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/tcp.h>
40 #include <linux/ipv6.h>
41 #include <linux/slab.h>
42 #include <net/checksum.h>
43 #include <net/ip6_checksum.h>
44 #include <linux/mii.h>
45 #include <linux/ethtool.h>
46 #include <linux/if_vlan.h>
47 #include <linux/cpu.h>
48 #include <linux/smp.h>
49 #include <linux/pm_qos_params.h>
50 #include <linux/pm_runtime.h>
51 #include <linux/aer.h>
52 #include <linux/prefetch.h>
56 #define DRV_EXTRAVERSION "-k2"
58 #define DRV_VERSION "1.3.10" DRV_EXTRAVERSION
59 char e1000e_driver_name[] = "e1000e";
60 const char e1000e_driver_version[] = DRV_VERSION;
62 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state);
64 static const struct e1000_info *e1000_info_tbl[] = {
65 [board_82571] = &e1000_82571_info,
66 [board_82572] = &e1000_82572_info,
67 [board_82573] = &e1000_82573_info,
68 [board_82574] = &e1000_82574_info,
69 [board_82583] = &e1000_82583_info,
70 [board_80003es2lan] = &e1000_es2_info,
71 [board_ich8lan] = &e1000_ich8_info,
72 [board_ich9lan] = &e1000_ich9_info,
73 [board_ich10lan] = &e1000_ich10_info,
74 [board_pchlan] = &e1000_pch_info,
75 [board_pch2lan] = &e1000_pch2_info,
78 struct e1000_reg_info {
83 #define E1000_RDFH 0x02410 /* Rx Data FIFO Head - RW */
84 #define E1000_RDFT 0x02418 /* Rx Data FIFO Tail - RW */
85 #define E1000_RDFHS 0x02420 /* Rx Data FIFO Head Saved - RW */
86 #define E1000_RDFTS 0x02428 /* Rx Data FIFO Tail Saved - RW */
87 #define E1000_RDFPC 0x02430 /* Rx Data FIFO Packet Count - RW */
89 #define E1000_TDFH 0x03410 /* Tx Data FIFO Head - RW */
90 #define E1000_TDFT 0x03418 /* Tx Data FIFO Tail - RW */
91 #define E1000_TDFHS 0x03420 /* Tx Data FIFO Head Saved - RW */
92 #define E1000_TDFTS 0x03428 /* Tx Data FIFO Tail Saved - RW */
93 #define E1000_TDFPC 0x03430 /* Tx Data FIFO Packet Count - RW */
95 static const struct e1000_reg_info e1000_reg_info_tbl[] = {
97 /* General Registers */
99 {E1000_STATUS, "STATUS"},
100 {E1000_CTRL_EXT, "CTRL_EXT"},
102 /* Interrupt Registers */
106 {E1000_RCTL, "RCTL"},
107 {E1000_RDLEN, "RDLEN"},
110 {E1000_RDTR, "RDTR"},
111 {E1000_RXDCTL(0), "RXDCTL"},
113 {E1000_RDBAL, "RDBAL"},
114 {E1000_RDBAH, "RDBAH"},
115 {E1000_RDFH, "RDFH"},
116 {E1000_RDFT, "RDFT"},
117 {E1000_RDFHS, "RDFHS"},
118 {E1000_RDFTS, "RDFTS"},
119 {E1000_RDFPC, "RDFPC"},
122 {E1000_TCTL, "TCTL"},
123 {E1000_TDBAL, "TDBAL"},
124 {E1000_TDBAH, "TDBAH"},
125 {E1000_TDLEN, "TDLEN"},
128 {E1000_TIDV, "TIDV"},
129 {E1000_TXDCTL(0), "TXDCTL"},
130 {E1000_TADV, "TADV"},
131 {E1000_TARC(0), "TARC"},
132 {E1000_TDFH, "TDFH"},
133 {E1000_TDFT, "TDFT"},
134 {E1000_TDFHS, "TDFHS"},
135 {E1000_TDFTS, "TDFTS"},
136 {E1000_TDFPC, "TDFPC"},
138 /* List Terminator */
143 * e1000_regdump - register printout routine
145 static void e1000_regdump(struct e1000_hw *hw, struct e1000_reg_info *reginfo)
151 switch (reginfo->ofs) {
152 case E1000_RXDCTL(0):
153 for (n = 0; n < 2; n++)
154 regs[n] = __er32(hw, E1000_RXDCTL(n));
156 case E1000_TXDCTL(0):
157 for (n = 0; n < 2; n++)
158 regs[n] = __er32(hw, E1000_TXDCTL(n));
161 for (n = 0; n < 2; n++)
162 regs[n] = __er32(hw, E1000_TARC(n));
165 printk(KERN_INFO "%-15s %08x\n",
166 reginfo->name, __er32(hw, reginfo->ofs));
170 snprintf(rname, 16, "%s%s", reginfo->name, "[0-1]");
171 printk(KERN_INFO "%-15s ", rname);
172 for (n = 0; n < 2; n++)
173 printk(KERN_CONT "%08x ", regs[n]);
174 printk(KERN_CONT "\n");
178 * e1000e_dump - Print registers, Tx-ring and Rx-ring
180 static void e1000e_dump(struct e1000_adapter *adapter)
182 struct net_device *netdev = adapter->netdev;
183 struct e1000_hw *hw = &adapter->hw;
184 struct e1000_reg_info *reginfo;
185 struct e1000_ring *tx_ring = adapter->tx_ring;
186 struct e1000_tx_desc *tx_desc;
191 struct e1000_buffer *buffer_info;
192 struct e1000_ring *rx_ring = adapter->rx_ring;
193 union e1000_rx_desc_packet_split *rx_desc_ps;
194 struct e1000_rx_desc *rx_desc;
204 if (!netif_msg_hw(adapter))
207 /* Print netdevice Info */
209 dev_info(&adapter->pdev->dev, "Net device Info\n");
210 printk(KERN_INFO "Device Name state "
211 "trans_start last_rx\n");
212 printk(KERN_INFO "%-15s %016lX %016lX %016lX\n",
213 netdev->name, netdev->state, netdev->trans_start,
217 /* Print Registers */
218 dev_info(&adapter->pdev->dev, "Register Dump\n");
219 printk(KERN_INFO " Register Name Value\n");
220 for (reginfo = (struct e1000_reg_info *)e1000_reg_info_tbl;
221 reginfo->name; reginfo++) {
222 e1000_regdump(hw, reginfo);
225 /* Print Tx Ring Summary */
226 if (!netdev || !netif_running(netdev))
229 dev_info(&adapter->pdev->dev, "Tx Ring Summary\n");
230 printk(KERN_INFO "Queue [NTU] [NTC] [bi(ntc)->dma ]"
231 " leng ntw timestamp\n");
232 buffer_info = &tx_ring->buffer_info[tx_ring->next_to_clean];
233 printk(KERN_INFO " %5d %5X %5X %016llX %04X %3X %016llX\n",
234 0, tx_ring->next_to_use, tx_ring->next_to_clean,
235 (unsigned long long)buffer_info->dma,
237 buffer_info->next_to_watch,
238 (unsigned long long)buffer_info->time_stamp);
241 if (!netif_msg_tx_done(adapter))
242 goto rx_ring_summary;
244 dev_info(&adapter->pdev->dev, "Tx Ring Dump\n");
246 /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
248 * Legacy Transmit Descriptor
249 * +--------------------------------------------------------------+
250 * 0 | Buffer Address [63:0] (Reserved on Write Back) |
251 * +--------------------------------------------------------------+
252 * 8 | Special | CSS | Status | CMD | CSO | Length |
253 * +--------------------------------------------------------------+
254 * 63 48 47 36 35 32 31 24 23 16 15 0
256 * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
257 * 63 48 47 40 39 32 31 16 15 8 7 0
258 * +----------------------------------------------------------------+
259 * 0 | TUCSE | TUCS0 | TUCSS | IPCSE | IPCS0 | IPCSS |
260 * +----------------------------------------------------------------+
261 * 8 | MSS | HDRLEN | RSV | STA | TUCMD | DTYP | PAYLEN |
262 * +----------------------------------------------------------------+
263 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
265 * Extended Data Descriptor (DTYP=0x1)
266 * +----------------------------------------------------------------+
267 * 0 | Buffer Address [63:0] |
268 * +----------------------------------------------------------------+
269 * 8 | VLAN tag | POPTS | Rsvd | Status | Command | DTYP | DTALEN |
270 * +----------------------------------------------------------------+
271 * 63 48 47 40 39 36 35 32 31 24 23 20 19 0
273 printk(KERN_INFO "Tl[desc] [address 63:0 ] [SpeCssSCmCsLen]"
274 " [bi->dma ] leng ntw timestamp bi->skb "
275 "<-- Legacy format\n");
276 printk(KERN_INFO "Tc[desc] [Ce CoCsIpceCoS] [MssHlRSCm0Plen]"
277 " [bi->dma ] leng ntw timestamp bi->skb "
278 "<-- Ext Context format\n");
279 printk(KERN_INFO "Td[desc] [address 63:0 ] [VlaPoRSCm1Dlen]"
280 " [bi->dma ] leng ntw timestamp bi->skb "
281 "<-- Ext Data format\n");
282 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
283 tx_desc = E1000_TX_DESC(*tx_ring, i);
284 buffer_info = &tx_ring->buffer_info[i];
285 u0 = (struct my_u0 *)tx_desc;
286 printk(KERN_INFO "T%c[0x%03X] %016llX %016llX %016llX "
287 "%04X %3X %016llX %p",
288 (!(le64_to_cpu(u0->b) & (1 << 29)) ? 'l' :
289 ((le64_to_cpu(u0->b) & (1 << 20)) ? 'd' : 'c')), i,
290 (unsigned long long)le64_to_cpu(u0->a),
291 (unsigned long long)le64_to_cpu(u0->b),
292 (unsigned long long)buffer_info->dma,
293 buffer_info->length, buffer_info->next_to_watch,
294 (unsigned long long)buffer_info->time_stamp,
296 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
297 printk(KERN_CONT " NTC/U\n");
298 else if (i == tx_ring->next_to_use)
299 printk(KERN_CONT " NTU\n");
300 else if (i == tx_ring->next_to_clean)
301 printk(KERN_CONT " NTC\n");
303 printk(KERN_CONT "\n");
305 if (netif_msg_pktdata(adapter) && buffer_info->dma != 0)
306 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS,
307 16, 1, phys_to_virt(buffer_info->dma),
308 buffer_info->length, true);
311 /* Print Rx Ring Summary */
313 dev_info(&adapter->pdev->dev, "Rx Ring Summary\n");
314 printk(KERN_INFO "Queue [NTU] [NTC]\n");
315 printk(KERN_INFO " %5d %5X %5X\n", 0,
316 rx_ring->next_to_use, rx_ring->next_to_clean);
319 if (!netif_msg_rx_status(adapter))
322 dev_info(&adapter->pdev->dev, "Rx Ring Dump\n");
323 switch (adapter->rx_ps_pages) {
327 /* [Extended] Packet Split Receive Descriptor Format
329 * +-----------------------------------------------------+
330 * 0 | Buffer Address 0 [63:0] |
331 * +-----------------------------------------------------+
332 * 8 | Buffer Address 1 [63:0] |
333 * +-----------------------------------------------------+
334 * 16 | Buffer Address 2 [63:0] |
335 * +-----------------------------------------------------+
336 * 24 | Buffer Address 3 [63:0] |
337 * +-----------------------------------------------------+
339 printk(KERN_INFO "R [desc] [buffer 0 63:0 ] "
341 "[buffer 2 63:0 ] [buffer 3 63:0 ] [bi->dma ] "
342 "[bi->skb] <-- Ext Pkt Split format\n");
343 /* [Extended] Receive Descriptor (Write-Back) Format
345 * 63 48 47 32 31 13 12 8 7 4 3 0
346 * +------------------------------------------------------+
347 * 0 | Packet | IP | Rsvd | MRQ | Rsvd | MRQ RSS |
348 * | Checksum | Ident | | Queue | | Type |
349 * +------------------------------------------------------+
350 * 8 | VLAN Tag | Length | Extended Error | Extended Status |
351 * +------------------------------------------------------+
352 * 63 48 47 32 31 20 19 0
354 printk(KERN_INFO "RWB[desc] [ck ipid mrqhsh] "
356 "[ l3 l2 l1 hs] [reserved ] ---------------- "
357 "[bi->skb] <-- Ext Rx Write-Back format\n");
358 for (i = 0; i < rx_ring->count; i++) {
359 buffer_info = &rx_ring->buffer_info[i];
360 rx_desc_ps = E1000_RX_DESC_PS(*rx_ring, i);
361 u1 = (struct my_u1 *)rx_desc_ps;
363 le32_to_cpu(rx_desc_ps->wb.middle.status_error);
364 if (staterr & E1000_RXD_STAT_DD) {
365 /* Descriptor Done */
366 printk(KERN_INFO "RWB[0x%03X] %016llX "
367 "%016llX %016llX %016llX "
368 "---------------- %p", i,
369 (unsigned long long)le64_to_cpu(u1->a),
370 (unsigned long long)le64_to_cpu(u1->b),
371 (unsigned long long)le64_to_cpu(u1->c),
372 (unsigned long long)le64_to_cpu(u1->d),
375 printk(KERN_INFO "R [0x%03X] %016llX "
376 "%016llX %016llX %016llX %016llX %p", i,
377 (unsigned long long)le64_to_cpu(u1->a),
378 (unsigned long long)le64_to_cpu(u1->b),
379 (unsigned long long)le64_to_cpu(u1->c),
380 (unsigned long long)le64_to_cpu(u1->d),
381 (unsigned long long)buffer_info->dma,
384 if (netif_msg_pktdata(adapter))
385 print_hex_dump(KERN_INFO, "",
386 DUMP_PREFIX_ADDRESS, 16, 1,
387 phys_to_virt(buffer_info->dma),
388 adapter->rx_ps_bsize0, true);
391 if (i == rx_ring->next_to_use)
392 printk(KERN_CONT " NTU\n");
393 else if (i == rx_ring->next_to_clean)
394 printk(KERN_CONT " NTC\n");
396 printk(KERN_CONT "\n");
401 /* Legacy Receive Descriptor Format
403 * +-----------------------------------------------------+
404 * | Buffer Address [63:0] |
405 * +-----------------------------------------------------+
406 * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
407 * +-----------------------------------------------------+
408 * 63 48 47 40 39 32 31 16 15 0
410 printk(KERN_INFO "Rl[desc] [address 63:0 ] "
411 "[vl er S cks ln] [bi->dma ] [bi->skb] "
412 "<-- Legacy format\n");
413 for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
414 rx_desc = E1000_RX_DESC(*rx_ring, i);
415 buffer_info = &rx_ring->buffer_info[i];
416 u0 = (struct my_u0 *)rx_desc;
417 printk(KERN_INFO "Rl[0x%03X] %016llX %016llX "
419 (unsigned long long)le64_to_cpu(u0->a),
420 (unsigned long long)le64_to_cpu(u0->b),
421 (unsigned long long)buffer_info->dma,
423 if (i == rx_ring->next_to_use)
424 printk(KERN_CONT " NTU\n");
425 else if (i == rx_ring->next_to_clean)
426 printk(KERN_CONT " NTC\n");
428 printk(KERN_CONT "\n");
430 if (netif_msg_pktdata(adapter))
431 print_hex_dump(KERN_INFO, "",
434 phys_to_virt(buffer_info->dma),
435 adapter->rx_buffer_len, true);
444 * e1000_desc_unused - calculate if we have unused descriptors
446 static int e1000_desc_unused(struct e1000_ring *ring)
448 if (ring->next_to_clean > ring->next_to_use)
449 return ring->next_to_clean - ring->next_to_use - 1;
451 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
455 * e1000_receive_skb - helper function to handle Rx indications
456 * @adapter: board private structure
457 * @status: descriptor status field as written by hardware
458 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
459 * @skb: pointer to sk_buff to be indicated to stack
461 static void e1000_receive_skb(struct e1000_adapter *adapter,
462 struct net_device *netdev, struct sk_buff *skb,
463 u8 status, __le16 vlan)
465 u16 tag = le16_to_cpu(vlan);
466 skb->protocol = eth_type_trans(skb, netdev);
468 if (status & E1000_RXD_STAT_VP)
469 __vlan_hwaccel_put_tag(skb, tag);
471 napi_gro_receive(&adapter->napi, skb);
475 * e1000_rx_checksum - Receive Checksum Offload
476 * @adapter: board private structure
477 * @status_err: receive descriptor status and error fields
478 * @csum: receive descriptor csum field
479 * @sk_buff: socket buffer with received data
481 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
482 u32 csum, struct sk_buff *skb)
484 u16 status = (u16)status_err;
485 u8 errors = (u8)(status_err >> 24);
487 skb_checksum_none_assert(skb);
489 /* Ignore Checksum bit is set */
490 if (status & E1000_RXD_STAT_IXSM)
492 /* TCP/UDP checksum error bit is set */
493 if (errors & E1000_RXD_ERR_TCPE) {
494 /* let the stack verify checksum errors */
495 adapter->hw_csum_err++;
499 /* TCP/UDP Checksum has not been calculated */
500 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
503 /* It must be a TCP or UDP packet with a valid checksum */
504 if (status & E1000_RXD_STAT_TCPCS) {
505 /* TCP checksum is good */
506 skb->ip_summed = CHECKSUM_UNNECESSARY;
509 * IP fragment with UDP payload
510 * Hardware complements the payload checksum, so we undo it
511 * and then put the value in host order for further stack use.
513 __sum16 sum = (__force __sum16)htons(csum);
514 skb->csum = csum_unfold(~sum);
515 skb->ip_summed = CHECKSUM_COMPLETE;
517 adapter->hw_csum_good++;
521 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
522 * @adapter: address of board private structure
524 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
527 struct net_device *netdev = adapter->netdev;
528 struct pci_dev *pdev = adapter->pdev;
529 struct e1000_ring *rx_ring = adapter->rx_ring;
530 struct e1000_rx_desc *rx_desc;
531 struct e1000_buffer *buffer_info;
534 unsigned int bufsz = adapter->rx_buffer_len;
536 i = rx_ring->next_to_use;
537 buffer_info = &rx_ring->buffer_info[i];
539 while (cleaned_count--) {
540 skb = buffer_info->skb;
546 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
548 /* Better luck next round */
549 adapter->alloc_rx_buff_failed++;
553 buffer_info->skb = skb;
555 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
556 adapter->rx_buffer_len,
558 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
559 dev_err(&pdev->dev, "Rx DMA map failed\n");
560 adapter->rx_dma_failed++;
564 rx_desc = E1000_RX_DESC(*rx_ring, i);
565 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
567 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
569 * Force memory writes to complete before letting h/w
570 * know there are new descriptors to fetch. (Only
571 * applicable for weak-ordered memory model archs,
575 writel(i, adapter->hw.hw_addr + rx_ring->tail);
578 if (i == rx_ring->count)
580 buffer_info = &rx_ring->buffer_info[i];
583 rx_ring->next_to_use = i;
587 * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
588 * @adapter: address of board private structure
590 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
593 struct net_device *netdev = adapter->netdev;
594 struct pci_dev *pdev = adapter->pdev;
595 union e1000_rx_desc_packet_split *rx_desc;
596 struct e1000_ring *rx_ring = adapter->rx_ring;
597 struct e1000_buffer *buffer_info;
598 struct e1000_ps_page *ps_page;
602 i = rx_ring->next_to_use;
603 buffer_info = &rx_ring->buffer_info[i];
605 while (cleaned_count--) {
606 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
608 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
609 ps_page = &buffer_info->ps_pages[j];
610 if (j >= adapter->rx_ps_pages) {
611 /* all unused desc entries get hw null ptr */
612 rx_desc->read.buffer_addr[j + 1] =
616 if (!ps_page->page) {
617 ps_page->page = alloc_page(GFP_ATOMIC);
618 if (!ps_page->page) {
619 adapter->alloc_rx_buff_failed++;
622 ps_page->dma = dma_map_page(&pdev->dev,
626 if (dma_mapping_error(&pdev->dev,
628 dev_err(&adapter->pdev->dev,
629 "Rx DMA page map failed\n");
630 adapter->rx_dma_failed++;
635 * Refresh the desc even if buffer_addrs
636 * didn't change because each write-back
639 rx_desc->read.buffer_addr[j + 1] =
640 cpu_to_le64(ps_page->dma);
643 skb = netdev_alloc_skb_ip_align(netdev,
644 adapter->rx_ps_bsize0);
647 adapter->alloc_rx_buff_failed++;
651 buffer_info->skb = skb;
652 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
653 adapter->rx_ps_bsize0,
655 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
656 dev_err(&pdev->dev, "Rx DMA map failed\n");
657 adapter->rx_dma_failed++;
659 dev_kfree_skb_any(skb);
660 buffer_info->skb = NULL;
664 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
666 if (unlikely(!(i & (E1000_RX_BUFFER_WRITE - 1)))) {
668 * Force memory writes to complete before letting h/w
669 * know there are new descriptors to fetch. (Only
670 * applicable for weak-ordered memory model archs,
674 writel(i << 1, adapter->hw.hw_addr + rx_ring->tail);
678 if (i == rx_ring->count)
680 buffer_info = &rx_ring->buffer_info[i];
684 rx_ring->next_to_use = i;
688 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
689 * @adapter: address of board private structure
690 * @cleaned_count: number of buffers to allocate this pass
693 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
696 struct net_device *netdev = adapter->netdev;
697 struct pci_dev *pdev = adapter->pdev;
698 struct e1000_rx_desc *rx_desc;
699 struct e1000_ring *rx_ring = adapter->rx_ring;
700 struct e1000_buffer *buffer_info;
703 unsigned int bufsz = 256 - 16 /* for skb_reserve */;
705 i = rx_ring->next_to_use;
706 buffer_info = &rx_ring->buffer_info[i];
708 while (cleaned_count--) {
709 skb = buffer_info->skb;
715 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
716 if (unlikely(!skb)) {
717 /* Better luck next round */
718 adapter->alloc_rx_buff_failed++;
722 buffer_info->skb = skb;
724 /* allocate a new page if necessary */
725 if (!buffer_info->page) {
726 buffer_info->page = alloc_page(GFP_ATOMIC);
727 if (unlikely(!buffer_info->page)) {
728 adapter->alloc_rx_buff_failed++;
733 if (!buffer_info->dma)
734 buffer_info->dma = dma_map_page(&pdev->dev,
735 buffer_info->page, 0,
739 rx_desc = E1000_RX_DESC(*rx_ring, i);
740 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
742 if (unlikely(++i == rx_ring->count))
744 buffer_info = &rx_ring->buffer_info[i];
747 if (likely(rx_ring->next_to_use != i)) {
748 rx_ring->next_to_use = i;
749 if (unlikely(i-- == 0))
750 i = (rx_ring->count - 1);
752 /* Force memory writes to complete before letting h/w
753 * know there are new descriptors to fetch. (Only
754 * applicable for weak-ordered memory model archs,
757 writel(i, adapter->hw.hw_addr + rx_ring->tail);
762 * e1000_clean_rx_irq - Send received data up the network stack; legacy
763 * @adapter: board private structure
765 * the return value indicates whether actual cleaning was done, there
766 * is no guarantee that everything was cleaned
768 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
769 int *work_done, int work_to_do)
771 struct net_device *netdev = adapter->netdev;
772 struct pci_dev *pdev = adapter->pdev;
773 struct e1000_hw *hw = &adapter->hw;
774 struct e1000_ring *rx_ring = adapter->rx_ring;
775 struct e1000_rx_desc *rx_desc, *next_rxd;
776 struct e1000_buffer *buffer_info, *next_buffer;
779 int cleaned_count = 0;
781 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
783 i = rx_ring->next_to_clean;
784 rx_desc = E1000_RX_DESC(*rx_ring, i);
785 buffer_info = &rx_ring->buffer_info[i];
787 while (rx_desc->status & E1000_RXD_STAT_DD) {
791 if (*work_done >= work_to_do)
794 rmb(); /* read descriptor and rx_buffer_info after status DD */
796 status = rx_desc->status;
797 skb = buffer_info->skb;
798 buffer_info->skb = NULL;
800 prefetch(skb->data - NET_IP_ALIGN);
803 if (i == rx_ring->count)
805 next_rxd = E1000_RX_DESC(*rx_ring, i);
808 next_buffer = &rx_ring->buffer_info[i];
812 dma_unmap_single(&pdev->dev,
814 adapter->rx_buffer_len,
816 buffer_info->dma = 0;
818 length = le16_to_cpu(rx_desc->length);
821 * !EOP means multiple descriptors were used to store a single
822 * packet, if that's the case we need to toss it. In fact, we
823 * need to toss every packet with the EOP bit clear and the
824 * next frame that _does_ have the EOP bit set, as it is by
825 * definition only a frame fragment
827 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
828 adapter->flags2 |= FLAG2_IS_DISCARDING;
830 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
831 /* All receives must fit into a single buffer */
832 e_dbg("Receive packet consumed multiple buffers\n");
834 buffer_info->skb = skb;
835 if (status & E1000_RXD_STAT_EOP)
836 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
840 if (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
842 buffer_info->skb = skb;
846 /* adjust length to remove Ethernet CRC */
847 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
850 total_rx_bytes += length;
854 * code added for copybreak, this should improve
855 * performance for small packets with large amounts
856 * of reassembly being done in the stack
858 if (length < copybreak) {
859 struct sk_buff *new_skb =
860 netdev_alloc_skb_ip_align(netdev, length);
862 skb_copy_to_linear_data_offset(new_skb,
868 /* save the skb in buffer_info as good */
869 buffer_info->skb = skb;
872 /* else just continue with the old one */
874 /* end copybreak code */
875 skb_put(skb, length);
877 /* Receive Checksum Offload */
878 e1000_rx_checksum(adapter,
880 ((u32)(rx_desc->errors) << 24),
881 le16_to_cpu(rx_desc->csum), skb);
883 e1000_receive_skb(adapter, netdev, skb,status,rx_desc->special);
888 /* return some buffers to hardware, one at a time is too slow */
889 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
890 adapter->alloc_rx_buf(adapter, cleaned_count);
894 /* use prefetched values */
896 buffer_info = next_buffer;
898 rx_ring->next_to_clean = i;
900 cleaned_count = e1000_desc_unused(rx_ring);
902 adapter->alloc_rx_buf(adapter, cleaned_count);
904 adapter->total_rx_bytes += total_rx_bytes;
905 adapter->total_rx_packets += total_rx_packets;
909 static void e1000_put_txbuf(struct e1000_adapter *adapter,
910 struct e1000_buffer *buffer_info)
912 if (buffer_info->dma) {
913 if (buffer_info->mapped_as_page)
914 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
915 buffer_info->length, DMA_TO_DEVICE);
917 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
918 buffer_info->length, DMA_TO_DEVICE);
919 buffer_info->dma = 0;
921 if (buffer_info->skb) {
922 dev_kfree_skb_any(buffer_info->skb);
923 buffer_info->skb = NULL;
925 buffer_info->time_stamp = 0;
928 static void e1000_print_hw_hang(struct work_struct *work)
930 struct e1000_adapter *adapter = container_of(work,
931 struct e1000_adapter,
933 struct e1000_ring *tx_ring = adapter->tx_ring;
934 unsigned int i = tx_ring->next_to_clean;
935 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
936 struct e1000_tx_desc *eop_desc = E1000_TX_DESC(*tx_ring, eop);
937 struct e1000_hw *hw = &adapter->hw;
938 u16 phy_status, phy_1000t_status, phy_ext_status;
941 if (test_bit(__E1000_DOWN, &adapter->state))
944 e1e_rphy(hw, PHY_STATUS, &phy_status);
945 e1e_rphy(hw, PHY_1000T_STATUS, &phy_1000t_status);
946 e1e_rphy(hw, PHY_EXT_STATUS, &phy_ext_status);
948 pci_read_config_word(adapter->pdev, PCI_STATUS, &pci_status);
950 /* detected Hardware unit hang */
951 e_err("Detected Hardware Unit Hang:\n"
954 " next_to_use <%x>\n"
955 " next_to_clean <%x>\n"
956 "buffer_info[next_to_clean]:\n"
957 " time_stamp <%lx>\n"
958 " next_to_watch <%x>\n"
960 " next_to_watch.status <%x>\n"
963 "PHY 1000BASE-T Status <%x>\n"
964 "PHY Extended Status <%x>\n"
966 readl(adapter->hw.hw_addr + tx_ring->head),
967 readl(adapter->hw.hw_addr + tx_ring->tail),
968 tx_ring->next_to_use,
969 tx_ring->next_to_clean,
970 tx_ring->buffer_info[eop].time_stamp,
973 eop_desc->upper.fields.status,
982 * e1000_clean_tx_irq - Reclaim resources after transmit completes
983 * @adapter: board private structure
985 * the return value indicates whether actual cleaning was done, there
986 * is no guarantee that everything was cleaned
988 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter)
990 struct net_device *netdev = adapter->netdev;
991 struct e1000_hw *hw = &adapter->hw;
992 struct e1000_ring *tx_ring = adapter->tx_ring;
993 struct e1000_tx_desc *tx_desc, *eop_desc;
994 struct e1000_buffer *buffer_info;
996 unsigned int count = 0;
997 unsigned int total_tx_bytes = 0, total_tx_packets = 0;
999 i = tx_ring->next_to_clean;
1000 eop = tx_ring->buffer_info[i].next_to_watch;
1001 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1003 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
1004 (count < tx_ring->count)) {
1005 bool cleaned = false;
1006 rmb(); /* read buffer_info after eop_desc */
1007 for (; !cleaned; count++) {
1008 tx_desc = E1000_TX_DESC(*tx_ring, i);
1009 buffer_info = &tx_ring->buffer_info[i];
1010 cleaned = (i == eop);
1013 total_tx_packets += buffer_info->segs;
1014 total_tx_bytes += buffer_info->bytecount;
1017 e1000_put_txbuf(adapter, buffer_info);
1018 tx_desc->upper.data = 0;
1021 if (i == tx_ring->count)
1025 if (i == tx_ring->next_to_use)
1027 eop = tx_ring->buffer_info[i].next_to_watch;
1028 eop_desc = E1000_TX_DESC(*tx_ring, eop);
1031 tx_ring->next_to_clean = i;
1033 #define TX_WAKE_THRESHOLD 32
1034 if (count && netif_carrier_ok(netdev) &&
1035 e1000_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD) {
1036 /* Make sure that anybody stopping the queue after this
1037 * sees the new next_to_clean.
