Merge git://bedivere.hansenpartnership.com/git/scsi-rc-fixes-2.6
[linux-2.6.git] / drivers / net / e1000e / netdev.c
1 /*******************************************************************************
2
3   Intel PRO/1000 Linux driver
4   Copyright(c) 1999 - 2011 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/interrupt.h>
40 #include <linux/tcp.h>
41 #include <linux/ipv6.h>
42 #include <linux/slab.h>
43 #include <net/checksum.h>
44 #include <net/ip6_checksum.h>
45 #include <linux/mii.h>
46 #include <linux/ethtool.h>
47 #include <linux/if_vlan.h>
48 #include <linux/cpu.h>
49 #include <linux/smp.h>
50 #include <linux/pm_qos_params.h>
51 #include <linux/pm_runtime.h>
52 #include <linux/aer.h>
53 #include <linux/prefetch.h>
54
55 #include "e1000.h"
56
57 #define DRV_EXTRAVERSION "-k"
58
59 #define DRV_VERSION "1.4.4" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
62
63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
64
65 static const struct e1000_info *e1000_info_tbl[] = {
66         [board_82571]           = &e1000_82571_info,
67         [board_82572]           = &e1000_82572_info,
68         [board_82573]           = &e1000_82573_info,
69         [board_82574]           = &e1000_82574_info,
70         [board_82583]           = &e1000_82583_info,
71         [board_80003es2lan]     = &e1000_es2_info,
72         [board_ich8lan]         = &e1000_ich8_info,
73         [board_ich9lan]         = &e1000_ich9_info,
74         [board_ich10lan]        = &e1000_ich10_info,
75         [board_pchlan]          = &e1000_pch_info,
76         [board_pch2lan]         = &e1000_pch2_info,
77 };
78
79 struct e1000_reg_info {
80         u32 ofs;
81         char *name;
82 };
83
84 #define E1000_RDFH      0x02410 /* Rx Data FIFO Head - RW */
85 #define E1000_RDFT      0x02418 /* Rx Data FIFO Tail - RW */
86 #define E1000_RDFHS     0x02420 /* Rx Data FIFO Head Saved - RW */
87 #define E1000_RDFTS     0x02428 /* Rx Data FIFO Tail Saved - RW */
88 #define E1000_RDFPC     0x02430 /* Rx Data FIFO Packet Count - RW */
89
90 #define E1000_TDFH      0x03410 /* Tx Data FIFO Head - RW */
91 #define E1000_TDFT      0x03418 /* Tx Data FIFO Tail - RW */
92 #define E1000_TDFHS     0x03420 /* Tx Data FIFO Head Saved - RW */
93 #define E1000_TDFTS     0x03428 /* Tx Data FIFO Tail Saved - RW */
94 #define E1000_TDFPC     0x03430 /* Tx Data FIFO Packet Count - RW */
95
96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
97
98         /* General Registers */
99         {E1000_CTRL, "CTRL"},
100         {E1000_STATUS, "STATUS"},
101         {E1000_CTRL_EXT, "CTRL_EXT"},
102
103         /* Interrupt Registers */
104         {E1000_ICR, "ICR"},
105
106         /* Rx Registers */
107         {E1000_RCTL, "RCTL"},
108         {E1000_RDLEN, "RDLEN"},
109         {E1000_RDH, "RDH"},
110         {E1000_RDT, "RDT"},
111         {E1000_RDTR, "RDTR"},
112         {E1000_RXDCTL(0), "RXDCTL"},
113         {E1000_ERT, "ERT"},
114         {E1000_RDBAL, "RDBAL"},
115         {E1000_RDBAH, "RDBAH"},
116         {E1000_RDFH, "RDFH"},
117         {E1000_RDFT, "RDFT"},
118         {E1000_RDFHS, "RDFHS"},
119         {E1000_RDFTS, "RDFTS"},
120         {E1000_RDFPC, "RDFPC"},
121
122         /* Tx Registers */
123         {E1000_TCTL, "TCTL"},
124         {E1000_TDBAL, "TDBAL"},
125         {E1000_TDBAH, "TDBAH"},
126         {E1000_TDLEN, "TDLEN"},
127         {E1000_TDH, "TDH"},
128         {E1000_TDT, "TDT"},
129         {E1000_TIDV, "TIDV"},
130         {E1000_TXDCTL(0), "TXDCTL"},
131         {E1000_TADV, "TADV"},
132         {E1000_TARC(0), "TARC"},
133         {E1000_TDFH, "TDFH"},
134         {E1000_TDFT, "TDFT"},
135         {E1000_TDFHS, "TDFHS"},
136         {E1000_TDFTS, "TDFTS"},
137         {E1000_TDFPC, "TDFPC"},
138
139         /* List Terminator */
140         {}
141 };
142
143 /*
144  * e1000_regdump - register printout routine
145  */
146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
147 {
148         int n = 0;
149         char rname[16];
150         u32 regs[8];
151
152         switch (reginfo->ofs) {
153         case E1000_RXDCTL(0):
154                 for (n = 0; n < 2; n++)
155                         regs[n] = __er32(hw, E1000_RXDCTL(n));
156                 break;
157         case E1000_TXDCTL(0):
158                 for (n = 0; n < 2; n++)
159                         regs[n] = __er32(hw, E1000_TXDCTL(n));
160                 break;
161         case E1000_TARC(0):
162                 for (n = 0; n < 2; n++)
163                         regs[n] = __er32(hw, E1000_TARC(n));
164                 break;
165         default:
166                 printk(KERN_INFO "%-15s %08x\n",
167                        reginfo->name, __er32(hw, reginfo->ofs));
168                 return;
169         }
170
171         snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
172         printk(KERN_INFO "%-15s ", rname);
173         for (n = 0; n < 2; n++)
174                 printk(KERN_CONT "%08x ", regs[n]);
175         printk(KERN_CONT "\n");
176 }
177
178 /*
179  * e1000e_dump - Print registers, Tx-ring and Rx-ring
180  */
181 static void e1000e_dump(struct e1000_adapter *adapter)
182 {
183         struct net_device *netdev = adapter->netdev;
184         struct e1000_hw *hw = &adapter->hw;
185         struct e1000_reg_info *reginfo;
186         struct e1000_ring *tx_ring = adapter->tx_ring;
187         struct e1000_tx_desc *tx_desc;
188         struct my_u0 {
189                 u64 a;
190                 u64 b;
191         } *u0;
192         struct e1000_buffer *buffer_info;
193         struct e1000_ring *rx_ring = adapter->rx_ring;
194         union e1000_rx_desc_packet_split *rx_desc_ps;
195         struct e1000_rx_desc *rx_desc;
196         struct my_u1 {
197                 u64 a;
198                 u64 b;
199                 u64 c;
200                 u64 d;
201         } *u1;
202         u32 staterr;
203         int i = 0;
204
205         if (!netif_msg_hw(adapter))
206                 return;
207
208         /* Print netdevice Info */
209         if (netdev) {
210                 dev_info(&adapter->pdev->dev, "Net device Info\n");
211                 printk(KERN_INFO "Device Name     state            "
212                        "trans_start      last_rx\n");
213                 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
214                        netdev->name, netdev->state, netdev->trans_start,
215                        netdev->last_rx);
216         }
217
218         /* Print Registers */
219         dev_info(&adapter->pdev->dev, "Register Dump\n");
220         printk(KERN_INFO " Register Name   Value\n");
221         for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
222              reginfo->name; reginfo++) {
223                 e1000_regdump(hw, reginfo);
224         }
225
226         /* Print Tx Ring Summary */
227         if (!netdev || !netif_running(netdev))
228                 goto exit;
229
230         dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
231         printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma  ]"
232                " leng ntw timestamp\n");
233         buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
234         printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
235                0, tx_ring->next_to_use, tx_ring->next_to_clean,
236                (unsigned long long)buffer_info->dma,
237                buffer_info->length,
238                buffer_info->next_to_watch,
239                (unsigned long long)buffer_info->time_stamp);
240
241         /* Print Tx Ring */
242         if (!netif_msg_tx_done(adapter))
243                 goto rx_ring_summary;
244
245         dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
246
247         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
248          *
249          * Legacy Transmit Descriptor
250          *   +--------------------------------------------------------------+
251          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
252          *   +--------------------------------------------------------------+
253          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
254          *   +--------------------------------------------------------------+
255          *   63       48 47        36 35    32 31     24 23    16 15        0
256          *
257          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
258          *   63      48 47    40 39       32 31             16 15    8 7      0
259          *   +----------------------------------------------------------------+
260          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
261          *   +----------------------------------------------------------------+
262          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
263          *   +----------------------------------------------------------------+
264          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
265          *
266          * Extended Data Descriptor (DTYP=0x1)
267          *   +----------------------------------------------------------------+
268          * 0 |                     Buffer Address [63:0]                      |
269          *   +----------------------------------------------------------------+
270          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
271          *   +----------------------------------------------------------------+
272          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
273          */
274         printk(KERN_INFO "Tl[desc]     [address 63:0  ] [SpeCssSCmCsLen]"
275                " [bi->dma       ] leng  ntw timestamp        bi->skb "
276                "<-- Legacy format\n");
277         printk(KERN_INFO "Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
278                " [bi->dma       ] leng  ntw timestamp        bi->skb "
279                "<-- Ext Context format\n");
280         printk(KERN_INFO "Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen]"
281                " [bi->dma       ] leng  ntw timestamp        bi->skb "
282                "<-- Ext Data format\n");
283         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
284                 tx_desc = E1000_TX_DESC(*tx_ring, i);
285                 buffer_info = &tx_ring->buffer_info[i];
286                 u0 = (struct my_u0 *)tx_desc;
287                 printk(KERN_INFO "T%c[0x%03X]    %016llX %016llX %016llX "
288                        "%04X  %3X %016llX %p",
289                        (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
290                         ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
291                        (unsigned long long)le64_to_cpu(u0->a),
292                        (unsigned long long)le64_to_cpu(u0->b),
293                        (unsigned long long)buffer_info->dma,
294                        buffer_info->length, buffer_info->next_to_watch,
295                        (unsigned long long)buffer_info->time_stamp,
296                        buffer_info->skb);
297                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
298                         printk(KERN_CONT " NTC/U\n");
299                 else if (i == tx_ring->next_to_use)
300                         printk(KERN_CONT " NTU\n");
301                 else if (i == tx_ring->next_to_clean)
302                         printk(KERN_CONT " NTC\n");
303                 else
304                         printk(KERN_CONT "\n");
305
306                 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
307                         print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
308                                        16, 1, phys_to_virt(buffer_info->dma),
309                                        buffer_info->length, true);
310         }
311
312         /* Print Rx Ring Summary */
313 rx_ring_summary:
314         dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
315         printk(KERN_INFO "Queue [NTU] [NTC]\n");
316         printk(KERN_INFO " %5d %5X %5X\n", 0,
317                rx_ring->next_to_use, rx_ring->next_to_clean);
318
319         /* Print Rx Ring */
320         if (!netif_msg_rx_status(adapter))
321                 goto exit;
322
323         dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
324         switch (adapter->rx_ps_pages) {
325         case 1:
326         case 2:
327         case 3:
328                 /* [Extended] Packet Split Receive Descriptor Format
329                  *
330                  *    +-----------------------------------------------------+
331                  *  0 |                Buffer Address 0 [63:0]              |
332                  *    +-----------------------------------------------------+
333                  *  8 |                Buffer Address 1 [63:0]              |
334                  *    +-----------------------------------------------------+
335                  * 16 |                Buffer Address 2 [63:0]              |
336                  *    +-----------------------------------------------------+
337                  * 24 |                Buffer Address 3 [63:0]              |
338                  *    +-----------------------------------------------------+
339                  */
340                 printk(KERN_INFO "R  [desc]      [buffer 0 63:0 ] "
341                        "[buffer 1 63:0 ] "
342                        "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma       ] "
343                        "[bi->skb] <-- Ext Pkt Split format\n");
344                 /* [Extended] Receive Descriptor (Write-Back) Format
345                  *
346                  *   63       48 47    32 31     13 12    8 7    4 3        0
347                  *   +------------------------------------------------------+
348                  * 0 | Packet   | IP     |  Rsvd   | MRQ   | Rsvd | MRQ RSS |
349                  *   | Checksum | Ident  |         | Queue |      |  Type   |
350                  *   +------------------------------------------------------+
351                  * 8 | VLAN Tag | Length | Extended Error | Extended Status |
352                  *   +------------------------------------------------------+
353                  *   63       48 47    32 31            20 19               0
354                  */
355                 printk(KERN_INFO "RWB[desc]      [ck ipid mrqhsh] "
356                        "[vl   l0 ee  es] "
357                        "[ l3  l2  l1 hs] [reserved      ] ---------------- "
358                        "[bi->skb] <-- Ext Rx Write-Back format\n");
359                 for (i = 0; i < rx_ring->count; i++) {
360                         buffer_info = &rx_ring->buffer_info[i];
361                         rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
362                         u1 = (struct my_u1 *)rx_desc_ps;
363                         staterr =
364                             le32_to_cpu(rx_desc_ps->wb.middle.status_error);
365                         if (staterr & E1000_RXD_STAT_DD) {
366                                 /* Descriptor Done */
367                                 printk(KERN_INFO "RWB[0x%03X]     %016llX "
368                                        "%016llX %016llX %016llX "
369                                        "---------------- %p", i,
370                                        (unsigned long long)le64_to_cpu(u1->a),
371                                        (unsigned long long)le64_to_cpu(u1->b),
372                                        (unsigned long long)le64_to_cpu(u1->c),
373                                        (unsigned long long)le64_to_cpu(u1->d),
374                                        buffer_info->skb);
375                         } else {
376                                 printk(KERN_INFO "R  [0x%03X]     %016llX "
377                                        "%016llX %016llX %016llX %016llX %p", i,
378                                        (unsigned long long)le64_to_cpu(u1->a),
379                                        (unsigned long long)le64_to_cpu(u1->b),
380                                        (unsigned long long)le64_to_cpu(u1->c),
381                                        (unsigned long long)le64_to_cpu(u1->d),
382                                        (unsigned long long)buffer_info->dma,
383                                        buffer_info->skb);
384
385                                 if (netif_msg_pktdata(adapter))
386                                         print_hex_dump(KERN_INFO, "",
387                                                 DUMP_PREFIX_ADDRESS, 16, 1,
388                                                 phys_to_virt(buffer_info->dma),
389                                                 adapter->rx_ps_bsize0, true);
390                         }
391
392                         if (i == rx_ring->next_to_use)
393                                 printk(KERN_CONT " NTU\n");
394                         else if (i == rx_ring->next_to_clean)
395                                 printk(KERN_CONT " NTC\n");
396                         else
397                                 printk(KERN_CONT "\n");
398                 }
399                 break;
400         default:
401         case 0:
402                 /* Legacy Receive Descriptor Format
403                  *
404                  * +-----------------------------------------------------+
405                  * |                Buffer Address [63:0]                |
406                  * +-----------------------------------------------------+
407                  * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
408                  * +-----------------------------------------------------+
409                  * 63       48 47    40 39      32 31         16 15      0
410                  */
411                 printk(KERN_INFO "Rl[desc]     [address 63:0  ] "
412                        "[vl er S cks ln] [bi->dma       ] [bi->skb] "
413                        "<-- Legacy format\n");
414                 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
415                         rx_desc = E1000_RX_DESC(*rx_ring, i);
416                         buffer_info = &rx_ring->buffer_info[i];
417                         u0 = (struct my_u0 *)rx_desc;
418                         printk(KERN_INFO "Rl[0x%03X]    %016llX %016llX "
419                                "%016llX %p", i,
420                                (unsigned long long)le64_to_cpu(u0->a),
421                                (unsigned long long)le64_to_cpu(u0->b),
422                                (unsigned long long)buffer_info->dma,
423                                buffer_info->skb);
424                         if (i == rx_ring->next_to_use)
425                                 printk(KERN_CONT " NTU\n");
426                         else if (i == rx_ring->next_to_clean)
427                                 printk(KERN_CONT " NTC\n");
428                         else
429                                 printk(KERN_CONT "\n");
430
431                         if (netif_msg_pktdata(adapter))
432                                 print_hex_dump(KERN_INFO, "",
433                                                DUMP_PREFIX_ADDRESS,
434                                                16, 1,
435                                                phys_to_virt(buffer_info->dma),
436                                                adapter->rx_buffer_len, true);
437                 }
438         }
439
440 exit:
441         return;
442 }
443
444 /**
445  * e1000_desc_unused - calculate if we have unused descriptors
446  **/
447 static int e1000_desc_unused(struct e1000_ring *ring)
448 {
449         if (ring->next_to_clean > ring->next_to_use)
450                 return ring->next_to_clean - ring->next_to_use - 1;
451
452         return ring->count + ring->next_to_clean - ring->next_to_use - 1;
453 }
454
455 /**
456  * e1000_receive_skb - helper function to handle Rx indications
457  * @adapter: board private structure
458  * @status: descriptor status field as written by hardware
459  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
460  * @skb: pointer to sk_buff to be indicated to stack
461  **/
462 static void e1000_receive_skb(struct e1000_adapter *adapter,
463                               struct net_device *netdev, struct sk_buff *skb,
464                               u8 status, __le16 vlan)
465 {
466         u16 tag = le16_to_cpu(vlan);
467         skb->protocol = eth_type_trans(skb, netdev);
468
469         if (status & E1000_RXD_STAT_VP)
470                 __vlan_hwaccel_put_tag(skb, tag);
471
472         napi_gro_receive(&adapter->napi, skb);
473 }
474
475 /**
476  * e1000_rx_checksum - Receive Checksum Offload
477  * @adapter:     board private structure
478  * @status_err:  receive descriptor status and error fields
479  * @csum:       receive descriptor csum field
480  * @sk_buff:     socket buffer with received data
481  **/
482 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
483                               u32 csum, struct sk_buff *skb)
484 {
485         u16 status = (u16)status_err;
486         u8 errors = (u8)(status_err >> 24);
487
488         skb_checksum_none_assert(skb);
489
490         /* Ignore Checksum bit is set */
491         if (status & E1000_RXD_STAT_IXSM)
492                 return;
493         /* TCP/UDP checksum error bit is set */
494         if (errors & E1000_RXD_ERR_TCPE) {
495                 /* let the stack verify checksum errors */
496                 adapter->hw_csum_err++;
497                 return;
498         }
499
500         /* TCP/UDP Checksum has not been calculated */
501         if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
502                 return;
503
504         /* It must be a TCP or UDP packet with a valid checksum */
505         if (status & E1000_RXD_STAT_TCPCS) {
506                 /* TCP checksum is good */
507                 skb->ip_summed = CHECKSUM_UNNECESSARY;
508         } else {
509                 /*
510                  * IP fragment with UDP payload
511                  * Hardware complements the payload checksum, so we undo it
512                  * and then put the value in host order for further stack use.
513                  */
514                 __sum16 sum = (__force __sum16)htons(csum);
515                 skb->csum = csum_unfold(~sum);
516                 skb->ip_summed = CHECKSUM_COMPLETE;
517         }
518         adapter->hw_csum_good++;
519 }
520
521 /**
522  * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
523  * @hw: pointer to the HW structure
524  * @tail: address of tail descriptor register
525  * @i: value to write to tail descriptor register
526  *
527  * When updating the tail register, the ME could be accessing Host CSR
528  * registers at the same time.  Normally, this is handled in h/w by an
529  * arbiter but on some parts there is a bug that acknowledges Host accesses
530  * later than it should which could result in the descriptor register to
531  * have an incorrect value.  Workaround this by checking the FWSM register
532  * which has bit 24 set while ME is accessing Host CSR registers, wait
533  * if it is set and try again a number of times.
534  **/
535 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, u8 __iomem * tail,
536                                         unsigned int i)
537 {
538         unsigned int j = 0;
539
540         while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
541                (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
542                 udelay(50);
543
544         writel(i, tail);
545
546         if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
547                 return E1000_ERR_SWFW_SYNC;
548
549         return 0;
550 }
551
552 static void e1000e_update_rdt_wa(struct e1000_adapter *adapter, unsigned int i)
553 {
554         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->rx_ring->tail);
555         struct e1000_hw *hw = &adapter->hw;
556
557         if (e1000e_update_tail_wa(hw, tail, i)) {
558                 u32 rctl = er32(RCTL);
559                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
560                 e_err("ME firmware caused invalid RDT - resetting\n");
561                 schedule_work(&adapter->reset_task);
562         }
563 }
564
565 static void e1000e_update_tdt_wa(struct e1000_adapter *adapter, unsigned int i)
566 {
567         u8 __iomem *tail = (adapter->hw.hw_addr + adapter->tx_ring->tail);
568         struct e1000_hw *hw = &adapter->hw;
569
570         if (e1000e_update_tail_wa(hw, tail, i)) {
571                 u32 tctl = er32(TCTL);
572                 ew32(TCTL, tctl & ~E1000_TCTL_EN);
573                 e_err("ME firmware caused invalid TDT - resetting\n");
574                 schedule_work(&adapter->reset_task);
575         }
576 }
577
578 /**
579  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
580  * @adapter: address of board private structure
581  **/
582 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
583                                    int cleaned_count, gfp_t gfp)
584 {
585         struct net_device *netdev = adapter->netdev;
586         struct pci_dev *pdev = adapter->pdev;
587         struct e1000_ring *rx_ring = adapter->rx_ring;
588         struct e1000_rx_desc *rx_desc;
589         struct e1000_buffer *buffer_info;
590         struct sk_buff *skb;
591         unsigned int i;
592         unsigned int bufsz = adapter->rx_buffer_len;
593
594         i = rx_ring->next_to_use;
595         buffer_info = &rx_ring->buffer_info[i];
596
597         while (cleaned_count--) {
598                 skb = buffer_info->skb;
599                 if (skb) {
600                         skb_trim(skb, 0);
601                         goto map_skb;
602                 }
603
604                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
605                 if (!skb) {
606                         /* Better luck next round */
607                         adapter->alloc_rx_buff_failed++;
608                         break;
609                 }
610
611                 buffer_info->skb = skb;
612 map_skb:
613                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
614                                                   adapter->rx_buffer_len,
615                                                   DMA_FROM_DEVICE);
616                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
617                         dev_err(&pdev->dev, "Rx DMA map failed\n");
618                         adapter->rx_dma_failed++;
619                         break;
620                 }
621
622                 rx_desc = E1000_RX_DESC(*rx_ring, i);
623                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
624
625                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
626                         /*
627                          * Force memory writes to complete before letting h/w
628                          * know there are new descriptors to fetch.  (Only
629                          * applicable for weak-ordered memory model archs,
630                          * such as IA-64).
631                          */
632                         wmb();
633                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
634                                 e1000e_update_rdt_wa(adapter, i);
635                         else
636                                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
637                 }
638                 i++;
639                 if (i == rx_ring->count)
640                         i = 0;
641                 buffer_info = &rx_ring->buffer_info[i];
642         }
643
644         rx_ring->next_to_use = i;
645 }
646
647 /**
648  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
649  * @adapter: address of board private structure
650  **/
651 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
652                                       int cleaned_count, gfp_t gfp)
653 {
654         struct net_device *netdev = adapter->netdev;
655         struct pci_dev *pdev = adapter->pdev;
656         union e1000_rx_desc_packet_split *rx_desc;
657         struct e1000_ring *rx_ring = adapter->rx_ring;
658         struct e1000_buffer *buffer_info;
659         struct e1000_ps_page *ps_page;
660         struct sk_buff *skb;
661         unsigned int i, j;
662
663         i = rx_ring->next_to_use;
664         buffer_info = &rx_ring->buffer_info[i];
665
666         while (cleaned_count--) {
667                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
668
669                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
670                         ps_page = &buffer_info->ps_pages[j];
671                         if (j >= adapter->rx_ps_pages) {
672                                 /* all unused desc entries get hw null ptr */
673                                 rx_desc->read.buffer_addr[j + 1] =
674                                     ~cpu_to_le64(0);
675                                 continue;
676                         }
677                         if (!ps_page->page) {
678                                 ps_page->page = alloc_page(gfp);
679                                 if (!ps_page->page) {
680                                         adapter->alloc_rx_buff_failed++;
681                                         goto no_buffers;
682                                 }
683                                 ps_page->dma = dma_map_page(&pdev->dev,
684                                                             ps_page->page,
685                                                             0, PAGE_SIZE,
686                                                             DMA_FROM_DEVICE);
687                                 if (dma_mapping_error(&pdev->dev,
688                                                       ps_page->dma)) {
689                                         dev_err(&adapter->pdev->dev,
690                                                 "Rx DMA page map failed\n");
691                                         adapter->rx_dma_failed++;
692                                         goto no_buffers;
693                                 }
694                         }
695                         /*
696                          * Refresh the desc even if buffer_addrs
697                          * didn't change because each write-back
698                          * erases this info.
