1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2011 Intel Corporation.
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.
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
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.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
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
27 *******************************************************************************/
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
57 #define DRV_EXTRAVERSION "-k"
59 #define DRV_VERSION "1.5.1" DRV_EXTRAVERSION
60 char e1000e_driver_name[] = "e1000e";
61 const char e1000e_driver_version[] = DRV_VERSION;
63 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
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,
79 struct e1000_reg_info {
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 */
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 */
96 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
98 /* General Registers */
100 {E1000_STATUS, "STATUS"},
101 {E1000_CTRL_EXT, "CTRL_EXT"},
103 /* Interrupt Registers */
107 {E1000_RCTL, "RCTL"},
108 {E1000_RDLEN, "RDLEN"},
111 {E1000_RDTR, "RDTR"},
112 {E1000_RXDCTL(0), "RXDCTL"},
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"},
123 {E1000_TCTL, "TCTL"},
124 {E1000_TDBAL, "TDBAL"},
125 {E1000_TDBAH, "TDBAH"},
126 {E1000_TDLEN, "TDLEN"},
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"},
139 /* List Terminator */
144 * e1000_regdump - register printout routine
146 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
152 switch (reginfo->ofs) {
153 case E1000_RXDCTL(0):
154 for (n = 0; n < 2; n++)
155 regs[n] = __er32(hw, E1000_RXDCTL(n));
157 case E1000_TXDCTL(0):
158 for (n = 0; n < 2; n++)
159 regs[n] = __er32(hw, E1000_TXDCTL(n));
162 for (n = 0; n < 2; n++)
163 regs[n] = __er32(hw, E1000_TARC(n));
166 printk(KERN_INFO "%-15s %08x\n",
167 reginfo->name, __er32(hw, reginfo->ofs));
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");
179 * e1000e_dump - Print registers, Tx-ring and Rx-ring
181 static void e1000e_dump(struct e1000_adapter *adapter)
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;
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 union e1000_rx_desc_extended *rx_desc;
205 if (!netif_msg_hw(adapter))
208 /* Print netdevice Info */
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,
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);
226 /* Print Tx Ring Summary */
227 if (!netdev || !netif_running(netdev))
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,
238 buffer_info->next_to_watch,
239 (unsigned long long)buffer_info->time_stamp);
242 if (!netif_msg_tx_done(adapter))
243 goto rx_ring_summary;
245 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
247 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
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
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
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
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,
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");
304 printk(KERN_CONT "\n");
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);
312 /* Print 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);
320 if (!netif_msg_rx_status(adapter))
323 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
324 switch (adapter->rx_ps_pages) {
328 /* [Extended] Packet Split Receive Descriptor Format
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 * +-----------------------------------------------------+
340 printk(KERN_INFO "R [desc] [buffer 0 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
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
355 printk(KERN_INFO "RWB[desc] [ck ipid mrqhsh] "
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;
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),
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,
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);
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");
397 printk(KERN_CONT "\n");
402 /* Extended Receive Descriptor (Read) Format
404 * +-----------------------------------------------------+
405 * 0 | Buffer Address [63:0] |
406 * +-----------------------------------------------------+
408 * +-----------------------------------------------------+
410 printk(KERN_INFO "R [desc] [buf addr 63:0 ] "
411 "[reserved 63:0 ] [bi->dma ] "
412 "[bi->skb] <-- Ext (Read) format\n");
413 /* Extended Receive Descriptor (Write-Back) Format
415 * 63 48 47 32 31 24 23 4 3 0
416 * +------------------------------------------------------+
418 * 0 +-------------------+ Rsvd | Reserved | MRQ RSS |
419 * | Packet | IP | | | Type |
420 * | Checksum | Ident | | | |
421 * +------------------------------------------------------+
422 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
423 * +------------------------------------------------------+
424 * 63 48 47 32 31 20 19 0
426 printk(KERN_INFO "RWB[desc] [cs ipid mrq] "
428 "[bi->skb] <-- Ext (Write-Back) format\n");
430 for (i = 0; i < rx_ring->count; i++) {
431 buffer_info = &rx_ring->buffer_info[i];
432 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
433 u1 = (struct my_u1 *)rx_desc;
434 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
435 if (staterr & E1000_RXD_STAT_DD) {
436 /* Descriptor Done */
437 printk(KERN_INFO "RWB[0x%03X] %016llX "
438 "%016llX ---------------- %p", i,
439 (unsigned long long)le64_to_cpu(u1->a),
440 (unsigned long long)le64_to_cpu(u1->b),
443 printk(KERN_INFO "R [0x%03X] %016llX "
444 "%016llX %016llX %p", i,
445 (unsigned long long)le64_to_cpu(u1->a),
446 (unsigned long long)le64_to_cpu(u1->b),
447 (unsigned long long)buffer_info->dma,
450 if (netif_msg_pktdata(adapter))
451 print_hex_dump(KERN_INFO, "",
452 DUMP_PREFIX_ADDRESS, 16,
456 adapter->rx_buffer_len,
460 if (i == rx_ring->next_to_use)
461 printk(KERN_CONT " NTU\n");
462 else if (i == rx_ring->next_to_clean)
463 printk(KERN_CONT " NTC\n");
465 printk(KERN_CONT "\n");
474 * e1000_desc_unused - calculate if we have unused descriptors
476 static int e1000_desc_unused(struct e1000_ring *ring)
478 if (ring->next_to_clean > ring->next_to_use)
479 return ring->next_to_clean - ring->next_to_use - 1;
481 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
485 * e1000_receive_skb - helper function to handle Rx indications
486 * @adapter: board private structure
487 * @status: descriptor status field as written by hardware
488 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
489 * @skb: pointer to sk_buff to be indicated to stack
491 static void e1000_receive_skb(struct e1000_adapter *adapter,
492 struct net_device *netdev, struct sk_buff *skb,
493 u8 status, __le16 vlan)
495 u16 tag = le16_to_cpu(vlan);
496 skb->protocol = eth_type_trans(skb, netdev);
498 if (status & E1000_RXD_STAT_VP)
499 __vlan_hwaccel_put_tag(skb, tag);
501 napi_gro_receive(&adapter->napi, skb);
505 * e1000_rx_checksum - Receive Checksum Offload
506 * @adapter: board private structure
507 * @status_err: receive descriptor status and error fields
508 * @csum: receive descriptor csum field
509 * @sk_buff: socket buffer with received data
511 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
512 u32 csum, struct sk_buff *skb)
514 u16 status = (u16)status_err;
515 u8 errors = (u8)(status_err >> 24);
517 skb_checksum_none_assert(skb);
519 /* Ignore Checksum bit is set */
520 if (status & E1000_RXD_STAT_IXSM)
522 /* TCP/UDP checksum error bit is set */
523 if (errors & E1000_RXD_ERR_TCPE) {
524 /* let the stack verify checksum errors */
525 adapter->hw_csum_err++;
529 /* TCP/UDP Checksum has not been calculated */
530 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
533 /* It must be a TCP or UDP packet with a valid checksum */
534 if (status & E1000_RXD_STAT_TCPCS) {
535 /* TCP checksum is good */
536 skb->ip_summed = CHECKSUM_UNNECESSARY;
539 * IP fragment with UDP payload
540 * Hardware complements the payload checksum, so we undo it
541 * and then put the value in host order for further stack use.
543 __sum16 sum = (__force __sum16)htons(csum);
544 skb->csum = csum_unfold(~sum);
545 skb->ip_summed = CHECKSUM_COMPLETE;
547 adapter->hw_csum_good++;
551 * e1000e_update_tail_wa - helper function for e1000e_update_[rt]dt_wa()
552 * @hw: pointer to the HW structure
553 * @tail: address of tail descriptor register
554 * @i: value to write to tail descriptor register
556 * When updating the tail register, the ME could be accessing Host CSR
557 * registers at the same time. Normally, this is handled in h/w by an
558 * arbiter but on some parts there is a bug that acknowledges Host accesses
559 * later than it should which could result in the descriptor register to
560 * have an incorrect value. Workaround this by checking the FWSM register
561 * which has bit 24 set while ME is accessing Host CSR registers, wait
562 * if it is set and try again a number of times.
564 static inline s32 e1000e_update_tail_wa(struct e1000_hw *hw, u8 __iomem * tail,
569 while ((j++ < E1000_ICH_FWSM_PCIM2PCI_COUNT) &&
570 (er32(FWSM) & E1000_ICH_FWSM_PCIM2PCI))
575 if ((j == E1000_ICH_FWSM_PCIM2PCI_COUNT) && (i != readl(tail)))
576 return E1000_ERR_SWFW_SYNC;
581 static void e1000e_update_rdt_wa(struct e1000_adapter *adapter, unsigned int i)
583 u8 __iomem *tail = (adapter->hw.hw_addr + adapter->rx_ring->tail);
584 struct e1000_hw *hw = &adapter->hw;
586 if (e1000e_update_tail_wa(hw, tail, i)) {
587 u32 rctl = er32(RCTL);
588 ew32(RCTL, rctl & ~E1000_RCTL_EN);
589 e_err("ME firmware caused invalid RDT - resetting\n");
590 schedule_work(&adapter->reset_task);
594 static void e1000e_update_tdt_wa(struct e1000_adapter *adapter, unsigned int i)
596 u8 __iomem *tail = (adapter->hw.hw_addr + adapter->tx_ring->tail);
597 struct e1000_hw *hw = &adapter->hw;
599 if (e1000e_update_tail_wa(hw, tail, i)) {
600 u32 tctl = er32(TCTL);
601 ew32(TCTL, tctl & ~E1000_TCTL_EN);
602 e_err("ME firmware caused invalid TDT - resetting\n");
603 schedule_work(&adapter->reset_task);
608 * e1000_alloc_rx_buffers - Replace used receive buffers
609 * @adapter: address of board private structure
611 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
612 int cleaned_count, gfp_t gfp)
614 struct net_device *netdev = adapter->netdev;
615 struct pci_dev *pdev = adapter->pdev;
616 struct e1000_ring *rx_ring = adapter->rx_ring;
617 union e1000_rx_desc_extended *rx_desc;
618 struct e1000_buffer *buffer_info;
621 unsigned int bufsz = adapter->rx_buffer_len;
623 i = rx_ring->next_to_use;
624 buffer_info = &rx_ring->buffer_info[i];
626 while (cleaned_count--) {
627 skb = buffer_info->skb;
633 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
635 /* Better luck next round */
636 adapter->alloc_rx_buff_failed++;
640 buffer_info->skb = skb;
642 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
643 adapter->rx_buffer_len,
645 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
646 dev_err(&pdev->dev, "Rx DMA map failed\n");
647 adapter->rx_dma_failed++;
651 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
652 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
654 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
656 * Force memory writes to complete before letting h/w
657 * know there are new descriptors to fetch. (Only
658 * applicable for weak-ordered memory model archs,
662 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
663 e1000e_update_rdt_wa(adapter, i);
665 writel(i, adapter->hw.hw_addr + rx_ring->tail);
668 if (i == rx_ring->count)
670 buffer_info = &rx_ring->buffer_info[i];
673 rx_ring->next_to_use = i;
677 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
678 * @adapter: address of board private structure
680 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
681 int cleaned_count, gfp_t gfp)
683 struct net_device *netdev = adapter->netdev;
684 struct pci_dev *pdev = adapter->pdev;
685 union e1000_rx_desc_packet_split *rx_desc;
686 struct e1000_ring *rx_ring = adapter->rx_ring;
687 struct e1000_buffer *buffer_info;
688 struct e1000_ps_page *ps_page;
692 i = rx_ring->next_to_use;
693 buffer_info = &rx_ring->buffer_info[i];
695 while (cleaned_count--) {
696 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
698 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
699 ps_page = &buffer_info->ps_pages[j];
700 if (j >= adapter->rx_ps_pages) {
701 /* all unused desc entries get hw null ptr */
702 rx_desc->read.buffer_addr[j + 1] =
706 if (!ps_page->page) {
707 ps_page->page = alloc_page(gfp);
708 if (!ps_page->page) {
709 adapter->alloc_rx_buff_failed++;
712 ps_page->dma = dma_map_page(&pdev->dev,
716 if (dma_mapping_error(&pdev->dev,
718 dev_err(&adapter->pdev->dev,
719 "Rx DMA page map failed\n");
720 adapter->rx_dma_failed++;
725 * Refresh the desc even if buffer_addrs
726 * didn't change because each write-back
729 rx_desc->read.buffer_addr[j + 1] =
730 cpu_to_le64(ps_page->dma);
733 skb = __netdev_alloc_skb_ip_align(netdev,
734 adapter->rx_ps_bsize0,
738 adapter->alloc_rx_buff_failed++;
742 buffer_info->skb = skb;
743 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
744 adapter->rx_ps_bsize0,
746 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
747 dev_err(&pdev->dev, "Rx DMA map failed\n");
748 adapter->rx_dma_failed++;
750 dev_kfree_skb_any(skb);
751 buffer_info->skb = NULL;
755 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
757 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
759 * Force memory writes to complete before letting h/w
760 * know there are new descriptors to fetch. (Only
761 * applicable for weak-ordered memory model archs,
765 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
766 e1000e_update_rdt_wa(adapter, i << 1);
769 adapter->hw.hw_addr + rx_ring->tail);
773 if (i == rx_ring->count)
775 buffer_info = &rx_ring->buffer_info[i];
779 rx_ring->next_to_use = i;
783 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
784 * @adapter: address of board private structure
785 * @cleaned_count: number of buffers to allocate this pass
788 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
789 int cleaned_count, gfp_t gfp)
791 struct net_device *netdev = adapter->netdev;
792 struct pci_dev *pdev = adapter->pdev;
793 union e1000_rx_desc_extended *rx_desc;
794 struct e1000_ring *rx_ring = adapter->rx_ring;
795 struct e1000_buffer *buffer_info;
798 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
800 i = rx_ring->next_to_use;
801 buffer_info = &rx_ring->buffer_info[i];
803 while (cleaned_count--) {
804 skb = buffer_info->skb;
810 skb = __netdev_alloc_skb_ip_align(netdev, bufsz, gfp);
811 if (unlikely(!skb)) {
812 /* Better luck next round */
813 adapter->alloc_rx_buff_failed++;
817 buffer_info->skb = skb;
819 /* allocate a new page if necessary */
820 if (!buffer_info->page) {
821 buffer_info->page = alloc_page(gfp);
822 if (unlikely(!buffer_info->page)) {
823 adapter->alloc_rx_buff_failed++;
828 if (!buffer_info->dma)
829 buffer_info->dma = dma_map_page(&pdev->dev,
830 buffer_info->page, 0,
834 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
835 rx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
837 if (unlikely(++i == rx_ring->count))
839 buffer_info = &rx_ring->buffer_info[i];
842 if (likely(rx_ring->next_to_use != i)) {
843 rx_ring->next_to_use = i;
844 if (unlikely(i-- == 0))
845 i = (rx_ring->count - 1);
847 /* Force memory writes to complete before letting h/w
848 * know there are new descriptors to fetch. (Only
849 * applicable for weak-ordered memory model archs,
852 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
853 e1000e_update_rdt_wa(adapter, i);
855 writel(i, adapter->hw.hw_addr + rx_ring->tail);
860 * e1000_clean_rx_irq - Send received data up the network stack; legacy
861 * @adapter: board private structure
863 * the return value indicates whether actual cleaning was done, there
864 * is no guarantee that everything was cleaned
866 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
867 int *work_done, int work_to_do)
869 struct net_device *netdev = adapter->netdev;
870 struct pci_dev *pdev = adapter->pdev;
871 struct e1000_hw *hw = &adapter->hw;
872 struct e1000_ring *rx_ring = adapter->rx_ring;
873 union e1000_rx_desc_extended *rx_desc, *next_rxd;
874 struct e1000_buffer *buffer_info, *next_buffer;
877 int cleaned_count = 0;
879 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
881 i = rx_ring->next_to_clean;
882 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
883 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
884 buffer_info = &rx_ring->buffer_info[i];
886 while (staterr & E1000_RXD_STAT_DD) {
889 if (*work_done >= work_to_do)
892 rmb(); /* read descriptor and rx_buffer_info after status DD */
894 skb = buffer_info->skb;
895 buffer_info->skb = NULL;
897 prefetch(skb->data - NET_IP_ALIGN);
900 if (i == rx_ring->count)
902 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
905 next_buffer = &rx_ring->buffer_info[i];
909 dma_unmap_single(&pdev->dev,
911 adapter->rx_buffer_len,
913 buffer_info->dma = 0;
915 length = le16_to_cpu(rx_desc->wb.upper.length);
918 * !EOP means multiple descriptors were used to store a single
919 * packet, if that's the case we need to toss it. In fact, we
920 * need to toss every packet with the EOP bit clear and the
921 * next frame that _does_ have the EOP bit set, as it is by
922 * definition only a frame fragment
924 if (unlikely(!(staterr & E1000_RXD_STAT_EOP)))
925 adapter->flags2 |= FLAG2_IS_DISCARDING;
927 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
928 /* All receives must fit into a single buffer */
929 e_dbg("Receive packet consumed multiple buffers\n");
931 buffer_info->skb = skb;
932 if (staterr & E1000_RXD_STAT_EOP)
933 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
937 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
939 buffer_info->skb = skb;
943 /* adjust length to remove Ethernet CRC */
944 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
947 total_rx_bytes += length;
951 * code added for copybreak, this should improve
952 * performance for small packets with large amounts
953 * of reassembly being done in the stack
955 if (length < copybreak) {
956 struct sk_buff *new_skb =
957 netdev_alloc_skb_ip_align(netdev, length);
959 skb_copy_to_linear_data_offset(new_skb,
965 /* save the skb in buffer_info as good */
966 buffer_info->skb = skb;
969 /* else just continue with the old one */
971 /* end copybreak code */
972 skb_put(skb, length);
974 /* Receive Checksum Offload */
975 e1000_rx_checksum(adapter, staterr,
976 le16_to_cpu(rx_desc->wb.lower.hi_dword.