1041 if (netif_queue_stopped(netdev) &&
1042 !(test_bit(__E1000_DOWN, &adapter->state))) {
1043 netif_wake_queue(netdev);
1044 ++adapter->restart_queue;
1048 if (adapter->detect_tx_hung) {
1050 * Detect a transmit hang in hardware, this serializes the
1051 * check with the clearing of time_stamp and movement of i
1053 adapter->detect_tx_hung = 0;
1054 if (tx_ring->buffer_info[i].time_stamp &&
1055 time_after(jiffies, tx_ring->buffer_info[i].time_stamp
1056 + (adapter->tx_timeout_factor * HZ)) &&
1057 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
1058 schedule_work(&adapter->print_hang_task);
1059 netif_stop_queue(netdev);
1062 adapter->total_tx_bytes += total_tx_bytes;
1063 adapter->total_tx_packets += total_tx_packets;
1064 return count < tx_ring->count;
1068 * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
1069 * @adapter: board private structure
1071 * the return value indicates whether actual cleaning was done, there
1072 * is no guarantee that everything was cleaned
1074 static bool e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
1075 int *work_done, int work_to_do)
1077 struct e1000_hw *hw = &adapter->hw;
1078 union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
1079 struct net_device *netdev = adapter->netdev;
1080 struct pci_dev *pdev = adapter->pdev;
1081 struct e1000_ring *rx_ring = adapter->rx_ring;
1082 struct e1000_buffer *buffer_info, *next_buffer;
1083 struct e1000_ps_page *ps_page;
1084 struct sk_buff *skb;
1086 u32 length, staterr;
1087 int cleaned_count = 0;
1089 unsigned int total_rx_bytes = 0, total_rx_packets = 0;
1091 i = rx_ring->next_to_clean;
1092 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
1093 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1094 buffer_info = &rx_ring->buffer_info[i];
1096 while (staterr & E1000_RXD_STAT_DD) {
1097 if (*work_done >= work_to_do)
1100 skb = buffer_info->skb;
1101 rmb(); /* read descriptor and rx_buffer_info after status DD */
1103 /* in the packet split case this is header only */
1104 prefetch(skb->data - NET_IP_ALIGN);
1107 if (i == rx_ring->count)
1109 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
1112 next_buffer = &rx_ring->buffer_info[i];
1116 dma_unmap_single(&pdev->dev, buffer_info->dma,
1117 adapter->rx_ps_bsize0, DMA_FROM_DEVICE);
1118 buffer_info->dma = 0;
1120 /* see !EOP comment in other Rx routine */
1121 if (!(staterr & E1000_RXD_STAT_EOP))
1122 adapter->flags2 |= FLAG2_IS_DISCARDING;
1124 if (adapter->flags2 & FLAG2_IS_DISCARDING) {
1125 e_dbg("Packet Split buffers didn't pick up the full "
1127 dev_kfree_skb_irq(skb);
1128 if (staterr & E1000_RXD_STAT_EOP)
1129 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1133 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
1134 dev_kfree_skb_irq(skb);
1138 length = le16_to_cpu(rx_desc->wb.middle.length0);
1141 e_dbg("Last part of the packet spanning multiple "
1143 dev_kfree_skb_irq(skb);
1148 skb_put(skb, length);
1152 * this looks ugly, but it seems compiler issues make it
1153 * more efficient than reusing j
1155 int l1 = le16_to_cpu(rx_desc->wb.upper.length[0]);
1158 * page alloc/put takes too long and effects small packet
1159 * throughput, so unsplit small packets and save the alloc/put
1160 * only valid in softirq (napi) context to call kmap_*
1162 if (l1 && (l1 <= copybreak) &&
1163 ((length + l1) <= adapter->rx_ps_bsize0)) {
1166 ps_page = &buffer_info->ps_pages[0];
1169 * there is no documentation about how to call
1170 * kmap_atomic, so we can't hold the mapping
1173 dma_sync_single_for_cpu(&pdev->dev, ps_page->dma,
1174 PAGE_SIZE, DMA_FROM_DEVICE);
1175 vaddr = kmap_atomic(ps_page->page, KM_SKB_DATA_SOFTIRQ);
1176 memcpy(skb_tail_pointer(skb), vaddr, l1);
1177 kunmap_atomic(vaddr, KM_SKB_DATA_SOFTIRQ);
1178 dma_sync_single_for_device(&pdev->dev, ps_page->dma,
1179 PAGE_SIZE, DMA_FROM_DEVICE);
1181 /* remove the CRC */
1182 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1190 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1191 length = le16_to_cpu(rx_desc->wb.upper.length[j]);
1195 ps_page = &buffer_info->ps_pages[j];
1196 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1199 skb_fill_page_desc(skb, j, ps_page->page, 0, length);
1200 ps_page->page = NULL;
1202 skb->data_len += length;
1203 skb->truesize += length;
1206 /* strip the ethernet crc, problem is we're using pages now so
1207 * this whole operation can get a little cpu intensive
1209 if (!(adapter->flags2 & FLAG2_CRC_STRIPPING))
1210 pskb_trim(skb, skb->len - 4);
1213 total_rx_bytes += skb->len;
1216 e1000_rx_checksum(adapter, staterr, le16_to_cpu(
1217 rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
1219 if (rx_desc->wb.upper.header_status &
1220 cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP))
1221 adapter->rx_hdr_split++;
1223 e1000_receive_skb(adapter, netdev, skb,
1224 staterr, rx_desc->wb.middle.vlan);
1227 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
1228 buffer_info->skb = NULL;
1230 /* return some buffers to hardware, one at a time is too slow */
1231 if (cleaned_count >= E1000_RX_BUFFER_WRITE) {
1232 adapter->alloc_rx_buf(adapter, cleaned_count);
1236 /* use prefetched values */
1238 buffer_info = next_buffer;
1240 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
1242 rx_ring->next_to_clean = i;
1244 cleaned_count = e1000_desc_unused(rx_ring);
1246 adapter->alloc_rx_buf(adapter, cleaned_count);
1248 adapter->total_rx_bytes += total_rx_bytes;
1249 adapter->total_rx_packets += total_rx_packets;
1254 * e1000_consume_page - helper function
1256 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
1261 skb->data_len += length;
1262 skb->truesize += length;
1266 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
1267 * @adapter: board private structure
1269 * the return value indicates whether actual cleaning was done, there
1270 * is no guarantee that everything was cleaned
1273 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
1274 int *work_done, int work_to_do)
1276 struct net_device *netdev = adapter->netdev;
1277 struct pci_dev *pdev = adapter->pdev;
1278 struct e1000_ring *rx_ring = adapter->rx_ring;
1279 struct e1000_rx_desc *rx_desc, *next_rxd;
1280 struct e1000_buffer *buffer_info, *next_buffer;
1283 int cleaned_count = 0;
1284 bool cleaned = false;
1285 unsigned int total_rx_bytes=0, total_rx_packets=0;
1287 i = rx_ring->next_to_clean;
1288 rx_desc = E1000_RX_DESC(*rx_ring, i);
1289 buffer_info = &rx_ring->buffer_info[i];
1291 while (rx_desc->status & E1000_RXD_STAT_DD) {
1292 struct sk_buff *skb;
1295 if (*work_done >= work_to_do)
1298 rmb(); /* read descriptor and rx_buffer_info after status DD */
1300 status = rx_desc->status;
1301 skb = buffer_info->skb;
1302 buffer_info->skb = NULL;
1305 if (i == rx_ring->count)
1307 next_rxd = E1000_RX_DESC(*rx_ring, i);
1310 next_buffer = &rx_ring->buffer_info[i];
1314 dma_unmap_page(&pdev->dev, buffer_info->dma, PAGE_SIZE,
1316 buffer_info->dma = 0;
1318 length = le16_to_cpu(rx_desc->length);
1320 /* errors is only valid for DD + EOP descriptors */
1321 if (unlikely((status & E1000_RXD_STAT_EOP) &&
1322 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
1323 /* recycle both page and skb */
1324 buffer_info->skb = skb;
1325 /* an error means any chain goes out the window
1327 if (rx_ring->rx_skb_top)
1328 dev_kfree_skb_irq(rx_ring->rx_skb_top);
1329 rx_ring->rx_skb_top = NULL;
1333 #define rxtop (rx_ring->rx_skb_top)
1334 if (!(status & E1000_RXD_STAT_EOP)) {
1335 /* this descriptor is only the beginning (or middle) */
1337 /* this is the beginning of a chain */
1339 skb_fill_page_desc(rxtop, 0, buffer_info->page,
1342 /* this is the middle of a chain */
1343 skb_fill_page_desc(rxtop,
1344 skb_shinfo(rxtop)->nr_frags,
1345 buffer_info->page, 0, length);
1346 /* re-use the skb, only consumed the page */
1347 buffer_info->skb = skb;
1349 e1000_consume_page(buffer_info, rxtop, length);
1353 /* end of the chain */
1354 skb_fill_page_desc(rxtop,
1355 skb_shinfo(rxtop)->nr_frags,
1356 buffer_info->page, 0, length);
1357 /* re-use the current skb, we only consumed the
1359 buffer_info->skb = skb;
1362 e1000_consume_page(buffer_info, skb, length);
1364 /* no chain, got EOP, this buf is the packet
1365 * copybreak to save the put_page/alloc_page */
1366 if (length <= copybreak &&
1367 skb_tailroom(skb) >= length) {
1369 vaddr = kmap_atomic(buffer_info->page,
1370 KM_SKB_DATA_SOFTIRQ);
1371 memcpy(skb_tail_pointer(skb), vaddr,
1373 kunmap_atomic(vaddr,
1374 KM_SKB_DATA_SOFTIRQ);
1375 /* re-use the page, so don't erase
1376 * buffer_info->page */
1377 skb_put(skb, length);
1379 skb_fill_page_desc(skb, 0,
1380 buffer_info->page, 0,
1382 e1000_consume_page(buffer_info, skb,
1388 /* Receive Checksum Offload XXX recompute due to CRC strip? */
1389 e1000_rx_checksum(adapter,
1391 ((u32)(rx_desc->errors) << 24),
1392 le16_to_cpu(rx_desc->csum), skb);
1394 /* probably a little skewed due to removing CRC */
1395 total_rx_bytes += skb->len;
1398 /* eth type trans needs skb->data to point to something */
1399 if (!pskb_may_pull(skb, ETH_HLEN)) {
1400 e_err("pskb_may_pull failed.\n");
1401 dev_kfree_skb_irq(skb);
1405 e1000_receive_skb(adapter, netdev, skb, status,
1409 rx_desc->status = 0;
1411 /* return some buffers to hardware, one at a time is too slow */
1412 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
1413 adapter->alloc_rx_buf(adapter, cleaned_count);
1417 /* use prefetched values */
1419 buffer_info = next_buffer;
1421 rx_ring->next_to_clean = i;
1423 cleaned_count = e1000_desc_unused(rx_ring);
1425 adapter->alloc_rx_buf(adapter, cleaned_count);
1427 adapter->total_rx_bytes += total_rx_bytes;
1428 adapter->total_rx_packets += total_rx_packets;
1433 * e1000_clean_rx_ring - Free Rx Buffers per Queue
1434 * @adapter: board private structure
1436 static void e1000_clean_rx_ring(struct e1000_adapter *adapter)
1438 struct e1000_ring *rx_ring = adapter->rx_ring;
1439 struct e1000_buffer *buffer_info;
1440 struct e1000_ps_page *ps_page;
1441 struct pci_dev *pdev = adapter->pdev;
1444 /* Free all the Rx ring sk_buffs */
1445 for (i = 0; i < rx_ring->count; i++) {
1446 buffer_info = &rx_ring->buffer_info[i];
1447 if (buffer_info->dma) {
1448 if (adapter->clean_rx == e1000_clean_rx_irq)
1449 dma_unmap_single(&pdev->dev, buffer_info->dma,
1450 adapter->rx_buffer_len,
1452 else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq)
1453 dma_unmap_page(&pdev->dev, buffer_info->dma,
1456 else if (adapter->clean_rx == e1000_clean_rx_irq_ps)
1457 dma_unmap_single(&pdev->dev, buffer_info->dma,
1458 adapter->rx_ps_bsize0,
1460 buffer_info->dma = 0;
1463 if (buffer_info->page) {
1464 put_page(buffer_info->page);
1465 buffer_info->page = NULL;
1468 if (buffer_info->skb) {
1469 dev_kfree_skb(buffer_info->skb);
1470 buffer_info->skb = NULL;
1473 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
1474 ps_page = &buffer_info->ps_pages[j];
1477 dma_unmap_page(&pdev->dev, ps_page->dma, PAGE_SIZE,
1480 put_page(ps_page->page);
1481 ps_page->page = NULL;
1485 /* there also may be some cached data from a chained receive */
1486 if (rx_ring->rx_skb_top) {
1487 dev_kfree_skb(rx_ring->rx_skb_top);
1488 rx_ring->rx_skb_top = NULL;
1491 /* Zero out the descriptor ring */
1492 memset(rx_ring->desc, 0, rx_ring->size);
1494 rx_ring->next_to_clean = 0;
1495 rx_ring->next_to_use = 0;
1496 adapter->flags2 &= ~FLAG2_IS_DISCARDING;
1498 writel(0, adapter->hw.hw_addr + rx_ring->head);
1499 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1502 static void e1000e_downshift_workaround(struct work_struct *work)
1504 struct e1000_adapter *adapter = container_of(work,
1505 struct e1000_adapter, downshift_task);
1507 if (test_bit(__E1000_DOWN, &adapter->state))
1510 e1000e_gig_downshift_workaround_ich8lan(&adapter->hw);
1514 * e1000_intr_msi - Interrupt Handler
1515 * @irq: interrupt number
1516 * @data: pointer to a network interface device structure
1518 static irqreturn_t e1000_intr_msi(int irq, void *data)
1520 struct net_device *netdev = data;
1521 struct e1000_adapter *adapter = netdev_priv(netdev);
1522 struct e1000_hw *hw = &adapter->hw;
1523 u32 icr = er32(ICR);
1526 * read ICR disables interrupts using IAM
1529 if (icr & E1000_ICR_LSC) {
1530 hw->mac.get_link_status = 1;
1532 * ICH8 workaround-- Call gig speed drop workaround on cable
1533 * disconnect (LSC) before accessing any PHY registers
1535 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1536 (!(er32(STATUS) & E1000_STATUS_LU)))
1537 schedule_work(&adapter->downshift_task);
1540 * 80003ES2LAN workaround-- For packet buffer work-around on
1541 * link down event; disable receives here in the ISR and reset
1542 * adapter in watchdog
1544 if (netif_carrier_ok(netdev) &&
1545 adapter->flags & FLAG_RX_NEEDS_RESTART) {
1546 /* disable receives */
1547 u32 rctl = er32(RCTL);
1548 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1549 adapter->flags |= FLAG_RX_RESTART_NOW;
1551 /* guard against interrupt when we're going down */
1552 if (!test_bit(__E1000_DOWN, &adapter->state))
1553 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1556 if (napi_schedule_prep(&adapter->napi)) {
1557 adapter->total_tx_bytes = 0;
1558 adapter->total_tx_packets = 0;
1559 adapter->total_rx_bytes = 0;
1560 adapter->total_rx_packets = 0;
1561 __napi_schedule(&adapter->napi);
1568 * e1000_intr - Interrupt Handler
1569 * @irq: interrupt number
1570 * @data: pointer to a network interface device structure
1572 static irqreturn_t e1000_intr(int irq, void *data)
1574 struct net_device *netdev = data;
1575 struct e1000_adapter *adapter = netdev_priv(netdev);
1576 struct e1000_hw *hw = &adapter->hw;
1577 u32 rctl, icr = er32(ICR);
1579 if (!icr || test_bit(__E1000_DOWN, &adapter->state))
1580 return IRQ_NONE; /* Not our interrupt */
1583 * IMS will not auto-mask if INT_ASSERTED is not set, and if it is
1584 * not set, then the adapter didn't send an interrupt
1586 if (!(icr & E1000_ICR_INT_ASSERTED))
1590 * Interrupt Auto-Mask...upon reading ICR,
1591 * interrupts are masked. No need for the
1595 if (icr & E1000_ICR_LSC) {
1596 hw->mac.get_link_status = 1;
1598 * ICH8 workaround-- Call gig speed drop workaround on cable
1599 * disconnect (LSC) before accessing any PHY registers
1601 if ((adapter->flags & FLAG_LSC_GIG_SPEED_DROP) &&
1602 (!(er32(STATUS) & E1000_STATUS_LU)))
1603 schedule_work(&adapter->downshift_task);
1606 * 80003ES2LAN workaround--
1607 * For packet buffer work-around on link down event;
1608 * disable receives here in the ISR and
1609 * reset adapter in watchdog
1611 if (netif_carrier_ok(netdev) &&
1612 (adapter->flags & FLAG_RX_NEEDS_RESTART)) {
1613 /* disable receives */
1615 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1616 adapter->flags |= FLAG_RX_RESTART_NOW;
1618 /* guard against interrupt when we're going down */
1619 if (!test_bit(__E1000_DOWN, &adapter->state))
1620 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1623 if (napi_schedule_prep(&adapter->napi)) {
1624 adapter->total_tx_bytes = 0;
1625 adapter->total_tx_packets = 0;
1626 adapter->total_rx_bytes = 0;
1627 adapter->total_rx_packets = 0;
1628 __napi_schedule(&adapter->napi);
1634 static irqreturn_t e1000_msix_other(int irq, void *data)
1636 struct net_device *netdev = data;
1637 struct e1000_adapter *adapter = netdev_priv(netdev);
1638 struct e1000_hw *hw = &adapter->hw;
1639 u32 icr = er32(ICR);
1641 if (!(icr & E1000_ICR_INT_ASSERTED)) {
1642 if (!test_bit(__E1000_DOWN, &adapter->state))
1643 ew32(IMS, E1000_IMS_OTHER);
1647 if (icr & adapter->eiac_mask)
1648 ew32(ICS, (icr & adapter->eiac_mask));
1650 if (icr & E1000_ICR_OTHER) {
1651 if (!(icr & E1000_ICR_LSC))
1652 goto no_link_interrupt;
1653 hw->mac.get_link_status = 1;
1654 /* guard against interrupt when we're going down */
1655 if (!test_bit(__E1000_DOWN, &adapter->state))
1656 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1660 if (!test_bit(__E1000_DOWN, &adapter->state))
1661 ew32(IMS, E1000_IMS_LSC | E1000_IMS_OTHER);
1667 static irqreturn_t e1000_intr_msix_tx(int irq, void *data)
1669 struct net_device *netdev = data;
1670 struct e1000_adapter *adapter = netdev_priv(netdev);
1671 struct e1000_hw *hw = &adapter->hw;
1672 struct e1000_ring *tx_ring = adapter->tx_ring;
1675 adapter->total_tx_bytes = 0;
1676 adapter->total_tx_packets = 0;
1678 if (!e1000_clean_tx_irq(adapter))
1679 /* Ring was not completely cleaned, so fire another interrupt */
1680 ew32(ICS, tx_ring->ims_val);
1685 static irqreturn_t e1000_intr_msix_rx(int irq, void *data)
1687 struct net_device *netdev = data;
1688 struct e1000_adapter *adapter = netdev_priv(netdev);
1690 /* Write the ITR value calculated at the end of the
1691 * previous interrupt.
