e1000e: move settting of flow control refresh timer to link setup code
[linux-2.6.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2009 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/pci.h>
35 #include <linux/vmalloc.h>
36 #include <linux/pagemap.h>
37 #include <linux/delay.h>
38 #include <linux/netdevice.h>
39 #include <linux/tcp.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/mii.h>
45 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/cpu.h>
48 #include <linux/smp.h>
49 #include <linux/pm_qos_params.h>
50 #include <linux/pm_runtime.h>
51 #include <linux/aer.h>
52
53 #include "e1000.h"
54
55 #define DRV_VERSION "1.0.2-k4"
56 char e1000e_driver_name[] = "e1000e";
57 const char e1000e_driver_version[] = DRV_VERSION;
58
59 static const struct e1000_info *e1000_info_tbl[] = {
60         [board_82571]           = &e1000_82571_info,
61         [board_82572]           = &e1000_82572_info,
62         [board_82573]           = &e1000_82573_info,
63         [board_82574]           = &e1000_82574_info,
64         [board_82583]           = &e1000_82583_info,
65         [board_80003es2lan]     = &e1000_es2_info,
66         [board_ich8lan]         = &e1000_ich8_info,
67         [board_ich9lan]         = &e1000_ich9_info,
68         [board_ich10lan]        = &e1000_ich10_info,
69         [board_pchlan]          = &e1000_pch_info,
70 };
71
72 struct e1000_reg_info {
73         u32 ofs;
74         char *name;
75 };
76
77 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
78 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
79 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
80 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
81 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
82
83 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
84 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
85 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
86 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
87 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
88
89 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
90
91         /* General Registers */
92         {E1000_CTRL, "CTRL"},
93         {E1000_STATUS, "STATUS"},
94         {E1000_CTRL_EXT, "CTRL_EXT"},
95
96         /* Interrupt Registers */
97         {E1000_ICR, "ICR"},
98
99         /* RX Registers */
100         {E1000_RCTL, "RCTL"},
101         {E1000_RDLEN, "RDLEN"},
102         {E1000_RDH, "RDH"},
103         {E1000_RDT, "RDT"},
104         {E1000_RDTR, "RDTR"},
105         {E1000_RXDCTL(0), "RXDCTL"},
106         {E1000_ERT, "ERT"},
107         {E1000_RDBAL, "RDBAL"},
108         {E1000_RDBAH, "RDBAH"},
109         {E1000_RDFH, "RDFH"},
110         {E1000_RDFT, "RDFT"},
111         {E1000_RDFHS, "RDFHS"},
112         {E1000_RDFTS, "RDFTS"},
113         {E1000_RDFPC, "RDFPC"},
114
115         /* TX Registers */
116         {E1000_TCTL, "TCTL"},
117         {E1000_TDBAL, "TDBAL"},
118         {E1000_TDBAH, "TDBAH"},
119         {E1000_TDLEN, "TDLEN"},
120         {E1000_TDH, "TDH"},
121         {E1000_TDT, "TDT"},
122         {E1000_TIDV, "TIDV"},
123         {E1000_TXDCTL(0), "TXDCTL"},
124         {E1000_TADV, "TADV"},
125         {E1000_TARC(0), "TARC"},
126         {E1000_TDFH, "TDFH"},
127         {E1000_TDFT, "TDFT"},
128         {E1000_TDFHS, "TDFHS"},
129         {E1000_TDFTS, "TDFTS"},
130         {E1000_TDFPC, "TDFPC"},
131
132         /* List Terminator */
133         {}
134 };
135
136 /*
137  * e1000_regdump - register printout routine
138  */
139 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
140 {
141         int n = 0;
142         char rname[16];
143         u32 regs[8];
144
145         switch (reginfo->ofs) {
146         case E1000_RXDCTL(0):
147                 for (n = 0; n < 2; n++)
148                         regs[n] = __er32(hw, E1000_RXDCTL(n));
149                 break;
150         case E1000_TXDCTL(0):
151                 for (n = 0; n < 2; n++)
152                         regs[n] = __er32(hw, E1000_TXDCTL(n));
153                 break;
154         case E1000_TARC(0):
155                 for (n = 0; n < 2; n++)
156                         regs[n] = __er32(hw, E1000_TARC(n));
157                 break;
158         default:
159                 printk(KERN_INFO "%-15s %08x\n",
160                         reginfo->name, __er32(hw, reginfo->ofs));
161                 return;
162         }
163
164         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
165         printk(KERN_INFO "%-15s ", rname);
166         for (n = 0; n < 2; n++)
167                 printk(KERN_CONT "%08x ", regs[n]);
168         printk(KERN_CONT "\n");
169 }
170
171
172 /*
173  * e1000e_dump - Print registers, tx-ring and rx-ring
174  */
175 static void e1000e_dump(struct e1000_adapter *adapter)
176 {
177         struct net_device *netdev = adapter->netdev;
178         struct e1000_hw *hw = &adapter->hw;
179         struct e1000_reg_info *reginfo;
180         struct e1000_ring *tx_ring = adapter->tx_ring;
181         struct e1000_tx_desc *tx_desc;
182         struct my_u0 { u64 a; u64 b; } *u0;
183         struct e1000_buffer *buffer_info;
184         struct e1000_ring *rx_ring = adapter->rx_ring;
185         union e1000_rx_desc_packet_split *rx_desc_ps;
186         struct e1000_rx_desc *rx_desc;
187         struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
188         u32 staterr;
189         int i = 0;
190
191         if (!netif_msg_hw(adapter))
192                 return;
193
194         /* Print netdevice Info */
195         if (netdev) {
196                 dev_info(&adapter->pdev->dev, "Net device Info\n");
197                 printk(KERN_INFO "Device Name     state            "
198                         "trans_start      last_rx\n");
199                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
200                         netdev->name,
201                         netdev->state,
202                         netdev->trans_start,
203                         netdev->last_rx);
204         }
205
206         /* Print Registers */
207         dev_info(&adapter->pdev->dev, "Register Dump\n");
208         printk(KERN_INFO " Register Name   Value\n");
209         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
210              reginfo->name; reginfo++) {
211                 e1000_regdump(hw, reginfo);
212         }
213
214         /* Print TX Ring Summary */
215         if (!netdev || !netif_running(netdev))
216                 goto exit;
217
218         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
219         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
220                 " leng ntw timestamp\n");
221         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
222         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
223                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
224                 (u64)buffer_info->dma,
225                 buffer_info->length,
226                 buffer_info->next_to_watch,
227                 (u64)buffer_info->time_stamp);
228
229         /* Print TX Rings */
230         if (!netif_msg_tx_done(adapter))
231                 goto rx_ring_summary;
232
233         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
234
235         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
236          *
237          * Legacy Transmit Descriptor
238          *   +--------------------------------------------------------------+
239          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
240          *   +--------------------------------------------------------------+
241          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
242          *   +--------------------------------------------------------------+
243          *   63       48 47        36 35    32 31     24 23    16 15        0
244          *
245          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
246          *   63      48 47    40 39       32 31             16 15    8 7      0
247          *   +----------------------------------------------------------------+
248          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
249          *   +----------------------------------------------------------------+
250          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
251          *   +----------------------------------------------------------------+
252          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
253          *
254          * Extended Data Descriptor (DTYP=0x1)
255          *   +----------------------------------------------------------------+
256          * 0 |                     Buffer Address [63:0]                      |
257          *   +----------------------------------------------------------------+
258          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
259          *   +----------------------------------------------------------------+
260          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
261          */
262         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
263                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
264                 "<-- Legacy format\n");
265         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
266                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
267                 "<-- Ext Context format\n");
268         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
269                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
270                 "<-- Ext Data format\n");
271         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
272                 tx_desc = E1000_TX_DESC(*tx_ring, i);
273                 buffer_info = &tx_ring->buffer_info[i];
274                 u0 = (struct my_u0 *)tx_desc;
275                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
276                         "%04X  %3X %016llX %p",
277                        (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
278                         ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
279                        le64_to_cpu(u0->a), le64_to_cpu(u0->b),
280                        (u64)buffer_info->dma, buffer_info->length,
281                        buffer_info->next_to_watch, (u64)buffer_info->time_stamp,
282                        buffer_info->skb);
283                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
284                         printk(KERN_CONT " NTC/U\n");
285                 else if (i == tx_ring->next_to_use)
286                         printk(KERN_CONT " NTU\n");
287                 else if (i == tx_ring->next_to_clean)
288                         printk(KERN_CONT " NTC\n");
289                 else
290                         printk(KERN_CONT "\n");
291
292                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
293                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
294                                         16, 1, phys_to_virt(buffer_info->dma),
295                                         buffer_info->length, true);
296         }
297
298         /* Print RX Rings Summary */
299 rx_ring_summary:
300         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
301         printk(KERN_INFO "Queue [NTU] [NTC]\n");
302         printk(KERN_INFO " %5d %5X %5X\n", 0,
303                 rx_ring->next_to_use, rx_ring->next_to_clean);
304
305         /* Print RX Rings */
306         if (!netif_msg_rx_status(adapter))
307                 goto exit;
308
309         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
310         switch (adapter->rx_ps_pages) {
311         case 1:
312         case 2:
313         case 3:
314                 /* [Extended] Packet Split Receive Descriptor Format
315                  *
316                  *    +-----------------------------------------------------+
317                  *  0 |                Buffer Address 0 [63:0]              |
318                  *    +-----------------------------------------------------+
319                  *  8 |                Buffer Address 1 [63:0]              |
320                  *    +-----------------------------------------------------+
321                  * 16 |                Buffer Address 2 [63:0]              |
322                  *    +-----------------------------------------------------+
323                  * 24 |                Buffer Address 3 [63:0]              |
324                  *    +-----------------------------------------------------+
325                  */
326                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
327                         "[buffer 1 63:0 ] "
328                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
329                        "[bi->skb] <-- Ext Pkt Split format\n");
330                 /* [Extended] Receive Descriptor (Write-Back) Format
331                  *
332                  *   63       48 47    32 31     13 12    8 7    4 3        0
333                  *   +------------------------------------------------------+
334                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
335                  *   | Checksum | Ident  |         | Queue |      |  Type   |
336                  *   +------------------------------------------------------+
337                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
338                  *   +------------------------------------------------------+
339                  *   63       48 47    32 31            20 19               0
340                  */
341                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
342                         "[vl   l0 ee  es] "
343                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
344                        "[bi->skb] <-- Ext Rx Write-Back format\n");
345                 for (i = 0; i < rx_ring->count; i++) {
346                         buffer_info = &rx_ring->buffer_info[i];
347                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
348                         u1 = (struct my_u1 *)rx_desc_ps;
349                         staterr =
350                                 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
351                         if (staterr & E1000_RXD_STAT_DD) {
352                                 /* Descriptor Done */
353                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
354                                         "%016llX %016llX %016llX "
355                                         "---------------- %p", i,
356                                         le64_to_cpu(u1->a),
357                                         le64_to_cpu(u1->b),
358                                         le64_to_cpu(u1->c),
359                                         le64_to_cpu(u1->d),
360                                         buffer_info->skb);
361                         } else {
362                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
363                                         "%016llX %016llX %016llX %016llX %p", i,
364                                         le64_to_cpu(u1->a),
365                                         le64_to_cpu(u1->b),
366                                         le64_to_cpu(u1->c),
367                                         le64_to_cpu(u1->d),
368                                         (u64)buffer_info->dma,
369                                         buffer_info->skb);
370
371                                 if (netif_msg_pktdata(adapter))
372                                         print_hex_dump(KERN_INFO, "",
373                                                 DUMP_PREFIX_ADDRESS, 16, 1,
374                                                 phys_to_virt(buffer_info->dma),
375                                                 adapter->rx_ps_bsize0, true);
376                         }
377
378                         if (i == rx_ring->next_to_use)
379                                 printk(KERN_CONT " NTU\n");
380                         else if (i == rx_ring->next_to_clean)
381                                 printk(KERN_CONT " NTC\n");
382                         else
383                                 printk(KERN_CONT "\n");
384                 }
385                 break;
386         default:
387         case 0:
388                 /* Legacy Receive Descriptor Format
389                  *
390                  * +-----------------------------------------------------+
391                  * |                Buffer Address [63:0]                |
392                  * +-----------------------------------------------------+
393                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
394                  * +-----------------------------------------------------+
395                  * 63       48 47    40 39      32 31         16 15      0
396                  */
397                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
398                         "[vl er S cks ln] [bi->dma       ] [bi->skb] "
399                         "<-- Legacy format\n");
400                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
401                         rx_desc = E1000_RX_DESC(*rx_ring, i);
402                         buffer_info = &rx_ring->buffer_info[i];
403                         u0 = (struct my_u0 *)rx_desc;
404                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
405                                 "%016llX %p",
406                                 i, le64_to_cpu(u0->a), le64_to_cpu(u0->b),
407                                 (u64)buffer_info->dma, buffer_info->skb);
408                         if (i == rx_ring->next_to_use)
409                                 printk(KERN_CONT " NTU\n");
410                         else if (i == rx_ring->next_to_clean)
411                                 printk(KERN_CONT " NTC\n");
412                         else
413                                 printk(KERN_CONT "\n");
414
415                         if (netif_msg_pktdata(adapter))
416                                 print_hex_dump(KERN_INFO, "",
417                                         DUMP_PREFIX_ADDRESS,
418                                         16, 1, phys_to_virt(buffer_info->dma),
419                                         adapter->rx_buffer_len, true);
420                 }
421         }
422
423 exit:
424         return;
425 }
426
427 /**
428  * e1000_desc_unused - calculate if we have unused descriptors
429  **/
430 static int e1000_desc_unused(struct e1000_ring *ring)
431 {
432         if (ring->next_to_clean > ring->next_to_use)
433                 return ring->next_to_clean - ring->next_to_use - 1;
434
435         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
436 }
437
438 /**
439  * e1000_receive_skb - helper function to handle Rx indications
440  * @adapter: board private structure
441  * @status: descriptor status field as written by hardware
442  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
443  * @skb: pointer to sk_buff to be indicated to stack
444  **/
445 static void e1000_receive_skb(struct e1000_adapter *adapter,
446                               struct net_device *netdev,
447                               struct sk_buff *skb,
448                               u8 status, __le16 vlan)
449 {
450         skb->protocol = eth_type_trans(skb, netdev);
451
452         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
453                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
454                                  le16_to_cpu(vlan), skb);
455         else
456                 napi_gro_receive(&adapter->napi, skb);
457 }
458
459 /**
460  * e1000_rx_checksum - Receive Checksum Offload for 82543
461  * @adapter:     board private structure
462  * @status_err:  receive descriptor status and error fields
463  * @csum:       receive descriptor csum field
464  * @sk_buff:     socket buffer with received data
465  **/
466 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
467                               u32 csum, struct sk_buff *skb)
468 {
469         u16 status = (u16)status_err;
470         u8 errors = (u8)(status_err >> 24);
471         skb->ip_summed = CHECKSUM_NONE;
472
473         /* Ignore Checksum bit is set */
474         if (status & E1000_RXD_STAT_IXSM)
475                 return;
476         /* TCP/UDP checksum error bit is set */
477         if (errors & E1000_RXD_ERR_TCPE) {
478                 /* let the stack verify checksum errors */
479                 adapter->hw_csum_err++;
480                 return;
481         }
482
483         /* TCP/UDP Checksum has not been calculated */
484         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
485                 return;
486
487         /* It must be a TCP or UDP packet with a valid checksum */
488         if (status & E1000_RXD_STAT_TCPCS) {
489                 /* TCP checksum is good */
490                 skb->ip_summed = CHECKSUM_UNNECESSARY;
491         } else {
492                 /*
493                  * IP fragment with UDP payload
494                  * Hardware complements the payload checksum, so we undo it
495                  * and then put the value in host order for further stack use.
496                  */
497                 __sum16 sum = (__force __sum16)htons(csum);
498                 skb->csum = csum_unfold(~sum);
499                 skb->ip_summed = CHECKSUM_COMPLETE;
500         }
501         adapter->hw_csum_good++;
502 }
503
504 /**
505  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
506  * @adapter: address of board private structure
507  **/
508 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
509                                    int cleaned_count)
510 {
511         struct net_device *netdev = adapter->netdev;
512         struct pci_dev *pdev = adapter->pdev;
513         struct e1000_ring *rx_ring = adapter->rx_ring;
514         struct e1000_rx_desc *rx_desc;
515         struct e1000_buffer *buffer_info;
516         struct sk_buff *skb;
517         unsigned int i;
518         unsigned int bufsz = adapter->rx_buffer_len;
519
520         i = rx_ring->next_to_use;
521         buffer_info = &rx_ring->buffer_info[i];
522
523         while (cleaned_count--) {
524                 skb = buffer_info->skb;
525                 if (skb) {
526                         skb_trim(skb, 0);
527                         goto map_skb;
528                 }
529
530                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
531                 if (!skb) {
532                         /* Better luck next round */
533                         adapter->alloc_rx_buff_failed++;
534                         break;
535                 }
536
537                 buffer_info->skb = skb;
538 map_skb:
539                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
540                                                   adapter->rx_buffer_len,
541                                                   DMA_FROM_DEVICE);
542                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
543                         dev_err(&pdev->dev, "RX DMA map failed\n");
544                         adapter->rx_dma_failed++;
545                         break;
546                 }
547
548                 rx_desc = E1000_RX_DESC(*rx_ring, i);
549                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
550
551                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
552                         /*
553                          * Force memory writes to complete before letting h/w
554                          * know there are new descriptors to fetch.  (Only
555                          * applicable for weak-ordered memory model archs,
556                          * such as IA-64).
557                          */
558                         wmb();
559                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
560                 }
561                 i++;
562                 if (i == rx_ring->count)
563                         i = 0;
564                 buffer_info = &rx_ring->buffer_info[i];
565         }
566
567         rx_ring->next_to_use = i;
568 }
569
570 /**
571  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
572  * @adapter: address of board private structure
573  **/
574 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
575                                       int cleaned_count)
576 {
577         struct net_device *netdev = adapter->netdev;
578         struct pci_dev *pdev = adapter->pdev;
579         union e1000_rx_desc_packet_split *rx_desc;
580         struct e1000_ring *rx_ring = adapter->rx_ring;
581         struct e1000_buffer *buffer_info;
582         struct e1000_ps_page *ps_page;
583         struct sk_buff *skb;
584         unsigned int i, j;
585
586         i = rx_ring->next_to_use;
587         buffer_info = &rx_ring->buffer_info[i];
588
589         while (cleaned_count--) {
590                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
591
592                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
593                         ps_page = &buffer_info->ps_pages[j];
594                         if (j >= adapter->rx_ps_pages) {
595                                 /* all unused desc entries get hw null ptr */
596                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
597                                 continue;
598                         }
599                         if (!ps_page->page) {
600                                 ps_page->page = alloc_page(GFP_ATOMIC);
601                                 if (!ps_page->page) {
602                                         adapter->alloc_rx_buff_failed++;
603                                         goto no_buffers;
604                                 }
605                                 ps_page->dma = dma_map_page(&pdev->dev,
606                                                             ps_page->page,
607                                                             0, PAGE_SIZE,
608                                                             DMA_FROM_DEVICE);
609                                 if (dma_mapping_error(&pdev->dev,
610                                                       ps_page->dma)) {
611                                         dev_err(&adapter->pdev->dev,
612                                           "RX DMA page map failed\n");
613                                         adapter->rx_dma_failed++;
614                                         goto no_buffers;
615                                 }
616                         }
617                         /*
618                          * Refresh the desc even if buffer_addrs
619                          * didn't change because each write-back
620                          * erases this info.
621                          */
622                         rx_desc->read.buffer_addr[j+1] =
623                              cpu_to_le64(ps_page->dma);
624                 }
625
626                 skb = netdev_alloc_skb_ip_align(netdev,
627                                                 adapter->rx_ps_bsize0);
628
629                 if (!skb) {
630                         adapter->alloc_rx_buff_failed++;
631                         break;
632                 }
633
634                 buffer_info->skb = skb;
635                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
636                                                   adapter->rx_ps_bsize0,
637                                                   DMA_FROM_DEVICE);
638                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
639                         dev_err(&pdev->dev, "RX DMA map failed\n");
640                         adapter->rx_dma_failed++;
641                         /* cleanup skb */
642                         dev_kfree_skb_any(skb);
643                         buffer_info->skb = NULL;
644                         break;
645                 }
646
647                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
648
649                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
650                         /*
651                          * Force memory writes to complete before letting h/w
652                          * know there are new descriptors to fetch.  (Only
653                          * applicable for weak-ordered memory model archs,
654                          * such as IA-64).