699                          */
700                         rx_desc->read.buffer_addr[j + 1] =
701                             cpu_to_le64(ps_page->dma);
702                 }
703
704                 skb = __netdev_alloc_skb_ip_align(netdev,
705                                                   adapter->rx_ps_bsize0,
706                                                   gfp);
707
708                 if (!skb) {
709                         adapter->alloc_rx_buff_failed++;
710                         break;
711                 }
712
713                 buffer_info->skb = skb;
714                 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
715                                                   adapter->rx_ps_bsize0,
716                                                   DMA_FROM_DEVICE);
717                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
718                         dev_err(&pdev->dev, "Rx DMA map failed\n");
719                         adapter->rx_dma_failed++;
720                         /* cleanup skb */
721                         dev_kfree_skb_any(skb);
722                         buffer_info->skb = NULL;
723                         break;
724                 }
725
726                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
727
728                 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
729                         /*
730                          * Force memory writes to complete before letting h/w
731                          * know there are new descriptors to fetch.  (Only
732                          * applicable for weak-ordered memory model archs,
733                          * such as IA-64).
734                          */
735                         wmb();
736                         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
737                                 e1000e_update_rdt_wa(adapter, i << 1);
738                         else
739                                 writel(i << 1,
740                                        adapter->hw.hw_addr + rx_ring->tail);
741                 }
742
743                 i++;
744                 if (i == rx_ring->count)
745                         i = 0;
746                 buffer_info = &rx_ring->buffer_info[i];
747         }
748
749 no_buffers:
750         rx_ring->next_to_use = i;
751 }
752
753 /**
754  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
755  * @adapter: address of board private structure
756  * @cleaned_count: number of buffers to allocate this pass
757  **/
758
759 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
760                                          int cleaned_count, gfp_t gfp)
761 {
762         struct net_device *netdev = adapter->netdev;
763         struct pci_dev *pdev = adapter->pdev;
764         struct e1000_rx_desc *rx_desc;
765         struct e1000_ring *rx_ring = adapter->rx_ring;
766         struct e1000_buffer *buffer_info;
767         struct sk_buff *skb;
768         unsigned int i;
769         unsigned int bufsz = 256 - 16 /* for skb_reserve */;
770
771         i = rx_ring->next_to_use;
772         buffer_info = &rx_ring->buffer_info[i];
773
774         while (cleaned_count--) {
775                 skb = buffer_info->skb;
776                 if (skb) {
777                         skb_trim(skb, 0);
778                         goto check_page;
779                 }
780
781                 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
782                 if (unlikely(!skb)) {
783                         /* Better luck next round */
784                         adapter->alloc_rx_buff_failed++;
785                         break;
786                 }
787
788                 buffer_info->skb = skb;
789 check_page:
790                 /* allocate a new page if necessary */
791                 if (!buffer_info->page) {
792                         buffer_info->page = alloc_page(gfp);
793                         if (unlikely(!buffer_info->page)) {
794                                 adapter->alloc_rx_buff_failed++;
795                                 break;
796                         }
797                 }
798
799                 if (!buffer_info->dma)
800                         buffer_info->dma = dma_map_page(&pdev->dev,
801                                                         buffer_info->page, 0,
802                                                         PAGE_SIZE,
803                                                         DMA_FROM_DEVICE);
804
805                 rx_desc = E1000_RX_DESC(*rx_ring, i);
806                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
807
808                 if (unlikely(++i == rx_ring->count))
809                         i = 0;
810                 buffer_info = &rx_ring->buffer_info[i];
811         }
812
813         if (likely(rx_ring->next_to_use != i)) {
814                 rx_ring->next_to_use = i;
815                 if (unlikely(i-- == 0))
816                         i = (rx_ring->count - 1);
817
818                 /* Force memory writes to complete before letting h/w
819                  * know there are new descriptors to fetch.  (Only
820                  * applicable for weak-ordered memory model archs,
821                  * such as IA-64). */
822                 wmb();
823                 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
824                         e1000e_update_rdt_wa(adapter, i);
825                 else
826                         writel(i, adapter->hw.hw_addr + rx_ring->tail);
827         }
828 }
829
830 /**
831  * e1000_clean_rx_irq - Send received data up the network stack; legacy
832  * @adapter: board private structure
833  *
834  * the return value indicates whether actual cleaning was done, there
835  * is no guarantee that everything was cleaned
836  **/
837 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
838                                int *work_done, int work_to_do)
839 {
840         struct net_device *netdev = adapter->netdev;
841         struct pci_dev *pdev = adapter->pdev;
842         struct e1000_hw *hw = &adapter->hw;
843         struct e1000_ring *rx_ring = adapter->rx_ring;
844         struct e1000_rx_desc *rx_desc, *next_rxd;
845         struct e1000_buffer *buffer_info, *next_buffer;
846         u32 length;
847         unsigned int i;
848         int cleaned_count = 0;
849         bool cleaned = 0;
850         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
851
852         i = rx_ring->next_to_clean;
853         rx_desc = E1000_RX_DESC(*rx_ring, i);
854         buffer_info = &rx_ring->buffer_info[i];
855
856         while (rx_desc->status & E1000_RXD_STAT_DD) {
857                 struct sk_buff *skb;
858                 u8 status;
859
860                 if (*work_done >= work_to_do)
861                         break;
862                 (*work_done)++;
863                 rmb();  /* read descriptor and rx_buffer_info after status DD */
864
865                 status = rx_desc->status;
866                 skb = buffer_info->skb;
867                 buffer_info->skb = NULL;
868
869                 prefetch(skb->data - NET_IP_ALIGN);
870
871                 i++;
872                 if (i == rx_ring->count)
873                         i = 0;
874                 next_rxd = E1000_RX_DESC(*rx_ring, i);
875                 prefetch(next_rxd);
876
877                 next_buffer = &rx_ring->buffer_info[i];
878
879                 cleaned = 1;
880                 cleaned_count++;
881                 dma_unmap_single(&pdev->dev,
882                                  buffer_info->dma,
883                                  adapter->rx_buffer_len,
884                                  DMA_FROM_DEVICE);
885                 buffer_info->dma = 0;
886
887                 length = le16_to_cpu(rx_desc->length);
888
889                 /*
890                  * !EOP means multiple descriptors were used to store a single
891                  * packet, if that's the case we need to toss it.  In fact, we
892                  * need to toss every packet with the EOP bit clear and the
893                  * next frame that _does_ have the EOP bit set, as it is by
894                  * definition only a frame fragment
895                  */
896                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
897                         adapter->flags2 |= FLAG2_IS_DISCARDING;
898
899                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
900                         /* All receives must fit into a single buffer */
901                         e_dbg("Receive packet consumed multiple buffers\n");
902                         /* recycle */
903                         buffer_info->skb = skb;
904                         if (status & E1000_RXD_STAT_EOP)
905                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
906                         goto next_desc;
907                 }
908
909                 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
910                         /* recycle */
911                         buffer_info->skb = skb;
912                         goto next_desc;
913                 }
914
915                 /* adjust length to remove Ethernet CRC */
916                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
917                         length -= 4;
918
919                 total_rx_bytes += length;
920                 total_rx_packets++;
921
922                 /*
923                  * code added for copybreak, this should improve
924                  * performance for small packets with large amounts
925                  * of reassembly being done in the stack
926                  */
927                 if (length < copybreak) {
928                         struct sk_buff *new_skb =
929                             netdev_alloc_skb_ip_align(netdev, length);
930                         if (new_skb) {
931                                 skb_copy_to_linear_data_offset(new_skb,
932                                                                -NET_IP_ALIGN,
933                                                                (skb->data -
934                                                                 NET_IP_ALIGN),
935                                                                (length +
936                                                                 NET_IP_ALIGN));
937                                 /* save the skb in buffer_info as good */
938                                 buffer_info->skb = skb;
939                                 skb = new_skb;
940                         }
941                         /* else just continue with the old one */
942                 }
943                 /* end copybreak code */
944                 skb_put(skb, length);
945
946                 /* Receive Checksum Offload */
947                 e1000_rx_checksum(adapter,
948                                   (u32)(status) |
949                                   ((u32)(rx_desc->errors) << 24),
950                                   le16_to_cpu(rx_desc->csum), skb);
951
952                 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
953
954 next_desc:
955                 rx_desc->status = 0;
956
957                 /* return some buffers to hardware, one at a time is too slow */
958                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
959                         adapter->alloc_rx_buf(adapter, cleaned_count,
960                                               GFP_ATOMIC);
961                         cleaned_count = 0;
962                 }
963
964                 /* use prefetched values */
965                 rx_desc = next_rxd;
966                 buffer_info = next_buffer;
967         }
968         rx_ring->next_to_clean = i;
969
970         cleaned_count = e1000_desc_unused(rx_ring);
971         if (cleaned_count)
972                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
973
974         adapter->total_rx_bytes += total_rx_bytes;
975         adapter->total_rx_packets += total_rx_packets;
976         return cleaned;
977 }
978
979 static void e1000_put_txbuf(struct e1000_adapter *adapter,
980                              struct e1000_buffer *buffer_info)
981 {
982         if (buffer_info->dma) {
983                 if (buffer_info->mapped_as_page)
984                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
985                                        buffer_info->length, DMA_TO_DEVICE);
986                 else
987                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
988                                          buffer_info->length, DMA_TO_DEVICE);
989                 buffer_info->dma = 0;
990         }
991         if (buffer_info->skb) {
992                 dev_kfree_skb_any(buffer_info->skb);
993                 buffer_info->skb = NULL;
994         }
995         buffer_info->time_stamp = 0;
996 }
997
998 static void e1000_print_hw_hang(struct work_struct *work)
999 {
1000         struct e1000_adapter *adapter = container_of(work,
1001                                                      struct e1000_adapter,
1002                                                      print_hang_task);
1003         struct e1000_ring *tx_ring = adapter->tx_ring;
1004         unsigned int i = tx_ring->next_to_clean;
1005         unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1006         struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1007         struct e1000_hw *hw = &adapter->hw;
1008         u16 phy_status, phy_1000t_status, phy_ext_status;
1009         u16 pci_status;
1010
1011         if (test_bit(__E1000_DOWN, &adapter->state))
1012                 return;
1013
1014         e1e_rphy(hw, PHY_STATUS, &phy_status);
1015         e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1016         e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1017
1018         pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1019
1020         /* detected Hardware unit hang */
1021         e_err("Detected Hardware Unit Hang:\n"
1022               "  TDH                  <%x>\n"
1023               "  TDT                  <%x>\n"
1024               "  next_to_use          <%x>\n"
1025               "  next_to_clean        <%x>\n"
1026               "buffer_info[next_to_clean]:\n"
1027               "  time_stamp           <%lx>\n"
1028               "  next_to_watch        <%x>\n"
1029               "  jiffies              <%lx>\n"
1030               "  next_to_watch.status <%x>\n"
1031               "MAC Status             <%x>\n"
1032               "PHY Status             <%x>\n"
1033               "PHY 1000BASE-T Status  <%x>\n"
1034               "PHY Extended Status    <%x>\n"
1035               "PCI Status             <%x>\n",
1036               readl(adapter->hw.hw_addr + tx_ring->head),
1037               readl(adapter->hw.hw_addr + tx_ring->tail),
1038               tx_ring->next_to_use,
1039               tx_ring->next_to_clean,
1040               tx_ring->buffer_info[eop].time_stamp,
1041               eop,
1042               jiffies,
1043               eop_desc->upper.fields.status,
1044               er32(STATUS),
1045               phy_status,
1046               phy_1000t_status,
1047               phy_ext_status,
1048               pci_status);
1049 }
1050
1051 /**
1052  * e1000_clean_tx_irq - Reclaim resources after transmit completes
1053  * @adapter: board private structure
1054  *
1055  * the return value indicates whether actual cleaning was done, there
1056  * is no guarantee that everything was cleaned
1057  **/
1058 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
1059 {
1060         struct net_device *netdev = adapter->netdev;
1061         struct e1000_hw *hw = &adapter->hw;
1062         struct e1000_ring *tx_ring = adapter->tx_ring;
1063         struct e1000_tx_desc *tx_desc, *eop_desc;
1064         struct e1000_buffer *buffer_info;
1065         unsigned int i, eop;
1066         unsigned int count = 0;
1067         unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1068
1069         i = tx_ring->next_to_clean;
1070         eop = tx_ring->buffer_info[i].next_to_watch;
1071         eop_desc = E1000_TX_DESC(*tx_ring, eop);
1072
1073         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1074                (count < tx_ring->count)) {
1075                 bool cleaned = false;
1076                 rmb(); /* read buffer_info after eop_desc */
1077                 for (; !cleaned; count++) {
1078                         tx_desc = E1000_TX_DESC(*tx_ring, i);
1079                         buffer_info = &tx_ring->buffer_info[i];
1080                         cleaned = (i == eop);
1081
1082                         if (cleaned) {
1083                                 total_tx_packets += buffer_info->segs;
1084                                 total_tx_bytes += buffer_info->bytecount;
1085                         }
1086
1087                         e1000_put_txbuf(adapter, buffer_info);
1088                         tx_desc->upper.data = 0;
1089
1090                         i++;
1091                         if (i == tx_ring->count)
1092                                 i = 0;
1093                 }
1094
1095                 if (i == tx_ring->next_to_use)
1096                         break;
1097                 eop = tx_ring->buffer_info[i].next_to_watch;
1098                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1099         }
1100
1101         tx_ring->next_to_clean = i;
1102
1103 #define TX_WAKE_THRESHOLD 32
1104         if (count && netif_carrier_ok(netdev) &&
1105             e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1106                 /* Make sure that anybody stopping the queue after this
1107                  * sees the new next_to_clean.
1108                  */
1109                 smp_mb();
1110
1111                 if (netif_queue_stopped(netdev) &&
1112                     !(test_bit(__E1000_DOWN, &adapter->state))) {
1113                         netif_wake_queue(netdev);
1114                         ++adapter->restart_queue;
1115                 }
1116         }
1117
1118         if (adapter->detect_tx_hung) {
1119                 /*
1120                  * Detect a transmit hang in hardware, this serializes the
1121                  * check with the clearing of time_stamp and movement of i
1122                  */
1123                 adapter->detect_tx_hung = 0;
1124                 if (tx_ring->buffer_info[i].time_stamp &&
1125                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1126                                + (adapter->tx_timeout_factor * HZ)) &&
1127                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1128                         schedule_work(&adapter->print_hang_task);
1129                         netif_stop_queue(netdev);
1130                 }
1131         }
1132         adapter->total_tx_bytes += total_tx_bytes;
1133         adapter->total_tx_packets += total_tx_packets;
1134         return count < tx_ring->count;
1135 }
1136
1137 /**
1138  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1139  * @adapter: board private structure
1140  *
1141  * the return value indicates whether actual cleaning was done, there
1142  * is no guarantee that everything was cleaned
1143  **/
1144 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1145                                   int *work_done, int work_to_do)
1146 {
1147         struct e1000_hw *hw = &adapter->hw;
1148         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1149         struct net_device *netdev = adapter->netdev;
1150         struct pci_dev *pdev = adapter->pdev;
1151         struct e1000_ring *rx_ring = adapter->rx_ring;
1152         struct e1000_buffer *buffer_info, *next_buffer;
1153         struct e1000_ps_page *ps_page;
1154         struct sk_buff *skb;
1155         unsigned int i, j;
1156         u32 length, staterr;
1157         int cleaned_count = 0;
1158         bool cleaned = 0;
1159         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1160
1161         i = rx_ring->next_to_clean;
1162         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1163         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1164         buffer_info = &rx_ring->buffer_info[i];
1165
1166         while (staterr & E1000_RXD_STAT_DD) {
1167                 if (*work_done >= work_to_do)
1168                         break;
1169                 (*work_done)++;
1170                 skb = buffer_info->skb;
1171                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1172
1173                 /* in the packet split case this is header only */
1174                 prefetch(skb->data - NET_IP_ALIGN);
1175
1176                 i++;
1177                 if (i == rx_ring->count)
1178                         i = 0;
1179                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1180                 prefetch(next_rxd);
1181
1182                 next_buffer = &rx_ring->buffer_info[i];
1183
1184                 cleaned = 1;
1185                 cleaned_count++;
1186                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1187                                  adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1188                 buffer_info->dma = 0;
1189
1190                 /* see !EOP comment in other Rx routine */
1191                 if (!(staterr & E1000_RXD_STAT_EOP))
1192                         adapter->flags2 |= FLAG2_IS_DISCARDING;
1193
1194                 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1195                         e_dbg("Packet Split buffers didn't pick up the full "
1196                               "packet\n");
1197                         dev_kfree_skb_irq(skb);
1198                         if (staterr & E1000_RXD_STAT_EOP)
1199                                 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1200                         goto next_desc;
1201                 }
1202
1203                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1204                         dev_kfree_skb_irq(skb);
1205                         goto next_desc;
1206                 }
1207
1208                 length = le16_to_cpu(rx_desc->wb.middle.length0);
1209
1210                 if (!length) {
1211                         e_dbg("Last part of the packet spanning multiple "
1212                               "descriptors\n");
1213                         dev_kfree_skb_irq(skb);
1214                         goto next_desc;
1215                 }
1216
1217                 /* Good Receive */
1218                 skb_put(skb, length);
1219
1220                 {
1221                 /*
1222                  * this looks ugly, but it seems compiler issues make it
1223                  * more efficient than reusing j
1224                  */
1225                 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1226
1227                 /*
1228                  * page alloc/put takes too long and effects small packet
1229                  * throughput, so unsplit small packets and save the alloc/put
1230                  * only valid in softirq (napi) context to call kmap_*
1231                  */
1232                 if (l1 && (l1 <= copybreak) &&
1233                     ((length + l1) <= adapter->rx_ps_bsize0)) {
1234                         u8 *vaddr;
1235
1236                         ps_page = &buffer_info->ps_pages[0];
1237
1238                         /*
1239                          * there is no documentation about how to call
1240                          * kmap_atomic, so we can't hold the mapping
1241                          * very long
1242                          */
1243                         dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1244                                                 PAGE_SIZE, DMA_FROM_DEVICE);
1245                         vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1246                         memcpy(skb_tail_pointer(skb), vaddr, l1);
1247                         kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1248                         dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1249                                                    PAGE_SIZE, DMA_FROM_DEVICE);
1250
1251                         /* remove the CRC */
1252                         if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1253                                 l1 -= 4;
1254
1255                         skb_put(skb, l1);
1256                         goto copydone;
1257                 } /* if */
1258                 }
1259
1260                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1261                         length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1262                         if (!length)
1263                                 break;
1264
1265                         ps_page = &buffer_info->ps_pages[j];
1266                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1267                                        DMA_FROM_DEVICE);
1268                         ps_page->dma = 0;
1269                         skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1270                         ps_page->page = NULL;
1271                         skb->len += length;
1272                         skb->data_len += length;
1273                         skb->truesize += length;
1274                 }
1275
1276                 /* strip the ethernet crc, problem is we're using pages now so
1277                  * this whole operation can get a little cpu intensive
1278                  */
1279                 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1280                         pskb_trim(skb, skb->len - 4);
1281
1282 copydone:
1283                 total_rx_bytes += skb->len;
1284                 total_rx_packets++;
1285
1286                 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1287                         rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1288
1289                 if (rx_desc->wb.upper.header_status &
1290                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1291                         adapter->rx_hdr_split++;
1292
1293                 e1000_receive_skb(adapter, netdev, skb,
1294                                   staterr, rx_desc->wb.middle.vlan);
1295
1296 next_desc:
1297                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1298                 buffer_info->skb = NULL;
1299
1300                 /* return some buffers to hardware, one at a time is too slow */
1301                 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1302                         adapter->alloc_rx_buf(adapter, cleaned_count,
1303                                               GFP_ATOMIC);
1304                         cleaned_count = 0;
1305                 }
1306
1307                 /* use prefetched values */
1308                 rx_desc = next_rxd;
1309                 buffer_info = next_buffer;
1310
1311                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1312         }
1313         rx_ring->next_to_clean = i;
1314
1315         cleaned_count = e1000_desc_unused(rx_ring);
1316         if (cleaned_count)
1317                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1318
1319         adapter->total_rx_bytes += total_rx_bytes;
1320         adapter->total_rx_packets += total_rx_packets;
1321         return cleaned;
1322 }
1323
1324 /**
1325  * e1000_consume_page - helper function
1326  **/
1327 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1328                                u16 length)
1329 {
1330         bi->page = NULL;
1331         skb->len += length;
1332         skb->data_len += length;
1333         skb->truesize += length;
1334 }
1335
1336 /**
1337  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1338  * @adapter: board private structure
1339  *
1340  * the return value indicates whether actual cleaning was done, there
1341  * is no guarantee that everything was cleaned
1342  **/
1343
1344 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1345                                      int *work_done, int work_to_do)
1346 {
1347         struct net_device *netdev = adapter->netdev;
1348         struct pci_dev *pdev = adapter->pdev;
1349         struct e1000_ring *rx_ring = adapter->rx_ring;
1350         struct e1000_rx_desc *rx_desc, *next_rxd;
1351         struct e1000_buffer *buffer_info, *next_buffer;
1352         u32 length;
1353         unsigned int i;
1354         int cleaned_count = 0;
1355         bool cleaned = false;
1356         unsigned int total_rx_bytes=0, total_rx_packets=0;
1357
1358         i = rx_ring->next_to_clean;
1359         rx_desc = E1000_RX_DESC(*rx_ring, i);
1360         buffer_info = &rx_ring->buffer_info[i];
1361
1362         while (rx_desc->status & E1000_RXD_STAT_DD) {
1363                 struct sk_buff *skb;
1364                 u8 status;
1365
1366                 if (*work_done >= work_to_do)
1367                         break;
1368                 (*work_done)++;
1369                 rmb();  /* read descriptor and rx_buffer_info after status DD */
1370
1371                 status = rx_desc->status;
1372                 skb = buffer_info->skb;
1373                 buffer_info->skb = NULL;
1374
1375                 ++i;
1376                 if (i == rx_ring->count)
1377                         i = 0;
1378                 next_rxd = E1000_RX_DESC(*rx_ring, i);
1379                 prefetch(next_rxd);
1380
1381                 next_buffer = &rx_ring->buffer_info[i];
1382
1383                 cleaned = true;
1384                 cleaned_count++;
1385                 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1386                                DMA_FROM_DEVICE);
1387                 buffer_info->dma = 0;
1388
1389                 length = le16_to_cpu(rx_desc->length);
1390
1391                 /* errors is only valid for DD + EOP descriptors */
1392                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1393                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1394                                 /* recycle both page and skb */
