net: remove interrupt.h inclusion from netdevice.h
[linux-3.10.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2011 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   Linux NICS <linux.nics@intel.com>
24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26
27 *******************************************************************************/
28
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31 #include <linux/module.h>
32 #include <linux/types.h>
33 #include <linux/init.h>
34 #include <linux/interrupt.h>
35 #include <linux/pci.h>
36 #include <linux/vmalloc.h>
37 #include <linux/pagemap.h>
38 #include <linux/delay.h>
39 #include <linux/netdevice.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/mii.h>
46 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/cpu.h>
49 #include <linux/smp.h>
50 #include <linux/pm_qos_params.h>
51 #include <linux/pm_runtime.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
54
55 #include "e1000.h"
56
57 #define DRV_EXTRAVERSION "-k2"
58
59 #define DRV_VERSION "1.3.10" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
62
63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
64
65 static const struct e1000_info *e1000_info_tbl[] = {
66         [board_82571]           = &e1000_82571_info,
67         [board_82572]           = &e1000_82572_info,
68         [board_82573]           = &e1000_82573_info,
69         [board_82574]           = &e1000_82574_info,
70         [board_82583]           = &e1000_82583_info,
71         [board_80003es2lan]     = &e1000_es2_info,
72         [board_ich8lan]         = &e1000_ich8_info,
73         [board_ich9lan]         = &e1000_ich9_info,
74         [board_ich10lan]        = &e1000_ich10_info,
75         [board_pchlan]          = &e1000_pch_info,
76         [board_pch2lan]         = &e1000_pch2_info,
77 };
78
79 struct e1000_reg_info {
80         u32 ofs;
81         char *name;
82 };
83
84 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
85 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
86 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
87 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
88 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
89
90 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
91 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
92 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
93 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
94 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
95
96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
97
98         /* General Registers */
99         {E1000_CTRL, "CTRL"},
100         {E1000_STATUS, "STATUS"},
101         {E1000_CTRL_EXT, "CTRL_EXT"},
102
103         /* Interrupt Registers */
104         {E1000_ICR, "ICR"},
105
106         /* Rx Registers */
107         {E1000_RCTL, "RCTL"},
108         {E1000_RDLEN, "RDLEN"},
109         {E1000_RDH, "RDH"},
110         {E1000_RDT, "RDT"},
111         {E1000_RDTR, "RDTR"},
112         {E1000_RXDCTL(0), "RXDCTL"},
113         {E1000_ERT, "ERT"},
114         {E1000_RDBAL, "RDBAL"},
115         {E1000_RDBAH, "RDBAH"},
116         {E1000_RDFH, "RDFH"},
117         {E1000_RDFT, "RDFT"},
118         {E1000_RDFHS, "RDFHS"},
119         {E1000_RDFTS, "RDFTS"},
120         {E1000_RDFPC, "RDFPC"},
121
122         /* Tx Registers */
123         {E1000_TCTL, "TCTL"},
124         {E1000_TDBAL, "TDBAL"},
125         {E1000_TDBAH, "TDBAH"},
126         {E1000_TDLEN, "TDLEN"},
127         {E1000_TDH, "TDH"},
128         {E1000_TDT, "TDT"},
129         {E1000_TIDV, "TIDV"},
130         {E1000_TXDCTL(0), "TXDCTL"},
131         {E1000_TADV, "TADV"},
132         {E1000_TARC(0), "TARC"},
133         {E1000_TDFH, "TDFH"},
134         {E1000_TDFT, "TDFT"},
135         {E1000_TDFHS, "TDFHS"},
136         {E1000_TDFTS, "TDFTS"},
137         {E1000_TDFPC, "TDFPC"},
138
139         /* List Terminator */
140         {}
141 };
142
143 /*
144  * e1000_regdump - register printout routine
145  */
146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
147 {
148         int n = 0;
149         char rname[16];
150         u32 regs[8];
151
152         switch (reginfo->ofs) {
153         case E1000_RXDCTL(0):
154                 for (n = 0; n < 2; n++)
155                         regs[n] = __er32(hw, E1000_RXDCTL(n));
156                 break;
157         case E1000_TXDCTL(0):
158                 for (n = 0; n < 2; n++)
159                         regs[n] = __er32(hw, E1000_TXDCTL(n));
160                 break;
161         case E1000_TARC(0):
162                 for (n = 0; n < 2; n++)
163                         regs[n] = __er32(hw, E1000_TARC(n));
164                 break;
165         default:
166                 printk(KERN_INFO "%-15s %08x\n",
167                        reginfo->name, __er32(hw, reginfo->ofs));
168                 return;
169         }
170
171         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
172         printk(KERN_INFO "%-15s ", rname);
173         for (n = 0; n < 2; n++)
174                 printk(KERN_CONT "%08x ", regs[n]);
175         printk(KERN_CONT "\n");
176 }
177
178 /*
179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
180  */
181 static void e1000e_dump(struct e1000_adapter *adapter)
182 {
183         struct net_device *netdev = adapter->netdev;
184         struct e1000_hw *hw = &adapter->hw;
185         struct e1000_reg_info *reginfo;
186         struct e1000_ring *tx_ring = adapter->tx_ring;
187         struct e1000_tx_desc *tx_desc;
188         struct my_u0 {
189                 u64 a;
190                 u64 b;
191         } *u0;
192         struct e1000_buffer *buffer_info;
193         struct e1000_ring *rx_ring = adapter->rx_ring;
194         union e1000_rx_desc_packet_split *rx_desc_ps;
195         struct e1000_rx_desc *rx_desc;
196         struct my_u1 {
197                 u64 a;
198                 u64 b;
199                 u64 c;
200                 u64 d;
201         } *u1;
202         u32 staterr;
203         int i = 0;
204
205         if (!netif_msg_hw(adapter))
206                 return;
207
208         /* Print netdevice Info */
209         if (netdev) {
210                 dev_info(&adapter->pdev->dev, "Net device Info\n");
211                 printk(KERN_INFO "Device Name     state            "
212                        "trans_start      last_rx\n");
213                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
214                        netdev->name, netdev->state, netdev->trans_start,
215                        netdev->last_rx);
216         }
217
218         /* Print Registers */
219         dev_info(&adapter->pdev->dev, "Register Dump\n");
220         printk(KERN_INFO " Register Name   Value\n");
221         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
222              reginfo->name; reginfo++) {
223                 e1000_regdump(hw, reginfo);
224         }
225
226         /* Print Tx Ring Summary */
227         if (!netdev || !netif_running(netdev))
228                 goto exit;
229
230         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
231         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
232                " leng ntw timestamp\n");
233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
234         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
235                0, tx_ring->next_to_use, tx_ring->next_to_clean,
236                (unsigned long long)buffer_info->dma,
237                buffer_info->length,
238                buffer_info->next_to_watch,
239                (unsigned long long)buffer_info->time_stamp);
240
241         /* Print Tx Ring */
242         if (!netif_msg_tx_done(adapter))
243                 goto rx_ring_summary;
244
245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
246
247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
248          *
249          * Legacy Transmit Descriptor
250          *   +--------------------------------------------------------------+
251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
252          *   +--------------------------------------------------------------+
253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
254          *   +--------------------------------------------------------------+
255          *   63       48 47        36 35    32 31     24 23    16 15        0
256          *
257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
258          *   63      48 47    40 39       32 31             16 15    8 7      0
259          *   +----------------------------------------------------------------+
260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
261          *   +----------------------------------------------------------------+
262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
263          *   +----------------------------------------------------------------+
264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
265          *
266          * Extended Data Descriptor (DTYP=0x1)
267          *   +----------------------------------------------------------------+
268          * 0 |                     Buffer Address [63:0]                      |
269          *   +----------------------------------------------------------------+
270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
271          *   +----------------------------------------------------------------+
272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
273          */
274         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
275                " [bi->dma       ] leng  ntw timestamp        bi->skb "
276                "<-- Legacy format\n");
277         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
278                " [bi->dma       ] leng  ntw timestamp        bi->skb "
279                "<-- Ext Context format\n");
280         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
281                " [bi->dma       ] leng  ntw timestamp        bi->skb "
282                "<-- Ext Data format\n");
283         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
284                 tx_desc = E1000_TX_DESC(*tx_ring, i);
285                 buffer_info = &tx_ring->buffer_info[i];
286                 u0 = (struct my_u0 *)tx_desc;
287                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
288                        "%04X  %3X %016llX %p",
289                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
290                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
291                        (unsigned long long)le64_to_cpu(u0->a),
292                        (unsigned long long)le64_to_cpu(u0->b),
293                        (unsigned long long)buffer_info->dma,
294                        buffer_info->length, buffer_info->next_to_watch,
295                        (unsigned long long)buffer_info->time_stamp,
296                        buffer_info->skb);
297                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
298                         printk(KERN_CONT " NTC/U\n");
299                 else if (i == tx_ring->next_to_use)
300                         printk(KERN_CONT " NTU\n");
301                 else if (i == tx_ring->next_to_clean)
302                         printk(KERN_CONT " NTC\n");
303                 else
304                         printk(KERN_CONT "\n");
305
306                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
307                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
308                                        16, 1, phys_to_virt(buffer_info->dma),
309                                        buffer_info->length, true);
310         }
311
312         /* Print Rx Ring Summary */
313 rx_ring_summary:
314         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
315         printk(KERN_INFO "Queue [NTU] [NTC]\n");
316         printk(KERN_INFO " %5d %5X %5X\n", 0,
317                rx_ring->next_to_use, rx_ring->next_to_clean);
318
319         /* Print Rx Ring */
320         if (!netif_msg_rx_status(adapter))
321                 goto exit;
322
323         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
324         switch (adapter->rx_ps_pages) {
325         case 1:
326         case 2:
327         case 3:
328                 /* [Extended] Packet Split Receive Descriptor Format
329                  *
330                  *    +-----------------------------------------------------+
331                  *  0 |                Buffer Address 0 [63:0]              |
332                  *    +-----------------------------------------------------+
333                  *  8 |                Buffer Address 1 [63:0]              |
334                  *    +-----------------------------------------------------+
335                  * 16 |                Buffer Address 2 [63:0]              |
336                  *    +-----------------------------------------------------+
337                  * 24 |                Buffer Address 3 [63:0]              |
338                  *    +-----------------------------------------------------+
339                  */
340                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
341                        "[buffer 1 63:0 ] "
342                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
343                        "[bi->skb] <-- Ext Pkt Split format\n");
344                 /* [Extended] Receive Descriptor (Write-Back) Format
345                  *
346                  *   63       48 47    32 31     13 12    8 7    4 3        0
347                  *   +------------------------------------------------------+
348                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
349                  *   | Checksum | Ident  |         | Queue |      |  Type   |
350                  *   +------------------------------------------------------+
351                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
352                  *   +------------------------------------------------------+
353                  *   63       48 47    32 31            20 19               0
354                  */
355                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
356                        "[vl   l0 ee  es] "
357                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
358                        "[bi->skb] <-- Ext Rx Write-Back format\n");
359                 for (i = 0; i < rx_ring->count; i++) {
360                         buffer_info = &rx_ring->buffer_info[i];
361                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
362                         u1 = (struct my_u1 *)rx_desc_ps;
363                         staterr =
364                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
365                         if (staterr & E1000_RXD_STAT_DD) {
366                                 /* Descriptor Done */
367                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
368                                        "%016llX %016llX %016llX "
369                                        "---------------- %p", i,
370                                        (unsigned long long)le64_to_cpu(u1->a),
371                                        (unsigned long long)le64_to_cpu(u1->b),
372                                        (unsigned long long)le64_to_cpu(u1->c),
373                                        (unsigned long long)le64_to_cpu(u1->d),
374                                        buffer_info->skb);
375                         } else {
376                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
377                                        "%016llX %016llX %016llX %016llX %p", i,
378                                        (unsigned long long)le64_to_cpu(u1->a),
379                                        (unsigned long long)le64_to_cpu(u1->b),
380                                        (unsigned long long)le64_to_cpu(u1->c),
381                                        (unsigned long long)le64_to_cpu(u1->d),
382                                        (unsigned long long)buffer_info->dma,
383                                        buffer_info->skb);
384
385                                 if (netif_msg_pktdata(adapter))
386                                         print_hex_dump(KERN_INFO, "",
387                                                 DUMP_PREFIX_ADDRESS, 16, 1,
388                                                 phys_to_virt(buffer_info->dma),
389                                                 adapter->rx_ps_bsize0, true);
390                         }
391
392                         if (i == rx_ring->next_to_use)
393                                 printk(KERN_CONT " NTU\n");
394                         else if (i == rx_ring->next_to_clean)
395                                 printk(KERN_CONT " NTC\n");
396                         else
397                                 printk(KERN_CONT "\n");
398                 }
399                 break;
400         default:
401         case 0:
402                 /* Legacy Receive Descriptor Format
403                  *
404                  * +-----------------------------------------------------+
405                  * |                Buffer Address [63:0]                |
406                  * +-----------------------------------------------------+
407                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
408                  * +-----------------------------------------------------+
409                  * 63       48 47    40 39      32 31         16 15      0
410                  */
411                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
412                        "[vl er S cks ln] [bi->dma       ] [bi->skb] "
413                        "<-- Legacy format\n");
414                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
415                         rx_desc = E1000_RX_DESC(*rx_ring, i);
416                         buffer_info = &rx_ring->buffer_info[i];
417                         u0 = (struct my_u0 *)rx_desc;
418                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
419                                "%016llX %p", i,
420                                (unsigned long long)le64_to_cpu(u0->a),
421                                (unsigned long long)le64_to_cpu(u0->b),
422                                (unsigned long long)buffer_info->dma,
423                                buffer_info->skb);
424                         if (i == rx_ring->next_to_use)
425                                 printk(KERN_CONT " NTU\n");
426                         else if (i == rx_ring->next_to_clean)
427                                 printk(KERN_CONT " NTC\n");
428                         else
429                                 printk(KERN_CONT "\n");
430
431                         if (netif_msg_pktdata(adapter))
432                                 print_hex_dump(KERN_INFO, "",
433                                                DUMP_PREFIX_ADDRESS,
434                                                16, 1,
435                                                phys_to_virt(buffer_info->dma),
436                                                adapter->rx_buffer_len, true);
437                 }
438         }
439
440 exit:
441         return;
442 }
443
444 /**
445  * e1000_desc_unused - calculate if we have unused descriptors
446  **/
447 static int e1000_desc_unused(struct e1000_ring *ring)
448 {
449         if (ring->next_to_clean > ring->next_to_use)
450                 return ring->next_to_clean - ring->next_to_use - 1;
451
452         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
453 }
454
455 /**
456  * e1000_receive_skb - helper function to handle Rx indications
457  * @adapter: board private structure
458  * @status: descriptor status field as written by hardware
459  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
460  * @skb: pointer to sk_buff to be indicated to stack
461  **/
462 static void e1000_receive_skb(struct e1000_adapter *adapter,
463                               struct net_device *netdev, struct sk_buff *skb,
464                               u8 status, __le16 vlan)
465 {
466         u16 tag = le16_to_cpu(vlan);
467         skb->protocol = eth_type_trans(skb, netdev);
468
469         if (status & E1000_RXD_STAT_VP)
470                 __vlan_hwaccel_put_tag(skb, tag);
471
472         napi_gro_receive(&adapter->napi, skb);
473 }
474
475 /**
476  * e1000_rx_checksum - Receive Checksum Offload
477  * @adapter:     board private structure
478  * @status_err:  receive descriptor status and error fields
479  * @csum:       receive descriptor csum field
480  * @sk_buff:     socket buffer with received data
481  **/
482 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
483                               u32 csum, struct sk_buff *skb)
484 {
485         u16 status = (u16)status_err;
486         u8 errors = (u8)(status_err >> 24);
487
488         skb_checksum_none_assert(skb);
489
490         /* Ignore Checksum bit is set */
491         if (status & E1000_RXD_STAT_IXSM)
492                 return;
493         /* TCP/UDP checksum error bit is set */
494         if (errors & E1000_RXD_ERR_TCPE) {
495                 /* let the stack verify checksum errors */
496                 adapter->hw_csum_err++;
497                 return;
498         }
499
500         /* TCP/UDP Checksum has not been calculated */
501         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
502                 return;
503
504         /* It must be a TCP or UDP packet with a valid checksum */
505         if (status & E1000_RXD_STAT_TCPCS) {
506                 /* TCP checksum is good */
507                 skb->ip_summed = CHECKSUM_UNNECESSARY;
508         } else {
509                 /*
510                  * IP fragment with UDP payload
511                  * Hardware complements the payload checksum, so we undo it
512                  * and then put the value in host order for further stack use.
513                  */
514                 __sum16 sum = (__force __sum16)htons(csum);
515                 skb->csum = csum_unfold(~sum);
516                 skb->ip_summed = CHECKSUM_COMPLETE;
517         }
518         adapter->hw_csum_good++;
519 }
520
521 /**
522  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
523  * @adapter: address of board private structure
524  **/
525 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
526                                    int cleaned_count)
527 {
528         struct net_device *netdev = adapter->netdev;
529         struct pci_dev *pdev = adapter->pdev;
530         struct e1000_ring *rx_ring = adapter->rx_ring;
531         struct e1000_rx_desc *rx_desc;
532         struct e1000_buffer *buffer_info;
533         struct sk_buff *skb;
534         unsigned int i;
535         unsigned int bufsz = adapter->rx_buffer_len;
536
537         i = rx_ring->next_to_use;
538         buffer_info = &rx_ring->buffer_info[i];
539
540         while (cleaned_count--) {
541                 skb = buffer_info->skb;
542                 if (skb) {
543                         skb_trim(skb, 0);
544                         goto map_skb;
545                 }
546
547                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
548                 if (!skb) {
549                         /* Better luck next round */
550                         adapter->alloc_rx_buff_failed++;
551                         break;
552                 }
553
554                 buffer_info->skb = skb;
555 map_skb:
556                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
557                                                   adapter->rx_buffer_len,
558                                                   DMA_FROM_DEVICE);
559                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
560                         dev_err(&pdev->dev, "Rx DMA map failed\n");
561                         adapter->rx_dma_failed++;
562                         break;
563                 }
564
565                 rx_desc = E1000_RX_DESC(*rx_ring, i);
566                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
567
568                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
569                         /*
570                          * Force memory writes to complete before letting h/w
571                          * know there are new descriptors to fetch.  (Only
572                          * applicable for weak-ordered memory model archs,
573                          * such as IA-64).
574                          */
575                         wmb();
576                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
577                 }
578                 i++;
579                 if (i == rx_ring->count)
580                         i = 0;
581                 buffer_info = &rx_ring->buffer_info[i];
582         }
583
584         rx_ring->next_to_use = i;
585 }
586
587 /**
588  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
589  * @adapter: address of board private structure
590  **/
591 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
592                                       int cleaned_count)
593 {
594         struct net_device *netdev = adapter->netdev;
595         struct pci_dev *pdev = adapter->pdev;
596         union e1000_rx_desc_packet_split *rx_desc;
597         struct e1000_ring *rx_ring = adapter->rx_ring;
598         struct e1000_buffer *buffer_info;
599         struct e1000_ps_page *ps_page;
600         struct sk_buff *skb;
601         unsigned int i, j;
602
603         i = rx_ring->next_to_use;
604         buffer_info = &rx_ring->buffer_info[i];
605
606         while (cleaned_count--) {
607                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
608
609                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
610                         ps_page = &buffer_info->ps_pages[j];
611                         if (j >= adapter->rx_ps_pages) {
612                                 /* all unused desc entries get hw null ptr */
613                                 rx_desc->read.buffer_addr[j + 1] =
614                                     ~cpu_to_le64(0);
615                                 continue;
616                         }
617                         if (!ps_page->page) {
618                                 ps_page->page = alloc_page(GFP_ATOMIC);
619                                 if (!ps_page->page) {
620                                         adapter->alloc_rx_buff_failed++;
621                                         goto no_buffers;
622                                 }
623                                 ps_page->dma = dma_map_page(&pdev->dev,
624                                                             ps_page->page,
625                                                             0, PAGE_SIZE,
626                                                             DMA_FROM_DEVICE);
627                                 if (dma_mapping_error(&pdev->dev,
628                                                       ps_page->dma)) {
629                                         dev_err(&adapter->pdev->dev,
630                                                 "Rx DMA page map failed\n");
631                                         adapter->rx_dma_failed++;
632                                         goto no_buffers;
633                                 }
634                         }
635                         /*
636                          * Refresh the desc even if buffer_addrs
637                          * didn't change because each write-back
638                          * erases this info.
639                          */
640                         rx_desc->read.buffer_addr[j + 1] =
641                             cpu_to_le64(ps_page->dma);
642                 }
643
644                 skb = netdev_alloc_skb_ip_align(netdev,
645                                                 adapter->rx_ps_bsize0);
646
647                 if (!skb) {
648                         adapter->alloc_rx_buff_failed++;
649                         break;
650                 }
651
652                 buffer_info->skb = skb;
653                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
654                                                   adapter->rx_ps_bsize0,
655                                                   DMA_FROM_DEVICE);
656                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
657                         dev_err(&pdev->dev, "Rx DMA map failed\n");
658                         adapter->rx_dma_failed++;
659                         /* cleanup skb */
660                         dev_kfree_skb_any(skb);
661                         buffer_info->skb = NULL;
662                         break;
663                 }
664
665                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
666
667                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
668                         /*
669                          * Force memory writes to complete before letting h/w
670                          * know there are new descriptors to fetch.  (Only
671                          * applicable for weak-ordered memory model archs,
672                          * such as IA-64).
