e1000e: Fix irq_synchronize in MSI-X case
[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
785                 status = rx_desc->status;
786                 skb = buffer_info->skb;
787                 buffer_info->skb = NULL;
788
789                 prefetch(skb->data - NET_IP_ALIGN);
790
791                 i++;
792                 if (i == rx_ring->count)
793                         i = 0;
794                 next_rxd = E1000_RX_DESC(*rx_ring, i);
795                 prefetch(next_rxd);
796
797                 next_buffer = &rx_ring->buffer_info[i];
798
799                 cleaned = 1;
800                 cleaned_count++;
801                 dma_unmap_single(&pdev->dev,
802                                  buffer_info->dma,
803                                  adapter->rx_buffer_len,
804                                  DMA_FROM_DEVICE);
805                 buffer_info->dma = 0;
806
807                 length = le16_to_cpu(rx_desc->length);
808
809                 /*
810                  * !EOP means multiple descriptors were used to store a single
811                  * packet, if that's the case we need to toss it.  In fact, we
812                  * need to toss every packet with the EOP bit clear and the
813                  * next frame that _does_ have the EOP bit set, as it is by
814                  * definition only a frame fragment
815                  */
816                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
817                         adapter->flags2 |= FLAG2_IS_DISCARDING;
818
819                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
820                         /* All receives must fit into a single buffer */
821                         e_dbg("Receive packet consumed multiple buffers\n");
822                         /* recycle */
823                         buffer_info->skb = skb;
824                         if (status & E1000_RXD_STAT_EOP)
825                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
826                         goto next_desc;
827                 }
828
829                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
830                         /* recycle */
831                         buffer_info->skb = skb;
832                         goto next_desc;
833                 }
834
835                 /* adjust length to remove Ethernet CRC */
836                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
837                         length -= 4;
838
839                 total_rx_bytes += length;
840                 total_rx_packets++;
841
842                 /*
843                  * code added for copybreak, this should improve
844                  * performance for small packets with large amounts
845                  * of reassembly being done in the stack
846                  */
847                 if (length < copybreak) {
848                         struct sk_buff *new_skb =
849                             netdev_alloc_skb_ip_align(netdev, length);
850                         if (new_skb) {
851                                 skb_copy_to_linear_data_offset(new_skb,
852                                                                -NET_IP_ALIGN,
853                                                                (skb->data -
854                                                                 NET_IP_ALIGN),
855                                                                (length +
856                                                                 NET_IP_ALIGN));
857                                 /* save the skb in buffer_info as good */
858                                 buffer_info->skb = skb;
859                                 skb = new_skb;
860                         }
861                         /* else just continue with the old one */
862                 }
863                 /* end copybreak code */
864                 skb_put(skb, length);
865
866                 /* Receive Checksum Offload */
867                 e1000_rx_checksum(adapter,
868                                   (u32)(status) |
869                                   ((u32)(rx_desc->errors) << 24),
870                                   le16_to_cpu(rx_desc->csum), skb);
871
872                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
873
874 next_desc:
875                 rx_desc->status = 0;
876
877                 /* return some buffers to hardware, one at a time is too slow */
878                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
879                         adapter->alloc_rx_buf(adapter, cleaned_count);
880                         cleaned_count = 0;
881                 }
882
883                 /* use prefetched values */
884                 rx_desc = next_rxd;
885                 buffer_info = next_buffer;
886         }
887         rx_ring->next_to_clean = i;
888
889         cleaned_count = e1000_desc_unused(rx_ring);
890         if (cleaned_count)
891                 adapter->alloc_rx_buf(adapter, cleaned_count);
892
893         adapter->total_rx_bytes += total_rx_bytes;
894         adapter->total_rx_packets += total_rx_packets;
895         netdev->stats.rx_bytes += total_rx_bytes;
896         netdev->stats.rx_packets += total_rx_packets;
897         return cleaned;
898 }
899
900 static void e1000_put_txbuf(struct e1000_adapter *adapter,
901                              struct e1000_buffer *buffer_info)
902 {
903         if (buffer_info->dma) {
904                 if (buffer_info->mapped_as_page)
905                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
906                                        buffer_info->length, DMA_TO_DEVICE);
907                 else
908                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
909                                          buffer_info->length, DMA_TO_DEVICE);
910                 buffer_info->dma = 0;
911         }
912         if (buffer_info->skb) {
913                 dev_kfree_skb_any(buffer_info->skb);
914                 buffer_info->skb = NULL;
915         }
916         buffer_info->time_stamp = 0;
917 }
918
919 static void e1000_print_hw_hang(struct work_struct *work)
920 {
921         struct e1000_adapter *adapter = container_of(work,
922                                                      struct e1000_adapter,
923                                                      print_hang_task);
924         struct e1000_ring *tx_ring = adapter->tx_ring;
925         unsigned int i = tx_ring->next_to_clean;
926         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
927         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
928         struct e1000_hw *hw = &adapter->hw;
929         u16 phy_status, phy_1000t_status, phy_ext_status;
930         u16 pci_status;
931
932         e1e_rphy(hw, PHY_STATUS, &phy_status);
933         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
934         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
935
936         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
937
938         /* detected Hardware unit hang */
939         e_err("Detected Hardware Unit Hang:\n"
940               "  TDH                  <%x>\n"
941               "  TDT                  <%x>\n"
942               "  next_to_use          <%x>\n"
943               "  next_to_clean        <%x>\n"
944               "buffer_info[next_to_clean]:\n"
945               "  time_stamp           <%lx>\n"
946               "  next_to_watch        <%x>\n"
947               "  jiffies              <%lx>\n"
948               "  next_to_watch.status <%x>\n"
949               "MAC Status             <%x>\n"
950               "PHY Status             <%x>\n"
951               "PHY 1000BASE-T Status  <%x>\n"
952               "PHY Extended Status    <%x>\n"
953               "PCI Status             <%x>\n",
954               readl(adapter->hw.hw_addr + tx_ring->head),
955               readl(adapter->hw.hw_addr + tx_ring->tail),
956               tx_ring->next_to_use,
957               tx_ring->next_to_clean,
958               tx_ring->buffer_info[eop].time_stamp,
959               eop,
960               jiffies,
961               eop_desc->upper.fields.status,
962               er32(STATUS),
963               phy_status,
964               phy_1000t_status,
965               phy_ext_status,
966               pci_status);
967 }
968
969 /**
970  * e1000_clean_tx_irq - Reclaim resources after transmit completes
971  * @adapter: board private structure
972  *
973  * the return value indicates whether actual cleaning was done, there
974  * is no guarantee that everything was cleaned
975  **/
976 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
977 {
978         struct net_device *netdev = adapter->netdev;
979         struct e1000_hw *hw = &adapter->hw;
980         struct e1000_ring *tx_ring = adapter->tx_ring;
981         struct e1000_tx_desc *tx_desc, *eop_desc;
982         struct e1000_buffer *buffer_info;
983         unsigned int i, eop;
984         unsigned int count = 0;
985         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
986
987         i = tx_ring->next_to_clean;
988         eop = tx_ring->buffer_info[i].next_to_watch;
989         eop_desc = E1000_TX_DESC(*tx_ring, eop);
990
991         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
992                (count < tx_ring->count)) {
993                 bool cleaned = false;
994                 for (; !cleaned; count++) {
995                         tx_desc = E1000_TX_DESC(*tx_ring, i);
996                         buffer_info = &tx_ring->buffer_info[i];
997                         cleaned = (i == eop);
998
999                         if (cleaned) {
1000                                 total_tx_packets += buffer_info->segs;
1001                                 total_tx_bytes += buffer_info->bytecount;
1002                         }
1003
1004                         e1000_put_txbuf(adapter, buffer_info);
1005                         tx_desc->upper.data = 0;
1006
1007                         i++;
1008                         if (i == tx_ring->count)
1009                                 i = 0;
1010                 }
1011
1012                 if (i == tx_ring->next_to_use)
1013                         break;
1014                 eop = tx_ring->buffer_info[i].next_to_watch;
1015                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1016         }
1017
1018         tx_ring->next_to_clean = i;
1019
1020 #define TX_WAKE_THRESHOLD 32
1021         if (count && netif_carrier_ok(netdev) &&
1022             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1023                 /* Make sure that anybody stopping the queue after this
1024                  * sees the new next_to_clean.
1025                  */
1026                 smp_mb();
1027
1028                 if (netif_queue_stopped(netdev) &&
1029                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1030                         netif_wake_queue(netdev);
1031                         ++adapter->restart_queue;
1032                 }
1033         }
1034
1035         if (adapter->detect_tx_hung) {
1036                 /*
1037                  * Detect a transmit hang in hardware, this serializes the
1038                  * check with the clearing of time_stamp and movement of i
1039                  */
1040                 adapter->detect_tx_hung = 0;
1041                 if (tx_ring->buffer_info[i].time_stamp &&
1042                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1043                                + (adapter->tx_timeout_factor * HZ)) &&
1044                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1045                         schedule_work(&adapter->print_hang_task);
1046                         netif_stop_queue(netdev);
1047                 }
1048         }
1049         adapter->total_tx_bytes += total_tx_bytes;
1050         adapter->total_tx_packets += total_tx_packets;
1051         netdev->stats.tx_bytes += total_tx_bytes;
1052         netdev->stats.tx_packets += total_tx_packets;
1053         return (count < tx_ring->count);
1054 }
1055
1056 /**
1057  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1058  * @adapter: board private structure
1059  *
1060  * the return value indicates whether actual cleaning was done, there
1061  * is no guarantee that everything was cleaned
1062  **/
1063 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1064                                   int *work_done, int work_to_do)
1065 {
1066         struct e1000_hw *hw = &adapter->hw;
1067         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1068         struct net_device *netdev = adapter->netdev;
1069         struct pci_dev *pdev = adapter->pdev;
1070         struct e1000_ring *rx_ring = adapter->rx_ring;
1071         struct e1000_buffer *buffer_info, *next_buffer;
1072         struct e1000_ps_page *ps_page;
1073         struct sk_buff *skb;
1074         unsigned int i, j;
1075         u32 length, staterr;
1076         int cleaned_count = 0;
1077         bool cleaned = 0;
1078         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1079
1080         i = rx_ring->next_to_clean;
1081         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1082         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1083         buffer_info = &rx_ring->buffer_info[i];
1084
1085         while (staterr & E1000_RXD_STAT_DD) {
1086                 if (*work_done >= work_to_do)
1087                         break;
1088                 (*work_done)++;
1089                 skb = buffer_info->skb;
1090
1091                 /* in the packet split case this is header only */
1092                 prefetch(skb->data - NET_IP_ALIGN);
1093
1094                 i++;
1095                 if (i == rx_ring->count)
1096                         i = 0;
1097                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1098                 prefetch(next_rxd);
1099
1100                 next_buffer = &rx_ring->buffer_info[i];
1101
1102                 cleaned = 1;
1103                 cleaned_count++;
1104                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1105                                  adapter->rx_ps_bsize0,
1106                                  DMA_FROM_DEVICE);
1107                 buffer_info->dma = 0;
1108
1109                 /* see !EOP comment in other rx routine */
1110                 if (!(staterr & E1000_RXD_STAT_EOP))
1111                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1112
1113                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1114                         e_dbg("Packet Split buffers didn't pick up the full "
1115                               "packet\n");
1116                         dev_kfree_skb_irq(skb);
1117                         if (staterr & E1000_RXD_STAT_EOP)
1118                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1119                         goto next_desc;
1120                 }
1121
1122                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1123                         dev_kfree_skb_irq(skb);
1124                         goto next_desc;
1125                 }
1126
1127                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1128
1129                 if (!length) {
1130                         e_dbg("Last part of the packet spanning multiple "
1131                               "descriptors\n");
1132                         dev_kfree_skb_irq(skb);
1133                         goto next_desc;
1134                 }
1135
1136                 /* Good Receive */
1137                 skb_put(skb, length);
1138
1139                 {
1140                 /*
1141                  * this looks ugly, but it seems compiler issues make it
1142                  * more efficient than reusing j
1143                  */
1144                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1145
1146                 /*
1147                  * page alloc/put takes too long and effects small packet
1148                  * throughput, so unsplit small packets and save the alloc/put
1149                  * only valid in softirq (napi) context to call kmap_*
1150                  */
1151                 if (l1 && (l1 <= copybreak) &&
1152                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1153                         u8 *vaddr;
1154
1155                         ps_page = &buffer_info->ps_pages[0];
1156
1157                         /*
1158                          * there is no documentation about how to call
1159                          * kmap_atomic, so we can't hold the mapping
1160                          * very long
1161                          */
1162                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1163                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1164                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1165                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1166                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1167                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1168                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1169
1170                         /* remove the CRC */
1171                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1172                                 l1 -= 4;
1173
1174                         skb_put(skb, l1);
1175                         goto copydone;
1176                 } /* if */
1177                 }
1178
1179                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1180                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1181                         if (!length)
1182                                 break;
1183
1184                         ps_page = &buffer_info->ps_pages[j];
1185                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1186                                        DMA_FROM_DEVICE);
1187                         ps_page->dma = 0;
1188                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1189                         ps_page->page = NULL;
1190                         skb->len += length;
1191                         skb->data_len += length;
1192                         skb->truesize += length;
1193                 }
1194
1195                 /* strip the ethernet crc, problem is we're using pages now so
1196                  * this whole operation can get a little cpu intensive
1197                  */
1198                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1199                         pskb_trim(skb, skb->len - 4);
1200
1201 copydone:
1202                 total_rx_bytes += skb->len;
1203                 total_rx_packets++;
1204
1205                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1206                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1207
1208                 if (rx_desc->wb.upper.header_status &
1209                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1210                         adapter->rx_hdr_split++;
1211
1212                 e1000_receive_skb(adapter, netdev, skb,
1213                                   staterr, rx_desc->wb.middle.vlan);
1214
1215 next_desc:
1216                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1217                 buffer_info->skb = NULL;
1218
1219                 /* return some buffers to hardware, one at a time is too slow */
1220                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1221                         adapter->alloc_rx_buf(adapter, cleaned_count);
1222                         cleaned_count = 0;
1223                 }
1224
1225                 /* use prefetched values */
1226                 rx_desc = next_rxd;
1227                 buffer_info = next_buffer;
1228
1229                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1230         }
1231         rx_ring->next_to_clean = i;
1232
1233         cleaned_count = e1000_desc_unused(rx_ring);
1234         if (cleaned_count)
1235                 adapter->alloc_rx_buf(adapter, cleaned_count);
1236
1237         adapter->total_rx_bytes += total_rx_bytes;
1238         adapter->total_rx_packets += total_rx_packets;
1239         netdev->stats.rx_bytes += total_rx_bytes;
1240         netdev->stats.rx_packets += total_rx_packets;
1241         return cleaned;
1242 }
1243
1244 /**
1245  * e1000_consume_page - helper function
1246  **/
1247 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1248                                u16 length)
1249 {
1250         bi->page = NULL;
1251         skb->len += length;
1252         skb->data_len += length;
1253         skb->truesize += length;
1254 }
1255
1256 /**
1257  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1258  * @adapter: board private structure
1259  *
1260  * the return value indicates whether actual cleaning was done, there
1261  * is no guarantee that everything was cleaned
1262  **/
1263
1264 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1265                                      int *work_done, int work_to_do)
1266 {
1267         struct net_device *netdev = adapter->netdev;
1268         struct pci_dev *pdev = adapter->pdev;
1269         struct e1000_ring *rx_ring = adapter->rx_ring;
1270         struct e1000_rx_desc *rx_desc, *next_rxd;
1271         struct e1000_buffer *buffer_info, *next_buffer;
1272         u32 length;
1273         unsigned int i;
1274         int cleaned_count = 0;
1275         bool cleaned = false;
1276         unsigned int total_rx_bytes=0, total_rx_packets=0;
1277
1278         i = rx_ring->next_to_clean;
1279         rx_desc = E1000_RX_DESC(*rx_ring, i);
1280         buffer_info = &rx_ring->buffer_info[i];
1281
1282         while (rx_desc->status & E1000_RXD_STAT_DD) {
1283                 struct sk_buff *skb;
1284                 u8 status;
1285
1286                 if (*work_done >= work_to_do)
1287                         break;
1288                 (*work_done)++;
1289
1290                 status = rx_desc->status;
1291                 skb = buffer_info->skb;
1292                 buffer_info->skb = NULL;
1293
1294                 ++i;
1295                 if (i == rx_ring->count)
1296                         i = 0;
1297                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1298                 prefetch(next_rxd);
1299
1300                 next_buffer = &rx_ring->buffer_info[i];
1301
1302                 cleaned = true;
1303                 cleaned_count++;
1304                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1305                                DMA_FROM_DEVICE);
1306                 buffer_info->dma = 0;
1307
1308                 length = le16_to_cpu(rx_desc->length);
1309
1310                 /* errors is only valid for DD + EOP descriptors */
1311                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1312                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1313                                 /* recycle both page and skb */
1314                                 buffer_info->skb = skb;
1315                                 /* an error means any chain goes out the window
1316                                  * too */
1317                                 if (rx_ring->rx_skb_top)
1318                                         dev_kfree_skb(rx_ring->rx_skb_top);
1319                                 rx_ring->rx_skb_top = NULL;
1320                                 goto next_desc;
1321                 }
1322
1323 #define rxtop rx_ring->rx_skb_top
1324                 if (!(status & E1000_RXD_STAT_EOP)) {
1325                         /* this descriptor is only the beginning (or middle) */
1326                         if (!rxtop) {
1327                                 /* this is the beginning of a chain */
1328                                 rxtop = skb;
1329                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1330                                                    0, length);
1331                         } else {
1332                                 /* this is the middle of a chain */
1333                                 skb_fill_page_desc(rxtop,
1334                                     skb_shinfo(rxtop)->nr_frags,
1335                                     buffer_info->page, 0, length);
1336                                 /* re-use the skb, only consumed the page */
1337                                 buffer_info->skb = skb;
1338                         }
1339                         e1000_consume_page(buffer_info, rxtop, length);
1340                         goto next_desc;
1341                 } else {
1342                         if (rxtop) {
1343                                 /* end of the chain */
1344                                 skb_fill_page_desc(rxtop,
1345                                     skb_shinfo(rxtop)->nr_frags,
1346                                     buffer_info->page, 0, length);
1347                                 /* re-use the current skb, we only consumed the
1348                                  * page */
1349                                 buffer_info->skb = skb;
1350                                 skb = rxtop;
1351                                 rxtop = NULL;
1352                                 e1000_consume_page(buffer_info, skb, length);
1353                         } else {
1354                                 /* no chain, got EOP, this buf is the packet
1355                                  * copybreak to save the put_page/alloc_page */
1356                                 if (length <= copybreak &&
1357                                     skb_tailroom(skb) >= length) {
1358                                         u8 *vaddr;
1359                                         vaddr = kmap_atomic(buffer_info->page,
1360                                                            KM_SKB_DATA_SOFTIRQ);
1361                                         memcpy(skb_tail_pointer(skb), vaddr,
1362                                                length);
1363                                         kunmap_atomic(vaddr,
1364                                                       KM_SKB_DATA_SOFTIRQ);
1365                                         /* re-use the page, so don't erase
1366                                          * buffer_info->page */
