Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[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         skb->ip_summed = CHECKSUM_NONE;
479
480         /* Ignore Checksum bit is set */
481         if (status & E1000_RXD_STAT_IXSM)
482                 return;
483         /* TCP/UDP checksum error bit is set */
484         if (errors & E1000_RXD_ERR_TCPE) {
485                 /* let the stack verify checksum errors */
486                 adapter->hw_csum_err++;
487                 return;
488         }
489
490         /* TCP/UDP Checksum has not been calculated */
491         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
492                 return;
493
494         /* It must be a TCP or UDP packet with a valid checksum */
495         if (status & E1000_RXD_STAT_TCPCS) {
496                 /* TCP checksum is good */
497                 skb->ip_summed = CHECKSUM_UNNECESSARY;
498         } else {
499                 /*
500                  * IP fragment with UDP payload
501                  * Hardware complements the payload checksum, so we undo it
502                  * and then put the value in host order for further stack use.
503                  */
504                 __sum16 sum = (__force __sum16)htons(csum);
505                 skb->csum = csum_unfold(~sum);
506                 skb->ip_summed = CHECKSUM_COMPLETE;
507         }
508         adapter->hw_csum_good++;
509 }
510
511 /**
512  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
513  * @adapter: address of board private structure
514  **/
515 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
516                                    int cleaned_count)
517 {
518         struct net_device *netdev = adapter->netdev;
519         struct pci_dev *pdev = adapter->pdev;
520         struct e1000_ring *rx_ring = adapter->rx_ring;
521         struct e1000_rx_desc *rx_desc;
522         struct e1000_buffer *buffer_info;
523         struct sk_buff *skb;
524         unsigned int i;
525         unsigned int bufsz = adapter->rx_buffer_len;
526
527         i = rx_ring->next_to_use;
528         buffer_info = &rx_ring->buffer_info[i];
529
530         while (cleaned_count--) {
531                 skb = buffer_info->skb;
532                 if (skb) {
533                         skb_trim(skb, 0);
534                         goto map_skb;
535                 }
536
537                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
538                 if (!skb) {
539                         /* Better luck next round */
540                         adapter->alloc_rx_buff_failed++;
541                         break;
542                 }
543
544                 buffer_info->skb = skb;
545 map_skb:
546                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
547                                                   adapter->rx_buffer_len,
548                                                   DMA_FROM_DEVICE);
549                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
550                         dev_err(&pdev->dev, "RX DMA map failed\n");
551                         adapter->rx_dma_failed++;
552                         break;
553                 }
554
555                 rx_desc = E1000_RX_DESC(*rx_ring, i);
556                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
557
558                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
559                         /*
560                          * Force memory writes to complete before letting h/w
561                          * know there are new descriptors to fetch.  (Only
562                          * applicable for weak-ordered memory model archs,
563                          * such as IA-64).
564                          */
565                         wmb();
566                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
567                 }
568                 i++;
569                 if (i == rx_ring->count)
570                         i = 0;
571                 buffer_info = &rx_ring->buffer_info[i];
572         }
573
574         rx_ring->next_to_use = i;
575 }
576
577 /**
578  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
579  * @adapter: address of board private structure
580  **/
581 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
582                                       int cleaned_count)
583 {
584         struct net_device *netdev = adapter->netdev;
585         struct pci_dev *pdev = adapter->pdev;
586         union e1000_rx_desc_packet_split *rx_desc;
587         struct e1000_ring *rx_ring = adapter->rx_ring;
588         struct e1000_buffer *buffer_info;
589         struct e1000_ps_page *ps_page;
590         struct sk_buff *skb;
591         unsigned int i, j;
592
593         i = rx_ring->next_to_use;
594         buffer_info = &rx_ring->buffer_info[i];
595
596         while (cleaned_count--) {
597                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
598
599                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
600                         ps_page = &buffer_info->ps_pages[j];
601                         if (j >= adapter->rx_ps_pages) {
602                                 /* all unused desc entries get hw null ptr */
603                                 rx_desc->read.buffer_addr[j+1] = ~cpu_to_le64(0);
604                                 continue;
605                         }
606                         if (!ps_page->page) {
607                                 ps_page->page = alloc_page(GFP_ATOMIC);
608                                 if (!ps_page->page) {
609                                         adapter->alloc_rx_buff_failed++;
610                                         goto no_buffers;
611                                 }
612                                 ps_page->dma = dma_map_page(&pdev->dev,
613                                                             ps_page->page,
614                                                             0, PAGE_SIZE,
615                                                             DMA_FROM_DEVICE);
616                                 if (dma_mapping_error(&pdev->dev,
617                                                       ps_page->dma)) {
618                                         dev_err(&adapter->pdev->dev,
619                                           "RX DMA page map failed\n");
620                                         adapter->rx_dma_failed++;
621                                         goto no_buffers;
622                                 }
623                         }
624                         /*
625                          * Refresh the desc even if buffer_addrs
626                          * didn't change because each write-back
627                          * erases this info.
628                          */
629                         rx_desc->read.buffer_addr[j+1] =
630                              cpu_to_le64(ps_page->dma);
631                 }
632
633                 skb = netdev_alloc_skb_ip_align(netdev,
634                                                 adapter->rx_ps_bsize0);
635
636                 if (!skb) {
637                         adapter->alloc_rx_buff_failed++;
638                         break;
639                 }
640
641                 buffer_info->skb = skb;
642                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
643                                                   adapter->rx_ps_bsize0,
644                                                   DMA_FROM_DEVICE);
645                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
646                         dev_err(&pdev->dev, "RX DMA map failed\n");
647                         adapter->rx_dma_failed++;
648                         /* cleanup skb */
649                         dev_kfree_skb_any(skb);
650                         buffer_info->skb = NULL;
651                         break;
652                 }
653
654                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
655
656                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
657                         /*
658                          * Force memory writes to complete before letting h/w
659                          * know there are new descriptors to fetch.  (Only
660                          * applicable for weak-ordered memory model archs,
661                          * such as IA-64).
662                          */
663                         wmb();
664                         writel(i<<1, adapter->hw.hw_addr + rx_ring->tail);
665                 }
666
667                 i++;
668                 if (i == rx_ring->count)
669                         i = 0;
670                 buffer_info = &rx_ring->buffer_info[i];
671         }
672
673 no_buffers:
674         rx_ring->next_to_use = i;
675 }
676
677 /**
678  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
679  * @adapter: address of board private structure
680  * @cleaned_count: number of buffers to allocate this pass
681  **/
682
683 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
684                                          int cleaned_count)
685 {
686         struct net_device *netdev = adapter->netdev;
687         struct pci_dev *pdev = adapter->pdev;
688         struct e1000_rx_desc *rx_desc;
689         struct e1000_ring *rx_ring = adapter->rx_ring;
690         struct e1000_buffer *buffer_info;
691         struct sk_buff *skb;
692         unsigned int i;
693         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
694
695         i = rx_ring->next_to_use;
696         buffer_info = &rx_ring->buffer_info[i];
697
698         while (cleaned_count--) {
699                 skb = buffer_info->skb;
700                 if (skb) {
701                         skb_trim(skb, 0);
702                         goto check_page;
703                 }
704
705                 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
706                 if (unlikely(!skb)) {
707                         /* Better luck next round */
708                         adapter->alloc_rx_buff_failed++;
709                         break;
710                 }
711
712                 buffer_info->skb = skb;
713 check_page:
714                 /* allocate a new page if necessary */
715                 if (!buffer_info->page) {
716                         buffer_info->page = alloc_page(GFP_ATOMIC);
717                         if (unlikely(!buffer_info->page)) {
718                                 adapter->alloc_rx_buff_failed++;
719                                 break;
720                         }
721                 }
722
723                 if (!buffer_info->dma)
724                         buffer_info->dma = dma_map_page(&pdev->dev,
725                                                         buffer_info->page, 0,
726                                                         PAGE_SIZE,
727                                                         DMA_FROM_DEVICE);
728
729                 rx_desc = E1000_RX_DESC(*rx_ring, i);
730                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
731
732                 if (unlikely(++i == rx_ring->count))
733                         i = 0;
734                 buffer_info = &rx_ring->buffer_info[i];
735         }
736
737         if (likely(rx_ring->next_to_use != i)) {
738                 rx_ring->next_to_use = i;
739                 if (unlikely(i-- == 0))
740                         i = (rx_ring->count - 1);
741
742                 /* Force memory writes to complete before letting h/w
743                  * know there are new descriptors to fetch.  (Only
744                  * applicable for weak-ordered memory model archs,
745                  * such as IA-64). */
746                 wmb();
747                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
748         }
749 }
750
751 /**
752  * e1000_clean_rx_irq - Send received data up the network stack; legacy
753  * @adapter: board private structure
754  *
755  * the return value indicates whether actual cleaning was done, there
756  * is no guarantee that everything was cleaned
757  **/
758 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
759                                int *work_done, int work_to_do)
760 {
761         struct net_device *netdev = adapter->netdev;
762         struct pci_dev *pdev = adapter->pdev;
763         struct e1000_hw *hw = &adapter->hw;
764         struct e1000_ring *rx_ring = adapter->rx_ring;
765         struct e1000_rx_desc *rx_desc, *next_rxd;
766         struct e1000_buffer *buffer_info, *next_buffer;
767         u32 length;
768         unsigned int i;
769         int cleaned_count = 0;
770         bool cleaned = 0;
771         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
772
773         i = rx_ring->next_to_clean;
774         rx_desc = E1000_RX_DESC(*rx_ring, i);
775         buffer_info = &rx_ring->buffer_info[i];
776
777         while (rx_desc->status & E1000_RXD_STAT_DD) {
778                 struct sk_buff *skb;
779                 u8 status;
780
781                 if (*work_done >= work_to_do)
782                         break;
783                 (*work_done)++;
784                 rmb();  /* read descriptor and rx_buffer_info after status DD */
785
786                 status = rx_desc->status;
787                 skb = buffer_info->skb;
788                 buffer_info->skb = NULL;
789
790                 prefetch(skb->data - NET_IP_ALIGN);
791
792                 i++;
793                 if (i == rx_ring->count)
794                         i = 0;
795                 next_rxd = E1000_RX_DESC(*rx_ring, i);
796                 prefetch(next_rxd);
797
798                 next_buffer = &rx_ring->buffer_info[i];
799
800                 cleaned = 1;
801                 cleaned_count++;
802                 dma_unmap_single(&pdev->dev,
803                                  buffer_info->dma,
804                                  adapter->rx_buffer_len,
805                                  DMA_FROM_DEVICE);
806                 buffer_info->dma = 0;
807
808                 length = le16_to_cpu(rx_desc->length);
809
810                 /*
811                  * !EOP means multiple descriptors were used to store a single
812                  * packet, if that's the case we need to toss it.  In fact, we
813                  * need to toss every packet with the EOP bit clear and the
814                  * next frame that _does_ have the EOP bit set, as it is by
815                  * definition only a frame fragment
816                  */
817                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
818                         adapter->flags2 |= FLAG2_IS_DISCARDING;
819
820                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
821                         /* All receives must fit into a single buffer */
822                         e_dbg("Receive packet consumed multiple buffers\n");
823                         /* recycle */
824                         buffer_info->skb = skb;
825                         if (status & E1000_RXD_STAT_EOP)
826                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
827                         goto next_desc;
828                 }
829
830                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
831                         /* recycle */
832                         buffer_info->skb = skb;
833                         goto next_desc;
834                 }
835
836                 /* adjust length to remove Ethernet CRC */
837                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
838                         length -= 4;
839
840                 total_rx_bytes += length;
841                 total_rx_packets++;
842
843                 /*
844                  * code added for copybreak, this should improve
845                  * performance for small packets with large amounts
846                  * of reassembly being done in the stack
847                  */
848                 if (length < copybreak) {
849                         struct sk_buff *new_skb =
850                             netdev_alloc_skb_ip_align(netdev, length);
851                         if (new_skb) {
852                                 skb_copy_to_linear_data_offset(new_skb,
853                                                                -NET_IP_ALIGN,
854                                                                (skb->data -
855                                                                 NET_IP_ALIGN),
856                                                                (length +
857                                                                 NET_IP_ALIGN));
858                                 /* save the skb in buffer_info as good */
859                                 buffer_info->skb = skb;
860                                 skb = new_skb;
861                         }
862                         /* else just continue with the old one */
863                 }
864                 /* end copybreak code */
865                 skb_put(skb, length);
866
867                 /* Receive Checksum Offload */
868                 e1000_rx_checksum(adapter,
869                                   (u32)(status) |
870                                   ((u32)(rx_desc->errors) << 24),
871                                   le16_to_cpu(rx_desc->csum), skb);
872
873                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
874
875 next_desc:
876                 rx_desc->status = 0;
877
878                 /* return some buffers to hardware, one at a time is too slow */
879                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
880                         adapter->alloc_rx_buf(adapter, cleaned_count);
881                         cleaned_count = 0;
882                 }
883
884                 /* use prefetched values */
885                 rx_desc = next_rxd;
886                 buffer_info = next_buffer;
887         }
888         rx_ring->next_to_clean = i;
889
890         cleaned_count = e1000_desc_unused(rx_ring);
891         if (cleaned_count)
892                 adapter->alloc_rx_buf(adapter, cleaned_count);
893
894         adapter->total_rx_bytes += total_rx_bytes;
895         adapter->total_rx_packets += total_rx_packets;
896         netdev->stats.rx_bytes += total_rx_bytes;
897         netdev->stats.rx_packets += total_rx_packets;
898         return cleaned;
899 }
900
901 static void e1000_put_txbuf(struct e1000_adapter *adapter,
902                              struct e1000_buffer *buffer_info)
903 {
904         if (buffer_info->dma) {
905                 if (buffer_info->mapped_as_page)
906                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
907                                        buffer_info->length, DMA_TO_DEVICE);
908                 else
909                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
910                                          buffer_info->length, DMA_TO_DEVICE);
911                 buffer_info->dma = 0;
912         }
913         if (buffer_info->skb) {
914                 dev_kfree_skb_any(buffer_info->skb);
915                 buffer_info->skb = NULL;
916         }
917         buffer_info->time_stamp = 0;
918 }
919
920 static void e1000_print_hw_hang(struct work_struct *work)
921 {
922         struct e1000_adapter *adapter = container_of(work,
923                                                      struct e1000_adapter,
924                                                      print_hang_task);
925         struct e1000_ring *tx_ring = adapter->tx_ring;
926         unsigned int i = tx_ring->next_to_clean;
927         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
928         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
929         struct e1000_hw *hw = &adapter->hw;
930         u16 phy_status, phy_1000t_status, phy_ext_status;
931         u16 pci_status;
932
933         e1e_rphy(hw, PHY_STATUS, &phy_status);
934         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
935         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
936
937         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
938
939         /* detected Hardware unit hang */
940         e_err("Detected Hardware Unit Hang:\n"
941               "  TDH                  <%x>\n"
942               "  TDT                  <%x>\n"
943               "  next_to_use          <%x>\n"
944               "  next_to_clean        <%x>\n"
945               "buffer_info[next_to_clean]:\n"
946               "  time_stamp           <%lx>\n"
947               "  next_to_watch        <%x>\n"
948               "  jiffies              <%lx>\n"
949               "  next_to_watch.status <%x>\n"
950               "MAC Status             <%x>\n"
951               "PHY Status             <%x>\n"
952               "PHY 1000BASE-T Status  <%x>\n"
953               "PHY Extended Status    <%x>\n"
954               "PCI Status             <%x>\n",
955               readl(adapter->hw.hw_addr + tx_ring->head),
956               readl(adapter->hw.hw_addr + tx_ring->tail),
957               tx_ring->next_to_use,
958               tx_ring->next_to_clean,
959               tx_ring->buffer_info[eop].time_stamp,
960               eop,
961               jiffies,
962               eop_desc->upper.fields.status,
963               er32(STATUS),
964               phy_status,
965               phy_1000t_status,
966               phy_ext_status,
967               pci_status);
968 }
969
970 /**
971  * e1000_clean_tx_irq - Reclaim resources after transmit completes
972  * @adapter: board private structure
973  *
974  * the return value indicates whether actual cleaning was done, there
975  * is no guarantee that everything was cleaned
976  **/
977 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
978 {
979         struct net_device *netdev = adapter->netdev;
980         struct e1000_hw *hw = &adapter->hw;
981         struct e1000_ring *tx_ring = adapter->tx_ring;
982         struct e1000_tx_desc *tx_desc, *eop_desc;
983         struct e1000_buffer *buffer_info;
984         unsigned int i, eop;
985         unsigned int count = 0;
986         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
987
988         i = tx_ring->next_to_clean;
989         eop = tx_ring->buffer_info[i].next_to_watch;
990         eop_desc = E1000_TX_DESC(*tx_ring, eop);
991
992         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
993                (count < tx_ring->count)) {
994                 bool cleaned = false;
995                 rmb(); /* read buffer_info after eop_desc */
996                 for (; !cleaned; count++) {
997                         tx_desc = E1000_TX_DESC(*tx_ring, i);
998                         buffer_info = &tx_ring->buffer_info[i];
999                         cleaned = (i == eop);
1000
1001                         if (cleaned) {
1002                                 total_tx_packets += buffer_info->segs;
1003                                 total_tx_bytes += buffer_info->bytecount;
1004                         }
1005
1006                         e1000_put_txbuf(adapter, buffer_info);
1007                         tx_desc->upper.data = 0;
1008
1009                         i++;
1010                         if (i == tx_ring->count)
1011                                 i = 0;
1012                 }
1013
1014                 if (i == tx_ring->next_to_use)
1015                         break;
1016                 eop = tx_ring->buffer_info[i].next_to_watch;
1017                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1018         }
1019
1020         tx_ring->next_to_clean = i;
1021
1022 #define TX_WAKE_THRESHOLD 32
1023         if (count && netif_carrier_ok(netdev) &&
1024             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1025                 /* Make sure that anybody stopping the queue after this
1026                  * sees the new next_to_clean.