673                          */
674                         wmb();
675                         writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
676                 }
677
678                 i++;
679                 if (i == rx_ring->count)
680                         i = 0;
681                 buffer_info = &rx_ring->buffer_info[i];
682         }
683
684 no_buffers:
685         rx_ring->next_to_use = i;
686 }
687
688 /**
689  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
690  * @adapter: address of board private structure
691  * @cleaned_count: number of buffers to allocate this pass
692  **/
693
694 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
695                                          int cleaned_count)
696 {
697         struct net_device *netdev = adapter->netdev;
698         struct pci_dev *pdev = adapter->pdev;
699         struct e1000_rx_desc *rx_desc;
700         struct e1000_ring *rx_ring = adapter->rx_ring;
701         struct e1000_buffer *buffer_info;
702         struct sk_buff *skb;
703         unsigned int i;
704         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
705
706         i = rx_ring->next_to_use;
707         buffer_info = &rx_ring->buffer_info[i];
708
709         while (cleaned_count--) {
710                 skb = buffer_info->skb;
711                 if (skb) {
712                         skb_trim(skb, 0);
713                         goto check_page;
714                 }
715
716                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
717                 if (unlikely(!skb)) {
718                         /* Better luck next round */
719                         adapter->alloc_rx_buff_failed++;
720                         break;
721                 }
722
723                 buffer_info->skb = skb;
724 check_page:
725                 /* allocate a new page if necessary */
726                 if (!buffer_info->page) {
727                         buffer_info->page = alloc_page(GFP_ATOMIC);
728                         if (unlikely(!buffer_info->page)) {
729                                 adapter->alloc_rx_buff_failed++;
730                                 break;
731                         }
732                 }
733
734                 if (!buffer_info->dma)
735                         buffer_info->dma = dma_map_page(&pdev->dev,
736                                                         buffer_info->page, 0,
737                                                         PAGE_SIZE,
738                                                         DMA_FROM_DEVICE);
739
740                 rx_desc = E1000_RX_DESC(*rx_ring, i);
741                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
742
743                 if (unlikely(++i == rx_ring->count))
744                         i = 0;
745                 buffer_info = &rx_ring->buffer_info[i];
746         }
747
748         if (likely(rx_ring->next_to_use != i)) {
749                 rx_ring->next_to_use = i;
750                 if (unlikely(i-- == 0))
751                         i = (rx_ring->count - 1);
752
753                 /* Force memory writes to complete before letting h/w
754                  * know there are new descriptors to fetch.  (Only
755                  * applicable for weak-ordered memory model archs,
756                  * such as IA-64). */
757                 wmb();
758                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
759         }
760 }
761
762 /**
763  * e1000_clean_rx_irq - Send received data up the network stack; legacy
764  * @adapter: board private structure
765  *
766  * the return value indicates whether actual cleaning was done, there
767  * is no guarantee that everything was cleaned
768  **/
769 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
770                                int *work_done, int work_to_do)
771 {
772         struct net_device *netdev = adapter->netdev;
773         struct pci_dev *pdev = adapter->pdev;
774         struct e1000_hw *hw = &adapter->hw;
775         struct e1000_ring *rx_ring = adapter->rx_ring;
776         struct e1000_rx_desc *rx_desc, *next_rxd;
777         struct e1000_buffer *buffer_info, *next_buffer;
778         u32 length;
779         unsigned int i;
780         int cleaned_count = 0;
781         bool cleaned = 0;
782         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
783
784         i = rx_ring->next_to_clean;
785         rx_desc = E1000_RX_DESC(*rx_ring, i);
786         buffer_info = &rx_ring->buffer_info[i];
787
788         while (rx_desc->status & E1000_RXD_STAT_DD) {
789                 struct sk_buff *skb;
790                 u8 status;
791
792                 if (*work_done >= work_to_do)
793                         break;
794                 (*work_done)++;
795                 rmb();  /* read descriptor and rx_buffer_info after status DD */
796
797                 status = rx_desc->status;
798                 skb = buffer_info->skb;
799                 buffer_info->skb = NULL;
800
801                 prefetch(skb->data - NET_IP_ALIGN);
802
803                 i++;
804                 if (i == rx_ring->count)
805                         i = 0;
806                 next_rxd = E1000_RX_DESC(*rx_ring, i);
807                 prefetch(next_rxd);
808
809                 next_buffer = &rx_ring->buffer_info[i];
810
811                 cleaned = 1;
812                 cleaned_count++;
813                 dma_unmap_single(&pdev->dev,
814                                  buffer_info->dma,
815                                  adapter->rx_buffer_len,
816                                  DMA_FROM_DEVICE);
817                 buffer_info->dma = 0;
818
819                 length = le16_to_cpu(rx_desc->length);
820
821                 /*
822                  * !EOP means multiple descriptors were used to store a single
823                  * packet, if that's the case we need to toss it.  In fact, we
824                  * need to toss every packet with the EOP bit clear and the
825                  * next frame that _does_ have the EOP bit set, as it is by
826                  * definition only a frame fragment
827                  */
828                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
829                         adapter->flags2 |= FLAG2_IS_DISCARDING;
830
831                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
832                         /* All receives must fit into a single buffer */
833                         e_dbg("Receive packet consumed multiple buffers\n");
834                         /* recycle */
835                         buffer_info->skb = skb;
836                         if (status & E1000_RXD_STAT_EOP)
837                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
838                         goto next_desc;
839                 }
840
841                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
842                         /* recycle */
843                         buffer_info->skb = skb;
844                         goto next_desc;
845                 }
846
847                 /* adjust length to remove Ethernet CRC */
848                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
849                         length -= 4;
850
851                 total_rx_bytes += length;
852                 total_rx_packets++;
853
854                 /*
855                  * code added for copybreak, this should improve
856                  * performance for small packets with large amounts
857                  * of reassembly being done in the stack
858                  */
859                 if (length < copybreak) {
860                         struct sk_buff *new_skb =
861                             netdev_alloc_skb_ip_align(netdev, length);
862                         if (new_skb) {
863                                 skb_copy_to_linear_data_offset(new_skb,
864                                                                -NET_IP_ALIGN,
865                                                                (skb->data -
866                                                                 NET_IP_ALIGN),
867                                                                (length +
868                                                                 NET_IP_ALIGN));
869                                 /* save the skb in buffer_info as good */
870                                 buffer_info->skb = skb;
871                                 skb = new_skb;
872                         }
873                         /* else just continue with the old one */
874                 }
875                 /* end copybreak code */
876                 skb_put(skb, length);
877
878                 /* Receive Checksum Offload */
879                 e1000_rx_checksum(adapter,
880                                   (u32)(status) |
881                                   ((u32)(rx_desc->errors) << 24),
882                                   le16_to_cpu(rx_desc->csum), skb);
883
884                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
885
886 next_desc:
887                 rx_desc->status = 0;
888
889                 /* return some buffers to hardware, one at a time is too slow */
890                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
891                         adapter->alloc_rx_buf(adapter, cleaned_count);
892                         cleaned_count = 0;
893                 }
894
895                 /* use prefetched values */
896                 rx_desc = next_rxd;
897                 buffer_info = next_buffer;
898         }
899         rx_ring->next_to_clean = i;
900
901         cleaned_count = e1000_desc_unused(rx_ring);
902         if (cleaned_count)
903                 adapter->alloc_rx_buf(adapter, cleaned_count);
904
905         adapter->total_rx_bytes += total_rx_bytes;
906         adapter->total_rx_packets += total_rx_packets;
907         return cleaned;
908 }
909
910 static void e1000_put_txbuf(struct e1000_adapter *adapter,
911                              struct e1000_buffer *buffer_info)
912 {
913         if (buffer_info->dma) {
914                 if (buffer_info->mapped_as_page)
915                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
916                                        buffer_info->length, DMA_TO_DEVICE);
917                 else
918                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
919                                          buffer_info->length, DMA_TO_DEVICE);
920                 buffer_info->dma = 0;
921         }
922         if (buffer_info->skb) {
923                 dev_kfree_skb_any(buffer_info->skb);
924                 buffer_info->skb = NULL;
925         }
926         buffer_info->time_stamp = 0;
927 }
928
929 static void e1000_print_hw_hang(struct work_struct *work)
930 {
931         struct e1000_adapter *adapter = container_of(work,
932                                                      struct e1000_adapter,
933                                                      print_hang_task);
934         struct e1000_ring *tx_ring = adapter->tx_ring;
935         unsigned int i = tx_ring->next_to_clean;
936         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
937         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
938         struct e1000_hw *hw = &adapter->hw;
939         u16 phy_status, phy_1000t_status, phy_ext_status;
940         u16 pci_status;
941
942         if (test_bit(__E1000_DOWN, &adapter->state))
943                 return;
944
945         e1e_rphy(hw, PHY_STATUS, &phy_status);
946         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
947         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
948
949         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
950
951         /* detected Hardware unit hang */
952         e_err("Detected Hardware Unit Hang:\n"
953               "  TDH                  <%x>\n"
954               "  TDT                  <%x>\n"
955               "  next_to_use          <%x>\n"
956               "  next_to_clean        <%x>\n"
957               "buffer_info[next_to_clean]:\n"
958               "  time_stamp           <%lx>\n"
959               "  next_to_watch        <%x>\n"
960               "  jiffies              <%lx>\n"
961               "  next_to_watch.status <%x>\n"
962               "MAC Status             <%x>\n"
963               "PHY Status             <%x>\n"
964               "PHY 1000BASE-T Status  <%x>\n"
965               "PHY Extended Status    <%x>\n"
966               "PCI Status             <%x>\n",
967               readl(adapter->hw.hw_addr + tx_ring->head),
968               readl(adapter->hw.hw_addr + tx_ring->tail),
969               tx_ring->next_to_use,
970               tx_ring->next_to_clean,
971               tx_ring->buffer_info[eop].time_stamp,
972               eop,
973               jiffies,
974               eop_desc->upper.fields.status,
975               er32(STATUS),
976               phy_status,
977               phy_1000t_status,
978               phy_ext_status,
979               pci_status);
980 }
981
982 /**
983  * e1000_clean_tx_irq - Reclaim resources after transmit completes
984  * @adapter: board private structure
985  *
986  * the return value indicates whether actual cleaning was done, there
987  * is no guarantee that everything was cleaned
988  **/
989 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
990 {
991         struct net_device *netdev = adapter->netdev;
992         struct e1000_hw *hw = &adapter->hw;
993         struct e1000_ring *tx_ring = adapter->tx_ring;
994         struct e1000_tx_desc *tx_desc, *eop_desc;
995         struct e1000_buffer *buffer_info;
996         unsigned int i, eop;
997         unsigned int count = 0;
998         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
999
1000         i = tx_ring->next_to_clean;
1001         eop = tx_ring->buffer_info[i].next_to_watch;
1002         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1003
1004         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1005                (count < tx_ring->count)) {
1006                 bool cleaned = false;
1007                 rmb(); /* read buffer_info after eop_desc */
1008                 for (; !cleaned; count++) {
1009                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1010                         buffer_info = &tx_ring->buffer_info[i];
1011                         cleaned = (i == eop);
1012
1013                         if (cleaned) {
1014                                 total_tx_packets += buffer_info->segs;
1015                                 total_tx_bytes += buffer_info->bytecount;
1016                         }
1017
1018                         e1000_put_txbuf(adapter, buffer_info);
1019                         tx_desc->upper.data = 0;
1020
1021                         i++;
1022                         if (i == tx_ring->count)
1023                                 i = 0;
1024                 }
1025
1026                 if (i == tx_ring->next_to_use)
1027                         break;
1028                 eop = tx_ring->buffer_info[i].next_to_watch;
1029                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1030         }
1031
1032         tx_ring->next_to_clean = i;
1033
1034 #define TX_WAKE_THRESHOLD 32
1035         if (count && netif_carrier_ok(netdev) &&
1036             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1037                 /* Make sure that anybody stopping the queue after this
1038                  * sees the new next_to_clean.
1039                  */
1040                 smp_mb();
1041
1042                 if (netif_queue_stopped(netdev) &&
1043                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1044                         netif_wake_queue(netdev);
1045                         ++adapter->restart_queue;
1046                 }
1047         }
1048
1049         if (adapter->detect_tx_hung) {
1050                 /*
1051                  * Detect a transmit hang in hardware, this serializes the
1052                  * check with the clearing of time_stamp and movement of i
1053                  */
1054                 adapter->detect_tx_hung = 0;
1055                 if (tx_ring->buffer_info[i].time_stamp &&
1056                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1057                                + (adapter->tx_timeout_factor * HZ)) &&
1058                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1059                         schedule_work(&adapter->print_hang_task);
1060                         netif_stop_queue(netdev);
1061                 }
1062         }
1063         adapter->total_tx_bytes += total_tx_bytes;
1064         adapter->total_tx_packets += total_tx_packets;
1065         return count < tx_ring->count;
1066 }
1067
1068 /**
1069  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1070  * @adapter: board private structure
1071  *
1072  * the return value indicates whether actual cleaning was done, there
1073  * is no guarantee that everything was cleaned
1074  **/
1075 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1076                                   int *work_done, int work_to_do)
1077 {
1078         struct e1000_hw *hw = &adapter->hw;
1079         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1080         struct net_device *netdev = adapter->netdev;
1081         struct pci_dev *pdev = adapter->pdev;
1082         struct e1000_ring *rx_ring = adapter->rx_ring;
1083         struct e1000_buffer *buffer_info, *next_buffer;
1084         struct e1000_ps_page *ps_page;
1085         struct sk_buff *skb;
1086         unsigned int i, j;
1087         u32 length, staterr;
1088         int cleaned_count = 0;
1089         bool cleaned = 0;
1090         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1091
1092         i = rx_ring->next_to_clean;
1093         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1094         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1095         buffer_info = &rx_ring->buffer_info[i];
1096
1097         while (staterr & E1000_RXD_STAT_DD) {
1098                 if (*work_done >= work_to_do)
1099                         break;
1100                 (*work_done)++;
1101                 skb = buffer_info->skb;
1102                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1103
1104                 /* in the packet split case this is header only */
1105                 prefetch(skb->data - NET_IP_ALIGN);
1106
1107                 i++;
1108                 if (i == rx_ring->count)
1109                         i = 0;
1110                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1111                 prefetch(next_rxd);
1112
1113                 next_buffer = &rx_ring->buffer_info[i];
1114
1115                 cleaned = 1;
1116                 cleaned_count++;
1117                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1118                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1119                 buffer_info->dma = 0;
1120
1121                 /* see !EOP comment in other Rx routine */
1122                 if (!(staterr & E1000_RXD_STAT_EOP))
1123                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1124
1125                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1126                         e_dbg("Packet Split buffers didn't pick up the full "
1127                               "packet\n");
1128                         dev_kfree_skb_irq(skb);
1129                         if (staterr & E1000_RXD_STAT_EOP)
1130                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1131                         goto next_desc;
1132                 }
1133
1134                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1135                         dev_kfree_skb_irq(skb);
1136                         goto next_desc;
1137                 }
1138
1139                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1140
1141                 if (!length) {
1142                         e_dbg("Last part of the packet spanning multiple "
1143                               "descriptors\n");
1144                         dev_kfree_skb_irq(skb);
1145                         goto next_desc;
1146                 }
1147
1148                 /* Good Receive */
1149                 skb_put(skb, length);
1150
1151                 {
1152                 /*
1153                  * this looks ugly, but it seems compiler issues make it
1154                  * more efficient than reusing j
1155                  */
1156                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1157
1158                 /*
1159                  * page alloc/put takes too long and effects small packet
1160                  * throughput, so unsplit small packets and save the alloc/put
1161                  * only valid in softirq (napi) context to call kmap_*
1162                  */
1163                 if (l1 && (l1 <= copybreak) &&
1164                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1165                         u8 *vaddr;
1166
1167                         ps_page = &buffer_info->ps_pages[0];
1168
1169                         /*
1170                          * there is no documentation about how to call
1171                          * kmap_atomic, so we can't hold the mapping
1172                          * very long
1173                          */
1174                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1175                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1176                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1177                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1178                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1179                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1180                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1181
1182                         /* remove the CRC */
1183                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1184                                 l1 -= 4;
1185
1186                         skb_put(skb, l1);
1187                         goto copydone;
1188                 } /* if */
1189                 }
1190
1191                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1192                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1193                         if (!length)
1194                                 break;
1195
1196                         ps_page = &buffer_info->ps_pages[j];
1197                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1198                                        DMA_FROM_DEVICE);
1199                         ps_page->dma = 0;
1200                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1201                         ps_page->page = NULL;
1202                         skb->len += length;
1203                         skb->data_len += length;
1204                         skb->truesize += length;
1205                 }
1206
1207                 /* strip the ethernet crc, problem is we're using pages now so
1208                  * this whole operation can get a little cpu intensive
1209                  */
1210                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1211                         pskb_trim(skb, skb->len - 4);
1212
1213 copydone:
1214                 total_rx_bytes += skb->len;
1215                 total_rx_packets++;
1216
1217                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1218                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1219
1220                 if (rx_desc->wb.upper.header_status &
1221                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1222                         adapter->rx_hdr_split++;
1223
1224                 e1000_receive_skb(adapter, netdev, skb,
1225                                   staterr, rx_desc->wb.middle.vlan);
1226
1227 next_desc:
1228                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1229                 buffer_info->skb = NULL;
1230
1231                 /* return some buffers to hardware, one at a time is too slow */
1232                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1233                         adapter->alloc_rx_buf(adapter, cleaned_count);
1234                         cleaned_count = 0;
1235                 }
1236
1237                 /* use prefetched values */
1238                 rx_desc = next_rxd;
1239                 buffer_info = next_buffer;
1240
1241                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1242         }
1243         rx_ring->next_to_clean = i;
1244
1245         cleaned_count = e1000_desc_unused(rx_ring);
1246         if (cleaned_count)
1247                 adapter->alloc_rx_buf(adapter, cleaned_count);
1248
1249         adapter->total_rx_bytes += total_rx_bytes;
1250         adapter->total_rx_packets += total_rx_packets;
1251         return cleaned;
1252 }
1253
1254 /**
1255  * e1000_consume_page - helper function
1256  **/
1257 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1258                                u16 length)
1259 {
1260         bi->page = NULL;
1261         skb->len += length;
1262         skb->data_len += length;
1263         skb->truesize += length;
1264 }
1265
1266 /**
1267  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1268  * @adapter: board private structure
1269  *
1270  * the return value indicates whether actual cleaning was done, there
1271  * is no guarantee that everything was cleaned
1272  **/
1273
1274 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1275                                      int *work_done, int work_to_do)
1276 {
1277         struct net_device *netdev = adapter->netdev;
1278         struct pci_dev *pdev = adapter->pdev;
1279         struct e1000_ring *rx_ring = adapter->rx_ring;
1280         struct e1000_rx_desc *rx_desc, *next_rxd;
1281         struct e1000_buffer *buffer_info, *next_buffer;
1282         u32 length;
1283         unsigned int i;
1284         int cleaned_count = 0;
1285         bool cleaned = false;
1286         unsigned int total_rx_bytes=0, total_rx_packets=0;
1287
1288         i = rx_ring->next_to_clean;
1289         rx_desc = E1000_RX_DESC(*rx_ring, i);
1290         buffer_info = &rx_ring->buffer_info[i];
1291
1292         while (rx_desc->status & E1000_RXD_STAT_DD) {
1293                 struct sk_buff *skb;
1294                 u8 status;
1295
1296                 if (*work_done >= work_to_do)
1297                         break;
1298                 (*work_done)++;
1299                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1300
1301                 status = rx_desc->status;
1302                 skb = buffer_info->skb;
1303                 buffer_info->skb = NULL;
1304
1305                 ++i;
1306                 if (i == rx_ring->count)
1307                         i = 0;
1308                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1309                 prefetch(next_rxd);
1310
1311                 next_buffer = &rx_ring->buffer_info[i];
1312
1313                 cleaned = true;
1314                 cleaned_count++;
1315                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1316                                DMA_FROM_DEVICE);
1317                 buffer_info->dma = 0;
1318
1319                 length = le16_to_cpu(rx_desc->length);
1320
1321                 /* errors is only valid for DD + EOP descriptors */
1322                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1323                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1324                                 /* recycle both page and skb */
1325                                 buffer_info->skb = skb;
1326                                 /* an error means any chain goes out the window
1327                                  * too */
1328                                 if (rx_ring->rx_skb_top)
1329                                         dev_kfree_skb_irq(rx_ring->rx_skb_top);
1330                                 rx_ring->rx_skb_top = NULL;
1331                                 goto next_desc;
1332                 }
1333
1334 #define rxtop (rx_ring->rx_skb_top)
1335                 if (!(status & E1000_RXD_STAT_EOP)) {
1336                         /* this descriptor is only the beginning (or middle) */
1337                         if (!rxtop) {
1338                                 /* this is the beginning of a chain */
1339                                 rxtop = skb;
1340                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1341                                                    0, length);
1342                         } else {
1343                                 /* this is the middle of a chain */
1344                                 skb_fill_page_desc(rxtop,
1345                                     skb_shinfo(rxtop)->nr_frags,
1346                                     buffer_info->page, 0, length);
1347                                 /* re-use the skb, only consumed the page */
1348                                 buffer_info->skb = skb;
1349                         }
1350                         e1000_consume_page(buffer_info, rxtop, length);
1351                         goto next_desc;
1352                 } else {
1353                         if (rxtop) {
1354                                 /* end of the chain */
1355                                 skb_fill_page_desc(rxtop,
1356                                     skb_shinfo(rxtop)->nr_frags,
1357                                     buffer_info->page, 0, length);
1358                                 /* re-use the current skb, we only consumed the
1359                                  * page */
1360                                 buffer_info->skb = skb;
1361                                 skb = rxtop;
1362                                 rxtop = NULL;
1363                                 e1000_consume_page(buffer_info, skb, length);
1364                         } else {
1365                                 /* no chain, got EOP, this buf is the packet
1366                                  * copybreak to save the put_page/alloc_page */
1367                                 if (length <= copybreak &&
1368                                     skb_tailroom(skb) >= length) {
1369                                         u8 *vaddr;
1370                                         vaddr = kmap_atomic(buffer_info->page,
1371                                                            KM_SKB_DATA_SOFTIRQ);
1372                                         memcpy(skb_tail_pointer(skb), vaddr,
1373                                                length);
1374                                         kunmap_atomic(vaddr,
1375                                                       KM_SKB_DATA_SOFTIRQ);
1376                                         /* re-use the page, so don't erase
1377                                          * buffer_info->page */
1378                                         skb_put(skb, length);
1379                                 } else {
1380                                         skb_fill_page_desc(skb, 0,
1381                                                            buffer_info->page, 0,
1382                                                            length);
1383                                         e1000_consume_page(buffer_info, skb,
1384                                                            length);
1385                                 }
1386                         }
1387                 }
1388
1389                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1390                 e1000_rx_checksum(adapter,
1391                                   (u32)(status) |
1392                                   ((u32)(rx_desc->errors) << 24),
1393                                   le16_to_cpu(rx_desc->csum), skb);
1394
1395                 /* probably a little skewed due to removing CRC */
1396                 total_rx_bytes += skb->len;
1397                 total_rx_packets++;
1398
1399                 /* eth type trans needs skb->data to point to something */
1400                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1401                         e_err("pskb_may_pull failed.\n");
1402                         dev_kfree_skb_irq(skb);
1403                         goto next_desc;
1404                 }
1405