1367                                         skb_put(skb, length);
1368                                 } else {
1369                                         skb_fill_page_desc(skb, 0,
1370                                                            buffer_info->page, 0,
1371                                                            length);
1372                                         e1000_consume_page(buffer_info, skb,
1373                                                            length);
1374                                 }
1375                         }
1376                 }
1377
1378                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1379                 e1000_rx_checksum(adapter,
1380                                   (u32)(status) |
1381                                   ((u32)(rx_desc->errors) << 24),
1382                                   le16_to_cpu(rx_desc->csum), skb);
1383
1384                 /* probably a little skewed due to removing CRC */
1385                 total_rx_bytes += skb->len;
1386                 total_rx_packets++;
1387
1388                 /* eth type trans needs skb->data to point to something */
1389                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1390                         e_err("pskb_may_pull failed.\n");
1391                         dev_kfree_skb(skb);
1392                         goto next_desc;
1393                 }
1394
1395                 e1000_receive_skb(adapter, netdev, skb, status,
1396                                   rx_desc->special);
1397
1398 next_desc:
1399                 rx_desc->status = 0;
1400
1401                 /* return some buffers to hardware, one at a time is too slow */
1402                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1403                         adapter->alloc_rx_buf(adapter, cleaned_count);
1404                         cleaned_count = 0;
1405                 }
1406
1407                 /* use prefetched values */
1408                 rx_desc = next_rxd;
1409                 buffer_info = next_buffer;
1410         }
1411         rx_ring->next_to_clean = i;
1412
1413         cleaned_count = e1000_desc_unused(rx_ring);
1414         if (cleaned_count)
1415                 adapter->alloc_rx_buf(adapter, cleaned_count);
1416
1417         adapter->total_rx_bytes += total_rx_bytes;
1418         adapter->total_rx_packets += total_rx_packets;
1419         netdev->stats.rx_bytes += total_rx_bytes;
1420         netdev->stats.rx_packets += total_rx_packets;
1421         return cleaned;
1422 }
1423
1424 /**
1425  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1426  * @adapter: board private structure
1427  **/
1428 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1429 {
1430         struct e1000_ring *rx_ring = adapter->rx_ring;
1431         struct e1000_buffer *buffer_info;
1432         struct e1000_ps_page *ps_page;
1433         struct pci_dev *pdev = adapter->pdev;
1434         unsigned int i, j;
1435
1436         /* Free all the Rx ring sk_buffs */
1437         for (i = 0; i < rx_ring->count; i++) {
1438                 buffer_info = &rx_ring->buffer_info[i];
1439                 if (buffer_info->dma) {
1440                         if (adapter->clean_rx == e1000_clean_rx_irq)
1441                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1442                                                  adapter->rx_buffer_len,
1443                                                  DMA_FROM_DEVICE);
1444                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1445                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1446                                                PAGE_SIZE,
1447                                                DMA_FROM_DEVICE);
1448                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1449                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1450                                                  adapter->rx_ps_bsize0,
1451                                                  DMA_FROM_DEVICE);
1452                         buffer_info->dma = 0;
1453                 }
1454
1455                 if (buffer_info->page) {
1456                         put_page(buffer_info->page);
1457                         buffer_info->page = NULL;
1458                 }
1459
1460                 if (buffer_info->skb) {
1461                         dev_kfree_skb(buffer_info->skb);
1462                         buffer_info->skb = NULL;
1463                 }
1464
1465                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1466                         ps_page = &buffer_info->ps_pages[j];
1467                         if (!ps_page->page)
1468                                 break;
1469                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1470                                        DMA_FROM_DEVICE);
1471                         ps_page->dma = 0;
1472                         put_page(ps_page->page);
1473                         ps_page->page = NULL;
1474                 }
1475         }
1476
1477         /* there also may be some cached data from a chained receive */
1478         if (rx_ring->rx_skb_top) {
1479                 dev_kfree_skb(rx_ring->rx_skb_top);
1480                 rx_ring->rx_skb_top = NULL;
1481         }
1482
1483         /* Zero out the descriptor ring */
1484         memset(rx_ring->desc, 0, rx_ring->size);
1485
1486         rx_ring->next_to_clean = 0;
1487         rx_ring->next_to_use = 0;
1488         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1489
1490         writel(0, adapter->hw.hw_addr + rx_ring->head);
1491         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1492 }
1493
1494 static void e1000e_downshift_workaround(struct work_struct *work)
1495 {
1496         struct e1000_adapter *adapter = container_of(work,
1497                                         struct e1000_adapter, downshift_task);
1498
1499         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1500 }
1501
1502 /**
1503  * e1000_intr_msi - Interrupt Handler
1504  * @irq: interrupt number
1505  * @data: pointer to a network interface device structure
1506  **/
1507 static irqreturn_t e1000_intr_msi(int irq, void *data)
1508 {
1509         struct net_device *netdev = data;
1510         struct e1000_adapter *adapter = netdev_priv(netdev);
1511         struct e1000_hw *hw = &adapter->hw;
1512         u32 icr = er32(ICR);
1513
1514         /*
1515          * read ICR disables interrupts using IAM
1516          */
1517
1518         if (icr & E1000_ICR_LSC) {
1519                 hw->mac.get_link_status = 1;
1520                 /*
1521                  * ICH8 workaround-- Call gig speed drop workaround on cable
1522                  * disconnect (LSC) before accessing any PHY registers
1523                  */
1524                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1525                     (!(er32(STATUS) & E1000_STATUS_LU)))
1526                         schedule_work(&adapter->downshift_task);
1527
1528                 /*
1529                  * 80003ES2LAN workaround-- For packet buffer work-around on
1530                  * link down event; disable receives here in the ISR and reset
1531                  * adapter in watchdog
1532                  */
1533                 if (netif_carrier_ok(netdev) &&
1534                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1535                         /* disable receives */
1536                         u32 rctl = er32(RCTL);
1537                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1538                         adapter->flags |= FLAG_RX_RESTART_NOW;
1539                 }
1540                 /* guard against interrupt when we're going down */
1541                 if (!test_bit(__E1000_DOWN, &adapter->state))
1542                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1543         }
1544
1545         if (napi_schedule_prep(&adapter->napi)) {
1546                 adapter->total_tx_bytes = 0;
1547                 adapter->total_tx_packets = 0;
1548                 adapter->total_rx_bytes = 0;
1549                 adapter->total_rx_packets = 0;
1550                 __napi_schedule(&adapter->napi);
1551         }
1552
1553         return IRQ_HANDLED;
1554 }
1555
1556 /**
1557  * e1000_intr - Interrupt Handler
1558  * @irq: interrupt number
1559  * @data: pointer to a network interface device structure
1560  **/
1561 static irqreturn_t e1000_intr(int irq, void *data)
1562 {
1563         struct net_device *netdev = data;
1564         struct e1000_adapter *adapter = netdev_priv(netdev);
1565         struct e1000_hw *hw = &adapter->hw;
1566         u32 rctl, icr = er32(ICR);
1567
1568         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1569                 return IRQ_NONE;  /* Not our interrupt */
1570
1571         /*
1572          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1573          * not set, then the adapter didn't send an interrupt
1574          */
1575         if (!(icr & E1000_ICR_INT_ASSERTED))
1576                 return IRQ_NONE;
1577
1578         /*
1579          * Interrupt Auto-Mask...upon reading ICR,
1580          * interrupts are masked.  No need for the
1581          * IMC write
1582          */
1583
1584         if (icr & E1000_ICR_LSC) {
1585                 hw->mac.get_link_status = 1;
1586                 /*
1587                  * ICH8 workaround-- Call gig speed drop workaround on cable
1588                  * disconnect (LSC) before accessing any PHY registers
1589                  */
1590                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1591                     (!(er32(STATUS) & E1000_STATUS_LU)))
1592                         schedule_work(&adapter->downshift_task);
1593
1594                 /*
1595                  * 80003ES2LAN workaround--
1596                  * For packet buffer work-around on link down event;
1597                  * disable receives here in the ISR and
1598                  * reset adapter in watchdog
1599                  */
1600                 if (netif_carrier_ok(netdev) &&
1601                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1602                         /* disable receives */
1603                         rctl = er32(RCTL);
1604                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1605                         adapter->flags |= FLAG_RX_RESTART_NOW;
1606                 }
1607                 /* guard against interrupt when we're going down */
1608                 if (!test_bit(__E1000_DOWN, &adapter->state))
1609                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1610         }
1611
1612         if (napi_schedule_prep(&adapter->napi)) {
1613                 adapter->total_tx_bytes = 0;
1614                 adapter->total_tx_packets = 0;
1615                 adapter->total_rx_bytes = 0;
1616                 adapter->total_rx_packets = 0;
1617                 __napi_schedule(&adapter->napi);
1618         }
1619
1620         return IRQ_HANDLED;
1621 }
1622
1623 static irqreturn_t e1000_msix_other(int irq, void *data)
1624 {
1625         struct net_device *netdev = data;
1626         struct e1000_adapter *adapter = netdev_priv(netdev);
1627         struct e1000_hw *hw = &adapter->hw;
1628         u32 icr = er32(ICR);
1629
1630         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1631                 if (!test_bit(__E1000_DOWN, &adapter->state))
1632                         ew32(IMS, E1000_IMS_OTHER);
1633                 return IRQ_NONE;
1634         }
1635
1636         if (icr & adapter->eiac_mask)
1637                 ew32(ICS, (icr & adapter->eiac_mask));
1638
1639         if (icr & E1000_ICR_OTHER) {
1640                 if (!(icr & E1000_ICR_LSC))
1641                         goto no_link_interrupt;
1642                 hw->mac.get_link_status = 1;
1643                 /* guard against interrupt when we're going down */
1644                 if (!test_bit(__E1000_DOWN, &adapter->state))
1645                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1646         }
1647
1648 no_link_interrupt:
1649         if (!test_bit(__E1000_DOWN, &adapter->state))
1650                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1651
1652         return IRQ_HANDLED;
1653 }
1654
1655
1656 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1657 {
1658         struct net_device *netdev = data;
1659         struct e1000_adapter *adapter = netdev_priv(netdev);
1660         struct e1000_hw *hw = &adapter->hw;
1661         struct e1000_ring *tx_ring = adapter->tx_ring;
1662
1663
1664         adapter->total_tx_bytes = 0;
1665         adapter->total_tx_packets = 0;
1666
1667         if (!e1000_clean_tx_irq(adapter))
1668                 /* Ring was not completely cleaned, so fire another interrupt */
1669                 ew32(ICS, tx_ring->ims_val);
1670
1671         return IRQ_HANDLED;
1672 }
1673
1674 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1675 {
1676         struct net_device *netdev = data;
1677         struct e1000_adapter *adapter = netdev_priv(netdev);
1678
1679         /* Write the ITR value calculated at the end of the
1680          * previous interrupt.
1681          */
1682         if (adapter->rx_ring->set_itr) {
1683                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1684                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1685                 adapter->rx_ring->set_itr = 0;
1686         }
1687
1688         if (napi_schedule_prep(&adapter->napi)) {
1689                 adapter->total_rx_bytes = 0;
1690                 adapter->total_rx_packets = 0;
1691                 __napi_schedule(&adapter->napi);
1692         }
1693         return IRQ_HANDLED;
1694 }
1695
1696 /**
1697  * e1000_configure_msix - Configure MSI-X hardware
1698  *
1699  * e1000_configure_msix sets up the hardware to properly
1700  * generate MSI-X interrupts.
1701  **/
1702 static void e1000_configure_msix(struct e1000_adapter *adapter)
1703 {
1704         struct e1000_hw *hw = &adapter->hw;
1705         struct e1000_ring *rx_ring = adapter->rx_ring;
1706         struct e1000_ring *tx_ring = adapter->tx_ring;
1707         int vector = 0;
1708         u32 ctrl_ext, ivar = 0;
1709
1710         adapter->eiac_mask = 0;
1711
1712         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1713         if (hw->mac.type == e1000_82574) {
1714                 u32 rfctl = er32(RFCTL);
1715                 rfctl |= E1000_RFCTL_ACK_DIS;
1716                 ew32(RFCTL, rfctl);
1717         }
1718
1719 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1720         /* Configure Rx vector */
1721         rx_ring->ims_val = E1000_IMS_RXQ0;
1722         adapter->eiac_mask |= rx_ring->ims_val;
1723         if (rx_ring->itr_val)
1724                 writel(1000000000 / (rx_ring->itr_val * 256),
1725                        hw->hw_addr + rx_ring->itr_register);
1726         else
1727                 writel(1, hw->hw_addr + rx_ring->itr_register);
1728         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1729
1730         /* Configure Tx vector */
1731         tx_ring->ims_val = E1000_IMS_TXQ0;
1732         vector++;
1733         if (tx_ring->itr_val)
1734                 writel(1000000000 / (tx_ring->itr_val * 256),
1735                        hw->hw_addr + tx_ring->itr_register);
1736         else
1737                 writel(1, hw->hw_addr + tx_ring->itr_register);
1738         adapter->eiac_mask |= tx_ring->ims_val;
1739         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1740
1741         /* set vector for Other Causes, e.g. link changes */
1742         vector++;
1743         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1744         if (rx_ring->itr_val)
1745                 writel(1000000000 / (rx_ring->itr_val * 256),
1746                        hw->hw_addr + E1000_EITR_82574(vector));
1747         else
1748                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1749
1750         /* Cause Tx interrupts on every write back */
1751         ivar |= (1 << 31);
1752
1753         ew32(IVAR, ivar);
1754
1755         /* enable MSI-X PBA support */
1756         ctrl_ext = er32(CTRL_EXT);
1757         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1758
1759         /* Auto-Mask Other interrupts upon ICR read */
1760 #define E1000_EIAC_MASK_82574   0x01F00000
1761         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1762         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1763         ew32(CTRL_EXT, ctrl_ext);
1764         e1e_flush();
1765 }
1766
1767 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1768 {
1769         if (adapter->msix_entries) {
1770                 pci_disable_msix(adapter->pdev);
1771                 kfree(adapter->msix_entries);
1772                 adapter->msix_entries = NULL;
1773         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1774                 pci_disable_msi(adapter->pdev);
1775                 adapter->flags &= ~FLAG_MSI_ENABLED;
1776         }
1777 }
1778
1779 /**
1780  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1781  *
1782  * Attempt to configure interrupts using the best available
1783  * capabilities of the hardware and kernel.
1784  **/
1785 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1786 {
1787         int err;
1788         int i;
1789
1790         switch (adapter->int_mode) {
1791         case E1000E_INT_MODE_MSIX:
1792                 if (adapter->flags & FLAG_HAS_MSIX) {
1793                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1794                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1795                                                       sizeof(struct msix_entry),
1796                                                       GFP_KERNEL);
1797                         if (adapter->msix_entries) {
1798                                 for (i = 0; i < adapter->num_vectors; i++)
1799                                         adapter->msix_entries[i].entry = i;
1800
1801                                 err = pci_enable_msix(adapter->pdev,
1802                                                       adapter->msix_entries,
1803                                                       adapter->num_vectors);
1804                                 if (err == 0) {
1805                                         return;
1806                                 }
1807                         }
1808                         /* MSI-X failed, so fall through and try MSI */
1809                         e_err("Failed to initialize MSI-X interrupts.  "
1810                               "Falling back to MSI interrupts.\n");
1811                         e1000e_reset_interrupt_capability(adapter);
1812                 }
1813                 adapter->int_mode = E1000E_INT_MODE_MSI;
1814                 /* Fall through */
1815         case E1000E_INT_MODE_MSI:
1816                 if (!pci_enable_msi(adapter->pdev)) {
1817                         adapter->flags |= FLAG_MSI_ENABLED;
1818                 } else {
1819                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1820                         e_err("Failed to initialize MSI interrupts.  Falling "
1821                               "back to legacy interrupts.\n");
1822                 }
1823                 /* Fall through */
1824         case E1000E_INT_MODE_LEGACY:
1825                 /* Don't do anything; this is the system default */
1826                 break;
1827         }
1828
1829         /* store the number of vectors being used */
1830         adapter->num_vectors = 1;
1831 }
1832
1833 /**
1834  * e1000_request_msix - Initialize MSI-X interrupts
1835  *
1836  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1837  * kernel.
1838  **/
1839 static int e1000_request_msix(struct e1000_adapter *adapter)
1840 {
1841         struct net_device *netdev = adapter->netdev;
1842         int err = 0, vector = 0;
1843
1844         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1845                 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1846         else
1847                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1848         err = request_irq(adapter->msix_entries[vector].vector,
1849                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1850                           netdev);
1851         if (err)
1852                 goto out;
1853         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1854         adapter->rx_ring->itr_val = adapter->itr;
1855         vector++;
1856
1857         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1858                 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1859         else
1860                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1861         err = request_irq(adapter->msix_entries[vector].vector,
1862                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1863                           netdev);
1864         if (err)
1865                 goto out;
1866         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1867         adapter->tx_ring->itr_val = adapter->itr;
1868         vector++;
1869
1870         err = request_irq(adapter->msix_entries[vector].vector,
1871                           e1000_msix_other, 0, netdev->name, netdev);
1872         if (err)
1873                 goto out;
1874
1875         e1000_configure_msix(adapter);
1876         return 0;
1877 out:
1878         return err;
1879 }
1880
1881 /**
1882  * e1000_request_irq - initialize interrupts
1883  *
1884  * Attempts to configure interrupts using the best available
1885  * capabilities of the hardware and kernel.
1886  **/
1887 static int e1000_request_irq(struct e1000_adapter *adapter)
1888 {
1889         struct net_device *netdev = adapter->netdev;
1890         int err;
1891
1892         if (adapter->msix_entries) {
1893                 err = e1000_request_msix(adapter);
1894                 if (!err)
1895                         return err;
1896                 /* fall back to MSI */
1897                 e1000e_reset_interrupt_capability(adapter);
1898                 adapter->int_mode = E1000E_INT_MODE_MSI;
1899                 e1000e_set_interrupt_capability(adapter);
1900         }
1901         if (adapter->flags & FLAG_MSI_ENABLED) {
1902                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1903                                   netdev->name, netdev);
1904                 if (!err)
1905                         return err;
1906
1907                 /* fall back to legacy interrupt */
1908                 e1000e_reset_interrupt_capability(adapter);
1909                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1910         }
1911
1912         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1913                           netdev->name, netdev);
1914         if (err)
1915                 e_err("Unable to allocate interrupt, Error: %d\n", err);
1916
1917         return err;
1918 }
1919
1920 static void e1000_free_irq(struct e1000_adapter *adapter)
1921 {
1922         struct net_device *netdev = adapter->netdev;
1923
1924         if (adapter->msix_entries) {
1925                 int vector = 0;
1926
1927                 free_irq(adapter->msix_entries[vector].vector, netdev);
1928                 vector++;
1929
1930                 free_irq(adapter->msix_entries[vector].vector, netdev);
1931                 vector++;
1932
1933                 /* Other Causes interrupt vector */
1934                 free_irq(adapter->msix_entries[vector].vector, netdev);
1935                 return;
1936         }
1937
1938         free_irq(adapter->pdev->irq, netdev);
1939 }
1940
1941 /**
1942  * e1000_irq_disable - Mask off interrupt generation on the NIC
1943  **/
1944 static void e1000_irq_disable(struct e1000_adapter *adapter)
1945 {
1946         struct e1000_hw *hw = &adapter->hw;
1947
1948         ew32(IMC, ~0);
1949         if (adapter->msix_entries)
1950                 ew32(EIAC_82574, 0);
1951         e1e_flush();
1952
1953         if (adapter->msix_entries) {
1954                 int i;
1955                 for (i = 0; i < adapter->num_vectors; i++)
1956                         synchronize_irq(adapter->msix_entries[i].vector);
1957         } else {
1958                 synchronize_irq(adapter->pdev->irq);
1959         }
1960 }
1961
1962 /**
1963  * e1000_irq_enable - Enable default interrupt generation settings
1964  **/
1965 static void e1000_irq_enable(struct e1000_adapter *adapter)
1966 {
1967         struct e1000_hw *hw = &adapter->hw;
1968
1969         if (adapter->msix_entries) {
1970                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1971                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1972         } else {
1973                 ew32(IMS, IMS_ENABLE_MASK);
1974         }
1975         e1e_flush();
1976 }
1977
1978 /**
1979  * e1000_get_hw_control - get control of the h/w from f/w
1980  * @adapter: address of board private structure
1981  *
1982  * e1000_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1983  * For ASF and Pass Through versions of f/w this means that
1984  * the driver is loaded. For AMT version (only with 82573)
1985  * of the f/w this means that the network i/f is open.
1986  **/
1987 static void e1000_get_hw_control(struct e1000_adapter *adapter)
1988 {
1989         struct e1000_hw *hw = &adapter->hw;
1990         u32 ctrl_ext;
1991         u32 swsm;
1992
1993         /* Let firmware know the driver has taken over */
1994         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
1995                 swsm = er32(SWSM);
1996                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
1997         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
1998                 ctrl_ext = er32(CTRL_EXT);
1999                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2000         }
2001 }
2002
2003 /**
2004  * e1000_release_hw_control - release control of the h/w to f/w
2005  * @adapter: address of board private structure
2006  *
2007  * e1000_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2008  * For ASF and Pass Through versions of f/w this means that the
2009  * driver is no longer loaded. For AMT version (only with 82573) i
2010  * of the f/w this means that the network i/f is closed.
