksum_help(skb
);
3409 /* update TX checksum flag */
3410 first
->tx_flags
|= IXGBE_TX_FLAGS_CSUM
;
3411 vlan_macip_lens
= skb_checksum_start_offset(skb
) -
3412 skb_network_offset(skb
);
3414 /* vlan_macip_lens: MACLEN, VLAN tag */
3415 vlan_macip_lens
|= skb_network_offset(skb
) << IXGBE_ADVTXD_MACLEN_SHIFT
;
3416 vlan_macip_lens
|= first
->tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
;
3418 ixgbevf_tx_ctxtdesc(tx_ring
, vlan_macip_lens
, type_tucmd
, 0);
3421 static __le32
ixgbevf_tx_cmd_type(u32 tx_flags
)
3423 /* set type for advanced descriptor with frame checksum insertion */
3424 __le32 cmd_type
= cpu_to_le32(IXGBE_ADVTXD_DTYP_DATA
|
3425 IXGBE_ADVTXD_DCMD_IFCS
|
3426 IXGBE_ADVTXD_DCMD_DEXT
);
3428 /* set HW VLAN bit if VLAN is present */
3429 if (tx_flags
& IXGBE_TX_FLAGS_VLAN
)
3430 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_DCMD_VLE
);
3432 /* set segmentation enable bits for TSO/FSO */
3433 if (tx_flags
& IXGBE_TX_FLAGS_TSO
)
3434 cmd_type
|= cpu_to_le32(IXGBE_ADVTXD_DCMD_TSE
);
3439 static void ixgbevf_tx_olinfo_status(union ixgbe_adv_tx_desc
*tx_desc
,
3440 u32 tx_flags
, unsigned int paylen
)
3442 __le32 olinfo_status
= cpu_to_le32(paylen
<< IXGBE_ADVTXD_PAYLEN_SHIFT
);
3444 /* enable L4 checksum for TSO and TX checksum offload */
3445 if (tx_flags
& IXGBE_TX_FLAGS_CSUM
)
3446 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_POPTS_TXSM
);
3448 /* enble IPv4 checksum for TSO */
3449 if (tx_flags
& IXGBE_TX_FLAGS_IPV4
)
3450 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_POPTS_IXSM
);
3452 /* use index 1 context for TSO/FSO/FCOE */
3453 if (tx_flags
& IXGBE_TX_FLAGS_TSO
)
3454 olinfo_status
|= cpu_to_le32(1u << IXGBE_ADVTXD_IDX_SHIFT
);
3456 /* Check Context must be set if Tx switch is enabled, which it
3457 * always is for case where virtual functions are running
3459 olinfo_status
|= cpu_to_le32(IXGBE_ADVTXD_CC
);
3461 tx_desc
->read
.olinfo_status
= olinfo_status
;
3464 static void ixgbevf_tx_map(struct ixgbevf_ring
*tx_ring
,
3465 struct ixgbevf_tx_buffer
*first
,
3469 struct sk_buff
*skb
= first
->skb
;
3470 struct ixgbevf_tx_buffer
*tx_buffer
;
3471 union ixgbe_adv_tx_desc
*tx_desc
;
3472 struct skb_frag_struct
*frag
= &skb_shinfo(skb
)->frags
[0];
3473 unsigned int data_len
= skb
->data_len
;
3474 unsigned int size
= skb_headlen(skb
);
3475 unsigned int paylen
= skb
->len
- hdr_len
;
3476 u32 tx_flags
= first
->tx_flags
;
3478 u16 i
= tx_ring
->next_to_use
;
3480 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, i
);
3482 ixgbevf_tx_olinfo_status(tx_desc
, tx_flags
, paylen
);
3483 cmd_type
= ixgbevf_tx_cmd_type(tx_flags
);
3485 dma
= dma_map_single(tx_ring
->dev
, skb
->data
, size
, DMA_TO_DEVICE