1027                  */
1028                 smp_mb();
1029
1030                 if (netif_queue_stopped(netdev) &&
1031                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1032                         netif_wake_queue(netdev);
1033                         ++adapter->restart_queue;
1034                 }
1035         }
1036
1037         if (adapter->detect_tx_hung) {
1038                 /*
1039                  * Detect a transmit hang in hardware, this serializes the
1040                  * check with the clearing of time_stamp and movement of i
1041                  */
1042                 adapter->detect_tx_hung = 0;
1043                 if (tx_ring->buffer_info[i].time_stamp &&
1044                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1045                                + (adapter->tx_timeout_factor * HZ)) &&
1046                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1047                         schedule_work(&adapter->print_hang_task);
1048                         netif_stop_queue(netdev);
1049                 }
1050         }
1051         adapter->total_tx_bytes += total_tx_bytes;
1052         adapter->total_tx_packets += total_tx_packets;
1053         netdev->stats.tx_bytes += total_tx_bytes;
1054         netdev->stats.tx_packets += total_tx_packets;
1055         return (count < tx_ring->count);
1056 }
1057
1058 /**
1059  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1060  * @adapter: board private structure
1061  *
1062  * the return value indicates whether actual cleaning was done, there
1063  * is no guarantee that everything was cleaned
1064  **/
1065 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1066                                   int *work_done, int work_to_do)
1067 {
1068         struct e1000_hw *hw = &adapter->hw;
1069         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1070         struct net_device *netdev = adapter->netdev;
1071         struct pci_dev *pdev = adapter->pdev;
1072         struct e1000_ring *rx_ring = adapter->rx_ring;
1073         struct e1000_buffer *buffer_info, *next_buffer;
1074         struct e1000_ps_page *ps_page;
1075         struct sk_buff *skb;
1076         unsigned int i, j;
1077         u32 length, staterr;
1078         int cleaned_count = 0;
1079         bool cleaned = 0;
1080         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1081
1082         i = rx_ring->next_to_clean;
1083         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1084         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1085         buffer_info = &rx_ring->buffer_info[i];
1086
1087         while (staterr & E1000_RXD_STAT_DD) {
1088                 if (*work_done >= work_to_do)
1089                         break;
1090                 (*work_done)++;
1091                 skb = buffer_info->skb;
1092                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1093
1094                 /* in the packet split case this is header only */
1095                 prefetch(skb->data - NET_IP_ALIGN);
1096
1097                 i++;
1098                 if (i == rx_ring->count)
1099                         i = 0;
1100                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1101                 prefetch(next_rxd);
1102
1103                 next_buffer = &rx_ring->buffer_info[i];
1104
1105                 cleaned = 1;
1106                 cleaned_count++;
1107                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1108                                  adapter->rx_ps_bsize0,
1109                                  DMA_FROM_DEVICE);
1110                 buffer_info->dma = 0;
1111
1112                 /* see !EOP comment in other rx routine */
1113                 if (!(staterr & E1000_RXD_STAT_EOP))
1114                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1115
1116                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1117                         e_dbg("Packet Split buffers didn't pick up the full "
1118                               "packet\n");
1119                         dev_kfree_skb_irq(skb);
1120                         if (staterr & E1000_RXD_STAT_EOP)
1121                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1122                         goto next_desc;
1123                 }
1124
1125                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1126                         dev_kfree_skb_irq(skb);
1127                         goto next_desc;
1128                 }
1129
1130                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1131
1132                 if (!length) {
1133                         e_dbg("Last part of the packet spanning multiple "
1134                               "descriptors\n");
1135                         dev_kfree_skb_irq(skb);
1136                         goto next_desc;
1137                 }
1138
1139                 /* Good Receive */
1140                 skb_put(skb, length);
1141
1142                 {
1143                 /*
1144                  * this looks ugly, but it seems compiler issues make it
1145                  * more efficient than reusing j
1146                  */
1147                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1148
1149                 /*
1150                  * page alloc/put takes too long and effects small packet
1151                  * throughput, so unsplit small packets and save the alloc/put
1152                  * only valid in softirq (napi) context to call kmap_*
1153                  */
1154                 if (l1 && (l1 <= copybreak) &&
1155                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1156                         u8 *vaddr;
1157
1158                         ps_page = &buffer_info->ps_pages[0];
1159
1160                         /*
1161                          * there is no documentation about how to call
1162                          * kmap_atomic, so we can't hold the mapping
1163                          * very long
1164                          */
1165                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1166                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1167                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1168                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1169                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1170                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1171                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1172
1173                         /* remove the CRC */
1174                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1175                                 l1 -= 4;
1176
1177                         skb_put(skb, l1);
1178                         goto copydone;
1179                 } /* if */
1180                 }
1181
1182                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1183                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1184                         if (!length)
1185                                 break;
1186
1187                         ps_page = &buffer_info->ps_pages[j];
1188                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1189                                        DMA_FROM_DEVICE);
1190                         ps_page->dma = 0;
1191                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1192                         ps_page->page = NULL;
1193                         skb->len += length;
1194                         skb->data_len += length;
1195                         skb->truesize += length;
1196                 }
1197
1198                 /* strip the ethernet crc, problem is we're using pages now so
1199                  * this whole operation can get a little cpu intensive
1200                  */
1201                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1202                         pskb_trim(skb, skb->len - 4);
1203
1204 copydone:
1205                 total_rx_bytes += skb->len;
1206                 total_rx_packets++;
1207
1208                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1209                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1210
1211                 if (rx_desc->wb.upper.header_status &
1212                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1213                         adapter->rx_hdr_split++;
1214
1215                 e1000_receive_skb(adapter, netdev, skb,
1216                                   staterr, rx_desc->wb.middle.vlan);
1217
1218 next_desc:
1219                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1220                 buffer_info->skb = NULL;
1221
1222                 /* return some buffers to hardware, one at a time is too slow */
1223                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1224                         adapter->alloc_rx_buf(adapter, cleaned_count);
1225                         cleaned_count = 0;
1226                 }
1227
1228                 /* use prefetched values */
1229                 rx_desc = next_rxd;
1230                 buffer_info = next_buffer;
1231
1232                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1233         }
1234         rx_ring->next_to_clean = i;
1235
1236         cleaned_count = e1000_desc_unused(rx_ring);
1237         if (cleaned_count)
1238                 adapter->alloc_rx_buf(adapter, cleaned_count);
1239
1240         adapter->total_rx_bytes += total_rx_bytes;
1241         adapter->total_rx_packets += total_rx_packets;
1242         netdev->stats.rx_bytes += total_rx_bytes;
1243         netdev->stats.rx_packets += total_rx_packets;
1244         return cleaned;
1245 }
1246
1247 /**
1248  * e1000_consume_page - helper function
1249  **/
1250 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1251                                u16 length)
1252 {
1253         bi->page = NULL;
1254         skb->len += length;
1255         skb->data_len += length;
1256         skb->truesize += length;
1257 }
1258
1259 /**
1260  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1261  * @adapter: board private structure
1262  *
1263  * the return value indicates whether actual cleaning was done, there
1264  * is no guarantee that everything was cleaned
1265  **/
1266
1267 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1268                                      int *work_done, int work_to_do)
1269 {
1270         struct net_device *netdev = adapter->netdev;
1271         struct pci_dev *pdev = adapter->pdev;
1272         struct e1000_ring *rx_ring = adapter->rx_ring;
1273         struct e1000_rx_desc *rx_desc, *next_rxd;
1274         struct e1000_buffer *buffer_info, *next_buffer;
1275         u32 length;
1276         unsigned int i;
1277         int cleaned_count = 0;
1278         bool cleaned = false;
1279         unsigned int total_rx_bytes=0, total_rx_packets=0;
1280
1281         i = rx_ring->next_to_clean;
1282         rx_desc = E1000_RX_DESC(*rx_ring, i);
1283         buffer_info = &rx_ring->buffer_info[i];
1284
1285         while (rx_desc->status & E1000_RXD_STAT_DD) {
1286                 struct sk_buff *skb;
1287                 u8 status;
1288
1289                 if (*work_done >= work_to_do)
1290                         break;
1291                 (*work_done)++;
1292                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1293
1294                 status = rx_desc->status;
1295                 skb = buffer_info->skb;
1296                 buffer_info->skb = NULL;
1297
1298                 ++i;
1299                 if (i == rx_ring->count)
1300                         i = 0;
1301                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1302                 prefetch(next_rxd);
1303
1304                 next_buffer = &rx_ring->buffer_info[i];
1305
1306                 cleaned = true;
1307                 cleaned_count++;
1308                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1309                                DMA_FROM_DEVICE);
1310                 buffer_info->dma = 0;
1311
1312                 length = le16_to_cpu(rx_desc->length);
1313
1314                 /* errors is only valid for DD + EOP descriptors */
1315                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1316                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1317                                 /* recycle both page and skb */
1318                                 buffer_info->skb = skb;
1319                                 /* an error means any chain goes out the window
1320                                  * too */
1321                                 if (rx_ring->rx_skb_top)
1322                                         dev_kfree_skb(rx_ring->rx_skb_top);
1323                                 rx_ring->rx_skb_top = NULL;
1324                                 goto next_desc;
1325                 }
1326
1327 #define rxtop rx_ring->rx_skb_top
1328                 if (!(status & E1000_RXD_STAT_EOP)) {
1329                         /* this descriptor is only the beginning (or middle) */
1330                         if (!rxtop) {
1331                                 /* this is the beginning of a chain */
1332                                 rxtop = skb;
1333                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1334                                                    0, length);
1335                         } else {
1336                                 /* this is the middle of a chain */
1337                                 skb_fill_page_desc(rxtop,
1338                                     skb_shinfo(rxtop)->nr_frags,
1339                                     buffer_info->page, 0, length);
1340                                 /* re-use the skb, only consumed the page */
1341                                 buffer_info->skb = skb;
1342                         }
1343                         e1000_consume_page(buffer_info, rxtop, length);
1344                         goto next_desc;
1345                 } else {
1346                         if (rxtop) {
1347                                 /* end of the chain */
1348                                 skb_fill_page_desc(rxtop,
1349                                     skb_shinfo(rxtop)->nr_frags,
1350                                     buffer_info->page, 0, length);
1351                                 /* re-use the current skb, we only consumed the
1352                                  * page */
1353                                 buffer_info->skb = skb;
1354                                 skb = rxtop;
1355                                 rxtop = NULL;
1356                                 e1000_consume_page(buffer_info, skb, length);
1357                         } else {
1358                                 /* no chain, got EOP, this buf is the packet
1359                                  * copybreak to save the put_page/alloc_page */
1360                                 if (length <= copybreak &&
1361                                     skb_tailroom(skb) >= length) {
1362                                         u8 *vaddr;
1363                                         vaddr = kmap_atomic(buffer_info->page,
1364                                                            KM_SKB_DATA_SOFTIRQ);
1365                                         memcpy(skb_tail_pointer(skb), vaddr,
1366                                                length);
1367                                         kunmap_atomic(vaddr,
1368                                                       KM_SKB_DATA_SOFTIRQ);
1369                                         /* re-use the page, so don't erase
1370                                          * buffer_info->page */
1371                                         skb_put(skb, length);
1372                                 } else {
1373                                         skb_fill_page_desc(skb, 0,
1374                                                            buffer_info->page, 0,
1375                                                            length);
1376                                         e1000_consume_page(buffer_info, skb,
1377                                                            length);
1378                                 }
1379                         }
1380                 }
1381
1382                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1383                 e1000_rx_checksum(adapter,
1384                                   (u32)(status) |
1385                                   ((u32)(rx_desc->errors) << 24),
1386                                   le16_to_cpu(rx_desc->csum), skb);
1387
1388                 /* probably a little skewed due to removing CRC */
1389                 total_rx_bytes += skb->len;
1390                 total_rx_packets++;
1391
1392                 /* eth type trans needs skb->data to point to something */
1393                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1394                         e_err("pskb_may_pull failed.\n");
1395                         dev_kfree_skb(skb);
1396                         goto next_desc;
1397                 }
1398
1399                 e1000_receive_skb(adapter, netdev, skb, status,
1400                                   rx_desc->special);
1401
1402 next_desc:
1403                 rx_desc->status = 0;
1404
1405                 /* return some buffers to hardware, one at a time is too slow */
1406                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1407                         adapter->alloc_rx_buf(adapter, cleaned_count);
1408                         cleaned_count = 0;
1409                 }
1410
1411                 /* use prefetched values */
1412                 rx_desc = next_rxd;
1413                 buffer_info = next_buffer;
1414         }
1415         rx_ring->next_to_clean = i;
1416
1417         cleaned_count = e1000_desc_unused(rx_ring);
1418         if (cleaned_count)
1419                 adapter->alloc_rx_buf(adapter, cleaned_count);
1420
1421         adapter->total_rx_bytes += total_rx_bytes;
1422         adapter->total_rx_packets += total_rx_packets;
1423         netdev->stats.rx_bytes += total_rx_bytes;
1424         netdev->stats.rx_packets += total_rx_packets;
1425         return cleaned;
1426 }
1427
1428 /**
1429  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1430  * @adapter: board private structure
1431  **/
1432 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1433 {
1434         struct e1000_ring *rx_ring = adapter->rx_ring;
1435         struct e1000_buffer *buffer_info;
1436         struct e1000_ps_page *ps_page;
1437         struct pci_dev *pdev = adapter->pdev;
1438         unsigned int i, j;
1439
1440         /* Free all the Rx ring sk_buffs */
1441         for (i = 0; i < rx_ring->count; i++) {
1442                 buffer_info = &rx_ring->buffer_info[i];
1443                 if (buffer_info->dma) {
1444                         if (adapter->clean_rx == e1000_clean_rx_irq)
1445                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1446                                                  adapter->rx_buffer_len,
1447                                                  DMA_FROM_DEVICE);
1448                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1449                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1450                                                PAGE_SIZE,
1451                                                DMA_FROM_DEVICE);
1452                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1453                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1454                                                  adapter->rx_ps_bsize0,
1455                                                  DMA_FROM_DEVICE);
1456                         buffer_info->dma = 0;
1457                 }
1458
1459                 if (buffer_info->page) {
1460                         put_page(buffer_info->page);
1461                         buffer_info->page = NULL;
1462                 }
1463
1464                 if (buffer_info->skb) {
1465                         dev_kfree_skb(buffer_info->skb);
1466                         buffer_info->skb = NULL;
1467                 }
1468
1469                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1470                         ps_page = &buffer_info->ps_pages[j];
1471                         if (!ps_page->page)
1472                                 break;
1473                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1474                                        DMA_FROM_DEVICE);
1475                         ps_page->dma = 0;
1476                         put_page(ps_page->page);
1477                         ps_page->page = NULL;
1478                 }
1479         }
1480
1481         /* there also may be some cached data from a chained receive */
1482         if (rx_ring->rx_skb_top) {
1483                 dev_kfree_skb(rx_ring->rx_skb_top);
1484                 rx_ring->rx_skb_top = NULL;
1485         }
1486
1487         /* Zero out the descriptor ring */
1488         memset(rx_ring->desc, 0, rx_ring->size);
1489
1490         rx_ring->next_to_clean = 0;
1491         rx_ring->next_to_use = 0;
1492         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1493
1494         writel(0, adapter->hw.hw_addr + rx_ring->head);
1495         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1496 }
1497
1498 static void e1000e_downshift_workaround(struct work_struct *work)
1499 {
1500         struct e1000_adapter *adapter = container_of(work,
1501                                         struct e1000_adapter, downshift_task);
1502
1503         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1504 }
1505
1506 /**
1507  * e1000_intr_msi - Interrupt Handler
1508  * @irq: interrupt number
1509  * @data: pointer to a network interface device structure
1510  **/
1511 static irqreturn_t e1000_intr_msi(int irq, void *data)
1512 {
1513         struct net_device *netdev = data;
1514         struct e1000_adapter *adapter = netdev_priv(netdev);
1515         struct e1000_hw *hw = &adapter->hw;
1516         u32 icr = er32(ICR);
1517
1518         /*
1519          * read ICR disables interrupts using IAM
1520          */
1521
1522         if (icr & E1000_ICR_LSC) {
1523                 hw->mac.get_link_status = 1;
1524                 /*
1525                  * ICH8 workaround-- Call gig speed drop workaround on cable
1526                  * disconnect (LSC) before accessing any PHY registers
1527                  */
1528                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1529                     (!(er32(STATUS) & E1000_STATUS_LU)))
1530                         schedule_work(&adapter->downshift_task);
1531
1532                 /*
1533                  * 80003ES2LAN workaround-- For packet buffer work-around on
1534                  * link down event; disable receives here in the ISR and reset
1535                  * adapter in watchdog
1536                  */
1537                 if (netif_carrier_ok(netdev) &&
1538                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1539                         /* disable receives */
1540                         u32 rctl = er32(RCTL);
1541                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1542                         adapter->flags |= FLAG_RX_RESTART_NOW;
1543                 }
1544                 /* guard against interrupt when we're going down */
1545                 if (!test_bit(__E1000_DOWN, &adapter->state))
1546                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1547         }
1548
1549         if (napi_schedule_prep(&adapter->napi)) {
1550                 adapter->total_tx_bytes = 0;
1551                 adapter->total_tx_packets = 0;
1552                 adapter->total_rx_bytes = 0;
1553                 adapter->total_rx_packets = 0;
1554                 __napi_schedule(&adapter->napi);
1555         }
1556
1557         return IRQ_HANDLED;
1558 }
1559
1560 /**
1561  * e1000_intr - Interrupt Handler
1562  * @irq: interrupt number
1563  * @data: pointer to a network interface device structure
1564  **/
1565 static irqreturn_t e1000_intr(int irq, void *data)
1566 {
1567         struct net_device *netdev = data;
1568         struct e1000_adapter *adapter = netdev_priv(netdev);
1569         struct e1000_hw *hw = &adapter->hw;
1570         u32 rctl, icr = er32(ICR);
1571
1572         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1573                 return IRQ_NONE;  /* Not our interrupt */
1574
1575         /*
1576          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1577          * not set, then the adapter didn't send an interrupt
1578          */
1579         if (!(icr & E1000_ICR_INT_ASSERTED))
1580                 return IRQ_NONE;
1581
1582         /*
1583          * Interrupt Auto-Mask...upon reading ICR,
1584          * interrupts are masked.  No need for the
1585          * IMC write
1586          */
1587
1588         if (icr & E1000_ICR_LSC) {
1589                 hw->mac.get_link_status = 1;
1590                 /*
1591                  * ICH8 workaround-- Call gig speed drop workaround on cable
1592                  * disconnect (LSC) before accessing any PHY registers
1593                  */
1594                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1595                     (!(er32(STATUS) & E1000_STATUS_LU)))
1596                         schedule_work(&adapter->downshift_task);
1597
1598                 /*
1599                  * 80003ES2LAN workaround--
1600                  * For packet buffer work-around on link down event;
1601                  * disable receives here in the ISR and
1602                  * reset adapter in watchdog
1603                  */
1604                 if (netif_carrier_ok(netdev) &&
1605                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1606                         /* disable receives */
1607                         rctl = er32(RCTL);
1608                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1609                         adapter->flags |= FLAG_RX_RESTART_NOW;
1610                 }
1611                 /* guard against interrupt when we're going down */
1612                 if (!test_bit(__E1000_DOWN, &adapter->state))
1613                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1614         }
1615
1616         if (napi_schedule_prep(&adapter->napi)) {
1617                 adapter->total_tx_bytes = 0;
1618                 adapter->total_tx_packets = 0;
1619                 adapter->total_rx_bytes = 0;
1620                 adapter->total_rx_packets = 0;
1621                 __napi_schedule(&adapter->napi);
1622         }
1623
1624         return IRQ_HANDLED;
1625 }
1626
1627 static irqreturn_t e1000_msix_other(int irq, void *data)
1628 {
1629         struct net_device *netdev = data;
1630         struct e1000_adapter *adapter = netdev_priv(netdev);
1631         struct e1000_hw *hw = &adapter->hw;
1632         u32 icr = er32(ICR);
1633
1634         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1635                 if (!test_bit(__E1000_DOWN, &adapter->state))
1636                         ew32(IMS, E1000_IMS_OTHER);
1637                 return IRQ_NONE;
1638         }
1639
1640         if (icr & adapter->eiac_mask)
1641                 ew32(ICS, (icr & adapter->eiac_mask));
1642
1643         if (icr & E1000_ICR_OTHER) {
1644                 if (!(icr & E1000_ICR_LSC))
1645                         goto no_link_interrupt;
1646                 hw->mac.get_link_status = 1;
1647                 /* guard against interrupt when we're going down */
1648                 if (!test_bit(__E1000_DOWN, &adapter->state))
1649                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1650         }
1651
1652 no_link_interrupt:
1653         if (!test_bit(__E1000_DOWN, &adapter->state))
1654                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1655
1656         return IRQ_HANDLED;
1657 }
1658
1659
1660 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1661 {
1662         struct net_device *netdev = data;
1663         struct e1000_adapter *adapter = netdev_priv(netdev);
1664         struct e1000_hw *hw = &adapter->hw;
1665         struct e1000_ring *tx_ring = adapter->tx_ring;
1666
1667
1668         adapter->total_tx_bytes = 0;
1669         adapter->total_tx_packets = 0;
1670
1671         if (!e1000_clean_tx_irq(adapter))
1672                 /* Ring was not completely cleaned, so fire another interrupt */
1673                 ew32(ICS, tx_ring->ims_val);
1674
1675         return IRQ_HANDLED;
1676 }
1677
1678 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1679 {
1680         struct net_device *netdev = data;
1681         struct e1000_adapter *adapter = netdev_priv(netdev);
1682
1683         /* Write the ITR value calculated at the end of the
1684          * previous interrupt.
1685          */
1686         if (adapter->rx_ring->set_itr) {
1687                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1688                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1689                 adapter->rx_ring->set_itr = 0;
1690         }
1691
1692         if (napi_schedule_prep(&adapter->napi)) {
1693                 adapter->total_rx_bytes = 0;
1694                 adapter->total_rx_packets = 0;
1695                 __napi_schedule(&adapter->napi);
1696         }
1697         return IRQ_HANDLED;
1698 }
1699
1700 /**
1701  * e1000_configure_msix - Configure MSI-X hardware
1702  *
1703  * e1000_configure_msix sets up the hardware to properly
1704  * generate MSI-X interrupts.
1705  **/
1706 static void e1000_configure_msix(struct e1000_adapter *adapter)
1707 {
1708         struct e1000_hw *hw = &adapter->hw;
1709         struct e1000_ring *rx_ring = adapter->rx_ring;
1710         struct e1000_ring *tx_ring = adapter->tx_ring;
1711         int vector = 0;
1712         u32 ctrl_ext, ivar = 0;
1713
1714         adapter->eiac_mask = 0;
1715
1716         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1717         if (hw->mac.type == e1000_82574) {
1718                 u32 rfctl = er32(RFCTL);
1719                 rfctl |= E1000_RFCTL_ACK_DIS;
1720                 ew32(RFCTL, rfctl);
1721         }
1722
1723 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1724         /* Configure Rx vector */
1725         rx_ring->ims_val = E1000_IMS_RXQ0;
1726         adapter->eiac_mask |= rx_ring->ims_val;
1727         if (rx_ring->itr_val)
1728                 writel(1000000000 / (rx_ring->itr_val * 256),
1729                        hw->hw_addr + rx_ring->itr_register);
1730         else
1731                 writel(1, hw->hw_addr + rx_ring->itr_register);
1732         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1733
1734         /* Configure Tx vector */
1735         tx_ring->ims_val = E1000_IMS_TXQ0;
1736         vector++;
1737         if (tx_ring->itr_val)
1738                 writel(1000000000 / (tx_ring->itr_val * 256),
1739                        hw->hw_addr + tx_ring->itr_register);
1740         else
1741                 writel(1, hw->hw_addr + tx_ring->itr_register);
1742         adapter->eiac_mask |= tx_ring->ims_val;
1743         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1744
1745         /* set vector for Other Causes, e.g. link changes */
1746         vector++;
1747         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1748         if (rx_ring->itr_val)
1749                 writel(1000000000 / (rx_ring->itr_val * 256),
1750                        hw->hw_addr + E1000_EITR_82574(vector));
1751         else
1752                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1753
1754         /* Cause Tx interrupts on every write back */
1755         ivar |= (1 << 31);
1756
1757         ew32(IVAR, ivar);
1758
1759         /* enable MSI-X PBA support */
1760         ctrl_ext = er32(CTRL_EXT);
1761         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1762
1763         /* Auto-Mask Other interrupts upon ICR read */
1764 #define E1000_EIAC_MASK_82574   0x01F00000
1765         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1766         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1767         ew32(CTRL_EXT, ctrl_ext);
1768         e1e_flush();
1769 }
1770
1771 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1772 {
1773         if (adapter->msix_entries) {
1774                 pci_disable_msix(adapter->pdev);
1775                 kfree(adapter->msix_entries);
1776                 adapter->msix_entries = NULL;
1777         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1778                 pci_disable_msi(adapter->pdev);
1779                 adapter->flags &= ~FLAG_MSI_ENABLED;
1780         }
1781 }
1782
1783 /**
1784  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1785  *
1786  * Attempt to configure interrupts using the best available
1787  * capabilities of the hardware and kernel.