1395                                 buffer_info->skb = skb;
1396                                 /* an error means any chain goes out the window
1397                                  * too */
1398                                 if (rx_ring->rx_skb_top)
1399                                         dev_kfree_skb_irq(rx_ring->rx_skb_top);
1400                                 rx_ring->rx_skb_top = NULL;
1401                                 goto next_desc;
1402                 }
1403
1404 #define rxtop (rx_ring->rx_skb_top)
1405                 if (!(status & E1000_RXD_STAT_EOP)) {
1406                         /* this descriptor is only the beginning (or middle) */
1407                         if (!rxtop) {
1408                                 /* this is the beginning of a chain */
1409                                 rxtop = skb;
1410                                 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1411                                                    0, length);
1412                         } else {
1413                                 /* this is the middle of a chain */
1414                                 skb_fill_page_desc(rxtop,
1415                                     skb_shinfo(rxtop)->nr_frags,
1416                                     buffer_info->page, 0, length);
1417                                 /* re-use the skb, only consumed the page */
1418                                 buffer_info->skb = skb;
1419                         }
1420                         e1000_consume_page(buffer_info, rxtop, length);
1421                         goto next_desc;
1422                 } else {
1423                         if (rxtop) {
1424                                 /* end of the chain */
1425                                 skb_fill_page_desc(rxtop,
1426                                     skb_shinfo(rxtop)->nr_frags,
1427                                     buffer_info->page, 0, length);
1428                                 /* re-use the current skb, we only consumed the
1429                                  * page */
1430                                 buffer_info->skb = skb;
1431                                 skb = rxtop;
1432                                 rxtop = NULL;
1433                                 e1000_consume_page(buffer_info, skb, length);
1434                         } else {
1435                                 /* no chain, got EOP, this buf is the packet
1436                                  * copybreak to save the put_page/alloc_page */
1437                                 if (length <= copybreak &&
1438                                     skb_tailroom(skb) >= length) {
1439                                         u8 *vaddr;
1440                                         vaddr = kmap_atomic(buffer_info->page,
1441                                                            KM_SKB_DATA_SOFTIRQ);
1442                                         memcpy(skb_tail_pointer(skb), vaddr,
1443                                                length);
1444                                         kunmap_atomic(vaddr,
1445                                                       KM_SKB_DATA_SOFTIRQ);
1446                                         /* re-use the page, so don't erase
1447                                          * buffer_info->page */
1448                                         skb_put(skb, length);
1449                                 } else {
1450                                         skb_fill_page_desc(skb, 0,
1451                                                            buffer_info->page, 0,
1452                                                            length);
1453                                         e1000_consume_page(buffer_info, skb,
1454                                                            length);
1455                                 }
1456                         }
1457                 }
1458
1459                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1460                 e1000_rx_checksum(adapter,
1461                                   (u32)(status) |
1462                                   ((u32)(rx_desc->errors) << 24),
1463                                   le16_to_cpu(rx_desc->csum), skb);
1464
1465                 /* probably a little skewed due to removing CRC */
1466                 total_rx_bytes += skb->len;
1467                 total_rx_packets++;
1468
1469                 /* eth type trans needs skb->data to point to something */
1470                 if (!pskb_may_pull(skb, ETH_HLEN)) {
1471                         e_err("pskb_may_pull failed.\n");
1472                         dev_kfree_skb_irq(skb);
1473                         goto next_desc;
1474                 }
1475
1476                 e1000_receive_skb(adapter, netdev, skb, status,
1477                                   rx_desc->special);
1478
1479 next_desc:
1480                 rx_desc->status = 0;
1481
1482                 /* return some buffers to hardware, one at a time is too slow */
1483                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1484                         adapter->alloc_rx_buf(adapter, cleaned_count,
1485                                               GFP_ATOMIC);
1486                         cleaned_count = 0;
1487                 }
1488
1489                 /* use prefetched values */
1490                 rx_desc = next_rxd;
1491                 buffer_info = next_buffer;
1492         }
1493         rx_ring->next_to_clean = i;
1494
1495         cleaned_count = e1000_desc_unused(rx_ring);
1496         if (cleaned_count)
1497                 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1498
1499         adapter->total_rx_bytes += total_rx_bytes;
1500         adapter->total_rx_packets += total_rx_packets;
1501         return cleaned;
1502 }
1503
1504 /**
1505  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1506  * @adapter: board private structure
1507  **/
1508 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1509 {
1510         struct e1000_ring *rx_ring = adapter->rx_ring;
1511         struct e1000_buffer *buffer_info;
1512         struct e1000_ps_page *ps_page;
1513         struct pci_dev *pdev = adapter->pdev;
1514         unsigned int i, j;
1515
1516         /* Free all the Rx ring sk_buffs */
1517         for (i = 0; i < rx_ring->count; i++) {
1518                 buffer_info = &rx_ring->buffer_info[i];
1519                 if (buffer_info->dma) {
1520                         if (adapter->clean_rx == e1000_clean_rx_irq)
1521                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1522                                                  adapter->rx_buffer_len,
1523                                                  DMA_FROM_DEVICE);
1524                         else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1525                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
1526                                                PAGE_SIZE,
1527                                                DMA_FROM_DEVICE);
1528                         else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1529                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
1530                                                  adapter->rx_ps_bsize0,
1531                                                  DMA_FROM_DEVICE);
1532                         buffer_info->dma = 0;
1533                 }
1534
1535                 if (buffer_info->page) {
1536                         put_page(buffer_info->page);
1537                         buffer_info->page = NULL;
1538                 }
1539
1540                 if (buffer_info->skb) {
1541                         dev_kfree_skb(buffer_info->skb);
1542                         buffer_info->skb = NULL;
1543                 }
1544
1545                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1546                         ps_page = &buffer_info->ps_pages[j];
1547                         if (!ps_page->page)
1548                                 break;
1549                         dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1550                                        DMA_FROM_DEVICE);
1551                         ps_page->dma = 0;
1552                         put_page(ps_page->page);
1553                         ps_page->page = NULL;
1554                 }
1555         }
1556
1557         /* there also may be some cached data from a chained receive */
1558         if (rx_ring->rx_skb_top) {
1559                 dev_kfree_skb(rx_ring->rx_skb_top);
1560                 rx_ring->rx_skb_top = NULL;
1561         }
1562
1563         /* Zero out the descriptor ring */
1564         memset(rx_ring->desc, 0, rx_ring->size);
1565
1566         rx_ring->next_to_clean = 0;
1567         rx_ring->next_to_use = 0;
1568         adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1569
1570         writel(0, adapter->hw.hw_addr + rx_ring->head);
1571         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1572 }
1573
1574 static void e1000e_downshift_workaround(struct work_struct *work)
1575 {
1576         struct e1000_adapter *adapter = container_of(work,
1577                                         struct e1000_adapter, downshift_task);
1578
1579         if (test_bit(__E1000_DOWN, &adapter->state))
1580                 return;
1581
1582         e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1583 }
1584
1585 /**
1586  * e1000_intr_msi - Interrupt Handler
1587  * @irq: interrupt number
1588  * @data: pointer to a network interface device structure
1589  **/
1590 static irqreturn_t e1000_intr_msi(int irq, void *data)
1591 {
1592         struct net_device *netdev = data;
1593         struct e1000_adapter *adapter = netdev_priv(netdev);
1594         struct e1000_hw *hw = &adapter->hw;
1595         u32 icr = er32(ICR);
1596
1597         /*
1598          * read ICR disables interrupts using IAM
1599          */
1600
1601         if (icr & E1000_ICR_LSC) {
1602                 hw->mac.get_link_status = 1;
1603                 /*
1604                  * ICH8 workaround-- Call gig speed drop workaround on cable
1605                  * disconnect (LSC) before accessing any PHY registers
1606                  */
1607                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1608                     (!(er32(STATUS) & E1000_STATUS_LU)))
1609                         schedule_work(&adapter->downshift_task);
1610
1611                 /*
1612                  * 80003ES2LAN workaround-- For packet buffer work-around on
1613                  * link down event; disable receives here in the ISR and reset
1614                  * adapter in watchdog
1615                  */
1616                 if (netif_carrier_ok(netdev) &&
1617                     adapter->flags & FLAG_RX_NEEDS_RESTART) {
1618                         /* disable receives */
1619                         u32 rctl = er32(RCTL);
1620                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1621                         adapter->flags |= FLAG_RX_RESTART_NOW;
1622                 }
1623                 /* guard against interrupt when we're going down */
1624                 if (!test_bit(__E1000_DOWN, &adapter->state))
1625                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1626         }
1627
1628         if (napi_schedule_prep(&adapter->napi)) {
1629                 adapter->total_tx_bytes = 0;
1630                 adapter->total_tx_packets = 0;
1631                 adapter->total_rx_bytes = 0;
1632                 adapter->total_rx_packets = 0;
1633                 __napi_schedule(&adapter->napi);
1634         }
1635
1636         return IRQ_HANDLED;
1637 }
1638
1639 /**
1640  * e1000_intr - Interrupt Handler
1641  * @irq: interrupt number
1642  * @data: pointer to a network interface device structure
1643  **/
1644 static irqreturn_t e1000_intr(int irq, void *data)
1645 {
1646         struct net_device *netdev = data;
1647         struct e1000_adapter *adapter = netdev_priv(netdev);
1648         struct e1000_hw *hw = &adapter->hw;
1649         u32 rctl, icr = er32(ICR);
1650
1651         if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1652                 return IRQ_NONE;  /* Not our interrupt */
1653
1654         /*
1655          * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1656          * not set, then the adapter didn't send an interrupt
1657          */
1658         if (!(icr & E1000_ICR_INT_ASSERTED))
1659                 return IRQ_NONE;
1660
1661         /*
1662          * Interrupt Auto-Mask...upon reading ICR,
1663          * interrupts are masked.  No need for the
1664          * IMC write
1665          */
1666
1667         if (icr & E1000_ICR_LSC) {
1668                 hw->mac.get_link_status = 1;
1669                 /*
1670                  * ICH8 workaround-- Call gig speed drop workaround on cable
1671                  * disconnect (LSC) before accessing any PHY registers
1672                  */
1673                 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1674                     (!(er32(STATUS) & E1000_STATUS_LU)))
1675                         schedule_work(&adapter->downshift_task);
1676
1677                 /*
1678                  * 80003ES2LAN workaround--
1679                  * For packet buffer work-around on link down event;
1680                  * disable receives here in the ISR and
1681                  * reset adapter in watchdog
1682                  */
1683                 if (netif_carrier_ok(netdev) &&
1684                     (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1685                         /* disable receives */
1686                         rctl = er32(RCTL);
1687                         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1688                         adapter->flags |= FLAG_RX_RESTART_NOW;
1689                 }
1690                 /* guard against interrupt when we're going down */
1691                 if (!test_bit(__E1000_DOWN, &adapter->state))
1692                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1693         }
1694
1695         if (napi_schedule_prep(&adapter->napi)) {
1696                 adapter->total_tx_bytes = 0;
1697                 adapter->total_tx_packets = 0;
1698                 adapter->total_rx_bytes = 0;
1699                 adapter->total_rx_packets = 0;
1700                 __napi_schedule(&adapter->napi);
1701         }
1702
1703         return IRQ_HANDLED;
1704 }
1705
1706 static irqreturn_t e1000_msix_other(int irq, void *data)
1707 {
1708         struct net_device *netdev = data;
1709         struct e1000_adapter *adapter = netdev_priv(netdev);
1710         struct e1000_hw *hw = &adapter->hw;
1711         u32 icr = er32(ICR);
1712
1713         if (!(icr & E1000_ICR_INT_ASSERTED)) {
1714                 if (!test_bit(__E1000_DOWN, &adapter->state))
1715                         ew32(IMS, E1000_IMS_OTHER);
1716                 return IRQ_NONE;
1717         }
1718
1719         if (icr & adapter->eiac_mask)
1720                 ew32(ICS, (icr & adapter->eiac_mask));
1721
1722         if (icr & E1000_ICR_OTHER) {
1723                 if (!(icr & E1000_ICR_LSC))
1724                         goto no_link_interrupt;
1725                 hw->mac.get_link_status = 1;
1726                 /* guard against interrupt when we're going down */
1727                 if (!test_bit(__E1000_DOWN, &adapter->state))
1728                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
1729         }
1730
1731 no_link_interrupt:
1732         if (!test_bit(__E1000_DOWN, &adapter->state))
1733                 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1734
1735         return IRQ_HANDLED;
1736 }
1737
1738
1739 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1740 {
1741         struct net_device *netdev = data;
1742         struct e1000_adapter *adapter = netdev_priv(netdev);
1743         struct e1000_hw *hw = &adapter->hw;
1744         struct e1000_ring *tx_ring = adapter->tx_ring;
1745
1746
1747         adapter->total_tx_bytes = 0;
1748         adapter->total_tx_packets = 0;
1749
1750         if (!e1000_clean_tx_irq(adapter))
1751                 /* Ring was not completely cleaned, so fire another interrupt */
1752                 ew32(ICS, tx_ring->ims_val);
1753
1754         return IRQ_HANDLED;
1755 }
1756
1757 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1758 {
1759         struct net_device *netdev = data;
1760         struct e1000_adapter *adapter = netdev_priv(netdev);
1761
1762         /* Write the ITR value calculated at the end of the
1763          * previous interrupt.
1764          */
1765         if (adapter->rx_ring->set_itr) {
1766                 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1767                        adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1768                 adapter->rx_ring->set_itr = 0;
1769         }
1770
1771         if (napi_schedule_prep(&adapter->napi)) {
1772                 adapter->total_rx_bytes = 0;
1773                 adapter->total_rx_packets = 0;
1774                 __napi_schedule(&adapter->napi);
1775         }
1776         return IRQ_HANDLED;
1777 }
1778
1779 /**
1780  * e1000_configure_msix - Configure MSI-X hardware
1781  *
1782  * e1000_configure_msix sets up the hardware to properly
1783  * generate MSI-X interrupts.
1784  **/
1785 static void e1000_configure_msix(struct e1000_adapter *adapter)
1786 {
1787         struct e1000_hw *hw = &adapter->hw;
1788         struct e1000_ring *rx_ring = adapter->rx_ring;
1789         struct e1000_ring *tx_ring = adapter->tx_ring;
1790         int vector = 0;
1791         u32 ctrl_ext, ivar = 0;
1792
1793         adapter->eiac_mask = 0;
1794
1795         /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1796         if (hw->mac.type == e1000_82574) {
1797                 u32 rfctl = er32(RFCTL);
1798                 rfctl |= E1000_RFCTL_ACK_DIS;
1799                 ew32(RFCTL, rfctl);
1800         }
1801
1802 #define E1000_IVAR_INT_ALLOC_VALID      0x8
1803         /* Configure Rx vector */
1804         rx_ring->ims_val = E1000_IMS_RXQ0;
1805         adapter->eiac_mask |= rx_ring->ims_val;
1806         if (rx_ring->itr_val)
1807                 writel(1000000000 / (rx_ring->itr_val * 256),
1808                        hw->hw_addr + rx_ring->itr_register);
1809         else
1810                 writel(1, hw->hw_addr + rx_ring->itr_register);
1811         ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1812
1813         /* Configure Tx vector */
1814         tx_ring->ims_val = E1000_IMS_TXQ0;
1815         vector++;
1816         if (tx_ring->itr_val)
1817                 writel(1000000000 / (tx_ring->itr_val * 256),
1818                        hw->hw_addr + tx_ring->itr_register);
1819         else
1820                 writel(1, hw->hw_addr + tx_ring->itr_register);
1821         adapter->eiac_mask |= tx_ring->ims_val;
1822         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1823
1824         /* set vector for Other Causes, e.g. link changes */
1825         vector++;
1826         ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1827         if (rx_ring->itr_val)
1828                 writel(1000000000 / (rx_ring->itr_val * 256),
1829                        hw->hw_addr + E1000_EITR_82574(vector));
1830         else
1831                 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1832
1833         /* Cause Tx interrupts on every write back */
1834         ivar |= (1 << 31);
1835
1836         ew32(IVAR, ivar);
1837
1838         /* enable MSI-X PBA support */
1839         ctrl_ext = er32(CTRL_EXT);
1840         ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1841
1842         /* Auto-Mask Other interrupts upon ICR read */
1843 #define E1000_EIAC_MASK_82574   0x01F00000
1844         ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1845         ctrl_ext |= E1000_CTRL_EXT_EIAME;
1846         ew32(CTRL_EXT, ctrl_ext);
1847         e1e_flush();
1848 }
1849
1850 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1851 {
1852         if (adapter->msix_entries) {
1853                 pci_disable_msix(adapter->pdev);
1854                 kfree(adapter->msix_entries);
1855                 adapter->msix_entries = NULL;
1856         } else if (adapter->flags & FLAG_MSI_ENABLED) {
1857                 pci_disable_msi(adapter->pdev);
1858                 adapter->flags &= ~FLAG_MSI_ENABLED;
1859         }
1860 }
1861
1862 /**
1863  * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1864  *
1865  * Attempt to configure interrupts using the best available
1866  * capabilities of the hardware and kernel.
1867  **/
1868 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1869 {
1870         int err;
1871         int i;
1872
1873         switch (adapter->int_mode) {
1874         case E1000E_INT_MODE_MSIX:
1875                 if (adapter->flags & FLAG_HAS_MSIX) {
1876                         adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1877                         adapter->msix_entries = kcalloc(adapter->num_vectors,
1878                                                       sizeof(struct msix_entry),
1879                                                       GFP_KERNEL);
1880                         if (adapter->msix_entries) {
1881                                 for (i = 0; i < adapter->num_vectors; i++)
1882                                         adapter->msix_entries[i].entry = i;
1883
1884                                 err = pci_enable_msix(adapter->pdev,
1885                                                       adapter->msix_entries,
1886                                                       adapter->num_vectors);
1887                                 if (err == 0)
1888                                         return;
1889                         }
1890                         /* MSI-X failed, so fall through and try MSI */
1891                         e_err("Failed to initialize MSI-X interrupts.  "
1892                               "Falling back to MSI interrupts.\n");
1893                         e1000e_reset_interrupt_capability(adapter);
1894                 }
1895                 adapter->int_mode = E1000E_INT_MODE_MSI;
1896                 /* Fall through */
1897         case E1000E_INT_MODE_MSI:
1898                 if (!pci_enable_msi(adapter->pdev)) {
1899                         adapter->flags |= FLAG_MSI_ENABLED;
1900                 } else {
1901                         adapter->int_mode = E1000E_INT_MODE_LEGACY;
1902                         e_err("Failed to initialize MSI interrupts.  Falling "
1903                               "back to legacy interrupts.\n");
1904                 }
1905                 /* Fall through */
1906         case E1000E_INT_MODE_LEGACY:
1907                 /* Don't do anything; this is the system default */
1908                 break;
1909         }
1910
1911         /* store the number of vectors being used */
1912         adapter->num_vectors = 1;
1913 }
1914
1915 /**
1916  * e1000_request_msix - Initialize MSI-X interrupts
1917  *
1918  * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1919  * kernel.
1920  **/
1921 static int e1000_request_msix(struct e1000_adapter *adapter)
1922 {
1923         struct net_device *netdev = adapter->netdev;
1924         int err = 0, vector = 0;
1925
1926         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1927                 snprintf(adapter->rx_ring->name,
1928                          sizeof(adapter->rx_ring->name) - 1,
1929                          "%s-rx-0", netdev->name);
1930         else
1931                 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1932         err = request_irq(adapter->msix_entries[vector].vector,
1933                           e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1934                           netdev);
1935         if (err)
1936                 goto out;
1937         adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1938         adapter->rx_ring->itr_val = adapter->itr;
1939         vector++;
1940
1941         if (strlen(netdev->name) < (IFNAMSIZ - 5))
1942                 snprintf(adapter->tx_ring->name,
1943                          sizeof(adapter->tx_ring->name) - 1,
1944                          "%s-tx-0", netdev->name);
1945         else
1946                 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1947         err = request_irq(adapter->msix_entries[vector].vector,
1948                           e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1949                           netdev);
1950         if (err)
1951                 goto out;
1952         adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1953         adapter->tx_ring->itr_val = adapter->itr;
1954         vector++;
1955
1956         err = request_irq(adapter->msix_entries[vector].vector,
1957                           e1000_msix_other, 0, netdev->name, netdev);
1958         if (err)
1959                 goto out;
1960
1961         e1000_configure_msix(adapter);
1962         return 0;
1963 out:
1964         return err;
1965 }
1966
1967 /**
1968  * e1000_request_irq - initialize interrupts
1969  *
1970  * Attempts to configure interrupts using the best available
1971  * capabilities of the hardware and kernel.
1972  **/
1973 static int e1000_request_irq(struct e1000_adapter *adapter)
1974 {
1975         struct net_device *netdev = adapter->netdev;
1976         int err;
1977
1978         if (adapter->msix_entries) {
1979                 err = e1000_request_msix(adapter);
1980                 if (!err)
1981                         return err;
1982                 /* fall back to MSI */
1983                 e1000e_reset_interrupt_capability(adapter);
1984                 adapter->int_mode = E1000E_INT_MODE_MSI;
1985                 e1000e_set_interrupt_capability(adapter);
1986         }
1987         if (adapter->flags & FLAG_MSI_ENABLED) {
1988                 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1989                                   netdev->name, netdev);
1990                 if (!err)
1991                         return err;
1992
1993                 /* fall back to legacy interrupt */
1994                 e1000e_reset_interrupt_capability(adapter);
1995                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1996         }
1997
1998         err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1999                           netdev->name, netdev);
2000         if (err)
2001                 e_err("Unable to allocate interrupt, Error: %d\n", err);
2002
2003         return err;
2004 }
2005
2006 static void e1000_free_irq(struct e1000_adapter *adapter)
2007 {
2008         struct net_device *netdev = adapter->netdev;
2009
2010         if (adapter->msix_entries) {
2011                 int vector = 0;
2012
2013                 free_irq(adapter->msix_entries[vector].vector, netdev);
2014                 vector++;
2015
2016                 free_irq(adapter->msix_entries[vector].vector, netdev);
2017                 vector++;
2018
2019                 /* Other Causes interrupt vector */
2020                 free_irq(adapter->msix_entries[vector].vector, netdev);
2021                 return;
2022         }
2023
2024         free_irq(adapter->pdev->irq, netdev);
2025 }
2026
2027 /**
2028  * e1000_irq_disable - Mask off interrupt generation on the NIC
2029  **/
2030 static void e1000_irq_disable(struct e1000_adapter *adapter)
2031 {
2032         struct e1000_hw *hw = &adapter->hw;
2033
2034         ew32(IMC, ~0);
2035         if (adapter->msix_entries)
2036                 ew32(EIAC_82574, 0);
2037         e1e_flush();
2038
2039         if (adapter->msix_entries) {
2040                 int i;
2041                 for (i = 0; i < adapter->num_vectors; i++)
2042                         synchronize_irq(adapter->msix_entries[i].vector);
2043         } else {
2044                 synchronize_irq(adapter->pdev->irq);
2045         }
2046 }
2047
2048 /**
2049  * e1000_irq_enable - Enable default interrupt generation settings
2050  **/
2051 static void e1000_irq_enable(struct e1000_adapter *adapter)
2052 {
2053         struct e1000_hw *hw = &adapter->hw;
2054
2055         if (adapter->msix_entries) {
2056                 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2057                 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2058         } else {
2059                 ew32(IMS, IMS_ENABLE_MASK);
2060         }
2061         e1e_flush();
2062 }
2063
2064 /**
2065  * e1000e_get_hw_control - get control of the h/w from f/w
2066  * @adapter: address of board private structure
2067  *
2068  * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2069  * For ASF and Pass Through versions of f/w this means that
2070  * the driver is loaded. For AMT version (only with 82573)
2071  * of the f/w this means that the network i/f is open.
2072  **/
2073 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2074 {
2075         struct e1000_hw *hw = &adapter->hw;
2076         u32 ctrl_ext;
2077         u32 swsm;
2078
2079         /* Let firmware know the driver has taken over */
2080         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2081                 swsm = er32(SWSM);
2082                 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2083         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2084                 ctrl_ext = er32(CTRL_EXT);
2085                 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2086         }
2087 }
2088
2089 /**
2090  * e1000e_release_hw_control - release control of the h/w to f/w
2091  * @adapter: address of board private structure
2092  *
2093  * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2094  * For ASF and Pass Through versions of f/w this means that the
2095  * driver is no longer loaded. For AMT version (only with 82573) i
2096  * of the f/w this means that the network i/f is closed.