655                          */
656                         wmb();
657                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
658                 }
659
660                 i++;
661                 if (i == rx_ring->count)
662                         i = 0;
663                 buffer_info = &rx_ring->buffer_info[i];
664         }
665
666 no_buffers:
667         rx_ring->next_to_use = i;
668 }
669
670 /**
671  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
672  * @adapter: address of board private structure
673  * @cleaned_count: number of buffers to allocate this pass
674  **/
675
676 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
677                                          int cleaned_count)
678 {
679         struct net_device *netdev = adapter->netdev;
680         struct pci_dev *pdev = adapter->pdev;
681         struct e1000_rx_desc *rx_desc;
682         struct e1000_ring *rx_ring = adapter->rx_ring;
683         struct e1000_buffer *buffer_info;
684         struct sk_buff *skb;
685         unsigned int i;
686         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
687
688         i = rx_ring->next_to_use;
689         buffer_info = &rx_ring->buffer_info[i];
690
691         while (cleaned_count--) {
692                 skb = buffer_info->skb;
693                 if (skb) {
694                         skb_trim(skb, 0);
695                         goto check_page;
696                 }
697
698                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
699                 if (unlikely(!skb)) {
700                         /* Better luck next round */
701                         adapter->alloc_rx_buff_failed++;
702                         break;
703                 }
704
705                 buffer_info->skb = skb;
706 check_page:
707                 /* allocate a new page if necessary */
708                 if (!buffer_info->page) {
709                         buffer_info->page = alloc_page(GFP_ATOMIC);
710                         if (unlikely(!buffer_info->page)) {
711                                 adapter->alloc_rx_buff_failed++;
712                                 break;
713                         }
714                 }
715
716                 if (!buffer_info->dma)
717                         buffer_info->dma = dma_map_page(&pdev->dev,
718                                                         buffer_info->page, 0,
719                                                         PAGE_SIZE,
720                                                         DMA_FROM_DEVICE);
721
722                 rx_desc = E1000_RX_DESC(*rx_ring, i);
723                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
724
725                 if (unlikely(++i == rx_ring->count))
726                         i = 0;
727                 buffer_info = &rx_ring->buffer_info[i];
728         }
729
730         if (likely(rx_ring->next_to_use != i)) {
731                 rx_ring->next_to_use = i;
732                 if (unlikely(i-- == 0))
733                         i = (rx_ring->count - 1);
734
735                 /* Force memory writes to complete before letting h/w
736                  * know there are new descriptors to fetch.  (Only
737                  * applicable for weak-ordered memory model archs,
738                  * such as IA-64). */
739                 wmb();
740                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
741         }
742 }
743
744 /**
745  * e1000_clean_rx_irq - Send received data up the network stack; legacy
746  * @adapter: board private structure
747  *
748  * the return value indicates whether actual cleaning was done, there
749  * is no guarantee that everything was cleaned
750  **/
751 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
752                                int *work_done, int work_to_do)
753 {
754         struct net_device *netdev = adapter->netdev;
755         struct pci_dev *pdev = adapter->pdev;
756         struct e1000_hw *hw = &adapter->hw;
757         struct e1000_ring *rx_ring = adapter->rx_ring;
758         struct e1000_rx_desc *rx_desc, *next_rxd;
759         struct e1000_buffer *buffer_info, *next_buffer;
760         u32 length;
761         unsigned int i;
762         int cleaned_count = 0;
763         bool cleaned = 0;
764         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
765
766         i = rx_ring->next_to_clean;
767         rx_desc = E1000_RX_DESC(*rx_ring, i);
768         buffer_info = &rx_ring->buffer_info[i];
769
770         while (rx_desc->status & E1000_RXD_STAT_DD) {
771                 struct sk_buff *skb;
772                 u8 status;
773
774                 if (*work_done >= work_to_do)
775                         break;
776                 (*work_done)++;
777
778                 status = rx_desc->status;
779                 skb = buffer_info->skb;
780                 buffer_info->skb = NULL;
781
782                 prefetch(skb->data - NET_IP_ALIGN);
783
784                 i++;
785                 if (i == rx_ring->count)
786                         i = 0;
787                 next_rxd = E1000_RX_DESC(*rx_ring, i);
788                 prefetch(next_rxd);
789
790                 next_buffer = &rx_ring->buffer_info[i];
791
792                 cleaned = 1;
793                 cleaned_count++;
794                 dma_unmap_single(&pdev->dev,
795                                  buffer_info->dma,
796                                  adapter->rx_buffer_len,
797                                  DMA_FROM_DEVICE);
798                 buffer_info->dma = 0;
799
800                 length = le16_to_cpu(rx_desc->length);
801
802                 /*
803                  * !EOP means multiple descriptors were used to store a single
804                  * packet, if that's the case we need to toss it.  In fact, we
805                  * need to toss every packet with the EOP bit clear and the
806                  * next frame that _does_ have the EOP bit set, as it is by
807                  * definition only a frame fragment
808                  */
809                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
810                         adapter->flags2 |= FLAG2_IS_DISCARDING;
811
812                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
813                         /* All receives must fit into a single buffer */
814                         e_dbg("Receive packet consumed multiple buffers\n");
815                         /* recycle */
816                         buffer_info->skb = skb;
817                         if (status & E1000_RXD_STAT_EOP)
818                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
819                         goto next_desc;
820                 }
821
822                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
823                         /* recycle */
824                         buffer_info->skb = skb;
825                         goto next_desc;
826                 }
827
828                 /* adjust length to remove Ethernet CRC */
829                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
830                         length -= 4;
831
832                 total_rx_bytes += length;
833                 total_rx_packets++;
834
835                 /*
836                  * code added for copybreak, this should improve
837                  * performance for small packets with large amounts
838                  * of reassembly being done in the stack
839                  */
840                 if (length < copybreak) {
841                         struct sk_buff *new_skb =
842                             netdev_alloc_skb_ip_align(netdev, length);
843                         if (new_skb) {
844                                 skb_copy_to_linear_data_offset(new_skb,
845                                                                -NET_IP_ALIGN,
846                                                                (skb->data -
847                                                                 NET_IP_ALIGN),
848                                                                (length +
849                                                                 NET_IP_ALIGN));
850                                 /* save the skb in buffer_info as good */
851                                 buffer_info->skb = skb;
852                                 skb = new_skb;
853                         }
854                         /* else just continue with the old one */
855                 }
856                 /* end copybreak code */
857                 skb_put(skb, length);
858
859                 /* Receive Checksum Offload */
860                 e1000_rx_checksum(adapter,
861                                   (u32)(status) |
862                                   ((u32)(rx_desc->errors) << 24),
863                                   le16_to_cpu(rx_desc->csum), skb);
864
865                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
866
867 next_desc:
868                 rx_desc->status = 0;
869
870                 /* return some buffers to hardware, one at a time is too slow */
871                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
872                         adapter->alloc_rx_buf(adapter, cleaned_count);
873                         cleaned_count = 0;
874                 }
875
876                 /* use prefetched values */
877                 rx_desc = next_rxd;
878                 buffer_info = next_buffer;
879         }
880         rx_ring->next_to_clean = i;
881
882         cleaned_count = e1000_desc_unused(rx_ring);
883         if (cleaned_count)
884                 adapter->alloc_rx_buf(adapter, cleaned_count);
885
886         adapter->total_rx_bytes += total_rx_bytes;
887         adapter->total_rx_packets += total_rx_packets;
888         netdev->stats.rx_bytes += total_rx_bytes;
889         netdev->stats.rx_packets += total_rx_packets;
890         return cleaned;
891 }
892
893 static void e1000_put_txbuf(struct e1000_adapter *adapter,
894                              struct e1000_buffer *buffer_info)
895 {
896         if (buffer_info->dma) {
897                 if (buffer_info->mapped_as_page)
898                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
899                                        buffer_info->length, DMA_TO_DEVICE);
900                 else
901                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
902                                          buffer_info->length, DMA_TO_DEVICE);
903                 buffer_info->dma = 0;
904         }
905         if (buffer_info->skb) {
906                 dev_kfree_skb_any(buffer_info->skb);
907                 buffer_info->skb = NULL;
908         }
909         buffer_info->time_stamp = 0;
910 }
911
912 static void e1000_print_hw_hang(struct work_struct *work)
913 {
914         struct e1000_adapter *adapter = container_of(work,
915                                                      struct e1000_adapter,
916                                                      print_hang_task);
917         struct e1000_ring *tx_ring = adapter->tx_ring;
918         unsigned int i = tx_ring->next_to_clean;
919         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
920         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
921         struct e1000_hw *hw = &adapter->hw;
922         u16 phy_status, phy_1000t_status, phy_ext_status;
923         u16 pci_status;
924
925         e1e_rphy(hw, PHY_STATUS, &phy_status);
926         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
927         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
928
929         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
930
931         /* detected Hardware unit hang */
932         e_err("Detected Hardware Unit Hang:\n"
933               "  TDH                  <%x>\n"
934               "  TDT                  <%x>\n"
935               "  next_to_use          <%x>\n"
936               "  next_to_clean        <%x>\n"
937               "buffer_info[next_to_clean]:\n"
938               "  time_stamp           <%lx>\n"
939               "  next_to_watch        <%x>\n"
940               "  jiffies              <%lx>\n"
941               "  next_to_watch.status <%x>\n"
942               "MAC Status             <%x>\n"
943               "PHY Status             <%x>\n"
944               "PHY 1000BASE-T Status  <%x>\n"
945               "PHY Extended Status    <%x>\n"
946               "PCI Status             <%x>\n",
947               readl(adapter->hw.hw_addr + tx_ring->head),
948               readl(adapter->hw.hw_addr + tx_ring->tail),
949               tx_ring->next_to_use,
950               tx_ring->next_to_clean,
951               tx_ring->buffer_info[eop].time_stamp,
952               eop,
953               jiffies,
954               eop_desc->upper.fields.status,
955               er32(STATUS),
956               phy_status,
957               phy_1000t_status,
958               phy_ext_status,
959               pci_status);
960 }
961
962 /**
963  * e1000_clean_tx_irq - Reclaim resources after transmit completes
964  * @adapter: board private structure
965  *
966  * the return value indicates whether actual cleaning was done, there
967  * is no guarantee that everything was cleaned
968  **/
969 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
970 {
971         struct net_device *netdev = adapter->netdev;
972         struct e1000_hw *hw = &adapter->hw;
973         struct e1000_ring *tx_ring = adapter->tx_ring;
974         struct e1000_tx_desc *tx_desc, *eop_desc;
975         struct e1000_buffer *buffer_info;
976         unsigned int i, eop;
977         unsigned int count = 0;
978         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
979
980         i = tx_ring->next_to_clean;
981         eop = tx_ring->buffer_info[i].next_to_watch;
982         eop_desc = E1000_TX_DESC(*tx_ring, eop);
983
984         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
985                (count < tx_ring->count)) {
986                 bool cleaned = false;
987                 for (; !cleaned; count++) {
988                         tx_desc = E1000_TX_DESC(*tx_ring, i);
989                         buffer_info = &tx_ring->buffer_info[i];
990                         cleaned = (i == eop);
991
992                         if (cleaned) {
993                                 total_tx_packets += buffer_info->segs;
994                                 total_tx_bytes += buffer_info->bytecount;
995                         }
996
997                         e1000_put_txbuf(adapter, buffer_info);
998                         tx_desc->upper.data = 0;
999
1000                         i++;
1001                         if (i == tx_ring->count)
1002                                 i = 0;
1003                 }
1004
1005                 if (i == tx_ring->next_to_use)
1006                         break;
1007                 eop = tx_ring->buffer_info[i].next_to_watch;
1008                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1009         }
1010
1011         tx_ring->next_to_clean = i;
1012
1013 #define TX_WAKE_THRESHOLD 32
1014         if (count && netif_carrier_ok(netdev) &&
1015             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1016                 /* Make sure that anybody stopping the queue after this
1017                  * sees the new next_to_clean.
1018                  */
1019                 smp_mb();
1020
1021                 if (netif_queue_stopped(netdev) &&
1022                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1023                         netif_wake_queue(netdev);
1024                         ++adapter->restart_queue;
1025                 }
1026         }
1027
1028         if (adapter->detect_tx_hung) {
1029                 /*
1030                  * Detect a transmit hang in hardware, this serializes the
1031                  * check with the clearing of time_stamp and movement of i
1032                  */
1033                 adapter->detect_tx_hung = 0;
1034                 if (tx_ring->buffer_info[i].time_stamp &&
1035                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1036                                + (adapter->tx_timeout_factor * HZ)) &&
1037                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1038                         schedule_work(&adapter->print_hang_task);
1039                         netif_stop_queue(netdev);
1040                 }
1041         }
1042         adapter->total_tx_bytes += total_tx_bytes;
1043         adapter->total_tx_packets += total_tx_packets;
1044         netdev->stats.tx_bytes += total_tx_bytes;
1045         netdev->stats.tx_packets += total_tx_packets;
1046         return (count < tx_ring->count);
1047 }
1048
1049 /**
1050  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1051  * @adapter: board private structure
1052  *
1053  * the return value indicates whether actual cleaning was done, there
1054  * is no guarantee that everything was cleaned
1055  **/
1056 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1057                                   int *work_done, int work_to_do)
1058 {
1059         struct e1000_hw *hw = &adapter->hw;
1060         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1061         struct net_device *netdev = adapter->netdev;
1062         struct pci_dev *pdev = adapter->pdev;
1063         struct e1000_ring *rx_ring = adapter->rx_ring;
1064         struct e1000_buffer *buffer_info, *next_buffer;
1065         struct e1000_ps_page *ps_page;
1066         struct sk_buff *skb;
1067         unsigned int i, j;
1068         u32 length, staterr;
1069         int cleaned_count = 0;
1070         bool cleaned = 0;
1071         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1072
1073         i = rx_ring->next_to_clean;
1074         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1075         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1076         buffer_info = &rx_ring->buffer_info[i];
1077
1078         while (staterr & E1000_RXD_STAT_DD) {
1079                 if (*work_done >= work_to_do)
1080                         break;
1081                 (*work_done)++;
1082                 skb = buffer_info->skb;
1083
1084                 /* in the packet split case this is header only */
1085                 prefetch(skb->data - NET_IP_ALIGN);
1086
1087                 i++;
1088                 if (i == rx_ring->count)
1089                         i = 0;
1090                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1091                 prefetch(next_rxd);
1092
1093                 next_buffer = &rx_ring->buffer_info[i];
1094
1095                 cleaned = 1;
1096                 cleaned_count++;
1097                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1098                                  adapter->rx_ps_bsize0,
1099                                  DMA_FROM_DEVICE);
1100                 buffer_info->dma = 0;
1101
1102                 /* see !EOP comment in other rx routine */
1103                 if (!(staterr & E1000_RXD_STAT_EOP))
1104                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1105
1106                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1107                         e_dbg("Packet Split buffers didn't pick up the full "
1108                               "packet\n");
1109                         dev_kfree_skb_irq(skb);
1110                         if (staterr & E1000_RXD_STAT_EOP)
1111                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1112                         goto next_desc;
1113                 }
1114
1115                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1116                         dev_kfree_skb_irq(skb);
1117                         goto next_desc;
1118                 }
1119
1120                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1121
1122                 if (!length) {
1123                         e_dbg("Last part of the packet spanning multiple "
1124                               "descriptors\n");
1125                         dev_kfree_skb_irq(skb);
1126                         goto next_desc;
1127                 }
1128
1129                 /* Good Receive */
1130                 skb_put(skb, length);
1131
1132                 {
1133                 /*
1134                  * this looks ugly, but it seems compiler issues make it
1135                  * more efficient than reusing j
1136                  */
1137                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1138
1139                 /*
1140                  * page alloc/put takes too long and effects small packet
1141                  * throughput, so unsplit small packets and save the alloc/put
1142                  * only valid in softirq (napi) context to call kmap_*
1143                  */
1144                 if (l1 && (l1 <= copybreak) &&
1145                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1146                         u8 *vaddr;
1147
1148                         ps_page = &buffer_info->ps_pages[0];
1149
1150                         /*
1151                          * there is no documentation about how to call
1152                          * kmap_atomic, so we can't hold the mapping
1153                          * very long
1154                          */
1155                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1156                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1157                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1158                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1159                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1160                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1161                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1162
1163                         /* remove the CRC */
1164                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1165                                 l1 -= 4;
1166
1167                         skb_put(skb, l1);
1168                         goto copydone;
1169                 } /* if */
1170                 }
1171
1172                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1173                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1174                         if (!length)
1175                                 break;
1176
1177                         ps_page = &buffer_info->ps_pages[j];
1178                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1179                                        DMA_FROM_DEVICE);
1180                         ps_page->dma = 0;
1181                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1182                         ps_page->page = NULL;
1183                         skb->len += length;
1184                         skb->data_len += length;
1185                         skb->truesize += length;
1186                 }
1187
1188                 /* strip the ethernet crc, problem is we're using pages now so
1189                  * this whole operation can get a little cpu intensive
1190                  */
1191                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1192                         pskb_trim(skb, skb->len - 4);
1193
1194 copydone:
1195                 total_rx_bytes += skb->len;
1196                 total_rx_packets++;
1197
1198                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1199                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1200
1201                 if (rx_desc->wb.upper.header_status &
1202                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1203                         adapter->rx_hdr_split++;
1204
1205                 e1000_receive_skb(adapter, netdev, skb,
1206                                   staterr, rx_desc->wb.middle.vlan);
1207
1208 next_desc:
1209                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1210                 buffer_info->skb = NULL;
1211
1212                 /* return some buffers to hardware, one at a time is too slow */
1213                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1214                         adapter->alloc_rx_buf(adapter, cleaned_count);
1215                         cleaned_count = 0;
1216                 }
1217
1218                 /* use prefetched values */
1219                 rx_desc = next_rxd;
1220                 buffer_info = next_buffer;
1221
1222                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1223         }
1224         rx_ring->next_to_clean = i;
1225
1226         cleaned_count = e1000_desc_unused(rx_ring);
1227         if (cleaned_count)
1228                 adapter->alloc_rx_buf(adapter, cleaned_count);
1229
1230         adapter->total_rx_bytes += total_rx_bytes;
1231         adapter->total_rx_packets += total_rx_packets;
1232         netdev->stats.rx_bytes += total_rx_bytes;
1233         netdev->stats.rx_packets += total_rx_packets;
1234         return cleaned;
1235 }
1236
1237 /**
1238  * e1000_consume_page - helper function
1239  **/
1240 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1241                                u16 length)
1242 {
1243         bi->page = NULL;
1244         skb->len += length;
1245         skb->data_len += length;
1246         skb->truesize += length;
1247 }
1248
1249 /**
1250  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1251  * @adapter: board private structure
1252  *
1253  * the return value indicates whether actual cleaning was done, there
1254  * is no guarantee that everything was cleaned
1255  **/
1256
1257 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1258                                      int *work_done, int work_to_do)
1259 {
1260         struct net_device *netdev = adapter->netdev;
1261         struct pci_dev *pdev = adapter->pdev;
1262         struct e1000_ring *rx_ring = adapter->rx_ring;
1263         struct e1000_rx_desc *rx_desc, *next_rxd;
1264         struct e1000_buffer *buffer_info, *next_buffer;
1265         u32 length;
1266         unsigned int i;
1267         int cleaned_count = 0;
1268         bool cleaned = false;
1269         unsigned int total_rx_bytes=0, total_rx_packets=0;
1270
1271         i = rx_ring->next_to_clean;
1272         rx_desc = E1000_RX_DESC(*rx_ring, i);
1273         buffer_info = &rx_ring->buffer_info[i];
1274
1275         while (rx_desc->status & E1000_RXD_STAT_DD) {
1276                 struct sk_buff *skb;
1277                 u8 status;
1278
1279                 if (*work_done >= work_to_do)
1280                         break;
1281                 (*work_done)++;
1282
1283                 status = rx_desc->status;
1284                 skb = buffer_info->skb;
1285                 buffer_info->skb = NULL;
1286
1287                 ++i;
1288                 if (i == rx_ring->count)
1289                         i = 0;
1290                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1291                 prefetch(next_rxd);
1292
1293                 next_buffer = &rx_ring->buffer_info[i];
1294
1295                 cleaned = true;
1296                 cleaned_count++;
1297                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1298                                DMA_FROM_DEVICE);
1299                 buffer_info->dma = 0;
1300
1301                 length = le16_to_cpu(rx_desc->length);
1302
1303                 /* errors is only valid for DD + EOP descriptors */
1304                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1305                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1306                                 /* recycle both page and skb */
1307                                 buffer_info->skb = skb;
1308                                 /* an error means any chain goes out the window
1309                                  * too */
1310                                 if (rx_ring->rx_skb_top)
1311                                         dev_kfree_skb(rx_ring->rx_skb_top);
1312                                 rx_ring->rx_skb_top = NULL;
1313                                 goto next_desc;
1314                 }
1315
1316 #define rxtop rx_ring->rx_skb_top
1317                 if (!(status & E1000_RXD_STAT_EOP)) {
1318                         /* this descriptor is only the beginning (or middle) */
1319                         if (!rxtop) {
1320                                 /* this is the beginning of a chain */
1321                                 rxtop = skb;
1322                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1323                                                    0, length);
1324                         } else {
1325                                 /* this is the middle of a chain */
1326                                 skb_fill_page_desc(rxtop,
1327                                     skb_shinfo(rxtop)->nr_frags,
1328                                     buffer_info->page, 0, length);
1329                                 /* re-use the skb, only consumed the page */
1330                                 buffer_info->skb = skb;
1331                         }
1332                         e1000_consume_page(buffer_info, rxtop, length);
1333                         goto next_desc;
1334                 } else {
1335                         if (rxtop) {
1336                                 /* end of the chain */
1337                                 skb_fill_page_desc(rxtop,
1338                                     skb_shinfo(rxtop)->nr_frags,
1339                                     buffer_info->page, 0, length);
1340                                 /* re-use the current skb, we only consumed the
1341                                  * page */
1342                                 buffer_info->skb = skb;
1343                                 skb = rxtop;
1344                                 rxtop = NULL;
1345                                 e1000_consume_page(buffer_info, skb, length);
1346                         } else {
1347                                 /* no chain, got EOP, this buf is the packet
1348                                  * copybreak to save the put_page/alloc_page */
1349                                 if (length <= copybreak &&
1350                                     skb_tailroom(skb) >= length) {
1351                                         u8 *vaddr;
1352                                         vaddr = kmap_atomic(buffer_info->page,
1353                                                            KM_SKB_DATA_SOFTIRQ);
1354                                         memcpy(skb_tail_pointer(skb), vaddr,
1355                                                length);
1356                                         kunmap_atomic(vaddr,
1357                                                       KM_SKB_DATA_SOFTIRQ);
1358                                         /* re-use the page, so don't erase
1359                                          * buffer_info->page */
1360                                         skb_put(skb, length);
1361                                 } else {
1362                                         skb_fill_page_desc(skb, 0,
1363                                                            buffer_info->page, 0,
1364                                                            length);
1365                                         e1000_consume_page(buffer_info, skb,
1366                                                            length);
1367                                 }
1368                         }
1369                 }
1370
1371                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1372                 e1000_rx_checksum(adapter,
1373                                   (u32)(status) |
1374                                   ((u32)(rx_desc->errors) << 24),
1375                                   le16_to_cpu(rx_desc->csum), skb);
1376
1377                 /* probably a little skewed due to removing CRC */
1378                 total_rx_bytes += skb->len;
1379                 total_rx_packets++;
1380
1381                 /* eth type trans needs skb->data to point to something */
1382                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1383                         e_err("pskb_may_pull failed.\n");