1406                 e1000_receive_skb(adapter, netdev, skb, status,
1407                                   rx_desc->special);
1408
1409 next_desc:
1410                 rx_desc->status = 0;
1411
1412                 /* return some buffers to hardware, one at a time is too slow */
1413                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1414                         adapter->alloc_rx_buf(adapter, cleaned_count);
1415                         cleaned_count = 0;
1416                 }
1417
1418                 /* use prefetched values */
1419                 rx_desc = next_rxd;
1420                 buffer_info = next_buffer;
1421         }
1422         rx_ring->next_to_clean = i;
1423
1424         cleaned_count = e1000_desc_unused(rx_ring);
1425         if (cleaned_count)
1426                 adapter->alloc_rx_buf(adapter, cleaned_count);
1427
1428         adapter->total_rx_bytes += total_rx_bytes;
1429         adapter->total_rx_packets += total_rx_packets;
1430         return cleaned;
1431 }
1432
1433 /**
1434  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1435  * @adapter: board private structure
1436  **/
1437 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1438 {
1439         struct e1000_ring *rx_ring = adapter->rx_ring;
1440         struct e1000_buffer *buffer_info;
1441         struct e1000_ps_page *ps_page;
1442         struct pci_dev *pdev = adapter->pdev;
1443         unsigned int i, j;
1444
1445         /* Free all the Rx ring sk_buffs */
1446         for (i = 0; i < rx_ring->count; i++) {
1447                 buffer_info = &rx_ring->buffer_info[i];
1448                 if (buffer_info->dma) {
1449                         if (adapter->clean_rx == e1000_clean_rx_irq)
1450                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1451                                                  adapter->rx_buffer_len,
1452                                                  DMA_FROM_DEVICE);
1453                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1454                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1455                                                PAGE_SIZE,
1456                                                DMA_FROM_DEVICE);
1457                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1458                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1459                                                  adapter->rx_ps_bsize0,
1460                                                  DMA_FROM_DEVICE);
1461                         buffer_info->dma = 0;
1462                 }
1463
1464                 if (buffer_info->page) {
1465                         put_page(buffer_info->page);
1466                         buffer_info->page = NULL;
1467                 }
1468
1469                 if (buffer_info->skb) {
1470                         dev_kfree_skb(buffer_info->skb);
1471                         buffer_info->skb = NULL;
1472                 }
1473
1474                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1475                         ps_page = &buffer_info->ps_pages[j];
1476                         if (!ps_page->page)
1477                                 break;
1478                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1479                                        DMA_FROM_DEVICE);
1480                         ps_page->dma = 0;
1481                         put_page(ps_page->page);
1482                         ps_page->page = NULL;
1483                 }
1484         }
1485
1486         /* there also may be some cached data from a chained receive */
1487         if (rx_ring->rx_skb_top) {
1488                 dev_kfree_skb(rx_ring->rx_skb_top);
1489                 rx_ring->rx_skb_top = NULL;
1490         }
1491
1492         /* Zero out the descriptor ring */
1493         memset(rx_ring->desc, 0, rx_ring->size);
1494
1495         rx_ring->next_to_clean = 0;
1496         rx_ring->next_to_use = 0;
1497         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1498
1499         writel(0, adapter->hw.hw_addr + rx_ring->head);
1500         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1501 }
1502
1503 static void e1000e_downshift_workaround(struct work_struct *work)
1504 {
1505         struct e1000_adapter *adapter = container_of(work,
1506                                         struct e1000_adapter, downshift_task);
1507
1508         if (test_bit(__E1000_DOWN, &adapter->state))
1509                 return;
1510
1511         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1512 }
1513
1514 /**
1515  * e1000_intr_msi - Interrupt Handler
1516  * @irq: interrupt number
1517  * @data: pointer to a network interface device structure
1518  **/
1519 static irqreturn_t e1000_intr_msi(int irq, void *data)
1520 {
1521         struct net_device *netdev = data;
1522         struct e1000_adapter *adapter = netdev_priv(netdev);
1523         struct e1000_hw *hw = &adapter->hw;
1524         u32 icr = er32(ICR);
1525
1526         /*
1527          * read ICR disables interrupts using IAM
1528          */
1529
1530         if (icr & E1000_ICR_LSC) {
1531                 hw->mac.get_link_status = 1;
1532                 /*
1533                  * ICH8 workaround-- Call gig speed drop workaround on cable
1534                  * disconnect (LSC) before accessing any PHY registers
1535                  */
1536                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1537                     (!(er32(STATUS) & E1000_STATUS_LU)))
1538                         schedule_work(&adapter->downshift_task);
1539
1540                 /*
1541                  * 80003ES2LAN workaround-- For packet buffer work-around on
1542                  * link down event; disable receives here in the ISR and reset
1543                  * adapter in watchdog
1544                  */
1545                 if (netif_carrier_ok(netdev) &&
1546                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1547                         /* disable receives */
1548                         u32 rctl = er32(RCTL);
1549                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1550                         adapter->flags |= FLAG_RX_RESTART_NOW;
1551                 }
1552                 /* guard against interrupt when we're going down */
1553                 if (!test_bit(__E1000_DOWN, &adapter->state))
1554                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1555         }
1556
1557         if (napi_schedule_prep(&adapter->napi)) {
1558                 adapter->total_tx_bytes = 0;
1559                 adapter->total_tx_packets = 0;
1560                 adapter->total_rx_bytes = 0;
1561                 adapter->total_rx_packets = 0;
1562                 __napi_schedule(&adapter->napi);
1563         }
1564
1565         return IRQ_HANDLED;
1566 }
1567
1568 /**
1569  * e1000_intr - Interrupt Handler
1570  * @irq: interrupt number
1571  * @data: pointer to a network interface device structure
1572  **/
1573 static irqreturn_t e1000_intr(int irq, void *data)
1574 {
1575         struct net_device *netdev = data;
1576         struct e1000_adapter *adapter = netdev_priv(netdev);
1577         struct e1000_hw *hw = &adapter->hw;
1578         u32 rctl, icr = er32(ICR);
1579
1580         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1581                 return IRQ_NONE;  /* Not our interrupt */
1582
1583         /*
1584          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1585          * not set, then the adapter didn't send an interrupt
1586          */
1587         if (!(icr & E1000_ICR_INT_ASSERTED))
1588                 return IRQ_NONE;
1589
1590         /*
1591          * Interrupt Auto-Mask...upon reading ICR,
1592          * interrupts are masked.  No need for the
1593          * IMC write
1594          */
1595
1596         if (icr & E1000_ICR_LSC) {
1597                 hw->mac.get_link_status = 1;
1598                 /*
1599                  * ICH8 workaround-- Call gig speed drop workaround on cable
1600                  * disconnect (LSC) before accessing any PHY registers
1601                  */
1602                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1603                     (!(er32(STATUS) & E1000_STATUS_LU)))
1604                         schedule_work(&adapter->downshift_task);
1605
1606                 /*
1607                  * 80003ES2LAN workaround--
1608                  * For packet buffer work-around on link down event;
1609                  * disable receives here in the ISR and
1610                  * reset adapter in watchdog
1611                  */
1612                 if (netif_carrier_ok(netdev) &&
1613                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1614                         /* disable receives */
1615                         rctl = er32(RCTL);
1616                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1617                         adapter->flags |= FLAG_RX_RESTART_NOW;
1618                 }
1619                 /* guard against interrupt when we're going down */
1620                 if (!test_bit(__E1000_DOWN, &adapter->state))
1621                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1622         }
1623
1624         if (napi_schedule_prep(&adapter->napi)) {
1625                 adapter->total_tx_bytes = 0;
1626                 adapter->total_tx_packets = 0;
1627                 adapter->total_rx_bytes = 0;
1628                 adapter->total_rx_packets = 0;
1629                 __napi_schedule(&adapter->napi);
1630         }
1631
1632         return IRQ_HANDLED;
1633 }
1634
1635 static irqreturn_t e1000_msix_other(int irq, void *data)
1636 {
1637         struct net_device *netdev = data;
1638         struct e1000_adapter *adapter = netdev_priv(netdev);
1639         struct e1000_hw *hw = &adapter->hw;
1640         u32 icr = er32(ICR);
1641
1642         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1643                 if (!test_bit(__E1000_DOWN, &adapter->state))
1644                         ew32(IMS, E1000_IMS_OTHER);
1645                 return IRQ_NONE;
1646         }
1647
1648         if (icr & adapter->eiac_mask)
1649                 ew32(ICS, (icr & adapter->eiac_mask));
1650
1651         if (icr & E1000_ICR_OTHER) {
1652                 if (!(icr & E1000_ICR_LSC))
1653                         goto no_link_interrupt;
1654                 hw->mac.get_link_status = 1;
1655                 /* guard against interrupt when we're going down */
1656                 if (!test_bit(__E1000_DOWN, &adapter->state))
1657                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1658         }
1659
1660 no_link_interrupt:
1661         if (!test_bit(__E1000_DOWN, &adapter->state))
1662                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1663
1664         return IRQ_HANDLED;
1665 }
1666
1667
1668 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1669 {
1670         struct net_device *netdev = data;
1671         struct e1000_adapter *adapter = netdev_priv(netdev);
1672         struct e1000_hw *hw = &adapter->hw;
1673         struct e1000_ring *tx_ring = adapter->tx_ring;
1674
1675
1676         adapter->total_tx_bytes = 0;
1677         adapter->total_tx_packets = 0;
1678
1679         if (!e1000_clean_tx_irq(adapter))
1680                 /* Ring was not completely cleaned, so fire another interrupt */
1681                 ew32(ICS, tx_ring->ims_val);
1682
1683         return IRQ_HANDLED;
1684 }
1685
1686 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1687 {
1688         struct net_device *netdev = data;
1689         struct e1000_adapter *adapter = netdev_priv(netdev);
1690
1691         /* Write the ITR value calculated at the end of the
1692          * previous interrupt.
1693          */
1694         if (adapter->rx_ring->set_itr) {
1695                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1696                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1697                 adapter->rx_ring->set_itr = 0;
1698         }
1699
1700         if (napi_schedule_prep(&adapter->napi)) {
1701                 adapter->total_rx_bytes = 0;
1702                 adapter->total_rx_packets = 0;
1703                 __napi_schedule(&adapter->napi);
1704         }
1705         return IRQ_HANDLED;
1706 }
1707
1708 /**
1709  * e1000_configure_msix - Configure MSI-X hardware
1710  *
1711  * e1000_configure_msix sets up the hardware to properly
1712  * generate MSI-X interrupts.
1713  **/
1714 static void e1000_configure_msix(struct e1000_adapter *adapter)
1715 {
1716         struct e1000_hw *hw = &adapter->hw;
1717         struct e1000_ring *rx_ring = adapter->rx_ring;
1718         struct e1000_ring *tx_ring = adapter->tx_ring;
1719         int vector = 0;
1720         u32 ctrl_ext, ivar = 0;
1721
1722         adapter->eiac_mask = 0;
1723
1724         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1725         if (hw->mac.type == e1000_82574) {
1726                 u32 rfctl = er32(RFCTL);
1727                 rfctl |= E1000_RFCTL_ACK_DIS;
1728                 ew32(RFCTL, rfctl);
1729         }
1730
1731 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1732         /* Configure Rx vector */
1733         rx_ring->ims_val = E1000_IMS_RXQ0;
1734         adapter->eiac_mask |= rx_ring->ims_val;
1735         if (rx_ring->itr_val)
1736                 writel(1000000000 / (rx_ring->itr_val * 256),
1737                        hw->hw_addr + rx_ring->itr_register);
1738         else
1739                 writel(1, hw->hw_addr + rx_ring->itr_register);
1740         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1741
1742         /* Configure Tx vector */
1743         tx_ring->ims_val = E1000_IMS_TXQ0;
1744         vector++;
1745         if (tx_ring->itr_val)
1746                 writel(1000000000 / (tx_ring->itr_val * 256),
1747                        hw->hw_addr + tx_ring->itr_register);
1748         else
1749                 writel(1, hw->hw_addr + tx_ring->itr_register);
1750         adapter->eiac_mask |= tx_ring->ims_val;
1751         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1752
1753         /* set vector for Other Causes, e.g. link changes */
1754         vector++;
1755         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1756         if (rx_ring->itr_val)
1757                 writel(1000000000 / (rx_ring->itr_val * 256),
1758                        hw->hw_addr + E1000_EITR_82574(vector));
1759         else
1760                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1761
1762         /* Cause Tx interrupts on every write back */
1763         ivar |= (1 << 31);
1764
1765         ew32(IVAR, ivar);
1766
1767         /* enable MSI-X PBA support */
1768         ctrl_ext = er32(CTRL_EXT);
1769         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1770
1771         /* Auto-Mask Other interrupts upon ICR read */
1772 #define E1000_EIAC_MASK_82574   0x01F00000
1773         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1774         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1775         ew32(CTRL_EXT, ctrl_ext);
1776         e1e_flush();
1777 }
1778
1779 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1780 {
1781         if (adapter->msix_entries) {
1782                 pci_disable_msix(adapter->pdev);
1783                 kfree(adapter->msix_entries);
1784                 adapter->msix_entries = NULL;
1785         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1786                 pci_disable_msi(adapter->pdev);
1787                 adapter->flags &= ~FLAG_MSI_ENABLED;
1788         }
1789 }
1790
1791 /**
1792  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1793  *
1794  * Attempt to configure interrupts using the best available
1795  * capabilities of the hardware and kernel.
1796  **/
1797 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1798 {
1799         int err;
1800         int i;
1801
1802         switch (adapter->int_mode) {
1803         case E1000E_INT_MODE_MSIX:
1804                 if (adapter->flags & FLAG_HAS_MSIX) {
1805                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1806                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1807                                                       sizeof(struct msix_entry),
1808                                                       GFP_KERNEL);
1809                         if (adapter->msix_entries) {
1810                                 for (i = 0; i < adapter->num_vectors; i++)
1811                                         adapter->msix_entries[i].entry = i;
1812
1813                                 err = pci_enable_msix(adapter->pdev,
1814                                                       adapter->msix_entries,
1815                                                       adapter->num_vectors);
1816                                 if (err == 0)
1817                                         return;
1818                         }
1819                         /* MSI-X failed, so fall through and try MSI */
1820                         e_err("Failed to initialize MSI-X interrupts.  "
1821                               "Falling back to MSI interrupts.\n");
1822                         e1000e_reset_interrupt_capability(adapter);
1823                 }
1824                 adapter->int_mode = E1000E_INT_MODE_MSI;
1825                 /* Fall through */
1826         case E1000E_INT_MODE_MSI:
1827                 if (!pci_enable_msi(adapter->pdev)) {
1828                         adapter->flags |= FLAG_MSI_ENABLED;
1829                 } else {
1830                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1831                         e_err("Failed to initialize MSI interrupts.  Falling "
1832                               "back to legacy interrupts.\n");
1833                 }
1834                 /* Fall through */
1835         case E1000E_INT_MODE_LEGACY:
1836                 /* Don't do anything; this is the system default */
1837                 break;
1838         }
1839
1840         /* store the number of vectors being used */
1841         adapter->num_vectors = 1;
1842 }
1843
1844 /**
1845  * e1000_request_msix - Initialize MSI-X interrupts
1846  *
1847  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1848  * kernel.
1849  **/
1850 static int e1000_request_msix(struct e1000_adapter *adapter)
1851 {
1852         struct net_device *netdev = adapter->netdev;
1853         int err = 0, vector = 0;
1854
1855         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1856                 snprintf(adapter->rx_ring->name,
1857                          sizeof(adapter->rx_ring->name) - 1,
1858                          "%s-rx-0", netdev->name);
1859         else
1860                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1861         err = request_irq(adapter->msix_entries[vector].vector,
1862                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1863                           netdev);
1864         if (err)
1865                 goto out;
1866         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1867         adapter->rx_ring->itr_val = adapter->itr;
1868         vector++;
1869
1870         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1871                 snprintf(adapter->tx_ring->name,
1872                          sizeof(adapter->tx_ring->name) - 1,
1873                          "%s-tx-0", netdev->name);
1874         else
1875                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1876         err = request_irq(adapter->msix_entries[vector].vector,
1877                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1878                           netdev);
1879         if (err)
1880                 goto out;
1881         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1882         adapter->tx_ring->itr_val = adapter->itr;
1883         vector++;
1884
1885         err = request_irq(adapter->msix_entries[vector].vector,
1886                           e1000_msix_other, 0, netdev->name, netdev);
1887         if (err)
1888                 goto out;
1889
1890         e1000_configure_msix(adapter);
1891         return 0;
1892 out:
1893         return err;
1894 }
1895
1896 /**
1897  * e1000_request_irq - initialize interrupts
1898  *
1899  * Attempts to configure interrupts using the best available
1900  * capabilities of the hardware and kernel.
1901  **/
1902 static int e1000_request_irq(struct e1000_adapter *adapter)
1903 {
1904         struct net_device *netdev = adapter->netdev;
1905         int err;
1906
1907         if (adapter->msix_entries) {
1908                 err = e1000_request_msix(adapter);
1909                 if (!err)
1910                         return err;
1911                 /* fall back to MSI */
1912                 e1000e_reset_interrupt_capability(adapter);
1913                 adapter->int_mode = E1000E_INT_MODE_MSI;
1914                 e1000e_set_interrupt_capability(adapter);
1915         }
1916         if (adapter->flags & FLAG_MSI_ENABLED) {
1917                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1918                                   netdev->name, netdev);
1919                 if (!err)
1920                         return err;
1921
1922                 /* fall back to legacy interrupt */
1923                 e1000e_reset_interrupt_capability(adapter);
1924                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1925         }
1926
1927         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1928                           netdev->name, netdev);
1929         if (err)
1930                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1931
1932         return err;
1933 }
1934
1935 static void e1000_free_irq(struct e1000_adapter *adapter)
1936 {
1937         struct net_device *netdev = adapter->netdev;
1938
1939         if (adapter->msix_entries) {
1940                 int vector = 0;
1941
1942                 free_irq(adapter->msix_entries[vector].vector, netdev);
1943                 vector++;
1944
1945                 free_irq(adapter->msix_entries[vector].vector, netdev);
1946                 vector++;
1947
1948                 /* Other Causes interrupt vector */
1949                 free_irq(adapter->msix_entries[vector].vector, netdev);
1950                 return;
1951         }
1952
1953         free_irq(adapter->pdev->irq, netdev);
1954 }
1955
1956 /**
1957  * e1000_irq_disable - Mask off interrupt generation on the NIC
1958  **/
1959 static void e1000_irq_disable(struct e1000_adapter *adapter)
1960 {
1961         struct e1000_hw *hw = &adapter->hw;
1962
1963         ew32(IMC, ~0);
1964         if (adapter->msix_entries)
1965                 ew32(EIAC_82574, 0);
1966         e1e_flush();
1967
1968         if (adapter->msix_entries) {
1969                 int i;
1970                 for (i = 0; i < adapter->num_vectors; i++)
1971                         synchronize_irq(adapter->msix_entries[i].vector);
1972         } else {
1973                 synchronize_irq(adapter->pdev->irq);
1974         }
1975 }
1976
1977 /**
1978  * e1000_irq_enable - Enable default interrupt generation settings
1979  **/
1980 static void e1000_irq_enable(struct e1000_adapter *adapter)
1981 {
1982         struct e1000_hw *hw = &adapter->hw;
1983
1984         if (adapter->msix_entries) {
1985                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1986                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1987         } else {
1988                 ew32(IMS, IMS_ENABLE_MASK);
1989         }
1990         e1e_flush();
1991 }
1992
1993 /**
1994  * e1000e_get_hw_control - get control of the h/w from f/w
1995  * @adapter: address of board private structure
1996  *
1997  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1998  * For ASF and Pass Through versions of f/w this means that
1999  * the driver is loaded. For AMT version (only with 82573)
2000  * of the f/w this means that the network i/f is open.
2001  **/
2002 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2003 {
2004         struct e1000_hw *hw = &adapter->hw;
2005         u32 ctrl_ext;
2006         u32 swsm;
2007
2008         /* Let firmware know the driver has taken over */
2009         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2010                 swsm = er32(SWSM);
2011                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2012         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2013                 ctrl_ext = er32(CTRL_EXT);
2014                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2015         }
2016 }
2017
2018 /**
2019  * e1000e_release_hw_control - release control of the h/w to f/w
2020  * @adapter: address of board private structure
2021  *
2022  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2023  * For ASF and Pass Through versions of f/w this means that the
2024  * driver is no longer loaded. For AMT version (only with 82573) i
2025  * of the f/w this means that the network i/f is closed.
2026  *
2027  **/
2028 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2029 {
2030         struct e1000_hw *hw = &adapter->hw;
2031         u32 ctrl_ext;
2032         u32 swsm;
2033
2034         /* Let firmware taken over control of h/w */
2035         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2036                 swsm = er32(SWSM);
2037                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2038         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2039                 ctrl_ext = er32(CTRL_EXT);
2040                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2041         }
2042 }
2043
2044 /**
2045  * @e1000_alloc_ring - allocate memory for a ring structure
2046  **/
2047 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2048                                 struct e1000_ring *ring)
2049 {
2050         struct pci_dev *pdev = adapter->pdev;
2051
2052         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2053                                         GFP_KERNEL);
2054         if (!ring->desc)
2055                 return -ENOMEM;
2056
2057         return 0;
2058 }
2059
2060 /**
2061  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2062  * @adapter: board private structure
2063  *
2064  * Return 0 on success, negative on failure
2065  **/
2066 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2067 {
2068         struct e1000_ring *tx_ring = adapter->tx_ring;
2069         int err = -ENOMEM, size;
2070
2071         size = sizeof(struct e1000_buffer) * tx_ring->count;
2072         tx_ring->buffer_info = vzalloc(size);
2073         if (!tx_ring->buffer_info)
2074                 goto err;
2075
2076         /* round up to nearest 4K */
2077         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2078         tx_ring->size = ALIGN(tx_ring->size, 4096);
2079
2080         err = e1000_alloc_ring_dma(adapter, tx_ring);
2081         if (err)
2082                 goto err;
2083
2084         tx_ring->next_to_use = 0;
2085         tx_ring->next_to_clean = 0;
2086
2087         return 0;
2088 err:
2089         vfree(tx_ring->buffer_info);
2090         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2091         return err;
2092 }
2093
2094 /**
2095  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2096  * @adapter: board private structure
2097  *
2098  * Returns 0 on success, negative on failure
2099  **/
2100 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2101 {
2102         struct e1000_ring *rx_ring = adapter->rx_ring;
2103         struct e1000_buffer *buffer_info;
2104         int i, size, desc_len, err = -ENOMEM;
2105
2106         size = sizeof(struct e1000_buffer) * rx_ring->count;
2107         rx_ring->buffer_info = vzalloc(size);
2108         if (!rx_ring->buffer_info)
2109                 goto err;
2110
2111         for (i = 0; i < rx_ring->count; i++) {
2112                 buffer_info = &rx_ring->buffer_info[i];
2113                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2114                                                 sizeof(struct e1000_ps_page),
2115                                                 GFP_KERNEL);
2116                 if (!buffer_info->ps_pages)
2117                         goto err_pages;
2118         }
2119
2120         desc_len = sizeof(union e1000_rx_desc_packet_split);
2121
2122         /* Round up to nearest 4K */
2123         rx_ring->size = rx_ring->count * desc_len;
2124         rx_ring->size = ALIGN(rx_ring->size, 4096);
2125
2126         err = e1000_alloc_ring_dma(adapter, rx_ring);
2127         if (err)
2128                 goto err_pages;
2129
2130         rx_ring->next_to_clean = 0;
2131         rx_ring->next_to_use = 0;
2132         rx_ring->rx_skb_top = NULL;
2133
2134         return 0;
2135
2136 err_pages:
2137         for (i = 0; i < rx_ring->count; i++) {
2138                 buffer_info = &rx_ring->buffer_info[i];
2139                 kfree(buffer_info->ps_pages);
2140         }
2141 err:
2142         vfree(rx_ring->buffer_info);
2143         e_err("Unable to allocate memory for the receive descriptor ring\n");
2144         return err;
2145 }
2146
2147 /**
2148  * e1000_clean_tx_ring - Free Tx Buffers
2149  * @adapter: board private structure
2150  **/
2151 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2152 {
2153         struct e1000_ring *tx_ring = adapter->tx_ring;
2154         struct e1000_buffer *buffer_info;
2155         unsigned long size;
2156         unsigned int i;
2157
2158         for (i = 0; i < tx_ring->count; i++) {
2159                 buffer_info = &tx_ring->buffer_info[i];
2160                 e1000_put_txbuf(adapter, buffer_info);
2161         }
2162
2163         size = sizeof(struct e1000_buffer) * tx_ring->count;
2164         memset(tx_ring->buffer_info, 0, size);
2165
2166         memset(tx_ring->desc, 0, tx_ring->size);
2167
2168         tx_ring->next_to_use = 0;
2169         tx_ring->next_to_clean = 0;
2170
2171         writel(0, adapter->hw.hw_addr + tx_ring->head);
2172         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2173 }
2174
2175 /**
2176  * e1000e_free_tx_resources - Free Tx Resources per Queue
2177  * @adapter: board private structure
2178  *
2179  * Free all transmit software resources
2180  **/
2181 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2182 {
2183         struct pci_dev *pdev = adapter->pdev;
2184         struct e1000_ring *tx_ring = adapter->tx_ring;
2185
2186         e1000_clean_tx_ring(adapter);
2187
2188         vfree(tx_ring->buffer_info);
2189         tx_ring->buffer_info = NULL;
2190
2191         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2192                           tx_ring->dma);
2193         tx_ring->desc = NULL;
2194 }
2195
2196 /**
2197  * e1000e_free_rx_resources - Free Rx Resources
2198  * @adapter: board private structure
2199  *
2200  * Free all receive software resources
2201  **/
2202
2203 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2204 {
2205         struct pci_dev *pdev = adapter->pdev;
2206         struct e1000_ring *rx_ring = adapter->rx_ring;
2207         int i;
2208
2209         e1000_clean_rx_ring(adapter);
2210
2211         for (i = 0; i < rx_ring->count; i++)
2212                 kfree(rx_ring->buffer_info[i].ps_pages);
2213
2214         vfree(rx_ring->buffer_info);
2215         rx_ring->buffer_info = NULL;
2216
2217         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2218                           rx_ring->dma);
2219         rx_ring->desc = NULL;
2220 }
2221
2222 /**
2223  * e1000_update_itr - update the dynamic ITR value based on statistics
2224  * @adapter: pointer to adapter
2225  * @itr_setting: current adapter->itr
2226  * @packets: the number of packets during this measurement interval
2227  * @bytes: the number of bytes during this measurement interval
2228  *
2229  *      Stores a new ITR value based on packets and byte
2230  *      counts during the last interrupt.  The advantage of per interrupt
2231  *      computation is faster updates and more accurate ITR for the current
2232  *      traffic pattern.  Constants in this function were computed
2233  *      based on theoretical maximum wire speed and thresholds were set based
2234  *      on testing data as well as attempting to minimize response time
2235  *      while increasing bulk throughput.  This functionality is controlled
2236  *      by the InterruptThrottleRate module parameter.