2097  *
2098  **/
2099 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2100 {
2101         struct e1000_hw *hw = &adapter->hw;
2102         u32 ctrl_ext;
2103         u32 swsm;
2104
2105         /* Let firmware taken over control of h/w */
2106         if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2107                 swsm = er32(SWSM);
2108                 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2109         } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2110                 ctrl_ext = er32(CTRL_EXT);
2111                 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2112         }
2113 }
2114
2115 /**
2116  * @e1000_alloc_ring - allocate memory for a ring structure
2117  **/
2118 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2119                                 struct e1000_ring *ring)
2120 {
2121         struct pci_dev *pdev = adapter->pdev;
2122
2123         ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2124                                         GFP_KERNEL);
2125         if (!ring->desc)
2126                 return -ENOMEM;
2127
2128         return 0;
2129 }
2130
2131 /**
2132  * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2133  * @adapter: board private structure
2134  *
2135  * Return 0 on success, negative on failure
2136  **/
2137 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2138 {
2139         struct e1000_ring *tx_ring = adapter->tx_ring;
2140         int err = -ENOMEM, size;
2141
2142         size = sizeof(struct e1000_buffer) * tx_ring->count;
2143         tx_ring->buffer_info = vzalloc(size);
2144         if (!tx_ring->buffer_info)
2145                 goto err;
2146
2147         /* round up to nearest 4K */
2148         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2149         tx_ring->size = ALIGN(tx_ring->size, 4096);
2150
2151         err = e1000_alloc_ring_dma(adapter, tx_ring);
2152         if (err)
2153                 goto err;
2154
2155         tx_ring->next_to_use = 0;
2156         tx_ring->next_to_clean = 0;
2157
2158         return 0;
2159 err:
2160         vfree(tx_ring->buffer_info);
2161         e_err("Unable to allocate memory for the transmit descriptor ring\n");
2162         return err;
2163 }
2164
2165 /**
2166  * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2167  * @adapter: board private structure
2168  *
2169  * Returns 0 on success, negative on failure
2170  **/
2171 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2172 {
2173         struct e1000_ring *rx_ring = adapter->rx_ring;
2174         struct e1000_buffer *buffer_info;
2175         int i, size, desc_len, err = -ENOMEM;
2176
2177         size = sizeof(struct e1000_buffer) * rx_ring->count;
2178         rx_ring->buffer_info = vzalloc(size);
2179         if (!rx_ring->buffer_info)
2180                 goto err;
2181
2182         for (i = 0; i < rx_ring->count; i++) {
2183                 buffer_info = &rx_ring->buffer_info[i];
2184                 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2185                                                 sizeof(struct e1000_ps_page),
2186                                                 GFP_KERNEL);
2187                 if (!buffer_info->ps_pages)
2188                         goto err_pages;
2189         }
2190
2191         desc_len = sizeof(union e1000_rx_desc_packet_split);
2192
2193         /* Round up to nearest 4K */
2194         rx_ring->size = rx_ring->count * desc_len;
2195         rx_ring->size = ALIGN(rx_ring->size, 4096);
2196
2197         err = e1000_alloc_ring_dma(adapter, rx_ring);
2198         if (err)
2199                 goto err_pages;
2200
2201         rx_ring->next_to_clean = 0;
2202         rx_ring->next_to_use = 0;
2203         rx_ring->rx_skb_top = NULL;
2204
2205         return 0;
2206
2207 err_pages:
2208         for (i = 0; i < rx_ring->count; i++) {
2209                 buffer_info = &rx_ring->buffer_info[i];
2210                 kfree(buffer_info->ps_pages);
2211         }
2212 err:
2213         vfree(rx_ring->buffer_info);
2214         e_err("Unable to allocate memory for the receive descriptor ring\n");
2215         return err;
2216 }
2217
2218 /**
2219  * e1000_clean_tx_ring - Free Tx Buffers
2220  * @adapter: board private structure
2221  **/
2222 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2223 {
2224         struct e1000_ring *tx_ring = adapter->tx_ring;
2225         struct e1000_buffer *buffer_info;
2226         unsigned long size;
2227         unsigned int i;
2228
2229         for (i = 0; i < tx_ring->count; i++) {
2230                 buffer_info = &tx_ring->buffer_info[i];
2231                 e1000_put_txbuf(adapter, buffer_info);
2232         }
2233
2234         size = sizeof(struct e1000_buffer) * tx_ring->count;
2235         memset(tx_ring->buffer_info, 0, size);
2236
2237         memset(tx_ring->desc, 0, tx_ring->size);
2238
2239         tx_ring->next_to_use = 0;
2240         tx_ring->next_to_clean = 0;
2241
2242         writel(0, adapter->hw.hw_addr + tx_ring->head);
2243         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2244 }
2245
2246 /**
2247  * e1000e_free_tx_resources - Free Tx Resources per Queue
2248  * @adapter: board private structure
2249  *
2250  * Free all transmit software resources
2251  **/
2252 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2253 {
2254         struct pci_dev *pdev = adapter->pdev;
2255         struct e1000_ring *tx_ring = adapter->tx_ring;
2256
2257         e1000_clean_tx_ring(adapter);
2258
2259         vfree(tx_ring->buffer_info);
2260         tx_ring->buffer_info = NULL;
2261
2262         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2263                           tx_ring->dma);
2264         tx_ring->desc = NULL;
2265 }
2266
2267 /**
2268  * e1000e_free_rx_resources - Free Rx Resources
2269  * @adapter: board private structure
2270  *
2271  * Free all receive software resources
2272  **/
2273
2274 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2275 {
2276         struct pci_dev *pdev = adapter->pdev;
2277         struct e1000_ring *rx_ring = adapter->rx_ring;
2278         int i;
2279
2280         e1000_clean_rx_ring(adapter);
2281
2282         for (i = 0; i < rx_ring->count; i++)
2283                 kfree(rx_ring->buffer_info[i].ps_pages);
2284
2285         vfree(rx_ring->buffer_info);
2286         rx_ring->buffer_info = NULL;
2287
2288         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2289                           rx_ring->dma);
2290         rx_ring->desc = NULL;
2291 }
2292
2293 /**
2294  * e1000_update_itr - update the dynamic ITR value based on statistics
2295  * @adapter: pointer to adapter
2296  * @itr_setting: current adapter->itr
2297  * @packets: the number of packets during this measurement interval
2298  * @bytes: the number of bytes during this measurement interval
2299  *
2300  *      Stores a new ITR value based on packets and byte
2301  *      counts during the last interrupt.  The advantage of per interrupt
2302  *      computation is faster updates and more accurate ITR for the current
2303  *      traffic pattern.  Constants in this function were computed
2304  *      based on theoretical maximum wire speed and thresholds were set based
2305  *      on testing data as well as attempting to minimize response time
2306  *      while increasing bulk throughput.  This functionality is controlled
2307  *      by the InterruptThrottleRate module parameter.
2308  **/
2309 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2310                                      u16 itr_setting, int packets,
2311                                      int bytes)
2312 {
2313         unsigned int retval = itr_setting;
2314
2315         if (packets == 0)
2316                 goto update_itr_done;
2317
2318         switch (itr_setting) {
2319         case lowest_latency:
2320                 /* handle TSO and jumbo frames */
2321                 if (bytes/packets > 8000)
2322                         retval = bulk_latency;
2323                 else if ((packets < 5) && (bytes > 512))
2324                         retval = low_latency;
2325                 break;
2326         case low_latency:  /* 50 usec aka 20000 ints/s */
2327                 if (bytes > 10000) {
2328                         /* this if handles the TSO accounting */
2329                         if (bytes/packets > 8000)
2330                                 retval = bulk_latency;
2331                         else if ((packets < 10) || ((bytes/packets) > 1200))
2332                                 retval = bulk_latency;
2333                         else if ((packets > 35))
2334                                 retval = lowest_latency;
2335                 } else if (bytes/packets > 2000) {
2336                         retval = bulk_latency;
2337                 } else if (packets <= 2 && bytes < 512) {
2338                         retval = lowest_latency;
2339                 }
2340                 break;
2341         case bulk_latency: /* 250 usec aka 4000 ints/s */
2342                 if (bytes > 25000) {
2343                         if (packets > 35)
2344                                 retval = low_latency;
2345                 } else if (bytes < 6000) {
2346                         retval = low_latency;
2347                 }
2348                 break;
2349         }
2350
2351 update_itr_done:
2352         return retval;
2353 }
2354
2355 static void e1000_set_itr(struct e1000_adapter *adapter)
2356 {
2357         struct e1000_hw *hw = &adapter->hw;
2358         u16 current_itr;
2359         u32 new_itr = adapter->itr;
2360
2361         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2362         if (adapter->link_speed != SPEED_1000) {
2363                 current_itr = 0;
2364                 new_itr = 4000;
2365                 goto set_itr_now;
2366         }
2367
2368         if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2369                 new_itr = 0;
2370                 goto set_itr_now;
2371         }
2372
2373         adapter->tx_itr = e1000_update_itr(adapter,
2374                                     adapter->tx_itr,
2375                                     adapter->total_tx_packets,
2376                                     adapter->total_tx_bytes);
2377         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2378         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2379                 adapter->tx_itr = low_latency;
2380
2381         adapter->rx_itr = e1000_update_itr(adapter,
2382                                     adapter->rx_itr,
2383                                     adapter->total_rx_packets,
2384                                     adapter->total_rx_bytes);
2385         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2386         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2387                 adapter->rx_itr = low_latency;
2388
2389         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2390
2391         switch (current_itr) {
2392         /* counts and packets in update_itr are dependent on these numbers */
2393         case lowest_latency:
2394                 new_itr = 70000;
2395                 break;
2396         case low_latency:
2397                 new_itr = 20000; /* aka hwitr = ~200 */
2398                 break;
2399         case bulk_latency:
2400                 new_itr = 4000;
2401                 break;
2402         default:
2403                 break;
2404         }
2405
2406 set_itr_now:
2407         if (new_itr != adapter->itr) {
2408                 /*
2409                  * this attempts to bias the interrupt rate towards Bulk
2410                  * by adding intermediate steps when interrupt rate is
2411                  * increasing
2412                  */
2413                 new_itr = new_itr > adapter->itr ?
2414                              min(adapter->itr + (new_itr >> 2), new_itr) :
2415                              new_itr;
2416                 adapter->itr = new_itr;
2417                 adapter->rx_ring->itr_val = new_itr;
2418                 if (adapter->msix_entries)
2419                         adapter->rx_ring->set_itr = 1;
2420                 else
2421                         if (new_itr)
2422                                 ew32(ITR, 1000000000 / (new_itr * 256));
2423                         else
2424                                 ew32(ITR, 0);
2425         }
2426 }
2427
2428 /**
2429  * e1000_alloc_queues - Allocate memory for all rings
2430  * @adapter: board private structure to initialize
2431  **/
2432 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2433 {
2434         adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2435         if (!adapter->tx_ring)
2436                 goto err;
2437
2438         adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2439         if (!adapter->rx_ring)
2440                 goto err;
2441
2442         return 0;
2443 err:
2444         e_err("Unable to allocate memory for queues\n");
2445         kfree(adapter->rx_ring);
2446         kfree(adapter->tx_ring);
2447         return -ENOMEM;
2448 }
2449
2450 /**
2451  * e1000_clean - NAPI Rx polling callback
2452  * @napi: struct associated with this polling callback
2453  * @budget: amount of packets driver is allowed to process this poll
2454  **/
2455 static int e1000_clean(struct napi_struct *napi, int budget)
2456 {
2457         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2458         struct e1000_hw *hw = &adapter->hw;
2459         struct net_device *poll_dev = adapter->netdev;
2460         int tx_cleaned = 1, work_done = 0;
2461
2462         adapter = netdev_priv(poll_dev);
2463
2464         if (adapter->msix_entries &&
2465             !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2466                 goto clean_rx;
2467
2468         tx_cleaned = e1000_clean_tx_irq(adapter);
2469
2470 clean_rx:
2471         adapter->clean_rx(adapter, &work_done, budget);
2472
2473         if (!tx_cleaned)
2474                 work_done = budget;
2475
2476         /* If budget not fully consumed, exit the polling mode */
2477         if (work_done < budget) {
2478                 if (adapter->itr_setting & 3)
2479                         e1000_set_itr(adapter);
2480                 napi_complete(napi);
2481                 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2482                         if (adapter->msix_entries)
2483                                 ew32(IMS, adapter->rx_ring->ims_val);
2484                         else
2485                                 e1000_irq_enable(adapter);
2486                 }
2487         }
2488
2489         return work_done;
2490 }
2491
2492 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2493 {
2494         struct e1000_adapter *adapter = netdev_priv(netdev);
2495         struct e1000_hw *hw = &adapter->hw;
2496         u32 vfta, index;
2497
2498         /* don't update vlan cookie if already programmed */
2499         if ((adapter->hw.mng_cookie.status &
2500              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2501             (vid == adapter->mng_vlan_id))
2502                 return;
2503
2504         /* add VID to filter table */
2505         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2506                 index = (vid >> 5) & 0x7F;
2507                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2508                 vfta |= (1 << (vid & 0x1F));
2509                 hw->mac.ops.write_vfta(hw, index, vfta);
2510         }
2511
2512         set_bit(vid, adapter->active_vlans);
2513 }
2514
2515 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2516 {
2517         struct e1000_adapter *adapter = netdev_priv(netdev);
2518         struct e1000_hw *hw = &adapter->hw;
2519         u32 vfta, index;
2520
2521         if ((adapter->hw.mng_cookie.status &
2522              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2523             (vid == adapter->mng_vlan_id)) {
2524                 /* release control to f/w */
2525                 e1000e_release_hw_control(adapter);
2526                 return;
2527         }
2528
2529         /* remove VID from filter table */
2530         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2531                 index = (vid >> 5) & 0x7F;
2532                 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2533                 vfta &= ~(1 << (vid & 0x1F));
2534                 hw->mac.ops.write_vfta(hw, index, vfta);
2535         }
2536
2537         clear_bit(vid, adapter->active_vlans);
2538 }
2539
2540 /**
2541  * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2542  * @adapter: board private structure to initialize
2543  **/
2544 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2545 {
2546         struct net_device *netdev = adapter->netdev;
2547         struct e1000_hw *hw = &adapter->hw;
2548         u32 rctl;
2549
2550         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2551                 /* disable VLAN receive filtering */
2552                 rctl = er32(RCTL);
2553                 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2554                 ew32(RCTL, rctl);
2555
2556                 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2557                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2558                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2559                 }
2560         }
2561 }
2562
2563 /**
2564  * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2565  * @adapter: board private structure to initialize
2566  **/
2567 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2568 {
2569         struct e1000_hw *hw = &adapter->hw;
2570         u32 rctl;
2571
2572         if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2573                 /* enable VLAN receive filtering */
2574                 rctl = er32(RCTL);
2575                 rctl |= E1000_RCTL_VFE;
2576                 rctl &= ~E1000_RCTL_CFIEN;
2577                 ew32(RCTL, rctl);
2578         }
2579 }
2580
2581 /**
2582  * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2583  * @adapter: board private structure to initialize
2584  **/
2585 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2586 {
2587         struct e1000_hw *hw = &adapter->hw;
2588         u32 ctrl;
2589
2590         /* disable VLAN tag insert/strip */
2591         ctrl = er32(CTRL);
2592         ctrl &= ~E1000_CTRL_VME;
2593         ew32(CTRL, ctrl);
2594 }
2595
2596 /**
2597  * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2598  * @adapter: board private structure to initialize
2599  **/
2600 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2601 {
2602         struct e1000_hw *hw = &adapter->hw;
2603         u32 ctrl;
2604
2605         /* enable VLAN tag insert/strip */
2606         ctrl = er32(CTRL);
2607         ctrl |= E1000_CTRL_VME;
2608         ew32(CTRL, ctrl);
2609 }
2610
2611 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2612 {
2613         struct net_device *netdev = adapter->netdev;
2614         u16 vid = adapter->hw.mng_cookie.vlan_id;
2615         u16 old_vid = adapter->mng_vlan_id;
2616
2617         if (adapter->hw.mng_cookie.status &
2618             E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2619                 e1000_vlan_rx_add_vid(netdev, vid);
2620                 adapter->mng_vlan_id = vid;
2621         }
2622
2623         if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2624                 e1000_vlan_rx_kill_vid(netdev, old_vid);
2625 }
2626
2627 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2628 {
2629         u16 vid;
2630
2631         e1000_vlan_rx_add_vid(adapter->netdev, 0);
2632
2633         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2634                 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2635 }
2636
2637 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2638 {
2639         struct e1000_hw *hw = &adapter->hw;
2640         u32 manc, manc2h, mdef, i, j;
2641
2642         if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2643                 return;
2644
2645         manc = er32(MANC);
2646
2647         /*
2648          * enable receiving management packets to the host. this will probably
2649          * generate destination unreachable messages from the host OS, but
2650          * the packets will be handled on SMBUS
2651          */
2652         manc |= E1000_MANC_EN_MNG2HOST;
2653         manc2h = er32(MANC2H);
2654
2655         switch (hw->mac.type) {
2656         default:
2657                 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2658                 break;
2659         case e1000_82574:
2660         case e1000_82583:
2661                 /*
2662                  * Check if IPMI pass-through decision filter already exists;
2663                  * if so, enable it.
2664                  */
2665                 for (i = 0, j = 0; i < 8; i++) {
2666                         mdef = er32(MDEF(i));
2667
2668                         /* Ignore filters with anything other than IPMI ports */
2669                         if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2670                                 continue;
2671
2672                         /* Enable this decision filter in MANC2H */
2673                         if (mdef)
2674                                 manc2h |= (1 << i);
2675
2676                         j |= mdef;
2677                 }
2678
2679                 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2680                         break;
2681
2682                 /* Create new decision filter in an empty filter */
2683                 for (i = 0, j = 0; i < 8; i++)
2684                         if (er32(MDEF(i)) == 0) {
2685                                 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2686                                                E1000_MDEF_PORT_664));
2687                                 manc2h |= (1 << 1);
2688                                 j++;
2689                                 break;
2690                         }
2691
2692                 if (!j)
2693                         e_warn("Unable to create IPMI pass-through filter\n");
2694                 break;
2695         }
2696
2697         ew32(MANC2H, manc2h);
2698         ew32(MANC, manc);
2699 }
2700
2701 /**
2702  * e1000_configure_tx - Configure Transmit Unit after Reset
2703  * @adapter: board private structure
2704  *
2705  * Configure the Tx unit of the MAC after a reset.
2706  **/
2707 static void e1000_configure_tx(struct e1000_adapter *adapter)
2708 {
2709         struct e1000_hw *hw = &adapter->hw;
2710         struct e1000_ring *tx_ring = adapter->tx_ring;
2711         u64 tdba;
2712         u32 tdlen, tctl, tipg, tarc;
2713         u32 ipgr1, ipgr2;
2714
2715         /* Setup the HW Tx Head and Tail descriptor pointers */
2716         tdba = tx_ring->dma;
2717         tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2718         ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2719         ew32(TDBAH, (tdba >> 32));
2720         ew32(TDLEN, tdlen);
2721         ew32(TDH, 0);
2722         ew32(TDT, 0);
2723         tx_ring->head = E1000_TDH;
2724         tx_ring->tail = E1000_TDT;
2725
2726         /* Set the default values for the Tx Inter Packet Gap timer */
2727         tipg = DEFAULT_82543_TIPG_IPGT_COPPER;          /*  8  */
2728         ipgr1 = DEFAULT_82543_TIPG_IPGR1;               /*  8  */
2729         ipgr2 = DEFAULT_82543_TIPG_IPGR2;               /*  6  */
2730
2731         if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2732                 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /*  7  */
2733
2734         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2735         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2736         ew32(TIPG, tipg);
2737
2738         /* Set the Tx Interrupt Delay register */
2739         ew32(TIDV, adapter->tx_int_delay);
2740         /* Tx irq moderation */
2741         ew32(TADV, adapter->tx_abs_int_delay);
2742
2743         if (adapter->flags2 & FLAG2_DMA_BURST) {
2744                 u32 txdctl = er32(TXDCTL(0));
2745                 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2746                             E1000_TXDCTL_WTHRESH);
2747                 /*
2748                  * set up some performance related parameters to encourage the
2749                  * hardware to use the bus more efficiently in bursts, depends
2750                  * on the tx_int_delay to be enabled,
2751                  * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2752                  * hthresh = 1 ==> prefetch when one or more available
2753                  * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2754                  * BEWARE: this seems to work but should be considered first if
2755                  * there are Tx hangs or other Tx related bugs
2756                  */
2757                 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2758                 ew32(TXDCTL(0), txdctl);
2759                 /* erratum work around: set txdctl the same for both queues */
2760                 ew32(TXDCTL(1), txdctl);
2761         }
2762
2763         /* Program the Transmit Control Register */
2764         tctl = er32(TCTL);
2765         tctl &= ~E1000_TCTL_CT;
2766         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2767                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2768
2769         if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2770                 tarc = er32(TARC(0));
2771                 /*
2772                  * set the speed mode bit, we'll clear it if we're not at
2773                  * gigabit link later
2774                  */
2775 #define SPEED_MODE_BIT (1 << 21)
2776                 tarc |= SPEED_MODE_BIT;
2777                 ew32(TARC(0), tarc);
2778         }
2779
2780         /* errata: program both queues to unweighted RR */
2781         if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2782                 tarc = er32(TARC(0));
2783                 tarc |= 1;
2784                 ew32(TARC(0), tarc);
2785                 tarc = er32(TARC(1));
2786                 tarc |= 1;
2787                 ew32(TARC(1), tarc);
2788         }
2789
2790         /* Setup Transmit Descriptor Settings for eop descriptor */
2791         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2792
2793         /* only set IDE if we are delaying interrupts using the timers */
2794         if (adapter->tx_int_delay)
2795                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2796
2797         /* enable Report Status bit */
2798         adapter->txd_cmd |= E1000_TXD_CMD_RS;
2799
2800         ew32(TCTL, tctl);
2801
2802         e1000e_config_collision_dist(hw);
2803 }
2804
2805 /**
2806  * e1000_setup_rctl - configure the receive control registers
2807  * @adapter: Board private structure
2808  **/
2809 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2810                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2811 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2812 {
2813         struct e1000_hw *hw = &adapter->hw;
2814         u32 rctl, rfctl;
2815         u32 pages = 0;
2816
2817         /* Workaround Si errata on 82579 - configure jumbo frame flow */
2818         if (hw->mac.type == e1000_pch2lan) {
2819                 s32 ret_val;
2820
2821                 if (adapter->netdev->mtu > ETH_DATA_LEN)
2822                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2823                 else
2824                         ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2825
2826                 if (ret_val)
2827                         e_dbg("failed to enable jumbo frame workaround mode\n");
2828         }
2829
2830         /* Program MC offset vector base */
2831         rctl = er32(RCTL);
2832         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2833         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2834                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2835                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2836
2837         /* Do not Store bad packets */
2838         rctl &= ~E1000_RCTL_SBP;
2839
2840         /* Enable Long Packet receive */
2841         if (adapter->netdev->mtu <= ETH_DATA_LEN)
2842                 rctl &= ~E1000_RCTL_LPE;
2843         else
2844                 rctl |= E1000_RCTL_LPE;
2845
2846         /* Some systems expect that the CRC is included in SMBUS traffic. The
2847          * hardware strips the CRC before sending to both SMBUS (BMC) and to
2848          * host memory when this is enabled
2849          */
2850         if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2851                 rctl |= E1000_RCTL_SECRC;
2852
2853         /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2854         if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2855                 u16 phy_data;
2856
2857                 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2858                 phy_data &= 0xfff8;
2859                 phy_data |= (1 << 2);
2860                 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2861
2862                 e1e_rphy(hw, 22, &phy_data);
2863                 phy_data &= 0x0fff;
2864                 phy_data |= (1 << 14);
2865                 e1e_wphy(hw, 0x10, 0x2823);
2866                 e1e_wphy(hw, 0x11, 0x0003);
2867                 e1e_wphy(hw, 22, phy_data);
2868         }
2869
2870         /* Setup buffer sizes */
2871         rctl &= ~E1000_RCTL_SZ_4096;
2872         rctl |= E1000_RCTL_BSEX;
2873         switch (adapter->rx_buffer_len) {
2874         case 2048:
2875         default:
2876                 rctl |= E1000_RCTL_SZ_2048;
2877                 rctl &= ~E1000_RCTL_BSEX;
2878                 break;
2879         case 4096:
2880                 rctl |= E1000_RCTL_SZ_4096;
2881                 break;
2882         case 8192:
2883                 rctl |= E1000_RCTL_SZ_8192;
2884                 break;
2885         case 16384:
2886                 rctl |= E1000_RCTL_SZ_16384;
2887                 break;
2888         }
2889
2890         /*
2891          * 82571 and greater support packet-split where the protocol
2892          * header is placed in skb->data and the packet data is
2893          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2894          * In the case of a non-split, skb->data is linearly filled,
2895          * followed by the page buffers.  Therefore, skb->data is
2896          * sized to hold the largest protocol header.
2897          *
2898          * allocations using alloc_page take too long for regular MTU
2899          * so only enable packet split for jumbo frames
2900          *
2901          * Using pages when the page size is greater than 16k wastes
2902          * a lot of memory, since we allocate 3 pages at all times
2903          * per packet.