2011  *
2012  **/
2013 static void e1000_release_hw_control(struct e1000_adapter *adapter)
2014 {
2015         struct e1000_hw *hw = &adapter->hw;
2016         u32 ctrl_ext;
2017         u32 swsm;
2018
2019         /* Let firmware taken over control of h/w */
2020         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2021                 swsm = er32(SWSM);
2022                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2023         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2024                 ctrl_ext = er32(CTRL_EXT);
2025                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2026         }
2027 }
2028
2029 /**
2030  * @e1000_alloc_ring - allocate memory for a ring structure
2031  **/
2032 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2033                                 struct e1000_ring *ring)
2034 {
2035         struct pci_dev *pdev = adapter->pdev;
2036
2037         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2038                                         GFP_KERNEL);
2039         if (!ring->desc)
2040                 return -ENOMEM;
2041
2042         return 0;
2043 }
2044
2045 /**
2046  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2047  * @adapter: board private structure
2048  *
2049  * Return 0 on success, negative on failure
2050  **/
2051 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2052 {
2053         struct e1000_ring *tx_ring = adapter->tx_ring;
2054         int err = -ENOMEM, size;
2055
2056         size = sizeof(struct e1000_buffer) * tx_ring->count;
2057         tx_ring->buffer_info = vmalloc(size);
2058         if (!tx_ring->buffer_info)
2059                 goto err;
2060         memset(tx_ring->buffer_info, 0, size);
2061
2062         /* round up to nearest 4K */
2063         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2064         tx_ring->size = ALIGN(tx_ring->size, 4096);
2065
2066         err = e1000_alloc_ring_dma(adapter, tx_ring);
2067         if (err)
2068                 goto err;
2069
2070         tx_ring->next_to_use = 0;
2071         tx_ring->next_to_clean = 0;
2072
2073         return 0;
2074 err:
2075         vfree(tx_ring->buffer_info);
2076         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2077         return err;
2078 }
2079
2080 /**
2081  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2082  * @adapter: board private structure
2083  *
2084  * Returns 0 on success, negative on failure
2085  **/
2086 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2087 {
2088         struct e1000_ring *rx_ring = adapter->rx_ring;
2089         struct e1000_buffer *buffer_info;
2090         int i, size, desc_len, err = -ENOMEM;
2091
2092         size = sizeof(struct e1000_buffer) * rx_ring->count;
2093         rx_ring->buffer_info = vmalloc(size);
2094         if (!rx_ring->buffer_info)
2095                 goto err;
2096         memset(rx_ring->buffer_info, 0, size);
2097
2098         for (i = 0; i < rx_ring->count; i++) {
2099                 buffer_info = &rx_ring->buffer_info[i];
2100                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2101                                                 sizeof(struct e1000_ps_page),
2102                                                 GFP_KERNEL);
2103                 if (!buffer_info->ps_pages)
2104                         goto err_pages;
2105         }
2106
2107         desc_len = sizeof(union e1000_rx_desc_packet_split);
2108
2109         /* Round up to nearest 4K */
2110         rx_ring->size = rx_ring->count * desc_len;
2111         rx_ring->size = ALIGN(rx_ring->size, 4096);
2112
2113         err = e1000_alloc_ring_dma(adapter, rx_ring);
2114         if (err)
2115                 goto err_pages;
2116
2117         rx_ring->next_to_clean = 0;
2118         rx_ring->next_to_use = 0;
2119         rx_ring->rx_skb_top = NULL;
2120
2121         return 0;
2122
2123 err_pages:
2124         for (i = 0; i < rx_ring->count; i++) {
2125                 buffer_info = &rx_ring->buffer_info[i];
2126                 kfree(buffer_info->ps_pages);
2127         }
2128 err:
2129         vfree(rx_ring->buffer_info);
2130         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2131         return err;
2132 }
2133
2134 /**
2135  * e1000_clean_tx_ring - Free Tx Buffers
2136  * @adapter: board private structure
2137  **/
2138 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2139 {
2140         struct e1000_ring *tx_ring = adapter->tx_ring;
2141         struct e1000_buffer *buffer_info;
2142         unsigned long size;
2143         unsigned int i;
2144
2145         for (i = 0; i < tx_ring->count; i++) {
2146                 buffer_info = &tx_ring->buffer_info[i];
2147                 e1000_put_txbuf(adapter, buffer_info);
2148         }
2149
2150         size = sizeof(struct e1000_buffer) * tx_ring->count;
2151         memset(tx_ring->buffer_info, 0, size);
2152
2153         memset(tx_ring->desc, 0, tx_ring->size);
2154
2155         tx_ring->next_to_use = 0;
2156         tx_ring->next_to_clean = 0;
2157
2158         writel(0, adapter->hw.hw_addr + tx_ring->head);
2159         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2160 }
2161
2162 /**
2163  * e1000e_free_tx_resources - Free Tx Resources per Queue
2164  * @adapter: board private structure
2165  *
2166  * Free all transmit software resources
2167  **/
2168 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2169 {
2170         struct pci_dev *pdev = adapter->pdev;
2171         struct e1000_ring *tx_ring = adapter->tx_ring;
2172
2173         e1000_clean_tx_ring(adapter);
2174
2175         vfree(tx_ring->buffer_info);
2176         tx_ring->buffer_info = NULL;
2177
2178         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2179                           tx_ring->dma);
2180         tx_ring->desc = NULL;
2181 }
2182
2183 /**
2184  * e1000e_free_rx_resources - Free Rx Resources
2185  * @adapter: board private structure
2186  *
2187  * Free all receive software resources
2188  **/
2189
2190 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2191 {
2192         struct pci_dev *pdev = adapter->pdev;
2193         struct e1000_ring *rx_ring = adapter->rx_ring;
2194         int i;
2195
2196         e1000_clean_rx_ring(adapter);
2197
2198         for (i = 0; i < rx_ring->count; i++) {
2199                 kfree(rx_ring->buffer_info[i].ps_pages);
2200         }
2201
2202         vfree(rx_ring->buffer_info);
2203         rx_ring->buffer_info = NULL;
2204
2205         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2206                           rx_ring->dma);
2207         rx_ring->desc = NULL;
2208 }
2209
2210 /**
2211  * e1000_update_itr - update the dynamic ITR value based on statistics
2212  * @adapter: pointer to adapter
2213  * @itr_setting: current adapter->itr
2214  * @packets: the number of packets during this measurement interval
2215  * @bytes: the number of bytes during this measurement interval
2216  *
2217  *      Stores a new ITR value based on packets and byte
2218  *      counts during the last interrupt.  The advantage of per interrupt
2219  *      computation is faster updates and more accurate ITR for the current
2220  *      traffic pattern.  Constants in this function were computed
2221  *      based on theoretical maximum wire speed and thresholds were set based
2222  *      on testing data as well as attempting to minimize response time
2223  *      while increasing bulk throughput.  This functionality is controlled
2224  *      by the InterruptThrottleRate module parameter.
2225  **/
2226 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2227                                      u16 itr_setting, int packets,
2228                                      int bytes)
2229 {
2230         unsigned int retval = itr_setting;
2231
2232         if (packets == 0)
2233                 goto update_itr_done;
2234
2235         switch (itr_setting) {
2236         case lowest_latency:
2237                 /* handle TSO and jumbo frames */
2238                 if (bytes/packets > 8000)
2239                         retval = bulk_latency;
2240                 else if ((packets < 5) && (bytes > 512)) {
2241                         retval = low_latency;
2242                 }
2243                 break;
2244         case low_latency:  /* 50 usec aka 20000 ints/s */
2245                 if (bytes > 10000) {
2246                         /* this if handles the TSO accounting */
2247                         if (bytes/packets > 8000) {
2248                                 retval = bulk_latency;
2249                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2250                                 retval = bulk_latency;
2251                         } else if ((packets > 35)) {
2252                                 retval = lowest_latency;
2253                         }
2254                 } else if (bytes/packets > 2000) {
2255                         retval = bulk_latency;
2256                 } else if (packets <= 2 && bytes < 512) {
2257                         retval = lowest_latency;
2258                 }
2259                 break;
2260         case bulk_latency: /* 250 usec aka 4000 ints/s */
2261                 if (bytes > 25000) {
2262                         if (packets > 35) {
2263                                 retval = low_latency;
2264                         }
2265                 } else if (bytes < 6000) {
2266                         retval = low_latency;
2267                 }
2268                 break;
2269         }
2270
2271 update_itr_done:
2272         return retval;
2273 }
2274
2275 static void e1000_set_itr(struct e1000_adapter *adapter)
2276 {
2277         struct e1000_hw *hw = &adapter->hw;
2278         u16 current_itr;
2279         u32 new_itr = adapter->itr;
2280
2281         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2282         if (adapter->link_speed != SPEED_1000) {
2283                 current_itr = 0;
2284                 new_itr = 4000;
2285                 goto set_itr_now;
2286         }
2287
2288         adapter->tx_itr = e1000_update_itr(adapter,
2289                                     adapter->tx_itr,
2290                                     adapter->total_tx_packets,
2291                                     adapter->total_tx_bytes);
2292         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2293         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2294                 adapter->tx_itr = low_latency;
2295
2296         adapter->rx_itr = e1000_update_itr(adapter,
2297                                     adapter->rx_itr,
2298                                     adapter->total_rx_packets,
2299                                     adapter->total_rx_bytes);
2300         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2301         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2302                 adapter->rx_itr = low_latency;
2303
2304         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2305
2306         switch (current_itr) {
2307         /* counts and packets in update_itr are dependent on these numbers */
2308         case lowest_latency:
2309                 new_itr = 70000;
2310                 break;
2311         case low_latency:
2312                 new_itr = 20000; /* aka hwitr = ~200 */
2313                 break;
2314         case bulk_latency:
2315                 new_itr = 4000;
2316                 break;
2317         default:
2318                 break;
2319         }
2320
2321 set_itr_now:
2322         if (new_itr != adapter->itr) {
2323                 /*
2324                  * this attempts to bias the interrupt rate towards Bulk
2325                  * by adding intermediate steps when interrupt rate is
2326                  * increasing
2327                  */
2328                 new_itr = new_itr > adapter->itr ?
2329                              min(adapter->itr + (new_itr >> 2), new_itr) :
2330                              new_itr;
2331                 adapter->itr = new_itr;
2332                 adapter->rx_ring->itr_val = new_itr;
2333                 if (adapter->msix_entries)
2334                         adapter->rx_ring->set_itr = 1;
2335                 else
2336                         ew32(ITR, 1000000000 / (new_itr * 256));
2337         }
2338 }
2339
2340 /**
2341  * e1000_alloc_queues - Allocate memory for all rings
2342  * @adapter: board private structure to initialize
2343  **/
2344 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2345 {
2346         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2347         if (!adapter->tx_ring)
2348                 goto err;
2349
2350         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2351         if (!adapter->rx_ring)
2352                 goto err;
2353
2354         return 0;
2355 err:
2356         e_err("Unable to allocate memory for queues\n");
2357         kfree(adapter->rx_ring);
2358         kfree(adapter->tx_ring);
2359         return -ENOMEM;
2360 }
2361
2362 /**
2363  * e1000_clean - NAPI Rx polling callback
2364  * @napi: struct associated with this polling callback
2365  * @budget: amount of packets driver is allowed to process this poll
2366  **/
2367 static int e1000_clean(struct napi_struct *napi, int budget)
2368 {
2369         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2370         struct e1000_hw *hw = &adapter->hw;
2371         struct net_device *poll_dev = adapter->netdev;
2372         int tx_cleaned = 1, work_done = 0;
2373
2374         adapter = netdev_priv(poll_dev);
2375
2376         if (adapter->msix_entries &&
2377             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2378                 goto clean_rx;
2379
2380         tx_cleaned = e1000_clean_tx_irq(adapter);
2381
2382 clean_rx:
2383         adapter->clean_rx(adapter, &work_done, budget);
2384
2385         if (!tx_cleaned)
2386                 work_done = budget;
2387
2388         /* If budget not fully consumed, exit the polling mode */
2389         if (work_done < budget) {
2390                 if (adapter->itr_setting & 3)
2391                         e1000_set_itr(adapter);
2392                 napi_complete(napi);
2393                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2394                         if (adapter->msix_entries)
2395                                 ew32(IMS, adapter->rx_ring->ims_val);
2396                         else
2397                                 e1000_irq_enable(adapter);
2398                 }
2399         }
2400
2401         return work_done;
2402 }
2403
2404 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2405 {
2406         struct e1000_adapter *adapter = netdev_priv(netdev);
2407         struct e1000_hw *hw = &adapter->hw;
2408         u32 vfta, index;
2409
2410         /* don't update vlan cookie if already programmed */
2411         if ((adapter->hw.mng_cookie.status &
2412              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2413             (vid == adapter->mng_vlan_id))
2414                 return;
2415
2416         /* add VID to filter table */
2417         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2418                 index = (vid >> 5) & 0x7F;
2419                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2420                 vfta |= (1 << (vid & 0x1F));
2421                 hw->mac.ops.write_vfta(hw, index, vfta);
2422         }
2423 }
2424
2425 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2426 {
2427         struct e1000_adapter *adapter = netdev_priv(netdev);
2428         struct e1000_hw *hw = &adapter->hw;
2429         u32 vfta, index;
2430
2431         if (!test_bit(__E1000_DOWN, &adapter->state))
2432                 e1000_irq_disable(adapter);
2433         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2434
2435         if (!test_bit(__E1000_DOWN, &adapter->state))
2436                 e1000_irq_enable(adapter);
2437
2438         if ((adapter->hw.mng_cookie.status &
2439              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2440             (vid == adapter->mng_vlan_id)) {
2441                 /* release control to f/w */
2442                 e1000_release_hw_control(adapter);
2443                 return;
2444         }
2445
2446         /* remove VID from filter table */
2447         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2448                 index = (vid >> 5) & 0x7F;
2449                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2450                 vfta &= ~(1 << (vid & 0x1F));
2451                 hw->mac.ops.write_vfta(hw, index, vfta);
2452         }
2453 }
2454
2455 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2456 {
2457         struct net_device *netdev = adapter->netdev;
2458         u16 vid = adapter->hw.mng_cookie.vlan_id;
2459         u16 old_vid = adapter->mng_vlan_id;
2460
2461         if (!adapter->vlgrp)
2462                 return;
2463
2464         if (!vlan_group_get_device(adapter->vlgrp, vid)) {
2465                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2466                 if (adapter->hw.mng_cookie.status &
2467                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2468                         e1000_vlan_rx_add_vid(netdev, vid);
2469                         adapter->mng_vlan_id = vid;
2470                 }
2471
2472                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
2473                                 (vid != old_vid) &&
2474                     !vlan_group_get_device(adapter->vlgrp, old_vid))
2475                         e1000_vlan_rx_kill_vid(netdev, old_vid);
2476         } else {
2477                 adapter->mng_vlan_id = vid;
2478         }
2479 }
2480
2481
2482 static void e1000_vlan_rx_register(struct net_device *netdev,
2483                                    struct vlan_group *grp)
2484 {
2485         struct e1000_adapter *adapter = netdev_priv(netdev);
2486         struct e1000_hw *hw = &adapter->hw;
2487         u32 ctrl, rctl;
2488
2489         if (!test_bit(__E1000_DOWN, &adapter->state))
2490                 e1000_irq_disable(adapter);
2491         adapter->vlgrp = grp;
2492
2493         if (grp) {
2494                 /* enable VLAN tag insert/strip */
2495                 ctrl = er32(CTRL);
2496                 ctrl |= E1000_CTRL_VME;
2497                 ew32(CTRL, ctrl);
2498
2499                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2500                         /* enable VLAN receive filtering */
2501                         rctl = er32(RCTL);
2502                         rctl &= ~E1000_RCTL_CFIEN;
2503                         ew32(RCTL, rctl);
2504                         e1000_update_mng_vlan(adapter);
2505                 }
2506         } else {
2507                 /* disable VLAN tag insert/strip */
2508                 ctrl = er32(CTRL);
2509                 ctrl &= ~E1000_CTRL_VME;
2510                 ew32(CTRL, ctrl);
2511
2512                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2513                         if (adapter->mng_vlan_id !=
2514                             (u16)E1000_MNG_VLAN_NONE) {
2515                                 e1000_vlan_rx_kill_vid(netdev,
2516                                                        adapter->mng_vlan_id);
2517                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2518                         }
2519                 }
2520         }
2521
2522         if (!test_bit(__E1000_DOWN, &adapter->state))
2523                 e1000_irq_enable(adapter);
2524 }
2525
2526 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2527 {
2528         u16 vid;
2529
2530         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2531
2532         if (!adapter->vlgrp)
2533                 return;
2534
2535         for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2536                 if (!vlan_group_get_device(adapter->vlgrp, vid))
2537                         continue;
2538                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2539         }
2540 }
2541
2542 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2543 {
2544         struct e1000_hw *hw = &adapter->hw;
2545         u32 manc, manc2h, mdef, i, j;
2546
2547         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2548                 return;
2549
2550         manc = er32(MANC);
2551
2552         /*
2553          * enable receiving management packets to the host. this will probably
2554          * generate destination unreachable messages from the host OS, but
2555          * the packets will be handled on SMBUS
2556          */
2557         manc |= E1000_MANC_EN_MNG2HOST;
2558         manc2h = er32(MANC2H);
2559
2560         switch (hw->mac.type) {
2561         default:
2562                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2563                 break;
2564         case e1000_82574:
2565         case e1000_82583:
2566                 /*
2567                  * Check if IPMI pass-through decision filter already exists;
2568                  * if so, enable it.
2569                  */
2570                 for (i = 0, j = 0; i < 8; i++) {
2571                         mdef = er32(MDEF(i));
2572
2573                         /* Ignore filters with anything other than IPMI ports */
2574                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2575                                 continue;
2576
2577                         /* Enable this decision filter in MANC2H */
2578                         if (mdef)
2579                                 manc2h |= (1 << i);
2580
2581                         j |= mdef;
2582                 }
2583
2584                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2585                         break;
2586
2587                 /* Create new decision filter in an empty filter */
2588                 for (i = 0, j = 0; i < 8; i++)
2589                         if (er32(MDEF(i)) == 0) {
2590                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2591                                                E1000_MDEF_PORT_664));
2592                                 manc2h |= (1 << 1);
2593                                 j++;
2594                                 break;
2595                         }
2596
2597                 if (!j)
2598                         e_warn("Unable to create IPMI pass-through filter\n");
2599                 break;
2600         }
2601
2602         ew32(MANC2H, manc2h);
2603         ew32(MANC, manc);
2604 }
2605
2606 /**
2607  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
2608  * @adapter: board private structure
2609  *
2610  * Configure the Tx unit of the MAC after a reset.