2237  **/
2238 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2239                                      u16 itr_setting, int packets,
2240                                      int bytes)
2241 {
2242         unsigned int retval = itr_setting;
2243
2244         if (packets == 0)
2245                 goto update_itr_done;
2246
2247         switch (itr_setting) {
2248         case lowest_latency:
2249                 /* handle TSO and jumbo frames */
2250                 if (bytes/packets > 8000)
2251                         retval = bulk_latency;
2252                 else if ((packets < 5) && (bytes > 512))
2253                         retval = low_latency;
2254                 break;
2255         case low_latency:  /* 50 usec aka 20000 ints/s */
2256                 if (bytes > 10000) {
2257                         /* this if handles the TSO accounting */
2258                         if (bytes/packets > 8000)
2259                                 retval = bulk_latency;
2260                         else if ((packets < 10) || ((bytes/packets) > 1200))
2261                                 retval = bulk_latency;
2262                         else if ((packets > 35))
2263                                 retval = lowest_latency;
2264                 } else if (bytes/packets > 2000) {
2265                         retval = bulk_latency;
2266                 } else if (packets <= 2 && bytes < 512) {
2267                         retval = lowest_latency;
2268                 }
2269                 break;
2270         case bulk_latency: /* 250 usec aka 4000 ints/s */
2271                 if (bytes > 25000) {
2272                         if (packets > 35)
2273                                 retval = low_latency;
2274                 } else if (bytes < 6000) {
2275                         retval = low_latency;
2276                 }
2277                 break;
2278         }
2279
2280 update_itr_done:
2281         return retval;
2282 }
2283
2284 static void e1000_set_itr(struct e1000_adapter *adapter)
2285 {
2286         struct e1000_hw *hw = &adapter->hw;
2287         u16 current_itr;
2288         u32 new_itr = adapter->itr;
2289
2290         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2291         if (adapter->link_speed != SPEED_1000) {
2292                 current_itr = 0;
2293                 new_itr = 4000;
2294                 goto set_itr_now;
2295         }
2296
2297         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2298                 new_itr = 0;
2299                 goto set_itr_now;
2300         }
2301
2302         adapter->tx_itr = e1000_update_itr(adapter,
2303                                     adapter->tx_itr,
2304                                     adapter->total_tx_packets,
2305                                     adapter->total_tx_bytes);
2306         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2307         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2308                 adapter->tx_itr = low_latency;
2309
2310         adapter->rx_itr = e1000_update_itr(adapter,
2311                                     adapter->rx_itr,
2312                                     adapter->total_rx_packets,
2313                                     adapter->total_rx_bytes);
2314         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2315         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2316                 adapter->rx_itr = low_latency;
2317
2318         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2319
2320         switch (current_itr) {
2321         /* counts and packets in update_itr are dependent on these numbers */
2322         case lowest_latency:
2323                 new_itr = 70000;
2324                 break;
2325         case low_latency:
2326                 new_itr = 20000; /* aka hwitr = ~200 */
2327                 break;
2328         case bulk_latency:
2329                 new_itr = 4000;
2330                 break;
2331         default:
2332                 break;
2333         }
2334
2335 set_itr_now:
2336         if (new_itr != adapter->itr) {
2337                 /*
2338                  * this attempts to bias the interrupt rate towards Bulk
2339                  * by adding intermediate steps when interrupt rate is
2340                  * increasing
2341                  */
2342                 new_itr = new_itr > adapter->itr ?
2343                              min(adapter->itr + (new_itr >> 2), new_itr) :
2344                              new_itr;
2345                 adapter->itr = new_itr;
2346                 adapter->rx_ring->itr_val = new_itr;
2347                 if (adapter->msix_entries)
2348                         adapter->rx_ring->set_itr = 1;
2349                 else
2350                         if (new_itr)
2351                                 ew32(ITR, 1000000000 / (new_itr * 256));
2352                         else
2353                                 ew32(ITR, 0);
2354         }
2355 }
2356
2357 /**
2358  * e1000_alloc_queues - Allocate memory for all rings
2359  * @adapter: board private structure to initialize
2360  **/
2361 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2362 {
2363         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2364         if (!adapter->tx_ring)
2365                 goto err;
2366
2367         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2368         if (!adapter->rx_ring)
2369                 goto err;
2370
2371         return 0;
2372 err:
2373         e_err("Unable to allocate memory for queues\n");
2374         kfree(adapter->rx_ring);
2375         kfree(adapter->tx_ring);
2376         return -ENOMEM;
2377 }
2378
2379 /**
2380  * e1000_clean - NAPI Rx polling callback
2381  * @napi: struct associated with this polling callback
2382  * @budget: amount of packets driver is allowed to process this poll
2383  **/
2384 static int e1000_clean(struct napi_struct *napi, int budget)
2385 {
2386         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2387         struct e1000_hw *hw = &adapter->hw;
2388         struct net_device *poll_dev = adapter->netdev;
2389         int tx_cleaned = 1, work_done = 0;
2390
2391         adapter = netdev_priv(poll_dev);
2392
2393         if (adapter->msix_entries &&
2394             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2395                 goto clean_rx;
2396
2397         tx_cleaned = e1000_clean_tx_irq(adapter);
2398
2399 clean_rx:
2400         adapter->clean_rx(adapter, &work_done, budget);
2401
2402         if (!tx_cleaned)
2403                 work_done = budget;
2404
2405         /* If budget not fully consumed, exit the polling mode */
2406         if (work_done < budget) {
2407                 if (adapter->itr_setting & 3)
2408                         e1000_set_itr(adapter);
2409                 napi_complete(napi);
2410                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2411                         if (adapter->msix_entries)
2412                                 ew32(IMS, adapter->rx_ring->ims_val);
2413                         else
2414                                 e1000_irq_enable(adapter);
2415                 }
2416         }
2417
2418         return work_done;
2419 }
2420
2421 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2422 {
2423         struct e1000_adapter *adapter = netdev_priv(netdev);
2424         struct e1000_hw *hw = &adapter->hw;
2425         u32 vfta, index;
2426
2427         /* don't update vlan cookie if already programmed */
2428         if ((adapter->hw.mng_cookie.status &
2429              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2430             (vid == adapter->mng_vlan_id))
2431                 return;
2432
2433         /* add VID to filter table */
2434         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2435                 index = (vid >> 5) & 0x7F;
2436                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2437                 vfta |= (1 << (vid & 0x1F));
2438                 hw->mac.ops.write_vfta(hw, index, vfta);
2439         }
2440
2441         set_bit(vid, adapter->active_vlans);
2442 }
2443
2444 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2445 {
2446         struct e1000_adapter *adapter = netdev_priv(netdev);
2447         struct e1000_hw *hw = &adapter->hw;
2448         u32 vfta, index;
2449
2450         if ((adapter->hw.mng_cookie.status &
2451              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2452             (vid == adapter->mng_vlan_id)) {
2453                 /* release control to f/w */
2454                 e1000e_release_hw_control(adapter);
2455                 return;
2456         }
2457
2458         /* remove VID from filter table */
2459         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2460                 index = (vid >> 5) & 0x7F;
2461                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2462                 vfta &= ~(1 << (vid & 0x1F));
2463                 hw->mac.ops.write_vfta(hw, index, vfta);
2464         }
2465
2466         clear_bit(vid, adapter->active_vlans);
2467 }
2468
2469 /**
2470  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2471  * @adapter: board private structure to initialize
2472  **/
2473 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2474 {
2475         struct net_device *netdev = adapter->netdev;
2476         struct e1000_hw *hw = &adapter->hw;
2477         u32 rctl;
2478
2479         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2480                 /* disable VLAN receive filtering */
2481                 rctl = er32(RCTL);
2482                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2483                 ew32(RCTL, rctl);
2484
2485                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2486                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2487                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2488                 }
2489         }
2490 }
2491
2492 /**
2493  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2494  * @adapter: board private structure to initialize
2495  **/
2496 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2497 {
2498         struct e1000_hw *hw = &adapter->hw;
2499         u32 rctl;
2500
2501         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2502                 /* enable VLAN receive filtering */
2503                 rctl = er32(RCTL);
2504                 rctl |= E1000_RCTL_VFE;
2505                 rctl &= ~E1000_RCTL_CFIEN;
2506                 ew32(RCTL, rctl);
2507         }
2508 }
2509
2510 /**
2511  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2512  * @adapter: board private structure to initialize
2513  **/
2514 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2515 {
2516         struct e1000_hw *hw = &adapter->hw;
2517         u32 ctrl;
2518
2519         /* disable VLAN tag insert/strip */
2520         ctrl = er32(CTRL);
2521         ctrl &= ~E1000_CTRL_VME;
2522         ew32(CTRL, ctrl);
2523 }
2524
2525 /**
2526  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2527  * @adapter: board private structure to initialize
2528  **/
2529 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2530 {
2531         struct e1000_hw *hw = &adapter->hw;
2532         u32 ctrl;
2533
2534         /* enable VLAN tag insert/strip */
2535         ctrl = er32(CTRL);
2536         ctrl |= E1000_CTRL_VME;
2537         ew32(CTRL, ctrl);
2538 }
2539
2540 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2541 {
2542         struct net_device *netdev = adapter->netdev;
2543         u16 vid = adapter->hw.mng_cookie.vlan_id;
2544         u16 old_vid = adapter->mng_vlan_id;
2545
2546         if (adapter->hw.mng_cookie.status &
2547             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2548                 e1000_vlan_rx_add_vid(netdev, vid);
2549                 adapter->mng_vlan_id = vid;
2550         }
2551
2552         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2553                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2554 }
2555
2556 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2557 {
2558         u16 vid;
2559
2560         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2561
2562         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2563                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2564 }
2565
2566 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2567 {
2568         struct e1000_hw *hw = &adapter->hw;
2569         u32 manc, manc2h, mdef, i, j;
2570
2571         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2572                 return;
2573
2574         manc = er32(MANC);
2575
2576         /*
2577          * enable receiving management packets to the host. this will probably
2578          * generate destination unreachable messages from the host OS, but
2579          * the packets will be handled on SMBUS
2580          */
2581         manc |= E1000_MANC_EN_MNG2HOST;
2582         manc2h = er32(MANC2H);
2583
2584         switch (hw->mac.type) {
2585         default:
2586                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2587                 break;
2588         case e1000_82574:
2589         case e1000_82583:
2590                 /*
2591                  * Check if IPMI pass-through decision filter already exists;
2592                  * if so, enable it.
2593                  */
2594                 for (i = 0, j = 0; i < 8; i++) {
2595                         mdef = er32(MDEF(i));
2596
2597                         /* Ignore filters with anything other than IPMI ports */
2598                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2599                                 continue;
2600
2601                         /* Enable this decision filter in MANC2H */
2602                         if (mdef)
2603                                 manc2h |= (1 << i);
2604
2605                         j |= mdef;
2606                 }
2607
2608                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2609                         break;
2610
2611                 /* Create new decision filter in an empty filter */
2612                 for (i = 0, j = 0; i < 8; i++)
2613                         if (er32(MDEF(i)) == 0) {
2614                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2615                                                E1000_MDEF_PORT_664));
2616                                 manc2h |= (1 << 1);
2617                                 j++;
2618                                 break;
2619                         }
2620
2621                 if (!j)
2622                         e_warn("Unable to create IPMI pass-through filter\n");
2623                 break;
2624         }
2625
2626         ew32(MANC2H, manc2h);
2627         ew32(MANC, manc);
2628 }
2629
2630 /**
2631  * e1000_configure_tx - Configure Transmit Unit after Reset
2632  * @adapter: board private structure
2633  *
2634  * Configure the Tx unit of the MAC after a reset.
2635  **/
2636 static void e1000_configure_tx(struct e1000_adapter *adapter)
2637 {
2638         struct e1000_hw *hw = &adapter->hw;
2639         struct e1000_ring *tx_ring = adapter->tx_ring;
2640         u64 tdba;
2641         u32 tdlen, tctl, tipg, tarc;
2642         u32 ipgr1, ipgr2;
2643
2644         /* Setup the HW Tx Head and Tail descriptor pointers */
2645         tdba = tx_ring->dma;
2646         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2647         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2648         ew32(TDBAH, (tdba >> 32));
2649         ew32(TDLEN, tdlen);
2650         ew32(TDH, 0);
2651         ew32(TDT, 0);
2652         tx_ring->head = E1000_TDH;
2653         tx_ring->tail = E1000_TDT;
2654
2655         /* Set the default values for the Tx Inter Packet Gap timer */
2656         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2657         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2658         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2659
2660         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2661                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2662
2663         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2664         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2665         ew32(TIPG, tipg);
2666
2667         /* Set the Tx Interrupt Delay register */
2668         ew32(TIDV, adapter->tx_int_delay);
2669         /* Tx irq moderation */
2670         ew32(TADV, adapter->tx_abs_int_delay);
2671
2672         if (adapter->flags2 & FLAG2_DMA_BURST) {
2673                 u32 txdctl = er32(TXDCTL(0));
2674                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2675                             E1000_TXDCTL_WTHRESH);
2676                 /*
2677                  * set up some performance related parameters to encourage the
2678                  * hardware to use the bus more efficiently in bursts, depends
2679                  * on the tx_int_delay to be enabled,
2680                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2681                  * hthresh = 1 ==> prefetch when one or more available
2682                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2683                  * BEWARE: this seems to work but should be considered first if
2684                  * there are Tx hangs or other Tx related bugs
2685                  */
2686                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2687                 ew32(TXDCTL(0), txdctl);
2688                 /* erratum work around: set txdctl the same for both queues */
2689                 ew32(TXDCTL(1), txdctl);
2690         }
2691
2692         /* Program the Transmit Control Register */
2693         tctl = er32(TCTL);
2694         tctl &= ~E1000_TCTL_CT;
2695         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2696                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2697
2698         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2699                 tarc = er32(TARC(0));
2700                 /*
2701                  * set the speed mode bit, we'll clear it if we're not at
2702                  * gigabit link later
2703                  */
2704 #define SPEED_MODE_BIT (1 << 21)
2705                 tarc |= SPEED_MODE_BIT;
2706                 ew32(TARC(0), tarc);
2707         }
2708
2709         /* errata: program both queues to unweighted RR */
2710         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2711                 tarc = er32(TARC(0));
2712                 tarc |= 1;
2713                 ew32(TARC(0), tarc);
2714                 tarc = er32(TARC(1));
2715                 tarc |= 1;
2716                 ew32(TARC(1), tarc);
2717         }
2718
2719         /* Setup Transmit Descriptor Settings for eop descriptor */
2720         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2721
2722         /* only set IDE if we are delaying interrupts using the timers */
2723         if (adapter->tx_int_delay)
2724                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2725
2726         /* enable Report Status bit */
2727         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2728
2729         ew32(TCTL, tctl);
2730
2731         e1000e_config_collision_dist(hw);
2732 }
2733
2734 /**
2735  * e1000_setup_rctl - configure the receive control registers
2736  * @adapter: Board private structure
2737  **/
2738 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2739                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2740 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2741 {
2742         struct e1000_hw *hw = &adapter->hw;
2743         u32 rctl, rfctl;
2744         u32 pages = 0;
2745
2746         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2747         if (hw->mac.type == e1000_pch2lan) {
2748                 s32 ret_val;
2749
2750                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2751                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2752                 else
2753                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2754
2755                 if (ret_val)
2756                         e_dbg("failed to enable jumbo frame workaround mode\n");
2757         }
2758
2759         /* Program MC offset vector base */
2760         rctl = er32(RCTL);
2761         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2762         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2763                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2764                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2765
2766         /* Do not Store bad packets */
2767         rctl &= ~E1000_RCTL_SBP;
2768
2769         /* Enable Long Packet receive */
2770         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2771                 rctl &= ~E1000_RCTL_LPE;
2772         else
2773                 rctl |= E1000_RCTL_LPE;
2774
2775         /* Some systems expect that the CRC is included in SMBUS traffic. The
2776          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2777          * host memory when this is enabled
2778          */
2779         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2780                 rctl |= E1000_RCTL_SECRC;
2781
2782         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2783         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2784                 u16 phy_data;
2785
2786                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2787                 phy_data &= 0xfff8;
2788                 phy_data |= (1 << 2);
2789                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2790
2791                 e1e_rphy(hw, 22, &phy_data);
2792                 phy_data &= 0x0fff;
2793                 phy_data |= (1 << 14);
2794                 e1e_wphy(hw, 0x10, 0x2823);
2795                 e1e_wphy(hw, 0x11, 0x0003);
2796                 e1e_wphy(hw, 22, phy_data);
2797         }
2798
2799         /* Setup buffer sizes */
2800         rctl &= ~E1000_RCTL_SZ_4096;
2801         rctl |= E1000_RCTL_BSEX;
2802         switch (adapter->rx_buffer_len) {
2803         case 2048:
2804         default:
2805                 rctl |= E1000_RCTL_SZ_2048;
2806                 rctl &= ~E1000_RCTL_BSEX;
2807                 break;
2808         case 4096:
2809                 rctl |= E1000_RCTL_SZ_4096;
2810                 break;
2811         case 8192:
2812                 rctl |= E1000_RCTL_SZ_8192;
2813                 break;
2814         case 16384:
2815                 rctl |= E1000_RCTL_SZ_16384;
2816                 break;
2817         }
2818
2819         /*
2820          * 82571 and greater support packet-split where the protocol
2821          * header is placed in skb->data and the packet data is
2822          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2823          * In the case of a non-split, skb->data is linearly filled,
2824          * followed by the page buffers.  Therefore, skb->data is
2825          * sized to hold the largest protocol header.
2826          *
2827          * allocations using alloc_page take too long for regular MTU
2828          * so only enable packet split for jumbo frames
2829          *
2830          * Using pages when the page size is greater than 16k wastes
2831          * a lot of memory, since we allocate 3 pages at all times
2832          * per packet.
2833          */
2834         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2835         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2836             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2837                 adapter->rx_ps_pages = pages;
2838         else
2839                 adapter->rx_ps_pages = 0;
2840
2841         if (adapter->rx_ps_pages) {
2842                 u32 psrctl = 0;
2843
2844                 /* Configure extra packet-split registers */
2845                 rfctl = er32(RFCTL);
2846                 rfctl |= E1000_RFCTL_EXTEN;
2847                 /*
2848                  * disable packet split support for IPv6 extension headers,
2849                  * because some malformed IPv6 headers can hang the Rx
2850                  */
2851                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2852                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2853
2854                 ew32(RFCTL, rfctl);
2855
2856                 /* Enable Packet split descriptors */
2857                 rctl |= E1000_RCTL_DTYP_PS;
2858
2859                 psrctl |= adapter->rx_ps_bsize0 >>
2860                         E1000_PSRCTL_BSIZE0_SHIFT;
2861
2862                 switch (adapter->rx_ps_pages) {
2863                 case 3:
2864                         psrctl |= PAGE_SIZE <<
2865                                 E1000_PSRCTL_BSIZE3_SHIFT;
2866                 case 2:
2867                         psrctl |= PAGE_SIZE <<
2868                                 E1000_PSRCTL_BSIZE2_SHIFT;
2869                 case 1:
2870                         psrctl |= PAGE_SIZE >>
2871                                 E1000_PSRCTL_BSIZE1_SHIFT;
2872                         break;
2873                 }
2874
2875                 ew32(PSRCTL, psrctl);
2876         }
2877
2878         ew32(RCTL, rctl);
2879         /* just started the receive unit, no need to restart */
2880         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2881 }
2882
2883 /**
2884  * e1000_configure_rx - Configure Receive Unit after Reset
2885  * @adapter: board private structure
2886  *
2887  * Configure the Rx unit of the MAC after a reset.