1384                         dev_kfree_skb(skb);
1385                         goto next_desc;
1386                 }
1387
1388                 e1000_receive_skb(adapter, netdev, skb, status,
1389                                   rx_desc->special);
1390
1391 next_desc:
1392                 rx_desc->status = 0;
1393
1394                 /* return some buffers to hardware, one at a time is too slow */
1395                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1396                         adapter->alloc_rx_buf(adapter, cleaned_count);
1397                         cleaned_count = 0;
1398                 }
1399
1400                 /* use prefetched values */
1401                 rx_desc = next_rxd;
1402                 buffer_info = next_buffer;
1403         }
1404         rx_ring->next_to_clean = i;
1405
1406         cleaned_count = e1000_desc_unused(rx_ring);
1407         if (cleaned_count)
1408                 adapter->alloc_rx_buf(adapter, cleaned_count);
1409
1410         adapter->total_rx_bytes += total_rx_bytes;
1411         adapter->total_rx_packets += total_rx_packets;
1412         netdev->stats.rx_bytes += total_rx_bytes;
1413         netdev->stats.rx_packets += total_rx_packets;
1414         return cleaned;
1415 }
1416
1417 /**
1418  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1419  * @adapter: board private structure
1420  **/
1421 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1422 {
1423         struct e1000_ring *rx_ring = adapter->rx_ring;
1424         struct e1000_buffer *buffer_info;
1425         struct e1000_ps_page *ps_page;
1426         struct pci_dev *pdev = adapter->pdev;
1427         unsigned int i, j;
1428
1429         /* Free all the Rx ring sk_buffs */
1430         for (i = 0; i < rx_ring->count; i++) {
1431                 buffer_info = &rx_ring->buffer_info[i];
1432                 if (buffer_info->dma) {
1433                         if (adapter->clean_rx == e1000_clean_rx_irq)
1434                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1435                                                  adapter->rx_buffer_len,
1436                                                  DMA_FROM_DEVICE);
1437                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1438                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1439                                                PAGE_SIZE,
1440                                                DMA_FROM_DEVICE);
1441                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1442                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1443                                                  adapter->rx_ps_bsize0,
1444                                                  DMA_FROM_DEVICE);
1445                         buffer_info->dma = 0;
1446                 }
1447
1448                 if (buffer_info->page) {
1449                         put_page(buffer_info->page);
1450                         buffer_info->page = NULL;
1451                 }
1452
1453                 if (buffer_info->skb) {
1454                         dev_kfree_skb(buffer_info->skb);
1455                         buffer_info->skb = NULL;
1456                 }
1457
1458                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1459                         ps_page = &buffer_info->ps_pages[j];
1460                         if (!ps_page->page)
1461                                 break;
1462                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1463                                        DMA_FROM_DEVICE);
1464                         ps_page->dma = 0;
1465                         put_page(ps_page->page);
1466                         ps_page->page = NULL;
1467                 }
1468         }
1469
1470         /* there also may be some cached data from a chained receive */
1471         if (rx_ring->rx_skb_top) {
1472                 dev_kfree_skb(rx_ring->rx_skb_top);
1473                 rx_ring->rx_skb_top = NULL;
1474         }
1475
1476         /* Zero out the descriptor ring */
1477         memset(rx_ring->desc, 0, rx_ring->size);
1478
1479         rx_ring->next_to_clean = 0;
1480         rx_ring->next_to_use = 0;
1481         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1482
1483         writel(0, adapter->hw.hw_addr + rx_ring->head);
1484         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1485 }
1486
1487 static void e1000e_downshift_workaround(struct work_struct *work)
1488 {
1489         struct e1000_adapter *adapter = container_of(work,
1490                                         struct e1000_adapter, downshift_task);
1491
1492         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1493 }
1494
1495 /**
1496  * e1000_intr_msi - Interrupt Handler
1497  * @irq: interrupt number
1498  * @data: pointer to a network interface device structure
1499  **/
1500 static irqreturn_t e1000_intr_msi(int irq, void *data)
1501 {
1502         struct net_device *netdev = data;
1503         struct e1000_adapter *adapter = netdev_priv(netdev);
1504         struct e1000_hw *hw = &adapter->hw;
1505         u32 icr = er32(ICR);
1506
1507         /*
1508          * read ICR disables interrupts using IAM
1509          */
1510
1511         if (icr & E1000_ICR_LSC) {
1512                 hw->mac.get_link_status = 1;
1513                 /*
1514                  * ICH8 workaround-- Call gig speed drop workaround on cable
1515                  * disconnect (LSC) before accessing any PHY registers
1516                  */
1517                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1518                     (!(er32(STATUS) & E1000_STATUS_LU)))
1519                         schedule_work(&adapter->downshift_task);
1520
1521                 /*
1522                  * 80003ES2LAN workaround-- For packet buffer work-around on
1523                  * link down event; disable receives here in the ISR and reset
1524                  * adapter in watchdog
1525                  */
1526                 if (netif_carrier_ok(netdev) &&
1527                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1528                         /* disable receives */
1529                         u32 rctl = er32(RCTL);
1530                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1531                         adapter->flags |= FLAG_RX_RESTART_NOW;
1532                 }
1533                 /* guard against interrupt when we're going down */
1534                 if (!test_bit(__E1000_DOWN, &adapter->state))
1535                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1536         }
1537
1538         if (napi_schedule_prep(&adapter->napi)) {
1539                 adapter->total_tx_bytes = 0;
1540                 adapter->total_tx_packets = 0;
1541                 adapter->total_rx_bytes = 0;
1542                 adapter->total_rx_packets = 0;
1543                 __napi_schedule(&adapter->napi);
1544         }
1545
1546         return IRQ_HANDLED;
1547 }
1548
1549 /**
1550  * e1000_intr - Interrupt Handler
1551  * @irq: interrupt number
1552  * @data: pointer to a network interface device structure
1553  **/
1554 static irqreturn_t e1000_intr(int irq, void *data)
1555 {
1556         struct net_device *netdev = data;
1557         struct e1000_adapter *adapter = netdev_priv(netdev);
1558         struct e1000_hw *hw = &adapter->hw;
1559         u32 rctl, icr = er32(ICR);
1560
1561         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1562                 return IRQ_NONE;  /* Not our interrupt */
1563
1564         /*
1565          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1566          * not set, then the adapter didn't send an interrupt
1567          */
1568         if (!(icr & E1000_ICR_INT_ASSERTED))
1569                 return IRQ_NONE;
1570
1571         /*
1572          * Interrupt Auto-Mask...upon reading ICR,
1573          * interrupts are masked.  No need for the
1574          * IMC write
1575          */
1576
1577         if (icr & E1000_ICR_LSC) {
1578                 hw->mac.get_link_status = 1;
1579                 /*
1580                  * ICH8 workaround-- Call gig speed drop workaround on cable
1581                  * disconnect (LSC) before accessing any PHY registers
1582                  */
1583                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1584                     (!(er32(STATUS) & E1000_STATUS_LU)))
1585                         schedule_work(&adapter->downshift_task);
1586
1587                 /*
1588                  * 80003ES2LAN workaround--
1589                  * For packet buffer work-around on link down event;
1590                  * disable receives here in the ISR and
1591                  * reset adapter in watchdog
1592                  */
1593                 if (netif_carrier_ok(netdev) &&
1594                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1595                         /* disable receives */
1596                         rctl = er32(RCTL);
1597                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1598                         adapter->flags |= FLAG_RX_RESTART_NOW;
1599                 }
1600                 /* guard against interrupt when we're going down */
1601                 if (!test_bit(__E1000_DOWN, &adapter->state))
1602                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1603         }
1604
1605         if (napi_schedule_prep(&adapter->napi)) {
1606                 adapter->total_tx_bytes = 0;
1607                 adapter->total_tx_packets = 0;
1608                 adapter->total_rx_bytes = 0;
1609                 adapter->total_rx_packets = 0;
1610                 __napi_schedule(&adapter->napi);
1611         }
1612
1613         return IRQ_HANDLED;
1614 }
1615
1616 static irqreturn_t e1000_msix_other(int irq, void *data)
1617 {
1618         struct net_device *netdev = data;
1619         struct e1000_adapter *adapter = netdev_priv(netdev);
1620         struct e1000_hw *hw = &adapter->hw;
1621         u32 icr = er32(ICR);
1622
1623         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1624                 if (!test_bit(__E1000_DOWN, &adapter->state))
1625                         ew32(IMS, E1000_IMS_OTHER);
1626                 return IRQ_NONE;
1627         }
1628
1629         if (icr & adapter->eiac_mask)
1630                 ew32(ICS, (icr & adapter->eiac_mask));
1631
1632         if (icr & E1000_ICR_OTHER) {
1633                 if (!(icr & E1000_ICR_LSC))
1634                         goto no_link_interrupt;
1635                 hw->mac.get_link_status = 1;
1636                 /* guard against interrupt when we're going down */
1637                 if (!test_bit(__E1000_DOWN, &adapter->state))
1638                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1639         }
1640
1641 no_link_interrupt:
1642         if (!test_bit(__E1000_DOWN, &adapter->state))
1643                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1644
1645         return IRQ_HANDLED;
1646 }
1647
1648
1649 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1650 {
1651         struct net_device *netdev = data;
1652         struct e1000_adapter *adapter = netdev_priv(netdev);
1653         struct e1000_hw *hw = &adapter->hw;
1654         struct e1000_ring *tx_ring = adapter->tx_ring;
1655
1656
1657         adapter->total_tx_bytes = 0;
1658         adapter->total_tx_packets = 0;
1659
1660         if (!e1000_clean_tx_irq(adapter))
1661                 /* Ring was not completely cleaned, so fire another interrupt */
1662                 ew32(ICS, tx_ring->ims_val);
1663
1664         return IRQ_HANDLED;
1665 }
1666
1667 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1668 {
1669         struct net_device *netdev = data;
1670         struct e1000_adapter *adapter = netdev_priv(netdev);
1671
1672         /* Write the ITR value calculated at the end of the
1673          * previous interrupt.
1674          */
1675         if (adapter->rx_ring->set_itr) {
1676                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1677                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1678                 adapter->rx_ring->set_itr = 0;
1679         }
1680
1681         if (napi_schedule_prep(&adapter->napi)) {
1682                 adapter->total_rx_bytes = 0;
1683                 adapter->total_rx_packets = 0;
1684                 __napi_schedule(&adapter->napi);
1685         }
1686         return IRQ_HANDLED;
1687 }
1688
1689 /**
1690  * e1000_configure_msix - Configure MSI-X hardware
1691  *
1692  * e1000_configure_msix sets up the hardware to properly
1693  * generate MSI-X interrupts.
1694  **/
1695 static void e1000_configure_msix(struct e1000_adapter *adapter)
1696 {
1697         struct e1000_hw *hw = &adapter->hw;
1698         struct e1000_ring *rx_ring = adapter->rx_ring;
1699         struct e1000_ring *tx_ring = adapter->tx_ring;
1700         int vector = 0;
1701         u32 ctrl_ext, ivar = 0;
1702
1703         adapter->eiac_mask = 0;
1704
1705         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1706         if (hw->mac.type == e1000_82574) {
1707                 u32 rfctl = er32(RFCTL);
1708                 rfctl |= E1000_RFCTL_ACK_DIS;
1709                 ew32(RFCTL, rfctl);
1710         }
1711
1712 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1713         /* Configure Rx vector */
1714         rx_ring->ims_val = E1000_IMS_RXQ0;
1715         adapter->eiac_mask |= rx_ring->ims_val;
1716         if (rx_ring->itr_val)
1717                 writel(1000000000 / (rx_ring->itr_val * 256),
1718                        hw->hw_addr + rx_ring->itr_register);
1719         else
1720                 writel(1, hw->hw_addr + rx_ring->itr_register);
1721         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1722
1723         /* Configure Tx vector */
1724         tx_ring->ims_val = E1000_IMS_TXQ0;
1725         vector++;
1726         if (tx_ring->itr_val)
1727                 writel(1000000000 / (tx_ring->itr_val * 256),
1728                        hw->hw_addr + tx_ring->itr_register);
1729         else
1730                 writel(1, hw->hw_addr + tx_ring->itr_register);
1731         adapter->eiac_mask |= tx_ring->ims_val;
1732         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1733
1734         /* set vector for Other Causes, e.g. link changes */
1735         vector++;
1736         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1737         if (rx_ring->itr_val)
1738                 writel(1000000000 / (rx_ring->itr_val * 256),
1739                        hw->hw_addr + E1000_EITR_82574(vector));
1740         else
1741                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1742
1743         /* Cause Tx interrupts on every write back */
1744         ivar |= (1 << 31);
1745
1746         ew32(IVAR, ivar);
1747
1748         /* enable MSI-X PBA support */
1749         ctrl_ext = er32(CTRL_EXT);
1750         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1751
1752         /* Auto-Mask Other interrupts upon ICR read */
1753 #define E1000_EIAC_MASK_82574   0x01F00000
1754         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1755         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1756         ew32(CTRL_EXT, ctrl_ext);
1757         e1e_flush();
1758 }
1759
1760 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1761 {
1762         if (adapter->msix_entries) {
1763                 pci_disable_msix(adapter->pdev);
1764                 kfree(adapter->msix_entries);
1765                 adapter->msix_entries = NULL;
1766         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1767                 pci_disable_msi(adapter->pdev);
1768                 adapter->flags &= ~FLAG_MSI_ENABLED;
1769         }
1770
1771         return;
1772 }
1773
1774 /**
1775  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1776  *
1777  * Attempt to configure interrupts using the best available
1778  * capabilities of the hardware and kernel.
1779  **/
1780 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1781 {
1782         int err;
1783         int numvecs, i;
1784
1785
1786         switch (adapter->int_mode) {
1787         case E1000E_INT_MODE_MSIX:
1788                 if (adapter->flags & FLAG_HAS_MSIX) {
1789                         numvecs = 3; /* RxQ0, TxQ0 and other */
1790                         adapter->msix_entries = kcalloc(numvecs,
1791                                                       sizeof(struct msix_entry),
1792                                                       GFP_KERNEL);
1793                         if (adapter->msix_entries) {
1794                                 for (i = 0; i < numvecs; i++)
1795                                         adapter->msix_entries[i].entry = i;
1796
1797                                 err = pci_enable_msix(adapter->pdev,
1798                                                       adapter->msix_entries,
1799                                                       numvecs);
1800                                 if (err == 0)
1801                                         return;
1802                         }
1803                         /* MSI-X failed, so fall through and try MSI */
1804                         e_err("Failed to initialize MSI-X interrupts.  "
1805                               "Falling back to MSI interrupts.\n");
1806                         e1000e_reset_interrupt_capability(adapter);
1807                 }
1808                 adapter->int_mode = E1000E_INT_MODE_MSI;
1809                 /* Fall through */
1810         case E1000E_INT_MODE_MSI:
1811                 if (!pci_enable_msi(adapter->pdev)) {
1812                         adapter->flags |= FLAG_MSI_ENABLED;
1813                 } else {
1814                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1815                         e_err("Failed to initialize MSI interrupts.  Falling "
1816                               "back to legacy interrupts.\n");
1817                 }
1818                 /* Fall through */
1819         case E1000E_INT_MODE_LEGACY:
1820                 /* Don't do anything; this is the system default */
1821                 break;
1822         }
1823
1824         return;
1825 }
1826
1827 /**
1828  * e1000_request_msix - Initialize MSI-X interrupts
1829  *
1830  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1831  * kernel.
1832  **/
1833 static int e1000_request_msix(struct e1000_adapter *adapter)
1834 {
1835         struct net_device *netdev = adapter->netdev;
1836         int err = 0, vector = 0;
1837
1838         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1839                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1840         else
1841                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1842         err = request_irq(adapter->msix_entries[vector].vector,
1843                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1844                           netdev);
1845         if (err)
1846                 goto out;
1847         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1848         adapter->rx_ring->itr_val = adapter->itr;
1849         vector++;
1850
1851         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1852                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1853         else
1854                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1855         err = request_irq(adapter->msix_entries[vector].vector,
1856                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1857                           netdev);
1858         if (err)
1859                 goto out;
1860         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1861         adapter->tx_ring->itr_val = adapter->itr;
1862         vector++;
1863
1864         err = request_irq(adapter->msix_entries[vector].vector,
1865                           e1000_msix_other, 0, netdev->name, netdev);
1866         if (err)
1867                 goto out;
1868
1869         e1000_configure_msix(adapter);
1870         return 0;
1871 out:
1872         return err;
1873 }
1874
1875 /**
1876  * e1000_request_irq - initialize interrupts
1877  *
1878  * Attempts to configure interrupts using the best available
1879  * capabilities of the hardware and kernel.
1880  **/
1881 static int e1000_request_irq(struct e1000_adapter *adapter)
1882 {
1883         struct net_device *netdev = adapter->netdev;
1884         int err;
1885
1886         if (adapter->msix_entries) {
1887                 err = e1000_request_msix(adapter);
1888                 if (!err)
1889                         return err;
1890                 /* fall back to MSI */
1891                 e1000e_reset_interrupt_capability(adapter);
1892                 adapter->int_mode = E1000E_INT_MODE_MSI;
1893                 e1000e_set_interrupt_capability(adapter);
1894         }
1895         if (adapter->flags & FLAG_MSI_ENABLED) {
1896                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1897                                   netdev->name, netdev);
1898                 if (!err)
1899                         return err;
1900
1901                 /* fall back to legacy interrupt */
1902                 e1000e_reset_interrupt_capability(adapter);
1903                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1904         }
1905
1906         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1907                           netdev->name, netdev);
1908         if (err)
1909                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1910
1911         return err;
1912 }
1913
1914 static void e1000_free_irq(struct e1000_adapter *adapter)
1915 {
1916         struct net_device *netdev = adapter->netdev;
1917
1918         if (adapter->msix_entries) {
1919                 int vector = 0;
1920
1921                 free_irq(adapter->msix_entries[vector].vector, netdev);
1922                 vector++;
1923
1924                 free_irq(adapter->msix_entries[vector].vector, netdev);
1925                 vector++;
1926
1927                 /* Other Causes interrupt vector */
1928                 free_irq(adapter->msix_entries[vector].vector, netdev);
1929                 return;
1930         }
1931
1932         free_irq(adapter->pdev->irq, netdev);
1933 }
1934
1935 /**
1936  * e1000_irq_disable - Mask off interrupt generation on the NIC
1937  **/
1938 static void e1000_irq_disable(struct e1000_adapter *adapter)
1939 {
1940         struct e1000_hw *hw = &adapter->hw;
1941
1942         ew32(IMC, ~0);
1943         if (adapter->msix_entries)
1944                 ew32(EIAC_82574, 0);
1945         e1e_flush();
1946         synchronize_irq(adapter->pdev->irq);
1947 }
1948
1949 /**
1950  * e1000_irq_enable - Enable default interrupt generation settings
1951  **/
1952 static void e1000_irq_enable(struct e1000_adapter *adapter)
1953 {
1954         struct e1000_hw *hw = &adapter->hw;
1955
1956         if (adapter->msix_entries) {
1957                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1958                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1959         } else {
1960                 ew32(IMS, IMS_ENABLE_MASK);
1961         }
1962         e1e_flush();
1963 }
1964
1965 /**
1966  * e1000_get_hw_control - get control of the h/w from f/w
1967  * @adapter: address of board private structure
1968  *
1969  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1970  * For ASF and Pass Through versions of f/w this means that
1971  * the driver is loaded. For AMT version (only with 82573)
1972  * of the f/w this means that the network i/f is open.
1973  **/
1974 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1975 {
1976         struct e1000_hw *hw = &adapter->hw;
1977         u32 ctrl_ext;
1978         u32 swsm;
1979
1980         /* Let firmware know the driver has taken over */
1981         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1982                 swsm = er32(SWSM);
1983                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1984         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1985                 ctrl_ext = er32(CTRL_EXT);
1986                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1987         }
1988 }
1989
1990 /**
1991  * e1000_release_hw_control - release control of the h/w to f/w
1992  * @adapter: address of board private structure
1993  *
1994  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1995  * For ASF and Pass Through versions of f/w this means that the
1996  * driver is no longer loaded. For AMT version (only with 82573) i
1997  * of the f/w this means that the network i/f is closed.
1998  *
1999  **/
2000 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2001 {
2002         struct e1000_hw *hw = &adapter->hw;
2003         u32 ctrl_ext;
2004         u32 swsm;
2005
2006         /* Let firmware taken over control of h/w */
2007         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2008                 swsm = er32(SWSM);
2009                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2010         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2011                 ctrl_ext = er32(CTRL_EXT);
2012                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2013         }
2014 }
2015
2016 /**
2017  * @e1000_alloc_ring - allocate memory for a ring structure
2018  **/
2019 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2020                                 struct e1000_ring *ring)
2021 {
2022         struct pci_dev *pdev = adapter->pdev;
2023
2024         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2025                                         GFP_KERNEL);
2026         if (!ring->desc)
2027                 return -ENOMEM;
2028
2029         return 0;
2030 }
2031
2032 /**
2033  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2034  * @adapter: board private structure
2035  *
2036  * Return 0 on success, negative on failure
2037  **/
2038 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2039 {
2040         struct e1000_ring *tx_ring = adapter->tx_ring;
2041         int err = -ENOMEM, size;
2042
2043         size = sizeof(struct e1000_buffer) * tx_ring->count;
2044         tx_ring->buffer_info = vmalloc(size);
2045         if (!tx_ring->buffer_info)
2046                 goto err;
2047         memset(tx_ring->buffer_info, 0, size);
2048
2049         /* round up to nearest 4K */
2050         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2051         tx_ring->size = ALIGN(tx_ring->size, 4096);
2052
2053         err = e1000_alloc_ring_dma(adapter, tx_ring);
2054         if (err)
2055                 goto err;
2056
2057         tx_ring->next_to_use = 0;
2058         tx_ring->next_to_clean = 0;
2059
2060         return 0;
2061 err:
2062         vfree(tx_ring->buffer_info);
2063         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2064         return err;
2065 }
2066
2067 /**
2068  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2069  * @adapter: board private structure
2070  *
2071  * Returns 0 on success, negative on failure
2072  **/
2073 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2074 {
2075         struct e1000_ring *rx_ring = adapter->rx_ring;
2076         struct e1000_buffer *buffer_info;
2077         int i, size, desc_len, err = -ENOMEM;
2078
2079         size = sizeof(struct e1000_buffer) * rx_ring->count;
2080         rx_ring->buffer_info = vmalloc(size);
2081         if (!rx_ring->buffer_info)
2082                 goto err;
2083         memset(rx_ring->buffer_info, 0, size);
2084
2085         for (i = 0; i < rx_ring->count; i++) {
2086                 buffer_info = &rx_ring->buffer_info[i];
2087                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2088                                                 sizeof(struct e1000_ps_page),
2089                                                 GFP_KERNEL);
2090                 if (!buffer_info->ps_pages)
2091                         goto err_pages;
2092         }
2093
2094         desc_len = sizeof(union e1000_rx_desc_packet_split);
2095
2096         /* Round up to nearest 4K */
2097         rx_ring->size = rx_ring->count * desc_len;
2098         rx_ring->size = ALIGN(rx_ring->size, 4096);
2099
2100         err = e1000_alloc_ring_dma(adapter, rx_ring);
2101         if (err)
2102                 goto err_pages;
2103
2104         rx_ring->next_to_clean = 0;
2105         rx_ring->next_to_use = 0;
2106         rx_ring->rx_skb_top = NULL;
2107
2108         return 0;
2109
2110 err_pages:
2111         for (i = 0; i < rx_ring->count; i++) {
2112                 buffer_info = &rx_ring->buffer_info[i];
2113                 kfree(buffer_info->ps_pages);
2114         }
2115 err:
2116         vfree(rx_ring->buffer_info);
2117         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2118         return err;
2119 }
2120
2121 /**
2122  * e1000_clean_tx_ring - Free Tx Buffers
2123  * @adapter: board private structure
2124  **/
2125 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2126 {
2127         struct e1000_ring *tx_ring = adapter->tx_ring;
2128         struct e1000_buffer *buffer_info;
2129         unsigned long size;
2130         unsigned int i;
2131
2132         for (i = 0; i < tx_ring->count; i++) {
2133                 buffer_info = &tx_ring->buffer_info[i];
2134                 e1000_put_txbuf(adapter, buffer_info);
2135         }
2136
2137         size = sizeof(struct e1000_buffer) * tx_ring->count;
2138         memset(tx_ring->buffer_info, 0, size);
2139
2140         memset(tx_ring->desc, 0, tx_ring->size);
2141
2142         tx_ring->next_to_use = 0;
2143         tx_ring->next_to_clean = 0;
2144
2145         writel(0, adapter->hw.hw_addr + tx_ring->head);
2146         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2147 }
2148
2149 /**
2150  * e1000e_free_tx_resources - Free Tx Resources per Queue
2151  * @adapter: board private structure
2152  *
2153  * Free all transmit software resources
2154  **/
2155 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2156 {
2157         struct pci_dev *pdev = adapter->pdev;
2158         struct e1000_ring *tx_ring = adapter->tx_ring;
2159
2160         e1000_clean_tx_ring(adapter);
2161
2162         vfree(tx_ring->buffer_info);
2163         tx_ring->buffer_info = NULL;
2164
2165         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2166                           tx_ring->dma);
2167         tx_ring->desc = NULL;
2168 }
2169
2170 /**
2171  * e1000e_free_rx_resources - Free Rx Resources
2172  * @adapter: board private structure
2173  *
2174  * Free all receive software resources
2175  **/
2176
2177 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2178 {
2179         struct pci_dev *pdev = adapter->pdev;
2180         struct e1000_ring *rx_ring = adapter->rx_ring;
2181         int i;
2182
2183         e1000_clean_rx_ring(adapter);
2184
2185         for (i = 0; i < rx_ring->count; i++) {
2186                 kfree(rx_ring->buffer_info[i].ps_pages);
2187         }
2188
2189         vfree(rx_ring->buffer_info);
2190         rx_ring->buffer_info = NULL;
2191
2192         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2193                           rx_ring->dma);
2194         rx_ring->desc = NULL;
2195 }
2196
2197 /**
2198  * e1000_update_itr - update the dynamic ITR value based on statistics
2199  * @adapter: pointer to adapter
2200  * @itr_setting: current adapter->itr
2201  * @packets: the number of packets during this measurement interval
2202  * @bytes: the number of bytes during this measurement interval
2203  *
2204  *      Stores a new ITR value based on packets and byte
2205  *      counts during the last interrupt.  The advantage of per interrupt
2206  *      computation is faster updates and more accurate ITR for the current
2207  *      traffic pattern.  Constants in this function were computed
2208  *      based on theoretical maximum wire speed and thresholds were set based
2209  *      on testing data as well as attempting to minimize response time
2210  *      while increasing bulk throughput.  This functionality is controlled
2211  *      by the InterruptThrottleRate module parameter.
