1 /*******************************************************************************
3 Intel 10 Gigabit PCI Express Linux driver
4 Copyright(c) 1999 - 2009 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/types.h>
29 #include <linux/module.h>
30 #include <linux/pci.h>
31 #include <linux/netdevice.h>
32 #include <linux/vmalloc.h>
33 #include <linux/string.h>
36 #include <linux/tcp.h>
37 #include <linux/pkt_sched.h>
38 #include <linux/ipv6.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/ethtool.h>
42 #include <linux/if_vlan.h>
43 #include <scsi/fc/fc_fcoe.h>
46 #include "ixgbe_common.h"
48 char ixgbe_driver_name
[] = "ixgbe";
49 static const char ixgbe_driver_string
[] =
50 "Intel(R) 10 Gigabit PCI Express Network Driver";
52 #define DRV_VERSION "2.0.37-k2"
53 const char ixgbe_driver_version
[] = DRV_VERSION
;
54 static char ixgbe_copyright
[] = "Copyright (c) 1999-2009 Intel Corporation.";
56 static const struct ixgbe_info
*ixgbe_info_tbl
[] = {
57 [board_82598
] = &ixgbe_82598_info
,
58 [board_82599
] = &ixgbe_82599_info
,
61 /* ixgbe_pci_tbl - PCI Device ID Table
63 * Wildcard entries (PCI_ANY_ID) should come last
64 * Last entry must be all 0s
66 * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
67 * Class, Class Mask, private data (not used) }
69 static struct pci_device_id ixgbe_pci_tbl
[] = {
70 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598
),
72 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_DUAL_PORT
),
74 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AF_SINGLE_PORT
),
76 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT
),
78 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598AT2
),
80 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_CX4
),
82 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_CX4_DUAL_PORT
),
84 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_DA_DUAL_PORT
),
86 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM
),
88 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_XF_LR
),
90 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598EB_SFP_LOM
),
92 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82598_BX
),
94 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_KX4
),
96 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_XAUI_LOM
),
98 {PCI_VDEVICE(INTEL
, IXGBE_DEV_ID_82599_SFP
),
101 /* required last entry */
104 MODULE_DEVICE_TABLE(pci
, ixgbe_pci_tbl
);
106 #ifdef CONFIG_IXGBE_DCA
107 static int ixgbe_notify_dca(struct notifier_block
*, unsigned long event
,
109 static struct notifier_block dca_notifier
= {
110 .notifier_call
= ixgbe_notify_dca
,
116 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
117 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
118 MODULE_LICENSE("GPL");
119 MODULE_VERSION(DRV_VERSION
);
121 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
123 static void ixgbe_release_hw_control(struct ixgbe_adapter
*adapter
)
127 /* Let firmware take over control of h/w */
128 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
129 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
130 ctrl_ext
& ~IXGBE_CTRL_EXT_DRV_LOAD
);
133 static void ixgbe_get_hw_control(struct ixgbe_adapter
*adapter
)
137 /* Let firmware know the driver has taken over */
138 ctrl_ext
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_CTRL_EXT
);
139 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_CTRL_EXT
,
140 ctrl_ext
| IXGBE_CTRL_EXT_DRV_LOAD
);
144 * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
145 * @adapter: pointer to adapter struct
146 * @direction: 0 for Rx, 1 for Tx, -1 for other causes
147 * @queue: queue to map the corresponding interrupt to
148 * @msix_vector: the vector to map to the corresponding queue
151 static void ixgbe_set_ivar(struct ixgbe_adapter
*adapter
, s8 direction
,
152 u8 queue
, u8 msix_vector
)
155 struct ixgbe_hw
*hw
= &adapter
->hw
;
156 switch (hw
->mac
.type
) {
157 case ixgbe_mac_82598EB
:
158 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
161 index
= (((direction
* 64) + queue
) >> 2) & 0x1F;
162 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(index
));
163 ivar
&= ~(0xFF << (8 * (queue
& 0x3)));
164 ivar
|= (msix_vector
<< (8 * (queue
& 0x3)));
165 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(index
), ivar
);
167 case ixgbe_mac_82599EB
:
168 if (direction
== -1) {
170 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
171 index
= ((queue
& 1) * 8);
172 ivar
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_IVAR_MISC
);
173 ivar
&= ~(0xFF << index
);
174 ivar
|= (msix_vector
<< index
);
175 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_IVAR_MISC
, ivar
);
178 /* tx or rx causes */
179 msix_vector
|= IXGBE_IVAR_ALLOC_VAL
;
180 index
= ((16 * (queue
& 1)) + (8 * direction
));
181 ivar
= IXGBE_READ_REG(hw
, IXGBE_IVAR(queue
>> 1));
182 ivar
&= ~(0xFF << index
);
183 ivar
|= (msix_vector
<< index
);
184 IXGBE_WRITE_REG(hw
, IXGBE_IVAR(queue
>> 1), ivar
);
192 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter
*adapter
,
197 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
198 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
199 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS
, mask
);
201 mask
= (qmask
& 0xFFFFFFFF);
202 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(0), mask
);
203 mask
= (qmask
>> 32);
204 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EICS_EX(1), mask
);
208 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter
*adapter
,
209 struct ixgbe_tx_buffer
212 tx_buffer_info
->dma
= 0;
213 if (tx_buffer_info
->skb
) {
214 skb_dma_unmap(&adapter
->pdev
->dev
, tx_buffer_info
->skb
,
216 dev_kfree_skb_any(tx_buffer_info
->skb
);
217 tx_buffer_info
->skb
= NULL
;
219 tx_buffer_info
->time_stamp
= 0;
220 /* tx_buffer_info must be completely set up in the transmit path */
223 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter
*adapter
,
224 struct ixgbe_ring
*tx_ring
,
227 struct ixgbe_hw
*hw
= &adapter
->hw
;
229 /* Detect a transmit hang in hardware, this serializes the
230 * check with the clearing of time_stamp and movement of eop */
231 adapter
->detect_tx_hung
= false;
232 if (tx_ring
->tx_buffer_info
[eop
].time_stamp
&&
233 time_after(jiffies
, tx_ring
->tx_buffer_info
[eop
].time_stamp
+ HZ
) &&
234 !(IXGBE_READ_REG(&adapter
->hw
, IXGBE_TFCS
) & IXGBE_TFCS_TXOFF
)) {
235 /* detected Tx unit hang */
236 union ixgbe_adv_tx_desc
*tx_desc
;
237 tx_desc
= IXGBE_TX_DESC_ADV(*tx_ring
, eop
);
238 DPRINTK(DRV
, ERR
, "Detected Tx Unit Hang\n"
240 " TDH, TDT <%x>, <%x>\n"
241 " next_to_use <%x>\n"
242 " next_to_clean <%x>\n"
243 "tx_buffer_info[next_to_clean]\n"
244 " time_stamp <%lx>\n"
246 tx_ring
->queue_index
,
247 IXGBE_READ_REG(hw
, tx_ring
->head
),
248 IXGBE_READ_REG(hw
, tx_ring
->tail
),
249 tx_ring
->next_to_use
, eop
,
250 tx_ring
->tx_buffer_info
[eop
].time_stamp
, jiffies
);
257 #define IXGBE_MAX_TXD_PWR 14
258 #define IXGBE_MAX_DATA_PER_TXD (1 << IXGBE_MAX_TXD_PWR)
260 /* Tx Descriptors needed, worst case */
261 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
262 (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
263 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
264 MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
266 static void ixgbe_tx_timeout(struct net_device
*netdev
);
269 * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
270 * @q_vector: structure containing interrupt and ring information
271 * @tx_ring: tx ring to clean
273 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector
*q_vector
,
274 struct ixgbe_ring
*tx_ring
)
276 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
277 struct net_device
*netdev
= adapter
->netdev
;
278 union ixgbe_adv_tx_desc
*tx_desc
, *eop_desc
;
279 struct ixgbe_tx_buffer
*tx_buffer_info
;
280 unsigned int i
, eop
, count
= 0;
281 unsigned int total_bytes
= 0, total_packets
= 0;
283 i
= tx_ring
->next_to_clean
;
284 eop
= tx_ring
->tx_buffer_info
[i
].next_to_watch
;
285 eop_desc
= IXGBE_TX_DESC_ADV(*tx_ring
, eop
);
287 while ((eop_desc
->wb
.status
& cpu_to_le32(IXGBE_TXD_STAT_DD
)) &&
288 (count
< tx_ring
->work_limit
)) {
289 bool cleaned
= false;
290 for ( ; !cleaned
; count
++) {
292 tx_desc
= IXGBE_TX_DESC_ADV(*tx_ring
, i
);
293 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
294 cleaned
= (i
== eop
);
295 skb
= tx_buffer_info
->skb
;
297 if (cleaned
&& skb
) {
298 unsigned int segs
, bytecount
;
299 unsigned int hlen
= skb_headlen(skb
);
301 /* gso_segs is currently only valid for tcp */
302 segs
= skb_shinfo(skb
)->gso_segs
?: 1;
304 /* adjust for FCoE Sequence Offload */
305 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)
306 && (skb
->protocol
== htons(ETH_P_FCOE
)) &&
308 hlen
= skb_transport_offset(skb
) +
309 sizeof(struct fc_frame_header
) +
310 sizeof(struct fcoe_crc_eof
);
311 segs
= DIV_ROUND_UP(skb
->len
- hlen
,
312 skb_shinfo(skb
)->gso_size
);
314 #endif /* IXGBE_FCOE */
315 /* multiply data chunks by size of headers */
316 bytecount
= ((segs
- 1) * hlen
) + skb
->len
;
317 total_packets
+= segs
;
318 total_bytes
+= bytecount
;
321 ixgbe_unmap_and_free_tx_resource(adapter
,
324 tx_desc
->wb
.status
= 0;
327 if (i
== tx_ring
->count
)
331 eop
= tx_ring
->tx_buffer_info
[i
].next_to_watch
;
332 eop_desc
= IXGBE_TX_DESC_ADV(*tx_ring
, eop
);
335 tx_ring
->next_to_clean
= i
;
337 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
338 if (unlikely(count
&& netif_carrier_ok(netdev
) &&
339 (IXGBE_DESC_UNUSED(tx_ring
) >= TX_WAKE_THRESHOLD
))) {
340 /* Make sure that anybody stopping the queue after this
341 * sees the new next_to_clean.
344 if (__netif_subqueue_stopped(netdev
, tx_ring
->queue_index
) &&
345 !test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
346 netif_wake_subqueue(netdev
, tx_ring
->queue_index
);
347 ++adapter
->restart_queue
;
351 if (adapter
->detect_tx_hung
) {
352 if (ixgbe_check_tx_hang(adapter
, tx_ring
, i
)) {
353 /* schedule immediate reset if we believe we hung */
355 "tx hang %d detected, resetting adapter\n",
356 adapter
->tx_timeout_count
+ 1);
357 ixgbe_tx_timeout(adapter
->netdev
);
361 /* re-arm the interrupt */
362 if (count
>= tx_ring
->work_limit
)
363 ixgbe_irq_rearm_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
365 tx_ring
->total_bytes
+= total_bytes
;
366 tx_ring
->total_packets
+= total_packets
;
367 tx_ring
->stats
.packets
+= total_packets
;
368 tx_ring
->stats
.bytes
+= total_bytes
;
369 adapter
->net_stats
.tx_bytes
+= total_bytes
;
370 adapter
->net_stats
.tx_packets
+= total_packets
;
371 return (count
< tx_ring
->work_limit
);
374 #ifdef CONFIG_IXGBE_DCA
375 static void ixgbe_update_rx_dca(struct ixgbe_adapter
*adapter
,
376 struct ixgbe_ring
*rx_ring
)
380 int q
= rx_ring
- adapter
->rx_ring
;
382 if (rx_ring
->cpu
!= cpu
) {
383 rxctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_DCA_RXCTRL(q
));
384 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
385 rxctrl
&= ~IXGBE_DCA_RXCTRL_CPUID_MASK
;
386 rxctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
387 } else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
388 rxctrl
&= ~IXGBE_DCA_RXCTRL_CPUID_MASK_82599
;
389 rxctrl
|= (dca3_get_tag(&adapter
->pdev
->dev
, cpu
) <<
390 IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599
);
392 rxctrl
|= IXGBE_DCA_RXCTRL_DESC_DCA_EN
;
393 rxctrl
|= IXGBE_DCA_RXCTRL_HEAD_DCA_EN
;
394 rxctrl
&= ~(IXGBE_DCA_RXCTRL_DESC_RRO_EN
);
395 rxctrl
&= ~(IXGBE_DCA_RXCTRL_DESC_WRO_EN
|
396 IXGBE_DCA_RXCTRL_DESC_HSRO_EN
);
397 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_RXCTRL(q
), rxctrl
);
403 static void ixgbe_update_tx_dca(struct ixgbe_adapter
*adapter
,
404 struct ixgbe_ring
*tx_ring
)
408 int q
= tx_ring
- adapter
->tx_ring
;
410 if (tx_ring
->cpu
!= cpu
) {
411 txctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_DCA_TXCTRL(q
));
412 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
413 txctrl
&= ~IXGBE_DCA_TXCTRL_CPUID_MASK
;
414 txctrl
|= dca3_get_tag(&adapter
->pdev
->dev
, cpu
);
415 } else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
416 txctrl
&= ~IXGBE_DCA_TXCTRL_CPUID_MASK_82599
;
417 txctrl
|= (dca3_get_tag(&adapter
->pdev
->dev
, cpu
) <<
418 IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599
);
420 txctrl
|= IXGBE_DCA_TXCTRL_DESC_DCA_EN
;
421 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_TXCTRL(q
), txctrl
);
427 static void ixgbe_setup_dca(struct ixgbe_adapter
*adapter
)
431 if (!(adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
))
434 /* always use CB2 mode, difference is masked in the CB driver */
435 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 2);
437 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
438 adapter
->tx_ring
[i
].cpu
= -1;
439 ixgbe_update_tx_dca(adapter
, &adapter
->tx_ring
[i
]);
441 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
442 adapter
->rx_ring
[i
].cpu
= -1;
443 ixgbe_update_rx_dca(adapter
, &adapter
->rx_ring
[i
]);
447 static int __ixgbe_notify_dca(struct device
*dev
, void *data
)
449 struct net_device
*netdev
= dev_get_drvdata(dev
);
450 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
451 unsigned long event
= *(unsigned long *)data
;
454 case DCA_PROVIDER_ADD
:
455 /* if we're already enabled, don't do it again */
456 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
458 if (dca_add_requester(dev
) == 0) {
459 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
460 ixgbe_setup_dca(adapter
);
463 /* Fall Through since DCA is disabled. */
464 case DCA_PROVIDER_REMOVE
:
465 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
466 dca_remove_requester(dev
);
467 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
468 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
476 #endif /* CONFIG_IXGBE_DCA */
478 * ixgbe_receive_skb - Send a completed packet up the stack
479 * @adapter: board private structure
480 * @skb: packet to send up
481 * @status: hardware indication of status of receive
482 * @rx_ring: rx descriptor ring (for a specific queue) to setup
483 * @rx_desc: rx descriptor
485 static void ixgbe_receive_skb(struct ixgbe_q_vector
*q_vector
,
486 struct sk_buff
*skb
, u8 status
,
487 struct ixgbe_ring
*ring
,
488 union ixgbe_adv_rx_desc
*rx_desc
)
490 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
491 struct napi_struct
*napi
= &q_vector
->napi
;
492 bool is_vlan
= (status
& IXGBE_RXD_STAT_VP
);
493 u16 tag
= le16_to_cpu(rx_desc
->wb
.upper
.vlan
);
495 skb_record_rx_queue(skb
, ring
->queue_index
);
496 if (!(adapter
->flags
& IXGBE_FLAG_IN_NETPOLL
)) {
497 if (adapter
->vlgrp
&& is_vlan
&& (tag
!= 0))
498 vlan_gro_receive(napi
, adapter
->vlgrp
, tag
, skb
);
500 napi_gro_receive(napi
, skb
);
502 if (adapter
->vlgrp
&& is_vlan
&& (tag
!= 0))
503 vlan_hwaccel_rx(skb
, adapter
->vlgrp
, tag
);
510 * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
511 * @adapter: address of board private structure
512 * @status_err: hardware indication of status of receive
513 * @skb: skb currently being received and modified
515 static inline void ixgbe_rx_checksum(struct ixgbe_adapter
*adapter
,
516 union ixgbe_adv_rx_desc
*rx_desc
,
519 u32 status_err
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
521 skb
->ip_summed
= CHECKSUM_NONE
;
523 /* Rx csum disabled */
524 if (!(adapter
->flags
& IXGBE_FLAG_RX_CSUM_ENABLED
))
527 /* if IP and error */
528 if ((status_err
& IXGBE_RXD_STAT_IPCS
) &&
529 (status_err
& IXGBE_RXDADV_ERR_IPE
)) {
530 adapter
->hw_csum_rx_error
++;
534 if (!(status_err
& IXGBE_RXD_STAT_L4CS
))
537 if (status_err
& IXGBE_RXDADV_ERR_TCPE
) {
538 u16 pkt_info
= rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
541 * 82599 errata, UDP frames with a 0 checksum can be marked as
544 if ((pkt_info
& IXGBE_RXDADV_PKTTYPE_UDP
) &&
545 (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
))
548 adapter
->hw_csum_rx_error
++;
552 /* It must be a TCP or UDP packet with a valid checksum */
553 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
554 adapter
->hw_csum_rx_good
++;
557 static inline void ixgbe_release_rx_desc(struct ixgbe_hw
*hw
,
558 struct ixgbe_ring
*rx_ring
, u32 val
)
561 * Force memory writes to complete before letting h/w
562 * know there are new descriptors to fetch. (Only
563 * applicable for weak-ordered memory model archs,
567 IXGBE_WRITE_REG(hw
, IXGBE_RDT(rx_ring
->reg_idx
), val
);
571 * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
572 * @adapter: address of board private structure
574 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter
*adapter
,
575 struct ixgbe_ring
*rx_ring
,
578 struct pci_dev
*pdev
= adapter
->pdev
;
579 union ixgbe_adv_rx_desc
*rx_desc
;
580 struct ixgbe_rx_buffer
*bi
;
583 i
= rx_ring
->next_to_use
;
584 bi
= &rx_ring
->rx_buffer_info
[i
];
586 while (cleaned_count
--) {
587 rx_desc
= IXGBE_RX_DESC_ADV(*rx_ring
, i
);
590 (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
)) {
592 bi
->page
= alloc_page(GFP_ATOMIC
);
594 adapter
->alloc_rx_page_failed
++;
599 /* use a half page if we're re-using */
600 bi
->page_offset
^= (PAGE_SIZE
/ 2);
603 bi
->page_dma
= pci_map_page(pdev
, bi
->page
,
611 skb
= netdev_alloc_skb(adapter
->netdev
,
612 (rx_ring
->rx_buf_len
+
616 adapter
->alloc_rx_buff_failed
++;
621 * Make buffer alignment 2 beyond a 16 byte boundary
622 * this will result in a 16 byte aligned IP header after
623 * the 14 byte MAC header is removed
625 skb_reserve(skb
, NET_IP_ALIGN
);
628 bi
->dma
= pci_map_single(pdev
, skb
->data
,
632 /* Refresh the desc even if buffer_addrs didn't change because
633 * each write-back erases this info. */
634 if (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
) {
635 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->page_dma
);
636 rx_desc
->read
.hdr_addr
= cpu_to_le64(bi
->dma
);
638 rx_desc
->read
.pkt_addr
= cpu_to_le64(bi
->dma
);
642 if (i
== rx_ring
->count
)
644 bi
= &rx_ring
->rx_buffer_info
[i
];
648 if (rx_ring
->next_to_use
!= i
) {
649 rx_ring
->next_to_use
= i
;
651 i
= (rx_ring
->count
- 1);
653 ixgbe_release_rx_desc(&adapter
->hw
, rx_ring
, i
);
657 static inline u16
ixgbe_get_hdr_info(union ixgbe_adv_rx_desc
*rx_desc
)
659 return rx_desc
->wb
.lower
.lo_dword
.hs_rss
.hdr_info
;
662 static inline u16
ixgbe_get_pkt_info(union ixgbe_adv_rx_desc
*rx_desc
)
664 return rx_desc
->wb
.lower
.lo_dword
.hs_rss
.pkt_info
;
667 static inline u32
ixgbe_get_rsc_count(union ixgbe_adv_rx_desc
*rx_desc
)
669 return (le32_to_cpu(rx_desc
->wb
.lower
.lo_dword
.data
) &
670 IXGBE_RXDADV_RSCCNT_MASK
) >>
671 IXGBE_RXDADV_RSCCNT_SHIFT
;
675 * ixgbe_transform_rsc_queue - change rsc queue into a full packet
676 * @skb: pointer to the last skb in the rsc queue
678 * This function changes a queue full of hw rsc buffers into a completed
679 * packet. It uses the ->prev pointers to find the first packet and then
680 * turns it into the frag list owner.
682 static inline struct sk_buff
*ixgbe_transform_rsc_queue(struct sk_buff
*skb
)
684 unsigned int frag_list_size
= 0;
687 struct sk_buff
*prev
= skb
->prev
;
688 frag_list_size
+= skb
->len
;
693 skb_shinfo(skb
)->frag_list
= skb
->next
;
695 skb
->len
+= frag_list_size
;
696 skb
->data_len
+= frag_list_size
;
697 skb
->truesize
+= frag_list_size
;
701 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector
*q_vector
,
702 struct ixgbe_ring
*rx_ring
,
703 int *work_done
, int work_to_do
)
705 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
706 struct pci_dev
*pdev
= adapter
->pdev
;
707 union ixgbe_adv_rx_desc
*rx_desc
, *next_rxd
;
708 struct ixgbe_rx_buffer
*rx_buffer_info
, *next_buffer
;
710 unsigned int i
, rsc_count
= 0;
713 bool cleaned
= false;
714 int cleaned_count
= 0;
715 unsigned int total_rx_bytes
= 0, total_rx_packets
= 0;
718 #endif /* IXGBE_FCOE */
720 i
= rx_ring
->next_to_clean
;
721 rx_desc
= IXGBE_RX_DESC_ADV(*rx_ring
, i
);
722 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
723 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
725 while (staterr
& IXGBE_RXD_STAT_DD
) {
727 if (*work_done
>= work_to_do
)
731 if (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
) {
732 hdr_info
= le16_to_cpu(ixgbe_get_hdr_info(rx_desc
));
733 len
= (hdr_info
& IXGBE_RXDADV_HDRBUFLEN_MASK
) >>
734 IXGBE_RXDADV_HDRBUFLEN_SHIFT
;
735 if (hdr_info
& IXGBE_RXDADV_SPH
)
736 adapter
->rx_hdr_split
++;
737 if (len
> IXGBE_RX_HDR_SIZE
)
738 len
= IXGBE_RX_HDR_SIZE
;
739 upper_len
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
741 len
= le16_to_cpu(rx_desc
->wb
.upper
.length
);
745 skb
= rx_buffer_info
->skb
;
746 prefetch(skb
->data
- NET_IP_ALIGN
);
747 rx_buffer_info
->skb
= NULL
;
749 if (rx_buffer_info
->dma
) {
750 pci_unmap_single(pdev
, rx_buffer_info
->dma
,
753 rx_buffer_info
->dma
= 0;
758 pci_unmap_page(pdev
, rx_buffer_info
->page_dma
,
759 PAGE_SIZE
/ 2, PCI_DMA_FROMDEVICE
);
760 rx_buffer_info
->page_dma
= 0;
761 skb_fill_page_desc(skb
, skb_shinfo(skb
)->nr_frags
,
762 rx_buffer_info
->page
,
763 rx_buffer_info
->page_offset
,
766 if ((rx_ring
->rx_buf_len
> (PAGE_SIZE
/ 2)) ||
767 (page_count(rx_buffer_info
->page
) != 1))
768 rx_buffer_info
->page
= NULL
;
770 get_page(rx_buffer_info
->page
);
772 skb
->len
+= upper_len
;
773 skb
->data_len
+= upper_len
;
774 skb
->truesize
+= upper_len
;
778 if (i
== rx_ring
->count
)
781 next_rxd
= IXGBE_RX_DESC_ADV(*rx_ring
, i
);
785 if (adapter
->flags2
& IXGBE_FLAG2_RSC_CAPABLE
)
786 rsc_count
= ixgbe_get_rsc_count(rx_desc
);
789 u32 nextp
= (staterr
& IXGBE_RXDADV_NEXTP_MASK
) >>
790 IXGBE_RXDADV_NEXTP_SHIFT
;
791 next_buffer
= &rx_ring
->rx_buffer_info
[nextp
];
792 rx_ring
->rsc_count
+= (rsc_count
- 1);
794 next_buffer
= &rx_ring
->rx_buffer_info
[i
];
797 if (staterr
& IXGBE_RXD_STAT_EOP
) {
799 skb
= ixgbe_transform_rsc_queue(skb
);
800 rx_ring
->stats
.packets
++;
801 rx_ring
->stats
.bytes
+= skb
->len
;
803 if (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
) {
804 rx_buffer_info
->skb
= next_buffer
->skb
;
805 rx_buffer_info
->dma
= next_buffer
->dma
;
806 next_buffer
->skb
= skb
;
807 next_buffer
->dma
= 0;
809 skb
->next
= next_buffer
->skb
;
810 skb
->next
->prev
= skb
;
812 adapter
->non_eop_descs
++;
816 if (staterr
& IXGBE_RXDADV_ERR_FRAME_ERR_MASK
) {
817 dev_kfree_skb_irq(skb
);
821 ixgbe_rx_checksum(adapter
, rx_desc
, skb
);
823 /* probably a little skewed due to removing CRC */
824 total_rx_bytes
+= skb
->len
;
827 skb
->protocol
= eth_type_trans(skb
, adapter
->netdev
);
829 /* if ddp, not passing to ULD unless for FCP_RSP or error */
830 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) {
831 ddp_bytes
= ixgbe_fcoe_ddp(adapter
, rx_desc
, skb
);
835 #endif /* IXGBE_FCOE */
836 ixgbe_receive_skb(q_vector
, skb
, staterr
, rx_ring
, rx_desc
);
839 rx_desc
->wb
.upper
.status_error
= 0;
841 /* return some buffers to hardware, one at a time is too slow */
842 if (cleaned_count
>= IXGBE_RX_BUFFER_WRITE
) {
843 ixgbe_alloc_rx_buffers(adapter
, rx_ring
, cleaned_count
);
847 /* use prefetched values */
849 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
851 staterr
= le32_to_cpu(rx_desc
->wb
.upper
.status_error
);
854 rx_ring
->next_to_clean
= i
;
855 cleaned_count
= IXGBE_DESC_UNUSED(rx_ring
);
858 ixgbe_alloc_rx_buffers(adapter
, rx_ring
, cleaned_count
);
861 /* include DDPed FCoE data */
865 mss
= adapter
->netdev
->mtu
- sizeof(struct fcoe_hdr
) -
866 sizeof(struct fc_frame_header
) -
867 sizeof(struct fcoe_crc_eof
);
870 total_rx_bytes
+= ddp_bytes
;
871 total_rx_packets
+= DIV_ROUND_UP(ddp_bytes
, mss
);
873 #endif /* IXGBE_FCOE */
875 rx_ring
->total_packets
+= total_rx_packets
;
876 rx_ring
->total_bytes
+= total_rx_bytes
;
877 adapter
->net_stats
.rx_bytes
+= total_rx_bytes
;
878 adapter
->net_stats
.rx_packets
+= total_rx_packets
;
883 static int ixgbe_clean_rxonly(struct napi_struct
*, int);
885 * ixgbe_configure_msix - Configure MSI-X hardware
886 * @adapter: board private structure
888 * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
891 static void ixgbe_configure_msix(struct ixgbe_adapter
*adapter
)
893 struct ixgbe_q_vector
*q_vector
;
894 int i
, j
, q_vectors
, v_idx
, r_idx
;
897 q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
900 * Populate the IVAR table and set the ITR values to the
901 * corresponding register.