2888  **/
2889 static void e1000_configure_rx(struct e1000_adapter *adapter)
2890 {
2891         struct e1000_hw *hw = &adapter->hw;
2892         struct e1000_ring *rx_ring = adapter->rx_ring;
2893         u64 rdba;
2894         u32 rdlen, rctl, rxcsum, ctrl_ext;
2895
2896         if (adapter->rx_ps_pages) {
2897                 /* this is a 32 byte descriptor */
2898                 rdlen = rx_ring->count *
2899                     sizeof(union e1000_rx_desc_packet_split);
2900                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2901                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2902         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2903                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2904                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2905                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2906         } else {
2907                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2908                 adapter->clean_rx = e1000_clean_rx_irq;
2909                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2910         }
2911
2912         /* disable receives while setting up the descriptors */
2913         rctl = er32(RCTL);
2914         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2915         e1e_flush();
2916         usleep_range(10000, 20000);
2917
2918         if (adapter->flags2 & FLAG2_DMA_BURST) {
2919                 /*
2920                  * set the writeback threshold (only takes effect if the RDTR
2921                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2922                  * enable prefetching of 0x20 Rx descriptors
2923                  * granularity = 01
2924                  * wthresh = 04,
2925                  * hthresh = 04,
2926                  * pthresh = 0x20
2927                  */
2928                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2929                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2930
2931                 /*
2932                  * override the delay timers for enabling bursting, only if
2933                  * the value was not set by the user via module options
2934                  */
2935                 if (adapter->rx_int_delay == DEFAULT_RDTR)
2936                         adapter->rx_int_delay = BURST_RDTR;
2937                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2938                         adapter->rx_abs_int_delay = BURST_RADV;
2939         }
2940
2941         /* set the Receive Delay Timer Register */
2942         ew32(RDTR, adapter->rx_int_delay);
2943
2944         /* irq moderation */
2945         ew32(RADV, adapter->rx_abs_int_delay);
2946         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2947                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2948
2949         ctrl_ext = er32(CTRL_EXT);
2950         /* Auto-Mask interrupts upon ICR access */
2951         ctrl_ext |= E1000_CTRL_EXT_IAME;
2952         ew32(IAM, 0xffffffff);
2953         ew32(CTRL_EXT, ctrl_ext);
2954         e1e_flush();
2955
2956         /*
2957          * Setup the HW Rx Head and Tail Descriptor Pointers and
2958          * the Base and Length of the Rx Descriptor Ring
2959          */
2960         rdba = rx_ring->dma;
2961         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2962         ew32(RDBAH, (rdba >> 32));
2963         ew32(RDLEN, rdlen);
2964         ew32(RDH, 0);
2965         ew32(RDT, 0);
2966         rx_ring->head = E1000_RDH;
2967         rx_ring->tail = E1000_RDT;
2968
2969         /* Enable Receive Checksum Offload for TCP and UDP */
2970         rxcsum = er32(RXCSUM);
2971         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2972                 rxcsum |= E1000_RXCSUM_TUOFL;
2973
2974                 /*
2975                  * IPv4 payload checksum for UDP fragments must be
2976                  * used in conjunction with packet-split.
2977                  */
2978                 if (adapter->rx_ps_pages)
2979                         rxcsum |= E1000_RXCSUM_IPPCSE;
2980         } else {
2981                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2982                 /* no need to clear IPPCSE as it defaults to 0 */
2983         }
2984         ew32(RXCSUM, rxcsum);
2985
2986         /*
2987          * Enable early receives on supported devices, only takes effect when
2988          * packet size is equal or larger than the specified value (in 8 byte
2989          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2990          */
2991         if ((adapter->flags & FLAG_HAS_ERT) ||
2992             (adapter->hw.mac.type == e1000_pch2lan)) {
2993                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2994                         u32 rxdctl = er32(RXDCTL(0));
2995                         ew32(RXDCTL(0), rxdctl | 0x3);
2996                         if (adapter->flags & FLAG_HAS_ERT)
2997                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2998                         /*
2999                          * With jumbo frames and early-receive enabled,
3000                          * excessive C-state transition latencies result in
3001                          * dropped transactions.
3002                          */
3003                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3004                 } else {
3005                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3006                                               PM_QOS_DEFAULT_VALUE);
3007                 }
3008         }
3009
3010         /* Enable Receives */
3011         ew32(RCTL, rctl);
3012 }
3013
3014 /**
3015  *  e1000_update_mc_addr_list - Update Multicast addresses
3016  *  @hw: pointer to the HW structure
3017  *  @mc_addr_list: array of multicast addresses to program
3018  *  @mc_addr_count: number of multicast addresses to program
3019  *
3020  *  Updates the Multicast Table Array.
3021  *  The caller must have a packed mc_addr_list of multicast addresses.
3022  **/
3023 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3024                                       u32 mc_addr_count)
3025 {
3026         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3027 }
3028
3029 /**
3030  * e1000_set_multi - Multicast and Promiscuous mode set
3031  * @netdev: network interface device structure
3032  *
3033  * The set_multi entry point is called whenever the multicast address
3034  * list or the network interface flags are updated.  This routine is
3035  * responsible for configuring the hardware for proper multicast,
3036  * promiscuous mode, and all-multi behavior.
3037  **/
3038 static void e1000_set_multi(struct net_device *netdev)
3039 {
3040         struct e1000_adapter *adapter = netdev_priv(netdev);
3041         struct e1000_hw *hw = &adapter->hw;
3042         struct netdev_hw_addr *ha;
3043         u8  *mta_list;
3044         u32 rctl;
3045
3046         /* Check for Promiscuous and All Multicast modes */
3047
3048         rctl = er32(RCTL);
3049
3050         if (netdev->flags & IFF_PROMISC) {
3051                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3052                 rctl &= ~E1000_RCTL_VFE;
3053                 /* Do not hardware filter VLANs in promisc mode */
3054                 e1000e_vlan_filter_disable(adapter);
3055         } else {
3056                 if (netdev->flags & IFF_ALLMULTI) {
3057                         rctl |= E1000_RCTL_MPE;
3058                         rctl &= ~E1000_RCTL_UPE;
3059                 } else {
3060                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3061                 }
3062                 e1000e_vlan_filter_enable(adapter);
3063         }
3064
3065         ew32(RCTL, rctl);
3066
3067         if (!netdev_mc_empty(netdev)) {
3068                 int i = 0;
3069
3070                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3071                 if (!mta_list)
3072                         return;
3073
3074                 /* prepare a packed array of only addresses. */
3075                 netdev_for_each_mc_addr(ha, netdev)
3076                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3077
3078                 e1000_update_mc_addr_list(hw, mta_list, i);
3079                 kfree(mta_list);
3080         } else {
3081                 /*
3082                  * if we're called from probe, we might not have
3083                  * anything to do here, so clear out the list
3084                  */
3085                 e1000_update_mc_addr_list(hw, NULL, 0);
3086         }
3087
3088         if (netdev->features & NETIF_F_HW_VLAN_RX)
3089                 e1000e_vlan_strip_enable(adapter);
3090         else
3091                 e1000e_vlan_strip_disable(adapter);
3092 }
3093
3094 /**
3095  * e1000_configure - configure the hardware for Rx and Tx
3096  * @adapter: private board structure
3097  **/
3098 static void e1000_configure(struct e1000_adapter *adapter)
3099 {
3100         e1000_set_multi(adapter->netdev);
3101
3102         e1000_restore_vlan(adapter);
3103         e1000_init_manageability_pt(adapter);
3104
3105         e1000_configure_tx(adapter);
3106         e1000_setup_rctl(adapter);
3107         e1000_configure_rx(adapter);
3108         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3109 }
3110
3111 /**
3112  * e1000e_power_up_phy - restore link in case the phy was powered down
3113  * @adapter: address of board private structure
3114  *
3115  * The phy may be powered down to save power and turn off link when the
3116  * driver is unloaded and wake on lan is not enabled (among others)
3117  * *** this routine MUST be followed by a call to e1000e_reset ***
3118  **/
3119 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3120 {
3121         if (adapter->hw.phy.ops.power_up)
3122                 adapter->hw.phy.ops.power_up(&adapter->hw);
3123
3124         adapter->hw.mac.ops.setup_link(&adapter->hw);
3125 }
3126
3127 /**
3128  * e1000_power_down_phy - Power down the PHY
3129  *
3130  * Power down the PHY so no link is implied when interface is down.
3131  * The PHY cannot be powered down if management or WoL is active.
3132  */
3133 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3134 {
3135         /* WoL is enabled */
3136         if (adapter->wol)
3137                 return;
3138
3139         if (adapter->hw.phy.ops.power_down)
3140                 adapter->hw.phy.ops.power_down(&adapter->hw);
3141 }
3142
3143 /**
3144  * e1000e_reset - bring the hardware into a known good state
3145  *
3146  * This function boots the hardware and enables some settings that
3147  * require a configuration cycle of the hardware - those cannot be
3148  * set/changed during runtime. After reset the device needs to be
3149  * properly configured for Rx, Tx etc.
3150  */
3151 void e1000e_reset(struct e1000_adapter *adapter)
3152 {
3153         struct e1000_mac_info *mac = &adapter->hw.mac;
3154         struct e1000_fc_info *fc = &adapter->hw.fc;
3155         struct e1000_hw *hw = &adapter->hw;
3156         u32 tx_space, min_tx_space, min_rx_space;
3157         u32 pba = adapter->pba;
3158         u16 hwm;
3159
3160         /* reset Packet Buffer Allocation to default */
3161         ew32(PBA, pba);
3162
3163         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3164                 /*
3165                  * To maintain wire speed transmits, the Tx FIFO should be
3166                  * large enough to accommodate two full transmit packets,
3167                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3168                  * the Rx FIFO should be large enough to accommodate at least
3169                  * one full receive packet and is similarly rounded up and
3170                  * expressed in KB.
3171                  */
3172                 pba = er32(PBA);
3173                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3174                 tx_space = pba >> 16;
3175                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3176                 pba &= 0xffff;
3177                 /*
3178                  * the Tx fifo also stores 16 bytes of information about the Tx
3179                  * but don't include ethernet FCS because hardware appends it
3180                  */
3181                 min_tx_space = (adapter->max_frame_size +
3182                                 sizeof(struct e1000_tx_desc) -
3183                                 ETH_FCS_LEN) * 2;
3184                 min_tx_space = ALIGN(min_tx_space, 1024);
3185                 min_tx_space >>= 10;
3186                 /* software strips receive CRC, so leave room for it */
3187                 min_rx_space = adapter->max_frame_size;
3188                 min_rx_space = ALIGN(min_rx_space, 1024);
3189                 min_rx_space >>= 10;
3190
3191                 /*
3192                  * If current Tx allocation is less than the min Tx FIFO size,
3193                  * and the min Tx FIFO size is less than the current Rx FIFO
3194                  * allocation, take space away from current Rx allocation
3195                  */
3196                 if ((tx_space < min_tx_space) &&
3197                     ((min_tx_space - tx_space) < pba)) {
3198                         pba -= min_tx_space - tx_space;
3199
3200                         /*
3201                          * if short on Rx space, Rx wins and must trump Tx
3202                          * adjustment or use Early Receive if available
3203                          */
3204                         if ((pba < min_rx_space) &&
3205                             (!(adapter->flags & FLAG_HAS_ERT)))
3206                                 /* ERT enabled in e1000_configure_rx */
3207                                 pba = min_rx_space;
3208                 }
3209
3210                 ew32(PBA, pba);
3211         }
3212
3213         /*
3214          * flow control settings
3215          *
3216          * The high water mark must be low enough to fit one full frame
3217          * (or the size used for early receive) above it in the Rx FIFO.
3218          * Set it to the lower of:
3219          * - 90% of the Rx FIFO size, and
3220          * - the full Rx FIFO size minus the early receive size (for parts
3221          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3222          * - the full Rx FIFO size minus one full frame
3223          */
3224         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3225                 fc->pause_time = 0xFFFF;
3226         else
3227                 fc->pause_time = E1000_FC_PAUSE_TIME;
3228         fc->send_xon = 1;
3229         fc->current_mode = fc->requested_mode;
3230
3231         switch (hw->mac.type) {
3232         default:
3233                 if ((adapter->flags & FLAG_HAS_ERT) &&
3234                     (adapter->netdev->mtu > ETH_DATA_LEN))
3235                         hwm = min(((pba << 10) * 9 / 10),
3236                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3237                 else
3238                         hwm = min(((pba << 10) * 9 / 10),
3239                                   ((pba << 10) - adapter->max_frame_size));
3240
3241                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3242                 fc->low_water = fc->high_water - 8;
3243                 break;
3244         case e1000_pchlan:
3245                 /*
3246                  * Workaround PCH LOM adapter hangs with certain network
3247                  * loads.  If hangs persist, try disabling Tx flow control.
3248                  */
3249                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3250                         fc->high_water = 0x3500;
3251                         fc->low_water  = 0x1500;
3252                 } else {
3253                         fc->high_water = 0x5000;
3254                         fc->low_water  = 0x3000;
3255                 }
3256                 fc->refresh_time = 0x1000;
3257                 break;
3258         case e1000_pch2lan:
3259                 fc->high_water = 0x05C20;
3260                 fc->low_water = 0x05048;
3261                 fc->pause_time = 0x0650;
3262                 fc->refresh_time = 0x0400;
3263                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3264                         pba = 14;
3265                         ew32(PBA, pba);
3266                 }
3267                 break;
3268         }
3269
3270         /*
3271          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3272          * fit in receive buffer and early-receive not supported.
3273          */
3274         if (adapter->itr_setting & 0x3) {
3275                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3276                     !(adapter->flags & FLAG_HAS_ERT)) {
3277                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3278                                 dev_info(&adapter->pdev->dev,
3279                                         "Interrupt Throttle Rate turned off\n");
3280                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3281                                 ew32(ITR, 0);
3282                         }
3283                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3284                         dev_info(&adapter->pdev->dev,
3285                                  "Interrupt Throttle Rate turned on\n");
3286                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3287                         adapter->itr = 20000;
3288                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3289                 }
3290         }
3291
3292         /* Allow time for pending master requests to run */
3293         mac->ops.reset_hw(hw);
3294
3295         /*
3296          * For parts with AMT enabled, let the firmware know
3297          * that the network interface is in control
3298          */
3299         if (adapter->flags & FLAG_HAS_AMT)
3300                 e1000e_get_hw_control(adapter);
3301
3302         ew32(WUC, 0);
3303
3304         if (mac->ops.init_hw(hw))
3305                 e_err("Hardware Error\n");
3306
3307         e1000_update_mng_vlan(adapter);
3308
3309         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3310         ew32(VET, ETH_P_8021Q);
3311
3312         e1000e_reset_adaptive(hw);
3313
3314         if (!netif_running(adapter->netdev) &&
3315             !test_bit(__E1000_TESTING, &adapter->state)) {
3316                 e1000_power_down_phy(adapter);
3317                 return;
3318         }
3319
3320         e1000_get_phy_info(hw);
3321
3322         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3323             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3324                 u16 phy_data = 0;
3325                 /*
3326                  * speed up time to link by disabling smart power down, ignore
3327                  * the return value of this function because there is nothing
3328                  * different we would do if it failed
3329                  */
3330                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3331                 phy_data &= ~IGP02E1000_PM_SPD;
3332                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3333         }
3334 }
3335
3336 int e1000e_up(struct e1000_adapter *adapter)
3337 {
3338         struct e1000_hw *hw = &adapter->hw;
3339
3340         /* hardware has been reset, we need to reload some things */
3341         e1000_configure(adapter);
3342
3343         clear_bit(__E1000_DOWN, &adapter->state);
3344
3345         napi_enable(&adapter->napi);
3346         if (adapter->msix_entries)
3347                 e1000_configure_msix(adapter);
3348         e1000_irq_enable(adapter);
3349
3350         netif_wake_queue(adapter->netdev);
3351
3352         /* fire a link change interrupt to start the watchdog */
3353         if (adapter->msix_entries)
3354                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3355         else
3356                 ew32(ICS, E1000_ICS_LSC);
3357
3358         return 0;
3359 }
3360
3361 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3362 {
3363         struct e1000_hw *hw = &adapter->hw;
3364
3365         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3366                 return;
3367
3368         /* flush pending descriptor writebacks to memory */
3369         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3370         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3371
3372         /* execute the writes immediately */
3373         e1e_flush();
3374 }
3375
3376 static void e1000e_update_stats(struct e1000_adapter *adapter);
3377
3378 void e1000e_down(struct e1000_adapter *adapter)
3379 {
3380         struct net_device *netdev = adapter->netdev;
3381         struct e1000_hw *hw = &adapter->hw;
3382         u32 tctl, rctl;
3383
3384         /*
3385          * signal that we're down so the interrupt handler does not
3386          * reschedule our watchdog timer
3387          */
3388         set_bit(__E1000_DOWN, &adapter->state);
3389
3390         /* disable receives in the hardware */
3391         rctl = er32(RCTL);
3392         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3393         /* flush and sleep below */
3394
3395         netif_stop_queue(netdev);
3396
3397         /* disable transmits in the hardware */
3398         tctl = er32(TCTL);
3399         tctl &= ~E1000_TCTL_EN;
3400         ew32(TCTL, tctl);
3401         /* flush both disables and wait for them to finish */
3402         e1e_flush();
3403         usleep_range(10000, 20000);
3404
3405         napi_disable(&adapter->napi);
3406         e1000_irq_disable(adapter);
3407
3408         del_timer_sync(&adapter->watchdog_timer);
3409         del_timer_sync(&adapter->phy_info_timer);
3410
3411         netif_carrier_off(netdev);
3412
3413         spin_lock(&adapter->stats64_lock);
3414         e1000e_update_stats(adapter);
3415         spin_unlock(&adapter->stats64_lock);
3416
3417         adapter->link_speed = 0;
3418         adapter->link_duplex = 0;
3419
3420         if (!pci_channel_offline(adapter->pdev))
3421                 e1000e_reset(adapter);
3422
3423         e1000e_flush_descriptors(adapter);
3424
3425         e1000_clean_tx_ring(adapter);
3426         e1000_clean_rx_ring(adapter);
3427
3428         /*
3429          * TODO: for power management, we could drop the link and
3430          * pci_disable_device here.
3431          */
3432 }
3433
3434 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3435 {
3436         might_sleep();
3437         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3438                 usleep_range(1000, 2000);
3439         e1000e_down(adapter);
3440         e1000e_up(adapter);
3441         clear_bit(__E1000_RESETTING, &adapter->state);
3442 }
3443
3444 /**
3445  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3446  * @adapter: board private structure to initialize
3447  *
3448  * e1000_sw_init initializes the Adapter private data structure.
3449  * Fields are initialized based on PCI device information and
3450  * OS network device settings (MTU size).
3451  **/
3452 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3453 {
3454         struct net_device *netdev = adapter->netdev;
3455
3456         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3457         adapter->rx_ps_bsize0 = 128;
3458         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3459         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3460
3461         spin_lock_init(&adapter->stats64_lock);
3462
3463         e1000e_set_interrupt_capability(adapter);
3464
3465         if (e1000_alloc_queues(adapter))
3466                 return -ENOMEM;
3467
3468         /* Explicitly disable IRQ since the NIC can be in any state. */
3469         e1000_irq_disable(adapter);
3470
3471         set_bit(__E1000_DOWN, &adapter->state);
3472         return 0;
3473 }
3474
3475 /**
3476  * e1000_intr_msi_test - Interrupt Handler
3477  * @irq: interrupt number
3478  * @data: pointer to a network interface device structure
3479  **/
3480 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3481 {
3482         struct net_device *netdev = data;
3483         struct e1000_adapter *adapter = netdev_priv(netdev);
3484         struct e1000_hw *hw = &adapter->hw;
3485         u32 icr = er32(ICR);
3486
3487         e_dbg("icr is %08X\n", icr);
3488         if (icr & E1000_ICR_RXSEQ) {
3489                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3490                 wmb();
3491         }
3492
3493         return IRQ_HANDLED;
3494 }
3495
3496 /**
3497  * e1000_test_msi_interrupt - Returns 0 for successful test
3498  * @adapter: board private struct
3499  *
3500  * code flow taken from tg3.c
3501  **/
3502 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3503 {
3504         struct net_device *netdev = adapter->netdev;
3505         struct e1000_hw *hw = &adapter->hw;
3506         int err;
3507
3508         /* poll_enable hasn't been called yet, so don't need disable */
3509         /* clear any pending events */
3510         er32(ICR);
3511
3512         /* free the real vector and request a test handler */
3513         e1000_free_irq(adapter);
3514         e1000e_reset_interrupt_capability(adapter);
3515
3516         /* Assume that the test fails, if it succeeds then the test
3517          * MSI irq handler will unset this flag */
3518         adapter->flags |= FLAG_MSI_TEST_FAILED;
3519
3520         err = pci_enable_msi(adapter->pdev);
3521         if (err)
3522                 goto msi_test_failed;
3523
3524         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3525                           netdev->name, netdev);
3526         if (err) {
3527                 pci_disable_msi(adapter->pdev);
3528                 goto msi_test_failed;
3529         }
3530
3531         wmb();
3532
3533         e1000_irq_enable(adapter);
3534
3535         /* fire an unusual interrupt on the test handler */
3536         ew32(ICS, E1000_ICS_RXSEQ);
3537         e1e_flush();
3538         msleep(50);
3539
3540         e1000_irq_disable(adapter);
3541
3542         rmb();
3543
3544         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3545                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3546                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3547         } else
3548                 e_dbg("MSI interrupt test succeeded!\n");
3549
3550         free_irq(adapter->pdev->irq, netdev);
3551         pci_disable_msi(adapter->pdev);
3552
3553 msi_test_failed:
3554         e1000e_set_interrupt_capability(adapter);
3555         return e1000_request_irq(adapter);
3556 }
3557
3558 /**
3559  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3560  * @adapter: board private struct
3561  *
3562  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3563  **/
3564 static int e1000_test_msi(struct e1000_adapter *adapter)
3565 {
3566         int err;
3567         u16 pci_cmd;
3568
3569         if (!(adapter->flags & FLAG_MSI_ENABLED))
3570                 return 0;
3571
3572         /* disable SERR in case the MSI write causes a master abort */
3573         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3574         if (pci_cmd & PCI_COMMAND_SERR)
3575                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3576                                       pci_cmd & ~PCI_COMMAND_SERR);
3577
3578         err = e1000_test_msi_interrupt(adapter);
3579
3580         /* re-enable SERR */
3581         if (pci_cmd & PCI_COMMAND_SERR) {
3582                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3583                 pci_cmd |= PCI_COMMAND_SERR;
3584                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3585         }
3586
3587         return err;
3588 }
3589
3590 /**
3591  * e1000_open - Called when a network interface is made active
3592  * @netdev: network interface device structure
3593  *
3594  * Returns 0 on success, negative value on failure
3595  *
3596  * The open entry point is called when a network interface is made
3597  * active by the system (IFF_UP).  At this point all resources needed
3598  * for transmit and receive operations are allocated, the interrupt
3599  * handler is registered with the OS, the watchdog timer is started,
3600  * and the stack is notified that the interface is ready.
3601  **/
3602 static int e1000_open(struct net_device *netdev)
3603 {
3604         struct e1000_adapter *adapter = netdev_priv(netdev);
3605         struct e1000_hw *hw = &adapter->hw;
3606         struct pci_dev *pdev = adapter->pdev;
3607         int err;
3608
3609         /* disallow open during test */
3610         if (test_bit(__E1000_TESTING, &adapter->state))
3611                 return -EBUSY;
3612
3613         pm_runtime_get_sync(&pdev->dev);
3614
3615         netif_carrier_off(netdev);
3616
3617         /* allocate transmit descriptors */
3618         err = e1000e_setup_tx_resources(adapter);
3619         if (err)
3620                 goto err_setup_tx;
3621
3622         /* allocate receive descriptors */
3623         err = e1000e_setup_rx_resources(adapter);
3624         if (err)
3625                 goto err_setup_rx;
3626
3627         /*
3628          * If AMT is enabled, let the firmware know that the network
3629          * interface is now open and reset the part to a known state.
3630          */
3631         if (adapter->flags & FLAG_HAS_AMT) {
3632                 e1000e_get_hw_control(adapter);
3633                 e1000e_reset(adapter);
3634         }
3635
3636         e1000e_power_up_phy(adapter);
3637
3638         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3639         if ((adapter->hw.mng_cookie.status &
3640              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3641                 e1000_update_mng_vlan(adapter);
3642
3643         /* DMA latency requirement to workaround early-receive/jumbo issue */
3644         if ((adapter->flags & FLAG_HAS_ERT) ||
3645             (adapter->hw.mac.type == e1000_pch2lan))
3646                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3647                                    PM_QOS_CPU_DMA_LATENCY,
3648                                    PM_QOS_DEFAULT_VALUE);
3649
3650         /*
3651          * before we allocate an interrupt, we must be ready to handle it.