2904          */
2905         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2906         if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2907             (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2908                 adapter->rx_ps_pages = pages;
2909         else
2910                 adapter->rx_ps_pages = 0;
2911
2912         if (adapter->rx_ps_pages) {
2913                 u32 psrctl = 0;
2914
2915                 /* Configure extra packet-split registers */
2916                 rfctl = er32(RFCTL);
2917                 rfctl |= E1000_RFCTL_EXTEN;
2918                 /*
2919                  * disable packet split support for IPv6 extension headers,
2920                  * because some malformed IPv6 headers can hang the Rx
2921                  */
2922                 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2923                           E1000_RFCTL_NEW_IPV6_EXT_DIS);
2924
2925                 ew32(RFCTL, rfctl);
2926
2927                 /* Enable Packet split descriptors */
2928                 rctl |= E1000_RCTL_DTYP_PS;
2929
2930                 psrctl |= adapter->rx_ps_bsize0 >>
2931                         E1000_PSRCTL_BSIZE0_SHIFT;
2932
2933                 switch (adapter->rx_ps_pages) {
2934                 case 3:
2935                         psrctl |= PAGE_SIZE <<
2936                                 E1000_PSRCTL_BSIZE3_SHIFT;
2937                 case 2:
2938                         psrctl |= PAGE_SIZE <<
2939                                 E1000_PSRCTL_BSIZE2_SHIFT;
2940                 case 1:
2941                         psrctl |= PAGE_SIZE >>
2942                                 E1000_PSRCTL_BSIZE1_SHIFT;
2943                         break;
2944                 }
2945
2946                 ew32(PSRCTL, psrctl);
2947         }
2948
2949         ew32(RCTL, rctl);
2950         /* just started the receive unit, no need to restart */
2951         adapter->flags &= ~FLAG_RX_RESTART_NOW;
2952 }
2953
2954 /**
2955  * e1000_configure_rx - Configure Receive Unit after Reset
2956  * @adapter: board private structure
2957  *
2958  * Configure the Rx unit of the MAC after a reset.
2959  **/
2960 static void e1000_configure_rx(struct e1000_adapter *adapter)
2961 {
2962         struct e1000_hw *hw = &adapter->hw;
2963         struct e1000_ring *rx_ring = adapter->rx_ring;
2964         u64 rdba;
2965         u32 rdlen, rctl, rxcsum, ctrl_ext;
2966
2967         if (adapter->rx_ps_pages) {
2968                 /* this is a 32 byte descriptor */
2969                 rdlen = rx_ring->count *
2970                     sizeof(union e1000_rx_desc_packet_split);
2971                 adapter->clean_rx = e1000_clean_rx_irq_ps;
2972                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2973         } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2974                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2975                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2976                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2977         } else {
2978                 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2979                 adapter->clean_rx = e1000_clean_rx_irq;
2980                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2981         }
2982
2983         /* disable receives while setting up the descriptors */
2984         rctl = er32(RCTL);
2985         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
2986                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2987         e1e_flush();
2988         usleep_range(10000, 20000);
2989
2990         if (adapter->flags2 & FLAG2_DMA_BURST) {
2991                 /*
2992                  * set the writeback threshold (only takes effect if the RDTR
2993                  * is set). set GRAN=1 and write back up to 0x4 worth, and
2994                  * enable prefetching of 0x20 Rx descriptors
2995                  * granularity = 01
2996                  * wthresh = 04,
2997                  * hthresh = 04,
2998                  * pthresh = 0x20
2999                  */
3000                 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3001                 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3002
3003                 /*
3004                  * override the delay timers for enabling bursting, only if
3005                  * the value was not set by the user via module options
3006                  */
3007                 if (adapter->rx_int_delay == DEFAULT_RDTR)
3008                         adapter->rx_int_delay = BURST_RDTR;
3009                 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3010                         adapter->rx_abs_int_delay = BURST_RADV;
3011         }
3012
3013         /* set the Receive Delay Timer Register */
3014         ew32(RDTR, adapter->rx_int_delay);
3015
3016         /* irq moderation */
3017         ew32(RADV, adapter->rx_abs_int_delay);
3018         if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3019                 ew32(ITR, 1000000000 / (adapter->itr * 256));
3020
3021         ctrl_ext = er32(CTRL_EXT);
3022         /* Auto-Mask interrupts upon ICR access */
3023         ctrl_ext |= E1000_CTRL_EXT_IAME;
3024         ew32(IAM, 0xffffffff);
3025         ew32(CTRL_EXT, ctrl_ext);
3026         e1e_flush();
3027
3028         /*
3029          * Setup the HW Rx Head and Tail Descriptor Pointers and
3030          * the Base and Length of the Rx Descriptor Ring
3031          */
3032         rdba = rx_ring->dma;
3033         ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3034         ew32(RDBAH, (rdba >> 32));
3035         ew32(RDLEN, rdlen);
3036         ew32(RDH, 0);
3037         ew32(RDT, 0);
3038         rx_ring->head = E1000_RDH;
3039         rx_ring->tail = E1000_RDT;
3040
3041         /* Enable Receive Checksum Offload for TCP and UDP */
3042         rxcsum = er32(RXCSUM);
3043         if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
3044                 rxcsum |= E1000_RXCSUM_TUOFL;
3045
3046                 /*
3047                  * IPv4 payload checksum for UDP fragments must be
3048                  * used in conjunction with packet-split.
3049                  */
3050                 if (adapter->rx_ps_pages)
3051                         rxcsum |= E1000_RXCSUM_IPPCSE;
3052         } else {
3053                 rxcsum &= ~E1000_RXCSUM_TUOFL;
3054                 /* no need to clear IPPCSE as it defaults to 0 */
3055         }
3056         ew32(RXCSUM, rxcsum);
3057
3058         /*
3059          * Enable early receives on supported devices, only takes effect when
3060          * packet size is equal or larger than the specified value (in 8 byte
3061          * units), e.g. using jumbo frames when setting to E1000_ERT_2048
3062          */
3063         if ((adapter->flags & FLAG_HAS_ERT) ||
3064             (adapter->hw.mac.type == e1000_pch2lan)) {
3065                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3066                         u32 rxdctl = er32(RXDCTL(0));
3067                         ew32(RXDCTL(0), rxdctl | 0x3);
3068                         if (adapter->flags & FLAG_HAS_ERT)
3069                                 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3070                         /*
3071                          * With jumbo frames and early-receive enabled,
3072                          * excessive C-state transition latencies result in
3073                          * dropped transactions.
3074                          */
3075                         pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3076                 } else {
3077                         pm_qos_update_request(&adapter->netdev->pm_qos_req,
3078                                               PM_QOS_DEFAULT_VALUE);
3079                 }
3080         }
3081
3082         /* Enable Receives */
3083         ew32(RCTL, rctl);
3084 }
3085
3086 /**
3087  *  e1000_update_mc_addr_list - Update Multicast addresses
3088  *  @hw: pointer to the HW structure
3089  *  @mc_addr_list: array of multicast addresses to program
3090  *  @mc_addr_count: number of multicast addresses to program
3091  *
3092  *  Updates the Multicast Table Array.
3093  *  The caller must have a packed mc_addr_list of multicast addresses.
3094  **/
3095 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3096                                       u32 mc_addr_count)
3097 {
3098         hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3099 }
3100
3101 /**
3102  * e1000_set_multi - Multicast and Promiscuous mode set
3103  * @netdev: network interface device structure
3104  *
3105  * The set_multi entry point is called whenever the multicast address
3106  * list or the network interface flags are updated.  This routine is
3107  * responsible for configuring the hardware for proper multicast,
3108  * promiscuous mode, and all-multi behavior.
3109  **/
3110 static void e1000_set_multi(struct net_device *netdev)
3111 {
3112         struct e1000_adapter *adapter = netdev_priv(netdev);
3113         struct e1000_hw *hw = &adapter->hw;
3114         struct netdev_hw_addr *ha;
3115         u8  *mta_list;
3116         u32 rctl;
3117
3118         /* Check for Promiscuous and All Multicast modes */
3119
3120         rctl = er32(RCTL);
3121
3122         if (netdev->flags & IFF_PROMISC) {
3123                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3124                 rctl &= ~E1000_RCTL_VFE;
3125                 /* Do not hardware filter VLANs in promisc mode */
3126                 e1000e_vlan_filter_disable(adapter);
3127         } else {
3128                 if (netdev->flags & IFF_ALLMULTI) {
3129                         rctl |= E1000_RCTL_MPE;
3130                         rctl &= ~E1000_RCTL_UPE;
3131                 } else {
3132                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3133                 }
3134                 e1000e_vlan_filter_enable(adapter);
3135         }
3136
3137         ew32(RCTL, rctl);
3138
3139         if (!netdev_mc_empty(netdev)) {
3140                 int i = 0;
3141
3142                 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3143                 if (!mta_list)
3144                         return;
3145
3146                 /* prepare a packed array of only addresses. */
3147                 netdev_for_each_mc_addr(ha, netdev)
3148                         memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3149
3150                 e1000_update_mc_addr_list(hw, mta_list, i);
3151                 kfree(mta_list);
3152         } else {
3153                 /*
3154                  * if we're called from probe, we might not have
3155                  * anything to do here, so clear out the list
3156                  */
3157                 e1000_update_mc_addr_list(hw, NULL, 0);
3158         }
3159
3160         if (netdev->features & NETIF_F_HW_VLAN_RX)
3161                 e1000e_vlan_strip_enable(adapter);
3162         else
3163                 e1000e_vlan_strip_disable(adapter);
3164 }
3165
3166 /**
3167  * e1000_configure - configure the hardware for Rx and Tx
3168  * @adapter: private board structure
3169  **/
3170 static void e1000_configure(struct e1000_adapter *adapter)
3171 {
3172         e1000_set_multi(adapter->netdev);
3173
3174         e1000_restore_vlan(adapter);
3175         e1000_init_manageability_pt(adapter);
3176
3177         e1000_configure_tx(adapter);
3178         e1000_setup_rctl(adapter);
3179         e1000_configure_rx(adapter);
3180         adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3181                               GFP_KERNEL);
3182 }
3183
3184 /**
3185  * e1000e_power_up_phy - restore link in case the phy was powered down
3186  * @adapter: address of board private structure
3187  *
3188  * The phy may be powered down to save power and turn off link when the
3189  * driver is unloaded and wake on lan is not enabled (among others)
3190  * *** this routine MUST be followed by a call to e1000e_reset ***
3191  **/
3192 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3193 {
3194         if (adapter->hw.phy.ops.power_up)
3195                 adapter->hw.phy.ops.power_up(&adapter->hw);
3196
3197         adapter->hw.mac.ops.setup_link(&adapter->hw);
3198 }
3199
3200 /**
3201  * e1000_power_down_phy - Power down the PHY
3202  *
3203  * Power down the PHY so no link is implied when interface is down.
3204  * The PHY cannot be powered down if management or WoL is active.
3205  */
3206 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3207 {
3208         /* WoL is enabled */
3209         if (adapter->wol)
3210                 return;
3211
3212         if (adapter->hw.phy.ops.power_down)
3213                 adapter->hw.phy.ops.power_down(&adapter->hw);
3214 }
3215
3216 /**
3217  * e1000e_reset - bring the hardware into a known good state
3218  *
3219  * This function boots the hardware and enables some settings that
3220  * require a configuration cycle of the hardware - those cannot be
3221  * set/changed during runtime. After reset the device needs to be
3222  * properly configured for Rx, Tx etc.
3223  */
3224 void e1000e_reset(struct e1000_adapter *adapter)
3225 {
3226         struct e1000_mac_info *mac = &adapter->hw.mac;
3227         struct e1000_fc_info *fc = &adapter->hw.fc;
3228         struct e1000_hw *hw = &adapter->hw;
3229         u32 tx_space, min_tx_space, min_rx_space;
3230         u32 pba = adapter->pba;
3231         u16 hwm;
3232
3233         /* reset Packet Buffer Allocation to default */
3234         ew32(PBA, pba);
3235
3236         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3237                 /*
3238                  * To maintain wire speed transmits, the Tx FIFO should be
3239                  * large enough to accommodate two full transmit packets,
3240                  * rounded up to the next 1KB and expressed in KB.  Likewise,
3241                  * the Rx FIFO should be large enough to accommodate at least
3242                  * one full receive packet and is similarly rounded up and
3243                  * expressed in KB.
3244                  */
3245                 pba = er32(PBA);
3246                 /* upper 16 bits has Tx packet buffer allocation size in KB */
3247                 tx_space = pba >> 16;
3248                 /* lower 16 bits has Rx packet buffer allocation size in KB */
3249                 pba &= 0xffff;
3250                 /*
3251                  * the Tx fifo also stores 16 bytes of information about the Tx
3252                  * but don't include ethernet FCS because hardware appends it
3253                  */
3254                 min_tx_space = (adapter->max_frame_size +
3255                                 sizeof(struct e1000_tx_desc) -
3256                                 ETH_FCS_LEN) * 2;
3257                 min_tx_space = ALIGN(min_tx_space, 1024);
3258                 min_tx_space >>= 10;
3259                 /* software strips receive CRC, so leave room for it */
3260                 min_rx_space = adapter->max_frame_size;
3261                 min_rx_space = ALIGN(min_rx_space, 1024);
3262                 min_rx_space >>= 10;
3263
3264                 /*
3265                  * If current Tx allocation is less than the min Tx FIFO size,
3266                  * and the min Tx FIFO size is less than the current Rx FIFO
3267                  * allocation, take space away from current Rx allocation
3268                  */
3269                 if ((tx_space < min_tx_space) &&
3270                     ((min_tx_space - tx_space) < pba)) {
3271                         pba -= min_tx_space - tx_space;
3272
3273                         /*
3274                          * if short on Rx space, Rx wins and must trump Tx
3275                          * adjustment or use Early Receive if available
3276                          */
3277                         if ((pba < min_rx_space) &&
3278                             (!(adapter->flags & FLAG_HAS_ERT)))
3279                                 /* ERT enabled in e1000_configure_rx */
3280                                 pba = min_rx_space;
3281                 }
3282
3283                 ew32(PBA, pba);
3284         }
3285
3286         /*
3287          * flow control settings
3288          *
3289          * The high water mark must be low enough to fit one full frame
3290          * (or the size used for early receive) above it in the Rx FIFO.
3291          * Set it to the lower of:
3292          * - 90% of the Rx FIFO size, and
3293          * - the full Rx FIFO size minus the early receive size (for parts
3294          *   with ERT support assuming ERT set to E1000_ERT_2048), or
3295          * - the full Rx FIFO size minus one full frame
3296          */
3297         if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3298                 fc->pause_time = 0xFFFF;
3299         else
3300                 fc->pause_time = E1000_FC_PAUSE_TIME;
3301         fc->send_xon = 1;
3302         fc->current_mode = fc->requested_mode;
3303
3304         switch (hw->mac.type) {
3305         default:
3306                 if ((adapter->flags & FLAG_HAS_ERT) &&
3307                     (adapter->netdev->mtu > ETH_DATA_LEN))
3308                         hwm = min(((pba << 10) * 9 / 10),
3309                                   ((pba << 10) - (E1000_ERT_2048 << 3)));
3310                 else
3311                         hwm = min(((pba << 10) * 9 / 10),
3312                                   ((pba << 10) - adapter->max_frame_size));
3313
3314                 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3315                 fc->low_water = fc->high_water - 8;
3316                 break;
3317         case e1000_pchlan:
3318                 /*
3319                  * Workaround PCH LOM adapter hangs with certain network
3320                  * loads.  If hangs persist, try disabling Tx flow control.
3321                  */
3322                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3323                         fc->high_water = 0x3500;
3324                         fc->low_water  = 0x1500;
3325                 } else {
3326                         fc->high_water = 0x5000;
3327                         fc->low_water  = 0x3000;
3328                 }
3329                 fc->refresh_time = 0x1000;
3330                 break;
3331         case e1000_pch2lan:
3332                 fc->high_water = 0x05C20;
3333                 fc->low_water = 0x05048;
3334                 fc->pause_time = 0x0650;
3335                 fc->refresh_time = 0x0400;
3336                 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3337                         pba = 14;
3338                         ew32(PBA, pba);
3339                 }
3340                 break;
3341         }
3342
3343         /*
3344          * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3345          * fit in receive buffer and early-receive not supported.
3346          */
3347         if (adapter->itr_setting & 0x3) {
3348                 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3349                     !(adapter->flags & FLAG_HAS_ERT)) {
3350                         if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3351                                 dev_info(&adapter->pdev->dev,
3352                                         "Interrupt Throttle Rate turned off\n");
3353                                 adapter->flags2 |= FLAG2_DISABLE_AIM;
3354                                 ew32(ITR, 0);
3355                         }
3356                 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3357                         dev_info(&adapter->pdev->dev,
3358                                  "Interrupt Throttle Rate turned on\n");
3359                         adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3360                         adapter->itr = 20000;
3361                         ew32(ITR, 1000000000 / (adapter->itr * 256));
3362                 }
3363         }
3364
3365         /* Allow time for pending master requests to run */
3366         mac->ops.reset_hw(hw);
3367
3368         /*
3369          * For parts with AMT enabled, let the firmware know
3370          * that the network interface is in control
3371          */
3372         if (adapter->flags & FLAG_HAS_AMT)
3373                 e1000e_get_hw_control(adapter);
3374
3375         ew32(WUC, 0);
3376
3377         if (mac->ops.init_hw(hw))
3378                 e_err("Hardware Error\n");
3379
3380         e1000_update_mng_vlan(adapter);
3381
3382         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3383         ew32(VET, ETH_P_8021Q);
3384
3385         e1000e_reset_adaptive(hw);
3386
3387         if (!netif_running(adapter->netdev) &&
3388             !test_bit(__E1000_TESTING, &adapter->state)) {
3389                 e1000_power_down_phy(adapter);
3390                 return;
3391         }
3392
3393         e1000_get_phy_info(hw);
3394
3395         if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3396             !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3397                 u16 phy_data = 0;
3398                 /*
3399                  * speed up time to link by disabling smart power down, ignore
3400                  * the return value of this function because there is nothing
3401                  * different we would do if it failed
3402                  */
3403                 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3404                 phy_data &= ~IGP02E1000_PM_SPD;
3405                 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3406         }
3407 }
3408
3409 int e1000e_up(struct e1000_adapter *adapter)
3410 {
3411         struct e1000_hw *hw = &adapter->hw;
3412
3413         /* hardware has been reset, we need to reload some things */
3414         e1000_configure(adapter);
3415
3416         clear_bit(__E1000_DOWN, &adapter->state);
3417
3418         napi_enable(&adapter->napi);
3419         if (adapter->msix_entries)
3420                 e1000_configure_msix(adapter);
3421         e1000_irq_enable(adapter);
3422
3423         netif_start_queue(adapter->netdev);
3424
3425         /* fire a link change interrupt to start the watchdog */
3426         if (adapter->msix_entries)
3427                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3428         else
3429                 ew32(ICS, E1000_ICS_LSC);
3430
3431         return 0;
3432 }
3433
3434 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3435 {
3436         struct e1000_hw *hw = &adapter->hw;
3437
3438         if (!(adapter->flags2 & FLAG2_DMA_BURST))
3439                 return;
3440
3441         /* flush pending descriptor writebacks to memory */
3442         ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3443         ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3444
3445         /* execute the writes immediately */
3446         e1e_flush();
3447 }
3448
3449 static void e1000e_update_stats(struct e1000_adapter *adapter);
3450
3451 void e1000e_down(struct e1000_adapter *adapter)
3452 {
3453         struct net_device *netdev = adapter->netdev;
3454         struct e1000_hw *hw = &adapter->hw;
3455         u32 tctl, rctl;
3456
3457         /*
3458          * signal that we're down so the interrupt handler does not
3459          * reschedule our watchdog timer
3460          */
3461         set_bit(__E1000_DOWN, &adapter->state);
3462
3463         /* disable receives in the hardware */
3464         rctl = er32(RCTL);
3465         if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3466                 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3467         /* flush and sleep below */
3468
3469         netif_stop_queue(netdev);
3470
3471         /* disable transmits in the hardware */
3472         tctl = er32(TCTL);
3473         tctl &= ~E1000_TCTL_EN;
3474         ew32(TCTL, tctl);
3475
3476         /* flush both disables and wait for them to finish */
3477         e1e_flush();
3478         usleep_range(10000, 20000);
3479
3480         napi_disable(&adapter->napi);
3481         e1000_irq_disable(adapter);
3482
3483         del_timer_sync(&adapter->watchdog_timer);
3484         del_timer_sync(&adapter->phy_info_timer);
3485
3486         netif_carrier_off(netdev);
3487
3488         spin_lock(&adapter->stats64_lock);
3489         e1000e_update_stats(adapter);
3490         spin_unlock(&adapter->stats64_lock);
3491
3492         e1000e_flush_descriptors(adapter);
3493         e1000_clean_tx_ring(adapter);
3494         e1000_clean_rx_ring(adapter);
3495
3496         adapter->link_speed = 0;
3497         adapter->link_duplex = 0;
3498
3499         if (!pci_channel_offline(adapter->pdev))
3500                 e1000e_reset(adapter);
3501
3502         /*
3503          * TODO: for power management, we could drop the link and
3504          * pci_disable_device here.
3505          */
3506 }
3507
3508 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3509 {
3510         might_sleep();
3511         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3512                 usleep_range(1000, 2000);
3513         e1000e_down(adapter);
3514         e1000e_up(adapter);
3515         clear_bit(__E1000_RESETTING, &adapter->state);
3516 }
3517
3518 /**
3519  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3520  * @adapter: board private structure to initialize
3521  *
3522  * e1000_sw_init initializes the Adapter private data structure.
3523  * Fields are initialized based on PCI device information and
3524  * OS network device settings (MTU size).
3525  **/
3526 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3527 {
3528         struct net_device *netdev = adapter->netdev;
3529
3530         adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3531         adapter->rx_ps_bsize0 = 128;
3532         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3533         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3534
3535         spin_lock_init(&adapter->stats64_lock);
3536
3537         e1000e_set_interrupt_capability(adapter);
3538
3539         if (e1000_alloc_queues(adapter))
3540                 return -ENOMEM;
3541
3542         /* Explicitly disable IRQ since the NIC can be in any state. */
3543         e1000_irq_disable(adapter);
3544
3545         set_bit(__E1000_DOWN, &adapter->state);
3546         return 0;
3547 }
3548
3549 /**
3550  * e1000_intr_msi_test - Interrupt Handler
3551  * @irq: interrupt number
3552  * @data: pointer to a network interface device structure
3553  **/
3554 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3555 {
3556         struct net_device *netdev = data;
3557         struct e1000_adapter *adapter = netdev_priv(netdev);
3558         struct e1000_hw *hw = &adapter->hw;
3559         u32 icr = er32(ICR);
3560
3561         e_dbg("icr is %08X\n", icr);
3562         if (icr & E1000_ICR_RXSEQ) {
3563                 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3564                 wmb();
3565         }
3566
3567         return IRQ_HANDLED;
3568 }
3569
3570 /**
3571  * e1000_test_msi_interrupt - Returns 0 for successful test
3572  * @adapter: board private struct
3573  *
3574  * code flow taken from tg3.c
3575  **/
3576 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3577 {
3578         struct net_device *netdev = adapter->netdev;
3579         struct e1000_hw *hw = &adapter->hw;
3580         int err;
3581
3582         /* poll_enable hasn't been called yet, so don't need disable */
3583         /* clear any pending events */
3584         er32(ICR);
3585
3586         /* free the real vector and request a test handler */
3587         e1000_free_irq(adapter);
3588         e1000e_reset_interrupt_capability(adapter);
3589
3590         /* Assume that the test fails, if it succeeds then the test
3591          * MSI irq handler will unset this flag */
3592         adapter->flags |= FLAG_MSI_TEST_FAILED;
3593
3594         err = pci_enable_msi(adapter->pdev);
3595         if (err)
3596                 goto msi_test_failed;
3597
3598         err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3599                           netdev->name, netdev);
3600         if (err) {
3601                 pci_disable_msi(adapter->pdev);
3602                 goto msi_test_failed;
3603         }
3604
3605         wmb();
3606
3607         e1000_irq_enable(adapter);
3608
3609         /* fire an unusual interrupt on the test handler */
3610         ew32(ICS, E1000_ICS_RXSEQ);
3611         e1e_flush();
3612         msleep(50);
3613
3614         e1000_irq_disable(adapter);
3615
3616         rmb();
3617
3618         if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3619                 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3620                 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3621         } else
3622                 e_dbg("MSI interrupt test succeeded!\n");
3623
3624         free_irq(adapter->pdev->irq, netdev);
3625         pci_disable_msi(adapter->pdev);
3626
3627 msi_test_failed:
3628         e1000e_set_interrupt_capability(adapter);
3629         return e1000_request_irq(adapter);
3630 }
3631
3632 /**
3633  * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3634  * @adapter: board private struct
3635  *
3636  * code flow taken from tg3.c, called with e1000 interrupts disabled.