2611  **/
2612 static void e1000_configure_tx(struct e1000_adapter *adapter)
2613 {
2614         struct e1000_hw *hw = &adapter->hw;
2615         struct e1000_ring *tx_ring = adapter->tx_ring;
2616         u64 tdba;
2617         u32 tdlen, tctl, tipg, tarc;
2618         u32 ipgr1, ipgr2;
2619
2620         /* Setup the HW Tx Head and Tail descriptor pointers */
2621         tdba = tx_ring->dma;
2622         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2623         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2624         ew32(TDBAH, (tdba >> 32));
2625         ew32(TDLEN, tdlen);
2626         ew32(TDH, 0);
2627         ew32(TDT, 0);
2628         tx_ring->head = E1000_TDH;
2629         tx_ring->tail = E1000_TDT;
2630
2631         /* Set the default values for the Tx Inter Packet Gap timer */
2632         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2633         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2634         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2635
2636         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2637                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2638
2639         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2640         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2641         ew32(TIPG, tipg);
2642
2643         /* Set the Tx Interrupt Delay register */
2644         ew32(TIDV, adapter->tx_int_delay);
2645         /* Tx irq moderation */
2646         ew32(TADV, adapter->tx_abs_int_delay);
2647
2648         /* Program the Transmit Control Register */
2649         tctl = er32(TCTL);
2650         tctl &= ~E1000_TCTL_CT;
2651         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2652                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2653
2654         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2655                 tarc = er32(TARC(0));
2656                 /*
2657                  * set the speed mode bit, we'll clear it if we're not at
2658                  * gigabit link later
2659                  */
2660 #define SPEED_MODE_BIT (1 << 21)
2661                 tarc |= SPEED_MODE_BIT;
2662                 ew32(TARC(0), tarc);
2663         }
2664
2665         /* errata: program both queues to unweighted RR */
2666         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2667                 tarc = er32(TARC(0));
2668                 tarc |= 1;
2669                 ew32(TARC(0), tarc);
2670                 tarc = er32(TARC(1));
2671                 tarc |= 1;
2672                 ew32(TARC(1), tarc);
2673         }
2674
2675         /* Setup Transmit Descriptor Settings for eop descriptor */
2676         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2677
2678         /* only set IDE if we are delaying interrupts using the timers */
2679         if (adapter->tx_int_delay)
2680                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2681
2682         /* enable Report Status bit */
2683         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2684
2685         ew32(TCTL, tctl);
2686
2687         e1000e_config_collision_dist(hw);
2688 }
2689
2690 /**
2691  * e1000_setup_rctl - configure the receive control registers
2692  * @adapter: Board private structure
2693  **/
2694 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2695                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2696 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2697 {
2698         struct e1000_hw *hw = &adapter->hw;
2699         u32 rctl, rfctl;
2700         u32 psrctl = 0;
2701         u32 pages = 0;
2702
2703         /* Program MC offset vector base */
2704         rctl = er32(RCTL);
2705         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2706         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2707                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2708                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2709
2710         /* Do not Store bad packets */
2711         rctl &= ~E1000_RCTL_SBP;
2712
2713         /* Enable Long Packet receive */
2714         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2715                 rctl &= ~E1000_RCTL_LPE;
2716         else
2717                 rctl |= E1000_RCTL_LPE;
2718
2719         /* Some systems expect that the CRC is included in SMBUS traffic. The
2720          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2721          * host memory when this is enabled
2722          */
2723         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2724                 rctl |= E1000_RCTL_SECRC;
2725
2726         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2727         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2728                 u16 phy_data;
2729
2730                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2731                 phy_data &= 0xfff8;
2732                 phy_data |= (1 << 2);
2733                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2734
2735                 e1e_rphy(hw, 22, &phy_data);
2736                 phy_data &= 0x0fff;
2737                 phy_data |= (1 << 14);
2738                 e1e_wphy(hw, 0x10, 0x2823);
2739                 e1e_wphy(hw, 0x11, 0x0003);
2740                 e1e_wphy(hw, 22, phy_data);
2741         }
2742
2743         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2744         if (hw->mac.type == e1000_pch2lan) {
2745                 s32 ret_val;
2746
2747                 if (rctl & E1000_RCTL_LPE)
2748                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2749                 else
2750                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2751         }
2752
2753         /* Setup buffer sizes */
2754         rctl &= ~E1000_RCTL_SZ_4096;
2755         rctl |= E1000_RCTL_BSEX;
2756         switch (adapter->rx_buffer_len) {
2757         case 2048:
2758         default:
2759                 rctl |= E1000_RCTL_SZ_2048;
2760                 rctl &= ~E1000_RCTL_BSEX;
2761                 break;
2762         case 4096:
2763                 rctl |= E1000_RCTL_SZ_4096;
2764                 break;
2765         case 8192:
2766                 rctl |= E1000_RCTL_SZ_8192;
2767                 break;
2768         case 16384:
2769                 rctl |= E1000_RCTL_SZ_16384;
2770                 break;
2771         }
2772
2773         /*
2774          * 82571 and greater support packet-split where the protocol
2775          * header is placed in skb->data and the packet data is
2776          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2777          * In the case of a non-split, skb->data is linearly filled,
2778          * followed by the page buffers.  Therefore, skb->data is
2779          * sized to hold the largest protocol header.
2780          *
2781          * allocations using alloc_page take too long for regular MTU
2782          * so only enable packet split for jumbo frames
2783          *
2784          * Using pages when the page size is greater than 16k wastes
2785          * a lot of memory, since we allocate 3 pages at all times
2786          * per packet.
2787          */
2788         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2789         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2790             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2791                 adapter->rx_ps_pages = pages;
2792         else
2793                 adapter->rx_ps_pages = 0;
2794
2795         if (adapter->rx_ps_pages) {
2796                 /* Configure extra packet-split registers */
2797                 rfctl = er32(RFCTL);
2798                 rfctl |= E1000_RFCTL_EXTEN;
2799                 /*
2800                  * disable packet split support for IPv6 extension headers,
2801                  * because some malformed IPv6 headers can hang the Rx
2802                  */
2803                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2804                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2805
2806                 ew32(RFCTL, rfctl);
2807
2808                 /* Enable Packet split descriptors */
2809                 rctl |= E1000_RCTL_DTYP_PS;
2810
2811                 psrctl |= adapter->rx_ps_bsize0 >>
2812                         E1000_PSRCTL_BSIZE0_SHIFT;
2813
2814                 switch (adapter->rx_ps_pages) {
2815                 case 3:
2816                         psrctl |= PAGE_SIZE <<
2817                                 E1000_PSRCTL_BSIZE3_SHIFT;
2818                 case 2:
2819                         psrctl |= PAGE_SIZE <<
2820                                 E1000_PSRCTL_BSIZE2_SHIFT;
2821                 case 1:
2822                         psrctl |= PAGE_SIZE >>
2823                                 E1000_PSRCTL_BSIZE1_SHIFT;
2824                         break;
2825                 }
2826
2827                 ew32(PSRCTL, psrctl);
2828         }
2829
2830         ew32(RCTL, rctl);
2831         /* just started the receive unit, no need to restart */
2832         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2833 }
2834
2835 /**
2836  * e1000_configure_rx - Configure Receive Unit after Reset
2837  * @adapter: board private structure
2838  *
2839  * Configure the Rx unit of the MAC after a reset.
2840  **/
2841 static void e1000_configure_rx(struct e1000_adapter *adapter)
2842 {
2843         struct e1000_hw *hw = &adapter->hw;
2844         struct e1000_ring *rx_ring = adapter->rx_ring;
2845         u64 rdba;
2846         u32 rdlen, rctl, rxcsum, ctrl_ext;
2847
2848         if (adapter->rx_ps_pages) {
2849                 /* this is a 32 byte descriptor */
2850                 rdlen = rx_ring->count *
2851                         sizeof(union e1000_rx_desc_packet_split);
2852                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2853                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2854         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2855                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2856                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2857                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2858         } else {
2859                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2860                 adapter->clean_rx = e1000_clean_rx_irq;
2861                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2862         }
2863
2864         /* disable receives while setting up the descriptors */
2865         rctl = er32(RCTL);
2866         ew32(RCTL, rctl & ~E1000_RCTL_EN);
2867         e1e_flush();
2868         msleep(10);
2869
2870         /* set the Receive Delay Timer Register */
2871         ew32(RDTR, adapter->rx_int_delay);
2872
2873         /* irq moderation */
2874         ew32(RADV, adapter->rx_abs_int_delay);
2875         if (adapter->itr_setting != 0)
2876                 ew32(ITR, 1000000000 / (adapter->itr * 256));
2877
2878         ctrl_ext = er32(CTRL_EXT);
2879         /* Auto-Mask interrupts upon ICR access */
2880         ctrl_ext |= E1000_CTRL_EXT_IAME;
2881         ew32(IAM, 0xffffffff);
2882         ew32(CTRL_EXT, ctrl_ext);
2883         e1e_flush();
2884
2885         /*
2886          * Setup the HW Rx Head and Tail Descriptor Pointers and
2887          * the Base and Length of the Rx Descriptor Ring
2888          */
2889         rdba = rx_ring->dma;
2890         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2891         ew32(RDBAH, (rdba >> 32));
2892         ew32(RDLEN, rdlen);
2893         ew32(RDH, 0);
2894         ew32(RDT, 0);
2895         rx_ring->head = E1000_RDH;
2896         rx_ring->tail = E1000_RDT;
2897
2898         /* Enable Receive Checksum Offload for TCP and UDP */
2899         rxcsum = er32(RXCSUM);
2900         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2901                 rxcsum |= E1000_RXCSUM_TUOFL;
2902
2903                 /*
2904                  * IPv4 payload checksum for UDP fragments must be
2905                  * used in conjunction with packet-split.
2906                  */
2907                 if (adapter->rx_ps_pages)
2908                         rxcsum |= E1000_RXCSUM_IPPCSE;
2909         } else {
2910                 rxcsum &= ~E1000_RXCSUM_TUOFL;
2911                 /* no need to clear IPPCSE as it defaults to 0 */
2912         }
2913         ew32(RXCSUM, rxcsum);
2914
2915         /*
2916          * Enable early receives on supported devices, only takes effect when
2917          * packet size is equal or larger than the specified value (in 8 byte
2918          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2919          */
2920         if (adapter->flags & FLAG_HAS_ERT) {
2921                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2922                         u32 rxdctl = er32(RXDCTL(0));
2923                         ew32(RXDCTL(0), rxdctl | 0x3);
2924                         ew32(ERT, E1000_ERT_2048 | (1 << 13));
2925                         /*
2926                          * With jumbo frames and early-receive enabled,
2927                          * excessive C-state transition latencies result in
2928                          * dropped transactions.
2929                          */
2930                         pm_qos_update_request(
2931                                 adapter->netdev->pm_qos_req, 55);
2932                 } else {
2933                         pm_qos_update_request(
2934                                 adapter->netdev->pm_qos_req,
2935                                 PM_QOS_DEFAULT_VALUE);
2936                 }
2937         }
2938
2939         /* Enable Receives */
2940         ew32(RCTL, rctl);
2941 }
2942
2943 /**
2944  *  e1000_update_mc_addr_list - Update Multicast addresses
2945  *  @hw: pointer to the HW structure
2946  *  @mc_addr_list: array of multicast addresses to program
2947  *  @mc_addr_count: number of multicast addresses to program
2948  *
2949  *  Updates the Multicast Table Array.
2950  *  The caller must have a packed mc_addr_list of multicast addresses.
2951  **/
2952 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
2953                                       u32 mc_addr_count)
2954 {
2955         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
2956 }
2957
2958 /**
2959  * e1000_set_multi - Multicast and Promiscuous mode set
2960  * @netdev: network interface device structure
2961  *
2962  * The set_multi entry point is called whenever the multicast address
2963  * list or the network interface flags are updated.  This routine is
2964  * responsible for configuring the hardware for proper multicast,
2965  * promiscuous mode, and all-multi behavior.
2966  **/
2967 static void e1000_set_multi(struct net_device *netdev)
2968 {
2969         struct e1000_adapter *adapter = netdev_priv(netdev);
2970         struct e1000_hw *hw = &adapter->hw;
2971         struct netdev_hw_addr *ha;
2972         u8  *mta_list;
2973         u32 rctl;
2974         int i;
2975
2976         /* Check for Promiscuous and All Multicast modes */
2977
2978         rctl = er32(RCTL);
2979
2980         if (netdev->flags & IFF_PROMISC) {
2981                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2982                 rctl &= ~E1000_RCTL_VFE;
2983         } else {
2984                 if (netdev->flags & IFF_ALLMULTI) {
2985                         rctl |= E1000_RCTL_MPE;
2986                         rctl &= ~E1000_RCTL_UPE;
2987                 } else {
2988                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2989                 }
2990                 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
2991                         rctl |= E1000_RCTL_VFE;
2992         }
2993
2994         ew32(RCTL, rctl);
2995
2996         if (!netdev_mc_empty(netdev)) {
2997                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
2998                 if (!mta_list)
2999                         return;
3000
3001                 /* prepare a packed array of only addresses. */
3002                 i = 0;
3003                 netdev_for_each_mc_addr(ha, netdev)
3004                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3005
3006                 e1000_update_mc_addr_list(hw, mta_list, i);
3007                 kfree(mta_list);
3008         } else {
3009                 /*
3010                  * if we're called from probe, we might not have
3011                  * anything to do here, so clear out the list
3012                  */
3013                 e1000_update_mc_addr_list(hw, NULL, 0);
3014         }
3015 }
3016
3017 /**
3018  * e1000_configure - configure the hardware for Rx and Tx
3019  * @adapter: private board structure
3020  **/
3021 static void e1000_configure(struct e1000_adapter *adapter)
3022 {
3023         e1000_set_multi(adapter->netdev);
3024
3025         e1000_restore_vlan(adapter);
3026         e1000_init_manageability_pt(adapter);
3027
3028         e1000_configure_tx(adapter);
3029         e1000_setup_rctl(adapter);
3030         e1000_configure_rx(adapter);
3031         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3032 }
3033
3034 /**
3035  * e1000e_power_up_phy - restore link in case the phy was powered down
3036  * @adapter: address of board private structure
3037  *
3038  * The phy may be powered down to save power and turn off link when the
3039  * driver is unloaded and wake on lan is not enabled (among others)
3040  * *** this routine MUST be followed by a call to e1000e_reset ***
3041  **/
3042 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3043 {
3044         if (adapter->hw.phy.ops.power_up)
3045                 adapter->hw.phy.ops.power_up(&adapter->hw);
3046
3047         adapter->hw.mac.ops.setup_link(&adapter->hw);
3048 }
3049
3050 /**
3051  * e1000_power_down_phy - Power down the PHY
3052  *
3053  * Power down the PHY so no link is implied when interface is down.
3054  * The PHY cannot be powered down if management or WoL is active.
3055  */
3056 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3057 {
3058         /* WoL is enabled */
3059         if (adapter->wol)
3060                 return;
3061
3062         if (adapter->hw.phy.ops.power_down)
3063                 adapter->hw.phy.ops.power_down(&adapter->hw);
3064 }
3065
3066 /**
3067  * e1000e_reset - bring the hardware into a known good state
3068  *
3069  * This function boots the hardware and enables some settings that
3070  * require a configuration cycle of the hardware - those cannot be
3071  * set/changed during runtime. After reset the device needs to be
3072  * properly configured for Rx, Tx etc.
3073  */
3074 void e1000e_reset(struct e1000_adapter *adapter)
3075 {
3076         struct e1000_mac_info *mac = &adapter->hw.mac;
3077         struct e1000_fc_info *fc = &adapter->hw.fc;
3078         struct e1000_hw *hw = &adapter->hw;
3079         u32 tx_space, min_tx_space, min_rx_space;
3080         u32 pba = adapter->pba;
3081         u16 hwm;
3082
3083         /* reset Packet Buffer Allocation to default */
3084         ew32(PBA, pba);
3085
3086         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3087                 /*
3088                  * To maintain wire speed transmits, the Tx FIFO should be
3089                  * large enough to accommodate two full transmit packets,
3090                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3091                  * the Rx FIFO should be large enough to accommodate at least
3092                  * one full receive packet and is similarly rounded up and
3093                  * expressed in KB.
3094                  */
3095                 pba = er32(PBA);
3096                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3097                 tx_space = pba >> 16;
3098                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3099                 pba &= 0xffff;
3100                 /*
3101                  * the Tx fifo also stores 16 bytes of information about the tx
3102                  * but don't include ethernet FCS because hardware appends it
3103                  */
3104                 min_tx_space = (adapter->max_frame_size +
3105                                 sizeof(struct e1000_tx_desc) -
3106                                 ETH_FCS_LEN) * 2;
3107                 min_tx_space = ALIGN(min_tx_space, 1024);
3108                 min_tx_space >>= 10;
3109                 /* software strips receive CRC, so leave room for it */
3110                 min_rx_space = adapter->max_frame_size;
3111                 min_rx_space = ALIGN(min_rx_space, 1024);
3112                 min_rx_space >>= 10;
3113
3114                 /*
3115                  * If current Tx allocation is less than the min Tx FIFO size,
3116                  * and the min Tx FIFO size is less than the current Rx FIFO
3117                  * allocation, take space away from current Rx allocation
3118                  */
3119                 if ((tx_space < min_tx_space) &&
3120                     ((min_tx_space - tx_space) < pba)) {
3121                         pba -= min_tx_space - tx_space;
3122
3123                         /*
3124                          * if short on Rx space, Rx wins and must trump tx
3125                          * adjustment or use Early Receive if available
3126                          */
3127                         if ((pba < min_rx_space) &&
3128                             (!(adapter->flags & FLAG_HAS_ERT)))
3129                                 /* ERT enabled in e1000_configure_rx */
3130                                 pba = min_rx_space;
3131                 }
3132
3133                 ew32(PBA, pba);
3134         }
3135
3136
3137         /*
3138          * flow control settings
3139          *
3140          * The high water mark must be low enough to fit one full frame
3141          * (or the size used for early receive) above it in the Rx FIFO.
3142          * Set it to the lower of:
3143          * - 90% of the Rx FIFO size, and
3144          * - the full Rx FIFO size minus the early receive size (for parts
3145          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3146          * - the full Rx FIFO size minus one full frame
3147          */
3148         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3149                 fc->pause_time = 0xFFFF;
3150         else
3151                 fc->pause_time = E1000_FC_PAUSE_TIME;
3152         fc->send_xon = 1;
3153         fc->current_mode = fc->requested_mode;
3154
3155         switch (hw->mac.type) {
3156         default:
3157                 if ((adapter->flags & FLAG_HAS_ERT) &&
3158                     (adapter->netdev->mtu > ETH_DATA_LEN))
3159                         hwm = min(((pba << 10) * 9 / 10),
3160                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3161                 else
3162                         hwm = min(((pba << 10) * 9 / 10),
3163                                   ((pba << 10) - adapter->max_frame_size));
3164
3165                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3166                 fc->low_water = fc->high_water - 8;
3167                 break;
3168         case e1000_pchlan:
3169                 /*
3170                  * Workaround PCH LOM adapter hangs with certain network
3171                  * loads.  If hangs persist, try disabling Tx flow control.
3172                  */
3173                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3174                         fc->high_water = 0x3500;
3175                         fc->low_water  = 0x1500;
3176                 } else {
3177                         fc->high_water = 0x5000;
3178                         fc->low_water  = 0x3000;
3179                 }
3180                 fc->refresh_time = 0x1000;
3181                 break;
3182         case e1000_pch2lan:
3183                 fc->high_water = 0x05C20;
3184                 fc->low_water = 0x05048;
3185                 fc->pause_time = 0x0650;
3186                 fc->refresh_time = 0x0400;
3187                 break;
3188         }
3189
3190         /* Allow time for pending master requests to run */
3191         mac->ops.reset_hw(hw);
3192
3193         /*
3194          * For parts with AMT enabled, let the firmware know
3195          * that the network interface is in control
3196          */
3197         if (adapter->flags & FLAG_HAS_AMT)
3198                 e1000_get_hw_control(adapter);
3199
3200         ew32(WUC, 0);
3201
3202         if (mac->ops.init_hw(hw))
3203                 e_err("Hardware Error\n");
3204
3205         e1000_update_mng_vlan(adapter);
3206
3207         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3208         ew32(VET, ETH_P_8021Q);
3209
3210         e1000e_reset_adaptive(hw);
3211         e1000_get_phy_info(hw);
3212
3213         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3214             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3215                 u16 phy_data = 0;
3216                 /*
3217                  * speed up time to link by disabling smart power down, ignore
3218                  * the return value of this function because there is nothing
3219                  * different we would do if it failed
3220                  */
3221                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3222                 phy_data &= ~IGP02E1000_PM_SPD;
3223                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3224         }
3225 }
3226
3227 int e1000e_up(struct e1000_adapter *adapter)
3228 {
3229         struct e1000_hw *hw = &adapter->hw;
3230
3231         /* hardware has been reset, we need to reload some things */
3232         e1000_configure(adapter);
3233
3234         clear_bit(__E1000_DOWN, &adapter->state);
3235
3236         napi_enable(&adapter->napi);
3237         if (adapter->msix_entries)
3238                 e1000_configure_msix(adapter);
3239         e1000_irq_enable(adapter);
3240
3241         netif_wake_queue(adapter->netdev);
3242
3243         /* fire a link change interrupt to start the watchdog */
3244         if (adapter->msix_entries)
3245                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3246         else
3247                 ew32(ICS, E1000_ICS_LSC);
3248
3249         return 0;
3250 }
3251
3252 void e1000e_down(struct e1000_adapter *adapter)
3253 {
3254         struct net_device *netdev = adapter->netdev;
3255         struct e1000_hw *hw = &adapter->hw;
3256         u32 tctl, rctl;
3257
3258         /*
3259          * signal that we're down so the interrupt handler does not
3260          * reschedule our watchdog timer
3261          */
3262         set_bit(__E1000_DOWN, &adapter->state);
3263
3264         /* disable receives in the hardware */
3265         rctl = er32(RCTL);
3266         ew32(RCTL, rctl & ~E1000_RCTL_EN);
3267         /* flush and sleep below */
3268
3269         netif_stop_queue(netdev);
3270
3271         /* disable transmits in the hardware */
3272         tctl = er32(TCTL);
3273         tctl &= ~E1000_TCTL_EN;
3274         ew32(TCTL, tctl);
3275         /* flush both disables and wait for them to finish */
3276         e1e_flush();
3277         msleep(10);
3278
3279         napi_disable(&adapter->napi);
3280         e1000_irq_disable(adapter);
3281
3282         del_timer_sync(&adapter->watchdog_timer);
3283         del_timer_sync(&adapter->phy_info_timer);
3284
3285         netif_carrier_off(netdev);
3286         adapter->link_speed = 0;
3287         adapter->link_duplex = 0;
3288
3289         if (!pci_channel_offline(adapter->pdev))
3290                 e1000e_reset(adapter);
3291         e1000_clean_tx_ring(adapter);
3292         e1000_clean_rx_ring(adapter);
3293
3294         /*
3295          * TODO: for power management, we could drop the link and
3296          * pci_disable_device here.