1788  **/
1789 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1790 {
1791         int err;
1792         int i;
1793
1794         switch (adapter->int_mode) {
1795         case E1000E_INT_MODE_MSIX:
1796                 if (adapter->flags & FLAG_HAS_MSIX) {
1797                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1798                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1799                                                       sizeof(struct msix_entry),
1800                                                       GFP_KERNEL);
1801                         if (adapter->msix_entries) {
1802                                 for (i = 0; i < adapter->num_vectors; i++)
1803                                         adapter->msix_entries[i].entry = i;
1804
1805                                 err = pci_enable_msix(adapter->pdev,
1806                                                       adapter->msix_entries,
1807                                                       adapter->num_vectors);
1808                                 if (err == 0) {
1809                                         return;
1810                                 }
1811                         }
1812                         /* MSI-X failed, so fall through and try MSI */
1813                         e_err("Failed to initialize MSI-X interrupts.  "
1814                               "Falling back to MSI interrupts.\n");
1815                         e1000e_reset_interrupt_capability(adapter);
1816                 }
1817                 adapter->int_mode = E1000E_INT_MODE_MSI;
1818                 /* Fall through */
1819         case E1000E_INT_MODE_MSI:
1820                 if (!pci_enable_msi(adapter->pdev)) {
1821                         adapter->flags |= FLAG_MSI_ENABLED;
1822                 } else {
1823                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1824                         e_err("Failed to initialize MSI interrupts.  Falling "
1825                               "back to legacy interrupts.\n");
1826                 }
1827                 /* Fall through */
1828         case E1000E_INT_MODE_LEGACY:
1829                 /* Don't do anything; this is the system default */
1830                 break;
1831         }
1832
1833         /* store the number of vectors being used */
1834         adapter->num_vectors = 1;
1835 }
1836
1837 /**
1838  * e1000_request_msix - Initialize MSI-X interrupts
1839  *
1840  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1841  * kernel.
1842  **/
1843 static int e1000_request_msix(struct e1000_adapter *adapter)
1844 {
1845         struct net_device *netdev = adapter->netdev;
1846         int err = 0, vector = 0;
1847
1848         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1849                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1850         else
1851                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1852         err = request_irq(adapter->msix_entries[vector].vector,
1853                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1854                           netdev);
1855         if (err)
1856                 goto out;
1857         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1858         adapter->rx_ring->itr_val = adapter->itr;
1859         vector++;
1860
1861         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1862                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1863         else
1864                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1865         err = request_irq(adapter->msix_entries[vector].vector,
1866                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1867                           netdev);
1868         if (err)
1869                 goto out;
1870         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1871         adapter->tx_ring->itr_val = adapter->itr;
1872         vector++;
1873
1874         err = request_irq(adapter->msix_entries[vector].vector,
1875                           e1000_msix_other, 0, netdev->name, netdev);
1876         if (err)
1877                 goto out;
1878
1879         e1000_configure_msix(adapter);
1880         return 0;
1881 out:
1882         return err;
1883 }
1884
1885 /**
1886  * e1000_request_irq - initialize interrupts
1887  *
1888  * Attempts to configure interrupts using the best available
1889  * capabilities of the hardware and kernel.
1890  **/
1891 static int e1000_request_irq(struct e1000_adapter *adapter)
1892 {
1893         struct net_device *netdev = adapter->netdev;
1894         int err;
1895
1896         if (adapter->msix_entries) {
1897                 err = e1000_request_msix(adapter);
1898                 if (!err)
1899                         return err;
1900                 /* fall back to MSI */
1901                 e1000e_reset_interrupt_capability(adapter);
1902                 adapter->int_mode = E1000E_INT_MODE_MSI;
1903                 e1000e_set_interrupt_capability(adapter);
1904         }
1905         if (adapter->flags & FLAG_MSI_ENABLED) {
1906                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1907                                   netdev->name, netdev);
1908                 if (!err)
1909                         return err;
1910
1911                 /* fall back to legacy interrupt */
1912                 e1000e_reset_interrupt_capability(adapter);
1913                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1914         }
1915
1916         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1917                           netdev->name, netdev);
1918         if (err)
1919                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1920
1921         return err;
1922 }
1923
1924 static void e1000_free_irq(struct e1000_adapter *adapter)
1925 {
1926         struct net_device *netdev = adapter->netdev;
1927
1928         if (adapter->msix_entries) {
1929                 int vector = 0;
1930
1931                 free_irq(adapter->msix_entries[vector].vector, netdev);
1932                 vector++;
1933
1934                 free_irq(adapter->msix_entries[vector].vector, netdev);
1935                 vector++;
1936
1937                 /* Other Causes interrupt vector */
1938                 free_irq(adapter->msix_entries[vector].vector, netdev);
1939                 return;
1940         }
1941
1942         free_irq(adapter->pdev->irq, netdev);
1943 }
1944
1945 /**
1946  * e1000_irq_disable - Mask off interrupt generation on the NIC
1947  **/
1948 static void e1000_irq_disable(struct e1000_adapter *adapter)
1949 {
1950         struct e1000_hw *hw = &adapter->hw;
1951
1952         ew32(IMC, ~0);
1953         if (adapter->msix_entries)
1954                 ew32(EIAC_82574, 0);
1955         e1e_flush();
1956
1957         if (adapter->msix_entries) {
1958                 int i;
1959                 for (i = 0; i < adapter->num_vectors; i++)
1960                         synchronize_irq(adapter->msix_entries[i].vector);
1961         } else {
1962                 synchronize_irq(adapter->pdev->irq);
1963         }
1964 }
1965
1966 /**
1967  * e1000_irq_enable - Enable default interrupt generation settings
1968  **/
1969 static void e1000_irq_enable(struct e1000_adapter *adapter)
1970 {
1971         struct e1000_hw *hw = &adapter->hw;
1972
1973         if (adapter->msix_entries) {
1974                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1975                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1976         } else {
1977                 ew32(IMS, IMS_ENABLE_MASK);
1978         }
1979         e1e_flush();
1980 }
1981
1982 /**
1983  * e1000_get_hw_control - get control of the h/w from f/w
1984  * @adapter: address of board private structure
1985  *
1986  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1987  * For ASF and Pass Through versions of f/w this means that
1988  * the driver is loaded. For AMT version (only with 82573)
1989  * of the f/w this means that the network i/f is open.
1990  **/
1991 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1992 {
1993         struct e1000_hw *hw = &adapter->hw;
1994         u32 ctrl_ext;
1995         u32 swsm;
1996
1997         /* Let firmware know the driver has taken over */
1998         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1999                 swsm = er32(SWSM);
2000                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2001         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2002                 ctrl_ext = er32(CTRL_EXT);
2003                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2004         }
2005 }
2006
2007 /**
2008  * e1000_release_hw_control - release control of the h/w to f/w
2009  * @adapter: address of board private structure
2010  *
2011  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2012  * For ASF and Pass Through versions of f/w this means that the
2013  * driver is no longer loaded. For AMT version (only with 82573) i
2014  * of the f/w this means that the network i/f is closed.
2015  *
2016  **/
2017 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2018 {
2019         struct e1000_hw *hw = &adapter->hw;
2020         u32 ctrl_ext;
2021         u32 swsm;
2022
2023         /* Let firmware taken over control of h/w */
2024         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2025                 swsm = er32(SWSM);
2026                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2027         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2028                 ctrl_ext = er32(CTRL_EXT);
2029                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2030         }
2031 }
2032
2033 /**
2034  * @e1000_alloc_ring - allocate memory for a ring structure
2035  **/
2036 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2037                                 struct e1000_ring *ring)
2038 {
2039         struct pci_dev *pdev = adapter->pdev;
2040
2041         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2042                                         GFP_KERNEL);
2043         if (!ring->desc)
2044                 return -ENOMEM;
2045
2046         return 0;
2047 }
2048
2049 /**
2050  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2051  * @adapter: board private structure
2052  *
2053  * Return 0 on success, negative on failure
2054  **/
2055 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2056 {
2057         struct e1000_ring *tx_ring = adapter->tx_ring;
2058         int err = -ENOMEM, size;
2059
2060         size = sizeof(struct e1000_buffer) * tx_ring->count;
2061         tx_ring->buffer_info = vmalloc(size);
2062         if (!tx_ring->buffer_info)
2063                 goto err;
2064         memset(tx_ring->buffer_info, 0, size);
2065
2066         /* round up to nearest 4K */
2067         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2068         tx_ring->size = ALIGN(tx_ring->size, 4096);
2069
2070         err = e1000_alloc_ring_dma(adapter, tx_ring);
2071         if (err)
2072                 goto err;
2073
2074         tx_ring->next_to_use = 0;
2075         tx_ring->next_to_clean = 0;
2076
2077         return 0;
2078 err:
2079         vfree(tx_ring->buffer_info);
2080         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2081         return err;
2082 }
2083
2084 /**
2085  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2086  * @adapter: board private structure
2087  *
2088  * Returns 0 on success, negative on failure
2089  **/
2090 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2091 {
2092         struct e1000_ring *rx_ring = adapter->rx_ring;
2093         struct e1000_buffer *buffer_info;
2094         int i, size, desc_len, err = -ENOMEM;
2095
2096         size = sizeof(struct e1000_buffer) * rx_ring->count;
2097         rx_ring->buffer_info = vmalloc(size);
2098         if (!rx_ring->buffer_info)
2099                 goto err;
2100         memset(rx_ring->buffer_info, 0, size);
2101
2102         for (i = 0; i < rx_ring->count; i++) {
2103                 buffer_info = &rx_ring->buffer_info[i];
2104                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2105                                                 sizeof(struct e1000_ps_page),
2106                                                 GFP_KERNEL);
2107                 if (!buffer_info->ps_pages)
2108                         goto err_pages;
2109         }
2110
2111         desc_len = sizeof(union e1000_rx_desc_packet_split);
2112
2113         /* Round up to nearest 4K */
2114         rx_ring->size = rx_ring->count * desc_len;
2115         rx_ring->size = ALIGN(rx_ring->size, 4096);
2116
2117         err = e1000_alloc_ring_dma(adapter, rx_ring);
2118         if (err)
2119                 goto err_pages;
2120
2121         rx_ring->next_to_clean = 0;
2122         rx_ring->next_to_use = 0;
2123         rx_ring->rx_skb_top = NULL;
2124
2125         return 0;
2126
2127 err_pages:
2128         for (i = 0; i < rx_ring->count; i++) {
2129                 buffer_info = &rx_ring->buffer_info[i];
2130                 kfree(buffer_info->ps_pages);
2131         }
2132 err:
2133         vfree(rx_ring->buffer_info);
2134         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2135         return err;
2136 }
2137
2138 /**
2139  * e1000_clean_tx_ring - Free Tx Buffers
2140  * @adapter: board private structure
2141  **/
2142 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2143 {
2144         struct e1000_ring *tx_ring = adapter->tx_ring;
2145         struct e1000_buffer *buffer_info;
2146         unsigned long size;
2147         unsigned int i;
2148
2149         for (i = 0; i < tx_ring->count; i++) {
2150                 buffer_info = &tx_ring->buffer_info[i];
2151                 e1000_put_txbuf(adapter, buffer_info);
2152         }
2153
2154         size = sizeof(struct e1000_buffer) * tx_ring->count;
2155         memset(tx_ring->buffer_info, 0, size);
2156
2157         memset(tx_ring->desc, 0, tx_ring->size);
2158
2159         tx_ring->next_to_use = 0;
2160         tx_ring->next_to_clean = 0;
2161
2162         writel(0, adapter->hw.hw_addr + tx_ring->head);
2163         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2164 }
2165
2166 /**
2167  * e1000e_free_tx_resources - Free Tx Resources per Queue
2168  * @adapter: board private structure
2169  *
2170  * Free all transmit software resources
2171  **/
2172 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2173 {
2174         struct pci_dev *pdev = adapter->pdev;
2175         struct e1000_ring *tx_ring = adapter->tx_ring;
2176
2177         e1000_clean_tx_ring(adapter);
2178
2179         vfree(tx_ring->buffer_info);
2180         tx_ring->buffer_info = NULL;
2181
2182         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2183                           tx_ring->dma);
2184         tx_ring->desc = NULL;
2185 }
2186
2187 /**
2188  * e1000e_free_rx_resources - Free Rx Resources
2189  * @adapter: board private structure
2190  *
2191  * Free all receive software resources
2192  **/
2193
2194 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2195 {
2196         struct pci_dev *pdev = adapter->pdev;
2197         struct e1000_ring *rx_ring = adapter->rx_ring;
2198         int i;
2199
2200         e1000_clean_rx_ring(adapter);
2201
2202         for (i = 0; i < rx_ring->count; i++) {
2203                 kfree(rx_ring->buffer_info[i].ps_pages);
2204         }
2205
2206         vfree(rx_ring->buffer_info);
2207         rx_ring->buffer_info = NULL;
2208
2209         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2210                           rx_ring->dma);
2211         rx_ring->desc = NULL;
2212 }
2213
2214 /**
2215  * e1000_update_itr - update the dynamic ITR value based on statistics
2216  * @adapter: pointer to adapter
2217  * @itr_setting: current adapter->itr
2218  * @packets: the number of packets during this measurement interval
2219  * @bytes: the number of bytes during this measurement interval
2220  *
2221  *      Stores a new ITR value based on packets and byte
2222  *      counts during the last interrupt.  The advantage of per interrupt
2223  *      computation is faster updates and more accurate ITR for the current
2224  *      traffic pattern.  Constants in this function were computed
2225  *      based on theoretical maximum wire speed and thresholds were set based
2226  *      on testing data as well as attempting to minimize response time
2227  *      while increasing bulk throughput.  This functionality is controlled
2228  *      by the InterruptThrottleRate module parameter.
2229  **/
2230 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2231                                      u16 itr_setting, int packets,
2232                                      int bytes)
2233 {
2234         unsigned int retval = itr_setting;
2235
2236         if (packets == 0)
2237                 goto update_itr_done;
2238
2239         switch (itr_setting) {
2240         case lowest_latency:
2241                 /* handle TSO and jumbo frames */
2242                 if (bytes/packets > 8000)
2243                         retval = bulk_latency;
2244                 else if ((packets < 5) && (bytes > 512)) {
2245                         retval = low_latency;
2246                 }
2247                 break;
2248         case low_latency:  /* 50 usec aka 20000 ints/s */
2249                 if (bytes > 10000) {
2250                         /* this if handles the TSO accounting */
2251                         if (bytes/packets > 8000) {
2252                                 retval = bulk_latency;
2253                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2254                                 retval = bulk_latency;
2255                         } else if ((packets > 35)) {
2256                                 retval = lowest_latency;
2257                         }
2258                 } else if (bytes/packets > 2000) {
2259                         retval = bulk_latency;
2260                 } else if (packets <= 2 && bytes < 512) {
2261                         retval = lowest_latency;
2262                 }
2263                 break;
2264         case bulk_latency: /* 250 usec aka 4000 ints/s */
2265                 if (bytes > 25000) {
2266                         if (packets > 35) {
2267                                 retval = low_latency;
2268                         }
2269                 } else if (bytes < 6000) {
2270                         retval = low_latency;
2271                 }
2272                 break;
2273         }
2274
2275 update_itr_done:
2276         return retval;
2277 }
2278
2279 static void e1000_set_itr(struct e1000_adapter *adapter)
2280 {
2281         struct e1000_hw *hw = &adapter->hw;
2282         u16 current_itr;
2283         u32 new_itr = adapter->itr;
2284
2285         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2286         if (adapter->link_speed != SPEED_1000) {
2287                 current_itr = 0;
2288                 new_itr = 4000;
2289                 goto set_itr_now;
2290         }
2291
2292         adapter->tx_itr = e1000_update_itr(adapter,
2293                                     adapter->tx_itr,
2294                                     adapter->total_tx_packets,
2295                                     adapter->total_tx_bytes);
2296         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2297         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2298                 adapter->tx_itr = low_latency;
2299
2300         adapter->rx_itr = e1000_update_itr(adapter,
2301                                     adapter->rx_itr,
2302                                     adapter->total_rx_packets,
2303                                     adapter->total_rx_bytes);
2304         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2305         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2306                 adapter->rx_itr = low_latency;
2307
2308         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2309
2310         switch (current_itr) {
2311         /* counts and packets in update_itr are dependent on these numbers */
2312         case lowest_latency:
2313                 new_itr = 70000;
2314                 break;
2315         case low_latency:
2316                 new_itr = 20000; /* aka hwitr = ~200 */
2317                 break;
2318         case bulk_latency:
2319                 new_itr = 4000;
2320                 break;
2321         default:
2322                 break;
2323         }
2324
2325 set_itr_now:
2326         if (new_itr != adapter->itr) {
2327                 /*
2328                  * this attempts to bias the interrupt rate towards Bulk
2329                  * by adding intermediate steps when interrupt rate is
2330                  * increasing
2331                  */
2332                 new_itr = new_itr > adapter->itr ?
2333                              min(adapter->itr + (new_itr >> 2), new_itr) :
2334                              new_itr;
2335                 adapter->itr = new_itr;
2336                 adapter->rx_ring->itr_val = new_itr;
2337                 if (adapter->msix_entries)
2338                         adapter->rx_ring->set_itr = 1;
2339                 else
2340                         ew32(ITR, 1000000000 / (new_itr * 256));
2341         }
2342 }
2343
2344 /**
2345  * e1000_alloc_queues - Allocate memory for all rings
2346  * @adapter: board private structure to initialize
2347  **/
2348 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2349 {
2350         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2351         if (!adapter->tx_ring)
2352                 goto err;
2353
2354         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2355         if (!adapter->rx_ring)
2356                 goto err;
2357
2358         return 0;
2359 err:
2360         e_err("Unable to allocate memory for queues\n");
2361         kfree(adapter->rx_ring);
2362         kfree(adapter->tx_ring);
2363         return -ENOMEM;
2364 }
2365
2366 /**
2367  * e1000_clean - NAPI Rx polling callback
2368  * @napi: struct associated with this polling callback
2369  * @budget: amount of packets driver is allowed to process this poll
2370  **/
2371 static int e1000_clean(struct napi_struct *napi, int budget)
2372 {
2373         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2374         struct e1000_hw *hw = &adapter->hw;
2375         struct net_device *poll_dev = adapter->netdev;
2376         int tx_cleaned = 1, work_done = 0;
2377
2378         adapter = netdev_priv(poll_dev);
2379
2380         if (adapter->msix_entries &&
2381             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2382                 goto clean_rx;
2383
2384         tx_cleaned = e1000_clean_tx_irq(adapter);
2385
2386 clean_rx:
2387         adapter->clean_rx(adapter, &work_done, budget);
2388
2389         if (!tx_cleaned)
2390                 work_done = budget;
2391
2392         /* If budget not fully consumed, exit the polling mode */
2393         if (work_done < budget) {
2394                 if (adapter->itr_setting & 3)
2395                         e1000_set_itr(adapter);
2396                 napi_complete(napi);
2397                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2398                         if (adapter->msix_entries)
2399                                 ew32(IMS, adapter->rx_ring->ims_val);
2400                         else
2401                                 e1000_irq_enable(adapter);
2402                 }
2403         }
2404
2405         return work_done;
2406 }
2407
2408 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2409 {
2410         struct e1000_adapter *adapter = netdev_priv(netdev);
2411         struct e1000_hw *hw = &adapter->hw;
2412         u32 vfta, index;
2413
2414         /* don't update vlan cookie if already programmed */
2415         if ((adapter->hw.mng_cookie.status &
2416              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2417             (vid == adapter->mng_vlan_id))
2418                 return;
2419
2420         /* add VID to filter table */
2421         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2422                 index = (vid >> 5) & 0x7F;
2423                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2424                 vfta |= (1 << (vid & 0x1F));
2425                 hw->mac.ops.write_vfta(hw, index, vfta);
2426         }
2427 }
2428
2429 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2430 {
2431         struct e1000_adapter *adapter = netdev_priv(netdev);
2432         struct e1000_hw *hw = &adapter->hw;
2433         u32 vfta, index;
2434
2435         if (!test_bit(__E1000_DOWN, &adapter->state))
2436                 e1000_irq_disable(adapter);
2437         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2438
2439         if (!test_bit(__E1000_DOWN, &adapter->state))
2440                 e1000_irq_enable(adapter);
2441
2442         if ((adapter->hw.mng_cookie.status &
2443              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2444             (vid == adapter->mng_vlan_id)) {
2445                 /* release control to f/w */
2446                 e1000_release_hw_control(adapter);
2447                 return;
2448         }
2449
2450         /* remove VID from filter table */
2451         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2452                 index = (vid >> 5) & 0x7F;
2453                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2454                 vfta &= ~(1 << (vid & 0x1F));
2455                 hw->mac.ops.write_vfta(hw, index, vfta);
2456         }
2457 }
2458
2459 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2460 {
2461         struct net_device *netdev = adapter->netdev;
2462         u16 vid = adapter->hw.mng_cookie.vlan_id;
2463         u16 old_vid = adapter->mng_vlan_id;
2464
2465         if (!adapter->vlgrp)
2466                 return;
2467
2468         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2469                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2470                 if (adapter->hw.mng_cookie.status &
2471                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2472                         e1000_vlan_rx_add_vid(netdev, vid);
2473                         adapter->mng_vlan_id = vid;
2474                 }
2475
2476                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2477                                 (vid != old_vid) &&
2478                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2479                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2480         } else {
2481                 adapter->mng_vlan_id = vid;
2482         }
2483 }
2484
2485
2486 static void e1000_vlan_rx_register(struct net_device *netdev,
2487                                    struct vlan_group *grp)
2488 {
2489         struct e1000_adapter *adapter = netdev_priv(netdev);
2490         struct e1000_hw *hw = &adapter->hw;
2491         u32 ctrl, rctl;
2492
2493         if (!test_bit(__E1000_DOWN, &adapter->state))
2494                 e1000_irq_disable(adapter);
2495         adapter->vlgrp = grp;
2496
2497         if (grp) {
2498                 /* enable VLAN tag insert/strip */
2499                 ctrl = er32(CTRL);
2500                 ctrl |= E1000_CTRL_VME;
2501                 ew32(CTRL, ctrl);
2502
2503                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2504                         /* enable VLAN receive filtering */
2505                         rctl = er32(RCTL);
2506                         rctl &= ~E1000_RCTL_CFIEN;
2507                         ew32(RCTL, rctl);
2508                         e1000_update_mng_vlan(adapter);
2509                 }
2510         } else {
2511                 /* disable VLAN tag insert/strip */
2512                 ctrl = er32(CTRL);
2513                 ctrl &= ~E1000_CTRL_VME;
2514                 ew32(CTRL, ctrl);
2515
2516                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2517                         if (adapter->mng_vlan_id !=
2518                             (u16)E1000_MNG_VLAN_NONE) {
2519                                 e1000_vlan_rx_kill_vid(netdev,
2520                                                        adapter->mng_vlan_id);
2521                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2522                         }
2523                 }
2524         }
2525
2526         if (!test_bit(__E1000_DOWN, &adapter->state))
2527                 e1000_irq_enable(adapter);
2528 }
2529
2530 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2531 {
2532         u16 vid;
2533
2534         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2535
2536         if (!adapter->vlgrp)
2537                 return;
2538
2539         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2540                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2541                         continue;
2542                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2543         }
2544 }
2545
2546 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2547 {
2548         struct e1000_hw *hw = &adapter->hw;
2549         u32 manc, manc2h, mdef, i, j;
2550
2551         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2552                 return;
2553
2554         manc = er32(MANC);
2555
2556         /*
2557          * enable receiving management packets to the host. this will probably
2558          * generate destination unreachable messages from the host OS, but
2559          * the packets will be handled on SMBUS
2560          */
2561         manc |= E1000_MANC_EN_MNG2HOST;
2562         manc2h = er32(MANC2H);
2563
2564         switch (hw->mac.type) {
2565         default:
2566                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2567                 break;
2568         case e1000_82574:
2569         case e1000_82583:
2570                 /*
2571                  * Check if IPMI pass-through decision filter already exists;
2572                  * if so, enable it.