2212  **/
2213 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2214                                      u16 itr_setting, int packets,
2215                                      int bytes)
2216 {
2217         unsigned int retval = itr_setting;
2218
2219         if (packets == 0)
2220                 goto update_itr_done;
2221
2222         switch (itr_setting) {
2223         case lowest_latency:
2224                 /* handle TSO and jumbo frames */
2225                 if (bytes/packets > 8000)
2226                         retval = bulk_latency;
2227                 else if ((packets < 5) && (bytes > 512)) {
2228                         retval = low_latency;
2229                 }
2230                 break;
2231         case low_latency:  /* 50 usec aka 20000 ints/s */
2232                 if (bytes > 10000) {
2233                         /* this if handles the TSO accounting */
2234                         if (bytes/packets > 8000) {
2235                                 retval = bulk_latency;
2236                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2237                                 retval = bulk_latency;
2238                         } else if ((packets > 35)) {
2239                                 retval = lowest_latency;
2240                         }
2241                 } else if (bytes/packets > 2000) {
2242                         retval = bulk_latency;
2243                 } else if (packets <= 2 && bytes < 512) {
2244                         retval = lowest_latency;
2245                 }
2246                 break;
2247         case bulk_latency: /* 250 usec aka 4000 ints/s */
2248                 if (bytes > 25000) {
2249                         if (packets > 35) {
2250                                 retval = low_latency;
2251                         }
2252                 } else if (bytes < 6000) {
2253                         retval = low_latency;
2254                 }
2255                 break;
2256         }
2257
2258 update_itr_done:
2259         return retval;
2260 }
2261
2262 static void e1000_set_itr(struct e1000_adapter *adapter)
2263 {
2264         struct e1000_hw *hw = &adapter->hw;
2265         u16 current_itr;
2266         u32 new_itr = adapter->itr;
2267
2268         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2269         if (adapter->link_speed != SPEED_1000) {
2270                 current_itr = 0;
2271                 new_itr = 4000;
2272                 goto set_itr_now;
2273         }
2274
2275         adapter->tx_itr = e1000_update_itr(adapter,
2276                                     adapter->tx_itr,
2277                                     adapter->total_tx_packets,
2278                                     adapter->total_tx_bytes);
2279         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2280         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2281                 adapter->tx_itr = low_latency;
2282
2283         adapter->rx_itr = e1000_update_itr(adapter,
2284                                     adapter->rx_itr,
2285                                     adapter->total_rx_packets,
2286                                     adapter->total_rx_bytes);
2287         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2288         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2289                 adapter->rx_itr = low_latency;
2290
2291         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2292
2293         switch (current_itr) {
2294         /* counts and packets in update_itr are dependent on these numbers */
2295         case lowest_latency:
2296                 new_itr = 70000;
2297                 break;
2298         case low_latency:
2299                 new_itr = 20000; /* aka hwitr = ~200 */
2300                 break;
2301         case bulk_latency:
2302                 new_itr = 4000;
2303                 break;
2304         default:
2305                 break;
2306         }
2307
2308 set_itr_now:
2309         if (new_itr != adapter->itr) {
2310                 /*
2311                  * this attempts to bias the interrupt rate towards Bulk
2312                  * by adding intermediate steps when interrupt rate is
2313                  * increasing
2314                  */
2315                 new_itr = new_itr > adapter->itr ?
2316                              min(adapter->itr + (new_itr >> 2), new_itr) :
2317                              new_itr;
2318                 adapter->itr = new_itr;
2319                 adapter->rx_ring->itr_val = new_itr;
2320                 if (adapter->msix_entries)
2321                         adapter->rx_ring->set_itr = 1;
2322                 else
2323                         ew32(ITR, 1000000000 / (new_itr * 256));
2324         }
2325 }
2326
2327 /**
2328  * e1000_alloc_queues - Allocate memory for all rings
2329  * @adapter: board private structure to initialize
2330  **/
2331 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2332 {
2333         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2334         if (!adapter->tx_ring)
2335                 goto err;
2336
2337         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2338         if (!adapter->rx_ring)
2339                 goto err;
2340
2341         return 0;
2342 err:
2343         e_err("Unable to allocate memory for queues\n");
2344         kfree(adapter->rx_ring);
2345         kfree(adapter->tx_ring);
2346         return -ENOMEM;
2347 }
2348
2349 /**
2350  * e1000_clean - NAPI Rx polling callback
2351  * @napi: struct associated with this polling callback
2352  * @budget: amount of packets driver is allowed to process this poll
2353  **/
2354 static int e1000_clean(struct napi_struct *napi, int budget)
2355 {
2356         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2357         struct e1000_hw *hw = &adapter->hw;
2358         struct net_device *poll_dev = adapter->netdev;
2359         int tx_cleaned = 1, work_done = 0;
2360
2361         adapter = netdev_priv(poll_dev);
2362
2363         if (adapter->msix_entries &&
2364             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2365                 goto clean_rx;
2366
2367         tx_cleaned = e1000_clean_tx_irq(adapter);
2368
2369 clean_rx:
2370         adapter->clean_rx(adapter, &work_done, budget);
2371
2372         if (!tx_cleaned)
2373                 work_done = budget;
2374
2375         /* If budget not fully consumed, exit the polling mode */
2376         if (work_done < budget) {
2377                 if (adapter->itr_setting & 3)
2378                         e1000_set_itr(adapter);
2379                 napi_complete(napi);
2380                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2381                         if (adapter->msix_entries)
2382                                 ew32(IMS, adapter->rx_ring->ims_val);
2383                         else
2384                                 e1000_irq_enable(adapter);
2385                 }
2386         }
2387
2388         return work_done;
2389 }
2390
2391 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2392 {
2393         struct e1000_adapter *adapter = netdev_priv(netdev);
2394         struct e1000_hw *hw = &adapter->hw;
2395         u32 vfta, index;
2396
2397         /* don't update vlan cookie if already programmed */
2398         if ((adapter->hw.mng_cookie.status &
2399              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2400             (vid == adapter->mng_vlan_id))
2401                 return;
2402
2403         /* add VID to filter table */
2404         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2405                 index = (vid >> 5) & 0x7F;
2406                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2407                 vfta |= (1 << (vid & 0x1F));
2408                 hw->mac.ops.write_vfta(hw, index, vfta);
2409         }
2410 }
2411
2412 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2413 {
2414         struct e1000_adapter *adapter = netdev_priv(netdev);
2415         struct e1000_hw *hw = &adapter->hw;
2416         u32 vfta, index;
2417
2418         if (!test_bit(__E1000_DOWN, &adapter->state))
2419                 e1000_irq_disable(adapter);
2420         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2421
2422         if (!test_bit(__E1000_DOWN, &adapter->state))
2423                 e1000_irq_enable(adapter);
2424
2425         if ((adapter->hw.mng_cookie.status &
2426              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2427             (vid == adapter->mng_vlan_id)) {
2428                 /* release control to f/w */
2429                 e1000_release_hw_control(adapter);
2430                 return;
2431         }
2432
2433         /* remove VID from 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
2442 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2443 {
2444         struct net_device *netdev = adapter->netdev;
2445         u16 vid = adapter->hw.mng_cookie.vlan_id;
2446         u16 old_vid = adapter->mng_vlan_id;
2447
2448         if (!adapter->vlgrp)
2449                 return;
2450
2451         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2452                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2453                 if (adapter->hw.mng_cookie.status &
2454                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2455                         e1000_vlan_rx_add_vid(netdev, vid);
2456                         adapter->mng_vlan_id = vid;
2457                 }
2458
2459                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2460                                 (vid != old_vid) &&
2461                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2462                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2463         } else {
2464                 adapter->mng_vlan_id = vid;
2465         }
2466 }
2467
2468
2469 static void e1000_vlan_rx_register(struct net_device *netdev,
2470                                    struct vlan_group *grp)
2471 {
2472         struct e1000_adapter *adapter = netdev_priv(netdev);
2473         struct e1000_hw *hw = &adapter->hw;
2474         u32 ctrl, rctl;
2475
2476         if (!test_bit(__E1000_DOWN, &adapter->state))
2477                 e1000_irq_disable(adapter);
2478         adapter->vlgrp = grp;
2479
2480         if (grp) {
2481                 /* enable VLAN tag insert/strip */
2482                 ctrl = er32(CTRL);
2483                 ctrl |= E1000_CTRL_VME;
2484                 ew32(CTRL, ctrl);
2485
2486                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2487                         /* enable VLAN receive filtering */
2488                         rctl = er32(RCTL);
2489                         rctl &= ~E1000_RCTL_CFIEN;
2490                         ew32(RCTL, rctl);
2491                         e1000_update_mng_vlan(adapter);
2492                 }
2493         } else {
2494                 /* disable VLAN tag insert/strip */
2495                 ctrl = er32(CTRL);
2496                 ctrl &= ~E1000_CTRL_VME;
2497                 ew32(CTRL, ctrl);
2498
2499                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2500                         if (adapter->mng_vlan_id !=
2501                             (u16)E1000_MNG_VLAN_NONE) {
2502                                 e1000_vlan_rx_kill_vid(netdev,
2503                                                        adapter->mng_vlan_id);
2504                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2505                         }
2506                 }
2507         }
2508
2509         if (!test_bit(__E1000_DOWN, &adapter->state))
2510                 e1000_irq_enable(adapter);
2511 }
2512
2513 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2514 {
2515         u16 vid;
2516
2517         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2518
2519         if (!adapter->vlgrp)
2520                 return;
2521
2522         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2523                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2524                         continue;
2525                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2526         }
2527 }
2528
2529 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2530 {
2531         struct e1000_hw *hw = &adapter->hw;
2532         u32 manc, manc2h, mdef, i, j;
2533
2534         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2535                 return;
2536
2537         manc = er32(MANC);
2538
2539         /*
2540          * enable receiving management packets to the host. this will probably
2541          * generate destination unreachable messages from the host OS, but
2542          * the packets will be handled on SMBUS
2543          */
2544         manc |= E1000_MANC_EN_MNG2HOST;
2545         manc2h = er32(MANC2H);
2546
2547         switch (hw->mac.type) {
2548         default:
2549                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2550                 break;
2551         case e1000_82574:
2552         case e1000_82583:
2553                 /*
2554                  * Check if IPMI pass-through decision filter already exists;
2555                  * if so, enable it.
2556                  */
2557                 for (i = 0, j = 0; i < 8; i++) {
2558                         mdef = er32(MDEF(i));
2559
2560                         /* Ignore filters with anything other than IPMI ports */
2561                         if (mdef & !(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2562                                 continue;
2563
2564                         /* Enable this decision filter in MANC2H */
2565                         if (mdef)
2566                                 manc2h |= (1 << i);
2567
2568                         j |= mdef;
2569                 }
2570
2571                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2572                         break;
2573
2574                 /* Create new decision filter in an empty filter */
2575                 for (i = 0, j = 0; i < 8; i++)
2576                         if (er32(MDEF(i)) == 0) {
2577                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2578                                                E1000_MDEF_PORT_664));
2579                                 manc2h |= (1 << 1);
2580                                 j++;
2581                                 break;
2582                         }
2583
2584                 if (!j)
2585                         e_warn("Unable to create IPMI pass-through filter\n");
2586                 break;
2587         }
2588
2589         ew32(MANC2H, manc2h);
2590         ew32(MANC, manc);
2591 }
2592
2593 /**
2594  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2595  * @adapter: board private structure
2596  *
2597  * Configure the Tx unit of the MAC after a reset.
2598  **/
2599 static void e1000_configure_tx(struct e1000_adapter *adapter)
2600 {
2601         struct e1000_hw *hw = &adapter->hw;
2602         struct e1000_ring *tx_ring = adapter->tx_ring;
2603         u64 tdba;
2604         u32 tdlen, tctl, tipg, tarc;
2605         u32 ipgr1, ipgr2;
2606
2607         /* Setup the HW Tx Head and Tail descriptor pointers */
2608         tdba = tx_ring->dma;
2609         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2610         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2611         ew32(TDBAH, (tdba >> 32));
2612         ew32(TDLEN, tdlen);
2613         ew32(TDH, 0);
2614         ew32(TDT, 0);
2615         tx_ring->head = E1000_TDH;
2616         tx_ring->tail = E1000_TDT;
2617
2618         /* Set the default values for the Tx Inter Packet Gap timer */
2619         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2620         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2621         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2622
2623         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2624                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2625
2626         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2627         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2628         ew32(TIPG, tipg);
2629
2630         /* Set the Tx Interrupt Delay register */
2631         ew32(TIDV, adapter->tx_int_delay);
2632         /* Tx irq moderation */
2633         ew32(TADV, adapter->tx_abs_int_delay);
2634
2635         /* Program the Transmit Control Register */
2636         tctl = er32(TCTL);
2637         tctl &= ~E1000_TCTL_CT;
2638         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2639                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2640
2641         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2642                 tarc = er32(TARC(0));
2643                 /*
2644                  * set the speed mode bit, we'll clear it if we're not at
2645                  * gigabit link later
2646                  */
2647 #define SPEED_MODE_BIT (1 << 21)
2648                 tarc |= SPEED_MODE_BIT;
2649                 ew32(TARC(0), tarc);
2650         }
2651
2652         /* errata: program both queues to unweighted RR */
2653         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2654                 tarc = er32(TARC(0));
2655                 tarc |= 1;
2656                 ew32(TARC(0), tarc);
2657                 tarc = er32(TARC(1));
2658                 tarc |= 1;
2659                 ew32(TARC(1), tarc);
2660         }
2661
2662         /* Setup Transmit Descriptor Settings for eop descriptor */
2663         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2664
2665         /* only set IDE if we are delaying interrupts using the timers */
2666         if (adapter->tx_int_delay)
2667                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2668
2669         /* enable Report Status bit */
2670         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2671
2672         ew32(TCTL, tctl);
2673
2674         e1000e_config_collision_dist(hw);
2675 }
2676
2677 /**
2678  * e1000_setup_rctl - configure the receive control registers
2679  * @adapter: Board private structure
2680  **/
2681 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2682                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2683 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2684 {
2685         struct e1000_hw *hw = &adapter->hw;
2686         u32 rctl, rfctl;
2687         u32 psrctl = 0;
2688         u32 pages = 0;
2689
2690         /* Program MC offset vector base */
2691         rctl = er32(RCTL);
2692         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2693         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2694                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2695                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2696
2697         /* Do not Store bad packets */
2698         rctl &= ~E1000_RCTL_SBP;
2699
2700         /* Enable Long Packet receive */
2701         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2702                 rctl &= ~E1000_RCTL_LPE;
2703         else
2704                 rctl |= E1000_RCTL_LPE;
2705
2706         /* Some systems expect that the CRC is included in SMBUS traffic. The
2707          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2708          * host memory when this is enabled
2709          */
2710         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2711                 rctl |= E1000_RCTL_SECRC;
2712
2713         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2714         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2715                 u16 phy_data;
2716
2717                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2718                 phy_data &= 0xfff8;
2719                 phy_data |= (1 << 2);
2720                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2721
2722                 e1e_rphy(hw, 22, &phy_data);
2723                 phy_data &= 0x0fff;
2724                 phy_data |= (1 << 14);
2725                 e1e_wphy(hw, 0x10, 0x2823);
2726                 e1e_wphy(hw, 0x11, 0x0003);
2727                 e1e_wphy(hw, 22, phy_data);
2728         }
2729
2730         /* Setup buffer sizes */
2731         rctl &= ~E1000_RCTL_SZ_4096;
2732         rctl |= E1000_RCTL_BSEX;
2733         switch (adapter->rx_buffer_len) {
2734         case 2048:
2735         default:
2736                 rctl |= E1000_RCTL_SZ_2048;
2737                 rctl &= ~E1000_RCTL_BSEX;
2738                 break;
2739         case 4096:
2740                 rctl |= E1000_RCTL_SZ_4096;
2741                 break;
2742         case 8192:
2743                 rctl |= E1000_RCTL_SZ_8192;
2744                 break;
2745         case 16384:
2746                 rctl |= E1000_RCTL_SZ_16384;
2747                 break;
2748         }
2749
2750         /*
2751          * 82571 and greater support packet-split where the protocol
2752          * header is placed in skb->data and the packet data is
2753          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2754          * In the case of a non-split, skb->data is linearly filled,
2755          * followed by the page buffers.  Therefore, skb->data is
2756          * sized to hold the largest protocol header.
2757          *
2758          * allocations using alloc_page take too long for regular MTU
2759          * so only enable packet split for jumbo frames
2760          *
2761          * Using pages when the page size is greater than 16k wastes
2762          * a lot of memory, since we allocate 3 pages at all times
2763          * per packet.
2764          */
2765         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2766         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2767             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2768                 adapter->rx_ps_pages = pages;
2769         else
2770                 adapter->rx_ps_pages = 0;
2771
2772         if (adapter->rx_ps_pages) {
2773                 /* Configure extra packet-split registers */
2774                 rfctl = er32(RFCTL);
2775                 rfctl |= E1000_RFCTL_EXTEN;
2776                 /*
2777                  * disable packet split support for IPv6 extension headers,
2778                  * because some malformed IPv6 headers can hang the Rx
2779                  */
2780                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2781                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2782
2783                 ew32(RFCTL, rfctl);
2784
2785                 /* Enable Packet split descriptors */
2786                 rctl |= E1000_RCTL_DTYP_PS;
2787
2788                 psrctl |= adapter->rx_ps_bsize0 >>
2789                         E1000_PSRCTL_BSIZE0_SHIFT;
2790
2791                 switch (adapter->rx_ps_pages) {
2792                 case 3:
2793                         psrctl |= PAGE_SIZE <<
2794                                 E1000_PSRCTL_BSIZE3_SHIFT;
2795                 case 2:
2796                         psrctl |= PAGE_SIZE <<
2797                                 E1000_PSRCTL_BSIZE2_SHIFT;
2798                 case 1:
2799                         psrctl |= PAGE_SIZE >>
2800                                 E1000_PSRCTL_BSIZE1_SHIFT;
2801                         break;
2802                 }
2803
2804                 ew32(PSRCTL, psrctl);
2805         }
2806
2807         ew32(RCTL, rctl);
2808         /* just started the receive unit, no need to restart */
2809         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2810 }
2811
2812 /**
2813  * e1000_configure_rx - Configure Receive Unit after Reset
2814  * @adapter: board private structure
2815  *
2816  * Configure the Rx unit of the MAC after a reset.
2817  **/
2818 static void e1000_configure_rx(struct e1000_adapter *adapter)
2819 {
2820         struct e1000_hw *hw = &adapter->hw;
2821         struct e1000_ring *rx_ring = adapter->rx_ring;
2822         u64 rdba;
2823         u32 rdlen, rctl, rxcsum, ctrl_ext;
2824
2825         if (adapter->rx_ps_pages) {
2826                 /* this is a 32 byte descriptor */
2827                 rdlen = rx_ring->count *
2828                         sizeof(union e1000_rx_desc_packet_split);
2829                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2830                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2831         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2832                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2833                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2834                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2835         } else {
2836                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2837                 adapter->clean_rx = e1000_clean_rx_irq;
2838                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2839         }
2840
2841         /* disable receives while setting up the descriptors */
2842         rctl = er32(RCTL);
2843         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2844         e1e_flush();
2845         msleep(10);
2846
2847         /* set the Receive Delay Timer Register */
2848         ew32(RDTR, adapter->rx_int_delay);
2849
2850         /* irq moderation */
2851         ew32(RADV, adapter->rx_abs_int_delay);
2852         if (adapter->itr_setting != 0)
2853                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2854
2855         ctrl_ext = er32(CTRL_EXT);
2856         /* Auto-Mask interrupts upon ICR access */
2857         ctrl_ext |= E1000_CTRL_EXT_IAME;
2858         ew32(IAM, 0xffffffff);
2859         ew32(CTRL_EXT, ctrl_ext);
2860         e1e_flush();
2861
2862         /*
2863          * Setup the HW Rx Head and Tail Descriptor Pointers and
2864          * the Base and Length of the Rx Descriptor Ring
2865          */
2866         rdba = rx_ring->dma;
2867         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2868         ew32(RDBAH, (rdba >> 32));
2869         ew32(RDLEN, rdlen);
2870         ew32(RDH, 0);
2871         ew32(RDT, 0);
2872         rx_ring->head = E1000_RDH;
2873         rx_ring->tail = E1000_RDT;
2874
2875         /* Enable Receive Checksum Offload for TCP and UDP */
2876         rxcsum = er32(RXCSUM);
2877         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2878                 rxcsum |= E1000_RXCSUM_TUOFL;
2879
2880                 /*
2881                  * IPv4 payload checksum for UDP fragments must be
2882                  * used in conjunction with packet-split.
2883                  */
2884                 if (adapter->rx_ps_pages)
2885                         rxcsum |= E1000_RXCSUM_IPPCSE;
2886         } else {
2887                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2888                 /* no need to clear IPPCSE as it defaults to 0 */
2889         }
2890         ew32(RXCSUM, rxcsum);
2891
2892         /*
2893          * Enable early receives on supported devices, only takes effect when
2894          * packet size is equal or larger than the specified value (in 8 byte
2895          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2896          */
2897         if (adapter->flags & FLAG_HAS_ERT) {
2898                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2899                         u32 rxdctl = er32(RXDCTL(0));
2900                         ew32(RXDCTL(0), rxdctl | 0x3);
2901                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2902                         /*
2903                          * With jumbo frames and early-receive enabled,
2904                          * excessive C-state transition latencies result in
2905                          * dropped transactions.
2906                          */
2907                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2908                                                   adapter->netdev->name, 55);
2909                 } else {
2910                         pm_qos_update_requirement(PM_QOS_CPU_DMA_LATENCY,
2911                                                   adapter->netdev->name,
2912                                                   PM_QOS_DEFAULT_VALUE);
2913                 }
2914         }
2915
2916         /* Enable Receives */
2917         ew32(RCTL, rctl);
2918 }
2919
2920 /**
2921  *  e1000_update_mc_addr_list - Update Multicast addresses
2922  *  @hw: pointer to the HW structure
2923  *  @mc_addr_list: array of multicast addresses to program
2924  *  @mc_addr_count: number of multicast addresses to program
2925  *
2926  *  Updates the Multicast Table Array.
2927  *  The caller must have a packed mc_addr_list of multicast addresses.