);
3486 if (dma_mapping_error(tx_ring
->dev
, dma
))
3489 /* record length, and DMA address */
3490 dma_unmap_len_set(first
, len
, size
);
3491 dma_unmap_addr_set(first
, dma
, dma
);
3493 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
3496 while (unlikely(size
> IXGBE_MAX_DATA_PER_TXD
)) {
3497 tx_desc
->read
.cmd_type_len
=
3498 cmd_type
| cpu_to_le32(IXGBE_MAX_DATA_PER_TXD
);
3502 if (i
== tx_ring
->count
) {
3503 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, 0);
3507 dma
+= IXGBE_MAX_DATA_PER_TXD
;
3508 size
-= IXGBE_MAX_DATA_PER_TXD
;
3510 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
3511 tx_desc
->read
.olinfo_status
= 0;
3514 if (likely(!data_len
))
3517 tx_desc
->read
.cmd_type_len
= cmd_type
| cpu_to_le32(size
);
3521 if (i
== tx_ring
->count
) {
3522 tx_desc
= IXGBEVF_TX_DESC(tx_ring
, 0);
3526 size
= skb_frag_size(frag
);
3529 dma
= skb_frag_dma_map(tx_ring
->dev
, frag
, 0, size
,
3531 if (dma_mapping_error(tx_ring
->dev
, dma
))
3534 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
3535 dma_unmap_len_set(tx_buffer
, len
, size
);
3536 dma_unmap_addr_set(tx_buffer
, dma
, dma
);
3538 tx_desc
->read
.buffer_addr
= cpu_to_le64(dma
);
3539 tx_desc
->read
.olinfo_status
= 0;
3544 /* write last descriptor with RS and EOP bits */
3545 cmd_type
|= cpu_to_le32(size
) | cpu_to_le32(IXGBE_TXD_CMD
);
3546 tx_desc
->read
.cmd_type_len
= cmd_type
;
3548 /* set the timestamp */
3549 first
->time_stamp
= jiffies
;
3551 /* Force memory writes to complete before letting h/w know there
3552 * are new descriptors to fetch. (Only applicable for weak-ordered
3553 * memory model archs, such as IA-64).
3555 * We also need this memory barrier (wmb) to make certain all of the
3556 * status bits have been updated before next_to_watch is written.
3560 /* set next_to_watch value indicating a packet is present */
3561 first
->next_to_watch
= tx_desc
;
3564 if (i
== tx_ring
->count
)
3567 tx_ring
->next_to_use
= i
;
3569 /* notify HW of packet */
3570 ixgbevf_write_tail(tx_ring
, i
);
3574 dev_err(tx_ring
->dev
, "TX DMA map failed\n");
3576 /* clear dma mappings for failed tx_buffer_info map */
3578 tx_buffer
= &tx_ring
->tx_buffer_info
[i
];
3579 ixgbevf_unmap_and_free_tx_resource(tx_ring
, tx_buffer
);
3580 if (tx_buffer
== first
)
3587 tx_ring
->next_to_use
= i
;
3590 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring
*tx_ring
, int size
)
3592 netif_stop_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
3593 /* Herbert's original patch had:
3594 * smp_mb__after_netif_stop_queue();
3595 * but since that doesn't exist yet, just open code it.
3599 /* We need to check again in a case another CPU has just
3600 * made room available.