3297          */
3298 }
3299
3300 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3301 {
3302         might_sleep();
3303         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3304                 msleep(1);
3305         e1000e_down(adapter);
3306         e1000e_up(adapter);
3307         clear_bit(__E1000_RESETTING, &adapter->state);
3308 }
3309
3310 /**
3311  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3312  * @adapter: board private structure to initialize
3313  *
3314  * e1000_sw_init initializes the Adapter private data structure.
3315  * Fields are initialized based on PCI device information and
3316  * OS network device settings (MTU size).
3317  **/
3318 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3319 {
3320         struct net_device *netdev = adapter->netdev;
3321
3322         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3323         adapter->rx_ps_bsize0 = 128;
3324         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3325         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3326
3327         e1000e_set_interrupt_capability(adapter);
3328
3329         if (e1000_alloc_queues(adapter))
3330                 return -ENOMEM;
3331
3332         /* Explicitly disable IRQ since the NIC can be in any state. */
3333         e1000_irq_disable(adapter);
3334
3335         set_bit(__E1000_DOWN, &adapter->state);
3336         return 0;
3337 }
3338
3339 /**
3340  * e1000_intr_msi_test - Interrupt Handler
3341  * @irq: interrupt number
3342  * @data: pointer to a network interface device structure
3343  **/
3344 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3345 {
3346         struct net_device *netdev = data;
3347         struct e1000_adapter *adapter = netdev_priv(netdev);
3348         struct e1000_hw *hw = &adapter->hw;
3349         u32 icr = er32(ICR);
3350
3351         e_dbg("icr is %08X\n", icr);
3352         if (icr & E1000_ICR_RXSEQ) {
3353                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3354                 wmb();
3355         }
3356
3357         return IRQ_HANDLED;
3358 }
3359
3360 /**
3361  * e1000_test_msi_interrupt - Returns 0 for successful test
3362  * @adapter: board private struct
3363  *
3364  * code flow taken from tg3.c
3365  **/
3366 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3367 {
3368         struct net_device *netdev = adapter->netdev;
3369         struct e1000_hw *hw = &adapter->hw;
3370         int err;
3371
3372         /* poll_enable hasn't been called yet, so don't need disable */
3373         /* clear any pending events */
3374         er32(ICR);
3375
3376         /* free the real vector and request a test handler */
3377         e1000_free_irq(adapter);
3378         e1000e_reset_interrupt_capability(adapter);
3379
3380         /* Assume that the test fails, if it succeeds then the test
3381          * MSI irq handler will unset this flag */
3382         adapter->flags |= FLAG_MSI_TEST_FAILED;
3383
3384         err = pci_enable_msi(adapter->pdev);
3385         if (err)
3386                 goto msi_test_failed;
3387
3388         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3389                           netdev->name, netdev);
3390         if (err) {
3391                 pci_disable_msi(adapter->pdev);
3392                 goto msi_test_failed;
3393         }
3394
3395         wmb();
3396
3397         e1000_irq_enable(adapter);
3398
3399         /* fire an unusual interrupt on the test handler */
3400         ew32(ICS, E1000_ICS_RXSEQ);
3401         e1e_flush();
3402         msleep(50);
3403
3404         e1000_irq_disable(adapter);
3405
3406         rmb();
3407
3408         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3409                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3410                 err = -EIO;
3411                 e_info("MSI interrupt test failed!\n");
3412         }
3413
3414         free_irq(adapter->pdev->irq, netdev);
3415         pci_disable_msi(adapter->pdev);
3416
3417         if (err == -EIO)
3418                 goto msi_test_failed;
3419
3420         /* okay so the test worked, restore settings */
3421         e_dbg("MSI interrupt test succeeded!\n");
3422 msi_test_failed:
3423         e1000e_set_interrupt_capability(adapter);
3424         e1000_request_irq(adapter);
3425         return err;
3426 }
3427
3428 /**
3429  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3430  * @adapter: board private struct
3431  *
3432  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3433  **/
3434 static int e1000_test_msi(struct e1000_adapter *adapter)
3435 {
3436         int err;
3437         u16 pci_cmd;
3438
3439         if (!(adapter->flags & FLAG_MSI_ENABLED))
3440                 return 0;
3441
3442         /* disable SERR in case the MSI write causes a master abort */
3443         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3444         if (pci_cmd & PCI_COMMAND_SERR)
3445                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3446                                       pci_cmd & ~PCI_COMMAND_SERR);
3447
3448         err = e1000_test_msi_interrupt(adapter);
3449
3450         /* re-enable SERR */
3451         if (pci_cmd & PCI_COMMAND_SERR) {
3452                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3453                 pci_cmd |= PCI_COMMAND_SERR;
3454                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3455         }
3456
3457         /* success ! */
3458         if (!err)
3459                 return 0;
3460
3461         /* EIO means MSI test failed */
3462         if (err != -EIO)
3463                 return err;
3464
3465         /* back to INTx mode */
3466         e_warn("MSI interrupt test failed, using legacy interrupt.\n");
3467
3468         e1000_free_irq(adapter);
3469
3470         err = e1000_request_irq(adapter);
3471
3472         return err;
3473 }
3474
3475 /**
3476  * e1000_open - Called when a network interface is made active
3477  * @netdev: network interface device structure
3478  *
3479  * Returns 0 on success, negative value on failure
3480  *
3481  * The open entry point is called when a network interface is made
3482  * active by the system (IFF_UP).  At this point all resources needed
3483  * for transmit and receive operations are allocated, the interrupt
3484  * handler is registered with the OS, the watchdog timer is started,
3485  * and the stack is notified that the interface is ready.
3486  **/
3487 static int e1000_open(struct net_device *netdev)
3488 {
3489         struct e1000_adapter *adapter = netdev_priv(netdev);
3490         struct e1000_hw *hw = &adapter->hw;
3491         struct pci_dev *pdev = adapter->pdev;
3492         int err;
3493
3494         /* disallow open during test */
3495         if (test_bit(__E1000_TESTING, &adapter->state))
3496                 return -EBUSY;
3497
3498         pm_runtime_get_sync(&pdev->dev);
3499
3500         netif_carrier_off(netdev);
3501
3502         /* allocate transmit descriptors */
3503         err = e1000e_setup_tx_resources(adapter);
3504         if (err)
3505                 goto err_setup_tx;
3506
3507         /* allocate receive descriptors */
3508         err = e1000e_setup_rx_resources(adapter);
3509         if (err)
3510                 goto err_setup_rx;
3511
3512         /*
3513          * If AMT is enabled, let the firmware know that the network
3514          * interface is now open and reset the part to a known state.
3515          */
3516         if (adapter->flags & FLAG_HAS_AMT) {
3517                 e1000_get_hw_control(adapter);
3518                 e1000e_reset(adapter);
3519         }
3520
3521         e1000e_power_up_phy(adapter);
3522
3523         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3524         if ((adapter->hw.mng_cookie.status &
3525              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3526                 e1000_update_mng_vlan(adapter);
3527
3528         /* DMA latency requirement to workaround early-receive/jumbo issue */
3529         if (adapter->flags & FLAG_HAS_ERT)
3530                 adapter->netdev->pm_qos_req =
3531                                     pm_qos_add_request(PM_QOS_CPU_DMA_LATENCY,
3532                                                        PM_QOS_DEFAULT_VALUE);
3533
3534         /*
3535          * before we allocate an interrupt, we must be ready to handle it.
3536          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3537          * as soon as we call pci_request_irq, so we have to setup our
3538          * clean_rx handler before we do so.
3539          */
3540         e1000_configure(adapter);
3541
3542         err = e1000_request_irq(adapter);
3543         if (err)
3544                 goto err_req_irq;
3545
3546         /*
3547          * Work around PCIe errata with MSI interrupts causing some chipsets to
3548          * ignore e1000e MSI messages, which means we need to test our MSI
3549          * interrupt now
3550          */
3551         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3552                 err = e1000_test_msi(adapter);
3553                 if (err) {
3554                         e_err("Interrupt allocation failed\n");
3555                         goto err_req_irq;
3556                 }
3557         }
3558
3559         /* From here on the code is the same as e1000e_up() */
3560         clear_bit(__E1000_DOWN, &adapter->state);
3561
3562         napi_enable(&adapter->napi);
3563
3564         e1000_irq_enable(adapter);
3565
3566         netif_start_queue(netdev);
3567
3568         adapter->idle_check = true;
3569         pm_runtime_put(&pdev->dev);
3570
3571         /* fire a link status change interrupt to start the watchdog */
3572         if (adapter->msix_entries)
3573                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3574         else
3575                 ew32(ICS, E1000_ICS_LSC);
3576
3577         return 0;
3578
3579 err_req_irq:
3580         e1000_release_hw_control(adapter);
3581         e1000_power_down_phy(adapter);
3582         e1000e_free_rx_resources(adapter);
3583 err_setup_rx:
3584         e1000e_free_tx_resources(adapter);
3585 err_setup_tx:
3586         e1000e_reset(adapter);
3587         pm_runtime_put_sync(&pdev->dev);
3588
3589         return err;
3590 }
3591
3592 /**
3593  * e1000_close - Disables a network interface
3594  * @netdev: network interface device structure
3595  *
3596  * Returns 0, this is not allowed to fail
3597  *
3598  * The close entry point is called when an interface is de-activated
3599  * by the OS.  The hardware is still under the drivers control, but
3600  * needs to be disabled.  A global MAC reset is issued to stop the
3601  * hardware, and all transmit and receive resources are freed.
3602  **/
3603 static int e1000_close(struct net_device *netdev)
3604 {
3605         struct e1000_adapter *adapter = netdev_priv(netdev);
3606         struct pci_dev *pdev = adapter->pdev;
3607
3608         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3609
3610         pm_runtime_get_sync(&pdev->dev);
3611
3612         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3613                 e1000e_down(adapter);
3614                 e1000_free_irq(adapter);
3615         }
3616         e1000_power_down_phy(adapter);
3617
3618         e1000e_free_tx_resources(adapter);
3619         e1000e_free_rx_resources(adapter);
3620
3621         /*
3622          * kill manageability vlan ID if supported, but not if a vlan with
3623          * the same ID is registered on the host OS (let 8021q kill it)
3624          */
3625         if ((adapter->hw.mng_cookie.status &
3626                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3627              !(adapter->vlgrp &&
3628                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
3629                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3630
3631         /*
3632          * If AMT is enabled, let the firmware know that the network
3633          * interface is now closed
3634          */
3635         if (adapter->flags & FLAG_HAS_AMT)
3636                 e1000_release_hw_control(adapter);
3637
3638         if (adapter->flags & FLAG_HAS_ERT) {
3639                 pm_qos_remove_request(adapter->netdev->pm_qos_req);
3640                 adapter->netdev->pm_qos_req = NULL;
3641         }
3642
3643         pm_runtime_put_sync(&pdev->dev);
3644
3645         return 0;
3646 }
3647 /**
3648  * e1000_set_mac - Change the Ethernet Address of the NIC
3649  * @netdev: network interface device structure
3650  * @p: pointer to an address structure
3651  *
3652  * Returns 0 on success, negative on failure
3653  **/
3654 static int e1000_set_mac(struct net_device *netdev, void *p)
3655 {
3656         struct e1000_adapter *adapter = netdev_priv(netdev);
3657         struct sockaddr *addr = p;
3658
3659         if (!is_valid_ether_addr(addr->sa_data))
3660                 return -EADDRNOTAVAIL;
3661
3662         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3663         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3664
3665         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3666
3667         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3668                 /* activate the work around */
3669                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3670
3671                 /*
3672                  * Hold a copy of the LAA in RAR[14] This is done so that
3673                  * between the time RAR[0] gets clobbered  and the time it
3674                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3675                  * of the RARs and no incoming packets directed to this port
3676                  * are dropped. Eventually the LAA will be in RAR[0] and
3677                  * RAR[14]
3678                  */
3679                 e1000e_rar_set(&adapter->hw,
3680                               adapter->hw.mac.addr,
3681                               adapter->hw.mac.rar_entry_count - 1);
3682         }
3683
3684         return 0;
3685 }
3686
3687 /**
3688  * e1000e_update_phy_task - work thread to update phy
3689  * @work: pointer to our work struct
3690  *
3691  * this worker thread exists because we must acquire a
3692  * semaphore to read the phy, which we could msleep while
3693  * waiting for it, and we can't msleep in a timer.
3694  **/
3695 static void e1000e_update_phy_task(struct work_struct *work)
3696 {
3697         struct e1000_adapter *adapter = container_of(work,
3698                                         struct e1000_adapter, update_phy_task);
3699         e1000_get_phy_info(&adapter->hw);
3700 }
3701
3702 /*
3703  * Need to wait a few seconds after link up to get diagnostic information from
3704  * the phy
3705  */
3706 static void e1000_update_phy_info(unsigned long data)
3707 {
3708         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3709         schedule_work(&adapter->update_phy_task);
3710 }
3711
3712 /**
3713  * e1000e_update_phy_stats - Update the PHY statistics counters
3714  * @adapter: board private structure
3715  **/
3716 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3717 {
3718         struct e1000_hw *hw = &adapter->hw;
3719         s32 ret_val;
3720         u16 phy_data;
3721
3722         ret_val = hw->phy.ops.acquire(hw);
3723         if (ret_val)
3724                 return;
3725
3726         hw->phy.addr = 1;
3727
3728 #define HV_PHY_STATS_PAGE       778
3729         /*
3730          * A page set is expensive so check if already on desired page.
3731          * If not, set to the page with the PHY status registers.
3732          */
3733         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3734                                            &phy_data);
3735         if (ret_val)
3736                 goto release;
3737         if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3738                 ret_val = e1000e_write_phy_reg_mdic(hw,
3739                                                     IGP01E1000_PHY_PAGE_SELECT,
3740                                                     (HV_PHY_STATS_PAGE <<
3741                                                      IGP_PAGE_SHIFT));
3742                 if (ret_val)
3743                         goto release;
3744         }
3745
3746         /* Read/clear the upper 16-bit registers and read/accumulate lower */
3747
3748         /* Single Collision Count */
3749         e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3750                                  &phy_data);
3751         ret_val = e1000e_read_phy_reg_mdic(hw,
3752                                            HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3753                                            &phy_data);
3754         if (!ret_val)
3755                 adapter->stats.scc += phy_data;
3756
3757         /* Excessive Collision Count */
3758         e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3759                                  &phy_data);
3760         ret_val = e1000e_read_phy_reg_mdic(hw,
3761                                            HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3762                                            &phy_data);
3763         if (!ret_val)
3764                 adapter->stats.ecol += phy_data;
3765
3766         /* Multiple Collision Count */
3767         e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3768                                  &phy_data);
3769         ret_val = e1000e_read_phy_reg_mdic(hw,
3770                                            HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3771                                            &phy_data);
3772         if (!ret_val)
3773                 adapter->stats.mcc += phy_data;
3774
3775         /* Late Collision Count */
3776         e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3777                                  &phy_data);
3778         ret_val = e1000e_read_phy_reg_mdic(hw,
3779                                            HV_LATECOL_LOWER &
3780                                            MAX_PHY_REG_ADDRESS,
3781                                            &phy_data);
3782         if (!ret_val)
3783                 adapter->stats.latecol += phy_data;
3784
3785         /* Collision Count - also used for adaptive IFS */
3786         e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3787                                  &phy_data);
3788         ret_val = e1000e_read_phy_reg_mdic(hw,
3789                                            HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3790                                            &phy_data);
3791         if (!ret_val)
3792                 hw->mac.collision_delta = phy_data;
3793
3794         /* Defer Count */
3795         e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3796                                  &phy_data);
3797         ret_val = e1000e_read_phy_reg_mdic(hw,
3798                                            HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3799                                            &phy_data);
3800         if (!ret_val)
3801                 adapter->stats.dc += phy_data;
3802
3803         /* Transmit with no CRS */
3804         e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3805                                  &phy_data);
3806         ret_val = e1000e_read_phy_reg_mdic(hw,
3807                                            HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3808                                            &phy_data);
3809         if (!ret_val)
3810                 adapter->stats.tncrs += phy_data;
3811
3812 release:
3813         hw->phy.ops.release(hw);
3814 }
3815
3816 /**
3817  * e1000e_update_stats - Update the board statistics counters
3818  * @adapter: board private structure
3819  **/
3820 void e1000e_update_stats(struct e1000_adapter *adapter)
3821 {
3822         struct net_device *netdev = adapter->netdev;
3823         struct e1000_hw *hw = &adapter->hw;
3824         struct pci_dev *pdev = adapter->pdev;
3825
3826         /*
3827          * Prevent stats update while adapter is being reset, or if the pci
3828          * connection is down.
3829          */
3830         if (adapter->link_speed == 0)
3831                 return;
3832         if (pci_channel_offline(pdev))
3833                 return;
3834
3835         adapter->stats.crcerrs += er32(CRCERRS);
3836         adapter->stats.gprc += er32(GPRC);
3837         adapter->stats.gorc += er32(GORCL);
3838         er32(GORCH); /* Clear gorc */
3839         adapter->stats.bprc += er32(BPRC);
3840         adapter->stats.mprc += er32(MPRC);
3841         adapter->stats.roc += er32(ROC);
3842
3843         adapter->stats.mpc += er32(MPC);
3844
3845         /* Half-duplex statistics */
3846         if (adapter->link_duplex == HALF_DUPLEX) {
3847                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3848                         e1000e_update_phy_stats(adapter);
3849                 } else {
3850                         adapter->stats.scc += er32(SCC);
3851                         adapter->stats.ecol += er32(ECOL);
3852                         adapter->stats.mcc += er32(MCC);
3853                         adapter->stats.latecol += er32(LATECOL);
3854                         adapter->stats.dc += er32(DC);
3855
3856                         hw->mac.collision_delta = er32(COLC);
3857
3858                         if ((hw->mac.type != e1000_82574) &&
3859                             (hw->mac.type != e1000_82583))
3860                                 adapter->stats.tncrs += er32(TNCRS);
3861                 }
3862                 adapter->stats.colc += hw->mac.collision_delta;
3863         }
3864
3865         adapter->stats.xonrxc += er32(XONRXC);
3866         adapter->stats.xontxc += er32(XONTXC);
3867         adapter->stats.xoffrxc += er32(XOFFRXC);
3868         adapter->stats.xofftxc += er32(XOFFTXC);
3869         adapter->stats.gptc += er32(GPTC);
3870         adapter->stats.gotc += er32(GOTCL);
3871         er32(GOTCH); /* Clear gotc */
3872         adapter->stats.rnbc += er32(RNBC);
3873         adapter->stats.ruc += er32(RUC);
3874
3875         adapter->stats.mptc += er32(MPTC);
3876         adapter->stats.bptc += er32(BPTC);
3877
3878         /* used for adaptive IFS */
3879
3880         hw->mac.tx_packet_delta = er32(TPT);
3881         adapter->stats.tpt += hw->mac.tx_packet_delta;
3882
3883         adapter->stats.algnerrc += er32(ALGNERRC);
3884         adapter->stats.rxerrc += er32(RXERRC);
3885         adapter->stats.cexterr += er32(CEXTERR);
3886         adapter->stats.tsctc += er32(TSCTC);
3887         adapter->stats.tsctfc += er32(TSCTFC);
3888
3889         /* Fill out the OS statistics structure */
3890         netdev->stats.multicast = adapter->stats.mprc;
3891         netdev->stats.collisions = adapter->stats.colc;
3892
3893         /* Rx Errors */
3894
3895         /*
3896          * RLEC on some newer hardware can be incorrect so build
3897          * our own version based on RUC and ROC
3898          */
3899         netdev->stats.rx_errors = adapter->stats.rxerrc +
3900                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3901                 adapter->stats.ruc + adapter->stats.roc +
3902                 adapter->stats.cexterr;
3903         netdev->stats.rx_length_errors = adapter->stats.ruc +
3904                                               adapter->stats.roc;
3905         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3906         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3907         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3908
3909         /* Tx Errors */
3910         netdev->stats.tx_errors = adapter->stats.ecol +
3911                                        adapter->stats.latecol;
3912         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3913         netdev->stats.tx_window_errors = adapter->stats.latecol;
3914         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3915
3916         /* Tx Dropped needs to be maintained elsewhere */
3917
3918         /* Management Stats */
3919         adapter->stats.mgptc += er32(MGTPTC);
3920         adapter->stats.mgprc += er32(MGTPRC);
3921         adapter->stats.mgpdc += er32(MGTPDC);
3922 }
3923
3924 /**
3925  * e1000_phy_read_status - Update the PHY register status snapshot
3926  * @adapter: board private structure
3927  **/
3928 static void e1000_phy_read_status(struct e1000_adapter *adapter)
3929 {
3930         struct e1000_hw *hw = &adapter->hw;
3931         struct e1000_phy_regs *phy = &adapter->phy_regs;
3932         int ret_val;
3933
3934         if ((er32(STATUS) & E1000_STATUS_LU) &&
3935             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
3936                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
3937                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
3938                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
3939                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
3940                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
3941                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
3942                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
3943                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
3944                 if (ret_val)
3945                         e_warn("Error reading PHY register\n");
3946         } else {
3947                 /*
3948                  * Do not read PHY registers if link is not up
3949                  * Set values to typical power-on defaults
3950                  */
3951                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
3952                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
3953                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
3954                              BMSR_ERCAP);
3955                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
3956                                   ADVERTISE_ALL | ADVERTISE_CSMA);
3957                 phy->lpa = 0;
3958                 phy->expansion = EXPANSION_ENABLENPAGE;
3959                 phy->ctrl1000 = ADVERTISE_1000FULL;
3960                 phy->stat1000 = 0;
3961                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
3962         }
3963 }
3964
3965 static void e1000_print_link_info(struct e1000_adapter *adapter)
3966 {
3967         struct e1000_hw *hw = &adapter->hw;
3968         u32 ctrl = er32(CTRL);
3969
3970         /* Link status message must follow this format for user tools */
3971         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
3972                "Flow Control: %s\n",
3973                adapter->netdev->name,
3974                adapter->link_speed,
3975                (adapter->link_duplex == FULL_DUPLEX) ?
