drivers/net: return operator cleanup
[linux-2.6.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2010 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_EXTRAVERSION "-k2"
56
57 #define DRV_VERSION "1.2.7" DRV_EXTRAVERSION
58 char e1000e_driver_name[] = "e1000e";
59 const char e1000e_driver_version[] = DRV_VERSION;
60
61 static const struct e1000_info *e1000_info_tbl[] = {
62         [board_82571]           = &e1000_82571_info,
63         [board_82572]           = &e1000_82572_info,
64         [board_82573]           = &e1000_82573_info,
65         [board_82574]           = &e1000_82574_info,
66         [board_82583]           = &e1000_82583_info,
67         [board_80003es2lan]     = &e1000_es2_info,
68         [board_ich8lan]         = &e1000_ich8_info,
69         [board_ich9lan]         = &e1000_ich9_info,
70         [board_ich10lan]        = &e1000_ich10_info,
71         [board_pchlan]          = &e1000_pch_info,
72         [board_pch2lan]         = &e1000_pch2_info,
73 };
74
75 struct e1000_reg_info {
76         u32 ofs;
77         char *name;
78 };
79
80 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
81 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
82 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
83 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
84 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
85
86 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
87 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
88 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
89 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
90 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
91
92 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
93
94         /* General Registers */
95         {E1000_CTRL, "CTRL"},
96         {E1000_STATUS, "STATUS"},
97         {E1000_CTRL_EXT, "CTRL_EXT"},
98
99         /* Interrupt Registers */
100         {E1000_ICR, "ICR"},
101
102         /* RX Registers */
103         {E1000_RCTL, "RCTL"},
104         {E1000_RDLEN, "RDLEN"},
105         {E1000_RDH, "RDH"},
106         {E1000_RDT, "RDT"},
107         {E1000_RDTR, "RDTR"},
108         {E1000_RXDCTL(0), "RXDCTL"},
109         {E1000_ERT, "ERT"},
110         {E1000_RDBAL, "RDBAL"},
111         {E1000_RDBAH, "RDBAH"},
112         {E1000_RDFH, "RDFH"},
113         {E1000_RDFT, "RDFT"},
114         {E1000_RDFHS, "RDFHS"},
115         {E1000_RDFTS, "RDFTS"},
116         {E1000_RDFPC, "RDFPC"},
117
118         /* TX Registers */
119         {E1000_TCTL, "TCTL"},
120         {E1000_TDBAL, "TDBAL"},
121         {E1000_TDBAH, "TDBAH"},
122         {E1000_TDLEN, "TDLEN"},
123         {E1000_TDH, "TDH"},
124         {E1000_TDT, "TDT"},
125         {E1000_TIDV, "TIDV"},
126         {E1000_TXDCTL(0), "TXDCTL"},
127         {E1000_TADV, "TADV"},
128         {E1000_TARC(0), "TARC"},
129         {E1000_TDFH, "TDFH"},
130         {E1000_TDFT, "TDFT"},
131         {E1000_TDFHS, "TDFHS"},
132         {E1000_TDFTS, "TDFTS"},
133         {E1000_TDFPC, "TDFPC"},
134
135         /* List Terminator */
136         {}
137 };
138
139 /*
140  * e1000_regdump - register printout routine
141  */
142 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
143 {
144         int n = 0;
145         char rname[16];
146         u32 regs[8];
147
148         switch (reginfo->ofs) {
149         case E1000_RXDCTL(0):
150                 for (n = 0; n < 2; n++)
151                         regs[n] = __er32(hw, E1000_RXDCTL(n));
152                 break;
153         case E1000_TXDCTL(0):
154                 for (n = 0; n < 2; n++)
155                         regs[n] = __er32(hw, E1000_TXDCTL(n));
156                 break;
157         case E1000_TARC(0):
158                 for (n = 0; n < 2; n++)
159                         regs[n] = __er32(hw, E1000_TARC(n));
160                 break;
161         default:
162                 printk(KERN_INFO "%-15s %08x\n",
163                         reginfo->name, __er32(hw, reginfo->ofs));
164                 return;
165         }
166
167         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
168         printk(KERN_INFO "%-15s ", rname);
169         for (n = 0; n < 2; n++)
170                 printk(KERN_CONT "%08x ", regs[n]);
171         printk(KERN_CONT "\n");
172 }
173
174
175 /*
176  * e1000e_dump - Print registers, tx-ring and rx-ring
177  */
178 static void e1000e_dump(struct e1000_adapter *adapter)
179 {
180         struct net_device *netdev = adapter->netdev;
181         struct e1000_hw *hw = &adapter->hw;
182         struct e1000_reg_info *reginfo;
183         struct e1000_ring *tx_ring = adapter->tx_ring;
184         struct e1000_tx_desc *tx_desc;
185         struct my_u0 { u64 a; u64 b; } *u0;
186         struct e1000_buffer *buffer_info;
187         struct e1000_ring *rx_ring = adapter->rx_ring;
188         union e1000_rx_desc_packet_split *rx_desc_ps;
189         struct e1000_rx_desc *rx_desc;
190         struct my_u1 { u64 a; u64 b; u64 c; u64 d; } *u1;
191         u32 staterr;
192         int i = 0;
193
194         if (!netif_msg_hw(adapter))
195                 return;
196
197         /* Print netdevice Info */
198         if (netdev) {
199                 dev_info(&adapter->pdev->dev, "Net device Info\n");
200                 printk(KERN_INFO "Device Name     state            "
201                         "trans_start      last_rx\n");
202                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
203                         netdev->name,
204                         netdev->state,
205                         netdev->trans_start,
206                         netdev->last_rx);
207         }
208
209         /* Print Registers */
210         dev_info(&adapter->pdev->dev, "Register Dump\n");
211         printk(KERN_INFO " Register Name   Value\n");
212         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
213              reginfo->name; reginfo++) {
214                 e1000_regdump(hw, reginfo);
215         }
216
217         /* Print TX Ring Summary */
218         if (!netdev || !netif_running(netdev))
219                 goto exit;
220
221         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
222         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
223                 " leng ntw timestamp\n");
224         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
225         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
226                 0, tx_ring->next_to_use, tx_ring->next_to_clean,
227                 (unsigned long long)buffer_info->dma,
228                 buffer_info->length,
229                 buffer_info->next_to_watch,
230                 (unsigned long long)buffer_info->time_stamp);
231
232         /* Print TX Rings */
233         if (!netif_msg_tx_done(adapter))
234                 goto rx_ring_summary;
235
236         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
237
238         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
239          *
240          * Legacy Transmit Descriptor
241          *   +--------------------------------------------------------------+
242          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
243          *   +--------------------------------------------------------------+
244          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
245          *   +--------------------------------------------------------------+
246          *   63       48 47        36 35    32 31     24 23    16 15        0
247          *
248          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
249          *   63      48 47    40 39       32 31             16 15    8 7      0
250          *   +----------------------------------------------------------------+
251          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
252          *   +----------------------------------------------------------------+
253          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
254          *   +----------------------------------------------------------------+
255          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
256          *
257          * Extended Data Descriptor (DTYP=0x1)
258          *   +----------------------------------------------------------------+
259          * 0 |                     Buffer Address [63:0]                      |
260          *   +----------------------------------------------------------------+
261          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
262          *   +----------------------------------------------------------------+
263          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
264          */
265         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
266                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
267                 "<-- Legacy format\n");
268         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
269                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
270                 "<-- Ext Context format\n");
271         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
272                 " [bi->dma       ] leng  ntw timestamp        bi->skb "
273                 "<-- Ext Data format\n");
274         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
275                 tx_desc = E1000_TX_DESC(*tx_ring, i);
276                 buffer_info = &tx_ring->buffer_info[i];
277                 u0 = (struct my_u0 *)tx_desc;
278                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
279                         "%04X  %3X %016llX %p",
280                        (!(le64_to_cpu(u0->b) & (1<<29)) ? 'l' :
281                         ((le64_to_cpu(u0->b) & (1<<20)) ? 'd' : 'c')), i,
282                        (unsigned long long)le64_to_cpu(u0->a),
283                        (unsigned long long)le64_to_cpu(u0->b),
284                        (unsigned long long)buffer_info->dma,
285                        buffer_info->length, buffer_info->next_to_watch,
286                        (unsigned long long)buffer_info->time_stamp,
287                        buffer_info->skb);
288                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
289                         printk(KERN_CONT " NTC/U\n");
290                 else if (i == tx_ring->next_to_use)
291                         printk(KERN_CONT " NTU\n");
292                 else if (i == tx_ring->next_to_clean)
293                         printk(KERN_CONT " NTC\n");
294                 else
295                         printk(KERN_CONT "\n");
296
297                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
298                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
299                                         16, 1, phys_to_virt(buffer_info->dma),
300                                         buffer_info->length, true);
301         }
302
303         /* Print RX Rings Summary */
304 rx_ring_summary:
305         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
306         printk(KERN_INFO "Queue [NTU] [NTC]\n");
307         printk(KERN_INFO " %5d %5X %5X\n", 0,
308                 rx_ring->next_to_use, rx_ring->next_to_clean);
309
310         /* Print RX Rings */
311         if (!netif_msg_rx_status(adapter))
312                 goto exit;
313
314         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
315         switch (adapter->rx_ps_pages) {
316         case 1:
317         case 2:
318         case 3:
319                 /* [Extended] Packet Split Receive Descriptor Format
320                  *
321                  *    +-----------------------------------------------------+
322                  *  0 |                Buffer Address 0 [63:0]              |
323                  *    +-----------------------------------------------------+
324                  *  8 |                Buffer Address 1 [63:0]              |
325                  *    +-----------------------------------------------------+
326                  * 16 |                Buffer Address 2 [63:0]              |
327                  *    +-----------------------------------------------------+
328                  * 24 |                Buffer Address 3 [63:0]              |
329                  *    +-----------------------------------------------------+
330                  */
331                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
332                         "[buffer 1 63:0 ] "
333                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
334                        "[bi->skb] <-- Ext Pkt Split format\n");
335                 /* [Extended] Receive Descriptor (Write-Back) Format
336                  *
337                  *   63       48 47    32 31     13 12    8 7    4 3        0
338                  *   +------------------------------------------------------+
339                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
340                  *   | Checksum | Ident  |         | Queue |      |  Type   |
341                  *   +------------------------------------------------------+
342                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
343                  *   +------------------------------------------------------+
344                  *   63       48 47    32 31            20 19               0
345                  */
346                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
347                         "[vl   l0 ee  es] "
348                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
349                        "[bi->skb] <-- Ext Rx Write-Back format\n");
350                 for (i = 0; i < rx_ring->count; i++) {
351                         buffer_info = &rx_ring->buffer_info[i];
352                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
353                         u1 = (struct my_u1 *)rx_desc_ps;
354                         staterr =
355                                 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
356                         if (staterr & E1000_RXD_STAT_DD) {
357                                 /* Descriptor Done */
358                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
359                                         "%016llX %016llX %016llX "
360                                         "---------------- %p", i,
361                                         (unsigned long long)le64_to_cpu(u1->a),
362                                         (unsigned long long)le64_to_cpu(u1->b),
363                                         (unsigned long long)le64_to_cpu(u1->c),
364                                         (unsigned long long)le64_to_cpu(u1->d),
365                                         buffer_info->skb);
366                         } else {
367                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
368                                         "%016llX %016llX %016llX %016llX %p", i,
369                                         (unsigned long long)le64_to_cpu(u1->a),
370                                         (unsigned long long)le64_to_cpu(u1->b),
371                                         (unsigned long long)le64_to_cpu(u1->c),
372                                         (unsigned long long)le64_to_cpu(u1->d),
373                                         (unsigned long long)buffer_info->dma,
374                                         buffer_info->skb);
375
376                                 if (netif_msg_pktdata(adapter))
377                                         print_hex_dump(KERN_INFO, "",
378                                                 DUMP_PREFIX_ADDRESS, 16, 1,
379                                                 phys_to_virt(buffer_info->dma),
380                                                 adapter->rx_ps_bsize0, true);
381                         }
382
383                         if (i == rx_ring->next_to_use)
384                                 printk(KERN_CONT " NTU\n");
385                         else if (i == rx_ring->next_to_clean)
386                                 printk(KERN_CONT " NTC\n");
387                         else
388                                 printk(KERN_CONT "\n");
389                 }
390                 break;
391         default:
392         case 0:
393                 /* Legacy Receive Descriptor Format
394                  *
395                  * +-----------------------------------------------------+
396                  * |                Buffer Address [63:0]                |
397                  * +-----------------------------------------------------+
398                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
399                  * +-----------------------------------------------------+
400                  * 63       48 47    40 39      32 31         16 15      0
401                  */
402                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
403                         "[vl er S cks ln] [bi->dma       ] [bi->skb] "
404                         "<-- Legacy format\n");
405                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
406                         rx_desc = E1000_RX_DESC(*rx_ring, i);
407                         buffer_info = &rx_ring->buffer_info[i];
408                         u0 = (struct my_u0 *)rx_desc;
409                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
410                                 "%016llX %p", i,
411                                 (unsigned long long)le64_to_cpu(u0->a),
412                                 (unsigned long long)le64_to_cpu(u0->b),
413                                 (unsigned long long)buffer_info->dma,
414                                 buffer_info->skb);
415                         if (i == rx_ring->next_to_use)
416                                 printk(KERN_CONT " NTU\n");
417                         else if (i == rx_ring->next_to_clean)
418                                 printk(KERN_CONT " NTC\n");
419                         else
420                                 printk(KERN_CONT "\n");
421
422                         if (netif_msg_pktdata(adapter))
423                                 print_hex_dump(KERN_INFO, "",
424                                         DUMP_PREFIX_ADDRESS,
425                                         16, 1, phys_to_virt(buffer_info->dma),
426                                         adapter->rx_buffer_len, true);
427                 }
428         }
429
430 exit:
431         return;
432 }
433
434 /**
435  * e1000_desc_unused - calculate if we have unused descriptors
436  **/
437 static int e1000_desc_unused(struct e1000_ring *ring)
438 {
439         if (ring->next_to_clean > ring->next_to_use)
440                 return ring->next_to_clean - ring->next_to_use - 1;
441
442         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
443 }
444
445 /**
446  * e1000_receive_skb - helper function to handle Rx indications
447  * @adapter: board private structure
448  * @status: descriptor status field as written by hardware
449  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
450  * @skb: pointer to sk_buff to be indicated to stack
451  **/
452 static void e1000_receive_skb(struct e1000_adapter *adapter,
453                               struct net_device *netdev,
454                               struct sk_buff *skb,
455                               u8 status, __le16 vlan)
456 {
457         skb->protocol = eth_type_trans(skb, netdev);
458
459         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
460                 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
461                                  le16_to_cpu(vlan), skb);
462         else
463                 napi_gro_receive(&adapter->napi, skb);
464 }
465
466 /**
467  * e1000_rx_checksum - Receive Checksum Offload for 82543
468  * @adapter:     board private structure
469  * @status_err:  receive descriptor status and error fields
470  * @csum:       receive descriptor csum field
471  * @sk_buff:     socket buffer with received data
472  **/
473 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
474                               u32 csum, struct sk_buff *skb)
475 {
476         u16 status = (u16)status_err;
477         u8 errors = (u8)(status_err >> 24);
478
479         skb_checksum_none_assert(skb);
480
481         /* Ignore Checksum bit is set */
482         if (status & E1000_RXD_STAT_IXSM)
483                 return;
484         /* TCP/UDP checksum error bit is set */
485         if (errors & E1000_RXD_ERR_TCPE) {
486                 /* let the stack verify checksum errors */
487                 adapter->hw_csum_err++;
488                 return;
489         }
490
491         /* TCP/UDP Checksum has not been calculated */
492         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
493                 return;
494
495         /* It must be a TCP or UDP packet with a valid checksum */
496         if (status & E1000_RXD_STAT_TCPCS) {
497                 /* TCP checksum is good */
498                 skb->ip_summed = CHECKSUM_UNNECESSARY;
499         } else {
500                 /*
501                  * IP fragment with UDP payload
502                  * Hardware complements the payload checksum, so we undo it
503                  * and then put the value in host order for further stack use.
504                  */
505                 __sum16 sum = (__force __sum16)htons(csum);
506                 skb->csum = csum_unfold(~sum);
507                 skb->ip_summed = CHECKSUM_COMPLETE;
508         }
509         adapter->hw_csum_good++;
510 }
511
512 /**
513  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
514  * @adapter: address of board private structure
515  **/
516 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
517                                    int cleaned_count)
518 {
519         struct net_device *netdev = adapter->netdev;
520         struct pci_dev *pdev = adapter->pdev;
521         struct e1000_ring *rx_ring = adapter->rx_ring;
522         struct e1000_rx_desc *rx_desc;
523         struct e1000_buffer *buffer_info;
524         struct sk_buff *skb;
525         unsigned int i;
526         unsigned int bufsz = adapter->rx_buffer_len;
527
528         i = rx_ring->next_to_use;
529         buffer_info = &rx_ring->buffer_info[i];
530
531         while (cleaned_count--) {
532                 skb = buffer_info->skb;
533                 if (skb) {
534                         skb_trim(skb, 0);
535                         goto map_skb;
536                 }
537
538                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
539                 if (!skb) {
540                         /* Better luck next round */
541                         adapter->alloc_rx_buff_failed++;
542                         break;
543                 }
544
545                 buffer_info->skb = skb;
546 map_skb:
547                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
548                                                   adapter->rx_buffer_len,
549                                                   DMA_FROM_DEVICE);
550                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
551                         dev_err(&pdev->dev, "RX DMA map failed\n");
552                         adapter->rx_dma_failed++;
553                         break;
554                 }
555
556                 rx_desc = E1000_RX_DESC(*rx_ring, i);
557                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
558
559                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
560                         /*
561                          * Force memory writes to complete before letting h/w
562                          * know there are new descriptors to fetch.  (Only
563                          * applicable for weak-ordered memory model archs,
564                          * such as IA-64).
565                          */
566                         wmb();
567                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
568                 }
569                 i++;
570                 if (i == rx_ring->count)
571                         i = 0;
572                 buffer_info = &rx_ring->buffer_info[i];
573         }
574
575         rx_ring->next_to_use = i;
576 }
577
578 /**
579  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
580  * @adapter: address of board private structure
581  **/
582 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
583                                       int cleaned_count)
584 {
585         struct net_device *netdev = adapter->netdev;
586         struct pci_dev *pdev = adapter->pdev;
587         union e1000_rx_desc_packet_split *rx_desc;
588         struct e1000_ring *rx_ring = adapter->rx_ring;
589         struct e1000_buffer *buffer_info;
590         struct e1000_ps_page *ps_page;
591         struct sk_buff *skb;
592         unsigned int i, j;
593
594         i = rx_ring->next_to_use;
595         buffer_info = &rx_ring->buffer_info[i];
596
597         while (cleaned_count--) {
598                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
599
600                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
601                         ps_page = &buffer_info->ps_pages[j];
602                         if (j >= adapter->rx_ps_pages) {
603                                 /* all unused desc entries get hw null ptr */
604                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
605                                 continue;
606                         }
607                         if (!ps_page->page) {
608                                 ps_page->page = alloc_page(GFP_ATOMIC);
609                                 if (!ps_page->page) {
610                                         adapter->alloc_rx_buff_failed++;
611                                         goto no_buffers;
612                                 }
613                                 ps_page->dma = dma_map_page(&pdev->dev,
614                                                             ps_page->page,
615                                                             0, PAGE_SIZE,
616                                                             DMA_FROM_DEVICE);
617                                 if (dma_mapping_error(&pdev->dev,
618                                                       ps_page->dma)) {
619                                         dev_err(&adapter->pdev->dev,
620                                           "RX DMA page map failed\n");
621                                         adapter->rx_dma_failed++;
622                                         goto no_buffers;
623                                 }
624                         }
625                         /*
626                          * Refresh the desc even if buffer_addrs
627                          * didn't change because each write-back
628                          * erases this info.
629                          */
630                         rx_desc->read.buffer_addr[j+1] =
631                              cpu_to_le64(ps_page->dma);
632                 }
633
634                 skb = netdev_alloc_skb_ip_align(netdev,
635                                                 adapter->rx_ps_bsize0);
636
637                 if (!skb) {
638                         adapter->alloc_rx_buff_failed++;
639                         break;
640                 }
641
642                 buffer_info->skb = skb;
643                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
644                                                   adapter->rx_ps_bsize0,
645                                                   DMA_FROM_DEVICE);
646                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
647                         dev_err(&pdev->dev, "RX DMA map failed\n");
648                         adapter->rx_dma_failed++;
649                         /* cleanup skb */
650                         dev_kfree_skb_any(skb);
651                         buffer_info->skb = NULL;
652                         break;
653                 }
654
655                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
656
657                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
658                         /*
659                          * Force memory writes to complete before letting h/w
660                          * know there are new descriptors to fetch.  (Only
661                          * applicable for weak-ordered memory model archs,
662                          * such as IA-64).