3602 if (likely(ixgbevf_desc_unused(tx_ring
) < size
))
3605 /* A reprieve! - use start_queue because it doesn't call schedule */
3606 netif_start_subqueue(tx_ring
->netdev
, tx_ring
->queue_index
);
3607 ++tx_ring
->tx_stats
.restart_queue
;
3612 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring
*tx_ring
, int size
)
3614 if (likely(ixgbevf_desc_unused(tx_ring
) >= size
))
3616 return __ixgbevf_maybe_stop_tx(tx_ring
, size
);
3619 static int ixgbevf_xmit_frame(struct sk_buff
*skb
, struct net_device
*netdev
)
3621 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3622 struct ixgbevf_tx_buffer
*first
;
3623 struct ixgbevf_ring
*tx_ring
;
3626 u16 count
= TXD_USE_COUNT(skb_headlen(skb
));
3627 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3631 u8
*dst_mac
= skb_header_pointer(skb
, 0, 0, NULL
);
3633 if (!dst_mac
|| is_link_local_ether_addr(dst_mac
)) {
3634 dev_kfree_skb_any(skb
);
3635 return NETDEV_TX_OK
;
3638 tx_ring
= adapter
->tx_ring
[skb
->queue_mapping
];
3640 /* need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3641 * + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3642 * + 2 desc gap to keep tail from touching head,
3643 * + 1 desc for context descriptor,
3644 * otherwise try next time
3646 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3647 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
3648 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
3650 count
+= skb_shinfo(skb
)->nr_frags
;
3652 if (ixgbevf_maybe_stop_tx(tx_ring
, count
+ 3)) {
3653 tx_ring
->tx_stats
.tx_busy
++;
3654 return NETDEV_TX_BUSY
;
3657 /* record the location of the first descriptor for this packet */
3658 first
= &tx_ring
->tx_buffer_info
[tx_ring
->next_to_use
];
3660 first
->bytecount
= skb
->len
;
3661 first
->gso_segs
= 1;
3663 if (skb_vlan_tag_present(skb
)) {
3664 tx_flags
|= skb_vlan_tag_get(skb
);
3665 tx_flags
<<= IXGBE_TX_FLAGS_VLAN_SHIFT
;
3666 tx_flags
|= IXGBE_TX_FLAGS_VLAN
;
3669 /* record initial flags and protocol */
3670 first
->tx_flags
= tx_flags
;
3671 first
->protocol
= vlan_get_protocol(skb
);
3673 tso
= ixgbevf_tso(tx_ring
, first
, &hdr_len
);
3677 ixgbevf_tx_csum(tx_ring
, first
);
3679 ixgbevf_tx_map(tx_ring
, first
, hdr_len
);
3681 ixgbevf_maybe_stop_tx(tx_ring
, DESC_NEEDED
);
3683 return NETDEV_TX_OK
;
3686 dev_kfree_skb_any(first
->skb
);
3689 return NETDEV_TX_OK
;
3693 * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3694 * @netdev: network interface device structure
3695 * @p: pointer to an address structure
3697 * Returns 0 on success, negative on failure
3699 static int ixgbevf_set_mac(struct net_device
*netdev
, void *p
)
3701 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3702 struct ixgbe_hw
*hw
= &adapter
->hw
;
3703 struct sockaddr
*addr
= p
;
3706 if (!is_valid_ether_addr(addr
->sa_data
))
3707 return -EADDRNOTAVAIL
;
3709 spin_lock_bh(&adapter
->mbx_lock
);
3711 err
= hw
->mac
.ops
.set_rar(hw
, 0, addr
->sa_data
, 0);
3713 spin_unlock_bh(&adapter
->mbx_lock
);
3718 ether_addr_copy(hw
->mac
.addr
, addr
->sa_data
);
3719 ether_addr_copy(netdev
->dev_addr
, addr
->sa_data
);
3725 * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3726 * @netdev: network interface device structure
3727 * @new_mtu: new value for maximum frame size
3729 * Returns 0 on success, negative on failure
3731 static int ixgbevf_change_mtu(struct net_device
*netdev
, int new_mtu
)
3733 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3734 struct ixgbe_hw
*hw
= &adapter
->hw
;
3735 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
3736 int max_possible_frame
= MAXIMUM_ETHERNET_VLAN_SIZE
;
3738 switch (adapter
->hw
.api_version
) {
3739 case ixgbe_mbox_api_11
:
3740 case ixgbe_mbox_api_12
:
3741 max_possible_frame
= IXGBE_MAX_JUMBO_FRAME_SIZE
;
3744 if (adapter
->hw
.mac
.type
!= ixgbe_mac_82599_vf
)
3745 max_possible_frame
= IXGBE_MAX_JUMBO_FRAME_SIZE
;
3749 /* MTU < 68 is an error and causes problems on some kernels */
3750 if ((new_mtu
< 68) || (max_frame
> max_possible_frame
))
3753 hw_dbg(hw
, "changing MTU from %d to %d\n",
3754 netdev
->mtu
, new_mtu
);
3755 /* must set new MTU before calling down or up */
3756 netdev
->mtu
= new_mtu
;
3758 /* notify the PF of our intent to use this size of frame */
3759 hw
->mac
.ops
.set_rlpml(hw
, max_frame
);
3764 #ifdef CONFIG_NET_POLL_CONTROLLER
3765 /* Polling 'interrupt' - used by things like netconsole to send skbs
3766 * without having to re-enable interrupts. It's not called while
3767 * the interrupt routine is executing.