2928  **/
2929 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2930                                       u32 mc_addr_count)
2931 {
2932         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2933 }
2934
2935 /**
2936  * e1000_set_multi - Multicast and Promiscuous mode set
2937  * @netdev: network interface device structure
2938  *
2939  * The set_multi entry point is called whenever the multicast address
2940  * list or the network interface flags are updated.  This routine is
2941  * responsible for configuring the hardware for proper multicast,
2942  * promiscuous mode, and all-multi behavior.
2943  **/
2944 static void e1000_set_multi(struct net_device *netdev)
2945 {
2946         struct e1000_adapter *adapter = netdev_priv(netdev);
2947         struct e1000_hw *hw = &adapter->hw;
2948         struct netdev_hw_addr *ha;
2949         u8  *mta_list;
2950         u32 rctl;
2951         int i;
2952
2953         /* Check for Promiscuous and All Multicast modes */
2954
2955         rctl = er32(RCTL);
2956
2957         if (netdev->flags & IFF_PROMISC) {
2958                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2959                 rctl &= ~E1000_RCTL_VFE;
2960         } else {
2961                 if (netdev->flags & IFF_ALLMULTI) {
2962                         rctl |= E1000_RCTL_MPE;
2963                         rctl &= ~E1000_RCTL_UPE;
2964                 } else {
2965                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2966                 }
2967                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2968                         rctl |= E1000_RCTL_VFE;
2969         }
2970
2971         ew32(RCTL, rctl);
2972
2973         if (!netdev_mc_empty(netdev)) {
2974                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2975                 if (!mta_list)
2976                         return;
2977
2978                 /* prepare a packed array of only addresses. */
2979                 i = 0;
2980                 netdev_for_each_mc_addr(ha, netdev)
2981                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2982
2983                 e1000_update_mc_addr_list(hw, mta_list, i);
2984                 kfree(mta_list);
2985         } else {
2986                 /*
2987                  * if we're called from probe, we might not have
2988                  * anything to do here, so clear out the list
2989                  */
2990                 e1000_update_mc_addr_list(hw, NULL, 0);
2991         }
2992 }
2993
2994 /**
2995  * e1000_configure - configure the hardware for Rx and Tx
2996  * @adapter: private board structure
2997  **/
2998 static void e1000_configure(struct e1000_adapter *adapter)
2999 {
3000         e1000_set_multi(adapter->netdev);
3001
3002         e1000_restore_vlan(adapter);
3003         e1000_init_manageability_pt(adapter);
3004
3005         e1000_configure_tx(adapter);
3006         e1000_setup_rctl(adapter);
3007         e1000_configure_rx(adapter);
3008         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3009 }
3010
3011 /**
3012  * e1000e_power_up_phy - restore link in case the phy was powered down
3013  * @adapter: address of board private structure
3014  *
3015  * The phy may be powered down to save power and turn off link when the
3016  * driver is unloaded and wake on lan is not enabled (among others)
3017  * *** this routine MUST be followed by a call to e1000e_reset ***
3018  **/
3019 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3020 {
3021         if (adapter->hw.phy.ops.power_up)
3022                 adapter->hw.phy.ops.power_up(&adapter->hw);
3023
3024         adapter->hw.mac.ops.setup_link(&adapter->hw);
3025 }
3026
3027 /**
3028  * e1000_power_down_phy - Power down the PHY
3029  *
3030  * Power down the PHY so no link is implied when interface is down.
3031  * The PHY cannot be powered down if management or WoL is active.
3032  */
3033 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3034 {
3035         /* WoL is enabled */
3036         if (adapter->wol)
3037                 return;
3038
3039         if (adapter->hw.phy.ops.power_down)
3040                 adapter->hw.phy.ops.power_down(&adapter->hw);
3041 }
3042
3043 /**
3044  * e1000e_reset - bring the hardware into a known good state
3045  *
3046  * This function boots the hardware and enables some settings that
3047  * require a configuration cycle of the hardware - those cannot be
3048  * set/changed during runtime. After reset the device needs to be
3049  * properly configured for Rx, Tx etc.
3050  */
3051 void e1000e_reset(struct e1000_adapter *adapter)
3052 {
3053         struct e1000_mac_info *mac = &adapter->hw.mac;
3054         struct e1000_fc_info *fc = &adapter->hw.fc;
3055         struct e1000_hw *hw = &adapter->hw;
3056         u32 tx_space, min_tx_space, min_rx_space;
3057         u32 pba = adapter->pba;
3058         u16 hwm;
3059
3060         /* reset Packet Buffer Allocation to default */
3061         ew32(PBA, pba);
3062
3063         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3064                 /*
3065                  * To maintain wire speed transmits, the Tx FIFO should be
3066                  * large enough to accommodate two full transmit packets,
3067                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3068                  * the Rx FIFO should be large enough to accommodate at least
3069                  * one full receive packet and is similarly rounded up and
3070                  * expressed in KB.
3071                  */
3072                 pba = er32(PBA);
3073                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3074                 tx_space = pba >> 16;
3075                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3076                 pba &= 0xffff;
3077                 /*
3078                  * the Tx fifo also stores 16 bytes of information about the tx
3079                  * but don't include ethernet FCS because hardware appends it
3080                  */
3081                 min_tx_space = (adapter->max_frame_size +
3082                                 sizeof(struct e1000_tx_desc) -
3083                                 ETH_FCS_LEN) * 2;
3084                 min_tx_space = ALIGN(min_tx_space, 1024);
3085                 min_tx_space >>= 10;
3086                 /* software strips receive CRC, so leave room for it */
3087                 min_rx_space = adapter->max_frame_size;
3088                 min_rx_space = ALIGN(min_rx_space, 1024);
3089                 min_rx_space >>= 10;
3090
3091                 /*
3092                  * If current Tx allocation is less than the min Tx FIFO size,
3093                  * and the min Tx FIFO size is less than the current Rx FIFO
3094                  * allocation, take space away from current Rx allocation
3095                  */
3096                 if ((tx_space < min_tx_space) &&
3097                     ((min_tx_space - tx_space) < pba)) {
3098                         pba -= min_tx_space - tx_space;
3099
3100                         /*
3101                          * if short on Rx space, Rx wins and must trump tx
3102                          * adjustment or use Early Receive if available
3103                          */
3104                         if ((pba < min_rx_space) &&
3105                             (!(adapter->flags & FLAG_HAS_ERT)))
3106                                 /* ERT enabled in e1000_configure_rx */
3107                                 pba = min_rx_space;
3108                 }
3109
3110                 ew32(PBA, pba);
3111         }
3112
3113
3114         /*
3115          * flow control settings
3116          *
3117          * The high water mark must be low enough to fit one full frame
3118          * (or the size used for early receive) above it in the Rx FIFO.
3119          * Set it to the lower of:
3120          * - 90% of the Rx FIFO size, and
3121          * - the full Rx FIFO size minus the early receive size (for parts
3122          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3123          * - the full Rx FIFO size minus one full frame
3124          */
3125         if (hw->mac.type == e1000_pchlan) {
3126                 /*
3127                  * Workaround PCH LOM adapter hangs with certain network
3128                  * loads.  If hangs persist, try disabling Tx flow control.
3129                  */
3130                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3131                         fc->high_water = 0x3500;
3132                         fc->low_water  = 0x1500;
3133                 } else {
3134                         fc->high_water = 0x5000;
3135                         fc->low_water  = 0x3000;
3136                 }
3137                 fc->refresh_time = 0x1000;
3138         } else {
3139                 if ((adapter->flags & FLAG_HAS_ERT) &&
3140                     (adapter->netdev->mtu > ETH_DATA_LEN))
3141                         hwm = min(((pba << 10) * 9 / 10),
3142                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3143                 else
3144                         hwm = min(((pba << 10) * 9 / 10),
3145                                   ((pba << 10) - adapter->max_frame_size));
3146
3147                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3148                 fc->low_water = fc->high_water - 8;
3149         }
3150
3151         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3152                 fc->pause_time = 0xFFFF;
3153         else
3154                 fc->pause_time = E1000_FC_PAUSE_TIME;
3155         fc->send_xon = 1;
3156         fc->current_mode = fc->requested_mode;
3157
3158         /* Allow time for pending master requests to run */
3159         mac->ops.reset_hw(hw);
3160
3161         /*
3162          * For parts with AMT enabled, let the firmware know
3163          * that the network interface is in control
3164          */
3165         if (adapter->flags & FLAG_HAS_AMT)
3166                 e1000_get_hw_control(adapter);
3167
3168         ew32(WUC, 0);
3169         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
3170                 e1e_wphy(&adapter->hw, BM_WUC, 0);
3171
3172         if (mac->ops.init_hw(hw))
3173                 e_err("Hardware Error\n");
3174
3175         e1000_update_mng_vlan(adapter);
3176
3177         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3178         ew32(VET, ETH_P_8021Q);
3179
3180         e1000e_reset_adaptive(hw);
3181         e1000_get_phy_info(hw);
3182
3183         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3184             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3185                 u16 phy_data = 0;
3186                 /*
3187                  * speed up time to link by disabling smart power down, ignore
3188                  * the return value of this function because there is nothing
3189                  * different we would do if it failed
3190                  */
3191                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3192                 phy_data &= ~IGP02E1000_PM_SPD;
3193                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3194         }
3195 }
3196
3197 int e1000e_up(struct e1000_adapter *adapter)
3198 {
3199         struct e1000_hw *hw = &adapter->hw;
3200
3201         /* DMA latency requirement to workaround early-receive/jumbo issue */
3202         if (adapter->flags & FLAG_HAS_ERT)
3203                 pm_qos_add_requirement(PM_QOS_CPU_DMA_LATENCY,
3204                                        adapter->netdev->name,
3205                                        PM_QOS_DEFAULT_VALUE);
3206
3207         /* hardware has been reset, we need to reload some things */
3208         e1000_configure(adapter);
3209
3210         clear_bit(__E1000_DOWN, &adapter->state);
3211
3212         napi_enable(&adapter->napi);
3213         if (adapter->msix_entries)
3214                 e1000_configure_msix(adapter);
3215         e1000_irq_enable(adapter);
3216
3217         netif_wake_queue(adapter->netdev);
3218
3219         /* fire a link change interrupt to start the watchdog */
3220         if (adapter->msix_entries)
3221                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3222         else
3223                 ew32(ICS, E1000_ICS_LSC);
3224
3225         return 0;
3226 }
3227
3228 void e1000e_down(struct e1000_adapter *adapter)
3229 {
3230         struct net_device *netdev = adapter->netdev;
3231         struct e1000_hw *hw = &adapter->hw;
3232         u32 tctl, rctl;
3233
3234         /*
3235          * signal that we're down so the interrupt handler does not
3236          * reschedule our watchdog timer
3237          */
3238         set_bit(__E1000_DOWN, &adapter->state);
3239
3240         /* disable receives in the hardware */
3241         rctl = er32(RCTL);
3242         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3243         /* flush and sleep below */
3244
3245         netif_stop_queue(netdev);
3246
3247         /* disable transmits in the hardware */
3248         tctl = er32(TCTL);
3249         tctl &= ~E1000_TCTL_EN;
3250         ew32(TCTL, tctl);
3251         /* flush both disables and wait for them to finish */
3252         e1e_flush();
3253         msleep(10);
3254
3255         napi_disable(&adapter->napi);
3256         e1000_irq_disable(adapter);
3257
3258         del_timer_sync(&adapter->watchdog_timer);
3259         del_timer_sync(&adapter->phy_info_timer);
3260
3261         netif_carrier_off(netdev);
3262         adapter->link_speed = 0;
3263         adapter->link_duplex = 0;
3264
3265         if (!pci_channel_offline(adapter->pdev))
3266                 e1000e_reset(adapter);
3267         e1000_clean_tx_ring(adapter);
3268         e1000_clean_rx_ring(adapter);
3269
3270         if (adapter->flags & FLAG_HAS_ERT)
3271                 pm_qos_remove_requirement(PM_QOS_CPU_DMA_LATENCY,
3272                                           adapter->netdev->name);
3273
3274         /*
3275          * TODO: for power management, we could drop the link and
3276          * pci_disable_device here.
3277          */
3278 }
3279
3280 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3281 {
3282         might_sleep();
3283         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3284                 msleep(1);
3285         e1000e_down(adapter);
3286         e1000e_up(adapter);
3287         clear_bit(__E1000_RESETTING, &adapter->state);
3288 }
3289
3290 /**
3291  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3292  * @adapter: board private structure to initialize
3293  *
3294  * e1000_sw_init initializes the Adapter private data structure.
3295  * Fields are initialized based on PCI device information and
3296  * OS network device settings (MTU size).
3297  **/
3298 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3299 {
3300         struct net_device *netdev = adapter->netdev;
3301
3302         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3303         adapter->rx_ps_bsize0 = 128;
3304         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3305         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3306
3307         e1000e_set_interrupt_capability(adapter);
3308
3309         if (e1000_alloc_queues(adapter))
3310                 return -ENOMEM;
3311
3312         /* Explicitly disable IRQ since the NIC can be in any state. */
3313         e1000_irq_disable(adapter);
3314
3315         set_bit(__E1000_DOWN, &adapter->state);
3316         return 0;
3317 }
3318
3319 /**
3320  * e1000_intr_msi_test - Interrupt Handler
3321  * @irq: interrupt number
3322  * @data: pointer to a network interface device structure
3323  **/
3324 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3325 {
3326         struct net_device *netdev = data;
3327         struct e1000_adapter *adapter = netdev_priv(netdev);
3328         struct e1000_hw *hw = &adapter->hw;
3329         u32 icr = er32(ICR);
3330
3331         e_dbg("icr is %08X\n", icr);
3332         if (icr & E1000_ICR_RXSEQ) {
3333                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3334                 wmb();
3335         }
3336
3337         return IRQ_HANDLED;
3338 }
3339
3340 /**
3341  * e1000_test_msi_interrupt - Returns 0 for successful test
3342  * @adapter: board private struct
3343  *
3344  * code flow taken from tg3.c
3345  **/
3346 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3347 {
3348         struct net_device *netdev = adapter->netdev;
3349         struct e1000_hw *hw = &adapter->hw;
3350         int err;
3351
3352         /* poll_enable hasn't been called yet, so don't need disable */
3353         /* clear any pending events */
3354         er32(ICR);
3355
3356         /* free the real vector and request a test handler */
3357         e1000_free_irq(adapter);
3358         e1000e_reset_interrupt_capability(adapter);
3359
3360         /* Assume that the test fails, if it succeeds then the test
3361          * MSI irq handler will unset this flag */
3362         adapter->flags |= FLAG_MSI_TEST_FAILED;
3363
3364         err = pci_enable_msi(adapter->pdev);
3365         if (err)
3366                 goto msi_test_failed;
3367
3368         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3369                           netdev->name, netdev);
3370         if (err) {
3371                 pci_disable_msi(adapter->pdev);
3372                 goto msi_test_failed;
3373         }
3374
3375         wmb();
3376
3377         e1000_irq_enable(adapter);
3378
3379         /* fire an unusual interrupt on the test handler */
3380         ew32(ICS, E1000_ICS_RXSEQ);
3381         e1e_flush();
3382         msleep(50);
3383
3384         e1000_irq_disable(adapter);
3385
3386         rmb();
3387
3388         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3389                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3390                 err = -EIO;
3391                 e_info("MSI interrupt test failed!\n");
3392         }
3393
3394         free_irq(adapter->pdev->irq, netdev);
3395         pci_disable_msi(adapter->pdev);
3396
3397         if (err == -EIO)
3398                 goto msi_test_failed;
3399
3400         /* okay so the test worked, restore settings */
3401         e_dbg("MSI interrupt test succeeded!\n");
3402 msi_test_failed:
3403         e1000e_set_interrupt_capability(adapter);
3404         e1000_request_irq(adapter);
3405         return err;
3406 }
3407
3408 /**
3409  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3410  * @adapter: board private struct
3411  *
3412  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3413  **/
3414 static int e1000_test_msi(struct e1000_adapter *adapter)
3415 {
3416         int err;
3417         u16 pci_cmd;
3418
3419         if (!(adapter->flags & FLAG_MSI_ENABLED))
3420                 return 0;
3421
3422         /* disable SERR in case the MSI write causes a master abort */
3423         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3424         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3425                               pci_cmd & ~PCI_COMMAND_SERR);
3426
3427         err = e1000_test_msi_interrupt(adapter);
3428
3429         /* restore previous setting of command word */
3430         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3431
3432         /* success ! */
3433         if (!err)
3434                 return 0;
3435
3436         /* EIO means MSI test failed */
3437         if (err != -EIO)
3438                 return err;
3439
3440         /* back to INTx mode */
3441         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3442
3443         e1000_free_irq(adapter);
3444
3445         err = e1000_request_irq(adapter);
3446
3447         return err;
3448 }
3449
3450 /**
3451  * e1000_open - Called when a network interface is made active
3452  * @netdev: network interface device structure
3453  *
3454  * Returns 0 on success, negative value on failure
3455  *
3456  * The open entry point is called when a network interface is made
3457  * active by the system (IFF_UP).  At this point all resources needed
3458  * for transmit and receive operations are allocated, the interrupt
3459  * handler is registered with the OS, the watchdog timer is started,
3460  * and the stack is notified that the interface is ready.
3461  **/
3462 static int e1000_open(struct net_device *netdev)
3463 {
3464         struct e1000_adapter *adapter = netdev_priv(netdev);
3465         struct e1000_hw *hw = &adapter->hw;
3466         struct pci_dev *pdev = adapter->pdev;
3467         int err;
3468
3469         /* disallow open during test */
3470         if (test_bit(__E1000_TESTING, &adapter->state))
3471                 return -EBUSY;
3472
3473         pm_runtime_get_sync(&pdev->dev);
3474
3475         netif_carrier_off(netdev);
3476
3477         /* allocate transmit descriptors */
3478         err = e1000e_setup_tx_resources(adapter);
3479         if (err)
3480                 goto err_setup_tx;
3481
3482         /* allocate receive descriptors */
3483         err = e1000e_setup_rx_resources(adapter);
3484         if (err)
3485                 goto err_setup_rx;
3486
3487         /*
3488          * If AMT is enabled, let the firmware know that the network
3489          * interface is now open and reset the part to a known state.
3490          */
3491         if (adapter->flags & FLAG_HAS_AMT) {
3492                 e1000_get_hw_control(adapter);
3493                 e1000e_reset(adapter);
3494         }
3495
3496         e1000e_power_up_phy(adapter);
3497
3498         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3499         if ((adapter->hw.mng_cookie.status &
3500              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3501                 e1000_update_mng_vlan(adapter);
3502
3503         /*
3504          * before we allocate an interrupt, we must be ready to handle it.
3505          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3506          * as soon as we call pci_request_irq, so we have to setup our
3507          * clean_rx handler before we do so.
3508          */
3509         e1000_configure(adapter);
3510
3511         err = e1000_request_irq(adapter);
3512         if (err)
3513                 goto err_req_irq;
3514
3515         /*
3516          * Work around PCIe errata with MSI interrupts causing some chipsets to
3517          * ignore e1000e MSI messages, which means we need to test our MSI
3518          * interrupt now
3519          */
3520         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3521                 err = e1000_test_msi(adapter);
3522                 if (err) {
3523                         e_err("Interrupt allocation failed\n");
3524                         goto err_req_irq;
3525                 }
3526         }
3527
3528         /* From here on the code is the same as e1000e_up() */
3529         clear_bit(__E1000_DOWN, &adapter->state);
3530
3531         napi_enable(&adapter->napi);
3532
3533         e1000_irq_enable(adapter);
3534
3535         netif_start_queue(netdev);
3536
3537         adapter->idle_check = true;
3538         pm_runtime_put(&pdev->dev);
3539
3540         /* fire a link status change interrupt to start the watchdog */
3541         if (adapter->msix_entries)
3542                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3543         else
3544                 ew32(ICS, E1000_ICS_LSC);
3545
3546         return 0;
3547
3548 err_req_irq:
3549         e1000_release_hw_control(adapter);
3550         e1000_power_down_phy(adapter);
3551         e1000e_free_rx_resources(adapter);
3552 err_setup_rx:
3553         e1000e_free_tx_resources(adapter);
3554 err_setup_tx:
3555         e1000e_reset(adapter);
3556         pm_runtime_put_sync(&pdev->dev);
3557
3558         return err;
3559 }
3560
3561 /**
3562  * e1000_close - Disables a network interface
3563  * @netdev: network interface device structure
3564  *
3565  * Returns 0, this is not allowed to fail
3566  *
3567  * The close entry point is called when an interface is de-activated
3568  * by the OS.  The hardware is still under the drivers control, but
3569  * needs to be disabled.  A global MAC reset is issued to stop the
3570  * hardware, and all transmit and receive resources are freed.
3571  **/
3572 static int e1000_close(struct net_device *netdev)
3573 {
3574         struct e1000_adapter *adapter = netdev_priv(netdev);
3575         struct pci_dev *pdev = adapter->pdev;
3576
3577         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3578
3579         pm_runtime_get_sync(&pdev->dev);
3580
3581         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3582                 e1000e_down(adapter);
3583                 e1000_free_irq(adapter);
3584         }
3585         e1000_power_down_phy(adapter);
3586
3587         e1000e_free_tx_resources(adapter);
3588         e1000e_free_rx_resources(adapter);
3589
3590         /*
3591          * kill manageability vlan ID if supported, but not if a vlan with
3592          * the same ID is registered on the host OS (let 8021q kill it)
3593          */
3594         if ((adapter->hw.mng_cookie.status &
3595                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3596              !(adapter->vlgrp &&
3597                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3598                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3599
3600         /*
3601          * If AMT is enabled, let the firmware know that the network
3602          * interface is now closed
3603          */
3604         if (adapter->flags & FLAG_HAS_AMT)
3605                 e1000_release_hw_control(adapter);
3606
3607         pm_runtime_put_sync(&pdev->dev);
3608
3609         return 0;
3610 }
3611 /**
3612  * e1000_set_mac - Change the Ethernet Address of the NIC
3613  * @netdev: network interface device structure
3614  * @p: pointer to an address structure
3615  *
3616  * Returns 0 on success, negative on failure
3617  **/
3618 static int e1000_set_mac(struct net_device *netdev, void *p)
3619 {
3620         struct e1000_adapter *adapter = netdev_priv(netdev);
3621         struct sockaddr *addr = p;
3622
3623         if (!is_valid_ether_addr(addr->sa_data))
3624                 return -EADDRNOTAVAIL;
3625
3626         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3627         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3628
3629         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3630
3631         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3632                 /* activate the work around */
3633                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3634
3635                 /*
3636                  * Hold a copy of the LAA in RAR[14] This is done so that
3637                  * between the time RAR[0] gets clobbered  and the time it
3638                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3639                  * of the RARs and no incoming packets directed to this port
3640                  * are dropped. Eventually the LAA will be in RAR[0] and
3641                  * RAR[14]
3642                  */
3643                 e1000e_rar_set(&adapter->hw,
3644                               adapter->hw.mac.addr,
3645                               adapter->hw.mac.rar_entry_count - 1);
3646         }
3647
3648         return 0;
3649 }
3650
3651 /**
3652  * e1000e_update_phy_task - work thread to update phy
3653  * @work: pointer to our work struct
3654  *
3655  * this worker thread exists because we must acquire a
3656  * semaphore to read the phy, which we could msleep while
3657  * waiting for it, and we can't msleep in a timer.
