PCI: change device runtime PM settings for probe and remove
[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
1772 /**
1773  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1774  *
1775  * Attempt to configure interrupts using the best available
1776  * capabilities of the hardware and kernel.
1777  **/
1778 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1779 {
1780         int err;
1781         int numvecs, i;
1782
1783
1784         switch (adapter->int_mode) {
1785         case E1000E_INT_MODE_MSIX:
1786                 if (adapter->flags & FLAG_HAS_MSIX) {
1787                         numvecs = 3; /* RxQ0, TxQ0 and other */
1788                         adapter->msix_entries = kcalloc(numvecs,
1789                                                       sizeof(struct msix_entry),
1790                                                       GFP_KERNEL);
1791                         if (adapter->msix_entries) {
1792                                 for (i = 0; i < numvecs; i++)
1793                                         adapter->msix_entries[i].entry = i;
1794
1795                                 err = pci_enable_msix(adapter->pdev,
1796                                                       adapter->msix_entries,
1797                                                       numvecs);
1798                                 if (err == 0)
1799                                         return;
1800                         }
1801                         /* MSI-X failed, so fall through and try MSI */
1802                         e_err("Failed to initialize MSI-X interrupts.  "
1803                               "Falling back to MSI interrupts.\n");
1804                         e1000e_reset_interrupt_capability(adapter);
1805                 }
1806                 adapter->int_mode = E1000E_INT_MODE_MSI;
1807                 /* Fall through */
1808         case E1000E_INT_MODE_MSI:
1809                 if (!pci_enable_msi(adapter->pdev)) {
1810                         adapter->flags |= FLAG_MSI_ENABLED;
1811                 } else {
1812                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1813                         e_err("Failed to initialize MSI interrupts.  Falling "
1814                               "back to legacy interrupts.\n");
1815                 }
1816                 /* Fall through */
1817         case E1000E_INT_MODE_LEGACY:
1818                 /* Don't do anything; this is the system default */
1819                 break;
1820         }
1821 }
1822
1823 /**
1824  * e1000_request_msix - Initialize MSI-X interrupts
1825  *
1826  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1827  * kernel.
1828  **/
1829 static int e1000_request_msix(struct e1000_adapter *adapter)
1830 {
1831         struct net_device *netdev = adapter->netdev;
1832         int err = 0, vector = 0;
1833
1834         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1835                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1836         else
1837                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1838         err = request_irq(adapter->msix_entries[vector].vector,
1839                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1840                           netdev);
1841         if (err)
1842                 goto out;
1843         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1844         adapter->rx_ring->itr_val = adapter->itr;
1845         vector++;
1846
1847         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1848                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1849         else
1850                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1851         err = request_irq(adapter->msix_entries[vector].vector,
1852                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1853                           netdev);
1854         if (err)
1855                 goto out;
1856         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1857         adapter->tx_ring->itr_val = adapter->itr;
1858         vector++;
1859
1860         err = request_irq(adapter->msix_entries[vector].vector,
1861                           e1000_msix_other, 0, netdev->name, netdev);
1862         if (err)
1863                 goto out;
1864
1865         e1000_configure_msix(adapter);
1866         return 0;
1867 out:
1868         return err;
1869 }
1870
1871 /**
1872  * e1000_request_irq - initialize interrupts
1873  *
1874  * Attempts to configure interrupts using the best available
1875  * capabilities of the hardware and kernel.
1876  **/
1877 static int e1000_request_irq(struct e1000_adapter *adapter)
1878 {
1879         struct net_device *netdev = adapter->netdev;
1880         int err;
1881
1882         if (adapter->msix_entries) {
1883                 err = e1000_request_msix(adapter);
1884                 if (!err)
1885                         return err;
1886                 /* fall back to MSI */
1887                 e1000e_reset_interrupt_capability(adapter);
1888                 adapter->int_mode = E1000E_INT_MODE_MSI;
1889                 e1000e_set_interrupt_capability(adapter);
1890         }
1891         if (adapter->flags & FLAG_MSI_ENABLED) {
1892                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1893                                   netdev->name, netdev);
1894                 if (!err)
1895                         return err;
1896
1897                 /* fall back to legacy interrupt */
1898                 e1000e_reset_interrupt_capability(adapter);
1899                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1900         }
1901
1902         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1903                           netdev->name, netdev);
1904         if (err)
1905                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1906
1907         return err;
1908 }
1909
1910 static void e1000_free_irq(struct e1000_adapter *adapter)
1911 {
1912         struct net_device *netdev = adapter->netdev;
1913
1914         if (adapter->msix_entries) {
1915                 int vector = 0;
1916
1917                 free_irq(adapter->msix_entries[vector].vector, netdev);
1918                 vector++;
1919
1920                 free_irq(adapter->msix_entries[vector].vector, netdev);
1921                 vector++;
1922
1923                 /* Other Causes interrupt vector */
1924                 free_irq(adapter->msix_entries[vector].vector, netdev);
1925                 return;
1926         }
1927
1928         free_irq(adapter->pdev->irq, netdev);
1929 }
1930
1931 /**
1932  * e1000_irq_disable - Mask off interrupt generation on the NIC
1933  **/
1934 static void e1000_irq_disable(struct e1000_adapter *adapter)
1935 {
1936         struct e1000_hw *hw = &adapter->hw;
1937
1938         ew32(IMC, ~0);
1939         if (adapter->msix_entries)
1940                 ew32(EIAC_82574, 0);
1941         e1e_flush();
1942         synchronize_irq(adapter->pdev->irq);
1943 }
1944
1945 /**
1946  * e1000_irq_enable - Enable default interrupt generation settings
1947  **/
1948 static void e1000_irq_enable(struct e1000_adapter *adapter)
1949 {
1950         struct e1000_hw *hw = &adapter->hw;
1951
1952         if (adapter->msix_entries) {
1953                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1954                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1955         } else {
1956                 ew32(IMS, IMS_ENABLE_MASK);
1957         }
1958         e1e_flush();
1959 }
1960
1961 /**
1962  * e1000_get_hw_control - get control of the h/w from f/w
1963  * @adapter: address of board private structure
1964  *
1965  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1966  * For ASF and Pass Through versions of f/w this means that
1967  * the driver is loaded. For AMT version (only with 82573)
1968  * of the f/w this means that the network i/f is open.
1969  **/
1970 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1971 {
1972         struct e1000_hw *hw = &adapter->hw;
1973         u32 ctrl_ext;
1974         u32 swsm;
1975
1976         /* Let firmware know the driver has taken over */
1977         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1978                 swsm = er32(SWSM);
1979                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1980         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1981                 ctrl_ext = er32(CTRL_EXT);
1982                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1983         }
1984 }
1985
1986 /**
1987  * e1000_release_hw_control - release control of the h/w to f/w
1988  * @adapter: address of board private structure
1989  *
1990  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1991  * For ASF and Pass Through versions of f/w this means that the
1992  * driver is no longer loaded. For AMT version (only with 82573) i
1993  * of the f/w this means that the network i/f is closed.
1994  *
1995  **/
1996 static void e1000_release_hw_control(struct e1000_adapter *adapter)
1997 {
1998         struct e1000_hw *hw = &adapter->hw;
1999         u32 ctrl_ext;
2000         u32 swsm;
2001
2002         /* Let firmware taken over control of h/w */
2003         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2004                 swsm = er32(SWSM);
2005                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2006         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2007                 ctrl_ext = er32(CTRL_EXT);
2008                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2009         }
2010 }
2011
2012 /**
2013  * @e1000_alloc_ring - allocate memory for a ring structure
2014  **/
2015 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2016                                 struct e1000_ring *ring)
2017 {
2018         struct pci_dev *pdev = adapter->pdev;
2019
2020         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2021                                         GFP_KERNEL);
2022         if (!ring->desc)
2023                 return -ENOMEM;
2024
2025         return 0;
2026 }
2027
2028 /**
2029  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2030  * @adapter: board private structure
2031  *
2032  * Return 0 on success, negative on failure
2033  **/
2034 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2035 {
2036         struct e1000_ring *tx_ring = adapter->tx_ring;
2037         int err = -ENOMEM, size;
2038
2039         size = sizeof(struct e1000_buffer) * tx_ring->count;
2040         tx_ring->buffer_info = vmalloc(size);
2041         if (!tx_ring->buffer_info)
2042                 goto err;
2043         memset(tx_ring->buffer_info, 0, size);
2044
2045         /* round up to nearest 4K */
2046         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2047         tx_ring->size = ALIGN(tx_ring->size, 4096);
2048
2049         err = e1000_alloc_ring_dma(adapter, tx_ring);
2050         if (err)
2051                 goto err;
2052
2053         tx_ring->next_to_use = 0;
2054         tx_ring->next_to_clean = 0;
2055
2056         return 0;
2057 err:
2058         vfree(tx_ring->buffer_info);
2059         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2060         return err;
2061 }
2062
2063 /**
2064  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2065  * @adapter: board private structure
2066  *
2067  * Returns 0 on success, negative on failure
2068  **/
2069 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2070 {
2071         struct e1000_ring *rx_ring = adapter->rx_ring;
2072         struct e1000_buffer *buffer_info;
2073         int i, size, desc_len, err = -ENOMEM;
2074
2075         size = sizeof(struct e1000_buffer) * rx_ring->count;
2076         rx_ring->buffer_info = vmalloc(size);
2077         if (!rx_ring->buffer_info)
2078                 goto err;
2079         memset(rx_ring->buffer_info, 0, size);
2080
2081         for (i = 0; i < rx_ring->count; i++) {
2082                 buffer_info = &rx_ring->buffer_info[i];
2083                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2084                                                 sizeof(struct e1000_ps_page),
2085                                                 GFP_KERNEL);
2086                 if (!buffer_info->ps_pages)
2087                         goto err_pages;
2088         }
2089
2090         desc_len = sizeof(union e1000_rx_desc_packet_split);
2091
2092         /* Round up to nearest 4K */
2093         rx_ring->size = rx_ring->count * desc_len;
2094         rx_ring->size = ALIGN(rx_ring->size, 4096);
2095
2096         err = e1000_alloc_ring_dma(adapter, rx_ring);
2097         if (err)
2098                 goto err_pages;
2099
2100         rx_ring->next_to_clean = 0;
2101         rx_ring->next_to_use = 0;
2102         rx_ring->rx_skb_top = NULL;
2103
2104         return 0;
2105
2106 err_pages:
2107         for (i = 0; i < rx_ring->count; i++) {
2108                 buffer_info = &rx_ring->buffer_info[i];
2109                 kfree(buffer_info->ps_pages);
2110         }
2111 err:
2112         vfree(rx_ring->buffer_info);
2113         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2114         return err;
2115 }
2116
2117 /**
2118  * e1000_clean_tx_ring - Free Tx Buffers
2119  * @adapter: board private structure
2120  **/
2121 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2122 {
2123         struct e1000_ring *tx_ring = adapter->tx_ring;
2124         struct e1000_buffer *buffer_info;
2125         unsigned long size;
2126         unsigned int i;
2127
2128         for (i = 0; i < tx_ring->count; i++) {
2129                 buffer_info = &tx_ring->buffer_info[i];
2130                 e1000_put_txbuf(adapter, buffer_info);
2131         }
2132
2133         size = sizeof(struct e1000_buffer) * tx_ring->count;
2134         memset(tx_ring->buffer_info, 0, size);
2135
2136         memset(tx_ring->desc, 0, tx_ring->size);
2137
2138         tx_ring->next_to_use = 0;
2139         tx_ring->next_to_clean = 0;
2140
2141         writel(0, adapter->hw.hw_addr + tx_ring->head);
2142         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2143 }
2144
2145 /**
2146  * e1000e_free_tx_resources - Free Tx Resources per Queue
2147  * @adapter: board private structure
2148  *
2149  * Free all transmit software resources
2150  **/
2151 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2152 {
2153         struct pci_dev *pdev = adapter->pdev;
2154         struct e1000_ring *tx_ring = adapter->tx_ring;
2155
2156         e1000_clean_tx_ring(adapter);
2157
2158         vfree(tx_ring->buffer_info);
2159         tx_ring->buffer_info = NULL;
2160
2161         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2162                           tx_ring->dma);
2163         tx_ring->desc = NULL;
2164 }
2165
2166 /**
2167  * e1000e_free_rx_resources - Free Rx Resources
2168  * @adapter: board private structure
2169  *
2170  * Free all receive software resources
2171  **/
2172
2173 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2174 {
2175         struct pci_dev *pdev = adapter->pdev;
2176         struct e1000_ring *rx_ring = adapter->rx_ring;
2177         int i;
2178
2179         e1000_clean_rx_ring(adapter);
2180
2181         for (i = 0; i < rx_ring->count; i++) {
2182                 kfree(rx_ring->buffer_info[i].ps_pages);
2183         }
2184
2185         vfree(rx_ring->buffer_info);
2186         rx_ring->buffer_info = NULL;
2187
2188         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2189                           rx_ring->dma);
2190         rx_ring->desc = NULL;
2191 }
2192
2193 /**
2194  * e1000_update_itr - update the dynamic ITR value based on statistics
2195  * @adapter: pointer to adapter
2196  * @itr_setting: current adapter->itr
2197  * @packets: the number of packets during this measurement interval
2198  * @bytes: the number of bytes during this measurement interval
2199  *
2200  *      Stores a new ITR value based on packets and byte
2201  *      counts during the last interrupt.  The advantage of per interrupt
2202  *      computation is faster updates and more accurate ITR for the current
2203  *      traffic pattern.  Constants in this function were computed
2204  *      based on theoretical maximum wire speed and thresholds were set based
2205  *      on testing data as well as attempting to minimize response time
2206  *      while increasing bulk throughput.  This functionality is controlled
2207  *      by the InterruptThrottleRate module parameter.
2208  **/
2209 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2210                                      u16 itr_setting, int packets,
2211                                      int bytes)
2212 {
2213         unsigned int retval = itr_setting;
2214
2215         if (packets == 0)
2216                 goto update_itr_done;
2217
2218         switch (itr_setting) {
2219         case lowest_latency:
2220                 /* handle TSO and jumbo frames */
2221                 if (bytes/packets > 8000)
2222                         retval = bulk_latency;
2223                 else if ((packets < 5) && (bytes > 512)) {
2224                         retval = low_latency;
2225                 }
2226                 break;
2227         case low_latency:  /* 50 usec aka 20000 ints/s */
2228                 if (bytes > 10000) {
2229                         /* this if handles the TSO accounting */
2230                         if (bytes/packets > 8000) {
2231                                 retval = bulk_latency;
2232                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2233                                 retval = bulk_latency;
2234                         } else if ((packets > 35)) {
2235                                 retval = lowest_latency;
2236                         }
2237                 } else if (bytes/packets > 2000) {
2238                         retval = bulk_latency;
2239                 } else if (packets <= 2 && bytes < 512) {
2240                         retval = lowest_latency;
2241                 }
2242                 break;
2243         case bulk_latency: /* 250 usec aka 4000 ints/s */
2244                 if (bytes > 25000) {
2245                         if (packets > 35) {
2246                                 retval = low_latency;
2247                         }
2248                 } else if (bytes < 6000) {
2249                         retval = low_latency;
2250                 }
2251                 break;
2252         }
2253
2254 update_itr_done:
2255         return retval;
2256 }
2257
2258 static void e1000_set_itr(struct e1000_adapter *adapter)
2259 {
2260         struct e1000_hw *hw = &adapter->hw;
2261         u16 current_itr;
2262         u32 new_itr = adapter->itr;
2263
2264         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2265         if (adapter->link_speed != SPEED_1000) {
2266                 current_itr = 0;
2267                 new_itr = 4000;
2268                 goto set_itr_now;
2269         }
2270
2271         adapter->tx_itr = e1000_update_itr(adapter,
2272                                     adapter->tx_itr,
2273                                     adapter->total_tx_packets,
2274                                     adapter->total_tx_bytes);
2275         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2276         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2277                 adapter->tx_itr = low_latency;
2278
2279         adapter->rx_itr = e1000_update_itr(adapter,
2280                                     adapter->rx_itr,
2281                                     adapter->total_rx_packets,
2282                                     adapter->total_rx_bytes);
2283         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2284         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2285                 adapter->rx_itr = low_latency;
2286
2287         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2288
2289         switch (current_itr) {
2290         /* counts and packets in update_itr are dependent on these numbers */
2291         case lowest_latency:
2292                 new_itr = 70000;
2293                 break;
2294         case low_latency:
2295                 new_itr = 20000; /* aka hwitr = ~200 */
2296                 break;
2297         case bulk_latency:
2298                 new_itr = 4000;
2299                 break;
2300         default:
2301                 break;
2302         }
2303
2304 set_itr_now:
2305         if (new_itr != adapter->itr) {
2306                 /*
2307                  * this attempts to bias the interrupt rate towards Bulk
2308                  * by adding intermediate steps when interrupt rate is
2309                  * increasing
2310                  */
2311                 new_itr = new_itr > adapter->itr ?