3769 static void ixgbevf_netpoll(struct net_device
*netdev
)
3771 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3774 /* if interface is down do nothing */
3775 if (test_bit(__IXGBEVF_DOWN
, &adapter
->state
))
3777 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3778 ixgbevf_msix_clean_rings(0, adapter
->q_vector
[i
]);
3780 #endif /* CONFIG_NET_POLL_CONTROLLER */
3782 static int ixgbevf_suspend(struct pci_dev
*pdev
, pm_message_t state
)
3784 struct net_device
*netdev
= pci_get_drvdata(pdev
);
3785 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3790 netif_device_detach(netdev
);
3792 if (netif_running(netdev
)) {
3794 ixgbevf_down(adapter
);
3795 ixgbevf_free_irq(adapter
);
3796 ixgbevf_free_all_tx_resources(adapter
);
3797 ixgbevf_free_all_rx_resources(adapter
);
3801 ixgbevf_clear_interrupt_scheme(adapter
);
3804 retval
= pci_save_state(pdev
);
3809 if (!test_and_set_bit(__IXGBEVF_DISABLED
, &adapter
->state
))
3810 pci_disable_device(pdev
);
3816 static int ixgbevf_resume(struct pci_dev
*pdev
)
3818 struct net_device
*netdev
= pci_get_drvdata(pdev
);
3819 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3822 pci_restore_state(pdev
);
3823 /* pci_restore_state clears dev->state_saved so call
3824 * pci_save_state to restore it.
3826 pci_save_state(pdev
);
3828 err
= pci_enable_device_mem(pdev
);
3830 dev_err(&pdev
->dev
, "Cannot enable PCI device from suspend\n");
3833 smp_mb__before_atomic();
3834 clear_bit(__IXGBEVF_DISABLED
, &adapter
->state
);
3835 pci_set_master(pdev
);
3837 ixgbevf_reset(adapter
);
3840 err
= ixgbevf_init_interrupt_scheme(adapter
);
3843 dev_err(&pdev
->dev
, "Cannot initialize interrupts\n");
3847 if (netif_running(netdev
)) {
3848 err
= ixgbevf_open(netdev
);
3853 netif_device_attach(netdev
);
3858 #endif /* CONFIG_PM */
3859 static void ixgbevf_shutdown(struct pci_dev
*pdev
)
3861 ixgbevf_suspend(pdev
, PMSG_SUSPEND
);
3864 static struct rtnl_link_stats64
*ixgbevf_get_stats(struct net_device
*netdev
,
3865 struct rtnl_link_stats64
*stats
)
3867 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
3870 const struct ixgbevf_ring
*ring
;
3873 ixgbevf_update_stats(adapter
);
3875 stats
->multicast
= adapter
->stats
.vfmprc
- adapter
->stats
.base_vfmprc
;
3877 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3878 ring
= adapter
->rx_ring
[i
];
3880 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
3881 bytes
= ring
->stats
.bytes
;
3882 packets
= ring
->stats
.packets
;
3883 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
3884 stats
->rx_bytes
+= bytes
;
3885 stats
->rx_packets
+= packets
;