3658  **/
3659 static void e1000e_update_phy_task(struct work_struct *work)
3660 {
3661         struct e1000_adapter *adapter = container_of(work,
3662                                         struct e1000_adapter, update_phy_task);
3663         e1000_get_phy_info(&adapter->hw);
3664 }
3665
3666 /*
3667  * Need to wait a few seconds after link up to get diagnostic information from
3668  * the phy
3669  */
3670 static void e1000_update_phy_info(unsigned long data)
3671 {
3672         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3673         schedule_work(&adapter->update_phy_task);
3674 }
3675
3676 /**
3677  * e1000e_update_stats - Update the board statistics counters
3678  * @adapter: board private structure
3679  **/
3680 void e1000e_update_stats(struct e1000_adapter *adapter)
3681 {
3682         struct net_device *netdev = adapter->netdev;
3683         struct e1000_hw *hw = &adapter->hw;
3684         struct pci_dev *pdev = adapter->pdev;
3685         u16 phy_data;
3686
3687         /*
3688          * Prevent stats update while adapter is being reset, or if the pci
3689          * connection is down.
3690          */
3691         if (adapter->link_speed == 0)
3692                 return;
3693         if (pci_channel_offline(pdev))
3694                 return;
3695
3696         adapter->stats.crcerrs += er32(CRCERRS);
3697         adapter->stats.gprc += er32(GPRC);
3698         adapter->stats.gorc += er32(GORCL);
3699         er32(GORCH); /* Clear gorc */
3700         adapter->stats.bprc += er32(BPRC);
3701         adapter->stats.mprc += er32(MPRC);
3702         adapter->stats.roc += er32(ROC);
3703
3704         adapter->stats.mpc += er32(MPC);
3705         if ((hw->phy.type == e1000_phy_82578) ||
3706             (hw->phy.type == e1000_phy_82577)) {
3707                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3708                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3709                         adapter->stats.scc += phy_data;
3710
3711                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3712                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3713                         adapter->stats.ecol += phy_data;
3714
3715                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3716                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3717                         adapter->stats.mcc += phy_data;
3718
3719                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3720                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3721                         adapter->stats.latecol += phy_data;
3722
3723                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3724                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3725                         adapter->stats.dc += phy_data;
3726         } else {
3727                 adapter->stats.scc += er32(SCC);
3728                 adapter->stats.ecol += er32(ECOL);
3729                 adapter->stats.mcc += er32(MCC);
3730                 adapter->stats.latecol += er32(LATECOL);
3731                 adapter->stats.dc += er32(DC);
3732         }
3733         adapter->stats.xonrxc += er32(XONRXC);
3734         adapter->stats.xontxc += er32(XONTXC);
3735         adapter->stats.xoffrxc += er32(XOFFRXC);
3736         adapter->stats.xofftxc += er32(XOFFTXC);
3737         adapter->stats.gptc += er32(GPTC);
3738         adapter->stats.gotc += er32(GOTCL);
3739         er32(GOTCH); /* Clear gotc */
3740         adapter->stats.rnbc += er32(RNBC);
3741         adapter->stats.ruc += er32(RUC);
3742
3743         adapter->stats.mptc += er32(MPTC);
3744         adapter->stats.bptc += er32(BPTC);
3745
3746         /* used for adaptive IFS */
3747
3748         hw->mac.tx_packet_delta = er32(TPT);
3749         adapter->stats.tpt += hw->mac.tx_packet_delta;
3750         if ((hw->phy.type == e1000_phy_82578) ||
3751             (hw->phy.type == e1000_phy_82577)) {
3752                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3753                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3754                         hw->mac.collision_delta = phy_data;
3755         } else {
3756                 hw->mac.collision_delta = er32(COLC);
3757         }
3758         adapter->stats.colc += hw->mac.collision_delta;
3759
3760         adapter->stats.algnerrc += er32(ALGNERRC);
3761         adapter->stats.rxerrc += er32(RXERRC);
3762         if ((hw->phy.type == e1000_phy_82578) ||
3763             (hw->phy.type == e1000_phy_82577)) {
3764                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3765                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3766                         adapter->stats.tncrs += phy_data;
3767         } else {
3768                 if ((hw->mac.type != e1000_82574) &&
3769                     (hw->mac.type != e1000_82583))
3770                         adapter->stats.tncrs += er32(TNCRS);
3771         }
3772         adapter->stats.cexterr += er32(CEXTERR);
3773         adapter->stats.tsctc += er32(TSCTC);
3774         adapter->stats.tsctfc += er32(TSCTFC);
3775
3776         /* Fill out the OS statistics structure */
3777         netdev->stats.multicast = adapter->stats.mprc;
3778         netdev->stats.collisions = adapter->stats.colc;
3779
3780         /* Rx Errors */
3781
3782         /*
3783          * RLEC on some newer hardware can be incorrect so build
3784          * our own version based on RUC and ROC
3785          */
3786         netdev->stats.rx_errors = adapter->stats.rxerrc +
3787                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3788                 adapter->stats.ruc + adapter->stats.roc +
3789                 adapter->stats.cexterr;
3790         netdev->stats.rx_length_errors = adapter->stats.ruc +
3791                                               adapter->stats.roc;
3792         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3793         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3794         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3795
3796         /* Tx Errors */
3797         netdev->stats.tx_errors = adapter->stats.ecol +
3798                                        adapter->stats.latecol;
3799         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3800         netdev->stats.tx_window_errors = adapter->stats.latecol;
3801         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3802
3803         /* Tx Dropped needs to be maintained elsewhere */
3804
3805         /* Management Stats */
3806         adapter->stats.mgptc += er32(MGTPTC);
3807         adapter->stats.mgprc += er32(MGTPRC);
3808         adapter->stats.mgpdc += er32(MGTPDC);
3809 }
3810
3811 /**
3812  * e1000_phy_read_status - Update the PHY register status snapshot
3813  * @adapter: board private structure
3814  **/
3815 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3816 {
3817         struct e1000_hw *hw = &adapter->hw;
3818         struct e1000_phy_regs *phy = &adapter->phy_regs;
3819         int ret_val;
3820
3821         if ((er32(STATUS) & E1000_STATUS_LU) &&
3822             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3823                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3824                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3825                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3826                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3827                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3828                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3829                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3830                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3831                 if (ret_val)
3832                         e_warn("Error reading PHY register\n");
3833         } else {
3834                 /*
3835                  * Do not read PHY registers if link is not up
3836                  * Set values to typical power-on defaults
3837                  */
3838                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3839                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3840                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3841                              BMSR_ERCAP);
3842                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3843                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3844                 phy->lpa = 0;
3845                 phy->expansion = EXPANSION_ENABLENPAGE;
3846                 phy->ctrl1000 = ADVERTISE_1000FULL;
3847                 phy->stat1000 = 0;
3848                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3849         }
3850 }
3851
3852 static void e1000_print_link_info(struct e1000_adapter *adapter)
3853 {
3854         struct e1000_hw *hw = &adapter->hw;
3855         u32 ctrl = er32(CTRL);
3856
3857         /* Link status message must follow this format for user tools */
3858         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3859                "Flow Control: %s\n",
3860                adapter->netdev->name,
3861                adapter->link_speed,
3862                (adapter->link_duplex == FULL_DUPLEX) ?
3863                                 "Full Duplex" : "Half Duplex",
3864                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3865                                 "RX/TX" :
3866                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3867                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3868 }
3869
3870 bool e1000e_has_link(struct e1000_adapter *adapter)
3871 {
3872         struct e1000_hw *hw = &adapter->hw;
3873         bool link_active = 0;
3874         s32 ret_val = 0;
3875
3876         /*
3877          * get_link_status is set on LSC (link status) interrupt or
3878          * Rx sequence error interrupt.  get_link_status will stay
3879          * false until the check_for_link establishes link
3880          * for copper adapters ONLY
3881          */
3882         switch (hw->phy.media_type) {
3883         case e1000_media_type_copper:
3884                 if (hw->mac.get_link_status) {
3885                         ret_val = hw->mac.ops.check_for_link(hw);
3886                         link_active = !hw->mac.get_link_status;
3887                 } else {
3888                         link_active = 1;
3889                 }
3890                 break;
3891         case e1000_media_type_fiber:
3892                 ret_val = hw->mac.ops.check_for_link(hw);
3893                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3894                 break;
3895         case e1000_media_type_internal_serdes:
3896                 ret_val = hw->mac.ops.check_for_link(hw);
3897                 link_active = adapter->hw.mac.serdes_has_link;
3898                 break;
3899         default:
3900         case e1000_media_type_unknown:
3901                 break;
3902         }
3903
3904         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3905             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3906                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3907                 e_info("Gigabit has been disabled, downgrading speed\n");
3908         }
3909
3910         return link_active;
3911 }
3912
3913 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3914 {
3915         /* make sure the receive unit is started */
3916         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3917             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3918                 struct e1000_hw *hw = &adapter->hw;
3919                 u32 rctl = er32(RCTL);
3920                 ew32(RCTL, rctl | E1000_RCTL_EN);
3921                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3922         }
3923 }
3924
3925 /**
3926  * e1000_watchdog - Timer Call-back
3927  * @data: pointer to adapter cast into an unsigned long
3928  **/
3929 static void e1000_watchdog(unsigned long data)
3930 {
3931         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3932
3933         /* Do the rest outside of interrupt context */
3934         schedule_work(&adapter->watchdog_task);
3935
3936         /* TODO: make this use queue_delayed_work() */
3937 }
3938
3939 static void e1000_watchdog_task(struct work_struct *work)
3940 {
3941         struct e1000_adapter *adapter = container_of(work,
3942                                         struct e1000_adapter, watchdog_task);
3943         struct net_device *netdev = adapter->netdev;
3944         struct e1000_mac_info *mac = &adapter->hw.mac;
3945         struct e1000_phy_info *phy = &adapter->hw.phy;
3946         struct e1000_ring *tx_ring = adapter->tx_ring;
3947         struct e1000_hw *hw = &adapter->hw;
3948         u32 link, tctl;
3949         int tx_pending = 0;
3950
3951         link = e1000e_has_link(adapter);
3952         if ((netif_carrier_ok(netdev)) && link) {
3953                 /* Cancel scheduled suspend requests. */
3954                 pm_runtime_resume(netdev->dev.parent);
3955
3956                 e1000e_enable_receives(adapter);
3957                 goto link_up;
3958         }
3959
3960         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3961             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3962                 e1000_update_mng_vlan(adapter);
3963
3964         if (link) {
3965                 if (!netif_carrier_ok(netdev)) {
3966                         bool txb2b = 1;
3967
3968                         /* Cancel scheduled suspend requests. */
3969                         pm_runtime_resume(netdev->dev.parent);
3970
3971                         /* update snapshot of PHY registers on LSC */
3972                         e1000_phy_read_status(adapter);
3973                         mac->ops.get_link_up_info(&adapter->hw,
3974                                                    &adapter->link_speed,
3975                                                    &adapter->link_duplex);
3976                         e1000_print_link_info(adapter);
3977                         /*
3978                          * On supported PHYs, check for duplex mismatch only
3979                          * if link has autonegotiated at 10/100 half
3980                          */
3981                         if ((hw->phy.type == e1000_phy_igp_3 ||
3982                              hw->phy.type == e1000_phy_bm) &&
3983                             (hw->mac.autoneg == true) &&
3984                             (adapter->link_speed == SPEED_10 ||
3985                              adapter->link_speed == SPEED_100) &&
3986                             (adapter->link_duplex == HALF_DUPLEX)) {
3987                                 u16 autoneg_exp;
3988
3989                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3990
3991                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3992                                         e_info("Autonegotiated half duplex but"
3993                                                " link partner cannot autoneg. "
3994                                                " Try forcing full duplex if "
3995                                                "link gets many collisions.\n");
3996                         }
3997
3998                         /* adjust timeout factor according to speed/duplex */
3999                         adapter->tx_timeout_factor = 1;
4000                         switch (adapter->link_speed) {
4001                         case SPEED_10:
4002                                 txb2b = 0;
4003                                 adapter->tx_timeout_factor = 16;
4004                                 break;
4005                         case SPEED_100:
4006                                 txb2b = 0;
4007                                 adapter->tx_timeout_factor = 10;
4008                                 break;
4009                         }
4010
4011                         /*
4012                          * workaround: re-program speed mode bit after
4013                          * link-up event
4014                          */
4015                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4016                             !txb2b) {
4017                                 u32 tarc0;
4018                                 tarc0 = er32(TARC(0));
4019                                 tarc0 &= ~SPEED_MODE_BIT;
4020                                 ew32(TARC(0), tarc0);
4021                         }
4022
4023                         /*
4024                          * disable TSO for pcie and 10/100 speeds, to avoid
4025                          * some hardware issues
4026                          */
4027                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4028                                 switch (adapter->link_speed) {
4029                                 case SPEED_10:
4030                                 case SPEED_100:
4031                                         e_info("10/100 speed: disabling TSO\n");
4032                                         netdev->features &= ~NETIF_F_TSO;
4033                                         netdev->features &= ~NETIF_F_TSO6;
4034                                         break;
4035                                 case SPEED_1000:
4036                                         netdev->features |= NETIF_F_TSO;
4037                                         netdev->features |= NETIF_F_TSO6;
4038                                         break;
4039                                 default:
4040                                         /* oops */
4041                                         break;
4042                                 }
4043                         }
4044
4045                         /*
4046                          * enable transmits in the hardware, need to do this
4047                          * after setting TARC(0)
4048                          */
4049                         tctl = er32(TCTL);
4050                         tctl |= E1000_TCTL_EN;
4051                         ew32(TCTL, tctl);
4052
4053                         /*
4054                          * Perform any post-link-up configuration before
4055                          * reporting link up.
4056                          */
4057                         if (phy->ops.cfg_on_link_up)
4058                                 phy->ops.cfg_on_link_up(hw);
4059
4060                         netif_carrier_on(netdev);
4061
4062                         if (!test_bit(__E1000_DOWN, &adapter->state))
4063                                 mod_timer(&adapter->phy_info_timer,
4064                                           round_jiffies(jiffies + 2 * HZ));
4065                 }
4066         } else {
4067                 if (netif_carrier_ok(netdev)) {
4068                         adapter->link_speed = 0;
4069                         adapter->link_duplex = 0;
4070                         /* Link status message must follow this format */
4071                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4072                                adapter->netdev->name);
4073                         netif_carrier_off(netdev);
4074                         if (!test_bit(__E1000_DOWN, &adapter->state))
4075                                 mod_timer(&adapter->phy_info_timer,
4076                                           round_jiffies(jiffies + 2 * HZ));
4077
4078                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4079                                 schedule_work(&adapter->reset_task);
4080                         else
4081                                 pm_schedule_suspend(netdev->dev.parent,
4082                                                         LINK_TIMEOUT);
4083                 }
4084         }
4085
4086 link_up:
4087         e1000e_update_stats(adapter);
4088
4089         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4090         adapter->tpt_old = adapter->stats.tpt;
4091         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4092         adapter->colc_old = adapter->stats.colc;
4093
4094         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4095         adapter->gorc_old = adapter->stats.gorc;
4096         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4097         adapter->gotc_old = adapter->stats.gotc;
4098
4099         e1000e_update_adaptive(&adapter->hw);
4100
4101         if (!netif_carrier_ok(netdev)) {
4102                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4103                                tx_ring->count);
4104                 if (tx_pending) {
4105                         /*
4106                          * We've lost link, so the controller stops DMA,
4107                          * but we've got queued Tx work that's never going
4108                          * to get done, so reset controller to flush Tx.
4109                          * (Do the reset outside of interrupt context).
4110                          */
4111                         adapter->tx_timeout_count++;
4112                         schedule_work(&adapter->reset_task);
4113                         /* return immediately since reset is imminent */
4114                         return;
4115                 }
4116         }
4117
4118         /* Simple mode for Interrupt Throttle Rate (ITR) */
4119         if (adapter->itr_setting == 4) {
4120                 /*
4121                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4122                  * Total asymmetrical Tx or Rx gets ITR=8000;
4123                  * everyone else is between 2000-8000.
4124                  */
4125                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4126                 u32 dif = (adapter->gotc > adapter->gorc ?
4127                             adapter->gotc - adapter->gorc :
4128                             adapter->gorc - adapter->gotc) / 10000;
4129                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4130
4131                 ew32(ITR, 1000000000 / (itr * 256));
4132         }
4133
4134         /* Cause software interrupt to ensure Rx ring is cleaned */
4135         if (adapter->msix_entries)
4136                 ew32(ICS, adapter->rx_ring->ims_val);
4137         else
4138                 ew32(ICS, E1000_ICS_RXDMT0);
4139
4140         /* Force detection of hung controller every watchdog period */
4141         adapter->detect_tx_hung = 1;
4142
4143         /*
4144          * With 82571 controllers, LAA may be overwritten due to controller
4145          * reset from the other port. Set the appropriate LAA in RAR[0]
4146          */
4147         if (e1000e_get_laa_state_82571(hw))
4148                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4149
4150         /* Reset the timer */
4151         if (!test_bit(__E1000_DOWN, &adapter->state))
4152                 mod_timer(&adapter->watchdog_timer,
4153                           round_jiffies(jiffies + 2 * HZ));
4154 }
4155
4156 #define E1000_TX_FLAGS_CSUM             0x00000001
4157 #define E1000_TX_FLAGS_VLAN             0x00000002
4158 #define E1000_TX_FLAGS_TSO              0x00000004
4159 #define E1000_TX_FLAGS_IPV4             0x00000008
4160 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4161 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4162
4163 static int e1000_tso(struct e1000_adapter *adapter,
4164                      struct sk_buff *skb)
4165 {
4166         struct e1000_ring *tx_ring = adapter->tx_ring;
4167         struct e1000_context_desc *context_desc;
4168         struct e1000_buffer *buffer_info;
4169         unsigned int i;
4170         u32 cmd_length = 0;
4171         u16 ipcse = 0, tucse, mss;
4172         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4173         int err;
4174
4175         if (!skb_is_gso(skb))
4176                 return 0;
4177
4178         if (skb_header_cloned(skb)) {
4179                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4180                 if (err)
4181                         return err;
4182         }
4183
4184         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4185         mss = skb_shinfo(skb)->gso_size;
4186         if (skb->protocol == htons(ETH_P_IP)) {
4187                 struct iphdr *iph = ip_hdr(skb);
4188                 iph->tot_len = 0;
4189                 iph->check = 0;
4190                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4191                                                          0, IPPROTO_TCP, 0);
4192                 cmd_length = E1000_TXD_CMD_IP;
4193                 ipcse = skb_transport_offset(skb) - 1;
4194         } else if (skb_is_gso_v6(skb)) {
4195                 ipv6_hdr(skb)->payload_len = 0;
4196                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4197                                                        &ipv6_hdr(skb)->daddr,
4198                                                        0, IPPROTO_TCP, 0);
4199                 ipcse = 0;
4200         }
4201         ipcss = skb_network_offset(skb);
4202         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4203         tucss = skb_transport_offset(skb);
4204         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4205         tucse = 0;
4206
4207         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4208                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4209
4210         i = tx_ring->next_to_use;
4211         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4212         buffer_info = &tx_ring->buffer_info[i];
4213
4214         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4215         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4216         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4217         context_desc->upper_setup.tcp_fields.tucss = tucss;
4218         context_desc->upper_setup.tcp_fields.tucso = tucso;
4219         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4220         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4221         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4222         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4223
4224         buffer_info->time_stamp = jiffies;
4225         buffer_info->next_to_watch = i;
4226
4227         i++;
4228         if (i == tx_ring->count)
4229                 i = 0;
4230         tx_ring->next_to_use = i;
4231
4232         return 1;
4233 }
4234
4235 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4236 {
4237         struct e1000_ring *tx_ring = adapter->tx_ring;
4238         struct e1000_context_desc *context_desc;
4239         struct e1000_buffer *buffer_info;
4240         unsigned int i;
4241         u8 css;
4242         u32 cmd_len = E1000_TXD_CMD_DEXT;
4243         __be16 protocol;
4244
4245         if (skb->ip_summed != CHECKSUM_PARTIAL)
4246                 return 0;
4247
4248         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4249                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4250         else
4251                 protocol = skb->protocol;
4252
4253         switch (protocol) {
4254         case cpu_to_be16(ETH_P_IP):
4255                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4256                         cmd_len |= E1000_TXD_CMD_TCP;
4257                 break;
4258         case cpu_to_be16(ETH_P_IPV6):
4259                 /* XXX not handling all IPV6 headers */
4260                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4261                         cmd_len |= E1000_TXD_CMD_TCP;
4262                 break;
4263         default:
4264                 if (unlikely(net_ratelimit()))
4265                         e_warn("checksum_partial proto=%x!\n",
4266                                be16_to_cpu(protocol));
4267                 break;
4268         }
4269
4270         css = skb_transport_offset(skb);
4271
4272         i = tx_ring->next_to_use;
4273         buffer_info = &tx_ring->buffer_info[i];
4274         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4275
4276         context_desc->lower_setup.ip_config = 0;
4277         context_desc->upper_setup.tcp_fields.tucss = css;
4278         context_desc->upper_setup.tcp_fields.tucso =
4279                                 css + skb->csum_offset;
4280         context_desc->upper_setup.tcp_fields.tucse = 0;
4281         context_desc->tcp_seg_setup.data = 0;
4282         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4283
4284         buffer_info->time_stamp = jiffies;
4285         buffer_info->next_to_watch = i;
4286
4287         i++;
4288         if (i == tx_ring->count)
4289                 i = 0;
4290         tx_ring->next_to_use = i;
4291
4292         return 1;
4293 }
4294
4295 #define E1000_MAX_PER_TXD       8192
4296 #define E1000_MAX_TXD_PWR       12
4297
4298 static int e1000_tx_map(struct e1000_adapter *adapter,
4299                         struct sk_buff *skb, unsigned int first,
4300                         unsigned int max_per_txd, unsigned int nr_frags,
4301                         unsigned int mss)
4302 {
4303         struct e1000_ring *tx_ring = adapter->tx_ring;
4304         struct pci_dev *pdev = adapter->pdev;
4305         struct e1000_buffer *buffer_info;
4306         unsigned int len = skb_headlen(skb);
4307         unsigned int offset = 0, size, count = 0, i;
4308         unsigned int f, bytecount, segs;
4309
4310         i = tx_ring->next_to_use;
4311
4312         while (len) {
4313                 buffer_info = &tx_ring->buffer_info[i];
4314                 size = min(len, max_per_txd);
4315
4316                 buffer_info->length = size;
4317                 buffer_info->time_stamp = jiffies;
4318                 buffer_info->next_to_watch = i;
4319                 buffer_info->dma = dma_map_single(&pdev->dev,
4320                                                   skb->data + offset,
4321                                                   size, DMA_TO_DEVICE);
4322                 buffer_info->mapped_as_page = false;
4323                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4324                         goto dma_error;
4325
4326                 len -= size;
4327                 offset += size;
4328                 count++;
4329
4330                 if (len) {
4331                         i++;
4332                         if (i == tx_ring->count)
4333                                 i = 0;
4334                 }
4335         }
4336
4337         for (f = 0; f < nr_frags; f++) {
4338                 struct skb_frag_struct *frag;
4339
4340                 frag = &skb_shinfo(skb)->frags[f];
4341                 len = frag->size;
4342                 offset = frag->page_offset;
4343
4344                 while (len) {
4345                         i++;
4346                         if (i == tx_ring->count)
4347                                 i = 0;
4348
4349                         buffer_info = &tx_ring->buffer_info[i];
4350                         size = min(len, max_per_txd);
4351
4352                         buffer_info->length = size;
4353                         buffer_info->time_stamp = jiffies;
4354                         buffer_info->next_to_watch = i;
4355                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4356                                                         offset, size,
4357                                                         DMA_TO_DEVICE);
4358                         buffer_info->mapped_as_page = true;
4359                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4360                                 goto dma_error;
4361
4362                         len -= size;
4363                         offset += size;
4364                         count++;
4365                 }
4366         }
4367
4368         segs = skb_shinfo(skb)->gso_segs ?: 1;
4369         /* multiply data chunks by size of headers */
4370         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4371
4372         tx_ring->buffer_info[i].skb = skb;
4373         tx_ring->buffer_info[i].segs = segs;
4374         tx_ring->buffer_info[i].bytecount = bytecount;
4375         tx_ring->buffer_info[first].next_to_watch = i;
4376
4377         return count;
4378
4379 dma_error:
4380         dev_err(&pdev->dev, "TX DMA map failed\n");
4381         buffer_info->dma = 0;
4382         if (count)
4383                 count--;
4384
4385         while (count--) {
4386                 if (i==0)
4387                         i += tx_ring->count;
4388                 i--;
4389                 buffer_info = &tx_ring->buffer_info[i];
4390                 e1000_put_txbuf(adapter, buffer_info);;
4391         }
4392
4393         return 0;
4394 }
4395
4396 static void e1000_tx_queue(struct e1000_adapter *adapter,
4397                            int tx_flags, int count)
4398 {
4399         struct e1000_ring *tx_ring = adapter->tx_ring;
4400         struct e1000_tx_desc *tx_desc = NULL;
4401         struct e1000_buffer *buffer_info;
4402         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4403         unsigned int i;
4404
4405         if (tx_flags & E1000_TX_FLAGS_TSO) {
4406                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4407                              E1000_TXD_CMD_TSE;
4408                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4409
4410                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4411                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4412         }
4413
4414         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4415                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4416                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4417         }
4418
4419         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4420                 txd_lower |= E1000_TXD_CMD_VLE;
4421                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4422         }
4423
4424         i = tx_ring->next_to_use;
4425
4426         while (count--) {
4427                 buffer_info = &tx_ring->buffer_info[i];
4428                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4429                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4430                 tx_desc->lower.data =
4431                         cpu_to_le32(txd_lower | buffer_info->length);
4432                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4433
4434                 i++;
4435                 if (i == tx_ring->count)
4436                         i = 0;
4437         }
4438
4439         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4440
4441         /*
4442          * Force memory writes to complete before letting h/w
4443          * know there are new descriptors to fetch.  (Only
4444          * applicable for weak-ordered memory model archs,
4445          * such as IA-64).