979 e1000_receive_skb(adapter, netdev, skb, staterr,
980 rx_desc->wb.upper.vlan);
983 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
985 /* return some buffers to hardware, one at a time is too slow */
986 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
987 adapter->alloc_rx_buf(adapter, cleaned_count,
992 /* use prefetched values */
994 buffer_info = next_buffer;
996 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
998 rx_ring->next_to_clean = i;
1000 cleaned_count = e1000_desc_unused(rx_ring);
1002 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1004 adapter->total_rx_bytes += total_rx_bytes;
1005 adapter->total_rx_packets += total_rx_packets;
1009 static void e1000_put_txbuf(struct e1000_adapter *adapter,
1010 struct e1000_buffer *buffer_info)
1012 if (buffer_info->dma) {
1013 if (buffer_info->mapped_as_page)
1014 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1015 buffer_info->length, DMA_TO_DEVICE);
1017 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1018 buffer_info->length, DMA_TO_DEVICE);
1019 buffer_info->dma = 0;
1021 if (buffer_info->skb) {
1022 dev_kfree_skb_any(buffer_info->skb);
1023 buffer_info->skb = NULL;
1025 buffer_info->time_stamp = 0;
1028 static void e1000_print_hw_hang(struct work_struct *work)
1030 struct e1000_adapter *adapter = container_of(work,
1031 struct e1000_adapter,
1033 struct e1000_ring *tx_ring = adapter->tx_ring;
1034 unsigned int i = tx_ring->next_to_clean;
1035 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
1036 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
1037 struct e1000_hw *hw = &adapter->hw;
1038 u16 phy_status, phy_1000t_status, phy_ext_status;
1041 if (test_bit(__E1000_DOWN, &adapter->state))
1044 e1e_rphy(hw, PHY_STATUS, &phy_status);
1045 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
1046 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
1048 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
1050 /* detected Hardware unit hang */
1051 e_err("Detected Hardware Unit Hang:\n"
1054 " next_to_use <%x>\n"
1055 " next_to_clean <%x>\n"
1056 "buffer_info[next_to_clean]:\n"
1057 " time_stamp <%lx>\n"
1058 " next_to_watch <%x>\n"
1060 " next_to_watch.status <%x>\n"
1063 "PHY 1000BASE-T Status <%x>\n"
1064 "PHY Extended Status <%x>\n"
1065 "PCI Status <%x>\n",
1066 readl(adapter->hw.hw_addr + tx_ring->head),
1067 readl(adapter->hw.hw_addr + tx_ring->tail),
1068 tx_ring->next_to_use,
1069 tx_ring->next_to_clean,
1070 tx_ring->buffer_info[eop].time_stamp,
1073 eop_desc->upper.fields.status,
1082 * e1000_clean_tx_irq - Reclaim resources after transmit completes
1083 * @adapter: board private structure
1085 * the return value indicates whether actual cleaning was done, there
1086 * is no guarantee that everything was cleaned
1088 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
1090 struct net_device *netdev = adapter->netdev;
1091 struct e1000_hw *hw = &adapter->hw;
1092 struct e1000_ring *tx_ring = adapter->tx_ring;
1093 struct e1000_tx_desc *tx_desc, *eop_desc;
1094 struct e1000_buffer *buffer_info;
1095 unsigned int i, eop;
1096 unsigned int count = 0;
1097 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
1099 i = tx_ring->next_to_clean;
1100 eop = tx_ring->buffer_info[i].next_to_watch;
1101 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1103 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1104 (count < tx_ring->count)) {
1105 bool cleaned = false;
1106 rmb(); /* read buffer_info after eop_desc */
1107 for (; !cleaned; count++) {
1108 tx_desc = E1000_TX_DESC(*tx_ring, i);
1109 buffer_info = &tx_ring->buffer_info[i];
1110 cleaned = (i == eop);
1113 total_tx_packets += buffer_info->segs;
1114 total_tx_bytes += buffer_info->bytecount;
1117 e1000_put_txbuf(adapter, buffer_info);
1118 tx_desc->upper.data = 0;
1121 if (i == tx_ring->count)
1125 if (i == tx_ring->next_to_use)
1127 eop = tx_ring->buffer_info[i].next_to_watch;
1128 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1131 tx_ring->next_to_clean = i;
1133 #define TX_WAKE_THRESHOLD 32
1134 if (count && netif_carrier_ok(netdev) &&
1135 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1136 /* Make sure that anybody stopping the queue after this
1137 * sees the new next_to_clean.
1141 if (netif_queue_stopped(netdev) &&
1142 !(test_bit(__E1000_DOWN, &adapter->state))) {
1143 netif_wake_queue(netdev);
1144 ++adapter->restart_queue;
1148 if (adapter->detect_tx_hung) {
1150 * Detect a transmit hang in hardware, this serializes the
1151 * check with the clearing of time_stamp and movement of i
1153 adapter->detect_tx_hung = 0;
1154 if (tx_ring->buffer_info[i].time_stamp &&
1155 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1156 + (adapter->tx_timeout_factor * HZ)) &&
1157 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1158 schedule_work(&adapter->print_hang_task);
1159 netif_stop_queue(netdev);
1162 adapter->total_tx_bytes += total_tx_bytes;
1163 adapter->total_tx_packets += total_tx_packets;
1164 return count < tx_ring->count;
1168 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1169 * @adapter: board private structure
1171 * the return value indicates whether actual cleaning was done, there
1172 * is no guarantee that everything was cleaned
1174 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1175 int *work_done, int work_to_do)
1177 struct e1000_hw *hw = &adapter->hw;
1178 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1179 struct net_device *netdev = adapter->netdev;
1180 struct pci_dev *pdev = adapter->pdev;
1181 struct e1000_ring *rx_ring = adapter->rx_ring;
1182 struct e1000_buffer *buffer_info, *next_buffer;
1183 struct e1000_ps_page *ps_page;
1184 struct sk_buff *skb;
1186 u32 length, staterr;
1187 int cleaned_count = 0;
1189 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1191 i = rx_ring->next_to_clean;
1192 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1193 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1194 buffer_info = &rx_ring->buffer_info[i];
1196 while (staterr & E1000_RXD_STAT_DD) {
1197 if (*work_done >= work_to_do)
1200 skb = buffer_info->skb;
1201 rmb(); /* read descriptor and rx_buffer_info after status DD */
1203 /* in the packet split case this is header only */
1204 prefetch(skb->data - NET_IP_ALIGN);
1207 if (i == rx_ring->count)
1209 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1212 next_buffer = &rx_ring->buffer_info[i];
1216 dma_unmap_single(&pdev->dev, buffer_info->dma,
1217 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1218 buffer_info->dma = 0;
1220 /* see !EOP comment in other Rx routine */
1221 if (!(staterr & E1000_RXD_STAT_EOP))
1222 adapter->flags2 |= FLAG2_IS_DISCARDING;
1224 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1225 e_dbg("Packet Split buffers didn't pick up the full "
1227 dev_kfree_skb_irq(skb);
1228 if (staterr & E1000_RXD_STAT_EOP)
1229 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1233 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1234 dev_kfree_skb_irq(skb);
1238 length = le16_to_cpu(rx_desc->wb.middle.length0);
1241 e_dbg("Last part of the packet spanning multiple "
1243 dev_kfree_skb_irq(skb);
1248 skb_put(skb, length);
1252 * this looks ugly, but it seems compiler issues make it
1253 * more efficient than reusing j
1255 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1258 * page alloc/put takes too long and effects small packet
1259 * throughput, so unsplit small packets and save the alloc/put
1260 * only valid in softirq (napi) context to call kmap_*
1262 if (l1 && (l1 <= copybreak) &&
1263 ((length + l1) <= adapter->rx_ps_bsize0)) {
1266 ps_page = &buffer_info->ps_pages[0];
1269 * there is no documentation about how to call
1270 * kmap_atomic, so we can't hold the mapping
1273 dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1274 PAGE_SIZE, DMA_FROM_DEVICE);
1275 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1276 memcpy(skb_tail_pointer(skb), vaddr, l1);
1277 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1278 dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1279 PAGE_SIZE, DMA_FROM_DEVICE);
1281 /* remove the CRC */
1282 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1290 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1291 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1295 ps_page = &buffer_info->ps_pages[j];
1296 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1299 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1300 ps_page->page = NULL;
1302 skb->data_len += length;
1303 skb->truesize += length;
1306 /* strip the ethernet crc, problem is we're using pages now so
1307 * this whole operation can get a little cpu intensive
1309 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1310 pskb_trim(skb, skb->len - 4);
1313 total_rx_bytes += skb->len;
1316 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1317 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1319 if (rx_desc->wb.upper.header_status &
1320 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1321 adapter->rx_hdr_split++;
1323 e1000_receive_skb(adapter, netdev, skb,
1324 staterr, rx_desc->wb.middle.vlan);
1327 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1328 buffer_info->skb = NULL;
1330 /* return some buffers to hardware, one at a time is too slow */
1331 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1332 adapter->alloc_rx_buf(adapter, cleaned_count,
1337 /* use prefetched values */
1339 buffer_info = next_buffer;
1341 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1343 rx_ring->next_to_clean = i;
1345 cleaned_count = e1000_desc_unused(rx_ring);
1347 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1349 adapter->total_rx_bytes += total_rx_bytes;
1350 adapter->total_rx_packets += total_rx_packets;
1355 * e1000_consume_page - helper function
1357 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1362 skb->data_len += length;
1363 skb->truesize += length;
1367 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1368 * @adapter: board private structure
1370 * the return value indicates whether actual cleaning was done, there
1371 * is no guarantee that everything was cleaned
1374 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1375 int *work_done, int work_to_do)
1377 struct net_device *netdev = adapter->netdev;
1378 struct pci_dev *pdev = adapter->pdev;
1379 struct e1000_ring *rx_ring = adapter->rx_ring;
1380 union e1000_rx_desc_extended *rx_desc, *next_rxd;
1381 struct e1000_buffer *buffer_info, *next_buffer;
1382 u32 length, staterr;
1384 int cleaned_count = 0;
1385 bool cleaned = false;
1386 unsigned int total_rx_bytes=0, total_rx_packets=0;
1388 i = rx_ring->next_to_clean;
1389 rx_desc = E1000_RX_DESC_EXT(*rx_ring, i);
1390 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1391 buffer_info = &rx_ring->buffer_info[i];
1393 while (staterr & E1000_RXD_STAT_DD) {
1394 struct sk_buff *skb;
1396 if (*work_done >= work_to_do)
1399 rmb(); /* read descriptor and rx_buffer_info after status DD */
1401 skb = buffer_info->skb;
1402 buffer_info->skb = NULL;
1405 if (i == rx_ring->count)
1407 next_rxd = E1000_RX_DESC_EXT(*rx_ring, i);
1410 next_buffer = &rx_ring->buffer_info[i];
1414 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1416 buffer_info->dma = 0;
1418 length = le16_to_cpu(rx_desc->wb.upper.length);
1420 /* errors is only valid for DD + EOP descriptors */
1421 if (unlikely((staterr & E1000_RXD_STAT_EOP) &&
1422 (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK))) {
1423 /* recycle both page and skb */
1424 buffer_info->skb = skb;
1425 /* an error means any chain goes out the window too */
1426 if (rx_ring->rx_skb_top)
1427 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1428 rx_ring->rx_skb_top = NULL;
1432 #define rxtop (rx_ring->rx_skb_top)
1433 if (!(staterr & E1000_RXD_STAT_EOP)) {
1434 /* this descriptor is only the beginning (or middle) */
1436 /* this is the beginning of a chain */
1438 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1441 /* this is the middle of a chain */
1442 skb_fill_page_desc(rxtop,
1443 skb_shinfo(rxtop)->nr_frags,
1444 buffer_info->page, 0, length);
1445 /* re-use the skb, only consumed the page */
1446 buffer_info->skb = skb;
1448 e1000_consume_page(buffer_info, rxtop, length);
1452 /* end of the chain */
1453 skb_fill_page_desc(rxtop,
1454 skb_shinfo(rxtop)->nr_frags,
1455 buffer_info->page, 0, length);
1456 /* re-use the current skb, we only consumed the
1458 buffer_info->skb = skb;
1461 e1000_consume_page(buffer_info, skb, length);
1463 /* no chain, got EOP, this buf is the packet
1464 * copybreak to save the put_page/alloc_page */
1465 if (length <= copybreak &&
1466 skb_tailroom(skb) >= length) {
1468 vaddr = kmap_atomic(buffer_info->page,
1469 KM_SKB_DATA_SOFTIRQ);
1470 memcpy(skb_tail_pointer(skb), vaddr,
1472 kunmap_atomic(vaddr,
1473 KM_SKB_DATA_SOFTIRQ);
1474 /* re-use the page, so don't erase
1475 * buffer_info->page */
1476 skb_put(skb, length);
1478 skb_fill_page_desc(skb, 0,
1479 buffer_info->page, 0,
1481 e1000_consume_page(buffer_info, skb,
1487 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1488 e1000_rx_checksum(adapter, staterr,
1489 le16_to_cpu(rx_desc->wb.lower.hi_dword.
1490 csum_ip.csum), skb);
1492 /* probably a little skewed due to removing CRC */
1493 total_rx_bytes += skb->len;
1496 /* eth type trans needs skb->data to point to something */
1497 if (!pskb_may_pull(skb, ETH_HLEN)) {
1498 e_err("pskb_may_pull failed.\n");
1499 dev_kfree_skb_irq(skb);
1503 e1000_receive_skb(adapter, netdev, skb, staterr,
1504 rx_desc->wb.upper.vlan);
1507 rx_desc->wb.upper.status_error &= cpu_to_le32(~0xFF);
1509 /* return some buffers to hardware, one at a time is too slow */
1510 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1511 adapter->alloc_rx_buf(adapter, cleaned_count,
1516 /* use prefetched values */
1518 buffer_info = next_buffer;
1520 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
1522 rx_ring->next_to_clean = i;
1524 cleaned_count = e1000_desc_unused(rx_ring);
1526 adapter->alloc_rx_buf(adapter, cleaned_count, GFP_ATOMIC);
1528 adapter->total_rx_bytes += total_rx_bytes;
1529 adapter->total_rx_packets += total_rx_packets;
1534 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1535 * @adapter: board private structure
1537 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1539 struct e1000_ring *rx_ring = adapter->rx_ring;
1540 struct e1000_buffer *buffer_info;
1541 struct e1000_ps_page *ps_page;
1542 struct pci_dev *pdev = adapter->pdev;
1545 /* Free all the Rx ring sk_buffs */
1546 for (i = 0; i < rx_ring->count; i++) {
1547 buffer_info = &rx_ring->buffer_info[i];
1548 if (buffer_info->dma) {
1549 if (adapter->clean_rx == e1000_clean_rx_irq)
1550 dma_unmap_single(&pdev->dev, buffer_info->dma,
1551 adapter->rx_buffer_len,
1553 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1554 dma_unmap_page(&pdev->dev, buffer_info->dma,
1557 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1558 dma_unmap_single(&pdev->dev, buffer_info->dma,
1559 adapter->rx_ps_bsize0,
1561 buffer_info->dma = 0;
1564 if (buffer_info->page) {
1565 put_page(buffer_info->page);
1566 buffer_info->page = NULL;
1569 if (buffer_info->skb) {
1570 dev_kfree_skb(buffer_info->skb);
1571 buffer_info->skb = NULL;
1574 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1575 ps_page = &buffer_info->ps_pages[j];
1578 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1581 put_page(ps_page->page);
1582 ps_page->page = NULL;
1586 /* there also may be some cached data from a chained receive */
1587 if (rx_ring->rx_skb_top) {
1588 dev_kfree_skb(rx_ring->rx_skb_top);
1589 rx_ring->rx_skb_top = NULL;
1592 /* Zero out the descriptor ring */
1593 memset(rx_ring->desc, 0, rx_ring->size);
1595 rx_ring->next_to_clean = 0;
1596 rx_ring->next_to_use = 0;
1597 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1599 writel(0, adapter->hw.hw_addr + rx_ring->head);
1600 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1603 static void e1000e_downshift_workaround(struct work_struct *work)
1605 struct e1000_adapter *adapter = container_of(work,
1606 struct e1000_adapter, downshift_task);
1608 if (test_bit(__E1000_DOWN, &adapter->state))
1611 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1615 * e1000_intr_msi - Interrupt Handler
1616 * @irq: interrupt number
1617 * @data: pointer to a network interface device structure
1619 static irqreturn_t e1000_intr_msi(int irq, void *data)
1621 struct net_device *netdev = data;
1622 struct e1000_adapter *adapter = netdev_priv(netdev);
1623 struct e1000_hw *hw = &adapter->hw;
1624 u32 icr = er32(ICR);
1627 * read ICR disables interrupts using IAM
1630 if (icr & E1000_ICR_LSC) {
1631 hw->mac.get_link_status = 1;
1633 * ICH8 workaround-- Call gig speed drop workaround on cable
1634 * disconnect (LSC) before accessing any PHY registers
1636 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1637 (!(er32(STATUS) & E1000_STATUS_LU)))
1638 schedule_work(&adapter->downshift_task);
1641 * 80003ES2LAN workaround-- For packet buffer work-around on
1642 * link down event; disable receives here in the ISR and reset
1643 * adapter in watchdog
1645 if (netif_carrier_ok(netdev) &&
1646 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1647 /* disable receives */
1648 u32 rctl = er32(RCTL);
1649 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1650 adapter->flags |= FLAG_RX_RESTART_NOW;
1652 /* guard against interrupt when we're going down */
1653 if (!test_bit(__E1000_DOWN, &adapter->state))
1654 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1657 if (napi_schedule_prep(&adapter->napi)) {
1658 adapter->total_tx_bytes = 0;
1659 adapter->total_tx_packets = 0;
1660 adapter->total_rx_bytes = 0;
1661 adapter->total_rx_packets = 0;
1662 __napi_schedule(&adapter->napi);
1669 * e1000_intr - Interrupt Handler
1670 * @irq: interrupt number
1671 * @data: pointer to a network interface device structure
1673 static irqreturn_t e1000_intr(int irq, void *data)
1675 struct net_device *netdev = data;
1676 struct e1000_adapter *adapter = netdev_priv(netdev);
1677 struct e1000_hw *hw = &adapter->hw;
1678 u32 rctl, icr = er32(ICR);
1680 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1681 return IRQ_NONE; /* Not our interrupt */
1684 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1685 * not set, then the adapter didn't send an interrupt
1687 if (!(icr & E1000_ICR_INT_ASSERTED))
1691 * Interrupt Auto-Mask...upon reading ICR,
1692 * interrupts are masked. No need for the
1696 if (icr & E1000_ICR_LSC) {
1697 hw->mac.get_link_status = 1;
1699 * ICH8 workaround-- Call gig speed drop workaround on cable
1700 * disconnect (LSC) before accessing any PHY registers
1702 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1703 (!(er32(STATUS) & E1000_STATUS_LU)))
1704 schedule_work(&adapter->downshift_task);
1707 * 80003ES2LAN workaround--
1708 * For packet buffer work-around on link down event;
1709 * disable receives here in the ISR and
1710 * reset adapter in watchdog
1712 if (netif_carrier_ok(netdev) &&
1713 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1714 /* disable receives */
1716 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1717 adapter->flags |= FLAG_RX_RESTART_NOW;
1719 /* guard against interrupt when we're going down */
1720 if (!test_bit(__E1000_DOWN, &adapter->state))
1721 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1724 if (napi_schedule_prep(&adapter->napi)) {
1725 adapter->total_tx_bytes = 0;
1726 adapter->total_tx_packets = 0;
1727 adapter->total_rx_bytes = 0;
1728 adapter->total_rx_packets = 0;
1729 __napi_schedule(&adapter->napi);
1735 static irqreturn_t e1000_msix_other(int irq, void *data)
1737 struct net_device *netdev = data;
1738 struct e1000_adapter *adapter = netdev_priv(netdev);
1739 struct e1000_hw *hw = &adapter->hw;
1740 u32 icr = er32(ICR);
1742 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1743 if (!test_bit(__E1000_DOWN, &adapter->state))
1744 ew32(IMS, E1000_IMS_OTHER);
1748 if (icr & adapter->eiac_mask)
1749 ew32(ICS, (icr & adapter->eiac_mask));
1751 if (icr & E1000_ICR_OTHER) {
1752 if (!(icr & E1000_ICR_LSC))
1753 goto no_link_interrupt;
1754 hw->mac.get_link_status = 1;
1755 /* guard against interrupt when we're going down */
1756 if (!test_bit(__E1000_DOWN, &adapter->state))
1757 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1761 if (!test_bit(__E1000_DOWN, &adapter->state))
1762 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1768 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1770 struct net_device *netdev = data;
1771 struct e1000_adapter *adapter = netdev_priv(netdev);
1772 struct e1000_hw *hw = &adapter->hw;
1773 struct e1000_ring *tx_ring = adapter->tx_ring;
1776 adapter->total_tx_bytes = 0;
1777 adapter->total_tx_packets = 0;
1779 if (!e1000_clean_tx_irq(adapter))
1780 /* Ring was not completely cleaned, so fire another interrupt */
1781 ew32(ICS, tx_ring->ims_val);
1786 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1788 struct net_device *netdev = data;
1789 struct e1000_adapter *adapter = netdev_priv(netdev);
1791 /* Write the ITR value calculated at the end of the
1792 * previous interrupt.