3637  **/
3638 static int e1000_test_msi(struct e1000_adapter *adapter)
3639 {
3640         int err;
3641         u16 pci_cmd;
3642
3643         if (!(adapter->flags & FLAG_MSI_ENABLED))
3644                 return 0;
3645
3646         /* disable SERR in case the MSI write causes a master abort */
3647         pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3648         if (pci_cmd & PCI_COMMAND_SERR)
3649                 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3650                                       pci_cmd & ~PCI_COMMAND_SERR);
3651
3652         err = e1000_test_msi_interrupt(adapter);
3653
3654         /* re-enable SERR */
3655         if (pci_cmd & PCI_COMMAND_SERR) {
3656                 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3657                 pci_cmd |= PCI_COMMAND_SERR;
3658                 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3659         }
3660
3661         return err;
3662 }
3663
3664 /**
3665  * e1000_open - Called when a network interface is made active
3666  * @netdev: network interface device structure
3667  *
3668  * Returns 0 on success, negative value on failure
3669  *
3670  * The open entry point is called when a network interface is made
3671  * active by the system (IFF_UP).  At this point all resources needed
3672  * for transmit and receive operations are allocated, the interrupt
3673  * handler is registered with the OS, the watchdog timer is started,
3674  * and the stack is notified that the interface is ready.
3675  **/
3676 static int e1000_open(struct net_device *netdev)
3677 {
3678         struct e1000_adapter *adapter = netdev_priv(netdev);
3679         struct e1000_hw *hw = &adapter->hw;
3680         struct pci_dev *pdev = adapter->pdev;
3681         int err;
3682
3683         /* disallow open during test */
3684         if (test_bit(__E1000_TESTING, &adapter->state))
3685                 return -EBUSY;
3686
3687         pm_runtime_get_sync(&pdev->dev);
3688
3689         netif_carrier_off(netdev);
3690
3691         /* allocate transmit descriptors */
3692         err = e1000e_setup_tx_resources(adapter);
3693         if (err)
3694                 goto err_setup_tx;
3695
3696         /* allocate receive descriptors */
3697         err = e1000e_setup_rx_resources(adapter);
3698         if (err)
3699                 goto err_setup_rx;
3700
3701         /*
3702          * If AMT is enabled, let the firmware know that the network
3703          * interface is now open and reset the part to a known state.
3704          */
3705         if (adapter->flags & FLAG_HAS_AMT) {
3706                 e1000e_get_hw_control(adapter);
3707                 e1000e_reset(adapter);
3708         }
3709
3710         e1000e_power_up_phy(adapter);
3711
3712         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3713         if ((adapter->hw.mng_cookie.status &
3714              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3715                 e1000_update_mng_vlan(adapter);
3716
3717         /* DMA latency requirement to workaround early-receive/jumbo issue */
3718         if ((adapter->flags & FLAG_HAS_ERT) ||
3719             (adapter->hw.mac.type == e1000_pch2lan))
3720                 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3721                                    PM_QOS_CPU_DMA_LATENCY,
3722                                    PM_QOS_DEFAULT_VALUE);
3723
3724         /*
3725          * before we allocate an interrupt, we must be ready to handle it.
3726          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3727          * as soon as we call pci_request_irq, so we have to setup our
3728          * clean_rx handler before we do so.
3729          */
3730         e1000_configure(adapter);
3731
3732         err = e1000_request_irq(adapter);
3733         if (err)
3734                 goto err_req_irq;
3735
3736         /*
3737          * Work around PCIe errata with MSI interrupts causing some chipsets to
3738          * ignore e1000e MSI messages, which means we need to test our MSI
3739          * interrupt now
3740          */
3741         if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3742                 err = e1000_test_msi(adapter);
3743                 if (err) {
3744                         e_err("Interrupt allocation failed\n");
3745                         goto err_req_irq;
3746                 }
3747         }
3748
3749         /* From here on the code is the same as e1000e_up() */
3750         clear_bit(__E1000_DOWN, &adapter->state);
3751
3752         napi_enable(&adapter->napi);
3753
3754         e1000_irq_enable(adapter);
3755
3756         netif_start_queue(netdev);
3757
3758         adapter->idle_check = true;
3759         pm_runtime_put(&pdev->dev);
3760
3761         /* fire a link status change interrupt to start the watchdog */
3762         if (adapter->msix_entries)
3763                 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3764         else
3765                 ew32(ICS, E1000_ICS_LSC);
3766
3767         return 0;
3768
3769 err_req_irq:
3770         e1000e_release_hw_control(adapter);
3771         e1000_power_down_phy(adapter);
3772         e1000e_free_rx_resources(adapter);
3773 err_setup_rx:
3774         e1000e_free_tx_resources(adapter);
3775 err_setup_tx:
3776         e1000e_reset(adapter);
3777         pm_runtime_put_sync(&pdev->dev);
3778
3779         return err;
3780 }
3781
3782 /**
3783  * e1000_close - Disables a network interface
3784  * @netdev: network interface device structure
3785  *
3786  * Returns 0, this is not allowed to fail
3787  *
3788  * The close entry point is called when an interface is de-activated
3789  * by the OS.  The hardware is still under the drivers control, but
3790  * needs to be disabled.  A global MAC reset is issued to stop the
3791  * hardware, and all transmit and receive resources are freed.
3792  **/
3793 static int e1000_close(struct net_device *netdev)
3794 {
3795         struct e1000_adapter *adapter = netdev_priv(netdev);
3796         struct pci_dev *pdev = adapter->pdev;
3797
3798         WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3799
3800         pm_runtime_get_sync(&pdev->dev);
3801
3802         if (!test_bit(__E1000_DOWN, &adapter->state)) {
3803                 e1000e_down(adapter);
3804                 e1000_free_irq(adapter);
3805         }
3806         e1000_power_down_phy(adapter);
3807
3808         e1000e_free_tx_resources(adapter);
3809         e1000e_free_rx_resources(adapter);
3810
3811         /*
3812          * kill manageability vlan ID if supported, but not if a vlan with
3813          * the same ID is registered on the host OS (let 8021q kill it)
3814          */
3815         if (adapter->hw.mng_cookie.status &
3816             E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3817                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3818
3819         /*
3820          * If AMT is enabled, let the firmware know that the network
3821          * interface is now closed
3822          */
3823         if ((adapter->flags & FLAG_HAS_AMT) &&
3824             !test_bit(__E1000_TESTING, &adapter->state))
3825                 e1000e_release_hw_control(adapter);
3826
3827         if ((adapter->flags & FLAG_HAS_ERT) ||
3828             (adapter->hw.mac.type == e1000_pch2lan))
3829                 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3830
3831         pm_runtime_put_sync(&pdev->dev);
3832
3833         return 0;
3834 }
3835 /**
3836  * e1000_set_mac - Change the Ethernet Address of the NIC
3837  * @netdev: network interface device structure
3838  * @p: pointer to an address structure
3839  *
3840  * Returns 0 on success, negative on failure
3841  **/
3842 static int e1000_set_mac(struct net_device *netdev, void *p)
3843 {
3844         struct e1000_adapter *adapter = netdev_priv(netdev);
3845         struct sockaddr *addr = p;
3846
3847         if (!is_valid_ether_addr(addr->sa_data))
3848                 return -EADDRNOTAVAIL;
3849
3850         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3851         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3852
3853         e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3854
3855         if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3856                 /* activate the work around */
3857                 e1000e_set_laa_state_82571(&adapter->hw, 1);
3858
3859                 /*
3860                  * Hold a copy of the LAA in RAR[14] This is done so that
3861                  * between the time RAR[0] gets clobbered  and the time it
3862                  * gets fixed (in e1000_watchdog), the actual LAA is in one
3863                  * of the RARs and no incoming packets directed to this port
3864                  * are dropped. Eventually the LAA will be in RAR[0] and
3865                  * RAR[14]
3866                  */
3867                 e1000e_rar_set(&adapter->hw,
3868                               adapter->hw.mac.addr,
3869                               adapter->hw.mac.rar_entry_count - 1);
3870         }
3871
3872         return 0;
3873 }
3874
3875 /**
3876  * e1000e_update_phy_task - work thread to update phy
3877  * @work: pointer to our work struct
3878  *
3879  * this worker thread exists because we must acquire a
3880  * semaphore to read the phy, which we could msleep while
3881  * waiting for it, and we can't msleep in a timer.
3882  **/
3883 static void e1000e_update_phy_task(struct work_struct *work)
3884 {
3885         struct e1000_adapter *adapter = container_of(work,
3886                                         struct e1000_adapter, update_phy_task);
3887
3888         if (test_bit(__E1000_DOWN, &adapter->state))
3889                 return;
3890
3891         e1000_get_phy_info(&adapter->hw);
3892 }
3893
3894 /*
3895  * Need to wait a few seconds after link up to get diagnostic information from
3896  * the phy
3897  */
3898 static void e1000_update_phy_info(unsigned long data)
3899 {
3900         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3901
3902         if (test_bit(__E1000_DOWN, &adapter->state))
3903                 return;
3904
3905         schedule_work(&adapter->update_phy_task);
3906 }
3907
3908 /**
3909  * e1000e_update_phy_stats - Update the PHY statistics counters
3910  * @adapter: board private structure
3911  *
3912  * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3913  **/
3914 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3915 {
3916         struct e1000_hw *hw = &adapter->hw;
3917         s32 ret_val;
3918         u16 phy_data;
3919
3920         ret_val = hw->phy.ops.acquire(hw);
3921         if (ret_val)
3922                 return;
3923
3924         /*
3925          * A page set is expensive so check if already on desired page.
3926          * If not, set to the page with the PHY status registers.
3927          */
3928         hw->phy.addr = 1;
3929         ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3930                                            &phy_data);
3931         if (ret_val)
3932                 goto release;
3933         if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
3934                 ret_val = hw->phy.ops.set_page(hw,
3935                                                HV_STATS_PAGE << IGP_PAGE_SHIFT);
3936                 if (ret_val)
3937                         goto release;
3938         }
3939
3940         /* Single Collision Count */
3941         hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
3942         ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
3943         if (!ret_val)
3944                 adapter->stats.scc += phy_data;
3945
3946         /* Excessive Collision Count */
3947         hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
3948         ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
3949         if (!ret_val)
3950                 adapter->stats.ecol += phy_data;
3951
3952         /* Multiple Collision Count */
3953         hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
3954         ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
3955         if (!ret_val)
3956                 adapter->stats.mcc += phy_data;
3957
3958         /* Late Collision Count */
3959         hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
3960         ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
3961         if (!ret_val)
3962                 adapter->stats.latecol += phy_data;
3963
3964         /* Collision Count - also used for adaptive IFS */
3965         hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
3966         ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
3967         if (!ret_val)
3968                 hw->mac.collision_delta = phy_data;
3969
3970         /* Defer Count */
3971         hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
3972         ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
3973         if (!ret_val)
3974                 adapter->stats.dc += phy_data;
3975
3976         /* Transmit with no CRS */
3977         hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
3978         ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
3979         if (!ret_val)
3980                 adapter->stats.tncrs += phy_data;
3981
3982 release:
3983         hw->phy.ops.release(hw);
3984 }
3985
3986 /**
3987  * e1000e_update_stats - Update the board statistics counters
3988  * @adapter: board private structure
3989  **/
3990 static void e1000e_update_stats(struct e1000_adapter *adapter)
3991 {
3992         struct net_device *netdev = adapter->netdev;
3993         struct e1000_hw *hw = &adapter->hw;
3994         struct pci_dev *pdev = adapter->pdev;
3995
3996         /*
3997          * Prevent stats update while adapter is being reset, or if the pci
3998          * connection is down.
3999          */
4000         if (adapter->link_speed == 0)
4001                 return;
4002         if (pci_channel_offline(pdev))
4003                 return;
4004
4005         adapter->stats.crcerrs += er32(CRCERRS);
4006         adapter->stats.gprc += er32(GPRC);
4007         adapter->stats.gorc += er32(GORCL);
4008         er32(GORCH); /* Clear gorc */
4009         adapter->stats.bprc += er32(BPRC);
4010         adapter->stats.mprc += er32(MPRC);
4011         adapter->stats.roc += er32(ROC);
4012
4013         adapter->stats.mpc += er32(MPC);
4014
4015         /* Half-duplex statistics */
4016         if (adapter->link_duplex == HALF_DUPLEX) {
4017                 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4018                         e1000e_update_phy_stats(adapter);
4019                 } else {
4020                         adapter->stats.scc += er32(SCC);
4021                         adapter->stats.ecol += er32(ECOL);
4022                         adapter->stats.mcc += er32(MCC);
4023                         adapter->stats.latecol += er32(LATECOL);
4024                         adapter->stats.dc += er32(DC);
4025
4026                         hw->mac.collision_delta = er32(COLC);
4027
4028                         if ((hw->mac.type != e1000_82574) &&
4029                             (hw->mac.type != e1000_82583))
4030                                 adapter->stats.tncrs += er32(TNCRS);
4031                 }
4032                 adapter->stats.colc += hw->mac.collision_delta;
4033         }
4034
4035         adapter->stats.xonrxc += er32(XONRXC);
4036         adapter->stats.xontxc += er32(XONTXC);
4037         adapter->stats.xoffrxc += er32(XOFFRXC);
4038         adapter->stats.xofftxc += er32(XOFFTXC);
4039         adapter->stats.gptc += er32(GPTC);
4040         adapter->stats.gotc += er32(GOTCL);
4041         er32(GOTCH); /* Clear gotc */
4042         adapter->stats.rnbc += er32(RNBC);
4043         adapter->stats.ruc += er32(RUC);
4044
4045         adapter->stats.mptc += er32(MPTC);
4046         adapter->stats.bptc += er32(BPTC);
4047
4048         /* used for adaptive IFS */
4049
4050         hw->mac.tx_packet_delta = er32(TPT);
4051         adapter->stats.tpt += hw->mac.tx_packet_delta;
4052
4053         adapter->stats.algnerrc += er32(ALGNERRC);
4054         adapter->stats.rxerrc += er32(RXERRC);
4055         adapter->stats.cexterr += er32(CEXTERR);
4056         adapter->stats.tsctc += er32(TSCTC);
4057         adapter->stats.tsctfc += er32(TSCTFC);
4058
4059         /* Fill out the OS statistics structure */
4060         netdev->stats.multicast = adapter->stats.mprc;
4061         netdev->stats.collisions = adapter->stats.colc;
4062
4063         /* Rx Errors */
4064
4065         /*
4066          * RLEC on some newer hardware can be incorrect so build
4067          * our own version based on RUC and ROC
4068          */
4069         netdev->stats.rx_errors = adapter->stats.rxerrc +
4070                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4071                 adapter->stats.ruc + adapter->stats.roc +
4072                 adapter->stats.cexterr;
4073         netdev->stats.rx_length_errors = adapter->stats.ruc +
4074                                               adapter->stats.roc;
4075         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4076         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4077         netdev->stats.rx_missed_errors = adapter->stats.mpc;
4078
4079         /* Tx Errors */
4080         netdev->stats.tx_errors = adapter->stats.ecol +
4081                                        adapter->stats.latecol;
4082         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4083         netdev->stats.tx_window_errors = adapter->stats.latecol;
4084         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4085
4086         /* Tx Dropped needs to be maintained elsewhere */
4087
4088         /* Management Stats */
4089         adapter->stats.mgptc += er32(MGTPTC);
4090         adapter->stats.mgprc += er32(MGTPRC);
4091         adapter->stats.mgpdc += er32(MGTPDC);
4092 }
4093
4094 /**
4095  * e1000_phy_read_status - Update the PHY register status snapshot
4096  * @adapter: board private structure
4097  **/
4098 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4099 {
4100         struct e1000_hw *hw = &adapter->hw;
4101         struct e1000_phy_regs *phy = &adapter->phy_regs;
4102
4103         if ((er32(STATUS) & E1000_STATUS_LU) &&
4104             (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4105                 int ret_val;
4106
4107                 ret_val  = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4108                 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4109                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4110                 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4111                 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4112                 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4113                 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4114                 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4115                 if (ret_val)
4116                         e_warn("Error reading PHY register\n");
4117         } else {
4118                 /*
4119                  * Do not read PHY registers if link is not up
4120                  * Set values to typical power-on defaults
4121                  */
4122                 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4123                 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4124                              BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4125                              BMSR_ERCAP);
4126                 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4127                                   ADVERTISE_ALL | ADVERTISE_CSMA);
4128                 phy->lpa = 0;
4129                 phy->expansion = EXPANSION_ENABLENPAGE;
4130                 phy->ctrl1000 = ADVERTISE_1000FULL;
4131                 phy->stat1000 = 0;
4132                 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4133         }
4134 }
4135
4136 static void e1000_print_link_info(struct e1000_adapter *adapter)
4137 {
4138         struct e1000_hw *hw = &adapter->hw;
4139         u32 ctrl = er32(CTRL);
4140
4141         /* Link status message must follow this format for user tools */
4142         printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4143                "Flow Control: %s\n",
4144                adapter->netdev->name,
4145                adapter->link_speed,
4146                (adapter->link_duplex == FULL_DUPLEX) ?
4147                "Full Duplex" : "Half Duplex",
4148                ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4149                "Rx/Tx" :
4150                ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4151                 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4152 }
4153
4154 static bool e1000e_has_link(struct e1000_adapter *adapter)
4155 {
4156         struct e1000_hw *hw = &adapter->hw;
4157         bool link_active = 0;
4158         s32 ret_val = 0;
4159
4160         /*
4161          * get_link_status is set on LSC (link status) interrupt or
4162          * Rx sequence error interrupt.  get_link_status will stay
4163          * false until the check_for_link establishes link
4164          * for copper adapters ONLY
4165          */
4166         switch (hw->phy.media_type) {
4167         case e1000_media_type_copper:
4168                 if (hw->mac.get_link_status) {
4169                         ret_val = hw->mac.ops.check_for_link(hw);
4170                         link_active = !hw->mac.get_link_status;
4171                 } else {
4172                         link_active = 1;
4173                 }
4174                 break;
4175         case e1000_media_type_fiber:
4176                 ret_val = hw->mac.ops.check_for_link(hw);
4177                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4178                 break;
4179         case e1000_media_type_internal_serdes:
4180                 ret_val = hw->mac.ops.check_for_link(hw);
4181                 link_active = adapter->hw.mac.serdes_has_link;
4182                 break;
4183         default:
4184         case e1000_media_type_unknown:
4185                 break;
4186         }
4187
4188         if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4189             (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4190                 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4191                 e_info("Gigabit has been disabled, downgrading speed\n");
4192         }
4193
4194         return link_active;
4195 }
4196
4197 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4198 {
4199         /* make sure the receive unit is started */
4200         if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4201             (adapter->flags & FLAG_RX_RESTART_NOW)) {
4202                 struct e1000_hw *hw = &adapter->hw;
4203                 u32 rctl = er32(RCTL);
4204                 ew32(RCTL, rctl | E1000_RCTL_EN);
4205                 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4206         }
4207 }
4208
4209 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4210 {
4211         struct e1000_hw *hw = &adapter->hw;
4212
4213         /*
4214          * With 82574 controllers, PHY needs to be checked periodically
4215          * for hung state and reset, if two calls return true
4216          */
4217         if (e1000_check_phy_82574(hw))
4218                 adapter->phy_hang_count++;
4219         else
4220                 adapter->phy_hang_count = 0;
4221
4222         if (adapter->phy_hang_count > 1) {
4223                 adapter->phy_hang_count = 0;
4224                 schedule_work(&adapter->reset_task);
4225         }
4226 }
4227
4228 /**
4229  * e1000_watchdog - Timer Call-back
4230  * @data: pointer to adapter cast into an unsigned long
4231  **/
4232 static void e1000_watchdog(unsigned long data)
4233 {
4234         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4235
4236         /* Do the rest outside of interrupt context */
4237         schedule_work(&adapter->watchdog_task);
4238
4239         /* TODO: make this use queue_delayed_work() */
4240 }
4241
4242 static void e1000_watchdog_task(struct work_struct *work)
4243 {
4244         struct e1000_adapter *adapter = container_of(work,
4245                                         struct e1000_adapter, watchdog_task);
4246         struct net_device *netdev = adapter->netdev;
4247         struct e1000_mac_info *mac = &adapter->hw.mac;
4248         struct e1000_phy_info *phy = &adapter->hw.phy;
4249         struct e1000_ring *tx_ring = adapter->tx_ring;
4250         struct e1000_hw *hw = &adapter->hw;
4251         u32 link, tctl;
4252
4253         if (test_bit(__E1000_DOWN, &adapter->state))
4254                 return;
4255
4256         link = e1000e_has_link(adapter);
4257         if ((netif_carrier_ok(netdev)) && link) {
4258                 /* Cancel scheduled suspend requests. */
4259                 pm_runtime_resume(netdev->dev.parent);
4260
4261                 e1000e_enable_receives(adapter);
4262                 goto link_up;
4263         }
4264
4265         if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4266             (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4267                 e1000_update_mng_vlan(adapter);
4268
4269         if (link) {
4270                 if (!netif_carrier_ok(netdev)) {
4271                         bool txb2b = 1;
4272
4273                         /* Cancel scheduled suspend requests. */
4274                         pm_runtime_resume(netdev->dev.parent);
4275
4276                         /* update snapshot of PHY registers on LSC */
4277                         e1000_phy_read_status(adapter);
4278                         mac->ops.get_link_up_info(&adapter->hw,
4279                                                    &adapter->link_speed,
4280                                                    &adapter->link_duplex);
4281                         e1000_print_link_info(adapter);
4282                         /*
4283                          * On supported PHYs, check for duplex mismatch only
4284                          * if link has autonegotiated at 10/100 half
4285                          */
4286                         if ((hw->phy.type == e1000_phy_igp_3 ||
4287                              hw->phy.type == e1000_phy_bm) &&
4288                             (hw->mac.autoneg == true) &&
4289                             (adapter->link_speed == SPEED_10 ||
4290                              adapter->link_speed == SPEED_100) &&
4291                             (adapter->link_duplex == HALF_DUPLEX)) {
4292                                 u16 autoneg_exp;
4293
4294                                 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4295
4296                                 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4297                                         e_info("Autonegotiated half duplex but"
4298                                                " link partner cannot autoneg. "
4299                                                " Try forcing full duplex if "
4300                                                "link gets many collisions.\n");
4301                         }
4302
4303                         /* adjust timeout factor according to speed/duplex */
4304                         adapter->tx_timeout_factor = 1;
4305                         switch (adapter->link_speed) {
4306                         case SPEED_10:
4307                                 txb2b = 0;
4308                                 adapter->tx_timeout_factor = 16;
4309                                 break;
4310                         case SPEED_100:
4311                                 txb2b = 0;
4312                                 adapter->tx_timeout_factor = 10;
4313                                 break;
4314                         }
4315
4316                         /*
4317                          * workaround: re-program speed mode bit after
4318                          * link-up event
4319                          */
4320                         if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4321                             !txb2b) {
4322                                 u32 tarc0;
4323                                 tarc0 = er32(TARC(0));
4324                                 tarc0 &= ~SPEED_MODE_BIT;
4325                                 ew32(TARC(0), tarc0);
4326                         }
4327
4328                         /*
4329                          * disable TSO for pcie and 10/100 speeds, to avoid
4330                          * some hardware issues
4331                          */
4332                         if (!(adapter->flags & FLAG_TSO_FORCE)) {
4333                                 switch (adapter->link_speed) {
4334                                 case SPEED_10:
4335                                 case SPEED_100:
4336                                         e_info("10/100 speed: disabling TSO\n");
4337                                         netdev->features &= ~NETIF_F_TSO;
4338                                         netdev->features &= ~NETIF_F_TSO6;
4339                                         break;
4340                                 case SPEED_1000:
4341                                         netdev->features |= NETIF_F_TSO;
4342                                         netdev->features |= NETIF_F_TSO6;
4343                                         break;
4344                                 default:
4345                                         /* oops */
4346                                         break;
4347                                 }
4348                         }
4349
4350                         /*
4351                          * enable transmits in the hardware, need to do this
4352                          * after setting TARC(0)
4353                          */
4354                         tctl = er32(TCTL);
4355                         tctl |= E1000_TCTL_EN;
4356                         ew32(TCTL, tctl);
4357
4358                         /*
4359                          * Perform any post-link-up configuration before
4360                          * reporting link up.