4446          */
4447         wmb();
4448
4449         tx_ring->next_to_use = i;
4450         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4451         /*
4452          * we need this if more than one processor can write to our tail
4453          * at a time, it synchronizes IO on IA64/Altix systems
4454          */
4455         mmiowb();
4456 }
4457
4458 #define MINIMUM_DHCP_PACKET_SIZE 282
4459 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4460                                     struct sk_buff *skb)
4461 {
4462         struct e1000_hw *hw =  &adapter->hw;
4463         u16 length, offset;
4464
4465         if (vlan_tx_tag_present(skb)) {
4466                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4467                     (adapter->hw.mng_cookie.status &
4468                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4469                         return 0;
4470         }
4471
4472         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4473                 return 0;
4474
4475         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4476                 return 0;
4477
4478         {
4479                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4480                 struct udphdr *udp;
4481
4482                 if (ip->protocol != IPPROTO_UDP)
4483                         return 0;
4484
4485                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4486                 if (ntohs(udp->dest) != 67)
4487                         return 0;
4488
4489                 offset = (u8 *)udp + 8 - skb->data;
4490                 length = skb->len - offset;
4491                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4492         }
4493
4494         return 0;
4495 }
4496
4497 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4498 {
4499         struct e1000_adapter *adapter = netdev_priv(netdev);
4500
4501         netif_stop_queue(netdev);
4502         /*
4503          * Herbert's original patch had:
4504          *  smp_mb__after_netif_stop_queue();
4505          * but since that doesn't exist yet, just open code it.
4506          */
4507         smp_mb();
4508
4509         /*
4510          * We need to check again in a case another CPU has just
4511          * made room available.
4512          */
4513         if (e1000_desc_unused(adapter->tx_ring) < size)
4514                 return -EBUSY;
4515
4516         /* A reprieve! */
4517         netif_start_queue(netdev);
4518         ++adapter->restart_queue;
4519         return 0;
4520 }
4521
4522 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4523 {
4524         struct e1000_adapter *adapter = netdev_priv(netdev);
4525
4526         if (e1000_desc_unused(adapter->tx_ring) >= size)
4527                 return 0;
4528         return __e1000_maybe_stop_tx(netdev, size);
4529 }
4530
4531 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4532 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4533                                     struct net_device *netdev)
4534 {
4535         struct e1000_adapter *adapter = netdev_priv(netdev);
4536         struct e1000_ring *tx_ring = adapter->tx_ring;
4537         unsigned int first;
4538         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4539         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4540         unsigned int tx_flags = 0;
4541         unsigned int len = skb_headlen(skb);
4542         unsigned int nr_frags;
4543         unsigned int mss;
4544         int count = 0;
4545         int tso;
4546         unsigned int f;
4547
4548         if (test_bit(__E1000_DOWN, &adapter->state)) {
4549                 dev_kfree_skb_any(skb);
4550                 return NETDEV_TX_OK;
4551         }
4552
4553         if (skb->len <= 0) {
4554                 dev_kfree_skb_any(skb);
4555                 return NETDEV_TX_OK;
4556         }
4557
4558         mss = skb_shinfo(skb)->gso_size;
4559         /*
4560          * The controller does a simple calculation to
4561          * make sure there is enough room in the FIFO before
4562          * initiating the DMA for each buffer.  The calc is:
4563          * 4 = ceil(buffer len/mss).  To make sure we don't
4564          * overrun the FIFO, adjust the max buffer len if mss
4565          * drops.
4566          */
4567         if (mss) {
4568                 u8 hdr_len;
4569                 max_per_txd = min(mss << 2, max_per_txd);
4570                 max_txd_pwr = fls(max_per_txd) - 1;
4571
4572                 /*
4573                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4574                  * points to just header, pull a few bytes of payload from
4575                  * frags into skb->data
4576                  */
4577                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4578                 /*
4579                  * we do this workaround for ES2LAN, but it is un-necessary,
4580                  * avoiding it could save a lot of cycles
4581                  */
4582                 if (skb->data_len && (hdr_len == len)) {
4583                         unsigned int pull_size;
4584
4585                         pull_size = min((unsigned int)4, skb->data_len);
4586                         if (!__pskb_pull_tail(skb, pull_size)) {
4587                                 e_err("__pskb_pull_tail failed.\n");
4588                                 dev_kfree_skb_any(skb);
4589                                 return NETDEV_TX_OK;
4590                         }
4591                         len = skb_headlen(skb);
4592                 }
4593         }
4594
4595         /* reserve a descriptor for the offload context */
4596         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4597                 count++;
4598         count++;
4599
4600         count += TXD_USE_COUNT(len, max_txd_pwr);
4601
4602         nr_frags = skb_shinfo(skb)->nr_frags;
4603         for (f = 0; f < nr_frags; f++)
4604                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4605                                        max_txd_pwr);
4606
4607         if (adapter->hw.mac.tx_pkt_filtering)
4608                 e1000_transfer_dhcp_info(adapter, skb);
4609
4610         /*
4611          * need: count + 2 desc gap to keep tail from touching
4612          * head, otherwise try next time
4613          */
4614         if (e1000_maybe_stop_tx(netdev, count + 2))
4615                 return NETDEV_TX_BUSY;
4616
4617         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4618                 tx_flags |= E1000_TX_FLAGS_VLAN;
4619                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4620         }
4621
4622         first = tx_ring->next_to_use;
4623
4624         tso = e1000_tso(adapter, skb);
4625         if (tso < 0) {
4626                 dev_kfree_skb_any(skb);
4627                 return NETDEV_TX_OK;
4628         }
4629
4630         if (tso)
4631                 tx_flags |= E1000_TX_FLAGS_TSO;
4632         else if (e1000_tx_csum(adapter, skb))
4633                 tx_flags |= E1000_TX_FLAGS_CSUM;
4634
4635         /*
4636          * Old method was to assume IPv4 packet by default if TSO was enabled.
4637          * 82571 hardware supports TSO capabilities for IPv6 as well...
4638          * no longer assume, we must.
4639          */
4640         if (skb->protocol == htons(ETH_P_IP))
4641                 tx_flags |= E1000_TX_FLAGS_IPV4;
4642
4643         /* if count is 0 then mapping error has occured */
4644         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4645         if (count) {
4646                 e1000_tx_queue(adapter, tx_flags, count);
4647                 /* Make sure there is space in the ring for the next send. */
4648                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4649
4650         } else {
4651                 dev_kfree_skb_any(skb);
4652                 tx_ring->buffer_info[first].time_stamp = 0;
4653                 tx_ring->next_to_use = first;
4654         }
4655
4656         return NETDEV_TX_OK;
4657 }
4658
4659 /**
4660  * e1000_tx_timeout - Respond to a Tx Hang
4661  * @netdev: network interface device structure
4662  **/
4663 static void e1000_tx_timeout(struct net_device *netdev)
4664 {
4665         struct e1000_adapter *adapter = netdev_priv(netdev);
4666
4667         /* Do the reset outside of interrupt context */
4668         adapter->tx_timeout_count++;
4669         schedule_work(&adapter->reset_task);
4670 }
4671
4672 static void e1000_reset_task(struct work_struct *work)
4673 {
4674         struct e1000_adapter *adapter;
4675         adapter = container_of(work, struct e1000_adapter, reset_task);
4676
4677         e1000e_dump(adapter);
4678         e_err("Reset adapter\n");
4679         e1000e_reinit_locked(adapter);
4680 }
4681
4682 /**
4683  * e1000_get_stats - Get System Network Statistics
4684  * @netdev: network interface device structure
4685  *
4686  * Returns the address of the device statistics structure.
4687  * The statistics are actually updated from the timer callback.
4688  **/
4689 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4690 {
4691         /* only return the current stats */
4692         return &netdev->stats;
4693 }
4694
4695 /**
4696  * e1000_change_mtu - Change the Maximum Transfer Unit
4697  * @netdev: network interface device structure
4698  * @new_mtu: new value for maximum frame size
4699  *
4700  * Returns 0 on success, negative on failure
4701  **/
4702 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4703 {
4704         struct e1000_adapter *adapter = netdev_priv(netdev);
4705         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4706
4707         /* Jumbo frame support */
4708         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4709             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4710                 e_err("Jumbo Frames not supported.\n");
4711                 return -EINVAL;
4712         }
4713
4714         /* Supported frame sizes */
4715         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4716             (max_frame > adapter->max_hw_frame_size)) {
4717                 e_err("Unsupported MTU setting\n");
4718                 return -EINVAL;
4719         }
4720
4721         /* 82573 Errata 17 */
4722         if (((adapter->hw.mac.type == e1000_82573) ||
4723              (adapter->hw.mac.type == e1000_82574)) &&
4724             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4725                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4726                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4727         }
4728
4729         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4730                 msleep(1);
4731         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4732         adapter->max_frame_size = max_frame;
4733         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4734         netdev->mtu = new_mtu;
4735         if (netif_running(netdev))
4736                 e1000e_down(adapter);
4737
4738         /*
4739          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4740          * means we reserve 2 more, this pushes us to allocate from the next
4741          * larger slab size.
4742          * i.e. RXBUFFER_2048 --> size-4096 slab
4743          * However with the new *_jumbo_rx* routines, jumbo receives will use
4744          * fragmented skbs
4745          */
4746
4747         if (max_frame <= 2048)
4748                 adapter->rx_buffer_len = 2048;
4749         else
4750                 adapter->rx_buffer_len = 4096;
4751
4752         /* adjust allocation if LPE protects us, and we aren't using SBP */
4753         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4754              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4755                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4756                                          + ETH_FCS_LEN;
4757
4758         if (netif_running(netdev))
4759                 e1000e_up(adapter);
4760         else
4761                 e1000e_reset(adapter);
4762
4763         clear_bit(__E1000_RESETTING, &adapter->state);
4764
4765         return 0;
4766 }
4767
4768 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4769                            int cmd)
4770 {
4771         struct e1000_adapter *adapter = netdev_priv(netdev);
4772         struct mii_ioctl_data *data = if_mii(ifr);
4773
4774         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4775                 return -EOPNOTSUPP;
4776
4777         switch (cmd) {
4778         case SIOCGMIIPHY:
4779                 data->phy_id = adapter->hw.phy.addr;
4780                 break;
4781         case SIOCGMIIREG:
4782                 e1000_phy_read_status(adapter);
4783
4784                 switch (data->reg_num & 0x1F) {
4785                 case MII_BMCR:
4786                         data->val_out = adapter->phy_regs.bmcr;
4787                         break;
4788                 case MII_BMSR:
4789                         data->val_out = adapter->phy_regs.bmsr;
4790                         break;
4791                 case MII_PHYSID1:
4792                         data->val_out = (adapter->hw.phy.id >> 16);
4793                         break;
4794                 case MII_PHYSID2:
4795                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4796                         break;
4797                 case MII_ADVERTISE:
4798                         data->val_out = adapter->phy_regs.advertise;
4799                         break;
4800                 case MII_LPA:
4801                         data->val_out = adapter->phy_regs.lpa;
4802                         break;
4803                 case MII_EXPANSION:
4804                         data->val_out = adapter->phy_regs.expansion;
4805                         break;
4806                 case MII_CTRL1000:
4807                         data->val_out = adapter->phy_regs.ctrl1000;
4808                         break;
4809                 case MII_STAT1000:
4810                         data->val_out = adapter->phy_regs.stat1000;
4811                         break;
4812                 case MII_ESTATUS:
4813                         data->val_out = adapter->phy_regs.estatus;
4814                         break;
4815                 default:
4816                         return -EIO;
4817                 }
4818                 break;
4819         case SIOCSMIIREG:
4820         default:
4821                 return -EOPNOTSUPP;
4822         }
4823         return 0;
4824 }
4825
4826 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4827 {
4828         switch (cmd) {
4829         case SIOCGMIIPHY:
4830         case SIOCGMIIREG:
4831         case SIOCSMIIREG:
4832                 return e1000_mii_ioctl(netdev, ifr, cmd);
4833         default:
4834                 return -EOPNOTSUPP;
4835         }
4836 }
4837
4838 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4839 {
4840         struct e1000_hw *hw = &adapter->hw;
4841         u32 i, mac_reg;
4842         u16 phy_reg;
4843         int retval = 0;
4844
4845         /* copy MAC RARs to PHY RARs */
4846         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4847                 mac_reg = er32(RAL(i));
4848                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4849                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4850                 mac_reg = er32(RAH(i));
4851                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4852                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4853         }
4854
4855         /* copy MAC MTA to PHY MTA */
4856         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4857                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4858                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4859                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4860         }
4861
4862         /* configure PHY Rx Control register */
4863         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4864         mac_reg = er32(RCTL);
4865         if (mac_reg & E1000_RCTL_UPE)
4866                 phy_reg |= BM_RCTL_UPE;
4867         if (mac_reg & E1000_RCTL_MPE)
4868                 phy_reg |= BM_RCTL_MPE;
4869         phy_reg &= ~(BM_RCTL_MO_MASK);
4870         if (mac_reg & E1000_RCTL_MO_3)
4871                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4872                                 << BM_RCTL_MO_SHIFT);
4873         if (mac_reg & E1000_RCTL_BAM)
4874                 phy_reg |= BM_RCTL_BAM;
4875         if (mac_reg & E1000_RCTL_PMCF)
4876                 phy_reg |= BM_RCTL_PMCF;
4877         mac_reg = er32(CTRL);
4878         if (mac_reg & E1000_CTRL_RFCE)
4879                 phy_reg |= BM_RCTL_RFCE;
4880         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4881
4882         /* enable PHY wakeup in MAC register */
4883         ew32(WUFC, wufc);
4884         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4885
4886         /* configure and enable PHY wakeup in PHY registers */
4887         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4888         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4889
4890         /* activate PHY wakeup */
4891         retval = hw->phy.ops.acquire(hw);
4892         if (retval) {
4893                 e_err("Could not acquire PHY\n");
4894                 return retval;
4895         }
4896         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4897                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4898         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4899         if (retval) {
4900                 e_err("Could not read PHY page 769\n");
4901                 goto out;
4902         }
4903         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4904         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4905         if (retval)
4906                 e_err("Could not set PHY Host Wakeup bit\n");
4907 out:
4908         hw->phy.ops.release(hw);
4909
4910         return retval;
4911 }
4912
4913 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4914                             bool runtime)
4915 {
4916         struct net_device *netdev = pci_get_drvdata(pdev);
4917         struct e1000_adapter *adapter = netdev_priv(netdev);
4918         struct e1000_hw *hw = &adapter->hw;
4919         u32 ctrl, ctrl_ext, rctl, status;
4920         /* Runtime suspend should only enable wakeup for link changes */
4921         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4922         int retval = 0;
4923
4924         netif_device_detach(netdev);
4925
4926         if (netif_running(netdev)) {
4927                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4928                 e1000e_down(adapter);
4929                 e1000_free_irq(adapter);
4930         }
4931         e1000e_reset_interrupt_capability(adapter);
4932
4933         retval = pci_save_state(pdev);
4934         if (retval)
4935                 return retval;
4936
4937         status = er32(STATUS);
4938         if (status & E1000_STATUS_LU)
4939                 wufc &= ~E1000_WUFC_LNKC;
4940
4941         if (wufc) {
4942                 e1000_setup_rctl(adapter);
4943                 e1000_set_multi(netdev);
4944
4945                 /* turn on all-multi mode if wake on multicast is enabled */
4946                 if (wufc & E1000_WUFC_MC) {
4947                         rctl = er32(RCTL);
4948                         rctl |= E1000_RCTL_MPE;
4949                         ew32(RCTL, rctl);
4950                 }
4951
4952                 ctrl = er32(CTRL);
4953                 /* advertise wake from D3Cold */
4954                 #define E1000_CTRL_ADVD3WUC 0x00100000
4955                 /* phy power management enable */
4956                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4957                 ctrl |= E1000_CTRL_ADVD3WUC;
4958                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4959                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4960                 ew32(CTRL, ctrl);
4961
4962                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4963                     adapter->hw.phy.media_type ==
4964                     e1000_media_type_internal_serdes) {
4965                         /* keep the laser running in D3 */
4966                         ctrl_ext = er32(CTRL_EXT);
4967                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4968                         ew32(CTRL_EXT, ctrl_ext);
4969                 }
4970
4971                 if (adapter->flags & FLAG_IS_ICH)
4972                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4973
4974                 /* Allow time for pending master requests to run */
4975                 e1000e_disable_pcie_master(&adapter->hw);
4976
4977                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4978                         /* enable wakeup by the PHY */
4979                         retval = e1000_init_phy_wakeup(adapter, wufc);
4980                         if (retval)
4981                                 return retval;
4982                 } else {
4983                         /* enable wakeup by the MAC */
4984                         ew32(WUFC, wufc);
4985                         ew32(WUC, E1000_WUC_PME_EN);
4986                 }
4987         } else {
4988                 ew32(WUC, 0);
4989                 ew32(WUFC, 0);
4990         }
4991
4992         *enable_wake = !!wufc;
4993
4994         /* make sure adapter isn't asleep if manageability is enabled */
4995         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4996             (hw->mac.ops.check_mng_mode(hw)))
4997                 *enable_wake = true;
4998
4999         if (adapter->hw.phy.type == e1000_phy_igp_3)
5000                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5001
5002         /*
5003          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5004          * would have already happened in close and is redundant.
5005          */
5006         e1000_release_hw_control(adapter);
5007
5008         pci_disable_device(pdev);
5009
5010         return 0;
5011 }
5012
5013 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5014 {
5015         if (sleep && wake) {
5016                 pci_prepare_to_sleep(pdev);
5017                 return;
5018         }
5019
5020         pci_wake_from_d3(pdev, wake);
5021         pci_set_power_state(pdev, PCI_D3hot);
5022 }
5023
5024 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5025                                     bool wake)
5026 {
5027         struct net_device *netdev = pci_get_drvdata(pdev);
5028         struct e1000_adapter *adapter = netdev_priv(netdev);
5029
5030         /*
5031          * The pci-e switch on some quad port adapters will report a
5032          * correctable error when the MAC transitions from D0 to D3.  To
5033          * prevent this we need to mask off the correctable errors on the
5034          * downstream port of the pci-e switch.
5035          */
5036         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5037                 struct pci_dev *us_dev = pdev->bus->self;
5038                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5039                 u16 devctl;
5040
5041                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5042                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5043                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5044
5045                 e1000_power_off(pdev, sleep, wake);
5046
5047                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5048         } else {
5049                 e1000_power_off(pdev, sleep, wake);
5050         }
5051 }
5052
5053 #ifdef CONFIG_PCIEASPM
5054 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5055 {
5056         pci_disable_link_state(pdev, state);
5057 }
5058 #else
5059 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5060 {
5061         int pos;
5062         u16 reg16;
5063
5064         /*
5065          * Both device and parent should have the same ASPM setting.
5066          * Disable ASPM in downstream component first and then upstream.
5067          */
5068         pos = pci_pcie_cap(pdev);
5069         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5070         reg16 &= ~state;
5071         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5072
5073         if (!pdev->bus->self)
5074                 return;
5075
5076         pos = pci_pcie_cap(pdev->bus->self);
5077         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5078         reg16 &= ~state;
5079         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5080 }
5081 #endif
5082 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5083 {
5084         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5085                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5086                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5087
5088         __e1000e_disable_aspm(pdev, state);
5089 }
5090
5091 #ifdef CONFIG_PM_OPS
5092 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5093 {
5094         return !!adapter->tx_ring->buffer_info;
5095 }
5096
5097 static int __e1000_resume(struct pci_dev *pdev)
5098 {
5099         struct net_device *netdev = pci_get_drvdata(pdev);
5100         struct e1000_adapter *adapter = netdev_priv(netdev);
5101         struct e1000_hw *hw = &adapter->hw;
5102         u32 err;
5103
5104         pci_set_power_state(pdev, PCI_D0);
5105         pci_restore_state(pdev);
5106         pci_save_state(pdev);
5107         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5108                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5109
5110         e1000e_set_interrupt_capability(adapter);
5111         if (netif_running(netdev)) {
5112                 err = e1000_request_irq(adapter);
5113                 if (err)
5114                         return err;
5115         }
5116
5117         e1000e_power_up_phy(adapter);
5118
5119         /* report the system wakeup cause from S3/S4 */
5120         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5121                 u16 phy_data;
5122
5123                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5124                 if (phy_data) {
5125                         e_info("PHY Wakeup cause - %s\n",
5126                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5127                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5128                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5129                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5130                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5131                                 " Change" : "other");
5132                 }
5133                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5134         } else {
5135                 u32 wus = er32(WUS);
5136                 if (wus) {
5137                         e_info("MAC Wakeup cause - %s\n",
5138                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5139                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5140                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5141                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5142                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5143                                 "other");
5144                 }
5145                 ew32(WUS, ~0);
5146         }
5147
5148         e1000e_reset(adapter);
5149
5150         e1000_init_manageability_pt(adapter);
5151
5152         if (netif_running(netdev))
5153                 e1000e_up(adapter);
5154
5155         netif_device_attach(netdev);
5156
5157         /*
5158          * If the controller has AMT, do not set DRV_LOAD until the interface
5159          * is up.  For all other cases, let the f/w know that the h/w is now
5160          * under the control of the driver.