663                          */
664                         wmb();
665                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
666                 }
667
668                 i++;
669                 if (i == rx_ring->count)
670                         i = 0;
671                 buffer_info = &rx_ring->buffer_info[i];
672         }
673
674 no_buffers:
675         rx_ring->next_to_use = i;
676 }
677
678 /**
679  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
680  * @adapter: address of board private structure
681  * @cleaned_count: number of buffers to allocate this pass
682  **/
683
684 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
685                                          int cleaned_count)
686 {
687         struct net_device *netdev = adapter->netdev;
688         struct pci_dev *pdev = adapter->pdev;
689         struct e1000_rx_desc *rx_desc;
690         struct e1000_ring *rx_ring = adapter->rx_ring;
691         struct e1000_buffer *buffer_info;
692         struct sk_buff *skb;
693         unsigned int i;
694         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
695
696         i = rx_ring->next_to_use;
697         buffer_info = &rx_ring->buffer_info[i];
698
699         while (cleaned_count--) {
700                 skb = buffer_info->skb;
701                 if (skb) {
702                         skb_trim(skb, 0);
703                         goto check_page;
704                 }
705
706                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
707                 if (unlikely(!skb)) {
708                         /* Better luck next round */
709                         adapter->alloc_rx_buff_failed++;
710                         break;
711                 }
712
713                 buffer_info->skb = skb;
714 check_page:
715                 /* allocate a new page if necessary */
716                 if (!buffer_info->page) {
717                         buffer_info->page = alloc_page(GFP_ATOMIC);
718                         if (unlikely(!buffer_info->page)) {
719                                 adapter->alloc_rx_buff_failed++;
720                                 break;
721                         }
722                 }
723
724                 if (!buffer_info->dma)
725                         buffer_info->dma = dma_map_page(&pdev->dev,
726                                                         buffer_info->page, 0,
727                                                         PAGE_SIZE,
728                                                         DMA_FROM_DEVICE);
729
730                 rx_desc = E1000_RX_DESC(*rx_ring, i);
731                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
732
733                 if (unlikely(++i == rx_ring->count))
734                         i = 0;
735                 buffer_info = &rx_ring->buffer_info[i];
736         }
737
738         if (likely(rx_ring->next_to_use != i)) {
739                 rx_ring->next_to_use = i;
740                 if (unlikely(i-- == 0))
741                         i = (rx_ring->count - 1);
742
743                 /* Force memory writes to complete before letting h/w
744                  * know there are new descriptors to fetch.  (Only
745                  * applicable for weak-ordered memory model archs,
746                  * such as IA-64). */
747                 wmb();
748                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
749         }
750 }
751
752 /**
753  * e1000_clean_rx_irq - Send received data up the network stack; legacy
754  * @adapter: board private structure
755  *
756  * the return value indicates whether actual cleaning was done, there
757  * is no guarantee that everything was cleaned
758  **/
759 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
760                                int *work_done, int work_to_do)
761 {
762         struct net_device *netdev = adapter->netdev;
763         struct pci_dev *pdev = adapter->pdev;
764         struct e1000_hw *hw = &adapter->hw;
765         struct e1000_ring *rx_ring = adapter->rx_ring;
766         struct e1000_rx_desc *rx_desc, *next_rxd;
767         struct e1000_buffer *buffer_info, *next_buffer;
768         u32 length;
769         unsigned int i;
770         int cleaned_count = 0;
771         bool cleaned = 0;
772         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
773
774         i = rx_ring->next_to_clean;
775         rx_desc = E1000_RX_DESC(*rx_ring, i);
776         buffer_info = &rx_ring->buffer_info[i];
777
778         while (rx_desc->status & E1000_RXD_STAT_DD) {
779                 struct sk_buff *skb;
780                 u8 status;
781
782                 if (*work_done >= work_to_do)
783                         break;
784                 (*work_done)++;
785                 rmb();  /* read descriptor and rx_buffer_info after status DD */
786
787                 status = rx_desc->status;
788                 skb = buffer_info->skb;
789                 buffer_info->skb = NULL;
790
791                 prefetch(skb->data - NET_IP_ALIGN);
792
793                 i++;
794                 if (i == rx_ring->count)
795                         i = 0;
796                 next_rxd = E1000_RX_DESC(*rx_ring, i);
797                 prefetch(next_rxd);
798
799                 next_buffer = &rx_ring->buffer_info[i];
800
801                 cleaned = 1;
802                 cleaned_count++;
803                 dma_unmap_single(&pdev->dev,
804                                  buffer_info->dma,
805                                  adapter->rx_buffer_len,
806                                  DMA_FROM_DEVICE);
807                 buffer_info->dma = 0;
808
809                 length = le16_to_cpu(rx_desc->length);
810
811                 /*
812                  * !EOP means multiple descriptors were used to store a single
813                  * packet, if that's the case we need to toss it.  In fact, we
814                  * need to toss every packet with the EOP bit clear and the
815                  * next frame that _does_ have the EOP bit set, as it is by
816                  * definition only a frame fragment
817                  */
818                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
819                         adapter->flags2 |= FLAG2_IS_DISCARDING;
820
821                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
822                         /* All receives must fit into a single buffer */
823                         e_dbg("Receive packet consumed multiple buffers\n");
824                         /* recycle */
825                         buffer_info->skb = skb;
826                         if (status & E1000_RXD_STAT_EOP)
827                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
828                         goto next_desc;
829                 }
830
831                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
832                         /* recycle */
833                         buffer_info->skb = skb;
834                         goto next_desc;
835                 }
836
837                 /* adjust length to remove Ethernet CRC */
838                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
839                         length -= 4;
840
841                 total_rx_bytes += length;
842                 total_rx_packets++;
843
844                 /*
845                  * code added for copybreak, this should improve
846                  * performance for small packets with large amounts
847                  * of reassembly being done in the stack
848                  */
849                 if (length < copybreak) {
850                         struct sk_buff *new_skb =
851                             netdev_alloc_skb_ip_align(netdev, length);
852                         if (new_skb) {
853                                 skb_copy_to_linear_data_offset(new_skb,
854                                                                -NET_IP_ALIGN,
855                                                                (skb->data -
856                                                                 NET_IP_ALIGN),
857                                                                (length +
858                                                                 NET_IP_ALIGN));
859                                 /* save the skb in buffer_info as good */
860                                 buffer_info->skb = skb;
861                                 skb = new_skb;
862                         }
863                         /* else just continue with the old one */
864                 }
865                 /* end copybreak code */
866                 skb_put(skb, length);
867
868                 /* Receive Checksum Offload */
869                 e1000_rx_checksum(adapter,
870                                   (u32)(status) |
871                                   ((u32)(rx_desc->errors) << 24),
872                                   le16_to_cpu(rx_desc->csum), skb);
873
874                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
875
876 next_desc:
877                 rx_desc->status = 0;
878
879                 /* return some buffers to hardware, one at a time is too slow */
880                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
881                         adapter->alloc_rx_buf(adapter, cleaned_count);
882                         cleaned_count = 0;
883                 }
884
885                 /* use prefetched values */
886                 rx_desc = next_rxd;
887                 buffer_info = next_buffer;
888         }
889         rx_ring->next_to_clean = i;
890
891         cleaned_count = e1000_desc_unused(rx_ring);
892         if (cleaned_count)
893                 adapter->alloc_rx_buf(adapter, cleaned_count);
894
895         adapter->total_rx_bytes += total_rx_bytes;
896         adapter->total_rx_packets += total_rx_packets;
897         netdev->stats.rx_bytes += total_rx_bytes;
898         netdev->stats.rx_packets += total_rx_packets;
899         return cleaned;
900 }
901
902 static void e1000_put_txbuf(struct e1000_adapter *adapter,
903                              struct e1000_buffer *buffer_info)
904 {
905         if (buffer_info->dma) {
906                 if (buffer_info->mapped_as_page)
907                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
908                                        buffer_info->length, DMA_TO_DEVICE);
909                 else
910                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
911                                          buffer_info->length, DMA_TO_DEVICE);
912                 buffer_info->dma = 0;
913         }
914         if (buffer_info->skb) {
915                 dev_kfree_skb_any(buffer_info->skb);
916                 buffer_info->skb = NULL;
917         }
918         buffer_info->time_stamp = 0;
919 }
920
921 static void e1000_print_hw_hang(struct work_struct *work)
922 {
923         struct e1000_adapter *adapter = container_of(work,
924                                                      struct e1000_adapter,
925                                                      print_hang_task);
926         struct e1000_ring *tx_ring = adapter->tx_ring;
927         unsigned int i = tx_ring->next_to_clean;
928         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
929         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
930         struct e1000_hw *hw = &adapter->hw;
931         u16 phy_status, phy_1000t_status, phy_ext_status;
932         u16 pci_status;
933
934         e1e_rphy(hw, PHY_STATUS, &phy_status);
935         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
936         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
937
938         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
939
940         /* detected Hardware unit hang */
941         e_err("Detected Hardware Unit Hang:\n"
942               "  TDH                  <%x>\n"
943               "  TDT                  <%x>\n"
944               "  next_to_use          <%x>\n"
945               "  next_to_clean        <%x>\n"
946               "buffer_info[next_to_clean]:\n"
947               "  time_stamp           <%lx>\n"
948               "  next_to_watch        <%x>\n"
949               "  jiffies              <%lx>\n"
950               "  next_to_watch.status <%x>\n"
951               "MAC Status             <%x>\n"
952               "PHY Status             <%x>\n"
953               "PHY 1000BASE-T Status  <%x>\n"
954               "PHY Extended Status    <%x>\n"
955               "PCI Status             <%x>\n",
956               readl(adapter->hw.hw_addr + tx_ring->head),
957               readl(adapter->hw.hw_addr + tx_ring->tail),
958               tx_ring->next_to_use,
959               tx_ring->next_to_clean,
960               tx_ring->buffer_info[eop].time_stamp,
961               eop,
962               jiffies,
963               eop_desc->upper.fields.status,
964               er32(STATUS),
965               phy_status,
966               phy_1000t_status,
967               phy_ext_status,
968               pci_status);
969 }
970
971 /**
972  * e1000_clean_tx_irq - Reclaim resources after transmit completes
973  * @adapter: board private structure
974  *
975  * the return value indicates whether actual cleaning was done, there
976  * is no guarantee that everything was cleaned
977  **/
978 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
979 {
980         struct net_device *netdev = adapter->netdev;
981         struct e1000_hw *hw = &adapter->hw;
982         struct e1000_ring *tx_ring = adapter->tx_ring;
983         struct e1000_tx_desc *tx_desc, *eop_desc;
984         struct e1000_buffer *buffer_info;
985         unsigned int i, eop;
986         unsigned int count = 0;
987         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
988
989         i = tx_ring->next_to_clean;
990         eop = tx_ring->buffer_info[i].next_to_watch;
991         eop_desc = E1000_TX_DESC(*tx_ring, eop);
992
993         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
994                (count < tx_ring->count)) {
995                 bool cleaned = false;
996                 rmb(); /* read buffer_info after eop_desc */
997                 for (; !cleaned; count++) {
998                         tx_desc = E1000_TX_DESC(*tx_ring, i);
999                         buffer_info = &tx_ring->buffer_info[i];
1000                         cleaned = (i == eop);
1001
1002                         if (cleaned) {
1003                                 total_tx_packets += buffer_info->segs;
1004                                 total_tx_bytes += buffer_info->bytecount;
1005                         }
1006
1007                         e1000_put_txbuf(adapter, buffer_info);
1008                         tx_desc->upper.data = 0;
1009
1010                         i++;
1011                         if (i == tx_ring->count)
1012                                 i = 0;
1013                 }
1014
1015                 if (i == tx_ring->next_to_use)
1016                         break;
1017                 eop = tx_ring->buffer_info[i].next_to_watch;
1018                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1019         }
1020
1021         tx_ring->next_to_clean = i;
1022
1023 #define TX_WAKE_THRESHOLD 32
1024         if (count && netif_carrier_ok(netdev) &&
1025             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1026                 /* Make sure that anybody stopping the queue after this
1027                  * sees the new next_to_clean.
1028                  */
1029                 smp_mb();
1030
1031                 if (netif_queue_stopped(netdev) &&
1032                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1033                         netif_wake_queue(netdev);
1034                         ++adapter->restart_queue;
1035                 }
1036         }
1037
1038         if (adapter->detect_tx_hung) {
1039                 /*
1040                  * Detect a transmit hang in hardware, this serializes the
1041                  * check with the clearing of time_stamp and movement of i
1042                  */
1043                 adapter->detect_tx_hung = 0;
1044                 if (tx_ring->buffer_info[i].time_stamp &&
1045                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1046                                + (adapter->tx_timeout_factor * HZ)) &&
1047                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1048                         schedule_work(&adapter->print_hang_task);
1049                         netif_stop_queue(netdev);
1050                 }
1051         }
1052         adapter->total_tx_bytes += total_tx_bytes;
1053         adapter->total_tx_packets += total_tx_packets;
1054         netdev->stats.tx_bytes += total_tx_bytes;
1055         netdev->stats.tx_packets += total_tx_packets;
1056         return count < tx_ring->count;
1057 }
1058
1059 /**
1060  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1061  * @adapter: board private structure
1062  *
1063  * the return value indicates whether actual cleaning was done, there
1064  * is no guarantee that everything was cleaned
1065  **/
1066 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1067                                   int *work_done, int work_to_do)
1068 {
1069         struct e1000_hw *hw = &adapter->hw;
1070         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1071         struct net_device *netdev = adapter->netdev;
1072         struct pci_dev *pdev = adapter->pdev;
1073         struct e1000_ring *rx_ring = adapter->rx_ring;
1074         struct e1000_buffer *buffer_info, *next_buffer;
1075         struct e1000_ps_page *ps_page;
1076         struct sk_buff *skb;
1077         unsigned int i, j;
1078         u32 length, staterr;
1079         int cleaned_count = 0;
1080         bool cleaned = 0;
1081         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1082
1083         i = rx_ring->next_to_clean;
1084         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1085         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1086         buffer_info = &rx_ring->buffer_info[i];
1087
1088         while (staterr & E1000_RXD_STAT_DD) {
1089                 if (*work_done >= work_to_do)
1090                         break;
1091                 (*work_done)++;
1092                 skb = buffer_info->skb;
1093                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1094
1095                 /* in the packet split case this is header only */
1096                 prefetch(skb->data - NET_IP_ALIGN);
1097
1098                 i++;
1099                 if (i == rx_ring->count)
1100                         i = 0;
1101                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1102                 prefetch(next_rxd);
1103
1104                 next_buffer = &rx_ring->buffer_info[i];
1105
1106                 cleaned = 1;
1107                 cleaned_count++;
1108                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1109                                  adapter->rx_ps_bsize0,
1110                                  DMA_FROM_DEVICE);
1111                 buffer_info->dma = 0;
1112
1113                 /* see !EOP comment in other rx routine */
1114                 if (!(staterr & E1000_RXD_STAT_EOP))
1115                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1116
1117                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1118                         e_dbg("Packet Split buffers didn't pick up the full "
1119                               "packet\n");
1120                         dev_kfree_skb_irq(skb);
1121                         if (staterr & E1000_RXD_STAT_EOP)
1122                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1123                         goto next_desc;
1124                 }
1125
1126                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1127                         dev_kfree_skb_irq(skb);
1128                         goto next_desc;
1129                 }
1130
1131                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1132
1133                 if (!length) {
1134                         e_dbg("Last part of the packet spanning multiple "
1135                               "descriptors\n");
1136                         dev_kfree_skb_irq(skb);
1137                         goto next_desc;
1138                 }
1139
1140                 /* Good Receive */
1141                 skb_put(skb, length);
1142
1143                 {
1144                 /*
1145                  * this looks ugly, but it seems compiler issues make it
1146                  * more efficient than reusing j
1147                  */
1148                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1149
1150                 /*
1151                  * page alloc/put takes too long and effects small packet
1152                  * throughput, so unsplit small packets and save the alloc/put
1153                  * only valid in softirq (napi) context to call kmap_*
1154                  */
1155                 if (l1 && (l1 <= copybreak) &&
1156                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1157                         u8 *vaddr;
1158
1159                         ps_page = &buffer_info->ps_pages[0];
1160
1161                         /*
1162                          * there is no documentation about how to call
1163                          * kmap_atomic, so we can't hold the mapping
1164                          * very long
1165                          */
1166                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1167                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1168                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1169                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1170                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1171                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1172                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1173
1174                         /* remove the CRC */
1175                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1176                                 l1 -= 4;
1177
1178                         skb_put(skb, l1);
1179                         goto copydone;
1180                 } /* if */
1181                 }
1182
1183                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1184                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1185                         if (!length)
1186                                 break;
1187
1188                         ps_page = &buffer_info->ps_pages[j];
1189                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1190                                        DMA_FROM_DEVICE);
1191                         ps_page->dma = 0;
1192                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1193                         ps_page->page = NULL;
1194                         skb->len += length;
1195                         skb->data_len += length;
1196                         skb->truesize += length;
1197                 }
1198
1199                 /* strip the ethernet crc, problem is we're using pages now so
1200                  * this whole operation can get a little cpu intensive
1201                  */
1202                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1203                         pskb_trim(skb, skb->len - 4);
1204
1205 copydone:
1206                 total_rx_bytes += skb->len;
1207                 total_rx_packets++;
1208
1209                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1210                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1211
1212                 if (rx_desc->wb.upper.header_status &
1213                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1214                         adapter->rx_hdr_split++;
1215
1216                 e1000_receive_skb(adapter, netdev, skb,
1217                                   staterr, rx_desc->wb.middle.vlan);
1218
1219 next_desc:
1220                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1221                 buffer_info->skb = NULL;
1222
1223                 /* return some buffers to hardware, one at a time is too slow */
1224                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1225                         adapter->alloc_rx_buf(adapter, cleaned_count);
1226                         cleaned_count = 0;
1227                 }
1228
1229                 /* use prefetched values */
1230                 rx_desc = next_rxd;
1231                 buffer_info = next_buffer;
1232
1233                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1234         }
1235         rx_ring->next_to_clean = i;
1236
1237         cleaned_count = e1000_desc_unused(rx_ring);
1238         if (cleaned_count)
1239                 adapter->alloc_rx_buf(adapter, cleaned_count);
1240
1241         adapter->total_rx_bytes += total_rx_bytes;
1242         adapter->total_rx_packets += total_rx_packets;
1243         netdev->stats.rx_bytes += total_rx_bytes;
1244         netdev->stats.rx_packets += total_rx_packets;
1245         return cleaned;
1246 }
1247
1248 /**
1249  * e1000_consume_page - helper function
1250  **/
1251 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1252                                u16 length)
1253 {
1254         bi->page = NULL;
1255         skb->len += length;
1256         skb->data_len += length;
1257         skb->truesize += length;
1258 }
1259
1260 /**
1261  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1262  * @adapter: board private structure
1263  *
1264  * the return value indicates whether actual cleaning was done, there
1265  * is no guarantee that everything was cleaned
1266  **/
1267
1268 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1269                                      int *work_done, int work_to_do)
1270 {
1271         struct net_device *netdev = adapter->netdev;
1272         struct pci_dev *pdev = adapter->pdev;
1273         struct e1000_ring *rx_ring = adapter->rx_ring;
1274         struct e1000_rx_desc *rx_desc, *next_rxd;
1275         struct e1000_buffer *buffer_info, *next_buffer;
1276         u32 length;
1277         unsigned int i;
1278         int cleaned_count = 0;
1279         bool cleaned = false;
1280         unsigned int total_rx_bytes=0, total_rx_packets=0;
1281
1282         i = rx_ring->next_to_clean;
1283         rx_desc = E1000_RX_DESC(*rx_ring, i);
1284         buffer_info = &rx_ring->buffer_info[i];
1285
1286         while (rx_desc->status & E1000_RXD_STAT_DD) {
1287                 struct sk_buff *skb;
1288                 u8 status;
1289
1290                 if (*work_done >= work_to_do)
1291                         break;
1292                 (*work_done)++;
1293                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1294
1295                 status = rx_desc->status;
1296                 skb = buffer_info->skb;
1297                 buffer_info->skb = NULL;
1298
1299                 ++i;
1300                 if (i == rx_ring->count)
1301                         i = 0;
1302                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1303                 prefetch(next_rxd);
1304
1305                 next_buffer = &rx_ring->buffer_info[i];
1306
1307                 cleaned = true;
1308                 cleaned_count++;
1309                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1310                                DMA_FROM_DEVICE);
1311                 buffer_info->dma = 0;
1312
1313                 length = le16_to_cpu(rx_desc->length);
1314
1315                 /* errors is only valid for DD + EOP descriptors */
1316                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1317                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1318                                 /* recycle both page and skb */
1319                                 buffer_info->skb = skb;
1320                                 /* an error means any chain goes out the window
1321                                  * too */
1322                                 if (rx_ring->rx_skb_top)
1323                                         dev_kfree_skb(rx_ring->rx_skb_top);
1324                                 rx_ring->rx_skb_top = NULL;
1325                                 goto next_desc;
1326                 }
1327
1328 #define rxtop rx_ring->rx_skb_top
1329                 if (!(status & E1000_RXD_STAT_EOP)) {
1330                         /* this descriptor is only the beginning (or middle) */
1331                         if (!rxtop) {
1332                                 /* this is the beginning of a chain */
1333                                 rxtop = skb;
1334                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1335                                                    0, length);
1336                         } else {
1337                                 /* this is the middle of a chain */
1338                                 skb_fill_page_desc(rxtop,
1339                                     skb_shinfo(rxtop)->nr_frags,
1340                                     buffer_info->page, 0, length);
1341                                 /* re-use the skb, only consumed the page */
1342                                 buffer_info->skb = skb;
1343                         }
1344                         e1000_consume_page(buffer_info, rxtop, length);
1345                         goto next_desc;
1346                 } else {
1347                         if (rxtop) {
1348                                 /* end of the chain */
1349                                 skb_fill_page_desc(rxtop,
1350                                     skb_shinfo(rxtop)->nr_frags,
1351                                     buffer_info->page, 0, length);
1352                                 /* re-use the current skb, we only consumed the
1353                                  * page */
1354                                 buffer_info->skb = skb;
1355                                 skb = rxtop;
1356                                 rxtop = NULL;
1357                                 e1000_consume_page(buffer_info, skb, length);
1358                         } else {
1359                                 /* no chain, got EOP, this buf is the packet
1360                                  * copybreak to save the put_page/alloc_page */
1361                                 if (length <= copybreak &&
1362                                     skb_tailroom(skb) >= length) {
1363                                         u8 *vaddr;
1364                                         vaddr = kmap_atomic(buffer_info->page,
1365                                                            KM_SKB_DATA_SOFTIRQ);
1366                                         memcpy(skb_tail_pointer(skb), vaddr,
1367                                                length);
1368                                         kunmap_atomic(vaddr,
1369                                                       KM_SKB_DATA_SOFTIRQ);
1370                                         /* re-use the page, so don't erase
1371                                          * buffer_info->page */
1372                                         skb_put(skb, length);
1373                                 } else {
1374                                         skb_fill_page_desc(skb, 0,
1375                                                            buffer_info->page, 0,
1376                                                            length);
1377                                         e1000_consume_page(buffer_info, skb,
1378                                                            length);
1379                                 }
1380                         }
1381                 }
1382
1383                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1384                 e1000_rx_checksum(adapter,
1385                                   (u32)(status) |
1386                                   ((u32)(rx_desc->errors) << 24),
1387                                   le16_to_cpu(rx_desc->csum), skb);
1388
1389                 /* probably a little skewed due to removing CRC */
1390                 total_rx_bytes += skb->len;
1391                 total_rx_packets++;
1392
1393                 /* eth type trans needs skb->data to point to something */
1394                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1395                         e_err("pskb_may_pull failed.\n");