1693 if (adapter->rx_ring->set_itr) {
1694 writel(1000000000 / (adapter->rx_ring->itr_val * 256),
1695 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
1696 adapter->rx_ring->set_itr = 0;
1699 if (napi_schedule_prep(&adapter->napi)) {
1700 adapter->total_rx_bytes = 0;
1701 adapter->total_rx_packets = 0;
1702 __napi_schedule(&adapter->napi);
1708 * e1000_configure_msix - Configure MSI-X hardware
1710 * e1000_configure_msix sets up the hardware to properly
1711 * generate MSI-X interrupts.
1713 static void e1000_configure_msix(struct e1000_adapter *adapter)
1715 struct e1000_hw *hw = &adapter->hw;
1716 struct e1000_ring *rx_ring = adapter->rx_ring;
1717 struct e1000_ring *tx_ring = adapter->tx_ring;
1719 u32 ctrl_ext, ivar = 0;
1721 adapter->eiac_mask = 0;
1723 /* Workaround issue with spurious interrupts on 82574 in MSI-X mode */
1724 if (hw->mac.type == e1000_82574) {
1725 u32 rfctl = er32(RFCTL);
1726 rfctl |= E1000_RFCTL_ACK_DIS;
1730 #define E1000_IVAR_INT_ALLOC_VALID 0x8
1731 /* Configure Rx vector */
1732 rx_ring->ims_val = E1000_IMS_RXQ0;
1733 adapter->eiac_mask |= rx_ring->ims_val;
1734 if (rx_ring->itr_val)
1735 writel(1000000000 / (rx_ring->itr_val * 256),
1736 hw->hw_addr + rx_ring->itr_register);
1738 writel(1, hw->hw_addr + rx_ring->itr_register);
1739 ivar = E1000_IVAR_INT_ALLOC_VALID | vector;
1741 /* Configure Tx vector */
1742 tx_ring->ims_val = E1000_IMS_TXQ0;
1744 if (tx_ring->itr_val)
1745 writel(1000000000 / (tx_ring->itr_val * 256),
1746 hw->hw_addr + tx_ring->itr_register);
1748 writel(1, hw->hw_addr + tx_ring->itr_register);
1749 adapter->eiac_mask |= tx_ring->ims_val;
1750 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 8);
1752 /* set vector for Other Causes, e.g. link changes */
1754 ivar |= ((E1000_IVAR_INT_ALLOC_VALID | vector) << 16);
1755 if (rx_ring->itr_val)
1756 writel(1000000000 / (rx_ring->itr_val * 256),
1757 hw->hw_addr + E1000_EITR_82574(vector));
1759 writel(1, hw->hw_addr + E1000_EITR_82574(vector));
1761 /* Cause Tx interrupts on every write back */
1766 /* enable MSI-X PBA support */
1767 ctrl_ext = er32(CTRL_EXT);
1768 ctrl_ext |= E1000_CTRL_EXT_PBA_CLR;
1770 /* Auto-Mask Other interrupts upon ICR read */
1771 #define E1000_EIAC_MASK_82574 0x01F00000
1772 ew32(IAM, ~E1000_EIAC_MASK_82574 | E1000_IMS_OTHER);
1773 ctrl_ext |= E1000_CTRL_EXT_EIAME;
1774 ew32(CTRL_EXT, ctrl_ext);
1778 void e1000e_reset_interrupt_capability(struct e1000_adapter *adapter)
1780 if (adapter->msix_entries) {
1781 pci_disable_msix(adapter->pdev);
1782 kfree(adapter->msix_entries);
1783 adapter->msix_entries = NULL;
1784 } else if (adapter->flags & FLAG_MSI_ENABLED) {
1785 pci_disable_msi(adapter->pdev);
1786 adapter->flags &= ~FLAG_MSI_ENABLED;
1791 * e1000e_set_interrupt_capability - set MSI or MSI-X if supported
1793 * Attempt to configure interrupts using the best available
1794 * capabilities of the hardware and kernel.
1796 void e1000e_set_interrupt_capability(struct e1000_adapter *adapter)
1801 switch (adapter->int_mode) {
1802 case E1000E_INT_MODE_MSIX:
1803 if (adapter->flags & FLAG_HAS_MSIX) {
1804 adapter->num_vectors = 3; /* RxQ0, TxQ0 and other */
1805 adapter->msix_entries = kcalloc(adapter->num_vectors,
1806 sizeof(struct msix_entry),
1808 if (adapter->msix_entries) {
1809 for (i = 0; i < adapter->num_vectors; i++)
1810 adapter->msix_entries[i].entry = i;
1812 err = pci_enable_msix(adapter->pdev,
1813 adapter->msix_entries,
1814 adapter->num_vectors);
1818 /* MSI-X failed, so fall through and try MSI */
1819 e_err("Failed to initialize MSI-X interrupts. "
1820 "Falling back to MSI interrupts.\n");
1821 e1000e_reset_interrupt_capability(adapter);
1823 adapter->int_mode = E1000E_INT_MODE_MSI;
1825 case E1000E_INT_MODE_MSI:
1826 if (!pci_enable_msi(adapter->pdev)) {
1827 adapter->flags |= FLAG_MSI_ENABLED;
1829 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1830 e_err("Failed to initialize MSI interrupts. Falling "
1831 "back to legacy interrupts.\n");
1834 case E1000E_INT_MODE_LEGACY:
1835 /* Don't do anything; this is the system default */
1839 /* store the number of vectors being used */
1840 adapter->num_vectors = 1;
1844 * e1000_request_msix - Initialize MSI-X interrupts
1846 * e1000_request_msix allocates MSI-X vectors and requests interrupts from the
1849 static int e1000_request_msix(struct e1000_adapter *adapter)
1851 struct net_device *netdev = adapter->netdev;
1852 int err = 0, vector = 0;
1854 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1855 snprintf(adapter->rx_ring->name,
1856 sizeof(adapter->rx_ring->name) - 1,
1857 "%s-rx-0", netdev->name);
1859 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1860 err = request_irq(adapter->msix_entries[vector].vector,
1861 e1000_intr_msix_rx, 0, adapter->rx_ring->name,
1865 adapter->rx_ring->itr_register = E1000_EITR_82574(vector);
1866 adapter->rx_ring->itr_val = adapter->itr;
1869 if (strlen(netdev->name) < (IFNAMSIZ - 5))
1870 snprintf(adapter->tx_ring->name,
1871 sizeof(adapter->tx_ring->name) - 1,
1872 "%s-tx-0", netdev->name);
1874 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1875 err = request_irq(adapter->msix_entries[vector].vector,
1876 e1000_intr_msix_tx, 0, adapter->tx_ring->name,
1880 adapter->tx_ring->itr_register = E1000_EITR_82574(vector);
1881 adapter->tx_ring->itr_val = adapter->itr;
1884 err = request_irq(adapter->msix_entries[vector].vector,
1885 e1000_msix_other, 0, netdev->name, netdev);
1889 e1000_configure_msix(adapter);
1896 * e1000_request_irq - initialize interrupts
1898 * Attempts to configure interrupts using the best available
1899 * capabilities of the hardware and kernel.
1901 static int e1000_request_irq(struct e1000_adapter *adapter)
1903 struct net_device *netdev = adapter->netdev;
1906 if (adapter->msix_entries) {
1907 err = e1000_request_msix(adapter);
1910 /* fall back to MSI */
1911 e1000e_reset_interrupt_capability(adapter);
1912 adapter->int_mode = E1000E_INT_MODE_MSI;
1913 e1000e_set_interrupt_capability(adapter);
1915 if (adapter->flags & FLAG_MSI_ENABLED) {
1916 err = request_irq(adapter->pdev->irq, e1000_intr_msi, 0,
1917 netdev->name, netdev);
1921 /* fall back to legacy interrupt */
1922 e1000e_reset_interrupt_capability(adapter);
1923 adapter->int_mode = E1000E_INT_MODE_LEGACY;
1926 err = request_irq(adapter->pdev->irq, e1000_intr, IRQF_SHARED,
1927 netdev->name, netdev);
1929 e_err("Unable to allocate interrupt, Error: %d\n", err);
1934 static void e1000_free_irq(struct e1000_adapter *adapter)
1936 struct net_device *netdev = adapter->netdev;
1938 if (adapter->msix_entries) {
1941 free_irq(adapter->msix_entries[vector].vector, netdev);
1944 free_irq(adapter->msix_entries[vector].vector, netdev);
1947 /* Other Causes interrupt vector */
1948 free_irq(adapter->msix_entries[vector].vector, netdev);
1952 free_irq(adapter->pdev->irq, netdev);
1956 * e1000_irq_disable - Mask off interrupt generation on the NIC
1958 static void e1000_irq_disable(struct e1000_adapter *adapter)
1960 struct e1000_hw *hw = &adapter->hw;
1963 if (adapter->msix_entries)
1964 ew32(EIAC_82574, 0);
1967 if (adapter->msix_entries) {
1969 for (i = 0; i < adapter->num_vectors; i++)
1970 synchronize_irq(adapter->msix_entries[i].vector);
1972 synchronize_irq(adapter->pdev->irq);
1977 * e1000_irq_enable - Enable default interrupt generation settings
1979 static void e1000_irq_enable(struct e1000_adapter *adapter)
1981 struct e1000_hw *hw = &adapter->hw;
1983 if (adapter->msix_entries) {
1984 ew32(EIAC_82574, adapter->eiac_mask & E1000_EIAC_MASK_82574);
1985 ew32(IMS, adapter->eiac_mask | E1000_IMS_OTHER | E1000_IMS_LSC);
1987 ew32(IMS, IMS_ENABLE_MASK);
1993 * e1000e_get_hw_control - get control of the h/w from f/w
1994 * @adapter: address of board private structure
1996 * e1000e_get_hw_control sets {CTRL_EXT|SWSM}:DRV_LOAD bit.
1997 * For ASF and Pass Through versions of f/w this means that
1998 * the driver is loaded. For AMT version (only with 82573)
1999 * of the f/w this means that the network i/f is open.
2001 void e1000e_get_hw_control(struct e1000_adapter *adapter)
2003 struct e1000_hw *hw = &adapter->hw;
2007 /* Let firmware know the driver has taken over */
2008 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2010 ew32(SWSM, swsm | E1000_SWSM_DRV_LOAD);
2011 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2012 ctrl_ext = er32(CTRL_EXT);
2013 ew32(CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
2018 * e1000e_release_hw_control - release control of the h/w to f/w
2019 * @adapter: address of board private structure
2021 * e1000e_release_hw_control resets {CTRL_EXT|SWSM}:DRV_LOAD bit.
2022 * For ASF and Pass Through versions of f/w this means that the
2023 * driver is no longer loaded. For AMT version (only with 82573) i
2024 * of the f/w this means that the network i/f is closed.
2027 void e1000e_release_hw_control(struct e1000_adapter *adapter)
2029 struct e1000_hw *hw = &adapter->hw;
2033 /* Let firmware taken over control of h/w */
2034 if (adapter->flags & FLAG_HAS_SWSM_ON_LOAD) {
2036 ew32(SWSM, swsm & ~E1000_SWSM_DRV_LOAD);
2037 } else if (adapter->flags & FLAG_HAS_CTRLEXT_ON_LOAD) {
2038 ctrl_ext = er32(CTRL_EXT);
2039 ew32(CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
2044 * @e1000_alloc_ring - allocate memory for a ring structure
2046 static int e1000_alloc_ring_dma(struct e1000_adapter *adapter,
2047 struct e1000_ring *ring)
2049 struct pci_dev *pdev = adapter->pdev;
2051 ring->desc = dma_alloc_coherent(&pdev->dev, ring->size, &ring->dma,
2060 * e1000e_setup_tx_resources - allocate Tx resources (Descriptors)
2061 * @adapter: board private structure
2063 * Return 0 on success, negative on failure
2065 int e1000e_setup_tx_resources(struct e1000_adapter *adapter)
2067 struct e1000_ring *tx_ring = adapter->tx_ring;
2068 int err = -ENOMEM, size;
2070 size = sizeof(struct e1000_buffer) * tx_ring->count;
2071 tx_ring->buffer_info = vzalloc(size);
2072 if (!tx_ring->buffer_info)
2075 /* round up to nearest 4K */
2076 tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc);
2077 tx_ring->size = ALIGN(tx_ring->size, 4096);
2079 err = e1000_alloc_ring_dma(adapter, tx_ring);
2083 tx_ring->next_to_use = 0;
2084 tx_ring->next_to_clean = 0;
2088 vfree(tx_ring->buffer_info);
2089 e_err("Unable to allocate memory for the transmit descriptor ring\n");
2094 * e1000e_setup_rx_resources - allocate Rx resources (Descriptors)
2095 * @adapter: board private structure
2097 * Returns 0 on success, negative on failure
2099 int e1000e_setup_rx_resources(struct e1000_adapter *adapter)
2101 struct e1000_ring *rx_ring = adapter->rx_ring;
2102 struct e1000_buffer *buffer_info;
2103 int i, size, desc_len, err = -ENOMEM;
2105 size = sizeof(struct e1000_buffer) * rx_ring->count;
2106 rx_ring->buffer_info = vzalloc(size);
2107 if (!rx_ring->buffer_info)
2110 for (i = 0; i < rx_ring->count; i++) {
2111 buffer_info = &rx_ring->buffer_info[i];
2112 buffer_info->ps_pages = kcalloc(PS_PAGE_BUFFERS,
2113 sizeof(struct e1000_ps_page),
2115 if (!buffer_info->ps_pages)
2119 desc_len = sizeof(union e1000_rx_desc_packet_split);
2121 /* Round up to nearest 4K */
2122 rx_ring->size = rx_ring->count * desc_len;
2123 rx_ring->size = ALIGN(rx_ring->size, 4096);
2125 err = e1000_alloc_ring_dma(adapter, rx_ring);
2129 rx_ring->next_to_clean = 0;
2130 rx_ring->next_to_use = 0;
2131 rx_ring->rx_skb_top = NULL;
2136 for (i = 0; i < rx_ring->count; i++) {
2137 buffer_info = &rx_ring->buffer_info[i];
2138 kfree(buffer_info->ps_pages);
2141 vfree(rx_ring->buffer_info);
2142 e_err("Unable to allocate memory for the receive descriptor ring\n");
2147 * e1000_clean_tx_ring - Free Tx Buffers
2148 * @adapter: board private structure
2150 static void e1000_clean_tx_ring(struct e1000_adapter *adapter)
2152 struct e1000_ring *tx_ring = adapter->tx_ring;
2153 struct e1000_buffer *buffer_info;
2157 for (i = 0; i < tx_ring->count; i++) {
2158 buffer_info = &tx_ring->buffer_info[i];
2159 e1000_put_txbuf(adapter, buffer_info);
2162 size = sizeof(struct e1000_buffer) * tx_ring->count;
2163 memset(tx_ring->buffer_info, 0, size);
2165 memset(tx_ring->desc, 0, tx_ring->size);
2167 tx_ring->next_to_use = 0;
2168 tx_ring->next_to_clean = 0;
2170 writel(0, adapter->hw.hw_addr + tx_ring->head);
2171 writel(0, adapter->hw.hw_addr + tx_ring->tail);
2175 * e1000e_free_tx_resources - Free Tx Resources per Queue
2176 * @adapter: board private structure
2178 * Free all transmit software resources
2180 void e1000e_free_tx_resources(struct e1000_adapter *adapter)
2182 struct pci_dev *pdev = adapter->pdev;
2183 struct e1000_ring *tx_ring = adapter->tx_ring;
2185 e1000_clean_tx_ring(adapter);
2187 vfree(tx_ring->buffer_info);
2188 tx_ring->buffer_info = NULL;
2190 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2192 tx_ring->desc = NULL;
2196 * e1000e_free_rx_resources - Free Rx Resources
2197 * @adapter: board private structure
2199 * Free all receive software resources
2202 void e1000e_free_rx_resources(struct e1000_adapter *adapter)
2204 struct pci_dev *pdev = adapter->pdev;
2205 struct e1000_ring *rx_ring = adapter->rx_ring;
2208 e1000_clean_rx_ring(adapter);
2210 for (i = 0; i < rx_ring->count; i++)
2211 kfree(rx_ring->buffer_info[i].ps_pages);
2213 vfree(rx_ring->buffer_info);
2214 rx_ring->buffer_info = NULL;
2216 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2218 rx_ring->desc = NULL;
2222 * e1000_update_itr - update the dynamic ITR value based on statistics
2223 * @adapter: pointer to adapter
2224 * @itr_setting: current adapter->itr
2225 * @packets: the number of packets during this measurement interval
2226 * @bytes: the number of bytes during this measurement interval
2228 * Stores a new ITR value based on packets and byte
2229 * counts during the last interrupt. The advantage of per interrupt
2230 * computation is faster updates and more accurate ITR for the current
2231 * traffic pattern. Constants in this function were computed
2232 * based on theoretical maximum wire speed and thresholds were set based
2233 * on testing data as well as attempting to minimize response time
2234 * while increasing bulk throughput. This functionality is controlled
2235 * by the InterruptThrottleRate module parameter.
2237 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2238 u16 itr_setting, int packets,
2241 unsigned int retval = itr_setting;
2244 goto update_itr_done;
2246 switch (itr_setting) {
2247 case lowest_latency:
2248 /* handle TSO and jumbo frames */
2249 if (bytes/packets > 8000)
2250 retval = bulk_latency;
2251 else if ((packets < 5) && (bytes > 512))
2252 retval = low_latency;
2254 case low_latency: /* 50 usec aka 20000 ints/s */
2255 if (bytes > 10000) {
2256 /* this if handles the TSO accounting */
2257 if (bytes/packets > 8000)
2258 retval = bulk_latency;
2259 else if ((packets < 10) || ((bytes/packets) > 1200))
2260 retval = bulk_latency;
2261 else if ((packets > 35))
2262 retval = lowest_latency;
2263 } else if (bytes/packets > 2000) {
2264 retval = bulk_latency;
2265 } else if (packets <= 2 && bytes < 512) {
2266 retval = lowest_latency;
2269 case bulk_latency: /* 250 usec aka 4000 ints/s */
2270 if (bytes > 25000) {
2272 retval = low_latency;
2273 } else if (bytes < 6000) {
2274 retval = low_latency;
2283 static void e1000_set_itr(struct e1000_adapter *adapter)
2285 struct e1000_hw *hw = &adapter->hw;
2287 u32 new_itr = adapter->itr;
2289 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2290 if (adapter->link_speed != SPEED_1000) {
2296 if (adapter->flags2 & FLAG2_DISABLE_AIM) {
2301 adapter->tx_itr = e1000_update_itr(adapter,
2303 adapter->total_tx_packets,
2304 adapter->total_tx_bytes);
2305 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2306 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2307 adapter->tx_itr = low_latency;
2309 adapter->rx_itr = e1000_update_itr(adapter,
2311 adapter->total_rx_packets,
2312 adapter->total_rx_bytes);
2313 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2314 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2315 adapter->rx_itr = low_latency;
2317 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2319 switch (current_itr) {
2320 /* counts and packets in update_itr are dependent on these numbers */
2321 case lowest_latency:
2325 new_itr = 20000; /* aka hwitr = ~200 */
2335 if (new_itr != adapter->itr) {
2337 * this attempts to bias the interrupt rate towards Bulk
2338 * by adding intermediate steps when interrupt rate is
2341 new_itr = new_itr > adapter->itr ?