3888 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3889 ring
= adapter
->tx_ring
[i
];
3891 start
= u64_stats_fetch_begin_irq(&ring
->syncp
);
3892 bytes
= ring
->stats
.bytes
;
3893 packets
= ring
->stats
.packets
;
3894 } while (u64_stats_fetch_retry_irq(&ring
->syncp
, start
));
3895 stats
->tx_bytes
+= bytes
;
3896 stats
->tx_packets
+= packets
;
3902 #define IXGBEVF_MAX_MAC_HDR_LEN 127
3903 #define IXGBEVF_MAX_NETWORK_HDR_LEN 511
3905 static netdev_features_t
3906 ixgbevf_features_check(struct sk_buff
*skb
, struct net_device
*dev
,
3907 netdev_features_t features
)
3909 unsigned int network_hdr_len
, mac_hdr_len
;
3911 /* Make certain the headers can be described by a context descriptor */
3912 mac_hdr_len
= skb_network_header(skb
) - skb
->data
;
3913 if (unlikely(mac_hdr_len
> IXGBEVF_MAX_MAC_HDR_LEN
))
3914 return features
& ~(NETIF_F_HW_CSUM
|
3916 NETIF_F_HW_VLAN_CTAG_TX
|
3920 network_hdr_len
= skb_checksum_start(skb
) - skb_network_header(skb
);
3921 if (unlikely(network_hdr_len
> IXGBEVF_MAX_NETWORK_HDR_LEN
))
3922 return features
& ~(NETIF_F_HW_CSUM
|
3927 /* We can only support IPV4 TSO in tunnels if we can mangle the
3928 * inner IP ID field, so strip TSO if MANGLEID is not supported.
3930 if (skb
->encapsulation
&& !(features
& NETIF_F_TSO_MANGLEID
))
3931 features
&= ~NETIF_F_TSO
;
3936 static const struct net_device_ops ixgbevf_netdev_ops
= {
3937 .ndo_open
= ixgbevf_open
,
3938 .ndo_stop
= ixgbevf_close
,
3939 .ndo_start_xmit
= ixgbevf_xmit_frame
,
3940 .ndo_set_rx_mode
= ixgbevf_set_rx_mode
,
3941 .ndo_get_stats64
= ixgbevf_get_stats
,
3942 .ndo_validate_addr
= eth_validate_addr
,
3943 .ndo_set_mac_address
= ixgbevf_set_mac
,
3944 .ndo_change_mtu
= ixgbevf_change_mtu
,
3945 .ndo_tx_timeout
= ixgbevf_tx_timeout
,
3946 .ndo_vlan_rx_add_vid
= ixgbevf_vlan_rx_add_vid
,
3947 .ndo_vlan_rx_kill_vid
= ixgbevf_vlan_rx_kill_vid
,
3948 #ifdef CONFIG_NET_RX_BUSY_POLL
3949 .ndo_busy_poll
= ixgbevf_busy_poll_recv
,
3951 #ifdef CONFIG_NET_POLL_CONTROLLER
3952 .ndo_poll_controller
= ixgbevf_netpoll
,
3954 .ndo_features_check
= ixgbevf_features_check
,
3957 static void ixgbevf_assign_netdev_ops(struct net_device
*dev
)
3959 dev
->netdev_ops
= &ixgbevf_netdev_ops
;
3960 ixgbevf_set_ethtool_ops(dev
);
3961 dev
->watchdog_timeo
= 5 * HZ
;
3965 * ixgbevf_probe - Device Initialization Routine
3966 * @pdev: PCI device information struct
3967 * @ent: entry in ixgbevf_pci_tbl
3969 * Returns 0 on success, negative on failure
3971 * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3972 * The OS initialization, configuring of the adapter private structure,
3973 * and a hardware reset occur.