1396                         dev_kfree_skb(skb);
1397                         goto next_desc;
1398                 }
1399
1400                 e1000_receive_skb(adapter, netdev, skb, status,
1401                                   rx_desc->special);
1402
1403 next_desc:
1404                 rx_desc->status = 0;
1405
1406                 /* return some buffers to hardware, one at a time is too slow */
1407                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1408                         adapter->alloc_rx_buf(adapter, cleaned_count);
1409                         cleaned_count = 0;
1410                 }
1411
1412                 /* use prefetched values */
1413                 rx_desc = next_rxd;
1414                 buffer_info = next_buffer;
1415         }
1416         rx_ring->next_to_clean = i;
1417
1418         cleaned_count = e1000_desc_unused(rx_ring);
1419         if (cleaned_count)
1420                 adapter->alloc_rx_buf(adapter, cleaned_count);
1421
1422         adapter->total_rx_bytes += total_rx_bytes;
1423         adapter->total_rx_packets += total_rx_packets;
1424         netdev->stats.rx_bytes += total_rx_bytes;
1425         netdev->stats.rx_packets += total_rx_packets;
1426         return cleaned;
1427 }
1428
1429 /**
1430  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1431  * @adapter: board private structure
1432  **/
1433 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1434 {
1435         struct e1000_ring *rx_ring = adapter->rx_ring;
1436         struct e1000_buffer *buffer_info;
1437         struct e1000_ps_page *ps_page;
1438         struct pci_dev *pdev = adapter->pdev;
1439         unsigned int i, j;
1440
1441         /* Free all the Rx ring sk_buffs */
1442         for (i = 0; i < rx_ring->count; i++) {
1443                 buffer_info = &rx_ring->buffer_info[i];
1444                 if (buffer_info->dma) {
1445                         if (adapter->clean_rx == e1000_clean_rx_irq)
1446                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1447                                                  adapter->rx_buffer_len,
1448                                                  DMA_FROM_DEVICE);
1449                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1450                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1451                                                PAGE_SIZE,
1452                                                DMA_FROM_DEVICE);
1453                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1454                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1455                                                  adapter->rx_ps_bsize0,
1456                                                  DMA_FROM_DEVICE);
1457                         buffer_info->dma = 0;
1458                 }
1459
1460                 if (buffer_info->page) {
1461                         put_page(buffer_info->page);
1462                         buffer_info->page = NULL;
1463                 }
1464
1465                 if (buffer_info->skb) {
1466                         dev_kfree_skb(buffer_info->skb);
1467                         buffer_info->skb = NULL;
1468                 }
1469
1470                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1471                         ps_page = &buffer_info->ps_pages[j];
1472                         if (!ps_page->page)
1473                                 break;
1474                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1475                                        DMA_FROM_DEVICE);
1476                         ps_page->dma = 0;
1477                         put_page(ps_page->page);
1478                         ps_page->page = NULL;
1479                 }
1480         }
1481
1482         /* there also may be some cached data from a chained receive */
1483         if (rx_ring->rx_skb_top) {
1484                 dev_kfree_skb(rx_ring->rx_skb_top);
1485                 rx_ring->rx_skb_top = NULL;
1486         }
1487
1488         /* Zero out the descriptor ring */
1489         memset(rx_ring->desc, 0, rx_ring->size);
1490
1491         rx_ring->next_to_clean = 0;
1492         rx_ring->next_to_use = 0;
1493         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1494
1495         writel(0, adapter->hw.hw_addr + rx_ring->head);
1496         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1497 }
1498
1499 static void e1000e_downshift_workaround(struct work_struct *work)
1500 {
1501         struct e1000_adapter *adapter = container_of(work,
1502                                         struct e1000_adapter, downshift_task);
1503
1504         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1505 }
1506
1507 /**
1508  * e1000_intr_msi - Interrupt Handler
1509  * @irq: interrupt number
1510  * @data: pointer to a network interface device structure
1511  **/
1512 static irqreturn_t e1000_intr_msi(int irq, void *data)
1513 {
1514         struct net_device *netdev = data;
1515         struct e1000_adapter *adapter = netdev_priv(netdev);
1516         struct e1000_hw *hw = &adapter->hw;
1517         u32 icr = er32(ICR);
1518
1519         /*
1520          * read ICR disables interrupts using IAM
1521          */
1522
1523         if (icr & E1000_ICR_LSC) {
1524                 hw->mac.get_link_status = 1;
1525                 /*
1526                  * ICH8 workaround-- Call gig speed drop workaround on cable
1527                  * disconnect (LSC) before accessing any PHY registers
1528                  */
1529                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1530                     (!(er32(STATUS) & E1000_STATUS_LU)))
1531                         schedule_work(&adapter->downshift_task);
1532
1533                 /*
1534                  * 80003ES2LAN workaround-- For packet buffer work-around on
1535                  * link down event; disable receives here in the ISR and reset
1536                  * adapter in watchdog
1537                  */
1538                 if (netif_carrier_ok(netdev) &&
1539                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1540                         /* disable receives */
1541                         u32 rctl = er32(RCTL);
1542                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1543                         adapter->flags |= FLAG_RX_RESTART_NOW;
1544                 }
1545                 /* guard against interrupt when we're going down */
1546                 if (!test_bit(__E1000_DOWN, &adapter->state))
1547                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1548         }
1549
1550         if (napi_schedule_prep(&adapter->napi)) {
1551                 adapter->total_tx_bytes = 0;
1552                 adapter->total_tx_packets = 0;
1553                 adapter->total_rx_bytes = 0;
1554                 adapter->total_rx_packets = 0;
1555                 __napi_schedule(&adapter->napi);
1556         }
1557
1558         return IRQ_HANDLED;
1559 }
1560
1561 /**
1562  * e1000_intr - Interrupt Handler
1563  * @irq: interrupt number
1564  * @data: pointer to a network interface device structure
1565  **/
1566 static irqreturn_t e1000_intr(int irq, void *data)
1567 {
1568         struct net_device *netdev = data;
1569         struct e1000_adapter *adapter = netdev_priv(netdev);
1570         struct e1000_hw *hw = &adapter->hw;
1571         u32 rctl, icr = er32(ICR);
1572
1573         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1574                 return IRQ_NONE;  /* Not our interrupt */
1575
1576         /*
1577          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1578          * not set, then the adapter didn't send an interrupt
1579          */
1580         if (!(icr & E1000_ICR_INT_ASSERTED))
1581                 return IRQ_NONE;
1582
1583         /*
1584          * Interrupt Auto-Mask...upon reading ICR,
1585          * interrupts are masked.  No need for the
1586          * IMC write
1587          */
1588
1589         if (icr & E1000_ICR_LSC) {
1590                 hw->mac.get_link_status = 1;
1591                 /*
1592                  * ICH8 workaround-- Call gig speed drop workaround on cable
1593                  * disconnect (LSC) before accessing any PHY registers
1594                  */
1595                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1596                     (!(er32(STATUS) & E1000_STATUS_LU)))
1597                         schedule_work(&adapter->downshift_task);
1598
1599                 /*
1600                  * 80003ES2LAN workaround--
1601                  * For packet buffer work-around on link down event;
1602                  * disable receives here in the ISR and
1603                  * reset adapter in watchdog
1604                  */
1605                 if (netif_carrier_ok(netdev) &&
1606                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1607                         /* disable receives */
1608                         rctl = er32(RCTL);
1609                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1610                         adapter->flags |= FLAG_RX_RESTART_NOW;
1611                 }
1612                 /* guard against interrupt when we're going down */
1613                 if (!test_bit(__E1000_DOWN, &adapter->state))
1614                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1615         }
1616
1617         if (napi_schedule_prep(&adapter->napi)) {
1618                 adapter->total_tx_bytes = 0;
1619                 adapter->total_tx_packets = 0;
1620                 adapter->total_rx_bytes = 0;
1621                 adapter->total_rx_packets = 0;
1622                 __napi_schedule(&adapter->napi);
1623         }
1624
1625         return IRQ_HANDLED;
1626 }
1627
1628 static irqreturn_t e1000_msix_other(int irq, void *data)
1629 {
1630         struct net_device *netdev = data;
1631         struct e1000_adapter *adapter = netdev_priv(netdev);
1632         struct e1000_hw *hw = &adapter->hw;
1633         u32 icr = er32(ICR);
1634
1635         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1636                 if (!test_bit(__E1000_DOWN, &adapter->state))
1637                         ew32(IMS, E1000_IMS_OTHER);
1638                 return IRQ_NONE;
1639         }
1640
1641         if (icr & adapter->eiac_mask)
1642                 ew32(ICS, (icr & adapter->eiac_mask));
1643
1644         if (icr & E1000_ICR_OTHER) {
1645                 if (!(icr & E1000_ICR_LSC))
1646                         goto no_link_interrupt;
1647                 hw->mac.get_link_status = 1;
1648                 /* guard against interrupt when we're going down */
1649                 if (!test_bit(__E1000_DOWN, &adapter->state))
1650                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1651         }
1652
1653 no_link_interrupt:
1654         if (!test_bit(__E1000_DOWN, &adapter->state))
1655                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1656
1657         return IRQ_HANDLED;
1658 }
1659
1660
1661 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1662 {
1663         struct net_device *netdev = data;
1664         struct e1000_adapter *adapter = netdev_priv(netdev);
1665         struct e1000_hw *hw = &adapter->hw;
1666         struct e1000_ring *tx_ring = adapter->tx_ring;
1667
1668
1669         adapter->total_tx_bytes = 0;
1670         adapter->total_tx_packets = 0;
1671
1672         if (!e1000_clean_tx_irq(adapter))
1673                 /* Ring was not completely cleaned, so fire another interrupt */
1674                 ew32(ICS, tx_ring->ims_val);
1675
1676         return IRQ_HANDLED;
1677 }
1678
1679 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1680 {
1681         struct net_device *netdev = data;
1682         struct e1000_adapter *adapter = netdev_priv(netdev);
1683
1684         /* Write the ITR value calculated at the end of the
1685          * previous interrupt.
1686          */
1687         if (adapter->rx_ring->set_itr) {
1688                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1689                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1690                 adapter->rx_ring->set_itr = 0;
1691         }
1692
1693         if (napi_schedule_prep(&adapter->napi)) {
1694                 adapter->total_rx_bytes = 0;
1695                 adapter->total_rx_packets = 0;
1696                 __napi_schedule(&adapter->napi);
1697         }
1698         return IRQ_HANDLED;
1699 }
1700
1701 /**
1702  * e1000_configure_msix - Configure MSI-X hardware
1703  *
1704  * e1000_configure_msix sets up the hardware to properly
1705  * generate MSI-X interrupts.
1706  **/
1707 static void e1000_configure_msix(struct e1000_adapter *adapter)
1708 {
1709         struct e1000_hw *hw = &adapter->hw;
1710         struct e1000_ring *rx_ring = adapter->rx_ring;
1711         struct e1000_ring *tx_ring = adapter->tx_ring;
1712         int vector = 0;
1713         u32 ctrl_ext, ivar = 0;
1714
1715         adapter->eiac_mask = 0;
1716
1717         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1718         if (hw->mac.type == e1000_82574) {
1719                 u32 rfctl = er32(RFCTL);
1720                 rfctl |= E1000_RFCTL_ACK_DIS;
1721                 ew32(RFCTL, rfctl);
1722         }
1723
1724 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1725         /* Configure Rx vector */
1726         rx_ring->ims_val = E1000_IMS_RXQ0;
1727         adapter->eiac_mask |= rx_ring->ims_val;
1728         if (rx_ring->itr_val)
1729                 writel(1000000000 / (rx_ring->itr_val * 256),
1730                        hw->hw_addr + rx_ring->itr_register);
1731         else
1732                 writel(1, hw->hw_addr + rx_ring->itr_register);
1733         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1734
1735         /* Configure Tx vector */
1736         tx_ring->ims_val = E1000_IMS_TXQ0;
1737         vector++;
1738         if (tx_ring->itr_val)
1739                 writel(1000000000 / (tx_ring->itr_val * 256),
1740                        hw->hw_addr + tx_ring->itr_register);
1741         else
1742                 writel(1, hw->hw_addr + tx_ring->itr_register);
1743         adapter->eiac_mask |= tx_ring->ims_val;
1744         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1745
1746         /* set vector for Other Causes, e.g. link changes */
1747         vector++;
1748         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1749         if (rx_ring->itr_val)
1750                 writel(1000000000 / (rx_ring->itr_val * 256),
1751                        hw->hw_addr + E1000_EITR_82574(vector));
1752         else
1753                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1754
1755         /* Cause Tx interrupts on every write back */
1756         ivar |= (1 << 31);
1757
1758         ew32(IVAR, ivar);
1759
1760         /* enable MSI-X PBA support */
1761         ctrl_ext = er32(CTRL_EXT);
1762         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1763
1764         /* Auto-Mask Other interrupts upon ICR read */
1765 #define E1000_EIAC_MASK_82574   0x01F00000
1766         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1767         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1768         ew32(CTRL_EXT, ctrl_ext);
1769         e1e_flush();
1770 }
1771
1772 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1773 {
1774         if (adapter->msix_entries) {
1775                 pci_disable_msix(adapter->pdev);
1776                 kfree(adapter->msix_entries);
1777                 adapter->msix_entries = NULL;
1778         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1779                 pci_disable_msi(adapter->pdev);
1780                 adapter->flags &= ~FLAG_MSI_ENABLED;
1781         }
1782 }
1783
1784 /**
1785  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1786  *
1787  * Attempt to configure interrupts using the best available
1788  * capabilities of the hardware and kernel.
1789  **/
1790 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1791 {
1792         int err;
1793         int i;
1794
1795         switch (adapter->int_mode) {
1796         case E1000E_INT_MODE_MSIX:
1797                 if (adapter->flags & FLAG_HAS_MSIX) {
1798                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1799                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1800                                                       sizeof(struct msix_entry),
1801                                                       GFP_KERNEL);
1802                         if (adapter->msix_entries) {
1803                                 for (i = 0; i < adapter->num_vectors; i++)
1804                                         adapter->msix_entries[i].entry = i;
1805
1806                                 err = pci_enable_msix(adapter->pdev,
1807                                                       adapter->msix_entries,
1808                                                       adapter->num_vectors);
1809                                 if (err == 0) {
1810                                         return;
1811                                 }
1812                         }
1813                         /* MSI-X failed, so fall through and try MSI */
1814                         e_err("Failed to initialize MSI-X interrupts.  "
1815                               "Falling back to MSI interrupts.\n");
1816                         e1000e_reset_interrupt_capability(adapter);
1817                 }
1818                 adapter->int_mode = E1000E_INT_MODE_MSI;
1819                 /* Fall through */
1820         case E1000E_INT_MODE_MSI:
1821                 if (!pci_enable_msi(adapter->pdev)) {
1822                         adapter->flags |= FLAG_MSI_ENABLED;
1823                 } else {
1824                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1825                         e_err("Failed to initialize MSI interrupts.  Falling "
1826                               "back to legacy interrupts.\n");
1827                 }
1828                 /* Fall through */
1829         case E1000E_INT_MODE_LEGACY:
1830                 /* Don't do anything; this is the system default */
1831                 break;
1832         }
1833
1834         /* store the number of vectors being used */
1835         adapter->num_vectors = 1;
1836 }
1837
1838 /**
1839  * e1000_request_msix - Initialize MSI-X interrupts
1840  *
1841  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1842  * kernel.
1843  **/
1844 static int e1000_request_msix(struct e1000_adapter *adapter)
1845 {
1846         struct net_device *netdev = adapter->netdev;
1847         int err = 0, vector = 0;
1848
1849         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1850                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1851         else
1852                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1853         err = request_irq(adapter->msix_entries[vector].vector,
1854                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1855                           netdev);
1856         if (err)
1857                 goto out;
1858         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1859         adapter->rx_ring->itr_val = adapter->itr;
1860         vector++;
1861
1862         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1863                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1864         else
1865                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1866         err = request_irq(adapter->msix_entries[vector].vector,
1867                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1868                           netdev);
1869         if (err)
1870                 goto out;
1871         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1872         adapter->tx_ring->itr_val = adapter->itr;
1873         vector++;
1874
1875         err = request_irq(adapter->msix_entries[vector].vector,
1876                           e1000_msix_other, 0, netdev->name, netdev);
1877         if (err)
1878                 goto out;
1879
1880         e1000_configure_msix(adapter);
1881         return 0;
1882 out:
1883         return err;
1884 }
1885
1886 /**
1887  * e1000_request_irq - initialize interrupts
1888  *
1889  * Attempts to configure interrupts using the best available
1890  * capabilities of the hardware and kernel.
1891  **/
1892 static int e1000_request_irq(struct e1000_adapter *adapter)
1893 {
1894         struct net_device *netdev = adapter->netdev;
1895         int err;
1896
1897         if (adapter->msix_entries) {
1898                 err = e1000_request_msix(adapter);
1899                 if (!err)
1900                         return err;
1901                 /* fall back to MSI */
1902                 e1000e_reset_interrupt_capability(adapter);
1903                 adapter->int_mode = E1000E_INT_MODE_MSI;
1904                 e1000e_set_interrupt_capability(adapter);
1905         }
1906         if (adapter->flags & FLAG_MSI_ENABLED) {
1907                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1908                                   netdev->name, netdev);
1909                 if (!err)
1910                         return err;
1911
1912                 /* fall back to legacy interrupt */
1913                 e1000e_reset_interrupt_capability(adapter);
1914                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1915         }
1916
1917         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1918                           netdev->name, netdev);
1919         if (err)
1920                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1921
1922         return err;
1923 }
1924
1925 static void e1000_free_irq(struct e1000_adapter *adapter)
1926 {
1927         struct net_device *netdev = adapter->netdev;
1928
1929         if (adapter->msix_entries) {
1930                 int vector = 0;
1931
1932                 free_irq(adapter->msix_entries[vector].vector, netdev);
1933                 vector++;
1934
1935                 free_irq(adapter->msix_entries[vector].vector, netdev);
1936                 vector++;
1937
1938                 /* Other Causes interrupt vector */
1939                 free_irq(adapter->msix_entries[vector].vector, netdev);
1940                 return;
1941         }
1942
1943         free_irq(adapter->pdev->irq, netdev);
1944 }
1945
1946 /**
1947  * e1000_irq_disable - Mask off interrupt generation on the NIC
1948  **/
1949 static void e1000_irq_disable(struct e1000_adapter *adapter)
1950 {
1951         struct e1000_hw *hw = &adapter->hw;
1952
1953         ew32(IMC, ~0);
1954         if (adapter->msix_entries)
1955                 ew32(EIAC_82574, 0);
1956         e1e_flush();
1957
1958         if (adapter->msix_entries) {
1959                 int i;
1960                 for (i = 0; i < adapter->num_vectors; i++)
1961                         synchronize_irq(adapter->msix_entries[i].vector);
1962         } else {
1963                 synchronize_irq(adapter->pdev->irq);
1964         }
1965 }
1966
1967 /**
1968  * e1000_irq_enable - Enable default interrupt generation settings
1969  **/
1970 static void e1000_irq_enable(struct e1000_adapter *adapter)
1971 {
1972         struct e1000_hw *hw = &adapter->hw;
1973
1974         if (adapter->msix_entries) {
1975                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1976                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1977         } else {
1978                 ew32(IMS, IMS_ENABLE_MASK);
1979         }
1980         e1e_flush();
1981 }
1982
1983 /**
1984  * e1000_get_hw_control - get control of the h/w from f/w
1985  * @adapter: address of board private structure
1986  *
1987  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1988  * For ASF and Pass Through versions of f/w this means that
1989  * the driver is loaded. For AMT version (only with 82573)
1990  * of the f/w this means that the network i/f is open.
1991  **/
1992 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1993 {
1994         struct e1000_hw *hw = &adapter->hw;
1995         u32 ctrl_ext;
1996         u32 swsm;
1997
1998         /* Let firmware know the driver has taken over */
1999         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2000                 swsm = er32(SWSM);
2001                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2002         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2003                 ctrl_ext = er32(CTRL_EXT);
2004                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2005         }
2006 }
2007
2008 /**
2009  * e1000_release_hw_control - release control of the h/w to f/w
2010  * @adapter: address of board private structure
2011  *
2012  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2013  * For ASF and Pass Through versions of f/w this means that the
2014  * driver is no longer loaded. For AMT version (only with 82573) i
2015  * of the f/w this means that the network i/f is closed.
2016  *
2017  **/
2018 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2019 {
2020         struct e1000_hw *hw = &adapter->hw;
2021         u32 ctrl_ext;
2022         u32 swsm;
2023
2024         /* Let firmware taken over control of h/w */
2025         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2026                 swsm = er32(SWSM);
2027                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2028         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2029                 ctrl_ext = er32(CTRL_EXT);
2030                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2031         }
2032 }
2033
2034 /**
2035  * @e1000_alloc_ring - allocate memory for a ring structure
2036  **/
2037 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2038                                 struct e1000_ring *ring)
2039 {
2040         struct pci_dev *pdev = adapter->pdev;
2041
2042         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2043                                         GFP_KERNEL);
2044         if (!ring->desc)
2045                 return -ENOMEM;
2046
2047         return 0;
2048 }
2049
2050 /**
2051  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2052  * @adapter: board private structure
2053  *
2054  * Return 0 on success, negative on failure
2055  **/
2056 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2057 {
2058         struct e1000_ring *tx_ring = adapter->tx_ring;
2059         int err = -ENOMEM, size;
2060
2061         size = sizeof(struct e1000_buffer) * tx_ring->count;
2062         tx_ring->buffer_info = vmalloc(size);
2063         if (!tx_ring->buffer_info)
2064                 goto err;
2065         memset(tx_ring->buffer_info, 0, size);
2066
2067         /* round up to nearest 4K */
2068         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2069         tx_ring->size = ALIGN(tx_ring->size, 4096);
2070
2071         err = e1000_alloc_ring_dma(adapter, tx_ring);
2072         if (err)
2073                 goto err;
2074
2075         tx_ring->next_to_use = 0;
2076         tx_ring->next_to_clean = 0;
2077
2078         return 0;
2079 err:
2080         vfree(tx_ring->buffer_info);
2081         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2082         return err;
2083 }
2084
2085 /**
2086  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2087  * @adapter: board private structure
2088  *
2089  * Returns 0 on success, negative on failure
2090  **/
2091 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2092 {
2093         struct e1000_ring *rx_ring = adapter->rx_ring;
2094         struct e1000_buffer *buffer_info;
2095         int i, size, desc_len, err = -ENOMEM;
2096
2097         size = sizeof(struct e1000_buffer) * rx_ring->count;
2098         rx_ring->buffer_info = vmalloc(size);
2099         if (!rx_ring->buffer_info)
2100                 goto err;
2101         memset(rx_ring->buffer_info, 0, size);
2102
2103         for (i = 0; i < rx_ring->count; i++) {
2104                 buffer_info = &rx_ring->buffer_info[i];
2105                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2106                                                 sizeof(struct e1000_ps_page),
2107                                                 GFP_KERNEL);
2108                 if (!buffer_info->ps_pages)
2109                         goto err_pages;
2110         }
2111
2112         desc_len = sizeof(union e1000_rx_desc_packet_split);
2113
2114         /* Round up to nearest 4K */
2115         rx_ring->size = rx_ring->count * desc_len;
2116         rx_ring->size = ALIGN(rx_ring->size, 4096);
2117
2118         err = e1000_alloc_ring_dma(adapter, rx_ring);
2119         if (err)
2120                 goto err_pages;
2121
2122         rx_ring->next_to_clean = 0;
2123         rx_ring->next_to_use = 0;
2124         rx_ring->rx_skb_top = NULL;
2125
2126         return 0;
2127
2128 err_pages:
2129         for (i = 0; i < rx_ring->count; i++) {
2130                 buffer_info = &rx_ring->buffer_info[i];
2131                 kfree(buffer_info->ps_pages);
2132         }
2133 err:
2134         vfree(rx_ring->buffer_info);
2135         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2136         return err;
2137 }
2138
2139 /**
2140  * e1000_clean_tx_ring - Free Tx Buffers
2141  * @adapter: board private structure
2142  **/
2143 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2144 {
2145         struct e1000_ring *tx_ring = adapter->tx_ring;
2146         struct e1000_buffer *buffer_info;
2147         unsigned long size;
2148         unsigned int i;
2149
2150         for (i = 0; i < tx_ring->count; i++) {
2151                 buffer_info = &tx_ring->buffer_info[i];
2152                 e1000_put_txbuf(adapter, buffer_info);
2153         }
2154
2155         size = sizeof(struct e1000_buffer) * tx_ring->count;
2156         memset(tx_ring->buffer_info, 0, size);
2157
2158         memset(tx_ring->desc, 0, tx_ring->size);
2159
2160         tx_ring->next_to_use = 0;
2161         tx_ring->next_to_clean = 0;
2162
2163         writel(0, adapter->hw.hw_addr + tx_ring->head);
2164         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2165 }
2166
2167 /**
2168  * e1000e_free_tx_resources - Free Tx Resources per Queue
2169  * @adapter: board private structure
2170  *
2171  * Free all transmit software resources
2172  **/
2173 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2174 {
2175         struct pci_dev *pdev = adapter->pdev;
2176         struct e1000_ring *tx_ring = adapter->tx_ring;
2177
2178         e1000_clean_tx_ring(adapter);
2179
2180         vfree(tx_ring->buffer_info);
2181         tx_ring->buffer_info = NULL;
2182
2183         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2184                           tx_ring->dma);
2185         tx_ring->desc = NULL;
2186 }
2187
2188 /**
2189  * e1000e_free_rx_resources - Free Rx Resources
2190  * @adapter: board private structure
2191  *
2192  * Free all receive software resources
2193  **/
2194
2195 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2196 {
2197         struct pci_dev *pdev = adapter->pdev;
2198         struct e1000_ring *rx_ring = adapter->rx_ring;
2199         int i;
2200
2201         e1000_clean_rx_ring(adapter);
2202
2203         for (i = 0; i < rx_ring->count; i++) {
2204                 kfree(rx_ring->buffer_info[i].ps_pages);
2205         }
2206
2207         vfree(rx_ring->buffer_info);
2208         rx_ring->buffer_info = NULL;
2209
2210         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2211                           rx_ring->dma);
2212         rx_ring->desc = NULL;
2213 }
2214
2215 /**
2216  * e1000_update_itr - update the dynamic ITR value based on statistics
2217  * @adapter: pointer to adapter
2218  * @itr_setting: current adapter->itr
2219  * @packets: the number of packets during this measurement interval
2220  * @bytes: the number of bytes during this measurement interval
2221  *
2222  *      Stores a new ITR value based on packets and byte
2223  *      counts during the last interrupt.  The advantage of per interrupt
2224  *      computation is faster updates and more accurate ITR for the current
2225  *      traffic pattern.  Constants in this function were computed
2226  *      based on theoretical maximum wire speed and thresholds were set based
2227  *      on testing data as well as attempting to minimize response time
2228  *      while increasing bulk throughput.  This functionality is controlled
2229  *      by the InterruptThrottleRate module parameter.