1794 if (adapter->rx_ring->set_itr) {
1795 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1796 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1797 adapter->rx_ring->set_itr = 0;
1800 if (napi_schedule_prep(&adapter->napi)) {
1801 adapter->total_rx_bytes = 0;
1802 adapter->total_rx_packets = 0;
1803 __napi_schedule(&adapter->napi);
1809 * e1000_configure_msix - Configure MSI-X hardware
1811 * e1000_configure_msix sets up the hardware to properly
1812 * generate MSI-X interrupts.
1814 static void e1000_configure_msix(struct e1000_adapter *adapter)
1816 struct e1000_hw *hw = &adapter->hw;
1817 struct e1000_ring *rx_ring = adapter->rx_ring;
1818 struct e1000_ring *tx_ring = adapter->tx_ring;
1820 u32 ctrl_ext, ivar = 0;
1822 adapter->eiac_mask = 0;
1824 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1825 if (hw->mac.type == e1000_82574) {
1826 u32 rfctl = er32(RFCTL);
1827 rfctl |= E1000_RFCTL_ACK_DIS;
1831 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1832 /* Configure Rx vector */
1833 rx_ring->ims_val = E1000_IMS_RXQ0;
1834 adapter->eiac_mask |= rx_ring->ims_val;
1835 if (rx_ring->itr_val)
1836 writel(1000000000 / (rx_ring->itr_val * 256),
1837 hw->hw_addr + rx_ring->itr_register);
1839 writel(1, hw->hw_addr + rx_ring->itr_register);
1840 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1842 /* Configure Tx vector */
1843 tx_ring->ims_val = E1000_IMS_TXQ0;
1845 if (tx_ring->itr_val)
1846 writel(1000000000 / (tx_ring->itr_val * 256),
1847 hw->hw_addr + tx_ring->itr_register);
1849 writel(1, hw->hw_addr + tx_ring->itr_register);
1850 adapter->eiac_mask |= tx_ring->ims_val;
1851 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1853 /* set vector for Other Causes, e.g. link changes */
1855 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1856 if (rx_ring->itr_val)
1857 writel(1000000000 / (rx_ring->itr_val * 256),
1858 hw->hw_addr + E1000_EITR_82574(vector));
1860 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1862 /* Cause Tx interrupts on every write back */
1867 /* enable MSI-X PBA support */
1868 ctrl_ext = er32(CTRL_EXT);
1869 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1871 /* Auto-Mask Other interrupts upon ICR read */
1872 #define E1000_EIAC_MASK_82574 0x01F00000
1873 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1874 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1875 ew32(CTRL_EXT, ctrl_ext);
1879 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1881 if (adapter->msix_entries) {
1882 pci_disable_msix(adapter->pdev);
1883 kfree(adapter->msix_entries);
1884 adapter->msix_entries = NULL;
1885 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1886 pci_disable_msi(adapter->pdev);
1887 adapter->flags &= ~FLAG_MSI_ENABLED;
1892 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1894 * Attempt to configure interrupts using the best available
1895 * capabilities of the hardware and kernel.
1897 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1902 switch (adapter->int_mode) {
1903 case E1000E_INT_MODE_MSIX:
1904 if (adapter->flags & FLAG_HAS_MSIX) {
1905 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1906 adapter->msix_entries = kcalloc(adapter->num_vectors,
1907 sizeof(struct msix_entry),
1909 if (adapter->msix_entries) {
1910 for (i = 0; i < adapter->num_vectors; i++)
1911 adapter->msix_entries[i].entry = i;
1913 err = pci_enable_msix(adapter->pdev,
1914 adapter->msix_entries,
1915 adapter->num_vectors);
1919 /* MSI-X failed, so fall through and try MSI */
1920 e_err("Failed to initialize MSI-X interrupts. "
1921 "Falling back to MSI interrupts.\n");
1922 e1000e_reset_interrupt_capability(adapter);
1924 adapter->int_mode = E1000E_INT_MODE_MSI;
1926 case E1000E_INT_MODE_MSI:
1927 if (!pci_enable_msi(adapter->pdev)) {
1928 adapter->flags |= FLAG_MSI_ENABLED;
1930 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1931 e_err("Failed to initialize MSI interrupts. Falling "
1932 "back to legacy interrupts.\n");
1935 case E1000E_INT_MODE_LEGACY:
1936 /* Don't do anything; this is the system default */
1940 /* store the number of vectors being used */
1941 adapter->num_vectors = 1;
1945 * e1000_request_msix - Initialize MSI-X interrupts
1947 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1950 static int e1000_request_msix(struct e1000_adapter *adapter)
1952 struct net_device *netdev = adapter->netdev;
1953 int err = 0, vector = 0;
1955 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1956 snprintf(adapter->rx_ring->name,
1957 sizeof(adapter->rx_ring->name) - 1,
1958 "%s-rx-0", netdev->name);
1960 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1961 err = request_irq(adapter->msix_entries[vector].vector,
1962 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1966 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1967 adapter->rx_ring->itr_val = adapter->itr;
1970 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1971 snprintf(adapter->tx_ring->name,
1972 sizeof(adapter->tx_ring->name) - 1,
1973 "%s-tx-0", netdev->name);
1975 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1976 err = request_irq(adapter->msix_entries[vector].vector,
1977 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1981 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1982 adapter->tx_ring->itr_val = adapter->itr;
1985 err = request_irq(adapter->msix_entries[vector].vector,
1986 e1000_msix_other, 0, netdev->name, netdev);
1990 e1000_configure_msix(adapter);
1997 * e1000_request_irq - initialize interrupts
1999 * Attempts to configure interrupts using the best available
2000 * capabilities of the hardware and kernel.
2002 static int e1000_request_irq(struct e1000_adapter *adapter)
2004 struct net_device *netdev = adapter->netdev;
2007 if (adapter->msix_entries) {
2008 err = e1000_request_msix(adapter);
2011 /* fall back to MSI */
2012 e1000e_reset_interrupt_capability(adapter);
2013 adapter->int_mode = E1000E_INT_MODE_MSI;
2014 e1000e_set_interrupt_capability(adapter);
2016 if (adapter->flags & FLAG_MSI_ENABLED) {
2017 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
2018 netdev->name, netdev);
2022 /* fall back to legacy interrupt */
2023 e1000e_reset_interrupt_capability(adapter);
2024 adapter->int_mode = E1000E_INT_MODE_LEGACY;
2027 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
2028 netdev->name, netdev);
2030 e_err("Unable to allocate interrupt, Error: %d\n", err);
2035 static void e1000_free_irq(struct e1000_adapter *adapter)
2037 struct net_device *netdev = adapter->netdev;
2039 if (adapter->msix_entries) {
2042 free_irq(adapter->msix_entries[vector].vector, netdev);
2045 free_irq(adapter->msix_entries[vector].vector, netdev);
2048 /* Other Causes interrupt vector */
2049 free_irq(adapter->msix_entries[vector].vector, netdev);
2053 free_irq(adapter->pdev->irq, netdev);
2057 * e1000_irq_disable - Mask off interrupt generation on the NIC
2059 static void e1000_irq_disable(struct e1000_adapter *adapter)
2061 struct e1000_hw *hw = &adapter->hw;
2064 if (adapter->msix_entries)
2065 ew32(EIAC_82574, 0);
2068 if (adapter->msix_entries) {
2070 for (i = 0; i < adapter->num_vectors; i++)
2071 synchronize_irq(adapter->msix_entries[i].vector);
2073 synchronize_irq(adapter->pdev->irq);
2078 * e1000_irq_enable - Enable default interrupt generation settings
2080 static void e1000_irq_enable(struct e1000_adapter *adapter)
2082 struct e1000_hw *hw = &adapter->hw;
2084 if (adapter->msix_entries) {
2085 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
2086 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
2088 ew32(IMS, IMS_ENABLE_MASK);
2094 * e1000e_get_hw_control - get control of the h/w from f/w
2095 * @adapter: address of board private structure
2097 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2098 * For ASF and Pass Through versions of f/w this means that
2099 * the driver is loaded. For AMT version (only with 82573)
2100 * of the f/w this means that the network i/f is open.
2102 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2104 struct e1000_hw *hw = &adapter->hw;
2108 /* Let firmware know the driver has taken over */
2109 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2111 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2112 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2113 ctrl_ext = er32(CTRL_EXT);
2114 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2119 * e1000e_release_hw_control - release control of the h/w to f/w
2120 * @adapter: address of board private structure
2122 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2123 * For ASF and Pass Through versions of f/w this means that the
2124 * driver is no longer loaded. For AMT version (only with 82573) i
2125 * of the f/w this means that the network i/f is closed.
2128 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2130 struct e1000_hw *hw = &adapter->hw;
2134 /* Let firmware taken over control of h/w */
2135 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2137 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2138 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2139 ctrl_ext = er32(CTRL_EXT);
2140 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2145 * @e1000_alloc_ring - allocate memory for a ring structure
2147 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2148 struct e1000_ring *ring)
2150 struct pci_dev *pdev = adapter->pdev;
2152 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2161 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2162 * @adapter: board private structure
2164 * Return 0 on success, negative on failure
2166 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2168 struct e1000_ring *tx_ring = adapter->tx_ring;
2169 int err = -ENOMEM, size;
2171 size = sizeof(struct e1000_buffer) * tx_ring->count;
2172 tx_ring->buffer_info = vzalloc(size);
2173 if (!tx_ring->buffer_info)
2176 /* round up to nearest 4K */
2177 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2178 tx_ring->size = ALIGN(tx_ring->size, 4096);
2180 err = e1000_alloc_ring_dma(adapter, tx_ring);
2184 tx_ring->next_to_use = 0;
2185 tx_ring->next_to_clean = 0;
2189 vfree(tx_ring->buffer_info);
2190 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2195 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2196 * @adapter: board private structure
2198 * Returns 0 on success, negative on failure
2200 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2202 struct e1000_ring *rx_ring = adapter->rx_ring;
2203 struct e1000_buffer *buffer_info;
2204 int i, size, desc_len, err = -ENOMEM;
2206 size = sizeof(struct e1000_buffer) * rx_ring->count;
2207 rx_ring->buffer_info = vzalloc(size);
2208 if (!rx_ring->buffer_info)
2211 for (i = 0; i < rx_ring->count; i++) {
2212 buffer_info = &rx_ring->buffer_info[i];
2213 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2214 sizeof(struct e1000_ps_page),
2216 if (!buffer_info->ps_pages)
2220 desc_len = sizeof(union e1000_rx_desc_packet_split);
2222 /* Round up to nearest 4K */
2223 rx_ring->size = rx_ring->count * desc_len;
2224 rx_ring->size = ALIGN(rx_ring->size, 4096);
2226 err = e1000_alloc_ring_dma(adapter, rx_ring);
2230 rx_ring->next_to_clean = 0;
2231 rx_ring->next_to_use = 0;
2232 rx_ring->rx_skb_top = NULL;
2237 for (i = 0; i < rx_ring->count; i++) {
2238 buffer_info = &rx_ring->buffer_info[i];
2239 kfree(buffer_info->ps_pages);
2242 vfree(rx_ring->buffer_info);
2243 e_err("Unable to allocate memory for the receive descriptor ring\n");
2248 * e1000_clean_tx_ring - Free Tx Buffers
2249 * @adapter: board private structure
2251 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2253 struct e1000_ring *tx_ring = adapter->tx_ring;
2254 struct e1000_buffer *buffer_info;
2258 for (i = 0; i < tx_ring->count; i++) {
2259 buffer_info = &tx_ring->buffer_info[i];
2260 e1000_put_txbuf(adapter, buffer_info);
2263 size = sizeof(struct e1000_buffer) * tx_ring->count;
2264 memset(tx_ring->buffer_info, 0, size);
2266 memset(tx_ring->desc, 0, tx_ring->size);
2268 tx_ring->next_to_use = 0;
2269 tx_ring->next_to_clean = 0;
2271 writel(0, adapter->hw.hw_addr + tx_ring->head);
2272 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2276 * e1000e_free_tx_resources - Free Tx Resources per Queue
2277 * @adapter: board private structure
2279 * Free all transmit software resources
2281 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2283 struct pci_dev *pdev = adapter->pdev;
2284 struct e1000_ring *tx_ring = adapter->tx_ring;
2286 e1000_clean_tx_ring(adapter);
2288 vfree(tx_ring->buffer_info);
2289 tx_ring->buffer_info = NULL;
2291 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2293 tx_ring->desc = NULL;
2297 * e1000e_free_rx_resources - Free Rx Resources
2298 * @adapter: board private structure
2300 * Free all receive software resources
2303 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2305 struct pci_dev *pdev = adapter->pdev;
2306 struct e1000_ring *rx_ring = adapter->rx_ring;
2309 e1000_clean_rx_ring(adapter);
2311 for (i = 0; i < rx_ring->count; i++)
2312 kfree(rx_ring->buffer_info[i].ps_pages);
2314 vfree(rx_ring->buffer_info);
2315 rx_ring->buffer_info = NULL;
2317 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2319 rx_ring->desc = NULL;
2323 * e1000_update_itr - update the dynamic ITR value based on statistics
2324 * @adapter: pointer to adapter
2325 * @itr_setting: current adapter->itr
2326 * @packets: the number of packets during this measurement interval
2327 * @bytes: the number of bytes during this measurement interval
2329 * Stores a new ITR value based on packets and byte
2330 * counts during the last interrupt. The advantage of per interrupt
2331 * computation is faster updates and more accurate ITR for the current
2332 * traffic pattern. Constants in this function were computed
2333 * based on theoretical maximum wire speed and thresholds were set based
2334 * on testing data as well as attempting to minimize response time
2335 * while increasing bulk throughput. This functionality is controlled
2336 * by the InterruptThrottleRate module parameter.
2338 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2339 u16 itr_setting, int packets,
2342 unsigned int retval = itr_setting;
2345 goto update_itr_done;
2347 switch (itr_setting) {
2348 case lowest_latency:
2349 /* handle TSO and jumbo frames */
2350 if (bytes/packets > 8000)
2351 retval = bulk_latency;
2352 else if ((packets < 5) && (bytes > 512))
2353 retval = low_latency;
2355 case low_latency: /* 50 usec aka 20000 ints/s */
2356 if (bytes > 10000) {
2357 /* this if handles the TSO accounting */
2358 if (bytes/packets > 8000)
2359 retval = bulk_latency;
2360 else if ((packets < 10) || ((bytes/packets) > 1200))
2361 retval = bulk_latency;
2362 else if ((packets > 35))
2363 retval = lowest_latency;
2364 } else if (bytes/packets > 2000) {
2365 retval = bulk_latency;
2366 } else if (packets <= 2 && bytes < 512) {
2367 retval = lowest_latency;
2370 case bulk_latency: /* 250 usec aka 4000 ints/s */
2371 if (bytes > 25000) {
2373 retval = low_latency;
2374 } else if (bytes < 6000) {
2375 retval = low_latency;
2384 static void e1000_set_itr(struct e1000_adapter *adapter)
2386 struct e1000_hw *hw = &adapter->hw;
2388 u32 new_itr = adapter->itr;
2390 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2391 if (adapter->link_speed != SPEED_1000) {
2397 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2402 adapter->tx_itr = e1000_update_itr(adapter,
2404 adapter->total_tx_packets,
2405 adapter->total_tx_bytes);
2406 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2407 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2408 adapter->tx_itr = low_latency;
2410 adapter->rx_itr = e1000_update_itr(adapter,
2412 adapter->total_rx_packets,
2413 adapter->total_rx_bytes);
2414 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2415 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2416 adapter->rx_itr = low_latency;
2418 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2420 switch (current_itr) {
2421 /* counts and packets in update_itr are dependent on these numbers */
2422 case lowest_latency:
2426 new_itr = 20000; /* aka hwitr = ~200 */
2436 if (new_itr != adapter->itr) {
2438 * this attempts to bias the interrupt rate towards Bulk
2439 * by adding intermediate steps when interrupt rate is
2442 new_itr = new_itr > adapter->itr ?
2443 min(adapter->itr + (new_itr >> 2), new_itr) :
2445 adapter->itr = new_itr;
2446 adapter->rx_ring->itr_val = new_itr;
2447 if (adapter->msix_entries)
2448 adapter->rx_ring->set_itr = 1;
2451 ew32(ITR, 1000000000 / (new_itr * 256));
2458 * e1000_alloc_queues - Allocate memory for all rings
2459 * @adapter: board private structure to initialize
2461 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2463 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2464 if (!adapter->tx_ring)
2467 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2468 if (!adapter->rx_ring)
2473 e_err("Unable to allocate memory for queues\n");
2474 kfree(adapter->rx_ring);
2475 kfree(adapter->tx_ring);
2480 * e1000_clean - NAPI Rx polling callback
2481 * @napi: struct associated with this polling callback
2482 * @budget: amount of packets driver is allowed to process this poll
2484 static int e1000_clean(struct napi_struct *napi, int budget)
2486 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2487 struct e1000_hw *hw = &adapter->hw;
2488 struct net_device *poll_dev = adapter->netdev;
2489 int tx_cleaned = 1, work_done = 0;
2491 adapter = netdev_priv(poll_dev);
2493 if (adapter->msix_entries &&
2494 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2497 tx_cleaned = e1000_clean_tx_irq(adapter);
2500 adapter->clean_rx(adapter, &work_done, budget);
2505 /* If budget not fully consumed, exit the polling mode */
2506 if (work_done < budget) {
2507 if (adapter->itr_setting & 3)
2508 e1000_set_itr(adapter);
2509 napi_complete(napi);
2510 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2511 if (adapter->msix_entries)
2512 ew32(IMS, adapter->rx_ring->ims_val);
2514 e1000_irq_enable(adapter);
2521 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2523 struct e1000_adapter *adapter = netdev_priv(netdev);
2524 struct e1000_hw *hw = &adapter->hw;
2527 /* don't update vlan cookie if already programmed */
2528 if ((adapter->hw.mng_cookie.status &
2529 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2530 (vid == adapter->mng_vlan_id))
2533 /* add VID to filter table */
2534 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2535 index = (vid >> 5) & 0x7F;
2536 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2537 vfta |= (1 << (vid & 0x1F));
2538 hw->mac.ops.write_vfta(hw, index, vfta);
2541 set_bit(vid, adapter->active_vlans);
2544 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2546 struct e1000_adapter *adapter = netdev_priv(netdev);
2547 struct e1000_hw *hw = &adapter->hw;
2550 if ((adapter->hw.mng_cookie.status &
2551 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2552 (vid == adapter->mng_vlan_id)) {
2553 /* release control to f/w */
2554 e1000e_release_hw_control(adapter);
2558 /* remove VID from filter table */
2559 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2560 index = (vid >> 5) & 0x7F;
2561 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2562 vfta &= ~(1 << (vid & 0x1F));
2563 hw->mac.ops.write_vfta(hw, index, vfta);
2566 clear_bit(vid, adapter->active_vlans);
2570 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2571 * @adapter: board private structure to initialize
2573 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2575 struct net_device *netdev = adapter->netdev;
2576 struct e1000_hw *hw = &adapter->hw;
2579 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2580 /* disable VLAN receive filtering */
2582 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2585 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2586 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2587 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2593 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2594 * @adapter: board private structure to initialize
2596 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2598 struct e1000_hw *hw = &adapter->hw;
2601 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2602 /* enable VLAN receive filtering */
2604 rctl |= E1000_RCTL_VFE;
2605 rctl &= ~E1000_RCTL_CFIEN;
2611 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2612 * @adapter: board private structure to initialize
2614 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2616 struct e1000_hw *hw = &adapter->hw;
2619 /* disable VLAN tag insert/strip */
2621 ctrl &= ~E1000_CTRL_VME;
2626 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2627 * @adapter: board private structure to initialize
2629 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2631 struct e1000_hw *hw = &adapter->hw;
2634 /* enable VLAN tag insert/strip */
2636 ctrl |= E1000_CTRL_VME;
2640 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2642 struct net_device *netdev = adapter->netdev;
2643 u16 vid = adapter->hw.mng_cookie.vlan_id;
2644 u16 old_vid = adapter->mng_vlan_id;
2646 if (adapter->hw.mng_cookie.status &
2647 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2648 e1000_vlan_rx_add_vid(netdev, vid);
2649 adapter->mng_vlan_id = vid;
2652 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2653 e1000_vlan_rx_kill_vid(netdev, old_vid);
2656 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2660 e1000_vlan_rx_add_vid(adapter->netdev, 0);
2662 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2663 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2666 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2668 struct e1000_hw *hw = &adapter->hw;
2669 u32 manc, manc2h, mdef, i, j;
2671 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2677 * enable receiving management packets to the host. this will probably
2678 * generate destination unreachable messages from the host OS, but
2679 * the packets will be handled on SMBUS
2681 manc |= E1000_MANC_EN_MNG2HOST;
2682 manc2h = er32(MANC2H);
2684 switch (hw->mac.type) {
2686 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2691 * Check if IPMI pass-through decision filter already exists;
2694 for (i = 0, j = 0; i < 8; i++) {
2695 mdef = er32(MDEF(i));
2697 /* Ignore filters with anything other than IPMI ports */
2698 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2701 /* Enable this decision filter in MANC2H */
2708 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2711 /* Create new decision filter in an empty filter */
2712 for (i = 0, j = 0; i < 8; i++)
2713 if (er32(MDEF(i)) == 0) {
2714 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2715 E1000_MDEF_PORT_664));
2722 e_warn("Unable to create IPMI pass-through filter\n");
2726 ew32(MANC2H, manc2h);
2731 * e1000_configure_tx - Configure Transmit Unit after Reset
2732 * @adapter: board private structure
2734 * Configure the Tx unit of the MAC after a reset.