2573                  */
2574                 for (i = 0, j = 0; i < 8; i++) {
2575                         mdef = er32(MDEF(i));
2576
2577                         /* Ignore filters with anything other than IPMI ports */
2578                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2579                                 continue;
2580
2581                         /* Enable this decision filter in MANC2H */
2582                         if (mdef)
2583                                 manc2h |= (1 << i);
2584
2585                         j |= mdef;
2586                 }
2587
2588                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2589                         break;
2590
2591                 /* Create new decision filter in an empty filter */
2592                 for (i = 0, j = 0; i < 8; i++)
2593                         if (er32(MDEF(i)) == 0) {
2594                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2595                                                E1000_MDEF_PORT_664));
2596                                 manc2h |= (1 << 1);
2597                                 j++;
2598                                 break;
2599                         }
2600
2601                 if (!j)
2602                         e_warn("Unable to create IPMI pass-through filter\n");
2603                 break;
2604         }
2605
2606         ew32(MANC2H, manc2h);
2607         ew32(MANC, manc);
2608 }
2609
2610 /**
2611  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2612  * @adapter: board private structure
2613  *
2614  * Configure the Tx unit of the MAC after a reset.
2615  **/
2616 static void e1000_configure_tx(struct e1000_adapter *adapter)
2617 {
2618         struct e1000_hw *hw = &adapter->hw;
2619         struct e1000_ring *tx_ring = adapter->tx_ring;
2620         u64 tdba;
2621         u32 tdlen, tctl, tipg, tarc;
2622         u32 ipgr1, ipgr2;
2623
2624         /* Setup the HW Tx Head and Tail descriptor pointers */
2625         tdba = tx_ring->dma;
2626         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2627         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2628         ew32(TDBAH, (tdba >> 32));
2629         ew32(TDLEN, tdlen);
2630         ew32(TDH, 0);
2631         ew32(TDT, 0);
2632         tx_ring->head = E1000_TDH;
2633         tx_ring->tail = E1000_TDT;
2634
2635         /* Set the default values for the Tx Inter Packet Gap timer */
2636         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2637         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2638         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2639
2640         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2641                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2642
2643         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2644         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2645         ew32(TIPG, tipg);
2646
2647         /* Set the Tx Interrupt Delay register */
2648         ew32(TIDV, adapter->tx_int_delay);
2649         /* Tx irq moderation */
2650         ew32(TADV, adapter->tx_abs_int_delay);
2651
2652         /* Program the Transmit Control Register */
2653         tctl = er32(TCTL);
2654         tctl &= ~E1000_TCTL_CT;
2655         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2656                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2657
2658         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2659                 tarc = er32(TARC(0));
2660                 /*
2661                  * set the speed mode bit, we'll clear it if we're not at
2662                  * gigabit link later
2663                  */
2664 #define SPEED_MODE_BIT (1 << 21)
2665                 tarc |= SPEED_MODE_BIT;
2666                 ew32(TARC(0), tarc);
2667         }
2668
2669         /* errata: program both queues to unweighted RR */
2670         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2671                 tarc = er32(TARC(0));
2672                 tarc |= 1;
2673                 ew32(TARC(0), tarc);
2674                 tarc = er32(TARC(1));
2675                 tarc |= 1;
2676                 ew32(TARC(1), tarc);
2677         }
2678
2679         /* Setup Transmit Descriptor Settings for eop descriptor */
2680         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2681
2682         /* only set IDE if we are delaying interrupts using the timers */
2683         if (adapter->tx_int_delay)
2684                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2685
2686         /* enable Report Status bit */
2687         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2688
2689         ew32(TCTL, tctl);
2690
2691         e1000e_config_collision_dist(hw);
2692 }
2693
2694 /**
2695  * e1000_setup_rctl - configure the receive control registers
2696  * @adapter: Board private structure
2697  **/
2698 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2699                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2700 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2701 {
2702         struct e1000_hw *hw = &adapter->hw;
2703         u32 rctl, rfctl;
2704         u32 psrctl = 0;
2705         u32 pages = 0;
2706
2707         /* Program MC offset vector base */
2708         rctl = er32(RCTL);
2709         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2710         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2711                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2712                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2713
2714         /* Do not Store bad packets */
2715         rctl &= ~E1000_RCTL_SBP;
2716
2717         /* Enable Long Packet receive */
2718         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2719                 rctl &= ~E1000_RCTL_LPE;
2720         else
2721                 rctl |= E1000_RCTL_LPE;
2722
2723         /* Some systems expect that the CRC is included in SMBUS traffic. The
2724          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2725          * host memory when this is enabled
2726          */
2727         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2728                 rctl |= E1000_RCTL_SECRC;
2729
2730         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2731         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2732                 u16 phy_data;
2733
2734                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2735                 phy_data &= 0xfff8;
2736                 phy_data |= (1 << 2);
2737                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2738
2739                 e1e_rphy(hw, 22, &phy_data);
2740                 phy_data &= 0x0fff;
2741                 phy_data |= (1 << 14);
2742                 e1e_wphy(hw, 0x10, 0x2823);
2743                 e1e_wphy(hw, 0x11, 0x0003);
2744                 e1e_wphy(hw, 22, phy_data);
2745         }
2746
2747         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2748         if (hw->mac.type == e1000_pch2lan) {
2749                 s32 ret_val;
2750
2751                 if (rctl & E1000_RCTL_LPE)
2752                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2753                 else
2754                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2755         }
2756
2757         /* Setup buffer sizes */
2758         rctl &= ~E1000_RCTL_SZ_4096;
2759         rctl |= E1000_RCTL_BSEX;
2760         switch (adapter->rx_buffer_len) {
2761         case 2048:
2762         default:
2763                 rctl |= E1000_RCTL_SZ_2048;
2764                 rctl &= ~E1000_RCTL_BSEX;
2765                 break;
2766         case 4096:
2767                 rctl |= E1000_RCTL_SZ_4096;
2768                 break;
2769         case 8192:
2770                 rctl |= E1000_RCTL_SZ_8192;
2771                 break;
2772         case 16384:
2773                 rctl |= E1000_RCTL_SZ_16384;
2774                 break;
2775         }
2776
2777         /*
2778          * 82571 and greater support packet-split where the protocol
2779          * header is placed in skb->data and the packet data is
2780          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2781          * In the case of a non-split, skb->data is linearly filled,
2782          * followed by the page buffers.  Therefore, skb->data is
2783          * sized to hold the largest protocol header.
2784          *
2785          * allocations using alloc_page take too long for regular MTU
2786          * so only enable packet split for jumbo frames
2787          *
2788          * Using pages when the page size is greater than 16k wastes
2789          * a lot of memory, since we allocate 3 pages at all times
2790          * per packet.
2791          */
2792         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2793         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2794             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2795                 adapter->rx_ps_pages = pages;
2796         else
2797                 adapter->rx_ps_pages = 0;
2798
2799         if (adapter->rx_ps_pages) {
2800                 /* Configure extra packet-split registers */
2801                 rfctl = er32(RFCTL);
2802                 rfctl |= E1000_RFCTL_EXTEN;
2803                 /*
2804                  * disable packet split support for IPv6 extension headers,
2805                  * because some malformed IPv6 headers can hang the Rx
2806                  */
2807                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2808                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2809
2810                 ew32(RFCTL, rfctl);
2811
2812                 /* Enable Packet split descriptors */
2813                 rctl |= E1000_RCTL_DTYP_PS;
2814
2815                 psrctl |= adapter->rx_ps_bsize0 >>
2816                         E1000_PSRCTL_BSIZE0_SHIFT;
2817
2818                 switch (adapter->rx_ps_pages) {
2819                 case 3:
2820                         psrctl |= PAGE_SIZE <<
2821                                 E1000_PSRCTL_BSIZE3_SHIFT;
2822                 case 2:
2823                         psrctl |= PAGE_SIZE <<
2824                                 E1000_PSRCTL_BSIZE2_SHIFT;
2825                 case 1:
2826                         psrctl |= PAGE_SIZE >>
2827                                 E1000_PSRCTL_BSIZE1_SHIFT;
2828                         break;
2829                 }
2830
2831                 ew32(PSRCTL, psrctl);
2832         }
2833
2834         ew32(RCTL, rctl);
2835         /* just started the receive unit, no need to restart */
2836         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2837 }
2838
2839 /**
2840  * e1000_configure_rx - Configure Receive Unit after Reset
2841  * @adapter: board private structure
2842  *
2843  * Configure the Rx unit of the MAC after a reset.
2844  **/
2845 static void e1000_configure_rx(struct e1000_adapter *adapter)
2846 {
2847         struct e1000_hw *hw = &adapter->hw;
2848         struct e1000_ring *rx_ring = adapter->rx_ring;
2849         u64 rdba;
2850         u32 rdlen, rctl, rxcsum, ctrl_ext;
2851
2852         if (adapter->rx_ps_pages) {
2853                 /* this is a 32 byte descriptor */
2854                 rdlen = rx_ring->count *
2855                         sizeof(union e1000_rx_desc_packet_split);
2856                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2857                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2858         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2859                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2860                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2861                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2862         } else {
2863                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2864                 adapter->clean_rx = e1000_clean_rx_irq;
2865                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2866         }
2867
2868         /* disable receives while setting up the descriptors */
2869         rctl = er32(RCTL);
2870         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2871         e1e_flush();
2872         msleep(10);
2873
2874         /* set the Receive Delay Timer Register */
2875         ew32(RDTR, adapter->rx_int_delay);
2876
2877         /* irq moderation */
2878         ew32(RADV, adapter->rx_abs_int_delay);
2879         if (adapter->itr_setting != 0)
2880                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2881
2882         ctrl_ext = er32(CTRL_EXT);
2883         /* Auto-Mask interrupts upon ICR access */
2884         ctrl_ext |= E1000_CTRL_EXT_IAME;
2885         ew32(IAM, 0xffffffff);
2886         ew32(CTRL_EXT, ctrl_ext);
2887         e1e_flush();
2888
2889         /*
2890          * Setup the HW Rx Head and Tail Descriptor Pointers and
2891          * the Base and Length of the Rx Descriptor Ring
2892          */
2893         rdba = rx_ring->dma;
2894         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2895         ew32(RDBAH, (rdba >> 32));
2896         ew32(RDLEN, rdlen);
2897         ew32(RDH, 0);
2898         ew32(RDT, 0);
2899         rx_ring->head = E1000_RDH;
2900         rx_ring->tail = E1000_RDT;
2901
2902         /* Enable Receive Checksum Offload for TCP and UDP */
2903         rxcsum = er32(RXCSUM);
2904         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2905                 rxcsum |= E1000_RXCSUM_TUOFL;
2906
2907                 /*
2908                  * IPv4 payload checksum for UDP fragments must be
2909                  * used in conjunction with packet-split.
2910                  */
2911                 if (adapter->rx_ps_pages)
2912                         rxcsum |= E1000_RXCSUM_IPPCSE;
2913         } else {
2914                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2915                 /* no need to clear IPPCSE as it defaults to 0 */
2916         }
2917         ew32(RXCSUM, rxcsum);
2918
2919         /*
2920          * Enable early receives on supported devices, only takes effect when
2921          * packet size is equal or larger than the specified value (in 8 byte
2922          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2923          */
2924         if (adapter->flags & FLAG_HAS_ERT) {
2925                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2926                         u32 rxdctl = er32(RXDCTL(0));
2927                         ew32(RXDCTL(0), rxdctl | 0x3);
2928                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2929                         /*
2930                          * With jumbo frames and early-receive enabled,
2931                          * excessive C-state transition latencies result in
2932                          * dropped transactions.
2933                          */
2934                         pm_qos_update_request(
2935                                 &adapter->netdev->pm_qos_req, 55);
2936                 } else {
2937                         pm_qos_update_request(
2938                                 &adapter->netdev->pm_qos_req,
2939                                 PM_QOS_DEFAULT_VALUE);
2940                 }
2941         }
2942
2943         /* Enable Receives */
2944         ew32(RCTL, rctl);
2945 }
2946
2947 /**
2948  *  e1000_update_mc_addr_list - Update Multicast addresses
2949  *  @hw: pointer to the HW structure
2950  *  @mc_addr_list: array of multicast addresses to program
2951  *  @mc_addr_count: number of multicast addresses to program
2952  *
2953  *  Updates the Multicast Table Array.
2954  *  The caller must have a packed mc_addr_list of multicast addresses.
2955  **/
2956 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2957                                       u32 mc_addr_count)
2958 {
2959         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2960 }
2961
2962 /**
2963  * e1000_set_multi - Multicast and Promiscuous mode set
2964  * @netdev: network interface device structure
2965  *
2966  * The set_multi entry point is called whenever the multicast address
2967  * list or the network interface flags are updated.  This routine is
2968  * responsible for configuring the hardware for proper multicast,
2969  * promiscuous mode, and all-multi behavior.
2970  **/
2971 static void e1000_set_multi(struct net_device *netdev)
2972 {
2973         struct e1000_adapter *adapter = netdev_priv(netdev);
2974         struct e1000_hw *hw = &adapter->hw;
2975         struct netdev_hw_addr *ha;
2976         u8  *mta_list;
2977         u32 rctl;
2978         int i;
2979
2980         /* Check for Promiscuous and All Multicast modes */
2981
2982         rctl = er32(RCTL);
2983
2984         if (netdev->flags & IFF_PROMISC) {
2985                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2986                 rctl &= ~E1000_RCTL_VFE;
2987         } else {
2988                 if (netdev->flags & IFF_ALLMULTI) {
2989                         rctl |= E1000_RCTL_MPE;
2990                         rctl &= ~E1000_RCTL_UPE;
2991                 } else {
2992                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2993                 }
2994                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2995                         rctl |= E1000_RCTL_VFE;
2996         }
2997
2998         ew32(RCTL, rctl);
2999
3000         if (!netdev_mc_empty(netdev)) {
3001                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3002                 if (!mta_list)
3003                         return;
3004
3005                 /* prepare a packed array of only addresses. */
3006                 i = 0;
3007                 netdev_for_each_mc_addr(ha, netdev)
3008                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3009
3010                 e1000_update_mc_addr_list(hw, mta_list, i);
3011                 kfree(mta_list);
3012         } else {
3013                 /*
3014                  * if we're called from probe, we might not have
3015                  * anything to do here, so clear out the list
3016                  */
3017                 e1000_update_mc_addr_list(hw, NULL, 0);
3018         }
3019 }
3020
3021 /**
3022  * e1000_configure - configure the hardware for Rx and Tx
3023  * @adapter: private board structure
3024  **/
3025 static void e1000_configure(struct e1000_adapter *adapter)
3026 {
3027         e1000_set_multi(adapter->netdev);
3028
3029         e1000_restore_vlan(adapter);
3030         e1000_init_manageability_pt(adapter);
3031
3032         e1000_configure_tx(adapter);
3033         e1000_setup_rctl(adapter);
3034         e1000_configure_rx(adapter);
3035         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3036 }
3037
3038 /**
3039  * e1000e_power_up_phy - restore link in case the phy was powered down
3040  * @adapter: address of board private structure
3041  *
3042  * The phy may be powered down to save power and turn off link when the
3043  * driver is unloaded and wake on lan is not enabled (among others)
3044  * *** this routine MUST be followed by a call to e1000e_reset ***
3045  **/
3046 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3047 {
3048         if (adapter->hw.phy.ops.power_up)
3049                 adapter->hw.phy.ops.power_up(&adapter->hw);
3050
3051         adapter->hw.mac.ops.setup_link(&adapter->hw);
3052 }
3053
3054 /**
3055  * e1000_power_down_phy - Power down the PHY
3056  *
3057  * Power down the PHY so no link is implied when interface is down.
3058  * The PHY cannot be powered down if management or WoL is active.
3059  */
3060 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3061 {
3062         /* WoL is enabled */
3063         if (adapter->wol)
3064                 return;
3065
3066         if (adapter->hw.phy.ops.power_down)
3067                 adapter->hw.phy.ops.power_down(&adapter->hw);
3068 }
3069
3070 /**
3071  * e1000e_reset - bring the hardware into a known good state
3072  *
3073  * This function boots the hardware and enables some settings that
3074  * require a configuration cycle of the hardware - those cannot be
3075  * set/changed during runtime. After reset the device needs to be
3076  * properly configured for Rx, Tx etc.
3077  */
3078 void e1000e_reset(struct e1000_adapter *adapter)
3079 {
3080         struct e1000_mac_info *mac = &adapter->hw.mac;
3081         struct e1000_fc_info *fc = &adapter->hw.fc;
3082         struct e1000_hw *hw = &adapter->hw;
3083         u32 tx_space, min_tx_space, min_rx_space;
3084         u32 pba = adapter->pba;
3085         u16 hwm;
3086
3087         /* reset Packet Buffer Allocation to default */
3088         ew32(PBA, pba);
3089
3090         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3091                 /*
3092                  * To maintain wire speed transmits, the Tx FIFO should be
3093                  * large enough to accommodate two full transmit packets,
3094                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3095                  * the Rx FIFO should be large enough to accommodate at least
3096                  * one full receive packet and is similarly rounded up and
3097                  * expressed in KB.