3975 static int ixgbevf_probe(struct pci_dev
*pdev
, const struct pci_device_id
*ent
)
3977 struct net_device
*netdev
;
3978 struct ixgbevf_adapter
*adapter
= NULL
;
3979 struct ixgbe_hw
*hw
= NULL
;
3980 const struct ixgbevf_info
*ii
= ixgbevf_info_tbl
[ent
->driver_data
];
3981 int err
, pci_using_dac
;
3982 bool disable_dev
= false;
3984 err
= pci_enable_device(pdev
);
3988 if (!dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(64))) {
3991 err
= dma_set_mask_and_coherent(&pdev
->dev
, DMA_BIT_MASK(32));
3993 dev_err(&pdev
->dev
, "No usable DMA configuration, aborting\n");
3999 err
= pci_request_regions(pdev
, ixgbevf_driver_name
);
4001 dev_err(&pdev
->dev
, "pci_request_regions failed 0x%x\n", err
);
4005 pci_set_master(pdev
);
4007 netdev
= alloc_etherdev_mq(sizeof(struct ixgbevf_adapter
),
4011 goto err_alloc_etherdev
;
4014 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
4016 adapter
= netdev_priv(netdev
);
4018 adapter
->netdev
= netdev
;
4019 adapter
->pdev
= pdev
;
4022 adapter
->msg_enable
= netif_msg_init(debug
, DEFAULT_MSG_ENABLE
);
4024 /* call save state here in standalone driver because it relies on
4025 * adapter struct to exist, and needs to call netdev_priv
4027 pci_save_state(pdev
);
4029 hw
->hw_addr
= ioremap(pci_resource_start(pdev
, 0),
4030 pci_resource_len(pdev
, 0));
4031 adapter
->io_addr
= hw
->hw_addr
;
4037 ixgbevf_assign_netdev_ops(netdev
);
4040 memcpy(&hw
->mac
.ops
, ii
->mac_ops
, sizeof(hw
->mac
.ops
));
4041 hw
->mac
.type
= ii
->mac
;
4043 memcpy(&hw
->mbx
.ops
, &ixgbevf_mbx_ops
,
4044 sizeof(struct ixgbe_mbx_operations
));
4046 /* setup the private structure */
4047 err
= ixgbevf_sw_init(adapter
);
4051 /* The HW MAC address was set and/or determined in sw_init */
4052 if (!is_valid_ether_addr(netdev
->dev_addr
)) {
4053 pr_err("invalid MAC address\n");
4058 netdev
->hw_features
= NETIF_F_SG
|
4065 #define IXGBEVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
4066 NETIF_F_GSO_GRE_CSUM | \
4067 NETIF_F_GSO_IPXIP4 | \
4068 NETIF_F_GSO_IPXIP6 | \
4069 NETIF_F_GSO_UDP_TUNNEL | \
4070 NETIF_F_GSO_UDP_TUNNEL_CSUM)
4072 netdev
->gso_partial_features
= IXGBEVF_GSO_PARTIAL_FEATURES
;
4073 netdev
->hw_features
|= NETIF_F_GSO_PARTIAL
|
4074 IXGBEVF_GSO_PARTIAL_FEATURES
;
4076 netdev
->features
= netdev
->hw_features
;
4079 netdev
->features
|= NETIF_F_HIGHDMA
;
4081 netdev
->vlan_features
|= netdev
->features
| NETIF_F_TSO_MANGLEID
;
4082 netdev
->mpls_features
|= NETIF_F_HW_CSUM
;
4083 netdev
->hw_enc_features
|= netdev
->vlan_features
;
4085 /* set this bit last since it cannot be part of vlan_features */
4086 netdev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
|
4087 NETIF_F_HW_VLAN_CTAG_RX
|
4088 NETIF_F_HW_VLAN_CTAG_TX
;
4090 netdev
->priv_flags
|= IFF_UNICAST_FLT
;
4092 if (IXGBE_REMOVED(hw
->hw_addr
)) {
4097 setup_timer(&adapter
->service_timer
, &ixgbevf_service_timer
,
4098 (unsigned long)adapter
);
4100 INIT_WORK(&adapter
->service_task
, ixgbevf_service_task
);
4101 set_bit(__IXGBEVF_SERVICE_INITED
, &adapter
->state
);
4102 clear_bit(__IXGBEVF_SERVICE_SCHED
, &adapter
->state
);