2230  **/
2231 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2232                                      u16 itr_setting, int packets,
2233                                      int bytes)
2234 {
2235         unsigned int retval = itr_setting;
2236
2237         if (packets == 0)
2238                 goto update_itr_done;
2239
2240         switch (itr_setting) {
2241         case lowest_latency:
2242                 /* handle TSO and jumbo frames */
2243                 if (bytes/packets > 8000)
2244                         retval = bulk_latency;
2245                 else if ((packets < 5) && (bytes > 512)) {
2246                         retval = low_latency;
2247                 }
2248                 break;
2249         case low_latency:  /* 50 usec aka 20000 ints/s */
2250                 if (bytes > 10000) {
2251                         /* this if handles the TSO accounting */
2252                         if (bytes/packets > 8000) {
2253                                 retval = bulk_latency;
2254                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2255                                 retval = bulk_latency;
2256                         } else if ((packets > 35)) {
2257                                 retval = lowest_latency;
2258                         }
2259                 } else if (bytes/packets > 2000) {
2260                         retval = bulk_latency;
2261                 } else if (packets <= 2 && bytes < 512) {
2262                         retval = lowest_latency;
2263                 }
2264                 break;
2265         case bulk_latency: /* 250 usec aka 4000 ints/s */
2266                 if (bytes > 25000) {
2267                         if (packets > 35) {
2268                                 retval = low_latency;
2269                         }
2270                 } else if (bytes < 6000) {
2271                         retval = low_latency;
2272                 }
2273                 break;
2274         }
2275
2276 update_itr_done:
2277         return retval;
2278 }
2279
2280 static void e1000_set_itr(struct e1000_adapter *adapter)
2281 {
2282         struct e1000_hw *hw = &adapter->hw;
2283         u16 current_itr;
2284         u32 new_itr = adapter->itr;
2285
2286         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2287         if (adapter->link_speed != SPEED_1000) {
2288                 current_itr = 0;
2289                 new_itr = 4000;
2290                 goto set_itr_now;
2291         }
2292
2293         adapter->tx_itr = e1000_update_itr(adapter,
2294                                     adapter->tx_itr,
2295                                     adapter->total_tx_packets,
2296                                     adapter->total_tx_bytes);
2297         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2298         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2299                 adapter->tx_itr = low_latency;
2300
2301         adapter->rx_itr = e1000_update_itr(adapter,
2302                                     adapter->rx_itr,
2303                                     adapter->total_rx_packets,
2304                                     adapter->total_rx_bytes);
2305         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2306         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2307                 adapter->rx_itr = low_latency;
2308
2309         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2310
2311         switch (current_itr) {
2312         /* counts and packets in update_itr are dependent on these numbers */
2313         case lowest_latency:
2314                 new_itr = 70000;
2315                 break;
2316         case low_latency:
2317                 new_itr = 20000; /* aka hwitr = ~200 */
2318                 break;
2319         case bulk_latency:
2320                 new_itr = 4000;
2321                 break;
2322         default:
2323                 break;
2324         }
2325
2326 set_itr_now:
2327         if (new_itr != adapter->itr) {
2328                 /*
2329                  * this attempts to bias the interrupt rate towards Bulk
2330                  * by adding intermediate steps when interrupt rate is
2331                  * increasing
2332                  */
2333                 new_itr = new_itr > adapter->itr ?
2334                              min(adapter->itr + (new_itr >> 2), new_itr) :
2335                              new_itr;
2336                 adapter->itr = new_itr;
2337                 adapter->rx_ring->itr_val = new_itr;
2338                 if (adapter->msix_entries)
2339                         adapter->rx_ring->set_itr = 1;
2340                 else
2341                         ew32(ITR, 1000000000 / (new_itr * 256));
2342         }
2343 }
2344
2345 /**
2346  * e1000_alloc_queues - Allocate memory for all rings
2347  * @adapter: board private structure to initialize
2348  **/
2349 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2350 {
2351         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2352         if (!adapter->tx_ring)
2353                 goto err;
2354
2355         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2356         if (!adapter->rx_ring)
2357                 goto err;
2358
2359         return 0;
2360 err:
2361         e_err("Unable to allocate memory for queues\n");
2362         kfree(adapter->rx_ring);
2363         kfree(adapter->tx_ring);
2364         return -ENOMEM;
2365 }
2366
2367 /**
2368  * e1000_clean - NAPI Rx polling callback
2369  * @napi: struct associated with this polling callback
2370  * @budget: amount of packets driver is allowed to process this poll
2371  **/
2372 static int e1000_clean(struct napi_struct *napi, int budget)
2373 {
2374         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2375         struct e1000_hw *hw = &adapter->hw;
2376         struct net_device *poll_dev = adapter->netdev;
2377         int tx_cleaned = 1, work_done = 0;
2378
2379         adapter = netdev_priv(poll_dev);
2380
2381         if (adapter->msix_entries &&
2382             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2383                 goto clean_rx;
2384
2385         tx_cleaned = e1000_clean_tx_irq(adapter);
2386
2387 clean_rx:
2388         adapter->clean_rx(adapter, &work_done, budget);
2389
2390         if (!tx_cleaned)
2391                 work_done = budget;
2392
2393         /* If budget not fully consumed, exit the polling mode */
2394         if (work_done < budget) {
2395                 if (adapter->itr_setting & 3)
2396                         e1000_set_itr(adapter);
2397                 napi_complete(napi);
2398                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2399                         if (adapter->msix_entries)
2400                                 ew32(IMS, adapter->rx_ring->ims_val);
2401                         else
2402                                 e1000_irq_enable(adapter);
2403                 }
2404         }
2405
2406         return work_done;
2407 }
2408
2409 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2410 {
2411         struct e1000_adapter *adapter = netdev_priv(netdev);
2412         struct e1000_hw *hw = &adapter->hw;
2413         u32 vfta, index;
2414
2415         /* don't update vlan cookie if already programmed */
2416         if ((adapter->hw.mng_cookie.status &
2417              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2418             (vid == adapter->mng_vlan_id))
2419                 return;
2420
2421         /* add VID to filter table */
2422         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2423                 index = (vid >> 5) & 0x7F;
2424                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2425                 vfta |= (1 << (vid & 0x1F));
2426                 hw->mac.ops.write_vfta(hw, index, vfta);
2427         }
2428 }
2429
2430 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2431 {
2432         struct e1000_adapter *adapter = netdev_priv(netdev);
2433         struct e1000_hw *hw = &adapter->hw;
2434         u32 vfta, index;
2435
2436         if (!test_bit(__E1000_DOWN, &adapter->state))
2437                 e1000_irq_disable(adapter);
2438         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2439
2440         if (!test_bit(__E1000_DOWN, &adapter->state))
2441                 e1000_irq_enable(adapter);
2442
2443         if ((adapter->hw.mng_cookie.status &
2444              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2445             (vid == adapter->mng_vlan_id)) {
2446                 /* release control to f/w */
2447                 e1000_release_hw_control(adapter);
2448                 return;
2449         }
2450
2451         /* remove VID from filter table */
2452         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2453                 index = (vid >> 5) & 0x7F;
2454                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2455                 vfta &= ~(1 << (vid & 0x1F));
2456                 hw->mac.ops.write_vfta(hw, index, vfta);
2457         }
2458 }
2459
2460 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2461 {
2462         struct net_device *netdev = adapter->netdev;
2463         u16 vid = adapter->hw.mng_cookie.vlan_id;
2464         u16 old_vid = adapter->mng_vlan_id;
2465
2466         if (!adapter->vlgrp)
2467                 return;
2468
2469         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2470                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2471                 if (adapter->hw.mng_cookie.status &
2472                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2473                         e1000_vlan_rx_add_vid(netdev, vid);
2474                         adapter->mng_vlan_id = vid;
2475                 }
2476
2477                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2478                                 (vid != old_vid) &&
2479                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2480                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2481         } else {
2482                 adapter->mng_vlan_id = vid;
2483         }
2484 }
2485
2486
2487 static void e1000_vlan_rx_register(struct net_device *netdev,
2488                                    struct vlan_group *grp)
2489 {
2490         struct e1000_adapter *adapter = netdev_priv(netdev);
2491         struct e1000_hw *hw = &adapter->hw;
2492         u32 ctrl, rctl;
2493
2494         if (!test_bit(__E1000_DOWN, &adapter->state))
2495                 e1000_irq_disable(adapter);
2496         adapter->vlgrp = grp;
2497
2498         if (grp) {
2499                 /* enable VLAN tag insert/strip */
2500                 ctrl = er32(CTRL);
2501                 ctrl |= E1000_CTRL_VME;
2502                 ew32(CTRL, ctrl);
2503
2504                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2505                         /* enable VLAN receive filtering */
2506                         rctl = er32(RCTL);
2507                         rctl &= ~E1000_RCTL_CFIEN;
2508                         ew32(RCTL, rctl);
2509                         e1000_update_mng_vlan(adapter);
2510                 }
2511         } else {
2512                 /* disable VLAN tag insert/strip */
2513                 ctrl = er32(CTRL);
2514                 ctrl &= ~E1000_CTRL_VME;
2515                 ew32(CTRL, ctrl);
2516
2517                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2518                         if (adapter->mng_vlan_id !=
2519                             (u16)E1000_MNG_VLAN_NONE) {
2520                                 e1000_vlan_rx_kill_vid(netdev,
2521                                                        adapter->mng_vlan_id);
2522                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2523                         }
2524                 }
2525         }
2526
2527         if (!test_bit(__E1000_DOWN, &adapter->state))
2528                 e1000_irq_enable(adapter);
2529 }
2530
2531 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2532 {
2533         u16 vid;
2534
2535         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2536
2537         if (!adapter->vlgrp)
2538                 return;
2539
2540         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2541                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2542                         continue;
2543                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2544         }
2545 }
2546
2547 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2548 {
2549         struct e1000_hw *hw = &adapter->hw;
2550         u32 manc, manc2h, mdef, i, j;
2551
2552         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2553                 return;
2554
2555         manc = er32(MANC);
2556
2557         /*
2558          * enable receiving management packets to the host. this will probably
2559          * generate destination unreachable messages from the host OS, but
2560          * the packets will be handled on SMBUS
2561          */
2562         manc |= E1000_MANC_EN_MNG2HOST;
2563         manc2h = er32(MANC2H);
2564
2565         switch (hw->mac.type) {
2566         default:
2567                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2568                 break;
2569         case e1000_82574:
2570         case e1000_82583:
2571                 /*
2572                  * Check if IPMI pass-through decision filter already exists;
2573                  * if so, enable it.
2574                  */
2575                 for (i = 0, j = 0; i < 8; i++) {
2576                         mdef = er32(MDEF(i));
2577
2578                         /* Ignore filters with anything other than IPMI ports */
2579                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2580                                 continue;
2581
2582                         /* Enable this decision filter in MANC2H */
2583                         if (mdef)
2584                                 manc2h |= (1 << i);
2585
2586                         j |= mdef;
2587                 }
2588
2589                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2590                         break;
2591
2592                 /* Create new decision filter in an empty filter */
2593                 for (i = 0, j = 0; i < 8; i++)
2594                         if (er32(MDEF(i)) == 0) {
2595                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2596                                                E1000_MDEF_PORT_664));
2597                                 manc2h |= (1 << 1);
2598                                 j++;
2599                                 break;
2600                         }
2601
2602                 if (!j)
2603                         e_warn("Unable to create IPMI pass-through filter\n");
2604                 break;
2605         }
2606
2607         ew32(MANC2H, manc2h);
2608         ew32(MANC, manc);
2609 }
2610
2611 /**
2612  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2613  * @adapter: board private structure
2614  *
2615  * Configure the Tx unit of the MAC after a reset.
2616  **/
2617 static void e1000_configure_tx(struct e1000_adapter *adapter)
2618 {
2619         struct e1000_hw *hw = &adapter->hw;
2620         struct e1000_ring *tx_ring = adapter->tx_ring;
2621         u64 tdba;
2622         u32 tdlen, tctl, tipg, tarc;
2623         u32 ipgr1, ipgr2;
2624
2625         /* Setup the HW Tx Head and Tail descriptor pointers */
2626         tdba = tx_ring->dma;
2627         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2628         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2629         ew32(TDBAH, (tdba >> 32));
2630         ew32(TDLEN, tdlen);
2631         ew32(TDH, 0);
2632         ew32(TDT, 0);
2633         tx_ring->head = E1000_TDH;
2634         tx_ring->tail = E1000_TDT;
2635
2636         /* Set the default values for the Tx Inter Packet Gap timer */
2637         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2638         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2639         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2640
2641         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2642                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2643
2644         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2645         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2646         ew32(TIPG, tipg);
2647
2648         /* Set the Tx Interrupt Delay register */
2649         ew32(TIDV, adapter->tx_int_delay);
2650         /* Tx irq moderation */
2651         ew32(TADV, adapter->tx_abs_int_delay);
2652
2653         /* Program the Transmit Control Register */
2654         tctl = er32(TCTL);
2655         tctl &= ~E1000_TCTL_CT;
2656         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2657                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2658
2659         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2660                 tarc = er32(TARC(0));
2661                 /*
2662                  * set the speed mode bit, we'll clear it if we're not at
2663                  * gigabit link later
2664                  */
2665 #define SPEED_MODE_BIT (1 << 21)
2666                 tarc |= SPEED_MODE_BIT;
2667                 ew32(TARC(0), tarc);
2668         }
2669
2670         /* errata: program both queues to unweighted RR */
2671         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2672                 tarc = er32(TARC(0));
2673                 tarc |= 1;
2674                 ew32(TARC(0), tarc);
2675                 tarc = er32(TARC(1));
2676                 tarc |= 1;
2677                 ew32(TARC(1), tarc);
2678         }
2679
2680         /* Setup Transmit Descriptor Settings for eop descriptor */
2681         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2682
2683         /* only set IDE if we are delaying interrupts using the timers */
2684         if (adapter->tx_int_delay)
2685                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2686
2687         /* enable Report Status bit */
2688         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2689
2690         ew32(TCTL, tctl);
2691
2692         e1000e_config_collision_dist(hw);
2693 }
2694
2695 /**
2696  * e1000_setup_rctl - configure the receive control registers
2697  * @adapter: Board private structure
2698  **/
2699 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2700                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2701 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2702 {
2703         struct e1000_hw *hw = &adapter->hw;
2704         u32 rctl, rfctl;
2705         u32 psrctl = 0;
2706         u32 pages = 0;
2707
2708         /* Program MC offset vector base */
2709         rctl = er32(RCTL);
2710         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2711         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2712                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2713                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2714
2715         /* Do not Store bad packets */
2716         rctl &= ~E1000_RCTL_SBP;
2717
2718         /* Enable Long Packet receive */
2719         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2720                 rctl &= ~E1000_RCTL_LPE;
2721         else
2722                 rctl |= E1000_RCTL_LPE;
2723
2724         /* Some systems expect that the CRC is included in SMBUS traffic. The
2725          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2726          * host memory when this is enabled
2727          */
2728         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2729                 rctl |= E1000_RCTL_SECRC;
2730
2731         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2732         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2733                 u16 phy_data;
2734
2735                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2736                 phy_data &= 0xfff8;
2737                 phy_data |= (1 << 2);
2738                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2739
2740                 e1e_rphy(hw, 22, &phy_data);
2741                 phy_data &= 0x0fff;
2742                 phy_data |= (1 << 14);
2743                 e1e_wphy(hw, 0x10, 0x2823);
2744                 e1e_wphy(hw, 0x11, 0x0003);
2745                 e1e_wphy(hw, 22, phy_data);
2746         }
2747
2748         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2749         if (hw->mac.type == e1000_pch2lan) {
2750                 s32 ret_val;
2751
2752                 if (rctl & E1000_RCTL_LPE)
2753                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2754                 else
2755                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2756         }
2757
2758         /* Setup buffer sizes */
2759         rctl &= ~E1000_RCTL_SZ_4096;
2760         rctl |= E1000_RCTL_BSEX;
2761         switch (adapter->rx_buffer_len) {
2762         case 2048:
2763         default:
2764                 rctl |= E1000_RCTL_SZ_2048;
2765                 rctl &= ~E1000_RCTL_BSEX;
2766                 break;
2767         case 4096:
2768                 rctl |= E1000_RCTL_SZ_4096;
2769                 break;
2770         case 8192:
2771                 rctl |= E1000_RCTL_SZ_8192;
2772                 break;
2773         case 16384:
2774                 rctl |= E1000_RCTL_SZ_16384;
2775                 break;
2776         }
2777
2778         /*
2779          * 82571 and greater support packet-split where the protocol
2780          * header is placed in skb->data and the packet data is
2781          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2782          * In the case of a non-split, skb->data is linearly filled,
2783          * followed by the page buffers.  Therefore, skb->data is
2784          * sized to hold the largest protocol header.
2785          *
2786          * allocations using alloc_page take too long for regular MTU
2787          * so only enable packet split for jumbo frames
2788          *
2789          * Using pages when the page size is greater than 16k wastes
2790          * a lot of memory, since we allocate 3 pages at all times
2791          * per packet.
2792          */
2793         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2794         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2795             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2796                 adapter->rx_ps_pages = pages;
2797         else
2798                 adapter->rx_ps_pages = 0;
2799
2800         if (adapter->rx_ps_pages) {
2801                 /* Configure extra packet-split registers */
2802                 rfctl = er32(RFCTL);
2803                 rfctl |= E1000_RFCTL_EXTEN;
2804                 /*
2805                  * disable packet split support for IPv6 extension headers,
2806                  * because some malformed IPv6 headers can hang the Rx
2807                  */
2808                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2809                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2810
2811                 ew32(RFCTL, rfctl);
2812
2813                 /* Enable Packet split descriptors */
2814                 rctl |= E1000_RCTL_DTYP_PS;
2815
2816                 psrctl |= adapter->rx_ps_bsize0 >>
2817                         E1000_PSRCTL_BSIZE0_SHIFT;
2818
2819                 switch (adapter->rx_ps_pages) {
2820                 case 3:
2821                         psrctl |= PAGE_SIZE <<
2822                                 E1000_PSRCTL_BSIZE3_SHIFT;
2823                 case 2:
2824                         psrctl |= PAGE_SIZE <<
2825                                 E1000_PSRCTL_BSIZE2_SHIFT;
2826                 case 1:
2827                         psrctl |= PAGE_SIZE >>
2828                                 E1000_PSRCTL_BSIZE1_SHIFT;
2829                         break;
2830                 }
2831
2832                 ew32(PSRCTL, psrctl);
2833         }
2834
2835         ew32(RCTL, rctl);
2836         /* just started the receive unit, no need to restart */
2837         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2838 }
2839
2840 /**
2841  * e1000_configure_rx - Configure Receive Unit after Reset
2842  * @adapter: board private structure
2843  *
2844  * Configure the Rx unit of the MAC after a reset.
2845  **/
2846 static void e1000_configure_rx(struct e1000_adapter *adapter)
2847 {
2848         struct e1000_hw *hw = &adapter->hw;
2849         struct e1000_ring *rx_ring = adapter->rx_ring;
2850         u64 rdba;
2851         u32 rdlen, rctl, rxcsum, ctrl_ext;
2852
2853         if (adapter->rx_ps_pages) {
2854                 /* this is a 32 byte descriptor */
2855                 rdlen = rx_ring->count *
2856                         sizeof(union e1000_rx_desc_packet_split);
2857                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2858                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2859         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2860                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2861                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2862                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2863         } else {
2864                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2865                 adapter->clean_rx = e1000_clean_rx_irq;
2866                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2867         }
2868
2869         /* disable receives while setting up the descriptors */
2870         rctl = er32(RCTL);
2871         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2872         e1e_flush();
2873         msleep(10);
2874
2875         /* set the Receive Delay Timer Register */
2876         ew32(RDTR, adapter->rx_int_delay);
2877
2878         /* irq moderation */
2879         ew32(RADV, adapter->rx_abs_int_delay);
2880         if (adapter->itr_setting != 0)
2881                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2882
2883         ctrl_ext = er32(CTRL_EXT);
2884         /* Auto-Mask interrupts upon ICR access */
2885         ctrl_ext |= E1000_CTRL_EXT_IAME;
2886         ew32(IAM, 0xffffffff);
2887         ew32(CTRL_EXT, ctrl_ext);
2888         e1e_flush();
2889
2890         /*
2891          * Setup the HW Rx Head and Tail Descriptor Pointers and
2892          * the Base and Length of the Rx Descriptor Ring
2893          */
2894         rdba = rx_ring->dma;
2895         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2896         ew32(RDBAH, (rdba >> 32));
2897         ew32(RDLEN, rdlen);
2898         ew32(RDH, 0);
2899         ew32(RDT, 0);
2900         rx_ring->head = E1000_RDH;
2901         rx_ring->tail = E1000_RDT;
2902
2903         /* Enable Receive Checksum Offload for TCP and UDP */
2904         rxcsum = er32(RXCSUM);
2905         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2906                 rxcsum |= E1000_RXCSUM_TUOFL;
2907
2908                 /*
2909                  * IPv4 payload checksum for UDP fragments must be
2910                  * used in conjunction with packet-split.