5161          */
5162         if (!(adapter->flags & FLAG_HAS_AMT))
5163                 e1000_get_hw_control(adapter);
5164
5165         return 0;
5166 }
5167
5168 #ifdef CONFIG_PM_SLEEP
5169 static int e1000_suspend(struct device *dev)
5170 {
5171         struct pci_dev *pdev = to_pci_dev(dev);
5172         int retval;
5173         bool wake;
5174
5175         retval = __e1000_shutdown(pdev, &wake, false);
5176         if (!retval)
5177                 e1000_complete_shutdown(pdev, true, wake);
5178
5179         return retval;
5180 }
5181
5182 static int e1000_resume(struct device *dev)
5183 {
5184         struct pci_dev *pdev = to_pci_dev(dev);
5185         struct net_device *netdev = pci_get_drvdata(pdev);
5186         struct e1000_adapter *adapter = netdev_priv(netdev);
5187
5188         if (e1000e_pm_ready(adapter))
5189                 adapter->idle_check = true;
5190
5191         return __e1000_resume(pdev);
5192 }
5193 #endif /* CONFIG_PM_SLEEP */
5194
5195 #ifdef CONFIG_PM_RUNTIME
5196 static int e1000_runtime_suspend(struct device *dev)
5197 {
5198         struct pci_dev *pdev = to_pci_dev(dev);
5199         struct net_device *netdev = pci_get_drvdata(pdev);
5200         struct e1000_adapter *adapter = netdev_priv(netdev);
5201
5202         if (e1000e_pm_ready(adapter)) {
5203                 bool wake;
5204
5205                 __e1000_shutdown(pdev, &wake, true);
5206         }
5207
5208         return 0;
5209 }
5210
5211 static int e1000_idle(struct device *dev)
5212 {
5213         struct pci_dev *pdev = to_pci_dev(dev);
5214         struct net_device *netdev = pci_get_drvdata(pdev);
5215         struct e1000_adapter *adapter = netdev_priv(netdev);
5216
5217         if (!e1000e_pm_ready(adapter))
5218                 return 0;
5219
5220         if (adapter->idle_check) {
5221                 adapter->idle_check = false;
5222                 if (!e1000e_has_link(adapter))
5223                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5224         }
5225
5226         return -EBUSY;
5227 }
5228
5229 static int e1000_runtime_resume(struct device *dev)
5230 {
5231         struct pci_dev *pdev = to_pci_dev(dev);
5232         struct net_device *netdev = pci_get_drvdata(pdev);
5233         struct e1000_adapter *adapter = netdev_priv(netdev);
5234
5235         if (!e1000e_pm_ready(adapter))
5236                 return 0;
5237
5238         adapter->idle_check = !dev->power.runtime_auto;
5239         return __e1000_resume(pdev);
5240 }
5241 #endif /* CONFIG_PM_RUNTIME */
5242 #endif /* CONFIG_PM_OPS */
5243
5244 static void e1000_shutdown(struct pci_dev *pdev)
5245 {
5246         bool wake = false;
5247
5248         __e1000_shutdown(pdev, &wake, false);
5249
5250         if (system_state == SYSTEM_POWER_OFF)
5251                 e1000_complete_shutdown(pdev, false, wake);
5252 }
5253
5254 #ifdef CONFIG_NET_POLL_CONTROLLER
5255 /*
5256  * Polling 'interrupt' - used by things like netconsole to send skbs
5257  * without having to re-enable interrupts. It's not called while
5258  * the interrupt routine is executing.
5259  */
5260 static void e1000_netpoll(struct net_device *netdev)
5261 {
5262         struct e1000_adapter *adapter = netdev_priv(netdev);
5263
5264         disable_irq(adapter->pdev->irq);
5265         e1000_intr(adapter->pdev->irq, netdev);
5266
5267         enable_irq(adapter->pdev->irq);
5268 }
5269 #endif
5270
5271 /**
5272  * e1000_io_error_detected - called when PCI error is detected
5273  * @pdev: Pointer to PCI device
5274  * @state: The current pci connection state
5275  *
5276  * This function is called after a PCI bus error affecting
5277  * this device has been detected.
5278  */
5279 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5280                                                 pci_channel_state_t state)
5281 {
5282         struct net_device *netdev = pci_get_drvdata(pdev);
5283         struct e1000_adapter *adapter = netdev_priv(netdev);
5284
5285         netif_device_detach(netdev);
5286
5287         if (state == pci_channel_io_perm_failure)
5288                 return PCI_ERS_RESULT_DISCONNECT;
5289
5290         if (netif_running(netdev))
5291                 e1000e_down(adapter);
5292         pci_disable_device(pdev);
5293
5294         /* Request a slot slot reset. */
5295         return PCI_ERS_RESULT_NEED_RESET;
5296 }
5297
5298 /**
5299  * e1000_io_slot_reset - called after the pci bus has been reset.
5300  * @pdev: Pointer to PCI device
5301  *
5302  * Restart the card from scratch, as if from a cold-boot. Implementation
5303  * resembles the first-half of the e1000_resume routine.
5304  */
5305 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5306 {
5307         struct net_device *netdev = pci_get_drvdata(pdev);
5308         struct e1000_adapter *adapter = netdev_priv(netdev);
5309         struct e1000_hw *hw = &adapter->hw;
5310         int err;
5311         pci_ers_result_t result;
5312
5313         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5314                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5315         err = pci_enable_device_mem(pdev);
5316         if (err) {
5317                 dev_err(&pdev->dev,
5318                         "Cannot re-enable PCI device after reset.\n");
5319                 result = PCI_ERS_RESULT_DISCONNECT;
5320         } else {
5321                 pci_set_master(pdev);
5322                 pdev->state_saved = true;
5323                 pci_restore_state(pdev);
5324
5325                 pci_enable_wake(pdev, PCI_D3hot, 0);
5326                 pci_enable_wake(pdev, PCI_D3cold, 0);
5327
5328                 e1000e_reset(adapter);
5329                 ew32(WUS, ~0);
5330                 result = PCI_ERS_RESULT_RECOVERED;
5331         }
5332
5333         pci_cleanup_aer_uncorrect_error_status(pdev);
5334
5335         return result;
5336 }
5337
5338 /**
5339  * e1000_io_resume - called when traffic can start flowing again.
5340  * @pdev: Pointer to PCI device
5341  *
5342  * This callback is called when the error recovery driver tells us that
5343  * its OK to resume normal operation. Implementation resembles the
5344  * second-half of the e1000_resume routine.
5345  */
5346 static void e1000_io_resume(struct pci_dev *pdev)
5347 {
5348         struct net_device *netdev = pci_get_drvdata(pdev);
5349         struct e1000_adapter *adapter = netdev_priv(netdev);
5350
5351         e1000_init_manageability_pt(adapter);
5352
5353         if (netif_running(netdev)) {
5354                 if (e1000e_up(adapter)) {
5355                         dev_err(&pdev->dev,
5356                                 "can't bring device back up after reset\n");
5357                         return;
5358                 }
5359         }
5360
5361         netif_device_attach(netdev);
5362
5363         /*
5364          * If the controller has AMT, do not set DRV_LOAD until the interface
5365          * is up.  For all other cases, let the f/w know that the h/w is now
5366          * under the control of the driver.
5367          */
5368         if (!(adapter->flags & FLAG_HAS_AMT))
5369                 e1000_get_hw_control(adapter);
5370
5371 }
5372
5373 static void e1000_print_device_info(struct e1000_adapter *adapter)
5374 {
5375         struct e1000_hw *hw = &adapter->hw;
5376         struct net_device *netdev = adapter->netdev;
5377         u32 pba_num;
5378
5379         /* print bus type/speed/width info */
5380         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5381                /* bus width */
5382                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5383                 "Width x1"),
5384                /* MAC address */
5385                netdev->dev_addr);
5386         e_info("Intel(R) PRO/%s Network Connection\n",
5387                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5388         e1000e_read_pba_num(hw, &pba_num);
5389         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5390                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5391 }
5392
5393 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5394 {
5395         struct e1000_hw *hw = &adapter->hw;
5396         int ret_val;
5397         u16 buf = 0;
5398
5399         if (hw->mac.type != e1000_82573)
5400                 return;
5401
5402         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5403         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5404                 /* Deep Smart Power Down (DSPD) */
5405                 dev_warn(&adapter->pdev->dev,
5406                          "Warning: detected DSPD enabled in EEPROM\n");
5407         }
5408 }
5409
5410 static const struct net_device_ops e1000e_netdev_ops = {
5411         .ndo_open               = e1000_open,
5412         .ndo_stop               = e1000_close,
5413         .ndo_start_xmit         = e1000_xmit_frame,
5414         .ndo_get_stats          = e1000_get_stats,
5415         .ndo_set_multicast_list = e1000_set_multi,
5416         .ndo_set_mac_address    = e1000_set_mac,
5417         .ndo_change_mtu         = e1000_change_mtu,
5418         .ndo_do_ioctl           = e1000_ioctl,
5419         .ndo_tx_timeout         = e1000_tx_timeout,
5420         .ndo_validate_addr      = eth_validate_addr,
5421
5422         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5423         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5424         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5425 #ifdef CONFIG_NET_POLL_CONTROLLER
5426         .ndo_poll_controller    = e1000_netpoll,
5427 #endif
5428 };
5429
5430 /**
5431  * e1000_probe - Device Initialization Routine
5432  * @pdev: PCI device information struct
5433  * @ent: entry in e1000_pci_tbl
5434  *
5435  * Returns 0 on success, negative on failure
5436  *
5437  * e1000_probe initializes an adapter identified by a pci_dev structure.
5438  * The OS initialization, configuring of the adapter private structure,
5439  * and a hardware reset occur.
5440  **/
5441 static int __devinit e1000_probe(struct pci_dev *pdev,
5442                                  const struct pci_device_id *ent)
5443 {
5444         struct net_device *netdev;
5445         struct e1000_adapter *adapter;
5446         struct e1000_hw *hw;
5447         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5448         resource_size_t mmio_start, mmio_len;
5449         resource_size_t flash_start, flash_len;
5450
5451         static int cards_found;
5452         int i, err, pci_using_dac;
5453         u16 eeprom_data = 0;
5454         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5455
5456         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5457                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5458
5459         err = pci_enable_device_mem(pdev);
5460         if (err)
5461                 return err;
5462
5463         pci_using_dac = 0;
5464         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5465         if (!err) {
5466                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5467                 if (!err)
5468                         pci_using_dac = 1;
5469         } else {
5470                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5471                 if (err) {
5472                         err = dma_set_coherent_mask(&pdev->dev,
5473                                                     DMA_BIT_MASK(32));
5474                         if (err) {
5475                                 dev_err(&pdev->dev, "No usable DMA "
5476                                         "configuration, aborting\n");
5477                                 goto err_dma;
5478                         }
5479                 }
5480         }
5481
5482         err = pci_request_selected_regions_exclusive(pdev,
5483                                           pci_select_bars(pdev, IORESOURCE_MEM),
5484                                           e1000e_driver_name);
5485         if (err)
5486                 goto err_pci_reg;
5487
5488         /* AER (Advanced Error Reporting) hooks */
5489         pci_enable_pcie_error_reporting(pdev);
5490
5491         pci_set_master(pdev);
5492         /* PCI config space info */
5493         err = pci_save_state(pdev);
5494         if (err)
5495                 goto err_alloc_etherdev;
5496
5497         err = -ENOMEM;
5498         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5499         if (!netdev)
5500                 goto err_alloc_etherdev;
5501
5502         SET_NETDEV_DEV(netdev, &pdev->dev);
5503
5504         netdev->irq = pdev->irq;
5505
5506         pci_set_drvdata(pdev, netdev);
5507         adapter = netdev_priv(netdev);
5508         hw = &adapter->hw;
5509         adapter->netdev = netdev;
5510         adapter->pdev = pdev;
5511         adapter->ei = ei;
5512         adapter->pba = ei->pba;
5513         adapter->flags = ei->flags;
5514         adapter->flags2 = ei->flags2;
5515         adapter->hw.adapter = adapter;
5516         adapter->hw.mac.type = ei->mac;
5517         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5518         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5519
5520         mmio_start = pci_resource_start(pdev, 0);
5521         mmio_len = pci_resource_len(pdev, 0);
5522
5523         err = -EIO;
5524         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5525         if (!adapter->hw.hw_addr)
5526                 goto err_ioremap;
5527
5528         if ((adapter->flags & FLAG_HAS_FLASH) &&
5529             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5530                 flash_start = pci_resource_start(pdev, 1);
5531                 flash_len = pci_resource_len(pdev, 1);
5532                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5533                 if (!adapter->hw.flash_address)
5534                         goto err_flashmap;
5535         }
5536
5537         /* construct the net_device struct */
5538         netdev->netdev_ops              = &e1000e_netdev_ops;
5539         e1000e_set_ethtool_ops(netdev);
5540         netdev->watchdog_timeo          = 5 * HZ;
5541         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5542         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5543
5544         netdev->mem_start = mmio_start;
5545         netdev->mem_end = mmio_start + mmio_len;
5546
5547         adapter->bd_number = cards_found++;
5548
5549         e1000e_check_options(adapter);
5550
5551         /* setup adapter struct */
5552         err = e1000_sw_init(adapter);
5553         if (err)
5554                 goto err_sw_init;
5555
5556         err = -EIO;
5557
5558         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5559         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5560         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5561
5562         err = ei->get_variants(adapter);
5563         if (err)
5564                 goto err_hw_init;
5565
5566         if ((adapter->flags & FLAG_IS_ICH) &&
5567             (adapter->flags & FLAG_READ_ONLY_NVM))
5568                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5569
5570         hw->mac.ops.get_bus_info(&adapter->hw);
5571
5572         adapter->hw.phy.autoneg_wait_to_complete = 0;
5573
5574         /* Copper options */
5575         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5576                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5577                 adapter->hw.phy.disable_polarity_correction = 0;
5578                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5579         }
5580
5581         if (e1000_check_reset_block(&adapter->hw))
5582                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5583
5584         netdev->features = NETIF_F_SG |
5585                            NETIF_F_HW_CSUM |
5586                            NETIF_F_HW_VLAN_TX |
5587                            NETIF_F_HW_VLAN_RX;
5588
5589         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5590                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5591
5592         netdev->features |= NETIF_F_TSO;
5593         netdev->features |= NETIF_F_TSO6;
5594
5595         netdev->vlan_features |= NETIF_F_TSO;
5596         netdev->vlan_features |= NETIF_F_TSO6;
5597         netdev->vlan_features |= NETIF_F_HW_CSUM;
5598         netdev->vlan_features |= NETIF_F_SG;
5599
5600         if (pci_using_dac)
5601                 netdev->features |= NETIF_F_HIGHDMA;
5602
5603         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5604                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5605
5606         /*
5607          * before reading the NVM, reset the controller to
5608          * put the device in a known good starting state
5609          */
5610         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5611
5612         /*
5613          * systems with ASPM and others may see the checksum fail on the first
5614          * attempt. Let's give it a few tries
5615          */
5616         for (i = 0;; i++) {
5617                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5618                         break;
5619                 if (i == 2) {
5620                         e_err("The NVM Checksum Is Not Valid\n");
5621                         err = -EIO;
5622                         goto err_eeprom;
5623                 }
5624         }
5625
5626         e1000_eeprom_checks(adapter);
5627
5628         /* copy the MAC address */
5629         if (e1000e_read_mac_addr(&adapter->hw))
5630                 e_err("NVM Read Error while reading MAC address\n");
5631
5632         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5633         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5634
5635         if (!is_valid_ether_addr(netdev->perm_addr)) {
5636                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5637                 err = -EIO;
5638                 goto err_eeprom;
5639         }
5640
5641         init_timer(&adapter->watchdog_timer);
5642         adapter->watchdog_timer.function = &e1000_watchdog;
5643         adapter->watchdog_timer.data = (unsigned long) adapter;
5644
5645         init_timer(&adapter->phy_info_timer);
5646         adapter->phy_info_timer.function = &e1000_update_phy_info;
5647         adapter->phy_info_timer.data = (unsigned long) adapter;
5648
5649         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5650         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5651         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5652         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5653         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5654
5655         /* Initialize link parameters. User can change them with ethtool */
5656         adapter->hw.mac.autoneg = 1;
5657         adapter->fc_autoneg = 1;
5658         adapter->hw.fc.requested_mode = e1000_fc_default;
5659         adapter->hw.fc.current_mode = e1000_fc_default;
5660         adapter->hw.phy.autoneg_advertised = 0x2f;
5661
5662         /* ring size defaults */
5663         adapter->rx_ring->count = 256;
5664         adapter->tx_ring->count = 256;
5665
5666         /*
5667          * Initial Wake on LAN setting - If APM wake is enabled in
5668          * the EEPROM, enable the ACPI Magic Packet filter
5669          */
5670         if (adapter->flags & FLAG_APME_IN_WUC) {
5671                 /* APME bit in EEPROM is mapped to WUC.APME */
5672                 eeprom_data = er32(WUC);
5673                 eeprom_apme_mask = E1000_WUC_APME;
5674                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5675                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5676         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5677                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5678                     (adapter->hw.bus.func == 1))
5679                         e1000_read_nvm(&adapter->hw,
5680                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5681                 else
5682                         e1000_read_nvm(&adapter->hw,
5683                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5684         }
5685
5686         /* fetch WoL from EEPROM */
5687         if (eeprom_data & eeprom_apme_mask)
5688                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5689
5690         /*
5691          * now that we have the eeprom settings, apply the special cases
5692          * where the eeprom may be wrong or the board simply won't support
5693          * wake on lan on a particular port
5694          */
5695         if (!(adapter->flags & FLAG_HAS_WOL))
5696                 adapter->eeprom_wol = 0;
5697
5698         /* initialize the wol settings based on the eeprom settings */
5699         adapter->wol = adapter->eeprom_wol;
5700         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5701
5702         /* save off EEPROM version number */
5703         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5704
5705         /* reset the hardware with the new settings */
5706         e1000e_reset(adapter);
5707
5708         /*
5709          * If the controller has AMT, do not set DRV_LOAD until the interface
5710          * is up.  For all other cases, let the f/w know that the h/w is now
5711          * under the control of the driver.
5712          */
5713         if (!(adapter->flags & FLAG_HAS_AMT))
5714                 e1000_get_hw_control(adapter);
5715
5716         strcpy(netdev->name, "eth%d");
5717         err = register_netdev(netdev);
5718         if (err)
5719                 goto err_register;
5720
5721         /* carrier off reporting is important to ethtool even BEFORE open */
5722         netif_carrier_off(netdev);
5723
5724         e1000_print_device_info(adapter);
5725
5726         if (pci_dev_run_wake(pdev)) {
5727                 pm_runtime_set_active(&pdev->dev);
5728                 pm_runtime_enable(&pdev->dev);
5729         }
5730         pm_schedule_suspend(&pdev->dev, MSEC_PER_SEC);
5731
5732         return 0;
5733
5734 err_register:
5735         if (!(adapter->flags & FLAG_HAS_AMT))
5736                 e1000_release_hw_control(adapter);
5737 err_eeprom:
5738         if (!e1000_check_reset_block(&adapter->hw))
5739                 e1000_phy_hw_reset(&adapter->hw);
5740 err_hw_init:
5741
5742         kfree(adapter->tx_ring);
5743         kfree(adapter->rx_ring);
5744 err_sw_init:
5745         if (adapter->hw.flash_address)
5746                 iounmap(adapter->hw.flash_address);
5747         e1000e_reset_interrupt_capability(adapter);
5748 err_flashmap:
5749         iounmap(adapter->hw.hw_addr);
5750 err_ioremap:
5751         free_netdev(netdev);
5752 err_alloc_etherdev:
5753         pci_release_selected_regions(pdev,
5754                                      pci_select_bars(pdev, IORESOURCE_MEM));
5755 err_pci_reg:
5756 err_dma:
5757         pci_disable_device(pdev);
5758         return err;
5759 }
5760
5761 /**
5762  * e1000_remove - Device Removal Routine
5763  * @pdev: PCI device information struct
5764  *
5765  * e1000_remove is called by the PCI subsystem to alert the driver
5766  * that it should release a PCI device.  The could be caused by a
5767  * Hot-Plug event, or because the driver is going to be removed from
5768  * memory.
5769  **/
5770 static void __devexit e1000_remove(struct pci_dev *pdev)
5771 {
5772         struct net_device *netdev = pci_get_drvdata(pdev);
5773         struct e1000_adapter *adapter = netdev_priv(netdev);
5774         bool down = test_bit(__E1000_DOWN, &adapter->state);
5775
5776         pm_runtime_get_sync(&pdev->dev);
5777
5778         /*
5779          * flush_scheduled work may reschedule our watchdog task, so
5780          * explicitly disable watchdog tasks from being rescheduled
5781          */
5782         if (!down)
5783                 set_bit(__E1000_DOWN, &adapter->state);
5784         del_timer_sync(&adapter->watchdog_timer);
5785         del_timer_sync(&adapter->phy_info_timer);
5786
5787         cancel_work_sync(&adapter->reset_task);
5788         cancel_work_sync(&adapter->watchdog_task);
5789         cancel_work_sync(&adapter->downshift_task);
5790         cancel_work_sync(&adapter->update_phy_task);
5791         cancel_work_sync(&adapter->print_hang_task);
5792         flush_scheduled_work();
5793
5794         if (!(netdev->flags & IFF_UP))
5795                 e1000_power_down_phy(adapter);
5796
5797         /* Don't lie to e1000_close() down the road. */
5798         if (!down)
5799                 clear_bit(__E1000_DOWN, &adapter->state);
5800         unregister_netdev(netdev);
5801
5802         if (pci_dev_run_wake(pdev)) {
5803                 pm_runtime_disable(&pdev->dev);
5804                 pm_runtime_set_suspended(&pdev->dev);
5805         }
5806         pm_runtime_put_noidle(&pdev->dev);
5807
5808         /*
5809          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5810          * would have already happened in close and is redundant.
5811          */
5812         e1000_release_hw_control(adapter);
5813
5814         e1000e_reset_interrupt_capability(adapter);
5815         kfree(adapter->tx_ring);
5816         kfree(adapter->rx_ring);
5817
5818         iounmap(adapter->hw.hw_addr);
5819         if (adapter->hw.flash_address)
5820                 iounmap(adapter->hw.flash_address);
5821         pci_release_selected_regions(pdev,
5822                                      pci_select_bars(pdev, IORESOURCE_MEM));
5823
5824         free_netdev(netdev);
5825
5826         /* AER disable */
5827         pci_disable_pcie_error_reporting(pdev);
5828
5829         pci_disable_device(pdev);
5830 }
5831