2736 static void e1000_configure_tx(struct e1000_adapter *adapter)
2738 struct e1000_hw *hw = &adapter->hw;
2739 struct e1000_ring *tx_ring = adapter->tx_ring;
2741 u32 tdlen, tctl, tipg, tarc;
2744 /* Setup the HW Tx Head and Tail descriptor pointers */
2745 tdba = tx_ring->dma;
2746 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2747 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2748 ew32(TDBAH, (tdba >> 32));
2752 tx_ring->head = E1000_TDH;
2753 tx_ring->tail = E1000_TDT;
2755 /* Set the default values for the Tx Inter Packet Gap timer */
2756 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2757 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2758 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2760 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2761 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2763 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2764 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2767 /* Set the Tx Interrupt Delay register */
2768 ew32(TIDV, adapter->tx_int_delay);
2769 /* Tx irq moderation */
2770 ew32(TADV, adapter->tx_abs_int_delay);
2772 if (adapter->flags2 & FLAG2_DMA_BURST) {
2773 u32 txdctl = er32(TXDCTL(0));
2774 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2775 E1000_TXDCTL_WTHRESH);
2777 * set up some performance related parameters to encourage the
2778 * hardware to use the bus more efficiently in bursts, depends
2779 * on the tx_int_delay to be enabled,
2780 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2781 * hthresh = 1 ==> prefetch when one or more available
2782 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2783 * BEWARE: this seems to work but should be considered first if
2784 * there are Tx hangs or other Tx related bugs
2786 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2787 ew32(TXDCTL(0), txdctl);
2788 /* erratum work around: set txdctl the same for both queues */
2789 ew32(TXDCTL(1), txdctl);
2792 /* Program the Transmit Control Register */
2794 tctl &= ~E1000_TCTL_CT;
2795 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2796 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2798 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2799 tarc = er32(TARC(0));
2801 * set the speed mode bit, we'll clear it if we're not at
2802 * gigabit link later
2804 #define SPEED_MODE_BIT (1 << 21)
2805 tarc |= SPEED_MODE_BIT;
2806 ew32(TARC(0), tarc);
2809 /* errata: program both queues to unweighted RR */
2810 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2811 tarc = er32(TARC(0));
2813 ew32(TARC(0), tarc);
2814 tarc = er32(TARC(1));
2816 ew32(TARC(1), tarc);
2819 /* Setup Transmit Descriptor Settings for eop descriptor */
2820 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2822 /* only set IDE if we are delaying interrupts using the timers */
2823 if (adapter->tx_int_delay)
2824 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2826 /* enable Report Status bit */
2827 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2831 e1000e_config_collision_dist(hw);
2835 * e1000_setup_rctl - configure the receive control registers
2836 * @adapter: Board private structure
2838 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2839 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2840 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2842 struct e1000_hw *hw = &adapter->hw;
2846 /* Workaround Si errata on 82579 - configure jumbo frame flow */
2847 if (hw->mac.type == e1000_pch2lan) {
2850 if (adapter->netdev->mtu > ETH_DATA_LEN)
2851 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2853 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2856 e_dbg("failed to enable jumbo frame workaround mode\n");
2859 /* Program MC offset vector base */
2861 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2862 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2863 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2864 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2866 /* Do not Store bad packets */
2867 rctl &= ~E1000_RCTL_SBP;
2869 /* Enable Long Packet receive */
2870 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2871 rctl &= ~E1000_RCTL_LPE;
2873 rctl |= E1000_RCTL_LPE;
2875 /* Some systems expect that the CRC is included in SMBUS traffic. The
2876 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2877 * host memory when this is enabled
2879 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2880 rctl |= E1000_RCTL_SECRC;
2882 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2883 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2886 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2888 phy_data |= (1 << 2);
2889 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2891 e1e_rphy(hw, 22, &phy_data);
2893 phy_data |= (1 << 14);
2894 e1e_wphy(hw, 0x10, 0x2823);
2895 e1e_wphy(hw, 0x11, 0x0003);
2896 e1e_wphy(hw, 22, phy_data);
2899 /* Setup buffer sizes */
2900 rctl &= ~E1000_RCTL_SZ_4096;
2901 rctl |= E1000_RCTL_BSEX;
2902 switch (adapter->rx_buffer_len) {
2905 rctl |= E1000_RCTL_SZ_2048;
2906 rctl &= ~E1000_RCTL_BSEX;
2909 rctl |= E1000_RCTL_SZ_4096;
2912 rctl |= E1000_RCTL_SZ_8192;
2915 rctl |= E1000_RCTL_SZ_16384;
2919 /* Enable Extended Status in all Receive Descriptors */
2920 rfctl = er32(RFCTL);
2921 rfctl |= E1000_RFCTL_EXTEN;
2924 * 82571 and greater support packet-split where the protocol
2925 * header is placed in skb->data and the packet data is
2926 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2927 * In the case of a non-split, skb->data is linearly filled,
2928 * followed by the page buffers. Therefore, skb->data is
2929 * sized to hold the largest protocol header.
2931 * allocations using alloc_page take too long for regular MTU
2932 * so only enable packet split for jumbo frames
2934 * Using pages when the page size is greater than 16k wastes
2935 * a lot of memory, since we allocate 3 pages at all times
2938 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2939 if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2940 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2941 adapter->rx_ps_pages = pages;
2943 adapter->rx_ps_pages = 0;
2945 if (adapter->rx_ps_pages) {
2949 * disable packet split support for IPv6 extension headers,
2950 * because some malformed IPv6 headers can hang the Rx
2952 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2953 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2955 /* Enable Packet split descriptors */
2956 rctl |= E1000_RCTL_DTYP_PS;
2958 psrctl |= adapter->rx_ps_bsize0 >>
2959 E1000_PSRCTL_BSIZE0_SHIFT;
2961 switch (adapter->rx_ps_pages) {
2963 psrctl |= PAGE_SIZE <<
2964 E1000_PSRCTL_BSIZE3_SHIFT;
2966 psrctl |= PAGE_SIZE <<
2967 E1000_PSRCTL_BSIZE2_SHIFT;
2969 psrctl |= PAGE_SIZE >>
2970 E1000_PSRCTL_BSIZE1_SHIFT;
2974 ew32(PSRCTL, psrctl);
2979 /* just started the receive unit, no need to restart */
2980 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2984 * e1000_configure_rx - Configure Receive Unit after Reset
2985 * @adapter: board private structure
2987 * Configure the Rx unit of the MAC after a reset.
2989 static void e1000_configure_rx(struct e1000_adapter *adapter)
2991 struct e1000_hw *hw = &adapter->hw;
2992 struct e1000_ring *rx_ring = adapter->rx_ring;
2994 u32 rdlen, rctl, rxcsum, ctrl_ext;
2996 if (adapter->rx_ps_pages) {
2997 /* this is a 32 byte descriptor */
2998 rdlen = rx_ring->count *
2999 sizeof(union e1000_rx_desc_packet_split);
3000 adapter->clean_rx = e1000_clean_rx_irq_ps;
3001 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
3002 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
3003 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3004 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
3005 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
3007 rdlen = rx_ring->count * sizeof(union e1000_rx_desc_extended);
3008 adapter->clean_rx = e1000_clean_rx_irq;
3009 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
3012 /* disable receives while setting up the descriptors */
3014 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3015 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3017 usleep_range(10000, 20000);
3019 if (adapter->flags2 & FLAG2_DMA_BURST) {
3021 * set the writeback threshold (only takes effect if the RDTR
3022 * is set). set GRAN=1 and write back up to 0x4 worth, and
3023 * enable prefetching of 0x20 Rx descriptors
3029 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
3030 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
3033 * override the delay timers for enabling bursting, only if
3034 * the value was not set by the user via module options
3036 if (adapter->rx_int_delay == DEFAULT_RDTR)
3037 adapter->rx_int_delay = BURST_RDTR;
3038 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
3039 adapter->rx_abs_int_delay = BURST_RADV;
3042 /* set the Receive Delay Timer Register */
3043 ew32(RDTR, adapter->rx_int_delay);
3045 /* irq moderation */
3046 ew32(RADV, adapter->rx_abs_int_delay);
3047 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
3048 ew32(ITR, 1000000000 / (adapter->itr * 256));
3050 ctrl_ext = er32(CTRL_EXT);
3051 /* Auto-Mask interrupts upon ICR access */
3052 ctrl_ext |= E1000_CTRL_EXT_IAME;
3053 ew32(IAM, 0xffffffff);
3054 ew32(CTRL_EXT, ctrl_ext);
3058 * Setup the HW Rx Head and Tail Descriptor Pointers and
3059 * the Base and Length of the Rx Descriptor Ring
3061 rdba = rx_ring->dma;
3062 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
3063 ew32(RDBAH, (rdba >> 32));
3067 rx_ring->head = E1000_RDH;
3068 rx_ring->tail = E1000_RDT;
3070 /* Enable Receive Checksum Offload for TCP and UDP */
3071 rxcsum = er32(RXCSUM);
3072 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
3073 rxcsum |= E1000_RXCSUM_TUOFL;
3076 * IPv4 payload checksum for UDP fragments must be
3077 * used in conjunction with packet-split.
3079 if (adapter->rx_ps_pages)
3080 rxcsum |= E1000_RXCSUM_IPPCSE;
3082 rxcsum &= ~E1000_RXCSUM_TUOFL;
3083 /* no need to clear IPPCSE as it defaults to 0 */
3085 ew32(RXCSUM, rxcsum);
3088 * Enable early receives on supported devices, only takes effect when
3089 * packet size is equal or larger than the specified value (in 8 byte
3090 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
3092 if ((adapter->flags & FLAG_HAS_ERT) ||
3093 (adapter->hw.mac.type == e1000_pch2lan)) {
3094 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3095 u32 rxdctl = er32(RXDCTL(0));
3096 ew32(RXDCTL(0), rxdctl | 0x3);
3097 if (adapter->flags & FLAG_HAS_ERT)
3098 ew32(ERT, E1000_ERT_2048 | (1 << 13));
3100 * With jumbo frames and early-receive enabled,
3101 * excessive C-state transition latencies result in
3102 * dropped transactions.
3104 pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3106 pm_qos_update_request(&adapter->netdev->pm_qos_req,
3107 PM_QOS_DEFAULT_VALUE);
3111 /* Enable Receives */
3116 * e1000_update_mc_addr_list - Update Multicast addresses
3117 * @hw: pointer to the HW structure
3118 * @mc_addr_list: array of multicast addresses to program
3119 * @mc_addr_count: number of multicast addresses to program
3121 * Updates the Multicast Table Array.
3122 * The caller must have a packed mc_addr_list of multicast addresses.
3124 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3127 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3131 * e1000_set_multi - Multicast and Promiscuous mode set
3132 * @netdev: network interface device structure
3134 * The set_multi entry point is called whenever the multicast address
3135 * list or the network interface flags are updated. This routine is
3136 * responsible for configuring the hardware for proper multicast,
3137 * promiscuous mode, and all-multi behavior.
3139 static void e1000_set_multi(struct net_device *netdev)
3141 struct e1000_adapter *adapter = netdev_priv(netdev);
3142 struct e1000_hw *hw = &adapter->hw;
3143 struct netdev_hw_addr *ha;
3147 /* Check for Promiscuous and All Multicast modes */
3151 if (netdev->flags & IFF_PROMISC) {
3152 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3153 rctl &= ~E1000_RCTL_VFE;
3154 /* Do not hardware filter VLANs in promisc mode */
3155 e1000e_vlan_filter_disable(adapter);
3157 if (netdev->flags & IFF_ALLMULTI) {
3158 rctl |= E1000_RCTL_MPE;
3159 rctl &= ~E1000_RCTL_UPE;
3161 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3163 e1000e_vlan_filter_enable(adapter);
3168 if (!netdev_mc_empty(netdev)) {
3171 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3175 /* prepare a packed array of only addresses. */
3176 netdev_for_each_mc_addr(ha, netdev)
3177 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3179 e1000_update_mc_addr_list(hw, mta_list, i);
3183 * if we're called from probe, we might not have
3184 * anything to do here, so clear out the list
3186 e1000_update_mc_addr_list(hw, NULL, 0);
3189 if (netdev->features & NETIF_F_HW_VLAN_RX)
3190 e1000e_vlan_strip_enable(adapter);
3192 e1000e_vlan_strip_disable(adapter);
3196 * e1000_configure - configure the hardware for Rx and Tx
3197 * @adapter: private board structure
3199 static void e1000_configure(struct e1000_adapter *adapter)
3201 e1000_set_multi(adapter->netdev);
3203 e1000_restore_vlan(adapter);
3204 e1000_init_manageability_pt(adapter);
3206 e1000_configure_tx(adapter);
3207 e1000_setup_rctl(adapter);
3208 e1000_configure_rx(adapter);
3209 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring),
3214 * e1000e_power_up_phy - restore link in case the phy was powered down
3215 * @adapter: address of board private structure
3217 * The phy may be powered down to save power and turn off link when the
3218 * driver is unloaded and wake on lan is not enabled (among others)
3219 * *** this routine MUST be followed by a call to e1000e_reset ***
3221 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3223 if (adapter->hw.phy.ops.power_up)
3224 adapter->hw.phy.ops.power_up(&adapter->hw);
3226 adapter->hw.mac.ops.setup_link(&adapter->hw);
3230 * e1000_power_down_phy - Power down the PHY
3232 * Power down the PHY so no link is implied when interface is down.
3233 * The PHY cannot be powered down if management or WoL is active.
3235 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3237 /* WoL is enabled */
3241 if (adapter->hw.phy.ops.power_down)
3242 adapter->hw.phy.ops.power_down(&adapter->hw);
3246 * e1000e_reset - bring the hardware into a known good state
3248 * This function boots the hardware and enables some settings that
3249 * require a configuration cycle of the hardware - those cannot be
3250 * set/changed during runtime. After reset the device needs to be
3251 * properly configured for Rx, Tx etc.
3253 void e1000e_reset(struct e1000_adapter *adapter)
3255 struct e1000_mac_info *mac = &adapter->hw.mac;
3256 struct e1000_fc_info *fc = &adapter->hw.fc;
3257 struct e1000_hw *hw = &adapter->hw;
3258 u32 tx_space, min_tx_space, min_rx_space;
3259 u32 pba = adapter->pba;
3262 /* reset Packet Buffer Allocation to default */
3265 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3267 * To maintain wire speed transmits, the Tx FIFO should be
3268 * large enough to accommodate two full transmit packets,
3269 * rounded up to the next 1KB and expressed in KB. Likewise,
3270 * the Rx FIFO should be large enough to accommodate at least
3271 * one full receive packet and is similarly rounded up and
3275 /* upper 16 bits has Tx packet buffer allocation size in KB */
3276 tx_space = pba >> 16;
3277 /* lower 16 bits has Rx packet buffer allocation size in KB */
3280 * the Tx fifo also stores 16 bytes of information about the Tx
3281 * but don't include ethernet FCS because hardware appends it
3283 min_tx_space = (adapter->max_frame_size +
3284 sizeof(struct e1000_tx_desc) -
3286 min_tx_space = ALIGN(min_tx_space, 1024);
3287 min_tx_space >>= 10;
3288 /* software strips receive CRC, so leave room for it */
3289 min_rx_space = adapter->max_frame_size;
3290 min_rx_space = ALIGN(min_rx_space, 1024);
3291 min_rx_space >>= 10;
3294 * If current Tx allocation is less than the min Tx FIFO size,
3295 * and the min Tx FIFO size is less than the current Rx FIFO
3296 * allocation, take space away from current Rx allocation
3298 if ((tx_space < min_tx_space) &&
3299 ((min_tx_space - tx_space) < pba)) {
3300 pba -= min_tx_space - tx_space;
3303 * if short on Rx space, Rx wins and must trump Tx
3304 * adjustment or use Early Receive if available
3306 if ((pba < min_rx_space) &&
3307 (!(adapter->flags & FLAG_HAS_ERT)))
3308 /* ERT enabled in e1000_configure_rx */
3316 * flow control settings
3318 * The high water mark must be low enough to fit one full frame
3319 * (or the size used for early receive) above it in the Rx FIFO.
3320 * Set it to the lower of:
3321 * - 90% of the Rx FIFO size, and
3322 * - the full Rx FIFO size minus the early receive size (for parts
3323 * with ERT support assuming ERT set to E1000_ERT_2048), or
3324 * - the full Rx FIFO size minus one full frame
3326 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3327 fc->pause_time = 0xFFFF;
3329 fc->pause_time = E1000_FC_PAUSE_TIME;
3331 fc->current_mode = fc->requested_mode;
3333 switch (hw->mac.type) {
3335 if ((adapter->flags & FLAG_HAS_ERT) &&
3336 (adapter->netdev->mtu > ETH_DATA_LEN))
3337 hwm = min(((pba << 10) * 9 / 10),
3338 ((pba << 10) - (E1000_ERT_2048 << 3)));
3340 hwm = min(((pba << 10) * 9 / 10),
3341 ((pba << 10) - adapter->max_frame_size));
3343 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3344 fc->low_water = fc->high_water - 8;
3348 * Workaround PCH LOM adapter hangs with certain network
3349 * loads. If hangs persist, try disabling Tx flow control.
3351 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3352 fc->high_water = 0x3500;
3353 fc->low_water = 0x1500;
3355 fc->high_water = 0x5000;
3356 fc->low_water = 0x3000;
3358 fc->refresh_time = 0x1000;
3361 fc->high_water = 0x05C20;
3362 fc->low_water = 0x05048;
3363 fc->pause_time = 0x0650;
3364 fc->refresh_time = 0x0400;
3365 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3373 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3374 * fit in receive buffer and early-receive not supported.