2911                  */
2912                 if (adapter->rx_ps_pages)
2913                         rxcsum |= E1000_RXCSUM_IPPCSE;
2914         } else {
2915                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2916                 /* no need to clear IPPCSE as it defaults to 0 */
2917         }
2918         ew32(RXCSUM, rxcsum);
2919
2920         /*
2921          * Enable early receives on supported devices, only takes effect when
2922          * packet size is equal or larger than the specified value (in 8 byte
2923          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2924          */
2925         if (adapter->flags & FLAG_HAS_ERT) {
2926                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2927                         u32 rxdctl = er32(RXDCTL(0));
2928                         ew32(RXDCTL(0), rxdctl | 0x3);
2929                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2930                         /*
2931                          * With jumbo frames and early-receive enabled,
2932                          * excessive C-state transition latencies result in
2933                          * dropped transactions.
2934                          */
2935                         pm_qos_update_request(
2936                                 &adapter->netdev->pm_qos_req, 55);
2937                 } else {
2938                         pm_qos_update_request(
2939                                 &adapter->netdev->pm_qos_req,
2940                                 PM_QOS_DEFAULT_VALUE);
2941                 }
2942         }
2943
2944         /* Enable Receives */
2945         ew32(RCTL, rctl);
2946 }
2947
2948 /**
2949  *  e1000_update_mc_addr_list - Update Multicast addresses
2950  *  @hw: pointer to the HW structure
2951  *  @mc_addr_list: array of multicast addresses to program
2952  *  @mc_addr_count: number of multicast addresses to program
2953  *
2954  *  Updates the Multicast Table Array.
2955  *  The caller must have a packed mc_addr_list of multicast addresses.
2956  **/
2957 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2958                                       u32 mc_addr_count)
2959 {
2960         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2961 }
2962
2963 /**
2964  * e1000_set_multi - Multicast and Promiscuous mode set
2965  * @netdev: network interface device structure
2966  *
2967  * The set_multi entry point is called whenever the multicast address
2968  * list or the network interface flags are updated.  This routine is
2969  * responsible for configuring the hardware for proper multicast,
2970  * promiscuous mode, and all-multi behavior.
2971  **/
2972 static void e1000_set_multi(struct net_device *netdev)
2973 {
2974         struct e1000_adapter *adapter = netdev_priv(netdev);
2975         struct e1000_hw *hw = &adapter->hw;
2976         struct netdev_hw_addr *ha;
2977         u8  *mta_list;
2978         u32 rctl;
2979         int i;
2980
2981         /* Check for Promiscuous and All Multicast modes */
2982
2983         rctl = er32(RCTL);
2984
2985         if (netdev->flags & IFF_PROMISC) {
2986                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2987                 rctl &= ~E1000_RCTL_VFE;
2988         } else {
2989                 if (netdev->flags & IFF_ALLMULTI) {
2990                         rctl |= E1000_RCTL_MPE;
2991                         rctl &= ~E1000_RCTL_UPE;
2992                 } else {
2993                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2994                 }
2995                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2996                         rctl |= E1000_RCTL_VFE;
2997         }
2998
2999         ew32(RCTL, rctl);
3000
3001         if (!netdev_mc_empty(netdev)) {
3002                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3003                 if (!mta_list)
3004                         return;
3005
3006                 /* prepare a packed array of only addresses. */
3007                 i = 0;
3008                 netdev_for_each_mc_addr(ha, netdev)
3009                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3010
3011                 e1000_update_mc_addr_list(hw, mta_list, i);
3012                 kfree(mta_list);
3013         } else {
3014                 /*
3015                  * if we're called from probe, we might not have
3016                  * anything to do here, so clear out the list
3017                  */
3018                 e1000_update_mc_addr_list(hw, NULL, 0);
3019         }
3020 }
3021
3022 /**
3023  * e1000_configure - configure the hardware for Rx and Tx
3024  * @adapter: private board structure
3025  **/
3026 static void e1000_configure(struct e1000_adapter *adapter)
3027 {
3028         e1000_set_multi(adapter->netdev);
3029
3030         e1000_restore_vlan(adapter);
3031         e1000_init_manageability_pt(adapter);
3032
3033         e1000_configure_tx(adapter);
3034         e1000_setup_rctl(adapter);
3035         e1000_configure_rx(adapter);
3036         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3037 }
3038
3039 /**
3040  * e1000e_power_up_phy - restore link in case the phy was powered down
3041  * @adapter: address of board private structure
3042  *
3043  * The phy may be powered down to save power and turn off link when the
3044  * driver is unloaded and wake on lan is not enabled (among others)
3045  * *** this routine MUST be followed by a call to e1000e_reset ***
3046  **/
3047 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3048 {
3049         if (adapter->hw.phy.ops.power_up)
3050                 adapter->hw.phy.ops.power_up(&adapter->hw);
3051
3052         adapter->hw.mac.ops.setup_link(&adapter->hw);
3053 }
3054
3055 /**
3056  * e1000_power_down_phy - Power down the PHY
3057  *
3058  * Power down the PHY so no link is implied when interface is down.
3059  * The PHY cannot be powered down if management or WoL is active.
3060  */
3061 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3062 {
3063         /* WoL is enabled */
3064         if (adapter->wol)
3065                 return;
3066
3067         if (adapter->hw.phy.ops.power_down)
3068                 adapter->hw.phy.ops.power_down(&adapter->hw);
3069 }
3070
3071 /**
3072  * e1000e_reset - bring the hardware into a known good state
3073  *
3074  * This function boots the hardware and enables some settings that
3075  * require a configuration cycle of the hardware - those cannot be
3076  * set/changed during runtime. After reset the device needs to be
3077  * properly configured for Rx, Tx etc.
3078  */
3079 void e1000e_reset(struct e1000_adapter *adapter)
3080 {
3081         struct e1000_mac_info *mac = &adapter->hw.mac;
3082         struct e1000_fc_info *fc = &adapter->hw.fc;
3083         struct e1000_hw *hw = &adapter->hw;
3084         u32 tx_space, min_tx_space, min_rx_space;
3085         u32 pba = adapter->pba;
3086         u16 hwm;
3087
3088         /* reset Packet Buffer Allocation to default */
3089         ew32(PBA, pba);
3090
3091         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3092                 /*
3093                  * To maintain wire speed transmits, the Tx FIFO should be
3094                  * large enough to accommodate two full transmit packets,
3095                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3096                  * the Rx FIFO should be large enough to accommodate at least
3097                  * one full receive packet and is similarly rounded up and
3098                  * expressed in KB.
3099                  */
3100                 pba = er32(PBA);
3101                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3102                 tx_space = pba >> 16;
3103                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3104                 pba &= 0xffff;
3105                 /*
3106                  * the Tx fifo also stores 16 bytes of information about the tx
3107                  * but don't include ethernet FCS because hardware appends it
3108                  */
3109                 min_tx_space = (adapter->max_frame_size +
3110                                 sizeof(struct e1000_tx_desc) -
3111                                 ETH_FCS_LEN) * 2;
3112                 min_tx_space = ALIGN(min_tx_space, 1024);
3113                 min_tx_space >>= 10;
3114                 /* software strips receive CRC, so leave room for it */
3115                 min_rx_space = adapter->max_frame_size;
3116                 min_rx_space = ALIGN(min_rx_space, 1024);
3117                 min_rx_space >>= 10;
3118
3119                 /*
3120                  * If current Tx allocation is less than the min Tx FIFO size,
3121                  * and the min Tx FIFO size is less than the current Rx FIFO
3122                  * allocation, take space away from current Rx allocation
3123                  */
3124                 if ((tx_space < min_tx_space) &&
3125                     ((min_tx_space - tx_space) < pba)) {
3126                         pba -= min_tx_space - tx_space;
3127
3128                         /*
3129                          * if short on Rx space, Rx wins and must trump tx
3130                          * adjustment or use Early Receive if available
3131                          */
3132                         if ((pba < min_rx_space) &&
3133                             (!(adapter->flags & FLAG_HAS_ERT)))
3134                                 /* ERT enabled in e1000_configure_rx */
3135                                 pba = min_rx_space;
3136                 }
3137
3138                 ew32(PBA, pba);
3139         }
3140
3141
3142         /*
3143          * flow control settings
3144          *
3145          * The high water mark must be low enough to fit one full frame
3146          * (or the size used for early receive) above it in the Rx FIFO.
3147          * Set it to the lower of:
3148          * - 90% of the Rx FIFO size, and
3149          * - the full Rx FIFO size minus the early receive size (for parts
3150          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3151          * - the full Rx FIFO size minus one full frame
3152          */
3153         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3154                 fc->pause_time = 0xFFFF;
3155         else
3156                 fc->pause_time = E1000_FC_PAUSE_TIME;
3157         fc->send_xon = 1;
3158         fc->current_mode = fc->requested_mode;
3159
3160         switch (hw->mac.type) {
3161         default:
3162                 if ((adapter->flags & FLAG_HAS_ERT) &&
3163                     (adapter->netdev->mtu > ETH_DATA_LEN))
3164                         hwm = min(((pba << 10) * 9 / 10),
3165                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3166                 else
3167                         hwm = min(((pba << 10) * 9 / 10),
3168                                   ((pba << 10) - adapter->max_frame_size));
3169
3170                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3171                 fc->low_water = fc->high_water - 8;
3172                 break;
3173         case e1000_pchlan:
3174                 /*
3175                  * Workaround PCH LOM adapter hangs with certain network
3176                  * loads.  If hangs persist, try disabling Tx flow control.
3177                  */
3178                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3179                         fc->high_water = 0x3500;
3180                         fc->low_water  = 0x1500;
3181                 } else {
3182                         fc->high_water = 0x5000;
3183                         fc->low_water  = 0x3000;
3184                 }
3185                 fc->refresh_time = 0x1000;
3186                 break;
3187         case e1000_pch2lan:
3188                 fc->high_water = 0x05C20;
3189                 fc->low_water = 0x05048;
3190                 fc->pause_time = 0x0650;
3191                 fc->refresh_time = 0x0400;
3192                 break;
3193         }
3194
3195         /* Allow time for pending master requests to run */
3196         mac->ops.reset_hw(hw);
3197
3198         /*
3199          * For parts with AMT enabled, let the firmware know
3200          * that the network interface is in control
3201          */
3202         if (adapter->flags & FLAG_HAS_AMT)
3203                 e1000_get_hw_control(adapter);
3204
3205         ew32(WUC, 0);
3206
3207         if (mac->ops.init_hw(hw))
3208                 e_err("Hardware Error\n");
3209
3210         e1000_update_mng_vlan(adapter);
3211
3212         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3213         ew32(VET, ETH_P_8021Q);
3214
3215         e1000e_reset_adaptive(hw);
3216         e1000_get_phy_info(hw);
3217
3218         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3219             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3220                 u16 phy_data = 0;
3221                 /*
3222                  * speed up time to link by disabling smart power down, ignore
3223                  * the return value of this function because there is nothing
3224                  * different we would do if it failed
3225                  */
3226                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3227                 phy_data &= ~IGP02E1000_PM_SPD;
3228                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3229         }
3230 }
3231
3232 int e1000e_up(struct e1000_adapter *adapter)
3233 {
3234         struct e1000_hw *hw = &adapter->hw;
3235
3236         /* hardware has been reset, we need to reload some things */
3237         e1000_configure(adapter);
3238
3239         clear_bit(__E1000_DOWN, &adapter->state);
3240
3241         napi_enable(&adapter->napi);
3242         if (adapter->msix_entries)
3243                 e1000_configure_msix(adapter);
3244         e1000_irq_enable(adapter);
3245
3246         netif_wake_queue(adapter->netdev);
3247
3248         /* fire a link change interrupt to start the watchdog */
3249         if (adapter->msix_entries)
3250                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3251         else
3252                 ew32(ICS, E1000_ICS_LSC);
3253
3254         return 0;
3255 }
3256
3257 void e1000e_down(struct e1000_adapter *adapter)
3258 {
3259         struct net_device *netdev = adapter->netdev;
3260         struct e1000_hw *hw = &adapter->hw;
3261         u32 tctl, rctl;
3262
3263         /*
3264          * signal that we're down so the interrupt handler does not
3265          * reschedule our watchdog timer
3266          */
3267         set_bit(__E1000_DOWN, &adapter->state);
3268
3269         /* disable receives in the hardware */
3270         rctl = er32(RCTL);
3271         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3272         /* flush and sleep below */
3273
3274         netif_stop_queue(netdev);
3275
3276         /* disable transmits in the hardware */
3277         tctl = er32(TCTL);
3278         tctl &= ~E1000_TCTL_EN;
3279         ew32(TCTL, tctl);
3280         /* flush both disables and wait for them to finish */
3281         e1e_flush();
3282         msleep(10);
3283
3284         napi_disable(&adapter->napi);
3285         e1000_irq_disable(adapter);
3286
3287         del_timer_sync(&adapter->watchdog_timer);
3288         del_timer_sync(&adapter->phy_info_timer);
3289
3290         netif_carrier_off(netdev);
3291         adapter->link_speed = 0;
3292         adapter->link_duplex = 0;
3293
3294         if (!pci_channel_offline(adapter->pdev))
3295                 e1000e_reset(adapter);
3296         e1000_clean_tx_ring(adapter);
3297         e1000_clean_rx_ring(adapter);
3298
3299         /*
3300          * TODO: for power management, we could drop the link and
3301          * pci_disable_device here.
3302          */
3303 }
3304
3305 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3306 {
3307         might_sleep();
3308         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3309                 msleep(1);
3310         e1000e_down(adapter);
3311         e1000e_up(adapter);
3312         clear_bit(__E1000_RESETTING, &adapter->state);
3313 }
3314
3315 /**
3316  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3317  * @adapter: board private structure to initialize
3318  *
3319  * e1000_sw_init initializes the Adapter private data structure.
3320  * Fields are initialized based on PCI device information and
3321  * OS network device settings (MTU size).
3322  **/
3323 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3324 {
3325         struct net_device *netdev = adapter->netdev;
3326
3327         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3328         adapter->rx_ps_bsize0 = 128;
3329         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3330         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3331
3332         e1000e_set_interrupt_capability(adapter);
3333
3334         if (e1000_alloc_queues(adapter))
3335                 return -ENOMEM;
3336
3337         /* Explicitly disable IRQ since the NIC can be in any state. */
3338         e1000_irq_disable(adapter);
3339
3340         set_bit(__E1000_DOWN, &adapter->state);
3341         return 0;
3342 }
3343
3344 /**
3345  * e1000_intr_msi_test - Interrupt Handler
3346  * @irq: interrupt number
3347  * @data: pointer to a network interface device structure
3348  **/
3349 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3350 {
3351         struct net_device *netdev = data;
3352         struct e1000_adapter *adapter = netdev_priv(netdev);
3353         struct e1000_hw *hw = &adapter->hw;
3354         u32 icr = er32(ICR);
3355
3356         e_dbg("icr is %08X\n", icr);
3357         if (icr & E1000_ICR_RXSEQ) {
3358                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3359                 wmb();
3360         }
3361
3362         return IRQ_HANDLED;
3363 }
3364
3365 /**
3366  * e1000_test_msi_interrupt - Returns 0 for successful test
3367  * @adapter: board private struct
3368  *
3369  * code flow taken from tg3.c
3370  **/
3371 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3372 {
3373         struct net_device *netdev = adapter->netdev;
3374         struct e1000_hw *hw = &adapter->hw;
3375         int err;
3376
3377         /* poll_enable hasn't been called yet, so don't need disable */
3378         /* clear any pending events */
3379         er32(ICR);
3380
3381         /* free the real vector and request a test handler */
3382         e1000_free_irq(adapter);
3383         e1000e_reset_interrupt_capability(adapter);
3384
3385         /* Assume that the test fails, if it succeeds then the test
3386          * MSI irq handler will unset this flag */
3387         adapter->flags |= FLAG_MSI_TEST_FAILED;
3388
3389         err = pci_enable_msi(adapter->pdev);
3390         if (err)
3391                 goto msi_test_failed;
3392
3393         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3394                           netdev->name, netdev);
3395         if (err) {
3396                 pci_disable_msi(adapter->pdev);
3397                 goto msi_test_failed;
3398         }
3399
3400         wmb();
3401
3402         e1000_irq_enable(adapter);
3403
3404         /* fire an unusual interrupt on the test handler */
3405         ew32(ICS, E1000_ICS_RXSEQ);
3406         e1e_flush();
3407         msleep(50);
3408
3409         e1000_irq_disable(adapter);
3410
3411         rmb();
3412
3413         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3414                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3415                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3416         } else
3417                 e_dbg("MSI interrupt test succeeded!\n");
3418
3419         free_irq(adapter->pdev->irq, netdev);
3420         pci_disable_msi(adapter->pdev);
3421
3422 msi_test_failed:
3423         e1000e_set_interrupt_capability(adapter);
3424         return e1000_request_irq(adapter);
3425 }
3426
3427 /**
3428  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3429  * @adapter: board private struct
3430  *
3431  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3432  **/
3433 static int e1000_test_msi(struct e1000_adapter *adapter)
3434 {
3435         int err;
3436         u16 pci_cmd;
3437
3438         if (!(adapter->flags & FLAG_MSI_ENABLED))
3439                 return 0;
3440
3441         /* disable SERR in case the MSI write causes a master abort */
3442         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3443         if (pci_cmd & PCI_COMMAND_SERR)
3444                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3445                                       pci_cmd & ~PCI_COMMAND_SERR);
3446
3447         err = e1000_test_msi_interrupt(adapter);
3448
3449         /* re-enable SERR */
3450         if (pci_cmd & PCI_COMMAND_SERR) {
3451                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3452                 pci_cmd |= PCI_COMMAND_SERR;
3453                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3454         }
3455
3456         return err;
3457 }
3458
3459 /**
3460  * e1000_open - Called when a network interface is made active
3461  * @netdev: network interface device structure
3462  *
3463  * Returns 0 on success, negative value on failure
3464  *
3465  * The open entry point is called when a network interface is made
3466  * active by the system (IFF_UP).  At this point all resources needed
3467  * for transmit and receive operations are allocated, the interrupt
3468  * handler is registered with the OS, the watchdog timer is started,
3469  * and the stack is notified that the interface is ready.
3470  **/
3471 static int e1000_open(struct net_device *netdev)
3472 {
3473         struct e1000_adapter *adapter = netdev_priv(netdev);
3474         struct e1000_hw *hw = &adapter->hw;
3475         struct pci_dev *pdev = adapter->pdev;
3476         int err;
3477
3478         /* disallow open during test */
3479         if (test_bit(__E1000_TESTING, &adapter->state))
3480                 return -EBUSY;
3481
3482         pm_runtime_get_sync(&pdev->dev);
3483
3484         netif_carrier_off(netdev);
3485
3486         /* allocate transmit descriptors */
3487         err = e1000e_setup_tx_resources(adapter);
3488         if (err)
3489                 goto err_setup_tx;
3490
3491         /* allocate receive descriptors */
3492         err = e1000e_setup_rx_resources(adapter);
3493         if (err)
3494                 goto err_setup_rx;
3495
3496         /*
3497          * If AMT is enabled, let the firmware know that the network
3498          * interface is now open and reset the part to a known state.
3499          */
3500         if (adapter->flags & FLAG_HAS_AMT) {
3501                 e1000_get_hw_control(adapter);
3502                 e1000e_reset(adapter);
3503         }
3504
3505         e1000e_power_up_phy(adapter);
3506
3507         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3508         if ((adapter->hw.mng_cookie.status &
3509              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3510                 e1000_update_mng_vlan(adapter);
3511
3512         /* DMA latency requirement to workaround early-receive/jumbo issue */
3513         if (adapter->flags & FLAG_HAS_ERT)
3514                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3515                                    PM_QOS_CPU_DMA_LATENCY,
3516                                    PM_QOS_DEFAULT_VALUE);
3517
3518         /*
3519          * before we allocate an interrupt, we must be ready to handle it.
3520          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3521          * as soon as we call pci_request_irq, so we have to setup our
3522          * clean_rx handler before we do so.
3523          */
3524         e1000_configure(adapter);
3525
3526         err = e1000_request_irq(adapter);
3527         if (err)
3528                 goto err_req_irq;
3529
3530         /*
3531          * Work around PCIe errata with MSI interrupts causing some chipsets to
3532          * ignore e1000e MSI messages, which means we need to test our MSI
3533          * interrupt now
3534          */
3535         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3536                 err = e1000_test_msi(adapter);
3537                 if (err) {
3538                         e_err("Interrupt allocation failed\n");
3539                         goto err_req_irq;
3540                 }
3541         }
3542
3543         /* From here on the code is the same as e1000e_up() */
3544         clear_bit(__E1000_DOWN, &adapter->state);
3545
3546         napi_enable(&adapter->napi);
3547
3548         e1000_irq_enable(adapter);
3549
3550         netif_start_queue(netdev);
3551
3552         adapter->idle_check = true;
3553         pm_runtime_put(&pdev->dev);
3554
3555         /* fire a link status change interrupt to start the watchdog */
3556         if (adapter->msix_entries)
3557                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3558         else
3559                 ew32(ICS, E1000_ICS_LSC);
3560
3561         return 0;
3562
3563 err_req_irq:
3564         e1000_release_hw_control(adapter);
3565         e1000_power_down_phy(adapter);
3566         e1000e_free_rx_resources(adapter);
3567 err_setup_rx:
3568         e1000e_free_tx_resources(adapter);
3569 err_setup_tx:
3570         e1000e_reset(adapter);
3571         pm_runtime_put_sync(&pdev->dev);
3572
3573         return err;
3574 }
3575
3576 /**
3577  * e1000_close - Disables a network interface
3578  * @netdev: network interface device structure
3579  *
3580  * Returns 0, this is not allowed to fail
3581  *
3582  * The close entry point is called when an interface is de-activated
3583  * by the OS.  The hardware is still under the drivers control, but
3584  * needs to be disabled.  A global MAC reset is issued to stop the
3585  * hardware, and all transmit and receive resources are freed.