2342 min(adapter->itr + (new_itr >> 2), new_itr) :
2344 adapter->itr = new_itr;
2345 adapter->rx_ring->itr_val = new_itr;
2346 if (adapter->msix_entries)
2347 adapter->rx_ring->set_itr = 1;
2350 ew32(ITR, 1000000000 / (new_itr * 256));
2357 * e1000_alloc_queues - Allocate memory for all rings
2358 * @adapter: board private structure to initialize
2360 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
2362 adapter->tx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2363 if (!adapter->tx_ring)
2366 adapter->rx_ring = kzalloc(sizeof(struct e1000_ring), GFP_KERNEL);
2367 if (!adapter->rx_ring)
2372 e_err("Unable to allocate memory for queues\n");
2373 kfree(adapter->rx_ring);
2374 kfree(adapter->tx_ring);
2379 * e1000_clean - NAPI Rx polling callback
2380 * @napi: struct associated with this polling callback
2381 * @budget: amount of packets driver is allowed to process this poll
2383 static int e1000_clean(struct napi_struct *napi, int budget)
2385 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
2386 struct e1000_hw *hw = &adapter->hw;
2387 struct net_device *poll_dev = adapter->netdev;
2388 int tx_cleaned = 1, work_done = 0;
2390 adapter = netdev_priv(poll_dev);
2392 if (adapter->msix_entries &&
2393 !(adapter->rx_ring->ims_val & adapter->tx_ring->ims_val))
2396 tx_cleaned = e1000_clean_tx_irq(adapter);
2399 adapter->clean_rx(adapter, &work_done, budget);
2404 /* If budget not fully consumed, exit the polling mode */
2405 if (work_done < budget) {
2406 if (adapter->itr_setting & 3)
2407 e1000_set_itr(adapter);
2408 napi_complete(napi);
2409 if (!test_bit(__E1000_DOWN, &adapter->state)) {
2410 if (adapter->msix_entries)
2411 ew32(IMS, adapter->rx_ring->ims_val);
2413 e1000_irq_enable(adapter);
2420 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2422 struct e1000_adapter *adapter = netdev_priv(netdev);
2423 struct e1000_hw *hw = &adapter->hw;
2426 /* don't update vlan cookie if already programmed */
2427 if ((adapter->hw.mng_cookie.status &
2428 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2429 (vid == adapter->mng_vlan_id))
2432 /* add VID to filter table */
2433 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2434 index = (vid >> 5) & 0x7F;
2435 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2436 vfta |= (1 << (vid & 0x1F));
2437 hw->mac.ops.write_vfta(hw, index, vfta);
2440 set_bit(vid, adapter->active_vlans);
2443 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2445 struct e1000_adapter *adapter = netdev_priv(netdev);
2446 struct e1000_hw *hw = &adapter->hw;
2449 if ((adapter->hw.mng_cookie.status &
2450 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
2451 (vid == adapter->mng_vlan_id)) {
2452 /* release control to f/w */
2453 e1000e_release_hw_control(adapter);
2457 /* remove VID from filter table */
2458 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2459 index = (vid >> 5) & 0x7F;
2460 vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, index);
2461 vfta &= ~(1 << (vid & 0x1F));
2462 hw->mac.ops.write_vfta(hw, index, vfta);
2465 clear_bit(vid, adapter->active_vlans);
2469 * e1000e_vlan_filter_disable - helper to disable hw VLAN filtering
2470 * @adapter: board private structure to initialize
2472 static void e1000e_vlan_filter_disable(struct e1000_adapter *adapter)
2474 struct net_device *netdev = adapter->netdev;
2475 struct e1000_hw *hw = &adapter->hw;
2478 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2479 /* disable VLAN receive filtering */
2481 rctl &= ~(E1000_RCTL_VFE | E1000_RCTL_CFIEN);
2484 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
2485 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
2486 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
2492 * e1000e_vlan_filter_enable - helper to enable HW VLAN filtering
2493 * @adapter: board private structure to initialize
2495 static void e1000e_vlan_filter_enable(struct e1000_adapter *adapter)
2497 struct e1000_hw *hw = &adapter->hw;
2500 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER) {
2501 /* enable VLAN receive filtering */
2503 rctl |= E1000_RCTL_VFE;
2504 rctl &= ~E1000_RCTL_CFIEN;
2510 * e1000e_vlan_strip_enable - helper to disable HW VLAN stripping
2511 * @adapter: board private structure to initialize
2513 static void e1000e_vlan_strip_disable(struct e1000_adapter *adapter)
2515 struct e1000_hw *hw = &adapter->hw;
2518 /* disable VLAN tag insert/strip */
2520 ctrl &= ~E1000_CTRL_VME;
2525 * e1000e_vlan_strip_enable - helper to enable HW VLAN stripping
2526 * @adapter: board private structure to initialize
2528 static void e1000e_vlan_strip_enable(struct e1000_adapter *adapter)
2530 struct e1000_hw *hw = &adapter->hw;
2533 /* enable VLAN tag insert/strip */
2535 ctrl |= E1000_CTRL_VME;
2539 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
2541 struct net_device *netdev = adapter->netdev;
2542 u16 vid = adapter->hw.mng_cookie.vlan_id;
2543 u16 old_vid = adapter->mng_vlan_id;
2545 if (adapter->hw.mng_cookie.status &
2546 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
2547 e1000_vlan_rx_add_vid(netdev, vid);
2548 adapter->mng_vlan_id = vid;
2551 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) && (vid != old_vid))
2552 e1000_vlan_rx_kill_vid(netdev, old_vid);
2555 static void e1000_restore_vlan(struct e1000_adapter *adapter)
2559 e1000_vlan_rx_add_vid(adapter->netdev, 0);
2561 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2562 e1000_vlan_rx_add_vid(adapter->netdev, vid);
2565 static void e1000_init_manageability_pt(struct e1000_adapter *adapter)
2567 struct e1000_hw *hw = &adapter->hw;
2568 u32 manc, manc2h, mdef, i, j;
2570 if (!(adapter->flags & FLAG_MNG_PT_ENABLED))
2576 * enable receiving management packets to the host. this will probably
2577 * generate destination unreachable messages from the host OS, but
2578 * the packets will be handled on SMBUS
2580 manc |= E1000_MANC_EN_MNG2HOST;
2581 manc2h = er32(MANC2H);
2583 switch (hw->mac.type) {
2585 manc2h |= (E1000_MANC2H_PORT_623 | E1000_MANC2H_PORT_664);
2590 * Check if IPMI pass-through decision filter already exists;
2593 for (i = 0, j = 0; i < 8; i++) {
2594 mdef = er32(MDEF(i));
2596 /* Ignore filters with anything other than IPMI ports */
2597 if (mdef & ~(E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2600 /* Enable this decision filter in MANC2H */
2607 if (j == (E1000_MDEF_PORT_623 | E1000_MDEF_PORT_664))
2610 /* Create new decision filter in an empty filter */
2611 for (i = 0, j = 0; i < 8; i++)
2612 if (er32(MDEF(i)) == 0) {
2613 ew32(MDEF(i), (E1000_MDEF_PORT_623 |
2614 E1000_MDEF_PORT_664));
2621 e_warn("Unable to create IPMI pass-through filter\n");
2625 ew32(MANC2H, manc2h);
2630 * e1000_configure_tx - Configure Transmit Unit after Reset
2631 * @adapter: board private structure
2633 * Configure the Tx unit of the MAC after a reset.
2635 static void e1000_configure_tx(struct e1000_adapter *adapter)
2637 struct e1000_hw *hw = &adapter->hw;
2638 struct e1000_ring *tx_ring = adapter->tx_ring;
2640 u32 tdlen, tctl, tipg, tarc;
2643 /* Setup the HW Tx Head and Tail descriptor pointers */
2644 tdba = tx_ring->dma;
2645 tdlen = tx_ring->count * sizeof(struct e1000_tx_desc);
2646 ew32(TDBAL, (tdba & DMA_BIT_MASK(32)));
2647 ew32(TDBAH, (tdba >> 32));
2651 tx_ring->head = E1000_TDH;
2652 tx_ring->tail = E1000_TDT;
2654 /* Set the default values for the Tx Inter Packet Gap timer */
2655 tipg = DEFAULT_82543_TIPG_IPGT_COPPER; /* 8 */
2656 ipgr1 = DEFAULT_82543_TIPG_IPGR1; /* 8 */
2657 ipgr2 = DEFAULT_82543_TIPG_IPGR2; /* 6 */
2659 if (adapter->flags & FLAG_TIPG_MEDIUM_FOR_80003ESLAN)
2660 ipgr2 = DEFAULT_80003ES2LAN_TIPG_IPGR2; /* 7 */
2662 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
2663 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
2666 /* Set the Tx Interrupt Delay register */
2667 ew32(TIDV, adapter->tx_int_delay);
2668 /* Tx irq moderation */
2669 ew32(TADV, adapter->tx_abs_int_delay);
2671 if (adapter->flags2 & FLAG2_DMA_BURST) {
2672 u32 txdctl = er32(TXDCTL(0));
2673 txdctl &= ~(E1000_TXDCTL_PTHRESH | E1000_TXDCTL_HTHRESH |
2674 E1000_TXDCTL_WTHRESH);
2676 * set up some performance related parameters to encourage the
2677 * hardware to use the bus more efficiently in bursts, depends
2678 * on the tx_int_delay to be enabled,
2679 * wthresh = 5 ==> burst write a cacheline (64 bytes) at a time
2680 * hthresh = 1 ==> prefetch when one or more available
2681 * pthresh = 0x1f ==> prefetch if internal cache 31 or less
2682 * BEWARE: this seems to work but should be considered first if
2683 * there are Tx hangs or other Tx related bugs
2685 txdctl |= E1000_TXDCTL_DMA_BURST_ENABLE;
2686 ew32(TXDCTL(0), txdctl);
2687 /* erratum work around: set txdctl the same for both queues */
2688 ew32(TXDCTL(1), txdctl);
2691 /* Program the Transmit Control Register */
2693 tctl &= ~E1000_TCTL_CT;
2694 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2695 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2697 if (adapter->flags & FLAG_TARC_SPEED_MODE_BIT) {
2698 tarc = er32(TARC(0));
2700 * set the speed mode bit, we'll clear it if we're not at
2701 * gigabit link later
2703 #define SPEED_MODE_BIT (1 << 21)
2704 tarc |= SPEED_MODE_BIT;
2705 ew32(TARC(0), tarc);
2708 /* errata: program both queues to unweighted RR */
2709 if (adapter->flags & FLAG_TARC_SET_BIT_ZERO) {
2710 tarc = er32(TARC(0));
2712 ew32(TARC(0), tarc);
2713 tarc = er32(TARC(1));
2715 ew32(TARC(1), tarc);
2718 /* Setup Transmit Descriptor Settings for eop descriptor */
2719 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
2721 /* only set IDE if we are delaying interrupts using the timers */
2722 if (adapter->tx_int_delay)
2723 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2725 /* enable Report Status bit */
2726 adapter->txd_cmd |= E1000_TXD_CMD_RS;
2730 e1000e_config_collision_dist(hw);
2734 * e1000_setup_rctl - configure the receive control registers
2735 * @adapter: Board private structure
2737 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
2738 (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
2739 static void e1000_setup_rctl(struct e1000_adapter *adapter)
2741 struct e1000_hw *hw = &adapter->hw;
2745 /* Workaround Si errata on 82579 - configure jumbo frame flow */
2746 if (hw->mac.type == e1000_pch2lan) {
2749 if (adapter->netdev->mtu > ETH_DATA_LEN)
2750 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, true);
2752 ret_val = e1000_lv_jumbo_workaround_ich8lan(hw, false);
2755 e_dbg("failed to enable jumbo frame workaround mode\n");
2758 /* Program MC offset vector base */
2760 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
2761 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
2762 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
2763 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
2765 /* Do not Store bad packets */
2766 rctl &= ~E1000_RCTL_SBP;
2768 /* Enable Long Packet receive */
2769 if (adapter->netdev->mtu <= ETH_DATA_LEN)
2770 rctl &= ~E1000_RCTL_LPE;
2772 rctl |= E1000_RCTL_LPE;
2774 /* Some systems expect that the CRC is included in SMBUS traffic. The
2775 * hardware strips the CRC before sending to both SMBUS (BMC) and to
2776 * host memory when this is enabled
2778 if (adapter->flags2 & FLAG2_CRC_STRIPPING)
2779 rctl |= E1000_RCTL_SECRC;
2781 /* Workaround Si errata on 82577 PHY - configure IPG for jumbos */
2782 if ((hw->phy.type == e1000_phy_82577) && (rctl & E1000_RCTL_LPE)) {
2785 e1e_rphy(hw, PHY_REG(770, 26), &phy_data);
2787 phy_data |= (1 << 2);
2788 e1e_wphy(hw, PHY_REG(770, 26), phy_data);
2790 e1e_rphy(hw, 22, &phy_data);
2792 phy_data |= (1 << 14);
2793 e1e_wphy(hw, 0x10, 0x2823);
2794 e1e_wphy(hw, 0x11, 0x0003);
2795 e1e_wphy(hw, 22, phy_data);
2798 /* Setup buffer sizes */
2799 rctl &= ~E1000_RCTL_SZ_4096;
2800 rctl |= E1000_RCTL_BSEX;
2801 switch (adapter->rx_buffer_len) {
2804 rctl |= E1000_RCTL_SZ_2048;
2805 rctl &= ~E1000_RCTL_BSEX;
2808 rctl |= E1000_RCTL_SZ_4096;
2811 rctl |= E1000_RCTL_SZ_8192;
2814 rctl |= E1000_RCTL_SZ_16384;
2819 * 82571 and greater support packet-split where the protocol
2820 * header is placed in skb->data and the packet data is
2821 * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2822 * In the case of a non-split, skb->data is linearly filled,
2823 * followed by the page buffers. Therefore, skb->data is
2824 * sized to hold the largest protocol header.
2826 * allocations using alloc_page take too long for regular MTU
2827 * so only enable packet split for jumbo frames
2829 * Using pages when the page size is greater than 16k wastes
2830 * a lot of memory, since we allocate 3 pages at all times
2833 pages = PAGE_USE_COUNT(adapter->netdev->mtu);
2834 if (!(adapter->flags & FLAG_HAS_ERT) && (pages <= 3) &&
2835 (PAGE_SIZE <= 16384) && (rctl & E1000_RCTL_LPE))
2836 adapter->rx_ps_pages = pages;
2838 adapter->rx_ps_pages = 0;
2840 if (adapter->rx_ps_pages) {
2843 /* Configure extra packet-split registers */
2844 rfctl = er32(RFCTL);
2845 rfctl |= E1000_RFCTL_EXTEN;
2847 * disable packet split support for IPv6 extension headers,
2848 * because some malformed IPv6 headers can hang the Rx
2850 rfctl |= (E1000_RFCTL_IPV6_EX_DIS |
2851 E1000_RFCTL_NEW_IPV6_EXT_DIS);
2855 /* Enable Packet split descriptors */
2856 rctl |= E1000_RCTL_DTYP_PS;
2858 psrctl |= adapter->rx_ps_bsize0 >>
2859 E1000_PSRCTL_BSIZE0_SHIFT;
2861 switch (adapter->rx_ps_pages) {
2863 psrctl |= PAGE_SIZE <<
2864 E1000_PSRCTL_BSIZE3_SHIFT;
2866 psrctl |= PAGE_SIZE <<
2867 E1000_PSRCTL_BSIZE2_SHIFT;
2869 psrctl |= PAGE_SIZE >>
2870 E1000_PSRCTL_BSIZE1_SHIFT;
2874 ew32(PSRCTL, psrctl);
2878 /* just started the receive unit, no need to restart */
2879 adapter->flags &= ~FLAG_RX_RESTART_NOW;
2883 * e1000_configure_rx - Configure Receive Unit after Reset
2884 * @adapter: board private structure
2886 * Configure the Rx unit of the MAC after a reset.
2888 static void e1000_configure_rx(struct e1000_adapter *adapter)
2890 struct e1000_hw *hw = &adapter->hw;
2891 struct e1000_ring *rx_ring = adapter->rx_ring;
2893 u32 rdlen, rctl, rxcsum, ctrl_ext;
2895 if (adapter->rx_ps_pages) {
2896 /* this is a 32 byte descriptor */
2897 rdlen = rx_ring->count *
2898 sizeof(union e1000_rx_desc_packet_split);
2899 adapter->clean_rx = e1000_clean_rx_irq_ps;
2900 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
2901 } else if (adapter->netdev->mtu > ETH_FRAME_LEN + ETH_FCS_LEN) {
2902 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2903 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
2904 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
2906 rdlen = rx_ring->count * sizeof(struct e1000_rx_desc);
2907 adapter->clean_rx = e1000_clean_rx_irq;
2908 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
2911 /* disable receives while setting up the descriptors */
2913 ew32(RCTL, rctl & ~E1000_RCTL_EN);
2915 usleep_range(10000, 20000);
2917 if (adapter->flags2 & FLAG2_DMA_BURST) {
2919 * set the writeback threshold (only takes effect if the RDTR
2920 * is set). set GRAN=1 and write back up to 0x4 worth, and
2921 * enable prefetching of 0x20 Rx descriptors
2927 ew32(RXDCTL(0), E1000_RXDCTL_DMA_BURST_ENABLE);
2928 ew32(RXDCTL(1), E1000_RXDCTL_DMA_BURST_ENABLE);
2931 * override the delay timers for enabling bursting, only if
2932 * the value was not set by the user via module options
2934 if (adapter->rx_int_delay == DEFAULT_RDTR)
2935 adapter->rx_int_delay = BURST_RDTR;
2936 if (adapter->rx_abs_int_delay == DEFAULT_RADV)
2937 adapter->rx_abs_int_delay = BURST_RADV;
2940 /* set the Receive Delay Timer Register */
2941 ew32(RDTR, adapter->rx_int_delay);
2943 /* irq moderation */
2944 ew32(RADV, adapter->rx_abs_int_delay);
2945 if ((adapter->itr_setting != 0) && (adapter->itr != 0))
2946 ew32(ITR, 1000000000 / (adapter->itr * 256));
2948 ctrl_ext = er32(CTRL_EXT);
2949 /* Auto-Mask interrupts upon ICR access */
2950 ctrl_ext |= E1000_CTRL_EXT_IAME;
2951 ew32(IAM, 0xffffffff);
2952 ew32(CTRL_EXT, ctrl_ext);
2956 * Setup the HW Rx Head and Tail Descriptor Pointers and
2957 * the Base and Length of the Rx Descriptor Ring
2959 rdba = rx_ring->dma;
2960 ew32(RDBAL, (rdba & DMA_BIT_MASK(32)));
2961 ew32(RDBAH, (rdba >> 32));
2965 rx_ring->head = E1000_RDH;
2966 rx_ring->tail = E1000_RDT;
2968 /* Enable Receive Checksum Offload for TCP and UDP */
2969 rxcsum = er32(RXCSUM);
2970 if (adapter->flags & FLAG_RX_CSUM_ENABLED) {
2971 rxcsum |= E1000_RXCSUM_TUOFL;
2974 * IPv4 payload checksum for UDP fragments must be
2975 * used in conjunction with packet-split.
2977 if (adapter->rx_ps_pages)
2978 rxcsum |= E1000_RXCSUM_IPPCSE;
2980 rxcsum &= ~E1000_RXCSUM_TUOFL;
2981 /* no need to clear IPPCSE as it defaults to 0 */
2983 ew32(RXCSUM, rxcsum);
2986 * Enable early receives on supported devices, only takes effect when
2987 * packet size is equal or larger than the specified value (in 8 byte
2988 * units), e.g. using jumbo frames when setting to E1000_ERT_2048
2990 if ((adapter->flags & FLAG_HAS_ERT) ||
2991 (adapter->hw.mac.type == e1000_pch2lan)) {
2992 if (adapter->netdev->mtu > ETH_DATA_LEN) {
2993 u32 rxdctl = er32(RXDCTL(0));
2994 ew32(RXDCTL(0), rxdctl | 0x3);
2995 if (adapter->flags & FLAG_HAS_ERT)
2996 ew32(ERT, E1000_ERT_2048 | (1 << 13));
2998 * With jumbo frames and early-receive enabled,
2999 * excessive C-state transition latencies result in
3000 * dropped transactions.
3002 pm_qos_update_request(&adapter->netdev->pm_qos_req, 55);
3004 pm_qos_update_request(&adapter->netdev->pm_qos_req,
3005 PM_QOS_DEFAULT_VALUE);
3009 /* Enable Receives */
3014 * e1000_update_mc_addr_list - Update Multicast addresses
3015 * @hw: pointer to the HW structure
3016 * @mc_addr_list: array of multicast addresses to program
3017 * @mc_addr_count: number of multicast addresses to program
3019 * Updates the Multicast Table Array.
3020 * The caller must have a packed mc_addr_list of multicast addresses.
3022 static void e1000_update_mc_addr_list(struct e1000_hw *hw, u8 *mc_addr_list,
3025 hw->mac.ops.update_mc_addr_list(hw, mc_addr_list, mc_addr_count);
3029 * e1000_set_multi - Multicast and Promiscuous mode set
3030 * @netdev: network interface device structure
3032 * The set_multi entry point is called whenever the multicast address
3033 * list or the network interface flags are updated. This routine is
3034 * responsible for configuring the hardware for proper multicast,
3035 * promiscuous mode, and all-multi behavior.
3037 static void e1000_set_multi(struct net_device *netdev)
3039 struct e1000_adapter *adapter = netdev_priv(netdev);
3040 struct e1000_hw *hw = &adapter->hw;
3041 struct netdev_hw_addr *ha;
3045 /* Check for Promiscuous and All Multicast modes */
3049 if (netdev->flags & IFF_PROMISC) {
3050 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3051 rctl &= ~E1000_RCTL_VFE;
3052 /* Do not hardware filter VLANs in promisc mode */
3053 e1000e_vlan_filter_disable(adapter);
3055 if (netdev->flags & IFF_ALLMULTI) {
3056 rctl |= E1000_RCTL_MPE;
3057 rctl &= ~E1000_RCTL_UPE;
3059 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
3061 e1000e_vlan_filter_enable(adapter);
3066 if (!netdev_mc_empty(netdev)) {
3069 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3073 /* prepare a packed array of only addresses. */
3074 netdev_for_each_mc_addr(ha, netdev)
3075 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3077 e1000_update_mc_addr_list(hw, mta_list, i);
3081 * if we're called from probe, we might not have
3082 * anything to do here, so clear out the list
3084 e1000_update_mc_addr_list(hw, NULL, 0);
3087 if (netdev->features & NETIF_F_HW_VLAN_RX)
3088 e1000e_vlan_strip_enable(adapter);
3090 e1000e_vlan_strip_disable(adapter);
3094 * e1000_configure - configure the hardware for Rx and Tx
3095 * @adapter: private board structure
3097 static void e1000_configure(struct e1000_adapter *adapter)
3099 e1000_set_multi(adapter->netdev);
3101 e1000_restore_vlan(adapter);
3102 e1000_init_manageability_pt(adapter);
3104 e1000_configure_tx(adapter);
3105 e1000_setup_rctl(adapter);
3106 e1000_configure_rx(adapter);
3107 adapter->alloc_rx_buf(adapter, e1000_desc_unused(adapter->rx_ring));
3111 * e1000e_power_up_phy - restore link in case the phy was powered down
3112 * @adapter: address of board private structure
3114 * The phy may be powered down to save power and turn off link when the
3115 * driver is unloaded and wake on lan is not enabled (among others)
3116 * *** this routine MUST be followed by a call to e1000e_reset ***
3118 void e1000e_power_up_phy(struct e1000_adapter *adapter)
3120 if (adapter->hw.phy.ops.power_up)
3121 adapter->hw.phy.ops.power_up(&adapter->hw);
3123 adapter->hw.mac.ops.setup_link(&adapter->hw);
3127 * e1000_power_down_phy - Power down the PHY
3129 * Power down the PHY so no link is implied when interface is down.
3130 * The PHY cannot be powered down if management or WoL is active.
3132 static void e1000_power_down_phy(struct e1000_adapter *adapter)
3134 /* WoL is enabled */
3138 if (adapter->hw.phy.ops.power_down)
3139 adapter->hw.phy.ops.power_down(&adapter->hw);
3143 * e1000e_reset - bring the hardware into a known good state
3145 * This function boots the hardware and enables some settings that
3146 * require a configuration cycle of the hardware - those cannot be
3147 * set/changed during runtime. After reset the device needs to be
3148 * properly configured for Rx, Tx etc.
3150 void e1000e_reset(struct e1000_adapter *adapter)
3152 struct e1000_mac_info *mac = &adapter->hw.mac;
3153 struct e1000_fc_info *fc = &adapter->hw.fc;
3154 struct e1000_hw *hw = &adapter->hw;
3155 u32 tx_space, min_tx_space, min_rx_space;
3156 u32 pba = adapter->pba;
3159 /* reset Packet Buffer Allocation to default */
3162 if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
3164 * To maintain wire speed transmits, the Tx FIFO should be
3165 * large enough to accommodate two full transmit packets,
3166 * rounded up to the next 1KB and expressed in KB. Likewise,
3167 * the Rx FIFO should be large enough to accommodate at least
3168 * one full receive packet and is similarly rounded up and
3172 /* upper 16 bits has Tx packet buffer allocation size in KB */
3173 tx_space = pba >> 16;
3174 /* lower 16 bits has Rx packet buffer allocation size in KB */
3177 * the Tx fifo also stores 16 bytes of information about the Tx
3178 * but don't include ethernet FCS because hardware appends it
3180 min_tx_space = (adapter->max_frame_size +
3181 sizeof(struct e1000_tx_desc) -
3183 min_tx_space = ALIGN(min_tx_space, 1024);
3184 min_tx_space >>= 10;
3185 /* software strips receive CRC, so leave room for it */
3186 min_rx_space = adapter->max_frame_size;
3187 min_rx_space = ALIGN(min_rx_space, 1024);
3188 min_rx_space >>= 10;
3191 * If current Tx allocation is less than the min Tx FIFO size,
3192 * and the min Tx FIFO size is less than the current Rx FIFO
3193 * allocation, take space away from current Rx allocation
3195 if ((tx_space < min_tx_space) &&
3196 ((min_tx_space - tx_space) < pba)) {
3197 pba -= min_tx_space - tx_space;
3200 * if short on Rx space, Rx wins and must trump Tx
3201 * adjustment or use Early Receive if available
3203 if ((pba < min_rx_space) &&
3204 (!(adapter->flags & FLAG_HAS_ERT)))
3205 /* ERT enabled in e1000_configure_rx */
3213 * flow control settings
3215 * The high water mark must be low enough to fit one full frame
3216 * (or the size used for early receive) above it in the Rx FIFO.