903 for (v_idx
= 0; v_idx
< q_vectors
; v_idx
++) {
904 q_vector
= adapter
->q_vector
[v_idx
];
905 /* XXX for_each_bit(...) */
906 r_idx
= find_first_bit(q_vector
->rxr_idx
,
907 adapter
->num_rx_queues
);
909 for (i
= 0; i
< q_vector
->rxr_count
; i
++) {
910 j
= adapter
->rx_ring
[r_idx
].reg_idx
;
911 ixgbe_set_ivar(adapter
, 0, j
, v_idx
);
912 r_idx
= find_next_bit(q_vector
->rxr_idx
,
913 adapter
->num_rx_queues
,
916 r_idx
= find_first_bit(q_vector
->txr_idx
,
917 adapter
->num_tx_queues
);
919 for (i
= 0; i
< q_vector
->txr_count
; i
++) {
920 j
= adapter
->tx_ring
[r_idx
].reg_idx
;
921 ixgbe_set_ivar(adapter
, 1, j
, v_idx
);
922 r_idx
= find_next_bit(q_vector
->txr_idx
,
923 adapter
->num_tx_queues
,
927 /* if this is a tx only vector halve the interrupt rate */
928 if (q_vector
->txr_count
&& !q_vector
->rxr_count
)
929 q_vector
->eitr
= (adapter
->eitr_param
>> 1);
930 else if (q_vector
->rxr_count
)
932 q_vector
->eitr
= adapter
->eitr_param
;
934 ixgbe_write_eitr(q_vector
);
937 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
)
938 ixgbe_set_ivar(adapter
, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX
,
940 else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
)
941 ixgbe_set_ivar(adapter
, -1, 1, v_idx
);
942 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EITR(v_idx
), 1950);
944 /* set up to autoclear timer, and the vectors */
945 mask
= IXGBE_EIMS_ENABLE_MASK
;
946 mask
&= ~(IXGBE_EIMS_OTHER
| IXGBE_EIMS_LSC
);
947 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIAC
, mask
);
954 latency_invalid
= 255
958 * ixgbe_update_itr - update the dynamic ITR value based on statistics
959 * @adapter: pointer to adapter
960 * @eitr: eitr setting (ints per sec) to give last timeslice
961 * @itr_setting: current throttle rate in ints/second
962 * @packets: the number of packets during this measurement interval
963 * @bytes: the number of bytes during this measurement interval
965 * Stores a new ITR value based on packets and byte
966 * counts during the last interrupt. The advantage of per interrupt
967 * computation is faster updates and more accurate ITR for the current
968 * traffic pattern. Constants in this function were computed
969 * based on theoretical maximum wire speed and thresholds were set based
970 * on testing data as well as attempting to minimize response time
971 * while increasing bulk throughput.
972 * this functionality is controlled by the InterruptThrottleRate module
973 * parameter (see ixgbe_param.c)
975 static u8
ixgbe_update_itr(struct ixgbe_adapter
*adapter
,
976 u32 eitr
, u8 itr_setting
,
977 int packets
, int bytes
)
979 unsigned int retval
= itr_setting
;
984 goto update_itr_done
;
987 /* simple throttlerate management
988 * 0-20MB/s lowest (100000 ints/s)
989 * 20-100MB/s low (20000 ints/s)
990 * 100-1249MB/s bulk (8000 ints/s)
992 /* what was last interrupt timeslice? */
993 timepassed_us
= 1000000/eitr
;
994 bytes_perint
= bytes
/ timepassed_us
; /* bytes/usec */
996 switch (itr_setting
) {
998 if (bytes_perint
> adapter
->eitr_low
)
999 retval
= low_latency
;
1002 if (bytes_perint
> adapter
->eitr_high
)
1003 retval
= bulk_latency
;
1004 else if (bytes_perint
<= adapter
->eitr_low
)
1005 retval
= lowest_latency
;
1008 if (bytes_perint
<= adapter
->eitr_high
)
1009 retval
= low_latency
;
1018 * ixgbe_write_eitr - write EITR register in hardware specific way
1019 * @q_vector: structure containing interrupt and ring information
1021 * This function is made to be called by ethtool and by the driver
1022 * when it needs to update EITR registers at runtime. Hardware
1023 * specific quirks/differences are taken care of here.
1025 void ixgbe_write_eitr(struct ixgbe_q_vector
*q_vector
)
1027 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1028 struct ixgbe_hw
*hw
= &adapter
->hw
;
1029 int v_idx
= q_vector
->v_idx
;
1030 u32 itr_reg
= EITR_INTS_PER_SEC_TO_REG(q_vector
->eitr
);
1032 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
1033 /* must write high and low 16 bits to reset counter */
1034 itr_reg
|= (itr_reg
<< 16);
1035 } else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
1037 * set the WDIS bit to not clear the timer bits and cause an
1038 * immediate assertion of the interrupt
1040 itr_reg
|= IXGBE_EITR_CNT_WDIS
;
1042 IXGBE_WRITE_REG(hw
, IXGBE_EITR(v_idx
), itr_reg
);
1045 static void ixgbe_set_itr_msix(struct ixgbe_q_vector
*q_vector
)
1047 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1049 u8 current_itr
, ret_itr
;
1051 struct ixgbe_ring
*rx_ring
, *tx_ring
;
1053 r_idx
= find_first_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
);
1054 for (i
= 0; i
< q_vector
->txr_count
; i
++) {
1055 tx_ring
= &(adapter
->tx_ring
[r_idx
]);
1056 ret_itr
= ixgbe_update_itr(adapter
, q_vector
->eitr
,
1058 tx_ring
->total_packets
,
1059 tx_ring
->total_bytes
);
1060 /* if the result for this queue would decrease interrupt
1061 * rate for this vector then use that result */
1062 q_vector
->tx_itr
= ((q_vector
->tx_itr
> ret_itr
) ?
1063 q_vector
->tx_itr
- 1 : ret_itr
);
1064 r_idx
= find_next_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
,
1068 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1069 for (i
= 0; i
< q_vector
->rxr_count
; i
++) {
1070 rx_ring
= &(adapter
->rx_ring
[r_idx
]);
1071 ret_itr
= ixgbe_update_itr(adapter
, q_vector
->eitr
,
1073 rx_ring
->total_packets
,
1074 rx_ring
->total_bytes
);
1075 /* if the result for this queue would decrease interrupt
1076 * rate for this vector then use that result */
1077 q_vector
->rx_itr
= ((q_vector
->rx_itr
> ret_itr
) ?
1078 q_vector
->rx_itr
- 1 : ret_itr
);
1079 r_idx
= find_next_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
,
1083 current_itr
= max(q_vector
->rx_itr
, q_vector
->tx_itr
);
1085 switch (current_itr
) {
1086 /* counts and packets in update_itr are dependent on these numbers */
1087 case lowest_latency
:
1091 new_itr
= 20000; /* aka hwitr = ~200 */
1099 if (new_itr
!= q_vector
->eitr
) {
1100 /* do an exponential smoothing */
1101 new_itr
= ((q_vector
->eitr
* 90)/100) + ((new_itr
* 10)/100);
1103 /* save the algorithm value here, not the smoothed one */
1104 q_vector
->eitr
= new_itr
;
1106 ixgbe_write_eitr(q_vector
);
1112 static void ixgbe_check_fan_failure(struct ixgbe_adapter
*adapter
, u32 eicr
)
1114 struct ixgbe_hw
*hw
= &adapter
->hw
;
1116 if ((adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) &&
1117 (eicr
& IXGBE_EICR_GPI_SDP1
)) {
1118 DPRINTK(PROBE
, CRIT
, "Fan has stopped, replace the adapter\n");
1119 /* write to clear the interrupt */
1120 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
1124 static void ixgbe_check_sfp_event(struct ixgbe_adapter
*adapter
, u32 eicr
)
1126 struct ixgbe_hw
*hw
= &adapter
->hw
;
1128 if (eicr
& IXGBE_EICR_GPI_SDP1
) {
1129 /* Clear the interrupt */
1130 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP1
);
1131 schedule_work(&adapter
->multispeed_fiber_task
);
1132 } else if (eicr
& IXGBE_EICR_GPI_SDP2
) {
1133 /* Clear the interrupt */
1134 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_GPI_SDP2
);
1135 schedule_work(&adapter
->sfp_config_module_task
);
1137 /* Interrupt isn't for us... */
1142 static void ixgbe_check_lsc(struct ixgbe_adapter
*adapter
)
1144 struct ixgbe_hw
*hw
= &adapter
->hw
;
1147 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
1148 adapter
->link_check_timeout
= jiffies
;
1149 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
)) {
1150 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_EIMC_LSC
);
1151 schedule_work(&adapter
->watchdog_task
);
1155 static irqreturn_t
ixgbe_msix_lsc(int irq
, void *data
)
1157 struct net_device
*netdev
= data
;
1158 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
1159 struct ixgbe_hw
*hw
= &adapter
->hw
;
1163 * Workaround for Silicon errata. Use clear-by-write instead
1164 * of clear-by-read. Reading with EICS will return the
1165 * interrupt causes without clearing, which later be done
1166 * with the write to EICR.
1168 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICS
);
1169 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, eicr
);
1171 if (eicr
& IXGBE_EICR_LSC
)
1172 ixgbe_check_lsc(adapter
);
1174 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
1175 ixgbe_check_fan_failure(adapter
, eicr
);
1177 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
1178 ixgbe_check_sfp_event(adapter
, eicr
);
1180 /* Handle Flow Director Full threshold interrupt */
1181 if (eicr
& IXGBE_EICR_FLOW_DIR
) {
1183 IXGBE_WRITE_REG(hw
, IXGBE_EICR
, IXGBE_EICR_FLOW_DIR
);
1184 /* Disable transmits before FDIR Re-initialization */
1185 netif_tx_stop_all_queues(netdev
);
1186 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
1187 struct ixgbe_ring
*tx_ring
=
1188 &adapter
->tx_ring
[i
];
1189 if (test_and_clear_bit(__IXGBE_FDIR_INIT_DONE
,
1190 &tx_ring
->reinit_state
))
1191 schedule_work(&adapter
->fdir_reinit_task
);
1195 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
1196 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMS_OTHER
);
1201 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter
*adapter
,
1206 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
1207 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
1208 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS
, mask
);
1210 mask
= (qmask
& 0xFFFFFFFF);
1211 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS_EX(0), mask
);
1212 mask
= (qmask
>> 32);
1213 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS_EX(1), mask
);
1215 /* skip the flush */
1218 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter
*adapter
,
1223 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
1224 mask
= (IXGBE_EIMS_RTX_QUEUE
& qmask
);
1225 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, mask
);
1227 mask
= (qmask
& 0xFFFFFFFF);
1228 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(0), mask
);
1229 mask
= (qmask
>> 32);
1230 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(1), mask
);
1232 /* skip the flush */
1235 static irqreturn_t
ixgbe_msix_clean_tx(int irq
, void *data
)
1237 struct ixgbe_q_vector
*q_vector
= data
;
1238 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1239 struct ixgbe_ring
*tx_ring
;
1242 if (!q_vector
->txr_count
)
1245 r_idx
= find_first_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
);
1246 for (i
= 0; i
< q_vector
->txr_count
; i
++) {
1247 tx_ring
= &(adapter
->tx_ring
[r_idx
]);
1248 tx_ring
->total_bytes
= 0;
1249 tx_ring
->total_packets
= 0;
1250 r_idx
= find_next_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
,
1254 /* disable interrupts on this vector only */
1255 ixgbe_irq_disable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
1256 napi_schedule(&q_vector
->napi
);
1262 * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
1264 * @data: pointer to our q_vector struct for this interrupt vector
1266 static irqreturn_t
ixgbe_msix_clean_rx(int irq
, void *data
)
1268 struct ixgbe_q_vector
*q_vector
= data
;
1269 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1270 struct ixgbe_ring
*rx_ring
;
1274 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1275 for (i
= 0; i
< q_vector
->rxr_count
; i
++) {
1276 rx_ring
= &(adapter
->rx_ring
[r_idx
]);
1277 rx_ring
->total_bytes
= 0;
1278 rx_ring
->total_packets
= 0;
1279 r_idx
= find_next_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
,
1283 if (!q_vector
->rxr_count
)
1286 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1287 rx_ring
= &(adapter
->rx_ring
[r_idx
]);
1288 /* disable interrupts on this vector only */
1289 ixgbe_irq_disable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
1290 napi_schedule(&q_vector
->napi
);
1295 static irqreturn_t
ixgbe_msix_clean_many(int irq
, void *data
)
1297 struct ixgbe_q_vector
*q_vector
= data
;
1298 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1299 struct ixgbe_ring
*ring
;
1303 if (!q_vector
->txr_count
&& !q_vector
->rxr_count
)
1306 r_idx
= find_first_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
);
1307 for (i
= 0; i
< q_vector
->txr_count
; i
++) {
1308 ring
= &(adapter
->tx_ring
[r_idx
]);
1309 ring
->total_bytes
= 0;
1310 ring
->total_packets
= 0;
1311 r_idx
= find_next_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
,
1315 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1316 for (i
= 0; i
< q_vector
->rxr_count
; i
++) {
1317 ring
= &(adapter
->rx_ring
[r_idx
]);
1318 ring
->total_bytes
= 0;
1319 ring
->total_packets
= 0;
1320 r_idx
= find_next_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
,
1324 /* disable interrupts on this vector only */
1325 ixgbe_irq_disable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
1326 napi_schedule(&q_vector
->napi
);
1332 * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1333 * @napi: napi struct with our devices info in it
1334 * @budget: amount of work driver is allowed to do this pass, in packets
1336 * This function is optimized for cleaning one queue only on a single
1339 static int ixgbe_clean_rxonly(struct napi_struct
*napi
, int budget
)
1341 struct ixgbe_q_vector
*q_vector
=
1342 container_of(napi
, struct ixgbe_q_vector
, napi
);
1343 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1344 struct ixgbe_ring
*rx_ring
= NULL
;
1348 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1349 rx_ring
= &(adapter
->rx_ring
[r_idx
]);
1350 #ifdef CONFIG_IXGBE_DCA
1351 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1352 ixgbe_update_rx_dca(adapter
, rx_ring
);
1355 ixgbe_clean_rx_irq(q_vector
, rx_ring
, &work_done
, budget
);
1357 /* If all Rx work done, exit the polling mode */
1358 if (work_done
< budget
) {
1359 napi_complete(napi
);
1360 if (adapter
->itr_setting
& 1)
1361 ixgbe_set_itr_msix(q_vector
);
1362 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
1363 ixgbe_irq_enable_queues(adapter
,
1364 ((u64
)1 << q_vector
->v_idx
));
1371 * ixgbe_clean_rxtx_many - msix (aka one shot) rx clean routine
1372 * @napi: napi struct with our devices info in it
1373 * @budget: amount of work driver is allowed to do this pass, in packets
1375 * This function will clean more than one rx queue associated with a
1378 static int ixgbe_clean_rxtx_many(struct napi_struct
*napi
, int budget
)
1380 struct ixgbe_q_vector
*q_vector
=
1381 container_of(napi
, struct ixgbe_q_vector
, napi
);
1382 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1383 struct ixgbe_ring
*ring
= NULL
;
1384 int work_done
= 0, i
;
1386 bool tx_clean_complete
= true;
1388 r_idx
= find_first_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
);
1389 for (i
= 0; i
< q_vector
->txr_count
; i
++) {
1390 ring
= &(adapter
->tx_ring
[r_idx
]);
1391 #ifdef CONFIG_IXGBE_DCA
1392 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1393 ixgbe_update_tx_dca(adapter
, ring
);
1395 tx_clean_complete
&= ixgbe_clean_tx_irq(q_vector
, ring
);
1396 r_idx
= find_next_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
,
1400 /* attempt to distribute budget to each queue fairly, but don't allow
1401 * the budget to go below 1 because we'll exit polling */
1402 budget
/= (q_vector
->rxr_count
?: 1);
1403 budget
= max(budget
, 1);
1404 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1405 for (i
= 0; i
< q_vector
->rxr_count
; i
++) {
1406 ring
= &(adapter
->rx_ring
[r_idx
]);
1407 #ifdef CONFIG_IXGBE_DCA
1408 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1409 ixgbe_update_rx_dca(adapter
, ring
);
1411 ixgbe_clean_rx_irq(q_vector
, ring
, &work_done
, budget
);
1412 r_idx
= find_next_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
,
1416 r_idx
= find_first_bit(q_vector
->rxr_idx
, adapter
->num_rx_queues
);
1417 ring
= &(adapter
->rx_ring
[r_idx
]);
1418 /* If all Rx work done, exit the polling mode */
1419 if (work_done
< budget
) {
1420 napi_complete(napi
);
1421 if (adapter
->itr_setting
& 1)
1422 ixgbe_set_itr_msix(q_vector
);
1423 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
1424 ixgbe_irq_enable_queues(adapter
,
1425 ((u64
)1 << q_vector
->v_idx
));
1433 * ixgbe_clean_txonly - msix (aka one shot) tx clean routine
1434 * @napi: napi struct with our devices info in it
1435 * @budget: amount of work driver is allowed to do this pass, in packets
1437 * This function is optimized for cleaning one queue only on a single
1440 static int ixgbe_clean_txonly(struct napi_struct
*napi
, int budget
)
1442 struct ixgbe_q_vector
*q_vector
=
1443 container_of(napi
, struct ixgbe_q_vector
, napi
);
1444 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
1445 struct ixgbe_ring
*tx_ring
= NULL
;
1449 r_idx
= find_first_bit(q_vector
->txr_idx
, adapter
->num_tx_queues
);
1450 tx_ring
= &(adapter
->tx_ring
[r_idx
]);
1451 #ifdef CONFIG_IXGBE_DCA
1452 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
)
1453 ixgbe_update_tx_dca(adapter
, tx_ring
);
1456 if (!ixgbe_clean_tx_irq(q_vector
, tx_ring
))
1459 /* If all Rx work done, exit the polling mode */
1460 if (work_done
< budget
) {
1461 napi_complete(napi
);
1462 if (adapter
->itr_setting
& 1)
1463 ixgbe_set_itr_msix(q_vector
);
1464 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
1465 ixgbe_irq_enable_queues(adapter
, ((u64
)1 << q_vector
->v_idx
));
1471 static inline void map_vector_to_rxq(struct ixgbe_adapter
*a
, int v_idx
,
1474 struct ixgbe_q_vector
*q_vector
= a
->q_vector
[v_idx
];
1476 set_bit(r_idx
, q_vector
->rxr_idx
);
1477 q_vector
->rxr_count
++;
1480 static inline void map_vector_to_txq(struct ixgbe_adapter
*a
, int v_idx
,
1483 struct ixgbe_q_vector
*q_vector
= a
->q_vector
[v_idx
];
1485 set_bit(t_idx
, q_vector
->txr_idx
);
1486 q_vector
->txr_count
++;
1490 * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1491 * @adapter: board private structure to initialize
1492 * @vectors: allotted vector count for descriptor rings
1494 * This function maps descriptor rings to the queue-specific vectors
1495 * we were allotted through the MSI-X enabling code. Ideally, we'd have
1496 * one vector per ring/queue, but on a constrained vector budget, we
1497 * group the rings as "efficiently" as possible. You would add new
1498 * mapping configurations in here.
1500 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter
*adapter
,
1504 int rxr_idx
= 0, txr_idx
= 0;
1505 int rxr_remaining
= adapter
->num_rx_queues
;
1506 int txr_remaining
= adapter
->num_tx_queues
;
1511 /* No mapping required if MSI-X is disabled. */
1512 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
))
1516 * The ideal configuration...
1517 * We have enough vectors to map one per queue.
1519 if (vectors
== adapter
->num_rx_queues
+ adapter
->num_tx_queues
) {
1520 for (; rxr_idx
< rxr_remaining
; v_start
++, rxr_idx
++)
1521 map_vector_to_rxq(adapter
, v_start
, rxr_idx
);
1523 for (; txr_idx
< txr_remaining
; v_start
++, txr_idx
++)
1524 map_vector_to_txq(adapter
, v_start
, txr_idx
);
1530 * If we don't have enough vectors for a 1-to-1
1531 * mapping, we'll have to group them so there are
1532 * multiple queues per vector.
1534 /* Re-adjusting *qpv takes care of the remainder. */
1535 for (i
= v_start
; i
< vectors
; i
++) {
1536 rqpv
= DIV_ROUND_UP(rxr_remaining
, vectors
- i
);
1537 for (j
= 0; j
< rqpv
; j
++) {
1538 map_vector_to_rxq(adapter
, i
, rxr_idx
);
1543 for (i
= v_start
; i
< vectors
; i
++) {
1544 tqpv
= DIV_ROUND_UP(txr_remaining
, vectors
- i
);
1545 for (j
= 0; j
< tqpv
; j
++) {
1546 map_vector_to_txq(adapter
, i
, txr_idx
);
1557 * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1558 * @adapter: board private structure
1560 * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1561 * interrupts from the kernel.
1563 static int ixgbe_request_msix_irqs(struct ixgbe_adapter
*adapter
)
1565 struct net_device
*netdev
= adapter
->netdev
;
1566 irqreturn_t (*handler
)(int, void *);
1567 int i
, vector
, q_vectors
, err
;
1570 /* Decrement for Other and TCP Timer vectors */
1571 q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1573 /* Map the Tx/Rx rings to the vectors we were allotted. */
1574 err
= ixgbe_map_rings_to_vectors(adapter
, q_vectors
);
1578 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1579 (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1580 &ixgbe_msix_clean_many)
1581 for (vector
= 0; vector
< q_vectors
; vector
++) {
1582 handler
= SET_HANDLER(adapter
->q_vector
[vector
]);
1584 if(handler
== &ixgbe_msix_clean_rx
) {
1585 sprintf(adapter
->name
[vector
], "%s-%s-%d",
1586 netdev
->name
, "rx", ri
++);
1588 else if(handler
== &ixgbe_msix_clean_tx
) {
1589 sprintf(adapter
->name
[vector
], "%s-%s-%d",
1590 netdev
->name
, "tx", ti
++);
1593 sprintf(adapter
->name
[vector
], "%s-%s-%d",
1594 netdev
->name
, "TxRx", vector
);
1596 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1597 handler
, 0, adapter
->name
[vector
],
1598 adapter
->q_vector
[vector
]);
1601 "request_irq failed for MSIX interrupt "
1602 "Error: %d\n", err
);
1603 goto free_queue_irqs
;
1607 sprintf(adapter
->name
[vector
], "%s:lsc", netdev
->name
);
1608 err
= request_irq(adapter
->msix_entries
[vector
].vector
,
1609 &ixgbe_msix_lsc
, 0, adapter
->name
[vector
], netdev
);
1612 "request_irq for msix_lsc failed: %d\n", err
);
1613 goto free_queue_irqs
;
1619 for (i
= vector
- 1; i
>= 0; i
--)
1620 free_irq(adapter
->msix_entries
[--vector
].vector
,
1621 adapter
->q_vector
[i
]);
1622 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
1623 pci_disable_msix(adapter
->pdev
);
1624 kfree(adapter
->msix_entries
);
1625 adapter
->msix_entries
= NULL
;
1630 static void ixgbe_set_itr(struct ixgbe_adapter
*adapter
)
1632 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
1634 u32 new_itr
= q_vector
->eitr
;
1635 struct ixgbe_ring
*rx_ring
= &adapter
->rx_ring
[0];
1636 struct ixgbe_ring
*tx_ring
= &adapter
->tx_ring
[0];
1638 q_vector
->tx_itr
= ixgbe_update_itr(adapter
, new_itr
,
1640 tx_ring
->total_packets
,
1641 tx_ring
->total_bytes
);
1642 q_vector
->rx_itr
= ixgbe_update_itr(adapter
, new_itr
,
1644 rx_ring
->total_packets
,
1645 rx_ring
->total_bytes
);
1647 current_itr
= max(q_vector
->rx_itr
, q_vector
->tx_itr
);
1649 switch (current_itr
) {
1650 /* counts and packets in update_itr are dependent on these numbers */
1651 case lowest_latency
:
1655 new_itr
= 20000; /* aka hwitr = ~200 */
1664 if (new_itr
!= q_vector
->eitr
) {
1665 /* do an exponential smoothing */
1666 new_itr
= ((q_vector
->eitr
* 90)/100) + ((new_itr
* 10)/100);
1668 /* save the algorithm value here, not the smoothed one */
1669 q_vector
->eitr
= new_itr
;
1671 ixgbe_write_eitr(q_vector
);
1678 * ixgbe_irq_enable - Enable default interrupt generation settings
1679 * @adapter: board private structure
1681 static inline void ixgbe_irq_enable(struct ixgbe_adapter
*adapter
)
1685 mask
= (IXGBE_EIMS_ENABLE_MASK
& ~IXGBE_EIMS_RTX_QUEUE
);
1686 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
)
1687 mask
|= IXGBE_EIMS_GPI_SDP1
;
1688 if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
1689 mask
|= IXGBE_EIMS_ECC
;
1690 mask
|= IXGBE_EIMS_GPI_SDP1
;
1691 mask
|= IXGBE_EIMS_GPI_SDP2
;
1693 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
1694 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
1695 mask
|= IXGBE_EIMS_FLOW_DIR
;
1697 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMS
, mask
);
1698 ixgbe_irq_enable_queues(adapter
, ~0);
1699 IXGBE_WRITE_FLUSH(&adapter
->hw
);
1703 * ixgbe_intr - legacy mode Interrupt Handler
1704 * @irq: interrupt number
1705 * @data: pointer to a network interface device structure
1707 static irqreturn_t
ixgbe_intr(int irq
, void *data
)
1709 struct net_device
*netdev
= data
;
1710 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
1711 struct ixgbe_hw
*hw
= &adapter
->hw
;
1712 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[0];
1716 * Workaround for silicon errata. Mask the interrupts
1717 * before the read of EICR.