2312                              min(adapter->itr + (new_itr >> 2), new_itr) :
2313                              new_itr;
2314                 adapter->itr = new_itr;
2315                 adapter->rx_ring->itr_val = new_itr;
2316                 if (adapter->msix_entries)
2317                         adapter->rx_ring->set_itr = 1;
2318                 else
2319                         ew32(ITR, 1000000000 / (new_itr * 256));
2320         }
2321 }
2322
2323 /**
2324  * e1000_alloc_queues - Allocate memory for all rings
2325  * @adapter: board private structure to initialize
2326  **/
2327 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2328 {
2329         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2330         if (!adapter->tx_ring)
2331                 goto err;
2332
2333         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2334         if (!adapter->rx_ring)
2335                 goto err;
2336
2337         return 0;
2338 err:
2339         e_err("Unable to allocate memory for queues\n");
2340         kfree(adapter->rx_ring);
2341         kfree(adapter->tx_ring);
2342         return -ENOMEM;
2343 }
2344
2345 /**
2346  * e1000_clean - NAPI Rx polling callback
2347  * @napi: struct associated with this polling callback
2348  * @budget: amount of packets driver is allowed to process this poll
2349  **/
2350 static int e1000_clean(struct napi_struct *napi, int budget)
2351 {
2352         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2353         struct e1000_hw *hw = &adapter->hw;
2354         struct net_device *poll_dev = adapter->netdev;
2355         int tx_cleaned = 1, work_done = 0;
2356
2357         adapter = netdev_priv(poll_dev);
2358
2359         if (adapter->msix_entries &&
2360             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2361                 goto clean_rx;
2362
2363         tx_cleaned = e1000_clean_tx_irq(adapter);
2364
2365 clean_rx:
2366         adapter->clean_rx(adapter, &work_done, budget);
2367
2368         if (!tx_cleaned)
2369                 work_done = budget;
2370
2371         /* If budget not fully consumed, exit the polling mode */
2372         if (work_done < budget) {
2373                 if (adapter->itr_setting & 3)
2374                         e1000_set_itr(adapter);
2375                 napi_complete(napi);
2376                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2377                         if (adapter->msix_entries)
2378                                 ew32(IMS, adapter->rx_ring->ims_val);
2379                         else
2380                                 e1000_irq_enable(adapter);
2381                 }
2382         }
2383
2384         return work_done;
2385 }
2386
2387 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2388 {
2389         struct e1000_adapter *adapter = netdev_priv(netdev);
2390         struct e1000_hw *hw = &adapter->hw;
2391         u32 vfta, index;
2392
2393         /* don't update vlan cookie if already programmed */
2394         if ((adapter->hw.mng_cookie.status &
2395              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2396             (vid == adapter->mng_vlan_id))
2397                 return;
2398
2399         /* add VID to filter table */
2400         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2401                 index = (vid >> 5) & 0x7F;
2402                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2403                 vfta |= (1 << (vid & 0x1F));
2404                 hw->mac.ops.write_vfta(hw, index, vfta);
2405         }
2406 }
2407
2408 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2409 {
2410         struct e1000_adapter *adapter = netdev_priv(netdev);
2411         struct e1000_hw *hw = &adapter->hw;
2412         u32 vfta, index;
2413
2414         if (!test_bit(__E1000_DOWN, &adapter->state))
2415                 e1000_irq_disable(adapter);
2416         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2417
2418         if (!test_bit(__E1000_DOWN, &adapter->state))
2419                 e1000_irq_enable(adapter);
2420
2421         if ((adapter->hw.mng_cookie.status &
2422              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2423             (vid == adapter->mng_vlan_id)) {
2424                 /* release control to f/w */
2425                 e1000_release_hw_control(adapter);
2426                 return;
2427         }
2428
2429         /* remove VID from filter table */
2430         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2431                 index = (vid >> 5) & 0x7F;
2432                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2433                 vfta &= ~(1 << (vid & 0x1F));
2434                 hw->mac.ops.write_vfta(hw, index, vfta);
2435         }
2436 }
2437
2438 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2439 {
2440         struct net_device *netdev = adapter->netdev;
2441         u16 vid = adapter->hw.mng_cookie.vlan_id;
2442         u16 old_vid = adapter->mng_vlan_id;
2443
2444         if (!adapter->vlgrp)
2445                 return;
2446
2447         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2448                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2449                 if (adapter->hw.mng_cookie.status &
2450                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2451                         e1000_vlan_rx_add_vid(netdev, vid);
2452                         adapter->mng_vlan_id = vid;
2453                 }
2454
2455                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2456                                 (vid != old_vid) &&
2457                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2458                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2459         } else {
2460                 adapter->mng_vlan_id = vid;
2461         }
2462 }
2463
2464
2465 static void e1000_vlan_rx_register(struct net_device *netdev,
2466                                    struct vlan_group *grp)
2467 {
2468         struct e1000_adapter *adapter = netdev_priv(netdev);
2469         struct e1000_hw *hw = &adapter->hw;
2470         u32 ctrl, rctl;
2471
2472         if (!test_bit(__E1000_DOWN, &adapter->state))
2473                 e1000_irq_disable(adapter);
2474         adapter->vlgrp = grp;
2475
2476         if (grp) {
2477                 /* enable VLAN tag insert/strip */
2478                 ctrl = er32(CTRL);
2479                 ctrl |= E1000_CTRL_VME;
2480                 ew32(CTRL, ctrl);
2481
2482                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2483                         /* enable VLAN receive filtering */
2484                         rctl = er32(RCTL);
2485                         rctl &= ~E1000_RCTL_CFIEN;
2486                         ew32(RCTL, rctl);
2487                         e1000_update_mng_vlan(adapter);
2488                 }
2489         } else {
2490                 /* disable VLAN tag insert/strip */
2491                 ctrl = er32(CTRL);
2492                 ctrl &= ~E1000_CTRL_VME;
2493                 ew32(CTRL, ctrl);
2494
2495                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2496                         if (adapter->mng_vlan_id !=
2497                             (u16)E1000_MNG_VLAN_NONE) {
2498                                 e1000_vlan_rx_kill_vid(netdev,
2499                                                        adapter->mng_vlan_id);
2500                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2501                         }
2502                 }
2503         }
2504
2505         if (!test_bit(__E1000_DOWN, &adapter->state))
2506                 e1000_irq_enable(adapter);
2507 }
2508
2509 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2510 {
2511         u16 vid;
2512
2513         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2514
2515         if (!adapter->vlgrp)
2516                 return;
2517
2518         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2519                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2520                         continue;
2521                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2522         }
2523 }
2524
2525 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2526 {
2527         struct e1000_hw *hw = &adapter->hw;
2528         u32 manc, manc2h, mdef, i, j;
2529
2530         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2531                 return;
2532
2533         manc = er32(MANC);
2534
2535         /*
2536          * enable receiving management packets to the host. this will probably
2537          * generate destination unreachable messages from the host OS, but
2538          * the packets will be handled on SMBUS
2539          */
2540         manc |= E1000_MANC_EN_MNG2HOST;
2541         manc2h = er32(MANC2H);
2542
2543         switch (hw->mac.type) {
2544         default:
2545                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2546                 break;
2547         case e1000_82574:
2548         case e1000_82583:
2549                 /*
2550                  * Check if IPMI pass-through decision filter already exists;
2551                  * if so, enable it.
2552                  */
2553                 for (i = 0, j = 0; i < 8; i++) {
2554                         mdef = er32(MDEF(i));
2555
2556                         /* Ignore filters with anything other than IPMI ports */
2557                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2558                                 continue;
2559
2560                         /* Enable this decision filter in MANC2H */
2561                         if (mdef)
2562                                 manc2h |= (1 << i);
2563
2564                         j |= mdef;
2565                 }
2566
2567                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2568                         break;
2569
2570                 /* Create new decision filter in an empty filter */
2571                 for (i = 0, j = 0; i < 8; i++)
2572                         if (er32(MDEF(i)) == 0) {
2573                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2574                                                E1000_MDEF_PORT_664));
2575                                 manc2h |= (1 << 1);
2576                                 j++;
2577                                 break;
2578                         }
2579
2580                 if (!j)
2581                         e_warn("Unable to create IPMI pass-through filter\n");
2582                 break;
2583         }
2584
2585         ew32(MANC2H, manc2h);
2586         ew32(MANC, manc);
2587 }
2588
2589 /**
2590  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2591  * @adapter: board private structure
2592  *
2593  * Configure the Tx unit of the MAC after a reset.
2594  **/
2595 static void e1000_configure_tx(struct e1000_adapter *adapter)
2596 {
2597         struct e1000_hw *hw = &adapter->hw;
2598         struct e1000_ring *tx_ring = adapter->tx_ring;
2599         u64 tdba;
2600         u32 tdlen, tctl, tipg, tarc;
2601         u32 ipgr1, ipgr2;
2602
2603         /* Setup the HW Tx Head and Tail descriptor pointers */
2604         tdba = tx_ring->dma;
2605         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2606         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2607         ew32(TDBAH, (tdba >> 32));
2608         ew32(TDLEN, tdlen);
2609         ew32(TDH, 0);
2610         ew32(TDT, 0);
2611         tx_ring->head = E1000_TDH;
2612         tx_ring->tail = E1000_TDT;
2613
2614         /* Set the default values for the Tx Inter Packet Gap timer */
2615         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2616         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2617         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2618
2619         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2620                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2621
2622         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2623         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2624         ew32(TIPG, tipg);
2625
2626         /* Set the Tx Interrupt Delay register */
2627         ew32(TIDV, adapter->tx_int_delay);
2628         /* Tx irq moderation */
2629         ew32(TADV, adapter->tx_abs_int_delay);
2630
2631         /* Program the Transmit Control Register */
2632         tctl = er32(TCTL);
2633         tctl &= ~E1000_TCTL_CT;
2634         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2635                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2636
2637         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2638                 tarc = er32(TARC(0));
2639                 /*
2640                  * set the speed mode bit, we'll clear it if we're not at
2641                  * gigabit link later
2642                  */
2643 #define SPEED_MODE_BIT (1 << 21)
2644                 tarc |= SPEED_MODE_BIT;
2645                 ew32(TARC(0), tarc);
2646         }
2647
2648         /* errata: program both queues to unweighted RR */
2649         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2650                 tarc = er32(TARC(0));
2651                 tarc |= 1;
2652                 ew32(TARC(0), tarc);
2653                 tarc = er32(TARC(1));
2654                 tarc |= 1;
2655                 ew32(TARC(1), tarc);
2656         }
2657
2658         /* Setup Transmit Descriptor Settings for eop descriptor */
2659         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2660
2661         /* only set IDE if we are delaying interrupts using the timers */
2662         if (adapter->tx_int_delay)
2663                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2664
2665         /* enable Report Status bit */
2666         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2667
2668         ew32(TCTL, tctl);
2669
2670         e1000e_config_collision_dist(hw);
2671 }
2672
2673 /**
2674  * e1000_setup_rctl - configure the receive control registers
2675  * @adapter: Board private structure
2676  **/
2677 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2678                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2679 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2680 {
2681         struct e1000_hw *hw = &adapter->hw;
2682         u32 rctl, rfctl;
2683         u32 psrctl = 0;
2684         u32 pages = 0;
2685
2686         /* Program MC offset vector base */
2687         rctl = er32(RCTL);
2688         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2689         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2690                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2691                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2692
2693         /* Do not Store bad packets */
2694         rctl &= ~E1000_RCTL_SBP;
2695
2696         /* Enable Long Packet receive */
2697         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2698                 rctl &= ~E1000_RCTL_LPE;
2699         else
2700                 rctl |= E1000_RCTL_LPE;
2701
2702         /* Some systems expect that the CRC is included in SMBUS traffic. The
2703          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2704          * host memory when this is enabled
2705          */
2706         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2707                 rctl |= E1000_RCTL_SECRC;
2708
2709         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2710         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2711                 u16 phy_data;
2712
2713                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2714                 phy_data &= 0xfff8;
2715                 phy_data |= (1 << 2);
2716                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2717
2718                 e1e_rphy(hw, 22, &phy_data);
2719                 phy_data &= 0x0fff;
2720                 phy_data |= (1 << 14);
2721                 e1e_wphy(hw, 0x10, 0x2823);
2722                 e1e_wphy(hw, 0x11, 0x0003);
2723                 e1e_wphy(hw, 22, phy_data);
2724         }
2725
2726         /* Setup buffer sizes */
2727         rctl &= ~E1000_RCTL_SZ_4096;
2728         rctl |= E1000_RCTL_BSEX;
2729         switch (adapter->rx_buffer_len) {
2730         case 2048:
2731         default:
2732                 rctl |= E1000_RCTL_SZ_2048;
2733                 rctl &= ~E1000_RCTL_BSEX;
2734                 break;
2735         case 4096:
2736                 rctl |= E1000_RCTL_SZ_4096;
2737                 break;
2738         case 8192:
2739                 rctl |= E1000_RCTL_SZ_8192;
2740                 break;
2741         case 16384:
2742                 rctl |= E1000_RCTL_SZ_16384;
2743                 break;
2744         }
2745
2746         /*
2747          * 82571 and greater support packet-split where the protocol
2748          * header is placed in skb->data and the packet data is
2749          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2750          * In the case of a non-split, skb->data is linearly filled,
2751          * followed by the page buffers.  Therefore, skb->data is
2752          * sized to hold the largest protocol header.
2753          *
2754          * allocations using alloc_page take too long for regular MTU
2755          * so only enable packet split for jumbo frames
2756          *
2757          * Using pages when the page size is greater than 16k wastes
2758          * a lot of memory, since we allocate 3 pages at all times
2759          * per packet.
2760          */
2761         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2762         if (!(adapter->flags & FLAG_IS_ICH) && (pages <= 3) &&
2763             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2764                 adapter->rx_ps_pages = pages;
2765         else
2766                 adapter->rx_ps_pages = 0;
2767
2768         if (adapter->rx_ps_pages) {
2769                 /* Configure extra packet-split registers */
2770                 rfctl = er32(RFCTL);
2771                 rfctl |= E1000_RFCTL_EXTEN;
2772                 /*
2773                  * disable packet split support for IPv6 extension headers,
2774                  * because some malformed IPv6 headers can hang the Rx
2775                  */
2776                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2777                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2778
2779                 ew32(RFCTL, rfctl);
2780
2781                 /* Enable Packet split descriptors */
2782                 rctl |= E1000_RCTL_DTYP_PS;
2783
2784                 psrctl |= adapter->rx_ps_bsize0 >>
2785                         E1000_PSRCTL_BSIZE0_SHIFT;
2786
2787                 switch (adapter->rx_ps_pages) {
2788                 case 3:
2789                         psrctl |= PAGE_SIZE <<
2790                                 E1000_PSRCTL_BSIZE3_SHIFT;
2791                 case 2:
2792                         psrctl |= PAGE_SIZE <<
2793                                 E1000_PSRCTL_BSIZE2_SHIFT;
2794                 case 1:
2795                         psrctl |= PAGE_SIZE >>
2796                                 E1000_PSRCTL_BSIZE1_SHIFT;
2797                         break;
2798                 }
2799
2800                 ew32(PSRCTL, psrctl);
2801         }
2802
2803         ew32(RCTL, rctl);
2804         /* just started the receive unit, no need to restart */
2805         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2806 }
2807
2808 /**
2809  * e1000_configure_rx - Configure Receive Unit after Reset
2810  * @adapter: board private structure
2811  *
2812  * Configure the Rx unit of the MAC after a reset.