3217 * Set it to the lower of:
3218 * - 90% of the Rx FIFO size, and
3219 * - the full Rx FIFO size minus the early receive size (for parts
3220 * with ERT support assuming ERT set to E1000_ERT_2048), or
3221 * - the full Rx FIFO size minus one full frame
3223 if (adapter->flags & FLAG_DISABLE_FC_PAUSE_TIME)
3224 fc->pause_time = 0xFFFF;
3226 fc->pause_time = E1000_FC_PAUSE_TIME;
3228 fc->current_mode = fc->requested_mode;
3230 switch (hw->mac.type) {
3232 if ((adapter->flags & FLAG_HAS_ERT) &&
3233 (adapter->netdev->mtu > ETH_DATA_LEN))
3234 hwm = min(((pba << 10) * 9 / 10),
3235 ((pba << 10) - (E1000_ERT_2048 << 3)));
3237 hwm = min(((pba << 10) * 9 / 10),
3238 ((pba << 10) - adapter->max_frame_size));
3240 fc->high_water = hwm & E1000_FCRTH_RTH; /* 8-byte granularity */
3241 fc->low_water = fc->high_water - 8;
3245 * Workaround PCH LOM adapter hangs with certain network
3246 * loads. If hangs persist, try disabling Tx flow control.
3248 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3249 fc->high_water = 0x3500;
3250 fc->low_water = 0x1500;
3252 fc->high_water = 0x5000;
3253 fc->low_water = 0x3000;
3255 fc->refresh_time = 0x1000;
3258 fc->high_water = 0x05C20;
3259 fc->low_water = 0x05048;
3260 fc->pause_time = 0x0650;
3261 fc->refresh_time = 0x0400;
3262 if (adapter->netdev->mtu > ETH_DATA_LEN) {
3270 * Disable Adaptive Interrupt Moderation if 2 full packets cannot
3271 * fit in receive buffer and early-receive not supported.
3273 if (adapter->itr_setting & 0x3) {
3274 if (((adapter->max_frame_size * 2) > (pba << 10)) &&
3275 !(adapter->flags & FLAG_HAS_ERT)) {
3276 if (!(adapter->flags2 & FLAG2_DISABLE_AIM)) {
3277 dev_info(&adapter->pdev->dev,
3278 "Interrupt Throttle Rate turned off\n");
3279 adapter->flags2 |= FLAG2_DISABLE_AIM;
3282 } else if (adapter->flags2 & FLAG2_DISABLE_AIM) {
3283 dev_info(&adapter->pdev->dev,
3284 "Interrupt Throttle Rate turned on\n");
3285 adapter->flags2 &= ~FLAG2_DISABLE_AIM;
3286 adapter->itr = 20000;
3287 ew32(ITR, 1000000000 / (adapter->itr * 256));
3291 /* Allow time for pending master requests to run */
3292 mac->ops.reset_hw(hw);
3295 * For parts with AMT enabled, let the firmware know
3296 * that the network interface is in control
3298 if (adapter->flags & FLAG_HAS_AMT)
3299 e1000e_get_hw_control(adapter);
3303 if (mac->ops.init_hw(hw))
3304 e_err("Hardware Error\n");
3306 e1000_update_mng_vlan(adapter);
3308 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
3309 ew32(VET, ETH_P_8021Q);
3311 e1000e_reset_adaptive(hw);
3313 if (!netif_running(adapter->netdev) &&
3314 !test_bit(__E1000_TESTING, &adapter->state)) {
3315 e1000_power_down_phy(adapter);
3319 e1000_get_phy_info(hw);
3321 if ((adapter->flags & FLAG_HAS_SMART_POWER_DOWN) &&
3322 !(adapter->flags & FLAG_SMART_POWER_DOWN)) {
3325 * speed up time to link by disabling smart power down, ignore
3326 * the return value of this function because there is nothing
3327 * different we would do if it failed
3329 e1e_rphy(hw, IGP02E1000_PHY_POWER_MGMT, &phy_data);
3330 phy_data &= ~IGP02E1000_PM_SPD;
3331 e1e_wphy(hw, IGP02E1000_PHY_POWER_MGMT, phy_data);
3335 int e1000e_up(struct e1000_adapter *adapter)
3337 struct e1000_hw *hw = &adapter->hw;
3339 /* hardware has been reset, we need to reload some things */
3340 e1000_configure(adapter);
3342 clear_bit(__E1000_DOWN, &adapter->state);
3344 napi_enable(&adapter->napi);
3345 if (adapter->msix_entries)
3346 e1000_configure_msix(adapter);
3347 e1000_irq_enable(adapter);
3349 netif_wake_queue(adapter->netdev);
3351 /* fire a link change interrupt to start the watchdog */
3352 if (adapter->msix_entries)
3353 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3355 ew32(ICS, E1000_ICS_LSC);
3360 static void e1000e_flush_descriptors(struct e1000_adapter *adapter)
3362 struct e1000_hw *hw = &adapter->hw;
3364 if (!(adapter->flags2 & FLAG2_DMA_BURST))
3367 /* flush pending descriptor writebacks to memory */
3368 ew32(TIDV, adapter->tx_int_delay | E1000_TIDV_FPD);
3369 ew32(RDTR, adapter->rx_int_delay | E1000_RDTR_FPD);
3371 /* execute the writes immediately */
3375 static void e1000e_update_stats(struct e1000_adapter *adapter);
3377 void e1000e_down(struct e1000_adapter *adapter)
3379 struct net_device *netdev = adapter->netdev;
3380 struct e1000_hw *hw = &adapter->hw;
3384 * signal that we're down so the interrupt handler does not
3385 * reschedule our watchdog timer
3387 set_bit(__E1000_DOWN, &adapter->state);
3389 /* disable receives in the hardware */
3391 ew32(RCTL, rctl & ~E1000_RCTL_EN);
3392 /* flush and sleep below */
3394 netif_stop_queue(netdev);
3396 /* disable transmits in the hardware */
3398 tctl &= ~E1000_TCTL_EN;
3400 /* flush both disables and wait for them to finish */
3402 usleep_range(10000, 20000);
3404 napi_disable(&adapter->napi);
3405 e1000_irq_disable(adapter);
3407 del_timer_sync(&adapter->watchdog_timer);
3408 del_timer_sync(&adapter->phy_info_timer);
3410 netif_carrier_off(netdev);
3412 spin_lock(&adapter->stats64_lock);
3413 e1000e_update_stats(adapter);
3414 spin_unlock(&adapter->stats64_lock);
3416 adapter->link_speed = 0;
3417 adapter->link_duplex = 0;
3419 if (!pci_channel_offline(adapter->pdev))
3420 e1000e_reset(adapter);
3422 e1000e_flush_descriptors(adapter);
3424 e1000_clean_tx_ring(adapter);
3425 e1000_clean_rx_ring(adapter);
3428 * TODO: for power management, we could drop the link and
3429 * pci_disable_device here.
3433 void e1000e_reinit_locked(struct e1000_adapter *adapter)
3436 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
3437 usleep_range(1000, 2000);
3438 e1000e_down(adapter);
3440 clear_bit(__E1000_RESETTING, &adapter->state);
3444 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
3445 * @adapter: board private structure to initialize
3447 * e1000_sw_init initializes the Adapter private data structure.
3448 * Fields are initialized based on PCI device information and
3449 * OS network device settings (MTU size).
3451 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
3453 struct net_device *netdev = adapter->netdev;
3455 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
3456 adapter->rx_ps_bsize0 = 128;
3457 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3458 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3460 spin_lock_init(&adapter->stats64_lock);
3462 e1000e_set_interrupt_capability(adapter);
3464 if (e1000_alloc_queues(adapter))
3467 /* Explicitly disable IRQ since the NIC can be in any state. */
3468 e1000_irq_disable(adapter);
3470 set_bit(__E1000_DOWN, &adapter->state);
3475 * e1000_intr_msi_test - Interrupt Handler
3476 * @irq: interrupt number
3477 * @data: pointer to a network interface device structure
3479 static irqreturn_t e1000_intr_msi_test(int irq, void *data)
3481 struct net_device *netdev = data;
3482 struct e1000_adapter *adapter = netdev_priv(netdev);
3483 struct e1000_hw *hw = &adapter->hw;
3484 u32 icr = er32(ICR);
3486 e_dbg("icr is %08X\n", icr);
3487 if (icr & E1000_ICR_RXSEQ) {
3488 adapter->flags &= ~FLAG_MSI_TEST_FAILED;
3496 * e1000_test_msi_interrupt - Returns 0 for successful test
3497 * @adapter: board private struct
3499 * code flow taken from tg3.c
3501 static int e1000_test_msi_interrupt(struct e1000_adapter *adapter)
3503 struct net_device *netdev = adapter->netdev;
3504 struct e1000_hw *hw = &adapter->hw;
3507 /* poll_enable hasn't been called yet, so don't need disable */
3508 /* clear any pending events */
3511 /* free the real vector and request a test handler */
3512 e1000_free_irq(adapter);
3513 e1000e_reset_interrupt_capability(adapter);
3515 /* Assume that the test fails, if it succeeds then the test
3516 * MSI irq handler will unset this flag */
3517 adapter->flags |= FLAG_MSI_TEST_FAILED;
3519 err = pci_enable_msi(adapter->pdev);
3521 goto msi_test_failed;
3523 err = request_irq(adapter->pdev->irq, e1000_intr_msi_test, 0,
3524 netdev->name, netdev);
3526 pci_disable_msi(adapter->pdev);
3527 goto msi_test_failed;
3532 e1000_irq_enable(adapter);
3534 /* fire an unusual interrupt on the test handler */
3535 ew32(ICS, E1000_ICS_RXSEQ);
3539 e1000_irq_disable(adapter);
3543 if (adapter->flags & FLAG_MSI_TEST_FAILED) {
3544 adapter->int_mode = E1000E_INT_MODE_LEGACY;
3545 e_info("MSI interrupt test failed, using legacy interrupt.\n");
3547 e_dbg("MSI interrupt test succeeded!\n");
3549 free_irq(adapter->pdev->irq, netdev);
3550 pci_disable_msi(adapter->pdev);
3553 e1000e_set_interrupt_capability(adapter);
3554 return e1000_request_irq(adapter);
3558 * e1000_test_msi - Returns 0 if MSI test succeeds or INTx mode is restored
3559 * @adapter: board private struct
3561 * code flow taken from tg3.c, called with e1000 interrupts disabled.
3563 static int e1000_test_msi(struct e1000_adapter *adapter)
3568 if (!(adapter->flags & FLAG_MSI_ENABLED))
3571 /* disable SERR in case the MSI write causes a master abort */
3572 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3573 if (pci_cmd & PCI_COMMAND_SERR)
3574 pci_write_config_word(adapter->pdev, PCI_COMMAND,
3575 pci_cmd & ~PCI_COMMAND_SERR);
3577 err = e1000_test_msi_interrupt(adapter);
3579 /* re-enable SERR */
3580 if (pci_cmd & PCI_COMMAND_SERR) {
3581 pci_read_config_word(adapter->pdev, PCI_COMMAND, &pci_cmd);
3582 pci_cmd |= PCI_COMMAND_SERR;
3583 pci_write_config_word(adapter->pdev, PCI_COMMAND, pci_cmd);
3590 * e1000_open - Called when a network interface is made active
3591 * @netdev: network interface device structure
3593 * Returns 0 on success, negative value on failure
3595 * The open entry point is called when a network interface is made
3596 * active by the system (IFF_UP). At this point all resources needed
3597 * for transmit and receive operations are allocated, the interrupt
3598 * handler is registered with the OS, the watchdog timer is started,
3599 * and the stack is notified that the interface is ready.
3601 static int e1000_open(struct net_device *netdev)
3603 struct e1000_adapter *adapter = netdev_priv(netdev);
3604 struct e1000_hw *hw = &adapter->hw;
3605 struct pci_dev *pdev = adapter->pdev;
3608 /* disallow open during test */
3609 if (test_bit(__E1000_TESTING, &adapter->state))
3612 pm_runtime_get_sync(&pdev->dev);
3614 netif_carrier_off(netdev);
3616 /* allocate transmit descriptors */
3617 err = e1000e_setup_tx_resources(adapter);
3621 /* allocate receive descriptors */
3622 err = e1000e_setup_rx_resources(adapter);
3627 * If AMT is enabled, let the firmware know that the network
3628 * interface is now open and reset the part to a known state.
3630 if (adapter->flags & FLAG_HAS_AMT) {
3631 e1000e_get_hw_control(adapter);
3632 e1000e_reset(adapter);
3635 e1000e_power_up_phy(adapter);
3637 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
3638 if ((adapter->hw.mng_cookie.status &
3639 E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
3640 e1000_update_mng_vlan(adapter);
3642 /* DMA latency requirement to workaround early-receive/jumbo issue */
3643 if ((adapter->flags & FLAG_HAS_ERT) ||
3644 (adapter->hw.mac.type == e1000_pch2lan))
3645 pm_qos_add_request(&adapter->netdev->pm_qos_req,
3646 PM_QOS_CPU_DMA_LATENCY,
3647 PM_QOS_DEFAULT_VALUE);
3650 * before we allocate an interrupt, we must be ready to handle it.
3651 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3652 * as soon as we call pci_request_irq, so we have to setup our
3653 * clean_rx handler before we do so.
3655 e1000_configure(adapter);
3657 err = e1000_request_irq(adapter);
3662 * Work around PCIe errata with MSI interrupts causing some chipsets to
3663 * ignore e1000e MSI messages, which means we need to test our MSI
3666 if (adapter->int_mode != E1000E_INT_MODE_LEGACY) {
3667 err = e1000_test_msi(adapter);
3669 e_err("Interrupt allocation failed\n");
3674 /* From here on the code is the same as e1000e_up() */
3675 clear_bit(__E1000_DOWN, &adapter->state);
3677 napi_enable(&adapter->napi);
3679 e1000_irq_enable(adapter);
3681 netif_start_queue(netdev);
3683 adapter->idle_check = true;
3684 pm_runtime_put(&pdev->dev);
3686 /* fire a link status change interrupt to start the watchdog */
3687 if (adapter->msix_entries)
3688 ew32(ICS, E1000_ICS_LSC | E1000_ICR_OTHER);
3690 ew32(ICS, E1000_ICS_LSC);
3695 e1000e_release_hw_control(adapter);
3696 e1000_power_down_phy(adapter);
3697 e1000e_free_rx_resources(adapter);
3699 e1000e_free_tx_resources(adapter);
3701 e1000e_reset(adapter);
3702 pm_runtime_put_sync(&pdev->dev);
3708 * e1000_close - Disables a network interface
3709 * @netdev: network interface device structure
3711 * Returns 0, this is not allowed to fail
3713 * The close entry point is called when an interface is de-activated
3714 * by the OS. The hardware is still under the drivers control, but
3715 * needs to be disabled. A global MAC reset is issued to stop the
3716 * hardware, and all transmit and receive resources are freed.
3718 static int e1000_close(struct net_device *netdev)
3720 struct e1000_adapter *adapter = netdev_priv(netdev);
3721 struct pci_dev *pdev = adapter->pdev;
3723 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
3725 pm_runtime_get_sync(&pdev->dev);
3727 if (!test_bit(__E1000_DOWN, &adapter->state)) {
3728 e1000e_down(adapter);
3729 e1000_free_irq(adapter);
3731 e1000_power_down_phy(adapter);
3733 e1000e_free_tx_resources(adapter);
3734 e1000e_free_rx_resources(adapter);
3737 * kill manageability vlan ID if supported, but not if a vlan with
3738 * the same ID is registered on the host OS (let 8021q kill it)
3740 if (adapter->hw.mng_cookie.status &
3741 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)
3742 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3745 * If AMT is enabled, let the firmware know that the network
3746 * interface is now closed
3748 if ((adapter->flags & FLAG_HAS_AMT) &&
3749 !test_bit(__E1000_TESTING, &adapter->state))
3750 e1000e_release_hw_control(adapter);
3752 if ((adapter->flags & FLAG_HAS_ERT) ||
3753 (adapter->hw.mac.type == e1000_pch2lan))
3754 pm_qos_remove_request(&adapter->netdev->pm_qos_req);
3756 pm_runtime_put_sync(&pdev->dev);
3761 * e1000_set_mac - Change the Ethernet Address of the NIC
3762 * @netdev: network interface device structure
3763 * @p: pointer to an address structure
3765 * Returns 0 on success, negative on failure
3767 static int e1000_set_mac(struct net_device *netdev, void *p)
3769 struct e1000_adapter *adapter = netdev_priv(netdev);
3770 struct sockaddr *addr = p;
3772 if (!is_valid_ether_addr(addr->sa_data))
3773 return -EADDRNOTAVAIL;
3775 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3776 memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
3778 e1000e_rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
3780 if (adapter->flags & FLAG_RESET_OVERWRITES_LAA) {
3781 /* activate the work around */
3782 e1000e_set_laa_state_82571(&adapter->hw, 1);
3785 * Hold a copy of the LAA in RAR[14] This is done so that
3786 * between the time RAR[0] gets clobbered and the time it
3787 * gets fixed (in e1000_watchdog), the actual LAA is in one
3788 * of the RARs and no incoming packets directed to this port
3789 * are dropped. Eventually the LAA will be in RAR[0] and
3792 e1000e_rar_set(&adapter->hw,
3793 adapter->hw.mac.addr,
3794 adapter->hw.mac.rar_entry_count - 1);
3801 * e1000e_update_phy_task - work thread to update phy
3802 * @work: pointer to our work struct
3804 * this worker thread exists because we must acquire a
3805 * semaphore to read the phy, which we could msleep while
3806 * waiting for it, and we can't msleep in a timer.
3808 static void e1000e_update_phy_task(struct work_struct *work)
3810 struct e1000_adapter *adapter = container_of(work,
3811 struct e1000_adapter, update_phy_task);
3813 if (test_bit(__E1000_DOWN, &adapter->state))
3816 e1000_get_phy_info(&adapter->hw);
3820 * Need to wait a few seconds after link up to get diagnostic information from
3823 static void e1000_update_phy_info(unsigned long data)
3825 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
3827 if (test_bit(__E1000_DOWN, &adapter->state))
3830 schedule_work(&adapter->update_phy_task);
3834 * e1000e_update_phy_stats - Update the PHY statistics counters
3835 * @adapter: board private structure
3837 static void e1000e_update_phy_stats(struct e1000_adapter *adapter)
3839 struct e1000_hw *hw = &adapter->hw;
3843 ret_val = hw->phy.ops.acquire(hw);
3849 #define HV_PHY_STATS_PAGE 778
3851 * A page set is expensive so check if already on desired page.
3852 * If not, set to the page with the PHY status registers.
3854 ret_val = e1000e_read_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
3858 if (phy_data != (HV_PHY_STATS_PAGE << IGP_PAGE_SHIFT)) {
3859 ret_val = e1000e_write_phy_reg_mdic(hw,
3860 IGP01E1000_PHY_PAGE_SELECT,
3861 (HV_PHY_STATS_PAGE <<
3867 /* Read/clear the upper 16-bit registers and read/accumulate lower */
3869 /* Single Collision Count */
3870 e1000e_read_phy_reg_mdic(hw, HV_SCC_UPPER & MAX_PHY_REG_ADDRESS,
3872 ret_val = e1000e_read_phy_reg_mdic(hw,
3873 HV_SCC_LOWER & MAX_PHY_REG_ADDRESS,
3876 adapter->stats.scc += phy_data;
3878 /* Excessive Collision Count */
3879 e1000e_read_phy_reg_mdic(hw, HV_ECOL_UPPER & MAX_PHY_REG_ADDRESS,
3881 ret_val = e1000e_read_phy_reg_mdic(hw,
3882 HV_ECOL_LOWER & MAX_PHY_REG_ADDRESS,
3885 adapter->stats.ecol += phy_data;
3887 /* Multiple Collision Count */
3888 e1000e_read_phy_reg_mdic(hw, HV_MCC_UPPER & MAX_PHY_REG_ADDRESS,
3890 ret_val = e1000e_read_phy_reg_mdic(hw,
3891 HV_MCC_LOWER & MAX_PHY_REG_ADDRESS,
3894 adapter->stats.mcc += phy_data;
3896 /* Late Collision Count */
3897 e1000e_read_phy_reg_mdic(hw, HV_LATECOL_UPPER & MAX_PHY_REG_ADDRESS,
3899 ret_val = e1000e_read_phy_reg_mdic(hw,
3901 MAX_PHY_REG_ADDRESS,
3904 adapter->stats.latecol += phy_data;
3906 /* Collision Count - also used for adaptive IFS */
3907 e1000e_read_phy_reg_mdic(hw, HV_COLC_UPPER & MAX_PHY_REG_ADDRESS,
3909 ret_val = e1000e_read_phy_reg_mdic(hw,
3910 HV_COLC_LOWER & MAX_PHY_REG_ADDRESS,
3913 hw->mac.collision_delta = phy_data;
3916 e1000e_read_phy_reg_mdic(hw, HV_DC_UPPER & MAX_PHY_REG_ADDRESS,
3918 ret_val = e1000e_read_phy_reg_mdic(hw,
3919 HV_DC_LOWER & MAX_PHY_REG_ADDRESS,
3922 adapter->stats.dc += phy_data;
3924 /* Transmit with no CRS */
3925 e1000e_read_phy_reg_mdic(hw, HV_TNCRS_UPPER & MAX_PHY_REG_ADDRESS,
3927 ret_val = e1000e_read_phy_reg_mdic(hw,
3928 HV_TNCRS_LOWER & MAX_PHY_REG_ADDRESS,
3931 adapter->stats.tncrs += phy_data;
3934 hw->phy.ops.release(hw);
3938 * e1000e_update_stats - Update the board statistics counters
3939 * @adapter: board private structure
3941 static void e1000e_update_stats(struct e1000_adapter *adapter)
3943 struct net_device *netdev = adapter->netdev;
3944 struct e1000_hw *hw = &adapter->hw;
3945 struct pci_dev *pdev = adapter->pdev;
3948 * Prevent stats update while adapter is being reset, or if the pci
3949 * connection is down.