1719 IXGBE_WRITE_REG(hw
, IXGBE_EIMC
, IXGBE_IRQ_CLEAR_MASK
);
1721 /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1722 * therefore no explict interrupt disable is necessary */
1723 eicr
= IXGBE_READ_REG(hw
, IXGBE_EICR
);
1725 /* shared interrupt alert!
1726 * make sure interrupts are enabled because the read will
1727 * have disabled interrupts due to EIAM */
1728 ixgbe_irq_enable(adapter
);
1729 return IRQ_NONE
; /* Not our interrupt */
1732 if (eicr
& IXGBE_EICR_LSC
)
1733 ixgbe_check_lsc(adapter
);
1735 if (hw
->mac
.type
== ixgbe_mac_82599EB
)
1736 ixgbe_check_sfp_event(adapter
, eicr
);
1738 ixgbe_check_fan_failure(adapter
, eicr
);
1740 if (napi_schedule_prep(&(q_vector
->napi
))) {
1741 adapter
->tx_ring
[0].total_packets
= 0;
1742 adapter
->tx_ring
[0].total_bytes
= 0;
1743 adapter
->rx_ring
[0].total_packets
= 0;
1744 adapter
->rx_ring
[0].total_bytes
= 0;
1745 /* would disable interrupts here but EIAM disabled it */
1746 __napi_schedule(&(q_vector
->napi
));
1752 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter
*adapter
)
1754 int i
, q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
1756 for (i
= 0; i
< q_vectors
; i
++) {
1757 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[i
];
1758 bitmap_zero(q_vector
->rxr_idx
, MAX_RX_QUEUES
);
1759 bitmap_zero(q_vector
->txr_idx
, MAX_TX_QUEUES
);
1760 q_vector
->rxr_count
= 0;
1761 q_vector
->txr_count
= 0;
1766 * ixgbe_request_irq - initialize interrupts
1767 * @adapter: board private structure
1769 * Attempts to configure interrupts using the best available
1770 * capabilities of the hardware and kernel.
1772 static int ixgbe_request_irq(struct ixgbe_adapter
*adapter
)
1774 struct net_device
*netdev
= adapter
->netdev
;
1777 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
1778 err
= ixgbe_request_msix_irqs(adapter
);
1779 } else if (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
) {
1780 err
= request_irq(adapter
->pdev
->irq
, &ixgbe_intr
, 0,
1781 netdev
->name
, netdev
);
1783 err
= request_irq(adapter
->pdev
->irq
, &ixgbe_intr
, IRQF_SHARED
,
1784 netdev
->name
, netdev
);
1788 DPRINTK(PROBE
, ERR
, "request_irq failed, Error %d\n", err
);
1793 static void ixgbe_free_irq(struct ixgbe_adapter
*adapter
)
1795 struct net_device
*netdev
= adapter
->netdev
;
1797 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
1800 q_vectors
= adapter
->num_msix_vectors
;
1803 free_irq(adapter
->msix_entries
[i
].vector
, netdev
);
1806 for (; i
>= 0; i
--) {
1807 free_irq(adapter
->msix_entries
[i
].vector
,
1808 adapter
->q_vector
[i
]);
1811 ixgbe_reset_q_vectors(adapter
);
1813 free_irq(adapter
->pdev
->irq
, netdev
);
1818 * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1819 * @adapter: board private structure
1821 static inline void ixgbe_irq_disable(struct ixgbe_adapter
*adapter
)
1823 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
1824 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, ~0);
1826 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC
, 0xFFFF0000);
1827 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(0), ~0);
1828 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_EIMC_EX(1), ~0);
1830 IXGBE_WRITE_FLUSH(&adapter
->hw
);
1831 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
1833 for (i
= 0; i
< adapter
->num_msix_vectors
; i
++)
1834 synchronize_irq(adapter
->msix_entries
[i
].vector
);
1836 synchronize_irq(adapter
->pdev
->irq
);
1841 * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1844 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter
*adapter
)
1846 struct ixgbe_hw
*hw
= &adapter
->hw
;
1848 IXGBE_WRITE_REG(hw
, IXGBE_EITR(0),
1849 EITR_INTS_PER_SEC_TO_REG(adapter
->eitr_param
));
1851 ixgbe_set_ivar(adapter
, 0, 0, 0);
1852 ixgbe_set_ivar(adapter
, 1, 0, 0);
1854 map_vector_to_rxq(adapter
, 0, 0);
1855 map_vector_to_txq(adapter
, 0, 0);
1857 DPRINTK(HW
, INFO
, "Legacy interrupt IVAR setup done\n");
1861 * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1862 * @adapter: board private structure
1864 * Configure the Tx unit of the MAC after a reset.
1866 static void ixgbe_configure_tx(struct ixgbe_adapter
*adapter
)
1869 struct ixgbe_hw
*hw
= &adapter
->hw
;
1870 u32 i
, j
, tdlen
, txctrl
;
1872 /* Setup the HW Tx Head and Tail descriptor pointers */
1873 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
1874 struct ixgbe_ring
*ring
= &adapter
->tx_ring
[i
];
1877 tdlen
= ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
1878 IXGBE_WRITE_REG(hw
, IXGBE_TDBAL(j
),
1879 (tdba
& DMA_BIT_MASK(32)));
1880 IXGBE_WRITE_REG(hw
, IXGBE_TDBAH(j
), (tdba
>> 32));
1881 IXGBE_WRITE_REG(hw
, IXGBE_TDLEN(j
), tdlen
);
1882 IXGBE_WRITE_REG(hw
, IXGBE_TDH(j
), 0);
1883 IXGBE_WRITE_REG(hw
, IXGBE_TDT(j
), 0);
1884 adapter
->tx_ring
[i
].head
= IXGBE_TDH(j
);
1885 adapter
->tx_ring
[i
].tail
= IXGBE_TDT(j
);
1886 /* Disable Tx Head Writeback RO bit, since this hoses
1887 * bookkeeping if things aren't delivered in order.
1889 txctrl
= IXGBE_READ_REG(hw
, IXGBE_DCA_TXCTRL(j
));
1890 txctrl
&= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN
;
1891 IXGBE_WRITE_REG(hw
, IXGBE_DCA_TXCTRL(j
), txctrl
);
1893 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
1894 /* We enable 8 traffic classes, DCB only */
1895 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)
1896 IXGBE_WRITE_REG(hw
, IXGBE_MTQC
, (IXGBE_MTQC_RT_ENA
|
1897 IXGBE_MTQC_8TC_8TQ
));
1901 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1903 static void ixgbe_configure_srrctl(struct ixgbe_adapter
*adapter
,
1904 struct ixgbe_ring
*rx_ring
)
1908 struct ixgbe_ring_feature
*feature
= adapter
->ring_feature
;
1910 index
= rx_ring
->reg_idx
;
1911 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
1913 mask
= (unsigned long) feature
[RING_F_RSS
].mask
;
1914 index
= index
& mask
;
1916 srrctl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_SRRCTL(index
));
1918 srrctl
&= ~IXGBE_SRRCTL_BSIZEHDR_MASK
;
1919 srrctl
&= ~IXGBE_SRRCTL_BSIZEPKT_MASK
;
1921 srrctl
|= (IXGBE_RX_HDR_SIZE
<< IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT
) &
1922 IXGBE_SRRCTL_BSIZEHDR_MASK
;
1924 if (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
) {
1925 #if (PAGE_SIZE / 2) > IXGBE_MAX_RXBUFFER
1926 srrctl
|= IXGBE_MAX_RXBUFFER
>> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
1928 srrctl
|= (PAGE_SIZE
/ 2) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
1930 srrctl
|= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS
;
1932 srrctl
|= ALIGN(rx_ring
->rx_buf_len
, 1024) >>
1933 IXGBE_SRRCTL_BSIZEPKT_SHIFT
;
1934 srrctl
|= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF
;
1937 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_SRRCTL(index
), srrctl
);
1940 static u32
ixgbe_setup_mrqc(struct ixgbe_adapter
*adapter
)
1945 if (!(adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
))
1948 mask
= adapter
->flags
& (IXGBE_FLAG_RSS_ENABLED
1949 #ifdef CONFIG_IXGBE_DCB
1950 | IXGBE_FLAG_DCB_ENABLED
1955 case (IXGBE_FLAG_RSS_ENABLED
):
1956 mrqc
= IXGBE_MRQC_RSSEN
;
1958 #ifdef CONFIG_IXGBE_DCB
1959 case (IXGBE_FLAG_DCB_ENABLED
):
1960 mrqc
= IXGBE_MRQC_RT8TCEN
;
1962 #endif /* CONFIG_IXGBE_DCB */
1971 * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1972 * @adapter: board private structure
1974 * Configure the Rx unit of the MAC after a reset.
1976 static void ixgbe_configure_rx(struct ixgbe_adapter
*adapter
)
1979 struct ixgbe_hw
*hw
= &adapter
->hw
;
1980 struct ixgbe_ring
*rx_ring
;
1981 struct net_device
*netdev
= adapter
->netdev
;
1982 int max_frame
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
1984 u32 rdlen
, rxctrl
, rxcsum
;
1985 static const u32 seed
[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1986 0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1987 0x6A3E67EA, 0x14364D17, 0x3BED200D};
1989 u32 reta
= 0, mrqc
= 0;
1994 /* Decide whether to use packet split mode or not */
1995 adapter
->flags
|= IXGBE_FLAG_RX_PS_ENABLED
;
1997 /* Set the RX buffer length according to the mode */
1998 if (adapter
->flags
& IXGBE_FLAG_RX_PS_ENABLED
) {
1999 rx_buf_len
= IXGBE_RX_HDR_SIZE
;
2000 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
2001 /* PSRTYPE must be initialized in 82599 */
2002 u32 psrtype
= IXGBE_PSRTYPE_TCPHDR
|
2003 IXGBE_PSRTYPE_UDPHDR
|
2004 IXGBE_PSRTYPE_IPV4HDR
|
2005 IXGBE_PSRTYPE_IPV6HDR
|
2006 IXGBE_PSRTYPE_L2HDR
;
2007 IXGBE_WRITE_REG(hw
, IXGBE_PSRTYPE(0), psrtype
);
2010 if (!(adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
) &&
2011 (netdev
->mtu
<= ETH_DATA_LEN
))
2012 rx_buf_len
= MAXIMUM_ETHERNET_VLAN_SIZE
;
2014 rx_buf_len
= ALIGN(max_frame
, 1024);
2017 fctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_FCTRL
);
2018 fctrl
|= IXGBE_FCTRL_BAM
;
2019 fctrl
|= IXGBE_FCTRL_DPF
; /* discard pause frames when FC enabled */
2020 fctrl
|= IXGBE_FCTRL_PMCF
;
2021 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_FCTRL
, fctrl
);
2023 hlreg0
= IXGBE_READ_REG(hw
, IXGBE_HLREG0
);
2024 if (adapter
->netdev
->mtu
<= ETH_DATA_LEN
)
2025 hlreg0
&= ~IXGBE_HLREG0_JUMBOEN
;
2027 hlreg0
|= IXGBE_HLREG0_JUMBOEN
;
2029 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)
2030 hlreg0
|= IXGBE_HLREG0_JUMBOEN
;
2032 IXGBE_WRITE_REG(hw
, IXGBE_HLREG0
, hlreg0
);
2034 rdlen
= adapter
->rx_ring
[0].count
* sizeof(union ixgbe_adv_rx_desc
);
2035 /* disable receives while setting up the descriptors */
2036 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
2037 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
2040 * Setup the HW Rx Head and Tail Descriptor Pointers and
2041 * the Base and Length of the Rx Descriptor Ring
2043 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2044 rx_ring
= &adapter
->rx_ring
[i
];
2045 rdba
= rx_ring
->dma
;
2046 j
= rx_ring
->reg_idx
;
2047 IXGBE_WRITE_REG(hw
, IXGBE_RDBAL(j
), (rdba
& DMA_BIT_MASK(32)));
2048 IXGBE_WRITE_REG(hw
, IXGBE_RDBAH(j
), (rdba
>> 32));
2049 IXGBE_WRITE_REG(hw
, IXGBE_RDLEN(j
), rdlen
);
2050 IXGBE_WRITE_REG(hw
, IXGBE_RDH(j
), 0);
2051 IXGBE_WRITE_REG(hw
, IXGBE_RDT(j
), 0);
2052 rx_ring
->head
= IXGBE_RDH(j
);
2053 rx_ring
->tail
= IXGBE_RDT(j
);
2054 rx_ring
->rx_buf_len
= rx_buf_len
;
2056 if (adapter
->flags
& IXGBE_FLAG_RX_PS_ENABLED
)
2057 rx_ring
->flags
|= IXGBE_RING_RX_PS_ENABLED
;
2060 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) {
2061 struct ixgbe_ring_feature
*f
;
2062 f
= &adapter
->ring_feature
[RING_F_FCOE
];
2063 if ((i
>= f
->mask
) && (i
< f
->mask
+ f
->indices
)) {
2064 rx_ring
->flags
&= ~IXGBE_RING_RX_PS_ENABLED
;
2065 if (rx_buf_len
< IXGBE_FCOE_JUMBO_FRAME_SIZE
)
2066 rx_ring
->rx_buf_len
=
2067 IXGBE_FCOE_JUMBO_FRAME_SIZE
;
2071 #endif /* IXGBE_FCOE */
2072 ixgbe_configure_srrctl(adapter
, rx_ring
);
2075 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
2077 * For VMDq support of different descriptor types or
2078 * buffer sizes through the use of multiple SRRCTL
2079 * registers, RDRXCTL.MVMEN must be set to 1
2081 * also, the manual doesn't mention it clearly but DCA hints
2082 * will only use queue 0's tags unless this bit is set. Side
2083 * effects of setting this bit are only that SRRCTL must be
2084 * fully programmed [0..15]
2086 rdrxctl
= IXGBE_READ_REG(hw
, IXGBE_RDRXCTL
);
2087 rdrxctl
|= IXGBE_RDRXCTL_MVMEN
;
2088 IXGBE_WRITE_REG(hw
, IXGBE_RDRXCTL
, rdrxctl
);
2091 /* Program MRQC for the distribution of queues */
2092 mrqc
= ixgbe_setup_mrqc(adapter
);
2094 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
) {
2095 /* Fill out redirection table */
2096 for (i
= 0, j
= 0; i
< 128; i
++, j
++) {
2097 if (j
== adapter
->ring_feature
[RING_F_RSS
].indices
)
2099 /* reta = 4-byte sliding window of
2100 * 0x00..(indices-1)(indices-1)00..etc. */
2101 reta
= (reta
<< 8) | (j
* 0x11);
2103 IXGBE_WRITE_REG(hw
, IXGBE_RETA(i
>> 2), reta
);
2106 /* Fill out hash function seeds */
2107 for (i
= 0; i
< 10; i
++)
2108 IXGBE_WRITE_REG(hw
, IXGBE_RSSRK(i
), seed
[i
]);
2110 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
2111 mrqc
|= IXGBE_MRQC_RSSEN
;
2112 /* Perform hash on these packet types */
2113 mrqc
|= IXGBE_MRQC_RSS_FIELD_IPV4
2114 | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
2115 | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
2116 | IXGBE_MRQC_RSS_FIELD_IPV6
2117 | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
2118 | IXGBE_MRQC_RSS_FIELD_IPV6_UDP
;
2120 IXGBE_WRITE_REG(hw
, IXGBE_MRQC
, mrqc
);
2122 rxcsum
= IXGBE_READ_REG(hw
, IXGBE_RXCSUM
);
2124 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
||
2125 adapter
->flags
& IXGBE_FLAG_RX_CSUM_ENABLED
) {
2126 /* Disable indicating checksum in descriptor, enables
2128 rxcsum
|= IXGBE_RXCSUM_PCSD
;
2130 if (!(rxcsum
& IXGBE_RXCSUM_PCSD
)) {
2131 /* Enable IPv4 payload checksum for UDP fragments
2132 * if PCSD is not set */
2133 rxcsum
|= IXGBE_RXCSUM_IPPCSE
;
2136 IXGBE_WRITE_REG(hw
, IXGBE_RXCSUM
, rxcsum
);
2138 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
2139 rdrxctl
= IXGBE_READ_REG(hw
, IXGBE_RDRXCTL
);
2140 rdrxctl
|= IXGBE_RDRXCTL_CRCSTRIP
;
2141 rdrxctl
&= ~IXGBE_RDRXCTL_RSCFRSTSIZE
;
2142 IXGBE_WRITE_REG(hw
, IXGBE_RDRXCTL
, rdrxctl
);
2145 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
) {
2146 /* Enable 82599 HW-RSC */
2147 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2148 rx_ring
= &adapter
->rx_ring
[i
];
2149 j
= rx_ring
->reg_idx
;
2150 rscctrl
= IXGBE_READ_REG(hw
, IXGBE_RSCCTL(j
));
2151 rscctrl
|= IXGBE_RSCCTL_RSCEN
;
2153 * we must limit the number of descriptors so that the
2154 * total size of max desc * buf_len is not greater
2157 if (rx_ring
->flags
& IXGBE_RING_RX_PS_ENABLED
) {
2158 #if (MAX_SKB_FRAGS > 16)
2159 rscctrl
|= IXGBE_RSCCTL_MAXDESC_16
;
2160 #elif (MAX_SKB_FRAGS > 8)
2161 rscctrl
|= IXGBE_RSCCTL_MAXDESC_8
;
2162 #elif (MAX_SKB_FRAGS > 4)
2163 rscctrl
|= IXGBE_RSCCTL_MAXDESC_4
;
2165 rscctrl
|= IXGBE_RSCCTL_MAXDESC_1
;
2168 if (rx_buf_len
< IXGBE_RXBUFFER_4096
)
2169 rscctrl
|= IXGBE_RSCCTL_MAXDESC_16
;
2170 else if (rx_buf_len
< IXGBE_RXBUFFER_8192
)
2171 rscctrl
|= IXGBE_RSCCTL_MAXDESC_8
;
2173 rscctrl
|= IXGBE_RSCCTL_MAXDESC_4
;
2175 IXGBE_WRITE_REG(hw
, IXGBE_RSCCTL(j
), rscctrl
);
2177 /* Disable RSC for ACK packets */
2178 IXGBE_WRITE_REG(hw
, IXGBE_RSCDBU
,
2179 (IXGBE_RSCDBU_RSCACKDIS
| IXGBE_READ_REG(hw
, IXGBE_RSCDBU
)));
2183 static void ixgbe_vlan_rx_add_vid(struct net_device
*netdev
, u16 vid
)
2185 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
2186 struct ixgbe_hw
*hw
= &adapter
->hw
;
2188 /* add VID to filter table */
2189 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, 0, true);
2192 static void ixgbe_vlan_rx_kill_vid(struct net_device
*netdev
, u16 vid
)
2194 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
2195 struct ixgbe_hw
*hw
= &adapter
->hw
;
2197 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2198 ixgbe_irq_disable(adapter
);
2200 vlan_group_set_device(adapter
->vlgrp
, vid
, NULL
);
2202 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2203 ixgbe_irq_enable(adapter
);
2205 /* remove VID from filter table */
2206 hw
->mac
.ops
.set_vfta(&adapter
->hw
, vid
, 0, false);
2209 static void ixgbe_vlan_rx_register(struct net_device
*netdev
,
2210 struct vlan_group
*grp
)
2212 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
2216 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2217 ixgbe_irq_disable(adapter
);
2218 adapter
->vlgrp
= grp
;
2221 * For a DCB driver, always enable VLAN tag stripping so we can
2222 * still receive traffic from a DCB-enabled host even if we're
2225 ctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_VLNCTRL
);
2226 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
2227 ctrl
|= IXGBE_VLNCTRL_VME
| IXGBE_VLNCTRL_VFE
;
2228 ctrl
&= ~IXGBE_VLNCTRL_CFIEN
;
2229 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_VLNCTRL
, ctrl
);
2230 } else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
2231 ctrl
|= IXGBE_VLNCTRL_VFE
;
2232 /* enable VLAN tag insert/strip */
2233 ctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_VLNCTRL
);
2234 ctrl
&= ~IXGBE_VLNCTRL_CFIEN
;
2235 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_VLNCTRL
, ctrl
);
2236 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2237 j
= adapter
->rx_ring
[i
].reg_idx
;
2238 ctrl
= IXGBE_READ_REG(&adapter
->hw
, IXGBE_RXDCTL(j
));
2239 ctrl
|= IXGBE_RXDCTL_VME
;
2240 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_RXDCTL(j
), ctrl
);
2243 ixgbe_vlan_rx_add_vid(netdev
, 0);
2245 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2246 ixgbe_irq_enable(adapter
);
2249 static void ixgbe_restore_vlan(struct ixgbe_adapter
*adapter
)
2251 ixgbe_vlan_rx_register(adapter
->netdev
, adapter
->vlgrp
);
2253 if (adapter
->vlgrp
) {
2255 for (vid
= 0; vid
< VLAN_GROUP_ARRAY_LEN
; vid
++) {
2256 if (!vlan_group_get_device(adapter
->vlgrp
, vid
))
2258 ixgbe_vlan_rx_add_vid(adapter
->netdev
, vid
);
2263 static u8
*ixgbe_addr_list_itr(struct ixgbe_hw
*hw
, u8
**mc_addr_ptr
, u32
*vmdq
)
2265 struct dev_mc_list
*mc_ptr
;
2266 u8
*addr
= *mc_addr_ptr
;
2269 mc_ptr
= container_of(addr
, struct dev_mc_list
, dmi_addr
[0]);
2271 *mc_addr_ptr
= mc_ptr
->next
->dmi_addr
;
2273 *mc_addr_ptr
= NULL
;
2279 * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
2280 * @netdev: network interface device structure
2282 * The set_rx_method entry point is called whenever the unicast/multicast
2283 * address list or the network interface flags are updated. This routine is
2284 * responsible for configuring the hardware for proper unicast, multicast and
2287 static void ixgbe_set_rx_mode(struct net_device
*netdev
)
2289 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
2290 struct ixgbe_hw
*hw
= &adapter
->hw
;
2292 u8
*addr_list
= NULL
;
2295 /* Check for Promiscuous and All Multicast modes */
2297 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
2298 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
2300 if (netdev
->flags
& IFF_PROMISC
) {
2301 hw
->addr_ctrl
.user_set_promisc
= 1;
2302 fctrl
|= (IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
2303 vlnctrl
&= ~IXGBE_VLNCTRL_VFE
;
2305 if (netdev
->flags
& IFF_ALLMULTI
) {
2306 fctrl
|= IXGBE_FCTRL_MPE
;
2307 fctrl
&= ~IXGBE_FCTRL_UPE
;
2309 fctrl
&= ~(IXGBE_FCTRL_UPE
| IXGBE_FCTRL_MPE
);
2311 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
2312 hw
->addr_ctrl
.user_set_promisc
= 0;
2315 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
2316 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
2318 /* reprogram secondary unicast list */
2319 hw
->mac
.ops
.update_uc_addr_list(hw
, &netdev
->uc
.list
);
2321 /* reprogram multicast list */
2322 addr_count
= netdev
->mc_count
;
2324 addr_list
= netdev
->mc_list
->dmi_addr
;
2325 hw
->mac
.ops
.update_mc_addr_list(hw
, addr_list
, addr_count
,
2326 ixgbe_addr_list_itr
);
2329 static void ixgbe_napi_enable_all(struct ixgbe_adapter
*adapter
)
2332 struct ixgbe_q_vector
*q_vector
;
2333 int q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
2335 /* legacy and MSI only use one vector */
2336 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
))
2339 for (q_idx
= 0; q_idx
< q_vectors
; q_idx
++) {
2340 struct napi_struct
*napi
;
2341 q_vector
= adapter
->q_vector
[q_idx
];
2342 napi
= &q_vector
->napi
;
2343 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
2344 if (!q_vector
->rxr_count
|| !q_vector
->txr_count
) {
2345 if (q_vector
->txr_count
== 1)
2346 napi
->poll
= &ixgbe_clean_txonly
;
2347 else if (q_vector
->rxr_count
== 1)
2348 napi
->poll
= &ixgbe_clean_rxonly
;
2356 static void ixgbe_napi_disable_all(struct ixgbe_adapter
*adapter
)
2359 struct ixgbe_q_vector
*q_vector
;
2360 int q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
2362 /* legacy and MSI only use one vector */
2363 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
))
2366 for (q_idx
= 0; q_idx
< q_vectors
; q_idx
++) {
2367 q_vector
= adapter
->q_vector
[q_idx
];
2368 napi_disable(&q_vector
->napi
);
2372 #ifdef CONFIG_IXGBE_DCB
2374 * ixgbe_configure_dcb - Configure DCB hardware
2375 * @adapter: ixgbe adapter struct
2377 * This is called by the driver on open to configure the DCB hardware.