3098                  */
3099                 pba = er32(PBA);
3100                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3101                 tx_space = pba >> 16;
3102                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3103                 pba &= 0xffff;
3104                 /*
3105                  * the Tx fifo also stores 16 bytes of information about the tx
3106                  * but don't include ethernet FCS because hardware appends it
3107                  */
3108                 min_tx_space = (adapter->max_frame_size +
3109                                 sizeof(struct e1000_tx_desc) -
3110                                 ETH_FCS_LEN) * 2;
3111                 min_tx_space = ALIGN(min_tx_space, 1024);
3112                 min_tx_space >>= 10;
3113                 /* software strips receive CRC, so leave room for it */
3114                 min_rx_space = adapter->max_frame_size;
3115                 min_rx_space = ALIGN(min_rx_space, 1024);
3116                 min_rx_space >>= 10;
3117
3118                 /*
3119                  * If current Tx allocation is less than the min Tx FIFO size,
3120                  * and the min Tx FIFO size is less than the current Rx FIFO
3121                  * allocation, take space away from current Rx allocation
3122                  */
3123                 if ((tx_space < min_tx_space) &&
3124                     ((min_tx_space - tx_space) < pba)) {
3125                         pba -= min_tx_space - tx_space;
3126
3127                         /*
3128                          * if short on Rx space, Rx wins and must trump tx
3129                          * adjustment or use Early Receive if available
3130                          */
3131                         if ((pba < min_rx_space) &&
3132                             (!(adapter->flags & FLAG_HAS_ERT)))
3133                                 /* ERT enabled in e1000_configure_rx */
3134                                 pba = min_rx_space;
3135                 }
3136
3137                 ew32(PBA, pba);
3138         }
3139
3140
3141         /*
3142          * flow control settings
3143          *
3144          * The high water mark must be low enough to fit one full frame
3145          * (or the size used for early receive) above it in the Rx FIFO.
3146          * Set it to the lower of:
3147          * - 90% of the Rx FIFO size, and
3148          * - the full Rx FIFO size minus the early receive size (for parts
3149          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3150          * - the full Rx FIFO size minus one full frame
3151          */
3152         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3153                 fc->pause_time = 0xFFFF;
3154         else
3155                 fc->pause_time = E1000_FC_PAUSE_TIME;
3156         fc->send_xon = 1;
3157         fc->current_mode = fc->requested_mode;
3158
3159         switch (hw->mac.type) {
3160         default:
3161                 if ((adapter->flags & FLAG_HAS_ERT) &&
3162                     (adapter->netdev->mtu > ETH_DATA_LEN))
3163                         hwm = min(((pba << 10) * 9 / 10),
3164                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3165                 else
3166                         hwm = min(((pba << 10) * 9 / 10),
3167                                   ((pba << 10) - adapter->max_frame_size));
3168
3169                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3170                 fc->low_water = fc->high_water - 8;
3171                 break;
3172         case e1000_pchlan:
3173                 /*
3174                  * Workaround PCH LOM adapter hangs with certain network
3175                  * loads.  If hangs persist, try disabling Tx flow control.
3176                  */
3177                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3178                         fc->high_water = 0x3500;
3179                         fc->low_water  = 0x1500;
3180                 } else {
3181                         fc->high_water = 0x5000;
3182                         fc->low_water  = 0x3000;
3183                 }
3184                 fc->refresh_time = 0x1000;
3185                 break;
3186         case e1000_pch2lan:
3187                 fc->high_water = 0x05C20;
3188                 fc->low_water = 0x05048;
3189                 fc->pause_time = 0x0650;
3190                 fc->refresh_time = 0x0400;
3191                 break;
3192         }
3193
3194         /* Allow time for pending master requests to run */
3195         mac->ops.reset_hw(hw);
3196
3197         /*
3198          * For parts with AMT enabled, let the firmware know
3199          * that the network interface is in control
3200          */
3201         if (adapter->flags & FLAG_HAS_AMT)
3202                 e1000_get_hw_control(adapter);
3203
3204         ew32(WUC, 0);
3205
3206         if (mac->ops.init_hw(hw))
3207                 e_err("Hardware Error\n");
3208
3209         e1000_update_mng_vlan(adapter);
3210
3211         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3212         ew32(VET, ETH_P_8021Q);
3213
3214         e1000e_reset_adaptive(hw);
3215         e1000_get_phy_info(hw);
3216
3217         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3218             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3219                 u16 phy_data = 0;
3220                 /*
3221                  * speed up time to link by disabling smart power down, ignore
3222                  * the return value of this function because there is nothing
3223                  * different we would do if it failed
3224                  */
3225                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3226                 phy_data &= ~IGP02E1000_PM_SPD;
3227                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3228         }
3229 }
3230
3231 int e1000e_up(struct e1000_adapter *adapter)
3232 {
3233         struct e1000_hw *hw = &adapter->hw;
3234
3235         /* hardware has been reset, we need to reload some things */
3236         e1000_configure(adapter);
3237
3238         clear_bit(__E1000_DOWN, &adapter->state);
3239
3240         napi_enable(&adapter->napi);
3241         if (adapter->msix_entries)
3242                 e1000_configure_msix(adapter);
3243         e1000_irq_enable(adapter);
3244
3245         netif_wake_queue(adapter->netdev);
3246
3247         /* fire a link change interrupt to start the watchdog */
3248         if (adapter->msix_entries)
3249                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3250         else
3251                 ew32(ICS, E1000_ICS_LSC);
3252
3253         return 0;
3254 }
3255
3256 void e1000e_down(struct e1000_adapter *adapter)
3257 {
3258         struct net_device *netdev = adapter->netdev;
3259         struct e1000_hw *hw = &adapter->hw;
3260         u32 tctl, rctl;
3261
3262         /*
3263          * signal that we're down so the interrupt handler does not
3264          * reschedule our watchdog timer
3265          */
3266         set_bit(__E1000_DOWN, &adapter->state);
3267
3268         /* disable receives in the hardware */
3269         rctl = er32(RCTL);
3270         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3271         /* flush and sleep below */
3272
3273         netif_stop_queue(netdev);
3274
3275         /* disable transmits in the hardware */
3276         tctl = er32(TCTL);
3277         tctl &= ~E1000_TCTL_EN;
3278         ew32(TCTL, tctl);
3279         /* flush both disables and wait for them to finish */
3280         e1e_flush();
3281         msleep(10);
3282
3283         napi_disable(&adapter->napi);
3284         e1000_irq_disable(adapter);
3285
3286         del_timer_sync(&adapter->watchdog_timer);
3287         del_timer_sync(&adapter->phy_info_timer);
3288
3289         netif_carrier_off(netdev);
3290         adapter->link_speed = 0;
3291         adapter->link_duplex = 0;
3292
3293         if (!pci_channel_offline(adapter->pdev))
3294                 e1000e_reset(adapter);
3295         e1000_clean_tx_ring(adapter);
3296         e1000_clean_rx_ring(adapter);
3297
3298         /*
3299          * TODO: for power management, we could drop the link and
3300          * pci_disable_device here.
3301          */
3302 }
3303
3304 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3305 {
3306         might_sleep();
3307         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3308                 msleep(1);
3309         e1000e_down(adapter);
3310         e1000e_up(adapter);
3311         clear_bit(__E1000_RESETTING, &adapter->state);
3312 }
3313
3314 /**
3315  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3316  * @adapter: board private structure to initialize
3317  *
3318  * e1000_sw_init initializes the Adapter private data structure.
3319  * Fields are initialized based on PCI device information and
3320  * OS network device settings (MTU size).
3321  **/
3322 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3323 {
3324         struct net_device *netdev = adapter->netdev;
3325
3326         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3327         adapter->rx_ps_bsize0 = 128;
3328         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3329         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3330
3331         e1000e_set_interrupt_capability(adapter);
3332
3333         if (e1000_alloc_queues(adapter))
3334                 return -ENOMEM;
3335
3336         /* Explicitly disable IRQ since the NIC can be in any state. */
3337         e1000_irq_disable(adapter);
3338
3339         set_bit(__E1000_DOWN, &adapter->state);
3340         return 0;
3341 }
3342
3343 /**
3344  * e1000_intr_msi_test - Interrupt Handler
3345  * @irq: interrupt number
3346  * @data: pointer to a network interface device structure
3347  **/
3348 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3349 {
3350         struct net_device *netdev = data;
3351         struct e1000_adapter *adapter = netdev_priv(netdev);
3352         struct e1000_hw *hw = &adapter->hw;
3353         u32 icr = er32(ICR);
3354
3355         e_dbg("icr is %08X\n", icr);
3356         if (icr & E1000_ICR_RXSEQ) {
3357                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3358                 wmb();
3359         }
3360
3361         return IRQ_HANDLED;
3362 }
3363
3364 /**
3365  * e1000_test_msi_interrupt - Returns 0 for successful test
3366  * @adapter: board private struct
3367  *
3368  * code flow taken from tg3.c
3369  **/
3370 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3371 {
3372         struct net_device *netdev = adapter->netdev;
3373         struct e1000_hw *hw = &adapter->hw;
3374         int err;
3375
3376         /* poll_enable hasn't been called yet, so don't need disable */
3377         /* clear any pending events */
3378         er32(ICR);
3379
3380         /* free the real vector and request a test handler */
3381         e1000_free_irq(adapter);
3382         e1000e_reset_interrupt_capability(adapter);
3383
3384         /* Assume that the test fails, if it succeeds then the test
3385          * MSI irq handler will unset this flag */
3386         adapter->flags |= FLAG_MSI_TEST_FAILED;
3387
3388         err = pci_enable_msi(adapter->pdev);
3389         if (err)
3390                 goto msi_test_failed;
3391
3392         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3393                           netdev->name, netdev);
3394         if (err) {
3395                 pci_disable_msi(adapter->pdev);
3396                 goto msi_test_failed;
3397         }
3398
3399         wmb();
3400
3401         e1000_irq_enable(adapter);
3402
3403         /* fire an unusual interrupt on the test handler */
3404         ew32(ICS, E1000_ICS_RXSEQ);
3405         e1e_flush();
3406         msleep(50);
3407
3408         e1000_irq_disable(adapter);
3409
3410         rmb();
3411
3412         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3413                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3414                 err = -EIO;
3415                 e_info("MSI interrupt test failed!\n");
3416         }
3417
3418         free_irq(adapter->pdev->irq, netdev);
3419         pci_disable_msi(adapter->pdev);
3420
3421         if (err == -EIO)
3422                 goto msi_test_failed;
3423
3424         /* okay so the test worked, restore settings */
3425         e_dbg("MSI interrupt test succeeded!\n");
3426 msi_test_failed:
3427         e1000e_set_interrupt_capability(adapter);
3428         e1000_request_irq(adapter);
3429         return err;
3430 }
3431
3432 /**
3433  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3434  * @adapter: board private struct
3435  *
3436  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3437  **/
3438 static int e1000_test_msi(struct e1000_adapter *adapter)
3439 {
3440         int err;
3441         u16 pci_cmd;
3442
3443         if (!(adapter->flags & FLAG_MSI_ENABLED))
3444                 return 0;
3445
3446         /* disable SERR in case the MSI write causes a master abort */
3447         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3448         if (pci_cmd & PCI_COMMAND_SERR)
3449                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3450                                       pci_cmd & ~PCI_COMMAND_SERR);
3451
3452         err = e1000_test_msi_interrupt(adapter);
3453
3454         /* re-enable SERR */
3455         if (pci_cmd & PCI_COMMAND_SERR) {
3456                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3457                 pci_cmd |= PCI_COMMAND_SERR;
3458                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3459         }
3460
3461         /* success ! */
3462         if (!err)
3463                 return 0;
3464
3465         /* EIO means MSI test failed */
3466         if (err != -EIO)
3467                 return err;
3468
3469         /* back to INTx mode */
3470         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3471
3472         e1000_free_irq(adapter);
3473
3474         err = e1000_request_irq(adapter);
3475
3476         return err;
3477 }
3478
3479 /**
3480  * e1000_open - Called when a network interface is made active
3481  * @netdev: network interface device structure
3482  *
3483  * Returns 0 on success, negative value on failure
3484  *
3485  * The open entry point is called when a network interface is made
3486  * active by the system (IFF_UP).  At this point all resources needed
3487  * for transmit and receive operations are allocated, the interrupt
3488  * handler is registered with the OS, the watchdog timer is started,
3489  * and the stack is notified that the interface is ready.
3490  **/
3491 static int e1000_open(struct net_device *netdev)
3492 {
3493         struct e1000_adapter *adapter = netdev_priv(netdev);
3494         struct e1000_hw *hw = &adapter->hw;
3495         struct pci_dev *pdev = adapter->pdev;
3496         int err;
3497
3498         /* disallow open during test */
3499         if (test_bit(__E1000_TESTING, &adapter->state))
3500                 return -EBUSY;
3501
3502         pm_runtime_get_sync(&pdev->dev);
3503
3504         netif_carrier_off(netdev);
3505
3506         /* allocate transmit descriptors */
3507         err = e1000e_setup_tx_resources(adapter);
3508         if (err)
3509                 goto err_setup_tx;
3510
3511         /* allocate receive descriptors */
3512         err = e1000e_setup_rx_resources(adapter);
3513         if (err)
3514                 goto err_setup_rx;
3515
3516         /*
3517          * If AMT is enabled, let the firmware know that the network
3518          * interface is now open and reset the part to a known state.
3519          */
3520         if (adapter->flags & FLAG_HAS_AMT) {
3521                 e1000_get_hw_control(adapter);
3522                 e1000e_reset(adapter);
3523         }
3524
3525         e1000e_power_up_phy(adapter);
3526
3527         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3528         if ((adapter->hw.mng_cookie.status &
3529              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3530                 e1000_update_mng_vlan(adapter);
3531
3532         /* DMA latency requirement to workaround early-receive/jumbo issue */
3533         if (adapter->flags & FLAG_HAS_ERT)
3534                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3535                                    PM_QOS_CPU_DMA_LATENCY,
3536                                    PM_QOS_DEFAULT_VALUE);
3537
3538         /*
3539          * before we allocate an interrupt, we must be ready to handle it.
3540          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3541          * as soon as we call pci_request_irq, so we have to setup our
3542          * clean_rx handler before we do so.
3543          */
3544         e1000_configure(adapter);
3545
3546         err = e1000_request_irq(adapter);
3547         if (err)
3548                 goto err_req_irq;
3549
3550         /*
3551          * Work around PCIe errata with MSI interrupts causing some chipsets to
3552          * ignore e1000e MSI messages, which means we need to test our MSI
3553          * interrupt now
3554          */
3555         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3556                 err = e1000_test_msi(adapter);
3557                 if (err) {
3558                         e_err("Interrupt allocation failed\n");
3559                         goto err_req_irq;
3560                 }
3561         }
3562
3563         /* From here on the code is the same as e1000e_up() */
3564         clear_bit(__E1000_DOWN, &adapter->state);
3565
3566         napi_enable(&adapter->napi);
3567
3568         e1000_irq_enable(adapter);
3569
3570         netif_start_queue(netdev);
3571
3572         adapter->idle_check = true;
3573         pm_runtime_put(&pdev->dev);
3574
3575         /* fire a link status change interrupt to start the watchdog */
3576         if (adapter->msix_entries)
3577                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3578         else
3579                 ew32(ICS, E1000_ICS_LSC);
3580
3581         return 0;
3582
3583 err_req_irq:
3584         e1000_release_hw_control(adapter);
3585         e1000_power_down_phy(adapter);
3586         e1000e_free_rx_resources(adapter);
3587 err_setup_rx:
3588         e1000e_free_tx_resources(adapter);
3589 err_setup_tx:
3590         e1000e_reset(adapter);
3591         pm_runtime_put_sync(&pdev->dev);
3592
3593         return err;
3594 }
3595
3596 /**
3597  * e1000_close - Disables a network interface
3598  * @netdev: network interface device structure
3599  *
3600  * Returns 0, this is not allowed to fail
3601  *
3602  * The close entry point is called when an interface is de-activated
3603  * by the OS.  The hardware is still under the drivers control, but
3604  * needs to be disabled.  A global MAC reset is issued to stop the
3605  * hardware, and all transmit and receive resources are freed.
3606  **/
3607 static int e1000_close(struct net_device *netdev)
3608 {
3609         struct e1000_adapter *adapter = netdev_priv(netdev);
3610         struct pci_dev *pdev = adapter->pdev;
3611
3612         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3613
3614         pm_runtime_get_sync(&pdev->dev);
3615
3616         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3617                 e1000e_down(adapter);
3618                 e1000_free_irq(adapter);
3619         }
3620         e1000_power_down_phy(adapter);
3621
3622         e1000e_free_tx_resources(adapter);
3623         e1000e_free_rx_resources(adapter);
3624
3625         /*
3626          * kill manageability vlan ID if supported, but not if a vlan with
3627          * the same ID is registered on the host OS (let 8021q kill it)
3628          */
3629         if ((adapter->hw.mng_cookie.status &
3630                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3631              !(adapter->vlgrp &&
3632                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3633                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3634
3635         /*
3636          * If AMT is enabled, let the firmware know that the network
3637          * interface is now closed
3638          */
3639         if (adapter->flags & FLAG_HAS_AMT)
3640                 e1000_release_hw_control(adapter);
3641
3642         if (adapter->flags & FLAG_HAS_ERT)
3643                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3644
3645         pm_runtime_put_sync(&pdev->dev);
3646
3647         return 0;
3648 }
3649 /**
3650  * e1000_set_mac - Change the Ethernet Address of the NIC
3651  * @netdev: network interface device structure
3652  * @p: pointer to an address structure
3653  *
3654  * Returns 0 on success, negative on failure
3655  **/
3656 static int e1000_set_mac(struct net_device *netdev, void *p)
3657 {
3658         struct e1000_adapter *adapter = netdev_priv(netdev);
3659         struct sockaddr *addr = p;
3660
3661         if (!is_valid_ether_addr(addr->sa_data))
3662                 return -EADDRNOTAVAIL;
3663
3664         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3665         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3666
3667         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3668
3669         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3670                 /* activate the work around */
3671                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3672
3673                 /*
3674                  * Hold a copy of the LAA in RAR[14] This is done so that
3675                  * between the time RAR[0] gets clobbered  and the time it
3676                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3677                  * of the RARs and no incoming packets directed to this port
3678                  * are dropped. Eventually the LAA will be in RAR[0] and
3679                  * RAR[14]
3680                  */
3681                 e1000e_rar_set(&adapter->hw,
3682                               adapter->hw.mac.addr,
3683                               adapter->hw.mac.rar_entry_count - 1);
3684         }
3685
3686         return 0;
3687 }
3688
3689 /**
3690  * e1000e_update_phy_task - work thread to update phy
3691  * @work: pointer to our work struct
3692  *
3693  * this worker thread exists because we must acquire a
3694  * semaphore to read the phy, which we could msleep while
3695  * waiting for it, and we can't msleep in a timer.
3696  **/
3697 static void e1000e_update_phy_task(struct work_struct *work)
3698 {
3699         struct e1000_adapter *adapter = container_of(work,
3700                                         struct e1000_adapter, update_phy_task);
3701         e1000_get_phy_info(&adapter->hw);
3702 }
3703
3704 /*
3705  * Need to wait a few seconds after link up to get diagnostic information from
3706  * the phy
3707  */
3708 static void e1000_update_phy_info(unsigned long data)
3709 {
3710         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3711         schedule_work(&adapter->update_phy_task);
3712 }
3713
3714 /**
3715  * e1000e_update_phy_stats - Update the PHY statistics counters
3716  * @adapter: board private structure
3717  **/
3718 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3719 {
3720         struct e1000_hw *hw = &adapter->hw;
3721         s32 ret_val;
3722         u16 phy_data;
3723
3724         ret_val = hw->phy.ops.acquire(hw);
3725         if (ret_val)
3726                 return;
3727
3728         hw->phy.addr = 1;
3729
3730 #define HV_PHY_STATS_PAGE       778
3731         /*
3732          * A page set is expensive so check if already on desired page.
3733          * If not, set to the page with the PHY status registers.
3734          */
3735         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3736                                            &phy_data);
3737         if (ret_val)
3738                 goto release;
3739         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3740                 ret_val = e1000e_write_phy_reg_mdic(hw,
3741                                                     IGP01E1000_PHY_PAGE_SELECT,
3742                                                     (HV_PHY_STATS_PAGE <<
3743                                                      IGP_PAGE_SHIFT));
3744                 if (ret_val)
3745                         goto release;
3746         }
3747
3748         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3749
3750         /* Single Collision Count */
3751         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3752                                  &phy_data);
3753         ret_val = e1000e_read_phy_reg_mdic(hw,
3754                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3755                                            &phy_data);
3756         if (!ret_val)
3757                 adapter->stats.scc += phy_data;
3758
3759         /* Excessive Collision Count */
3760         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3761                                  &phy_data);
3762         ret_val = e1000e_read_phy_reg_mdic(hw,
3763                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3764                                            &phy_data);
3765         if (!ret_val)
3766                 adapter->stats.ecol += phy_data;
3767
3768         /* Multiple Collision Count */
3769         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3770                                  &phy_data);
3771         ret_val = e1000e_read_phy_reg_mdic(hw,
3772                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3773                                            &phy_data);
3774         if (!ret_val)
3775                 adapter->stats.mcc += phy_data;
3776
3777         /* Late Collision Count */
3778         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3779                                  &phy_data);
3780         ret_val = e1000e_read_phy_reg_mdic(hw,
3781                                            HV_LATECOL_LOWER &
3782                                            MAX_PHY_REG_ADDRESS,
3783                                            &phy_data);
3784         if (!ret_val)
3785                 adapter->stats.latecol += phy_data;
3786
3787         /* Collision Count - also used for adaptive IFS */
3788         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3789                                  &phy_data);
3790         ret_val = e1000e_read_phy_reg_mdic(hw,
3791                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3792                                            &phy_data);
3793         if (!ret_val)
3794                 hw->mac.collision_delta = phy_data;
3795
3796         /* Defer Count */
3797         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3798                                  &phy_data);
3799         ret_val = e1000e_read_phy_reg_mdic(hw,
3800                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3801                                            &phy_data);
3802         if (!ret_val)
3803                 adapter->stats.dc += phy_data;
3804
3805         /* Transmit with no CRS */
3806         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3807                                  &phy_data);
3808         ret_val = e1000e_read_phy_reg_mdic(hw,
3809                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3810                                            &phy_data);
3811         if (!ret_val)
3812                 adapter->stats.tncrs += phy_data;
3813
3814 release:
3815         hw->phy.ops.release(hw);
3816 }
3817
3818 /**
3819  * e1000e_update_stats - Update the board statistics counters
3820  * @adapter: board private structure
3821  **/
3822 void e1000e_update_stats(struct e1000_adapter *adapter)
3823 {
3824         struct net_device *netdev = adapter->netdev;
3825         struct e1000_hw *hw = &adapter->hw;
3826         struct pci_dev *pdev = adapter->pdev;
3827
3828         /*
3829          * Prevent stats update while adapter is being reset, or if the pci
3830          * connection is down.