3376 if (adapter->itr_setting & 0x3) {
3377 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3378 !(adapter->flags & FLAG_HAS_ERT)) {
3379 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3380 dev_info(&adapter->pdev->dev,
3381 "Interrupt Throttle Rate turned off\n");
3382 adapter->flags2 |= FLAG2_DISABLE_AIM;
3385 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3386 dev_info(&adapter->pdev->dev,
3387 "Interrupt Throttle Rate turned on\n");
3388 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3389 adapter->itr = 20000;
3390 ew32(ITR, 1000000000 / (adapter->itr * 256));
3394 /* Allow time for pending master requests to run */
3395 mac->ops.reset_hw(hw);
3398 * For parts with AMT enabled, let the firmware know
3399 * that the network interface is in control
3401 if (adapter->flags & FLAG_HAS_AMT)
3402 e1000e_get_hw_control(adapter);
3406 if (mac->ops.init_hw(hw))
3407 e_err("Hardware Error\n");
3409 e1000_update_mng_vlan(adapter);
3411 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3412 ew32(VET, ETH_P_8021Q);
3414 e1000e_reset_adaptive(hw);
3416 if (!netif_running(adapter->netdev) &&
3417 !test_bit(__E1000_TESTING, &adapter->state)) {
3418 e1000_power_down_phy(adapter);
3422 e1000_get_phy_info(hw);
3424 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3425 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3428 * speed up time to link by disabling smart power down, ignore
3429 * the return value of this function because there is nothing
3430 * different we would do if it failed
3432 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3433 phy_data &= ~IGP02E1000_PM_SPD;
3434 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3438 int e1000e_up(struct e1000_adapter *adapter)
3440 struct e1000_hw *hw = &adapter->hw;
3442 /* hardware has been reset, we need to reload some things */
3443 e1000_configure(adapter);
3445 clear_bit(__E1000_DOWN, &adapter->state);
3447 napi_enable(&adapter->napi);
3448 if (adapter->msix_entries)
3449 e1000_configure_msix(adapter);
3450 e1000_irq_enable(adapter);
3452 netif_start_queue(adapter->netdev);
3454 /* fire a link change interrupt to start the watchdog */
3455 if (adapter->msix_entries)
3456 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3458 ew32(ICS, E1000_ICS_LSC);
3463 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3465 struct e1000_hw *hw = &adapter->hw;
3467 if (!(adapter->flags2 & FLAG2_DMA_BURST))
3470 /* flush pending descriptor writebacks to memory */
3471 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3472 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3474 /* execute the writes immediately */
3478 static void e1000e_update_stats(struct e1000_adapter *adapter);
3480 void e1000e_down(struct e1000_adapter *adapter)
3482 struct net_device *netdev = adapter->netdev;
3483 struct e1000_hw *hw = &adapter->hw;
3487 * signal that we're down so the interrupt handler does not
3488 * reschedule our watchdog timer
3490 set_bit(__E1000_DOWN, &adapter->state);
3492 /* disable receives in the hardware */
3494 if (!(adapter->flags2 & FLAG2_NO_DISABLE_RX))
3495 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3496 /* flush and sleep below */
3498 netif_stop_queue(netdev);
3500 /* disable transmits in the hardware */
3502 tctl &= ~E1000_TCTL_EN;
3505 /* flush both disables and wait for them to finish */
3507 usleep_range(10000, 20000);
3509 napi_disable(&adapter->napi);
3510 e1000_irq_disable(adapter);
3512 del_timer_sync(&adapter->watchdog_timer);
3513 del_timer_sync(&adapter->phy_info_timer);
3515 netif_carrier_off(netdev);
3517 spin_lock(&adapter->stats64_lock);
3518 e1000e_update_stats(adapter);
3519 spin_unlock(&adapter->stats64_lock);
3521 e1000e_flush_descriptors(adapter);
3522 e1000_clean_tx_ring(adapter);
3523 e1000_clean_rx_ring(adapter);
3525 adapter->link_speed = 0;
3526 adapter->link_duplex = 0;
3528 if (!pci_channel_offline(adapter->pdev))
3529 e1000e_reset(adapter);
3532 * TODO: for power management, we could drop the link and
3533 * pci_disable_device here.
3537 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3540 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3541 usleep_range(1000, 2000);
3542 e1000e_down(adapter);
3544 clear_bit(__E1000_RESETTING, &adapter->state);
3548 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3549 * @adapter: board private structure to initialize
3551 * e1000_sw_init initializes the Adapter private data structure.
3552 * Fields are initialized based on PCI device information and
3553 * OS network device settings (MTU size).
3555 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3557 struct net_device *netdev = adapter->netdev;
3559 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3560 adapter->rx_ps_bsize0 = 128;
3561 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3562 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3564 spin_lock_init(&adapter->stats64_lock);
3566 e1000e_set_interrupt_capability(adapter);
3568 if (e1000_alloc_queues(adapter))
3571 /* Explicitly disable IRQ since the NIC can be in any state. */
3572 e1000_irq_disable(adapter);
3574 set_bit(__E1000_DOWN, &adapter->state);
3579 * e1000_intr_msi_test - Interrupt Handler
3580 * @irq: interrupt number
3581 * @data: pointer to a network interface device structure
3583 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3585 struct net_device *netdev = data;
3586 struct e1000_adapter *adapter = netdev_priv(netdev);
3587 struct e1000_hw *hw = &adapter->hw;
3588 u32 icr = er32(ICR);
3590 e_dbg("icr is %08X\n", icr);
3591 if (icr & E1000_ICR_RXSEQ) {
3592 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3600 * e1000_test_msi_interrupt - Returns 0 for successful test
3601 * @adapter: board private struct
3603 * code flow taken from tg3.c
3605 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3607 struct net_device *netdev = adapter->netdev;
3608 struct e1000_hw *hw = &adapter->hw;
3611 /* poll_enable hasn't been called yet, so don't need disable */
3612 /* clear any pending events */
3615 /* free the real vector and request a test handler */
3616 e1000_free_irq(adapter);
3617 e1000e_reset_interrupt_capability(adapter);
3619 /* Assume that the test fails, if it succeeds then the test
3620 * MSI irq handler will unset this flag */
3621 adapter->flags |= FLAG_MSI_TEST_FAILED;
3623 err = pci_enable_msi(adapter->pdev);
3625 goto msi_test_failed;
3627 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3628 netdev->name, netdev);
3630 pci_disable_msi(adapter->pdev);
3631 goto msi_test_failed;
3636 e1000_irq_enable(adapter);
3638 /* fire an unusual interrupt on the test handler */
3639 ew32(ICS, E1000_ICS_RXSEQ);
3643 e1000_irq_disable(adapter);
3647 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3648 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3649 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3651 e_dbg("MSI interrupt test succeeded!\n");
3653 free_irq(adapter->pdev->irq, netdev);
3654 pci_disable_msi(adapter->pdev);
3657 e1000e_set_interrupt_capability(adapter);
3658 return e1000_request_irq(adapter);
3662 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3663 * @adapter: board private struct
3665 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3667 static int e1000_test_msi(struct e1000_adapter *adapter)
3672 if (!(adapter->flags & FLAG_MSI_ENABLED))
3675 /* disable SERR in case the MSI write causes a master abort */
3676 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3677 if (pci_cmd & PCI_COMMAND_SERR)
3678 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3679 pci_cmd & ~PCI_COMMAND_SERR);
3681 err = e1000_test_msi_interrupt(adapter);
3683 /* re-enable SERR */
3684 if (pci_cmd & PCI_COMMAND_SERR) {
3685 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3686 pci_cmd |= PCI_COMMAND_SERR;
3687 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3694 * e1000_open - Called when a network interface is made active
3695 * @netdev: network interface device structure
3697 * Returns 0 on success, negative value on failure
3699 * The open entry point is called when a network interface is made
3700 * active by the system (IFF_UP). At this point all resources needed
3701 * for transmit and receive operations are allocated, the interrupt
3702 * handler is registered with the OS, the watchdog timer is started,
3703 * and the stack is notified that the interface is ready.
3705 static int e1000_open(struct net_device *netdev)
3707 struct e1000_adapter *adapter = netdev_priv(netdev);
3708 struct e1000_hw *hw = &adapter->hw;
3709 struct pci_dev *pdev = adapter->pdev;
3712 /* disallow open during test */
3713 if (test_bit(__E1000_TESTING, &adapter->state))
3716 pm_runtime_get_sync(&pdev->dev);
3718 netif_carrier_off(netdev);
3720 /* allocate transmit descriptors */
3721 err = e1000e_setup_tx_resources(adapter);
3725 /* allocate receive descriptors */
3726 err = e1000e_setup_rx_resources(adapter);
3731 * If AMT is enabled, let the firmware know that the network
3732 * interface is now open and reset the part to a known state.
3734 if (adapter->flags & FLAG_HAS_AMT) {
3735 e1000e_get_hw_control(adapter);
3736 e1000e_reset(adapter);
3739 e1000e_power_up_phy(adapter);
3741 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3742 if ((adapter->hw.mng_cookie.status &
3743 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3744 e1000_update_mng_vlan(adapter);
3746 /* DMA latency requirement to workaround early-receive/jumbo issue */
3747 if ((adapter->flags & FLAG_HAS_ERT) ||
3748 (adapter->hw.mac.type == e1000_pch2lan))
3749 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3750 PM_QOS_CPU_DMA_LATENCY,
3751 PM_QOS_DEFAULT_VALUE);
3754 * before we allocate an interrupt, we must be ready to handle it.
3755 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3756 * as soon as we call pci_request_irq, so we have to setup our
3757 * clean_rx handler before we do so.
3759 e1000_configure(adapter);
3761 err = e1000_request_irq(adapter);
3766 * Work around PCIe errata with MSI interrupts causing some chipsets to
3767 * ignore e1000e MSI messages, which means we need to test our MSI
3770 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3771 err = e1000_test_msi(adapter);
3773 e_err("Interrupt allocation failed\n");
3778 /* From here on the code is the same as e1000e_up() */
3779 clear_bit(__E1000_DOWN, &adapter->state);
3781 napi_enable(&adapter->napi);
3783 e1000_irq_enable(adapter);
3785 netif_start_queue(netdev);
3787 adapter->idle_check = true;
3788 pm_runtime_put(&pdev->dev);
3790 /* fire a link status change interrupt to start the watchdog */
3791 if (adapter->msix_entries)
3792 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3794 ew32(ICS, E1000_ICS_LSC);
3799 e1000e_release_hw_control(adapter);
3800 e1000_power_down_phy(adapter);
3801 e1000e_free_rx_resources(adapter);
3803 e1000e_free_tx_resources(adapter);
3805 e1000e_reset(adapter);
3806 pm_runtime_put_sync(&pdev->dev);
3812 * e1000_close - Disables a network interface
3813 * @netdev: network interface device structure
3815 * Returns 0, this is not allowed to fail
3817 * The close entry point is called when an interface is de-activated
3818 * by the OS. The hardware is still under the drivers control, but
3819 * needs to be disabled. A global MAC reset is issued to stop the
3820 * hardware, and all transmit and receive resources are freed.
3822 static int e1000_close(struct net_device *netdev)
3824 struct e1000_adapter *adapter = netdev_priv(netdev);
3825 struct pci_dev *pdev = adapter->pdev;
3827 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3829 pm_runtime_get_sync(&pdev->dev);
3831 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3832 e1000e_down(adapter);
3833 e1000_free_irq(adapter);
3835 e1000_power_down_phy(adapter);
3837 e1000e_free_tx_resources(adapter);
3838 e1000e_free_rx_resources(adapter);
3841 * kill manageability vlan ID if supported, but not if a vlan with
3842 * the same ID is registered on the host OS (let 8021q kill it)
3844 if (adapter->hw.mng_cookie.status &
3845 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3846 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3849 * If AMT is enabled, let the firmware know that the network
3850 * interface is now closed
3852 if ((adapter->flags & FLAG_HAS_AMT) &&
3853 !test_bit(__E1000_TESTING, &adapter->state))
3854 e1000e_release_hw_control(adapter);
3856 if ((adapter->flags & FLAG_HAS_ERT) ||
3857 (adapter->hw.mac.type == e1000_pch2lan))
3858 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3860 pm_runtime_put_sync(&pdev->dev);
3865 * e1000_set_mac - Change the Ethernet Address of the NIC
3866 * @netdev: network interface device structure
3867 * @p: pointer to an address structure
3869 * Returns 0 on success, negative on failure
3871 static int e1000_set_mac(struct net_device *netdev, void *p)
3873 struct e1000_adapter *adapter = netdev_priv(netdev);
3874 struct sockaddr *addr = p;
3876 if (!is_valid_ether_addr(addr->sa_data))
3877 return -EADDRNOTAVAIL;
3879 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3880 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3882 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3884 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3885 /* activate the work around */
3886 e1000e_set_laa_state_82571(&adapter->hw, 1);
3889 * Hold a copy of the LAA in RAR[14] This is done so that
3890 * between the time RAR[0] gets clobbered and the time it
3891 * gets fixed (in e1000_watchdog), the actual LAA is in one
3892 * of the RARs and no incoming packets directed to this port
3893 * are dropped. Eventually the LAA will be in RAR[0] and
3896 e1000e_rar_set(&adapter->hw,
3897 adapter->hw.mac.addr,
3898 adapter->hw.mac.rar_entry_count - 1);
3905 * e1000e_update_phy_task - work thread to update phy
3906 * @work: pointer to our work struct
3908 * this worker thread exists because we must acquire a
3909 * semaphore to read the phy, which we could msleep while
3910 * waiting for it, and we can't msleep in a timer.
3912 static void e1000e_update_phy_task(struct work_struct *work)
3914 struct e1000_adapter *adapter = container_of(work,
3915 struct e1000_adapter, update_phy_task);
3917 if (test_bit(__E1000_DOWN, &adapter->state))
3920 e1000_get_phy_info(&adapter->hw);
3924 * Need to wait a few seconds after link up to get diagnostic information from
3927 static void e1000_update_phy_info(unsigned long data)
3929 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3931 if (test_bit(__E1000_DOWN, &adapter->state))
3934 schedule_work(&adapter->update_phy_task);
3938 * e1000e_update_phy_stats - Update the PHY statistics counters
3939 * @adapter: board private structure
3941 * Read/clear the upper 16-bit PHY registers and read/accumulate lower
3943 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3945 struct e1000_hw *hw = &adapter->hw;
3949 ret_val = hw->phy.ops.acquire(hw);
3954 * A page set is expensive so check if already on desired page.
3955 * If not, set to the page with the PHY status registers.
3958 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3962 if (phy_data != (HV_STATS_PAGE << IGP_PAGE_SHIFT)) {
3963 ret_val = hw->phy.ops.set_page(hw,
3964 HV_STATS_PAGE << IGP_PAGE_SHIFT);
3969 /* Single Collision Count */
3970 hw->phy.ops.read_reg_page(hw, HV_SCC_UPPER, &phy_data);
3971 ret_val = hw->phy.ops.read_reg_page(hw, HV_SCC_LOWER, &phy_data);
3973 adapter->stats.scc += phy_data;
3975 /* Excessive Collision Count */
3976 hw->phy.ops.read_reg_page(hw, HV_ECOL_UPPER, &phy_data);
3977 ret_val = hw->phy.ops.read_reg_page(hw, HV_ECOL_LOWER, &phy_data);
3979 adapter->stats.ecol += phy_data;
3981 /* Multiple Collision Count */
3982 hw->phy.ops.read_reg_page(hw, HV_MCC_UPPER, &phy_data);
3983 ret_val = hw->phy.ops.read_reg_page(hw, HV_MCC_LOWER, &phy_data);
3985 adapter->stats.mcc += phy_data;
3987 /* Late Collision Count */
3988 hw->phy.ops.read_reg_page(hw, HV_LATECOL_UPPER, &phy_data);
3989 ret_val = hw->phy.ops.read_reg_page(hw, HV_LATECOL_LOWER, &phy_data);
3991 adapter->stats.latecol += phy_data;
3993 /* Collision Count - also used for adaptive IFS */
3994 hw->phy.ops.read_reg_page(hw, HV_COLC_UPPER, &phy_data);
3995 ret_val = hw->phy.ops.read_reg_page(hw, HV_COLC_LOWER, &phy_data);
3997 hw->mac.collision_delta = phy_data;
4000 hw->phy.ops.read_reg_page(hw, HV_DC_UPPER, &phy_data);
4001 ret_val = hw->phy.ops.read_reg_page(hw, HV_DC_LOWER, &phy_data);
4003 adapter->stats.dc += phy_data;
4005 /* Transmit with no CRS */
4006 hw->phy.ops.read_reg_page(hw, HV_TNCRS_UPPER, &phy_data);
4007 ret_val = hw->phy.ops.read_reg_page(hw, HV_TNCRS_LOWER, &phy_data);
4009 adapter->stats.tncrs += phy_data;
4012 hw->phy.ops.release(hw);
4016 * e1000e_update_stats - Update the board statistics counters
4017 * @adapter: board private structure
4019 static void e1000e_update_stats(struct e1000_adapter *adapter)
4021 struct net_device *netdev = adapter->netdev;
4022 struct e1000_hw *hw = &adapter->hw;
4023 struct pci_dev *pdev = adapter->pdev;
4026 * Prevent stats update while adapter is being reset, or if the pci
4027 * connection is down.
4029 if (adapter->link_speed == 0)
4031 if (pci_channel_offline(pdev))
4034 adapter->stats.crcerrs += er32(CRCERRS);
4035 adapter->stats.gprc += er32(GPRC);
4036 adapter->stats.gorc += er32(GORCL);
4037 er32(GORCH); /* Clear gorc */
4038 adapter->stats.bprc += er32(BPRC);
4039 adapter->stats.mprc += er32(MPRC);
4040 adapter->stats.roc += er32(ROC);
4042 adapter->stats.mpc += er32(MPC);
4044 /* Half-duplex statistics */
4045 if (adapter->link_duplex == HALF_DUPLEX) {
4046 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
4047 e1000e_update_phy_stats(adapter);
4049 adapter->stats.scc += er32(SCC);
4050 adapter->stats.ecol += er32(ECOL);
4051 adapter->stats.mcc += er32(MCC);
4052 adapter->stats.latecol += er32(LATECOL);
4053 adapter->stats.dc += er32(DC);
4055 hw->mac.collision_delta = er32(COLC);
4057 if ((hw->mac.type != e1000_82574) &&
4058 (hw->mac.type != e1000_82583))
4059 adapter->stats.tncrs += er32(TNCRS);
4061 adapter->stats.colc += hw->mac.collision_delta;
4064 adapter->stats.xonrxc += er32(XONRXC);
4065 adapter->stats.xontxc += er32(XONTXC);
4066 adapter->stats.xoffrxc += er32(XOFFRXC);
4067 adapter->stats.xofftxc += er32(XOFFTXC);
4068 adapter->stats.gptc += er32(GPTC);
4069 adapter->stats.gotc += er32(GOTCL);
4070 er32(GOTCH); /* Clear gotc */
4071 adapter->stats.rnbc += er32(RNBC);
4072 adapter->stats.ruc += er32(RUC);
4074 adapter->stats.mptc += er32(MPTC);
4075 adapter->stats.bptc += er32(BPTC);
4077 /* used for adaptive IFS */
4079 hw->mac.tx_packet_delta = er32(TPT);
4080 adapter->stats.tpt += hw->mac.tx_packet_delta;
4082 adapter->stats.algnerrc += er32(ALGNERRC);
4083 adapter->stats.rxerrc += er32(RXERRC);
4084 adapter->stats.cexterr += er32(CEXTERR);
4085 adapter->stats.tsctc += er32(TSCTC);
4086 adapter->stats.tsctfc += er32(TSCTFC);
4088 /* Fill out the OS statistics structure */
4089 netdev->stats.multicast = adapter->stats.mprc;
4090 netdev->stats.collisions = adapter->stats.colc;
4095 * RLEC on some newer hardware can be incorrect so build
4096 * our own version based on RUC and ROC
4098 netdev->stats.rx_errors = adapter->stats.rxerrc +
4099 adapter->stats.crcerrs + adapter->stats.algnerrc +
4100 adapter->stats.ruc + adapter->stats.roc +
4101 adapter->stats.cexterr;
4102 netdev->stats.rx_length_errors = adapter->stats.ruc +
4104 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4105 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4106 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4109 netdev->stats.tx_errors = adapter->stats.ecol +
4110 adapter->stats.latecol;
4111 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4112 netdev->stats.tx_window_errors = adapter->stats.latecol;
4113 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4115 /* Tx Dropped needs to be maintained elsewhere */
4117 /* Management Stats */
4118 adapter->stats.mgptc += er32(MGTPTC);
4119 adapter->stats.mgprc += er32(MGTPRC);
4120 adapter->stats.mgpdc += er32(MGTPDC);
4124 * e1000_phy_read_status - Update the PHY register status snapshot
4125 * @adapter: board private structure
4127 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4129 struct e1000_hw *hw = &adapter->hw;
4130 struct e1000_phy_regs *phy = &adapter->phy_regs;
4132 if ((er32(STATUS) & E1000_STATUS_LU) &&
4133 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4136 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4137 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4138 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4139 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4140 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4141 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4142 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4143 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4145 e_warn("Error reading PHY register\n");
4148 * Do not read PHY registers if link is not up
4149 * Set values to typical power-on defaults
4151 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4152 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4153 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4155 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4156 ADVERTISE_ALL | ADVERTISE_CSMA);
4158 phy->expansion = EXPANSION_ENABLENPAGE;
4159 phy->ctrl1000 = ADVERTISE_1000FULL;
4161 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4165 static void e1000_print_link_info(struct e1000_adapter *adapter)
4167 struct e1000_hw *hw = &adapter->hw;
4168 u32 ctrl = er32(CTRL);
4170 /* Link status message must follow this format for user tools */
4171 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4172 "Flow Control: %s\n",
4173 adapter->netdev->name,
4174 adapter->link_speed,
4175 (adapter->link_duplex == FULL_DUPLEX) ?