2813  **/
2814 static void e1000_configure_rx(struct e1000_adapter *adapter)
2815 {
2816         struct e1000_hw *hw = &adapter->hw;
2817         struct e1000_ring *rx_ring = adapter->rx_ring;
2818         u64 rdba;
2819         u32 rdlen, rctl, rxcsum, ctrl_ext;
2820
2821         if (adapter->rx_ps_pages) {
2822                 /* this is a 32 byte descriptor */
2823                 rdlen = rx_ring->count *
2824                         sizeof(union e1000_rx_desc_packet_split);
2825                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2826                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2827         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2828                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2829                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2830                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2831         } else {
2832                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2833                 adapter->clean_rx = e1000_clean_rx_irq;
2834                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2835         }
2836
2837         /* disable receives while setting up the descriptors */
2838         rctl = er32(RCTL);
2839         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2840         e1e_flush();
2841         msleep(10);
2842
2843         /* set the Receive Delay Timer Register */
2844         ew32(RDTR, adapter->rx_int_delay);
2845
2846         /* irq moderation */
2847         ew32(RADV, adapter->rx_abs_int_delay);
2848         if (adapter->itr_setting != 0)
2849                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2850
2851         ctrl_ext = er32(CTRL_EXT);
2852         /* Auto-Mask interrupts upon ICR access */
2853         ctrl_ext |= E1000_CTRL_EXT_IAME;
2854         ew32(IAM, 0xffffffff);
2855         ew32(CTRL_EXT, ctrl_ext);
2856         e1e_flush();
2857
2858         /*
2859          * Setup the HW Rx Head and Tail Descriptor Pointers and
2860          * the Base and Length of the Rx Descriptor Ring
2861          */
2862         rdba = rx_ring->dma;
2863         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2864         ew32(RDBAH, (rdba >> 32));
2865         ew32(RDLEN, rdlen);
2866         ew32(RDH, 0);
2867         ew32(RDT, 0);
2868         rx_ring->head = E1000_RDH;
2869         rx_ring->tail = E1000_RDT;
2870
2871         /* Enable Receive Checksum Offload for TCP and UDP */
2872         rxcsum = er32(RXCSUM);
2873         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2874                 rxcsum |= E1000_RXCSUM_TUOFL;
2875
2876                 /*
2877                  * IPv4 payload checksum for UDP fragments must be
2878                  * used in conjunction with packet-split.
2879                  */
2880                 if (adapter->rx_ps_pages)
2881                         rxcsum |= E1000_RXCSUM_IPPCSE;
2882         } else {
2883                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2884                 /* no need to clear IPPCSE as it defaults to 0 */
2885         }
2886         ew32(RXCSUM, rxcsum);
2887
2888         /*
2889          * Enable early receives on supported devices, only takes effect when
2890          * packet size is equal or larger than the specified value (in 8 byte
2891          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2892          */
2893         if (adapter->flags & FLAG_HAS_ERT) {
2894                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2895                         u32 rxdctl = er32(RXDCTL(0));
2896                         ew32(RXDCTL(0), rxdctl | 0x3);
2897                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2898                         /*
2899                          * With jumbo frames and early-receive enabled,
2900                          * excessive C-state transition latencies result in
2901                          * dropped transactions.
2902                          */
2903                         pm_qos_update_request(
2904                                 adapter->netdev->pm_qos_req, 55);
2905                 } else {
2906                         pm_qos_update_request(
2907                                 adapter->netdev->pm_qos_req,
2908                                 PM_QOS_DEFAULT_VALUE);
2909                 }
2910         }
2911
2912         /* Enable Receives */
2913         ew32(RCTL, rctl);
2914 }
2915
2916 /**
2917  *  e1000_update_mc_addr_list - Update Multicast addresses
2918  *  @hw: pointer to the HW structure
2919  *  @mc_addr_list: array of multicast addresses to program
2920  *  @mc_addr_count: number of multicast addresses to program
2921  *
2922  *  Updates the Multicast Table Array.
2923  *  The caller must have a packed mc_addr_list of multicast addresses.
2924  **/
2925 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2926                                       u32 mc_addr_count)
2927 {
2928         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2929 }
2930
2931 /**
2932  * e1000_set_multi - Multicast and Promiscuous mode set
2933  * @netdev: network interface device structure
2934  *
2935  * The set_multi entry point is called whenever the multicast address
2936  * list or the network interface flags are updated.  This routine is
2937  * responsible for configuring the hardware for proper multicast,
2938  * promiscuous mode, and all-multi behavior.
2939  **/
2940 static void e1000_set_multi(struct net_device *netdev)
2941 {
2942         struct e1000_adapter *adapter = netdev_priv(netdev);
2943         struct e1000_hw *hw = &adapter->hw;
2944         struct netdev_hw_addr *ha;
2945         u8  *mta_list;
2946         u32 rctl;
2947         int i;
2948
2949         /* Check for Promiscuous and All Multicast modes */
2950
2951         rctl = er32(RCTL);
2952
2953         if (netdev->flags & IFF_PROMISC) {
2954                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2955                 rctl &= ~E1000_RCTL_VFE;
2956         } else {
2957                 if (netdev->flags & IFF_ALLMULTI) {
2958                         rctl |= E1000_RCTL_MPE;
2959                         rctl &= ~E1000_RCTL_UPE;
2960                 } else {
2961                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2962                 }
2963                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2964                         rctl |= E1000_RCTL_VFE;
2965         }
2966
2967         ew32(RCTL, rctl);
2968
2969         if (!netdev_mc_empty(netdev)) {
2970                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2971                 if (!mta_list)
2972                         return;
2973
2974                 /* prepare a packed array of only addresses. */
2975                 i = 0;
2976                 netdev_for_each_mc_addr(ha, netdev)
2977                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
2978
2979                 e1000_update_mc_addr_list(hw, mta_list, i);
2980                 kfree(mta_list);
2981         } else {
2982                 /*
2983                  * if we're called from probe, we might not have
2984                  * anything to do here, so clear out the list
2985                  */
2986                 e1000_update_mc_addr_list(hw, NULL, 0);
2987         }
2988 }
2989
2990 /**
2991  * e1000_configure - configure the hardware for Rx and Tx
2992  * @adapter: private board structure
2993  **/
2994 static void e1000_configure(struct e1000_adapter *adapter)
2995 {
2996         e1000_set_multi(adapter->netdev);
2997
2998         e1000_restore_vlan(adapter);
2999         e1000_init_manageability_pt(adapter);
3000
3001         e1000_configure_tx(adapter);
3002         e1000_setup_rctl(adapter);
3003         e1000_configure_rx(adapter);
3004         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3005 }
3006
3007 /**
3008  * e1000e_power_up_phy - restore link in case the phy was powered down
3009  * @adapter: address of board private structure
3010  *
3011  * The phy may be powered down to save power and turn off link when the
3012  * driver is unloaded and wake on lan is not enabled (among others)
3013  * *** this routine MUST be followed by a call to e1000e_reset ***
3014  **/
3015 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3016 {
3017         if (adapter->hw.phy.ops.power_up)
3018                 adapter->hw.phy.ops.power_up(&adapter->hw);
3019
3020         adapter->hw.mac.ops.setup_link(&adapter->hw);
3021 }
3022
3023 /**
3024  * e1000_power_down_phy - Power down the PHY
3025  *
3026  * Power down the PHY so no link is implied when interface is down.
3027  * The PHY cannot be powered down if management or WoL is active.
3028  */
3029 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3030 {
3031         /* WoL is enabled */
3032         if (adapter->wol)
3033                 return;
3034
3035         if (adapter->hw.phy.ops.power_down)
3036                 adapter->hw.phy.ops.power_down(&adapter->hw);
3037 }
3038
3039 /**
3040  * e1000e_reset - bring the hardware into a known good state
3041  *
3042  * This function boots the hardware and enables some settings that
3043  * require a configuration cycle of the hardware - those cannot be
3044  * set/changed during runtime. After reset the device needs to be
3045  * properly configured for Rx, Tx etc.
3046  */
3047 void e1000e_reset(struct e1000_adapter *adapter)
3048 {
3049         struct e1000_mac_info *mac = &adapter->hw.mac;
3050         struct e1000_fc_info *fc = &adapter->hw.fc;
3051         struct e1000_hw *hw = &adapter->hw;
3052         u32 tx_space, min_tx_space, min_rx_space;
3053         u32 pba = adapter->pba;
3054         u16 hwm;
3055
3056         /* reset Packet Buffer Allocation to default */
3057         ew32(PBA, pba);
3058
3059         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3060                 /*
3061                  * To maintain wire speed transmits, the Tx FIFO should be
3062                  * large enough to accommodate two full transmit packets,
3063                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3064                  * the Rx FIFO should be large enough to accommodate at least
3065                  * one full receive packet and is similarly rounded up and
3066                  * expressed in KB.
3067                  */
3068                 pba = er32(PBA);
3069                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3070                 tx_space = pba >> 16;
3071                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3072                 pba &= 0xffff;
3073                 /*
3074                  * the Tx fifo also stores 16 bytes of information about the tx
3075                  * but don't include ethernet FCS because hardware appends it
3076                  */
3077                 min_tx_space = (adapter->max_frame_size +
3078                                 sizeof(struct e1000_tx_desc) -
3079                                 ETH_FCS_LEN) * 2;
3080                 min_tx_space = ALIGN(min_tx_space, 1024);
3081                 min_tx_space >>= 10;
3082                 /* software strips receive CRC, so leave room for it */
3083                 min_rx_space = adapter->max_frame_size;
3084                 min_rx_space = ALIGN(min_rx_space, 1024);
3085                 min_rx_space >>= 10;
3086
3087                 /*
3088                  * If current Tx allocation is less than the min Tx FIFO size,
3089                  * and the min Tx FIFO size is less than the current Rx FIFO
3090                  * allocation, take space away from current Rx allocation
3091                  */
3092                 if ((tx_space < min_tx_space) &&
3093                     ((min_tx_space - tx_space) < pba)) {
3094                         pba -= min_tx_space - tx_space;
3095
3096                         /*
3097                          * if short on Rx space, Rx wins and must trump tx
3098                          * adjustment or use Early Receive if available
3099                          */
3100                         if ((pba < min_rx_space) &&
3101                             (!(adapter->flags & FLAG_HAS_ERT)))
3102                                 /* ERT enabled in e1000_configure_rx */
3103                                 pba = min_rx_space;
3104                 }
3105
3106                 ew32(PBA, pba);
3107         }
3108
3109
3110         /*
3111          * flow control settings
3112          *
3113          * The high water mark must be low enough to fit one full frame
3114          * (or the size used for early receive) above it in the Rx FIFO.
3115          * Set it to the lower of:
3116          * - 90% of the Rx FIFO size, and
3117          * - the full Rx FIFO size minus the early receive size (for parts
3118          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3119          * - the full Rx FIFO size minus one full frame
3120          */
3121         if (hw->mac.type == e1000_pchlan) {
3122                 /*
3123                  * Workaround PCH LOM adapter hangs with certain network
3124                  * loads.  If hangs persist, try disabling Tx flow control.
3125                  */
3126                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3127                         fc->high_water = 0x3500;
3128                         fc->low_water  = 0x1500;
3129                 } else {
3130                         fc->high_water = 0x5000;
3131                         fc->low_water  = 0x3000;
3132                 }
3133                 fc->refresh_time = 0x1000;
3134         } else {
3135                 if ((adapter->flags & FLAG_HAS_ERT) &&
3136                     (adapter->netdev->mtu > ETH_DATA_LEN))
3137                         hwm = min(((pba << 10) * 9 / 10),
3138                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3139                 else
3140                         hwm = min(((pba << 10) * 9 / 10),
3141                                   ((pba << 10) - adapter->max_frame_size));
3142
3143                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3144                 fc->low_water = fc->high_water - 8;
3145         }
3146
3147         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3148                 fc->pause_time = 0xFFFF;
3149         else
3150                 fc->pause_time = E1000_FC_PAUSE_TIME;
3151         fc->send_xon = 1;
3152         fc->current_mode = fc->requested_mode;
3153
3154         /* Allow time for pending master requests to run */
3155         mac->ops.reset_hw(hw);
3156
3157         /*
3158          * For parts with AMT enabled, let the firmware know
3159          * that the network interface is in control
3160          */
3161         if (adapter->flags & FLAG_HAS_AMT)
3162                 e1000_get_hw_control(adapter);
3163
3164         ew32(WUC, 0);
3165         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP)
3166                 e1e_wphy(&adapter->hw, BM_WUC, 0);
3167
3168         if (mac->ops.init_hw(hw))
3169                 e_err("Hardware Error\n");
3170
3171         e1000_update_mng_vlan(adapter);
3172
3173         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3174         ew32(VET, ETH_P_8021Q);
3175
3176         e1000e_reset_adaptive(hw);
3177         e1000_get_phy_info(hw);
3178
3179         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3180             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3181                 u16 phy_data = 0;
3182                 /*
3183                  * speed up time to link by disabling smart power down, ignore
3184                  * the return value of this function because there is nothing
3185                  * different we would do if it failed
3186                  */
3187                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3188                 phy_data &= ~IGP02E1000_PM_SPD;
3189                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3190         }
3191 }
3192
3193 int e1000e_up(struct e1000_adapter *adapter)
3194 {
3195         struct e1000_hw *hw = &adapter->hw;
3196
3197         /* DMA latency requirement to workaround early-receive/jumbo issue */
3198         if (adapter->flags & FLAG_HAS_ERT)
3199                 adapter->netdev->pm_qos_req =
3200                         pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
3201                                        PM_QOS_DEFAULT_VALUE);
3202
3203         /* hardware has been reset, we need to reload some things */
3204         e1000_configure(adapter);
3205
3206         clear_bit(__E1000_DOWN, &adapter->state);
3207
3208         napi_enable(&adapter->napi);
3209         if (adapter->msix_entries)
3210                 e1000_configure_msix(adapter);
3211         e1000_irq_enable(adapter);
3212
3213         netif_wake_queue(adapter->netdev);
3214
3215         /* fire a link change interrupt to start the watchdog */
3216         if (adapter->msix_entries)
3217                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3218         else
3219                 ew32(ICS, E1000_ICS_LSC);
3220
3221         return 0;
3222 }
3223
3224 void e1000e_down(struct e1000_adapter *adapter)
3225 {
3226         struct net_device *netdev = adapter->netdev;
3227         struct e1000_hw *hw = &adapter->hw;
3228         u32 tctl, rctl;
3229
3230         /*
3231          * signal that we're down so the interrupt handler does not
3232          * reschedule our watchdog timer
3233          */
3234         set_bit(__E1000_DOWN, &adapter->state);
3235
3236         /* disable receives in the hardware */
3237         rctl = er32(RCTL);
3238         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3239         /* flush and sleep below */
3240
3241         netif_stop_queue(netdev);
3242
3243         /* disable transmits in the hardware */
3244         tctl = er32(TCTL);
3245         tctl &= ~E1000_TCTL_EN;
3246         ew32(TCTL, tctl);
3247         /* flush both disables and wait for them to finish */
3248         e1e_flush();
3249         msleep(10);
3250
3251         napi_disable(&adapter->napi);
3252         e1000_irq_disable(adapter);
3253
3254         del_timer_sync(&adapter->watchdog_timer);
3255         del_timer_sync(&adapter->phy_info_timer);
3256
3257         netif_carrier_off(netdev);
3258         adapter->link_speed = 0;
3259         adapter->link_duplex = 0;
3260
3261         if (!pci_channel_offline(adapter->pdev))
3262                 e1000e_reset(adapter);
3263         e1000_clean_tx_ring(adapter);
3264         e1000_clean_rx_ring(adapter);
3265
3266         if (adapter->flags & FLAG_HAS_ERT) {
3267                 pm_qos_remove_request(
3268                               adapter->netdev->pm_qos_req);
3269                 adapter->netdev->pm_qos_req = NULL;
3270         }
3271
3272         /*
3273          * TODO: for power management, we could drop the link and
3274          * pci_disable_device here.
3275          */
3276 }
3277
3278 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3279 {
3280         might_sleep();
3281         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3282                 msleep(1);
3283         e1000e_down(adapter);
3284         e1000e_up(adapter);
3285         clear_bit(__E1000_RESETTING, &adapter->state);
3286 }
3287
3288 /**
3289  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3290  * @adapter: board private structure to initialize
3291  *
3292  * e1000_sw_init initializes the Adapter private data structure.
3293  * Fields are initialized based on PCI device information and
3294  * OS network device settings (MTU size).