3831          */
3832         if (adapter->link_speed == 0)
3833                 return;
3834         if (pci_channel_offline(pdev))
3835                 return;
3836
3837         adapter->stats.crcerrs += er32(CRCERRS);
3838         adapter->stats.gprc += er32(GPRC);
3839         adapter->stats.gorc += er32(GORCL);
3840         er32(GORCH); /* Clear gorc */
3841         adapter->stats.bprc += er32(BPRC);
3842         adapter->stats.mprc += er32(MPRC);
3843         adapter->stats.roc += er32(ROC);
3844
3845         adapter->stats.mpc += er32(MPC);
3846
3847         /* Half-duplex statistics */
3848         if (adapter->link_duplex == HALF_DUPLEX) {
3849                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3850                         e1000e_update_phy_stats(adapter);
3851                 } else {
3852                         adapter->stats.scc += er32(SCC);
3853                         adapter->stats.ecol += er32(ECOL);
3854                         adapter->stats.mcc += er32(MCC);
3855                         adapter->stats.latecol += er32(LATECOL);
3856                         adapter->stats.dc += er32(DC);
3857
3858                         hw->mac.collision_delta = er32(COLC);
3859
3860                         if ((hw->mac.type != e1000_82574) &&
3861                             (hw->mac.type != e1000_82583))
3862                                 adapter->stats.tncrs += er32(TNCRS);
3863                 }
3864                 adapter->stats.colc += hw->mac.collision_delta;
3865         }
3866
3867         adapter->stats.xonrxc += er32(XONRXC);
3868         adapter->stats.xontxc += er32(XONTXC);
3869         adapter->stats.xoffrxc += er32(XOFFRXC);
3870         adapter->stats.xofftxc += er32(XOFFTXC);
3871         adapter->stats.gptc += er32(GPTC);
3872         adapter->stats.gotc += er32(GOTCL);
3873         er32(GOTCH); /* Clear gotc */
3874         adapter->stats.rnbc += er32(RNBC);
3875         adapter->stats.ruc += er32(RUC);
3876
3877         adapter->stats.mptc += er32(MPTC);
3878         adapter->stats.bptc += er32(BPTC);
3879
3880         /* used for adaptive IFS */
3881
3882         hw->mac.tx_packet_delta = er32(TPT);
3883         adapter->stats.tpt += hw->mac.tx_packet_delta;
3884
3885         adapter->stats.algnerrc += er32(ALGNERRC);
3886         adapter->stats.rxerrc += er32(RXERRC);
3887         adapter->stats.cexterr += er32(CEXTERR);
3888         adapter->stats.tsctc += er32(TSCTC);
3889         adapter->stats.tsctfc += er32(TSCTFC);
3890
3891         /* Fill out the OS statistics structure */
3892         netdev->stats.multicast = adapter->stats.mprc;
3893         netdev->stats.collisions = adapter->stats.colc;
3894
3895         /* Rx Errors */
3896
3897         /*
3898          * RLEC on some newer hardware can be incorrect so build
3899          * our own version based on RUC and ROC
3900          */
3901         netdev->stats.rx_errors = adapter->stats.rxerrc +
3902                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3903                 adapter->stats.ruc + adapter->stats.roc +
3904                 adapter->stats.cexterr;
3905         netdev->stats.rx_length_errors = adapter->stats.ruc +
3906                                               adapter->stats.roc;
3907         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3908         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3909         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3910
3911         /* Tx Errors */
3912         netdev->stats.tx_errors = adapter->stats.ecol +
3913                                        adapter->stats.latecol;
3914         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3915         netdev->stats.tx_window_errors = adapter->stats.latecol;
3916         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3917
3918         /* Tx Dropped needs to be maintained elsewhere */
3919
3920         /* Management Stats */
3921         adapter->stats.mgptc += er32(MGTPTC);
3922         adapter->stats.mgprc += er32(MGTPRC);
3923         adapter->stats.mgpdc += er32(MGTPDC);
3924 }
3925
3926 /**
3927  * e1000_phy_read_status - Update the PHY register status snapshot
3928  * @adapter: board private structure
3929  **/
3930 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3931 {
3932         struct e1000_hw *hw = &adapter->hw;
3933         struct e1000_phy_regs *phy = &adapter->phy_regs;
3934         int ret_val;
3935
3936         if ((er32(STATUS) & E1000_STATUS_LU) &&
3937             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3938                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3939                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3940                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3941                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3942                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3943                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3944                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3945                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3946                 if (ret_val)
3947                         e_warn("Error reading PHY register\n");
3948         } else {
3949                 /*
3950                  * Do not read PHY registers if link is not up
3951                  * Set values to typical power-on defaults
3952                  */
3953                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3954                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3955                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3956                              BMSR_ERCAP);
3957                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3958                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3959                 phy->lpa = 0;
3960                 phy->expansion = EXPANSION_ENABLENPAGE;
3961                 phy->ctrl1000 = ADVERTISE_1000FULL;
3962                 phy->stat1000 = 0;
3963                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3964         }
3965 }
3966
3967 static void e1000_print_link_info(struct e1000_adapter *adapter)
3968 {
3969         struct e1000_hw *hw = &adapter->hw;
3970         u32 ctrl = er32(CTRL);
3971
3972         /* Link status message must follow this format for user tools */
3973         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3974                "Flow Control: %s\n",
3975                adapter->netdev->name,
3976                adapter->link_speed,
3977                (adapter->link_duplex == FULL_DUPLEX) ?
3978                                 "Full Duplex" : "Half Duplex",
3979                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3980                                 "RX/TX" :
3981                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3982                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3983 }
3984
3985 static bool e1000e_has_link(struct e1000_adapter *adapter)
3986 {
3987         struct e1000_hw *hw = &adapter->hw;
3988         bool link_active = 0;
3989         s32 ret_val = 0;
3990
3991         /*
3992          * get_link_status is set on LSC (link status) interrupt or
3993          * Rx sequence error interrupt.  get_link_status will stay
3994          * false until the check_for_link establishes link
3995          * for copper adapters ONLY
3996          */
3997         switch (hw->phy.media_type) {
3998         case e1000_media_type_copper:
3999                 if (hw->mac.get_link_status) {
4000                         ret_val = hw->mac.ops.check_for_link(hw);
4001                         link_active = !hw->mac.get_link_status;
4002                 } else {
4003                         link_active = 1;
4004                 }
4005                 break;
4006         case e1000_media_type_fiber:
4007                 ret_val = hw->mac.ops.check_for_link(hw);
4008                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4009                 break;
4010         case e1000_media_type_internal_serdes:
4011                 ret_val = hw->mac.ops.check_for_link(hw);
4012                 link_active = adapter->hw.mac.serdes_has_link;
4013                 break;
4014         default:
4015         case e1000_media_type_unknown:
4016                 break;
4017         }
4018
4019         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4020             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4021                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4022                 e_info("Gigabit has been disabled, downgrading speed\n");
4023         }
4024
4025         return link_active;
4026 }
4027
4028 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4029 {
4030         /* make sure the receive unit is started */
4031         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4032             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4033                 struct e1000_hw *hw = &adapter->hw;
4034                 u32 rctl = er32(RCTL);
4035                 ew32(RCTL, rctl | E1000_RCTL_EN);
4036                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4037         }
4038 }
4039
4040 /**
4041  * e1000_watchdog - Timer Call-back
4042  * @data: pointer to adapter cast into an unsigned long
4043  **/
4044 static void e1000_watchdog(unsigned long data)
4045 {
4046         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4047
4048         /* Do the rest outside of interrupt context */
4049         schedule_work(&adapter->watchdog_task);
4050
4051         /* TODO: make this use queue_delayed_work() */
4052 }
4053
4054 static void e1000_watchdog_task(struct work_struct *work)
4055 {
4056         struct e1000_adapter *adapter = container_of(work,
4057                                         struct e1000_adapter, watchdog_task);
4058         struct net_device *netdev = adapter->netdev;
4059         struct e1000_mac_info *mac = &adapter->hw.mac;
4060         struct e1000_phy_info *phy = &adapter->hw.phy;
4061         struct e1000_ring *tx_ring = adapter->tx_ring;
4062         struct e1000_hw *hw = &adapter->hw;
4063         u32 link, tctl;
4064         int tx_pending = 0;
4065
4066         link = e1000e_has_link(adapter);
4067         if ((netif_carrier_ok(netdev)) && link) {
4068                 /* Cancel scheduled suspend requests. */
4069                 pm_runtime_resume(netdev->dev.parent);
4070
4071                 e1000e_enable_receives(adapter);
4072                 goto link_up;
4073         }
4074
4075         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4076             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4077                 e1000_update_mng_vlan(adapter);
4078
4079         if (link) {
4080                 if (!netif_carrier_ok(netdev)) {
4081                         bool txb2b = 1;
4082
4083                         /* Cancel scheduled suspend requests. */
4084                         pm_runtime_resume(netdev->dev.parent);
4085
4086                         /* update snapshot of PHY registers on LSC */
4087                         e1000_phy_read_status(adapter);
4088                         mac->ops.get_link_up_info(&adapter->hw,
4089                                                    &adapter->link_speed,
4090                                                    &adapter->link_duplex);
4091                         e1000_print_link_info(adapter);
4092                         /*
4093                          * On supported PHYs, check for duplex mismatch only
4094                          * if link has autonegotiated at 10/100 half
4095                          */
4096                         if ((hw->phy.type == e1000_phy_igp_3 ||
4097                              hw->phy.type == e1000_phy_bm) &&
4098                             (hw->mac.autoneg == true) &&
4099                             (adapter->link_speed == SPEED_10 ||
4100                              adapter->link_speed == SPEED_100) &&
4101                             (adapter->link_duplex == HALF_DUPLEX)) {
4102                                 u16 autoneg_exp;
4103
4104                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4105
4106                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4107                                         e_info("Autonegotiated half duplex but"
4108                                                " link partner cannot autoneg. "
4109                                                " Try forcing full duplex if "
4110                                                "link gets many collisions.\n");
4111                         }
4112
4113                         /* adjust timeout factor according to speed/duplex */
4114                         adapter->tx_timeout_factor = 1;
4115                         switch (adapter->link_speed) {
4116                         case SPEED_10:
4117                                 txb2b = 0;
4118                                 adapter->tx_timeout_factor = 16;
4119                                 break;
4120                         case SPEED_100:
4121                                 txb2b = 0;
4122                                 adapter->tx_timeout_factor = 10;
4123                                 break;
4124                         }
4125
4126                         /*
4127                          * workaround: re-program speed mode bit after
4128                          * link-up event
4129                          */
4130                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4131                             !txb2b) {
4132                                 u32 tarc0;
4133                                 tarc0 = er32(TARC(0));
4134                                 tarc0 &= ~SPEED_MODE_BIT;
4135                                 ew32(TARC(0), tarc0);
4136                         }
4137
4138                         /*
4139                          * disable TSO for pcie and 10/100 speeds, to avoid
4140                          * some hardware issues
4141                          */
4142                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4143                                 switch (adapter->link_speed) {
4144                                 case SPEED_10:
4145                                 case SPEED_100:
4146                                         e_info("10/100 speed: disabling TSO\n");
4147                                         netdev->features &= ~NETIF_F_TSO;
4148                                         netdev->features &= ~NETIF_F_TSO6;
4149                                         break;
4150                                 case SPEED_1000:
4151                                         netdev->features |= NETIF_F_TSO;
4152                                         netdev->features |= NETIF_F_TSO6;
4153                                         break;
4154                                 default:
4155                                         /* oops */
4156                                         break;
4157                                 }
4158                         }
4159
4160                         /*
4161                          * enable transmits in the hardware, need to do this
4162                          * after setting TARC(0)
4163                          */
4164                         tctl = er32(TCTL);
4165                         tctl |= E1000_TCTL_EN;
4166                         ew32(TCTL, tctl);
4167
4168                         /*
4169                          * Perform any post-link-up configuration before
4170                          * reporting link up.
4171                          */
4172                         if (phy->ops.cfg_on_link_up)
4173                                 phy->ops.cfg_on_link_up(hw);
4174
4175                         netif_carrier_on(netdev);
4176
4177                         if (!test_bit(__E1000_DOWN, &adapter->state))
4178                                 mod_timer(&adapter->phy_info_timer,
4179                                           round_jiffies(jiffies + 2 * HZ));
4180                 }
4181         } else {
4182                 if (netif_carrier_ok(netdev)) {
4183                         adapter->link_speed = 0;
4184                         adapter->link_duplex = 0;
4185                         /* Link status message must follow this format */
4186                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4187                                adapter->netdev->name);
4188                         netif_carrier_off(netdev);
4189                         if (!test_bit(__E1000_DOWN, &adapter->state))
4190                                 mod_timer(&adapter->phy_info_timer,
4191                                           round_jiffies(jiffies + 2 * HZ));
4192
4193                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4194                                 schedule_work(&adapter->reset_task);
4195                         else
4196                                 pm_schedule_suspend(netdev->dev.parent,
4197                                                         LINK_TIMEOUT);
4198                 }
4199         }
4200
4201 link_up:
4202         e1000e_update_stats(adapter);
4203
4204         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4205         adapter->tpt_old = adapter->stats.tpt;
4206         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4207         adapter->colc_old = adapter->stats.colc;
4208
4209         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4210         adapter->gorc_old = adapter->stats.gorc;
4211         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4212         adapter->gotc_old = adapter->stats.gotc;
4213
4214         e1000e_update_adaptive(&adapter->hw);
4215
4216         if (!netif_carrier_ok(netdev)) {
4217                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4218                                tx_ring->count);
4219                 if (tx_pending) {
4220                         /*
4221                          * We've lost link, so the controller stops DMA,
4222                          * but we've got queued Tx work that's never going
4223                          * to get done, so reset controller to flush Tx.
4224                          * (Do the reset outside of interrupt context).
4225                          */
4226                         adapter->tx_timeout_count++;
4227                         schedule_work(&adapter->reset_task);
4228                         /* return immediately since reset is imminent */
4229                         return;
4230                 }
4231         }
4232
4233         /* Simple mode for Interrupt Throttle Rate (ITR) */
4234         if (adapter->itr_setting == 4) {
4235                 /*
4236                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4237                  * Total asymmetrical Tx or Rx gets ITR=8000;
4238                  * everyone else is between 2000-8000.
4239                  */
4240                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4241                 u32 dif = (adapter->gotc > adapter->gorc ?
4242                             adapter->gotc - adapter->gorc :
4243                             adapter->gorc - adapter->gotc) / 10000;
4244                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4245
4246                 ew32(ITR, 1000000000 / (itr * 256));
4247         }
4248
4249         /* Cause software interrupt to ensure Rx ring is cleaned */
4250         if (adapter->msix_entries)
4251                 ew32(ICS, adapter->rx_ring->ims_val);
4252         else
4253                 ew32(ICS, E1000_ICS_RXDMT0);
4254
4255         /* Force detection of hung controller every watchdog period */
4256         adapter->detect_tx_hung = 1;
4257
4258         /*
4259          * With 82571 controllers, LAA may be overwritten due to controller
4260          * reset from the other port. Set the appropriate LAA in RAR[0]
4261          */
4262         if (e1000e_get_laa_state_82571(hw))
4263                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4264
4265         /* Reset the timer */
4266         if (!test_bit(__E1000_DOWN, &adapter->state))
4267                 mod_timer(&adapter->watchdog_timer,
4268                           round_jiffies(jiffies + 2 * HZ));
4269 }
4270
4271 #define E1000_TX_FLAGS_CSUM             0x00000001
4272 #define E1000_TX_FLAGS_VLAN             0x00000002
4273 #define E1000_TX_FLAGS_TSO              0x00000004
4274 #define E1000_TX_FLAGS_IPV4             0x00000008
4275 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4276 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4277
4278 static int e1000_tso(struct e1000_adapter *adapter,
4279                      struct sk_buff *skb)
4280 {
4281         struct e1000_ring *tx_ring = adapter->tx_ring;
4282         struct e1000_context_desc *context_desc;
4283         struct e1000_buffer *buffer_info;
4284         unsigned int i;
4285         u32 cmd_length = 0;
4286         u16 ipcse = 0, tucse, mss;
4287         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4288         int err;
4289
4290         if (!skb_is_gso(skb))
4291                 return 0;
4292
4293         if (skb_header_cloned(skb)) {
4294                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4295                 if (err)
4296                         return err;
4297         }
4298
4299         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4300         mss = skb_shinfo(skb)->gso_size;
4301         if (skb->protocol == htons(ETH_P_IP)) {
4302                 struct iphdr *iph = ip_hdr(skb);
4303                 iph->tot_len = 0;
4304                 iph->check = 0;
4305                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4306                                                          0, IPPROTO_TCP, 0);
4307                 cmd_length = E1000_TXD_CMD_IP;
4308                 ipcse = skb_transport_offset(skb) - 1;
4309         } else if (skb_is_gso_v6(skb)) {
4310                 ipv6_hdr(skb)->payload_len = 0;
4311                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4312                                                        &ipv6_hdr(skb)->daddr,
4313                                                        0, IPPROTO_TCP, 0);
4314                 ipcse = 0;
4315         }
4316         ipcss = skb_network_offset(skb);
4317         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4318         tucss = skb_transport_offset(skb);
4319         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4320         tucse = 0;
4321
4322         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4323                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4324
4325         i = tx_ring->next_to_use;
4326         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4327         buffer_info = &tx_ring->buffer_info[i];
4328
4329         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4330         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4331         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4332         context_desc->upper_setup.tcp_fields.tucss = tucss;
4333         context_desc->upper_setup.tcp_fields.tucso = tucso;
4334         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4335         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4336         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4337         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4338
4339         buffer_info->time_stamp = jiffies;
4340         buffer_info->next_to_watch = i;
4341
4342         i++;
4343         if (i == tx_ring->count)
4344                 i = 0;
4345         tx_ring->next_to_use = i;
4346
4347         return 1;
4348 }
4349
4350 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4351 {
4352         struct e1000_ring *tx_ring = adapter->tx_ring;
4353         struct e1000_context_desc *context_desc;
4354         struct e1000_buffer *buffer_info;
4355         unsigned int i;
4356         u8 css;
4357         u32 cmd_len = E1000_TXD_CMD_DEXT;
4358         __be16 protocol;
4359
4360         if (skb->ip_summed != CHECKSUM_PARTIAL)
4361                 return 0;
4362
4363         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4364                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4365         else
4366                 protocol = skb->protocol;
4367
4368         switch (protocol) {
4369         case cpu_to_be16(ETH_P_IP):
4370                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4371                         cmd_len |= E1000_TXD_CMD_TCP;
4372                 break;
4373         case cpu_to_be16(ETH_P_IPV6):
4374                 /* XXX not handling all IPV6 headers */
4375                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4376                         cmd_len |= E1000_TXD_CMD_TCP;
4377                 break;
4378         default:
4379                 if (unlikely(net_ratelimit()))
4380                         e_warn("checksum_partial proto=%x!\n",
4381                                be16_to_cpu(protocol));
4382                 break;
4383         }
4384
4385         css = skb_transport_offset(skb);
4386
4387         i = tx_ring->next_to_use;
4388         buffer_info = &tx_ring->buffer_info[i];
4389         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4390
4391         context_desc->lower_setup.ip_config = 0;
4392         context_desc->upper_setup.tcp_fields.tucss = css;
4393         context_desc->upper_setup.tcp_fields.tucso =
4394                                 css + skb->csum_offset;
4395         context_desc->upper_setup.tcp_fields.tucse = 0;
4396         context_desc->tcp_seg_setup.data = 0;
4397         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4398
4399         buffer_info->time_stamp = jiffies;
4400         buffer_info->next_to_watch = i;
4401
4402         i++;
4403         if (i == tx_ring->count)
4404                 i = 0;
4405         tx_ring->next_to_use = i;
4406
4407         return 1;
4408 }
4409
4410 #define E1000_MAX_PER_TXD       8192
4411 #define E1000_MAX_TXD_PWR       12
4412
4413 static int e1000_tx_map(struct e1000_adapter *adapter,
4414                         struct sk_buff *skb, unsigned int first,
4415                         unsigned int max_per_txd, unsigned int nr_frags,
4416                         unsigned int mss)
4417 {
4418         struct e1000_ring *tx_ring = adapter->tx_ring;
4419         struct pci_dev *pdev = adapter->pdev;
4420         struct e1000_buffer *buffer_info;
4421         unsigned int len = skb_headlen(skb);
4422         unsigned int offset = 0, size, count = 0, i;
4423         unsigned int f, bytecount, segs;
4424
4425         i = tx_ring->next_to_use;
4426
4427         while (len) {
4428                 buffer_info = &tx_ring->buffer_info[i];
4429                 size = min(len, max_per_txd);
4430
4431                 buffer_info->length = size;
4432                 buffer_info->time_stamp = jiffies;
4433                 buffer_info->next_to_watch = i;
4434                 buffer_info->dma = dma_map_single(&pdev->dev,
4435                                                   skb->data + offset,
4436                                                   size, DMA_TO_DEVICE);
4437                 buffer_info->mapped_as_page = false;
4438                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4439                         goto dma_error;
4440
4441                 len -= size;
4442                 offset += size;
4443                 count++;
4444
4445                 if (len) {
4446                         i++;
4447                         if (i == tx_ring->count)
4448                                 i = 0;
4449                 }
4450         }
4451
4452         for (f = 0; f < nr_frags; f++) {
4453                 struct skb_frag_struct *frag;
4454
4455                 frag = &skb_shinfo(skb)->frags[f];
4456                 len = frag->size;
4457                 offset = frag->page_offset;
4458
4459                 while (len) {
4460                         i++;
4461                         if (i == tx_ring->count)
4462                                 i = 0;
4463
4464                         buffer_info = &tx_ring->buffer_info[i];
4465                         size = min(len, max_per_txd);
4466
4467                         buffer_info->length = size;
4468                         buffer_info->time_stamp = jiffies;
4469                         buffer_info->next_to_watch = i;
4470                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4471                                                         offset, size,
4472                                                         DMA_TO_DEVICE);
4473                         buffer_info->mapped_as_page = true;
4474                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4475                                 goto dma_error;
4476
4477                         len -= size;
4478                         offset += size;
4479                         count++;
4480                 }
4481         }
4482
4483         segs = skb_shinfo(skb)->gso_segs ?: 1;
4484         /* multiply data chunks by size of headers */
4485         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4486
4487         tx_ring->buffer_info[i].skb = skb;
4488         tx_ring->buffer_info[i].segs = segs;
4489         tx_ring->buffer_info[i].bytecount = bytecount;
4490         tx_ring->buffer_info[first].next_to_watch = i;
4491
4492         return count;
4493
4494 dma_error:
4495         dev_err(&pdev->dev, "TX DMA map failed\n");
4496         buffer_info->dma = 0;
4497         if (count)
4498                 count--;
4499
4500         while (count--) {
4501                 if (i==0)
4502                         i += tx_ring->count;
4503                 i--;
4504                 buffer_info = &tx_ring->buffer_info[i];
4505                 e1000_put_txbuf(adapter, buffer_info);;
4506         }
4507
4508         return 0;
4509 }
4510
4511 static void e1000_tx_queue(struct e1000_adapter *adapter,
4512                            int tx_flags, int count)
4513 {
4514         struct e1000_ring *tx_ring = adapter->tx_ring;
4515         struct e1000_tx_desc *tx_desc = NULL;
4516         struct e1000_buffer *buffer_info;
4517         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4518         unsigned int i;
4519
4520         if (tx_flags & E1000_TX_FLAGS_TSO) {
4521                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4522                              E1000_TXD_CMD_TSE;
4523                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4524
4525                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4526                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4527         }
4528
4529         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4530                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4531                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4532         }
4533
4534         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4535                 txd_lower |= E1000_TXD_CMD_VLE;
4536                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4537         }
4538
4539         i = tx_ring->next_to_use;
4540
4541         while (count--) {
4542                 buffer_info = &tx_ring->buffer_info[i];
4543                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4544                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4545                 tx_desc->lower.data =
4546                         cpu_to_le32(txd_lower | buffer_info->length);
4547                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4548
4549                 i++;
4550                 if (i == tx_ring->count)
4551                         i = 0;
4552         }
4553
4554         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4555
4556         /*
4557          * Force memory writes to complete before letting h/w
4558          * know there are new descriptors to fetch.  (Only
4559          * applicable for weak-ordered memory model archs,
4560          * such as IA-64).