3951 if (adapter->link_speed == 0)
3953 if (pci_channel_offline(pdev))
3956 adapter->stats.crcerrs += er32(CRCERRS);
3957 adapter->stats.gprc += er32(GPRC);
3958 adapter->stats.gorc += er32(GORCL);
3959 er32(GORCH); /* Clear gorc */
3960 adapter->stats.bprc += er32(BPRC);
3961 adapter->stats.mprc += er32(MPRC);
3962 adapter->stats.roc += er32(ROC);
3964 adapter->stats.mpc += er32(MPC);
3966 /* Half-duplex statistics */
3967 if (adapter->link_duplex == HALF_DUPLEX) {
3968 if (adapter->flags2 & FLAG2_HAS_PHY_STATS) {
3969 e1000e_update_phy_stats(adapter);
3971 adapter->stats.scc += er32(SCC);
3972 adapter->stats.ecol += er32(ECOL);
3973 adapter->stats.mcc += er32(MCC);
3974 adapter->stats.latecol += er32(LATECOL);
3975 adapter->stats.dc += er32(DC);
3977 hw->mac.collision_delta = er32(COLC);
3979 if ((hw->mac.type != e1000_82574) &&
3980 (hw->mac.type != e1000_82583))
3981 adapter->stats.tncrs += er32(TNCRS);
3983 adapter->stats.colc += hw->mac.collision_delta;
3986 adapter->stats.xonrxc += er32(XONRXC);
3987 adapter->stats.xontxc += er32(XONTXC);
3988 adapter->stats.xoffrxc += er32(XOFFRXC);
3989 adapter->stats.xofftxc += er32(XOFFTXC);
3990 adapter->stats.gptc += er32(GPTC);
3991 adapter->stats.gotc += er32(GOTCL);
3992 er32(GOTCH); /* Clear gotc */
3993 adapter->stats.rnbc += er32(RNBC);
3994 adapter->stats.ruc += er32(RUC);
3996 adapter->stats.mptc += er32(MPTC);
3997 adapter->stats.bptc += er32(BPTC);
3999 /* used for adaptive IFS */
4001 hw->mac.tx_packet_delta = er32(TPT);
4002 adapter->stats.tpt += hw->mac.tx_packet_delta;
4004 adapter->stats.algnerrc += er32(ALGNERRC);
4005 adapter->stats.rxerrc += er32(RXERRC);
4006 adapter->stats.cexterr += er32(CEXTERR);
4007 adapter->stats.tsctc += er32(TSCTC);
4008 adapter->stats.tsctfc += er32(TSCTFC);
4010 /* Fill out the OS statistics structure */
4011 netdev->stats.multicast = adapter->stats.mprc;
4012 netdev->stats.collisions = adapter->stats.colc;
4017 * RLEC on some newer hardware can be incorrect so build
4018 * our own version based on RUC and ROC
4020 netdev->stats.rx_errors = adapter->stats.rxerrc +
4021 adapter->stats.crcerrs + adapter->stats.algnerrc +
4022 adapter->stats.ruc + adapter->stats.roc +
4023 adapter->stats.cexterr;
4024 netdev->stats.rx_length_errors = adapter->stats.ruc +
4026 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
4027 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
4028 netdev->stats.rx_missed_errors = adapter->stats.mpc;
4031 netdev->stats.tx_errors = adapter->stats.ecol +
4032 adapter->stats.latecol;
4033 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
4034 netdev->stats.tx_window_errors = adapter->stats.latecol;
4035 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
4037 /* Tx Dropped needs to be maintained elsewhere */
4039 /* Management Stats */
4040 adapter->stats.mgptc += er32(MGTPTC);
4041 adapter->stats.mgprc += er32(MGTPRC);
4042 adapter->stats.mgpdc += er32(MGTPDC);
4046 * e1000_phy_read_status - Update the PHY register status snapshot
4047 * @adapter: board private structure
4049 static void e1000_phy_read_status(struct e1000_adapter *adapter)
4051 struct e1000_hw *hw = &adapter->hw;
4052 struct e1000_phy_regs *phy = &adapter->phy_regs;
4054 if ((er32(STATUS) & E1000_STATUS_LU) &&
4055 (adapter->hw.phy.media_type == e1000_media_type_copper)) {
4058 ret_val = e1e_rphy(hw, PHY_CONTROL, &phy->bmcr);
4059 ret_val |= e1e_rphy(hw, PHY_STATUS, &phy->bmsr);
4060 ret_val |= e1e_rphy(hw, PHY_AUTONEG_ADV, &phy->advertise);
4061 ret_val |= e1e_rphy(hw, PHY_LP_ABILITY, &phy->lpa);
4062 ret_val |= e1e_rphy(hw, PHY_AUTONEG_EXP, &phy->expansion);
4063 ret_val |= e1e_rphy(hw, PHY_1000T_CTRL, &phy->ctrl1000);
4064 ret_val |= e1e_rphy(hw, PHY_1000T_STATUS, &phy->stat1000);
4065 ret_val |= e1e_rphy(hw, PHY_EXT_STATUS, &phy->estatus);
4067 e_warn("Error reading PHY register\n");
4070 * Do not read PHY registers if link is not up
4071 * Set values to typical power-on defaults
4073 phy->bmcr = (BMCR_SPEED1000 | BMCR_ANENABLE | BMCR_FULLDPLX);
4074 phy->bmsr = (BMSR_100FULL | BMSR_100HALF | BMSR_10FULL |
4075 BMSR_10HALF | BMSR_ESTATEN | BMSR_ANEGCAPABLE |
4077 phy->advertise = (ADVERTISE_PAUSE_ASYM | ADVERTISE_PAUSE_CAP |
4078 ADVERTISE_ALL | ADVERTISE_CSMA);
4080 phy->expansion = EXPANSION_ENABLENPAGE;
4081 phy->ctrl1000 = ADVERTISE_1000FULL;
4083 phy->estatus = (ESTATUS_1000_TFULL | ESTATUS_1000_THALF);
4087 static void e1000_print_link_info(struct e1000_adapter *adapter)
4089 struct e1000_hw *hw = &adapter->hw;
4090 u32 ctrl = er32(CTRL);
4092 /* Link status message must follow this format for user tools */
4093 printk(KERN_INFO "e1000e: %s NIC Link is Up %d Mbps %s, "
4094 "Flow Control: %s\n",
4095 adapter->netdev->name,
4096 adapter->link_speed,
4097 (adapter->link_duplex == FULL_DUPLEX) ?
4098 "Full Duplex" : "Half Duplex",
4099 ((ctrl & E1000_CTRL_TFCE) && (ctrl & E1000_CTRL_RFCE)) ?
4101 ((ctrl & E1000_CTRL_RFCE) ? "Rx" :
4102 ((ctrl & E1000_CTRL_TFCE) ? "Tx" : "None")));
4105 static bool e1000e_has_link(struct e1000_adapter *adapter)
4107 struct e1000_hw *hw = &adapter->hw;
4108 bool link_active = 0;
4112 * get_link_status is set on LSC (link status) interrupt or
4113 * Rx sequence error interrupt. get_link_status will stay
4114 * false until the check_for_link establishes link
4115 * for copper adapters ONLY
4117 switch (hw->phy.media_type) {
4118 case e1000_media_type_copper:
4119 if (hw->mac.get_link_status) {
4120 ret_val = hw->mac.ops.check_for_link(hw);
4121 link_active = !hw->mac.get_link_status;
4126 case e1000_media_type_fiber:
4127 ret_val = hw->mac.ops.check_for_link(hw);
4128 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
4130 case e1000_media_type_internal_serdes:
4131 ret_val = hw->mac.ops.check_for_link(hw);
4132 link_active = adapter->hw.mac.serdes_has_link;
4135 case e1000_media_type_unknown:
4139 if ((ret_val == E1000_ERR_PHY) && (hw->phy.type == e1000_phy_igp_3) &&
4140 (er32(CTRL) & E1000_PHY_CTRL_GBE_DISABLE)) {
4141 /* See e1000_kmrn_lock_loss_workaround_ich8lan() */
4142 e_info("Gigabit has been disabled, downgrading speed\n");
4148 static void e1000e_enable_receives(struct e1000_adapter *adapter)
4150 /* make sure the receive unit is started */
4151 if ((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4152 (adapter->flags & FLAG_RX_RESTART_NOW)) {
4153 struct e1000_hw *hw = &adapter->hw;
4154 u32 rctl = er32(RCTL);
4155 ew32(RCTL, rctl | E1000_RCTL_EN);
4156 adapter->flags &= ~FLAG_RX_RESTART_NOW;
4160 static void e1000e_check_82574_phy_workaround(struct e1000_adapter *adapter)
4162 struct e1000_hw *hw = &adapter->hw;
4165 * With 82574 controllers, PHY needs to be checked periodically
4166 * for hung state and reset, if two calls return true
4168 if (e1000_check_phy_82574(hw))
4169 adapter->phy_hang_count++;
4171 adapter->phy_hang_count = 0;
4173 if (adapter->phy_hang_count > 1) {
4174 adapter->phy_hang_count = 0;
4175 schedule_work(&adapter->reset_task);
4180 * e1000_watchdog - Timer Call-back
4181 * @data: pointer to adapter cast into an unsigned long
4183 static void e1000_watchdog(unsigned long data)
4185 struct e1000_adapter *adapter = (struct e1000_adapter *) data;
4187 /* Do the rest outside of interrupt context */
4188 schedule_work(&adapter->watchdog_task);
4190 /* TODO: make this use queue_delayed_work() */
4193 static void e1000_watchdog_task(struct work_struct *work)
4195 struct e1000_adapter *adapter = container_of(work,
4196 struct e1000_adapter, watchdog_task);
4197 struct net_device *netdev = adapter->netdev;
4198 struct e1000_mac_info *mac = &adapter->hw.mac;
4199 struct e1000_phy_info *phy = &adapter->hw.phy;
4200 struct e1000_ring *tx_ring = adapter->tx_ring;
4201 struct e1000_hw *hw = &adapter->hw;
4204 if (test_bit(__E1000_DOWN, &adapter->state))
4207 link = e1000e_has_link(adapter);
4208 if ((netif_carrier_ok(netdev)) && link) {
4209 /* Cancel scheduled suspend requests. */
4210 pm_runtime_resume(netdev->dev.parent);
4212 e1000e_enable_receives(adapter);
4216 if ((e1000e_enable_tx_pkt_filtering(hw)) &&
4217 (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id))
4218 e1000_update_mng_vlan(adapter);
4221 if (!netif_carrier_ok(netdev)) {
4224 /* Cancel scheduled suspend requests. */
4225 pm_runtime_resume(netdev->dev.parent);
4227 /* update snapshot of PHY registers on LSC */
4228 e1000_phy_read_status(adapter);
4229 mac->ops.get_link_up_info(&adapter->hw,
4230 &adapter->link_speed,
4231 &adapter->link_duplex);
4232 e1000_print_link_info(adapter);
4234 * On supported PHYs, check for duplex mismatch only
4235 * if link has autonegotiated at 10/100 half
4237 if ((hw->phy.type == e1000_phy_igp_3 ||
4238 hw->phy.type == e1000_phy_bm) &&
4239 (hw->mac.autoneg == true) &&
4240 (adapter->link_speed == SPEED_10 ||
4241 adapter->link_speed == SPEED_100) &&
4242 (adapter->link_duplex == HALF_DUPLEX)) {
4245 e1e_rphy(hw, PHY_AUTONEG_EXP, &autoneg_exp);
4247 if (!(autoneg_exp & NWAY_ER_LP_NWAY_CAPS))
4248 e_info("Autonegotiated half duplex but"
4249 " link partner cannot autoneg. "
4250 " Try forcing full duplex if "
4251 "link gets many collisions.\n");
4254 /* adjust timeout factor according to speed/duplex */
4255 adapter->tx_timeout_factor = 1;
4256 switch (adapter->link_speed) {
4259 adapter->tx_timeout_factor = 16;
4263 adapter->tx_timeout_factor = 10;
4268 * workaround: re-program speed mode bit after
4271 if ((adapter->flags & FLAG_TARC_SPEED_MODE_BIT) &&
4274 tarc0 = er32(TARC(0));
4275 tarc0 &= ~SPEED_MODE_BIT;
4276 ew32(TARC(0), tarc0);
4280 * disable TSO for pcie and 10/100 speeds, to avoid
4281 * some hardware issues
4283 if (!(adapter->flags & FLAG_TSO_FORCE)) {
4284 switch (adapter->link_speed) {
4287 e_info("10/100 speed: disabling TSO\n");
4288 netdev->features &= ~NETIF_F_TSO;
4289 netdev->features &= ~NETIF_F_TSO6;
4292 netdev->features |= NETIF_F_TSO;
4293 netdev->features |= NETIF_F_TSO6;
4302 * enable transmits in the hardware, need to do this
4303 * after setting TARC(0)
4306 tctl |= E1000_TCTL_EN;
4310 * Perform any post-link-up configuration before
4311 * reporting link up.
4313 if (phy->ops.cfg_on_link_up)
4314 phy->ops.cfg_on_link_up(hw);
4316 netif_carrier_on(netdev);
4318 if (!test_bit(__E1000_DOWN, &adapter->state))
4319 mod_timer(&adapter->phy_info_timer,
4320 round_jiffies(jiffies + 2 * HZ));
4323 if (netif_carrier_ok(netdev)) {
4324 adapter->link_speed = 0;
4325 adapter->link_duplex = 0;
4326 /* Link status message must follow this format */
4327 printk(KERN_INFO "e1000e: %s NIC Link is Down\n",
4328 adapter->netdev->name);
4329 netif_carrier_off(netdev);
4330 if (!test_bit(__E1000_DOWN, &adapter->state))
4331 mod_timer(&adapter->phy_info_timer,
4332 round_jiffies(jiffies + 2 * HZ));
4334 if (adapter->flags & FLAG_RX_NEEDS_RESTART)
4335 schedule_work(&adapter->reset_task);
4337 pm_schedule_suspend(netdev->dev.parent,
4343 spin_lock(&adapter->stats64_lock);
4344 e1000e_update_stats(adapter);
4346 mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
4347 adapter->tpt_old = adapter->stats.tpt;
4348 mac->collision_delta = adapter->stats.colc - adapter->colc_old;
4349 adapter->colc_old = adapter->stats.colc;
4351 adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
4352 adapter->gorc_old = adapter->stats.gorc;
4353 adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
4354 adapter->gotc_old = adapter->stats.gotc;
4355 spin_unlock(&adapter->stats64_lock);
4357 e1000e_update_adaptive(&adapter->hw);
4359 if (!netif_carrier_ok(netdev) &&
4360 (e1000_desc_unused(tx_ring) + 1 < tx_ring->count)) {
4362 * We've lost link, so the controller stops DMA,
4363 * but we've got queued Tx work that's never going
4364 * to get done, so reset controller to flush Tx.
4365 * (Do the reset outside of interrupt context).
4367 schedule_work(&adapter->reset_task);
4368 /* return immediately since reset is imminent */
4372 /* Simple mode for Interrupt Throttle Rate (ITR) */
4373 if (adapter->itr_setting == 4) {
4375 * Symmetric Tx/Rx gets a reduced ITR=2000;
4376 * Total asymmetrical Tx or Rx gets ITR=8000;
4377 * everyone else is between 2000-8000.
4379 u32 goc = (adapter->gotc + adapter->gorc) / 10000;
4380 u32 dif = (adapter->gotc > adapter->gorc ?
4381 adapter->gotc - adapter->gorc :
4382 adapter->gorc - adapter->gotc) / 10000;
4383 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
4385 ew32(ITR, 1000000000 / (itr * 256));
4388 /* Cause software interrupt to ensure Rx ring is cleaned */
4389 if (adapter->msix_entries)
4390 ew32(ICS, adapter->rx_ring->ims_val);
4392 ew32(ICS, E1000_ICS_RXDMT0);
4394 /* flush pending descriptors to memory before detecting Tx hang */
4395 e1000e_flush_descriptors(adapter);
4397 /* Force detection of hung controller every watchdog period */
4398 adapter->detect_tx_hung = 1;
4401 * With 82571 controllers, LAA may be overwritten due to controller
4402 * reset from the other port. Set the appropriate LAA in RAR[0]
4404 if (e1000e_get_laa_state_82571(hw))
4405 e1000e_rar_set(hw, adapter->hw.mac.addr, 0);
4407 if (adapter->flags2 & FLAG2_CHECK_PHY_HANG)
4408 e1000e_check_82574_phy_workaround(adapter);
4410 /* Reset the timer */
4411 if (!test_bit(__E1000_DOWN, &adapter->state))
4412 mod_timer(&adapter->watchdog_timer,
4413 round_jiffies(jiffies + 2 * HZ));
4416 #define E1000_TX_FLAGS_CSUM 0x00000001
4417 #define E1000_TX_FLAGS_VLAN 0x00000002
4418 #define E1000_TX_FLAGS_TSO 0x00000004
4419 #define E1000_TX_FLAGS_IPV4 0x00000008
4420 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
4421 #define E1000_TX_FLAGS_VLAN_SHIFT 16
4423 static int e1000_tso(struct e1000_adapter *adapter,
4424 struct sk_buff *skb)
4426 struct e1000_ring *tx_ring = adapter->tx_ring;
4427 struct e1000_context_desc *context_desc;
4428 struct e1000_buffer *buffer_info;
4431 u16 ipcse = 0, tucse, mss;
4432 u8 ipcss, ipcso, tucss, tucso, hdr_len;
4434 if (!skb_is_gso(skb))
4437 if (skb_header_cloned(skb)) {
4438 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4444 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4445 mss = skb_shinfo(skb)->gso_size;
4446 if (skb->protocol == htons(ETH_P_IP)) {
4447 struct iphdr *iph = ip_hdr(skb);
4450 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
4452 cmd_length = E1000_TXD_CMD_IP;
4453 ipcse = skb_transport_offset(skb) - 1;
4454 } else if (skb_is_gso_v6(skb)) {
4455 ipv6_hdr(skb)->payload_len = 0;
4456 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4457 &ipv6_hdr(skb)->daddr,
4461 ipcss = skb_network_offset(skb);
4462 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
4463 tucss = skb_transport_offset(skb);
4464 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
4467 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
4468 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
4470 i = tx_ring->next_to_use;
4471 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4472 buffer_info = &tx_ring->buffer_info[i];
4474 context_desc->lower_setup.ip_fields.ipcss = ipcss;
4475 context_desc->lower_setup.ip_fields.ipcso = ipcso;
4476 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
4477 context_desc->upper_setup.tcp_fields.tucss = tucss;
4478 context_desc->upper_setup.tcp_fields.tucso = tucso;
4479 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
4480 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
4481 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
4482 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
4484 buffer_info->time_stamp = jiffies;
4485 buffer_info->next_to_watch = i;
4488 if (i == tx_ring->count)
4490 tx_ring->next_to_use = i;
4495 static bool e1000_tx_csum(struct e1000_adapter *adapter, struct sk_buff *skb)
4497 struct e1000_ring *tx_ring = adapter->tx_ring;
4498 struct e1000_context_desc *context_desc;
4499 struct e1000_buffer *buffer_info;
4502 u32 cmd_len = E1000_TXD_CMD_DEXT;
4505 if (skb->ip_summed != CHECKSUM_PARTIAL)
4508 if (skb->protocol == cpu_to_be16(ETH_P_8021Q))
4509 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
4511 protocol = skb->protocol;
4514 case cpu_to_be16(ETH_P_IP):
4515 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
4516 cmd_len |= E1000_TXD_CMD_TCP;
4518 case cpu_to_be16(ETH_P_IPV6):
4519 /* XXX not handling all IPV6 headers */
4520 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
4521 cmd_len |= E1000_TXD_CMD_TCP;
4524 if (unlikely(net_ratelimit()))
4525 e_warn("checksum_partial proto=%x!\n",
4526 be16_to_cpu(protocol));
4530 css = skb_checksum_start_offset(skb);
4532 i = tx_ring->next_to_use;
4533 buffer_info = &tx_ring->buffer_info[i];
4534 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
4536 context_desc->lower_setup.ip_config = 0;
4537 context_desc->upper_setup.tcp_fields.tucss = css;
4538 context_desc->upper_setup.tcp_fields.tucso =
4539 css + skb->csum_offset;
4540 context_desc->upper_setup.tcp_fields.tucse = 0;
4541 context_desc->tcp_seg_setup.data = 0;
4542 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
4544 buffer_info->time_stamp = jiffies;
4545 buffer_info->next_to_watch = i;
4548 if (i == tx_ring->count)
4550 tx_ring->next_to_use = i;
4555 #define E1000_MAX_PER_TXD 8192
4556 #define E1000_MAX_TXD_PWR 12
4558 static int e1000_tx_map(struct e1000_adapter *adapter,
4559 struct sk_buff *skb, unsigned int first,
4560 unsigned int max_per_txd, unsigned int nr_frags,
4563 struct e1000_ring *tx_ring = adapter->tx_ring;
4564 struct pci_dev *pdev = adapter->pdev;
4565 struct e1000_buffer *buffer_info;
4566 unsigned int len = skb_headlen(skb);
4567 unsigned int offset = 0, size, count = 0, i;
4568 unsigned int f, bytecount, segs;
4570 i = tx_ring->next_to_use;
4573 buffer_info = &tx_ring->buffer_info[i];
4574 size = min(len, max_per_txd);
4576 buffer_info->length = size;
4577 buffer_info->time_stamp = jiffies;
4578 buffer_info->next_to_watch = i;
4579 buffer_info->dma = dma_map_single(&pdev->dev,
4581 size, DMA_TO_DEVICE);
4582 buffer_info->mapped_as_page = false;
4583 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4592 if (i == tx_ring->count)
4597 for (f = 0; f < nr_frags; f++) {
4598 struct skb_frag_struct *frag;
4600 frag = &skb_shinfo(skb)->frags[f];
4602 offset = frag->page_offset;
4606 if (i == tx_ring->count)
4609 buffer_info = &tx_ring->buffer_info[i];
4610 size = min(len, max_per_txd);
4612 buffer_info->length = size;
4613 buffer_info->time_stamp = jiffies;
4614 buffer_info->next_to_watch = i;
4615 buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
4618 buffer_info->mapped_as_page = true;
4619 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
4628 segs = skb_shinfo(skb)->gso_segs ? : 1;
4629 /* multiply data chunks by size of headers */
4630 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
4632 tx_ring->buffer_info[i].skb = skb;
4633 tx_ring->buffer_info[i].segs = segs;
4634 tx_ring->buffer_info[i].bytecount = bytecount;
4635 tx_ring->buffer_info[first].next_to_watch = i;
4640 dev_err(&pdev->dev, "Tx DMA map failed\n");
4641 buffer_info->dma = 0;
4647 i += tx_ring->count;
4649 buffer_info = &tx_ring->buffer_info[i];
4650 e1000_put_txbuf(adapter, buffer_info);
4656 static void e1000_tx_queue(struct e1000_adapter *adapter,
4657 int tx_flags, int count)
4659 struct e1000_ring *tx_ring = adapter->tx_ring;
4660 struct e1000_tx_desc *tx_desc = NULL;
4661 struct e1000_buffer *buffer_info;
4662 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
4665 if (tx_flags & E1000_TX_FLAGS_TSO) {
4666 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
4668 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4670 if (tx_flags & E1000_TX_FLAGS_IPV4)
4671 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
4674 if (tx_flags & E1000_TX_FLAGS_CSUM) {
4675 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
4676 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
4679 if (tx_flags & E1000_TX_FLAGS_VLAN) {
4680 txd_lower |= E1000_TXD_CMD_VLE;
4681 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
4684 i = tx_ring->next_to_use;
4687 buffer_info = &tx_ring->buffer_info[i];
4688 tx_desc = E1000_TX_DESC(*tx_ring, i);
4689 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4690 tx_desc->lower.data =
4691 cpu_to_le32(txd_lower | buffer_info->length);
4692 tx_desc->upper.data = cpu_to_le32(txd_upper);
4695 if (i == tx_ring->count)
4697 } while (--count > 0);
4699 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
4702 * Force memory writes to complete before letting h/w
4703 * know there are new descriptors to fetch. (Only
4704 * applicable for weak-ordered memory model archs,
4709 tx_ring->next_to_use = i;
4710 writel(i, adapter->hw.hw_addr + tx_ring->tail);
4712 * we need this if more than one processor can write to our tail
4713 * at a time, it synchronizes IO on IA64/Altix systems
4718 #define MINIMUM_DHCP_PACKET_SIZE 282
4719 static int e1000_transfer_dhcp_info(struct e1000_adapter *adapter,
4720 struct sk_buff *skb)
4722 struct e1000_hw *hw = &adapter->hw;
4725 if (vlan_tx_tag_present(skb)) {
4726 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
4727 (adapter->hw.mng_cookie.status &
4728 E1000_MNG_DHCP_COOKIE_STATUS_VLAN)))
4732 if (skb->len <= MINIMUM_DHCP_PACKET_SIZE)
4735 if (((struct ethhdr *) skb->data)->h_proto != htons(ETH_P_IP))
4739 const struct iphdr *ip = (struct iphdr *)((u8 *)skb->data+14);
4742 if (ip->protocol != IPPROTO_UDP)
4745 udp = (struct udphdr *)((u8 *)ip + (ip->ihl << 2));
4746 if (ntohs(udp->dest) != 67)
4749 offset = (u8 *)udp + 8 - skb->data;
4750 length = skb->len - offset;
4751 return e1000e_mng_write_dhcp_info(hw, (u8 *)udp + 8, length);
4757 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
4759 struct e1000_adapter *adapter = netdev_priv(netdev);
4761 netif_stop_queue(netdev);
4763 * Herbert's original patch had:
4764 * smp_mb__after_netif_stop_queue();
4765 * but since that doesn't exist yet, just open code it.
4770 * We need to check again in a case another CPU has just
4771 * made room available.