4361                          */
4362                         if (phy->ops.cfg_on_link_up)
4363                                 phy->ops.cfg_on_link_up(hw);
4364
4365                         netif_carrier_on(netdev);
4366
4367                         if (!test_bit(__E1000_DOWN, &adapter->state))
4368                                 mod_timer(&adapter->phy_info_timer,
4369                                           round_jiffies(jiffies + 2 * HZ));
4370                 }
4371         } else {
4372                 if (netif_carrier_ok(netdev)) {
4373                         adapter->link_speed = 0;
4374                         adapter->link_duplex = 0;
4375                         /* Link status message must follow this format */
4376                         printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4377                                adapter->netdev->name);
4378                         netif_carrier_off(netdev);
4379                         if (!test_bit(__E1000_DOWN, &adapter->state))
4380                                 mod_timer(&adapter->phy_info_timer,
4381                                           round_jiffies(jiffies + 2 * HZ));
4382
4383                         if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4384                                 schedule_work(&adapter->reset_task);
4385                         else
4386                                 pm_schedule_suspend(netdev->dev.parent,
4387                                                         LINK_TIMEOUT);
4388                 }
4389         }
4390
4391 link_up:
4392         spin_lock(&adapter->stats64_lock);
4393         e1000e_update_stats(adapter);
4394
4395         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4396         adapter->tpt_old = adapter->stats.tpt;
4397         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4398         adapter->colc_old = adapter->stats.colc;
4399
4400         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4401         adapter->gorc_old = adapter->stats.gorc;
4402         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4403         adapter->gotc_old = adapter->stats.gotc;
4404         spin_unlock(&adapter->stats64_lock);
4405
4406         e1000e_update_adaptive(&adapter->hw);
4407
4408         if (!netif_carrier_ok(netdev) &&
4409             (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4410                 /*
4411                  * We've lost link, so the controller stops DMA,
4412                  * but we've got queued Tx work that's never going
4413                  * to get done, so reset controller to flush Tx.
4414                  * (Do the reset outside of interrupt context).
4415                  */
4416                 schedule_work(&adapter->reset_task);
4417                 /* return immediately since reset is imminent */
4418                 return;
4419         }
4420
4421         /* Simple mode for Interrupt Throttle Rate (ITR) */
4422         if (adapter->itr_setting == 4) {
4423                 /*
4424                  * Symmetric Tx/Rx gets a reduced ITR=2000;
4425                  * Total asymmetrical Tx or Rx gets ITR=8000;
4426                  * everyone else is between 2000-8000.
4427                  */
4428                 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4429                 u32 dif = (adapter->gotc > adapter->gorc ?
4430                             adapter->gotc - adapter->gorc :
4431                             adapter->gorc - adapter->gotc) / 10000;
4432                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4433
4434                 ew32(ITR, 1000000000 / (itr * 256));
4435         }
4436
4437         /* Cause software interrupt to ensure Rx ring is cleaned */
4438         if (adapter->msix_entries)
4439                 ew32(ICS, adapter->rx_ring->ims_val);
4440         else
4441                 ew32(ICS, E1000_ICS_RXDMT0);
4442
4443         /* flush pending descriptors to memory before detecting Tx hang */
4444         e1000e_flush_descriptors(adapter);
4445
4446         /* Force detection of hung controller every watchdog period */
4447         adapter->detect_tx_hung = 1;
4448
4449         /*
4450          * With 82571 controllers, LAA may be overwritten due to controller
4451          * reset from the other port. Set the appropriate LAA in RAR[0]
4452          */
4453         if (e1000e_get_laa_state_82571(hw))
4454                 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4455
4456         if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4457                 e1000e_check_82574_phy_workaround(adapter);
4458
4459         /* Reset the timer */
4460         if (!test_bit(__E1000_DOWN, &adapter->state))
4461                 mod_timer(&adapter->watchdog_timer,
4462                           round_jiffies(jiffies + 2 * HZ));
4463 }
4464
4465 #define E1000_TX_FLAGS_CSUM             0x00000001
4466 #define E1000_TX_FLAGS_VLAN             0x00000002
4467 #define E1000_TX_FLAGS_TSO              0x00000004
4468 #define E1000_TX_FLAGS_IPV4             0x00000008
4469 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
4470 #define E1000_TX_FLAGS_VLAN_SHIFT       16
4471
4472 static int e1000_tso(struct e1000_adapter *adapter,
4473                      struct sk_buff *skb)
4474 {
4475         struct e1000_ring *tx_ring = adapter->tx_ring;
4476         struct e1000_context_desc *context_desc;
4477         struct e1000_buffer *buffer_info;
4478         unsigned int i;
4479         u32 cmd_length = 0;
4480         u16 ipcse = 0, tucse, mss;
4481         u8 ipcss, ipcso, tucss, tucso, hdr_len;
4482
4483         if (!skb_is_gso(skb))
4484                 return 0;
4485
4486         if (skb_header_cloned(skb)) {
4487                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4488
4489                 if (err)
4490                         return err;
4491         }
4492
4493         hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4494         mss = skb_shinfo(skb)->gso_size;
4495         if (skb->protocol == htons(ETH_P_IP)) {
4496                 struct iphdr *iph = ip_hdr(skb);
4497                 iph->tot_len = 0;
4498                 iph->check = 0;
4499                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4500                                                          0, IPPROTO_TCP, 0);
4501                 cmd_length = E1000_TXD_CMD_IP;
4502                 ipcse = skb_transport_offset(skb) - 1;
4503         } else if (skb_is_gso_v6(skb)) {
4504                 ipv6_hdr(skb)->payload_len = 0;
4505                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4506                                                        &ipv6_hdr(skb)->daddr,
4507                                                        0, IPPROTO_TCP, 0);
4508                 ipcse = 0;
4509         }
4510         ipcss = skb_network_offset(skb);
4511         ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4512         tucss = skb_transport_offset(skb);
4513         tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4514         tucse = 0;
4515
4516         cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4517                        E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4518
4519         i = tx_ring->next_to_use;
4520         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4521         buffer_info = &tx_ring->buffer_info[i];
4522
4523         context_desc->lower_setup.ip_fields.ipcss  = ipcss;
4524         context_desc->lower_setup.ip_fields.ipcso  = ipcso;
4525         context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
4526         context_desc->upper_setup.tcp_fields.tucss = tucss;
4527         context_desc->upper_setup.tcp_fields.tucso = tucso;
4528         context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4529         context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
4530         context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4531         context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4532
4533         buffer_info->time_stamp = jiffies;
4534         buffer_info->next_to_watch = i;
4535
4536         i++;
4537         if (i == tx_ring->count)
4538                 i = 0;
4539         tx_ring->next_to_use = i;
4540
4541         return 1;
4542 }
4543
4544 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4545 {
4546         struct e1000_ring *tx_ring = adapter->tx_ring;
4547         struct e1000_context_desc *context_desc;
4548         struct e1000_buffer *buffer_info;
4549         unsigned int i;
4550         u8 css;
4551         u32 cmd_len = E1000_TXD_CMD_DEXT;
4552         __be16 protocol;
4553
4554         if (skb->ip_summed != CHECKSUM_PARTIAL)
4555                 return 0;
4556
4557         if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4558                 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4559         else
4560                 protocol = skb->protocol;
4561
4562         switch (protocol) {
4563         case cpu_to_be16(ETH_P_IP):
4564                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4565                         cmd_len |= E1000_TXD_CMD_TCP;
4566                 break;
4567         case cpu_to_be16(ETH_P_IPV6):
4568                 /* XXX not handling all IPV6 headers */
4569                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4570                         cmd_len |= E1000_TXD_CMD_TCP;
4571                 break;
4572         default:
4573                 if (unlikely(net_ratelimit()))
4574                         e_warn("checksum_partial proto=%x!\n",
4575                                be16_to_cpu(protocol));
4576                 break;
4577         }
4578
4579         css = skb_checksum_start_offset(skb);
4580
4581         i = tx_ring->next_to_use;
4582         buffer_info = &tx_ring->buffer_info[i];
4583         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4584
4585         context_desc->lower_setup.ip_config = 0;
4586         context_desc->upper_setup.tcp_fields.tucss = css;
4587         context_desc->upper_setup.tcp_fields.tucso =
4588                                 css + skb->csum_offset;
4589         context_desc->upper_setup.tcp_fields.tucse = 0;
4590         context_desc->tcp_seg_setup.data = 0;
4591         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4592
4593         buffer_info->time_stamp = jiffies;
4594         buffer_info->next_to_watch = i;
4595
4596         i++;
4597         if (i == tx_ring->count)
4598                 i = 0;
4599         tx_ring->next_to_use = i;
4600
4601         return 1;
4602 }
4603
4604 #define E1000_MAX_PER_TXD       8192
4605 #define E1000_MAX_TXD_PWR       12
4606
4607 static int e1000_tx_map(struct e1000_adapter *adapter,
4608                         struct sk_buff *skb, unsigned int first,
4609                         unsigned int max_per_txd, unsigned int nr_frags,
4610                         unsigned int mss)
4611 {
4612         struct e1000_ring *tx_ring = adapter->tx_ring;
4613         struct pci_dev *pdev = adapter->pdev;
4614         struct e1000_buffer *buffer_info;
4615         unsigned int len = skb_headlen(skb);
4616         unsigned int offset = 0, size, count = 0, i;
4617         unsigned int f, bytecount, segs;
4618
4619         i = tx_ring->next_to_use;
4620
4621         while (len) {
4622                 buffer_info = &tx_ring->buffer_info[i];
4623                 size = min(len, max_per_txd);
4624
4625                 buffer_info->length = size;
4626                 buffer_info->time_stamp = jiffies;
4627                 buffer_info->next_to_watch = i;
4628                 buffer_info->dma = dma_map_single(&pdev->dev,
4629                                                   skb->data + offset,
4630                                                   size, DMA_TO_DEVICE);
4631                 buffer_info->mapped_as_page = false;
4632                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4633                         goto dma_error;
4634
4635                 len -= size;
4636                 offset += size;
4637                 count++;
4638
4639                 if (len) {
4640                         i++;
4641                         if (i == tx_ring->count)
4642                                 i = 0;
4643                 }
4644         }
4645
4646         for (f = 0; f < nr_frags; f++) {
4647                 struct skb_frag_struct *frag;
4648
4649                 frag = &skb_shinfo(skb)->frags[f];
4650                 len = frag->size;
4651                 offset = frag->page_offset;
4652
4653                 while (len) {
4654                         i++;
4655                         if (i == tx_ring->count)
4656                                 i = 0;
4657
4658                         buffer_info = &tx_ring->buffer_info[i];
4659                         size = min(len, max_per_txd);
4660
4661                         buffer_info->length = size;
4662                         buffer_info->time_stamp = jiffies;
4663                         buffer_info->next_to_watch = i;
4664                         buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4665                                                         offset, size,
4666                                                         DMA_TO_DEVICE);
4667                         buffer_info->mapped_as_page = true;
4668                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4669                                 goto dma_error;
4670
4671                         len -= size;
4672                         offset += size;
4673                         count++;
4674                 }
4675         }
4676
4677         segs = skb_shinfo(skb)->gso_segs ? : 1;
4678         /* multiply data chunks by size of headers */
4679         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4680
4681         tx_ring->buffer_info[i].skb = skb;
4682         tx_ring->buffer_info[i].segs = segs;
4683         tx_ring->buffer_info[i].bytecount = bytecount;
4684         tx_ring->buffer_info[first].next_to_watch = i;
4685
4686         return count;
4687
4688 dma_error:
4689         dev_err(&pdev->dev, "Tx DMA map failed\n");
4690         buffer_info->dma = 0;
4691         if (count)
4692                 count--;
4693
4694         while (count--) {
4695                 if (i == 0)
4696                         i += tx_ring->count;
4697                 i--;
4698                 buffer_info = &tx_ring->buffer_info[i];
4699                 e1000_put_txbuf(adapter, buffer_info);
4700         }
4701
4702         return 0;
4703 }
4704
4705 static void e1000_tx_queue(struct e1000_adapter *adapter,
4706                            int tx_flags, int count)
4707 {
4708         struct e1000_ring *tx_ring = adapter->tx_ring;
4709         struct e1000_tx_desc *tx_desc = NULL;
4710         struct e1000_buffer *buffer_info;
4711         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4712         unsigned int i;
4713
4714         if (tx_flags & E1000_TX_FLAGS_TSO) {
4715                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4716                              E1000_TXD_CMD_TSE;
4717                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4718
4719                 if (tx_flags & E1000_TX_FLAGS_IPV4)
4720                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4721         }
4722
4723         if (tx_flags & E1000_TX_FLAGS_CSUM) {
4724                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4725                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4726         }
4727
4728         if (tx_flags & E1000_TX_FLAGS_VLAN) {
4729                 txd_lower |= E1000_TXD_CMD_VLE;
4730                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4731         }
4732
4733         i = tx_ring->next_to_use;
4734
4735         do {
4736                 buffer_info = &tx_ring->buffer_info[i];
4737                 tx_desc = E1000_TX_DESC(*tx_ring, i);
4738                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4739                 tx_desc->lower.data =
4740                         cpu_to_le32(txd_lower | buffer_info->length);
4741                 tx_desc->upper.data = cpu_to_le32(txd_upper);
4742
4743                 i++;
4744                 if (i == tx_ring->count)
4745                         i = 0;
4746         } while (--count > 0);
4747
4748         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4749
4750         /*
4751          * Force memory writes to complete before letting h/w
4752          * know there are new descriptors to fetch.  (Only
4753          * applicable for weak-ordered memory model archs,
4754          * such as IA-64).
4755          */
4756         wmb();
4757
4758         tx_ring->next_to_use = i;
4759
4760         if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4761                 e1000e_update_tdt_wa(adapter, i);
4762         else
4763                 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4764
4765         /*
4766          * we need this if more than one processor can write to our tail
4767          * at a time, it synchronizes IO on IA64/Altix systems
4768          */
4769         mmiowb();
4770 }
4771
4772 #define MINIMUM_DHCP_PACKET_SIZE 282
4773 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4774                                     struct sk_buff *skb)
4775 {
4776         struct e1000_hw *hw =  &adapter->hw;
4777         u16 length, offset;
4778
4779         if (vlan_tx_tag_present(skb)) {
4780                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4781                     (adapter->hw.mng_cookie.status &
4782                         E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4783                         return 0;
4784         }
4785
4786         if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4787                 return 0;
4788
4789         if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4790                 return 0;
4791
4792         {
4793                 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4794                 struct udphdr *udp;
4795
4796                 if (ip->protocol != IPPROTO_UDP)
4797                         return 0;
4798
4799                 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4800                 if (ntohs(udp->dest) != 67)
4801                         return 0;
4802
4803                 offset = (u8 *)udp + 8 - skb->data;
4804                 length = skb->len - offset;
4805                 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4806         }
4807
4808         return 0;
4809 }
4810
4811 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4812 {
4813         struct e1000_adapter *adapter = netdev_priv(netdev);
4814
4815         netif_stop_queue(netdev);
4816         /*
4817          * Herbert's original patch had:
4818          *  smp_mb__after_netif_stop_queue();
4819          * but since that doesn't exist yet, just open code it.
4820          */
4821         smp_mb();
4822
4823         /*
4824          * We need to check again in a case another CPU has just
4825          * made room available.
4826          */
4827         if (e1000_desc_unused(adapter->tx_ring) < size)
4828                 return -EBUSY;
4829
4830         /* A reprieve! */
4831         netif_start_queue(netdev);
4832         ++adapter->restart_queue;
4833         return 0;
4834 }
4835
4836 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4837 {
4838         struct e1000_adapter *adapter = netdev_priv(netdev);
4839
4840         if (e1000_desc_unused(adapter->tx_ring) >= size)
4841                 return 0;
4842         return __e1000_maybe_stop_tx(netdev, size);
4843 }
4844
4845 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4846 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4847                                     struct net_device *netdev)
4848 {
4849         struct e1000_adapter *adapter = netdev_priv(netdev);
4850         struct e1000_ring *tx_ring = adapter->tx_ring;
4851         unsigned int first;
4852         unsigned int max_per_txd = E1000_MAX_PER_TXD;
4853         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4854         unsigned int tx_flags = 0;
4855         unsigned int len = skb_headlen(skb);
4856         unsigned int nr_frags;
4857         unsigned int mss;
4858         int count = 0;
4859         int tso;
4860         unsigned int f;
4861
4862         if (test_bit(__E1000_DOWN, &adapter->state)) {
4863                 dev_kfree_skb_any(skb);
4864                 return NETDEV_TX_OK;
4865         }
4866
4867         if (skb->len <= 0) {
4868                 dev_kfree_skb_any(skb);
4869                 return NETDEV_TX_OK;
4870         }
4871
4872         mss = skb_shinfo(skb)->gso_size;
4873         /*
4874          * The controller does a simple calculation to
4875          * make sure there is enough room in the FIFO before
4876          * initiating the DMA for each buffer.  The calc is:
4877          * 4 = ceil(buffer len/mss).  To make sure we don't
4878          * overrun the FIFO, adjust the max buffer len if mss
4879          * drops.
4880          */
4881         if (mss) {
4882                 u8 hdr_len;
4883                 max_per_txd = min(mss << 2, max_per_txd);
4884                 max_txd_pwr = fls(max_per_txd) - 1;
4885
4886                 /*
4887                  * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4888                  * points to just header, pull a few bytes of payload from
4889                  * frags into skb->data
4890                  */
4891                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4892                 /*
4893                  * we do this workaround for ES2LAN, but it is un-necessary,
4894                  * avoiding it could save a lot of cycles
4895                  */
4896                 if (skb->data_len && (hdr_len == len)) {
4897                         unsigned int pull_size;
4898
4899                         pull_size = min((unsigned int)4, skb->data_len);
4900                         if (!__pskb_pull_tail(skb, pull_size)) {
4901                                 e_err("__pskb_pull_tail failed.\n");
4902                                 dev_kfree_skb_any(skb);
4903                                 return NETDEV_TX_OK;
4904                         }
4905                         len = skb_headlen(skb);
4906                 }
4907         }
4908
4909         /* reserve a descriptor for the offload context */
4910         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4911                 count++;
4912         count++;
4913
4914         count += TXD_USE_COUNT(len, max_txd_pwr);
4915
4916         nr_frags = skb_shinfo(skb)->nr_frags;
4917         for (f = 0; f < nr_frags; f++)
4918                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4919                                        max_txd_pwr);
4920
4921         if (adapter->hw.mac.tx_pkt_filtering)
4922                 e1000_transfer_dhcp_info(adapter, skb);
4923
4924         /*
4925          * need: count + 2 desc gap to keep tail from touching
4926          * head, otherwise try next time
4927          */
4928         if (e1000_maybe_stop_tx(netdev, count + 2))
4929                 return NETDEV_TX_BUSY;
4930
4931         if (vlan_tx_tag_present(skb)) {
4932                 tx_flags |= E1000_TX_FLAGS_VLAN;
4933                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4934         }
4935
4936         first = tx_ring->next_to_use;
4937
4938         tso = e1000_tso(adapter, skb);
4939         if (tso < 0) {
4940                 dev_kfree_skb_any(skb);
4941                 return NETDEV_TX_OK;
4942         }
4943
4944         if (tso)
4945                 tx_flags |= E1000_TX_FLAGS_TSO;
4946         else if (e1000_tx_csum(adapter, skb))
4947                 tx_flags |= E1000_TX_FLAGS_CSUM;
4948
4949         /*
4950          * Old method was to assume IPv4 packet by default if TSO was enabled.
4951          * 82571 hardware supports TSO capabilities for IPv6 as well...
4952          * no longer assume, we must.
4953          */
4954         if (skb->protocol == htons(ETH_P_IP))
4955                 tx_flags |= E1000_TX_FLAGS_IPV4;
4956
4957         /* if count is 0 then mapping error has occurred */
4958         count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4959         if (count) {
4960                 e1000_tx_queue(adapter, tx_flags, count);
4961                 /* Make sure there is space in the ring for the next send. */
4962                 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4963
4964         } else {
4965                 dev_kfree_skb_any(skb);
4966                 tx_ring->buffer_info[first].time_stamp = 0;
4967                 tx_ring->next_to_use = first;
4968         }
4969
4970         return NETDEV_TX_OK;
4971 }
4972
4973 /**
4974  * e1000_tx_timeout - Respond to a Tx Hang
4975  * @netdev: network interface device structure
4976  **/
4977 static void e1000_tx_timeout(struct net_device *netdev)
4978 {
4979         struct e1000_adapter *adapter = netdev_priv(netdev);
4980
4981         /* Do the reset outside of interrupt context */
4982         adapter->tx_timeout_count++;
4983         schedule_work(&adapter->reset_task);
4984 }
4985
4986 static void e1000_reset_task(struct work_struct *work)
4987 {
4988         struct e1000_adapter *adapter;
4989         adapter = container_of(work, struct e1000_adapter, reset_task);
4990
4991         /* don't run the task if already down */
4992         if (test_bit(__E1000_DOWN, &adapter->state))
4993                 return;
4994
4995         if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4996               (adapter->flags & FLAG_RX_RESTART_NOW))) {
4997                 e1000e_dump(adapter);
4998                 e_err("Reset adapter\n");
4999         }
5000         e1000e_reinit_locked(adapter);
5001 }
5002
5003 /**
5004  * e1000_get_stats64 - Get System Network Statistics
5005  * @netdev: network interface device structure
5006  * @stats: rtnl_link_stats64 pointer
5007  *
5008  * Returns the address of the device statistics structure.
5009  **/
5010 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5011                                              struct rtnl_link_stats64 *stats)
5012 {
5013         struct e1000_adapter *adapter = netdev_priv(netdev);
5014
5015         memset(stats, 0, sizeof(struct rtnl_link_stats64));
5016         spin_lock(&adapter->stats64_lock);
5017         e1000e_update_stats(adapter);
5018         /* Fill out the OS statistics structure */
5019         stats->rx_bytes = adapter->stats.gorc;
5020         stats->rx_packets = adapter->stats.gprc;
5021         stats->tx_bytes = adapter->stats.gotc;
5022         stats->tx_packets = adapter->stats.gptc;
5023         stats->multicast = adapter->stats.mprc;
5024         stats->collisions = adapter->stats.colc;
5025
5026         /* Rx Errors */
5027
5028         /*
5029          * RLEC on some newer hardware can be incorrect so build
5030          * our own version based on RUC and ROC
5031          */
5032         stats->rx_errors = adapter->stats.rxerrc +
5033                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5034                 adapter->stats.ruc + adapter->stats.roc +
5035                 adapter->stats.cexterr;
5036         stats->rx_length_errors = adapter->stats.ruc +
5037                                               adapter->stats.roc;
5038         stats->rx_crc_errors = adapter->stats.crcerrs;
5039         stats->rx_frame_errors = adapter->stats.algnerrc;
5040         stats->rx_missed_errors = adapter->stats.mpc;
5041
5042         /* Tx Errors */
5043         stats->tx_errors = adapter->stats.ecol +
5044                                        adapter->stats.latecol;
5045         stats->tx_aborted_errors = adapter->stats.ecol;
5046         stats->tx_window_errors = adapter->stats.latecol;
5047         stats->tx_carrier_errors = adapter->stats.tncrs;
5048
5049         /* Tx Dropped needs to be maintained elsewhere */
5050
5051         spin_unlock(&adapter->stats64_lock);
5052         return stats;
5053 }
5054
5055 /**
5056  * e1000_change_mtu - Change the Maximum Transfer Unit
5057  * @netdev: network interface device structure
5058  * @new_mtu: new value for maximum frame size
5059  *
5060  * Returns 0 on success, negative on failure
5061  **/
5062 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5063 {
5064         struct e1000_adapter *adapter = netdev_priv(netdev);
5065         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5066
5067         /* Jumbo frame support */
5068         if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5069             !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5070                 e_err("Jumbo Frames not supported.\n");
5071                 return -EINVAL;
5072         }
5073
5074         /* Supported frame sizes */
5075         if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5076             (max_frame > adapter->max_hw_frame_size)) {
5077                 e_err("Unsupported MTU setting\n");
5078                 return -EINVAL;
5079         }
5080
5081         /* Jumbo frame workaround on 82579 requires CRC be stripped */
5082         if ((adapter->hw.mac.type == e1000_pch2lan) &&
5083             !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5084             (new_mtu > ETH_DATA_LEN)) {
5085                 e_err("Jumbo Frames not supported on 82579 when CRC "
5086                       "stripping is disabled.\n");
5087                 return -EINVAL;
5088         }
5089
5090         /* 82573 Errata 17 */
5091         if (((adapter->hw.mac.type == e1000_82573) ||
5092              (adapter->hw.mac.type == e1000_82574)) &&
5093             (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5094                 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5095                 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5096         }
5097
5098         while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5099                 usleep_range(1000, 2000);
5100         /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5101         adapter->max_frame_size = max_frame;
5102         e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5103         netdev->mtu = new_mtu;
5104         if (netif_running(netdev))
5105                 e1000e_down(adapter);
5106
5107         /*
5108          * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5109          * means we reserve 2 more, this pushes us to allocate from the next
5110          * larger slab size.