3295  **/
3296 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3297 {
3298         struct net_device *netdev = adapter->netdev;
3299
3300         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3301         adapter->rx_ps_bsize0 = 128;
3302         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3303         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3304
3305         e1000e_set_interrupt_capability(adapter);
3306
3307         if (e1000_alloc_queues(adapter))
3308                 return -ENOMEM;
3309
3310         /* Explicitly disable IRQ since the NIC can be in any state. */
3311         e1000_irq_disable(adapter);
3312
3313         set_bit(__E1000_DOWN, &adapter->state);
3314         return 0;
3315 }
3316
3317 /**
3318  * e1000_intr_msi_test - Interrupt Handler
3319  * @irq: interrupt number
3320  * @data: pointer to a network interface device structure
3321  **/
3322 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3323 {
3324         struct net_device *netdev = data;
3325         struct e1000_adapter *adapter = netdev_priv(netdev);
3326         struct e1000_hw *hw = &adapter->hw;
3327         u32 icr = er32(ICR);
3328
3329         e_dbg("icr is %08X\n", icr);
3330         if (icr & E1000_ICR_RXSEQ) {
3331                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3332                 wmb();
3333         }
3334
3335         return IRQ_HANDLED;
3336 }
3337
3338 /**
3339  * e1000_test_msi_interrupt - Returns 0 for successful test
3340  * @adapter: board private struct
3341  *
3342  * code flow taken from tg3.c
3343  **/
3344 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3345 {
3346         struct net_device *netdev = adapter->netdev;
3347         struct e1000_hw *hw = &adapter->hw;
3348         int err;
3349
3350         /* poll_enable hasn't been called yet, so don't need disable */
3351         /* clear any pending events */
3352         er32(ICR);
3353
3354         /* free the real vector and request a test handler */
3355         e1000_free_irq(adapter);
3356         e1000e_reset_interrupt_capability(adapter);
3357
3358         /* Assume that the test fails, if it succeeds then the test
3359          * MSI irq handler will unset this flag */
3360         adapter->flags |= FLAG_MSI_TEST_FAILED;
3361
3362         err = pci_enable_msi(adapter->pdev);
3363         if (err)
3364                 goto msi_test_failed;
3365
3366         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3367                           netdev->name, netdev);
3368         if (err) {
3369                 pci_disable_msi(adapter->pdev);
3370                 goto msi_test_failed;
3371         }
3372
3373         wmb();
3374
3375         e1000_irq_enable(adapter);
3376
3377         /* fire an unusual interrupt on the test handler */
3378         ew32(ICS, E1000_ICS_RXSEQ);
3379         e1e_flush();
3380         msleep(50);
3381
3382         e1000_irq_disable(adapter);
3383
3384         rmb();
3385
3386         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3387                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3388                 err = -EIO;
3389                 e_info("MSI interrupt test failed!\n");
3390         }
3391
3392         free_irq(adapter->pdev->irq, netdev);
3393         pci_disable_msi(adapter->pdev);
3394
3395         if (err == -EIO)
3396                 goto msi_test_failed;
3397
3398         /* okay so the test worked, restore settings */
3399         e_dbg("MSI interrupt test succeeded!\n");
3400 msi_test_failed:
3401         e1000e_set_interrupt_capability(adapter);
3402         e1000_request_irq(adapter);
3403         return err;
3404 }
3405
3406 /**
3407  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3408  * @adapter: board private struct
3409  *
3410  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3411  **/
3412 static int e1000_test_msi(struct e1000_adapter *adapter)
3413 {
3414         int err;
3415         u16 pci_cmd;
3416
3417         if (!(adapter->flags & FLAG_MSI_ENABLED))
3418                 return 0;
3419
3420         /* disable SERR in case the MSI write causes a master abort */
3421         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3422         pci_write_config_word(adapter->pdev, PCI_COMMAND,
3423                               pci_cmd & ~PCI_COMMAND_SERR);
3424
3425         err = e1000_test_msi_interrupt(adapter);
3426
3427         /* restore previous setting of command word */
3428         pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3429
3430         /* success ! */
3431         if (!err)
3432                 return 0;
3433
3434         /* EIO means MSI test failed */
3435         if (err != -EIO)
3436                 return err;
3437
3438         /* back to INTx mode */
3439         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3440
3441         e1000_free_irq(adapter);
3442
3443         err = e1000_request_irq(adapter);
3444
3445         return err;
3446 }
3447
3448 /**
3449  * e1000_open - Called when a network interface is made active
3450  * @netdev: network interface device structure
3451  *
3452  * Returns 0 on success, negative value on failure
3453  *
3454  * The open entry point is called when a network interface is made
3455  * active by the system (IFF_UP).  At this point all resources needed
3456  * for transmit and receive operations are allocated, the interrupt
3457  * handler is registered with the OS, the watchdog timer is started,
3458  * and the stack is notified that the interface is ready.
3459  **/
3460 static int e1000_open(struct net_device *netdev)
3461 {
3462         struct e1000_adapter *adapter = netdev_priv(netdev);
3463         struct e1000_hw *hw = &adapter->hw;
3464         struct pci_dev *pdev = adapter->pdev;
3465         int err;
3466
3467         /* disallow open during test */
3468         if (test_bit(__E1000_TESTING, &adapter->state))
3469                 return -EBUSY;
3470
3471         pm_runtime_get_sync(&pdev->dev);
3472
3473         netif_carrier_off(netdev);
3474
3475         /* allocate transmit descriptors */
3476         err = e1000e_setup_tx_resources(adapter);
3477         if (err)
3478                 goto err_setup_tx;
3479
3480         /* allocate receive descriptors */
3481         err = e1000e_setup_rx_resources(adapter);
3482         if (err)
3483                 goto err_setup_rx;
3484
3485         /*
3486          * If AMT is enabled, let the firmware know that the network
3487          * interface is now open and reset the part to a known state.
3488          */
3489         if (adapter->flags & FLAG_HAS_AMT) {
3490                 e1000_get_hw_control(adapter);
3491                 e1000e_reset(adapter);
3492         }
3493
3494         e1000e_power_up_phy(adapter);
3495
3496         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3497         if ((adapter->hw.mng_cookie.status &
3498              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3499                 e1000_update_mng_vlan(adapter);
3500
3501         /*
3502          * before we allocate an interrupt, we must be ready to handle it.
3503          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3504          * as soon as we call pci_request_irq, so we have to setup our
3505          * clean_rx handler before we do so.
3506          */
3507         e1000_configure(adapter);
3508
3509         err = e1000_request_irq(adapter);
3510         if (err)
3511                 goto err_req_irq;
3512
3513         /*
3514          * Work around PCIe errata with MSI interrupts causing some chipsets to
3515          * ignore e1000e MSI messages, which means we need to test our MSI
3516          * interrupt now
3517          */
3518         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3519                 err = e1000_test_msi(adapter);
3520                 if (err) {
3521                         e_err("Interrupt allocation failed\n");
3522                         goto err_req_irq;
3523                 }
3524         }
3525
3526         /* From here on the code is the same as e1000e_up() */
3527         clear_bit(__E1000_DOWN, &adapter->state);
3528
3529         napi_enable(&adapter->napi);
3530
3531         e1000_irq_enable(adapter);
3532
3533         netif_start_queue(netdev);
3534
3535         adapter->idle_check = true;
3536         pm_runtime_put(&pdev->dev);
3537
3538         /* fire a link status change interrupt to start the watchdog */
3539         if (adapter->msix_entries)
3540                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3541         else
3542                 ew32(ICS, E1000_ICS_LSC);
3543
3544         return 0;
3545
3546 err_req_irq:
3547         e1000_release_hw_control(adapter);
3548         e1000_power_down_phy(adapter);
3549         e1000e_free_rx_resources(adapter);
3550 err_setup_rx:
3551         e1000e_free_tx_resources(adapter);
3552 err_setup_tx:
3553         e1000e_reset(adapter);
3554         pm_runtime_put_sync(&pdev->dev);
3555
3556         return err;
3557 }
3558
3559 /**
3560  * e1000_close - Disables a network interface
3561  * @netdev: network interface device structure
3562  *
3563  * Returns 0, this is not allowed to fail
3564  *
3565  * The close entry point is called when an interface is de-activated
3566  * by the OS.  The hardware is still under the drivers control, but
3567  * needs to be disabled.  A global MAC reset is issued to stop the
3568  * hardware, and all transmit and receive resources are freed.
3569  **/
3570 static int e1000_close(struct net_device *netdev)
3571 {
3572         struct e1000_adapter *adapter = netdev_priv(netdev);
3573         struct pci_dev *pdev = adapter->pdev;
3574
3575         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3576
3577         pm_runtime_get_sync(&pdev->dev);
3578
3579         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3580                 e1000e_down(adapter);
3581                 e1000_free_irq(adapter);
3582         }
3583         e1000_power_down_phy(adapter);
3584
3585         e1000e_free_tx_resources(adapter);
3586         e1000e_free_rx_resources(adapter);
3587
3588         /*
3589          * kill manageability vlan ID if supported, but not if a vlan with
3590          * the same ID is registered on the host OS (let 8021q kill it)
3591          */
3592         if ((adapter->hw.mng_cookie.status &
3593                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3594              !(adapter->vlgrp &&
3595                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3596                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3597
3598         /*
3599          * If AMT is enabled, let the firmware know that the network
3600          * interface is now closed
3601          */
3602         if (adapter->flags & FLAG_HAS_AMT)
3603                 e1000_release_hw_control(adapter);
3604
3605         pm_runtime_put_sync(&pdev->dev);
3606
3607         return 0;
3608 }
3609 /**
3610  * e1000_set_mac - Change the Ethernet Address of the NIC
3611  * @netdev: network interface device structure
3612  * @p: pointer to an address structure
3613  *
3614  * Returns 0 on success, negative on failure
3615  **/
3616 static int e1000_set_mac(struct net_device *netdev, void *p)
3617 {
3618         struct e1000_adapter *adapter = netdev_priv(netdev);
3619         struct sockaddr *addr = p;
3620
3621         if (!is_valid_ether_addr(addr->sa_data))
3622                 return -EADDRNOTAVAIL;
3623
3624         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3625         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3626
3627         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3628
3629         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3630                 /* activate the work around */
3631                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3632
3633                 /*
3634                  * Hold a copy of the LAA in RAR[14] This is done so that
3635                  * between the time RAR[0] gets clobbered  and the time it
3636                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3637                  * of the RARs and no incoming packets directed to this port
3638                  * are dropped. Eventually the LAA will be in RAR[0] and
3639                  * RAR[14]
3640                  */
3641                 e1000e_rar_set(&adapter->hw,
3642                               adapter->hw.mac.addr,
3643                               adapter->hw.mac.rar_entry_count - 1);
3644         }
3645
3646         return 0;
3647 }
3648
3649 /**
3650  * e1000e_update_phy_task - work thread to update phy
3651  * @work: pointer to our work struct
3652  *
3653  * this worker thread exists because we must acquire a
3654  * semaphore to read the phy, which we could msleep while
3655  * waiting for it, and we can't msleep in a timer.
3656  **/
3657 static void e1000e_update_phy_task(struct work_struct *work)
3658 {
3659         struct e1000_adapter *adapter = container_of(work,
3660                                         struct e1000_adapter, update_phy_task);
3661         e1000_get_phy_info(&adapter->hw);
3662 }
3663
3664 /*
3665  * Need to wait a few seconds after link up to get diagnostic information from
3666  * the phy
3667  */
3668 static void e1000_update_phy_info(unsigned long data)
3669 {
3670         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3671         schedule_work(&adapter->update_phy_task);
3672 }
3673
3674 /**
3675  * e1000e_update_stats - Update the board statistics counters
3676  * @adapter: board private structure
3677  **/
3678 void e1000e_update_stats(struct e1000_adapter *adapter)
3679 {
3680         struct net_device *netdev = adapter->netdev;
3681         struct e1000_hw *hw = &adapter->hw;
3682         struct pci_dev *pdev = adapter->pdev;
3683         u16 phy_data;
3684
3685         /*
3686          * Prevent stats update while adapter is being reset, or if the pci
3687          * connection is down.
3688          */
3689         if (adapter->link_speed == 0)
3690                 return;
3691         if (pci_channel_offline(pdev))
3692                 return;
3693
3694         adapter->stats.crcerrs += er32(CRCERRS);
3695         adapter->stats.gprc += er32(GPRC);
3696         adapter->stats.gorc += er32(GORCL);
3697         er32(GORCH); /* Clear gorc */
3698         adapter->stats.bprc += er32(BPRC);
3699         adapter->stats.mprc += er32(MPRC);
3700         adapter->stats.roc += er32(ROC);
3701
3702         adapter->stats.mpc += er32(MPC);
3703         if ((hw->phy.type == e1000_phy_82578) ||
3704             (hw->phy.type == e1000_phy_82577)) {
3705                 e1e_rphy(hw, HV_SCC_UPPER, &phy_data);
3706                 if (!e1e_rphy(hw, HV_SCC_LOWER, &phy_data))
3707                         adapter->stats.scc += phy_data;
3708
3709                 e1e_rphy(hw, HV_ECOL_UPPER, &phy_data);
3710                 if (!e1e_rphy(hw, HV_ECOL_LOWER, &phy_data))
3711                         adapter->stats.ecol += phy_data;
3712
3713                 e1e_rphy(hw, HV_MCC_UPPER, &phy_data);
3714                 if (!e1e_rphy(hw, HV_MCC_LOWER, &phy_data))
3715                         adapter->stats.mcc += phy_data;
3716
3717                 e1e_rphy(hw, HV_LATECOL_UPPER, &phy_data);
3718                 if (!e1e_rphy(hw, HV_LATECOL_LOWER, &phy_data))
3719                         adapter->stats.latecol += phy_data;
3720
3721                 e1e_rphy(hw, HV_DC_UPPER, &phy_data);
3722                 if (!e1e_rphy(hw, HV_DC_LOWER, &phy_data))
3723                         adapter->stats.dc += phy_data;
3724         } else {
3725                 adapter->stats.scc += er32(SCC);
3726                 adapter->stats.ecol += er32(ECOL);
3727                 adapter->stats.mcc += er32(MCC);
3728                 adapter->stats.latecol += er32(LATECOL);
3729                 adapter->stats.dc += er32(DC);
3730         }
3731         adapter->stats.xonrxc += er32(XONRXC);
3732         adapter->stats.xontxc += er32(XONTXC);
3733         adapter->stats.xoffrxc += er32(XOFFRXC);
3734         adapter->stats.xofftxc += er32(XOFFTXC);
3735         adapter->stats.gptc += er32(GPTC);
3736         adapter->stats.gotc += er32(GOTCL);
3737         er32(GOTCH); /* Clear gotc */
3738         adapter->stats.rnbc += er32(RNBC);
3739         adapter->stats.ruc += er32(RUC);
3740
3741         adapter->stats.mptc += er32(MPTC);
3742         adapter->stats.bptc += er32(BPTC);
3743
3744         /* used for adaptive IFS */
3745
3746         hw->mac.tx_packet_delta = er32(TPT);
3747         adapter->stats.tpt += hw->mac.tx_packet_delta;
3748         if ((hw->phy.type == e1000_phy_82578) ||
3749             (hw->phy.type == e1000_phy_82577)) {
3750                 e1e_rphy(hw, HV_COLC_UPPER, &phy_data);
3751                 if (!e1e_rphy(hw, HV_COLC_LOWER, &phy_data))
3752                         hw->mac.collision_delta = phy_data;
3753         } else {
3754                 hw->mac.collision_delta = er32(COLC);
3755         }
3756         adapter->stats.colc += hw->mac.collision_delta;
3757
3758         adapter->stats.algnerrc += er32(ALGNERRC);
3759         adapter->stats.rxerrc += er32(RXERRC);
3760         if ((hw->phy.type == e1000_phy_82578) ||
3761             (hw->phy.type == e1000_phy_82577)) {
3762                 e1e_rphy(hw, HV_TNCRS_UPPER, &phy_data);
3763                 if (!e1e_rphy(hw, HV_TNCRS_LOWER, &phy_data))
3764                         adapter->stats.tncrs += phy_data;
3765         } else {
3766                 if ((hw->mac.type != e1000_82574) &&
3767                     (hw->mac.type != e1000_82583))
3768                         adapter->stats.tncrs += er32(TNCRS);
3769         }
3770         adapter->stats.cexterr += er32(CEXTERR);
3771         adapter->stats.tsctc += er32(TSCTC);
3772         adapter->stats.tsctfc += er32(TSCTFC);
3773
3774         /* Fill out the OS statistics structure */
3775         netdev->stats.multicast = adapter->stats.mprc;
3776         netdev->stats.collisions = adapter->stats.colc;
3777
3778         /* Rx Errors */
3779
3780         /*
3781          * RLEC on some newer hardware can be incorrect so build
3782          * our own version based on RUC and ROC
3783          */
3784         netdev->stats.rx_errors = adapter->stats.rxerrc +
3785                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3786                 adapter->stats.ruc + adapter->stats.roc +
3787                 adapter->stats.cexterr;
3788         netdev->stats.rx_length_errors = adapter->stats.ruc +
3789                                               adapter->stats.roc;
3790         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3791         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3792         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3793
3794         /* Tx Errors */
3795         netdev->stats.tx_errors = adapter->stats.ecol +
3796                                        adapter->stats.latecol;
3797         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3798         netdev->stats.tx_window_errors = adapter->stats.latecol;
3799         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3800
3801         /* Tx Dropped needs to be maintained elsewhere */
3802
3803         /* Management Stats */
3804         adapter->stats.mgptc += er32(MGTPTC);
3805         adapter->stats.mgprc += er32(MGTPRC);
3806         adapter->stats.mgpdc += er32(MGTPDC);
3807 }
3808
3809 /**
3810  * e1000_phy_read_status - Update the PHY register status snapshot
3811  * @adapter: board private structure
3812  **/
3813 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3814 {
3815         struct e1000_hw *hw = &adapter->hw;
3816         struct e1000_phy_regs *phy = &adapter->phy_regs;
3817         int ret_val;
3818
3819         if ((er32(STATUS) & E1000_STATUS_LU) &&
3820             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3821                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3822                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3823                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3824                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3825                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3826                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3827                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3828                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3829                 if (ret_val)
3830                         e_warn("Error reading PHY register\n");
3831         } else {
3832                 /*
3833                  * Do not read PHY registers if link is not up
3834                  * Set values to typical power-on defaults
3835                  */
3836                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3837                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3838                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3839                              BMSR_ERCAP);
3840                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3841                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3842                 phy->lpa = 0;
3843                 phy->expansion = EXPANSION_ENABLENPAGE;
3844                 phy->ctrl1000 = ADVERTISE_1000FULL;
3845                 phy->stat1000 = 0;
3846                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3847         }
3848 }
3849
3850 static void e1000_print_link_info(struct e1000_adapter *adapter)
3851 {
3852         struct e1000_hw *hw = &adapter->hw;
3853         u32 ctrl = er32(CTRL);
3854
3855         /* Link status message must follow this format for user tools */
3856         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3857                "Flow Control: %s\n",
3858                adapter->netdev->name,
3859                adapter->link_speed,
3860                (adapter->link_duplex == FULL_DUPLEX) ?