4104 err
= ixgbevf_init_interrupt_scheme(adapter
);
4108 strcpy(netdev
->name
, "eth%d");
4110 err
= register_netdev(netdev
);
4114 pci_set_drvdata(pdev
, netdev
);
4115 netif_carrier_off(netdev
);
4117 ixgbevf_init_last_counter_stats(adapter
);
4119 /* print the VF info */
4120 dev_info(&pdev
->dev
, "%pM\n", netdev
->dev_addr
);
4121 dev_info(&pdev
->dev
, "MAC: %d\n", hw
->mac
.type
);
4123 switch (hw
->mac
.type
) {
4124 case ixgbe_mac_X550_vf
:
4125 dev_info(&pdev
->dev
, "Intel(R) X550 Virtual Function\n");
4127 case ixgbe_mac_X540_vf
:
4128 dev_info(&pdev
->dev
, "Intel(R) X540 Virtual Function\n");
4130 case ixgbe_mac_82599_vf
:
4132 dev_info(&pdev
->dev
, "Intel(R) 82599 Virtual Function\n");
4139 ixgbevf_clear_interrupt_scheme(adapter
);
4141 ixgbevf_reset_interrupt_capability(adapter
);
4142 iounmap(adapter
->io_addr
);
4144 disable_dev
= !test_and_set_bit(__IXGBEVF_DISABLED
, &adapter
->state
);
4145 free_netdev(netdev
);
4147 pci_release_regions(pdev
);
4150 if (!adapter
|| disable_dev
)
4151 pci_disable_device(pdev
);
4156 * ixgbevf_remove - Device Removal Routine
4157 * @pdev: PCI device information struct
4159 * ixgbevf_remove is called by the PCI subsystem to alert the driver
4160 * that it should release a PCI device. The could be caused by a
4161 * Hot-Plug event, or because the driver is going to be removed from
4164 static void ixgbevf_remove(struct pci_dev
*pdev
)
4166 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4167 struct ixgbevf_adapter
*adapter
;
4173 adapter
= netdev_priv(netdev
);
4175 set_bit(__IXGBEVF_REMOVING
, &adapter
->state
);
4176 cancel_work_sync(&adapter
->service_task
);
4178 if (netdev
->reg_state
== NETREG_REGISTERED
)
4179 unregister_netdev(netdev
);
4181 ixgbevf_clear_interrupt_scheme(adapter
);
4182 ixgbevf_reset_interrupt_capability(adapter
);
4184 iounmap(adapter
->io_addr
);
4185 pci_release_regions(pdev
);
4187 hw_dbg(&adapter
->hw
, "Remove complete\n");
4189 disable_dev
= !test_and_set_bit(__IXGBEVF_DISABLED
, &adapter
->state
);
4190 free_netdev(netdev
);
4193 pci_disable_device(pdev
);
4197 * ixgbevf_io_error_detected - called when PCI error is detected
4198 * @pdev: Pointer to PCI device
4199 * @state: The current pci connection state
4201 * This function is called after a PCI bus error affecting
4202 * this device has been detected.
4204 static pci_ers_result_t
ixgbevf_io_error_detected(struct pci_dev
*pdev
,
4205 pci_channel_state_t state
)
4207 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4208 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
4210 if (!test_bit(__IXGBEVF_SERVICE_INITED
, &adapter
->state
))
4211 return PCI_ERS_RESULT_DISCONNECT
;
4214 netif_device_detach(netdev
);
4216 if (state
== pci_channel_io_perm_failure
) {
4218 return PCI_ERS_RESULT_DISCONNECT
;
4221 if (netif_running(netdev
))
4222 ixgbevf_down(adapter
);
4224 if (!test_and_set_bit(__IXGBEVF_DISABLED
, &adapter
->state
))
4225 pci_disable_device(pdev
);
4228 /* Request a slot slot reset. */
4229 return PCI_ERS_RESULT_NEED_RESET
;
4233 * ixgbevf_io_slot_reset - called after the pci bus has been reset.