2378 * This is also called by the gennetlink interface when reconfiguring
2381 static void ixgbe_configure_dcb(struct ixgbe_adapter
*adapter
)
2383 struct ixgbe_hw
*hw
= &adapter
->hw
;
2384 u32 txdctl
, vlnctrl
;
2387 ixgbe_dcb_check_config(&adapter
->dcb_cfg
);
2388 ixgbe_dcb_calculate_tc_credits(&adapter
->dcb_cfg
, DCB_TX_CONFIG
);
2389 ixgbe_dcb_calculate_tc_credits(&adapter
->dcb_cfg
, DCB_RX_CONFIG
);
2391 /* reconfigure the hardware */
2392 ixgbe_dcb_hw_config(&adapter
->hw
, &adapter
->dcb_cfg
);
2394 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2395 j
= adapter
->tx_ring
[i
].reg_idx
;
2396 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(j
));
2397 /* PThresh workaround for Tx hang with DFP enabled. */
2399 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(j
), txdctl
);
2401 /* Enable VLAN tag insert/strip */
2402 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_VLNCTRL
);
2403 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
2404 vlnctrl
|= IXGBE_VLNCTRL_VME
| IXGBE_VLNCTRL_VFE
;
2405 vlnctrl
&= ~IXGBE_VLNCTRL_CFIEN
;
2406 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
2407 } else if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
2408 vlnctrl
|= IXGBE_VLNCTRL_VFE
;
2409 vlnctrl
&= ~IXGBE_VLNCTRL_CFIEN
;
2410 IXGBE_WRITE_REG(hw
, IXGBE_VLNCTRL
, vlnctrl
);
2411 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
2412 j
= adapter
->rx_ring
[i
].reg_idx
;
2413 vlnctrl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
2414 vlnctrl
|= IXGBE_RXDCTL_VME
;
2415 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), vlnctrl
);
2418 hw
->mac
.ops
.set_vfta(&adapter
->hw
, 0, 0, true);
2422 static void ixgbe_configure(struct ixgbe_adapter
*adapter
)
2424 struct net_device
*netdev
= adapter
->netdev
;
2425 struct ixgbe_hw
*hw
= &adapter
->hw
;
2428 ixgbe_set_rx_mode(netdev
);
2430 ixgbe_restore_vlan(adapter
);
2431 #ifdef CONFIG_IXGBE_DCB
2432 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
2433 netif_set_gso_max_size(netdev
, 32768);
2434 ixgbe_configure_dcb(adapter
);
2436 netif_set_gso_max_size(netdev
, 65536);
2439 netif_set_gso_max_size(netdev
, 65536);
2443 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)
2444 ixgbe_configure_fcoe(adapter
);
2446 #endif /* IXGBE_FCOE */
2447 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) {
2448 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2449 adapter
->tx_ring
[i
].atr_sample_rate
=
2450 adapter
->atr_sample_rate
;
2451 ixgbe_init_fdir_signature_82599(hw
, adapter
->fdir_pballoc
);
2452 } else if (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
) {
2453 ixgbe_init_fdir_perfect_82599(hw
, adapter
->fdir_pballoc
);
2456 ixgbe_configure_tx(adapter
);
2457 ixgbe_configure_rx(adapter
);
2458 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2459 ixgbe_alloc_rx_buffers(adapter
, &adapter
->rx_ring
[i
],
2460 (adapter
->rx_ring
[i
].count
- 1));
2463 static inline bool ixgbe_is_sfp(struct ixgbe_hw
*hw
)
2465 switch (hw
->phy
.type
) {
2466 case ixgbe_phy_sfp_avago
:
2467 case ixgbe_phy_sfp_ftl
:
2468 case ixgbe_phy_sfp_intel
:
2469 case ixgbe_phy_sfp_unknown
:
2470 case ixgbe_phy_tw_tyco
:
2471 case ixgbe_phy_tw_unknown
:
2479 * ixgbe_sfp_link_config - set up SFP+ link
2480 * @adapter: pointer to private adapter struct
2482 static void ixgbe_sfp_link_config(struct ixgbe_adapter
*adapter
)
2484 struct ixgbe_hw
*hw
= &adapter
->hw
;
2486 if (hw
->phy
.multispeed_fiber
) {
2488 * In multispeed fiber setups, the device may not have
2489 * had a physical connection when the driver loaded.
2490 * If that's the case, the initial link configuration
2491 * couldn't get the MAC into 10G or 1G mode, so we'll
2492 * never have a link status change interrupt fire.
2493 * We need to try and force an autonegotiation
2494 * session, then bring up link.
2496 hw
->mac
.ops
.setup_sfp(hw
);
2497 if (!(adapter
->flags
& IXGBE_FLAG_IN_SFP_LINK_TASK
))
2498 schedule_work(&adapter
->multispeed_fiber_task
);
2501 * Direct Attach Cu and non-multispeed fiber modules
2502 * still need to be configured properly prior to
2505 if (!(adapter
->flags
& IXGBE_FLAG_IN_SFP_MOD_TASK
))
2506 schedule_work(&adapter
->sfp_config_module_task
);
2511 * ixgbe_non_sfp_link_config - set up non-SFP+ link
2512 * @hw: pointer to private hardware struct
2514 * Returns 0 on success, negative on failure
2516 static int ixgbe_non_sfp_link_config(struct ixgbe_hw
*hw
)
2519 bool link_up
= false;
2520 u32 ret
= IXGBE_ERR_LINK_SETUP
;
2522 if (hw
->mac
.ops
.check_link
)
2523 ret
= hw
->mac
.ops
.check_link(hw
, &autoneg
, &link_up
, false);
2528 if (hw
->mac
.ops
.get_link_capabilities
)
2529 ret
= hw
->mac
.ops
.get_link_capabilities(hw
, &autoneg
,
2534 if (hw
->mac
.ops
.setup_link_speed
)
2535 ret
= hw
->mac
.ops
.setup_link_speed(hw
, autoneg
, true, link_up
);
2540 #define IXGBE_MAX_RX_DESC_POLL 10
2541 static inline void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter
*adapter
,
2544 int j
= adapter
->rx_ring
[rxr
].reg_idx
;
2547 for (k
= 0; k
< IXGBE_MAX_RX_DESC_POLL
; k
++) {
2548 if (IXGBE_READ_REG(&adapter
->hw
,
2549 IXGBE_RXDCTL(j
)) & IXGBE_RXDCTL_ENABLE
)
2554 if (k
>= IXGBE_MAX_RX_DESC_POLL
) {
2555 DPRINTK(DRV
, ERR
, "RXDCTL.ENABLE on Rx queue %d "
2556 "not set within the polling period\n", rxr
);
2558 ixgbe_release_rx_desc(&adapter
->hw
, &adapter
->rx_ring
[rxr
],
2559 (adapter
->rx_ring
[rxr
].count
- 1));
2562 static int ixgbe_up_complete(struct ixgbe_adapter
*adapter
)
2564 struct net_device
*netdev
= adapter
->netdev
;
2565 struct ixgbe_hw
*hw
= &adapter
->hw
;
2567 int num_rx_rings
= adapter
->num_rx_queues
;
2569 int max_frame
= netdev
->mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
2570 u32 txdctl
, rxdctl
, mhadd
;
2574 ixgbe_get_hw_control(adapter
);
2576 if ((adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) ||
2577 (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
)) {
2578 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
2579 gpie
= (IXGBE_GPIE_MSIX_MODE
| IXGBE_GPIE_EIAME
|
2580 IXGBE_GPIE_PBA_SUPPORT
| IXGBE_GPIE_OCD
);
2585 /* XXX: to interrupt immediately for EICS writes, enable this */
2586 /* gpie |= IXGBE_GPIE_EIMEN; */
2587 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
2590 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
2591 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
2592 * specifically only auto mask tx and rx interrupts */
2593 IXGBE_WRITE_REG(hw
, IXGBE_EIAM
, IXGBE_EICS_RTX_QUEUE
);
2596 /* Enable fan failure interrupt if media type is copper */
2597 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
2598 gpie
= IXGBE_READ_REG(hw
, IXGBE_GPIE
);
2599 gpie
|= IXGBE_SDP1_GPIEN
;
2600 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
2603 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
2604 gpie
= IXGBE_READ_REG(hw
, IXGBE_GPIE
);
2605 gpie
|= IXGBE_SDP1_GPIEN
;
2606 gpie
|= IXGBE_SDP2_GPIEN
;
2607 IXGBE_WRITE_REG(hw
, IXGBE_GPIE
, gpie
);
2611 /* adjust max frame to be able to do baby jumbo for FCoE */
2612 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) &&
2613 (max_frame
< IXGBE_FCOE_JUMBO_FRAME_SIZE
))
2614 max_frame
= IXGBE_FCOE_JUMBO_FRAME_SIZE
;
2616 #endif /* IXGBE_FCOE */
2617 mhadd
= IXGBE_READ_REG(hw
, IXGBE_MHADD
);
2618 if (max_frame
!= (mhadd
>> IXGBE_MHADD_MFS_SHIFT
)) {
2619 mhadd
&= ~IXGBE_MHADD_MFS_MASK
;
2620 mhadd
|= max_frame
<< IXGBE_MHADD_MFS_SHIFT
;
2622 IXGBE_WRITE_REG(hw
, IXGBE_MHADD
, mhadd
);
2625 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2626 j
= adapter
->tx_ring
[i
].reg_idx
;
2627 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(j
));
2628 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
2629 txdctl
|= (8 << 16);
2630 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(j
), txdctl
);
2633 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
2634 /* DMATXCTL.EN must be set after all Tx queue config is done */
2635 dmatxctl
= IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
);
2636 dmatxctl
|= IXGBE_DMATXCTL_TE
;
2637 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
, dmatxctl
);
2639 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2640 j
= adapter
->tx_ring
[i
].reg_idx
;
2641 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(j
));
2642 txdctl
|= IXGBE_TXDCTL_ENABLE
;
2643 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(j
), txdctl
);
2646 for (i
= 0; i
< num_rx_rings
; i
++) {
2647 j
= adapter
->rx_ring
[i
].reg_idx
;
2648 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXDCTL(j
));
2649 /* enable PTHRESH=32 descriptors (half the internal cache)
2650 * and HTHRESH=0 descriptors (to minimize latency on fetch),
2651 * this also removes a pesky rx_no_buffer_count increment */
2653 rxdctl
|= IXGBE_RXDCTL_ENABLE
;
2654 IXGBE_WRITE_REG(hw
, IXGBE_RXDCTL(j
), rxdctl
);
2655 if (hw
->mac
.type
== ixgbe_mac_82599EB
)
2656 ixgbe_rx_desc_queue_enable(adapter
, i
);
2658 /* enable all receives */
2659 rxdctl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
2660 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
2661 rxdctl
|= (IXGBE_RXCTRL_DMBYPS
| IXGBE_RXCTRL_RXEN
);
2663 rxdctl
|= IXGBE_RXCTRL_RXEN
;
2664 hw
->mac
.ops
.enable_rx_dma(hw
, rxdctl
);
2666 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
2667 ixgbe_configure_msix(adapter
);
2669 ixgbe_configure_msi_and_legacy(adapter
);
2671 clear_bit(__IXGBE_DOWN
, &adapter
->state
);
2672 ixgbe_napi_enable_all(adapter
);
2674 /* clear any pending interrupts, may auto mask */
2675 IXGBE_READ_REG(hw
, IXGBE_EICR
);
2677 ixgbe_irq_enable(adapter
);
2680 * If this adapter has a fan, check to see if we had a failure
2681 * before we enabled the interrupt.
2683 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
2684 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
2685 if (esdp
& IXGBE_ESDP_SDP1
)
2687 "Fan has stopped, replace the adapter\n");
2691 * For hot-pluggable SFP+ devices, a new SFP+ module may have
2692 * arrived before interrupts were enabled but after probe. Such
2693 * devices wouldn't have their type identified yet. We need to
2694 * kick off the SFP+ module setup first, then try to bring up link.
2695 * If we're not hot-pluggable SFP+, we just need to configure link
2698 if (hw
->phy
.type
== ixgbe_phy_unknown
) {
2699 err
= hw
->phy
.ops
.identify(hw
);
2700 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
2702 * Take the device down and schedule the sfp tasklet
2703 * which will unregister_netdev and log it.
2705 ixgbe_down(adapter
);
2706 schedule_work(&adapter
->sfp_config_module_task
);
2711 if (ixgbe_is_sfp(hw
)) {
2712 ixgbe_sfp_link_config(adapter
);
2714 err
= ixgbe_non_sfp_link_config(hw
);
2716 DPRINTK(PROBE
, ERR
, "link_config FAILED %d\n", err
);
2719 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2720 set_bit(__IXGBE_FDIR_INIT_DONE
,
2721 &(adapter
->tx_ring
[i
].reinit_state
));
2723 /* enable transmits */
2724 netif_tx_start_all_queues(netdev
);
2726 /* bring the link up in the watchdog, this could race with our first
2727 * link up interrupt but shouldn't be a problem */
2728 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
2729 adapter
->link_check_timeout
= jiffies
;
2730 mod_timer(&adapter
->watchdog_timer
, jiffies
);
2734 void ixgbe_reinit_locked(struct ixgbe_adapter
*adapter
)
2736 WARN_ON(in_interrupt());
2737 while (test_and_set_bit(__IXGBE_RESETTING
, &adapter
->state
))
2739 ixgbe_down(adapter
);
2741 clear_bit(__IXGBE_RESETTING
, &adapter
->state
);
2744 int ixgbe_up(struct ixgbe_adapter
*adapter
)
2746 /* hardware has been reset, we need to reload some things */
2747 ixgbe_configure(adapter
);
2749 return ixgbe_up_complete(adapter
);
2752 void ixgbe_reset(struct ixgbe_adapter
*adapter
)
2754 struct ixgbe_hw
*hw
= &adapter
->hw
;
2757 err
= hw
->mac
.ops
.init_hw(hw
);
2760 case IXGBE_ERR_SFP_NOT_PRESENT
:
2762 case IXGBE_ERR_MASTER_REQUESTS_PENDING
:
2763 dev_err(&adapter
->pdev
->dev
, "master disable timed out\n");
2765 case IXGBE_ERR_EEPROM_VERSION
:
2766 /* We are running on a pre-production device, log a warning */
2767 dev_warn(&adapter
->pdev
->dev
, "This device is a pre-production "
2768 "adapter/LOM. Please be aware there may be issues "
2769 "associated with your hardware. If you are "
2770 "experiencing problems please contact your Intel or "
2771 "hardware representative who provided you with this "
2775 dev_err(&adapter
->pdev
->dev
, "Hardware Error: %d\n", err
);
2778 /* reprogram the RAR[0] in case user changed it. */
2779 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.addr
, 0, IXGBE_RAH_AV
);
2783 * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2784 * @adapter: board private structure
2785 * @rx_ring: ring to free buffers from
2787 static void ixgbe_clean_rx_ring(struct ixgbe_adapter
*adapter
,
2788 struct ixgbe_ring
*rx_ring
)
2790 struct pci_dev
*pdev
= adapter
->pdev
;
2794 /* Free all the Rx ring sk_buffs */
2796 for (i
= 0; i
< rx_ring
->count
; i
++) {
2797 struct ixgbe_rx_buffer
*rx_buffer_info
;
2799 rx_buffer_info
= &rx_ring
->rx_buffer_info
[i
];
2800 if (rx_buffer_info
->dma
) {
2801 pci_unmap_single(pdev
, rx_buffer_info
->dma
,
2802 rx_ring
->rx_buf_len
,
2803 PCI_DMA_FROMDEVICE
);
2804 rx_buffer_info
->dma
= 0;
2806 if (rx_buffer_info
->skb
) {
2807 struct sk_buff
*skb
= rx_buffer_info
->skb
;
2808 rx_buffer_info
->skb
= NULL
;
2810 struct sk_buff
*this = skb
;
2812 dev_kfree_skb(this);
2815 if (!rx_buffer_info
->page
)
2817 if (rx_buffer_info
->page_dma
) {
2818 pci_unmap_page(pdev
, rx_buffer_info
->page_dma
,
2819 PAGE_SIZE
/ 2, PCI_DMA_FROMDEVICE
);
2820 rx_buffer_info
->page_dma
= 0;
2822 put_page(rx_buffer_info
->page
);
2823 rx_buffer_info
->page
= NULL
;
2824 rx_buffer_info
->page_offset
= 0;
2827 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
2828 memset(rx_ring
->rx_buffer_info
, 0, size
);
2830 /* Zero out the descriptor ring */
2831 memset(rx_ring
->desc
, 0, rx_ring
->size
);
2833 rx_ring
->next_to_clean
= 0;
2834 rx_ring
->next_to_use
= 0;
2837 writel(0, adapter
->hw
.hw_addr
+ rx_ring
->head
);
2839 writel(0, adapter
->hw
.hw_addr
+ rx_ring
->tail
);
2843 * ixgbe_clean_tx_ring - Free Tx Buffers
2844 * @adapter: board private structure
2845 * @tx_ring: ring to be cleaned
2847 static void ixgbe_clean_tx_ring(struct ixgbe_adapter
*adapter
,
2848 struct ixgbe_ring
*tx_ring
)
2850 struct ixgbe_tx_buffer
*tx_buffer_info
;
2854 /* Free all the Tx ring sk_buffs */
2856 for (i
= 0; i
< tx_ring
->count
; i
++) {
2857 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
2858 ixgbe_unmap_and_free_tx_resource(adapter
, tx_buffer_info
);
2861 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
2862 memset(tx_ring
->tx_buffer_info
, 0, size
);
2864 /* Zero out the descriptor ring */
2865 memset(tx_ring
->desc
, 0, tx_ring
->size
);
2867 tx_ring
->next_to_use
= 0;
2868 tx_ring
->next_to_clean
= 0;
2871 writel(0, adapter
->hw
.hw_addr
+ tx_ring
->head
);
2873 writel(0, adapter
->hw
.hw_addr
+ tx_ring
->tail
);
2877 * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2878 * @adapter: board private structure
2880 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter
*adapter
)
2884 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
2885 ixgbe_clean_rx_ring(adapter
, &adapter
->rx_ring
[i
]);
2889 * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2890 * @adapter: board private structure
2892 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter
*adapter
)
2896 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
2897 ixgbe_clean_tx_ring(adapter
, &adapter
->tx_ring
[i
]);
2900 void ixgbe_down(struct ixgbe_adapter
*adapter
)
2902 struct net_device
*netdev
= adapter
->netdev
;
2903 struct ixgbe_hw
*hw
= &adapter
->hw
;
2908 /* signal that we are down to the interrupt handler */
2909 set_bit(__IXGBE_DOWN
, &adapter
->state
);
2911 /* disable receives */
2912 rxctrl
= IXGBE_READ_REG(hw
, IXGBE_RXCTRL
);
2913 IXGBE_WRITE_REG(hw
, IXGBE_RXCTRL
, rxctrl
& ~IXGBE_RXCTRL_RXEN
);
2915 netif_tx_disable(netdev
);
2917 IXGBE_WRITE_FLUSH(hw
);
2920 netif_tx_stop_all_queues(netdev
);
2922 ixgbe_irq_disable(adapter
);
2924 ixgbe_napi_disable_all(adapter
);
2926 del_timer_sync(&adapter
->watchdog_timer
);
2927 cancel_work_sync(&adapter
->watchdog_task
);
2929 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
2930 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
2931 cancel_work_sync(&adapter
->fdir_reinit_task
);
2933 /* disable transmits in the hardware now that interrupts are off */
2934 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
2935 j
= adapter
->tx_ring
[i
].reg_idx
;
2936 txdctl
= IXGBE_READ_REG(hw
, IXGBE_TXDCTL(j
));
2937 IXGBE_WRITE_REG(hw
, IXGBE_TXDCTL(j
),
2938 (txdctl
& ~IXGBE_TXDCTL_ENABLE
));
2940 /* Disable the Tx DMA engine on 82599 */
2941 if (hw
->mac
.type
== ixgbe_mac_82599EB
)
2942 IXGBE_WRITE_REG(hw
, IXGBE_DMATXCTL
,
2943 (IXGBE_READ_REG(hw
, IXGBE_DMATXCTL
) &
2944 ~IXGBE_DMATXCTL_TE
));
2946 netif_carrier_off(netdev
);
2948 if (!pci_channel_offline(adapter
->pdev
))
2949 ixgbe_reset(adapter
);
2950 ixgbe_clean_all_tx_rings(adapter
);
2951 ixgbe_clean_all_rx_rings(adapter
);
2953 #ifdef CONFIG_IXGBE_DCA
2954 /* since we reset the hardware DCA settings were cleared */
2955 ixgbe_setup_dca(adapter
);
2960 * ixgbe_poll - NAPI Rx polling callback
2961 * @napi: structure for representing this polling device
2962 * @budget: how many packets driver is allowed to clean
2964 * This function is used for legacy and MSI, NAPI mode
2966 static int ixgbe_poll(struct napi_struct
*napi
, int budget
)
2968 struct ixgbe_q_vector
*q_vector
=
2969 container_of(napi
, struct ixgbe_q_vector
, napi
);
2970 struct ixgbe_adapter
*adapter
= q_vector
->adapter
;
2971 int tx_clean_complete
, work_done
= 0;
2973 #ifdef CONFIG_IXGBE_DCA
2974 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
2975 ixgbe_update_tx_dca(adapter
, adapter
->tx_ring
);
2976 ixgbe_update_rx_dca(adapter
, adapter
->rx_ring
);
2980 tx_clean_complete
= ixgbe_clean_tx_irq(q_vector
, adapter
->tx_ring
);
2981 ixgbe_clean_rx_irq(q_vector
, adapter
->rx_ring
, &work_done
, budget
);
2983 if (!tx_clean_complete
)
2986 /* If budget not fully consumed, exit the polling mode */
2987 if (work_done
< budget
) {
2988 napi_complete(napi
);
2989 if (adapter
->itr_setting
& 1)
2990 ixgbe_set_itr(adapter
);
2991 if (!test_bit(__IXGBE_DOWN
, &adapter
->state
))
2992 ixgbe_irq_enable_queues(adapter
, IXGBE_EIMS_RTX_QUEUE
);
2998 * ixgbe_tx_timeout - Respond to a Tx Hang
2999 * @netdev: network interface device structure
3001 static void ixgbe_tx_timeout(struct net_device
*netdev
)
3003 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
3005 /* Do the reset outside of interrupt context */
3006 schedule_work(&adapter
->reset_task
);
3009 static void ixgbe_reset_task(struct work_struct
*work
)
3011 struct ixgbe_adapter
*adapter
;
3012 adapter
= container_of(work
, struct ixgbe_adapter
, reset_task
);
3014 /* If we're already down or resetting, just bail */
3015 if (test_bit(__IXGBE_DOWN
, &adapter
->state
) ||
3016 test_bit(__IXGBE_RESETTING
, &adapter
->state
))
3019 adapter
->tx_timeout_count
++;
3021 ixgbe_reinit_locked(adapter
);
3024 #ifdef CONFIG_IXGBE_DCB
3025 static inline bool ixgbe_set_dcb_queues(struct ixgbe_adapter
*adapter
)
3028 struct ixgbe_ring_feature
*f
= &adapter
->ring_feature
[RING_F_DCB
];
3030 if (!(adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
))
3034 adapter
->num_rx_queues
= f
->indices
;
3035 adapter
->num_tx_queues
= f
->indices
;
3043 * ixgbe_set_rss_queues: Allocate queues for RSS
3044 * @adapter: board private structure to initialize
3046 * This is our "base" multiqueue mode. RSS (Receive Side Scaling) will try
3047 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
3050 static inline bool ixgbe_set_rss_queues(struct ixgbe_adapter
*adapter
)
3053 struct ixgbe_ring_feature
*f
= &adapter
->ring_feature
[RING_F_RSS
];
3055 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
) {
3057 adapter
->num_rx_queues
= f
->indices
;
3058 adapter
->num_tx_queues
= f
->indices
;
3068 * ixgbe_set_fdir_queues: Allocate queues for Flow Director
3069 * @adapter: board private structure to initialize
3071 * Flow Director is an advanced Rx filter, attempting to get Rx flows back
3072 * to the original CPU that initiated the Tx session. This runs in addition
3073 * to RSS, so if a packet doesn't match an FDIR filter, we can still spread the
3074 * Rx load across CPUs using RSS.
3077 static bool inline ixgbe_set_fdir_queues(struct ixgbe_adapter
*adapter
)
3080 struct ixgbe_ring_feature
*f_fdir
= &adapter
->ring_feature
[RING_F_FDIR
];
3082 f_fdir
->indices
= min((int)num_online_cpus(), f_fdir
->indices
);
3085 /* Flow Director must have RSS enabled */
3086 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
&&
3087 ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
3088 (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)))) {
3089 adapter
->num_tx_queues
= f_fdir
->indices
;
3090 adapter
->num_rx_queues
= f_fdir
->indices
;
3093 adapter
->flags
&= ~IXGBE_FLAG_FDIR_HASH_CAPABLE
;
3094 adapter
->flags
&= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
3101 * ixgbe_set_fcoe_queues: Allocate queues for Fiber Channel over Ethernet (FCoE)
3102 * @adapter: board private structure to initialize
3104 * FCoE RX FCRETA can use up to 8 rx queues for up to 8 different exchanges.
3105 * The ring feature mask is not used as a mask for FCoE, as it can take any 8
3106 * rx queues out of the max number of rx queues, instead, it is used as the
3107 * index of the first rx queue used by FCoE.
3110 static inline bool ixgbe_set_fcoe_queues(struct ixgbe_adapter
*adapter
)
3113 struct ixgbe_ring_feature
*f
= &adapter
->ring_feature
[RING_F_FCOE
];
3115 f
->indices
= min((int)num_online_cpus(), f
->indices
);
3116 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) {
3117 #ifdef CONFIG_IXGBE_DCB
3118 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
3119 DPRINTK(PROBE
, INFO
, "FCOE enabled with DCB \n");
3120 ixgbe_set_dcb_queues(adapter
);
3123 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
) {
3124 DPRINTK(PROBE
, INFO
, "FCOE enabled with RSS \n");
3125 if ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) ||
3126 (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))
3127 ixgbe_set_fdir_queues(adapter
);
3129 ixgbe_set_rss_queues(adapter
);
3131 /* adding FCoE rx rings to the end */
3132 f
->mask
= adapter
->num_rx_queues
;
3133 adapter
->num_rx_queues
+= f
->indices
;
3134 if (adapter
->num_tx_queues
== 0)
3135 adapter
->num_tx_queues
= f
->indices
;
3143 #endif /* IXGBE_FCOE */
3145 * ixgbe_set_num_queues: Allocate queues for device, feature dependant
3146 * @adapter: board private structure to initialize
3148 * This is the top level queue allocation routine. The order here is very
3149 * important, starting with the "most" number of features turned on at once,
3150 * and ending with the smallest set of features. This way large combinations
3151 * can be allocated if they're turned on, and smaller combinations are the
3152 * fallthrough conditions.