3652          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3653          * as soon as we call pci_request_irq, so we have to setup our
3654          * clean_rx handler before we do so.
3655          */
3656         e1000_configure(adapter);
3657
3658         err = e1000_request_irq(adapter);
3659         if (err)
3660                 goto err_req_irq;
3661
3662         /*
3663          * Work around PCIe errata with MSI interrupts causing some chipsets to
3664          * ignore e1000e MSI messages, which means we need to test our MSI
3665          * interrupt now
3666          */
3667         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3668                 err = e1000_test_msi(adapter);
3669                 if (err) {
3670                         e_err("Interrupt allocation failed\n");
3671                         goto err_req_irq;
3672                 }
3673         }
3674
3675         /* From here on the code is the same as e1000e_up() */
3676         clear_bit(__E1000_DOWN, &adapter->state);
3677
3678         napi_enable(&adapter->napi);
3679
3680         e1000_irq_enable(adapter);
3681
3682         netif_start_queue(netdev);
3683
3684         adapter->idle_check = true;
3685         pm_runtime_put(&pdev->dev);
3686
3687         /* fire a link status change interrupt to start the watchdog */
3688         if (adapter->msix_entries)
3689                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3690         else
3691                 ew32(ICS, E1000_ICS_LSC);
3692
3693         return 0;
3694
3695 err_req_irq:
3696         e1000e_release_hw_control(adapter);
3697         e1000_power_down_phy(adapter);
3698         e1000e_free_rx_resources(adapter);
3699 err_setup_rx:
3700         e1000e_free_tx_resources(adapter);
3701 err_setup_tx:
3702         e1000e_reset(adapter);
3703         pm_runtime_put_sync(&pdev->dev);
3704
3705         return err;
3706 }
3707
3708 /**
3709  * e1000_close - Disables a network interface
3710  * @netdev: network interface device structure
3711  *
3712  * Returns 0, this is not allowed to fail
3713  *
3714  * The close entry point is called when an interface is de-activated
3715  * by the OS.  The hardware is still under the drivers control, but
3716  * needs to be disabled.  A global MAC reset is issued to stop the
3717  * hardware, and all transmit and receive resources are freed.
3718  **/
3719 static int e1000_close(struct net_device *netdev)
3720 {
3721         struct e1000_adapter *adapter = netdev_priv(netdev);
3722         struct pci_dev *pdev = adapter->pdev;
3723
3724         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3725
3726         pm_runtime_get_sync(&pdev->dev);
3727
3728         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3729                 e1000e_down(adapter);
3730                 e1000_free_irq(adapter);
3731         }
3732         e1000_power_down_phy(adapter);
3733
3734         e1000e_free_tx_resources(adapter);
3735         e1000e_free_rx_resources(adapter);
3736
3737         /*
3738          * kill manageability vlan ID if supported, but not if a vlan with
3739          * the same ID is registered on the host OS (let 8021q kill it)
3740          */
3741         if (adapter->hw.mng_cookie.status &
3742             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3743                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3744
3745         /*
3746          * If AMT is enabled, let the firmware know that the network
3747          * interface is now closed
3748          */
3749         if ((adapter->flags & FLAG_HAS_AMT) &&
3750             !test_bit(__E1000_TESTING, &adapter->state))
3751                 e1000e_release_hw_control(adapter);
3752
3753         if ((adapter->flags & FLAG_HAS_ERT) ||
3754             (adapter->hw.mac.type == e1000_pch2lan))
3755                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3756
3757         pm_runtime_put_sync(&pdev->dev);
3758
3759         return 0;
3760 }
3761 /**
3762  * e1000_set_mac - Change the Ethernet Address of the NIC
3763  * @netdev: network interface device structure
3764  * @p: pointer to an address structure
3765  *
3766  * Returns 0 on success, negative on failure
3767  **/
3768 static int e1000_set_mac(struct net_device *netdev, void *p)
3769 {
3770         struct e1000_adapter *adapter = netdev_priv(netdev);
3771         struct sockaddr *addr = p;
3772
3773         if (!is_valid_ether_addr(addr->sa_data))
3774                 return -EADDRNOTAVAIL;
3775
3776         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3777         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3778
3779         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3780
3781         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3782                 /* activate the work around */
3783                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3784
3785                 /*
3786                  * Hold a copy of the LAA in RAR[14] This is done so that
3787                  * between the time RAR[0] gets clobbered  and the time it
3788                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3789                  * of the RARs and no incoming packets directed to this port
3790                  * are dropped. Eventually the LAA will be in RAR[0] and
3791                  * RAR[14]
3792                  */
3793                 e1000e_rar_set(&adapter->hw,
3794                               adapter->hw.mac.addr,
3795                               adapter->hw.mac.rar_entry_count - 1);
3796         }
3797
3798         return 0;
3799 }
3800
3801 /**
3802  * e1000e_update_phy_task - work thread to update phy
3803  * @work: pointer to our work struct
3804  *
3805  * this worker thread exists because we must acquire a
3806  * semaphore to read the phy, which we could msleep while
3807  * waiting for it, and we can't msleep in a timer.
3808  **/
3809 static void e1000e_update_phy_task(struct work_struct *work)
3810 {
3811         struct e1000_adapter *adapter = container_of(work,
3812                                         struct e1000_adapter, update_phy_task);
3813
3814         if (test_bit(__E1000_DOWN, &adapter->state))
3815                 return;
3816
3817         e1000_get_phy_info(&adapter->hw);
3818 }
3819
3820 /*
3821  * Need to wait a few seconds after link up to get diagnostic information from
3822  * the phy
3823  */
3824 static void e1000_update_phy_info(unsigned long data)
3825 {
3826         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3827
3828         if (test_bit(__E1000_DOWN, &adapter->state))
3829                 return;
3830
3831         schedule_work(&adapter->update_phy_task);
3832 }
3833
3834 /**
3835  * e1000e_update_phy_stats - Update the PHY statistics counters
3836  * @adapter: board private structure
3837  **/
3838 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3839 {
3840         struct e1000_hw *hw = &adapter->hw;
3841         s32 ret_val;
3842         u16 phy_data;
3843
3844         ret_val = hw->phy.ops.acquire(hw);
3845         if (ret_val)
3846                 return;
3847
3848         hw->phy.addr = 1;
3849
3850 #define HV_PHY_STATS_PAGE       778
3851         /*
3852          * A page set is expensive so check if already on desired page.
3853          * If not, set to the page with the PHY status registers.
3854          */
3855         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3856                                            &phy_data);
3857         if (ret_val)
3858                 goto release;
3859         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3860                 ret_val = e1000e_write_phy_reg_mdic(hw,
3861                                                     IGP01E1000_PHY_PAGE_SELECT,
3862                                                     (HV_PHY_STATS_PAGE <<
3863                                                      IGP_PAGE_SHIFT));
3864                 if (ret_val)
3865                         goto release;
3866         }
3867
3868         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3869
3870         /* Single Collision Count */
3871         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3872                                  &phy_data);
3873         ret_val = e1000e_read_phy_reg_mdic(hw,
3874                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3875                                            &phy_data);
3876         if (!ret_val)
3877                 adapter->stats.scc += phy_data;
3878
3879         /* Excessive Collision Count */
3880         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3881                                  &phy_data);
3882         ret_val = e1000e_read_phy_reg_mdic(hw,
3883                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3884                                            &phy_data);
3885         if (!ret_val)
3886                 adapter->stats.ecol += phy_data;
3887
3888         /* Multiple Collision Count */
3889         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3890                                  &phy_data);
3891         ret_val = e1000e_read_phy_reg_mdic(hw,
3892                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3893                                            &phy_data);
3894         if (!ret_val)
3895                 adapter->stats.mcc += phy_data;
3896
3897         /* Late Collision Count */
3898         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3899                                  &phy_data);
3900         ret_val = e1000e_read_phy_reg_mdic(hw,
3901                                            HV_LATECOL_LOWER &
3902                                            MAX_PHY_REG_ADDRESS,
3903                                            &phy_data);
3904         if (!ret_val)
3905                 adapter->stats.latecol += phy_data;
3906
3907         /* Collision Count - also used for adaptive IFS */
3908         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3909                                  &phy_data);
3910         ret_val = e1000e_read_phy_reg_mdic(hw,
3911                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3912                                            &phy_data);
3913         if (!ret_val)
3914                 hw->mac.collision_delta = phy_data;
3915
3916         /* Defer Count */
3917         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3918                                  &phy_data);
3919         ret_val = e1000e_read_phy_reg_mdic(hw,
3920                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3921                                            &phy_data);
3922         if (!ret_val)
3923                 adapter->stats.dc += phy_data;
3924
3925         /* Transmit with no CRS */
3926         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3927                                  &phy_data);
3928         ret_val = e1000e_read_phy_reg_mdic(hw,
3929                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3930                                            &phy_data);
3931         if (!ret_val)
3932                 adapter->stats.tncrs += phy_data;
3933
3934 release:
3935         hw->phy.ops.release(hw);
3936 }
3937
3938 /**
3939  * e1000e_update_stats - Update the board statistics counters
3940  * @adapter: board private structure
3941  **/
3942 static void e1000e_update_stats(struct e1000_adapter *adapter)
3943 {
3944         struct net_device *netdev = adapter->netdev;
3945         struct e1000_hw *hw = &adapter->hw;
3946         struct pci_dev *pdev = adapter->pdev;
3947
3948         /*
3949          * Prevent stats update while adapter is being reset, or if the pci
3950          * connection is down.
3951          */
3952         if (adapter->link_speed == 0)
3953                 return;
3954         if (pci_channel_offline(pdev))
3955                 return;
3956
3957         adapter->stats.crcerrs += er32(CRCERRS);
3958         adapter->stats.gprc += er32(GPRC);
3959         adapter->stats.gorc += er32(GORCL);
3960         er32(GORCH); /* Clear gorc */
3961         adapter->stats.bprc += er32(BPRC);
3962         adapter->stats.mprc += er32(MPRC);
3963         adapter->stats.roc += er32(ROC);
3964
3965         adapter->stats.mpc += er32(MPC);
3966
3967         /* Half-duplex statistics */
3968         if (adapter->link_duplex == HALF_DUPLEX) {
3969                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3970                         e1000e_update_phy_stats(adapter);
3971                 } else {
3972                         adapter->stats.scc += er32(SCC);
3973                         adapter->stats.ecol += er32(ECOL);
3974                         adapter->stats.mcc += er32(MCC);
3975                         adapter->stats.latecol += er32(LATECOL);
3976                         adapter->stats.dc += er32(DC);
3977
3978                         hw->mac.collision_delta = er32(COLC);
3979
3980                         if ((hw->mac.type != e1000_82574) &&
3981                             (hw->mac.type != e1000_82583))
3982                                 adapter->stats.tncrs += er32(TNCRS);
3983                 }
3984                 adapter->stats.colc += hw->mac.collision_delta;
3985         }
3986
3987         adapter->stats.xonrxc += er32(XONRXC);
3988         adapter->stats.xontxc += er32(XONTXC);
3989         adapter->stats.xoffrxc += er32(XOFFRXC);
3990         adapter->stats.xofftxc += er32(XOFFTXC);
3991         adapter->stats.gptc += er32(GPTC);
3992         adapter->stats.gotc += er32(GOTCL);
3993         er32(GOTCH); /* Clear gotc */
3994         adapter->stats.rnbc += er32(RNBC);
3995         adapter->stats.ruc += er32(RUC);
3996
3997         adapter->stats.mptc += er32(MPTC);
3998         adapter->stats.bptc += er32(BPTC);
3999
4000         /* used for adaptive IFS */
4001
4002         hw->mac.tx_packet_delta = er32(TPT);
4003         adapter->stats.tpt += hw->mac.tx_packet_delta;
4004
4005         adapter->stats.algnerrc += er32(ALGNERRC);
4006         adapter->stats.rxerrc += er32(RXERRC);
4007         adapter->stats.cexterr += er32(CEXTERR);
4008         adapter->stats.tsctc += er32(TSCTC);
4009         adapter->stats.tsctfc += er32(TSCTFC);
4010
4011         /* Fill out the OS statistics structure */
4012         netdev->stats.multicast = adapter->stats.mprc;
4013         netdev->stats.collisions = adapter->stats.colc;
4014
4015         /* Rx Errors */
4016
4017         /*
4018          * RLEC on some newer hardware can be incorrect so build
4019          * our own version based on RUC and ROC
4020          */
4021         netdev->stats.rx_errors = adapter->stats.rxerrc +
4022                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4023                 adapter->stats.ruc + adapter->stats.roc +
4024                 adapter->stats.cexterr;
4025         netdev->stats.rx_length_errors = adapter->stats.ruc +
4026                                               adapter->stats.roc;
4027         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4028         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4029         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4030
4031         /* Tx Errors */
4032         netdev->stats.tx_errors = adapter->stats.ecol +
4033                                        adapter->stats.latecol;
4034         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4035         netdev->stats.tx_window_errors = adapter->stats.latecol;
4036         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4037
4038         /* Tx Dropped needs to be maintained elsewhere */
4039
4040         /* Management Stats */
4041         adapter->stats.mgptc += er32(MGTPTC);
4042         adapter->stats.mgprc += er32(MGTPRC);
4043         adapter->stats.mgpdc += er32(MGTPDC);
4044 }
4045
4046 /**
4047  * e1000_phy_read_status - Update the PHY register status snapshot
4048  * @adapter: board private structure
4049  **/
4050 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4051 {
4052         struct e1000_hw *hw = &adapter->hw;
4053         struct e1000_phy_regs *phy = &adapter->phy_regs;
4054
4055         if ((er32(STATUS) & E1000_STATUS_LU) &&
4056             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4057                 int ret_val;
4058
4059                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4060                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4061                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4062                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4063                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4064                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4065                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4066                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4067                 if (ret_val)
4068                         e_warn("Error reading PHY register\n");
4069         } else {
4070                 /*
4071                  * Do not read PHY registers if link is not up
4072                  * Set values to typical power-on defaults
4073                  */
4074                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4075                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4076                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4077                              BMSR_ERCAP);
4078                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4079                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4080                 phy->lpa = 0;
4081                 phy->expansion = EXPANSION_ENABLENPAGE;
4082                 phy->ctrl1000 = ADVERTISE_1000FULL;
4083                 phy->stat1000 = 0;
4084                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4085         }
4086 }
4087
4088 static void e1000_print_link_info(struct e1000_adapter *adapter)
4089 {
4090         struct e1000_hw *hw = &adapter->hw;
4091         u32 ctrl = er32(CTRL);
4092
4093         /* Link status message must follow this format for user tools */
4094         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4095                "Flow Control: %s\n",
4096                adapter->netdev->name,
4097                adapter->link_speed,
4098                (adapter->link_duplex == FULL_DUPLEX) ?
4099                "Full Duplex" : "Half Duplex",
4100                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4101                "Rx/Tx" :
4102                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4103                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4104 }
4105
4106 static bool e1000e_has_link(struct e1000_adapter *adapter)
4107 {
4108         struct e1000_hw *hw = &adapter->hw;
4109         bool link_active = 0;
4110         s32 ret_val = 0;
4111
4112         /*
4113          * get_link_status is set on LSC (link status) interrupt or
4114          * Rx sequence error interrupt.  get_link_status will stay
4115          * false until the check_for_link establishes link
4116          * for copper adapters ONLY
4117          */
4118         switch (hw->phy.media_type) {
4119         case e1000_media_type_copper:
4120                 if (hw->mac.get_link_status) {
4121                         ret_val = hw->mac.ops.check_for_link(hw);
4122                         link_active = !hw->mac.get_link_status;
4123                 } else {
4124                         link_active = 1;
4125                 }
4126                 break;
4127         case e1000_media_type_fiber:
4128                 ret_val = hw->mac.ops.check_for_link(hw);
4129                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4130                 break;
4131         case e1000_media_type_internal_serdes:
4132                 ret_val = hw->mac.ops.check_for_link(hw);
4133                 link_active = adapter->hw.mac.serdes_has_link;
4134                 break;
4135         default:
4136         case e1000_media_type_unknown:
4137                 break;
4138         }
4139
4140         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4141             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4142                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4143                 e_info("Gigabit has been disabled, downgrading speed\n");
4144         }
4145
4146         return link_active;
4147 }
4148
4149 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4150 {
4151         /* make sure the receive unit is started */
4152         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4153             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4154                 struct e1000_hw *hw = &adapter->hw;
4155                 u32 rctl = er32(RCTL);
4156                 ew32(RCTL, rctl | E1000_RCTL_EN);
4157                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4158         }
4159 }
4160
4161 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4162 {
4163         struct e1000_hw *hw = &adapter->hw;
4164
4165         /*
4166          * With 82574 controllers, PHY needs to be checked periodically
4167          * for hung state and reset, if two calls return true
4168          */
4169         if (e1000_check_phy_82574(hw))
4170                 adapter->phy_hang_count++;
4171         else
4172                 adapter->phy_hang_count = 0;
4173
4174         if (adapter->phy_hang_count > 1) {
4175                 adapter->phy_hang_count = 0;
4176                 schedule_work(&adapter->reset_task);
4177         }
4178 }
4179
4180 /**
4181  * e1000_watchdog - Timer Call-back
4182  * @data: pointer to adapter cast into an unsigned long
4183  **/
4184 static void e1000_watchdog(unsigned long data)
4185 {
4186         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4187
4188         /* Do the rest outside of interrupt context */
4189         schedule_work(&adapter->watchdog_task);
4190
4191         /* TODO: make this use queue_delayed_work() */
4192 }
4193
4194 static void e1000_watchdog_task(struct work_struct *work)
4195 {
4196         struct e1000_adapter *adapter = container_of(work,
4197                                         struct e1000_adapter, watchdog_task);
4198         struct net_device *netdev = adapter->netdev;
4199         struct e1000_mac_info *mac = &adapter->hw.mac;
4200         struct e1000_phy_info *phy = &adapter->hw.phy;
4201         struct e1000_ring *tx_ring = adapter->tx_ring;
4202         struct e1000_hw *hw = &adapter->hw;
4203         u32 link, tctl;
4204
4205         if (test_bit(__E1000_DOWN, &adapter->state))
4206                 return;
4207
4208         link = e1000e_has_link(adapter);
4209         if ((netif_carrier_ok(netdev)) && link) {
4210                 /* Cancel scheduled suspend requests. */
4211                 pm_runtime_resume(netdev->dev.parent);
4212
4213                 e1000e_enable_receives(adapter);
4214                 goto link_up;
4215         }
4216
4217         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4218             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4219                 e1000_update_mng_vlan(adapter);
4220
4221         if (link) {
4222                 if (!netif_carrier_ok(netdev)) {
4223                         bool txb2b = 1;
4224
4225                         /* Cancel scheduled suspend requests. */
4226                         pm_runtime_resume(netdev->dev.parent);
4227
4228                         /* update snapshot of PHY registers on LSC */
4229                         e1000_phy_read_status(adapter);
4230                         mac->ops.get_link_up_info(&adapter->hw,
4231                                                    &adapter->link_speed,
4232                                                    &adapter->link_duplex);
4233                         e1000_print_link_info(adapter);
4234                         /*
4235                          * On supported PHYs, check for duplex mismatch only
4236                          * if link has autonegotiated at 10/100 half
4237                          */
4238                         if ((hw->phy.type == e1000_phy_igp_3 ||
4239                              hw->phy.type == e1000_phy_bm) &&
4240                             (hw->mac.autoneg == true) &&
4241                             (adapter->link_speed == SPEED_10 ||
4242                              adapter->link_speed == SPEED_100) &&
4243                             (adapter->link_duplex == HALF_DUPLEX)) {
4244                                 u16 autoneg_exp;
4245
4246                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4247
4248                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4249                                         e_info("Autonegotiated half duplex but"
4250                                                " link partner cannot autoneg. "
4251                                                " Try forcing full duplex if "
4252                                                "link gets many collisions.\n");
4253                         }
4254
4255                         /* adjust timeout factor according to speed/duplex */
4256                         adapter->tx_timeout_factor = 1;
4257                         switch (adapter->link_speed) {
4258                         case SPEED_10:
4259                                 txb2b = 0;
4260                                 adapter->tx_timeout_factor = 16;
4261                                 break;
4262                         case SPEED_100:
4263                                 txb2b = 0;
4264                                 adapter->tx_timeout_factor = 10;
4265                                 break;
4266                         }
4267
4268                         /*
4269                          * workaround: re-program speed mode bit after
4270                          * link-up event
4271                          */
4272                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4273                             !txb2b) {
4274                                 u32 tarc0;
4275                                 tarc0 = er32(TARC(0));
4276                                 tarc0 &= ~SPEED_MODE_BIT;
4277                                 ew32(TARC(0), tarc0);
4278                         }
4279
4280                         /*
4281                          * disable TSO for pcie and 10/100 speeds, to avoid
4282                          * some hardware issues
4283                          */
4284                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4285                                 switch (adapter->link_speed) {
4286                                 case SPEED_10:
4287                                 case SPEED_100:
4288                                         e_info("10/100 speed: disabling TSO\n");
4289                                         netdev->features &= ~NETIF_F_TSO;
4290                                         netdev->features &= ~NETIF_F_TSO6;
4291                                         break;
4292                                 case SPEED_1000:
4293                                         netdev->features |= NETIF_F_TSO;
4294                                         netdev->features |= NETIF_F_TSO6;
4295                                         break;
4296                                 default:
4297                                         /* oops */
4298                                         break;
4299                                 }
4300                         }
4301
4302                         /*
4303                          * enable transmits in the hardware, need to do this
4304                          * after setting TARC(0)
4305                          */
4306                         tctl = er32(TCTL);
4307                         tctl |= E1000_TCTL_EN;
4308                         ew32(TCTL, tctl);
4309
4310                         /*
4311                          * Perform any post-link-up configuration before
4312                          * reporting link up.
4313                          */
4314                         if (phy->ops.cfg_on_link_up)
4315                                 phy->ops.cfg_on_link_up(hw);
4316
4317                         netif_carrier_on(netdev);
4318
4319                         if (!test_bit(__E1000_DOWN, &adapter->state))
4320                                 mod_timer(&adapter->phy_info_timer,
4321                                           round_jiffies(jiffies + 2 * HZ));
4322                 }
4323         } else {
4324                 if (netif_carrier_ok(netdev)) {
4325                         adapter->link_speed = 0;
4326                         adapter->link_duplex = 0;
4327                         /* Link status message must follow this format */
4328                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4329                                adapter->netdev->name);
4330                         netif_carrier_off(netdev);
4331                         if (!test_bit(__E1000_DOWN, &adapter->state))
4332                                 mod_timer(&adapter->phy_info_timer,
4333                                           round_jiffies(jiffies + 2 * HZ));
4334
4335                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4336                                 schedule_work(&adapter->reset_task);
4337                         else
4338                                 pm_schedule_suspend(netdev->dev.parent,
4339                                                         LINK_TIMEOUT);
4340                 }
4341         }
4342
4343 link_up:
4344         spin_lock(&adapter->stats64_lock);
4345         e1000e_update_stats(adapter);
4346
4347         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4348         adapter->tpt_old = adapter->stats.tpt;
4349         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4350         adapter->colc_old = adapter->stats.colc;
4351
4352         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4353         adapter->gorc_old = adapter->stats.gorc;
4354         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4355         adapter->gotc_old = adapter->stats.gotc;
4356         spin_unlock(&adapter->stats64_lock);
4357
4358         e1000e_update_adaptive(&adapter->hw);
4359
4360         if (!netif_carrier_ok(netdev) &&
4361             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4362                 /*
4363                  * We've lost link, so the controller stops DMA,
4364                  * but we've got queued Tx work that's never going
4365                  * to get done, so reset controller to flush Tx.