3861                                 "Full Duplex" : "Half Duplex",
3862                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3863                                 "RX/TX" :
3864                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3865                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3866 }
3867
3868 bool e1000e_has_link(struct e1000_adapter *adapter)
3869 {
3870         struct e1000_hw *hw = &adapter->hw;
3871         bool link_active = 0;
3872         s32 ret_val = 0;
3873
3874         /*
3875          * get_link_status is set on LSC (link status) interrupt or
3876          * Rx sequence error interrupt.  get_link_status will stay
3877          * false until the check_for_link establishes link
3878          * for copper adapters ONLY
3879          */
3880         switch (hw->phy.media_type) {
3881         case e1000_media_type_copper:
3882                 if (hw->mac.get_link_status) {
3883                         ret_val = hw->mac.ops.check_for_link(hw);
3884                         link_active = !hw->mac.get_link_status;
3885                 } else {
3886                         link_active = 1;
3887                 }
3888                 break;
3889         case e1000_media_type_fiber:
3890                 ret_val = hw->mac.ops.check_for_link(hw);
3891                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3892                 break;
3893         case e1000_media_type_internal_serdes:
3894                 ret_val = hw->mac.ops.check_for_link(hw);
3895                 link_active = adapter->hw.mac.serdes_has_link;
3896                 break;
3897         default:
3898         case e1000_media_type_unknown:
3899                 break;
3900         }
3901
3902         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
3903             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
3904                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
3905                 e_info("Gigabit has been disabled, downgrading speed\n");
3906         }
3907
3908         return link_active;
3909 }
3910
3911 static void e1000e_enable_receives(struct e1000_adapter *adapter)
3912 {
3913         /* make sure the receive unit is started */
3914         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
3915             (adapter->flags & FLAG_RX_RESTART_NOW)) {
3916                 struct e1000_hw *hw = &adapter->hw;
3917                 u32 rctl = er32(RCTL);
3918                 ew32(RCTL, rctl | E1000_RCTL_EN);
3919                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
3920         }
3921 }
3922
3923 /**
3924  * e1000_watchdog - Timer Call-back
3925  * @data: pointer to adapter cast into an unsigned long
3926  **/
3927 static void e1000_watchdog(unsigned long data)
3928 {
3929         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3930
3931         /* Do the rest outside of interrupt context */
3932         schedule_work(&adapter->watchdog_task);
3933
3934         /* TODO: make this use queue_delayed_work() */
3935 }
3936
3937 static void e1000_watchdog_task(struct work_struct *work)
3938 {
3939         struct e1000_adapter *adapter = container_of(work,
3940                                         struct e1000_adapter, watchdog_task);
3941         struct net_device *netdev = adapter->netdev;
3942         struct e1000_mac_info *mac = &adapter->hw.mac;
3943         struct e1000_phy_info *phy = &adapter->hw.phy;
3944         struct e1000_ring *tx_ring = adapter->tx_ring;
3945         struct e1000_hw *hw = &adapter->hw;
3946         u32 link, tctl;
3947         int tx_pending = 0;
3948
3949         link = e1000e_has_link(adapter);
3950         if ((netif_carrier_ok(netdev)) && link) {
3951                 /* Cancel scheduled suspend requests. */
3952                 pm_runtime_resume(netdev->dev.parent);
3953
3954                 e1000e_enable_receives(adapter);
3955                 goto link_up;
3956         }
3957
3958         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
3959             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
3960                 e1000_update_mng_vlan(adapter);
3961
3962         if (link) {
3963                 if (!netif_carrier_ok(netdev)) {
3964                         bool txb2b = 1;
3965
3966                         /* Cancel scheduled suspend requests. */
3967                         pm_runtime_resume(netdev->dev.parent);
3968
3969                         /* update snapshot of PHY registers on LSC */
3970                         e1000_phy_read_status(adapter);
3971                         mac->ops.get_link_up_info(&adapter->hw,
3972                                                    &adapter->link_speed,
3973                                                    &adapter->link_duplex);
3974                         e1000_print_link_info(adapter);
3975                         /*
3976                          * On supported PHYs, check for duplex mismatch only
3977                          * if link has autonegotiated at 10/100 half
3978                          */
3979                         if ((hw->phy.type == e1000_phy_igp_3 ||
3980                              hw->phy.type == e1000_phy_bm) &&
3981                             (hw->mac.autoneg == true) &&
3982                             (adapter->link_speed == SPEED_10 ||
3983                              adapter->link_speed == SPEED_100) &&
3984                             (adapter->link_duplex == HALF_DUPLEX)) {
3985                                 u16 autoneg_exp;
3986
3987                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
3988
3989                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
3990                                         e_info("Autonegotiated half duplex but"
3991                                                " link partner cannot autoneg. "
3992                                                " Try forcing full duplex if "
3993                                                "link gets many collisions.\n");
3994                         }
3995
3996                         /* adjust timeout factor according to speed/duplex */
3997                         adapter->tx_timeout_factor = 1;
3998                         switch (adapter->link_speed) {
3999                         case SPEED_10:
4000                                 txb2b = 0;
4001                                 adapter->tx_timeout_factor = 16;
4002                                 break;
4003                         case SPEED_100:
4004                                 txb2b = 0;
4005                                 adapter->tx_timeout_factor = 10;
4006                                 break;
4007                         }
4008
4009                         /*
4010                          * workaround: re-program speed mode bit after
4011                          * link-up event
4012                          */
4013                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4014                             !txb2b) {
4015                                 u32 tarc0;
4016                                 tarc0 = er32(TARC(0));
4017                                 tarc0 &= ~SPEED_MODE_BIT;
4018                                 ew32(TARC(0), tarc0);
4019                         }
4020
4021                         /*
4022                          * disable TSO for pcie and 10/100 speeds, to avoid
4023                          * some hardware issues
4024                          */
4025                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4026                                 switch (adapter->link_speed) {
4027                                 case SPEED_10:
4028                                 case SPEED_100:
4029                                         e_info("10/100 speed: disabling TSO\n");
4030                                         netdev->features &= ~NETIF_F_TSO;
4031                                         netdev->features &= ~NETIF_F_TSO6;
4032                                         break;
4033                                 case SPEED_1000:
4034                                         netdev->features |= NETIF_F_TSO;
4035                                         netdev->features |= NETIF_F_TSO6;
4036                                         break;
4037                                 default:
4038                                         /* oops */
4039                                         break;
4040                                 }
4041                         }
4042
4043                         /*
4044                          * enable transmits in the hardware, need to do this
4045                          * after setting TARC(0)
4046                          */
4047                         tctl = er32(TCTL);
4048                         tctl |= E1000_TCTL_EN;
4049                         ew32(TCTL, tctl);
4050
4051                         /*
4052                          * Perform any post-link-up configuration before
4053                          * reporting link up.
4054                          */
4055                         if (phy->ops.cfg_on_link_up)
4056                                 phy->ops.cfg_on_link_up(hw);
4057
4058                         netif_carrier_on(netdev);
4059
4060                         if (!test_bit(__E1000_DOWN, &adapter->state))
4061                                 mod_timer(&adapter->phy_info_timer,
4062                                           round_jiffies(jiffies + 2 * HZ));
4063                 }
4064         } else {
4065                 if (netif_carrier_ok(netdev)) {
4066                         adapter->link_speed = 0;
4067                         adapter->link_duplex = 0;
4068                         /* Link status message must follow this format */
4069                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4070                                adapter->netdev->name);
4071                         netif_carrier_off(netdev);
4072                         if (!test_bit(__E1000_DOWN, &adapter->state))
4073                                 mod_timer(&adapter->phy_info_timer,
4074                                           round_jiffies(jiffies + 2 * HZ));
4075
4076                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4077                                 schedule_work(&adapter->reset_task);
4078                         else
4079                                 pm_schedule_suspend(netdev->dev.parent,
4080                                                         LINK_TIMEOUT);
4081                 }
4082         }
4083
4084 link_up:
4085         e1000e_update_stats(adapter);
4086
4087         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4088         adapter->tpt_old = adapter->stats.tpt;
4089         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4090         adapter->colc_old = adapter->stats.colc;
4091
4092         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4093         adapter->gorc_old = adapter->stats.gorc;
4094         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4095         adapter->gotc_old = adapter->stats.gotc;
4096
4097         e1000e_update_adaptive(&adapter->hw);
4098
4099         if (!netif_carrier_ok(netdev)) {
4100                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4101                                tx_ring->count);
4102                 if (tx_pending) {
4103                         /*
4104                          * We've lost link, so the controller stops DMA,
4105                          * but we've got queued Tx work that's never going
4106                          * to get done, so reset controller to flush Tx.
4107                          * (Do the reset outside of interrupt context).
4108                          */
4109                         adapter->tx_timeout_count++;
4110                         schedule_work(&adapter->reset_task);
4111                         /* return immediately since reset is imminent */
4112                         return;
4113                 }
4114         }
4115
4116         /* Simple mode for Interrupt Throttle Rate (ITR) */
4117         if (adapter->itr_setting == 4) {
4118                 /*
4119                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4120                  * Total asymmetrical Tx or Rx gets ITR=8000;
4121                  * everyone else is between 2000-8000.
4122                  */
4123                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4124                 u32 dif = (adapter->gotc > adapter->gorc ?
4125                             adapter->gotc - adapter->gorc :
4126                             adapter->gorc - adapter->gotc) / 10000;
4127                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4128
4129                 ew32(ITR, 1000000000 / (itr * 256));
4130         }
4131
4132         /* Cause software interrupt to ensure Rx ring is cleaned */
4133         if (adapter->msix_entries)
4134                 ew32(ICS, adapter->rx_ring->ims_val);
4135         else
4136                 ew32(ICS, E1000_ICS_RXDMT0);
4137
4138         /* Force detection of hung controller every watchdog period */
4139         adapter->detect_tx_hung = 1;
4140
4141         /*
4142          * With 82571 controllers, LAA may be overwritten due to controller
4143          * reset from the other port. Set the appropriate LAA in RAR[0]
4144          */
4145         if (e1000e_get_laa_state_82571(hw))
4146                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4147
4148         /* Reset the timer */
4149         if (!test_bit(__E1000_DOWN, &adapter->state))
4150                 mod_timer(&adapter->watchdog_timer,
4151                           round_jiffies(jiffies + 2 * HZ));
4152 }
4153
4154 #define E1000_TX_FLAGS_CSUM             0x00000001
4155 #define E1000_TX_FLAGS_VLAN             0x00000002
4156 #define E1000_TX_FLAGS_TSO              0x00000004
4157 #define E1000_TX_FLAGS_IPV4             0x00000008
4158 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4159 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4160
4161 static int e1000_tso(struct e1000_adapter *adapter,
4162                      struct sk_buff *skb)
4163 {
4164         struct e1000_ring *tx_ring = adapter->tx_ring;
4165         struct e1000_context_desc *context_desc;
4166         struct e1000_buffer *buffer_info;
4167         unsigned int i;
4168         u32 cmd_length = 0;
4169         u16 ipcse = 0, tucse, mss;
4170         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4171         int err;
4172
4173         if (!skb_is_gso(skb))
4174                 return 0;
4175
4176         if (skb_header_cloned(skb)) {
4177                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4178                 if (err)
4179                         return err;
4180         }
4181
4182         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4183         mss = skb_shinfo(skb)->gso_size;
4184         if (skb->protocol == htons(ETH_P_IP)) {
4185                 struct iphdr *iph = ip_hdr(skb);
4186                 iph->tot_len = 0;
4187                 iph->check = 0;
4188                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4189                                                          0, IPPROTO_TCP, 0);
4190                 cmd_length = E1000_TXD_CMD_IP;
4191                 ipcse = skb_transport_offset(skb) - 1;
4192         } else if (skb_is_gso_v6(skb)) {
4193                 ipv6_hdr(skb)->payload_len = 0;
4194                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4195                                                        &ipv6_hdr(skb)->daddr,
4196                                                        0, IPPROTO_TCP, 0);
4197                 ipcse = 0;
4198         }
4199         ipcss = skb_network_offset(skb);
4200         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4201         tucss = skb_transport_offset(skb);
4202         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4203         tucse = 0;
4204
4205         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4206                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4207
4208         i = tx_ring->next_to_use;
4209         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4210         buffer_info = &tx_ring->buffer_info[i];
4211
4212         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4213         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4214         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4215         context_desc->upper_setup.tcp_fields.tucss = tucss;
4216         context_desc->upper_setup.tcp_fields.tucso = tucso;
4217         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4218         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4219         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4220         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4221
4222         buffer_info->time_stamp = jiffies;
4223         buffer_info->next_to_watch = i;
4224
4225         i++;
4226         if (i == tx_ring->count)
4227                 i = 0;
4228         tx_ring->next_to_use = i;
4229
4230         return 1;
4231 }
4232
4233 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4234 {
4235         struct e1000_ring *tx_ring = adapter->tx_ring;
4236         struct e1000_context_desc *context_desc;
4237         struct e1000_buffer *buffer_info;
4238         unsigned int i;
4239         u8 css;
4240         u32 cmd_len = E1000_TXD_CMD_DEXT;
4241         __be16 protocol;
4242
4243         if (skb->ip_summed != CHECKSUM_PARTIAL)
4244                 return 0;
4245
4246         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4247                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4248         else
4249                 protocol = skb->protocol;
4250
4251         switch (protocol) {
4252         case cpu_to_be16(ETH_P_IP):
4253                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4254                         cmd_len |= E1000_TXD_CMD_TCP;
4255                 break;
4256         case cpu_to_be16(ETH_P_IPV6):
4257                 /* XXX not handling all IPV6 headers */
4258                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4259                         cmd_len |= E1000_TXD_CMD_TCP;
4260                 break;
4261         default:
4262                 if (unlikely(net_ratelimit()))