3155 static void ixgbe_set_num_queues(struct ixgbe_adapter
*adapter
)
3158 if (ixgbe_set_fcoe_queues(adapter
))
3161 #endif /* IXGBE_FCOE */
3162 #ifdef CONFIG_IXGBE_DCB
3163 if (ixgbe_set_dcb_queues(adapter
))
3167 if (ixgbe_set_fdir_queues(adapter
))
3170 if (ixgbe_set_rss_queues(adapter
))
3173 /* fallback to base case */
3174 adapter
->num_rx_queues
= 1;
3175 adapter
->num_tx_queues
= 1;
3178 /* Notify the stack of the (possibly) reduced Tx Queue count. */
3179 adapter
->netdev
->real_num_tx_queues
= adapter
->num_tx_queues
;
3182 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter
*adapter
,
3185 int err
, vector_threshold
;
3187 /* We'll want at least 3 (vector_threshold):
3190 * 3) Other (Link Status Change, etc.)
3191 * 4) TCP Timer (optional)
3193 vector_threshold
= MIN_MSIX_COUNT
;
3195 /* The more we get, the more we will assign to Tx/Rx Cleanup
3196 * for the separate queues...where Rx Cleanup >= Tx Cleanup.
3197 * Right now, we simply care about how many we'll get; we'll
3198 * set them up later while requesting irq's.
3200 while (vectors
>= vector_threshold
) {
3201 err
= pci_enable_msix(adapter
->pdev
, adapter
->msix_entries
,
3203 if (!err
) /* Success in acquiring all requested vectors. */
3206 vectors
= 0; /* Nasty failure, quit now */
3207 else /* err == number of vectors we should try again with */
3211 if (vectors
< vector_threshold
) {
3212 /* Can't allocate enough MSI-X interrupts? Oh well.
3213 * This just means we'll go with either a single MSI
3214 * vector or fall back to legacy interrupts.
3216 DPRINTK(HW
, DEBUG
, "Unable to allocate MSI-X interrupts\n");
3217 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
3218 kfree(adapter
->msix_entries
);
3219 adapter
->msix_entries
= NULL
;
3221 adapter
->flags
|= IXGBE_FLAG_MSIX_ENABLED
; /* Woot! */
3223 * Adjust for only the vectors we'll use, which is minimum
3224 * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
3225 * vectors we were allocated.
3227 adapter
->num_msix_vectors
= min(vectors
,
3228 adapter
->max_msix_q_vectors
+ NON_Q_VECTORS
);
3233 * ixgbe_cache_ring_rss - Descriptor ring to register mapping for RSS
3234 * @adapter: board private structure to initialize
3236 * Cache the descriptor ring offsets for RSS to the assigned rings.
3239 static inline bool ixgbe_cache_ring_rss(struct ixgbe_adapter
*adapter
)
3244 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
) {
3245 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3246 adapter
->rx_ring
[i
].reg_idx
= i
;
3247 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3248 adapter
->tx_ring
[i
].reg_idx
= i
;
3257 #ifdef CONFIG_IXGBE_DCB
3259 * ixgbe_cache_ring_dcb - Descriptor ring to register mapping for DCB
3260 * @adapter: board private structure to initialize
3262 * Cache the descriptor ring offsets for DCB to the assigned rings.
3265 static inline bool ixgbe_cache_ring_dcb(struct ixgbe_adapter
*adapter
)
3269 int dcb_i
= adapter
->ring_feature
[RING_F_DCB
].indices
;
3271 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
3272 if (adapter
->hw
.mac
.type
== ixgbe_mac_82598EB
) {
3273 /* the number of queues is assumed to be symmetric */
3274 for (i
= 0; i
< dcb_i
; i
++) {
3275 adapter
->rx_ring
[i
].reg_idx
= i
<< 3;
3276 adapter
->tx_ring
[i
].reg_idx
= i
<< 2;
3279 } else if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
) {
3282 * Tx TC0 starts at: descriptor queue 0
3283 * Tx TC1 starts at: descriptor queue 32
3284 * Tx TC2 starts at: descriptor queue 64
3285 * Tx TC3 starts at: descriptor queue 80
3286 * Tx TC4 starts at: descriptor queue 96
3287 * Tx TC5 starts at: descriptor queue 104
3288 * Tx TC6 starts at: descriptor queue 112
3289 * Tx TC7 starts at: descriptor queue 120
3291 * Rx TC0-TC7 are offset by 16 queues each
3293 for (i
= 0; i
< 3; i
++) {
3294 adapter
->tx_ring
[i
].reg_idx
= i
<< 5;
3295 adapter
->rx_ring
[i
].reg_idx
= i
<< 4;
3297 for ( ; i
< 5; i
++) {
3298 adapter
->tx_ring
[i
].reg_idx
=
3300 adapter
->rx_ring
[i
].reg_idx
= i
<< 4;
3302 for ( ; i
< dcb_i
; i
++) {
3303 adapter
->tx_ring
[i
].reg_idx
=
3305 adapter
->rx_ring
[i
].reg_idx
= i
<< 4;
3309 } else if (dcb_i
== 4) {
3311 * Tx TC0 starts at: descriptor queue 0
3312 * Tx TC1 starts at: descriptor queue 64
3313 * Tx TC2 starts at: descriptor queue 96
3314 * Tx TC3 starts at: descriptor queue 112
3316 * Rx TC0-TC3 are offset by 32 queues each
3318 adapter
->tx_ring
[0].reg_idx
= 0;
3319 adapter
->tx_ring
[1].reg_idx
= 64;
3320 adapter
->tx_ring
[2].reg_idx
= 96;
3321 adapter
->tx_ring
[3].reg_idx
= 112;
3322 for (i
= 0 ; i
< dcb_i
; i
++)
3323 adapter
->rx_ring
[i
].reg_idx
= i
<< 5;
3341 * ixgbe_cache_ring_fdir - Descriptor ring to register mapping for Flow Director
3342 * @adapter: board private structure to initialize
3344 * Cache the descriptor ring offsets for Flow Director to the assigned rings.
3347 static bool inline ixgbe_cache_ring_fdir(struct ixgbe_adapter
*adapter
)
3352 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
&&
3353 ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) ||
3354 (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))) {
3355 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
3356 adapter
->rx_ring
[i
].reg_idx
= i
;
3357 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
3358 adapter
->tx_ring
[i
].reg_idx
= i
;
3367 * ixgbe_cache_ring_fcoe - Descriptor ring to register mapping for the FCoE
3368 * @adapter: board private structure to initialize
3370 * Cache the descriptor ring offsets for FCoE mode to the assigned rings.
3373 static inline bool ixgbe_cache_ring_fcoe(struct ixgbe_adapter
*adapter
)
3377 struct ixgbe_ring_feature
*f
= &adapter
->ring_feature
[RING_F_FCOE
];
3379 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) {
3380 #ifdef CONFIG_IXGBE_DCB
3381 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
3382 ixgbe_cache_ring_dcb(adapter
);
3383 fcoe_i
= adapter
->rx_ring
[0].reg_idx
+ 1;
3385 #endif /* CONFIG_IXGBE_DCB */
3386 if (adapter
->flags
& IXGBE_FLAG_RSS_ENABLED
) {
3387 if ((adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
) ||
3388 (adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
))
3389 ixgbe_cache_ring_fdir(adapter
);
3391 ixgbe_cache_ring_rss(adapter
);
3395 for (i
= 0; i
< f
->indices
; i
++, fcoe_i
++)
3396 adapter
->rx_ring
[f
->mask
+ i
].reg_idx
= fcoe_i
;
3402 #endif /* IXGBE_FCOE */
3404 * ixgbe_cache_ring_register - Descriptor ring to register mapping
3405 * @adapter: board private structure to initialize
3407 * Once we know the feature-set enabled for the device, we'll cache
3408 * the register offset the descriptor ring is assigned to.
3410 * Note, the order the various feature calls is important. It must start with
3411 * the "most" features enabled at the same time, then trickle down to the
3412 * least amount of features turned on at once.
3414 static void ixgbe_cache_ring_register(struct ixgbe_adapter
*adapter
)
3416 /* start with default case */
3417 adapter
->rx_ring
[0].reg_idx
= 0;
3418 adapter
->tx_ring
[0].reg_idx
= 0;
3421 if (ixgbe_cache_ring_fcoe(adapter
))
3424 #endif /* IXGBE_FCOE */
3425 #ifdef CONFIG_IXGBE_DCB
3426 if (ixgbe_cache_ring_dcb(adapter
))
3430 if (ixgbe_cache_ring_fdir(adapter
))
3433 if (ixgbe_cache_ring_rss(adapter
))
3438 * ixgbe_alloc_queues - Allocate memory for all rings
3439 * @adapter: board private structure to initialize
3441 * We allocate one ring per queue at run-time since we don't know the
3442 * number of queues at compile-time. The polling_netdev array is
3443 * intended for Multiqueue, but should work fine with a single queue.
3445 static int ixgbe_alloc_queues(struct ixgbe_adapter
*adapter
)
3449 adapter
->tx_ring
= kcalloc(adapter
->num_tx_queues
,
3450 sizeof(struct ixgbe_ring
), GFP_KERNEL
);
3451 if (!adapter
->tx_ring
)
3452 goto err_tx_ring_allocation
;
3454 adapter
->rx_ring
= kcalloc(adapter
->num_rx_queues
,
3455 sizeof(struct ixgbe_ring
), GFP_KERNEL
);
3456 if (!adapter
->rx_ring
)
3457 goto err_rx_ring_allocation
;
3459 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3460 adapter
->tx_ring
[i
].count
= adapter
->tx_ring_count
;
3461 adapter
->tx_ring
[i
].queue_index
= i
;
3464 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3465 adapter
->rx_ring
[i
].count
= adapter
->rx_ring_count
;
3466 adapter
->rx_ring
[i
].queue_index
= i
;
3469 ixgbe_cache_ring_register(adapter
);
3473 err_rx_ring_allocation
:
3474 kfree(adapter
->tx_ring
);
3475 err_tx_ring_allocation
:
3480 * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
3481 * @adapter: board private structure to initialize
3483 * Attempt to configure the interrupts using the best available
3484 * capabilities of the hardware and the kernel.
3486 static int ixgbe_set_interrupt_capability(struct ixgbe_adapter
*adapter
)
3488 struct ixgbe_hw
*hw
= &adapter
->hw
;
3490 int vector
, v_budget
;
3493 * It's easy to be greedy for MSI-X vectors, but it really
3494 * doesn't do us much good if we have a lot more vectors
3495 * than CPU's. So let's be conservative and only ask for
3496 * (roughly) twice the number of vectors as there are CPU's.
3498 v_budget
= min(adapter
->num_rx_queues
+ adapter
->num_tx_queues
,
3499 (int)(num_online_cpus() * 2)) + NON_Q_VECTORS
;
3502 * At the same time, hardware can only support a maximum of
3503 * hw.mac->max_msix_vectors vectors. With features
3504 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
3505 * descriptor queues supported by our device. Thus, we cap it off in
3506 * those rare cases where the cpu count also exceeds our vector limit.
3508 v_budget
= min(v_budget
, (int)hw
->mac
.max_msix_vectors
);
3510 /* A failure in MSI-X entry allocation isn't fatal, but it does
3511 * mean we disable MSI-X capabilities of the adapter. */
3512 adapter
->msix_entries
= kcalloc(v_budget
,
3513 sizeof(struct msix_entry
), GFP_KERNEL
);
3514 if (adapter
->msix_entries
) {
3515 for (vector
= 0; vector
< v_budget
; vector
++)
3516 adapter
->msix_entries
[vector
].entry
= vector
;
3518 ixgbe_acquire_msix_vectors(adapter
, v_budget
);
3520 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
3524 adapter
->flags
&= ~IXGBE_FLAG_DCB_ENABLED
;
3525 adapter
->flags
&= ~IXGBE_FLAG_RSS_ENABLED
;
3526 adapter
->flags
&= ~IXGBE_FLAG_FDIR_HASH_CAPABLE
;
3527 adapter
->flags
&= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE
;
3528 adapter
->atr_sample_rate
= 0;
3529 ixgbe_set_num_queues(adapter
);
3531 err
= pci_enable_msi(adapter
->pdev
);
3533 adapter
->flags
|= IXGBE_FLAG_MSI_ENABLED
;
3535 DPRINTK(HW
, DEBUG
, "Unable to allocate MSI interrupt, "
3536 "falling back to legacy. Error: %d\n", err
);
3546 * ixgbe_alloc_q_vectors - Allocate memory for interrupt vectors
3547 * @adapter: board private structure to initialize
3549 * We allocate one q_vector per queue interrupt. If allocation fails we
3552 static int ixgbe_alloc_q_vectors(struct ixgbe_adapter
*adapter
)
3554 int q_idx
, num_q_vectors
;
3555 struct ixgbe_q_vector
*q_vector
;
3557 int (*poll
)(struct napi_struct
*, int);
3559 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
3560 num_q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
3561 napi_vectors
= adapter
->num_rx_queues
;
3562 poll
= &ixgbe_clean_rxtx_many
;
3569 for (q_idx
= 0; q_idx
< num_q_vectors
; q_idx
++) {
3570 q_vector
= kzalloc(sizeof(struct ixgbe_q_vector
), GFP_KERNEL
);
3573 q_vector
->adapter
= adapter
;
3574 q_vector
->eitr
= adapter
->eitr_param
;
3575 q_vector
->v_idx
= q_idx
;
3576 netif_napi_add(adapter
->netdev
, &q_vector
->napi
, (*poll
), 64);
3577 adapter
->q_vector
[q_idx
] = q_vector
;
3585 q_vector
= adapter
->q_vector
[q_idx
];
3586 netif_napi_del(&q_vector
->napi
);
3588 adapter
->q_vector
[q_idx
] = NULL
;
3594 * ixgbe_free_q_vectors - Free memory allocated for interrupt vectors
3595 * @adapter: board private structure to initialize
3597 * This function frees the memory allocated to the q_vectors. In addition if
3598 * NAPI is enabled it will delete any references to the NAPI struct prior
3599 * to freeing the q_vector.
3601 static void ixgbe_free_q_vectors(struct ixgbe_adapter
*adapter
)
3603 int q_idx
, num_q_vectors
;
3605 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)
3606 num_q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
3610 for (q_idx
= 0; q_idx
< num_q_vectors
; q_idx
++) {
3611 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[q_idx
];
3612 adapter
->q_vector
[q_idx
] = NULL
;
3613 netif_napi_del(&q_vector
->napi
);
3618 void ixgbe_reset_interrupt_capability(struct ixgbe_adapter
*adapter
)
3620 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
3621 adapter
->flags
&= ~IXGBE_FLAG_MSIX_ENABLED
;
3622 pci_disable_msix(adapter
->pdev
);
3623 kfree(adapter
->msix_entries
);
3624 adapter
->msix_entries
= NULL
;
3625 } else if (adapter
->flags
& IXGBE_FLAG_MSI_ENABLED
) {
3626 adapter
->flags
&= ~IXGBE_FLAG_MSI_ENABLED
;
3627 pci_disable_msi(adapter
->pdev
);
3633 * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
3634 * @adapter: board private structure to initialize
3636 * We determine which interrupt scheme to use based on...
3637 * - Kernel support (MSI, MSI-X)
3638 * - which can be user-defined (via MODULE_PARAM)
3639 * - Hardware queue count (num_*_queues)
3640 * - defined by miscellaneous hardware support/features (RSS, etc.)
3642 int ixgbe_init_interrupt_scheme(struct ixgbe_adapter
*adapter
)
3646 /* Number of supported queues */
3647 ixgbe_set_num_queues(adapter
);
3649 err
= ixgbe_set_interrupt_capability(adapter
);
3651 DPRINTK(PROBE
, ERR
, "Unable to setup interrupt capabilities\n");
3652 goto err_set_interrupt
;
3655 err
= ixgbe_alloc_q_vectors(adapter
);
3657 DPRINTK(PROBE
, ERR
, "Unable to allocate memory for queue "
3659 goto err_alloc_q_vectors
;
3662 err
= ixgbe_alloc_queues(adapter
);
3664 DPRINTK(PROBE
, ERR
, "Unable to allocate memory for queues\n");
3665 goto err_alloc_queues
;
3668 DPRINTK(DRV
, INFO
, "Multiqueue %s: Rx Queue count = %u, "
3669 "Tx Queue count = %u\n",
3670 (adapter
->num_rx_queues
> 1) ? "Enabled" :
3671 "Disabled", adapter
->num_rx_queues
, adapter
->num_tx_queues
);
3673 set_bit(__IXGBE_DOWN
, &adapter
->state
);
3678 ixgbe_free_q_vectors(adapter
);
3679 err_alloc_q_vectors
:
3680 ixgbe_reset_interrupt_capability(adapter
);
3686 * ixgbe_clear_interrupt_scheme - Clear the current interrupt scheme settings
3687 * @adapter: board private structure to clear interrupt scheme on
3689 * We go through and clear interrupt specific resources and reset the structure
3690 * to pre-load conditions
3692 void ixgbe_clear_interrupt_scheme(struct ixgbe_adapter
*adapter
)
3694 kfree(adapter
->tx_ring
);
3695 kfree(adapter
->rx_ring
);
3696 adapter
->tx_ring
= NULL
;
3697 adapter
->rx_ring
= NULL
;
3699 ixgbe_free_q_vectors(adapter
);
3700 ixgbe_reset_interrupt_capability(adapter
);
3704 * ixgbe_sfp_timer - worker thread to find a missing module
3705 * @data: pointer to our adapter struct
3707 static void ixgbe_sfp_timer(unsigned long data
)
3709 struct ixgbe_adapter
*adapter
= (struct ixgbe_adapter
*)data
;
3712 * Do the sfp_timer outside of interrupt context due to the
3713 * delays that sfp+ detection requires
3715 schedule_work(&adapter
->sfp_task
);
3719 * ixgbe_sfp_task - worker thread to find a missing module
3720 * @work: pointer to work_struct containing our data
3722 static void ixgbe_sfp_task(struct work_struct
*work
)
3724 struct ixgbe_adapter
*adapter
= container_of(work
,
3725 struct ixgbe_adapter
,
3727 struct ixgbe_hw
*hw
= &adapter
->hw
;
3729 if ((hw
->phy
.type
== ixgbe_phy_nl
) &&
3730 (hw
->phy
.sfp_type
== ixgbe_sfp_type_not_present
)) {
3731 s32 ret
= hw
->phy
.ops
.identify_sfp(hw
);
3732 if (ret
== IXGBE_ERR_SFP_NOT_PRESENT
)
3734 ret
= hw
->phy
.ops
.reset(hw
);
3735 if (ret
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
3736 dev_err(&adapter
->pdev
->dev
, "failed to initialize "
3737 "because an unsupported SFP+ module type "
3739 "Reload the driver after installing a "
3740 "supported module.\n");
3741 unregister_netdev(adapter
->netdev
);
3743 DPRINTK(PROBE
, INFO
, "detected SFP+: %d\n",
3746 /* don't need this routine any more */
3747 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND
, &adapter
->state
);
3751 if (test_bit(__IXGBE_SFP_MODULE_NOT_FOUND
, &adapter
->state
))
3752 mod_timer(&adapter
->sfp_timer
,
3753 round_jiffies(jiffies
+ (2 * HZ
)));
3757 * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
3758 * @adapter: board private structure to initialize
3760 * ixgbe_sw_init initializes the Adapter private data structure.
3761 * Fields are initialized based on PCI device information and
3762 * OS network device settings (MTU size).
3764 static int __devinit
ixgbe_sw_init(struct ixgbe_adapter
*adapter
)
3766 struct ixgbe_hw
*hw
= &adapter
->hw
;
3767 struct pci_dev
*pdev
= adapter
->pdev
;
3769 #ifdef CONFIG_IXGBE_DCB
3771 struct tc_configuration
*tc
;
3774 /* PCI config space info */
3776 hw
->vendor_id
= pdev
->vendor
;
3777 hw
->device_id
= pdev
->device
;
3778 hw
->revision_id
= pdev
->revision
;
3779 hw
->subsystem_vendor_id
= pdev
->subsystem_vendor
;
3780 hw
->subsystem_device_id
= pdev
->subsystem_device
;
3782 /* Set capability flags */
3783 rss
= min(IXGBE_MAX_RSS_INDICES
, (int)num_online_cpus());
3784 adapter
->ring_feature
[RING_F_RSS
].indices
= rss
;
3785 adapter
->flags
|= IXGBE_FLAG_RSS_ENABLED
;
3786 adapter
->ring_feature
[RING_F_DCB
].indices
= IXGBE_MAX_DCB_INDICES
;
3787 if (hw
->mac
.type
== ixgbe_mac_82598EB
) {
3788 if (hw
->device_id
== IXGBE_DEV_ID_82598AT
)
3789 adapter
->flags
|= IXGBE_FLAG_FAN_FAIL_CAPABLE
;
3790 adapter
->max_msix_q_vectors
= MAX_MSIX_Q_VECTORS_82598
;
3791 } else if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
3792 adapter
->max_msix_q_vectors
= MAX_MSIX_Q_VECTORS_82599
;
3793 adapter
->flags2
|= IXGBE_FLAG2_RSC_CAPABLE
;
3794 adapter
->flags2
|= IXGBE_FLAG2_RSC_ENABLED
;
3795 adapter
->flags
|= IXGBE_FLAG_FDIR_HASH_CAPABLE
;
3796 adapter
->ring_feature
[RING_F_FDIR
].indices
=
3797 IXGBE_MAX_FDIR_INDICES
;
3798 adapter
->atr_sample_rate
= 20;
3799 adapter
->fdir_pballoc
= 0;
3801 adapter
->flags
|= IXGBE_FLAG_FCOE_CAPABLE
;
3802 adapter
->flags
&= ~IXGBE_FLAG_FCOE_ENABLED
;
3803 adapter
->ring_feature
[RING_F_FCOE
].indices
= 0;
3804 #endif /* IXGBE_FCOE */
3807 #ifdef CONFIG_IXGBE_DCB
3808 /* Configure DCB traffic classes */
3809 for (j
= 0; j
< MAX_TRAFFIC_CLASS
; j
++) {
3810 tc
= &adapter
->dcb_cfg
.tc_config
[j
];
3811 tc
->path
[DCB_TX_CONFIG
].bwg_id
= 0;
3812 tc
->path
[DCB_TX_CONFIG
].bwg_percent
= 12 + (j
& 1);
3813 tc
->path
[DCB_RX_CONFIG
].bwg_id
= 0;
3814 tc
->path
[DCB_RX_CONFIG
].bwg_percent
= 12 + (j
& 1);
3815 tc
->dcb_pfc
= pfc_disabled
;
3817 adapter
->dcb_cfg
.bw_percentage
[DCB_TX_CONFIG
][0] = 100;
3818 adapter
->dcb_cfg
.bw_percentage
[DCB_RX_CONFIG
][0] = 100;
3819 adapter
->dcb_cfg
.rx_pba_cfg
= pba_equal
;
3820 adapter
->dcb_cfg
.pfc_mode_enable
= false;
3821 adapter
->dcb_cfg
.round_robin_enable
= false;
3822 adapter
->dcb_set_bitmap
= 0x00;
3823 ixgbe_copy_dcb_cfg(&adapter
->dcb_cfg
, &adapter
->temp_dcb_cfg
,
3824 adapter
->ring_feature
[RING_F_DCB
].indices
);
3828 /* default flow control settings */
3829 hw
->fc
.requested_mode
= ixgbe_fc_full
;
3830 hw
->fc
.current_mode
= ixgbe_fc_full
; /* init for ethtool output */
3832 adapter
->last_lfc_mode
= hw
->fc
.current_mode
;
3834 hw
->fc
.high_water
= IXGBE_DEFAULT_FCRTH
;
3835 hw
->fc
.low_water
= IXGBE_DEFAULT_FCRTL
;
3836 hw
->fc
.pause_time
= IXGBE_DEFAULT_FCPAUSE
;
3837 hw
->fc
.send_xon
= true;
3838 hw
->fc
.disable_fc_autoneg
= false;
3840 /* enable itr by default in dynamic mode */
3841 adapter
->itr_setting
= 1;
3842 adapter
->eitr_param
= 20000;
3844 /* set defaults for eitr in MegaBytes */
3845 adapter
->eitr_low
= 10;
3846 adapter
->eitr_high
= 20;
3848 /* set default ring sizes */
3849 adapter
->tx_ring_count
= IXGBE_DEFAULT_TXD
;
3850 adapter
->rx_ring_count
= IXGBE_DEFAULT_RXD
;
3852 /* initialize eeprom parameters */
3853 if (ixgbe_init_eeprom_params_generic(hw
)) {
3854 dev_err(&pdev
->dev
, "EEPROM initialization failed\n");
3858 /* enable rx csum by default */
3859 adapter
->flags
|= IXGBE_FLAG_RX_CSUM_ENABLED
;
3861 set_bit(__IXGBE_DOWN
, &adapter
->state
);
3867 * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
3868 * @adapter: board private structure
3869 * @tx_ring: tx descriptor ring (for a specific queue) to setup
3871 * Return 0 on success, negative on failure
3873 int ixgbe_setup_tx_resources(struct ixgbe_adapter
*adapter
,
3874 struct ixgbe_ring
*tx_ring
)
3876 struct pci_dev
*pdev
= adapter
->pdev
;
3879 size
= sizeof(struct ixgbe_tx_buffer
) * tx_ring
->count
;
3880 tx_ring
->tx_buffer_info
= vmalloc(size
);
3881 if (!tx_ring
->tx_buffer_info
)
3883 memset(tx_ring
->tx_buffer_info
, 0, size
);
3885 /* round up to nearest 4K */
3886 tx_ring
->size
= tx_ring
->count
* sizeof(union ixgbe_adv_tx_desc
);
3887 tx_ring
->size
= ALIGN(tx_ring
->size
, 4096);
3889 tx_ring
->desc
= pci_alloc_consistent(pdev
, tx_ring
->size
,
3894 tx_ring
->next_to_use
= 0;
3895 tx_ring
->next_to_clean
= 0;
3896 tx_ring
->work_limit
= tx_ring
->count
;
3900 vfree(tx_ring
->tx_buffer_info
);
3901 tx_ring
->tx_buffer_info
= NULL
;
3902 DPRINTK(PROBE
, ERR
, "Unable to allocate memory for the transmit "
3903 "descriptor ring\n");
3908 * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
3909 * @adapter: board private structure
3911 * If this function returns with an error, then it's possible one or
3912 * more of the rings is populated (while the rest are not). It is the
3913 * callers duty to clean those orphaned rings.