4773 if (e1000_desc_unused(adapter->tx_ring) < size)
4777 netif_start_queue(netdev);
4778 ++adapter->restart_queue;
4782 static int e1000_maybe_stop_tx(struct net_device *netdev, int size)
4784 struct e1000_adapter *adapter = netdev_priv(netdev);
4786 if (e1000_desc_unused(adapter->tx_ring) >= size)
4788 return __e1000_maybe_stop_tx(netdev, size);
4791 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
4792 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
4793 struct net_device *netdev)
4795 struct e1000_adapter *adapter = netdev_priv(netdev);
4796 struct e1000_ring *tx_ring = adapter->tx_ring;
4798 unsigned int max_per_txd = E1000_MAX_PER_TXD;
4799 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
4800 unsigned int tx_flags = 0;
4801 unsigned int len = skb_headlen(skb);
4802 unsigned int nr_frags;
4808 if (test_bit(__E1000_DOWN, &adapter->state)) {
4809 dev_kfree_skb_any(skb);
4810 return NETDEV_TX_OK;
4813 if (skb->len <= 0) {
4814 dev_kfree_skb_any(skb);
4815 return NETDEV_TX_OK;
4818 mss = skb_shinfo(skb)->gso_size;
4820 * The controller does a simple calculation to
4821 * make sure there is enough room in the FIFO before
4822 * initiating the DMA for each buffer. The calc is:
4823 * 4 = ceil(buffer len/mss). To make sure we don't
4824 * overrun the FIFO, adjust the max buffer len if mss
4829 max_per_txd = min(mss << 2, max_per_txd);
4830 max_txd_pwr = fls(max_per_txd) - 1;
4833 * TSO Workaround for 82571/2/3 Controllers -- if skb->data
4834 * points to just header, pull a few bytes of payload from
4835 * frags into skb->data
4837 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
4839 * we do this workaround for ES2LAN, but it is un-necessary,
4840 * avoiding it could save a lot of cycles
4842 if (skb->data_len && (hdr_len == len)) {
4843 unsigned int pull_size;
4845 pull_size = min((unsigned int)4, skb->data_len);
4846 if (!__pskb_pull_tail(skb, pull_size)) {
4847 e_err("__pskb_pull_tail failed.\n");
4848 dev_kfree_skb_any(skb);
4849 return NETDEV_TX_OK;
4851 len = skb_headlen(skb);
4855 /* reserve a descriptor for the offload context */
4856 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
4860 count += TXD_USE_COUNT(len, max_txd_pwr);
4862 nr_frags = skb_shinfo(skb)->nr_frags;
4863 for (f = 0; f < nr_frags; f++)
4864 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
4867 if (adapter->hw.mac.tx_pkt_filtering)
4868 e1000_transfer_dhcp_info(adapter, skb);
4871 * need: count + 2 desc gap to keep tail from touching
4872 * head, otherwise try next time
4874 if (e1000_maybe_stop_tx(netdev, count + 2))
4875 return NETDEV_TX_BUSY;
4877 if (vlan_tx_tag_present(skb)) {
4878 tx_flags |= E1000_TX_FLAGS_VLAN;
4879 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
4882 first = tx_ring->next_to_use;
4884 tso = e1000_tso(adapter, skb);
4886 dev_kfree_skb_any(skb);
4887 return NETDEV_TX_OK;
4891 tx_flags |= E1000_TX_FLAGS_TSO;
4892 else if (e1000_tx_csum(adapter, skb))
4893 tx_flags |= E1000_TX_FLAGS_CSUM;
4896 * Old method was to assume IPv4 packet by default if TSO was enabled.
4897 * 82571 hardware supports TSO capabilities for IPv6 as well...
4898 * no longer assume, we must.
4900 if (skb->protocol == htons(ETH_P_IP))
4901 tx_flags |= E1000_TX_FLAGS_IPV4;
4903 /* if count is 0 then mapping error has occurred */
4904 count = e1000_tx_map(adapter, skb, first, max_per_txd, nr_frags, mss);
4906 e1000_tx_queue(adapter, tx_flags, count);
4907 /* Make sure there is space in the ring for the next send. */
4908 e1000_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 2);
4911 dev_kfree_skb_any(skb);
4912 tx_ring->buffer_info[first].time_stamp = 0;
4913 tx_ring->next_to_use = first;
4916 return NETDEV_TX_OK;
4920 * e1000_tx_timeout - Respond to a Tx Hang
4921 * @netdev: network interface device structure
4923 static void e1000_tx_timeout(struct net_device *netdev)
4925 struct e1000_adapter *adapter = netdev_priv(netdev);
4927 /* Do the reset outside of interrupt context */
4928 adapter->tx_timeout_count++;
4929 schedule_work(&adapter->reset_task);
4932 static void e1000_reset_task(struct work_struct *work)
4934 struct e1000_adapter *adapter;
4935 adapter = container_of(work, struct e1000_adapter, reset_task);
4937 /* don't run the task if already down */
4938 if (test_bit(__E1000_DOWN, &adapter->state))
4941 if (!((adapter->flags & FLAG_RX_NEEDS_RESTART) &&
4942 (adapter->flags & FLAG_RX_RESTART_NOW))) {
4943 e1000e_dump(adapter);
4944 e_err("Reset adapter\n");
4946 e1000e_reinit_locked(adapter);
4950 * e1000_get_stats64 - Get System Network Statistics
4951 * @netdev: network interface device structure
4952 * @stats: rtnl_link_stats64 pointer
4954 * Returns the address of the device statistics structure.
4956 struct rtnl_link_stats64 *e1000e_get_stats64(struct net_device *netdev,
4957 struct rtnl_link_stats64 *stats)
4959 struct e1000_adapter *adapter = netdev_priv(netdev);
4961 memset(stats, 0, sizeof(struct rtnl_link_stats64));
4962 spin_lock(&adapter->stats64_lock);
4963 e1000e_update_stats(adapter);
4964 /* Fill out the OS statistics structure */
4965 stats->rx_bytes = adapter->stats.gorc;
4966 stats->rx_packets = adapter->stats.gprc;
4967 stats->tx_bytes = adapter->stats.gotc;
4968 stats->tx_packets = adapter->stats.gptc;
4969 stats->multicast = adapter->stats.mprc;
4970 stats->collisions = adapter->stats.colc;
4975 * RLEC on some newer hardware can be incorrect so build
4976 * our own version based on RUC and ROC
4978 stats->rx_errors = adapter->stats.rxerrc +
4979 adapter->stats.crcerrs + adapter->stats.algnerrc +
4980 adapter->stats.ruc + adapter->stats.roc +
4981 adapter->stats.cexterr;
4982 stats->rx_length_errors = adapter->stats.ruc +
4984 stats->rx_crc_errors = adapter->stats.crcerrs;
4985 stats->rx_frame_errors = adapter->stats.algnerrc;
4986 stats->rx_missed_errors = adapter->stats.mpc;
4989 stats->tx_errors = adapter->stats.ecol +
4990 adapter->stats.latecol;
4991 stats->tx_aborted_errors = adapter->stats.ecol;
4992 stats->tx_window_errors = adapter->stats.latecol;
4993 stats->tx_carrier_errors = adapter->stats.tncrs;
4995 /* Tx Dropped needs to be maintained elsewhere */
4997 spin_unlock(&adapter->stats64_lock);
5002 * e1000_change_mtu - Change the Maximum Transfer Unit
5003 * @netdev: network interface device structure
5004 * @new_mtu: new value for maximum frame size
5006 * Returns 0 on success, negative on failure
5008 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
5010 struct e1000_adapter *adapter = netdev_priv(netdev);
5011 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5013 /* Jumbo frame support */
5014 if ((max_frame > ETH_FRAME_LEN + ETH_FCS_LEN) &&
5015 !(adapter->flags & FLAG_HAS_JUMBO_FRAMES)) {
5016 e_err("Jumbo Frames not supported.\n");
5020 /* Supported frame sizes */
5021 if ((new_mtu < ETH_ZLEN + ETH_FCS_LEN + VLAN_HLEN) ||
5022 (max_frame > adapter->max_hw_frame_size)) {
5023 e_err("Unsupported MTU setting\n");
5027 /* Jumbo frame workaround on 82579 requires CRC be stripped */
5028 if ((adapter->hw.mac.type == e1000_pch2lan) &&
5029 !(adapter->flags2 & FLAG2_CRC_STRIPPING) &&
5030 (new_mtu > ETH_DATA_LEN)) {
5031 e_err("Jumbo Frames not supported on 82579 when CRC "
5032 "stripping is disabled.\n");
5036 /* 82573 Errata 17 */
5037 if (((adapter->hw.mac.type == e1000_82573) ||
5038 (adapter->hw.mac.type == e1000_82574)) &&
5039 (max_frame > ETH_FRAME_LEN + ETH_FCS_LEN)) {
5040 adapter->flags2 |= FLAG2_DISABLE_ASPM_L1;
5041 e1000e_disable_aspm(adapter->pdev, PCIE_LINK_STATE_L1);
5044 while (test_and_set_bit(__E1000_RESETTING, &adapter->state))
5045 usleep_range(1000, 2000);
5046 /* e1000e_down -> e1000e_reset dependent on max_frame_size & mtu */
5047 adapter->max_frame_size = max_frame;
5048 e_info("changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5049 netdev->mtu = new_mtu;
5050 if (netif_running(netdev))
5051 e1000e_down(adapter);
5054 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
5055 * means we reserve 2 more, this pushes us to allocate from the next
5057 * i.e. RXBUFFER_2048 --> size-4096 slab
5058 * However with the new *_jumbo_rx* routines, jumbo receives will use
5062 if (max_frame <= 2048)
5063 adapter->rx_buffer_len = 2048;
5065 adapter->rx_buffer_len = 4096;
5067 /* adjust allocation if LPE protects us, and we aren't using SBP */
5068 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
5069 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
5070 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN
5073 if (netif_running(netdev))
5076 e1000e_reset(adapter);
5078 clear_bit(__E1000_RESETTING, &adapter->state);
5083 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
5086 struct e1000_adapter *adapter = netdev_priv(netdev);
5087 struct mii_ioctl_data *data = if_mii(ifr);
5089 if (adapter->hw.phy.media_type != e1000_media_type_copper)
5094 data->phy_id = adapter->hw.phy.addr;
5097 e1000_phy_read_status(adapter);
5099 switch (data->reg_num & 0x1F) {
5101 data->val_out = adapter->phy_regs.bmcr;
5104 data->val_out = adapter->phy_regs.bmsr;
5107 data->val_out = (adapter->hw.phy.id >> 16);
5110 data->val_out = (adapter->hw.phy.id & 0xFFFF);
5113 data->val_out = adapter->phy_regs.advertise;
5116 data->val_out = adapter->phy_regs.lpa;
5119 data->val_out = adapter->phy_regs.expansion;
5122 data->val_out = adapter->phy_regs.ctrl1000;
5125 data->val_out = adapter->phy_regs.stat1000;
5128 data->val_out = adapter->phy_regs.estatus;
5141 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5147 return e1000_mii_ioctl(netdev, ifr, cmd);
5153 static int e1000_init_phy_wakeup(struct e1000_adapter *adapter, u32 wufc)
5155 struct e1000_hw *hw = &adapter->hw;
5160 /* copy MAC RARs to PHY RARs */
5161 e1000_copy_rx_addrs_to_phy_ich8lan(hw);
5163 /* copy MAC MTA to PHY MTA */
5164 for (i = 0; i < adapter->hw.mac.mta_reg_count; i++) {
5165 mac_reg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i);
5166 e1e_wphy(hw, BM_MTA(i), (u16)(mac_reg & 0xFFFF));
5167 e1e_wphy(hw, BM_MTA(i) + 1, (u16)((mac_reg >> 16) & 0xFFFF));
5170 /* configure PHY Rx Control register */
5171 e1e_rphy(&adapter->hw, BM_RCTL, &phy_reg);
5172 mac_reg = er32(RCTL);
5173 if (mac_reg & E1000_RCTL_UPE)
5174 phy_reg |= BM_RCTL_UPE;
5175 if (mac_reg & E1000_RCTL_MPE)
5176 phy_reg |= BM_RCTL_MPE;
5177 phy_reg &= ~(BM_RCTL_MO_MASK);
5178 if (mac_reg & E1000_RCTL_MO_3)
5179 phy_reg |= (((mac_reg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT)
5180 << BM_RCTL_MO_SHIFT);
5181 if (mac_reg & E1000_RCTL_BAM)
5182 phy_reg |= BM_RCTL_BAM;
5183 if (mac_reg & E1000_RCTL_PMCF)
5184 phy_reg |= BM_RCTL_PMCF;
5185 mac_reg = er32(CTRL);
5186 if (mac_reg & E1000_CTRL_RFCE)
5187 phy_reg |= BM_RCTL_RFCE;
5188 e1e_wphy(&adapter->hw, BM_RCTL, phy_reg);
5190 /* enable PHY wakeup in MAC register */
5192 ew32(WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN);
5194 /* configure and enable PHY wakeup in PHY registers */
5195 e1e_wphy(&adapter->hw, BM_WUFC, wufc);
5196 e1e_wphy(&adapter->hw, BM_WUC, E1000_WUC_PME_EN);
5198 /* activate PHY wakeup */
5199 retval = hw->phy.ops.acquire(hw);
5201 e_err("Could not acquire PHY\n");
5204 e1000e_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT,
5205 (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT));
5206 retval = e1000e_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &phy_reg);
5208 e_err("Could not read PHY page 769\n");
5211 phy_reg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT;
5212 retval = e1000e_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, phy_reg);
5214 e_err("Could not set PHY Host Wakeup bit\n");
5216 hw->phy.ops.release(hw);
5221 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake,
5224 struct net_device *netdev = pci_get_drvdata(pdev);
5225 struct e1000_adapter *adapter = netdev_priv(netdev);
5226 struct e1000_hw *hw = &adapter->hw;
5227 u32 ctrl, ctrl_ext, rctl, status;
5228 /* Runtime suspend should only enable wakeup for link changes */
5229 u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
5232 netif_device_detach(netdev);
5234 if (netif_running(netdev)) {
5235 WARN_ON(test_bit(__E1000_RESETTING, &adapter->state));
5236 e1000e_down(adapter);
5237 e1000_free_irq(adapter);
5239 e1000e_reset_interrupt_capability(adapter);
5241 retval = pci_save_state(pdev);
5245 status = er32(STATUS);
5246 if (status & E1000_STATUS_LU)
5247 wufc &= ~E1000_WUFC_LNKC;
5250 e1000_setup_rctl(adapter);
5251 e1000_set_multi(netdev);
5253 /* turn on all-multi mode if wake on multicast is enabled */
5254 if (wufc & E1000_WUFC_MC) {
5256 rctl |= E1000_RCTL_MPE;
5261 /* advertise wake from D3Cold */
5262 #define E1000_CTRL_ADVD3WUC 0x00100000
5263 /* phy power management enable */
5264 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5265 ctrl |= E1000_CTRL_ADVD3WUC;
5266 if (!(adapter->flags2 & FLAG2_HAS_PHY_WAKEUP))
5267 ctrl |= E1000_CTRL_EN_PHY_PWR_MGMT;
5270 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
5271 adapter->hw.phy.media_type ==
5272 e1000_media_type_internal_serdes) {
5273 /* keep the laser running in D3 */
5274 ctrl_ext = er32(CTRL_EXT);
5275 ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA;
5276 ew32(CTRL_EXT, ctrl_ext);
5279 if (adapter->flags & FLAG_IS_ICH)
5280 e1000e_disable_gig_wol_ich8lan(&adapter->hw);
5282 /* Allow time for pending master requests to run */
5283 e1000e_disable_pcie_master(&adapter->hw);
5285 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5286 /* enable wakeup by the PHY */
5287 retval = e1000_init_phy_wakeup(adapter, wufc);
5291 /* enable wakeup by the MAC */
5293 ew32(WUC, E1000_WUC_PME_EN);
5300 *enable_wake = !!wufc;
5302 /* make sure adapter isn't asleep if manageability is enabled */
5303 if ((adapter->flags & FLAG_MNG_PT_ENABLED) ||
5304 (hw->mac.ops.check_mng_mode(hw)))
5305 *enable_wake = true;
5307 if (adapter->hw.phy.type == e1000_phy_igp_3)
5308 e1000e_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw);
5311 * Release control of h/w to f/w. If f/w is AMT enabled, this
5312 * would have already happened in close and is redundant.
5314 e1000e_release_hw_control(adapter);
5316 pci_disable_device(pdev);
5321 static void e1000_power_off(struct pci_dev *pdev, bool sleep, bool wake)
5323 if (sleep && wake) {
5324 pci_prepare_to_sleep(pdev);
5328 pci_wake_from_d3(pdev, wake);
5329 pci_set_power_state(pdev, PCI_D3hot);
5332 static void e1000_complete_shutdown(struct pci_dev *pdev, bool sleep,
5335 struct net_device *netdev = pci_get_drvdata(pdev);
5336 struct e1000_adapter *adapter = netdev_priv(netdev);
5339 * The pci-e switch on some quad port adapters will report a
5340 * correctable error when the MAC transitions from D0 to D3. To
5341 * prevent this we need to mask off the correctable errors on the
5342 * downstream port of the pci-e switch.
5344 if (adapter->flags & FLAG_IS_QUAD_PORT) {
5345 struct pci_dev *us_dev = pdev->bus->self;
5346 int pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
5349 pci_read_config_word(us_dev, pos + PCI_EXP_DEVCTL, &devctl);
5350 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL,
5351 (devctl & ~PCI_EXP_DEVCTL_CERE));
5353 e1000_power_off(pdev, sleep, wake);
5355 pci_write_config_word(us_dev, pos + PCI_EXP_DEVCTL, devctl);
5357 e1000_power_off(pdev, sleep, wake);
5361 #ifdef CONFIG_PCIEASPM
5362 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5364 pci_disable_link_state_locked(pdev, state);
5367 static void __e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5373 * Both device and parent should have the same ASPM setting.
5374 * Disable ASPM in downstream component first and then upstream.
5376 pos = pci_pcie_cap(pdev);
5377 pci_read_config_word(pdev, pos + PCI_EXP_LNKCTL, ®16);
5379 pci_write_config_word(pdev, pos + PCI_EXP_LNKCTL, reg16);
5381 if (!pdev->bus->self)
5384 pos = pci_pcie_cap(pdev->bus->self);
5385 pci_read_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, ®16);
5387 pci_write_config_word(pdev->bus->self, pos + PCI_EXP_LNKCTL, reg16);
5390 static void e1000e_disable_aspm(struct pci_dev *pdev, u16 state)
5392 dev_info(&pdev->dev, "Disabling ASPM %s %s\n",
5393 (state & PCIE_LINK_STATE_L0S) ? "L0s" : "",
5394 (state & PCIE_LINK_STATE_L1) ? "L1" : "");
5396 __e1000e_disable_aspm(pdev, state);
5400 static bool e1000e_pm_ready(struct e1000_adapter *adapter)
5402 return !!adapter->tx_ring->buffer_info;
5405 static int __e1000_resume(struct pci_dev *pdev)
5407 struct net_device *netdev = pci_get_drvdata(pdev);
5408 struct e1000_adapter *adapter = netdev_priv(netdev);
5409 struct e1000_hw *hw = &adapter->hw;
5410 u16 aspm_disable_flag = 0;
5413 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5414 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5415 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5416 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5417 if (aspm_disable_flag)
5418 e1000e_disable_aspm(pdev, aspm_disable_flag);
5420 pci_set_power_state(pdev, PCI_D0);
5421 pci_restore_state(pdev);
5422 pci_save_state(pdev);
5424 e1000e_set_interrupt_capability(adapter);
5425 if (netif_running(netdev)) {
5426 err = e1000_request_irq(adapter);
5431 e1000e_power_up_phy(adapter);
5433 /* report the system wakeup cause from S3/S4 */
5434 if (adapter->flags2 & FLAG2_HAS_PHY_WAKEUP) {
5437 e1e_rphy(&adapter->hw, BM_WUS, &phy_data);
5439 e_info("PHY Wakeup cause - %s\n",
5440 phy_data & E1000_WUS_EX ? "Unicast Packet" :
5441 phy_data & E1000_WUS_MC ? "Multicast Packet" :
5442 phy_data & E1000_WUS_BC ? "Broadcast Packet" :
5443 phy_data & E1000_WUS_MAG ? "Magic Packet" :
5444 phy_data & E1000_WUS_LNKC ? "Link Status "
5445 " Change" : "other");
5447 e1e_wphy(&adapter->hw, BM_WUS, ~0);
5449 u32 wus = er32(WUS);
5451 e_info("MAC Wakeup cause - %s\n",
5452 wus & E1000_WUS_EX ? "Unicast Packet" :
5453 wus & E1000_WUS_MC ? "Multicast Packet" :
5454 wus & E1000_WUS_BC ? "Broadcast Packet" :
5455 wus & E1000_WUS_MAG ? "Magic Packet" :
5456 wus & E1000_WUS_LNKC ? "Link Status Change" :
5462 e1000e_reset(adapter);
5464 e1000_init_manageability_pt(adapter);
5466 if (netif_running(netdev))
5469 netif_device_attach(netdev);
5472 * If the controller has AMT, do not set DRV_LOAD until the interface
5473 * is up. For all other cases, let the f/w know that the h/w is now
5474 * under the control of the driver.
5476 if (!(adapter->flags & FLAG_HAS_AMT))
5477 e1000e_get_hw_control(adapter);
5482 #ifdef CONFIG_PM_SLEEP
5483 static int e1000_suspend(struct device *dev)
5485 struct pci_dev *pdev = to_pci_dev(dev);
5489 retval = __e1000_shutdown(pdev, &wake, false);
5491 e1000_complete_shutdown(pdev, true, wake);
5496 static int e1000_resume(struct device *dev)
5498 struct pci_dev *pdev = to_pci_dev(dev);
5499 struct net_device *netdev = pci_get_drvdata(pdev);
5500 struct e1000_adapter *adapter = netdev_priv(netdev);
5502 if (e1000e_pm_ready(adapter))
5503 adapter->idle_check = true;
5505 return __e1000_resume(pdev);
5507 #endif /* CONFIG_PM_SLEEP */
5509 #ifdef CONFIG_PM_RUNTIME
5510 static int e1000_runtime_suspend(struct device *dev)
5512 struct pci_dev *pdev = to_pci_dev(dev);
5513 struct net_device *netdev = pci_get_drvdata(pdev);
5514 struct e1000_adapter *adapter = netdev_priv(netdev);
5516 if (e1000e_pm_ready(adapter)) {
5519 __e1000_shutdown(pdev, &wake, true);
5525 static int e1000_idle(struct device *dev)
5527 struct pci_dev *pdev = to_pci_dev(dev);
5528 struct net_device *netdev = pci_get_drvdata(pdev);
5529 struct e1000_adapter *adapter = netdev_priv(netdev);
5531 if (!e1000e_pm_ready(adapter))
5534 if (adapter->idle_check) {
5535 adapter->idle_check = false;
5536 if (!e1000e_has_link(adapter))
5537 pm_schedule_suspend(dev, MSEC_PER_SEC);
5543 static int e1000_runtime_resume(struct device *dev)
5545 struct pci_dev *pdev = to_pci_dev(dev);
5546 struct net_device *netdev = pci_get_drvdata(pdev);
5547 struct e1000_adapter *adapter = netdev_priv(netdev);
5549 if (!e1000e_pm_ready(adapter))
5552 adapter->idle_check = !dev->power.runtime_auto;
5553 return __e1000_resume(pdev);
5555 #endif /* CONFIG_PM_RUNTIME */
5556 #endif /* CONFIG_PM */
5558 static void e1000_shutdown(struct pci_dev *pdev)
5562 __e1000_shutdown(pdev, &wake, false);
5564 if (system_state == SYSTEM_POWER_OFF)
5565 e1000_complete_shutdown(pdev, false, wake);
5568 #ifdef CONFIG_NET_POLL_CONTROLLER
5570 static irqreturn_t e1000_intr_msix(int irq, void *data)
5572 struct net_device *netdev = data;
5573 struct e1000_adapter *adapter = netdev_priv(netdev);
5575 if (adapter->msix_entries) {
5576 int vector, msix_irq;
5579 msix_irq = adapter->msix_entries[vector].vector;
5580 disable_irq(msix_irq);
5581 e1000_intr_msix_rx(msix_irq, netdev);
5582 enable_irq(msix_irq);
5585 msix_irq = adapter->msix_entries[vector].vector;
5586 disable_irq(msix_irq);
5587 e1000_intr_msix_tx(msix_irq, netdev);
5588 enable_irq(msix_irq);
5591 msix_irq = adapter->msix_entries[vector].vector;
5592 disable_irq(msix_irq);
5593 e1000_msix_other(msix_irq, netdev);
5594 enable_irq(msix_irq);
5601 * Polling 'interrupt' - used by things like netconsole to send skbs
5602 * without having to re-enable interrupts. It's not called while
5603 * the interrupt routine is executing.