4263                         e_warn("checksum_partial proto=%x!\n",
4264                                be16_to_cpu(protocol));
4265                 break;
4266         }
4267
4268         css = skb_transport_offset(skb);
4269
4270         i = tx_ring->next_to_use;
4271         buffer_info = &tx_ring->buffer_info[i];
4272         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4273
4274         context_desc->lower_setup.ip_config = 0;
4275         context_desc->upper_setup.tcp_fields.tucss = css;
4276         context_desc->upper_setup.tcp_fields.tucso =
4277                                 css + skb->csum_offset;
4278         context_desc->upper_setup.tcp_fields.tucse = 0;
4279         context_desc->tcp_seg_setup.data = 0;
4280         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4281
4282         buffer_info->time_stamp = jiffies;
4283         buffer_info->next_to_watch = i;
4284
4285         i++;
4286         if (i == tx_ring->count)
4287                 i = 0;
4288         tx_ring->next_to_use = i;
4289
4290         return 1;
4291 }
4292
4293 #define E1000_MAX_PER_TXD       8192
4294 #define E1000_MAX_TXD_PWR       12
4295
4296 static int e1000_tx_map(struct e1000_adapter *adapter,
4297                         struct sk_buff *skb, unsigned int first,
4298                         unsigned int max_per_txd, unsigned int nr_frags,
4299                         unsigned int mss)
4300 {
4301         struct e1000_ring *tx_ring = adapter->tx_ring;
4302         struct pci_dev *pdev = adapter->pdev;
4303         struct e1000_buffer *buffer_info;
4304         unsigned int len = skb_headlen(skb);
4305         unsigned int offset = 0, size, count = 0, i;
4306         unsigned int f, bytecount, segs;
4307
4308         i = tx_ring->next_to_use;
4309
4310         while (len) {
4311                 buffer_info = &tx_ring->buffer_info[i];
4312                 size = min(len, max_per_txd);
4313
4314                 buffer_info->length = size;
4315                 buffer_info->time_stamp = jiffies;
4316                 buffer_info->next_to_watch = i;
4317                 buffer_info->dma = dma_map_single(&pdev->dev,
4318                                                   skb->data + offset,
4319                                                   size, DMA_TO_DEVICE);
4320                 buffer_info->mapped_as_page = false;
4321                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4322                         goto dma_error;
4323
4324                 len -= size;
4325                 offset += size;
4326                 count++;
4327
4328                 if (len) {
4329                         i++;
4330                         if (i == tx_ring->count)
4331                                 i = 0;
4332                 }
4333         }
4334
4335         for (f = 0; f < nr_frags; f++) {
4336                 struct skb_frag_struct *frag;
4337
4338                 frag = &skb_shinfo(skb)->frags[f];
4339                 len = frag->size;
4340                 offset = frag->page_offset;
4341
4342                 while (len) {
4343                         i++;
4344                         if (i == tx_ring->count)
4345                                 i = 0;
4346
4347                         buffer_info = &tx_ring->buffer_info[i];
4348                         size = min(len, max_per_txd);
4349
4350                         buffer_info->length = size;
4351                         buffer_info->time_stamp = jiffies;
4352                         buffer_info->next_to_watch = i;
4353                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4354                                                         offset, size,
4355                                                         DMA_TO_DEVICE);
4356                         buffer_info->mapped_as_page = true;
4357                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4358                                 goto dma_error;
4359
4360                         len -= size;
4361                         offset += size;
4362                         count++;
4363                 }
4364         }
4365
4366         segs = skb_shinfo(skb)->gso_segs ?: 1;
4367         /* multiply data chunks by size of headers */
4368         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4369
4370         tx_ring->buffer_info[i].skb = skb;
4371         tx_ring->buffer_info[i].segs = segs;
4372         tx_ring->buffer_info[i].bytecount = bytecount;
4373         tx_ring->buffer_info[first].next_to_watch = i;
4374
4375         return count;
4376
4377 dma_error:
4378         dev_err(&pdev->dev, "TX DMA map failed\n");
4379         buffer_info->dma = 0;
4380         if (count)
4381                 count--;
4382
4383         while (count--) {
4384                 if (i==0)
4385                         i += tx_ring->count;
4386                 i--;
4387                 buffer_info = &tx_ring->buffer_info[i];
4388                 e1000_put_txbuf(adapter, buffer_info);;
4389         }
4390
4391         return 0;
4392 }
4393
4394 static void e1000_tx_queue(struct e1000_adapter *adapter,
4395                            int tx_flags, int count)
4396 {
4397         struct e1000_ring *tx_ring = adapter->tx_ring;
4398         struct e1000_tx_desc *tx_desc = NULL;
4399         struct e1000_buffer *buffer_info;
4400         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4401         unsigned int i;
4402
4403         if (tx_flags & E1000_TX_FLAGS_TSO) {
4404                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4405                              E1000_TXD_CMD_TSE;
4406                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4407
4408                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4409                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4410         }
4411
4412         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4413                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4414                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4415         }
4416
4417         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4418                 txd_lower |= E1000_TXD_CMD_VLE;
4419                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4420         }
4421
4422         i = tx_ring->next_to_use;
4423
4424         while (count--) {
4425                 buffer_info = &tx_ring->buffer_info[i];
4426                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4427                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4428                 tx_desc->lower.data =
4429                         cpu_to_le32(txd_lower | buffer_info->length);
4430                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4431
4432                 i++;
4433                 if (i == tx_ring->count)
4434                         i = 0;
4435         }
4436
4437         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4438
4439         /*
4440          * Force memory writes to complete before letting h/w
4441          * know there are new descriptors to fetch.  (Only
4442          * applicable for weak-ordered memory model archs,
4443          * such as IA-64).
4444          */
4445         wmb();
4446
4447         tx_ring->next_to_use = i;
4448         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4449         /*
4450          * we need this if more than one processor can write to our tail
4451          * at a time, it synchronizes IO on IA64/Altix systems
4452          */
4453         mmiowb();
4454 }
4455
4456 #define MINIMUM_DHCP_PACKET_SIZE 282
4457 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4458                                     struct sk_buff *skb)
4459 {
4460         struct e1000_hw *hw =  &adapter->hw;
4461         u16 length, offset;
4462
4463         if (vlan_tx_tag_present(skb)) {
4464                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4465                     (adapter->hw.mng_cookie.status &
4466                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4467                         return 0;
4468         }
4469
4470         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4471                 return 0;
4472
4473         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4474                 return 0;
4475
4476         {
4477                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4478                 struct udphdr *udp;
4479
4480                 if (ip->protocol != IPPROTO_UDP)
4481                         return 0;
4482
4483                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4484                 if (ntohs(udp->dest) != 67)
4485                         return 0;
4486
4487                 offset = (u8 *)udp + 8 - skb->data;
4488                 length = skb->len - offset;
4489                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4490         }
4491
4492         return 0;
4493 }
4494
4495 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4496 {
4497         struct e1000_adapter *adapter = netdev_priv(netdev);
4498
4499         netif_stop_queue(netdev);
4500         /*
4501          * Herbert's original patch had:
4502          *  smp_mb__after_netif_stop_queue();
4503          * but since that doesn't exist yet, just open code it.
4504          */
4505         smp_mb();
4506
4507         /*
4508          * We need to check again in a case another CPU has just
4509          * made room available.
4510          */
4511         if (e1000_desc_unused(adapter->tx_ring) < size)
4512                 return -EBUSY;
4513
4514         /* A reprieve! */
4515         netif_start_queue(netdev);
4516         ++adapter->restart_queue;
4517         return 0;
4518 }
4519
4520 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4521 {
4522         struct e1000_adapter *adapter = netdev_priv(netdev);
4523
4524         if (e1000_desc_unused(adapter->tx_ring) >= size)
4525                 return 0;
4526         return __e1000_maybe_stop_tx(netdev, size);
4527 }
4528
4529 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4530 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4531                                     struct net_device *netdev)
4532 {
4533         struct e1000_adapter *adapter = netdev_priv(netdev);
4534         struct e1000_ring *tx_ring = adapter->tx_ring;
4535         unsigned int first;
4536         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4537         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4538         unsigned int tx_flags = 0;
4539         unsigned int len = skb_headlen(skb);
4540         unsigned int nr_frags;
4541         unsigned int mss;
4542         int count = 0;
4543         int tso;
4544         unsigned int f;
4545
4546         if (test_bit(__E1000_DOWN, &adapter->state)) {
4547                 dev_kfree_skb_any(skb);
4548                 return NETDEV_TX_OK;
4549         }
4550
4551         if (skb->len <= 0) {
4552                 dev_kfree_skb_any(skb);
4553                 return NETDEV_TX_OK;
4554         }
4555
4556         mss = skb_shinfo(skb)->gso_size;
4557         /*
4558          * The controller does a simple calculation to
4559          * make sure there is enough room in the FIFO before
4560          * initiating the DMA for each buffer.  The calc is:
4561          * 4 = ceil(buffer len/mss).  To make sure we don't
4562          * overrun the FIFO, adjust the max buffer len if mss
4563          * drops.
4564          */
4565         if (mss) {
4566                 u8 hdr_len;
4567                 max_per_txd = min(mss << 2, max_per_txd);
4568                 max_txd_pwr = fls(max_per_txd) - 1;
4569
4570                 /*
4571                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4572                  * points to just header, pull a few bytes of payload from
4573                  * frags into skb->data
4574                  */
4575                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4576                 /*
4577                  * we do this workaround for ES2LAN, but it is un-necessary,
4578                  * avoiding it could save a lot of cycles
4579                  */
4580                 if (skb->data_len && (hdr_len == len)) {
4581                         unsigned int pull_size;
4582
4583                         pull_size = min((unsigned int)4, skb->data_len);
4584                         if (!__pskb_pull_tail(skb, pull_size)) {
4585                                 e_err("__pskb_pull_tail failed.\n");
4586                                 dev_kfree_skb_any(skb);
4587                                 return NETDEV_TX_OK;
4588                         }
4589                         len = skb_headlen(skb);
4590                 }
4591         }
4592
4593         /* reserve a descriptor for the offload context */
4594         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4595                 count++;
4596         count++;
4597
4598         count += TXD_USE_COUNT(len, max_txd_pwr);
4599
4600         nr_frags = skb_shinfo(skb)->nr_frags;
4601         for (f = 0; f < nr_frags; f++)
4602                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4603                                        max_txd_pwr);
4604
4605         if (adapter->hw.mac.tx_pkt_filtering)
4606                 e1000_transfer_dhcp_info(adapter, skb);
4607
4608         /*
4609          * need: count + 2 desc gap to keep tail from touching
4610          * head, otherwise try next time
4611          */
4612         if (e1000_maybe_stop_tx(netdev, count + 2))
4613                 return NETDEV_TX_BUSY;
4614
4615         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4616                 tx_flags |= E1000_TX_FLAGS_VLAN;
4617                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4618         }
4619
4620         first = tx_ring->next_to_use;
4621
4622         tso = e1000_tso(adapter, skb);
4623         if (tso < 0) {
4624                 dev_kfree_skb_any(skb);
4625                 return NETDEV_TX_OK;
4626         }
4627
4628         if (tso)
4629                 tx_flags |= E1000_TX_FLAGS_TSO;
4630         else if (e1000_tx_csum(adapter, skb))
4631                 tx_flags |= E1000_TX_FLAGS_CSUM;
4632
4633         /*
4634          * Old method was to assume IPv4 packet by default if TSO was enabled.
4635          * 82571 hardware supports TSO capabilities for IPv6 as well...
4636          * no longer assume, we must.
4637          */
4638         if (skb->protocol == htons(ETH_P_IP))
4639                 tx_flags |= E1000_TX_FLAGS_IPV4;
4640
4641         /* if count is 0 then mapping error has occured */
4642         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4643         if (count) {
4644                 e1000_tx_queue(adapter, tx_flags, count);
4645                 /* Make sure there is space in the ring for the next send. */
4646                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4647
4648         } else {
4649                 dev_kfree_skb_any(skb);
4650                 tx_ring->buffer_info[first].time_stamp = 0;
4651                 tx_ring->next_to_use = first;
4652         }
4653
4654         return NETDEV_TX_OK;
4655 }
4656
4657 /**
4658  * e1000_tx_timeout - Respond to a Tx Hang
4659  * @netdev: network interface device structure
4660  **/
4661 static void e1000_tx_timeout(struct net_device *netdev)
4662 {
4663         struct e1000_adapter *adapter = netdev_priv(netdev);
4664
4665         /* Do the reset outside of interrupt context */
4666         adapter->tx_timeout_count++;
4667         schedule_work(&adapter->reset_task);
4668 }
4669
4670 static void e1000_reset_task(struct work_struct *work)
4671 {
4672         struct e1000_adapter *adapter;
4673         adapter = container_of(work, struct e1000_adapter, reset_task);
4674
4675         e1000e_dump(adapter);
4676         e_err("Reset adapter\n");
4677         e1000e_reinit_locked(adapter);
4678 }
4679
4680 /**
4681  * e1000_get_stats - Get System Network Statistics
4682  * @netdev: network interface device structure
4683  *
4684  * Returns the address of the device statistics structure.
4685  * The statistics are actually updated from the timer callback.
4686  **/
4687 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4688 {
4689         /* only return the current stats */
4690         return &netdev->stats;
4691 }
4692
4693 /**
4694  * e1000_change_mtu - Change the Maximum Transfer Unit
4695  * @netdev: network interface device structure
4696  * @new_mtu: new value for maximum frame size
4697  *
4698  * Returns 0 on success, negative on failure
4699  **/
4700 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4701 {
4702         struct e1000_adapter *adapter = netdev_priv(netdev);
4703         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4704
4705         /* Jumbo frame support */
4706         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4707             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4708                 e_err("Jumbo Frames not supported.\n");
4709                 return -EINVAL;
4710         }
4711
4712         /* Supported frame sizes */
4713         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4714             (max_frame > adapter->max_hw_frame_size)) {
4715                 e_err("Unsupported MTU setting\n");
4716                 return -EINVAL;
4717         }
4718
4719         /* 82573 Errata 17 */
4720         if (((adapter->hw.mac.type == e1000_82573) ||
4721              (adapter->hw.mac.type == e1000_82574)) &&
4722             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4723                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4724                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4725         }
4726
4727         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4728                 msleep(1);
4729         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4730         adapter->max_frame_size = max_frame;
4731         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4732         netdev->mtu = new_mtu;
4733         if (netif_running(netdev))
4734                 e1000e_down(adapter);
4735
4736         /*
4737          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4738          * means we reserve 2 more, this pushes us to allocate from the next
4739          * larger slab size.