3976                                 "Full Duplex" : "Half Duplex",
3977                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
3978                                 "RX/TX" :
3979                ((ctrl & E1000_CTRL_RFCE) ? "RX" :
3980                ((ctrl & E1000_CTRL_TFCE) ? "TX" : "None" )));
3981 }
3982
3983 static bool e1000e_has_link(struct e1000_adapter *adapter)
3984 {
3985         struct e1000_hw *hw = &adapter->hw;
3986         bool link_active = 0;
3987         s32 ret_val = 0;
3988
3989         /*
3990          * get_link_status is set on LSC (link status) interrupt or
3991          * Rx sequence error interrupt.  get_link_status will stay
3992          * false until the check_for_link establishes link
3993          * for copper adapters ONLY
3994          */
3995         switch (hw->phy.media_type) {
3996         case e1000_media_type_copper:
3997                 if (hw->mac.get_link_status) {
3998                         ret_val = hw->mac.ops.check_for_link(hw);
3999                         link_active = !hw->mac.get_link_status;
4000                 } else {
4001                         link_active = 1;
4002                 }
4003                 break;
4004         case e1000_media_type_fiber:
4005                 ret_val = hw->mac.ops.check_for_link(hw);
4006                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4007                 break;
4008         case e1000_media_type_internal_serdes:
4009                 ret_val = hw->mac.ops.check_for_link(hw);
4010                 link_active = adapter->hw.mac.serdes_has_link;
4011                 break;
4012         default:
4013         case e1000_media_type_unknown:
4014                 break;
4015         }
4016
4017         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4018             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4019                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4020                 e_info("Gigabit has been disabled, downgrading speed\n");
4021         }
4022
4023         return link_active;
4024 }
4025
4026 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4027 {
4028         /* make sure the receive unit is started */
4029         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4030             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4031                 struct e1000_hw *hw = &adapter->hw;
4032                 u32 rctl = er32(RCTL);
4033                 ew32(RCTL, rctl | E1000_RCTL_EN);
4034                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4035         }
4036 }
4037
4038 /**
4039  * e1000_watchdog - Timer Call-back
4040  * @data: pointer to adapter cast into an unsigned long
4041  **/
4042 static void e1000_watchdog(unsigned long data)
4043 {
4044         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4045
4046         /* Do the rest outside of interrupt context */
4047         schedule_work(&adapter->watchdog_task);
4048
4049         /* TODO: make this use queue_delayed_work() */
4050 }
4051
4052 static void e1000_watchdog_task(struct work_struct *work)
4053 {
4054         struct e1000_adapter *adapter = container_of(work,
4055                                         struct e1000_adapter, watchdog_task);
4056         struct net_device *netdev = adapter->netdev;
4057         struct e1000_mac_info *mac = &adapter->hw.mac;
4058         struct e1000_phy_info *phy = &adapter->hw.phy;
4059         struct e1000_ring *tx_ring = adapter->tx_ring;
4060         struct e1000_hw *hw = &adapter->hw;
4061         u32 link, tctl;
4062         int tx_pending = 0;
4063
4064         link = e1000e_has_link(adapter);
4065         if ((netif_carrier_ok(netdev)) && link) {
4066                 /* Cancel scheduled suspend requests. */
4067                 pm_runtime_resume(netdev->dev.parent);
4068
4069                 e1000e_enable_receives(adapter);
4070                 goto link_up;
4071         }
4072
4073         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4074             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4075                 e1000_update_mng_vlan(adapter);
4076
4077         if (link) {
4078                 if (!netif_carrier_ok(netdev)) {
4079                         bool txb2b = 1;
4080
4081                         /* Cancel scheduled suspend requests. */
4082                         pm_runtime_resume(netdev->dev.parent);
4083
4084                         /* update snapshot of PHY registers on LSC */
4085                         e1000_phy_read_status(adapter);
4086                         mac->ops.get_link_up_info(&adapter->hw,
4087                                                    &adapter->link_speed,
4088                                                    &adapter->link_duplex);
4089                         e1000_print_link_info(adapter);
4090                         /*
4091                          * On supported PHYs, check for duplex mismatch only
4092                          * if link has autonegotiated at 10/100 half
4093                          */
4094                         if ((hw->phy.type == e1000_phy_igp_3 ||
4095                              hw->phy.type == e1000_phy_bm) &&
4096                             (hw->mac.autoneg == true) &&
4097                             (adapter->link_speed == SPEED_10 ||
4098                              adapter->link_speed == SPEED_100) &&
4099                             (adapter->link_duplex == HALF_DUPLEX)) {
4100                                 u16 autoneg_exp;
4101
4102                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4103
4104                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4105                                         e_info("Autonegotiated half duplex but"
4106                                                " link partner cannot autoneg. "
4107                                                " Try forcing full duplex if "
4108                                                "link gets many collisions.\n");
4109                         }
4110
4111                         /* adjust timeout factor according to speed/duplex */
4112                         adapter->tx_timeout_factor = 1;
4113                         switch (adapter->link_speed) {
4114                         case SPEED_10:
4115                                 txb2b = 0;
4116                                 adapter->tx_timeout_factor = 16;
4117                                 break;
4118                         case SPEED_100:
4119                                 txb2b = 0;
4120                                 adapter->tx_timeout_factor = 10;
4121                                 break;
4122                         }
4123
4124                         /*
4125                          * workaround: re-program speed mode bit after
4126                          * link-up event
4127                          */
4128                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4129                             !txb2b) {
4130                                 u32 tarc0;
4131                                 tarc0 = er32(TARC(0));
4132                                 tarc0 &= ~SPEED_MODE_BIT;
4133                                 ew32(TARC(0), tarc0);
4134                         }
4135
4136                         /*
4137                          * disable TSO for pcie and 10/100 speeds, to avoid
4138                          * some hardware issues
4139                          */
4140                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4141                                 switch (adapter->link_speed) {
4142                                 case SPEED_10:
4143                                 case SPEED_100:
4144                                         e_info("10/100 speed: disabling TSO\n");
4145                                         netdev->features &= ~NETIF_F_TSO;
4146                                         netdev->features &= ~NETIF_F_TSO6;
4147                                         break;
4148                                 case SPEED_1000:
4149                                         netdev->features |= NETIF_F_TSO;
4150                                         netdev->features |= NETIF_F_TSO6;
4151                                         break;
4152                                 default:
4153                                         /* oops */
4154                                         break;
4155                                 }
4156                         }
4157
4158                         /*
4159                          * enable transmits in the hardware, need to do this
4160                          * after setting TARC(0)
4161                          */
4162                         tctl = er32(TCTL);
4163                         tctl |= E1000_TCTL_EN;
4164                         ew32(TCTL, tctl);
4165
4166                         /*
4167                          * Perform any post-link-up configuration before
4168                          * reporting link up.
4169                          */
4170                         if (phy->ops.cfg_on_link_up)
4171                                 phy->ops.cfg_on_link_up(hw);
4172
4173                         netif_carrier_on(netdev);
4174
4175                         if (!test_bit(__E1000_DOWN, &adapter->state))
4176                                 mod_timer(&adapter->phy_info_timer,
4177                                           round_jiffies(jiffies + 2 * HZ));
4178                 }
4179         } else {
4180                 if (netif_carrier_ok(netdev)) {
4181                         adapter->link_speed = 0;
4182                         adapter->link_duplex = 0;
4183                         /* Link status message must follow this format */
4184                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4185                                adapter->netdev->name);
4186                         netif_carrier_off(netdev);
4187                         if (!test_bit(__E1000_DOWN, &adapter->state))
4188                                 mod_timer(&adapter->phy_info_timer,
4189                                           round_jiffies(jiffies + 2 * HZ));
4190
4191                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4192                                 schedule_work(&adapter->reset_task);
4193                         else
4194                                 pm_schedule_suspend(netdev->dev.parent,
4195                                                         LINK_TIMEOUT);
4196                 }
4197         }
4198
4199 link_up:
4200         e1000e_update_stats(adapter);
4201
4202         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4203         adapter->tpt_old = adapter->stats.tpt;
4204         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4205         adapter->colc_old = adapter->stats.colc;
4206
4207         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4208         adapter->gorc_old = adapter->stats.gorc;
4209         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4210         adapter->gotc_old = adapter->stats.gotc;
4211
4212         e1000e_update_adaptive(&adapter->hw);
4213
4214         if (!netif_carrier_ok(netdev)) {
4215                 tx_pending = (e1000_desc_unused(tx_ring) + 1 <
4216                                tx_ring->count);
4217                 if (tx_pending) {
4218                         /*
4219                          * We've lost link, so the controller stops DMA,
4220                          * but we've got queued Tx work that's never going
4221                          * to get done, so reset controller to flush Tx.
4222                          * (Do the reset outside of interrupt context).
4223                          */
4224                         adapter->tx_timeout_count++;
4225                         schedule_work(&adapter->reset_task);
4226                         /* return immediately since reset is imminent */
4227                         return;
4228                 }
4229         }
4230
4231         /* Simple mode for Interrupt Throttle Rate (ITR) */
4232         if (adapter->itr_setting == 4) {
4233                 /*
4234                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4235                  * Total asymmetrical Tx or Rx gets ITR=8000;
4236                  * everyone else is between 2000-8000.
4237                  */
4238                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4239                 u32 dif = (adapter->gotc > adapter->gorc ?
4240                             adapter->gotc - adapter->gorc :
4241                             adapter->gorc - adapter->gotc) / 10000;
4242                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4243
4244                 ew32(ITR, 1000000000 / (itr * 256));
4245         }
4246
4247         /* Cause software interrupt to ensure Rx ring is cleaned */
4248         if (adapter->msix_entries)
4249                 ew32(ICS, adapter->rx_ring->ims_val);
4250         else
4251                 ew32(ICS, E1000_ICS_RXDMT0);
4252
4253         /* Force detection of hung controller every watchdog period */
4254         adapter->detect_tx_hung = 1;
4255
4256         /*
4257          * With 82571 controllers, LAA may be overwritten due to controller
4258          * reset from the other port. Set the appropriate LAA in RAR[0]
4259          */
4260         if (e1000e_get_laa_state_82571(hw))
4261                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4262
4263         /* Reset the timer */
4264         if (!test_bit(__E1000_DOWN, &adapter->state))
4265                 mod_timer(&adapter->watchdog_timer,
4266                           round_jiffies(jiffies + 2 * HZ));
4267 }
4268
4269 #define E1000_TX_FLAGS_CSUM             0x00000001
4270 #define E1000_TX_FLAGS_VLAN             0x00000002
4271 #define E1000_TX_FLAGS_TSO              0x00000004
4272 #define E1000_TX_FLAGS_IPV4             0x00000008
4273 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4274 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4275
4276 static int e1000_tso(struct e1000_adapter *adapter,
4277                      struct sk_buff *skb)
4278 {
4279         struct e1000_ring *tx_ring = adapter->tx_ring;
4280         struct e1000_context_desc *context_desc;
4281         struct e1000_buffer *buffer_info;
4282         unsigned int i;
4283         u32 cmd_length = 0;
4284         u16 ipcse = 0, tucse, mss;
4285         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4286         int err;
4287
4288         if (!skb_is_gso(skb))
4289                 return 0;
4290
4291         if (skb_header_cloned(skb)) {
4292                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4293                 if (err)
4294                         return err;
4295         }
4296
4297         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4298         mss = skb_shinfo(skb)->gso_size;
4299         if (skb->protocol == htons(ETH_P_IP)) {
4300                 struct iphdr *iph = ip_hdr(skb);
4301                 iph->tot_len = 0;
4302                 iph->check = 0;
4303                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4304                                                          0, IPPROTO_TCP, 0);
4305                 cmd_length = E1000_TXD_CMD_IP;
4306                 ipcse = skb_transport_offset(skb) - 1;
4307         } else if (skb_is_gso_v6(skb)) {
4308                 ipv6_hdr(skb)->payload_len = 0;
4309                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4310                                                        &ipv6_hdr(skb)->daddr,
4311                                                        0, IPPROTO_TCP, 0);
4312                 ipcse = 0;
4313         }
4314         ipcss = skb_network_offset(skb);
4315         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4316         tucss = skb_transport_offset(skb);
4317         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4318         tucse = 0;
4319
4320         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4321                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4322
4323         i = tx_ring->next_to_use;
4324         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4325         buffer_info = &tx_ring->buffer_info[i];
4326
4327         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4328         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4329         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4330         context_desc->upper_setup.tcp_fields.tucss = tucss;
4331         context_desc->upper_setup.tcp_fields.tucso = tucso;
4332         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4333         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4334         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4335         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4336
4337         buffer_info->time_stamp = jiffies;
4338         buffer_info->next_to_watch = i;
4339
4340         i++;
4341         if (i == tx_ring->count)
4342                 i = 0;
4343         tx_ring->next_to_use = i;
4344
4345         return 1;
4346 }
4347
4348 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4349 {
4350         struct e1000_ring *tx_ring = adapter->tx_ring;
4351         struct e1000_context_desc *context_desc;
4352         struct e1000_buffer *buffer_info;
4353         unsigned int i;
4354         u8 css;
4355         u32 cmd_len = E1000_TXD_CMD_DEXT;
4356         __be16 protocol;
4357
4358         if (skb->ip_summed != CHECKSUM_PARTIAL)
4359                 return 0;
4360
4361         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4362                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4363         else
4364                 protocol = skb->protocol;
4365
4366         switch (protocol) {
4367         case cpu_to_be16(ETH_P_IP):
4368                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4369                         cmd_len |= E1000_TXD_CMD_TCP;
4370                 break;
4371         case cpu_to_be16(ETH_P_IPV6):
4372                 /* XXX not handling all IPV6 headers */
4373                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4374                         cmd_len |= E1000_TXD_CMD_TCP;
4375                 break;
4376         default:
4377                 if (unlikely(net_ratelimit()))
4378                         e_warn("checksum_partial proto=%x!\n",
4379                                be16_to_cpu(protocol));
4380                 break;
4381         }
4382
4383         css = skb_transport_offset(skb);
4384
4385         i = tx_ring->next_to_use;
4386         buffer_info = &tx_ring->buffer_info[i];
4387         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4388
4389         context_desc->lower_setup.ip_config = 0;
4390         context_desc->upper_setup.tcp_fields.tucss = css;
4391         context_desc->upper_setup.tcp_fields.tucso =
4392                                 css + skb->csum_offset;
4393         context_desc->upper_setup.tcp_fields.tucse = 0;
4394         context_desc->tcp_seg_setup.data = 0;
4395         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4396
4397         buffer_info->time_stamp = jiffies;
4398         buffer_info->next_to_watch = i;
4399
4400         i++;
4401         if (i == tx_ring->count)
4402                 i = 0;
4403         tx_ring->next_to_use = i;
4404
4405         return 1;
4406 }
4407
4408 #define E1000_MAX_PER_TXD       8192
4409 #define E1000_MAX_TXD_PWR       12
4410
4411 static int e1000_tx_map(struct e1000_adapter *adapter,
4412                         struct sk_buff *skb, unsigned int first,
4413                         unsigned int max_per_txd, unsigned int nr_frags,
4414                         unsigned int mss)
4415 {
4416         struct e1000_ring *tx_ring = adapter->tx_ring;
4417         struct pci_dev *pdev = adapter->pdev;
4418         struct e1000_buffer *buffer_info;
4419         unsigned int len = skb_headlen(skb);
4420         unsigned int offset = 0, size, count = 0, i;
4421         unsigned int f, bytecount, segs;
4422
4423         i = tx_ring->next_to_use;
4424
4425         while (len) {
4426                 buffer_info = &tx_ring->buffer_info[i];
4427                 size = min(len, max_per_txd);
4428
4429                 buffer_info->length = size;
4430                 buffer_info->time_stamp = jiffies;
4431                 buffer_info->next_to_watch = i;
4432                 buffer_info->dma = dma_map_single(&pdev->dev,
4433                                                   skb->data + offset,
4434                                                   size, DMA_TO_DEVICE);
4435                 buffer_info->mapped_as_page = false;
4436                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4437                         goto dma_error;
4438
4439                 len -= size;
4440                 offset += size;
4441                 count++;
4442
4443                 if (len) {
4444                         i++;
4445                         if (i == tx_ring->count)
4446                                 i = 0;
4447                 }
4448         }
4449
4450         for (f = 0; f < nr_frags; f++) {
4451                 struct skb_frag_struct *frag;
4452
4453                 frag = &skb_shinfo(skb)->frags[f];
4454                 len = frag->size;
4455                 offset = frag->page_offset;
4456
4457                 while (len) {
4458                         i++;
4459                         if (i == tx_ring->count)
4460                                 i = 0;
4461
4462                         buffer_info = &tx_ring->buffer_info[i];
4463                         size = min(len, max_per_txd);
4464
4465                         buffer_info->length = size;
4466                         buffer_info->time_stamp = jiffies;
4467                         buffer_info->next_to_watch = i;
4468                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4469                                                         offset, size,
4470                                                         DMA_TO_DEVICE);
4471                         buffer_info->mapped_as_page = true;
4472                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4473                                 goto dma_error;
4474
4475                         len -= size;
4476                         offset += size;
4477                         count++;
4478                 }
4479         }
4480
4481         segs = skb_shinfo(skb)->gso_segs ?: 1;
4482         /* multiply data chunks by size of headers */
4483         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4484
4485         tx_ring->buffer_info[i].skb = skb;
4486         tx_ring->buffer_info[i].segs = segs;
4487         tx_ring->buffer_info[i].bytecount = bytecount;
4488         tx_ring->buffer_info[first].next_to_watch = i;
4489
4490         return count;
4491
4492 dma_error:
4493         dev_err(&pdev->dev, "TX DMA map failed\n");
4494         buffer_info->dma = 0;
4495         if (count)
4496                 count--;
4497
4498         while (count--) {
4499                 if (i==0)
4500                         i += tx_ring->count;
4501                 i--;
4502                 buffer_info = &tx_ring->buffer_info[i];
4503                 e1000_put_txbuf(adapter, buffer_info);;
4504         }
4505
4506         return 0;
4507 }
4508
4509 static void e1000_tx_queue(struct e1000_adapter *adapter,
4510                            int tx_flags, int count)
4511 {
4512         struct e1000_ring *tx_ring = adapter->tx_ring;
4513         struct e1000_tx_desc *tx_desc = NULL;
4514         struct e1000_buffer *buffer_info;
4515         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4516         unsigned int i;
4517
4518         if (tx_flags & E1000_TX_FLAGS_TSO) {
4519                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4520                              E1000_TXD_CMD_TSE;
4521                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4522
4523                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4524                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4525         }
4526
4527         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4528                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4529                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4530         }
4531
4532         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4533                 txd_lower |= E1000_TXD_CMD_VLE;
4534                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4535         }
4536
4537         i = tx_ring->next_to_use;
4538
4539         while (count--) {
4540                 buffer_info = &tx_ring->buffer_info[i];
4541                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4542                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4543                 tx_desc->lower.data =
4544                         cpu_to_le32(txd_lower | buffer_info->length);
4545                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4546
4547                 i++;
4548                 if (i == tx_ring->count)
4549                         i = 0;
4550         }
4551
4552         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4553
4554         /*
4555          * Force memory writes to complete before letting h/w
4556          * know there are new descriptors to fetch.  (Only
4557          * applicable for weak-ordered memory model archs,
4558          * such as IA-64).