4234 * @pdev: Pointer to PCI device
4236 * Restart the card from scratch, as if from a cold-boot. Implementation
4237 * resembles the first-half of the ixgbevf_resume routine.
4239 static pci_ers_result_t
ixgbevf_io_slot_reset(struct pci_dev
*pdev
)
4241 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4242 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
4244 if (pci_enable_device_mem(pdev
)) {
4246 "Cannot re-enable PCI device after reset.\n");
4247 return PCI_ERS_RESULT_DISCONNECT
;
4250 smp_mb__before_atomic();
4251 clear_bit(__IXGBEVF_DISABLED
, &adapter
->state
);
4252 pci_set_master(pdev
);
4254 ixgbevf_reset(adapter
);
4256 return PCI_ERS_RESULT_RECOVERED
;
4260 * ixgbevf_io_resume - called when traffic can start flowing again.
4261 * @pdev: Pointer to PCI device
4263 * This callback is called when the error recovery driver tells us that
4264 * its OK to resume normal operation. Implementation resembles the
4265 * second-half of the ixgbevf_resume routine.
4267 static void ixgbevf_io_resume(struct pci_dev
*pdev
)
4269 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4270 struct ixgbevf_adapter
*adapter
= netdev_priv(netdev
);
4272 if (netif_running(netdev
))
4273 ixgbevf_up(adapter
);
4275 netif_device_attach(netdev
);
4278 /* PCI Error Recovery (ERS) */
4279 static const struct pci_error_handlers ixgbevf_err_handler
= {
4280 .error_detected
= ixgbevf_io_error_detected
,
4281 .slot_reset
= ixgbevf_io_slot_reset
,
4282 .resume
= ixgbevf_io_resume
,
4285 static struct pci_driver ixgbevf_driver
= {
4286 .name
= ixgbevf_driver_name
,
4287 .id_table
= ixgbevf_pci_tbl
,
4288 .probe
= ixgbevf_probe
,
4289 .remove
= ixgbevf_remove
,
4291 /* Power Management Hooks */
4292 .suspend
= ixgbevf_suspend
,
4293 .resume
= ixgbevf_resume
,
4295 .shutdown
= ixgbevf_shutdown
,
4296 .err_handler
= &ixgbevf_err_handler
4300 * ixgbevf_init_module - Driver Registration Routine
4302 * ixgbevf_init_module is the first routine called when the driver is
4303 * loaded. All it does is register with the PCI subsystem.
4305 static int __init
ixgbevf_init_module(void)
4307 pr_info("%s - version %s\n", ixgbevf_driver_string
,
4308 ixgbevf_driver_version
);
4310 pr_info("%s\n", ixgbevf_copyright
);
4311 ixgbevf_wq
= create_singlethread_workqueue(ixgbevf_driver_name
);
4313 pr_err("%s: Failed to create workqueue\n", ixgbevf_driver_name
);
4317 return pci_register_driver(&ixgbevf_driver
);
4320 module_init(ixgbevf_init_module
);
4323 * ixgbevf_exit_module - Driver Exit Cleanup Routine
4325 * ixgbevf_exit_module is called just before the driver is removed
4328 static void __exit
ixgbevf_exit_module(void)
4330 pci_unregister_driver(&ixgbevf_driver
);
4332 destroy_workqueue(ixgbevf_wq
);
4339 * ixgbevf_get_hw_dev_name - return device name string
4340 * used by hardware layer to print debugging information
4342 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw
*hw
)
4344 struct ixgbevf_adapter
*adapter
= hw
->back
;
4346 return adapter
->netdev
->name
;
4350 module_exit(ixgbevf_exit_module
);
4352 /* ixgbevf_main.c */