4366                  * (Do the reset outside of interrupt context).
4367                  */
4368                 schedule_work(&adapter->reset_task);
4369                 /* return immediately since reset is imminent */
4370                 return;
4371         }
4372
4373         /* Simple mode for Interrupt Throttle Rate (ITR) */
4374         if (adapter->itr_setting == 4) {
4375                 /*
4376                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4377                  * Total asymmetrical Tx or Rx gets ITR=8000;
4378                  * everyone else is between 2000-8000.
4379                  */
4380                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4381                 u32 dif = (adapter->gotc > adapter->gorc ?
4382                             adapter->gotc - adapter->gorc :
4383                             adapter->gorc - adapter->gotc) / 10000;
4384                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4385
4386                 ew32(ITR, 1000000000 / (itr * 256));
4387         }
4388
4389         /* Cause software interrupt to ensure Rx ring is cleaned */
4390         if (adapter->msix_entries)
4391                 ew32(ICS, adapter->rx_ring->ims_val);
4392         else
4393                 ew32(ICS, E1000_ICS_RXDMT0);
4394
4395         /* flush pending descriptors to memory before detecting Tx hang */
4396         e1000e_flush_descriptors(adapter);
4397
4398         /* Force detection of hung controller every watchdog period */
4399         adapter->detect_tx_hung = 1;
4400
4401         /*
4402          * With 82571 controllers, LAA may be overwritten due to controller
4403          * reset from the other port. Set the appropriate LAA in RAR[0]
4404          */
4405         if (e1000e_get_laa_state_82571(hw))
4406                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4407
4408         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4409                 e1000e_check_82574_phy_workaround(adapter);
4410
4411         /* Reset the timer */
4412         if (!test_bit(__E1000_DOWN, &adapter->state))
4413                 mod_timer(&adapter->watchdog_timer,
4414                           round_jiffies(jiffies + 2 * HZ));
4415 }
4416
4417 #define E1000_TX_FLAGS_CSUM             0x00000001
4418 #define E1000_TX_FLAGS_VLAN             0x00000002
4419 #define E1000_TX_FLAGS_TSO              0x00000004
4420 #define E1000_TX_FLAGS_IPV4             0x00000008
4421 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4422 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4423
4424 static int e1000_tso(struct e1000_adapter *adapter,
4425                      struct sk_buff *skb)
4426 {
4427         struct e1000_ring *tx_ring = adapter->tx_ring;
4428         struct e1000_context_desc *context_desc;
4429         struct e1000_buffer *buffer_info;
4430         unsigned int i;
4431         u32 cmd_length = 0;
4432         u16 ipcse = 0, tucse, mss;
4433         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4434
4435         if (!skb_is_gso(skb))
4436                 return 0;
4437
4438         if (skb_header_cloned(skb)) {
4439                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4440
4441                 if (err)
4442                         return err;
4443         }
4444
4445         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4446         mss = skb_shinfo(skb)->gso_size;
4447         if (skb->protocol == htons(ETH_P_IP)) {
4448                 struct iphdr *iph = ip_hdr(skb);
4449                 iph->tot_len = 0;
4450                 iph->check = 0;
4451                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4452                                                          0, IPPROTO_TCP, 0);
4453                 cmd_length = E1000_TXD_CMD_IP;
4454                 ipcse = skb_transport_offset(skb) - 1;
4455         } else if (skb_is_gso_v6(skb)) {
4456                 ipv6_hdr(skb)->payload_len = 0;
4457                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4458                                                        &ipv6_hdr(skb)->daddr,
4459                                                        0, IPPROTO_TCP, 0);
4460                 ipcse = 0;
4461         }
4462         ipcss = skb_network_offset(skb);
4463         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4464         tucss = skb_transport_offset(skb);
4465         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4466         tucse = 0;
4467
4468         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4469                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4470
4471         i = tx_ring->next_to_use;
4472         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4473         buffer_info = &tx_ring->buffer_info[i];
4474
4475         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4476         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4477         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4478         context_desc->upper_setup.tcp_fields.tucss = tucss;
4479         context_desc->upper_setup.tcp_fields.tucso = tucso;
4480         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4481         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4482         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4483         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4484
4485         buffer_info->time_stamp = jiffies;
4486         buffer_info->next_to_watch = i;
4487
4488         i++;
4489         if (i == tx_ring->count)
4490                 i = 0;
4491         tx_ring->next_to_use = i;
4492
4493         return 1;
4494 }
4495
4496 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4497 {
4498         struct e1000_ring *tx_ring = adapter->tx_ring;
4499         struct e1000_context_desc *context_desc;
4500         struct e1000_buffer *buffer_info;
4501         unsigned int i;
4502         u8 css;
4503         u32 cmd_len = E1000_TXD_CMD_DEXT;
4504         __be16 protocol;
4505
4506         if (skb->ip_summed != CHECKSUM_PARTIAL)
4507                 return 0;
4508
4509         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4510                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4511         else
4512                 protocol = skb->protocol;
4513
4514         switch (protocol) {
4515         case cpu_to_be16(ETH_P_IP):
4516                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4517                         cmd_len |= E1000_TXD_CMD_TCP;
4518                 break;
4519         case cpu_to_be16(ETH_P_IPV6):
4520                 /* XXX not handling all IPV6 headers */
4521                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4522                         cmd_len |= E1000_TXD_CMD_TCP;
4523                 break;
4524         default:
4525                 if (unlikely(net_ratelimit()))
4526                         e_warn("checksum_partial proto=%x!\n",
4527                                be16_to_cpu(protocol));
4528                 break;
4529         }
4530
4531         css = skb_checksum_start_offset(skb);
4532
4533         i = tx_ring->next_to_use;
4534         buffer_info = &tx_ring->buffer_info[i];
4535         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4536
4537         context_desc->lower_setup.ip_config = 0;
4538         context_desc->upper_setup.tcp_fields.tucss = css;
4539         context_desc->upper_setup.tcp_fields.tucso =
4540                                 css + skb->csum_offset;
4541         context_desc->upper_setup.tcp_fields.tucse = 0;
4542         context_desc->tcp_seg_setup.data = 0;
4543         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4544
4545         buffer_info->time_stamp = jiffies;
4546         buffer_info->next_to_watch = i;
4547
4548         i++;
4549         if (i == tx_ring->count)
4550                 i = 0;
4551         tx_ring->next_to_use = i;
4552
4553         return 1;
4554 }
4555
4556 #define E1000_MAX_PER_TXD       8192
4557 #define E1000_MAX_TXD_PWR       12
4558
4559 static int e1000_tx_map(struct e1000_adapter *adapter,
4560                         struct sk_buff *skb, unsigned int first,
4561                         unsigned int max_per_txd, unsigned int nr_frags,
4562                         unsigned int mss)
4563 {
4564         struct e1000_ring *tx_ring = adapter->tx_ring;
4565         struct pci_dev *pdev = adapter->pdev;
4566         struct e1000_buffer *buffer_info;
4567         unsigned int len = skb_headlen(skb);
4568         unsigned int offset = 0, size, count = 0, i;
4569         unsigned int f, bytecount, segs;
4570
4571         i = tx_ring->next_to_use;
4572
4573         while (len) {
4574                 buffer_info = &tx_ring->buffer_info[i];
4575                 size = min(len, max_per_txd);
4576
4577                 buffer_info->length = size;
4578                 buffer_info->time_stamp = jiffies;
4579                 buffer_info->next_to_watch = i;
4580                 buffer_info->dma = dma_map_single(&pdev->dev,
4581                                                   skb->data + offset,
4582                                                   size, DMA_TO_DEVICE);
4583                 buffer_info->mapped_as_page = false;
4584                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4585                         goto dma_error;
4586
4587                 len -= size;
4588                 offset += size;
4589                 count++;
4590
4591                 if (len) {
4592                         i++;
4593                         if (i == tx_ring->count)
4594                                 i = 0;
4595                 }
4596         }
4597
4598         for (f = 0; f < nr_frags; f++) {
4599                 struct skb_frag_struct *frag;
4600
4601                 frag = &skb_shinfo(skb)->frags[f];
4602                 len = frag->size;
4603                 offset = frag->page_offset;
4604
4605                 while (len) {
4606                         i++;
4607                         if (i == tx_ring->count)
4608                                 i = 0;
4609
4610                         buffer_info = &tx_ring->buffer_info[i];
4611                         size = min(len, max_per_txd);
4612
4613                         buffer_info->length = size;
4614                         buffer_info->time_stamp = jiffies;
4615                         buffer_info->next_to_watch = i;
4616                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4617                                                         offset, size,
4618                                                         DMA_TO_DEVICE);
4619                         buffer_info->mapped_as_page = true;
4620                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4621                                 goto dma_error;
4622
4623                         len -= size;
4624                         offset += size;
4625                         count++;
4626                 }
4627         }
4628
4629         segs = skb_shinfo(skb)->gso_segs ? : 1;
4630         /* multiply data chunks by size of headers */
4631         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4632
4633         tx_ring->buffer_info[i].skb = skb;
4634         tx_ring->buffer_info[i].segs = segs;
4635         tx_ring->buffer_info[i].bytecount = bytecount;
4636         tx_ring->buffer_info[first].next_to_watch = i;
4637
4638         return count;
4639
4640 dma_error:
4641         dev_err(&pdev->dev, "Tx DMA map failed\n");
4642         buffer_info->dma = 0;
4643         if (count)
4644                 count--;
4645
4646         while (count--) {
4647                 if (i == 0)
4648                         i += tx_ring->count;
4649                 i--;
4650                 buffer_info = &tx_ring->buffer_info[i];
4651                 e1000_put_txbuf(adapter, buffer_info);
4652         }
4653
4654         return 0;
4655 }
4656
4657 static void e1000_tx_queue(struct e1000_adapter *adapter,
4658                            int tx_flags, int count)
4659 {
4660         struct e1000_ring *tx_ring = adapter->tx_ring;
4661         struct e1000_tx_desc *tx_desc = NULL;
4662         struct e1000_buffer *buffer_info;
4663         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4664         unsigned int i;
4665
4666         if (tx_flags & E1000_TX_FLAGS_TSO) {
4667                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4668                              E1000_TXD_CMD_TSE;
4669                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4670
4671                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4672                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4673         }
4674
4675         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4676                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4677                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4678         }
4679
4680         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4681                 txd_lower |= E1000_TXD_CMD_VLE;
4682                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4683         }
4684
4685         i = tx_ring->next_to_use;
4686
4687         do {
4688                 buffer_info = &tx_ring->buffer_info[i];
4689                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4690                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4691                 tx_desc->lower.data =
4692                         cpu_to_le32(txd_lower | buffer_info->length);
4693                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4694
4695                 i++;
4696                 if (i == tx_ring->count)
4697                         i = 0;
4698         } while (--count > 0);
4699
4700         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4701
4702         /*
4703          * Force memory writes to complete before letting h/w
4704          * know there are new descriptors to fetch.  (Only
4705          * applicable for weak-ordered memory model archs,
4706          * such as IA-64).
4707          */
4708         wmb();
4709
4710         tx_ring->next_to_use = i;
4711         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4712         /*
4713          * we need this if more than one processor can write to our tail
4714          * at a time, it synchronizes IO on IA64/Altix systems
4715          */
4716         mmiowb();
4717 }
4718
4719 #define MINIMUM_DHCP_PACKET_SIZE 282
4720 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4721                                     struct sk_buff *skb)
4722 {
4723         struct e1000_hw *hw =  &adapter->hw;
4724         u16 length, offset;
4725
4726         if (vlan_tx_tag_present(skb)) {
4727                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4728                     (adapter->hw.mng_cookie.status &
4729                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4730                         return 0;
4731         }
4732
4733         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4734                 return 0;
4735
4736         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4737                 return 0;
4738
4739         {
4740                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4741                 struct udphdr *udp;
4742
4743                 if (ip->protocol != IPPROTO_UDP)
4744                         return 0;
4745
4746                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4747                 if (ntohs(udp->dest) != 67)
4748                         return 0;
4749
4750                 offset = (u8 *)udp + 8 - skb->data;
4751                 length = skb->len - offset;
4752                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4753         }
4754
4755         return 0;
4756 }
4757
4758 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4759 {
4760         struct e1000_adapter *adapter = netdev_priv(netdev);
4761
4762         netif_stop_queue(netdev);
4763         /*
4764          * Herbert's original patch had:
4765          *  smp_mb__after_netif_stop_queue();
4766          * but since that doesn't exist yet, just open code it.
4767          */
4768         smp_mb();
4769
4770         /*
4771          * We need to check again in a case another CPU has just
4772          * made room available.
4773          */
4774         if (e1000_desc_unused(adapter->tx_ring) < size)
4775                 return -EBUSY;
4776
4777         /* A reprieve! */
4778         netif_start_queue(netdev);
4779         ++adapter->restart_queue;
4780         return 0;
4781 }
4782
4783 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4784 {
4785         struct e1000_adapter *adapter = netdev_priv(netdev);
4786
4787         if (e1000_desc_unused(adapter->tx_ring) >= size)
4788                 return 0;
4789         return __e1000_maybe_stop_tx(netdev, size);
4790 }
4791
4792 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4793 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4794                                     struct net_device *netdev)
4795 {
4796         struct e1000_adapter *adapter = netdev_priv(netdev);
4797         struct e1000_ring *tx_ring = adapter->tx_ring;
4798         unsigned int first;
4799         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4800         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4801         unsigned int tx_flags = 0;
4802         unsigned int len = skb_headlen(skb);
4803         unsigned int nr_frags;
4804         unsigned int mss;
4805         int count = 0;
4806         int tso;
4807         unsigned int f;
4808
4809         if (test_bit(__E1000_DOWN, &adapter->state)) {
4810                 dev_kfree_skb_any(skb);
4811                 return NETDEV_TX_OK;
4812         }
4813
4814         if (skb->len <= 0) {
4815                 dev_kfree_skb_any(skb);
4816                 return NETDEV_TX_OK;
4817         }
4818
4819         mss = skb_shinfo(skb)->gso_size;
4820         /*
4821          * The controller does a simple calculation to
4822          * make sure there is enough room in the FIFO before
4823          * initiating the DMA for each buffer.  The calc is:
4824          * 4 = ceil(buffer len/mss).  To make sure we don't
4825          * overrun the FIFO, adjust the max buffer len if mss
4826          * drops.
4827          */
4828         if (mss) {
4829                 u8 hdr_len;
4830                 max_per_txd = min(mss << 2, max_per_txd);
4831                 max_txd_pwr = fls(max_per_txd) - 1;
4832
4833                 /*
4834                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4835                  * points to just header, pull a few bytes of payload from
4836                  * frags into skb->data
4837                  */
4838                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4839                 /*
4840                  * we do this workaround for ES2LAN, but it is un-necessary,
4841                  * avoiding it could save a lot of cycles
4842                  */
4843                 if (skb->data_len && (hdr_len == len)) {
4844                         unsigned int pull_size;
4845
4846                         pull_size = min((unsigned int)4, skb->data_len);
4847                         if (!__pskb_pull_tail(skb, pull_size)) {
4848                                 e_err("__pskb_pull_tail failed.\n");
4849                                 dev_kfree_skb_any(skb);
4850                                 return NETDEV_TX_OK;
4851                         }
4852                         len = skb_headlen(skb);
4853                 }
4854         }
4855
4856         /* reserve a descriptor for the offload context */
4857         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4858                 count++;
4859         count++;
4860
4861         count += TXD_USE_COUNT(len, max_txd_pwr);
4862
4863         nr_frags = skb_shinfo(skb)->nr_frags;
4864         for (f = 0; f < nr_frags; f++)
4865                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4866                                        max_txd_pwr);
4867
4868         if (adapter->hw.mac.tx_pkt_filtering)
4869                 e1000_transfer_dhcp_info(adapter, skb);
4870
4871         /*
4872          * need: count + 2 desc gap to keep tail from touching
4873          * head, otherwise try next time
4874          */
4875         if (e1000_maybe_stop_tx(netdev, count + 2))
4876                 return NETDEV_TX_BUSY;
4877
4878         if (vlan_tx_tag_present(skb)) {
4879                 tx_flags |= E1000_TX_FLAGS_VLAN;
4880                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4881         }
4882
4883         first = tx_ring->next_to_use;
4884
4885         tso = e1000_tso(adapter, skb);
4886         if (tso < 0) {
4887                 dev_kfree_skb_any(skb);
4888                 return NETDEV_TX_OK;
4889         }
4890
4891         if (tso)
4892                 tx_flags |= E1000_TX_FLAGS_TSO;
4893         else if (e1000_tx_csum(adapter, skb))
4894                 tx_flags |= E1000_TX_FLAGS_CSUM;
4895
4896         /*
4897          * Old method was to assume IPv4 packet by default if TSO was enabled.
4898          * 82571 hardware supports TSO capabilities for IPv6 as well...
4899          * no longer assume, we must.
4900          */
4901         if (skb->protocol == htons(ETH_P_IP))
4902                 tx_flags |= E1000_TX_FLAGS_IPV4;
4903
4904         /* if count is 0 then mapping error has occurred */
4905         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4906         if (count) {
4907                 e1000_tx_queue(adapter, tx_flags, count);
4908                 /* Make sure there is space in the ring for the next send. */
4909                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4910
4911         } else {
4912                 dev_kfree_skb_any(skb);
4913                 tx_ring->buffer_info[first].time_stamp = 0;
4914                 tx_ring->next_to_use = first;
4915         }
4916
4917         return NETDEV_TX_OK;
4918 }
4919
4920 /**
4921  * e1000_tx_timeout - Respond to a Tx Hang
4922  * @netdev: network interface device structure
4923  **/
4924 static void e1000_tx_timeout(struct net_device *netdev)
4925 {
4926         struct e1000_adapter *adapter = netdev_priv(netdev);
4927
4928         /* Do the reset outside of interrupt context */
4929         adapter->tx_timeout_count++;
4930         schedule_work(&adapter->reset_task);
4931 }
4932
4933 static void e1000_reset_task(struct work_struct *work)
4934 {
4935         struct e1000_adapter *adapter;
4936         adapter = container_of(work, struct e1000_adapter, reset_task);
4937
4938         /* don't run the task if already down */
4939         if (test_bit(__E1000_DOWN, &adapter->state))
4940                 return;
4941
4942         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4943               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4944                 e1000e_dump(adapter);
4945                 e_err("Reset adapter\n");
4946         }
4947         e1000e_reinit_locked(adapter);
4948 }
4949
4950 /**
4951  * e1000_get_stats64 - Get System Network Statistics
4952  * @netdev: network interface device structure
4953  * @stats: rtnl_link_stats64 pointer
4954  *
4955  * Returns the address of the device statistics structure.
4956  **/
4957 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4958                                              struct rtnl_link_stats64 *stats)
4959 {
4960         struct e1000_adapter *adapter = netdev_priv(netdev);
4961
4962         memset(stats, 0, sizeof(struct rtnl_link_stats64));
4963         spin_lock(&adapter->stats64_lock);
4964         e1000e_update_stats(adapter);
4965         /* Fill out the OS statistics structure */
4966         stats->rx_bytes = adapter->stats.gorc;
4967         stats->rx_packets = adapter->stats.gprc;
4968         stats->tx_bytes = adapter->stats.gotc;
4969         stats->tx_packets = adapter->stats.gptc;
4970         stats->multicast = adapter->stats.mprc;
4971         stats->collisions = adapter->stats.colc;
4972
4973         /* Rx Errors */
4974
4975         /*
4976          * RLEC on some newer hardware can be incorrect so build
4977          * our own version based on RUC and ROC
4978          */
4979         stats->rx_errors = adapter->stats.rxerrc +
4980                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4981                 adapter->stats.ruc + adapter->stats.roc +
4982                 adapter->stats.cexterr;
4983         stats->rx_length_errors = adapter->stats.ruc +
4984                                               adapter->stats.roc;
4985         stats->rx_crc_errors = adapter->stats.crcerrs;
4986         stats->rx_frame_errors = adapter->stats.algnerrc;
4987         stats->rx_missed_errors = adapter->stats.mpc;
4988
4989         /* Tx Errors */
4990         stats->tx_errors = adapter->stats.ecol +
4991                                        adapter->stats.latecol;
4992         stats->tx_aborted_errors = adapter->stats.ecol;
4993         stats->tx_window_errors = adapter->stats.latecol;
4994         stats->tx_carrier_errors = adapter->stats.tncrs;
4995
4996         /* Tx Dropped needs to be maintained elsewhere */
4997
4998         spin_unlock(&adapter->stats64_lock);
4999         return stats;
5000 }
5001
5002 /**
5003  * e1000_change_mtu - Change the Maximum Transfer Unit
5004  * @netdev: network interface device structure
5005  * @new_mtu: new value for maximum frame size
5006  *
5007  * Returns 0 on success, negative on failure
5008  **/
5009 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5010 {
5011         struct e1000_adapter *adapter = netdev_priv(netdev);
5012         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5013
5014         /* Jumbo frame support */
5015         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5016             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5017                 e_err("Jumbo Frames not supported.\n");
5018                 return -EINVAL;
5019         }
5020
5021         /* Supported frame sizes */
5022         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5023             (max_frame > adapter->max_hw_frame_size)) {
5024                 e_err("Unsupported MTU setting\n");
5025                 return -EINVAL;
5026         }
5027
5028         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5029         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5030             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5031             (new_mtu > ETH_DATA_LEN)) {
5032                 e_err("Jumbo Frames not supported on 82579 when CRC "
5033                       "stripping is disabled.\n");
5034                 return -EINVAL;
5035         }
5036
5037         /* 82573 Errata 17 */
5038         if (((adapter->hw.mac.type == e1000_82573) ||
5039              (adapter->hw.mac.type == e1000_82574)) &&
5040             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5041                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5042                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5043         }
5044
5045         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5046                 usleep_range(1000, 2000);
5047         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5048         adapter->max_frame_size = max_frame;
5049         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5050         netdev->mtu = new_mtu;
5051         if (netif_running(netdev))
5052                 e1000e_down(adapter);
5053
5054         /*
5055          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5056          * means we reserve 2 more, this pushes us to allocate from the next
5057          * larger slab size.