4176 "Full Duplex" : "Half Duplex",
4177 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4179 ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4180 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4183 static bool e1000e_has_link(struct e1000_adapter *adapter)
4185 struct e1000_hw *hw = &adapter->hw;
4186 bool link_active = 0;
4190 * get_link_status is set on LSC (link status) interrupt or
4191 * Rx sequence error interrupt. get_link_status will stay
4192 * false until the check_for_link establishes link
4193 * for copper adapters ONLY
4195 switch (hw->phy.media_type) {
4196 case e1000_media_type_copper:
4197 if (hw->mac.get_link_status) {
4198 ret_val = hw->mac.ops.check_for_link(hw);
4199 link_active = !hw->mac.get_link_status;
4204 case e1000_media_type_fiber:
4205 ret_val = hw->mac.ops.check_for_link(hw);
4206 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4208 case e1000_media_type_internal_serdes:
4209 ret_val = hw->mac.ops.check_for_link(hw);
4210 link_active = adapter->hw.mac.serdes_has_link;
4213 case e1000_media_type_unknown:
4217 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4218 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4219 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4220 e_info("Gigabit has been disabled, downgrading speed\n");
4226 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4228 /* make sure the receive unit is started */
4229 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4230 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4231 struct e1000_hw *hw = &adapter->hw;
4232 u32 rctl = er32(RCTL);
4233 ew32(RCTL, rctl | E1000_RCTL_EN);
4234 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4238 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4240 struct e1000_hw *hw = &adapter->hw;
4243 * With 82574 controllers, PHY needs to be checked periodically
4244 * for hung state and reset, if two calls return true
4246 if (e1000_check_phy_82574(hw))
4247 adapter->phy_hang_count++;
4249 adapter->phy_hang_count = 0;
4251 if (adapter->phy_hang_count > 1) {
4252 adapter->phy_hang_count = 0;
4253 schedule_work(&adapter->reset_task);
4258 * e1000_watchdog - Timer Call-back
4259 * @data: pointer to adapter cast into an unsigned long
4261 static void e1000_watchdog(unsigned long data)
4263 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4265 /* Do the rest outside of interrupt context */
4266 schedule_work(&adapter->watchdog_task);
4268 /* TODO: make this use queue_delayed_work() */
4271 static void e1000_watchdog_task(struct work_struct *work)
4273 struct e1000_adapter *adapter = container_of(work,
4274 struct e1000_adapter, watchdog_task);
4275 struct net_device *netdev = adapter->netdev;
4276 struct e1000_mac_info *mac = &adapter->hw.mac;
4277 struct e1000_phy_info *phy = &adapter->hw.phy;
4278 struct e1000_ring *tx_ring = adapter->tx_ring;
4279 struct e1000_hw *hw = &adapter->hw;
4282 if (test_bit(__E1000_DOWN, &adapter->state))
4285 link = e1000e_has_link(adapter);
4286 if ((netif_carrier_ok(netdev)) && link) {
4287 /* Cancel scheduled suspend requests. */
4288 pm_runtime_resume(netdev->dev.parent);
4290 e1000e_enable_receives(adapter);
4294 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4295 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4296 e1000_update_mng_vlan(adapter);
4299 if (!netif_carrier_ok(netdev)) {
4302 /* Cancel scheduled suspend requests. */
4303 pm_runtime_resume(netdev->dev.parent);
4305 /* update snapshot of PHY registers on LSC */
4306 e1000_phy_read_status(adapter);
4307 mac->ops.get_link_up_info(&adapter->hw,
4308 &adapter->link_speed,
4309 &adapter->link_duplex);
4310 e1000_print_link_info(adapter);
4312 * On supported PHYs, check for duplex mismatch only
4313 * if link has autonegotiated at 10/100 half
4315 if ((hw->phy.type == e1000_phy_igp_3 ||
4316 hw->phy.type == e1000_phy_bm) &&
4317 (hw->mac.autoneg == true) &&
4318 (adapter->link_speed == SPEED_10 ||
4319 adapter->link_speed == SPEED_100) &&
4320 (adapter->link_duplex == HALF_DUPLEX)) {
4323 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4325 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4326 e_info("Autonegotiated half duplex but"
4327 " link partner cannot autoneg. "
4328 " Try forcing full duplex if "
4329 "link gets many collisions.\n");
4332 /* adjust timeout factor according to speed/duplex */
4333 adapter->tx_timeout_factor = 1;
4334 switch (adapter->link_speed) {
4337 adapter->tx_timeout_factor = 16;
4341 adapter->tx_timeout_factor = 10;
4346 * workaround: re-program speed mode bit after
4349 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4352 tarc0 = er32(TARC(0));
4353 tarc0 &= ~SPEED_MODE_BIT;
4354 ew32(TARC(0), tarc0);
4358 * disable TSO for pcie and 10/100 speeds, to avoid
4359 * some hardware issues
4361 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4362 switch (adapter->link_speed) {
4365 e_info("10/100 speed: disabling TSO\n");
4366 netdev->features &= ~NETIF_F_TSO;
4367 netdev->features &= ~NETIF_F_TSO6;
4370 netdev->features |= NETIF_F_TSO;
4371 netdev->features |= NETIF_F_TSO6;
4380 * enable transmits in the hardware, need to do this
4381 * after setting TARC(0)
4384 tctl |= E1000_TCTL_EN;
4388 * Perform any post-link-up configuration before
4389 * reporting link up.
4391 if (phy->ops.cfg_on_link_up)
4392 phy->ops.cfg_on_link_up(hw);
4394 netif_carrier_on(netdev);
4396 if (!test_bit(__E1000_DOWN, &adapter->state))
4397 mod_timer(&adapter->phy_info_timer,
4398 round_jiffies(jiffies + 2 * HZ));
4401 if (netif_carrier_ok(netdev)) {
4402 adapter->link_speed = 0;
4403 adapter->link_duplex = 0;
4404 /* Link status message must follow this format */
4405 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4406 adapter->netdev->name);
4407 netif_carrier_off(netdev);
4408 if (!test_bit(__E1000_DOWN, &adapter->state))
4409 mod_timer(&adapter->phy_info_timer,
4410 round_jiffies(jiffies + 2 * HZ));
4412 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4413 schedule_work(&adapter->reset_task);
4415 pm_schedule_suspend(netdev->dev.parent,
4421 spin_lock(&adapter->stats64_lock);
4422 e1000e_update_stats(adapter);
4424 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4425 adapter->tpt_old = adapter->stats.tpt;
4426 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4427 adapter->colc_old = adapter->stats.colc;
4429 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4430 adapter->gorc_old = adapter->stats.gorc;
4431 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4432 adapter->gotc_old = adapter->stats.gotc;
4433 spin_unlock(&adapter->stats64_lock);
4435 e1000e_update_adaptive(&adapter->hw);
4437 if (!netif_carrier_ok(netdev) &&
4438 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4440 * We've lost link, so the controller stops DMA,
4441 * but we've got queued Tx work that's never going
4442 * to get done, so reset controller to flush Tx.
4443 * (Do the reset outside of interrupt context).
4445 schedule_work(&adapter->reset_task);
4446 /* return immediately since reset is imminent */
4450 /* Simple mode for Interrupt Throttle Rate (ITR) */
4451 if (adapter->itr_setting == 4) {
4453 * Symmetric Tx/Rx gets a reduced ITR=2000;
4454 * Total asymmetrical Tx or Rx gets ITR=8000;
4455 * everyone else is between 2000-8000.
4457 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4458 u32 dif = (adapter->gotc > adapter->gorc ?
4459 adapter->gotc - adapter->gorc :
4460 adapter->gorc - adapter->gotc) / 10000;
4461 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4463 ew32(ITR, 1000000000 / (itr * 256));
4466 /* Cause software interrupt to ensure Rx ring is cleaned */
4467 if (adapter->msix_entries)
4468 ew32(ICS, adapter->rx_ring->ims_val);
4470 ew32(ICS, E1000_ICS_RXDMT0);
4472 /* flush pending descriptors to memory before detecting Tx hang */
4473 e1000e_flush_descriptors(adapter);
4475 /* Force detection of hung controller every watchdog period */
4476 adapter->detect_tx_hung = 1;
4479 * With 82571 controllers, LAA may be overwritten due to controller
4480 * reset from the other port. Set the appropriate LAA in RAR[0]
4482 if (e1000e_get_laa_state_82571(hw))
4483 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4485 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4486 e1000e_check_82574_phy_workaround(adapter);
4488 /* Reset the timer */
4489 if (!test_bit(__E1000_DOWN, &adapter->state))
4490 mod_timer(&adapter->watchdog_timer,
4491 round_jiffies(jiffies + 2 * HZ));
4494 #define E1000_TX_FLAGS_CSUM 0x00000001
4495 #define E1000_TX_FLAGS_VLAN 0x00000002
4496 #define E1000_TX_FLAGS_TSO 0x00000004
4497 #define E1000_TX_FLAGS_IPV4 0x00000008
4498 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4499 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4501 static int e1000_tso(struct e1000_adapter *adapter,
4502 struct sk_buff *skb)
4504 struct e1000_ring *tx_ring = adapter->tx_ring;
4505 struct e1000_context_desc *context_desc;
4506 struct e1000_buffer *buffer_info;
4509 u16 ipcse = 0, tucse, mss;
4510 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4512 if (!skb_is_gso(skb))
4515 if (skb_header_cloned(skb)) {
4516 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4522 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4523 mss = skb_shinfo(skb)->gso_size;
4524 if (skb->protocol == htons(ETH_P_IP)) {
4525 struct iphdr *iph = ip_hdr(skb);
4528 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4530 cmd_length = E1000_TXD_CMD_IP;
4531 ipcse = skb_transport_offset(skb) - 1;
4532 } else if (skb_is_gso_v6(skb)) {
4533 ipv6_hdr(skb)->payload_len = 0;
4534 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4535 &ipv6_hdr(skb)->daddr,
4539 ipcss = skb_network_offset(skb);
4540 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4541 tucss = skb_transport_offset(skb);
4542 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4545 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4546 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4548 i = tx_ring->next_to_use;
4549 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4550 buffer_info = &tx_ring->buffer_info[i];
4552 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4553 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4554 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4555 context_desc->upper_setup.tcp_fields.tucss = tucss;
4556 context_desc->upper_setup.tcp_fields.tucso = tucso;
4557 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4558 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4559 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4560 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4562 buffer_info->time_stamp = jiffies;
4563 buffer_info->next_to_watch = i;
4566 if (i == tx_ring->count)
4568 tx_ring->next_to_use = i;
4573 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4575 struct e1000_ring *tx_ring = adapter->tx_ring;
4576 struct e1000_context_desc *context_desc;
4577 struct e1000_buffer *buffer_info;
4580 u32 cmd_len = E1000_TXD_CMD_DEXT;
4583 if (skb->ip_summed != CHECKSUM_PARTIAL)
4586 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4587 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4589 protocol = skb->protocol;
4592 case cpu_to_be16(ETH_P_IP):
4593 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4594 cmd_len |= E1000_TXD_CMD_TCP;
4596 case cpu_to_be16(ETH_P_IPV6):
4597 /* XXX not handling all IPV6 headers */
4598 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4599 cmd_len |= E1000_TXD_CMD_TCP;
4602 if (unlikely(net_ratelimit()))
4603 e_warn("checksum_partial proto=%x!\n",
4604 be16_to_cpu(protocol));
4608 css = skb_checksum_start_offset(skb);
4610 i = tx_ring->next_to_use;
4611 buffer_info = &tx_ring->buffer_info[i];
4612 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4614 context_desc->lower_setup.ip_config = 0;
4615 context_desc->upper_setup.tcp_fields.tucss = css;
4616 context_desc->upper_setup.tcp_fields.tucso =
4617 css + skb->csum_offset;
4618 context_desc->upper_setup.tcp_fields.tucse = 0;
4619 context_desc->tcp_seg_setup.data = 0;
4620 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4622 buffer_info->time_stamp = jiffies;
4623 buffer_info->next_to_watch = i;
4626 if (i == tx_ring->count)
4628 tx_ring->next_to_use = i;
4633 #define E1000_MAX_PER_TXD 8192
4634 #define E1000_MAX_TXD_PWR 12
4636 static int e1000_tx_map(struct e1000_adapter *adapter,
4637 struct sk_buff *skb, unsigned int first,
4638 unsigned int max_per_txd, unsigned int nr_frags,
4641 struct e1000_ring *tx_ring = adapter->tx_ring;
4642 struct pci_dev *pdev = adapter->pdev;
4643 struct e1000_buffer *buffer_info;
4644 unsigned int len = skb_headlen(skb);
4645 unsigned int offset = 0, size, count = 0, i;
4646 unsigned int f, bytecount, segs;
4648 i = tx_ring->next_to_use;
4651 buffer_info = &tx_ring->buffer_info[i];
4652 size = min(len, max_per_txd);
4654 buffer_info->length = size;
4655 buffer_info->time_stamp = jiffies;
4656 buffer_info->next_to_watch = i;
4657 buffer_info->dma = dma_map_single(&pdev->dev,
4659 size, DMA_TO_DEVICE);
4660 buffer_info->mapped_as_page = false;
4661 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4670 if (i == tx_ring->count)
4675 for (f = 0; f < nr_frags; f++) {
4676 struct skb_frag_struct *frag;
4678 frag = &skb_shinfo(skb)->frags[f];
4680 offset = frag->page_offset;
4684 if (i == tx_ring->count)
4687 buffer_info = &tx_ring->buffer_info[i];
4688 size = min(len, max_per_txd);
4690 buffer_info->length = size;
4691 buffer_info->time_stamp = jiffies;
4692 buffer_info->next_to_watch = i;
4693 buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4696 buffer_info->mapped_as_page = true;
4697 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4706 segs = skb_shinfo(skb)->gso_segs ? : 1;
4707 /* multiply data chunks by size of headers */
4708 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4710 tx_ring->buffer_info[i].skb = skb;
4711 tx_ring->buffer_info[i].segs = segs;
4712 tx_ring->buffer_info[i].bytecount = bytecount;
4713 tx_ring->buffer_info[first].next_to_watch = i;
4718 dev_err(&pdev->dev, "Tx DMA map failed\n");
4719 buffer_info->dma = 0;
4725 i += tx_ring->count;
4727 buffer_info = &tx_ring->buffer_info[i];
4728 e1000_put_txbuf(adapter, buffer_info);
4734 static void e1000_tx_queue(struct e1000_adapter *adapter,
4735 int tx_flags, int count)
4737 struct e1000_ring *tx_ring = adapter->tx_ring;
4738 struct e1000_tx_desc *tx_desc = NULL;
4739 struct e1000_buffer *buffer_info;
4740 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4743 if (tx_flags & E1000_TX_FLAGS_TSO) {
4744 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4746 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4748 if (tx_flags & E1000_TX_FLAGS_IPV4)
4749 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4752 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4753 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4754 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4757 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4758 txd_lower |= E1000_TXD_CMD_VLE;
4759 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4762 i = tx_ring->next_to_use;
4765 buffer_info = &tx_ring->buffer_info[i];
4766 tx_desc = E1000_TX_DESC(*tx_ring, i);
4767 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4768 tx_desc->lower.data =
4769 cpu_to_le32(txd_lower | buffer_info->length);
4770 tx_desc->upper.data = cpu_to_le32(txd_upper);
4773 if (i == tx_ring->count)
4775 } while (--count > 0);
4777 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4780 * Force memory writes to complete before letting h/w
4781 * know there are new descriptors to fetch. (Only
4782 * applicable for weak-ordered memory model archs,
4787 tx_ring->next_to_use = i;
4789 if (adapter->flags2 & FLAG2_PCIM2PCI_ARBITER_WA)
4790 e1000e_update_tdt_wa(adapter, i);
4792 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4795 * we need this if more than one processor can write to our tail
4796 * at a time, it synchronizes IO on IA64/Altix systems
4801 #define MINIMUM_DHCP_PACKET_SIZE 282
4802 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4803 struct sk_buff *skb)
4805 struct e1000_hw *hw = &adapter->hw;
4808 if (vlan_tx_tag_present(skb)) {
4809 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4810 (adapter->hw.mng_cookie.status &
4811 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4815 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4818 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4822 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4825 if (ip->protocol != IPPROTO_UDP)
4828 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4829 if (ntohs(udp->dest) != 67)
4832 offset = (u8 *)udp + 8 - skb->data;
4833 length = skb->len - offset;
4834 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4840 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4842 struct e1000_adapter *adapter = netdev_priv(netdev);
4844 netif_stop_queue(netdev);
4846 * Herbert's original patch had:
4847 * smp_mb__after_netif_stop_queue();
4848 * but since that doesn't exist yet, just open code it.
4853 * We need to check again in a case another CPU has just
4854 * made room available.