5605 static void e1000_netpoll(struct net_device *netdev)
5607 struct e1000_adapter *adapter = netdev_priv(netdev);
5609 switch (adapter->int_mode) {
5610 case E1000E_INT_MODE_MSIX:
5611 e1000_intr_msix(adapter->pdev->irq, netdev);
5613 case E1000E_INT_MODE_MSI:
5614 disable_irq(adapter->pdev->irq);
5615 e1000_intr_msi(adapter->pdev->irq, netdev);
5616 enable_irq(adapter->pdev->irq);
5618 default: /* E1000E_INT_MODE_LEGACY */
5619 disable_irq(adapter->pdev->irq);
5620 e1000_intr(adapter->pdev->irq, netdev);
5621 enable_irq(adapter->pdev->irq);
5628 * e1000_io_error_detected - called when PCI error is detected
5629 * @pdev: Pointer to PCI device
5630 * @state: The current pci connection state
5632 * This function is called after a PCI bus error affecting
5633 * this device has been detected.
5635 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5636 pci_channel_state_t state)
5638 struct net_device *netdev = pci_get_drvdata(pdev);
5639 struct e1000_adapter *adapter = netdev_priv(netdev);
5641 netif_device_detach(netdev);
5643 if (state == pci_channel_io_perm_failure)
5644 return PCI_ERS_RESULT_DISCONNECT;
5646 if (netif_running(netdev))
5647 e1000e_down(adapter);
5648 pci_disable_device(pdev);
5650 /* Request a slot slot reset. */
5651 return PCI_ERS_RESULT_NEED_RESET;
5655 * e1000_io_slot_reset - called after the pci bus has been reset.
5656 * @pdev: Pointer to PCI device
5658 * Restart the card from scratch, as if from a cold-boot. Implementation
5659 * resembles the first-half of the e1000_resume routine.
5661 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5663 struct net_device *netdev = pci_get_drvdata(pdev);
5664 struct e1000_adapter *adapter = netdev_priv(netdev);
5665 struct e1000_hw *hw = &adapter->hw;
5666 u16 aspm_disable_flag = 0;
5668 pci_ers_result_t result;
5670 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L0S)
5671 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5672 if (adapter->flags2 & FLAG2_DISABLE_ASPM_L1)
5673 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5674 if (aspm_disable_flag)
5675 e1000e_disable_aspm(pdev, aspm_disable_flag);
5677 err = pci_enable_device_mem(pdev);
5680 "Cannot re-enable PCI device after reset.\n");
5681 result = PCI_ERS_RESULT_DISCONNECT;
5683 pci_set_master(pdev);
5684 pdev->state_saved = true;
5685 pci_restore_state(pdev);
5687 pci_enable_wake(pdev, PCI_D3hot, 0);
5688 pci_enable_wake(pdev, PCI_D3cold, 0);
5690 e1000e_reset(adapter);
5692 result = PCI_ERS_RESULT_RECOVERED;
5695 pci_cleanup_aer_uncorrect_error_status(pdev);
5701 * e1000_io_resume - called when traffic can start flowing again.
5702 * @pdev: Pointer to PCI device
5704 * This callback is called when the error recovery driver tells us that
5705 * its OK to resume normal operation. Implementation resembles the
5706 * second-half of the e1000_resume routine.
5708 static void e1000_io_resume(struct pci_dev *pdev)
5710 struct net_device *netdev = pci_get_drvdata(pdev);
5711 struct e1000_adapter *adapter = netdev_priv(netdev);
5713 e1000_init_manageability_pt(adapter);
5715 if (netif_running(netdev)) {
5716 if (e1000e_up(adapter)) {
5718 "can't bring device back up after reset\n");
5723 netif_device_attach(netdev);
5726 * If the controller has AMT, do not set DRV_LOAD until the interface
5727 * is up. For all other cases, let the f/w know that the h/w is now
5728 * under the control of the driver.
5730 if (!(adapter->flags & FLAG_HAS_AMT))
5731 e1000e_get_hw_control(adapter);
5735 static void e1000_print_device_info(struct e1000_adapter *adapter)
5737 struct e1000_hw *hw = &adapter->hw;
5738 struct net_device *netdev = adapter->netdev;
5740 u8 pba_str[E1000_PBANUM_LENGTH];
5742 /* print bus type/speed/width info */
5743 e_info("(PCI Express:2.5GT/s:%s) %pM\n",
5745 ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
5749 e_info("Intel(R) PRO/%s Network Connection\n",
5750 (hw->phy.type == e1000_phy_ife) ? "10/100" : "1000");
5751 ret_val = e1000_read_pba_string_generic(hw, pba_str,
5752 E1000_PBANUM_LENGTH);
5754 strncpy((char *)pba_str, "Unknown", sizeof(pba_str) - 1);
5755 e_info("MAC: %d, PHY: %d, PBA No: %s\n",
5756 hw->mac.type, hw->phy.type, pba_str);
5759 static void e1000_eeprom_checks(struct e1000_adapter *adapter)
5761 struct e1000_hw *hw = &adapter->hw;
5765 if (hw->mac.type != e1000_82573)
5768 ret_val = e1000_read_nvm(hw, NVM_INIT_CONTROL2_REG, 1, &buf);
5769 if (!ret_val && (!(le16_to_cpu(buf) & (1 << 0)))) {
5770 /* Deep Smart Power Down (DSPD) */
5771 dev_warn(&adapter->pdev->dev,
5772 "Warning: detected DSPD enabled in EEPROM\n");
5776 static const struct net_device_ops e1000e_netdev_ops = {
5777 .ndo_open = e1000_open,
5778 .ndo_stop = e1000_close,
5779 .ndo_start_xmit = e1000_xmit_frame,
5780 .ndo_get_stats64 = e1000e_get_stats64,
5781 .ndo_set_multicast_list = e1000_set_multi,
5782 .ndo_set_mac_address = e1000_set_mac,
5783 .ndo_change_mtu = e1000_change_mtu,
5784 .ndo_do_ioctl = e1000_ioctl,
5785 .ndo_tx_timeout = e1000_tx_timeout,
5786 .ndo_validate_addr = eth_validate_addr,
5788 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
5789 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
5790 #ifdef CONFIG_NET_POLL_CONTROLLER
5791 .ndo_poll_controller = e1000_netpoll,
5796 * e1000_probe - Device Initialization Routine
5797 * @pdev: PCI device information struct
5798 * @ent: entry in e1000_pci_tbl
5800 * Returns 0 on success, negative on failure
5802 * e1000_probe initializes an adapter identified by a pci_dev structure.
5803 * The OS initialization, configuring of the adapter private structure,
5804 * and a hardware reset occur.
5806 static int __devinit e1000_probe(struct pci_dev *pdev,
5807 const struct pci_device_id *ent)
5809 struct net_device *netdev;
5810 struct e1000_adapter *adapter;
5811 struct e1000_hw *hw;
5812 const struct e1000_info *ei = e1000_info_tbl[ent->driver_data];
5813 resource_size_t mmio_start, mmio_len;
5814 resource_size_t flash_start, flash_len;
5816 static int cards_found;
5817 u16 aspm_disable_flag = 0;
5818 int i, err, pci_using_dac;
5819 u16 eeprom_data = 0;
5820 u16 eeprom_apme_mask = E1000_EEPROM_APME;
5822 if (ei->flags2 & FLAG2_DISABLE_ASPM_L0S)
5823 aspm_disable_flag = PCIE_LINK_STATE_L0S;
5824 if (ei->flags2 & FLAG2_DISABLE_ASPM_L1)
5825 aspm_disable_flag |= PCIE_LINK_STATE_L1;
5826 if (aspm_disable_flag)
5827 e1000e_disable_aspm(pdev, aspm_disable_flag);
5829 err = pci_enable_device_mem(pdev);
5834 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
5836 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
5840 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
5842 err = dma_set_coherent_mask(&pdev->dev,
5845 dev_err(&pdev->dev, "No usable DMA "
5846 "configuration, aborting\n");
5852 err = pci_request_selected_regions_exclusive(pdev,
5853 pci_select_bars(pdev, IORESOURCE_MEM),
5854 e1000e_driver_name);
5858 /* AER (Advanced Error Reporting) hooks */
5859 pci_enable_pcie_error_reporting(pdev);
5861 pci_set_master(pdev);
5862 /* PCI config space info */
5863 err = pci_save_state(pdev);
5865 goto err_alloc_etherdev;
5868 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
5870 goto err_alloc_etherdev;
5872 SET_NETDEV_DEV(netdev, &pdev->dev);
5874 netdev->irq = pdev->irq;
5876 pci_set_drvdata(pdev, netdev);
5877 adapter = netdev_priv(netdev);
5879 adapter->netdev = netdev;
5880 adapter->pdev = pdev;
5882 adapter->pba = ei->pba;
5883 adapter->flags = ei->flags;
5884 adapter->flags2 = ei->flags2;
5885 adapter->hw.adapter = adapter;
5886 adapter->hw.mac.type = ei->mac;
5887 adapter->max_hw_frame_size = ei->max_hw_frame_size;
5888 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
5890 mmio_start = pci_resource_start(pdev, 0);
5891 mmio_len = pci_resource_len(pdev, 0);
5894 adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
5895 if (!adapter->hw.hw_addr)
5898 if ((adapter->flags & FLAG_HAS_FLASH) &&
5899 (pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
5900 flash_start = pci_resource_start(pdev, 1);
5901 flash_len = pci_resource_len(pdev, 1);
5902 adapter->hw.flash_address = ioremap(flash_start, flash_len);
5903 if (!adapter->hw.flash_address)
5907 /* construct the net_device struct */
5908 netdev->netdev_ops = &e1000e_netdev_ops;
5909 e1000e_set_ethtool_ops(netdev);
5910 netdev->watchdog_timeo = 5 * HZ;
5911 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
5912 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
5914 netdev->mem_start = mmio_start;
5915 netdev->mem_end = mmio_start + mmio_len;
5917 adapter->bd_number = cards_found++;
5919 e1000e_check_options(adapter);
5921 /* setup adapter struct */
5922 err = e1000_sw_init(adapter);
5926 memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
5927 memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
5928 memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
5930 err = ei->get_variants(adapter);
5934 if ((adapter->flags & FLAG_IS_ICH) &&
5935 (adapter->flags & FLAG_READ_ONLY_NVM))
5936 e1000e_write_protect_nvm_ich8lan(&adapter->hw);
5938 hw->mac.ops.get_bus_info(&adapter->hw);
5940 adapter->hw.phy.autoneg_wait_to_complete = 0;
5942 /* Copper options */
5943 if (adapter->hw.phy.media_type == e1000_media_type_copper) {
5944 adapter->hw.phy.mdix = AUTO_ALL_MODES;
5945 adapter->hw.phy.disable_polarity_correction = 0;
5946 adapter->hw.phy.ms_type = e1000_ms_hw_default;
5949 if (e1000_check_reset_block(&adapter->hw))
5950 e_info("PHY reset is blocked due to SOL/IDER session.\n");
5952 netdev->features = NETIF_F_SG |
5954 NETIF_F_HW_VLAN_TX |
5957 if (adapter->flags & FLAG_HAS_HW_VLAN_FILTER)
5958 netdev->features |= NETIF_F_HW_VLAN_FILTER;
5960 netdev->features |= NETIF_F_TSO;
5961 netdev->features |= NETIF_F_TSO6;
5963 netdev->vlan_features |= NETIF_F_TSO;
5964 netdev->vlan_features |= NETIF_F_TSO6;
5965 netdev->vlan_features |= NETIF_F_HW_CSUM;
5966 netdev->vlan_features |= NETIF_F_SG;
5968 if (pci_using_dac) {
5969 netdev->features |= NETIF_F_HIGHDMA;
5970 netdev->vlan_features |= NETIF_F_HIGHDMA;
5973 if (e1000e_enable_mng_pass_thru(&adapter->hw))
5974 adapter->flags |= FLAG_MNG_PT_ENABLED;
5977 * before reading the NVM, reset the controller to
5978 * put the device in a known good starting state
5980 adapter->hw.mac.ops.reset_hw(&adapter->hw);
5983 * systems with ASPM and others may see the checksum fail on the first
5984 * attempt. Let's give it a few tries
5987 if (e1000_validate_nvm_checksum(&adapter->hw) >= 0)
5990 e_err("The NVM Checksum Is Not Valid\n");
5996 e1000_eeprom_checks(adapter);
5998 /* copy the MAC address */
5999 if (e1000e_read_mac_addr(&adapter->hw))
6000 e_err("NVM Read Error while reading MAC address\n");
6002 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
6003 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
6005 if (!is_valid_ether_addr(netdev->perm_addr)) {
6006 e_err("Invalid MAC Address: %pM\n", netdev->perm_addr);
6011 init_timer(&adapter->watchdog_timer);
6012 adapter->watchdog_timer.function = e1000_watchdog;
6013 adapter->watchdog_timer.data = (unsigned long) adapter;
6015 init_timer(&adapter->phy_info_timer);
6016 adapter->phy_info_timer.function = e1000_update_phy_info;
6017 adapter->phy_info_timer.data = (unsigned long) adapter;
6019 INIT_WORK(&adapter->reset_task, e1000_reset_task);
6020 INIT_WORK(&adapter->watchdog_task, e1000_watchdog_task);
6021 INIT_WORK(&adapter->downshift_task, e1000e_downshift_workaround);
6022 INIT_WORK(&adapter->update_phy_task, e1000e_update_phy_task);
6023 INIT_WORK(&adapter->print_hang_task, e1000_print_hw_hang);
6025 /* Initialize link parameters. User can change them with ethtool */
6026 adapter->hw.mac.autoneg = 1;
6027 adapter->fc_autoneg = 1;
6028 adapter->hw.fc.requested_mode = e1000_fc_default;
6029 adapter->hw.fc.current_mode = e1000_fc_default;
6030 adapter->hw.phy.autoneg_advertised = 0x2f;
6032 /* ring size defaults */
6033 adapter->rx_ring->count = 256;
6034 adapter->tx_ring->count = 256;
6037 * Initial Wake on LAN setting - If APM wake is enabled in
6038 * the EEPROM, enable the ACPI Magic Packet filter
6040 if (adapter->flags & FLAG_APME_IN_WUC) {
6041 /* APME bit in EEPROM is mapped to WUC.APME */
6042 eeprom_data = er32(WUC);
6043 eeprom_apme_mask = E1000_WUC_APME;
6044 if ((hw->mac.type > e1000_ich10lan) &&
6045 (eeprom_data & E1000_WUC_PHY_WAKE))
6046 adapter->flags2 |= FLAG2_HAS_PHY_WAKEUP;
6047 } else if (adapter->flags & FLAG_APME_IN_CTRL3) {
6048 if (adapter->flags & FLAG_APME_CHECK_PORT_B &&
6049 (adapter->hw.bus.func == 1))
6050 e1000_read_nvm(&adapter->hw,
6051 NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
6053 e1000_read_nvm(&adapter->hw,
6054 NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
6057 /* fetch WoL from EEPROM */
6058 if (eeprom_data & eeprom_apme_mask)
6059 adapter->eeprom_wol |= E1000_WUFC_MAG;
6062 * now that we have the eeprom settings, apply the special cases
6063 * where the eeprom may be wrong or the board simply won't support
6064 * wake on lan on a particular port
6066 if (!(adapter->flags & FLAG_HAS_WOL))
6067 adapter->eeprom_wol = 0;
6069 /* initialize the wol settings based on the eeprom settings */
6070 adapter->wol = adapter->eeprom_wol;
6071 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
6073 /* save off EEPROM version number */
6074 e1000_read_nvm(&adapter->hw, 5, 1, &adapter->eeprom_vers);
6076 /* reset the hardware with the new settings */
6077 e1000e_reset(adapter);
6080 * If the controller has AMT, do not set DRV_LOAD until the interface
6081 * is up. For all other cases, let the f/w know that the h/w is now
6082 * under the control of the driver.
6084 if (!(adapter->flags & FLAG_HAS_AMT))
6085 e1000e_get_hw_control(adapter);
6087 strncpy(netdev->name, "eth%d", sizeof(netdev->name) - 1);
6088 err = register_netdev(netdev);
6092 /* carrier off reporting is important to ethtool even BEFORE open */
6093 netif_carrier_off(netdev);
6095 e1000_print_device_info(adapter);
6097 if (pci_dev_run_wake(pdev))
6098 pm_runtime_put_noidle(&pdev->dev);
6103 if (!(adapter->flags & FLAG_HAS_AMT))
6104 e1000e_release_hw_control(adapter);
6106 if (!e1000_check_reset_block(&adapter->hw))
6107 e1000_phy_hw_reset(&adapter->hw);
6109 kfree(adapter->tx_ring);
6110 kfree(adapter->rx_ring);
6112 if (adapter->hw.flash_address)
6113 iounmap(adapter->hw.flash_address);
6114 e1000e_reset_interrupt_capability(adapter);
6116 iounmap(adapter->hw.hw_addr);
6118 free_netdev(netdev);
6120 pci_release_selected_regions(pdev,
6121 pci_select_bars(pdev, IORESOURCE_MEM));
6124 pci_disable_device(pdev);
6129 * e1000_remove - Device Removal Routine
6130 * @pdev: PCI device information struct
6132 * e1000_remove is called by the PCI subsystem to alert the driver
6133 * that it should release a PCI device. The could be caused by a
6134 * Hot-Plug event, or because the driver is going to be removed from
6137 static void __devexit e1000_remove(struct pci_dev *pdev)
6139 struct net_device *netdev = pci_get_drvdata(pdev);
6140 struct e1000_adapter *adapter = netdev_priv(netdev);
6141 bool down = test_bit(__E1000_DOWN, &adapter->state);
6144 * The timers may be rescheduled, so explicitly disable them
6145 * from being rescheduled.
6148 set_bit(__E1000_DOWN, &adapter->state);
6149 del_timer_sync(&adapter->watchdog_timer);
6150 del_timer_sync(&adapter->phy_info_timer);
6152 cancel_work_sync(&adapter->reset_task);
6153 cancel_work_sync(&adapter->watchdog_task);
6154 cancel_work_sync(&adapter->downshift_task);
6155 cancel_work_sync(&adapter->update_phy_task);
6156 cancel_work_sync(&adapter->print_hang_task);
6158 if (!(netdev->flags & IFF_UP))
6159 e1000_power_down_phy(adapter);
6161 /* Don't lie to e1000_close() down the road. */
6163 clear_bit(__E1000_DOWN, &adapter->state);
6164 unregister_netdev(netdev);
6166 if (pci_dev_run_wake(pdev))
6167 pm_runtime_get_noresume(&pdev->dev);
6170 * Release control of h/w to f/w. If f/w is AMT enabled, this
6171 * would have already happened in close and is redundant.
6173 e1000e_release_hw_control(adapter);
6175 e1000e_reset_interrupt_capability(adapter);
6176 kfree(adapter->tx_ring);
6177 kfree(adapter->rx_ring);
6179 iounmap(adapter->hw.hw_addr);
6180 if (adapter->hw.flash_address)
6181 iounmap(adapter->hw.flash_address);
6182 pci_release_selected_regions(pdev,
6183 pci_select_bars(pdev, IORESOURCE_MEM));
6185 free_netdev(netdev);
6188 pci_disable_pcie_error_reporting(pdev);
6190 pci_disable_device(pdev);
6193 /* PCI Error Recovery (ERS) */
6194 static struct pci_error_handlers e1000_err_handler = {
6195 .error_detected = e1000_io_error_detected,
6196 .slot_reset = e1000_io_slot_reset,
6197 .resume = e1000_io_resume,
6200 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
6201 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_COPPER), board_82571 },
6202 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_FIBER), board_82571 },
6203 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER), board_82571 },
6204 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_COPPER_LP), board_82571 },
6205 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_QUAD_FIBER), board_82571 },
6206 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES), board_82571 },
6207 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_DUAL), board_82571 },
6208 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571EB_SERDES_QUAD), board_82571 },
6209 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82571PT_QUAD_COPPER), board_82571 },
6211 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI), board_82572 },
6212 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_COPPER), board_82572 },
6213 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_FIBER), board_82572 },
6214 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82572EI_SERDES), board_82572 },
6216 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E), board_82573 },
6217 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573E_IAMT), board_82573 },
6218 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82573L), board_82573 },
6220 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574L), board_82574 },
6221 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82574LA), board_82574 },
6222 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82583V), board_82583 },
6224 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_DPT),
6225 board_80003es2lan },
6226 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_COPPER_SPT),
6227 board_80003es2lan },
6228 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_DPT),
6229 board_80003es2lan },
6230 { PCI_VDEVICE(INTEL, E1000_DEV_ID_80003ES2LAN_SERDES_SPT),
6231 board_80003es2lan },
6233 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE), board_ich8lan },
6234 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_G), board_ich8lan },
6235 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IFE_GT), board_ich8lan },
6236 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_AMT), board_ich8lan },
6237 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_C), board_ich8lan },
6238 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M), board_ich8lan },
6239 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_IGP_M_AMT), board_ich8lan },
6240 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH8_82567V_3), board_ich8lan },
6242 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE), board_ich9lan },
6243 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_G), board_ich9lan },
6244 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IFE_GT), board_ich9lan },
6245 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_AMT), board_ich9lan },
6246 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_C), board_ich9lan },
6247 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_BM), board_ich9lan },
6248 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M), board_ich9lan },
6249 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_AMT), board_ich9lan },
6250 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH9_IGP_M_V), board_ich9lan },
6252 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LM), board_ich9lan },
6253 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_LF), board_ich9lan },
6254 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_R_BM_V), board_ich9lan },
6256 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LM), board_ich10lan },
6257 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_LF), board_ich10lan },
6258 { PCI_VDEVICE(INTEL, E1000_DEV_ID_ICH10_D_BM_V), board_ich10lan },
6260 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LM), board_pchlan },
6261 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_M_HV_LC), board_pchlan },
6262 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DM), board_pchlan },
6263 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH_D_HV_DC), board_pchlan },
6265 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_LM), board_pch2lan },
6266 { PCI_VDEVICE(INTEL, E1000_DEV_ID_PCH2_LV_V), board_pch2lan },
6268 { } /* terminate list */
6270 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
6273 static const struct dev_pm_ops e1000_pm_ops = {
6274 SET_SYSTEM_SLEEP_PM_OPS(e1000_suspend, e1000_resume)
6275 SET_RUNTIME_PM_OPS(e1000_runtime_suspend,
6276 e1000_runtime_resume, e1000_idle)
6280 /* PCI Device API Driver */
6281 static struct pci_driver e1000_driver = {
6282 .name = e1000e_driver_name,
6283 .id_table = e1000_pci_tbl,
6284 .probe = e1000_probe,
6285 .remove = __devexit_p(e1000_remove),
6287 .driver.pm = &e1000_pm_ops,
6289 .shutdown = e1000_shutdown,
6290 .err_handler = &e1000_err_handler
6294 * e1000_init_module - Driver Registration Routine
6296 * e1000_init_module is the first routine called when the driver is
6297 * loaded. All it does is register with the PCI subsystem.
6299 static int __init e1000_init_module(void)
6302 pr_info("Intel(R) PRO/1000 Network Driver - %s\n",
6303 e1000e_driver_version);
6304 pr_info("Copyright(c) 1999 - 2011 Intel Corporation.\n");
6305 ret = pci_register_driver(&e1000_driver);
6309 module_init(e1000_init_module);
6312 * e1000_exit_module - Driver Exit Cleanup Routine
6314 * e1000_exit_module is called just before the driver is removed
6317 static void __exit e1000_exit_module(void)
6319 pci_unregister_driver(&e1000_driver);
6321 module_exit(e1000_exit_module);
6324 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
6325 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
6326 MODULE_LICENSE("GPL");
6327 MODULE_VERSION(DRV_VERSION);