3915 * Return 0 on success, negative on failure
3917 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter
*adapter
)
3921 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
3922 err
= ixgbe_setup_tx_resources(adapter
, &adapter
->tx_ring
[i
]);
3925 DPRINTK(PROBE
, ERR
, "Allocation for Tx Queue %u failed\n", i
);
3933 * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
3934 * @adapter: board private structure
3935 * @rx_ring: rx descriptor ring (for a specific queue) to setup
3937 * Returns 0 on success, negative on failure
3939 int ixgbe_setup_rx_resources(struct ixgbe_adapter
*adapter
,
3940 struct ixgbe_ring
*rx_ring
)
3942 struct pci_dev
*pdev
= adapter
->pdev
;
3945 size
= sizeof(struct ixgbe_rx_buffer
) * rx_ring
->count
;
3946 rx_ring
->rx_buffer_info
= vmalloc(size
);
3947 if (!rx_ring
->rx_buffer_info
) {
3949 "vmalloc allocation failed for the rx desc ring\n");
3952 memset(rx_ring
->rx_buffer_info
, 0, size
);
3954 /* Round up to nearest 4K */
3955 rx_ring
->size
= rx_ring
->count
* sizeof(union ixgbe_adv_rx_desc
);
3956 rx_ring
->size
= ALIGN(rx_ring
->size
, 4096);
3958 rx_ring
->desc
= pci_alloc_consistent(pdev
, rx_ring
->size
, &rx_ring
->dma
);
3960 if (!rx_ring
->desc
) {
3962 "Memory allocation failed for the rx desc ring\n");
3963 vfree(rx_ring
->rx_buffer_info
);
3967 rx_ring
->next_to_clean
= 0;
3968 rx_ring
->next_to_use
= 0;
3977 * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
3978 * @adapter: board private structure
3980 * If this function returns with an error, then it's possible one or
3981 * more of the rings is populated (while the rest are not). It is the
3982 * callers duty to clean those orphaned rings.
3984 * Return 0 on success, negative on failure
3987 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter
*adapter
)
3991 for (i
= 0; i
< adapter
->num_rx_queues
; i
++) {
3992 err
= ixgbe_setup_rx_resources(adapter
, &adapter
->rx_ring
[i
]);
3995 DPRINTK(PROBE
, ERR
, "Allocation for Rx Queue %u failed\n", i
);
4003 * ixgbe_free_tx_resources - Free Tx Resources per Queue
4004 * @adapter: board private structure
4005 * @tx_ring: Tx descriptor ring for a specific queue
4007 * Free all transmit software resources
4009 void ixgbe_free_tx_resources(struct ixgbe_adapter
*adapter
,
4010 struct ixgbe_ring
*tx_ring
)
4012 struct pci_dev
*pdev
= adapter
->pdev
;
4014 ixgbe_clean_tx_ring(adapter
, tx_ring
);
4016 vfree(tx_ring
->tx_buffer_info
);
4017 tx_ring
->tx_buffer_info
= NULL
;
4019 pci_free_consistent(pdev
, tx_ring
->size
, tx_ring
->desc
, tx_ring
->dma
);
4021 tx_ring
->desc
= NULL
;
4025 * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
4026 * @adapter: board private structure
4028 * Free all transmit software resources
4030 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter
*adapter
)
4034 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
4035 if (adapter
->tx_ring
[i
].desc
)
4036 ixgbe_free_tx_resources(adapter
, &adapter
->tx_ring
[i
]);
4040 * ixgbe_free_rx_resources - Free Rx Resources
4041 * @adapter: board private structure
4042 * @rx_ring: ring to clean the resources from
4044 * Free all receive software resources
4046 void ixgbe_free_rx_resources(struct ixgbe_adapter
*adapter
,
4047 struct ixgbe_ring
*rx_ring
)
4049 struct pci_dev
*pdev
= adapter
->pdev
;
4051 ixgbe_clean_rx_ring(adapter
, rx_ring
);
4053 vfree(rx_ring
->rx_buffer_info
);
4054 rx_ring
->rx_buffer_info
= NULL
;
4056 pci_free_consistent(pdev
, rx_ring
->size
, rx_ring
->desc
, rx_ring
->dma
);
4058 rx_ring
->desc
= NULL
;
4062 * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
4063 * @adapter: board private structure
4065 * Free all receive software resources
4067 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter
*adapter
)
4071 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4072 if (adapter
->rx_ring
[i
].desc
)
4073 ixgbe_free_rx_resources(adapter
, &adapter
->rx_ring
[i
]);
4077 * ixgbe_change_mtu - Change the Maximum Transfer Unit
4078 * @netdev: network interface device structure
4079 * @new_mtu: new value for maximum frame size
4081 * Returns 0 on success, negative on failure
4083 static int ixgbe_change_mtu(struct net_device
*netdev
, int new_mtu
)
4085 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4086 int max_frame
= new_mtu
+ ETH_HLEN
+ ETH_FCS_LEN
;
4088 /* MTU < 68 is an error and causes problems on some kernels */
4089 if ((new_mtu
< 68) || (max_frame
> IXGBE_MAX_JUMBO_FRAME_SIZE
))
4092 DPRINTK(PROBE
, INFO
, "changing MTU from %d to %d\n",
4093 netdev
->mtu
, new_mtu
);
4094 /* must set new MTU before calling down or up */
4095 netdev
->mtu
= new_mtu
;
4097 if (netif_running(netdev
))
4098 ixgbe_reinit_locked(adapter
);
4104 * ixgbe_open - Called when a network interface is made active
4105 * @netdev: network interface device structure
4107 * Returns 0 on success, negative value on failure
4109 * The open entry point is called when a network interface is made
4110 * active by the system (IFF_UP). At this point all resources needed
4111 * for transmit and receive operations are allocated, the interrupt
4112 * handler is registered with the OS, the watchdog timer is started,
4113 * and the stack is notified that the interface is ready.
4115 static int ixgbe_open(struct net_device
*netdev
)
4117 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4120 /* disallow open during test */
4121 if (test_bit(__IXGBE_TESTING
, &adapter
->state
))
4124 netif_carrier_off(netdev
);
4126 /* allocate transmit descriptors */
4127 err
= ixgbe_setup_all_tx_resources(adapter
);
4131 /* allocate receive descriptors */
4132 err
= ixgbe_setup_all_rx_resources(adapter
);
4136 ixgbe_configure(adapter
);
4138 err
= ixgbe_request_irq(adapter
);
4142 err
= ixgbe_up_complete(adapter
);
4146 netif_tx_start_all_queues(netdev
);
4151 ixgbe_release_hw_control(adapter
);
4152 ixgbe_free_irq(adapter
);
4155 ixgbe_free_all_rx_resources(adapter
);
4157 ixgbe_free_all_tx_resources(adapter
);
4158 ixgbe_reset(adapter
);
4164 * ixgbe_close - Disables a network interface
4165 * @netdev: network interface device structure
4167 * Returns 0, this is not allowed to fail
4169 * The close entry point is called when an interface is de-activated
4170 * by the OS. The hardware is still under the drivers control, but
4171 * needs to be disabled. A global MAC reset is issued to stop the
4172 * hardware, and all transmit and receive resources are freed.
4174 static int ixgbe_close(struct net_device
*netdev
)
4176 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4178 ixgbe_down(adapter
);
4179 ixgbe_free_irq(adapter
);
4181 ixgbe_free_all_tx_resources(adapter
);
4182 ixgbe_free_all_rx_resources(adapter
);
4184 ixgbe_release_hw_control(adapter
);
4190 static int ixgbe_resume(struct pci_dev
*pdev
)
4192 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4193 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4196 pci_set_power_state(pdev
, PCI_D0
);
4197 pci_restore_state(pdev
);
4199 err
= pci_enable_device_mem(pdev
);
4201 printk(KERN_ERR
"ixgbe: Cannot enable PCI device from "
4205 pci_set_master(pdev
);
4207 pci_wake_from_d3(pdev
, false);
4209 err
= ixgbe_init_interrupt_scheme(adapter
);
4211 printk(KERN_ERR
"ixgbe: Cannot initialize interrupts for "
4216 ixgbe_reset(adapter
);
4218 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
4220 if (netif_running(netdev
)) {
4221 err
= ixgbe_open(adapter
->netdev
);
4226 netif_device_attach(netdev
);
4230 #endif /* CONFIG_PM */
4232 static int __ixgbe_shutdown(struct pci_dev
*pdev
, bool *enable_wake
)
4234 struct net_device
*netdev
= pci_get_drvdata(pdev
);
4235 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
4236 struct ixgbe_hw
*hw
= &adapter
->hw
;
4238 u32 wufc
= adapter
->wol
;
4243 netif_device_detach(netdev
);
4245 if (netif_running(netdev
)) {
4246 ixgbe_down(adapter
);
4247 ixgbe_free_irq(adapter
);
4248 ixgbe_free_all_tx_resources(adapter
);
4249 ixgbe_free_all_rx_resources(adapter
);
4251 ixgbe_clear_interrupt_scheme(adapter
);
4254 retval
= pci_save_state(pdev
);
4260 ixgbe_set_rx_mode(netdev
);
4262 /* turn on all-multi mode if wake on multicast is enabled */
4263 if (wufc
& IXGBE_WUFC_MC
) {
4264 fctrl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
4265 fctrl
|= IXGBE_FCTRL_MPE
;
4266 IXGBE_WRITE_REG(hw
, IXGBE_FCTRL
, fctrl
);
4269 ctrl
= IXGBE_READ_REG(hw
, IXGBE_CTRL
);
4270 ctrl
|= IXGBE_CTRL_GIO_DIS
;
4271 IXGBE_WRITE_REG(hw
, IXGBE_CTRL
, ctrl
);
4273 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, wufc
);
4275 IXGBE_WRITE_REG(hw
, IXGBE_WUC
, 0);
4276 IXGBE_WRITE_REG(hw
, IXGBE_WUFC
, 0);
4279 if (wufc
&& hw
->mac
.type
== ixgbe_mac_82599EB
)
4280 pci_wake_from_d3(pdev
, true);
4282 pci_wake_from_d3(pdev
, false);
4284 *enable_wake
= !!wufc
;
4286 ixgbe_release_hw_control(adapter
);
4288 pci_disable_device(pdev
);
4294 static int ixgbe_suspend(struct pci_dev
*pdev
, pm_message_t state
)
4299 retval
= __ixgbe_shutdown(pdev
, &wake
);
4304 pci_prepare_to_sleep(pdev
);
4306 pci_wake_from_d3(pdev
, false);
4307 pci_set_power_state(pdev
, PCI_D3hot
);
4312 #endif /* CONFIG_PM */
4314 static void ixgbe_shutdown(struct pci_dev
*pdev
)
4318 __ixgbe_shutdown(pdev
, &wake
);
4320 if (system_state
== SYSTEM_POWER_OFF
) {
4321 pci_wake_from_d3(pdev
, wake
);
4322 pci_set_power_state(pdev
, PCI_D3hot
);
4327 * ixgbe_update_stats - Update the board statistics counters.
4328 * @adapter: board private structure
4330 void ixgbe_update_stats(struct ixgbe_adapter
*adapter
)
4332 struct ixgbe_hw
*hw
= &adapter
->hw
;
4334 u32 i
, missed_rx
= 0, mpc
, bprc
, lxon
, lxoff
, xon_off_tot
;
4336 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4338 for (i
= 0; i
< 16; i
++)
4339 adapter
->hw_rx_no_dma_resources
+=
4340 IXGBE_READ_REG(hw
, IXGBE_QPRDC(i
));
4341 for (i
= 0; i
< adapter
->num_rx_queues
; i
++)
4342 rsc_count
+= adapter
->rx_ring
[i
].rsc_count
;
4343 adapter
->rsc_count
= rsc_count
;
4346 adapter
->stats
.crcerrs
+= IXGBE_READ_REG(hw
, IXGBE_CRCERRS
);
4347 for (i
= 0; i
< 8; i
++) {
4348 /* for packet buffers not used, the register should read 0 */
4349 mpc
= IXGBE_READ_REG(hw
, IXGBE_MPC(i
));
4351 adapter
->stats
.mpc
[i
] += mpc
;
4352 total_mpc
+= adapter
->stats
.mpc
[i
];
4353 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4354 adapter
->stats
.rnbc
[i
] += IXGBE_READ_REG(hw
, IXGBE_RNBC(i
));
4355 adapter
->stats
.qptc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPTC(i
));
4356 adapter
->stats
.qbtc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBTC(i
));
4357 adapter
->stats
.qprc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPRC(i
));
4358 adapter
->stats
.qbrc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QBRC(i
));
4359 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4360 adapter
->stats
.pxonrxc
[i
] += IXGBE_READ_REG(hw
,
4361 IXGBE_PXONRXCNT(i
));
4362 adapter
->stats
.pxoffrxc
[i
] += IXGBE_READ_REG(hw
,
4363 IXGBE_PXOFFRXCNT(i
));
4364 adapter
->stats
.qprdc
[i
] += IXGBE_READ_REG(hw
, IXGBE_QPRDC(i
));
4366 adapter
->stats
.pxonrxc
[i
] += IXGBE_READ_REG(hw
,
4368 adapter
->stats
.pxoffrxc
[i
] += IXGBE_READ_REG(hw
,
4371 adapter
->stats
.pxontxc
[i
] += IXGBE_READ_REG(hw
,
4373 adapter
->stats
.pxofftxc
[i
] += IXGBE_READ_REG(hw
,
4376 adapter
->stats
.gprc
+= IXGBE_READ_REG(hw
, IXGBE_GPRC
);
4377 /* work around hardware counting issue */
4378 adapter
->stats
.gprc
-= missed_rx
;
4380 /* 82598 hardware only has a 32 bit counter in the high register */
4381 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4382 adapter
->stats
.gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCL
);
4383 IXGBE_READ_REG(hw
, IXGBE_GORCH
); /* to clear */
4384 adapter
->stats
.gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCL
);
4385 IXGBE_READ_REG(hw
, IXGBE_GOTCH
); /* to clear */
4386 adapter
->stats
.tor
+= IXGBE_READ_REG(hw
, IXGBE_TORL
);
4387 IXGBE_READ_REG(hw
, IXGBE_TORH
); /* to clear */
4388 adapter
->stats
.lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXCNT
);
4389 adapter
->stats
.lxoffrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXCNT
);
4390 adapter
->stats
.fdirmatch
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMATCH
);
4391 adapter
->stats
.fdirmiss
+= IXGBE_READ_REG(hw
, IXGBE_FDIRMISS
);
4393 adapter
->stats
.fccrc
+= IXGBE_READ_REG(hw
, IXGBE_FCCRC
);
4394 adapter
->stats
.fcoerpdc
+= IXGBE_READ_REG(hw
, IXGBE_FCOERPDC
);
4395 adapter
->stats
.fcoeprc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPRC
);
4396 adapter
->stats
.fcoeptc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEPTC
);
4397 adapter
->stats
.fcoedwrc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWRC
);
4398 adapter
->stats
.fcoedwtc
+= IXGBE_READ_REG(hw
, IXGBE_FCOEDWTC
);
4399 #endif /* IXGBE_FCOE */
4401 adapter
->stats
.lxonrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXONRXC
);
4402 adapter
->stats
.lxoffrxc
+= IXGBE_READ_REG(hw
, IXGBE_LXOFFRXC
);
4403 adapter
->stats
.gorc
+= IXGBE_READ_REG(hw
, IXGBE_GORCH
);
4404 adapter
->stats
.gotc
+= IXGBE_READ_REG(hw
, IXGBE_GOTCH
);
4405 adapter
->stats
.tor
+= IXGBE_READ_REG(hw
, IXGBE_TORH
);
4407 bprc
= IXGBE_READ_REG(hw
, IXGBE_BPRC
);
4408 adapter
->stats
.bprc
+= bprc
;
4409 adapter
->stats
.mprc
+= IXGBE_READ_REG(hw
, IXGBE_MPRC
);
4410 if (hw
->mac
.type
== ixgbe_mac_82598EB
)
4411 adapter
->stats
.mprc
-= bprc
;
4412 adapter
->stats
.roc
+= IXGBE_READ_REG(hw
, IXGBE_ROC
);
4413 adapter
->stats
.prc64
+= IXGBE_READ_REG(hw
, IXGBE_PRC64
);
4414 adapter
->stats
.prc127
+= IXGBE_READ_REG(hw
, IXGBE_PRC127
);
4415 adapter
->stats
.prc255
+= IXGBE_READ_REG(hw
, IXGBE_PRC255
);
4416 adapter
->stats
.prc511
+= IXGBE_READ_REG(hw
, IXGBE_PRC511
);
4417 adapter
->stats
.prc1023
+= IXGBE_READ_REG(hw
, IXGBE_PRC1023
);
4418 adapter
->stats
.prc1522
+= IXGBE_READ_REG(hw
, IXGBE_PRC1522
);
4419 adapter
->stats
.rlec
+= IXGBE_READ_REG(hw
, IXGBE_RLEC
);
4420 lxon
= IXGBE_READ_REG(hw
, IXGBE_LXONTXC
);
4421 adapter
->stats
.lxontxc
+= lxon
;
4422 lxoff
= IXGBE_READ_REG(hw
, IXGBE_LXOFFTXC
);
4423 adapter
->stats
.lxofftxc
+= lxoff
;
4424 adapter
->stats
.ruc
+= IXGBE_READ_REG(hw
, IXGBE_RUC
);
4425 adapter
->stats
.gptc
+= IXGBE_READ_REG(hw
, IXGBE_GPTC
);
4426 adapter
->stats
.mptc
+= IXGBE_READ_REG(hw
, IXGBE_MPTC
);
4428 * 82598 errata - tx of flow control packets is included in tx counters
4430 xon_off_tot
= lxon
+ lxoff
;
4431 adapter
->stats
.gptc
-= xon_off_tot
;
4432 adapter
->stats
.mptc
-= xon_off_tot
;
4433 adapter
->stats
.gotc
-= (xon_off_tot
* (ETH_ZLEN
+ ETH_FCS_LEN
));
4434 adapter
->stats
.ruc
+= IXGBE_READ_REG(hw
, IXGBE_RUC
);
4435 adapter
->stats
.rfc
+= IXGBE_READ_REG(hw
, IXGBE_RFC
);
4436 adapter
->stats
.rjc
+= IXGBE_READ_REG(hw
, IXGBE_RJC
);
4437 adapter
->stats
.tpr
+= IXGBE_READ_REG(hw
, IXGBE_TPR
);
4438 adapter
->stats
.ptc64
+= IXGBE_READ_REG(hw
, IXGBE_PTC64
);
4439 adapter
->stats
.ptc64
-= xon_off_tot
;
4440 adapter
->stats
.ptc127
+= IXGBE_READ_REG(hw
, IXGBE_PTC127
);
4441 adapter
->stats
.ptc255
+= IXGBE_READ_REG(hw
, IXGBE_PTC255
);
4442 adapter
->stats
.ptc511
+= IXGBE_READ_REG(hw
, IXGBE_PTC511
);
4443 adapter
->stats
.ptc1023
+= IXGBE_READ_REG(hw
, IXGBE_PTC1023
);
4444 adapter
->stats
.ptc1522
+= IXGBE_READ_REG(hw
, IXGBE_PTC1522
);
4445 adapter
->stats
.bptc
+= IXGBE_READ_REG(hw
, IXGBE_BPTC
);
4447 /* Fill out the OS statistics structure */
4448 adapter
->net_stats
.multicast
= adapter
->stats
.mprc
;
4451 adapter
->net_stats
.rx_errors
= adapter
->stats
.crcerrs
+
4452 adapter
->stats
.rlec
;
4453 adapter
->net_stats
.rx_dropped
= 0;
4454 adapter
->net_stats
.rx_length_errors
= adapter
->stats
.rlec
;
4455 adapter
->net_stats
.rx_crc_errors
= adapter
->stats
.crcerrs
;
4456 adapter
->net_stats
.rx_missed_errors
= total_mpc
;
4460 * ixgbe_watchdog - Timer Call-back
4461 * @data: pointer to adapter cast into an unsigned long
4463 static void ixgbe_watchdog(unsigned long data
)
4465 struct ixgbe_adapter
*adapter
= (struct ixgbe_adapter
*)data
;
4466 struct ixgbe_hw
*hw
= &adapter
->hw
;
4471 * Do the watchdog outside of interrupt context due to the lovely
4472 * delays that some of the newer hardware requires
4475 if (test_bit(__IXGBE_DOWN
, &adapter
->state
))
4476 goto watchdog_short_circuit
;
4478 if (!(adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
)) {
4480 * for legacy and MSI interrupts don't set any bits
4481 * that are enabled for EIAM, because this operation
4482 * would set *both* EIMS and EICS for any bit in EIAM
4484 IXGBE_WRITE_REG(hw
, IXGBE_EICS
,
4485 (IXGBE_EICS_TCP_TIMER
| IXGBE_EICS_OTHER
));
4486 goto watchdog_reschedule
;
4489 /* get one bit for every active tx/rx interrupt vector */
4490 for (i
= 0; i
< adapter
->num_msix_vectors
- NON_Q_VECTORS
; i
++) {
4491 struct ixgbe_q_vector
*qv
= adapter
->q_vector
[i
];
4492 if (qv
->rxr_count
|| qv
->txr_count
)
4493 eics
|= ((u64
)1 << i
);
4496 /* Cause software interrupt to ensure rx rings are cleaned */
4497 ixgbe_irq_rearm_queues(adapter
, eics
);
4499 watchdog_reschedule
:
4500 /* Reset the timer */
4501 mod_timer(&adapter
->watchdog_timer
, round_jiffies(jiffies
+ 2 * HZ
));
4503 watchdog_short_circuit
:
4504 schedule_work(&adapter
->watchdog_task
);
4508 * ixgbe_multispeed_fiber_task - worker thread to configure multispeed fiber
4509 * @work: pointer to work_struct containing our data
4511 static void ixgbe_multispeed_fiber_task(struct work_struct
*work
)
4513 struct ixgbe_adapter
*adapter
= container_of(work
,
4514 struct ixgbe_adapter
,
4515 multispeed_fiber_task
);
4516 struct ixgbe_hw
*hw
= &adapter
->hw
;
4519 adapter
->flags
|= IXGBE_FLAG_IN_SFP_LINK_TASK
;
4520 autoneg
= hw
->phy
.autoneg_advertised
;
4521 if ((!autoneg
) && (hw
->mac
.ops
.get_link_capabilities
))
4522 hw
->mac
.ops
.get_link_capabilities(hw
, &autoneg
,
4524 if (hw
->mac
.ops
.setup_link_speed
)
4525 hw
->mac
.ops
.setup_link_speed(hw
, autoneg
, true, true);
4526 adapter
->flags
|= IXGBE_FLAG_NEED_LINK_UPDATE
;
4527 adapter
->flags
&= ~IXGBE_FLAG_IN_SFP_LINK_TASK
;
4531 * ixgbe_sfp_config_module_task - worker thread to configure a new SFP+ module
4532 * @work: pointer to work_struct containing our data
4534 static void ixgbe_sfp_config_module_task(struct work_struct
*work
)
4536 struct ixgbe_adapter
*adapter
= container_of(work
,
4537 struct ixgbe_adapter
,
4538 sfp_config_module_task
);
4539 struct ixgbe_hw
*hw
= &adapter
->hw
;
4542 adapter
->flags
|= IXGBE_FLAG_IN_SFP_MOD_TASK
;
4544 /* Time for electrical oscillations to settle down */
4546 err
= hw
->phy
.ops
.identify_sfp(hw
);
4548 if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
4549 dev_err(&adapter
->pdev
->dev
, "failed to initialize because "
4550 "an unsupported SFP+ module type was detected.\n"
4551 "Reload the driver after installing a supported "
4553 unregister_netdev(adapter
->netdev
);
4556 hw
->mac
.ops
.setup_sfp(hw
);
4558 if (!(adapter
->flags
& IXGBE_FLAG_IN_SFP_LINK_TASK
))
4559 /* This will also work for DA Twinax connections */
4560 schedule_work(&adapter
->multispeed_fiber_task
);
4561 adapter
->flags
&= ~IXGBE_FLAG_IN_SFP_MOD_TASK
;
4565 * ixgbe_fdir_reinit_task - worker thread to reinit FDIR filter table
4566 * @work: pointer to work_struct containing our data
4568 static void ixgbe_fdir_reinit_task(struct work_struct
*work
)
4570 struct ixgbe_adapter
*adapter
= container_of(work
,
4571 struct ixgbe_adapter
,
4573 struct ixgbe_hw
*hw
= &adapter
->hw
;
4576 if (ixgbe_reinit_fdir_tables_82599(hw
) == 0) {
4577 for (i
= 0; i
< adapter
->num_tx_queues
; i
++)
4578 set_bit(__IXGBE_FDIR_INIT_DONE
,
4579 &(adapter
->tx_ring
[i
].reinit_state
));
4581 DPRINTK(PROBE
, ERR
, "failed to finish FDIR re-initialization, "
4582 "ignored adding FDIR ATR filters \n");
4584 /* Done FDIR Re-initialization, enable transmits */
4585 netif_tx_start_all_queues(adapter
->netdev
);
4589 * ixgbe_watchdog_task - worker thread to bring link up
4590 * @work: pointer to work_struct containing our data
4592 static void ixgbe_watchdog_task(struct work_struct
*work
)
4594 struct ixgbe_adapter
*adapter
= container_of(work
,
4595 struct ixgbe_adapter
,
4597 struct net_device
*netdev
= adapter
->netdev
;
4598 struct ixgbe_hw
*hw
= &adapter
->hw
;
4599 u32 link_speed
= adapter
->link_speed
;
4600 bool link_up
= adapter
->link_up
;
4602 struct ixgbe_ring
*tx_ring
;
4603 int some_tx_pending
= 0;
4605 adapter
->flags
|= IXGBE_FLAG_IN_WATCHDOG_TASK
;
4607 if (adapter
->flags
& IXGBE_FLAG_NEED_LINK_UPDATE
) {
4608 hw
->mac
.ops
.check_link(hw
, &link_speed
, &link_up
, false);
4611 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
4612 for (i
= 0; i
< MAX_TRAFFIC_CLASS
; i
++)
4613 hw
->mac
.ops
.fc_enable(hw
, i
);
4615 hw
->mac
.ops
.fc_enable(hw
, 0);
4618 hw
->mac
.ops
.fc_enable(hw
, 0);
4623 time_after(jiffies
, (adapter
->link_check_timeout
+
4624 IXGBE_TRY_LINK_TIMEOUT
))) {
4625 adapter
->flags
&= ~IXGBE_FLAG_NEED_LINK_UPDATE
;
4626 IXGBE_WRITE_REG(hw
, IXGBE_EIMS
, IXGBE_EIMC_LSC
);
4628 adapter
->link_up
= link_up
;
4629 adapter
->link_speed
= link_speed
;
4633 if (!netif_carrier_ok(netdev
)) {
4634 bool flow_rx
, flow_tx
;
4636 if (hw
->mac
.type
== ixgbe_mac_82599EB
) {
4637 u32 mflcn
= IXGBE_READ_REG(hw
, IXGBE_MFLCN
);
4638 u32 fccfg
= IXGBE_READ_REG(hw
, IXGBE_FCCFG
);
4639 flow_rx
= !!(mflcn
& IXGBE_MFLCN_RFCE
);
4640 flow_tx
= !!(fccfg
& IXGBE_FCCFG_TFCE_802_3X
);
4642 u32 frctl
= IXGBE_READ_REG(hw
, IXGBE_FCTRL
);
4643 u32 rmcs
= IXGBE_READ_REG(hw
, IXGBE_RMCS
);
4644 flow_rx
= !!(frctl
& IXGBE_FCTRL_RFCE
);
4645 flow_tx
= !!(rmcs
& IXGBE_RMCS_TFCE_802_3X
);
4648 printk(KERN_INFO
"ixgbe: %s NIC Link is Up %s, "
4649 "Flow Control: %s\n",
4651 (link_speed
== IXGBE_LINK_SPEED_10GB_FULL
?