4559          */
4560         wmb();
4561
4562         tx_ring->next_to_use = i;
4563         writel(i, adapter->hw.hw_addr + tx_ring->tail);
4564         /*
4565          * we need this if more than one processor can write to our tail
4566          * at a time, it synchronizes IO on IA64/Altix systems
4567          */
4568         mmiowb();
4569 }
4570
4571 #define MINIMUM_DHCP_PACKET_SIZE 282
4572 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4573                                     struct sk_buff *skb)
4574 {
4575         struct e1000_hw *hw =  &adapter->hw;
4576         u16 length, offset;
4577
4578         if (vlan_tx_tag_present(skb)) {
4579                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4580                     (adapter->hw.mng_cookie.status &
4581                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4582                         return 0;
4583         }
4584
4585         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4586                 return 0;
4587
4588         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4589                 return 0;
4590
4591         {
4592                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4593                 struct udphdr *udp;
4594
4595                 if (ip->protocol != IPPROTO_UDP)
4596                         return 0;
4597
4598                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4599                 if (ntohs(udp->dest) != 67)
4600                         return 0;
4601
4602                 offset = (u8 *)udp + 8 - skb->data;
4603                 length = skb->len - offset;
4604                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4605         }
4606
4607         return 0;
4608 }
4609
4610 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4611 {
4612         struct e1000_adapter *adapter = netdev_priv(netdev);
4613
4614         netif_stop_queue(netdev);
4615         /*
4616          * Herbert's original patch had:
4617          *  smp_mb__after_netif_stop_queue();
4618          * but since that doesn't exist yet, just open code it.
4619          */
4620         smp_mb();
4621
4622         /*
4623          * We need to check again in a case another CPU has just
4624          * made room available.
4625          */
4626         if (e1000_desc_unused(adapter->tx_ring) < size)
4627                 return -EBUSY;
4628
4629         /* A reprieve! */
4630         netif_start_queue(netdev);
4631         ++adapter->restart_queue;
4632         return 0;
4633 }
4634
4635 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4636 {
4637         struct e1000_adapter *adapter = netdev_priv(netdev);
4638
4639         if (e1000_desc_unused(adapter->tx_ring) >= size)
4640                 return 0;
4641         return __e1000_maybe_stop_tx(netdev, size);
4642 }
4643
4644 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4645 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4646                                     struct net_device *netdev)
4647 {
4648         struct e1000_adapter *adapter = netdev_priv(netdev);
4649         struct e1000_ring *tx_ring = adapter->tx_ring;
4650         unsigned int first;
4651         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4652         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4653         unsigned int tx_flags = 0;
4654         unsigned int len = skb_headlen(skb);
4655         unsigned int nr_frags;
4656         unsigned int mss;
4657         int count = 0;
4658         int tso;
4659         unsigned int f;
4660
4661         if (test_bit(__E1000_DOWN, &adapter->state)) {
4662                 dev_kfree_skb_any(skb);
4663                 return NETDEV_TX_OK;
4664         }
4665
4666         if (skb->len <= 0) {
4667                 dev_kfree_skb_any(skb);
4668                 return NETDEV_TX_OK;
4669         }
4670
4671         mss = skb_shinfo(skb)->gso_size;
4672         /*
4673          * The controller does a simple calculation to
4674          * make sure there is enough room in the FIFO before
4675          * initiating the DMA for each buffer.  The calc is:
4676          * 4 = ceil(buffer len/mss).  To make sure we don't
4677          * overrun the FIFO, adjust the max buffer len if mss
4678          * drops.
4679          */
4680         if (mss) {
4681                 u8 hdr_len;
4682                 max_per_txd = min(mss << 2, max_per_txd);
4683                 max_txd_pwr = fls(max_per_txd) - 1;
4684
4685                 /*
4686                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4687                  * points to just header, pull a few bytes of payload from
4688                  * frags into skb->data
4689                  */
4690                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4691                 /*
4692                  * we do this workaround for ES2LAN, but it is un-necessary,
4693                  * avoiding it could save a lot of cycles
4694                  */
4695                 if (skb->data_len && (hdr_len == len)) {
4696                         unsigned int pull_size;
4697
4698                         pull_size = min((unsigned int)4, skb->data_len);
4699                         if (!__pskb_pull_tail(skb, pull_size)) {
4700                                 e_err("__pskb_pull_tail failed.\n");
4701                                 dev_kfree_skb_any(skb);
4702                                 return NETDEV_TX_OK;
4703                         }
4704                         len = skb_headlen(skb);
4705                 }
4706         }
4707
4708         /* reserve a descriptor for the offload context */
4709         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4710                 count++;
4711         count++;
4712
4713         count += TXD_USE_COUNT(len, max_txd_pwr);
4714
4715         nr_frags = skb_shinfo(skb)->nr_frags;
4716         for (f = 0; f < nr_frags; f++)
4717                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4718                                        max_txd_pwr);
4719
4720         if (adapter->hw.mac.tx_pkt_filtering)
4721                 e1000_transfer_dhcp_info(adapter, skb);
4722
4723         /*
4724          * need: count + 2 desc gap to keep tail from touching
4725          * head, otherwise try next time
4726          */
4727         if (e1000_maybe_stop_tx(netdev, count + 2))
4728                 return NETDEV_TX_BUSY;
4729
4730         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
4731                 tx_flags |= E1000_TX_FLAGS_VLAN;
4732                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4733         }
4734
4735         first = tx_ring->next_to_use;
4736
4737         tso = e1000_tso(adapter, skb);
4738         if (tso < 0) {
4739                 dev_kfree_skb_any(skb);
4740                 return NETDEV_TX_OK;
4741         }
4742
4743         if (tso)
4744                 tx_flags |= E1000_TX_FLAGS_TSO;
4745         else if (e1000_tx_csum(adapter, skb))
4746                 tx_flags |= E1000_TX_FLAGS_CSUM;
4747
4748         /*
4749          * Old method was to assume IPv4 packet by default if TSO was enabled.
4750          * 82571 hardware supports TSO capabilities for IPv6 as well...
4751          * no longer assume, we must.
4752          */
4753         if (skb->protocol == htons(ETH_P_IP))
4754                 tx_flags |= E1000_TX_FLAGS_IPV4;
4755
4756         /* if count is 0 then mapping error has occured */
4757         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4758         if (count) {
4759                 e1000_tx_queue(adapter, tx_flags, count);
4760                 /* Make sure there is space in the ring for the next send. */
4761                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4762
4763         } else {
4764                 dev_kfree_skb_any(skb);
4765                 tx_ring->buffer_info[first].time_stamp = 0;
4766                 tx_ring->next_to_use = first;
4767         }
4768
4769         return NETDEV_TX_OK;
4770 }
4771
4772 /**
4773  * e1000_tx_timeout - Respond to a Tx Hang
4774  * @netdev: network interface device structure
4775  **/
4776 static void e1000_tx_timeout(struct net_device *netdev)
4777 {
4778         struct e1000_adapter *adapter = netdev_priv(netdev);
4779
4780         /* Do the reset outside of interrupt context */
4781         adapter->tx_timeout_count++;
4782         schedule_work(&adapter->reset_task);
4783 }
4784
4785 static void e1000_reset_task(struct work_struct *work)
4786 {
4787         struct e1000_adapter *adapter;
4788         adapter = container_of(work, struct e1000_adapter, reset_task);
4789
4790         e1000e_dump(adapter);
4791         e_err("Reset adapter\n");
4792         e1000e_reinit_locked(adapter);
4793 }
4794
4795 /**
4796  * e1000_get_stats - Get System Network Statistics
4797  * @netdev: network interface device structure
4798  *
4799  * Returns the address of the device statistics structure.
4800  * The statistics are actually updated from the timer callback.
4801  **/
4802 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
4803 {
4804         /* only return the current stats */
4805         return &netdev->stats;
4806 }
4807
4808 /**
4809  * e1000_change_mtu - Change the Maximum Transfer Unit
4810  * @netdev: network interface device structure
4811  * @new_mtu: new value for maximum frame size
4812  *
4813  * Returns 0 on success, negative on failure
4814  **/
4815 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
4816 {
4817         struct e1000_adapter *adapter = netdev_priv(netdev);
4818         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
4819
4820         /* Jumbo frame support */
4821         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
4822             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
4823                 e_err("Jumbo Frames not supported.\n");
4824                 return -EINVAL;
4825         }
4826
4827         /* Supported frame sizes */
4828         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
4829             (max_frame > adapter->max_hw_frame_size)) {
4830                 e_err("Unsupported MTU setting\n");
4831                 return -EINVAL;
4832         }
4833
4834         /* 82573 Errata 17 */
4835         if (((adapter->hw.mac.type == e1000_82573) ||
4836              (adapter->hw.mac.type == e1000_82574)) &&
4837             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
4838                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
4839                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
4840         }
4841
4842         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
4843                 msleep(1);
4844         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
4845         adapter->max_frame_size = max_frame;
4846         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4847         netdev->mtu = new_mtu;
4848         if (netif_running(netdev))
4849                 e1000e_down(adapter);
4850
4851         /*
4852          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
4853          * means we reserve 2 more, this pushes us to allocate from the next
4854          * larger slab size.
4855          * i.e. RXBUFFER_2048 --> size-4096 slab
4856          * However with the new *_jumbo_rx* routines, jumbo receives will use
4857          * fragmented skbs
4858          */
4859
4860         if (max_frame <= 2048)
4861                 adapter->rx_buffer_len = 2048;
4862         else
4863                 adapter->rx_buffer_len = 4096;
4864
4865         /* adjust allocation if LPE protects us, and we aren't using SBP */
4866         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
4867              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
4868                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
4869                                          + ETH_FCS_LEN;
4870
4871         if (netif_running(netdev))
4872                 e1000e_up(adapter);
4873         else
4874                 e1000e_reset(adapter);
4875
4876         clear_bit(__E1000_RESETTING, &adapter->state);
4877
4878         return 0;
4879 }
4880
4881 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4882                            int cmd)
4883 {
4884         struct e1000_adapter *adapter = netdev_priv(netdev);
4885         struct mii_ioctl_data *data = if_mii(ifr);
4886
4887         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4888                 return -EOPNOTSUPP;
4889
4890         switch (cmd) {
4891         case SIOCGMIIPHY:
4892                 data->phy_id = adapter->hw.phy.addr;
4893                 break;
4894         case SIOCGMIIREG:
4895                 e1000_phy_read_status(adapter);
4896
4897                 switch (data->reg_num & 0x1F) {
4898                 case MII_BMCR:
4899                         data->val_out = adapter->phy_regs.bmcr;
4900                         break;
4901                 case MII_BMSR:
4902                         data->val_out = adapter->phy_regs.bmsr;
4903                         break;
4904                 case MII_PHYSID1:
4905                         data->val_out = (adapter->hw.phy.id >> 16);
4906                         break;
4907                 case MII_PHYSID2:
4908                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
4909                         break;
4910                 case MII_ADVERTISE:
4911                         data->val_out = adapter->phy_regs.advertise;
4912                         break;
4913                 case MII_LPA:
4914                         data->val_out = adapter->phy_regs.lpa;
4915                         break;
4916                 case MII_EXPANSION:
4917                         data->val_out = adapter->phy_regs.expansion;
4918                         break;
4919                 case MII_CTRL1000:
4920                         data->val_out = adapter->phy_regs.ctrl1000;
4921                         break;
4922                 case MII_STAT1000:
4923                         data->val_out = adapter->phy_regs.stat1000;
4924                         break;
4925                 case MII_ESTATUS:
4926                         data->val_out = adapter->phy_regs.estatus;
4927                         break;
4928                 default:
4929                         return -EIO;
4930                 }
4931                 break;
4932         case SIOCSMIIREG:
4933         default:
4934                 return -EOPNOTSUPP;
4935         }
4936         return 0;
4937 }
4938
4939 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4940 {
4941         switch (cmd) {
4942         case SIOCGMIIPHY:
4943         case SIOCGMIIREG:
4944         case SIOCSMIIREG:
4945                 return e1000_mii_ioctl(netdev, ifr, cmd);
4946         default:
4947                 return -EOPNOTSUPP;
4948         }
4949 }
4950
4951 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
4952 {
4953         struct e1000_hw *hw = &adapter->hw;
4954         u32 i, mac_reg;
4955         u16 phy_reg;
4956         int retval = 0;
4957
4958         /* copy MAC RARs to PHY RARs */
4959         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
4960
4961         /* copy MAC MTA to PHY MTA */
4962         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
4963                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
4964                 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
4965                 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
4966         }
4967
4968         /* configure PHY Rx Control register */
4969         e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
4970         mac_reg = er32(RCTL);
4971         if (mac_reg & E1000_RCTL_UPE)
4972                 phy_reg |= BM_RCTL_UPE;
4973         if (mac_reg & E1000_RCTL_MPE)
4974                 phy_reg |= BM_RCTL_MPE;
4975         phy_reg &= ~(BM_RCTL_MO_MASK);
4976         if (mac_reg & E1000_RCTL_MO_3)
4977                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
4978                                 << BM_RCTL_MO_SHIFT);
4979         if (mac_reg & E1000_RCTL_BAM)
4980                 phy_reg |= BM_RCTL_BAM;
4981         if (mac_reg & E1000_RCTL_PMCF)
4982                 phy_reg |= BM_RCTL_PMCF;
4983         mac_reg = er32(CTRL);
4984         if (mac_reg & E1000_CTRL_RFCE)
4985                 phy_reg |= BM_RCTL_RFCE;
4986         e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
4987
4988         /* enable PHY wakeup in MAC register */
4989         ew32(WUFC, wufc);
4990         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
4991
4992         /* configure and enable PHY wakeup in PHY registers */
4993         e1e_wphy(&adapter->hw, BM_WUFC, wufc);
4994         e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
4995
4996         /* activate PHY wakeup */
4997         retval = hw->phy.ops.acquire(hw);
4998         if (retval) {
4999                 e_err("Could not acquire PHY\n");
5000                 return retval;
5001         }
5002         e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5003                                  (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5004         retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5005         if (retval) {
5006                 e_err("Could not read PHY page 769\n");
5007                 goto out;
5008         }
5009         phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5010         retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5011         if (retval)
5012                 e_err("Could not set PHY Host Wakeup bit\n");
5013 out:
5014         hw->phy.ops.release(hw);
5015
5016         return retval;
5017 }
5018
5019 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5020                             bool runtime)
5021 {
5022         struct net_device *netdev = pci_get_drvdata(pdev);
5023         struct e1000_adapter *adapter = netdev_priv(netdev);
5024         struct e1000_hw *hw = &adapter->hw;
5025         u32 ctrl, ctrl_ext, rctl, status;
5026         /* Runtime suspend should only enable wakeup for link changes */
5027         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5028         int retval = 0;
5029
5030         netif_device_detach(netdev);
5031
5032         if (netif_running(netdev)) {
5033                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5034                 e1000e_down(adapter);
5035                 e1000_free_irq(adapter);
5036         }
5037         e1000e_reset_interrupt_capability(adapter);
5038
5039         retval = pci_save_state(pdev);
5040         if (retval)
5041                 return retval;
5042
5043         status = er32(STATUS);
5044         if (status & E1000_STATUS_LU)
5045                 wufc &= ~E1000_WUFC_LNKC;
5046
5047         if (wufc) {
5048                 e1000_setup_rctl(adapter);
5049                 e1000_set_multi(netdev);
5050
5051                 /* turn on all-multi mode if wake on multicast is enabled */
5052                 if (wufc & E1000_WUFC_MC) {
5053                         rctl = er32(RCTL);
5054                         rctl |= E1000_RCTL_MPE;
5055                         ew32(RCTL, rctl);
5056                 }
5057
5058                 ctrl = er32(CTRL);
5059                 /* advertise wake from D3Cold */
5060                 #define E1000_CTRL_ADVD3WUC 0x00100000
5061                 /* phy power management enable */
5062                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5063                 ctrl |= E1000_CTRL_ADVD3WUC;
5064                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5065                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5066                 ew32(CTRL, ctrl);
5067
5068                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5069                     adapter->hw.phy.media_type ==
5070                     e1000_media_type_internal_serdes) {
5071                         /* keep the laser running in D3 */
5072                         ctrl_ext = er32(CTRL_EXT);
5073                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5074                         ew32(CTRL_EXT, ctrl_ext);
5075                 }
5076
5077                 if (adapter->flags & FLAG_IS_ICH)
5078                         e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5079
5080                 /* Allow time for pending master requests to run */
5081                 e1000e_disable_pcie_master(&adapter->hw);
5082
5083                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5084                         /* enable wakeup by the PHY */
5085                         retval = e1000_init_phy_wakeup(adapter, wufc);
5086                         if (retval)
5087                                 return retval;
5088                 } else {
5089                         /* enable wakeup by the MAC */
5090                         ew32(WUFC, wufc);
5091                         ew32(WUC, E1000_WUC_PME_EN);
5092                 }
5093         } else {
5094                 ew32(WUC, 0);
5095                 ew32(WUFC, 0);
5096         }
5097
5098         *enable_wake = !!wufc;
5099
5100         /* make sure adapter isn't asleep if manageability is enabled */
5101         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5102             (hw->mac.ops.check_mng_mode(hw)))
5103                 *enable_wake = true;
5104
5105         if (adapter->hw.phy.type == e1000_phy_igp_3)
5106                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5107
5108         /*
5109          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5110          * would have already happened in close and is redundant.
5111          */
5112         e1000_release_hw_control(adapter);
5113
5114         pci_disable_device(pdev);
5115
5116         return 0;
5117 }
5118
5119 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5120 {
5121         if (sleep && wake) {
5122                 pci_prepare_to_sleep(pdev);
5123                 return;
5124         }
5125
5126         pci_wake_from_d3(pdev, wake);
5127         pci_set_power_state(pdev, PCI_D3hot);
5128 }
5129
5130 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5131                                     bool wake)
5132 {
5133         struct net_device *netdev = pci_get_drvdata(pdev);
5134         struct e1000_adapter *adapter = netdev_priv(netdev);
5135
5136         /*
5137          * The pci-e switch on some quad port adapters will report a
5138          * correctable error when the MAC transitions from D0 to D3.  To
5139          * prevent this we need to mask off the correctable errors on the
5140          * downstream port of the pci-e switch.
5141          */
5142         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5143                 struct pci_dev *us_dev = pdev->bus->self;
5144                 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5145                 u16 devctl;
5146
5147                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5148                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5149                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5150
5151                 e1000_power_off(pdev, sleep, wake);
5152
5153                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5154         } else {
5155                 e1000_power_off(pdev, sleep, wake);
5156         }
5157 }
5158
5159 #ifdef CONFIG_PCIEASPM
5160 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5161 {
5162         pci_disable_link_state(pdev, state);
5163 }
5164 #else
5165 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5166 {
5167         int pos;
5168         u16 reg16;
5169
5170         /*
5171          * Both device and parent should have the same ASPM setting.
5172          * Disable ASPM in downstream component first and then upstream.