4740          * i.e. RXBUFFER_2048 --> size-4096 slab
4741          * However with the new *_jumbo_rx* routines, jumbo receives will use
4742          * fragmented skbs
4743          */
4744
4745         if (max_frame <= 2048)
4746                 adapter->rx_buffer_len = 2048;
4747         else
4748                 adapter->rx_buffer_len = 4096;
4749
4750         /* adjust allocation if LPE protects us, and we aren't using SBP */
4751         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4752              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4753                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4754                                          + ETH_FCS_LEN;
4755
4756         if (netif_running(netdev))
4757                 e1000e_up(adapter);
4758         else
4759                 e1000e_reset(adapter);
4760
4761         clear_bit(__E1000_RESETTING, &adapter->state);
4762
4763         return 0;
4764 }
4765
4766 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4767                            int cmd)
4768 {
4769         struct e1000_adapter *adapter = netdev_priv(netdev);
4770         struct mii_ioctl_data *data = if_mii(ifr);
4771
4772         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4773                 return -EOPNOTSUPP;
4774
4775         switch (cmd) {
4776         case SIOCGMIIPHY:
4777                 data->phy_id = adapter->hw.phy.addr;
4778                 break;
4779         case SIOCGMIIREG:
4780                 e1000_phy_read_status(adapter);
4781
4782                 switch (data->reg_num & 0x1F) {
4783                 case MII_BMCR:
4784                         data->val_out = adapter->phy_regs.bmcr;
4785                         break;
4786                 case MII_BMSR:
4787                         data->val_out = adapter->phy_regs.bmsr;
4788                         break;
4789                 case MII_PHYSID1:
4790                         data->val_out = (adapter->hw.phy.id >> 16);
4791                         break;
4792                 case MII_PHYSID2:
4793                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4794                         break;
4795                 case MII_ADVERTISE:
4796                         data->val_out = adapter->phy_regs.advertise;
4797                         break;
4798                 case MII_LPA:
4799                         data->val_out = adapter->phy_regs.lpa;
4800                         break;
4801                 case MII_EXPANSION:
4802                         data->val_out = adapter->phy_regs.expansion;
4803                         break;
4804                 case MII_CTRL1000:
4805                         data->val_out = adapter->phy_regs.ctrl1000;
4806                         break;
4807                 case MII_STAT1000:
4808                         data->val_out = adapter->phy_regs.stat1000;
4809                         break;
4810                 case MII_ESTATUS:
4811                         data->val_out = adapter->phy_regs.estatus;
4812                         break;
4813                 default:
4814                         return -EIO;
4815                 }
4816                 break;
4817         case SIOCSMIIREG:
4818         default:
4819                 return -EOPNOTSUPP;
4820         }
4821         return 0;
4822 }
4823
4824 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4825 {
4826         switch (cmd) {
4827         case SIOCGMIIPHY:
4828         case SIOCGMIIREG:
4829         case SIOCSMIIREG:
4830                 return e1000_mii_ioctl(netdev, ifr, cmd);
4831         default:
4832                 return -EOPNOTSUPP;
4833         }
4834 }
4835
4836 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4837 {
4838         struct e1000_hw *hw = &adapter->hw;
4839         u32 i, mac_reg;
4840         u16 phy_reg;
4841         int retval = 0;
4842
4843         /* copy MAC RARs to PHY RARs */
4844         for (i = 0; i < adapter->hw.mac.rar_entry_count; i++) {
4845                 mac_reg = er32(RAL(i));
4846                 e1e_wphy(hw, BM_RAR_L(i), (u16)(mac_reg & 0xFFFF));
4847                 e1e_wphy(hw, BM_RAR_M(i), (u16)((mac_reg >> 16) & 0xFFFF));
4848                 mac_reg = er32(RAH(i));
4849                 e1e_wphy(hw, BM_RAR_H(i), (u16)(mac_reg & 0xFFFF));
4850                 e1e_wphy(hw, BM_RAR_CTRL(i), (u16)((mac_reg >> 16) & 0xFFFF));
4851         }
4852
4853         /* copy MAC MTA to PHY MTA */
4854         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4855                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4856                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4857                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4858         }
4859
4860         /* configure PHY Rx Control register */
4861         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4862         mac_reg = er32(RCTL);
4863         if (mac_reg & E1000_RCTL_UPE)
4864                 phy_reg |= BM_RCTL_UPE;
4865         if (mac_reg & E1000_RCTL_MPE)
4866                 phy_reg |= BM_RCTL_MPE;
4867         phy_reg &= ~(BM_RCTL_MO_MASK);
4868         if (mac_reg & E1000_RCTL_MO_3)
4869                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4870                                 << BM_RCTL_MO_SHIFT);
4871         if (mac_reg & E1000_RCTL_BAM)
4872                 phy_reg |= BM_RCTL_BAM;
4873         if (mac_reg & E1000_RCTL_PMCF)
4874                 phy_reg |= BM_RCTL_PMCF;
4875         mac_reg = er32(CTRL);
4876         if (mac_reg & E1000_CTRL_RFCE)
4877                 phy_reg |= BM_RCTL_RFCE;
4878         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4879
4880         /* enable PHY wakeup in MAC register */
4881         ew32(WUFC, wufc);
4882         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4883
4884         /* configure and enable PHY wakeup in PHY registers */
4885         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4886         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4887
4888         /* activate PHY wakeup */
4889         retval = hw->phy.ops.acquire(hw);
4890         if (retval) {
4891                 e_err("Could not acquire PHY\n");
4892                 return retval;
4893         }
4894         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4895                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4896         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4897         if (retval) {
4898                 e_err("Could not read PHY page 769\n");
4899                 goto out;
4900         }
4901         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4902         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4903         if (retval)
4904                 e_err("Could not set PHY Host Wakeup bit\n");
4905 out:
4906         hw->phy.ops.release(hw);
4907
4908         return retval;
4909 }
4910
4911 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
4912                             bool runtime)
4913 {
4914         struct net_device *netdev = pci_get_drvdata(pdev);
4915         struct e1000_adapter *adapter = netdev_priv(netdev);
4916         struct e1000_hw *hw = &adapter->hw;
4917         u32 ctrl, ctrl_ext, rctl, status;
4918         /* Runtime suspend should only enable wakeup for link changes */
4919         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
4920         int retval = 0;
4921
4922         netif_device_detach(netdev);
4923
4924         if (netif_running(netdev)) {
4925                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
4926                 e1000e_down(adapter);
4927                 e1000_free_irq(adapter);
4928         }
4929         e1000e_reset_interrupt_capability(adapter);
4930
4931         retval = pci_save_state(pdev);
4932         if (retval)
4933                 return retval;
4934
4935         status = er32(STATUS);
4936         if (status & E1000_STATUS_LU)
4937                 wufc &= ~E1000_WUFC_LNKC;
4938
4939         if (wufc) {
4940                 e1000_setup_rctl(adapter);
4941                 e1000_set_multi(netdev);
4942
4943                 /* turn on all-multi mode if wake on multicast is enabled */
4944                 if (wufc & E1000_WUFC_MC) {
4945                         rctl = er32(RCTL);
4946                         rctl |= E1000_RCTL_MPE;
4947                         ew32(RCTL, rctl);
4948                 }
4949
4950                 ctrl = er32(CTRL);
4951                 /* advertise wake from D3Cold */
4952                 #define E1000_CTRL_ADVD3WUC 0x00100000
4953                 /* phy power management enable */
4954                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4955                 ctrl |= E1000_CTRL_ADVD3WUC;
4956                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
4957                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
4958                 ew32(CTRL, ctrl);
4959
4960                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
4961                     adapter->hw.phy.media_type ==
4962                     e1000_media_type_internal_serdes) {
4963                         /* keep the laser running in D3 */
4964                         ctrl_ext = er32(CTRL_EXT);
4965                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
4966                         ew32(CTRL_EXT, ctrl_ext);
4967                 }
4968
4969                 if (adapter->flags & FLAG_IS_ICH)
4970                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
4971
4972                 /* Allow time for pending master requests to run */
4973                 e1000e_disable_pcie_master(&adapter->hw);
4974
4975                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
4976                         /* enable wakeup by the PHY */
4977                         retval = e1000_init_phy_wakeup(adapter, wufc);
4978                         if (retval)
4979                                 return retval;
4980                 } else {
4981                         /* enable wakeup by the MAC */
4982                         ew32(WUFC, wufc);
4983                         ew32(WUC, E1000_WUC_PME_EN);
4984                 }
4985         } else {
4986                 ew32(WUC, 0);
4987                 ew32(WUFC, 0);
4988         }
4989
4990         *enable_wake = !!wufc;
4991
4992         /* make sure adapter isn't asleep if manageability is enabled */
4993         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
4994             (hw->mac.ops.check_mng_mode(hw)))
4995                 *enable_wake = true;
4996
4997         if (adapter->hw.phy.type == e1000_phy_igp_3)
4998                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
4999
5000         /*
5001          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5002          * would have already happened in close and is redundant.
5003          */
5004         e1000_release_hw_control(adapter);
5005
5006         pci_disable_device(pdev);
5007
5008         return 0;
5009 }
5010
5011 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5012 {
5013         if (sleep && wake) {
5014                 pci_prepare_to_sleep(pdev);
5015                 return;
5016         }
5017
5018         pci_wake_from_d3(pdev, wake);
5019         pci_set_power_state(pdev, PCI_D3hot);
5020 }
5021
5022 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5023                                     bool wake)
5024 {
5025         struct net_device *netdev = pci_get_drvdata(pdev);
5026         struct e1000_adapter *adapter = netdev_priv(netdev);
5027
5028         /*
5029          * The pci-e switch on some quad port adapters will report a
5030          * correctable error when the MAC transitions from D0 to D3.  To
5031          * prevent this we need to mask off the correctable errors on the
5032          * downstream port of the pci-e switch.
5033          */
5034         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5035                 struct pci_dev *us_dev = pdev->bus->self;
5036                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5037                 u16 devctl;
5038
5039                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5040                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5041                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5042
5043                 e1000_power_off(pdev, sleep, wake);
5044
5045                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5046         } else {
5047                 e1000_power_off(pdev, sleep, wake);
5048         }
5049 }
5050
5051 #ifdef CONFIG_PCIEASPM
5052 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5053 {
5054         pci_disable_link_state(pdev, state);
5055 }
5056 #else
5057 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5058 {
5059         int pos;
5060         u16 reg16;
5061
5062         /*
5063          * Both device and parent should have the same ASPM setting.
5064          * Disable ASPM in downstream component first and then upstream.
5065          */
5066         pos = pci_pcie_cap(pdev);
5067         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5068         reg16 &= ~state;
5069         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5070
5071         if (!pdev->bus->self)
5072                 return;
5073
5074         pos = pci_pcie_cap(pdev->bus->self);
5075         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5076         reg16 &= ~state;
5077         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5078 }
5079 #endif
5080 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5081 {
5082         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5083                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5084                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5085
5086         __e1000e_disable_aspm(pdev, state);
5087 }
5088
5089 #ifdef CONFIG_PM_OPS
5090 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5091 {
5092         return !!adapter->tx_ring->buffer_info;
5093 }
5094
5095 static int __e1000_resume(struct pci_dev *pdev)
5096 {
5097         struct net_device *netdev = pci_get_drvdata(pdev);
5098         struct e1000_adapter *adapter = netdev_priv(netdev);
5099         struct e1000_hw *hw = &adapter->hw;
5100         u32 err;
5101
5102         pci_set_power_state(pdev, PCI_D0);
5103         pci_restore_state(pdev);
5104         pci_save_state(pdev);
5105         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5106                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5107
5108         e1000e_set_interrupt_capability(adapter);
5109         if (netif_running(netdev)) {
5110                 err = e1000_request_irq(adapter);
5111                 if (err)
5112                         return err;
5113         }
5114
5115         e1000e_power_up_phy(adapter);
5116
5117         /* report the system wakeup cause from S3/S4 */
5118         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5119                 u16 phy_data;
5120
5121                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5122                 if (phy_data) {
5123                         e_info("PHY Wakeup cause - %s\n",
5124                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5125                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5126                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5127                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5128                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5129                                 " Change" : "other");
5130                 }
5131                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5132         } else {
5133                 u32 wus = er32(WUS);
5134                 if (wus) {
5135                         e_info("MAC Wakeup cause - %s\n",
5136                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5137                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5138                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5139                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5140                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5141                                 "other");
5142                 }
5143                 ew32(WUS, ~0);
5144         }
5145
5146         e1000e_reset(adapter);
5147
5148         e1000_init_manageability_pt(adapter);
5149
5150         if (netif_running(netdev))
5151                 e1000e_up(adapter);
5152
5153         netif_device_attach(netdev);
5154
5155         /*
5156          * If the controller has AMT, do not set DRV_LOAD until the interface
5157          * is up.  For all other cases, let the f/w know that the h/w is now
5158          * under the control of the driver.
5159          */
5160         if (!(adapter->flags & FLAG_HAS_AMT))
5161                 e1000_get_hw_control(adapter);
5162
5163         return 0;
5164 }
5165
5166 #ifdef CONFIG_PM_SLEEP
5167 static int e1000_suspend(struct device *dev)
5168 {
5169         struct pci_dev *pdev = to_pci_dev(dev);
5170         int retval;
5171         bool wake;
5172
5173         retval = __e1000_shutdown(pdev, &wake, false);
5174         if (!retval)
5175                 e1000_complete_shutdown(pdev, true, wake);
5176
5177         return retval;
5178 }
5179
5180 static int e1000_resume(struct device *dev)
5181 {
5182         struct pci_dev *pdev = to_pci_dev(dev);
5183         struct net_device *netdev = pci_get_drvdata(pdev);
5184         struct e1000_adapter *adapter = netdev_priv(netdev);
5185
5186         if (e1000e_pm_ready(adapter))
5187                 adapter->idle_check = true;
5188
5189         return __e1000_resume(pdev);
5190 }
5191 #endif /* CONFIG_PM_SLEEP */
5192
5193 #ifdef CONFIG_PM_RUNTIME
5194 static int e1000_runtime_suspend(struct device *dev)
5195 {
5196         struct pci_dev *pdev = to_pci_dev(dev);
5197         struct net_device *netdev = pci_get_drvdata(pdev);
5198         struct e1000_adapter *adapter = netdev_priv(netdev);
5199
5200         if (e1000e_pm_ready(adapter)) {
5201                 bool wake;
5202
5203                 __e1000_shutdown(pdev, &wake, true);
5204         }
5205
5206         return 0;
5207 }
5208
5209 static int e1000_idle(struct device *dev)
5210 {
5211         struct pci_dev *pdev = to_pci_dev(dev);
5212         struct net_device *netdev = pci_get_drvdata(pdev);
5213         struct e1000_adapter *adapter = netdev_priv(netdev);
5214
5215         if (!e1000e_pm_ready(adapter))
5216                 return 0;
5217
5218         if (adapter->idle_check) {
5219                 adapter->idle_check = false;
5220                 if (!e1000e_has_link(adapter))
5221                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5222         }
5223
5224         return -EBUSY;
5225 }
5226
5227 static int e1000_runtime_resume(struct device *dev)
5228 {
5229         struct pci_dev *pdev = to_pci_dev(dev);
5230         struct net_device *netdev = pci_get_drvdata(pdev);
5231         struct e1000_adapter *adapter = netdev_priv(netdev);
5232
5233         if (!e1000e_pm_ready(adapter))
5234                 return 0;
5235
5236         adapter->idle_check = !dev->power.runtime_auto;
5237         return __e1000_resume(pdev);
5238 }
5239 #endif /* CONFIG_PM_RUNTIME */
5240 #endif /* CONFIG_PM_OPS */
5241
5242 static void e1000_shutdown(struct pci_dev *pdev)
5243 {
5244         bool wake = false;
5245
5246         __e1000_shutdown(pdev, &wake, false);
5247
5248         if (system_state == SYSTEM_POWER_OFF)
5249                 e1000_complete_shutdown(pdev, false, wake);
5250 }
5251
5252 #ifdef CONFIG_NET_POLL_CONTROLLER
5253 /*
5254  * Polling 'interrupt' - used by things like netconsole to send skbs
5255  * without having to re-enable interrupts. It's not called while
5256  * the interrupt routine is executing.
5257  */
5258 static void e1000_netpoll(struct net_device *netdev)
5259 {
5260         struct e1000_adapter *adapter = netdev_priv(netdev);
5261
5262         disable_irq(adapter->pdev->irq);
5263         e1000_intr(adapter->pdev->irq, netdev);
5264
5265         enable_irq(adapter->pdev->irq);
5266 }
5267 #endif
5268
5269 /**
5270  * e1000_io_error_detected - called when PCI error is detected
5271  * @pdev: Pointer to PCI device
5272  * @state: The current pci connection state
5273  *
5274  * This function is called after a PCI bus error affecting
5275  * this device has been detected.
5276  */
5277 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5278                                                 pci_channel_state_t state)
5279 {
5280         struct net_device *netdev = pci_get_drvdata(pdev);
5281         struct e1000_adapter *adapter = netdev_priv(netdev);
5282
5283         netif_device_detach(netdev);
5284
5285         if (state == pci_channel_io_perm_failure)
5286                 return PCI_ERS_RESULT_DISCONNECT;
5287
5288         if (netif_running(netdev))
5289                 e1000e_down(adapter);
5290         pci_disable_device(pdev);
5291
5292         /* Request a slot slot reset. */
5293         return PCI_ERS_RESULT_NEED_RESET;
5294 }
5295
5296 /**
5297  * e1000_io_slot_reset - called after the pci bus has been reset.
5298  * @pdev: Pointer to PCI device
5299  *
5300  * Restart the card from scratch, as if from a cold-boot. Implementation
5301  * resembles the first-half of the e1000_resume routine.
5302  */
5303 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5304 {
5305         struct net_device *netdev = pci_get_drvdata(pdev);
5306         struct e1000_adapter *adapter = netdev_priv(netdev);
5307         struct e1000_hw *hw = &adapter->hw;
5308         int err;
5309         pci_ers_result_t result;
5310
5311         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5312                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5313         err = pci_enable_device_mem(pdev);
5314         if (err) {
5315                 dev_err(&pdev->dev,
5316                         "Cannot re-enable PCI device after reset.\n");
5317                 result = PCI_ERS_RESULT_DISCONNECT;
5318         } else {
5319                 pci_set_master(pdev);
5320                 pdev->state_saved = true;
5321                 pci_restore_state(pdev);
5322
5323                 pci_enable_wake(pdev, PCI_D3hot, 0);
5324                 pci_enable_wake(pdev, PCI_D3cold, 0);
5325
5326                 e1000e_reset(adapter);
5327                 ew32(WUS, ~0);
5328                 result = PCI_ERS_RESULT_RECOVERED;
5329         }
5330
5331         pci_cleanup_aer_uncorrect_error_status(pdev);
5332
5333         return result;
5334 }
5335
5336 /**
5337  * e1000_io_resume - called when traffic can start flowing again.