4653 (link_speed
== IXGBE_LINK_SPEED_1GB_FULL
?
4654 "1 Gbps" : "unknown speed")),
4655 ((flow_rx
&& flow_tx
) ? "RX/TX" :
4657 (flow_tx
? "TX" : "None"))));
4659 netif_carrier_on(netdev
);
4661 /* Force detection of hung controller */
4662 adapter
->detect_tx_hung
= true;
4665 adapter
->link_up
= false;
4666 adapter
->link_speed
= 0;
4667 if (netif_carrier_ok(netdev
)) {
4668 printk(KERN_INFO
"ixgbe: %s NIC Link is Down\n",
4670 netif_carrier_off(netdev
);
4674 if (!netif_carrier_ok(netdev
)) {
4675 for (i
= 0; i
< adapter
->num_tx_queues
; i
++) {
4676 tx_ring
= &adapter
->tx_ring
[i
];
4677 if (tx_ring
->next_to_use
!= tx_ring
->next_to_clean
) {
4678 some_tx_pending
= 1;
4683 if (some_tx_pending
) {
4684 /* We've lost link, so the controller stops DMA,
4685 * but we've got queued Tx work that's never going
4686 * to get done, so reset controller to flush Tx.
4687 * (Do the reset outside of interrupt context).
4689 schedule_work(&adapter
->reset_task
);
4693 ixgbe_update_stats(adapter
);
4694 adapter
->flags
&= ~IXGBE_FLAG_IN_WATCHDOG_TASK
;
4697 static int ixgbe_tso(struct ixgbe_adapter
*adapter
,
4698 struct ixgbe_ring
*tx_ring
, struct sk_buff
*skb
,
4699 u32 tx_flags
, u8
*hdr_len
)
4701 struct ixgbe_adv_tx_context_desc
*context_desc
;
4704 struct ixgbe_tx_buffer
*tx_buffer_info
;
4705 u32 vlan_macip_lens
= 0, type_tucmd_mlhl
;
4706 u32 mss_l4len_idx
, l4len
;
4708 if (skb_is_gso(skb
)) {
4709 if (skb_header_cloned(skb
)) {
4710 err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
4714 l4len
= tcp_hdrlen(skb
);
4717 if (skb
->protocol
== htons(ETH_P_IP
)) {
4718 struct iphdr
*iph
= ip_hdr(skb
);
4721 tcp_hdr(skb
)->check
= ~csum_tcpudp_magic(iph
->saddr
,
4725 adapter
->hw_tso_ctxt
++;
4726 } else if (skb_shinfo(skb
)->gso_type
== SKB_GSO_TCPV6
) {
4727 ipv6_hdr(skb
)->payload_len
= 0;
4728 tcp_hdr(skb
)->check
=
4729 ~csum_ipv6_magic(&ipv6_hdr(skb
)->saddr
,
4730 &ipv6_hdr(skb
)->daddr
,
4732 adapter
->hw_tso6_ctxt
++;
4735 i
= tx_ring
->next_to_use
;
4737 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4738 context_desc
= IXGBE_TX_CTXTDESC_ADV(*tx_ring
, i
);
4740 /* VLAN MACLEN IPLEN */
4741 if (tx_flags
& IXGBE_TX_FLAGS_VLAN
)
4743 (tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
);
4744 vlan_macip_lens
|= ((skb_network_offset(skb
)) <<
4745 IXGBE_ADVTXD_MACLEN_SHIFT
);
4746 *hdr_len
+= skb_network_offset(skb
);
4748 (skb_transport_header(skb
) - skb_network_header(skb
));
4750 (skb_transport_header(skb
) - skb_network_header(skb
));
4751 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4752 context_desc
->seqnum_seed
= 0;
4754 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4755 type_tucmd_mlhl
= (IXGBE_TXD_CMD_DEXT
|
4756 IXGBE_ADVTXD_DTYP_CTXT
);
4758 if (skb
->protocol
== htons(ETH_P_IP
))
4759 type_tucmd_mlhl
|= IXGBE_ADVTXD_TUCMD_IPV4
;
4760 type_tucmd_mlhl
|= IXGBE_ADVTXD_TUCMD_L4T_TCP
;
4761 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd_mlhl
);
4765 (skb_shinfo(skb
)->gso_size
<< IXGBE_ADVTXD_MSS_SHIFT
);
4766 mss_l4len_idx
|= (l4len
<< IXGBE_ADVTXD_L4LEN_SHIFT
);
4767 /* use index 1 for TSO */
4768 mss_l4len_idx
|= (1 << IXGBE_ADVTXD_IDX_SHIFT
);
4769 context_desc
->mss_l4len_idx
= cpu_to_le32(mss_l4len_idx
);
4771 tx_buffer_info
->time_stamp
= jiffies
;
4772 tx_buffer_info
->next_to_watch
= i
;
4775 if (i
== tx_ring
->count
)
4777 tx_ring
->next_to_use
= i
;
4784 static bool ixgbe_tx_csum(struct ixgbe_adapter
*adapter
,
4785 struct ixgbe_ring
*tx_ring
,
4786 struct sk_buff
*skb
, u32 tx_flags
)
4788 struct ixgbe_adv_tx_context_desc
*context_desc
;
4790 struct ixgbe_tx_buffer
*tx_buffer_info
;
4791 u32 vlan_macip_lens
= 0, type_tucmd_mlhl
= 0;
4793 if (skb
->ip_summed
== CHECKSUM_PARTIAL
||
4794 (tx_flags
& IXGBE_TX_FLAGS_VLAN
)) {
4795 i
= tx_ring
->next_to_use
;
4796 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4797 context_desc
= IXGBE_TX_CTXTDESC_ADV(*tx_ring
, i
);
4799 if (tx_flags
& IXGBE_TX_FLAGS_VLAN
)
4801 (tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
);
4802 vlan_macip_lens
|= (skb_network_offset(skb
) <<
4803 IXGBE_ADVTXD_MACLEN_SHIFT
);
4804 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
4805 vlan_macip_lens
|= (skb_transport_header(skb
) -
4806 skb_network_header(skb
));
4808 context_desc
->vlan_macip_lens
= cpu_to_le32(vlan_macip_lens
);
4809 context_desc
->seqnum_seed
= 0;
4811 type_tucmd_mlhl
|= (IXGBE_TXD_CMD_DEXT
|
4812 IXGBE_ADVTXD_DTYP_CTXT
);
4814 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
4815 switch (skb
->protocol
) {
4816 case cpu_to_be16(ETH_P_IP
):
4817 type_tucmd_mlhl
|= IXGBE_ADVTXD_TUCMD_IPV4
;
4818 if (ip_hdr(skb
)->protocol
== IPPROTO_TCP
)
4820 IXGBE_ADVTXD_TUCMD_L4T_TCP
;
4821 else if (ip_hdr(skb
)->protocol
== IPPROTO_SCTP
)
4823 IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
4825 case cpu_to_be16(ETH_P_IPV6
):
4826 /* XXX what about other V6 headers?? */
4827 if (ipv6_hdr(skb
)->nexthdr
== IPPROTO_TCP
)
4829 IXGBE_ADVTXD_TUCMD_L4T_TCP
;
4830 else if (ipv6_hdr(skb
)->nexthdr
== IPPROTO_SCTP
)
4832 IXGBE_ADVTXD_TUCMD_L4T_SCTP
;
4835 if (unlikely(net_ratelimit())) {
4836 DPRINTK(PROBE
, WARNING
,
4837 "partial checksum but proto=%x!\n",
4844 context_desc
->type_tucmd_mlhl
= cpu_to_le32(type_tucmd_mlhl
);
4845 /* use index zero for tx checksum offload */
4846 context_desc
->mss_l4len_idx
= 0;
4848 tx_buffer_info
->time_stamp
= jiffies
;
4849 tx_buffer_info
->next_to_watch
= i
;
4851 adapter
->hw_csum_tx_good
++;
4853 if (i
== tx_ring
->count
)
4855 tx_ring
->next_to_use
= i
;
4863 static int ixgbe_tx_map(struct ixgbe_adapter
*adapter
,
4864 struct ixgbe_ring
*tx_ring
,
4865 struct sk_buff
*skb
, u32 tx_flags
,
4868 struct ixgbe_tx_buffer
*tx_buffer_info
;
4870 unsigned int total
= skb
->len
;
4871 unsigned int offset
= 0, size
, count
= 0, i
;
4872 unsigned int nr_frags
= skb_shinfo(skb
)->nr_frags
;
4876 i
= tx_ring
->next_to_use
;
4878 if (skb_dma_map(&adapter
->pdev
->dev
, skb
, DMA_TO_DEVICE
)) {
4879 dev_err(&adapter
->pdev
->dev
, "TX DMA map failed\n");
4883 map
= skb_shinfo(skb
)->dma_maps
;
4885 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
)
4886 /* excluding fcoe_crc_eof for FCoE */
4887 total
-= sizeof(struct fcoe_crc_eof
);
4889 len
= min(skb_headlen(skb
), total
);
4891 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4892 size
= min(len
, (uint
)IXGBE_MAX_DATA_PER_TXD
);
4894 tx_buffer_info
->length
= size
;
4895 tx_buffer_info
->dma
= skb_shinfo(skb
)->dma_head
+ offset
;
4896 tx_buffer_info
->time_stamp
= jiffies
;
4897 tx_buffer_info
->next_to_watch
= i
;
4906 if (i
== tx_ring
->count
)
4911 for (f
= 0; f
< nr_frags
; f
++) {
4912 struct skb_frag_struct
*frag
;
4914 frag
= &skb_shinfo(skb
)->frags
[f
];
4915 len
= min((unsigned int)frag
->size
, total
);
4920 if (i
== tx_ring
->count
)
4923 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4924 size
= min(len
, (uint
)IXGBE_MAX_DATA_PER_TXD
);
4926 tx_buffer_info
->length
= size
;
4927 tx_buffer_info
->dma
= map
[f
] + offset
;
4928 tx_buffer_info
->time_stamp
= jiffies
;
4929 tx_buffer_info
->next_to_watch
= i
;
4940 tx_ring
->tx_buffer_info
[i
].skb
= skb
;
4941 tx_ring
->tx_buffer_info
[first
].next_to_watch
= i
;
4946 static void ixgbe_tx_queue(struct ixgbe_adapter
*adapter
,
4947 struct ixgbe_ring
*tx_ring
,
4948 int tx_flags
, int count
, u32 paylen
, u8 hdr_len
)
4950 union ixgbe_adv_tx_desc
*tx_desc
= NULL
;
4951 struct ixgbe_tx_buffer
*tx_buffer_info
;
4952 u32 olinfo_status
= 0, cmd_type_len
= 0;
4954 u32 txd_cmd
= IXGBE_TXD_CMD_EOP
| IXGBE_TXD_CMD_RS
| IXGBE_TXD_CMD_IFCS
;
4956 cmd_type_len
|= IXGBE_ADVTXD_DTYP_DATA
;
4958 cmd_type_len
|= IXGBE_ADVTXD_DCMD_IFCS
| IXGBE_ADVTXD_DCMD_DEXT
;
4960 if (tx_flags
& IXGBE_TX_FLAGS_VLAN
)
4961 cmd_type_len
|= IXGBE_ADVTXD_DCMD_VLE
;
4963 if (tx_flags
& IXGBE_TX_FLAGS_TSO
) {
4964 cmd_type_len
|= IXGBE_ADVTXD_DCMD_TSE
;
4966 olinfo_status
|= IXGBE_TXD_POPTS_TXSM
<<
4967 IXGBE_ADVTXD_POPTS_SHIFT
;
4969 /* use index 1 context for tso */
4970 olinfo_status
|= (1 << IXGBE_ADVTXD_IDX_SHIFT
);
4971 if (tx_flags
& IXGBE_TX_FLAGS_IPV4
)
4972 olinfo_status
|= IXGBE_TXD_POPTS_IXSM
<<
4973 IXGBE_ADVTXD_POPTS_SHIFT
;
4975 } else if (tx_flags
& IXGBE_TX_FLAGS_CSUM
)
4976 olinfo_status
|= IXGBE_TXD_POPTS_TXSM
<<
4977 IXGBE_ADVTXD_POPTS_SHIFT
;
4979 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
) {
4980 olinfo_status
|= IXGBE_ADVTXD_CC
;
4981 olinfo_status
|= (1 << IXGBE_ADVTXD_IDX_SHIFT
);
4982 if (tx_flags
& IXGBE_TX_FLAGS_FSO
)
4983 cmd_type_len
|= IXGBE_ADVTXD_DCMD_TSE
;
4986 olinfo_status
|= ((paylen
- hdr_len
) << IXGBE_ADVTXD_PAYLEN_SHIFT
);
4988 i
= tx_ring
->next_to_use
;
4990 tx_buffer_info
= &tx_ring
->tx_buffer_info
[i
];
4991 tx_desc
= IXGBE_TX_DESC_ADV(*tx_ring
, i
);
4992 tx_desc
->read
.buffer_addr
= cpu_to_le64(tx_buffer_info
->dma
);
4993 tx_desc
->read
.cmd_type_len
=
4994 cpu_to_le32(cmd_type_len
| tx_buffer_info
->length
);
4995 tx_desc
->read
.olinfo_status
= cpu_to_le32(olinfo_status
);
4997 if (i
== tx_ring
->count
)
5001 tx_desc
->read
.cmd_type_len
|= cpu_to_le32(txd_cmd
);
5004 * Force memory writes to complete before letting h/w
5005 * know there are new descriptors to fetch. (Only
5006 * applicable for weak-ordered memory model archs,
5011 tx_ring
->next_to_use
= i
;
5012 writel(i
, adapter
->hw
.hw_addr
+ tx_ring
->tail
);
5015 static void ixgbe_atr(struct ixgbe_adapter
*adapter
, struct sk_buff
*skb
,
5016 int queue
, u32 tx_flags
)
5018 /* Right now, we support IPv4 only */
5019 struct ixgbe_atr_input atr_input
;
5022 struct iphdr
*iph
= ip_hdr(skb
);
5023 struct ethhdr
*eth
= (struct ethhdr
*)skb
->data
;
5024 u16 vlan_id
, src_port
, dst_port
, flex_bytes
;
5025 u32 src_ipv4_addr
, dst_ipv4_addr
;
5028 /* check if we're UDP or TCP */
5029 if (iph
->protocol
== IPPROTO_TCP
) {
5031 src_port
= th
->source
;
5032 dst_port
= th
->dest
;
5033 l4type
|= IXGBE_ATR_L4TYPE_TCP
;
5034 /* l4type IPv4 type is 0, no need to assign */
5035 } else if(iph
->protocol
== IPPROTO_UDP
) {
5037 src_port
= uh
->source
;
5038 dst_port
= uh
->dest
;
5039 l4type
|= IXGBE_ATR_L4TYPE_UDP
;
5040 /* l4type IPv4 type is 0, no need to assign */
5042 /* Unsupported L4 header, just bail here */
5046 memset(&atr_input
, 0, sizeof(struct ixgbe_atr_input
));
5048 vlan_id
= (tx_flags
& IXGBE_TX_FLAGS_VLAN_MASK
) >>
5049 IXGBE_TX_FLAGS_VLAN_SHIFT
;
5050 src_ipv4_addr
= iph
->saddr
;
5051 dst_ipv4_addr
= iph
->daddr
;
5052 flex_bytes
= eth
->h_proto
;
5054 ixgbe_atr_set_vlan_id_82599(&atr_input
, vlan_id
);
5055 ixgbe_atr_set_src_port_82599(&atr_input
, dst_port
);
5056 ixgbe_atr_set_dst_port_82599(&atr_input
, src_port
);
5057 ixgbe_atr_set_flex_byte_82599(&atr_input
, flex_bytes
);
5058 ixgbe_atr_set_l4type_82599(&atr_input
, l4type
);
5059 /* src and dst are inverted, think how the receiver sees them */
5060 ixgbe_atr_set_src_ipv4_82599(&atr_input
, dst_ipv4_addr
);
5061 ixgbe_atr_set_dst_ipv4_82599(&atr_input
, src_ipv4_addr
);
5063 /* This assumes the Rx queue and Tx queue are bound to the same CPU */
5064 ixgbe_fdir_add_signature_filter_82599(&adapter
->hw
, &atr_input
, queue
);
5067 static int __ixgbe_maybe_stop_tx(struct net_device
*netdev
,
5068 struct ixgbe_ring
*tx_ring
, int size
)
5070 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5072 netif_stop_subqueue(netdev
, tx_ring
->queue_index
);
5073 /* Herbert's original patch had:
5074 * smp_mb__after_netif_stop_queue();
5075 * but since that doesn't exist yet, just open code it. */
5078 /* We need to check again in a case another CPU has just
5079 * made room available. */
5080 if (likely(IXGBE_DESC_UNUSED(tx_ring
) < size
))
5083 /* A reprieve! - use start_queue because it doesn't call schedule */
5084 netif_start_subqueue(netdev
, tx_ring
->queue_index
);
5085 ++adapter
->restart_queue
;
5089 static int ixgbe_maybe_stop_tx(struct net_device
*netdev
,
5090 struct ixgbe_ring
*tx_ring
, int size
)
5092 if (likely(IXGBE_DESC_UNUSED(tx_ring
) >= size
))
5094 return __ixgbe_maybe_stop_tx(netdev
, tx_ring
, size
);
5097 static u16
ixgbe_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
5099 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
5101 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
)
5102 return smp_processor_id();
5104 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)
5105 return (skb
->vlan_tci
& IXGBE_TX_FLAGS_VLAN_PRIO_MASK
) >> 13;
5107 return skb_tx_hash(dev
, skb
);
5110 static int ixgbe_xmit_frame(struct sk_buff
*skb
, struct net_device
*netdev
)
5112 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5113 struct ixgbe_ring
*tx_ring
;
5115 unsigned int tx_flags
= 0;
5121 if (adapter
->vlgrp
&& vlan_tx_tag_present(skb
)) {
5122 tx_flags
|= vlan_tx_tag_get(skb
);
5123 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
5124 tx_flags
&= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK
;
5125 tx_flags
|= (skb
->queue_mapping
<< 13);
5127 tx_flags
<<= IXGBE_TX_FLAGS_VLAN_SHIFT
;
5128 tx_flags
|= IXGBE_TX_FLAGS_VLAN
;
5129 } else if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
) {
5130 if (skb
->priority
!= TC_PRIO_CONTROL
) {
5131 tx_flags
|= (skb
->queue_mapping
<< 13);
5132 tx_flags
<<= IXGBE_TX_FLAGS_VLAN_SHIFT
;
5133 tx_flags
|= IXGBE_TX_FLAGS_VLAN
;
5135 skb
->queue_mapping
=
5136 adapter
->ring_feature
[RING_F_DCB
].indices
-1;
5140 r_idx
= skb
->queue_mapping
;
5141 tx_ring
= &adapter
->tx_ring
[r_idx
];
5143 if ((adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
) &&
5144 (skb
->protocol
== htons(ETH_P_FCOE
)))
5145 tx_flags
|= IXGBE_TX_FLAGS_FCOE
;
5147 /* four things can cause us to need a context descriptor */
5148 if (skb_is_gso(skb
) ||
5149 (skb
->ip_summed
== CHECKSUM_PARTIAL
) ||
5150 (tx_flags
& IXGBE_TX_FLAGS_VLAN
) ||
5151 (tx_flags
& IXGBE_TX_FLAGS_FCOE
))
5154 count
+= TXD_USE_COUNT(skb_headlen(skb
));
5155 for (f
= 0; f
< skb_shinfo(skb
)->nr_frags
; f
++)
5156 count
+= TXD_USE_COUNT(skb_shinfo(skb
)->frags
[f
].size
);
5158 if (ixgbe_maybe_stop_tx(netdev
, tx_ring
, count
)) {
5160 return NETDEV_TX_BUSY
;
5163 first
= tx_ring
->next_to_use
;
5164 if (tx_flags
& IXGBE_TX_FLAGS_FCOE
) {
5166 /* setup tx offload for FCoE */
5167 tso
= ixgbe_fso(adapter
, tx_ring
, skb
, tx_flags
, &hdr_len
);
5169 dev_kfree_skb_any(skb
);
5170 return NETDEV_TX_OK
;
5173 tx_flags
|= IXGBE_TX_FLAGS_FSO
;
5174 #endif /* IXGBE_FCOE */
5176 if (skb
->protocol
== htons(ETH_P_IP
))
5177 tx_flags
|= IXGBE_TX_FLAGS_IPV4
;
5178 tso
= ixgbe_tso(adapter
, tx_ring
, skb
, tx_flags
, &hdr_len
);
5180 dev_kfree_skb_any(skb
);
5181 return NETDEV_TX_OK
;
5185 tx_flags
|= IXGBE_TX_FLAGS_TSO
;
5186 else if (ixgbe_tx_csum(adapter
, tx_ring
, skb
, tx_flags
) &&
5187 (skb
->ip_summed
== CHECKSUM_PARTIAL
))
5188 tx_flags
|= IXGBE_TX_FLAGS_CSUM
;
5191 count
= ixgbe_tx_map(adapter
, tx_ring
, skb
, tx_flags
, first
);
5193 /* add the ATR filter if ATR is on */
5194 if (tx_ring
->atr_sample_rate
) {
5195 ++tx_ring
->atr_count
;
5196 if ((tx_ring
->atr_count
>= tx_ring
->atr_sample_rate
) &&
5197 test_bit(__IXGBE_FDIR_INIT_DONE
,
5198 &tx_ring
->reinit_state
)) {
5199 ixgbe_atr(adapter
, skb
, tx_ring
->queue_index
,
5201 tx_ring
->atr_count
= 0;
5204 ixgbe_tx_queue(adapter
, tx_ring
, tx_flags
, count
, skb
->len
,
5206 ixgbe_maybe_stop_tx(netdev
, tx_ring
, DESC_NEEDED
);
5209 dev_kfree_skb_any(skb
);
5210 tx_ring
->tx_buffer_info
[first
].time_stamp
= 0;
5211 tx_ring
->next_to_use
= first
;
5214 return NETDEV_TX_OK
;
5218 * ixgbe_get_stats - Get System Network Statistics
5219 * @netdev: network interface device structure
5221 * Returns the address of the device statistics structure.
5222 * The statistics are actually updated from the timer callback.