3586  **/
3587 static int e1000_close(struct net_device *netdev)
3588 {
3589         struct e1000_adapter *adapter = netdev_priv(netdev);
3590         struct pci_dev *pdev = adapter->pdev;
3591
3592         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3593
3594         pm_runtime_get_sync(&pdev->dev);
3595
3596         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3597                 e1000e_down(adapter);
3598                 e1000_free_irq(adapter);
3599         }
3600         e1000_power_down_phy(adapter);
3601
3602         e1000e_free_tx_resources(adapter);
3603         e1000e_free_rx_resources(adapter);
3604
3605         /*
3606          * kill manageability vlan ID if supported, but not if a vlan with
3607          * the same ID is registered on the host OS (let 8021q kill it)
3608          */
3609         if ((adapter->hw.mng_cookie.status &
3610                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3611              !(adapter->vlgrp &&
3612                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3613                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3614
3615         /*
3616          * If AMT is enabled, let the firmware know that the network
3617          * interface is now closed
3618          */
3619         if (adapter->flags & FLAG_HAS_AMT)
3620                 e1000_release_hw_control(adapter);
3621
3622         if (adapter->flags & FLAG_HAS_ERT)
3623                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3624
3625         pm_runtime_put_sync(&pdev->dev);
3626
3627         return 0;
3628 }
3629 /**
3630  * e1000_set_mac - Change the Ethernet Address of the NIC
3631  * @netdev: network interface device structure
3632  * @p: pointer to an address structure
3633  *
3634  * Returns 0 on success, negative on failure
3635  **/
3636 static int e1000_set_mac(struct net_device *netdev, void *p)
3637 {
3638         struct e1000_adapter *adapter = netdev_priv(netdev);
3639         struct sockaddr *addr = p;
3640
3641         if (!is_valid_ether_addr(addr->sa_data))
3642                 return -EADDRNOTAVAIL;
3643
3644         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3645         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3646
3647         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3648
3649         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3650                 /* activate the work around */
3651                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3652
3653                 /*
3654                  * Hold a copy of the LAA in RAR[14] This is done so that
3655                  * between the time RAR[0] gets clobbered  and the time it
3656                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3657                  * of the RARs and no incoming packets directed to this port
3658                  * are dropped. Eventually the LAA will be in RAR[0] and
3659                  * RAR[14]
3660                  */
3661                 e1000e_rar_set(&adapter->hw,
3662                               adapter->hw.mac.addr,
3663                               adapter->hw.mac.rar_entry_count - 1);
3664         }
3665
3666         return 0;
3667 }
3668
3669 /**
3670  * e1000e_update_phy_task - work thread to update phy
3671  * @work: pointer to our work struct
3672  *
3673  * this worker thread exists because we must acquire a
3674  * semaphore to read the phy, which we could msleep while
3675  * waiting for it, and we can't msleep in a timer.
3676  **/
3677 static void e1000e_update_phy_task(struct work_struct *work)
3678 {
3679         struct e1000_adapter *adapter = container_of(work,
3680                                         struct e1000_adapter, update_phy_task);
3681         e1000_get_phy_info(&adapter->hw);
3682 }
3683
3684 /*
3685  * Need to wait a few seconds after link up to get diagnostic information from
3686  * the phy
3687  */
3688 static void e1000_update_phy_info(unsigned long data)
3689 {
3690         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3691         schedule_work(&adapter->update_phy_task);
3692 }
3693
3694 /**
3695  * e1000e_update_phy_stats - Update the PHY statistics counters
3696  * @adapter: board private structure
3697  **/
3698 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3699 {
3700         struct e1000_hw *hw = &adapter->hw;
3701         s32 ret_val;
3702         u16 phy_data;
3703
3704         ret_val = hw->phy.ops.acquire(hw);
3705         if (ret_val)
3706                 return;
3707
3708         hw->phy.addr = 1;
3709
3710 #define HV_PHY_STATS_PAGE       778
3711         /*
3712          * A page set is expensive so check if already on desired page.
3713          * If not, set to the page with the PHY status registers.
3714          */
3715         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3716                                            &phy_data);
3717         if (ret_val)
3718                 goto release;
3719         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3720                 ret_val = e1000e_write_phy_reg_mdic(hw,
3721                                                     IGP01E1000_PHY_PAGE_SELECT,
3722                                                     (HV_PHY_STATS_PAGE <<
3723                                                      IGP_PAGE_SHIFT));
3724                 if (ret_val)
3725                         goto release;
3726         }
3727
3728         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3729
3730         /* Single Collision Count */
3731         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3732                                  &phy_data);
3733         ret_val = e1000e_read_phy_reg_mdic(hw,
3734                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3735                                            &phy_data);
3736         if (!ret_val)
3737                 adapter->stats.scc += phy_data;
3738
3739         /* Excessive Collision Count */
3740         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3741                                  &phy_data);
3742         ret_val = e1000e_read_phy_reg_mdic(hw,
3743                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3744                                            &phy_data);
3745         if (!ret_val)
3746                 adapter->stats.ecol += phy_data;
3747
3748         /* Multiple Collision Count */
3749         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3750                                  &phy_data);
3751         ret_val = e1000e_read_phy_reg_mdic(hw,
3752                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3753                                            &phy_data);
3754         if (!ret_val)
3755                 adapter->stats.mcc += phy_data;
3756
3757         /* Late Collision Count */
3758         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3759                                  &phy_data);
3760         ret_val = e1000e_read_phy_reg_mdic(hw,
3761                                            HV_LATECOL_LOWER &
3762                                            MAX_PHY_REG_ADDRESS,
3763                                            &phy_data);
3764         if (!ret_val)
3765                 adapter->stats.latecol += phy_data;
3766
3767         /* Collision Count - also used for adaptive IFS */
3768         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3769                                  &phy_data);
3770         ret_val = e1000e_read_phy_reg_mdic(hw,
3771                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3772                                            &phy_data);
3773         if (!ret_val)
3774                 hw->mac.collision_delta = phy_data;
3775
3776         /* Defer Count */
3777         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3778                                  &phy_data);
3779         ret_val = e1000e_read_phy_reg_mdic(hw,
3780                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3781                                            &phy_data);
3782         if (!ret_val)
3783                 adapter->stats.dc += phy_data;
3784
3785         /* Transmit with no CRS */
3786         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3787                                  &phy_data);
3788         ret_val = e1000e_read_phy_reg_mdic(hw,
3789                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3790                                            &phy_data);
3791         if (!ret_val)
3792                 adapter->stats.tncrs += phy_data;
3793
3794 release:
3795         hw->phy.ops.release(hw);
3796 }
3797
3798 /**
3799  * e1000e_update_stats - Update the board statistics counters
3800  * @adapter: board private structure
3801  **/
3802 void e1000e_update_stats(struct e1000_adapter *adapter)
3803 {
3804         struct net_device *netdev = adapter->netdev;
3805         struct e1000_hw *hw = &adapter->hw;
3806         struct pci_dev *pdev = adapter->pdev;
3807
3808         /*
3809          * Prevent stats update while adapter is being reset, or if the pci
3810          * connection is down.
3811          */
3812         if (adapter->link_speed == 0)
3813                 return;
3814         if (pci_channel_offline(pdev))
3815                 return;
3816
3817         adapter->stats.crcerrs += er32(CRCERRS);
3818         adapter->stats.gprc += er32(GPRC);
3819         adapter->stats.gorc += er32(GORCL);
3820         er32(GORCH); /* Clear gorc */
3821         adapter->stats.bprc += er32(BPRC);
3822         adapter->stats.mprc += er32(MPRC);
3823         adapter->stats.roc += er32(ROC);
3824
3825         adapter->stats.mpc += er32(MPC);
3826
3827         /* Half-duplex statistics */
3828         if (adapter->link_duplex == HALF_DUPLEX) {
3829                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3830                         e1000e_update_phy_stats(adapter);
3831                 } else {
3832                         adapter->stats.scc += er32(SCC);
3833                         adapter->stats.ecol += er32(ECOL);
3834                         adapter->stats.mcc += er32(MCC);
3835                         adapter->stats.latecol += er32(LATECOL);
3836                         adapter->stats.dc += er32(DC);
3837
3838                         hw->mac.collision_delta = er32(COLC);
3839
3840                         if ((hw->mac.type != e1000_82574) &&
3841                             (hw->mac.type != e1000_82583))
3842                                 adapter->stats.tncrs += er32(TNCRS);
3843                 }
3844                 adapter->stats.colc += hw->mac.collision_delta;
3845         }
3846
3847         adapter->stats.xonrxc += er32(XONRXC);
3848         adapter->stats.xontxc += er32(XONTXC);
3849         adapter->stats.xoffrxc += er32(XOFFRXC);
3850         adapter->stats.xofftxc += er32(XOFFTXC);
3851         adapter->stats.gptc += er32(GPTC);
3852         adapter->stats.gotc += er32(GOTCL);
3853         er32(GOTCH); /* Clear gotc */
3854         adapter->stats.rnbc += er32(RNBC);
3855         adapter->stats.ruc += er32(RUC);
3856
3857         adapter->stats.mptc += er32(MPTC);
3858         adapter->stats.bptc += er32(BPTC);
3859
3860         /* used for adaptive IFS */
3861
3862         hw->mac.tx_packet_delta = er32(TPT);
3863         adapter->stats.tpt += hw->mac.tx_packet_delta;
3864
3865         adapter->stats.algnerrc += er32(ALGNERRC);
3866         adapter->stats.rxerrc += er32(RXERRC);
3867         adapter->stats.cexterr += er32(CEXTERR);
3868         adapter->stats.tsctc += er32(TSCTC);
3869         adapter->stats.tsctfc += er32(TSCTFC);
3870
3871         /* Fill out the OS statistics structure */
3872         netdev->stats.multicast = adapter->stats.mprc;
3873         netdev->stats.collisions = adapter->stats.colc;
3874
3875         /* Rx Errors */
3876
3877         /*
3878          * RLEC on some newer hardware can be incorrect so build
3879          * our own version based on RUC and ROC
3880          */
3881         netdev->stats.rx_errors = adapter->stats.rxerrc +
3882                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3883                 adapter->stats.ruc + adapter->stats.roc +
3884                 adapter->stats.cexterr;
3885         netdev->stats.rx_length_errors = adapter->stats.ruc +
3886                                               adapter->stats.roc;
3887         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3888         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3889         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3890
3891         /* Tx Errors */
3892         netdev->stats.tx_errors = adapter->stats.ecol +
3893                                        adapter->stats.latecol;
3894         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3895         netdev->stats.tx_window_errors = adapter->stats.latecol;
3896         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3897
3898         /* Tx Dropped needs to be maintained elsewhere */
3899
3900         /* Management Stats */
3901         adapter->stats.mgptc += er32(MGTPTC);
3902         adapter->stats.mgprc += er32(MGTPRC);
3903         adapter->stats.mgpdc += er32(MGTPDC);
3904 }
3905
3906 /**
3907  * e1000_phy_read_status - Update the PHY register status snapshot
3908  * @adapter: board private structure
3909  **/
3910 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3911 {
3912         struct e1000_hw *hw = &adapter->hw;
3913         struct e1000_phy_regs *phy = &adapter->phy_regs;
3914         int ret_val;
3915
3916         if ((er32(STATUS) & E1000_STATUS_LU) &&
3917             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3918                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3919                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3920                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3921                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3922                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3923                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3924                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3925                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3926                 if (ret_val)
3927                         e_warn("Error reading PHY register\n");
3928         } else {
3929                 /*
3930                  * Do not read PHY registers if link is not up
3931                  * Set values to typical power-on defaults
3932                  */
3933                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3934                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3935                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3936                              BMSR_ERCAP);
3937                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3938                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3939                 phy->lpa = 0;
3940                 phy->expansion = EXPANSION_ENABLENPAGE;
3941                 phy->ctrl1000 = ADVERTISE_1000FULL;
3942                 phy->stat1000 = 0;
3943                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3944         }
3945 }
3946
3947 static void e1000_print_link_info(struct e1000_adapter *adapter)
3948 {
3949         struct e1000_hw *hw = &adapter->hw;
3950         u32 ctrl = er32(CTRL);
3951
3952         /* Link status message must follow this format for user tools */
3953         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3954                "Flow Control: %s\n",
3955                adapter->netdev->name,
3956                adapter->link_speed,
3957                (adapter->link_duplex == FULL_DUPLEX) ?
3958                                 "Full Duplex" : "Half Duplex",
3959                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3960                                 "RX/TX" :
3961                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3962                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3963 }
3964
3965 static bool e1000e_has_link(struct e1000_adapter *adapter)
3966 {
3967         struct e1000_hw *hw = &adapter->hw;
3968         bool link_active = 0;
3969         s32 ret_val = 0;
3970
3971         /*
3972          * get_link_status is set on LSC (link status) interrupt or
3973          * Rx sequence error interrupt.  get_link_status will stay
3974          * false until the check_for_link establishes link
3975          * for copper adapters ONLY
3976          */
3977         switch (hw->phy.media_type) {
3978         case e1000_media_type_copper:
3979                 if (hw->mac.get_link_status) {
3980                         ret_val = hw->mac.ops.check_for_link(hw);
3981                         link_active = !hw->mac.get_link_status;
3982                 } else {
3983                         link_active = 1;
3984                 }
3985                 break;
3986         case e1000_media_type_fiber:
3987                 ret_val = hw->mac.ops.check_for_link(hw);
3988                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
3989                 break;
3990         case e1000_media_type_internal_serdes:
3991                 ret_val = hw->mac.ops.check_for_link(hw);
3992                 link_active = adapter->hw.mac.serdes_has_link;
3993                 break;
3994         default:
3995         case e1000_media_type_unknown:
3996                 break;
3997         }
3998
3999         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4000             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4001                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4002                 e_info("Gigabit has been disabled, downgrading speed\n");
4003         }
4004
4005         return link_active;
4006 }
4007
4008 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4009 {
4010         /* make sure the receive unit is started */
4011         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4012             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4013                 struct e1000_hw *hw = &adapter->hw;
4014                 u32 rctl = er32(RCTL);
4015                 ew32(RCTL, rctl | E1000_RCTL_EN);
4016                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4017         }
4018 }
4019
4020 /**
4021  * e1000_watchdog - Timer Call-back
4022  * @data: pointer to adapter cast into an unsigned long
4023  **/
4024 static void e1000_watchdog(unsigned long data)
4025 {
4026         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4027
4028         /* Do the rest outside of interrupt context */
4029         schedule_work(&adapter->watchdog_task);
4030
4031         /* TODO: make this use queue_delayed_work() */
4032 }
4033
4034 static void e1000_watchdog_task(struct work_struct *work)
4035 {
4036         struct e1000_adapter *adapter = container_of(work,
4037                                         struct e1000_adapter, watchdog_task);
4038         struct net_device *netdev = adapter->netdev;
4039         struct e1000_mac_info *mac = &adapter->hw.mac;
4040         struct e1000_phy_info *phy = &adapter->hw.phy;
4041         struct e1000_ring *tx_ring = adapter->tx_ring;
4042         struct e1000_hw *hw = &adapter->hw;
4043         u32 link, tctl;
4044         int tx_pending = 0;
4045
4046         link = e1000e_has_link(adapter);
4047         if ((netif_carrier_ok(netdev)) && link) {
4048                 /* Cancel scheduled suspend requests. */
4049                 pm_runtime_resume(netdev->dev.parent);
4050
4051                 e1000e_enable_receives(adapter);
4052                 goto link_up;
4053         }
4054
4055         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4056             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4057                 e1000_update_mng_vlan(adapter);
4058
4059         if (link) {
4060                 if (!netif_carrier_ok(netdev)) {
4061                         bool txb2b = 1;
4062
4063                         /* Cancel scheduled suspend requests. */
4064                         pm_runtime_resume(netdev->dev.parent);
4065
4066                         /* update snapshot of PHY registers on LSC */
4067                         e1000_phy_read_status(adapter);
4068                         mac->ops.get_link_up_info(&adapter->hw,
4069                                                    &adapter->link_speed,
4070                                                    &adapter->link_duplex);
4071                         e1000_print_link_info(adapter);
4072                         /*
4073                          * On supported PHYs, check for duplex mismatch only
4074                          * if link has autonegotiated at 10/100 half
4075                          */
4076                         if ((hw->phy.type == e1000_phy_igp_3 ||
4077                              hw->phy.type == e1000_phy_bm) &&
4078                             (hw->mac.autoneg == true) &&
4079                             (adapter->link_speed == SPEED_10 ||
4080                              adapter->link_speed == SPEED_100) &&
4081                             (adapter->link_duplex == HALF_DUPLEX)) {
4082                                 u16 autoneg_exp;
4083
4084                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4085
4086                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4087                                         e_info("Autonegotiated half duplex but"
4088                                                " link partner cannot autoneg. "
4089                                                " Try forcing full duplex if "
4090                                                "link gets many collisions.\n");
4091                         }
4092
4093                         /* adjust timeout factor according to speed/duplex */
4094                         adapter->tx_timeout_factor = 1;
4095                         switch (adapter->link_speed) {
4096                         case SPEED_10:
4097                                 txb2b = 0;
4098                                 adapter->tx_timeout_factor = 16;
4099                                 break;
4100                         case SPEED_100:
4101                                 txb2b = 0;
4102                                 adapter->tx_timeout_factor = 10;
4103                                 break;
4104                         }
4105
4106                         /*
4107                          * workaround: re-program speed mode bit after
4108                          * link-up event
4109                          */
4110                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4111                             !txb2b) {
4112                                 u32 tarc0;
4113                                 tarc0 = er32(TARC(0));
4114                                 tarc0 &= ~SPEED_MODE_BIT;
4115                                 ew32(TARC(0), tarc0);
4116                         }
4117
4118                         /*
4119                          * disable TSO for pcie and 10/100 speeds, to avoid
4120                          * some hardware issues
4121                          */
4122                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4123                                 switch (adapter->link_speed) {
4124                                 case SPEED_10:
4125                                 case SPEED_100:
4126                                         e_info("10/100 speed: disabling TSO\n");
4127                                         netdev->features &= ~NETIF_F_TSO;
4128                                         netdev->features &= ~NETIF_F_TSO6;
4129                                         break;
4130                                 case SPEED_1000:
4131                                         netdev->features |= NETIF_F_TSO;
4132                                         netdev->features |= NETIF_F_TSO6;
4133                                         break;
4134                                 default:
4135                                         /* oops */
4136                                         break;
4137                                 }
4138                         }
4139
4140                         /*
4141                          * enable transmits in the hardware, need to do this
4142                          * after setting TARC(0)
4143                          */
4144                         tctl = er32(TCTL);
4145                         tctl |= E1000_TCTL_EN;
4146                         ew32(TCTL, tctl);
4147
4148                         /*
4149                          * Perform any post-link-up configuration before
4150                          * reporting link up.
4151                          */
4152                         if (phy->ops.cfg_on_link_up)
4153                                 phy->ops.cfg_on_link_up(hw);
4154
4155                         netif_carrier_on(netdev);
4156
4157                         if (!test_bit(__E1000_DOWN, &adapter->state))
4158                                 mod_timer(&adapter->phy_info_timer,
4159                                           round_jiffies(jiffies + 2 * HZ));
4160                 }
4161         } else {
4162                 if (netif_carrier_ok(netdev)) {
4163                         adapter->link_speed = 0;
4164                         adapter->link_duplex = 0;
4165                         /* Link status message must follow this format */
4166                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4167                                adapter->netdev->name);
4168                         netif_carrier_off(netdev);
4169                         if (!test_bit(__E1000_DOWN, &adapter->state))
4170                                 mod_timer(&adapter->phy_info_timer,
4171                                           round_jiffies(jiffies + 2 * HZ));
4172
4173                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4174                                 schedule_work(&adapter->reset_task);
4175                         else
4176                                 pm_schedule_suspend(netdev->dev.parent,
4177                                                         LINK_TIMEOUT);
4178                 }
4179         }
4180
4181 link_up:
4182         e1000e_update_stats(adapter);
4183
4184         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4185         adapter->tpt_old = adapter->stats.tpt;
4186         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4187         adapter->colc_old = adapter->stats.colc;
4188
4189         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4190         adapter->gorc_old = adapter->stats.gorc;
4191         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4192         adapter->gotc_old = adapter->stats.gotc;
4193
4194         e1000e_update_adaptive(&adapter->hw);
4195
4196         if (!netif_carrier_ok(netdev)) {
4197                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4198                                tx_ring->count);
4199                 if (tx_pending) {
4200                         /*
4201                          * We've lost link, so the controller stops DMA,
4202                          * but we've got queued Tx work that's never going
4203                          * to get done, so reset controller to flush Tx.
4204                          * (Do the reset outside of interrupt context).
4205                          */
4206                         adapter->tx_timeout_count++;
4207                         schedule_work(&adapter->reset_task);
4208                         /* return immediately since reset is imminent */
4209                         return;
4210                 }
4211         }
4212
4213         /* Simple mode for Interrupt Throttle Rate (ITR) */
4214         if (adapter->itr_setting == 4) {
4215                 /*
4216                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4217                  * Total asymmetrical Tx or Rx gets ITR=8000;
4218                  * everyone else is between 2000-8000.
4219                  */
4220                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4221                 u32 dif = (adapter->gotc > adapter->gorc ?