5058          * i.e. RXBUFFER_2048 --> size-4096 slab
5059          * However with the new *_jumbo_rx* routines, jumbo receives will use
5060          * fragmented skbs
5061          */
5062
5063         if (max_frame <= 2048)
5064                 adapter->rx_buffer_len = 2048;
5065         else
5066                 adapter->rx_buffer_len = 4096;
5067
5068         /* adjust allocation if LPE protects us, and we aren't using SBP */
5069         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5070              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5071                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5072                                          + ETH_FCS_LEN;
5073
5074         if (netif_running(netdev))
5075                 e1000e_up(adapter);
5076         else
5077                 e1000e_reset(adapter);
5078
5079         clear_bit(__E1000_RESETTING, &adapter->state);
5080
5081         return 0;
5082 }
5083
5084 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5085                            int cmd)
5086 {
5087         struct e1000_adapter *adapter = netdev_priv(netdev);
5088         struct mii_ioctl_data *data = if_mii(ifr);
5089
5090         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5091                 return -EOPNOTSUPP;
5092
5093         switch (cmd) {
5094         case SIOCGMIIPHY:
5095                 data->phy_id = adapter->hw.phy.addr;
5096                 break;
5097         case SIOCGMIIREG:
5098                 e1000_phy_read_status(adapter);
5099
5100                 switch (data->reg_num & 0x1F) {
5101                 case MII_BMCR:
5102                         data->val_out = adapter->phy_regs.bmcr;
5103                         break;
5104                 case MII_BMSR:
5105                         data->val_out = adapter->phy_regs.bmsr;
5106                         break;
5107                 case MII_PHYSID1:
5108                         data->val_out = (adapter->hw.phy.id >> 16);
5109                         break;
5110                 case MII_PHYSID2:
5111                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5112                         break;
5113                 case MII_ADVERTISE:
5114                         data->val_out = adapter->phy_regs.advertise;
5115                         break;
5116                 case MII_LPA:
5117                         data->val_out = adapter->phy_regs.lpa;
5118                         break;
5119                 case MII_EXPANSION:
5120                         data->val_out = adapter->phy_regs.expansion;
5121                         break;
5122                 case MII_CTRL1000:
5123                         data->val_out = adapter->phy_regs.ctrl1000;
5124                         break;
5125                 case MII_STAT1000:
5126                         data->val_out = adapter->phy_regs.stat1000;
5127                         break;
5128                 case MII_ESTATUS:
5129                         data->val_out = adapter->phy_regs.estatus;
5130                         break;
5131                 default:
5132                         return -EIO;
5133                 }
5134                 break;
5135         case SIOCSMIIREG:
5136         default:
5137                 return -EOPNOTSUPP;
5138         }
5139         return 0;
5140 }
5141
5142 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5143 {
5144         switch (cmd) {
5145         case SIOCGMIIPHY:
5146         case SIOCGMIIREG:
5147         case SIOCSMIIREG:
5148                 return e1000_mii_ioctl(netdev, ifr, cmd);
5149         default:
5150                 return -EOPNOTSUPP;
5151         }
5152 }
5153
5154 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5155 {
5156         struct e1000_hw *hw = &adapter->hw;
5157         u32 i, mac_reg;
5158         u16 phy_reg;
5159         int retval = 0;
5160
5161         /* copy MAC RARs to PHY RARs */
5162         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5163
5164         /* copy MAC MTA to PHY MTA */
5165         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5166                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5167                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5168                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5169         }
5170
5171         /* configure PHY Rx Control register */
5172         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5173         mac_reg = er32(RCTL);
5174         if (mac_reg & E1000_RCTL_UPE)
5175                 phy_reg |= BM_RCTL_UPE;
5176         if (mac_reg & E1000_RCTL_MPE)
5177                 phy_reg |= BM_RCTL_MPE;
5178         phy_reg &= ~(BM_RCTL_MO_MASK);
5179         if (mac_reg & E1000_RCTL_MO_3)
5180                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5181                                 << BM_RCTL_MO_SHIFT);
5182         if (mac_reg & E1000_RCTL_BAM)
5183                 phy_reg |= BM_RCTL_BAM;
5184         if (mac_reg & E1000_RCTL_PMCF)
5185                 phy_reg |= BM_RCTL_PMCF;
5186         mac_reg = er32(CTRL);
5187         if (mac_reg & E1000_CTRL_RFCE)
5188                 phy_reg |= BM_RCTL_RFCE;
5189         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5190
5191         /* enable PHY wakeup in MAC register */
5192         ew32(WUFC, wufc);
5193         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5194
5195         /* configure and enable PHY wakeup in PHY registers */
5196         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5197         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5198
5199         /* activate PHY wakeup */
5200         retval = hw->phy.ops.acquire(hw);
5201         if (retval) {
5202                 e_err("Could not acquire PHY\n");
5203                 return retval;
5204         }
5205         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5206                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5207         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5208         if (retval) {
5209                 e_err("Could not read PHY page 769\n");
5210                 goto out;
5211         }
5212         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5213         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5214         if (retval)
5215                 e_err("Could not set PHY Host Wakeup bit\n");
5216 out:
5217         hw->phy.ops.release(hw);
5218
5219         return retval;
5220 }
5221
5222 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5223                             bool runtime)
5224 {
5225         struct net_device *netdev = pci_get_drvdata(pdev);
5226         struct e1000_adapter *adapter = netdev_priv(netdev);
5227         struct e1000_hw *hw = &adapter->hw;
5228         u32 ctrl, ctrl_ext, rctl, status;
5229         /* Runtime suspend should only enable wakeup for link changes */
5230         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5231         int retval = 0;
5232
5233         netif_device_detach(netdev);
5234
5235         if (netif_running(netdev)) {
5236                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5237                 e1000e_down(adapter);
5238                 e1000_free_irq(adapter);
5239         }
5240         e1000e_reset_interrupt_capability(adapter);
5241
5242         retval = pci_save_state(pdev);
5243         if (retval)
5244                 return retval;
5245
5246         status = er32(STATUS);
5247         if (status & E1000_STATUS_LU)
5248                 wufc &= ~E1000_WUFC_LNKC;
5249
5250         if (wufc) {
5251                 e1000_setup_rctl(adapter);
5252                 e1000_set_multi(netdev);
5253
5254                 /* turn on all-multi mode if wake on multicast is enabled */
5255                 if (wufc & E1000_WUFC_MC) {
5256                         rctl = er32(RCTL);
5257                         rctl |= E1000_RCTL_MPE;
5258                         ew32(RCTL, rctl);
5259                 }
5260
5261                 ctrl = er32(CTRL);
5262                 /* advertise wake from D3Cold */
5263                 #define E1000_CTRL_ADVD3WUC 0x00100000
5264                 /* phy power management enable */
5265                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5266                 ctrl |= E1000_CTRL_ADVD3WUC;
5267                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5268                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5269                 ew32(CTRL, ctrl);
5270
5271                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5272                     adapter->hw.phy.media_type ==
5273                     e1000_media_type_internal_serdes) {
5274                         /* keep the laser running in D3 */
5275                         ctrl_ext = er32(CTRL_EXT);
5276                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5277                         ew32(CTRL_EXT, ctrl_ext);
5278                 }
5279
5280                 if (adapter->flags & FLAG_IS_ICH)
5281                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5282
5283                 /* Allow time for pending master requests to run */
5284                 e1000e_disable_pcie_master(&adapter->hw);
5285
5286                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5287                         /* enable wakeup by the PHY */
5288                         retval = e1000_init_phy_wakeup(adapter, wufc);
5289                         if (retval)
5290                                 return retval;
5291                 } else {
5292                         /* enable wakeup by the MAC */
5293                         ew32(WUFC, wufc);
5294                         ew32(WUC, E1000_WUC_PME_EN);
5295                 }
5296         } else {
5297                 ew32(WUC, 0);
5298                 ew32(WUFC, 0);
5299         }
5300
5301         *enable_wake = !!wufc;
5302
5303         /* make sure adapter isn't asleep if manageability is enabled */
5304         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5305             (hw->mac.ops.check_mng_mode(hw)))
5306                 *enable_wake = true;
5307
5308         if (adapter->hw.phy.type == e1000_phy_igp_3)
5309                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5310
5311         /*
5312          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5313          * would have already happened in close and is redundant.
5314          */
5315         e1000e_release_hw_control(adapter);
5316
5317         pci_disable_device(pdev);
5318
5319         return 0;
5320 }
5321
5322 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5323 {
5324         if (sleep && wake) {
5325                 pci_prepare_to_sleep(pdev);
5326                 return;
5327         }
5328
5329         pci_wake_from_d3(pdev, wake);
5330         pci_set_power_state(pdev, PCI_D3hot);
5331 }
5332
5333 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5334                                     bool wake)
5335 {
5336         struct net_device *netdev = pci_get_drvdata(pdev);
5337         struct e1000_adapter *adapter = netdev_priv(netdev);
5338
5339         /*
5340          * The pci-e switch on some quad port adapters will report a
5341          * correctable error when the MAC transitions from D0 to D3.  To
5342          * prevent this we need to mask off the correctable errors on the
5343          * downstream port of the pci-e switch.
5344          */
5345         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5346                 struct pci_dev *us_dev = pdev->bus->self;
5347                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5348                 u16 devctl;
5349
5350                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5351                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5352                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5353
5354                 e1000_power_off(pdev, sleep, wake);
5355
5356                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5357         } else {
5358                 e1000_power_off(pdev, sleep, wake);
5359         }
5360 }
5361
5362 #ifdef CONFIG_PCIEASPM
5363 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5364 {
5365         pci_disable_link_state_locked(pdev, state);
5366 }
5367 #else
5368 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5369 {
5370         int pos;
5371         u16 reg16;
5372
5373         /*
5374          * Both device and parent should have the same ASPM setting.
5375          * Disable ASPM in downstream component first and then upstream.
5376          */
5377         pos = pci_pcie_cap(pdev);
5378         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5379         reg16 &= ~state;
5380         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5381
5382         if (!pdev->bus->self)
5383                 return;
5384
5385         pos = pci_pcie_cap(pdev->bus->self);
5386         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5387         reg16 &= ~state;
5388         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5389 }
5390 #endif
5391 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5392 {
5393         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5394                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5395                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5396
5397         __e1000e_disable_aspm(pdev, state);
5398 }
5399
5400 #ifdef CONFIG_PM
5401 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5402 {
5403         return !!adapter->tx_ring->buffer_info;
5404 }
5405
5406 static int __e1000_resume(struct pci_dev *pdev)
5407 {
5408         struct net_device *netdev = pci_get_drvdata(pdev);
5409         struct e1000_adapter *adapter = netdev_priv(netdev);
5410         struct e1000_hw *hw = &adapter->hw;
5411         u16 aspm_disable_flag = 0;
5412         u32 err;
5413
5414         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5415                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5416         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5417                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5418         if (aspm_disable_flag)
5419                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5420
5421         pci_set_power_state(pdev, PCI_D0);
5422         pci_restore_state(pdev);
5423         pci_save_state(pdev);
5424
5425         e1000e_set_interrupt_capability(adapter);
5426         if (netif_running(netdev)) {
5427                 err = e1000_request_irq(adapter);
5428                 if (err)
5429                         return err;
5430         }
5431
5432         e1000e_power_up_phy(adapter);
5433
5434         /* report the system wakeup cause from S3/S4 */
5435         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5436                 u16 phy_data;
5437
5438                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5439                 if (phy_data) {
5440                         e_info("PHY Wakeup cause - %s\n",
5441                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5442                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5443                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5444                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5445                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5446                                 " Change" : "other");
5447                 }
5448                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5449         } else {
5450                 u32 wus = er32(WUS);
5451                 if (wus) {
5452                         e_info("MAC Wakeup cause - %s\n",
5453                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5454                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5455                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5456                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5457                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5458                                 "other");
5459                 }
5460                 ew32(WUS, ~0);
5461         }
5462
5463         e1000e_reset(adapter);
5464
5465         e1000_init_manageability_pt(adapter);
5466
5467         if (netif_running(netdev))
5468                 e1000e_up(adapter);
5469
5470         netif_device_attach(netdev);
5471
5472         /*
5473          * If the controller has AMT, do not set DRV_LOAD until the interface
5474          * is up.  For all other cases, let the f/w know that the h/w is now
5475          * under the control of the driver.
5476          */
5477         if (!(adapter->flags & FLAG_HAS_AMT))
5478                 e1000e_get_hw_control(adapter);
5479
5480         return 0;
5481 }
5482
5483 #ifdef CONFIG_PM_SLEEP
5484 static int e1000_suspend(struct device *dev)
5485 {
5486         struct pci_dev *pdev = to_pci_dev(dev);
5487         int retval;
5488         bool wake;
5489
5490         retval = __e1000_shutdown(pdev, &wake, false);
5491         if (!retval)
5492                 e1000_complete_shutdown(pdev, true, wake);
5493
5494         return retval;
5495 }
5496
5497 static int e1000_resume(struct device *dev)
5498 {
5499         struct pci_dev *pdev = to_pci_dev(dev);
5500         struct net_device *netdev = pci_get_drvdata(pdev);
5501         struct e1000_adapter *adapter = netdev_priv(netdev);
5502
5503         if (e1000e_pm_ready(adapter))
5504                 adapter->idle_check = true;
5505
5506         return __e1000_resume(pdev);
5507 }
5508 #endif /* CONFIG_PM_SLEEP */
5509
5510 #ifdef CONFIG_PM_RUNTIME
5511 static int e1000_runtime_suspend(struct device *dev)
5512 {
5513         struct pci_dev *pdev = to_pci_dev(dev);
5514         struct net_device *netdev = pci_get_drvdata(pdev);
5515         struct e1000_adapter *adapter = netdev_priv(netdev);
5516
5517         if (e1000e_pm_ready(adapter)) {
5518                 bool wake;
5519
5520                 __e1000_shutdown(pdev, &wake, true);
5521         }
5522
5523         return 0;
5524 }
5525
5526 static int e1000_idle(struct device *dev)
5527 {
5528         struct pci_dev *pdev = to_pci_dev(dev);
5529         struct net_device *netdev = pci_get_drvdata(pdev);
5530         struct e1000_adapter *adapter = netdev_priv(netdev);
5531
5532         if (!e1000e_pm_ready(adapter))
5533                 return 0;
5534
5535         if (adapter->idle_check) {
5536                 adapter->idle_check = false;
5537                 if (!e1000e_has_link(adapter))
5538                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5539         }
5540
5541         return -EBUSY;
5542 }
5543
5544 static int e1000_runtime_resume(struct device *dev)
5545 {
5546         struct pci_dev *pdev = to_pci_dev(dev);
5547         struct net_device *netdev = pci_get_drvdata(pdev);
5548         struct e1000_adapter *adapter = netdev_priv(netdev);
5549
5550         if (!e1000e_pm_ready(adapter))
5551                 return 0;
5552
5553         adapter->idle_check = !dev->power.runtime_auto;
5554         return __e1000_resume(pdev);
5555 }
5556 #endif /* CONFIG_PM_RUNTIME */
5557 #endif /* CONFIG_PM */
5558
5559 static void e1000_shutdown(struct pci_dev *pdev)
5560 {
5561         bool wake = false;
5562
5563         __e1000_shutdown(pdev, &wake, false);
5564
5565         if (system_state == SYSTEM_POWER_OFF)
5566                 e1000_complete_shutdown(pdev, false, wake);
5567 }
5568
5569 #ifdef CONFIG_NET_POLL_CONTROLLER
5570
5571 static irqreturn_t e1000_intr_msix(int irq, void *data)
5572 {
5573         struct net_device *netdev = data;
5574         struct e1000_adapter *adapter = netdev_priv(netdev);
5575
5576         if (adapter->msix_entries) {
5577                 int vector, msix_irq;
5578
5579                 vector = 0;
5580                 msix_irq = adapter->msix_entries[vector].vector;
5581                 disable_irq(msix_irq);
5582                 e1000_intr_msix_rx(msix_irq, netdev);
5583                 enable_irq(msix_irq);
5584
5585                 vector++;
5586                 msix_irq = adapter->msix_entries[vector].vector;
5587                 disable_irq(msix_irq);
5588                 e1000_intr_msix_tx(msix_irq, netdev);
5589                 enable_irq(msix_irq);
5590
5591                 vector++;
5592                 msix_irq = adapter->msix_entries[vector].vector;
5593                 disable_irq(msix_irq);
5594                 e1000_msix_other(msix_irq, netdev);
5595                 enable_irq(msix_irq);
5596         }
5597
5598         return IRQ_HANDLED;
5599 }
5600
5601 /*
5602  * Polling 'interrupt' - used by things like netconsole to send skbs
5603  * without having to re-enable interrupts. It's not called while
5604  * the interrupt routine is executing.
5605  */
5606 static void e1000_netpoll(struct net_device *netdev)
5607 {
5608         struct e1000_adapter *adapter = netdev_priv(netdev);
5609
5610         switch (adapter->int_mode) {
5611         case E1000E_INT_MODE_MSIX:
5612                 e1000_intr_msix(adapter->pdev->irq, netdev);
5613                 break;
5614         case E1000E_INT_MODE_MSI:
5615                 disable_irq(adapter->pdev->irq);
5616                 e1000_intr_msi(adapter->pdev->irq, netdev);
5617                 enable_irq(adapter->pdev->irq);
5618                 break;
5619         default: /* E1000E_INT_MODE_LEGACY */
5620                 disable_irq(adapter->pdev->irq);
5621                 e1000_intr(adapter->pdev->irq, netdev);
5622                 enable_irq(adapter->pdev->irq);
5623                 break;
5624         }
5625 }
5626 #endif
5627
5628 /**
5629  * e1000_io_error_detected - called when PCI error is detected
5630  * @pdev: Pointer to PCI device
5631  * @state: The current pci connection state
5632  *
5633  * This function is called after a PCI bus error affecting
5634  * this device has been detected.
5635  */
5636 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5637                                                 pci_channel_state_t state)
5638 {
5639         struct net_device *netdev = pci_get_drvdata(pdev);
5640         struct e1000_adapter *adapter = netdev_priv(netdev);
5641
5642         netif_device_detach(netdev);
5643
5644         if (state == pci_channel_io_perm_failure)
5645                 return PCI_ERS_RESULT_DISCONNECT;
5646
5647         if (netif_running(netdev))
5648                 e1000e_down(adapter);
5649         pci_disable_device(pdev);
5650
5651         /* Request a slot slot reset. */
5652         return PCI_ERS_RESULT_NEED_RESET;
5653 }
5654
5655 /**
5656  * e1000_io_slot_reset - called after the pci bus has been reset.
5657  * @pdev: Pointer to PCI device
5658  *
5659  * Restart the card from scratch, as if from a cold-boot. Implementation
5660  * resembles the first-half of the e1000_resume routine.
5661  */
5662 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5663 {
5664         struct net_device *netdev = pci_get_drvdata(pdev);
5665         struct e1000_adapter *adapter = netdev_priv(netdev);
5666         struct e1000_hw *hw = &adapter->hw;
5667         u16 aspm_disable_flag = 0;
5668         int err;
5669         pci_ers_result_t result;
5670
5671         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5672                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5673         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5674                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5675         if (aspm_disable_flag)
5676                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5677
5678         err = pci_enable_device_mem(pdev);
5679         if (err) {
5680                 dev_err(&pdev->dev,
5681                         "Cannot re-enable PCI device after reset.\n");
5682                 result = PCI_ERS_RESULT_DISCONNECT;
5683         } else {
5684                 pci_set_master(pdev);
5685                 pdev->state_saved = true;
5686                 pci_restore_state(pdev);
5687
5688                 pci_enable_wake(pdev, PCI_D3hot, 0);
5689                 pci_enable_wake(pdev, PCI_D3cold, 0);
5690
5691                 e1000e_reset(adapter);
5692                 ew32(WUS, ~0);
5693                 result = PCI_ERS_RESULT_RECOVERED;
5694         }
5695
5696         pci_cleanup_aer_uncorrect_error_status(pdev);
5697
5698         return result;
5699 }
5700
5701 /**
5702  * e1000_io_resume - called when traffic can start flowing again.
5703  * @pdev: Pointer to PCI device
5704  *
5705  * This callback is called when the error recovery driver tells us that
5706  * its OK to resume normal operation. Implementation resembles the
5707  * second-half of the e1000_resume routine.
5708  */
5709 static void e1000_io_resume(struct pci_dev *pdev)
5710 {
5711         struct net_device *netdev = pci_get_drvdata(pdev);
5712         struct e1000_adapter *adapter = netdev_priv(netdev);
5713
5714         e1000_init_manageability_pt(adapter);
5715
5716         if (netif_running(netdev)) {
5717                 if (e1000e_up(adapter)) {
5718                         dev_err(&pdev->dev,
5719                                 "can't bring device back up after reset\n");
5720                         return;
5721                 }
5722         }
5723
5724         netif_device_attach(netdev);
5725
5726         /*
5727          * If the controller has AMT, do not set DRV_LOAD until the interface
5728          * is up.  For all other cases, let the f/w know that the h/w is now
5729          * under the control of the driver.
5730          */
5731         if (!(adapter->flags & FLAG_HAS_AMT))
5732                 e1000e_get_hw_control(adapter);
5733
5734 }
5735
5736 static void e1000_print_device_info(struct e1000_adapter *adapter)
5737 {
5738         struct e1000_hw *hw = &adapter->hw;
5739         struct net_device *netdev = adapter->netdev;
5740         u32 ret_val;
5741         u8 pba_str[E1000_PBANUM_LENGTH];
5742
5743         /* print bus type/speed/width info */
5744         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5745                /* bus width */
5746                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5747                 "Width x1"),
5748                /* MAC address */
5749                netdev->dev_addr);
5750         e_info("Intel(R) PRO/%s Network Connection\n",
5751                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5752         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5753                                                 E1000_PBANUM_LENGTH);
5754         if (ret_val)
5755                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5756         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5757                hw->mac.type, hw->phy.type, pba_str);
5758 }
5759
5760 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5761 {
5762         struct e1000_hw *hw = &adapter->hw;
5763         int ret_val;
5764         u16 buf = 0;
5765
5766         if (hw->mac.type != e1000_82573)
5767                 return;
5768
5769         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5770         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5771                 /* Deep Smart Power Down (DSPD) */
5772                 dev_warn(&adapter->pdev->dev,
5773                          "Warning: detected DSPD enabled in EEPROM\n");
5774         }
5775 }
5776
5777 static const struct net_device_ops e1000e_netdev_ops = {
5778         .ndo_open               = e1000_open,
5779         .ndo_stop               = e1000_close,
5780         .ndo_start_xmit         = e1000_xmit_frame,
5781         .ndo_get_stats64        = e1000e_get_stats64,
5782         .ndo_set_multicast_list = e1000_set_multi,
5783         .ndo_set_mac_address    = e1000_set_mac,
5784         .ndo_change_mtu         = e1000_change_mtu,
5785         .ndo_do_ioctl           = e1000_ioctl,
5786         .ndo_tx_timeout         = e1000_tx_timeout,
5787         .ndo_validate_addr      = eth_validate_addr,
5788
5789         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5790         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5791 #ifdef CONFIG_NET_POLL_CONTROLLER
5792         .ndo_poll_controller    = e1000_netpoll,
5793 #endif
5794 };
5795
5796 /**
5797  * e1000_probe - Device Initialization Routine
5798  * @pdev: PCI device information struct
5799  * @ent: entry in e1000_pci_tbl
5800  *
5801  * Returns 0 on success, negative on failure
5802  *
5803  * e1000_probe initializes an adapter identified by a pci_dev structure.
5804  * The OS initialization, configuring of the adapter private structure,
580