5111          * i.e. RXBUFFER_2048 --> size-4096 slab
5112          * However with the new *_jumbo_rx* routines, jumbo receives will use
5113          * fragmented skbs
5114          */
5115
5116         if (max_frame <= 2048)
5117                 adapter->rx_buffer_len = 2048;
5118         else
5119                 adapter->rx_buffer_len = 4096;
5120
5121         /* adjust allocation if LPE protects us, and we aren't using SBP */
5122         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5123              (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5124                 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5125                                          + ETH_FCS_LEN;
5126
5127         if (netif_running(netdev))
5128                 e1000e_up(adapter);
5129         else
5130                 e1000e_reset(adapter);
5131
5132         clear_bit(__E1000_RESETTING, &adapter->state);
5133
5134         return 0;
5135 }
5136
5137 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5138                            int cmd)
5139 {
5140         struct e1000_adapter *adapter = netdev_priv(netdev);
5141         struct mii_ioctl_data *data = if_mii(ifr);
5142
5143         if (adapter->hw.phy.media_type != e1000_media_type_copper)
5144                 return -EOPNOTSUPP;
5145
5146         switch (cmd) {
5147         case SIOCGMIIPHY:
5148                 data->phy_id = adapter->hw.phy.addr;
5149                 break;
5150         case SIOCGMIIREG:
5151                 e1000_phy_read_status(adapter);
5152
5153                 switch (data->reg_num & 0x1F) {
5154                 case MII_BMCR:
5155                         data->val_out = adapter->phy_regs.bmcr;
5156                         break;
5157                 case MII_BMSR:
5158                         data->val_out = adapter->phy_regs.bmsr;
5159                         break;
5160                 case MII_PHYSID1:
5161                         data->val_out = (adapter->hw.phy.id >> 16);
5162                         break;
5163                 case MII_PHYSID2:
5164                         data->val_out = (adapter->hw.phy.id & 0xFFFF);
5165                         break;
5166                 case MII_ADVERTISE:
5167                         data->val_out = adapter->phy_regs.advertise;
5168                         break;
5169                 case MII_LPA:
5170                         data->val_out = adapter->phy_regs.lpa;
5171                         break;
5172                 case MII_EXPANSION:
5173                         data->val_out = adapter->phy_regs.expansion;
5174                         break;
5175                 case MII_CTRL1000:
5176                         data->val_out = adapter->phy_regs.ctrl1000;
5177                         break;
5178                 case MII_STAT1000:
5179                         data->val_out = adapter->phy_regs.stat1000;
5180                         break;
5181                 case MII_ESTATUS:
5182                         data->val_out = adapter->phy_regs.estatus;
5183                         break;
5184                 default:
5185                         return -EIO;
5186                 }
5187                 break;
5188         case SIOCSMIIREG:
5189         default:
5190                 return -EOPNOTSUPP;
5191         }
5192         return 0;
5193 }
5194
5195 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5196 {
5197         switch (cmd) {
5198         case SIOCGMIIPHY:
5199         case SIOCGMIIREG:
5200         case SIOCSMIIREG:
5201                 return e1000_mii_ioctl(netdev, ifr, cmd);
5202         default:
5203                 return -EOPNOTSUPP;
5204         }
5205 }
5206
5207 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5208 {
5209         struct e1000_hw *hw = &adapter->hw;
5210         u32 i, mac_reg;
5211         u16 phy_reg, wuc_enable;
5212         int retval = 0;
5213
5214         /* copy MAC RARs to PHY RARs */
5215         e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5216
5217         retval = hw->phy.ops.acquire(hw);
5218         if (retval) {
5219                 e_err("Could not acquire PHY\n");
5220                 return retval;
5221         }
5222
5223         /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5224         retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5225         if (retval)
5226                 goto out;
5227
5228         /* copy MAC MTA to PHY MTA - only needed for pchlan */
5229         for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5230                 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5231                 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5232                                            (u16)(mac_reg & 0xFFFF));
5233                 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5234                                            (u16)((mac_reg >> 16) & 0xFFFF));
5235         }
5236
5237         /* configure PHY Rx Control register */
5238         hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5239         mac_reg = er32(RCTL);
5240         if (mac_reg & E1000_RCTL_UPE)
5241                 phy_reg |= BM_RCTL_UPE;
5242         if (mac_reg & E1000_RCTL_MPE)
5243                 phy_reg |= BM_RCTL_MPE;
5244         phy_reg &= ~(BM_RCTL_MO_MASK);
5245         if (mac_reg & E1000_RCTL_MO_3)
5246                 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5247                                 << BM_RCTL_MO_SHIFT);
5248         if (mac_reg & E1000_RCTL_BAM)
5249                 phy_reg |= BM_RCTL_BAM;
5250         if (mac_reg & E1000_RCTL_PMCF)
5251                 phy_reg |= BM_RCTL_PMCF;
5252         mac_reg = er32(CTRL);
5253         if (mac_reg & E1000_CTRL_RFCE)
5254                 phy_reg |= BM_RCTL_RFCE;
5255         hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5256
5257         /* enable PHY wakeup in MAC register */
5258         ew32(WUFC, wufc);
5259         ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5260
5261         /* configure and enable PHY wakeup in PHY registers */
5262         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5263         hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5264
5265         /* activate PHY wakeup */
5266         wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5267         retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5268         if (retval)
5269                 e_err("Could not set PHY Host Wakeup bit\n");
5270 out:
5271         hw->phy.ops.release(hw);
5272
5273         return retval;
5274 }
5275
5276 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5277                             bool runtime)
5278 {
5279         struct net_device *netdev = pci_get_drvdata(pdev);
5280         struct e1000_adapter *adapter = netdev_priv(netdev);
5281         struct e1000_hw *hw = &adapter->hw;
5282         u32 ctrl, ctrl_ext, rctl, status;
5283         /* Runtime suspend should only enable wakeup for link changes */
5284         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5285         int retval = 0;
5286
5287         netif_device_detach(netdev);
5288
5289         if (netif_running(netdev)) {
5290                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5291                 e1000e_down(adapter);
5292                 e1000_free_irq(adapter);
5293         }
5294         e1000e_reset_interrupt_capability(adapter);
5295
5296         retval = pci_save_state(pdev);
5297         if (retval)
5298                 return retval;
5299
5300         status = er32(STATUS);
5301         if (status & E1000_STATUS_LU)
5302                 wufc &= ~E1000_WUFC_LNKC;
5303
5304         if (wufc) {
5305                 e1000_setup_rctl(adapter);
5306                 e1000_set_multi(netdev);
5307
5308                 /* turn on all-multi mode if wake on multicast is enabled */
5309                 if (wufc & E1000_WUFC_MC) {
5310                         rctl = er32(RCTL);
5311                         rctl |= E1000_RCTL_MPE;
5312                         ew32(RCTL, rctl);
5313                 }
5314
5315                 ctrl = er32(CTRL);
5316                 /* advertise wake from D3Cold */
5317                 #define E1000_CTRL_ADVD3WUC 0x00100000
5318                 /* phy power management enable */
5319                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5320                 ctrl |= E1000_CTRL_ADVD3WUC;
5321                 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5322                         ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5323                 ew32(CTRL, ctrl);
5324
5325                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5326                     adapter->hw.phy.media_type ==
5327                     e1000_media_type_internal_serdes) {
5328                         /* keep the laser running in D3 */
5329                         ctrl_ext = er32(CTRL_EXT);
5330                         ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5331                         ew32(CTRL_EXT, ctrl_ext);
5332                 }
5333
5334                 if (adapter->flags & FLAG_IS_ICH)
5335                         e1000_suspend_workarounds_ich8lan(&adapter->hw);
5336
5337                 /* Allow time for pending master requests to run */
5338                 e1000e_disable_pcie_master(&adapter->hw);
5339
5340                 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5341                         /* enable wakeup by the PHY */
5342                         retval = e1000_init_phy_wakeup(adapter, wufc);
5343                         if (retval)
5344                                 return retval;
5345                 } else {
5346                         /* enable wakeup by the MAC */
5347                         ew32(WUFC, wufc);
5348                         ew32(WUC, E1000_WUC_PME_EN);
5349                 }
5350         } else {
5351                 ew32(WUC, 0);
5352                 ew32(WUFC, 0);
5353         }
5354
5355         *enable_wake = !!wufc;
5356
5357         /* make sure adapter isn't asleep if manageability is enabled */
5358         if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5359             (hw->mac.ops.check_mng_mode(hw)))
5360                 *enable_wake = true;
5361
5362         if (adapter->hw.phy.type == e1000_phy_igp_3)
5363                 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5364
5365         /*
5366          * Release control of h/w to f/w.  If f/w is AMT enabled, this
5367          * would have already happened in close and is redundant.
5368          */
5369         e1000e_release_hw_control(adapter);
5370
5371         pci_disable_device(pdev);
5372
5373         return 0;
5374 }
5375
5376 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5377 {
5378         if (sleep && wake) {
5379                 pci_prepare_to_sleep(pdev);
5380                 return;
5381         }
5382
5383         pci_wake_from_d3(pdev, wake);
5384         pci_set_power_state(pdev, PCI_D3hot);
5385 }
5386
5387 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5388                                     bool wake)
5389 {
5390         struct net_device *netdev = pci_get_drvdata(pdev);
5391         struct e1000_adapter *adapter = netdev_priv(netdev);
5392
5393         /*
5394          * The pci-e switch on some quad port adapters will report a
5395          * correctable error when the MAC transitions from D0 to D3.  To
5396          * prevent this we need to mask off the correctable errors on the
5397          * downstream port of the pci-e switch.
5398          */
5399         if (adapter->flags & FLAG_IS_QUAD_PORT) {
5400                 struct pci_dev *us_dev = pdev->bus->self;
5401                 int pos = pci_pcie_cap(us_dev);
5402                 u16 devctl;
5403
5404                 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5405                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5406                                       (devctl & ~PCI_EXP_DEVCTL_CERE));
5407
5408                 e1000_power_off(pdev, sleep, wake);
5409
5410                 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5411         } else {
5412                 e1000_power_off(pdev, sleep, wake);
5413         }
5414 }
5415
5416 #ifdef CONFIG_PCIEASPM
5417 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5418 {
5419         pci_disable_link_state_locked(pdev, state);
5420 }
5421 #else
5422 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5423 {
5424         int pos;
5425         u16 reg16;
5426
5427         /*
5428          * Both device and parent should have the same ASPM setting.
5429          * Disable ASPM in downstream component first and then upstream.
5430          */
5431         pos = pci_pcie_cap(pdev);
5432         pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, &reg16);
5433         reg16 &= ~state;
5434         pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5435
5436         if (!pdev->bus->self)
5437                 return;
5438
5439         pos = pci_pcie_cap(pdev->bus->self);
5440         pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, &reg16);
5441         reg16 &= ~state;
5442         pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5443 }
5444 #endif
5445 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5446 {
5447         dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5448                  (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5449                  (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5450
5451         __e1000e_disable_aspm(pdev, state);
5452 }
5453
5454 #ifdef CONFIG_PM
5455 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5456 {
5457         return !!adapter->tx_ring->buffer_info;
5458 }
5459
5460 static int __e1000_resume(struct pci_dev *pdev)
5461 {
5462         struct net_device *netdev = pci_get_drvdata(pdev);
5463         struct e1000_adapter *adapter = netdev_priv(netdev);
5464         struct e1000_hw *hw = &adapter->hw;
5465         u16 aspm_disable_flag = 0;
5466         u32 err;
5467
5468         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5469                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5470         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5471                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5472         if (aspm_disable_flag)
5473                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5474
5475         pci_set_power_state(pdev, PCI_D0);
5476         pci_restore_state(pdev);
5477         pci_save_state(pdev);
5478
5479         e1000e_set_interrupt_capability(adapter);
5480         if (netif_running(netdev)) {
5481                 err = e1000_request_irq(adapter);
5482                 if (err)
5483                         return err;
5484         }
5485
5486         if (hw->mac.type == e1000_pch2lan)
5487                 e1000_resume_workarounds_pchlan(&adapter->hw);
5488
5489         e1000e_power_up_phy(adapter);
5490
5491         /* report the system wakeup cause from S3/S4 */
5492         if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5493                 u16 phy_data;
5494
5495                 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5496                 if (phy_data) {
5497                         e_info("PHY Wakeup cause - %s\n",
5498                                 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5499                                 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5500                                 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5501                                 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5502                                 phy_data & E1000_WUS_LNKC ? "Link Status "
5503                                 " Change" : "other");
5504                 }
5505                 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5506         } else {
5507                 u32 wus = er32(WUS);
5508                 if (wus) {
5509                         e_info("MAC Wakeup cause - %s\n",
5510                                 wus & E1000_WUS_EX ? "Unicast Packet" :
5511                                 wus & E1000_WUS_MC ? "Multicast Packet" :
5512                                 wus & E1000_WUS_BC ? "Broadcast Packet" :
5513                                 wus & E1000_WUS_MAG ? "Magic Packet" :
5514                                 wus & E1000_WUS_LNKC ? "Link Status Change" :
5515                                 "other");
5516                 }
5517                 ew32(WUS, ~0);
5518         }
5519
5520         e1000e_reset(adapter);
5521
5522         e1000_init_manageability_pt(adapter);
5523
5524         if (netif_running(netdev))
5525                 e1000e_up(adapter);
5526
5527         netif_device_attach(netdev);
5528
5529         /*
5530          * If the controller has AMT, do not set DRV_LOAD until the interface
5531          * is up.  For all other cases, let the f/w know that the h/w is now
5532          * under the control of the driver.
5533          */
5534         if (!(adapter->flags & FLAG_HAS_AMT))
5535                 e1000e_get_hw_control(adapter);
5536
5537         return 0;
5538 }
5539
5540 #ifdef CONFIG_PM_SLEEP
5541 static int e1000_suspend(struct device *dev)
5542 {
5543         struct pci_dev *pdev = to_pci_dev(dev);
5544         int retval;
5545         bool wake;
5546
5547         retval = __e1000_shutdown(pdev, &wake, false);
5548         if (!retval)
5549                 e1000_complete_shutdown(pdev, true, wake);
5550
5551         return retval;
5552 }
5553
5554 static int e1000_resume(struct device *dev)
5555 {
5556         struct pci_dev *pdev = to_pci_dev(dev);
5557         struct net_device *netdev = pci_get_drvdata(pdev);
5558         struct e1000_adapter *adapter = netdev_priv(netdev);
5559
5560         if (e1000e_pm_ready(adapter))
5561                 adapter->idle_check = true;
5562
5563         return __e1000_resume(pdev);
5564 }
5565 #endif /* CONFIG_PM_SLEEP */
5566
5567 #ifdef CONFIG_PM_RUNTIME
5568 static int e1000_runtime_suspend(struct device *dev)
5569 {
5570         struct pci_dev *pdev = to_pci_dev(dev);
5571         struct net_device *netdev = pci_get_drvdata(pdev);
5572         struct e1000_adapter *adapter = netdev_priv(netdev);
5573
5574         if (e1000e_pm_ready(adapter)) {
5575                 bool wake;
5576
5577                 __e1000_shutdown(pdev, &wake, true);
5578         }
5579
5580         return 0;
5581 }
5582
5583 static int e1000_idle(struct device *dev)
5584 {
5585         struct pci_dev *pdev = to_pci_dev(dev);
5586         struct net_device *netdev = pci_get_drvdata(pdev);
5587         struct e1000_adapter *adapter = netdev_priv(netdev);
5588
5589         if (!e1000e_pm_ready(adapter))
5590                 return 0;
5591
5592         if (adapter->idle_check) {
5593                 adapter->idle_check = false;
5594                 if (!e1000e_has_link(adapter))
5595                         pm_schedule_suspend(dev, MSEC_PER_SEC);
5596         }
5597
5598         return -EBUSY;
5599 }
5600
5601 static int e1000_runtime_resume(struct device *dev)
5602 {
5603         struct pci_dev *pdev = to_pci_dev(dev);
5604         struct net_device *netdev = pci_get_drvdata(pdev);
5605         struct e1000_adapter *adapter = netdev_priv(netdev);
5606
5607         if (!e1000e_pm_ready(adapter))
5608                 return 0;
5609
5610         adapter->idle_check = !dev->power.runtime_auto;
5611         return __e1000_resume(pdev);
5612 }
5613 #endif /* CONFIG_PM_RUNTIME */
5614 #endif /* CONFIG_PM */
5615
5616 static void e1000_shutdown(struct pci_dev *pdev)
5617 {
5618         bool wake = false;
5619
5620         __e1000_shutdown(pdev, &wake, false);
5621
5622         if (system_state == SYSTEM_POWER_OFF)
5623                 e1000_complete_shutdown(pdev, false, wake);
5624 }
5625
5626 #ifdef CONFIG_NET_POLL_CONTROLLER
5627
5628 static irqreturn_t e1000_intr_msix(int irq, void *data)
5629 {
5630         struct net_device *netdev = data;
5631         struct e1000_adapter *adapter = netdev_priv(netdev);
5632
5633         if (adapter->msix_entries) {
5634                 int vector, msix_irq;
5635
5636                 vector = 0;
5637                 msix_irq = adapter->msix_entries[vector].vector;
5638                 disable_irq(msix_irq);
5639                 e1000_intr_msix_rx(msix_irq, netdev);
5640                 enable_irq(msix_irq);
5641
5642                 vector++;
5643                 msix_irq = adapter->msix_entries[vector].vector;
5644                 disable_irq(msix_irq);
5645                 e1000_intr_msix_tx(msix_irq, netdev);
5646                 enable_irq(msix_irq);
5647
5648                 vector++;
5649                 msix_irq = adapter->msix_entries[vector].vector;
5650                 disable_irq(msix_irq);
5651                 e1000_msix_other(msix_irq, netdev);
5652                 enable_irq(msix_irq);
5653         }
5654
5655         return IRQ_HANDLED;
5656 }
5657
5658 /*
5659  * Polling 'interrupt' - used by things like netconsole to send skbs
5660  * without having to re-enable interrupts. It's not called while
5661  * the interrupt routine is executing.
5662  */
5663 static void e1000_netpoll(struct net_device *netdev)
5664 {
5665         struct e1000_adapter *adapter = netdev_priv(netdev);
5666
5667         switch (adapter->int_mode) {
5668         case E1000E_INT_MODE_MSIX:
5669                 e1000_intr_msix(adapter->pdev->irq, netdev);
5670                 break;
5671         case E1000E_INT_MODE_MSI:
5672                 disable_irq(adapter->pdev->irq);
5673                 e1000_intr_msi(adapter->pdev->irq, netdev);
5674                 enable_irq(adapter->pdev->irq);
5675                 break;
5676         default: /* E1000E_INT_MODE_LEGACY */
5677                 disable_irq(adapter->pdev->irq);
5678                 e1000_intr(adapter->pdev->irq, netdev);
5679                 enable_irq(adapter->pdev->irq);
5680                 break;
5681         }
5682 }
5683 #endif
5684
5685 /**
5686  * e1000_io_error_detected - called when PCI error is detected
5687  * @pdev: Pointer to PCI device
5688  * @state: The current pci connection state
5689  *
5690  * This function is called after a PCI bus error affecting
5691  * this device has been detected.
5692  */
5693 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5694                                                 pci_channel_state_t state)
5695 {
5696         struct net_device *netdev = pci_get_drvdata(pdev);
5697         struct e1000_adapter *adapter = netdev_priv(netdev);
5698
5699         netif_device_detach(netdev);
5700
5701         if (state == pci_channel_io_perm_failure)
5702                 return PCI_ERS_RESULT_DISCONNECT;
5703
5704         if (netif_running(netdev))
5705                 e1000e_down(adapter);
5706         pci_disable_device(pdev);
5707
5708         /* Request a slot slot reset. */
5709         return PCI_ERS_RESULT_NEED_RESET;
5710 }
5711
5712 /**
5713  * e1000_io_slot_reset - called after the pci bus has been reset.
5714  * @pdev: Pointer to PCI device
5715  *
5716  * Restart the card from scratch, as if from a cold-boot. Implementation
5717  * resembles the first-half of the e1000_resume routine.
5718  */
5719 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5720 {
5721         struct net_device *netdev = pci_get_drvdata(pdev);
5722         struct e1000_adapter *adapter = netdev_priv(netdev);
5723         struct e1000_hw *hw = &adapter->hw;
5724         u16 aspm_disable_flag = 0;
5725         int err;
5726         pci_ers_result_t result;
5727
5728         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5729                 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5730         if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5731                 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5732         if (aspm_disable_flag)
5733                 e1000e_disable_aspm(pdev, aspm_disable_flag);
5734
5735         err = pci_enable_device_mem(pdev);
5736         if (err) {
5737                 dev_err(&pdev->dev,
5738                         "Cannot re-enable PCI device after reset.\n");
5739                 result = PCI_ERS_RESULT_DISCONNECT;
5740         } else {
5741                 pci_set_master(pdev);
5742                 pdev->state_saved = true;
5743                 pci_restore_state(pdev);
5744
5745                 pci_enable_wake(pdev, PCI_D3hot, 0);
5746                 pci_enable_wake(pdev, PCI_D3cold, 0);
5747
5748                 e1000e_reset(adapter);
5749                 ew32(WUS, ~0);
5750                 result = PCI_ERS_RESULT_RECOVERED;
5751         }
5752
5753         pci_cleanup_aer_uncorrect_error_status(pdev);
5754
5755         return result;
5756 }
5757
5758 /**
5759  * e1000_io_resume - called when traffic can start flowing again.
5760  * @pdev: Pointer to PCI device
5761  *
5762  * This callback is called when the error recovery driver tells us that
5763  * its OK to resume normal operation. Implementation resembles the
5764  * second-half of the e1000_resume routine.
5765  */
5766 static void e1000_io_resume(struct pci_dev *pdev)
5767 {
5768         struct net_device *netdev = pci_get_drvdata(pdev);
5769         struct e1000_adapter *adapter = netdev_priv(netdev);
5770
5771         e1000_init_manageability_pt(adapter);
5772
5773         if (netif_running(netdev)) {
5774                 if (e1000e_up(adapter)) {
5775                         dev_err(&pdev->dev,
5776                                 "can't bring device back up after reset\n");
5777                         return;
5778                 }
5779         }
5780
5781         netif_device_attach(netdev);
5782
5783         /*
5784          * If the controller has AMT, do not set DRV_LOAD until the interface
5785          * is up.  For all other cases, let the f/w know that the h/w is now
5786          * under the control of the driver.
5787          */
5788         if (!(adapter->flags & FLAG_HAS_AMT))
5789                 e1000e_get_hw_control(adapter);
5790
5791 }
5792
5793 static void e1000_print_device_info(struct e1000_adapter *adapter)
5794 {
5795         struct e1000_hw *hw = &adapter->hw;
5796         struct net_device *netdev = adapter->netdev;
5797         u32 ret_val;
5798         u8 pba_str[E1000_PBANUM_LENGTH];
5799
5800         /* print bus type/speed/width info */
5801         e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5802                /* bus width */
5803                ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5804                 "Width x1"),
5805                /* MAC address */
5806                netdev->dev_addr);
5807         e_info("Intel(R) PRO/%s Network Connection\n",
5808                (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5809         ret_val = e1000_read_pba_string_generic(hw, pba_str,
5810                                                 E1000_PBANUM_LENGTH);
5811         if (ret_val)
5812                 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5813         e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5814