4561          */
4562         wmb();
4563
4564         tx_ring->next_to_use = i;
4565         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4566         /*
4567          * we need this if more than one processor can write to our tail
4568          * at a time, it synchronizes IO on IA64/Altix systems
4569          */
4570         mmiowb();
4571 }
4572
4573 #define MINIMUM_DHCP_PACKET_SIZE 282
4574 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4575                                     struct sk_buff *skb)
4576 {
4577         struct e1000_hw *hw =  &adapter->hw;
4578         u16 length, offset;
4579
4580         if (vlan_tx_tag_present(skb)) {
4581                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4582                     (adapter->hw.mng_cookie.status &
4583                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4584                         return 0;
4585         }
4586
4587         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4588                 return 0;
4589
4590         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4591                 return 0;
4592
4593         {
4594                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4595                 struct udphdr *udp;
4596
4597                 if (ip->protocol != IPPROTO_UDP)
4598                         return 0;
4599
4600                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4601                 if (ntohs(udp->dest) != 67)
4602                         return 0;
4603
4604                 offset = (u8 *)udp + 8 - skb->data;
4605                 length = skb->len - offset;
4606                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4607         }
4608
4609         return 0;
4610 }
4611
4612 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4613 {
4614         struct e1000_adapter *adapter = netdev_priv(netdev);
4615
4616         netif_stop_queue(netdev);
4617         /*
4618          * Herbert's original patch had:
4619          *  smp_mb__after_netif_stop_queue();
4620          * but since that doesn't exist yet, just open code it.
4621          */
4622         smp_mb();
4623
4624         /*
4625          * We need to check again in a case another CPU has just
4626          * made room available.
4627          */
4628         if (e1000_desc_unused(adapter->tx_ring) < size)
4629                 return -EBUSY;
4630
4631         /* A reprieve! */
4632         netif_start_queue(netdev);
4633         ++adapter->restart_queue;
4634         return 0;
4635 }
4636
4637 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4638 {
4639         struct e1000_adapter *adapter = netdev_priv(netdev);
4640
4641         if (e1000_desc_unused(adapter->tx_ring) >= size)
4642                 return 0;
4643         return __e1000_maybe_stop_tx(netdev, size);
4644 }
4645
4646 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4647 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4648                                     struct net_device *netdev)
4649 {
4650         struct e1000_adapter *adapter = netdev_priv(netdev);
4651         struct e1000_ring *tx_ring = adapter->tx_ring;
4652         unsigned int first;
4653         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4654         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4655         unsigned int tx_flags = 0;
4656         unsigned int len = skb_headlen(skb);
4657         unsigned int nr_frags;
4658         unsigned int mss;
4659         int count = 0;
4660         int tso;
4661         unsigned int f;
4662
4663         if (test_bit(__E1000_DOWN, &adapter->state)) {
4664                 dev_kfree_skb_any(skb);
4665                 return NETDEV_TX_OK;
4666         }
4667
4668         if (skb->len <= 0) {
4669                 dev_kfree_skb_any(skb);
4670                 return NETDEV_TX_OK;
4671         }
4672
4673         mss = skb_shinfo(skb)->gso_size;
4674         /*
4675          * The controller does a simple calculation to
4676          * make sure there is enough room in the FIFO before
4677          * initiating the DMA for each buffer.  The calc is:
4678          * 4 = ceil(buffer len/mss).  To make sure we don't
4679          * overrun the FIFO, adjust the max buffer len if mss
4680          * drops.
4681          */
4682         if (mss) {
4683                 u8 hdr_len;
4684                 max_per_txd = min(mss << 2, max_per_txd);
4685                 max_txd_pwr = fls(max_per_txd) - 1;
4686
4687                 /*
4688                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4689                  * points to just header, pull a few bytes of payload from
4690                  * frags into skb->data
4691                  */
4692                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4693                 /*
4694                  * we do this workaround for ES2LAN, but it is un-necessary,
4695                  * avoiding it could save a lot of cycles
4696                  */
4697                 if (skb->data_len && (hdr_len == len)) {
4698                         unsigned int pull_size;
4699
4700                         pull_size = min((unsigned int)4, skb->data_len);
4701                         if (!__pskb_pull_tail(skb, pull_size)) {
4702                                 e_err("__pskb_pull_tail failed.\n");
4703                                 dev_kfree_skb_any(skb);
4704                                 return NETDEV_TX_OK;
4705                         }
4706                         len = skb_headlen(skb);
4707                 }
4708         }
4709
4710         /* reserve a descriptor for the offload context */
4711         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4712                 count++;
4713         count++;
4714
4715         count += TXD_USE_COUNT(len, max_txd_pwr);
4716
4717         nr_frags = skb_shinfo(skb)->nr_frags;
4718         for (f = 0; f < nr_frags; f++)
4719                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4720                                        max_txd_pwr);
4721
4722         if (adapter->hw.mac.tx_pkt_filtering)
4723                 e1000_transfer_dhcp_info(adapter, skb);
4724
4725         /*
4726          * need: count + 2 desc gap to keep tail from touching
4727          * head, otherwise try next time
4728          */
4729         if (e1000_maybe_stop_tx(netdev, count + 2))
4730                 return NETDEV_TX_BUSY;
4731
4732         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4733                 tx_flags |= E1000_TX_FLAGS_VLAN;
4734                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4735         }
4736
4737         first = tx_ring->next_to_use;
4738
4739         tso = e1000_tso(adapter, skb);
4740         if (tso < 0) {
4741                 dev_kfree_skb_any(skb);
4742                 return NETDEV_TX_OK;
4743         }
4744
4745         if (tso)
4746                 tx_flags |= E1000_TX_FLAGS_TSO;
4747         else if (e1000_tx_csum(adapter, skb))
4748                 tx_flags |= E1000_TX_FLAGS_CSUM;
4749
4750         /*
4751          * Old method was to assume IPv4 packet by default if TSO was enabled.
4752          * 82571 hardware supports TSO capabilities for IPv6 as well...
4753          * no longer assume, we must.
4754          */
4755         if (skb->protocol == htons(ETH_P_IP))
4756                 tx_flags |= E1000_TX_FLAGS_IPV4;
4757
4758         /* if count is 0 then mapping error has occured */
4759         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4760         if (count) {
4761                 e1000_tx_queue(adapter, tx_flags, count);
4762                 /* Make sure there is space in the ring for the next send. */
4763                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4764
4765         } else {
4766                 dev_kfree_skb_any(skb);
4767                 tx_ring->buffer_info[first].time_stamp = 0;
4768                 tx_ring->next_to_use = first;
4769         }
4770
4771         return NETDEV_TX_OK;
4772 }
4773
4774 /**
4775  * e1000_tx_timeout - Respond to a Tx Hang
4776  * @netdev: network interface device structure
4777  **/
4778 static void e1000_tx_timeout(struct net_device *netdev)
4779 {
4780         struct e1000_adapter *adapter = netdev_priv(netdev);
4781
4782         /* Do the reset outside of interrupt context */
4783         adapter->tx_timeout_count++;
4784         schedule_work(&adapter->reset_task);
4785 }
4786
4787 static void e1000_reset_task(struct work_struct *work)
4788 {
4789         struct e1000_adapter *adapter;
4790         adapter = container_of(work, struct e1000_adapter, reset_task);
4791
4792         e1000e_dump(adapter);
4793         e_err("Reset adapter\n");
4794         e1000e_reinit_locked(adapter);
4795 }
4796
4797 /**
4798  * e1000_get_stats - Get System Network Statistics
4799  * @netdev: network interface device structure
4800  *
4801  * Returns the address of the device statistics structure.
4802  * The statistics are actually updated from the timer callback.
4803  **/
4804 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4805 {
4806         /* only return the current stats */
4807         return &netdev->stats;
4808 }
4809
4810 /**
4811  * e1000_change_mtu - Change the Maximum Transfer Unit
4812  * @netdev: network interface device structure
4813  * @new_mtu: new value for maximum frame size
4814  *
4815  * Returns 0 on success, negative on failure
4816  **/
4817 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4818 {
4819         struct e1000_adapter *adapter = netdev_priv(netdev);
4820         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4821
4822         /* Jumbo frame support */
4823         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4824             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4825                 e_err("Jumbo Frames not supported.\n");
4826                 return -EINVAL;
4827         }
4828
4829         /* Supported frame sizes */
4830         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4831             (max_frame > adapter->max_hw_frame_size)) {
4832                 e_err("Unsupported MTU setting\n");
4833                 return -EINVAL;
4834         }
4835
4836         /* 82573 Errata 17 */
4837         if (((adapter->hw.mac.type == e1000_82573) ||
4838              (adapter->hw.mac.type == e1000_82574)) &&
4839             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4840                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4841                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4842         }
4843
4844         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4845                 msleep(1);
4846         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4847         adapter->max_frame_size = max_frame;
4848         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4849         netdev->mtu = new_mtu;
4850         if (netif_running(netdev))
4851                 e1000e_down(adapter);
4852
4853         /*
4854          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4855          * means we reserve 2 more, this pushes us to allocate from the next
4856          * larger slab size.
4857          * i.e. RXBUFFER_2048 --> size-4096 slab
4858          * However with the new *_jumbo_rx* routines, jumbo receives will use
4859          * fragmented skbs
4860          */
4861
4862         if (max_frame <= 2048)
4863                 adapter->rx_buffer_len = 2048;
4864         else
4865                 adapter->rx_buffer_len = 4096;
4866
4867         /* adjust allocation if LPE protects us, and we aren't using SBP */
4868         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4869              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4870                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4871                                          + ETH_FCS_LEN;
4872
4873         if (netif_running(netdev))
4874                 e1000e_up(adapter);
4875         else
4876                 e1000e_reset(adapter);
4877
4878         clear_bit(__E1000_RESETTING, &adapter->state);
4879
4880         return 0;
4881 }
4882
4883 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4884                            int cmd)
4885 {
4886         struct e1000_adapter *adapter = netdev_priv(netdev);
4887         struct mii_ioctl_data *data = if_mii(ifr);
4888
4889         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4890                 return -EOPNOTSUPP;
4891
4892         switch (cmd) {
4893         case SIOCGMIIPHY:
4894                 data->phy_id = adapter->hw.phy.addr;
4895                 break;
4896         case SIOCGMIIREG:
4897                 e1000_phy_read_status(adapter);
4898
4899                 switch (data->reg_num & 0x1F) {
4900                 case MII_BMCR:
4901                         data->val_out = adapter->phy_regs.bmcr;
4902                         break;
4903                 case MII_BMSR:
4904                         data->val_out = adapter->phy_regs.bmsr;
4905                         break;
4906                 case MII_PHYSID1:
4907                         data->val_out = (adapter->hw.phy.id >> 16);
4908                         break;
4909                 case MII_PHYSID2:
4910                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4911                         break;
4912                 case MII_ADVERTISE:
4913                         data->val_out = adapter->phy_regs.advertise;
4914                         break;
4915                 case MII_LPA:
4916                         data->val_out = adapter->phy_regs.lpa;
4917                         break;
4918                 case MII_EXPANSION:
4919                         data->val_out = adapter->phy_regs.expansion;
4920                         break;
4921                 case MII_CTRL1000:
4922                         data->val_out = adapter->phy_regs.ctrl1000;
4923                         break;
4924                 case MII_STAT1000:
4925                         data->val_out = adapter->phy_regs.stat1000;
4926                         break;
4927                 case MII_ESTATUS:
4928                         data->val_out = adapter->phy_regs.estatus;
4929                         break;
4930                 default:
4931                         return -EIO;
4932                 }
4933                 break;
4934         case SIOCSMIIREG:
4935         default:
4936                 return -EOPNOTSUPP;
4937         }
4938         return 0;
4939 }
4940
4941 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4942 {
4943         switch (cmd) {
4944         case SIOCGMIIPHY:
4945         case SIOCGMIIREG:
4946         case SIOCSMIIREG:
4947                 return e1000_mii_ioctl(netdev, ifr, cmd);
4948         default:
4949                 return -EOPNOTSUPP;
4950         }
4951 }
4952
4953 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4954 {
4955         struct e1000_hw *hw = &adapter->hw;
4956         u32 i, mac_reg;
4957         u16 phy_reg;
4958         int retval = 0;
4959
4960         /* copy MAC RARs to PHY RARs */
4961         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
4962
4963         /* copy MAC MTA to PHY MTA */
4964         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4965                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4966                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4967                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4968         }
4969
4970         /* configure PHY Rx Control register */
4971         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4972         mac_reg = er32(RCTL);
4973         if (mac_reg & E1000_RCTL_UPE)
4974                 phy_reg |= BM_RCTL_UPE;
4975         if (mac_reg & E1000_RCTL_MPE)
4976                 phy_reg |= BM_RCTL_MPE;
4977         phy_reg &= ~(BM_RCTL_MO_MASK);
4978         if (mac_reg & E1000_RCTL_MO_3)
4979                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4980                                 << BM_RCTL_MO_SHIFT);
4981         if (mac_reg & E1000_RCTL_BAM)
4982                 phy_reg |= BM_RCTL_BAM;
4983         if (mac_reg & E1000_RCTL_PMCF)
4984                 phy_reg |= BM_RCTL_PMCF;
4985         mac_reg = er32(CTRL);
4986         if (mac_reg & E1000_CTRL_RFCE)
4987                 phy_reg |= BM_RCTL_RFCE;
4988         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4989
4990         /* enable PHY wakeup in MAC register */
4991         ew32(WUFC, wufc);
4992         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4993
4994         /* configure and enable PHY wakeup in PHY registers */
4995         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4996         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4997
4998         /* activate PHY wakeup */
4999         retval = hw->phy.ops.acquire(hw);
5000         if (retval) {
5001                 e_err("Could not acquire PHY\n");
5002                 return retval;
5003         }
5004         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5005                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5006         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5007         if (retval) {
5008                 e_err("Could not read PHY page 769\n");
5009                 goto out;
5010         }
5011         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5012         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5013         if (retval)
5014                 e_err("Could not set PHY Host Wakeup bit\n");
5015 out:
5016         hw->phy.ops.release(hw);
5017
5018         return retval;
5019 }
5020
5021 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5022                             bool runtime)
5023 {
5024         struct net_device *netdev = pci_get_drvdata(pdev);
5025         struct e1000_adapter *adapter = netdev_priv(netdev);
5026         struct e1000_hw *hw = &adapter->hw;
5027         u32 ctrl, ctrl_ext, rctl, status;
5028         /* Runtime suspend should only enable wakeup for link changes */
5029         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5030         int retval = 0;
5031
5032         netif_device_detach(netdev);
5033
5034         if (netif_running(netdev)) {
5035                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5036                 e1000e_down(adapter);
5037                 e1000_free_irq(adapter);
5038         }
5039         e1000e_reset_interrupt_capability(adapter);
5040
5041         retval = pci_save_state(pdev);
5042         if (retval)
5043                 return retval;
5044
5045         status = er32(STATUS);
5046         if (status & E1000_STATUS_LU)
5047                 wufc &= ~E1000_WUFC_LNKC;
5048
5049         if (wufc) {
5050                 e1000_setup_rctl(adapter);
5051                 e1000_set_multi(netdev);
5052
5053                 /* turn on all-multi mode if wake on multicast is enabled */
5054                 if (wufc & E1000_WUFC_MC) {
5055                         rctl = er32(RCTL);
5056                         rctl |= E1000_RCTL_MPE;
5057                         ew32(RCTL, rctl);
5058                 }
5059
5060                 ctrl = er32(CTRL);
5061                 /* advertise wake from D3Cold */
5062                 #define E1000_CTRL_ADVD3WUC 0x00100000
5063                 /* phy power management enable */
5064                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5065                 ctrl |= E1000_CTRL_ADVD3WUC;
5066                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5067                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5068                 ew32(CTRL, ctrl);
5069
5070                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5071                     adapter->hw.phy.media_type ==
5072                     e1000_media_type_internal_serdes) {
5073                         /* keep the laser running in D3 */
5074                         ctrl_ext = er32(CTRL_EXT);
5075                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5076                         ew32(CTRL_EXT, ctrl_ext);
5077                 }
5078
5079                 if (adapter->flags & FLAG_IS_ICH)
5080                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5081
5082                 /* Allow time for pending master requests to run */
5083                 e1000e_disable_pcie_master(&adapter->hw);
5084
5085                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5086                         /* enable wakeup by the PHY */
5087                         retval = e1000_init_phy_wakeup(adapter, wufc);
5088                         if (retval)
5089                                 return retval;
5090                 } else {
5091                         /* enable wakeup by the MAC */
5092                         ew32(WUFC, wufc);
5093                         ew32(WUC, E1000_WUC_PME_EN);
5094                 }
5095         } else {
5096                 ew32(WUC, 0);
5097                 ew32(WUFC, 0);
5098         }
5099
5100         *enable_wake = !!wufc;
5101
5102         /* make sure adapter isn't asleep if manageability is enabled */
5103         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5104             (hw->mac.ops.check_mng_mode(hw)))
5105                 *enable_wake = true;
5106
5107         if (adapter->hw.phy.type == e1000_phy_igp_3)
5108                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5109
5110         /*
5111          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5112          * would have already happened in close and is redundant.