5224 static struct net_device_stats
*ixgbe_get_stats(struct net_device
*netdev
)
5226 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5228 /* only return the current stats */
5229 return &adapter
->net_stats
;
5233 * ixgbe_set_mac - Change the Ethernet Address of the NIC
5234 * @netdev: network interface device structure
5235 * @p: pointer to an address structure
5237 * Returns 0 on success, negative on failure
5239 static int ixgbe_set_mac(struct net_device
*netdev
, void *p
)
5241 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5242 struct ixgbe_hw
*hw
= &adapter
->hw
;
5243 struct sockaddr
*addr
= p
;
5245 if (!is_valid_ether_addr(addr
->sa_data
))
5246 return -EADDRNOTAVAIL
;
5248 memcpy(netdev
->dev_addr
, addr
->sa_data
, netdev
->addr_len
);
5249 memcpy(hw
->mac
.addr
, addr
->sa_data
, netdev
->addr_len
);
5251 hw
->mac
.ops
.set_rar(hw
, 0, hw
->mac
.addr
, 0, IXGBE_RAH_AV
);
5257 ixgbe_mdio_read(struct net_device
*netdev
, int prtad
, int devad
, u16 addr
)
5259 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5260 struct ixgbe_hw
*hw
= &adapter
->hw
;
5264 if (prtad
!= hw
->phy
.mdio
.prtad
)
5266 rc
= hw
->phy
.ops
.read_reg(hw
, addr
, devad
, &value
);
5272 static int ixgbe_mdio_write(struct net_device
*netdev
, int prtad
, int devad
,
5273 u16 addr
, u16 value
)
5275 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5276 struct ixgbe_hw
*hw
= &adapter
->hw
;
5278 if (prtad
!= hw
->phy
.mdio
.prtad
)
5280 return hw
->phy
.ops
.write_reg(hw
, addr
, devad
, value
);
5283 static int ixgbe_ioctl(struct net_device
*netdev
, struct ifreq
*req
, int cmd
)
5285 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5287 return mdio_mii_ioctl(&adapter
->hw
.phy
.mdio
, if_mii(req
), cmd
);
5291 * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
5293 * @netdev: network interface device structure
5295 * Returns non-zero on failure
5297 static int ixgbe_add_sanmac_netdev(struct net_device
*dev
)
5300 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
5301 struct ixgbe_mac_info
*mac
= &adapter
->hw
.mac
;
5303 if (is_valid_ether_addr(mac
->san_addr
)) {
5305 err
= dev_addr_add(dev
, mac
->san_addr
, NETDEV_HW_ADDR_T_SAN
);
5312 * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
5314 * @netdev: network interface device structure
5316 * Returns non-zero on failure
5318 static int ixgbe_del_sanmac_netdev(struct net_device
*dev
)
5321 struct ixgbe_adapter
*adapter
= netdev_priv(dev
);
5322 struct ixgbe_mac_info
*mac
= &adapter
->hw
.mac
;
5324 if (is_valid_ether_addr(mac
->san_addr
)) {
5326 err
= dev_addr_del(dev
, mac
->san_addr
, NETDEV_HW_ADDR_T_SAN
);
5332 #ifdef CONFIG_NET_POLL_CONTROLLER
5334 * Polling 'interrupt' - used by things like netconsole to send skbs
5335 * without having to re-enable interrupts. It's not called while
5336 * the interrupt routine is executing.
5338 static void ixgbe_netpoll(struct net_device
*netdev
)
5340 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5343 adapter
->flags
|= IXGBE_FLAG_IN_NETPOLL
;
5344 if (adapter
->flags
& IXGBE_FLAG_MSIX_ENABLED
) {
5345 int num_q_vectors
= adapter
->num_msix_vectors
- NON_Q_VECTORS
;
5346 for (i
= 0; i
< num_q_vectors
; i
++) {
5347 struct ixgbe_q_vector
*q_vector
= adapter
->q_vector
[i
];
5348 ixgbe_msix_clean_many(0, q_vector
);
5351 ixgbe_intr(adapter
->pdev
->irq
, netdev
);
5353 adapter
->flags
&= ~IXGBE_FLAG_IN_NETPOLL
;
5357 static const struct net_device_ops ixgbe_netdev_ops
= {
5358 .ndo_open
= ixgbe_open
,
5359 .ndo_stop
= ixgbe_close
,
5360 .ndo_start_xmit
= ixgbe_xmit_frame
,
5361 .ndo_select_queue
= ixgbe_select_queue
,
5362 .ndo_get_stats
= ixgbe_get_stats
,
5363 .ndo_set_rx_mode
= ixgbe_set_rx_mode
,
5364 .ndo_set_multicast_list
= ixgbe_set_rx_mode
,
5365 .ndo_validate_addr
= eth_validate_addr
,
5366 .ndo_set_mac_address
= ixgbe_set_mac
,
5367 .ndo_change_mtu
= ixgbe_change_mtu
,
5368 .ndo_tx_timeout
= ixgbe_tx_timeout
,
5369 .ndo_vlan_rx_register
= ixgbe_vlan_rx_register
,
5370 .ndo_vlan_rx_add_vid
= ixgbe_vlan_rx_add_vid
,
5371 .ndo_vlan_rx_kill_vid
= ixgbe_vlan_rx_kill_vid
,
5372 .ndo_do_ioctl
= ixgbe_ioctl
,
5373 #ifdef CONFIG_NET_POLL_CONTROLLER
5374 .ndo_poll_controller
= ixgbe_netpoll
,
5377 .ndo_fcoe_ddp_setup
= ixgbe_fcoe_ddp_get
,
5378 .ndo_fcoe_ddp_done
= ixgbe_fcoe_ddp_put
,
5379 #endif /* IXGBE_FCOE */
5383 * ixgbe_probe - Device Initialization Routine
5384 * @pdev: PCI device information struct
5385 * @ent: entry in ixgbe_pci_tbl
5387 * Returns 0 on success, negative on failure
5389 * ixgbe_probe initializes an adapter identified by a pci_dev structure.
5390 * The OS initialization, configuring of the adapter private structure,
5391 * and a hardware reset occur.
5393 static int __devinit
ixgbe_probe(struct pci_dev
*pdev
,
5394 const struct pci_device_id
*ent
)
5396 struct net_device
*netdev
;
5397 struct ixgbe_adapter
*adapter
= NULL
;
5398 struct ixgbe_hw
*hw
;
5399 const struct ixgbe_info
*ii
= ixgbe_info_tbl
[ent
->driver_data
];
5400 static int cards_found
;
5401 int i
, err
, pci_using_dac
;
5407 err
= pci_enable_device_mem(pdev
);
5411 if (!pci_set_dma_mask(pdev
, DMA_BIT_MASK(64)) &&
5412 !pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(64))) {
5415 err
= pci_set_dma_mask(pdev
, DMA_BIT_MASK(32));
5417 err
= pci_set_consistent_dma_mask(pdev
, DMA_BIT_MASK(32));
5419 dev_err(&pdev
->dev
, "No usable DMA "
5420 "configuration, aborting\n");
5427 err
= pci_request_selected_regions(pdev
, pci_select_bars(pdev
,
5428 IORESOURCE_MEM
), ixgbe_driver_name
);
5431 "pci_request_selected_regions failed 0x%x\n", err
);
5435 err
= pci_enable_pcie_error_reporting(pdev
);
5437 dev_err(&pdev
->dev
, "pci_enable_pcie_error_reporting failed "
5439 /* non-fatal, continue */
5442 pci_set_master(pdev
);
5443 pci_save_state(pdev
);
5445 netdev
= alloc_etherdev_mq(sizeof(struct ixgbe_adapter
), MAX_TX_QUEUES
);
5448 goto err_alloc_etherdev
;
5451 SET_NETDEV_DEV(netdev
, &pdev
->dev
);
5453 pci_set_drvdata(pdev
, netdev
);
5454 adapter
= netdev_priv(netdev
);
5456 adapter
->netdev
= netdev
;
5457 adapter
->pdev
= pdev
;
5460 adapter
->msg_enable
= (1 << DEFAULT_DEBUG_LEVEL_SHIFT
) - 1;
5462 hw
->hw_addr
= ioremap(pci_resource_start(pdev
, 0),
5463 pci_resource_len(pdev
, 0));
5469 for (i
= 1; i
<= 5; i
++) {
5470 if (pci_resource_len(pdev
, i
) == 0)
5474 netdev
->netdev_ops
= &ixgbe_netdev_ops
;
5475 ixgbe_set_ethtool_ops(netdev
);
5476 netdev
->watchdog_timeo
= 5 * HZ
;
5477 strcpy(netdev
->name
, pci_name(pdev
));
5479 adapter
->bd_number
= cards_found
;
5482 memcpy(&hw
->mac
.ops
, ii
->mac_ops
, sizeof(hw
->mac
.ops
));
5483 hw
->mac
.type
= ii
->mac
;
5486 memcpy(&hw
->eeprom
.ops
, ii
->eeprom_ops
, sizeof(hw
->eeprom
.ops
));
5487 eec
= IXGBE_READ_REG(hw
, IXGBE_EEC
);
5488 /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
5489 if (!(eec
& (1 << 8)))
5490 hw
->eeprom
.ops
.read
= &ixgbe_read_eeprom_bit_bang_generic
;
5493 memcpy(&hw
->phy
.ops
, ii
->phy_ops
, sizeof(hw
->phy
.ops
));
5494 hw
->phy
.sfp_type
= ixgbe_sfp_type_unknown
;
5495 /* ixgbe_identify_phy_generic will set prtad and mmds properly */
5496 hw
->phy
.mdio
.prtad
= MDIO_PRTAD_NONE
;
5497 hw
->phy
.mdio
.mmds
= 0;
5498 hw
->phy
.mdio
.mode_support
= MDIO_SUPPORTS_C45
| MDIO_EMULATE_C22
;
5499 hw
->phy
.mdio
.dev
= netdev
;
5500 hw
->phy
.mdio
.mdio_read
= ixgbe_mdio_read
;
5501 hw
->phy
.mdio
.mdio_write
= ixgbe_mdio_write
;
5503 /* set up this timer and work struct before calling get_invariants
5504 * which might start the timer
5506 init_timer(&adapter
->sfp_timer
);
5507 adapter
->sfp_timer
.function
= &ixgbe_sfp_timer
;
5508 adapter
->sfp_timer
.data
= (unsigned long) adapter
;
5510 INIT_WORK(&adapter
->sfp_task
, ixgbe_sfp_task
);
5512 /* multispeed fiber has its own tasklet, called from GPI SDP1 context */
5513 INIT_WORK(&adapter
->multispeed_fiber_task
, ixgbe_multispeed_fiber_task
);
5515 /* a new SFP+ module arrival, called from GPI SDP2 context */
5516 INIT_WORK(&adapter
->sfp_config_module_task
,
5517 ixgbe_sfp_config_module_task
);
5519 ii
->get_invariants(hw
);
5521 /* setup the private structure */
5522 err
= ixgbe_sw_init(adapter
);
5527 * If there is a fan on this device and it has failed log the
5530 if (adapter
->flags
& IXGBE_FLAG_FAN_FAIL_CAPABLE
) {
5531 u32 esdp
= IXGBE_READ_REG(hw
, IXGBE_ESDP
);
5532 if (esdp
& IXGBE_ESDP_SDP1
)
5533 DPRINTK(PROBE
, CRIT
,
5534 "Fan has stopped, replace the adapter\n");
5537 /* reset_hw fills in the perm_addr as well */
5538 err
= hw
->mac
.ops
.reset_hw(hw
);
5539 if (err
== IXGBE_ERR_SFP_NOT_PRESENT
&&
5540 hw
->mac
.type
== ixgbe_mac_82598EB
) {
5542 * Start a kernel thread to watch for a module to arrive.
5543 * Only do this for 82598, since 82599 will generate
5544 * interrupts on module arrival.
5546 set_bit(__IXGBE_SFP_MODULE_NOT_FOUND
, &adapter
->state
);
5547 mod_timer(&adapter
->sfp_timer
,
5548 round_jiffies(jiffies
+ (2 * HZ
)));
5550 } else if (err
== IXGBE_ERR_SFP_NOT_SUPPORTED
) {
5551 dev_err(&adapter
->pdev
->dev
, "failed to initialize because "
5552 "an unsupported SFP+ module type was detected.\n"
5553 "Reload the driver after installing a supported "
5557 dev_err(&adapter
->pdev
->dev
, "HW Init failed: %d\n", err
);
5561 netdev
->features
= NETIF_F_SG
|
5563 NETIF_F_HW_VLAN_TX
|
5564 NETIF_F_HW_VLAN_RX
|
5565 NETIF_F_HW_VLAN_FILTER
;
5567 netdev
->features
|= NETIF_F_IPV6_CSUM
;
5568 netdev
->features
|= NETIF_F_TSO
;
5569 netdev
->features
|= NETIF_F_TSO6
;
5570 netdev
->features
|= NETIF_F_GRO
;
5572 if (adapter
->hw
.mac
.type
== ixgbe_mac_82599EB
)
5573 netdev
->features
|= NETIF_F_SCTP_CSUM
;
5575 netdev
->vlan_features
|= NETIF_F_TSO
;
5576 netdev
->vlan_features
|= NETIF_F_TSO6
;
5577 netdev
->vlan_features
|= NETIF_F_IP_CSUM
;
5578 netdev
->vlan_features
|= NETIF_F_SG
;
5580 if (adapter
->flags
& IXGBE_FLAG_DCB_ENABLED
)
5581 adapter
->flags
&= ~IXGBE_FLAG_RSS_ENABLED
;
5583 #ifdef CONFIG_IXGBE_DCB
5584 netdev
->dcbnl_ops
= &dcbnl_ops
;
5588 if (adapter
->flags
& IXGBE_FLAG_FCOE_CAPABLE
) {
5589 if (hw
->mac
.ops
.get_device_caps
) {
5590 hw
->mac
.ops
.get_device_caps(hw
, &device_caps
);
5591 if (device_caps
& IXGBE_DEVICE_CAPS_FCOE_OFFLOADS
)
5592 adapter
->flags
&= ~IXGBE_FLAG_FCOE_CAPABLE
;
5595 #endif /* IXGBE_FCOE */
5597 netdev
->features
|= NETIF_F_HIGHDMA
;
5599 if (adapter
->flags2
& IXGBE_FLAG2_RSC_ENABLED
)
5600 netdev
->features
|= NETIF_F_LRO
;
5602 /* make sure the EEPROM is good */
5603 if (hw
->eeprom
.ops
.validate_checksum(hw
, NULL
) < 0) {
5604 dev_err(&pdev
->dev
, "The EEPROM Checksum Is Not Valid\n");
5609 memcpy(netdev
->dev_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
5610 memcpy(netdev
->perm_addr
, hw
->mac
.perm_addr
, netdev
->addr_len
);
5612 if (ixgbe_validate_mac_addr(netdev
->perm_addr
)) {
5613 dev_err(&pdev
->dev
, "invalid MAC address\n");
5618 init_timer(&adapter
->watchdog_timer
);
5619 adapter
->watchdog_timer
.function
= &ixgbe_watchdog
;
5620 adapter
->watchdog_timer
.data
= (unsigned long)adapter
;
5622 INIT_WORK(&adapter
->reset_task
, ixgbe_reset_task
);
5623 INIT_WORK(&adapter
->watchdog_task
, ixgbe_watchdog_task
);
5625 err
= ixgbe_init_interrupt_scheme(adapter
);
5629 switch (pdev
->device
) {
5630 case IXGBE_DEV_ID_82599_KX4
:
5631 adapter
->wol
= (IXGBE_WUFC_MAG
| IXGBE_WUFC_EX
|
5632 IXGBE_WUFC_MC
| IXGBE_WUFC_BC
);
5633 /* Enable ACPI wakeup in GRC */
5634 IXGBE_WRITE_REG(hw
, IXGBE_GRC
,
5635 (IXGBE_READ_REG(hw
, IXGBE_GRC
) & ~IXGBE_GRC_APME
));
5641 device_set_wakeup_enable(&adapter
->pdev
->dev
, adapter
->wol
);
5643 /* pick up the PCI bus settings for reporting later */
5644 hw
->mac
.ops
.get_bus_info(hw
);
5646 /* print bus type/speed/width info */
5647 dev_info(&pdev
->dev
, "(PCI Express:%s:%s) %pM\n",
5648 ((hw
->bus
.speed
== ixgbe_bus_speed_5000
) ? "5.0Gb/s":
5649 (hw
->bus
.speed
== ixgbe_bus_speed_2500
) ? "2.5Gb/s":"Unknown"),
5650 ((hw
->bus
.width
== ixgbe_bus_width_pcie_x8
) ? "Width x8" :
5651 (hw
->bus
.width
== ixgbe_bus_width_pcie_x4
) ? "Width x4" :
5652 (hw
->bus
.width
== ixgbe_bus_width_pcie_x1
) ? "Width x1" :
5655 ixgbe_read_pba_num_generic(hw
, &part_num
);
5656 if (ixgbe_is_sfp(hw
) && hw
->phy
.sfp_type
!= ixgbe_sfp_type_not_present
)
5657 dev_info(&pdev
->dev
, "MAC: %d, PHY: %d, SFP+: %d, PBA No: %06x-%03x\n",
5658 hw
->mac
.type
, hw
->phy
.type
, hw
->phy
.sfp_type
,
5659 (part_num
>> 8), (part_num
& 0xff));
5661 dev_info(&pdev
->dev
, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5662 hw
->mac
.type
, hw
->phy
.type
,
5663 (part_num
>> 8), (part_num
& 0xff));
5665 if (hw
->bus
.width
<= ixgbe_bus_width_pcie_x4
) {
5666 dev_warn(&pdev
->dev
, "PCI-Express bandwidth available for "
5667 "this card is not sufficient for optimal "
5669 dev_warn(&pdev
->dev
, "For optimal performance a x8 "
5670 "PCI-Express slot is required.\n");
5673 /* save off EEPROM version number */
5674 hw
->eeprom
.ops
.read(hw
, 0x29, &adapter
->eeprom_version
);
5676 /* reset the hardware with the new settings */
5677 err
= hw
->mac
.ops
.start_hw(hw
);
5679 if (err
== IXGBE_ERR_EEPROM_VERSION
) {
5680 /* We are running on a pre-production device, log a warning */
5681 dev_warn(&pdev
->dev
, "This device is a pre-production "
5682 "adapter/LOM. Please be aware there may be issues "
5683 "associated with your hardware. If you are "
5684 "experiencing problems please contact your Intel or "
5685 "hardware representative who provided you with this "
5688 strcpy(netdev
->name
, "eth%d");
5689 err
= register_netdev(netdev
);
5693 /* carrier off reporting is important to ethtool even BEFORE open */
5694 netif_carrier_off(netdev
);
5696 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
5697 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
5698 INIT_WORK(&adapter
->fdir_reinit_task
, ixgbe_fdir_reinit_task
);
5700 #ifdef CONFIG_IXGBE_DCA
5701 if (dca_add_requester(&pdev
->dev
) == 0) {
5702 adapter
->flags
|= IXGBE_FLAG_DCA_ENABLED
;
5703 ixgbe_setup_dca(adapter
);
5706 /* add san mac addr to netdev */
5707 ixgbe_add_sanmac_netdev(netdev
);
5709 dev_info(&pdev
->dev
, "Intel(R) 10 Gigabit Network Connection\n");
5714 ixgbe_release_hw_control(adapter
);
5715 ixgbe_clear_interrupt_scheme(adapter
);
5718 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND
, &adapter
->state
);
5719 del_timer_sync(&adapter
->sfp_timer
);
5720 cancel_work_sync(&adapter
->sfp_task
);
5721 cancel_work_sync(&adapter
->multispeed_fiber_task
);
5722 cancel_work_sync(&adapter
->sfp_config_module_task
);
5723 iounmap(hw
->hw_addr
);
5725 free_netdev(netdev
);
5727 pci_release_selected_regions(pdev
, pci_select_bars(pdev
,
5731 pci_disable_device(pdev
);
5736 * ixgbe_remove - Device Removal Routine
5737 * @pdev: PCI device information struct
5739 * ixgbe_remove is called by the PCI subsystem to alert the driver
5740 * that it should release a PCI device. The could be caused by a
5741 * Hot-Plug event, or because the driver is going to be removed from
5744 static void __devexit
ixgbe_remove(struct pci_dev
*pdev
)
5746 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5747 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5750 set_bit(__IXGBE_DOWN
, &adapter
->state
);
5751 /* clear the module not found bit to make sure the worker won't
5754 clear_bit(__IXGBE_SFP_MODULE_NOT_FOUND
, &adapter
->state
);
5755 del_timer_sync(&adapter
->watchdog_timer
);
5757 del_timer_sync(&adapter
->sfp_timer
);
5758 cancel_work_sync(&adapter
->watchdog_task
);
5759 cancel_work_sync(&adapter
->sfp_task
);
5760 cancel_work_sync(&adapter
->multispeed_fiber_task
);
5761 cancel_work_sync(&adapter
->sfp_config_module_task
);
5762 if (adapter
->flags
& IXGBE_FLAG_FDIR_HASH_CAPABLE
||
5763 adapter
->flags
& IXGBE_FLAG_FDIR_PERFECT_CAPABLE
)
5764 cancel_work_sync(&adapter
->fdir_reinit_task
);
5765 flush_scheduled_work();
5767 #ifdef CONFIG_IXGBE_DCA
5768 if (adapter
->flags
& IXGBE_FLAG_DCA_ENABLED
) {
5769 adapter
->flags
&= ~IXGBE_FLAG_DCA_ENABLED
;
5770 dca_remove_requester(&pdev
->dev
);
5771 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_DCA_CTRL
, 1);
5776 if (adapter
->flags
& IXGBE_FLAG_FCOE_ENABLED
)
5777 ixgbe_cleanup_fcoe(adapter
);
5779 #endif /* IXGBE_FCOE */
5781 /* remove the added san mac */
5782 ixgbe_del_sanmac_netdev(netdev
);
5784 if (netdev
->reg_state
== NETREG_REGISTERED
)
5785 unregister_netdev(netdev
);
5787 ixgbe_clear_interrupt_scheme(adapter
);
5789 ixgbe_release_hw_control(adapter
);
5791 iounmap(adapter
->hw
.hw_addr
);
5792 pci_release_selected_regions(pdev
, pci_select_bars(pdev
,
5795 DPRINTK(PROBE
, INFO
, "complete\n");
5797 free_netdev(netdev
);
5799 err
= pci_disable_pcie_error_reporting(pdev
);
5802 "pci_disable_pcie_error_reporting failed 0x%x\n", err
);
5804 pci_disable_device(pdev
);
5808 * ixgbe_io_error_detected - called when PCI error is detected
5809 * @pdev: Pointer to PCI device
5810 * @state: The current pci connection state
5812 * This function is called after a PCI bus error affecting
5813 * this device has been detected.
5815 static pci_ers_result_t
ixgbe_io_error_detected(struct pci_dev
*pdev
,
5816 pci_channel_state_t state
)
5818 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5819 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5821 netif_device_detach(netdev
);
5823 if (state
== pci_channel_io_perm_failure
)
5824 return PCI_ERS_RESULT_DISCONNECT
;
5826 if (netif_running(netdev
))
5827 ixgbe_down(adapter
);
5828 pci_disable_device(pdev
);
5830 /* Request a slot reset. */
5831 return PCI_ERS_RESULT_NEED_RESET
;
5835 * ixgbe_io_slot_reset - called after the pci bus has been reset.
5836 * @pdev: Pointer to PCI device
5838 * Restart the card from scratch, as if from a cold-boot.
5840 static pci_ers_result_t
ixgbe_io_slot_reset(struct pci_dev
*pdev
)
5842 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5843 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5844 pci_ers_result_t result
;
5847 if (pci_enable_device_mem(pdev
)) {
5849 "Cannot re-enable PCI device after reset.\n");
5850 result
= PCI_ERS_RESULT_DISCONNECT
;
5852 pci_set_master(pdev
);
5853 pci_restore_state(pdev
);
5855 pci_wake_from_d3(pdev
, false);
5857 ixgbe_reset(adapter
);
5858 IXGBE_WRITE_REG(&adapter
->hw
, IXGBE_WUS
, ~0);
5859 result
= PCI_ERS_RESULT_RECOVERED
;
5862 err
= pci_cleanup_aer_uncorrect_error_status(pdev
);
5865 "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n", err
);
5866 /* non-fatal, continue */
5873 * ixgbe_io_resume - called when traffic can start flowing again.
5874 * @pdev: Pointer to PCI device
5876 * This callback is called when the error recovery driver tells us that
5877 * its OK to resume normal operation.
5879 static void ixgbe_io_resume(struct pci_dev
*pdev
)
5881 struct net_device
*netdev
= pci_get_drvdata(pdev
);
5882 struct ixgbe_adapter
*adapter
= netdev_priv(netdev
);
5884 if (netif_running(netdev
)) {
5885 if (ixgbe_up(adapter
)) {
5886 DPRINTK(PROBE
, INFO
, "ixgbe_up failed after reset\n");
5891 netif_device_attach(netdev
);
5894 static struct pci_error_handlers ixgbe_err_handler
= {
5895 .error_detected
= ixgbe_io_error_detected
,
5896 .slot_reset
= ixgbe_io_slot_reset
,
5897 .resume
= ixgbe_io_resume
,
5900 static struct pci_driver ixgbe_driver
= {
5901 .name
= ixgbe_driver_name
,
5902 .id_table
= ixgbe_pci_tbl
,
5903 .probe
= ixgbe_probe
,
5904 .remove
= __devexit_p(ixgbe_remove
),
5906 .suspend
= ixgbe_suspend
,
5907 .resume
= ixgbe_resume
,
5909 .shutdown
= ixgbe_shutdown
,
5910 .err_handler
= &ixgbe_err_handler
5914 * ixgbe_init_module - Driver Registration Routine
5916 * ixgbe_init_module is the first routine called when the driver is
5917 * loaded. All it does is register with the PCI subsystem.
5919 static int __init
ixgbe_init_module(void)
5922 printk(KERN_INFO
"%s: %s - version %s\n", ixgbe_driver_name
,
5923 ixgbe_driver_string
, ixgbe_driver_version
);
5925 printk(KERN_INFO
"%s: %s\n", ixgbe_driver_name
, ixgbe_copyright
);
5927 #ifdef CONFIG_IXGBE_DCA
5928 dca_register_notify(&dca_notifier
);
5931 ret
= pci_register_driver(&ixgbe_driver
);
5935 module_init(ixgbe_init_module
);
5938 * ixgbe_exit_module - Driver Exit Cleanup Routine
5940 * ixgbe_exit_module is called just before the driver is removed
5943 static void __exit
ixgbe_exit_module(void)
5945 #ifdef CONFIG_IXGBE_DCA
5946 dca_unregister_notify(&dca_notifier
);
5948 pci_unregister_driver(&ixgbe_driver
);
5951 #ifdef CONFIG_IXGBE_DCA
5952 static int ixgbe_notify_dca(struct notifier_block
*nb
, unsigned long event
,
5957 ret_val
= driver_for_each_device(&ixgbe_driver
.driver
, NULL
, &event
,
5958 __ixgbe_notify_dca
);
5960 return ret_val
? NOTIFY_BAD
: NOTIFY_DONE
;
5963 #endif /* CONFIG_IXGBE_DCA */
5966 * ixgbe_get_hw_dev_name - return device name string
5967 * used by hardware layer to print debugging information
5969 char *ixgbe_get_hw_dev_name(struct ixgbe_hw
*hw
)
5971 struct ixgbe_adapter
*adapter
= hw
->back
;
5972 return adapter
->netdev
->name
;
5976 module_exit(ixgbe_exit_module
);