4856 if (e1000_desc_unused(adapter->tx_ring) < size)
4860 netif_start_queue(netdev);
4861 ++adapter->restart_queue;
4865 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4867 struct e1000_adapter *adapter = netdev_priv(netdev);
4869 if (e1000_desc_unused(adapter->tx_ring) >= size)
4871 return __e1000_maybe_stop_tx(netdev, size);
4874 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4875 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4876 struct net_device *netdev)
4878 struct e1000_adapter *adapter = netdev_priv(netdev);
4879 struct e1000_ring *tx_ring = adapter->tx_ring;
4881 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4882 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4883 unsigned int tx_flags = 0;
4884 unsigned int len = skb_headlen(skb);
4885 unsigned int nr_frags;
4891 if (test_bit(__E1000_DOWN, &adapter->state)) {
4892 dev_kfree_skb_any(skb);
4893 return NETDEV_TX_OK;
4896 if (skb->len <= 0) {
4897 dev_kfree_skb_any(skb);
4898 return NETDEV_TX_OK;
4901 mss = skb_shinfo(skb)->gso_size;
4903 * The controller does a simple calculation to
4904 * make sure there is enough room in the FIFO before
4905 * initiating the DMA for each buffer. The calc is:
4906 * 4 = ceil(buffer len/mss). To make sure we don't
4907 * overrun the FIFO, adjust the max buffer len if mss
4912 max_per_txd = min(mss << 2, max_per_txd);
4913 max_txd_pwr = fls(max_per_txd) - 1;
4916 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4917 * points to just header, pull a few bytes of payload from
4918 * frags into skb->data
4920 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4922 * we do this workaround for ES2LAN, but it is un-necessary,
4923 * avoiding it could save a lot of cycles
4925 if (skb->data_len && (hdr_len == len)) {
4926 unsigned int pull_size;
4928 pull_size = min((unsigned int)4, skb->data_len);
4929 if (!__pskb_pull_tail(skb, pull_size)) {
4930 e_err("__pskb_pull_tail failed.\n");
4931 dev_kfree_skb_any(skb);
4932 return NETDEV_TX_OK;
4934 len = skb_headlen(skb);
4938 /* reserve a descriptor for the offload context */
4939 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4943 count += TXD_USE_COUNT(len, max_txd_pwr);
4945 nr_frags = skb_shinfo(skb)->nr_frags;
4946 for (f = 0; f < nr_frags; f++)
4947 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4950 if (adapter->hw.mac.tx_pkt_filtering)
4951 e1000_transfer_dhcp_info(adapter, skb);
4954 * need: count + 2 desc gap to keep tail from touching
4955 * head, otherwise try next time
4957 if (e1000_maybe_stop_tx(netdev, count + 2))
4958 return NETDEV_TX_BUSY;
4960 if (vlan_tx_tag_present(skb)) {
4961 tx_flags |= E1000_TX_FLAGS_VLAN;
4962 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4965 first = tx_ring->next_to_use;
4967 tso = e1000_tso(adapter, skb);
4969 dev_kfree_skb_any(skb);
4970 return NETDEV_TX_OK;
4974 tx_flags |= E1000_TX_FLAGS_TSO;
4975 else if (e1000_tx_csum(adapter, skb))
4976 tx_flags |= E1000_TX_FLAGS_CSUM;
4979 * Old method was to assume IPv4 packet by default if TSO was enabled.
4980 * 82571 hardware supports TSO capabilities for IPv6 as well...
4981 * no longer assume, we must.
4983 if (skb->protocol == htons(ETH_P_IP))
4984 tx_flags |= E1000_TX_FLAGS_IPV4;
4986 /* if count is 0 then mapping error has occurred */
4987 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4989 e1000_tx_queue(adapter, tx_flags, count);
4990 /* Make sure there is space in the ring for the next send. */
4991 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4994 dev_kfree_skb_any(skb);
4995 tx_ring->buffer_info[first].time_stamp = 0;
4996 tx_ring->next_to_use = first;
4999 return NETDEV_TX_OK;
5003 * e1000_tx_timeout - Respond to a Tx Hang
5004 * @netdev: network interface device structure
5006 static void e1000_tx_timeout(struct net_device *netdev)
5008 struct e1000_adapter *adapter = netdev_priv(netdev);
5010 /* Do the reset outside of interrupt context */
5011 adapter->tx_timeout_count++;
5012 schedule_work(&adapter->reset_task);
5015 static void e1000_reset_task(struct work_struct *work)
5017 struct e1000_adapter *adapter;
5018 adapter = container_of(work, struct e1000_adapter, reset_task);
5020 /* don't run the task if already down */
5021 if (test_bit(__E1000_DOWN, &adapter->state))
5024 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
5025 (adapter->flags & FLAG_RX_RESTART_NOW))) {
5026 e1000e_dump(adapter);
5027 e_err("Reset adapter\n");
5029 e1000e_reinit_locked(adapter);
5033 * e1000_get_stats64 - Get System Network Statistics
5034 * @netdev: network interface device structure
5035 * @stats: rtnl_link_stats64 pointer
5037 * Returns the address of the device statistics structure.
5039 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
5040 struct rtnl_link_stats64 *stats)
5042 struct e1000_adapter *adapter = netdev_priv(netdev);
5044 memset(stats, 0, sizeof(struct rtnl_link_stats64));
5045 spin_lock(&adapter->stats64_lock);
5046 e1000e_update_stats(adapter);
5047 /* Fill out the OS statistics structure */
5048 stats->rx_bytes = adapter->stats.gorc;
5049 stats->rx_packets = adapter->stats.gprc;
5050 stats->tx_bytes = adapter->stats.gotc;
5051 stats->tx_packets = adapter->stats.gptc;
5052 stats->multicast = adapter->stats.mprc;
5053 stats->collisions = adapter->stats.colc;
5058 * RLEC on some newer hardware can be incorrect so build
5059 * our own version based on RUC and ROC
5061 stats->rx_errors = adapter->stats.rxerrc +
5062 adapter->stats.crcerrs + adapter->stats.algnerrc +
5063 adapter->stats.ruc + adapter->stats.roc +
5064 adapter->stats.cexterr;
5065 stats->rx_length_errors = adapter->stats.ruc +
5067 stats->rx_crc_errors = adapter->stats.crcerrs;
5068 stats->rx_frame_errors = adapter->stats.algnerrc;
5069 stats->rx_missed_errors = adapter->stats.mpc;
5072 stats->tx_errors = adapter->stats.ecol +
5073 adapter->stats.latecol;
5074 stats->tx_aborted_errors = adapter->stats.ecol;
5075 stats->tx_window_errors = adapter->stats.latecol;
5076 stats->tx_carrier_errors = adapter->stats.tncrs;
5078 /* Tx Dropped needs to be maintained elsewhere */
5080 spin_unlock(&adapter->stats64_lock);
5085 * e1000_change_mtu - Change the Maximum Transfer Unit
5086 * @netdev: network interface device structure
5087 * @new_mtu: new value for maximum frame size
5089 * Returns 0 on success, negative on failure
5091 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5093 struct e1000_adapter *adapter = netdev_priv(netdev);
5094 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5096 /* Jumbo frame support */
5097 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5098 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5099 e_err("Jumbo Frames not supported.\n");
5103 /* Supported frame sizes */
5104 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5105 (max_frame > adapter->max_hw_frame_size)) {
5106 e_err("Unsupported MTU setting\n");
5110 /* Jumbo frame workaround on 82579 requires CRC be stripped */
5111 if ((adapter->hw.mac.type == e1000_pch2lan) &&
5112 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5113 (new_mtu > ETH_DATA_LEN)) {
5114 e_err("Jumbo Frames not supported on 82579 when CRC "
5115 "stripping is disabled.\n");
5119 /* 82573 Errata 17 */
5120 if (((adapter->hw.mac.type == e1000_82573) ||
5121 (adapter->hw.mac.type == e1000_82574)) &&
5122 (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5123 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5124 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5127 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5128 usleep_range(1000, 2000);
5129 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5130 adapter->max_frame_size = max_frame;
5131 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5132 netdev->mtu = new_mtu;
5133 if (netif_running(netdev))
5134 e1000e_down(adapter);
5137 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5138 * means we reserve 2 more, this pushes us to allocate from the next
5140 * i.e. RXBUFFER_2048 --> size-4096 slab
5141 * However with the new *_jumbo_rx* routines, jumbo receives will use
5145 if (max_frame <= 2048)
5146 adapter->rx_buffer_len = 2048;
5148 adapter->rx_buffer_len = 4096;
5150 /* adjust allocation if LPE protects us, and we aren't using SBP */
5151 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5152 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5153 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5156 if (netif_running(netdev))
5159 e1000e_reset(adapter);
5161 clear_bit(__E1000_RESETTING, &adapter->state);
5166 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5169 struct e1000_adapter *adapter = netdev_priv(netdev);
5170 struct mii_ioctl_data *data = if_mii(ifr);
5172 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5177 data->phy_id = adapter->hw.phy.addr;
5180 e1000_phy_read_status(adapter);
5182 switch (data->reg_num & 0x1F) {
5184 data->val_out = adapter->phy_regs.bmcr;
5187 data->val_out = adapter->phy_regs.bmsr;
5190 data->val_out = (adapter->hw.phy.id >> 16);
5193 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5196 data->val_out = adapter->phy_regs.advertise;
5199 data->val_out = adapter->phy_regs.lpa;
5202 data->val_out = adapter->phy_regs.expansion;
5205 data->val_out = adapter->phy_regs.ctrl1000;
5208 data->val_out = adapter->phy_regs.stat1000;
5211 data->val_out = adapter->phy_regs.estatus;
5224 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5230 return e1000_mii_ioctl(netdev, ifr, cmd);
5236 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5238 struct e1000_hw *hw = &adapter->hw;
5240 u16 phy_reg, wuc_enable;
5243 /* copy MAC RARs to PHY RARs */
5244 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5246 retval = hw->phy.ops.acquire(hw);
5248 e_err("Could not acquire PHY\n");
5252 /* Enable access to wakeup registers on and set page to BM_WUC_PAGE */
5253 retval = e1000_enable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5257 /* copy MAC MTA to PHY MTA - only needed for pchlan */
5258 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5259 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5260 hw->phy.ops.write_reg_page(hw, BM_MTA(i),
5261 (u16)(mac_reg & 0xFFFF));
5262 hw->phy.ops.write_reg_page(hw, BM_MTA(i) + 1,
5263 (u16)((mac_reg >> 16) & 0xFFFF));
5266 /* configure PHY Rx Control register */
5267 hw->phy.ops.read_reg_page(&adapter->hw, BM_RCTL, &phy_reg);
5268 mac_reg = er32(RCTL);
5269 if (mac_reg & E1000_RCTL_UPE)
5270 phy_reg |= BM_RCTL_UPE;
5271 if (mac_reg & E1000_RCTL_MPE)
5272 phy_reg |= BM_RCTL_MPE;
5273 phy_reg &= ~(BM_RCTL_MO_MASK);
5274 if (mac_reg & E1000_RCTL_MO_3)
5275 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5276 << BM_RCTL_MO_SHIFT);
5277 if (mac_reg & E1000_RCTL_BAM)
5278 phy_reg |= BM_RCTL_BAM;
5279 if (mac_reg & E1000_RCTL_PMCF)
5280 phy_reg |= BM_RCTL_PMCF;
5281 mac_reg = er32(CTRL);
5282 if (mac_reg & E1000_CTRL_RFCE)
5283 phy_reg |= BM_RCTL_RFCE;
5284 hw->phy.ops.write_reg_page(&adapter->hw, BM_RCTL, phy_reg);
5286 /* enable PHY wakeup in MAC register */
5288 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5290 /* configure and enable PHY wakeup in PHY registers */
5291 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUFC, wufc);
5292 hw->phy.ops.write_reg_page(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5294 /* activate PHY wakeup */
5295 wuc_enable |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5296 retval = e1000_disable_phy_wakeup_reg_access_bm(hw, &wuc_enable);
5298 e_err("Could not set PHY Host Wakeup bit\n");
5300 hw->phy.ops.release(hw);
5305 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5308 struct net_device *netdev = pci_get_drvdata(pdev);
5309 struct e1000_adapter *adapter = netdev_priv(netdev);
5310 struct e1000_hw *hw = &adapter->hw;
5311 u32 ctrl, ctrl_ext, rctl, status;
5312 /* Runtime suspend should only enable wakeup for link changes */
5313 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5316 netif_device_detach(netdev);
5318 if (netif_running(netdev)) {
5319 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5320 e1000e_down(adapter);
5321 e1000_free_irq(adapter);
5323 e1000e_reset_interrupt_capability(adapter);
5325 retval = pci_save_state(pdev);
5329 status = er32(STATUS);
5330 if (status & E1000_STATUS_LU)
5331 wufc &= ~E1000_WUFC_LNKC;
5334 e1000_setup_rctl(adapter);
5335 e1000_set_multi(netdev);
5337 /* turn on all-multi mode if wake on multicast is enabled */
5338 if (wufc & E1000_WUFC_MC) {
5340 rctl |= E1000_RCTL_MPE;
5345 /* advertise wake from D3Cold */
5346 #define E1000_CTRL_ADVD3WUC 0x00100000
5347 /* phy power management enable */
5348 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5349 ctrl |= E1000_CTRL_ADVD3WUC;
5350 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5351 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5354 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5355 adapter->hw.phy.media_type ==
5356 e1000_media_type_internal_serdes) {
5357 /* keep the laser running in D3 */
5358 ctrl_ext = er32(CTRL_EXT);
5359 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5360 ew32(CTRL_EXT, ctrl_ext);
5363 if (adapter->flags & FLAG_IS_ICH)
5364 e1000_suspend_workarounds_ich8lan(&adapter->hw);
5366 /* Allow time for pending master requests to run */
5367 e1000e_disable_pcie_master(&adapter->hw);
5369 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5370 /* enable wakeup by the PHY */
5371 retval = e1000_init_phy_wakeup(adapter, wufc);
5375 /* enable wakeup by the MAC */
5377 ew32(WUC, E1000_WUC_PME_EN);
5384 *enable_wake = !!wufc;
5386 /* make sure adapter isn't asleep if manageability is enabled */
5387 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5388 (hw->mac.ops.check_mng_mode(hw)))
5389 *enable_wake = true;
5391 if (adapter->hw.phy.type == e1000_phy_igp_3)
5392 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5395 * Release control of h/w to f/w. If f/w is AMT enabled, this
5396 * would have already happened in close and is redundant.
5398 e1000e_release_hw_control(adapter);
5400 pci_disable_device(pdev);
5405 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5407 if (sleep && wake) {
5408 pci_prepare_to_sleep(pdev);
5412 pci_wake_from_d3(pdev, wake);
5413 pci_set_power_state(pdev, PCI_D3hot);
5416 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5419 struct net_device *netdev = pci_get_drvdata(pdev);
5420 struct e1000_adapter *adapter = netdev_priv(netdev);
5423 * The pci-e switch on some quad port adapters will report a
5424 * correctable error when the MAC transitions from D0 to D3. To
5425 * prevent this we need to mask off the correctable errors on the
5426 * downstream port of the pci-e switch.
5428 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5429 struct pci_dev *us_dev = pdev->bus->self;
5430 int pos = pci_pcie_cap(us_dev);
5433 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5434 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5435 (devctl & ~PCI_EXP_DEVCTL_CERE));
5437 e1000_power_off(pdev, sleep, wake);
5439 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5441 e1000_power_off(pdev, sleep, wake);
5445 #ifdef CONFIG_PCIEASPM
5446 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5448 pci_disable_link_state_locked(pdev, state);
5451 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5457 * Both device and parent should have the same ASPM setting.
5458 * Disable ASPM in downstream component first and then upstream.
5460 pos = pci_pcie_cap(pdev);
5461 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, ®16);
5463 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5465 if (!pdev->bus->self)
5468 pos = pci_pcie_cap(pdev->bus->self);
5469 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, ®16);
5471 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5474 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5476 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5477 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5478 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5480 __e1000e_disable_aspm(pdev, state);
5484 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5486 return !!adapter->tx_ring->buffer_info;
5489 static int __e1000_resume(struct pci_dev *pdev)
5491 struct net_device *netdev = pci_get_drvdata(pdev);
5492 struct e1000_adapter *adapter = netdev_priv(netdev);
5493 struct e1000_hw *hw = &adapter->hw;
5494 u16 aspm_disable_flag = 0;
5497 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5498 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5499 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5500 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5501 if (aspm_disable_flag)
5502 e1000e_disable_aspm(pdev, aspm_disable_flag);
5504 pci_set_power_state(pdev, PCI_D0);
5505 pci_restore_state(pdev);
5506 pci_save_state(pdev);
5508 e1000e_set_interrupt_capability(adapter);
5509 if (netif_running(netdev)) {
5510 err = e1000_request_irq(adapter);
5515 if (hw->mac.type == e1000_pch2lan)
5516 e1000_resume_workarounds_pchlan(&adapter->hw);
5518 e1000e_power_up_phy(adapter);
5520 /* report the system wakeup cause from S3/S4 */
5521 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5524 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5526 e_info("PHY Wakeup cause - %s\n",
5527 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5528 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5529 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5530 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5531 phy_data & E1000_WUS_LNKC ? "Link Status "
5532 " Change" : "other");
5534 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5536 u32 wus = er32(WUS);
5538 e_info("MAC Wakeup cause - %s\n",
5539 wus & E1000_WUS_EX ? "Unicast Packet" :
5540 wus & E1000_WUS_MC ? "Multicast Packet" :
5541 wus & E1000_WUS_BC ? "Broadcast Packet" :
5542 wus & E1000_WUS_MAG ? "Magic Packet" :
5543 wus & E1000_WUS_LNKC ? "Link Status Change" :
5549 e1000e_reset(adapter);
5551 e1000_init_manageability_pt(adapter);
5553 if (netif_running(netdev))
5556 netif_device_attach(netdev);
5559 * If the controller has AMT, do not set DRV_LOAD until the interface
5560 * is up. For all other cases, let the f/w know that the h/w is now
5561 * under the control of the driver.
5563 if (!(adapter->flags & FLAG_HAS_AMT))
5564 e1000e_get_hw_control(adapter);
5569 #ifdef CONFIG_PM_SLEEP
5570 static int e1000_suspend(struct device *dev)
5572 struct pci_dev *pdev = to_pci_dev(dev);
5576 retval = __e1000_shutdown(pdev, &wake, false);
5578 e1000_complete_shutdown(pdev, true, wake);
5583 static int e1000_resume(struct device *dev)
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);
5589 if (e1000e_pm_ready(adapter))
5590 adapter->idle_check = true;
5592 return __e1000_resume(pdev);
5594 #endif /* CONFIG_PM_SLEEP */
5596 #ifdef CONFIG_PM_RUNTIME
5597 static int e1000_runtime_suspend(struct device *dev)
5599 struct pci_dev *pdev = to_pci_dev(dev);
5600 struct net_device *netdev = pci_get_drvdata(pdev);
5601 struct e1000_adapter *adapter = netdev_priv(netdev);
5603 if (e1000e_pm_ready(adapter)) {
5606 __e1000_shutdown(pdev, &wake, true);
5612 static int e1000_idle(struct device *dev)
5614 struct pci_dev *pdev = to_pci_dev(dev);
5615 struct net_device *netdev = pci_get_drvdata(pdev);
5616 struct e1000_adapter *adapter = netdev_priv(netdev);
5618 if (!e1000e_pm_ready(adapter))
5621 if (adapter->idle_check) {
5622 adapter->idle_check = false;
5623 if (!e1000e_has_link(adapter))
5624 pm_schedule_suspend(dev, MSEC_PER_SEC);
5630 static int e1000_runtime_resume(struct device *dev)
5632 struct pci_dev *pdev = to_pci_dev(dev);
5633 struct net_device *netdev = pci_get_drvdata(pdev);
5634 struct e1000_adapter *adapter = netdev_priv(netdev);
5636 if (!e1000e_pm_ready(adapter))
5639 adapter->idle_check = !dev->power.runtime_auto;
5640 return __e1000_resume(pdev);
5642 #endif /* CONFIG_PM_RUNTIME */
5643 #endif /* CONFIG_PM */
5645 static void e1000_shutdown(struct pci_dev *pdev)
5649 __e1000_shutdown(pdev, &wake, false);
5651 if (system_state == SYSTEM_POWER_OFF)
5652 e1000_complete_shutdown(pdev, false, wake);
5655 #ifdef CONFIG_NET_POLL_CONTROLLER
5657 static irqreturn_t e1000_intr_msix(int irq, void *data)
5659 struct net_device *netdev = data;
5660 struct e1000_adapter *adapter = netdev_priv(netdev);
5662 if (adapter->msix_entries) {
5663 int vector, msix_irq;
5666 msix_irq = adapter->msix_entries[vector].vector;
5667 disable_irq(msix_irq);
5668 e1000_intr_msix_rx(msix_irq, netdev);
5669 enable_irq(msix_irq);
5672 msix_irq = adapter->msix_entries[vector].vector;
5673 disable_irq(msix_irq);
5674 e1000_intr_msix_tx(msix_irq, netdev);
5675 enable_irq(msix_irq);
5678 msix_irq = adapter->msix_entries[vector].vector;
5679 disable_irq(msix_irq);
5680 e1000_msix_other(msix_irq, netdev);
5681 enable_irq(msix_irq);
5688 * Polling 'interrupt' - used by things like netconsole to send skbs
5689 * without having to re-enable interrupts. It's not called while
5690 * the interrupt routine is executing.