5173          */
5174         pos = pci_pcie_cap(pdev);
5175         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5176         reg16 &= ~state;
5177         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5178
5179         if (!pdev->bus->self)
5180                 return;
5181
5182         pos = pci_pcie_cap(pdev->bus->self);
5183         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5184         reg16 &= ~state;
5185         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5186 }
5187 #endif
5188 void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5189 {
5190         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5191                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5192                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5193
5194         __e1000e_disable_aspm(pdev, state);
5195 }
5196
5197 #ifdef CONFIG_PM_OPS
5198 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5199 {
5200         return !!adapter->tx_ring->buffer_info;
5201 }
5202
5203 static int __e1000_resume(struct pci_dev *pdev)
5204 {
5205         struct net_device *netdev = pci_get_drvdata(pdev);
5206         struct e1000_adapter *adapter = netdev_priv(netdev);
5207         struct e1000_hw *hw = &adapter->hw;
5208         u32 err;
5209
5210         pci_set_power_state(pdev, PCI_D0);
5211         pci_restore_state(pdev);
5212         pci_save_state(pdev);
5213         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5214                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5215
5216         e1000e_set_interrupt_capability(adapter);
5217         if (netif_running(netdev)) {
5218                 err = e1000_request_irq(adapter);
5219                 if (err)
5220                         return err;
5221         }
5222
5223         e1000e_power_up_phy(adapter);
5224
5225         /* report the system wakeup cause from S3/S4 */
5226         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5227                 u16 phy_data;
5228
5229                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5230                 if (phy_data) {
5231                         e_info("PHY Wakeup cause - %s\n",
5232                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5233                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5234                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5235                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5236                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5237                                 " Change" : "other");
5238                 }
5239                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5240         } else {
5241                 u32 wus = er32(WUS);
5242                 if (wus) {
5243                         e_info("MAC Wakeup cause - %s\n",
5244                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5245                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5246                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5247                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5248                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5249                                 "other");
5250                 }
5251                 ew32(WUS, ~0);
5252         }
5253
5254         e1000e_reset(adapter);
5255
5256         e1000_init_manageability_pt(adapter);
5257
5258         if (netif_running(netdev))
5259                 e1000e_up(adapter);
5260
5261         netif_device_attach(netdev);
5262
5263         /*
5264          * If the controller has AMT, do not set DRV_LOAD until the interface
5265          * is up.  For all other cases, let the f/w know that the h/w is now
5266          * under the control of the driver.
5267          */
5268         if (!(adapter->flags & FLAG_HAS_AMT))
5269                 e1000_get_hw_control(adapter);
5270
5271         return 0;
5272 }
5273
5274 #ifdef CONFIG_PM_SLEEP
5275 static int e1000_suspend(struct device *dev)
5276 {
5277         struct pci_dev *pdev = to_pci_dev(dev);
5278         int retval;
5279         bool wake;
5280
5281         retval = __e1000_shutdown(pdev, &wake, false);
5282         if (!retval)
5283                 e1000_complete_shutdown(pdev, true, wake);
5284
5285         return retval;
5286 }
5287
5288 static int e1000_resume(struct device *dev)
5289 {
5290         struct pci_dev *pdev = to_pci_dev(dev);
5291         struct net_device *netdev = pci_get_drvdata(pdev);
5292         struct e1000_adapter *adapter = netdev_priv(netdev);
5293
5294         if (e1000e_pm_ready(adapter))
5295                 adapter->idle_check = true;
5296
5297         return __e1000_resume(pdev);
5298 }
5299 #endif /* CONFIG_PM_SLEEP */
5300
5301 #ifdef CONFIG_PM_RUNTIME
5302 static int e1000_runtime_suspend(struct device *dev)
5303 {
5304         struct pci_dev *pdev = to_pci_dev(dev);
5305         struct net_device *netdev = pci_get_drvdata(pdev);
5306         struct e1000_adapter *adapter = netdev_priv(netdev);
5307
5308         if (e1000e_pm_ready(adapter)) {
5309                 bool wake;
5310
5311                 __e1000_shutdown(pdev, &wake, true);
5312         }
5313
5314         return 0;
5315 }
5316
5317 static int e1000_idle(struct device *dev)
5318 {
5319         struct pci_dev *pdev = to_pci_dev(dev);
5320         struct net_device *netdev = pci_get_drvdata(pdev);
5321         struct e1000_adapter *adapter = netdev_priv(netdev);
5322
5323         if (!e1000e_pm_ready(adapter))
5324                 return 0;
5325
5326         if (adapter->idle_check) {
5327                 adapter->idle_check = false;
5328                 if (!e1000e_has_link(adapter))
5329                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5330         }
5331
5332         return -EBUSY;
5333 }
5334
5335 static int e1000_runtime_resume(struct device *dev)
5336 {
5337         struct pci_dev *pdev = to_pci_dev(dev);
5338         struct net_device *netdev = pci_get_drvdata(pdev);
5339         struct e1000_adapter *adapter = netdev_priv(netdev);
5340
5341         if (!e1000e_pm_ready(adapter))
5342                 return 0;
5343
5344         adapter->idle_check = !dev->power.runtime_auto;
5345         return __e1000_resume(pdev);
5346 }
5347 #endif /* CONFIG_PM_RUNTIME */
5348 #endif /* CONFIG_PM_OPS */
5349
5350 static void e1000_shutdown(struct pci_dev *pdev)
5351 {
5352         bool wake = false;
5353
5354         __e1000_shutdown(pdev, &wake, false);
5355
5356         if (system_state == SYSTEM_POWER_OFF)
5357                 e1000_complete_shutdown(pdev, false, wake);
5358 }
5359
5360 #ifdef CONFIG_NET_POLL_CONTROLLER
5361 /*
5362  * Polling 'interrupt' - used by things like netconsole to send skbs
5363  * without having to re-enable interrupts. It's not called while
5364  * the interrupt routine is executing.
5365  */
5366 static void e1000_netpoll(struct net_device *netdev)
5367 {
5368         struct e1000_adapter *adapter = netdev_priv(netdev);
5369
5370         disable_irq(adapter->pdev->irq);
5371         e1000_intr(adapter->pdev->irq, netdev);
5372
5373         enable_irq(adapter->pdev->irq);
5374 }
5375 #endif
5376
5377 /**
5378  * e1000_io_error_detected - called when PCI error is detected
5379  * @pdev: Pointer to PCI device
5380  * @state: The current pci connection state
5381  *
5382  * This function is called after a PCI bus error affecting
5383  * this device has been detected.
5384  */
5385 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5386                                                 pci_channel_state_t state)
5387 {
5388         struct net_device *netdev = pci_get_drvdata(pdev);
5389         struct e1000_adapter *adapter = netdev_priv(netdev);
5390
5391         netif_device_detach(netdev);
5392
5393         if (state == pci_channel_io_perm_failure)
5394                 return PCI_ERS_RESULT_DISCONNECT;
5395
5396         if (netif_running(netdev))
5397                 e1000e_down(adapter);
5398         pci_disable_device(pdev);
5399
5400         /* Request a slot slot reset. */
5401         return PCI_ERS_RESULT_NEED_RESET;
5402 }
5403
5404 /**
5405  * e1000_io_slot_reset - called after the pci bus has been reset.
5406  * @pdev: Pointer to PCI device
5407  *
5408  * Restart the card from scratch, as if from a cold-boot. Implementation
5409  * resembles the first-half of the e1000_resume routine.
5410  */
5411 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5412 {
5413         struct net_device *netdev = pci_get_drvdata(pdev);
5414         struct e1000_adapter *adapter = netdev_priv(netdev);
5415         struct e1000_hw *hw = &adapter->hw;
5416         int err;
5417         pci_ers_result_t result;
5418
5419         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5420                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5421         err = pci_enable_device_mem(pdev);
5422         if (err) {
5423                 dev_err(&pdev->dev,
5424                         "Cannot re-enable PCI device after reset.\n");
5425                 result = PCI_ERS_RESULT_DISCONNECT;
5426         } else {
5427                 pci_set_master(pdev);
5428                 pdev->state_saved = true;
5429                 pci_restore_state(pdev);
5430
5431                 pci_enable_wake(pdev, PCI_D3hot, 0);
5432                 pci_enable_wake(pdev, PCI_D3cold, 0);
5433
5434                 e1000e_reset(adapter);
5435                 ew32(WUS, ~0);
5436                 result = PCI_ERS_RESULT_RECOVERED;
5437         }
5438
5439         pci_cleanup_aer_uncorrect_error_status(pdev);
5440
5441         return result;
5442 }
5443
5444 /**
5445  * e1000_io_resume - called when traffic can start flowing again.
5446  * @pdev: Pointer to PCI device
5447  *
5448  * This callback is called when the error recovery driver tells us that
5449  * its OK to resume normal operation. Implementation resembles the
5450  * second-half of the e1000_resume routine.
5451  */
5452 static void e1000_io_resume(struct pci_dev *pdev)
5453 {
5454         struct net_device *netdev = pci_get_drvdata(pdev);
5455         struct e1000_adapter *adapter = netdev_priv(netdev);
5456
5457         e1000_init_manageability_pt(adapter);
5458
5459         if (netif_running(netdev)) {
5460                 if (e1000e_up(adapter)) {
5461                         dev_err(&pdev->dev,
5462                                 "can't bring device back up after reset\n");
5463                         return;
5464                 }
5465         }
5466
5467         netif_device_attach(netdev);
5468
5469         /*
5470          * If the controller has AMT, do not set DRV_LOAD until the interface
5471          * is up.  For all other cases, let the f/w know that the h/w is now
5472          * under the control of the driver.
5473          */
5474         if (!(adapter->flags & FLAG_HAS_AMT))
5475                 e1000_get_hw_control(adapter);
5476
5477 }
5478
5479 static void e1000_print_device_info(struct e1000_adapter *adapter)
5480 {
5481         struct e1000_hw *hw = &adapter->hw;
5482         struct net_device *netdev = adapter->netdev;
5483         u32 pba_num;
5484
5485         /* print bus type/speed/width info */
5486         e_info("(PCI Express:2.5GB/s:%s) %pM\n",
5487                /* bus width */
5488                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5489                 "Width x1"),
5490                /* MAC address */
5491                netdev->dev_addr);
5492         e_info("Intel(R) PRO/%s Network Connection\n",
5493                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5494         e1000e_read_pba_num(hw, &pba_num);
5495         e_info("MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
5496                hw->mac.type, hw->phy.type, (pba_num >> 8), (pba_num & 0xff));
5497 }
5498
5499 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5500 {
5501         struct e1000_hw *hw = &adapter->hw;
5502         int ret_val;
5503         u16 buf = 0;
5504
5505         if (hw->mac.type != e1000_82573)
5506                 return;
5507
5508         ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5509         if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5510                 /* Deep Smart Power Down (DSPD) */
5511                 dev_warn(&adapter->pdev->dev,
5512                          "Warning: detected DSPD enabled in EEPROM\n");
5513         }
5514 }
5515
5516 static const struct net_device_ops e1000e_netdev_ops = {
5517         .ndo_open               = e1000_open,
5518         .ndo_stop               = e1000_close,
5519         .ndo_start_xmit         = e1000_xmit_frame,
5520         .ndo_get_stats          = e1000_get_stats,
5521         .ndo_set_multicast_list = e1000_set_multi,
5522         .ndo_set_mac_address    = e1000_set_mac,
5523         .ndo_change_mtu         = e1000_change_mtu,
5524         .ndo_do_ioctl           = e1000_ioctl,
5525         .ndo_tx_timeout         = e1000_tx_timeout,
5526         .ndo_validate_addr      = eth_validate_addr,
5527
5528         .ndo_vlan_rx_register   = e1000_vlan_rx_register,
5529         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
5530         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
5531 #ifdef CONFIG_NET_POLL_CONTROLLER
5532         .ndo_poll_controller    = e1000_netpoll,
5533 #endif
5534 };
5535
5536 /**
5537  * e1000_probe - Device Initialization Routine
5538  * @pdev: PCI device information struct
5539  * @ent: entry in e1000_pci_tbl
5540  *
5541  * Returns 0 on success, negative on failure
5542  *
5543  * e1000_probe initializes an adapter identified by a pci_dev structure.
5544  * The OS initialization, configuring of the adapter private structure,
5545  * and a hardware reset occur.
5546  **/
5547 static int __devinit e1000_probe(struct pci_dev *pdev,
5548                                  const struct pci_device_id *ent)
5549 {
5550         struct net_device *netdev;
5551         struct e1000_adapter *adapter;
5552         struct e1000_hw *hw;
5553         const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5554         resource_size_t mmio_start, mmio_len;
5555         resource_size_t flash_start, flash_len;
5556
5557         static int cards_found;
5558         int i, err, pci_using_dac;
5559         u16 eeprom_data = 0;
5560         u16 eeprom_apme_mask = E1000_EEPROM_APME;
5561
5562         if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5563                 e1000e_disable_aspm(pdev, PCIE_LINK_STATE_L1);
5564
5565         err = pci_enable_device_mem(pdev);
5566         if (err)
5567                 return err;
5568
5569         pci_using_dac = 0;
5570         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5571         if (!err) {
5572                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5573                 if (!err)
5574                         pci_using_dac = 1;
5575         } else {
5576                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5577                 if (err) {
5578                         err = dma_set_coherent_mask(&pdev->dev,
5579                                                     DMA_BIT_MASK(32));
5580                         if (err) {
5581                                 dev_err(&pdev->dev, "No usable DMA "
5582                                         "configuration, aborting\n");
5583                                 goto err_dma;
5584                         }
5585                 }
5586         }
5587
5588         err = pci_request_selected_regions_exclusive(pdev,
5589                                           pci_select_bars(pdev, IORESOURCE_MEM),
5590                                           e1000e_driver_name);
5591         if (err)
5592                 goto err_pci_reg;
5593
5594         /* AER (Advanced Error Reporting) hooks */
5595         pci_enable_pcie_error_reporting(pdev);
5596
5597         pci_set_master(pdev);
5598         /* PCI config space info */
5599         err = pci_save_state(pdev);
5600         if (err)
5601                 goto err_alloc_etherdev;
5602
5603         err = -ENOMEM;
5604         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5605         if (!netdev)
5606                 goto err_alloc_etherdev;
5607
5608         SET_NETDEV_DEV(netdev, &pdev->dev);
5609
5610         netdev->irq = pdev->irq;
5611
5612         pci_set_drvdata(pdev, netdev);
5613         adapter = netdev_priv(netdev);
5614         hw = &adapter->hw;
5615         adapter->netdev = netdev;
5616         adapter->pdev = pdev;
5617         adapter->ei = ei;
5618         adapter->pba = ei->pba;
5619         adapter->flags = ei->flags;
5620         adapter->flags2 = ei->flags2;
5621         adapter->hw.adapter = adapter;
5622         adapter->hw.mac.type = ei->mac;
5623         adapter->max_hw_frame_size = ei->max_hw_frame_size;
5624         adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5625
5626         mmio_start = pci_resource_start(pdev, 0);
5627         mmio_len = pci_resource_len(pdev, 0);
5628
5629         err = -EIO;
5630         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5631         if (!adapter->hw.hw_addr)
5632                 goto err_ioremap;
5633
5634         if ((adapter->flags & FLAG_HAS_FLASH) &&
5635             (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5636                 flash_start = pci_resource_start(pdev, 1);
5637                 flash_len = pci_resource_len(pdev, 1);
5638                 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5639                 if (!adapter->hw.flash_address)
5640                         goto err_flashmap;
5641         }
5642
5643         /* construct the net_device struct */
5644         netdev->netdev_ops              = &e1000e_netdev_ops;
5645         e1000e_set_ethtool_ops(netdev);
5646         netdev->watchdog_timeo          = 5 * HZ;
5647         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5648         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5649
5650         netdev->mem_start = mmio_start;
5651         netdev->mem_end = mmio_start + mmio_len;
5652
5653         adapter->bd_number = cards_found++;
5654
5655         e1000e_check_options(adapter);
5656
5657         /* setup adapter struct */
5658         err = e1000_sw_init(adapter);
5659         if (err)
5660                 goto err_sw_init;
5661
5662         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5663         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5664         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5665
5666         err = ei->get_variants(adapter);
5667         if (err)
5668                 goto err_hw_init;
5669
5670         if ((adapter->flags & FLAG_IS_ICH) &&
5671             (adapter->flags & FLAG_READ_ONLY_NVM))
5672                 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5673
5674         hw->mac.ops.get_bus_info(&adapter->hw);
5675
5676         adapter->hw.phy.autoneg_wait_to_complete = 0;
5677
5678         /* Copper options */
5679         if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5680                 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5681                 adapter->hw.phy.disable_polarity_correction = 0;
5682                 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5683         }
5684
5685         if (e1000_check_reset_block(&adapter->hw))
5686                 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5687
5688         netdev->features = NETIF_F_SG |
5689                            NETIF_F_HW_CSUM |
5690                            NETIF_F_HW_VLAN_TX |
5691                            NETIF_F_HW_VLAN_RX;
5692
5693         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5694                 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5695
5696         netdev->features |= NETIF_F_TSO;
5697         netdev->features |= NETIF_F_TSO6;
5698
5699         netdev->vlan_features |= NETIF_F_TSO;
5700         netdev->vlan_features |= NETIF_F_TSO6;
5701         netdev->vlan_features |= NETIF_F_HW_CSUM;
5702         netdev->vlan_features |= NETIF_F_SG;
5703
5704         if (pci_using_dac)
5705                 netdev->features |= NETIF_F_HIGHDMA;
5706
5707         if (e1000e_enable_mng_pass_thru(&adapter->hw))
5708                 adapter->flags |= FLAG_MNG_PT_ENABLED;
5709
5710         /*
5711          * before reading the NVM, reset the controller to
5712          * put the device in a known good starting state
5713          */
5714         adapter->hw.mac.ops.reset_hw(&adapter->hw);
5715
5716         /*
5717          * systems with ASPM and others may see the checksum fail on the first
5718          * attempt. Let's give it a few tries
5719          */
5720         for (i = 0;; i++) {
5721                 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5722                         break;
5723                 if (i == 2) {
5724                         e_err("The NVM Checksum Is Not Valid\n");
5725                         err = -EIO;
5726                         goto err_eeprom;
5727                 }
5728         }
5729
5730         e1000_eeprom_checks(adapter);
5731
5732         /* copy the MAC address */
5733         if (e1000e_read_mac_addr(&adapter->hw))
5734                 e_err("NVM Read Error while reading MAC address\n");
5735
5736         memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
5737         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
5738
5739         if (!is_valid_ether_addr(netdev->perm_addr)) {
5740                 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
5741                 err = -EIO;
5742                 goto err_eeprom;
5743         }
5744
5745         init_timer(&adapter->watchdog_timer);
5746         adapter->watchdog_timer.function = &e1000_watchdog;
5747         adapter->watchdog_timer.data = (unsigned long) adapter;
5748
5749         init_timer(&adapter->phy_info_timer);
5750         adapter->phy_info_timer.function = &e1000_update_phy_info;
5751         adapter->phy_info_timer.data = (unsigned long) adapter;
5752
5753         INIT_WORK(&adapter->reset_task, e1000_reset_task);
5754         INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
5755         INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
5756         INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
5757         INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
5758
5759         /* Initialize link parameters. User can change them with ethtool */
5760         adapter->hw.mac.autoneg = 1;
5761         adapter->fc_autoneg = 1;
5762         adapter->hw.fc.requested_mode = e1000_fc_default;
5763         adapter->hw.fc.current_mode = e1000_fc_default;
5764         adapter->hw.phy.autoneg_advertised = 0x2f;
5765
5766         /* ring size defaults */
5767         adapter->rx_ring->count = 256;
5768         adapter->tx_ring->count = 256;
5769
5770         /*
5771          * Initial Wake on LAN setting - If APM wake is enabled in
5772          * the EEPROM, enable the ACPI Magic Packet filter
5773          */
5774         if (adapter->flags & FLAG_APME_IN_WUC) {
5775                 /* APME bit in EEPROM is mapped to WUC.APME */
5776                 eeprom_data = er32(WUC);
5777                 eeprom_apme_mask = E1000_WUC_APME;
5778                 if (eeprom_data & E1000_WUC_PHY_WAKE)
5779                         adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
5780         } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
5781                 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
5782                     (adapter->hw.bus.func == 1))
5783                         e1000_read_nvm(&adapter->hw,
5784                                 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
5785                 else
5786                         e1000_read_nvm(&adapter->hw,
5787                                 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
5788         }
5789
5790         /* fetch WoL from EEPROM */
5791         if (eeprom_data & eeprom_apme_mask)
5792                 adapter->eeprom_wol |= E1000_WUFC_MAG;
5793
5794         /*
5795          * now that we have&n