5113          */
5114         e1000_release_hw_control(adapter);
5115
5116         pci_disable_device(pdev);
5117
5118         return 0;
5119 }
5120
5121 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5122 {
5123         if (sleep && wake) {
5124                 pci_prepare_to_sleep(pdev);
5125                 return;
5126         }
5127
5128         pci_wake_from_d3(pdev, wake);
5129         pci_set_power_state(pdev, PCI_D3hot);
5130 }
5131
5132 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5133                                     bool wake)
5134 {
5135         struct net_device *netdev = pci_get_drvdata(pdev);
5136         struct e1000_adapter *adapter = netdev_priv(netdev);
5137
5138         /*
5139          * The pci-e switch on some quad port adapters will report a
5140          * correctable error when the MAC transitions from D0 to D3.  To
5141          * prevent this we need to mask off the correctable errors on the
5142          * downstream port of the pci-e switch.
5143          */
5144         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5145                 struct pci_dev *us_dev = pdev->bus->self;
5146                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5147                 u16 devctl;
5148
5149                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5150                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5151                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5152
5153                 e1000_power_off(pdev, sleep, wake);
5154
5155                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5156         } else {
5157                 e1000_power_off(pdev, sleep, wake);
5158         }
5159 }
5160
5161 #ifdef CONFIG_PCIEASPM
5162 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5163 {
5164         pci_disable_link_state(pdev, state);
5165 }
5166 #else
5167 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5168 {
5169         int pos;
5170         u16 reg16;
5171
5172         /*
5173          * Both device and parent should have the same ASPM setting.
5174          * Disable ASPM in downstream component first and then upstream.
5175          */
5176         pos = pci_pcie_cap(pdev);
5177         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5178         reg16 &= ~state;
5179         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5180
5181         if (!pdev->bus->self)
5182                 return;
5183
5184         pos = pci_pcie_cap(pdev->bus->self);
5185         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5186         reg16 &= ~state;
5187         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5188 }
5189 #endif
5190 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5191 {
5192         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5193                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5194                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5195
5196         __e1000e_disable_aspm(pdev, state);
5197 }
5198
5199 #ifdef CONFIG_PM_OPS
5200 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5201 {
5202         return !!adapter->tx_ring->buffer_info;
5203 }
5204
5205 static int __e1000_resume(struct pci_dev *pdev)
5206 {
5207         struct net_device *netdev = pci_get_drvdata(pdev);
5208         struct e1000_adapter *adapter = netdev_priv(netdev);
5209         struct e1000_hw *hw = &adapter->hw;
5210         u32 err;
5211
5212         pci_set_power_state(pdev, PCI_D0);
5213         pci_restore_state(pdev);
5214         pci_save_state(pdev);
5215         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5216                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5217
5218         e1000e_set_interrupt_capability(adapter);
5219         if (netif_running(netdev)) {
5220                 err = e1000_request_irq(adapter);
5221                 if (err)
5222                         return err;
5223         }
5224
5225         e1000e_power_up_phy(adapter);
5226
5227         /* report the system wakeup cause from S3/S4 */
5228         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5229                 u16 phy_data;
5230
5231                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5232                 if (phy_data) {
5233                         e_info("PHY Wakeup cause - %s\n",
5234                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5235                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5236                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5237                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5238                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5239                                 " Change" : "other");
5240                 }
5241                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5242         } else {
5243                 u32 wus = er32(WUS);
5244                 if (wus) {
5245                         e_info("MAC Wakeup cause - %s\n",
5246                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5247                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5248                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5249                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5250                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5251                                 "other");
5252                 }
5253                 ew32(WUS, ~0);
5254         }
5255
5256         e1000e_reset(adapter);
5257
5258         e1000_init_manageability_pt(adapter);
5259
5260         if (netif_running(netdev))
5261                 e1000e_up(adapter);
5262
5263         netif_device_attach(netdev);
5264
5265         /*
5266          * If the controller has AMT, do not set DRV_LOAD until the interface
5267          * is up.  For all other cases, let the f/w know that the h/w is now
5268          * under the control of the driver.
5269          */
5270         if (!(adapter->flags & FLAG_HAS_AMT))
5271                 e1000_get_hw_control(adapter);
5272
5273         return 0;
5274 }
5275
5276 #ifdef CONFIG_PM_SLEEP
5277 static int e1000_suspend(struct device *dev)
5278 {
5279         struct pci_dev *pdev = to_pci_dev(dev);
5280         int retval;
5281         bool wake;
5282
5283         retval = __e1000_shutdown(pdev, &wake, false);
5284         if (!retval)
5285                 e1000_complete_shutdown(pdev, true, wake);
5286
5287         return retval;
5288 }
5289
5290 static int e1000_resume(struct device *dev)
5291 {
5292         struct pci_dev *pdev = to_pci_dev(dev);
5293         struct net_device *netdev = pci_get_drvdata(pdev);
5294         struct e1000_adapter *adapter = netdev_priv(netdev);
5295
5296         if (e1000e_pm_ready(adapter))
5297                 adapter->idle_check = true;
5298
5299         return __e1000_resume(pdev);
5300 }
5301 #endif /* CONFIG_PM_SLEEP */
5302
5303 #ifdef CONFIG_PM_RUNTIME
5304 static int e1000_runtime_suspend(struct device *dev)
5305 {
5306         struct pci_dev *pdev = to_pci_dev(dev);
5307         struct net_device *netdev = pci_get_drvdata(pdev);
5308         struct e1000_adapter *adapter = netdev_priv(netdev);
5309
5310         if (e1000e_pm_ready(adapter)) {
5311                 bool wake;
5312
5313                 __e1000_shutdown(pdev, &wake, true);
5314         }
5315
5316         return 0;
5317 }
5318
5319 static int e1000_idle(struct device *dev)
5320 {
5321         struct pci_dev *pdev = to_pci_dev(dev);
5322         struct net_device *netdev = pci_get_drvdata(pdev);
5323         struct e1000_adapter *adapter = netdev_priv(netdev);
5324
5325         if (!e1000e_pm_ready(adapter))
5326                 return 0;
5327
5328         if (adapter->idle_check) {
5329                 adapter->idle_check = false;
5330                 if (!e1000e_has_link(adapter))
5331                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5332         }
5333
5334         return -EBUSY;
5335 }
5336
5337 static int e1000_runtime_resume(struct device *dev)
5338 {
5339         struct pci_dev *pdev = to_pci_dev(dev);
5340         struct net_device *netdev = pci_get_drvdata(pdev);
5341         struct e1000_adapter *adapter = netdev_priv(netdev);
5342
5343         if (!e1000e_pm_ready(adapter))
5344                 return 0;
5345
5346         adapter->idle_check = !dev->power.runtime_auto;
5347         return __e1000_resume(pdev);
5348 }
5349 #endif /* CONFIG_PM_RUNTIME */
5350 #endif /* CONFIG_PM_OPS */
5351
5352 static void e1000_shutdown(struct pci_dev *pdev)
5353 {
5354         bool wake = false;
5355
5356         __e1000_shutdown(pdev, &wake, false);
5357
5358         if (system_state == SYSTEM_POWER_OFF)
5359                 e1000_complete_shutdown(pdev, false, wake);
5360 }
5361
5362 #ifdef CONFIG_NET_POLL_CONTROLLER
5363 /*
5364  * Polling 'interrupt' - used by things like netconsole to send skbs
5365  * without having to re-enable interrupts. It's not called while
5366  * the interrupt routine is executing.
5367  */
5368 static void e1000_netpoll(struct net_device *netdev)
5369 {
5370         struct e1000_adapter *adapter = netdev_priv(netdev);
5371
5372         disable_irq(adapter->pdev->irq);
5373         e1000_intr(adapter->pdev->irq, netdev);
5374
5375         enable_irq(adapter->pdev->irq);
5376 }
5377 #endif
5378
5379 /**
5380  * e1000_io_error_detected - called when PCI error is detected
5381  * @pdev: Pointer to PCI device
5382  * @state: The current pci connection state
5383  *
5384  * This function is called after a PCI bus error affecting
5385  * this device has been detected.
5386  */
5387 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5388                                                 pci_channel_state_t state)
5389 {
5390         struct net_device *netdev = pci_get_drvdata(pdev);
5391         struct e1000_adapter *adapter = netdev_priv(netdev);
5392
5393         netif_device_detach(netdev);
5394
5395         if (state == pci_channel_io_perm_failure)
5396                 return PCI_ERS_RESULT_DISCONNECT;
5397
5398         if (netif_running(netdev))
5399                 e1000e_down(adapter);
5400         pci_disable_device(pdev);
5401
5402         /* Request a slot slot reset. */
5403         return PCI_ERS_RESULT_NEED_RESET;
5404 }
5405
5406 /**
5407  * e1000_io_slot_reset - called after the pci bus has been reset.
5408  * @pdev: Pointer to PCI device
5409  *
5410  * Restart the card from scratch, as if from a cold-boot. Implementation
5411  * resembles the first-half of the e1000_resume routine.
5412  */
5413 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5414 {
5415         struct net_device *netdev = pci_get_drvdata(pdev);
5416         struct e1000_adapter *adapter = netdev_priv(netdev);
5417         struct e1000_hw *hw = &adapter->hw;
5418         int err;
5419         pci_ers_result_t result;
5420
5421         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5422                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5423         err = pci_enable_device_mem(pdev);
5424         if (err) {
5425                 dev_err(&pdev->dev,
5426                         "Cannot re-enable PCI device after reset.\n");
5427                 result = PCI_ERS_RESULT_DISCONNECT;
5428         } else {
5429                 pci_set_master(pdev);
5430                 pdev->state_saved = true;
5431                 pci_restore_state(pdev);
5432
5433                 pci_enable_wake(pdev, PCI_D3hot, 0);
5434                 pci_enable_wake(pdev, PCI_D3cold, 0);
5435
5436                 e1000e_reset(adapter);
5437                 ew32(WUS, ~0);
5438                 result = PCI_ERS_RESULT_RECOVERED;
5439         }
5440
5441         pci_cleanup_aer_uncorrect_error_status(pdev);
5442
5443         return result;
5444 }
5445
5446 /**
5447  * e1000_io_resume - called when traffic can start flowing again.
5448  * @pdev: Pointer to PCI device
5449  *
5450  * This callback is called when the error recovery driver tells us that
5451  * its OK to resume normal operation. Implementation resembles the
5452  * second-half of the e1000_resume routine.
5453  */
5454 static void e1000_io_resume(struct pci_dev *pdev)
5455 {
5456         struct net_device *netdev = pci_get_drvdata(pdev);
5457         struct e1000_adapter *adapter = netdev_priv(netdev);
5458
5459         e1000_init_manageability_pt(adapter);
5460
5461         if (netif_running(netdev)) {
5462                 if (e1000e_up(adapter)) {
5463                         dev_err(&pdev->dev,
5464                                 "can't bring device back up after reset\n");
5465                         return;
5466                 }
5467         }
5468
5469         netif_device_attach(netdev);
5470
5471         /*
5472          * If the controller has AMT, do not set DRV_LOAD until the interface
5473          * is up.  For all other cases, let the f/w know that the h/w is now
5474          * under the control of the driver.
5475          */
5476         if (!(adapter->flags & FLAG_HAS_AMT))
5477                 e1000_get_hw_control(adapter);
5478
5479 }
5480
5481 static void e1000_print_device_info(struct e1000_adapter *adapter)
5482 {
5483         struct e1000_hw *hw = &adapter->hw;
5484         struct net_device *netdev = adapter->netdev;
5485         u32 pba_num;
5486
5487         /* print bus type/speed/width info */
5488         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5489                /* bus width */
5490                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5491                 "Width x1"),
5492                /* MAC address */
5493                netdev->dev_addr);
5494         e_info("Intel(R) PRO/%s Network Connection\n",
5495                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5496         e1000e_read_pba_num(hw, &pba_num);
5497         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5498                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5499 }
5500
5501 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5502 {
5503         struct e1000_hw *hw = &adapter->hw;
5504         int ret_val;
5505         u16 buf = 0;
5506
5507         if (hw->mac.type != e1000_82573)
5508                 return;
5509
5510         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5511         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5512                 /* Deep Smart Power Down (DSPD) */
5513                 dev_warn(&adapter->pdev->dev,
5514                          "Warning: detected DSPD enabled in EEPROM\n");
5515         }
5516 }
5517
5518 static const struct net_device_ops e1000e_netdev_ops = {
5519         .ndo_open               = e1000_open,
5520         .ndo_stop               = e1000_close,
5521         .ndo_start_xmit         = e1000_xmit_frame,
5522         .ndo_get_stats          = e1000_get_stats,
5523         .ndo_set_multicast_list = e1000_set_multi,
5524         .ndo_set_mac_address    = e1000_set_mac,
5525         .ndo_change_mtu         = e1000_change_mtu,
5526         .ndo_do_ioctl           = e1000_ioctl,
5527         .ndo_tx_timeout         = e1000_tx_timeout,
5528         .ndo_validate_addr      = eth_validate_addr,
5529
5530         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5531         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5532         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5533 #ifdef CONFIG_NET_POLL_CONTROLLER
5534         .ndo_poll_controller    = e1000_netpoll,
5535 #endif
5536 };
5537
5538 /**
5539  * e1000_probe - Device Initialization Routine
5540  * @pdev: PCI device information struct
5541  * @ent: entry in e1000_pci_tbl
5542  *
5543  * Returns 0 on success, negative on failure
5544  *
5545  * e1000_probe initializes an adapter identified by a pci_dev structure.
5546  * The OS initialization, configuring of the adapter private structure,
5547  * and a hardware reset occur.
5548  **/
5549 static int __devinit e1000_probe(struct pci_dev *pdev,
5550                                  const struct pci_device_id *ent)
5551 {
5552         struct net_device *netdev;
5553         struct e1000_adapter *adapter;
5554         struct e1000_hw *hw;
5555         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5556         resource_size_t mmio_start, mmio_len;
5557         resource_size_t flash_start, flash_len;
5558
5559         static int cards_found;
5560         int i, err, pci_using_dac;
5561         u16 eeprom_data = 0;
5562         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5563
5564         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5565                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5566
5567         err = pci_enable_device_mem(pdev);
5568         if (err)
5569                 return err;
5570
5571         pci_using_dac = 0;
5572         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5573         if (!err) {
5574                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5575                 if (!err)
5576                         pci_using_dac = 1;
5577         } else {
5578                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5579                 if (err) {
5580                         err = dma_set_coherent_mask(&pdev->dev,
5581                                                     DMA_BIT_MASK(32));
5582                         if (err) {
5583                                 dev_err(&pdev->dev, "No usable DMA "
5584                                         "configuration, aborting\n");
5585                                 goto err_dma;
5586                         }
5587                 }
5588         }
5589
5590         err = pci_request_selected_regions_exclusive(pdev,
5591                                           pci_select_bars(pdev, IORESOURCE_MEM),
5592                                           e1000e_driver_name);
5593         if (err)
5594                 goto err_pci_reg;
5595
5596         /* AER (Advanced Error Reporting) hooks */
5597         pci_enable_pcie_error_reporting(pdev);
5598
5599         pci_set_master(pdev);
5600         /* PCI config space info */
5601         err = pci_save_state(pdev);
5602         if (err)
5603                 goto err_alloc_etherdev;
5604
5605         err = -ENOMEM;
5606         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5607         if (!netdev)
5608                 goto err_alloc_etherdev;
5609
5610         SET_NETDEV_DEV(netdev, &pdev->dev);
5611
5612         netdev->irq = pdev->irq;
5613
5614         pci_set_drvdata(pdev, netdev);
5615         adapter = netdev_priv(netdev);
5616         hw = &adapter->hw;
5617         adapter->netdev = netdev;
5618         adapter->pdev = pdev;
5619         adapter->ei = ei;
5620         adapter->pba = ei->pba;
5621         adapter->flags = ei->flags;
5622         adapter->flags2 = ei->flags2;
5623         adapter->hw.adapter = adapter;
5624         adapter->hw.mac.type = ei->mac;
5625         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5626         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5627
5628         mmio_start = pci_resource_start(pdev, 0);
5629         mmio_len = pci_resource_len(pdev, 0);
5630
5631         err = -EIO;
5632         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5633         if (!adapter->hw.hw_addr)
5634                 goto err_ioremap;
5635
5636         if ((adapter->flags & FLAG_HAS_FLASH) &&
5637             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5638                 flash_start = pci_resource_start(pdev, 1);
5639                 flash_len = pci_resource_len(pdev, 1);
5640                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5641                 if (!adapter->hw.flash_address)
5642                         goto err_flashmap;
5643         }
5644
5645         /* construct the net_device struct */
5646         netdev->netdev_ops              = &e1000e_netdev_ops;
5647         e1000e_set_ethtool_ops(netdev);
5648         netdev->watchdog_timeo          = 5 * HZ;
5649         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5650         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5651
5652         netdev->mem_start = mmio_start;
5653         netdev->mem_end = mmio_start + mmio_len;
5654
5655         adapter->bd_number = cards_found++;
5656
5657         e1000e_check_options(adapter);
5658
5659         /* setup adapter struct */
5660         err = e1000_sw_init(adapter);
5661         if (err)
5662                 goto err_sw_init;
5663
5664         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5665         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5666         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5667
5668         err = ei->get_variants(adapter);
5669         if (err)
5670                 goto err_hw_init;
5671
5672         if ((adapter->flags & FLAG_IS_ICH) &&
5673             (adapter->flags & FLAG_READ_ONLY_NVM))
5674                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5675
5676         hw->mac.ops.get_bus_info(&adapter->hw);
5677
5678         adapter->hw.phy.autoneg_wait_to_complete = 0;
5679
5680         /* Copper options */
5681         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5682                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5683                 adapter->hw.phy.disable_polarity_correction = 0;
5684                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5685         }
5686
5687         if (e1000_check_reset_block(&adapter->hw))
5688                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5689
5690         netdev->features = NETIF_F_SG |
5691                            NETIF_F_HW_CSUM |
5692                            NETIF_F_HW_VLAN_TX |
5693                            NETIF_F_HW_VLAN_RX;
5694
5695         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5696                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5697
5698         netdev->features |= NETIF_F_TSO;
5699         netdev->features |= NETIF_F_TSO6;
5700
5701         netdev->vlan_features |= NETIF_F_TSO;
5702         netdev->vlan_features |= NETIF_F_TSO6;
5703         netdev->vlan_features |= NETIF_F_HW_CSUM;
5704         netdev->vlan_features |= NETIF_F_SG;
5705
5706         if (pci_using_dac)
5707                 netdev->features |= NETIF_F_HIGHDMA;
5708
5709         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5710                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5711
5712         /*
5713          * before reading the NVM, reset the controller to
5714          * put the device in a known good starting state
5715          */
5716         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5717
5718         /*
5719          * systems with ASPM and others may see the checksum fail on the first
5720          * attempt. Let's give it a few tries
5721          */
5722         for (i = 0;; i++) {
5723                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5724                         break;
5725                 if (i == 2) {
5726                         e_err("The NVM Checksum Is Not Valid\n");
5727                         err = -EIO;
5728                         goto err_eeprom;
5729                 }
5730         }
5731
5732         e1000_eeprom_checks(adapter);
5733
5734         /* copy the MAC address */
5735         if (e1000e_read_mac_addr(&adapter->hw))
5736                 e_err("NVM Read Error while reading MAC address\n");
5737
5738         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5739         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5740
5741         if (!is_valid_ether_addr(netdev->perm_addr)) {
5742                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5743                 err = -EIO;
5744                 goto err_eeprom;
5745         }
5746
5747         init_timer(&adapter->watchdog_timer);
5748         adapter->watchdog_timer.function = &e1000_watchdog;
5749         adapter->watchdog_timer.data = (unsigned long) adapter;
5750
5751         init_timer(&adapter->phy_info_timer);
5752         adapter->phy_info_timer.function = &e1000_update_phy_info;
5753         adapter->phy_info_timer.data = (unsigned long) adapter;
5754
5755         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5756         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5757         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5758         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5759         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5760
5761         /* Initialize link parameters. User can change them with ethtool */
5762         adapter->hw.mac.autoneg = 1;
5763         adapter->fc_autoneg = 1;
5764         adapter->hw.fc.requested_mode = e1000_fc_default;
5765         adapter->hw.fc.current_mode = e1000_fc_default;
5766         adapter->hw.phy.autoneg_advertised = 0x2f;
5767
5768         /* ring size defaults */
5769         adapter->rx_ring->count = 256;
5770         adapter->tx_ring->count = 256;
5771
5772         /*
5773          * Initial Wake on LAN setting - If APM wake is enabled in
5774          * the EEPROM, enable the ACPI Magic Packet filter
5775          */
5776         if (adapter->flags & FLAG_APME_IN_WUC) {
5777                 /* APME bit in EEPROM is mapped to WUC.APME */
5778                 eeprom_data = er32(WUC);
5779                 eeprom_apme_mask = E1000_WUC_APME;
5780                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5781                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5782         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5783                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5784                     (adapter->hw.bus.func == 1))
5785                         e1000_read_nvm(&adapter->hw,
5786                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5787                 else
5788                         e1000_read_nvm(&adapter->hw,
5789                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5790         }
5791
5792         /* fetch WoL from EEPROM */
5793