5692 static void e1000_netpoll(struct net_device *netdev)
5694 struct e1000_adapter *adapter = netdev_priv(netdev);
5696 switch (adapter->int_mode) {
5697 case E1000E_INT_MODE_MSIX:
5698 e1000_intr_msix(adapter->pdev->irq, netdev);
5700 case E1000E_INT_MODE_MSI:
5701 disable_irq(adapter->pdev->irq);
5702 e1000_intr_msi(adapter->pdev->irq, netdev);
5703 enable_irq(adapter->pdev->irq);
5705 default: /* E1000E_INT_MODE_LEGACY */
5706 disable_irq(adapter->pdev->irq);
5707 e1000_intr(adapter->pdev->irq, netdev);
5708 enable_irq(adapter->pdev->irq);
5715 * e1000_io_error_detected - called when PCI error is detected
5716 * @pdev: Pointer to PCI device
5717 * @state: The current pci connection state
5719 * This function is called after a PCI bus error affecting
5720 * this device has been detected.
5722 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5723 pci_channel_state_t state)
5725 struct net_device *netdev = pci_get_drvdata(pdev);
5726 struct e1000_adapter *adapter = netdev_priv(netdev);
5728 netif_device_detach(netdev);
5730 if (state == pci_channel_io_perm_failure)
5731 return PCI_ERS_RESULT_DISCONNECT;
5733 if (netif_running(netdev))
5734 e1000e_down(adapter);
5735 pci_disable_device(pdev);
5737 /* Request a slot slot reset. */
5738 return PCI_ERS_RESULT_NEED_RESET;
5742 * e1000_io_slot_reset - called after the pci bus has been reset.
5743 * @pdev: Pointer to PCI device
5745 * Restart the card from scratch, as if from a cold-boot. Implementation
5746 * resembles the first-half of the e1000_resume routine.
5748 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5750 struct net_device *netdev = pci_get_drvdata(pdev);
5751 struct e1000_adapter *adapter = netdev_priv(netdev);
5752 struct e1000_hw *hw = &adapter->hw;
5753 u16 aspm_disable_flag = 0;
5755 pci_ers_result_t result;
5757 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5758 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5759 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5760 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5761 if (aspm_disable_flag)
5762 e1000e_disable_aspm(pdev, aspm_disable_flag);
5764 err = pci_enable_device_mem(pdev);
5767 "Cannot re-enable PCI device after reset.\n");
5768 result = PCI_ERS_RESULT_DISCONNECT;
5770 pci_set_master(pdev);
5771 pdev->state_saved = true;
5772 pci_restore_state(pdev);
5774 pci_enable_wake(pdev, PCI_D3hot, 0);
5775 pci_enable_wake(pdev, PCI_D3cold, 0);
5777 e1000e_reset(adapter);
5779 result = PCI_ERS_RESULT_RECOVERED;
5782 pci_cleanup_aer_uncorrect_error_status(pdev);
5788 * e1000_io_resume - called when traffic can start flowing again.
5789 * @pdev: Pointer to PCI device
5791 * This callback is called when the error recovery driver tells us that
5792 * its OK to resume normal operation. Implementation resembles the
5793 * second-half of the e1000_resume routine.
5795 static void e1000_io_resume(struct pci_dev *pdev)
5797 struct net_device *netdev = pci_get_drvdata(pdev);
5798 struct e1000_adapter *adapter = netdev_priv(netdev);
5800 e1000_init_manageability_pt(adapter);
5802 if (netif_running(netdev)) {
5803 if (e1000e_up(adapter)) {
5805 "can't bring device back up after reset\n");
5810 netif_device_attach(netdev);
5813 * If the controller has AMT, do not set DRV_LOAD until the interface
5814 * is up. For all other cases, let the f/w know that the h/w is now
5815 * under the control of the driver.
5817 if (!(adapter->flags & FLAG_HAS_AMT))
5818 e1000e_get_hw_control(adapter);
5822 static void e1000_print_device_info(struct e1000_adapter *adapter)
5824 struct e1000_hw *hw = &adapter->hw;
5825 struct net_device *netdev = adapter->netdev;
5827 u8 pba_str[E1000_PBANUM_LENGTH];
5829 /* print bus type/speed/width info */
5830 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5832 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5836 e_info("Intel(R) PRO/%s Network Connection\n",
5837 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5838 ret_val = e1000_read_pba_string_generic(hw, pba_str,
5839 E1000_PBANUM_LENGTH);
5841 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5842 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5843 hw->mac.type, hw->phy.type, pba_str);
5846 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5848 struct e1000_hw *hw = &adapter->hw;
5852 if (hw->mac.type != e1000_82573)
5855 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5856 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5857 /* Deep Smart Power Down (DSPD) */
5858 dev_warn(&adapter->pdev->dev,
5859 "Warning: detected DSPD enabled in EEPROM\n");
5863 static const struct net_device_ops e1000e_netdev_ops = {
5864 .ndo_open = e1000_open,
5865 .ndo_stop = e1000_close,
5866 .ndo_start_xmit = e1000_xmit_frame,
5867 .ndo_get_stats64 = e1000e_get_stats64,
5868 .ndo_set_rx_mode = e1000_set_multi,
5869 .ndo_set_mac_address = e1000_set_mac,
5870 .ndo_change_mtu = e1000_change_mtu,
5871 .ndo_do_ioctl = e1000_ioctl,
5872 .ndo_tx_timeout = e1000_tx_timeout,
5873 .ndo_validate_addr = eth_validate_addr,
5875 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5876 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5877 #ifdef CONFIG_NET_POLL_CONTROLLER
5878 .ndo_poll_controller = e1000_netpoll,
5883 * e1000_probe - Device Initialization Routine
5884 * @pdev: PCI device information struct
5885 * @ent: entry in e1000_pci_tbl
5887 * Returns 0 on success, negative on failure
5889 * e1000_probe initializes an adapter identified by a pci_dev structure.
5890 * The OS initialization, configuring of the adapter private structure,
5891 * and a hardware reset occur.
5893 static int __devinit e1000_probe(struct pci_dev *pdev,
5894 const struct pci_device_id *ent)
5896 struct net_device *netdev;
5897 struct e1000_adapter *adapter;
5898 struct e1000_hw *hw;
5899 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5900 resource_size_t mmio_start, mmio_len;
5901 resource_size_t flash_start, flash_len;
5903 static int cards_found;
5904 u16 aspm_disable_flag = 0;
5905 int i, err, pci_using_dac;
5906 u16 eeprom_data = 0;
5907 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5909 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5910 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5911 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5912 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5913 if (aspm_disable_flag)
5914 e1000e_disable_aspm(pdev, aspm_disable_flag);
5916 err = pci_enable_device_mem(pdev);
5921 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5923 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5927 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5929 err = dma_set_coherent_mask(&pdev->dev,
5932 dev_err(&pdev->dev, "No usable DMA "
5933 "configuration, aborting\n");
5939 err = pci_request_selected_regions_exclusive(pdev,
5940 pci_select_bars(pdev, IORESOURCE_MEM),
5941 e1000e_driver_name);
5945 /* AER (Advanced Error Reporting) hooks */
5946 pci_enable_pcie_error_reporting(pdev);
5948 pci_set_master(pdev);
5949 /* PCI config space info */
5950 err = pci_save_state(pdev);
5952 goto err_alloc_etherdev;
5955 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5957 goto err_alloc_etherdev;
5959 SET_NETDEV_DEV(netdev, &pdev->dev);
5961 netdev->irq = pdev->irq;
5963 pci_set_drvdata(pdev, netdev);
5964 adapter = netdev_priv(netdev);
5966 adapter->netdev = netdev;
5967 adapter->pdev = pdev;
5969 adapter->pba = ei->pba;
5970 adapter->flags = ei->flags;
5971 adapter->flags2 = ei->flags2;
5972 adapter->hw.adapter = adapter;
5973 adapter->hw.mac.type = ei->mac;
5974 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5975 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5977 mmio_start = pci_resource_start(pdev, 0);
5978 mmio_len = pci_resource_len(pdev, 0);
5981 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5982 if (!adapter->hw.hw_addr)
5985 if ((adapter->flags & FLAG_HAS_FLASH) &&
5986 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5987 flash_start = pci_resource_start(pdev, 1);
5988 flash_len = pci_resource_len(pdev, 1);
5989 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5990 if (!adapter->hw.flash_address)
5994 /* construct the net_device struct */
5995 netdev->netdev_ops = &e1000e_netdev_ops;
5996 e1000e_set_ethtool_ops(netdev);
5997 netdev->watchdog_timeo = 5 * HZ;
5998 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5999 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
6001 netdev->mem_start = mmio_start;
6002 netdev->mem_end = mmio_start + mmio_len;
6004 adapter->bd_number = cards_found++;
6006 e1000e_check_options(adapter);
6008 /* setup adapter struct */
6009 err = e1000_sw_init(adapter);
6013 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6014 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
6015 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6017 err = ei->get_variants(adapter);
6021 if ((adapter->flags & FLAG_IS_ICH) &&
6022 (adapter->flags & FLAG_READ_ONLY_NVM))
6023 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
6025 hw->mac.ops.get_bus_info(&adapter->hw);
6027 adapter->hw.phy.autoneg_wait_to_complete = 0;
6029 /* Copper options */
6030 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
6031 adapter->hw.phy.mdix = AUTO_ALL_MODES;
6032 adapter->hw.phy.disable_polarity_correction = 0;
6033 adapter->hw.phy.ms_type = e1000_ms_hw_default;
6036 if (e1000_check_reset_block(&adapter->hw))
6037 e_info("PHY reset is blocked due to SOL/IDER session.\n");
6039 netdev->features = NETIF_F_SG |
6041 NETIF_F_HW_VLAN_TX |
6044 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
6045 netdev->features |= NETIF_F_HW_VLAN_FILTER;
6047 netdev->features |= NETIF_F_TSO;
6048 netdev->features |= NETIF_F_TSO6;
6050 netdev->vlan_features |= NETIF_F_TSO;
6051 netdev->vlan_features |= NETIF_F_TSO6;
6052 netdev->vlan_features |= NETIF_F_HW_CSUM;
6053 netdev->vlan_features |= NETIF_F_SG;
6055 if (pci_using_dac) {
6056 netdev->features |= NETIF_F_HIGHDMA;
6057 netdev->vlan_features |= NETIF_F_HIGHDMA;
6060 if (e1000e_enable_mng_pass_thru(&adapter->hw))
6061 adapter->flags |= FLAG_MNG_PT_ENABLED;
6064 * before reading the NVM, reset the controller to
6065 * put the device in a known good starting state
6067 adapter->hw.mac.ops.reset_hw(&adapter->hw);
6070 * systems with ASPM and others may see the checksum fail on the first
6071 * attempt. Let's give it a few tries
6074 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
6077 e_err("The NVM Checksum Is Not Valid\n");
6083 e1000_eeprom_checks(adapter);
6085 /* copy the MAC address */
6086 if (e1000e_read_mac_addr(&adapter->hw))
6087 e_err("NVM Read Error while reading MAC address\n");
6089 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6090 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6092 if (!is_valid_ether_addr(netdev->perm_addr)) {
6093 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6098 init_timer(&adapter->watchdog_timer);
6099 adapter->watchdog_timer.function = e1000_watchdog;
6100 adapter->watchdog_timer.data = (unsigned long) adapter;
6102 init_timer(&adapter->phy_info_timer);
6103 adapter->phy_info_timer.function = e1000_update_phy_info;
6104 adapter->phy_info_timer.data = (unsigned long) adapter;
6106 INIT_WORK(&adapter->reset_task, e1000_reset_task);
6107 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6108 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6109 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6110 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6112 /* Initialize link parameters. User can change them with ethtool */
6113 adapter->hw.mac.autoneg = 1;
6114 adapter->fc_autoneg = 1;
6115 adapter->hw.fc.requested_mode = e1000_fc_default;
6116 adapter->hw.fc.current_mode = e1000_fc_default;
6117 adapter->hw.phy.autoneg_advertised = 0x2f;
6119 /* ring size defaults */
6120 adapter->rx_ring->count = 256;
6121 adapter->tx_ring->count = 256;
6124 * Initial Wake on LAN setting - If APM wake is enabled in
6125 * the EEPROM, enable the ACPI Magic Packet filter
6127 if (adapter->flags & FLAG_APME_IN_WUC) {
6128 /* APME bit in EEPROM is mapped to WUC.APME */
6129 eeprom_data = er32(WUC);
6130 eeprom_apme_mask = E1000_WUC_APME;
6131 if ((hw->mac.type > e1000_ich10lan) &&
6132 (eeprom_data & E1000_WUC_PHY_WAKE))
6133 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6134 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6135 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6136 (adapter->hw.bus.func == 1))
6137 e1000_read_nvm(&adapter->hw,
6138 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6140 e1000_read_nvm(&adapter->hw,
6141 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6144 /* fetch WoL from EEPROM */
6145 if (eeprom_data & eeprom_apme_mask)
6146 adapter->eeprom_wol |= E1000_WUFC_MAG;
6149 * now that we have the eeprom settings, apply the special cases
6150 * where the eeprom may be wrong or the board simply won't support
6151 * wake on lan on a particular port
6153 if (!(adapter->flags & FLAG_HAS_WOL))
6154 adapter->eeprom_wol = 0;
6156 /* initialize the wol settings based on the eeprom settings */
6157 adapter->wol = adapter->eeprom_wol;
6158 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6160 /* save off EEPROM version number */
6161 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6163 /* reset the hardware with the new settings */
6164 e1000e_reset(adapter);
6167 * If the controller has AMT, do not set DRV_LOAD until the interface
6168 * is up. For all other cases, let the f/w know that the h/w is now
6169 * under the control of the driver.
6171 if (!(adapter->flags & FLAG_HAS_AMT))
6172 e1000e_get_hw_control(adapter);
6174 strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6175 err = register_netdev(netdev);
6179 /* carrier off reporting is important to ethtool even BEFORE open */
6180 netif_carrier_off(netdev);
6182 e1000_print_device_info(adapter);
6184 if (pci_dev_run_wake(pdev))
6185 pm_runtime_put_noidle(&pdev->dev);
6190 if (!(adapter->flags & FLAG_HAS_AMT))
6191 e1000e_release_hw_control(adapter);
6193 if (!e1000_check_reset_block(&adapter->hw))
6194 e1000_phy_hw_reset(&adapter->hw);
6196 kfree(adapter->tx_ring);
6197 kfree(adapter->rx_ring);
6199 if (adapter->hw.flash_address)
6200 iounmap(adapter->hw.flash_address);
6201 e1000e_reset_interrupt_capability(adapter);
6203 iounmap(adapter->hw.hw_addr);
6205 free_netdev(netdev);
6207 pci_release_selected_regions(pdev,
6208 pci_select_bars(pdev, IORESOURCE_MEM));
6211 pci_disable_device(pdev);
6216 * e1000_remove - Device Removal Routine
6217 * @pdev: PCI device information struct
6219 * e1000_remove is called by the PCI subsystem to alert the driver
6220 * that it should release a PCI device. The could be caused by a
6221 * Hot-Plug event, or because the driver is going to be removed from
6224 static void __devexit e1000_remove(struct pci_dev *pdev)
6226 struct net_device *netdev = pci_get_drvdata(pdev);
6227 struct e1000_adapter *adapter = netdev_priv(netdev);
6228 bool down = test_bit(__E1000_DOWN, &adapter->state);
6231 * The timers may be rescheduled, so explicitly disable them
6232 * from being rescheduled.
6235 set_bit(__E1000_DOWN, &adapter->state);
6236 del_timer_sync(&adapter->watchdog_timer);
6237 del_timer_sync(&adapter->phy_info_timer);
6239 cancel_work_sync(&adapter->reset_task);
6240 cancel_work_sync(&adapter->watchdog_task);
6241 cancel_work_sync(&adapter->downshift_task);
6242 cancel_work_sync(&adapter->update_phy_task);
6243 cancel_work_sync(&adapter->print_hang_task);
6245 if (!(netdev->flags & IFF_UP))
6246 e1000_power_down_phy(adapter);
6248 /* Don't lie to e1000_close() down the road. */
6250 clear_bit(__E1000_DOWN, &adapter->state);
6251 unregister_netdev(netdev);
6253 if (pci_dev_run_wake(pdev))
6254 pm_runtime_get_noresume(&pdev->dev);
6257 * Release control of h/w to f/w. If f/w is AMT enabled, this
6258 * would have already happened in close and is redundant.
6260 e1000e_release_hw_control(adapter);
6262 e1000e_reset_interrupt_capability(adapter);
6263 kfree(adapter->tx_ring);
6264 kfree(adapter->rx_ring);
6266 iounmap(adapter->hw.hw_addr);
6267 if (adapter->hw.flash_address)
6268 iounmap(adapter->hw.flash_address);
6269 pci_release_selected_regions(pdev,
6270 pci_select_bars(pdev, IORESOURCE_MEM));
6272 free_netdev(netdev);
6275 pci_disable_pcie_error_reporting(pdev);
6277 pci_disable_device(pdev);
6280 /* PCI Error Recovery (ERS) */
6281 static struct pci_error_handlers e1000_err_handler = {
6282 .error_detected = e1000_io_error_detected,
6283 .slot_reset = e1000_io_slot_reset,
6284 .resume = e1000_io_resume,
6287 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6288 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6289 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6290 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6291 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6292 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6293 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6294 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6295 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6296 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6298 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6299 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6300 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6301 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6303 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6304 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6305 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6307 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6308 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6309 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6311 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6312 board_80003es2lan },
6313 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6314 board_80003es2lan },
6315 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6316 board_80003es2lan },
6317 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6318 board_80003es2lan },
6320 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6321 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6322 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6323 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6324 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6325 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6326 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6327 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6329 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6330 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6331 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6332 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6333 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6334 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6335 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6336 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6337 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6339 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6340 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6341 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6343 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6344 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6345 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6347 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6348 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6349 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6350 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6352 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6353 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6355 { } /* terminate list */
6357 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6360 static const struct dev_pm_ops e1000_pm_ops = {
6361 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6362 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6363 e1000_runtime_resume, e1000_idle)
6367 /* PCI Device API Driver */
6368 static struct pci_driver e1000_driver = {
6369 .name = e1000e_driver_name,
6370 .id_table = e1000_pci_tbl,
6371 .probe = e1000_probe,
6372 .remove = __devexit_p(e1000_remove),
6374 .driver.pm = &e1000_pm_ops,
6376 .shutdown = e1000_shutdown,
6377 .err_handler = &e1000_err_handler
6381 * e1000_init_module - Driver Registration Routine
6383 * e1000_init_module is the first routine called when the driver is
6384 * loaded. All it does is register with the PCI subsystem.
6386 static int __init e1000_init_module(void)
6389 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6390 e1000e_driver_version);
6391 pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6392 ret = pci_register_driver(&e1000_driver);
6396 module_init(e1000_init_module);
6399 * e1000_exit_module - Driver Exit Cleanup Routine
6401 * e1000_exit_module is called just before the driver is removed
6404 static void __exit e1000_exit_module(void)
6406 pci_unregister_driver(&e1000_driver);
6408 module_exit(e1000_exit_module);
6411 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6412 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6413 MODULE_LICENSE("GPL");
6414 MODULE_VERSION(DRV_VERSION);