4222                             adapter->gotc - adapter->gorc :
4223                             adapter->gorc - adapter->gotc) / 10000;
4224                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4225
4226                 ew32(ITR, 1000000000 / (itr * 256));
4227         }
4228
4229         /* Cause software interrupt to ensure Rx ring is cleaned */
4230         if (adapter->msix_entries)
4231                 ew32(ICS, adapter->rx_ring->ims_val);
4232         else
4233                 ew32(ICS, E1000_ICS_RXDMT0);
4234
4235         /* Force detection of hung controller every watchdog period */
4236         adapter->detect_tx_hung = 1;
4237
4238         /*
4239          * With 82571 controllers, LAA may be overwritten due to controller
4240          * reset from the other port. Set the appropriate LAA in RAR[0]
4241          */
4242         if (e1000e_get_laa_state_82571(hw))
4243                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4244
4245         /* Reset the timer */
4246         if (!test_bit(__E1000_DOWN, &adapter->state))
4247                 mod_timer(&adapter->watchdog_timer,
4248                           round_jiffies(jiffies + 2 * HZ));
4249 }
4250
4251 #define E1000_TX_FLAGS_CSUM             0x00000001
4252 #define E1000_TX_FLAGS_VLAN             0x00000002
4253 #define E1000_TX_FLAGS_TSO              0x00000004
4254 #define E1000_TX_FLAGS_IPV4             0x00000008
4255 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4256 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4257
4258 static int e1000_tso(struct e1000_adapter *adapter,
4259                      struct sk_buff *skb)
4260 {
4261         struct e1000_ring *tx_ring = adapter->tx_ring;
4262         struct e1000_context_desc *context_desc;
4263         struct e1000_buffer *buffer_info;
4264         unsigned int i;
4265         u32 cmd_length = 0;
4266         u16 ipcse = 0, tucse, mss;
4267         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4268         int err;
4269
4270         if (!skb_is_gso(skb))
4271                 return 0;
4272
4273         if (skb_header_cloned(skb)) {
4274                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4275                 if (err)
4276                         return err;
4277         }
4278
4279         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4280         mss = skb_shinfo(skb)->gso_size;
4281         if (skb->protocol == htons(ETH_P_IP)) {
4282                 struct iphdr *iph = ip_hdr(skb);
4283                 iph->tot_len = 0;
4284                 iph->check = 0;
4285                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4286                                                          0, IPPROTO_TCP, 0);
4287                 cmd_length = E1000_TXD_CMD_IP;
4288                 ipcse = skb_transport_offset(skb) - 1;
4289         } else if (skb_is_gso_v6(skb)) {
4290                 ipv6_hdr(skb)->payload_len = 0;
4291                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4292                                                        &ipv6_hdr(skb)->daddr,
4293                                                        0, IPPROTO_TCP, 0);
4294                 ipcse = 0;
4295         }
4296         ipcss = skb_network_offset(skb);
4297         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4298         tucss = skb_transport_offset(skb);
4299         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4300         tucse = 0;
4301
4302         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4303                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4304
4305         i = tx_ring->next_to_use;
4306         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4307         buffer_info = &tx_ring->buffer_info[i];
4308
4309         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4310         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4311         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4312         context_desc->upper_setup.tcp_fields.tucss = tucss;
4313         context_desc->upper_setup.tcp_fields.tucso = tucso;
4314         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4315         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4316         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4317         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4318
4319         buffer_info->time_stamp = jiffies;
4320         buffer_info->next_to_watch = i;
4321
4322         i++;
4323         if (i == tx_ring->count)
4324                 i = 0;
4325         tx_ring->next_to_use = i;
4326
4327         return 1;
4328 }
4329
4330 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4331 {
4332         struct e1000_ring *tx_ring = adapter->tx_ring;
4333         struct e1000_context_desc *context_desc;
4334         struct e1000_buffer *buffer_info;
4335         unsigned int i;
4336         u8 css;
4337         u32 cmd_len = E1000_TXD_CMD_DEXT;
4338         __be16 protocol;
4339
4340         if (skb->ip_summed != CHECKSUM_PARTIAL)
4341                 return 0;
4342
4343         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4344                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4345         else
4346                 protocol = skb->protocol;
4347
4348         switch (protocol) {
4349         case cpu_to_be16(ETH_P_IP):
4350                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4351                         cmd_len |= E1000_TXD_CMD_TCP;
4352                 break;
4353         case cpu_to_be16(ETH_P_IPV6):
4354                 /* XXX not handling all IPV6 headers */
4355                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4356                         cmd_len |= E1000_TXD_CMD_TCP;
4357                 break;
4358         default:
4359                 if (unlikely(net_ratelimit()))
4360                         e_warn("checksum_partial proto=%x!\n",
4361                                be16_to_cpu(protocol));
4362                 break;
4363         }
4364
4365         css = skb_transport_offset(skb);
4366
4367         i = tx_ring->next_to_use;
4368         buffer_info = &tx_ring->buffer_info[i];
4369         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4370
4371         context_desc->lower_setup.ip_config = 0;
4372         context_desc->upper_setup.tcp_fields.tucss = css;
4373         context_desc->upper_setup.tcp_fields.tucso =
4374                                 css + skb->csum_offset;
4375         context_desc->upper_setup.tcp_fields.tucse = 0;
4376         context_desc->tcp_seg_setup.data = 0;
4377         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4378
4379         buffer_info->time_stamp = jiffies;
4380         buffer_info->next_to_watch = i;
4381
4382         i++;
4383         if (i == tx_ring->count)
4384                 i = 0;
4385         tx_ring->next_to_use = i;
4386
4387         return 1;
4388 }
4389
4390 #define E1000_MAX_PER_TXD       8192
4391 #define E1000_MAX_TXD_PWR       12
4392
4393 static int e1000_tx_map(struct e1000_adapter *adapter,
4394                         struct sk_buff *skb, unsigned int first,
4395                         unsigned int max_per_txd, unsigned int nr_frags,
4396                         unsigned int mss)
4397 {
4398         struct e1000_ring *tx_ring = adapter->tx_ring;
4399         struct pci_dev *pdev = adapter->pdev;
4400         struct e1000_buffer *buffer_info;
4401         unsigned int len = skb_headlen(skb);
4402         unsigned int offset = 0, size, count = 0, i;
4403         unsigned int f, bytecount, segs;
4404
4405         i = tx_ring->next_to_use;
4406
4407         while (len) {
4408                 buffer_info = &tx_ring->buffer_info[i];
4409                 size = min(len, max_per_txd);
4410
4411                 buffer_info->length = size;
4412                 buffer_info->time_stamp = jiffies;
4413                 buffer_info->next_to_watch = i;
4414                 buffer_info->dma = dma_map_single(&pdev->dev,
4415                                                   skb->data + offset,
4416                                                   size, DMA_TO_DEVICE);
4417                 buffer_info->mapped_as_page = false;
4418                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4419                         goto dma_error;
4420
4421                 len -= size;
4422                 offset += size;
4423                 count++;
4424
4425                 if (len) {
4426                         i++;
4427                         if (i == tx_ring->count)
4428                                 i = 0;
4429                 }
4430         }
4431
4432         for (f = 0; f < nr_frags; f++) {
4433                 struct skb_frag_struct *frag;
4434
4435                 frag = &skb_shinfo(skb)->frags[f];
4436                 len = frag->size;
4437                 offset = frag->page_offset;
4438
4439                 while (len) {
4440                         i++;
4441                         if (i == tx_ring->count)
4442                                 i = 0;
4443
4444                         buffer_info = &tx_ring->buffer_info[i];
4445                         size = min(len, max_per_txd);
4446
4447                         buffer_info->length = size;
4448                         buffer_info->time_stamp = jiffies;
4449                         buffer_info->next_to_watch = i;
4450                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4451                                                         offset, size,
4452                                                         DMA_TO_DEVICE);
4453                         buffer_info->mapped_as_page = true;
4454                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4455                                 goto dma_error;
4456
4457                         len -= size;
4458                         offset += size;
4459                         count++;
4460                 }
4461         }
4462
4463         segs = skb_shinfo(skb)->gso_segs ?: 1;
4464         /* multiply data chunks by size of headers */
4465         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4466
4467         tx_ring->buffer_info[i].skb = skb;
4468         tx_ring->buffer_info[i].segs = segs;
4469         tx_ring->buffer_info[i].bytecount = bytecount;
4470         tx_ring->buffer_info[first].next_to_watch = i;
4471
4472         return count;
4473
4474 dma_error:
4475         dev_err(&pdev->dev, "TX DMA map failed\n");
4476         buffer_info->dma = 0;
4477         if (count)
4478                 count--;
4479
4480         while (count--) {
4481                 if (i==0)
4482                         i += tx_ring->count;
4483                 i--;
4484                 buffer_info = &tx_ring->buffer_info[i];
4485                 e1000_put_txbuf(adapter, buffer_info);;
4486         }
4487
4488         return 0;
4489 }
4490
4491 static void e1000_tx_queue(struct e1000_adapter *adapter,
4492                            int tx_flags, int count)
4493 {
4494         struct e1000_ring *tx_ring = adapter->tx_ring;
4495         struct e1000_tx_desc *tx_desc = NULL;
4496         struct e1000_buffer *buffer_info;
4497         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4498         unsigned int i;
4499
4500         if (tx_flags & E1000_TX_FLAGS_TSO) {
4501                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4502                              E1000_TXD_CMD_TSE;
4503                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4504
4505                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4506                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4507         }
4508
4509         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4510                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4511                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4512         }
4513
4514         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4515                 txd_lower |= E1000_TXD_CMD_VLE;
4516                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4517         }
4518
4519         i = tx_ring->next_to_use;
4520
4521         while (count--) {
4522                 buffer_info = &tx_ring->buffer_info[i];
4523                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4524                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4525                 tx_desc->lower.data =
4526                         cpu_to_le32(txd_lower | buffer_info->length);
4527                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4528
4529                 i++;
4530                 if (i == tx_ring->count)
4531                         i = 0;
4532         }
4533
4534         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4535
4536         /*
4537          * Force memory writes to complete before letting h/w
4538          * know there are new descriptors to fetch.  (Only
4539          * applicable for weak-ordered memory model archs,
4540          * such as IA-64).
4541          */
4542         wmb();
4543
4544         tx_ring->next_to_use = i;
4545         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4546         /*
4547          * we need this if more than one processor can write to our tail
4548          * at a time, it synchronizes IO on IA64/Altix systems
4549          */
4550         mmiowb();
4551 }
4552
4553 #define MINIMUM_DHCP_PACKET_SIZE 282
4554 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4555                                     struct sk_buff *skb)
4556 {
4557         struct e1000_hw *hw =  &adapter->hw;
4558         u16 length, offset;
4559
4560         if (vlan_tx_tag_present(skb)) {
4561                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4562                     (adapter->hw.mng_cookie.status &
4563                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4564                         return 0;
4565         }
4566
4567         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4568                 return 0;
4569
4570         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4571                 return 0;
4572
4573         {
4574                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4575                 struct udphdr *udp;
4576
4577                 if (ip->protocol != IPPROTO_UDP)
4578                         return 0;
4579
4580                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4581                 if (ntohs(udp->dest) != 67)
4582                         return 0;
4583
4584                 offset = (u8 *)udp + 8 - skb->data;
4585                 length = skb->len - offset;
4586                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4587         }
4588
4589         return 0;
4590 }
4591
4592 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4593 {
4594         struct e1000_adapter *adapter = netdev_priv(netdev);
4595
4596         netif_stop_queue(netdev);
4597         /*
4598          * Herbert's original patch had:
4599          *  smp_mb__after_netif_stop_queue();
4600          * but since that doesn't exist yet, just open code it.
4601          */
4602         smp_mb();
4603
4604         /*
4605          * We need to check again in a case another CPU has just
4606          * made room available.
4607          */
4608         if (e1000_desc_unused(adapter->tx_ring) < size)
4609                 return -EBUSY;
4610
4611         /* A reprieve! */
4612         netif_start_queue(netdev);
4613         ++adapter->restart_queue;
4614         return 0;
4615 }
4616
4617 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4618 {
4619         struct e1000_adapter *adapter = netdev_priv(netdev);
4620
4621         if (e1000_desc_unused(adapter->tx_ring) >= size)
4622                 return 0;
4623         return __e1000_maybe_stop_tx(netdev, size);
4624 }
4625
4626 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4627 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4628                                     struct net_device *netdev)
4629 {
4630         struct e1000_adapter *adapter = netdev_priv(netdev);
4631         struct e1000_ring *tx_ring = adapter->tx_ring;
4632         unsigned int first;
4633         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4634         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4635         unsigned int tx_flags = 0;
4636         unsigned int len = skb_headlen(skb);
4637         unsigned int nr_frags;
4638         unsigned int mss;
4639         int count = 0;
4640         int tso;
4641         unsigned int f;
4642
4643         if (test_bit(__E1000_DOWN, &adapter->state)) {
4644                 dev_kfree_skb_any(skb);
4645                 return NETDEV_TX_OK;
4646         }
4647
4648         if (skb->len <= 0) {
4649                 dev_kfree_skb_any(skb);
4650                 return NETDEV_TX_OK;
4651         }
4652
4653         mss = skb_shinfo(skb)->gso_size;
4654         /*
4655          * The controller does a simple calculation to
4656          * make sure there is enough room in the FIFO before
4657          * initiating the DMA for each buffer.  The calc is:
4658          * 4 = ceil(buffer len/mss).  To make sure we don't
4659          * overrun the FIFO, adjust the max buffer len if mss
4660          * drops.
4661          */
4662         if (mss) {
4663                 u8 hdr_len;
4664                 max_per_txd = min(mss << 2, max_per_txd);
4665                 max_txd_pwr = fls(max_per_txd) - 1;
4666
4667                 /*
4668                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4669                  * points to just header, pull a few bytes of payload from
4670                  * frags into skb->data
4671                  */
4672                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4673                 /*
4674                  * we do this workaround for ES2LAN, but it is un-necessary,
4675                  * avoiding it could save a lot of cycles
4676                  */
4677                 if (skb->data_len && (hdr_len == len)) {
4678                         unsigned int pull_size;
4679
4680                         pull_size = min((unsigned int)4, skb->data_len);
4681                         if (!__pskb_pull_tail(skb, pull_size)) {
4682                                 e_err("__pskb_pull_tail failed.\n");
4683                                 dev_kfree_skb_any(skb);
4684                                 return NETDEV_TX_OK;
4685                         }
4686                         len = skb_headlen(skb);
4687                 }
4688         }
4689
4690         /* reserve a descriptor for the offload context */
4691         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4692                 count++;
4693         count++;
4694
4695         count += TXD_USE_COUNT(len, max_txd_pwr);
4696
4697         nr_frags = skb_shinfo(skb)->nr_frags;
4698         for (f = 0; f < nr_frags; f++)
4699                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4700                                        max_txd_pwr);
4701
4702         if (adapter->hw.mac.tx_pkt_filtering)
4703                 e1000_transfer_dhcp_info(adapter, skb);
4704
4705         /*
4706          * need: count + 2 desc gap to keep tail from touching
4707          * head, otherwise try next time
4708          */
4709         if (e1000_maybe_stop_tx(netdev, count + 2))
4710                 return NETDEV_TX_BUSY;
4711
4712         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4713                 tx_flags |= E1000_TX_FLAGS_VLAN;
4714                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4715         }
4716
4717         first = tx_ring->next_to_use;
4718
4719         tso = e1000_tso(adapter, skb);
4720         if (tso < 0) {
4721                 dev_kfree_skb_any(skb);
4722                 return NETDEV_TX_OK;
4723         }
4724
4725         if (tso)
4726                 tx_flags |= E1000_TX_FLAGS_TSO;
4727         else if (e1000_tx_csum(adapter, skb))
4728                 tx_flags |= E1000_TX_FLAGS_CSUM;
4729
4730         /*
4731          * Old method was to assume IPv4 packet by default if TSO was enabled.
4732          * 82571 hardware supports TSO capabilities for IPv6 as well...
4733          * no longer assume, we must.
4734          */
4735         if (skb->protocol == htons(ETH_P_IP))
4736                 tx_flags |= E1000_TX_FLAGS_IPV4;
4737
4738         /* if count is 0 then mapping error has occured */
4739         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4740         if (count) {
4741                 e1000_tx_queue(adapter, tx_flags, count);
4742                 /* Make sure there is space in the ring for the next send. */
4743                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4744
4745         } else {
4746                 dev_kfree_skb_any(skb);
4747                 tx_ring->buffer_info[first].time_stamp = 0;
4748                 tx_ring->next_to_use = first;
4749         }
4750
4751         return NETDEV_TX_OK;
4752 }
4753
4754 /**
4755  * e1000_tx_timeout - Respond to a Tx Hang
4756  * @netdev: network interface device structure
4757  **/
4758 static void e1000_tx_timeout(struct net_device *netdev)
4759 {
4760         struct e1000_adapter *adapter = netdev_priv(netdev);
4761
4762         /* Do the reset outside of interrupt context */
4763         adapter->tx_timeout_count++;
4764         schedule_work(&adapter->reset_task);
4765 }
4766
4767 static void e1000_reset_task(struct work_struct *work)
4768 {
4769         struct e1000_adapter *adapter;
4770         adapter = container_of(work, struct e1000_adapter, reset_task);
4771
4772         e1000e_dump(adapter);
4773         e_err("Reset adapter\n");
4774         e1000e_reinit_locked(adapter);
4775 }
4776
4777 /**
4778  * e1000_get_stats - Get System Network Statistics
4779  * @netdev: network interface device structure
4780  *
4781  * Returns the address of the device statistics structure.
4782  * The statistics are actually updated from the timer callback.
4783  **/
4784 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4785 {
4786         /* only return the current stats */
4787         return &netdev->stats;
4788 }
4789
4790 /**
4791  * e1000_change_mtu - Change the Maximum Transfer Unit
4792  * @netdev: network interface device structure
4793  * @new_mtu: new value for maximum frame size
4794  *
4795  * Returns 0 on success, negative on failure
4796  **/
4797 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4798 {
4799         struct e1000_adapter *adapter = netdev_priv(netdev);
4800         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4801
4802         /* Jumbo frame support */
4803         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4804             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4805                 e_err("Jumbo Frames not supported.\n");
4806                 return -EINVAL;
4807         }
4808
4809         /* Supported frame sizes */
4810         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4811             (max_frame > adapter->max_hw_frame_size)) {
4812                 e_err("Unsupported MTU setting\n");
4813                 return -EINVAL;
4814         }
4815
4816         /* 82573 Errata 17 */
4817         if (((adapter->hw.mac.type == e1000_82573) ||
4818              (adapter->hw.mac.type == e1000_82574)) &&
4819             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4820                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4821                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4822         }
4823
4824         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4825                 msleep(1);
4826         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4827         adapter->max_frame_size = max_frame;
4828         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4829         netdev->mtu = new_mtu;
4830         if (netif_running(netdev))
4831                 e1000e_down(adapter);
4832
4833         /*
4834          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4835          * means we reserve 2 more, this pushes us to allocate from the next
4836          * larger slab size.
4837          * i.e. RXBUFFER_2048 --> size-4096 slab
4838          * However with the new *_jumbo_rx* routines, jumbo receives will use
4839          * fragmented skbs
4840          */
4841
4842         if (max_frame <= 2048)
4843                 adapter->rx_buffer_len = 2048;
4844         else
4845                 adapter->rx_buffer_len = 4096;
4846
4847         /* adjust allocation if LPE protects us, and we aren't using SBP */
4848         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4849              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4850                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4851                                          + ETH_FCS_LEN;
4852
4853         if (netif_running(netdev))
4854                 e1000e_up(adapter);
4855         else
4856                 e1000e_reset(adapter);
4857
4858         clear_bit(__E1000_RESETTING, &adapter->state);
4859
4860         return 0;
4861 }
4862
4863 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4864                            int cmd)
4865 {
4866         struct e1000_adapter *adapter = netdev_priv(netdev);
4867         struct mii_ioctl_data *data = if_mii(ifr);
4868
4869         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4870                 return -EOPNOTSUPP;
4871
4872         switch (cmd) {
4873         case SIOCGMIIPHY:
4874                 data->phy_id = adapter->hw.phy.addr;
4875                 break;
4876         case SIOCGMIIREG:
4877                 e1000_phy_read_status(adapter);
4878
4879                 switch (data->reg_num & 0x1F) {
4880                 case MII_BMCR:
4881                         data->val_out = adapter->phy_regs.bmcr;
4882                         break;
4883                 case MII_BMSR:
4884                         data->val_out = adapter->phy_regs.bmsr;
4885                         break;
4886                 case MII_PHYSID1:
4887                         data->val_out = (adapter->hw.phy.id >> 16);
4888                         break;
4889                 case MII_PHYSID2:
4890                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4891                         break;
4892                 case MII_ADVERTISE:
4893                         data->val_out = adapter->phy_regs.advertise;
4894                         break;
4895                 case MII_LPA:
4896                         data->val_out = adapter->phy_regs.lpa;
4897                         break;
4898                 case MII_EXPANSION:
4899                         data->val_out = adapter->phy_regs.expansion;
4900                         break;
4901                 case MII_CTRL1000:
4902                         data->val_out = adapter->phy_regs.ctrl1000;
4903                         break;
4904                 case MII_STAT1000:
4905                         data->val_out = adapter->phy_regs.stat1000;
4906                         break;
4907                 case MII_ESTATUS:
4908                         data->val_out = adapter->phy_regs.estatus;
4909                         break;
4910                 default:
4911                         return -EIO;
4912                 }
4913                 break;
4914         case SIOCSMIIREG:
4915         default:
4916                 return -EOPNOTSUPP;
4917         }
4918         return 0;
4919 }
4920
4921 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4922 {
4923         switch (cmd) {
4924         case SIOCGMIIPHY:
4925         case SIOCGMIIREG:
4926         case SIOCSMIIREG:
4927                 return e1000_mii_ioctl(netdev, ifr, cmd);
4928         default:
4929                 return -EOPNOTSUPP;
4930         }
4931 }
4932
4933 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4934 {
4935         struct e1000_hw *hw = &adapter->hw;
4936         u32 i, mac_reg;
4937         u16 phy_reg;
4938         int retval = 0;
4939
4940         /* copy MAC RARs to PHY RARs */
4941         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
4942
4943         /* copy MAC MTA to PHY MTA */
4944         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4945                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4946                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4947                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4948         }
4949
4950         /* configure PHY Rx Control register */
4951         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4952         mac_reg = er32(RCTL);
4953         if (mac_reg & E1000_RCTL_UPE)
4954                 phy_reg |= BM_RCTL_UPE;
4955         if (mac_reg & E1000_RCTL_MPE)
4956                 phy_reg |= BM_RCTL_MPE;
4957         phy_reg &= ~(BM_RCTL_MO_MASK);
4958         if (mac_reg & E1000_RCTL_MO_3)
4959                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4960                                 << BM_RCTL_MO_SHIFT);
4961         if (mac_reg & E1000_RCTL_BAM)
4962                 phy_reg |= BM_RCTL_BAM;
4963         if (mac_reg & E1000_RCTL_PMCF)
4964                 phy_reg |= BM_RCTL_PMCF;
4965         mac_reg = er32(CTRL);
4966         if (mac_reg & E1000_CTRL_RFCE)
4967                 phy_reg |= BM_RCTL_RFCE;
4968         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4969
4970         /* enable PHY wakeup in MAC register */
4971         ew32(WUFC, wufc);
4972         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4973
4974         /* configure and enable PHY wakeup in PHY registers */
4975         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4976         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4977
4978         /* activate PHY wakeup */
4979         retval = hw->phy.ops.acquire(hw);
4980         if (retval) {
4981                 e_err("Could not acquire PHY\n");
4982                 return retval;
4983         }
4984         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
4985                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
4986         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
4987         if (retval) {
4988                 e_err("Could not read PHY page 769\n");
4989                 goto out;
4990         }
4991         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
4992         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
4993         if (retval)
4994                 e_err("Could not set PHY Host Wakeup bit\n");
4995 out:
4996         hw->phy.ops.release(hw);
4997
4998         return retval;
4999 }
5000
5001 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5002                             bool runtime)
5003 {
5004         struct net_device *netdev = pci_get_drvdata(pdev);
5005         struct e1000_adapter *adapter = netdev_priv(netdev);
5006         struct e1000_hw *hw = &adapter->hw;
5007         u32 ctrl, ctrl_ext, rctl, status;
5008         /* Runtime suspend should only enable wakeup for link changes */
5009         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5010         int retval = 0;
5011
5012         netif_device_detach(netdev);
5013
5014         if (netif_running(netdev)) {
5015                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5016                 e1000e_down(adapter);
5017                 e1000_free_irq(adapter);
5018         }
5019         e1000e_reset_interrupt_capability(adapter);
5020
5021         retval = pci_save_state(pdev);
5022         if (retval)
5023                 return retval;
5024
5025         status = er32(STATUS);
5026         if (status & E1000_STATUS_LU)
5027                 wufc &= ~E1000_WUFC_LNKC;
5028
5029         if (wufc) {
5030                 e1000_setup_rctl(adapter);
5031                 e1000_set_multi(netdev);
5032
5033                 /* turn on all-multi mode if wake on multicast is enabled */
5034                 if (wufc & E1000_WUFC_MC) {
5035                         rctl = er32(RCTL);
5036                         rctl |= E1000_RCTL_MPE;
5037                         ew32(RCTL, rctl);
5038                 }
5039
5040                 ctrl = er32(CTRL);
5041                 /* advertise wake from D3Cold */
5042                 #define E1000_CTRL_ADVD3WUC 0x00100000
5043                 /* phy power management enable */
5044                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5045                 ctrl |= E1000_CTRL_ADVD3WUC;
5046                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5047                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5048                 ew32(CTRL, ctrl);
5049
5050                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5051                     adapter->hw.phy.media_type ==
5052                     e1000_media_type_internal_serdes) {
5053                         /* keep the laser running in D3 */
5054                         ctrl_ext = er32(CTRL_EXT);
5055                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5056                         ew32(CTRL_EXT, ctrl_ext);
5057                 }
5058
5059                 if (adapter->flags & FLAG_IS_ICH)
5060                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5061
5062                 /* Allow time for pending master requests to run */
5063                 e1000e_disable_pcie_master(&adapter->hw);
5064
5065                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5066                         /* enable wakeup by the PHY */
5067                         retval = e1000_init_phy_wakeup(adapter, wufc);
5068                         if (retval)
5069                                 return retval;
5070                 } else {
5071                         /* enable wakeup by the MAC */
5072                         ew32(WUFC, wufc);
5073                         ew32(WUC, E1000_WUC_PME_EN);
5074                 }
5075         } else {
5076                 ew32(WUC, 0);
5077                 ew32(WUFC, 0);
5078         }
5079
5080         *enable_wake = !!wufc;
5081
5082         /* make sure adapter isn't asleep if manageability is enabled */
5083         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5084             (hw->mac.ops.check_mng_mode(hw)))
5085                 *enable_wake = true;
5086
5087         if (adapter->hw.phy.type == e1000_phy_igp_3)
5088                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5089
5090         /*
5091          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5092          * would have already happened in close and is redundant.
5093          */
5094         e1000_release_hw_control(adapter);
5095
5096         pci_disable_device(pdev);
5097
5098         return 0;
5099 }
5100
5101 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5102 {
5103         if (sleep && wake) {
5104                 pci_prepare_to_sleep(pdev);
5105                 return;
5106         }
5107
5108         pci_wake_from_d3(pdev, wake);
5109         pci_set_power_state(pdev, PCI_D3hot);
5110 }
5111
5112 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5113                                     bool wake)
5114 {
5115         struct net_device *netdev = pci_get_drvdata(pdev);
5116         struct e1000_adapter *adapter = netdev_priv(netdev);
5117
5118         /*
5119          * The pci-e switch on some quad port adapters will report a
5120          * correctable error when the MAC transitions from D0 to D3.  To
5121          * prevent this we need to mask off the correctable errors on the
5122          * downstream port of the pci-e switch.
5123          */
5124         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5125                 struct pci_dev *us_dev = pdev->bus->self;
5126                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5127                 u16 devctl;
5128
5129                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5130                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5131                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5132
5133                 e1000_power_off(pdev, sleep, wake);
5134
5135                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5136         } else {
5137                 e1000_power_off(pdev, sleep, wake);
5138         }
5139 }
5140
5141 #ifdef CONFIG_PCIEASPM
5142 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5143 {
5144         pci_disable_link_state(pdev, state);
5145 }
5146 #else
5147 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5148 {
5149         int pos;
5150         u16 reg16;
5151
5152         /*
5153          * Both device and parent should have the same ASPM setting.
5154          * Disable ASPM in downstream component first and then upstream.
5155          */
5156         pos = pci_pcie_cap(pdev);
5157         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5158         reg16 &= ~state;
5159         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5160
5161         if (!pdev->bus->self)
5162                 return;
5163
5164         pos = pci_pcie_cap(pdev->bus->self);
5165         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5166         reg16 &= ~state;
5167         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5168 }
5169 #endif
5170 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5171 {
5172         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5173                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5174                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5175
5176         __e1000e_disable_aspm(pdev, state);
5177 }
5178
5179 #ifdef CONFIG_PM_OPS
5180 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5181 {
5182         return !!adapter->tx_ring->buffer_info;
5183 }
5184
5185 static int __e1000_resume(struct pci_dev *pdev)
5186 {
5187         struct net_device *netdev = pci_get_drvdata(pdev);
5188         struct e1000_adapter *adapter = netdev_priv(netdev);
5189         struct e1000_hw *hw = &adapter->hw;
5190         u32 err;
5191
5192         pci_set_power_state(pdev, PCI_D0);
5193         pci_restore_state(pdev);
5194         pci_save_state(pdev);
5195         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5196                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5197
5198         e1000e_set_interrupt_capability(adapter);
5199         if (netif_running(netdev)) {
5200                 err = e1000_request_irq(adapter);
5201                 if (err)
5202                         return err;
5203         }
5204
5205         e1000e_power_up_phy(adapter);
5206
5207         /* report the system wakeup cause from S3/S4 */
5208         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5209                 u16 phy_data;
5210
5211                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5212                 if (phy_data) {
5213                         e_info("PHY Wakeup cause - %s\n",
5214                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5215                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5216                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5217                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5218                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5219                                 " Change" : "other");
5220                 }
5221                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5222         } else {
5223                 u32 wus = er32(WUS);
5224                 if (wus) {
5225                         e_info("MAC Wakeup cause - %s\n",
5226                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5227                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5228                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5229                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5230                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5231                                 "other");
5232                 }
5233                 ew32(WUS, ~0);
5234         }
5235
5236         e1000e_reset(adapter);
5237
5238         e1000_init_manageability_pt(adapter);
5239
5240         if (netif_running(netdev))
5241                 e1000e_up(adapter);
5242
5243         netif_device_attach(netdev);
5244
5245         /*
5246          * If the controller has AMT, do not set DRV_LOAD until the interface
5247          * is up.  For all other cases, let the f/w know that the h/w is now
5248          * under the control of the driver.
5249          */
5250         if (!(adapter->flags & FLAG_HAS_AMT))
5251                 e1000_get_hw_control(adapter);
5252
5253         return 0;
5254 }
5255
5256 #ifdef CONFIG_PM_SLEEP
5257 static int e1000_suspend(struct device *dev)
5258 {
5259         struct pci_dev *pdev = to_pci_dev(dev);
5260         int retval;
5261         bool wake;
5262
5263         retval = __e1000_shutdown(pdev, &wake, false);
5264         if (!retval)
5265                 e1000_complete_shutdown(pdev, true, wake);
5266
5267         return retval;
5268 }
5269
5270 static int e1000_resume(struct device *dev)
5271 {
5272         struct pci_dev *pdev = to_pci_dev(dev);
5273         struct net_device *netdev = pci_get_drvdata(pdev);
5274         struct e1000_adapter *adapter = netdev_priv(netdev);
5275
5276         if (e1000e_pm_ready(adapter))
5277                 adapter->idle_check = true;
5278
5279         return __e1000_resume(pdev);
5280 }
5281 #endif /* CONFIG_PM_SLEEP */
5282
5283 #ifdef CONFIG_PM_RUNTIME
5284 static int e1000_runtime_suspend(struct device *dev)
5285 {
5286         struct pci_dev *pdev = to_pci_dev(dev);
5287         struct net_device *netdev = pci_get_drvdata(pdev);
5288         struct e1000_adapter *adapter = netdev_priv(netdev);
5289
5290         if (e1000e_pm_ready(adapter)) {
5291                 bool wake;
5292
5293                 __e1000_shutdown(pdev, &wake, true);
5294         }
5295
5296         return 0;
5297 }
5298
5299 static int e1000_idle(struct device *dev)
5300 {
5301         struct pci_dev *pdev = to_pci_dev(dev);
5302         struct net_device *netdev = pci_get_drvdata(pdev);
5303         struct e1000_adapter *adapter = netdev_priv(netdev);
5304
5305         if (!e1000e_pm_ready(adapter))
5306                 return 0;
5307
5308         if (adapter->idle_check) {
5309                 adapter->idle_check = false;
5310                 if (!e1000e_has_link(adapter))
5311                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5312         }
5313
5314         return -EBUSY;
5315 }
5316
5317 static int e1000_runtime_resume(struct device *dev)
5318 {
5319         struct pci_dev *pdev = to_pci_dev(dev);
5320         struct net_device *netdev = pci_get_drvdata(pdev);
5321         struct e1000_adapter *adapter = netdev_priv(netdev);
5322
5323         if (!e1000e_pm_ready(adapter))
5324                 return 0;
5325
5326         adapter->idle_check = !dev->power.runtime_auto;
5327         return __e1000_resume(pdev);
5328 }
5329 #endif /* CONFIG_PM_RUNTIME */
5330 #endif /* CONFIG_PM_OPS */
5331
5332 static void e1000_shutdown(struct pci_dev *pdev)
5333 {
5334         bool wake = false;
5335
5336         __e1000_shutdown(pdev, &wake, false);
5337
5338         if (system_state == SYSTEM_POWER_OFF)
5339                 e1000_complete_shutdown(pdev, false, wake);
5340 }
5341
5342 #ifdef CONFIG_NET_POLL_CONTROLLER
5343 /*
5344  * Polling 'interrupt' - used by things like netconsole to send skbs
5345  * without having to re-enable interrupts. It's not called while
5346  * the interrupt routine is executing.
5347  */
5348 static void e1000_netpoll(struct net_device *netdev)
5349 {
5350         struct e1000_adapter *adapter = netdev_priv(netdev);
5351
5352         disable_irq(adapter->pdev->irq);
5353         e1000_intr(adapter->pdev->irq, netdev);
5354
5355         enable_irq(adapter->pdev->irq);
5356 }
5357 #endif
5358
5359 /**
5360  * e1000_io_error_detected - called when PCI error is detected
5361  * @pdev: Pointer to PCI device
5362  * @state: The current pci connection state
5363  *
5364  * This function is called after a PCI bus error affecting
5365  * this device has been detected.
5366  */
5367 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5368                                                 pci_channel_state_t state)
5369 {
5370         struct net_device *netdev = pci_get_drvdata(pdev);
5371         struct e1000_adapter *adapter = netdev_priv(netdev);
5372
5373         netif_device_detach(netdev);
5374
5375         if (state == pci_channel_io_perm_failure)
5376                 return PCI_ERS_RESULT_DISCONNECT;
5377
5378         if (netif_running(netdev))
5379                 e1000e_down(adapter);
5380         pci_disable_device(pdev);
5381
5382         /* Request a slot slot reset. */
5383         return PCI_ERS_RESULT_NEED_RESET;
5384 }
5385
5386 /**
5387  * e1000_io_slot_reset - called after the pci bus has been reset.
5388  * @pdev: Pointer to PCI device
5389  *
5390  * Restart the card from scratch, as if from a cold-boot. Implementation
5391  * resembles the first-half of the e1000_resume routine.
5392  */
5393 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5394 {
5395         struct net_device *netdev = pci_get_drvdata(pdev);
5396         struct e1000_adapter *adapter = netdev_priv(netdev);
5397         struct e1000_hw *hw = &adapter->hw;
5398         int err;
5399         pci_ers_result_t result;
5400
5401         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5402                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5403         err = pci_enable_device_mem(pdev);
5404         if (err) {
5405                 dev_err(&pdev->dev,
5406                         "Cannot re-enable PCI device after reset.\n");
5407                 result = PCI_ERS_RESULT_DISCONNECT;
5408         } else {
5409                 pci_set_master(pdev);
5410                 pdev->state_saved = true;
5411                 pci_restore_state(pdev);
5412
5413                 pci_enable_wake(pdev, PCI_D3hot, 0);
5414                 pci_enable_wake(pdev, PCI_D3cold, 0);
5415
5416                 e1000e_reset(adapter);
5417                 ew32(WUS, ~0);
5418                 result = PCI_ERS_RESULT_RECOVERED;
5419         }
5420
5421         pci_cleanup_aer_uncorrect_error_status(pdev);
5422
5423         return result;
5424 }
5425
5426 /**
5427  * e1000_io_resume - called when traffic can start flowing again.
5428  * @pdev: Pointer to PCI device
5429  *
5430  * This callback is called when the error recovery driver tells us that
5431  * its OK to resume normal operation. Implementation resembles the
5432  * second-half of the e1000_resume routine.
5433  */
5434 static void e1000_io_resume(struct pci_dev *pdev)
5435 {
5436         struct net_device *netdev = pci_get_drvdata(pdev);
5437         struct e1000_adapter *adapter = netdev_priv(netdev);
5438
5439         e1000_init_manageability_pt(adapter);
5440
5441         if (netif_running(netdev)) {
5442                 if (e1000e_up(adapter)) {
5443                         dev_err(&pdev->dev,
5444                                 "can't bring device back up after reset\n");
5445                         return;
5446                 }
5447         }
5448
5449         netif_device_attach(netdev);
5450
5451         /*
5452          * If the controller has AMT, do not set DRV_LOAD until the interface
5453          * is up.  For all other cases, let the f/w know that the h/w is now
5454          * under the control of the driver.
5455          */
5456         if (!(adapter->flags & FLAG_HAS_AMT))
5457                 e1000_get_hw_control(adapter);
5458
5459 }
5460
5461 static void e1000_print_device_info(struct e1000_adapter *adapter)
5462 {
5463         struct e1000_hw *hw = &adapter->hw;
5464         struct net_device *netdev = adapter->netdev;
5465         u32 pba_num;
5466
5467         /* print bus type/speed/width info */
5468         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5469                /* bus width */
5470                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5471                 "Width x1"),
5472                /* MAC address */
5473                netdev->dev_addr);
5474         e_info("Intel(R) PRO/%s Network Connection\n",
5475                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5476         e1000e_read_pba_num(hw, &pba_num);
5477         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5478                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5479 }
5480
5481 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5482 {
5483         struct e1000_hw *hw = &adapter->hw;
5484         int ret_val;
5485         u16 buf = 0;
5486
5487         if (hw->mac.type != e1000_82573)
5488                 return;
5489
5490         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5491         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5492                 /* Deep Smart Power Down (DSPD) */
5493                 dev_warn(&adapter->pdev->dev,
5494                          "Warning: detected DSPD enabled in EEPROM\n");
5495         }
5496 }
5497
5498 static const struct net_device_ops e1000e_netdev_ops = {
5499         .ndo_open               = e1000_open,
5500         .ndo_stop               = e1000_close,
5501         .ndo_start_xmit         = e1000_xmit_frame,
5502         .ndo_get_stats          = e1000_get_stats,
5503         .ndo_set_multicast_list = e1000_set_multi,
5504         .ndo_set_mac_address    = e1000_set_mac,
5505         .ndo_change_mtu         = e1000_change_mtu,
5506         .ndo_do_ioctl           = e1000_ioctl,
5507         .ndo_tx_timeout         = e1000_tx_timeout,
5508         .ndo_validate_addr      = eth_validate_addr,
5509
5510         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5511         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5512         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5513 #ifdef CONFIG_NET_POLL_CONTROLLER
5514         .ndo_poll_controller    = e1000_netpoll,
5515 #endif
5516 };
5517
5518 /**
5519  * e1000_probe - Device Initialization Routine
5520  * @pdev: PCI device information struct
5521  * @ent: entry in e1000_pci_tbl
5522  *
5523  * Returns 0 on success, negative on failure
5524  *
5525  * e1000_probe initializes an adapter identified by a pci_dev structure.
5526  * The OS initialization, configuring of the adapter private structure,
5527  * and a hardware reset occur.
5528  **/
5529 static int __devinit e1000_probe(struct pci_dev *pdev,
5530                                  const struct pci_device_id *ent)
5531 {
5532         struct net_device *netdev;
5533         struct e1000_adapter *adapter;
5534         struct e1000_hw *hw;
5535         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5536         resource_size_t mmio_start, mmio_len;
5537         resource_size_t flash_start, flash_len;
5538
5539         static int cards_found;
5540         int i, err, pci_using_dac;
5541         u16 eeprom_data = 0;
5542         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5543
5544         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5545                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5546
5547         err = pci_enable_device_mem(pdev);
5548         if (err)
5549                 return err;
5550
5551         pci_using_dac = 0;
5552         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5553         if (!err) {
5554                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5555                 if (!err)
5556                         pci_using_dac = 1;
5557         } else {
5558                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5559                 if (err) {
5560                         err = dma_set_coherent_mask(&pdev->dev,
5561                                                     DMA_BIT_MASK(32));
5562                         if (err) {
5563                                 dev_err(&pdev->dev, "No usable DMA "
5564                                         "configuration, aborting\n");
5565                                 goto err_dma;
5566                         }
5567                 }
5568         }
5569
5570         err = pci_request_selected_regions_exclusive(pdev,
5571                                           pci_select_bars(pdev, IORESOURCE_MEM),
5572                                           e1000e_driver_name);
5573         if (err)
5574                 goto err_pci_reg;
5575
5576         /* AER (Advanced Error Reporting) hooks */
5577         pci_enable_pcie_error_reporting(pdev);
5578
5579         pci_set_master(pdev);
5580         /* PCI config space info */
5581         err = pci_save_state(pdev);
5582         if (err)
5583                 goto err_alloc_etherdev;
5584
5585         err = -ENOMEM;
5586         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5587         if (!netdev)
5588                 goto err_alloc_etherdev;
5589
5590         SET_NETDEV_DEV(netdev, &pdev->dev);
5591
5592         netdev->irq = pdev->irq;
5593
5594         pci_set_drvdata(pdev, netdev);
5595         adapter = netdev_priv(netdev);
5596         hw = &adapter->hw;
5597         adapter->netdev = netdev;
5598         adapter->pdev = pdev;
5599         adapter->ei = ei;
5600         adapter->pba = ei->pba;
5601         adapter->flags = ei->flags;
5602         adapter->flags2 = ei->flags2;
5603         adapter->hw.adapter = adapter;
5604         adapter->hw.mac.type = ei->mac;
5605         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5606         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5607
5608         mmio_start = pci_resource_start(pdev, 0);
5609         mmio_len = pci_resource_len(pdev, 0);
5610
5611         err = -EIO;
5612         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5613         if (!adapter->hw.hw_addr)
5614                 goto err_ioremap;
5615
5616         if ((adapter->flags & FLAG_HAS_FLASH) &&
5617             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5618                 flash_start = pci_resource_start(pdev, 1);
5619                 flash_len = pci_resource_len(pdev, 1);
5620                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5621                 if (!adapter->hw.flash_address)
5622                         goto err_flashmap;
5623         }
5624
5625         /* construct the net_device struct */
5626         netdev->netdev_ops              = &e1000e_netdev_ops;
5627         e1000e_set_ethtool_ops(netdev);
5628         netdev->watchdog_timeo          = 5 * HZ;
5629         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5630         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5631
5632         netdev->mem_start = mmio_start;
5633         netdev->mem_end = mmio_start + mmio_len;
5634
5635         adapter->bd_number = cards_found++;
5636
5637         e1000e_check_options(adapter);
5638
5639         /* setup adapter struct */
5640         err = e1000_sw_init(adapter);
5641         if (err)
5642                 goto err_sw_init;
5643
5644         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5645         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5646         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5647
5648         err = ei->get_variants(adapter);
5649         if (err)
5650                 goto err_hw_init;
5651
5652         if ((adapter->flags & FLAG_IS_ICH) &&
5653             (adapter->flags & FLAG_READ_ONLY_NVM))
5654                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5655
5656         hw->mac.ops.get_bus_info(&adapter->hw);
5657
5658         adapter->hw.phy.autoneg_wait_to_complete = 0;
5659
5660         /* Copper options */
5661         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5662                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5663                 adapter->hw.phy.disable_polarity_correction = 0;
5664                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5665         }
5666
5667         if (e1000_check_reset_block(&adapter->hw))
5668                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5669
5670         netdev->features = NETIF_F_SG |
5671                            NETIF_F_HW_CSUM |
5672                            NETIF_F_HW_VLAN_TX |
5673                            NETIF_F_HW_VLAN_RX;
5674
5675         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5676                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5677
5678         netdev->features |= NETIF_F_TSO;
5679         netdev->features |= NETIF_F_TSO6;
5680
5681         netdev->vlan_features |= NETIF_F_TSO;
5682         netdev->vlan_features |= NETIF_F_TSO6;
5683         netdev->vlan_features |= NETIF_F_HW_CSUM;
5684         netdev->vlan_features |= NETIF_F_SG;
5685
5686         if (pci_using_dac) {
5687                 netdev->features |= NETIF_F_HIGHDMA;
5688                 netdev->vlan_features |= NETIF_F_HIGHDMA;
5689         }
5690
5691         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5692                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5693
5694         /*
5695          * before reading the NVM, reset the controller to
5696          * put the device in a known good starting state
5697          */
5698         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5699
5700         /*
5701          * systems with ASPM and others may see the checksum fail on the first
5702          * attempt. Let's give it a few tries
5703          */
5704         for (i = 0;; i++) {
5705                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5706                         break;
5707                 if (i == 2) {
5708                         e_err("The NVM Checksum Is Not Valid\n");
5709                         err = -EIO;
5710                         goto err_eeprom;
5711                 }
5712         }
5713
5714         e1000_eeprom_checks(adapter);
5715
5716         /* copy the MAC address */
5717         if (e1000e_read_mac_addr(&adapter->hw))
5718                 e_err("NVM Read Error while reading MAC address\n");
5719
5720         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5721         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5722
5723         if (!is_valid_ether_addr(netdev->perm_addr)) {
5724                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5725                 err = -EIO;
5726                 goto err_eeprom;
5727         }
5728
5729         init_timer(&adapter->watchdog_timer);
5730         adapter->watchdog_timer.function = e1000_watchdog;
5731         adapter->watchdog_timer.data = (unsigned long) adapter;
5732
5733         init_timer(&adapter->phy_info_timer);
5734         adapter->phy_info_timer.function = e1000_update_phy_info;
5735         adapter->phy_info_timer.data = (unsigned long) adapter;
5736
5737         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5738         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5739         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5740         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5741         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5742
5743         /* Initialize link parameters. User can change them with ethtool */
5744         adapter->hw.mac.autoneg = 1;
5745         adapter->fc_autoneg = 1;
5746         adapter->hw.fc.requested_mode = e1000_fc_default;
5747         adapter->hw.fc.current_mode = e1000_fc_default;
5748         adapter->hw.phy.autoneg_advertised = 0x2f;
5749
5750         /* ring size defaults */
5751         adapter->rx_ring->count = 256;
5752         adapter->tx_ring->count = 256;
5753
5754         /*
5755          * Initial Wake on LAN setting - If APM wake is enabled in
5756          * the EEPROM, enable the ACPI Magic Packet filter
5757          */
5758         if (adapter->flags & FLAG_APME_IN_WUC) {
5759                 /* APME bit in EEPROM is mapped to WUC.APME */
5760                 eeprom_data = er32(WUC);
5761                 eeprom_apme_mask = E1000_WUC_APME;
5762                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5763                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5764         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5765                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5766                     (adapter->hw.bus.func == 1))
5767                         e1000_read_nvm(&adapter->hw,
5768                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5769                 else
5770                         e1000_read_nvm(&adapter->hw,
5771                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5772         }
5773
5774         /* fetch WoL from EEPROM */
5775         if (eeprom_data & eeprom_apme_mask)
5776                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5777
5778         /*
5779          * now that we have the eeprom settings, apply the special cases
5780          * where the eeprom may be wrong or the board simply won't support
5781          * wake on lan on a particular port
5782          */
5783         if (!(adapter->flags & FLAG_HAS_WOL))
5784                 adapter->eeprom_wol = 0;
5785
5786         /* initialize the wol settings based on the eeprom settings */
5787         adapter->wol = adapter->eeprom_wol;
5788         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
5789
5790     &nb