5338  * @pdev: Pointer to PCI device
5339  *
5340  * This callback is called when the error recovery driver tells us that
5341  * its OK to resume normal operation. Implementation resembles the
5342  * second-half of the e1000_resume routine.
5343  */
5344 static void e1000_io_resume(struct pci_dev *pdev)
5345 {
5346         struct net_device *netdev = pci_get_drvdata(pdev);
5347         struct e1000_adapter *adapter = netdev_priv(netdev);
5348
5349         e1000_init_manageability_pt(adapter);
5350
5351         if (netif_running(netdev)) {
5352                 if (e1000e_up(adapter)) {
5353                         dev_err(&pdev->dev,
5354                                 "can't bring device back up after reset\n");
5355                         return;
5356                 }
5357         }
5358
5359         netif_device_attach(netdev);
5360
5361         /*
5362          * If the controller has AMT, do not set DRV_LOAD until the interface
5363          * is up.  For all other cases, let the f/w know that the h/w is now
5364          * under the control of the driver.
5365          */
5366         if (!(adapter->flags & FLAG_HAS_AMT))
5367                 e1000_get_hw_control(adapter);
5368
5369 }
5370
5371 static void e1000_print_device_info(struct e1000_adapter *adapter)
5372 {
5373         struct e1000_hw *hw = &adapter->hw;
5374         struct net_device *netdev = adapter->netdev;
5375         u32 pba_num;
5376
5377         /* print bus type/speed/width info */
5378         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5379                /* bus width */
5380                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5381                 "Width x1"),
5382                /* MAC address */
5383                netdev->dev_addr);
5384         e_info("Intel(R) PRO/%s Network Connection\n",
5385                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5386         e1000e_read_pba_num(hw, &pba_num);
5387         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5388                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5389 }
5390
5391 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5392 {
5393         struct e1000_hw *hw = &adapter->hw;
5394         int ret_val;
5395         u16 buf = 0;
5396
5397         if (hw->mac.type != e1000_82573)
5398                 return;
5399
5400         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5401         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5402                 /* Deep Smart Power Down (DSPD) */
5403                 dev_warn(&adapter->pdev->dev,
5404                          "Warning: detected DSPD enabled in EEPROM\n");
5405         }
5406 }
5407
5408 static const struct net_device_ops e1000e_netdev_ops = {
5409         .ndo_open               = e1000_open,
5410         .ndo_stop               = e1000_close,
5411         .ndo_start_xmit         = e1000_xmit_frame,
5412         .ndo_get_stats          = e1000_get_stats,
5413         .ndo_set_multicast_list = e1000_set_multi,
5414         .ndo_set_mac_address    = e1000_set_mac,
5415         .ndo_change_mtu         = e1000_change_mtu,
5416         .ndo_do_ioctl           = e1000_ioctl,
5417         .ndo_tx_timeout         = e1000_tx_timeout,
5418         .ndo_validate_addr      = eth_validate_addr,
5419
5420         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5421         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5422         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5423 #ifdef CONFIG_NET_POLL_CONTROLLER
5424         .ndo_poll_controller    = e1000_netpoll,
5425 #endif
5426 };
5427
5428 /**
5429  * e1000_probe - Device Initialization Routine
5430  * @pdev: PCI device information struct
5431  * @ent: entry in e1000_pci_tbl
5432  *
5433  * Returns 0 on success, negative on failure
5434  *
5435  * e1000_probe initializes an adapter identified by a pci_dev structure.
5436  * The OS initialization, configuring of the adapter private structure,
5437  * and a hardware reset occur.
5438  **/
5439 static int __devinit e1000_probe(struct pci_dev *pdev,
5440                                  const struct pci_device_id *ent)
5441 {
5442         struct net_device *netdev;
5443         struct e1000_adapter *adapter;
5444         struct e1000_hw *hw;
5445         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5446         resource_size_t mmio_start, mmio_len;
5447         resource_size_t flash_start, flash_len;
5448
5449         static int cards_found;
5450         int i, err, pci_using_dac;
5451         u16 eeprom_data = 0;
5452         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5453
5454         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5455                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5456
5457         err = pci_enable_device_mem(pdev);
5458         if (err)
5459                 return err;
5460
5461         pci_using_dac = 0;
5462         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5463         if (!err) {
5464                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5465                 if (!err)
5466                         pci_using_dac = 1;
5467         } else {
5468                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5469                 if (err) {
5470                         err = dma_set_coherent_mask(&pdev->dev,
5471                                                     DMA_BIT_MASK(32));
5472                         if (err) {
5473                                 dev_err(&pdev->dev, "No usable DMA "
5474                                         "configuration, aborting\n");
5475                                 goto err_dma;
5476                         }
5477                 }
5478         }
5479
5480         err = pci_request_selected_regions_exclusive(pdev,
5481                                           pci_select_bars(pdev, IORESOURCE_MEM),
5482                                           e1000e_driver_name);
5483         if (err)
5484                 goto err_pci_reg;
5485
5486         /* AER (Advanced Error Reporting) hooks */
5487         pci_enable_pcie_error_reporting(pdev);
5488
5489         pci_set_master(pdev);
5490         /* PCI config space info */
5491         err = pci_save_state(pdev);
5492         if (err)
5493                 goto err_alloc_etherdev;
5494
5495         err = -ENOMEM;
5496         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5497         if (!netdev)
5498                 goto err_alloc_etherdev;
5499
5500         SET_NETDEV_DEV(netdev, &pdev->dev);
5501
5502         netdev->irq = pdev->irq;
5503
5504         pci_set_drvdata(pdev, netdev);
5505         adapter = netdev_priv(netdev);
5506         hw = &adapter->hw;
5507         adapter->netdev = netdev;
5508         adapter->pdev = pdev;
5509         adapter->ei = ei;
5510         adapter->pba = ei->pba;
5511         adapter->flags = ei->flags;
5512         adapter->flags2 = ei->flags2;
5513         adapter->hw.adapter = adapter;
5514         adapter->hw.mac.type = ei->mac;
5515         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5516         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5517
5518         mmio_start = pci_resource_start(pdev, 0);
5519         mmio_len = pci_resource_len(pdev, 0);
5520
5521         err = -EIO;
5522         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5523         if (!adapter->hw.hw_addr)
5524                 goto err_ioremap;
5525
5526         if ((adapter->flags & FLAG_HAS_FLASH) &&
5527             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5528                 flash_start = pci_resource_start(pdev, 1);
5529                 flash_len = pci_resource_len(pdev, 1);
5530                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5531                 if (!adapter->hw.flash_address)
5532                         goto err_flashmap;
5533         }
5534
5535         /* construct the net_device struct */
5536         netdev->netdev_ops              = &e1000e_netdev_ops;
5537         e1000e_set_ethtool_ops(netdev);
5538         netdev->watchdog_timeo          = 5 * HZ;
5539         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5540         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5541
5542         netdev->mem_start = mmio_start;
5543         netdev->mem_end = mmio_start + mmio_len;
5544
5545         adapter->bd_number = cards_found++;
5546
5547         e1000e_check_options(adapter);
5548
5549         /* setup adapter struct */
5550         err = e1000_sw_init(adapter);
5551         if (err)
5552                 goto err_sw_init;
5553
5554         err = -EIO;
5555
5556         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5557         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5558         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5559
5560         err = ei->get_variants(adapter);
5561         if (err)
5562                 goto err_hw_init;
5563
5564         if ((adapter->flags & FLAG_IS_ICH) &&
5565             (adapter->flags & FLAG_READ_ONLY_NVM))
5566                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5567
5568         hw->mac.ops.get_bus_info(&adapter->hw);
5569
5570         adapter->hw.phy.autoneg_wait_to_complete = 0;
5571
5572         /* Copper options */
5573         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5574                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5575                 adapter->hw.phy.disable_polarity_correction = 0;
5576                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5577         }
5578
5579         if (e1000_check_reset_block(&adapter->hw))
5580                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5581
5582         netdev->features = NETIF_F_SG |
5583                            NETIF_F_HW_CSUM |
5584                            NETIF_F_HW_VLAN_TX |
5585                            NETIF_F_HW_VLAN_RX;
5586
5587         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5588                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5589
5590         netdev->features |= NETIF_F_TSO;
5591         netdev->features |= NETIF_F_TSO6;
5592
5593         netdev->vlan_features |= NETIF_F_TSO;
5594         netdev->vlan_features |= NETIF_F_TSO6;
5595         netdev->vlan_features |= NETIF_F_HW_CSUM;
5596         netdev->vlan_features |= NETIF_F_SG;
5597
5598         if (pci_using_dac)
5599                 netdev->features |= NETIF_F_HIGHDMA;
5600
5601         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5602                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5603
5604         /*
5605          * before reading the NVM, reset the controller to
5606          * put the device in a known good starting state
5607          */
5608         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5609
5610         /*
5611          * systems with ASPM and others may see the checksum fail on the first
5612          * attempt. Let's give it a few tries
5613          */
5614         for (i = 0;; i++) {
5615                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5616                         break;
5617                 if (i == 2) {
5618                         e_err("The NVM Checksum Is Not Valid\n");
5619                         err = -EIO;
5620                         goto err_eeprom;
5621                 }
5622         }
5623
5624         e1000_eeprom_checks(adapter);
5625
5626         /* copy the MAC address */
5627         if (e1000e_read_mac_addr(&adapter->hw))
5628                 e_err("NVM Read Error while reading MAC address\n");
5629
5630         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5631         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5632
5633         if (!is_valid_ether_addr(netdev->perm_addr)) {
5634                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5635                 err = -EIO;
5636                 goto err_eeprom;
5637         }
5638
5639         init_timer(&adapter->watchdog_timer);
5640         adapter->watchdog_timer.function = &e1000_watchdog;
5641         adapter->watchdog_timer.data = (unsigned long) adapter;
5642
5643         init_timer(&adapter->phy_info_timer);
5644         adapter->phy_info_timer.function = &e1000_update_phy_info;
5645         adapter->phy_info_timer.data = (unsigned long) adapter;
5646
5647         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5648         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5649         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5650         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5651         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5652
5653         /* Initialize link parameters. User can change them with ethtool */
5654         adapter->hw.mac.autoneg = 1;
5655         adapter->fc_autoneg = 1;
5656         adapter->hw.fc.requested_mode = e1000_fc_default;
5657         adapter->hw.fc.current_mode = e1000_fc_default;
5658         adapter->hw.phy.autoneg_advertised = 0x2f;
5659
5660         /* ring size defaults */
5661         adapter->rx_ring->count = 256;
5662         adapter->tx_ring->count = 256;
5663
5664         /*
5665          * Initial Wake on LAN setting - If APM wake is enabled in
5666          * the EEPROM, enable the ACPI Magic Packet filter
5667          */
5668         if (adapter->flags & FLAG_APME_IN_WUC) {
5669                 /* APME bit in EEPROM is mapped to WUC.APME */
5670                 eeprom_data = er32(WUC);
5671                 eeprom_apme_mask = E1000_WUC_APME;
5672                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5673                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5674         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5675                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5676                     (adapter->hw.bus.func == 1))
5677                         e1000_read_nvm(&adapter->hw,
5678                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5679                 else
5680                         e1000_read_nvm(&adapter->hw,
5681                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5682         }
5683
5684         /* fetch WoL from EEPROM */
5685         if (eeprom_data & eeprom_apme_mask)
5686                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5687
5688         /*
5689          * now that we have the eeprom settings, apply the special cases
5690          * where the eeprom may be wrong or the board simply won't support
5691          * wake on lan on a particular port
5692          */
5693         if (!(adapter->flags & FLAG_HAS_WOL))
5694                 adapter->eeprom_wol = 0;
5695
5696         /* initialize the wol settings based on the eeprom settings */
5697         adapter->wol = adapter->eeprom_wol;
5698         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5699
5700         /* save off EEPROM version number */
5701         e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
5702
5703         /* reset the hardware with the new settings */
5704         e1000e_reset(adapter);
5705
5706         /*
5707          * If the controller has AMT, do not set DRV_LOAD until the interface
5708          * is up.  For all other cases, let the f/w know that the h/w is now
5709          * under the control of the driver.
5710          */
5711         if (!(adapter->flags & FLAG_HAS_AMT))
5712                 e1000_get_hw_control(adapter);
5713
5714         strcpy(netdev->name, "eth%d");
5715         err = register_netdev(netdev);
5716         if (err)
5717                 goto err_register;
5718
5719         /* carrier off reporting is important to ethtool even BEFORE open */
5720         netif_carrier_off(netdev);
5721
5722         e1000_print_device_info(adapter);
5723
5724         if (pci_dev_run_wake(pdev))
5725                 pm_runtime_put_noidle(&pdev->dev);
5726
5727         return 0;
5728
5729 err_register:
5730         if (!(adapter->flags & FLAG_HAS_AMT))
5731                 e1000_release_hw_control(adapter);
5732 err_eeprom:
5733         if (!e1000_check_reset_block(&adapter->hw))
5734                 e1000_phy_hw_reset(&adapter->hw);
5735 err_hw_init:
5736
5737         kfree(adapter->tx_ring);
5738         kfree(adapter->rx_ring);
5739 err_sw_init:
5740         if (adapter->hw.flash_address)
5741                 iounmap(adapter->hw.flash_address);
5742         e1000e_reset_interrupt_capability(adapter);
5743 err_flashmap:
5744         iounmap(adapter->hw.hw_addr);
5745 err_ioremap:
5746         free_netdev(netdev);
5747 err_alloc_etherdev:
5748         pci_release_selected_regions(pdev,
5749                                      pci_select_bars(pdev, IORESOURCE_MEM));
5750 err_pci_reg:
5751 err_dma:
5752         pci_disable_device(pdev);
5753         return err;
5754 }
5755
5756 /**
5757  * e1000_remove - Device Removal Routine
5758  * @pdev: PCI device information struct
5759  *
5760  * e1000_remove is called by the PCI subsystem to alert the driver
5761  * that it should release a PCI device.  The could be caused by a
5762  * Hot-Plug event, or because the driver is going to be removed from
5763  * memory.
5764  **/
5765 static void __devexit e1000_remove(struct pci_dev *pdev)
5766 {
5767         struct net_device *netdev = pci_get_drvdata(pdev);
5768         struct e1000_adapter *adapter = netdev_priv(netdev);
5769         bool down = test_bit(__E1000_DOWN, &adapter->state);
5770
5771         /*
5772          * flush_scheduled work may reschedule our watchdog task, so
5773          * explicitly disable watchdog tasks from being rescheduled
5774          */
5775         if (!down)
5776                 set_bit(__E1000_DOWN, &adapter->state);
5777         del_timer_sync(&adapter->watchdog_timer);
5778         del_timer_sync(&adapter->phy_info_timer);
5779
5780         cancel_work_sync(&adapter->reset_task);
5781         cancel_work_sync(&adapter->watchdog_task);
5782         cancel_work_sync(&adapter->downshift_task);
5783         cancel_work_sync(&adapter->update_phy_task);
5784         cancel_work_sync(&adapter->print_hang_task);
5785         flush_scheduled_work();
5786
5787         if (!(netdev->flags & IFF_UP))
5788                 e1000_power_down_phy(adapter);
5789
5790         /* Don't lie to e1000_close() down the road. */
5791         if (!down)
5792                 clear_bit(__E1000_DOWN, &adapter->state);
5793         unregister_netdev(netdev);
5794
5795         if (pci_dev_run_wake(pdev))
5796                 pm_runtime_get_noresume(&pdev->dev);
5797
5798         /*
5799          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5800          * would have already happened in close and is redundant.
5801          */
5802         e1000_release_hw_control(adapter);
5803
5804         e1000e_reset_interrupt_capability(adapter);
5805         kfree(adapter->tx_ring);
5806         kfree(adapter->rx_ring);
5807
5808         iounmap(adapter->hw.hw_addr);
5809         if (adapter->hw.flash_address)
5810                 iounmap(adapter->hw.flash_address);
5811         pci_release_selected_regions(pdev,
5812                                      pci_select_bars(pdev, IORESOURCE_MEM));
5813
5814         free_netdev(netdev);
5815
5816         /* AER disable */
5817         pci_disable_pcie_error_reporting(pdev);
5818
5819         pci_disable_device(pdev);
5820 }
5821
5822 /* PCI Error Recovery (ERS) */
5823 static struct pci_error_handlers e1000_err_handler = {
5824         .error_detected = e1000_io_error_detected,
5825         .slot_reset = e1000_io_slot_reset,
5826         .resume = e1000_io_resume,
5827 };
5828
5829 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
5830         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
5831         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
5832         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
5833         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },